Bioseparation and Bioprocessing : Biochromatography,
Membrane Separations, Modeling, Validation Edited by G. Subramanian
@ WILEY-VCH
Further Reading from WILEY-VCH A Practical Approach to Chiral Separations by Liquid Chromatography Edited by G. Subramanian Fundamentals and Applications, 2nd Edition 1994, 422 pp., hardcover, VCH, ISBN 3-527-28288-2
Process Scale Liquid Chromatography Edited by G. Subramanian 1994, XVI, 225 pp., hardcover, VCH, ISBN 3-527-28672-1
Bioseparation and Bioprocessing Volume I: Biochromatography, Membrane Separations, Modeling, Validation Edited by G. Subramanian
@3WILEYVCH Weinheim New York Chichester Brisbane Singapore Toronto
Ganapathy Subramanian 60 B Jubilee Road Littlebourne Canterbury Kent CT3 lTP, UK
This book was carefully produced. Nevertheless, authors, editor and publisher do not warrant the information contained therein to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate.
Cover illustration: Three-dimensional model of human choriogonadotropin. The model is based on the crystal structure of deglycosylated hCG9 (PDB code lhrp). The protein part of the molecule (ribbon) and the four N-linked carbohydrate chains (spheres) are shown on the same scale. The oligosaccharides are attached to Am52 (top, right) and Am78 (bottom) of the a-subunit (green), and to A d 3 and 30 (top, left) of the P-subunit (blue). The binding region is indicated in red. It should be noted that the spatial orientation of the carbohydrate chains is arbitrarily set as they are not present in the crystal structure. The carboxy-terminal peptide of the P-subunit (amino acid residues 131-145) is not depicted because its 3D-structure could not be deduced from the crystal [Figure reproduced by courtesy of Prof. Dr. P.D. J. Grootenhuis (Dept. of Computational Medicinal Chemistry, N.V. Organon, Oss)]. See also Volume 11, Chapter 5 .
Library of Congress Card No. applied for British Library Cataloguing-in-Publication Data: A catalogue record for this book is available from the British Library Die Deutsche Bibliothek - CIP-Einheitsaufnahme Bioseparation and bioprocessing / ed. by Ganapathy Subramanian. - Weinheim ; New York : Chichester ; Brisbane ; Singapore ; Toronto : Wiley-VCH ISBN 3-527-28876-7 Vol. I. Biochromatography, membrane separations, modeling, validation. - 1998
Q WILEY-VCH Verlag GmbH, D-69469 Weinheim (Federal Republic of Germany), 1998
Printed on acid-free and chlorine-free paper All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form - by photoprinting, microfilm, or any other means - nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not specifically marked as such, are not to be considered unprotected by law. Composition: Hagedorn Kommunikation, D-68519 Viernheim Printing: straws offsetdruck GmbH, D-69509 Mhrlenbach Bookbinding: Wilhelm Osswald + Co., D-67433 Neustadt Printed in the Federal Republic of Germany
Preface
Biotechnology represents the confluence of several disciplines. The European Federation of Biotechnology has defined biotechnology as an integrated use of biochemistry, microbiology and chemical engineering in order to achieve the technological (industrial) application of the capacities of microbes and cultured cells. Thus, to produce purified biologically active components really depends on the effective separation process. Within this versatile area of separation it would be incorrect to claim that this book covers the entire field of separation technology comprehensively; it does not; nor is it intended to be used as a textbook for a specific course. This book is intended to project an overview on selected techniques that are actively applied in the biotechnology industries. Volume 1 of Bioseparation and Bioprocessing is organised into four parts containing seventeen chapters contributed by experienced scientists. The nine chapters in part one gives an overview of different chromatographic methods that are applied in the bioseparation. Membrane technology and its application in the separation of bioactive components are addressed in chapters 10, 11 and 12 (part two). Part three consisting of chapters 13, 14 and 15 deals with modelling aspects as applied in the product separation in the biotechnology, and finally part four deals with validations as applied chromatographic process (chapter 16) and for virus removal is described in the last chapter. It is my hope that this volume will bring together accumulated knowledge in a way which will promote the advancement of separation technology, which will continue to grow and develop on the basis of fascinating discoveries in the control and separation of biomolecules to create technologies that are useful to society. I gratefully acknowledge the authors for their time and motivation in preparing their contributions, without which this volume would not have been possible. I should be most grateful for any suggestions which could serve to improve future editions of this book. Finally I would like to thank the staff of WILEY-VCH for their help. Canterbury, Kent January 1998
G. Subramanian.
Contents
Part One: Biochromatography 1
Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology . . . . . . . . . . . . . . . . . .
3
Roger M . Nicoud
1.1 1.2 1.3 1.3.1 1.3.2 1.3.3 1.4 1.5 1.5.1 1.5.2 1.6 1.6.1 1.6.2 1.6.3 1.6.4 1.6.5 1.6.6 1.7 1.7.1 1.7.2 1.7.3 1.8
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . How many Zones? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Three-Zone Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Five-Zone Scheme .......................................... Conclusion Regarding the Number of Zones .................... Technical Aspects ........................................... Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Equilibrium Adsorption Isotherms ............................. TMB and SMB: Two Equivalent Processes ..................... Main Applications and Developments . . . . . . . . . . . . . . . . . . . . . . . . . . Separation of Sugars ......................................... Desalting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Purification of Proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Separation of Ionic Molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Separation in Organic Solvents ................................ Separation of Optical Isomers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison with Batch Chromatography ....................... Influence of Flow Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Influence of Number of Theoretical Plates ...................... SMB is Continuous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviations and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 4 7 8 9 11 12 16 17 19 25 25 27 28 29 29 31 32 33 34 36 36 37 38
VIII
Contents
2
Systematic Development of Chromatographic Processes using Perfusion Chromatography Technology .......................
41
Scott Fulton and Thomas Londo 2.1 2.2 2.3 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5 2.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Perfusion Chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Principle of Systematic Development ........................... Define . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Evaluate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Implement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Troubleshoot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
Hydrophobic Interaction Chromatography of Proteins
..........
41 42 44 45 50 57 60 63 64 64 65
Eric Grund 3.1 3.2 3.2.1 3.2.2 3.2.3 3.2.4 3.2.5 3.3 3.3.1 3.3.2 3.3.3 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 3.4.6 3.5 3.6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Solvent Additives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . HIC versus Reversed-Phase Chromatography .................... Selectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Purification Strategies ........................................ Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Intermediate Purification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Polishing ................................................... Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stability Window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chromatographic Medium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Binding and Wash Conditions ................................. Elution Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sample Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cleaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scaling-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65 65 66 67 68 69 70 71 73 74 74 75 76 77 81 83 83 84 84 86 87
Contents
4
Displacement Chromatography: Application to Downstream Processing in Biotechnology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IX
89
Ruth Freitag
4.1 4.2 4.3 4.3.1 4.3.2 4.4 4.5 4.5.1 4.6 4.6.1 4.6.2 4.6.3 4.7 4.7.1 4.7.2 4.7.3 4.8
5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Product Isolation in Biotechnology .The Downstream Process . . . . Modes of Chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displacement Chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Use and Advantages of the Displacement Mode in Preparative Chromatography of Biopolymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modeling and Theory of Displacement Chromatography . . . . . . . . . . Displacers for Displacement Biochromatography . . . . . . . . . . . . . . . . . The Rational Design of Protein Displacers ...................... Special Forms of Displacement Chromatography . . . . . . . . . . . . . . . . . Analytical Aspects of Displacement Chromatography . . . . . . . . . . . . . Separation of Isomers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Miscellaneous .............................................. Applications of Displacement Chromatography in Biotechnology . . . Separation and Isolation of Peptides and Antibiotics . . . . . . . . . . . . . . Protein Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References .................................................
Affinity Chromatography
....................................
89 89 91 92
95 96 99 101 102 103 103 104 105 105 106 109 110 110 113
Jim Pearson 5.1 5.2 5.3 5.4 5.4.1 5.4.2 5.4.3 5.4.4 5.4.5 5.4.6 5.5 5.5.1 5.6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Ligand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Particle Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chemical Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Particle Shape and Rigidity ................................... Pore Size and Accessible Internal Volume ...................... Surface Chemistry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Low Non-Specific Binding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ligand Selection and Development ............................ Ligand Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Combinatorial Ligand Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
113 114 115 116 117 117 117 118 119 119 120 120 123 123
X
Contents
6
Large-Scale Chromatography: Design and Operation . . . . . . . . . . . 125 C. J . A . Davis
6.1 6.2 6.2.1 6.2.2 6.2.3 6.3 6.3.1 6.3.2 6.3.3 6.3.4 6.4 6.4.1 6.4.2 6.4.3 6.4.4 6.5 6.5.1 6.5.2 6.5.3 6.6 6.6.1 6.6.2 6.6.3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chromatography Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Automation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Column Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Axial Flow Chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Radial Flow Chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Column Operating Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Operating Philosophies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hygienic Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Column Qualification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Operational Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Economics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bioprocessing Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
125 126 126 126 131 132 132 133 135 136 137 137 138 138 139 139 139 140 140 141 141 142 142 143
7
Radial Flow Chromatography: Developments and Application in Bioseparations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
145
Denise M . Wallworth 7.1 7.2 7.3 7.4 7.5
Radial Flow Column design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Slurry Packing of Radial Flow Columns ........................ Scale-up Using Radial Flow Columns .......................... Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References .................................................
146 148 150 152 156 156
8
Contents
XI
Enhanced Diffusion Chromatography and Related Sorbents for Biopurification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
157
Egisto Boschetti and John L. CofSman 8.1 8.2 8.2.1 8.2.2 8.3 8.3.1 8.3.2 8.4 8.4.1 8.4.2 8.4.3 8.4.4 8.4.5 8.5 8.5.1 8.5.2 8.5.3 8.5.4 8.6 8.6.1 8.6.2 8.6.3 8.7
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of Intraparticle Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pore Diffusion versus Particle Diffusion ........................ Effect of Pore Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Enhanced Diffusion Situation ................................. History of Enhanced Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modeling Enhanced Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chromatographic Media for Enhanced Diffusion . . . . . . . . . . . . . . . . . From a Theoretical Model to Practical Media . . . . . . . . . . . . . . . . . . . The Nature of Porous Rigid Material ........................... The Gel-Filled Porous Structures and their Preparation . . . . . . . . . . . HyperD Microstructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Properties of HyperD Ion Exchangers ..................... Experimental Evidence of Enhanced Diffusion in HyperD . . . . . . . . . Macromolecular Conformation of Hydrogel Network . . . . . . . . . . . . . Batch Uptake Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shallow Bed Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Breakthrough Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Benefits and Applications of Enhanced Diffusion . . . . . . . . . . . . . . . . Protein Capture with Dilute Solutions .......................... Capture with Large Particles .................................. Capture with Dense Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviations and Symbols ................................... References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
157 159 160 161 162 162 164 169 169 172 173 174 175 180 180 183 185 187 189 189 190 191 194 195 196
9
Expanded Bed Adsorption Chromatography
...................
199
Rolf Hjorth. Patrik Leijon. Ann-Kristin Barnfield Frej and Christina Jagersten 9.1 9.1.1 9.1.2 9.1.3 9.2 9.2.1 9.2.2 9.2.3 9.2.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Clarification and Recovery Techniques ......................... Fluidized Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Expanded Bed Adsorption .................................... Theoretical Background for Expanded Bed Adsorption . . . . . . . . . . . Hydrodynamics of Expanded Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Distribution System ......................................... Adsorbent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Evaluation of Bed Stability ...................................
199 199 200 201 202 202 203 204 206
XI1
Contents
9.2.5 9.3 9.3.1 9.3.2 9.3.3 9.3.4 9.3.5 9.4 9.4.1
Scale-up of Expanded Beds ................................... Development and Operation of Expanded Bed Processes . . . . . . . . . . Adsorbents and Equipment for Expanded Bed Adsorption . . . . . . . . . Experimental Strategy for Expanded Bed Adsorption . . . . . . . . . . . . . Operation of Expanded Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Feedstocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cleaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications of Expanded Bed Adsorption ...................... Adsorption of Recombinant Protein Expressed in E . coli Periplasm using STREAMLINE DEAE ......................... Adsorption of Recombinant Protein Expressed in E . coli Cells using STREAMLINE DEAE ............................. Purification of Native gp120 from HIV-Infected T Cells using STREAMLINE SP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Capture of Two Aprotinin variants Produced in Hansenula polymorpha on STREAMLINE SP ............................ Affinity Purification of G6PDH from Unclarified Yeast Homogenate using STREAMLINE Red H-7B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Capture of Humanized IgG4 Directly from the Fermenter using STREAMLINE rProtein A .................................... Process for Purifying Recombinant Human Serum Albumin . . . . . . . Capture of Native Alcohol Dehydrogenase from Baker’s Yeast by Hydrophobic Interaction Expanded Bed Chromatography . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.4.2 9.4.3 9.4.4 9.4.5 9.4.6 9.4.7 9.4.8
208 209 210 212 215 217 218 220 221 222 222 223 223 223 224 224 225
Part Two: Membrane Separations 10
Application of Membrane Bioseparation Processes in the Beverage and Food Industries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 Dan Donnelly. Joe Bergin. Tom Duane and Niall McNulty
10.1 10.2 10.2.1 10.2.2 10.2.3 10.3 10.3.1 10.3.2 10.3.3 10.3.4 10.3.5 10.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Membrane technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pressure-Driven processes .................................... Non-Pressure-Driven Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Membrane Reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Process Conditions for Crossflow Filtration ..................... Membrane Materials ......................................... Membrane Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flux Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Membrane Cleaning and Disinfection . . . . . . . . . . . . . . . . . . . . . . . . . . Catalytic Membrane Reactors for the Food Industry . . . . . . . . . . . . . .
229 230 230 232 234 234 234 237 237 238 238 239
Contents
10.4.1 10.4.2 10.4.3 10.4.4 10.4.5 10.4.6 10.4.7 10.4.8 10.4.9 10.5 10.5.1 10.5.2 10.5.3 10.6 10.6.1 10.6.2 10.6.3 10.6.4 10.7 10.7.1 10.7.2 10.7.3 10.8 10.8.1 10.8.2 10.8.3 10.9
11
XI11
Continuous Stirred Tank Membrane Reactors (CSTMR) . . . . . . . . . . Membrane Reactor Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Enzyme Immobilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Catalytic Membrane Reactors versus Conventional Bioreactor Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Food-Based Applications of Catalytic Membrane Technology . . . . . . Catalytic Membrane Reactors in Dairy Processing . . . . . . . . . . . . . . . Catalytic Membrane Reactors in the Hydrolysis of Fats and Oils . . . Catalytic Membrane Reactors in Fruit Juice Processing . . . . . . . . . . . Catalytic Membrane Reactors in Carbohydrate Processing . . . . . . . . . Applications in the Beer and Alcoholic Beverage Industries . . . . . . . Removal of Alcohol using Reverse Osmosis .................... Removal of Microorganisms by Microfiltration . . . . . . . . . . . . . . . . . . Gas Exchange using Crossflow Filtration ....................... Applications of Membrane Separations in Dairy Processing . . . . . . . Historical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Whey Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ultrafiltration of Milk for Ice Cream Manufacture . . . . . . . . . . . . . . . Microfiltration of Raw Milk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications of Membrane Separations to the Fruit Juice Industry . . Apple Juice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Citrus Juice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tomato Juice Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications of Membrane Separations to Cereal Processing . . . . . . . Corn Syrups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Protein Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Waste Stream Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
240 241 241
Recovery of Biological Products by Liquid Emulsion Membranes
267
243 244 244 246 246 247 248 248 250 253 254 254 255 258 258 259 259 260 261 261 261 262 262 263 264
P. R . Patnaik 11.1 11.2 11.3 11.3.1 11.3.2 11.3.3 11.4 11.4.1 11.4.2 11.5 11.5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Principle and Advantage of LEMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Types of LEMs and their Preparation .......................... Emulsion Liquid Membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Immobilized Liquid Membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contained Liquid Membranes ................................. Types of Separations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Type I . Physical Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Type I1. Facilitated Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Factors Affecting LEM and SLM Performance . . . . . . . . . . . . . . . . . . Stability of Membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
267 268 270 270 271 272 273 273 274 276 276
XIV
Contents
11.5.2 11.5.3 11.5.4 11.6 11.7 11.7.1 11.7.2 11.7.3 11.7.4 11.8 11.8.1 11.8.2 11.9
Membrane Swelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Carrier Concentration and Selectivity .......................... Rate Controlling Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modeling of LEM Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selected Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Separation of L-Phenylalanine and its Derivatives . . . . . . . . . . . . . . . . Extraction of Lactic Acid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Citric Acid Recovery ........................................ Extraction (and Subsequent Reaction) of Penicillin G . . . . . . . . . . . . . Separation of Proteins via Reversed Micelles in LEM Systems . . . . . Factors Affecting Protein Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Economic Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
Membranes Modified for Biochromatography
277 279 280 282 286 286 289 290 292 294 295 298 299 300
. . . . . . . . . . . . . . . . . . 305
Egbert Mutter and Etias Ktein
12.1 12.2 12.2.1 12.2.2 12.3 12.3.1 12.3.2 12.4 12.5 12.6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 Membrane Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 Sheet Membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 Hollow Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 Binding Chemistries ......................................... 307 Differences Between Gel and Membrane Derivatization Chemistry . 307 Different Ligands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310 Kinetic Advantages of Membranes ............................. 316 Packaging of Membranes ..................................... 318 Ion Exchanger Hollow Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Part Three: Modeling 13
Computer Modeling of Chromatographic Bioseparation . . . . . . . . . 329 Andreas Spieker, Ernst Kloppenburg and Ernst-Dieter Gilles
13.1 13.2 13.2.1 13.2.2 13.2.3 13.2.4 13.2.5
Introduction ................................................ Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modeling Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Model Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The l + l d Mass Transfer Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Id Mass Transfer Model ................................. The 1d Equilibrium Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
329 330 330 332 333 340 342
Contents
XV
13.2.6 13.2.7 13.2.8 13.3 13.3.1 13.3.2 13.3.3 13.4 13.4.1 13.4.2 13.4.3
Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simulated Moving Bed Chromatography . . . . . . . . . . . . . . . . . . . . . . . . Equilibrium Isotherms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Empiric Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlations for the Dispersion Coefficient ...................... Correlations for the Mass Transfer Coefficient . . . . . . . . . . . . . . . . . . . Correlations for Diffusivities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . From Modeling to Numeric Simulation . . . . . . . . . . . . . . . . . . . . . . . . . Solution via Semi-discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Numeric Solution of the Discretized System .................... Solution using a Specialized PDAE Solver ...................... Abbreviations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14
Neural Network Applications to Fermentation Processes . . . . . . . . 363
343 345 348 350 351 351 351 352 353 357 358 358 360
P. R. Patnaik
14.1 14.2 14.2.1 14.2.2 14.2.3 14.3 14.3.1 14.3.2 14.3.3 14.3.4 14.3.5 14.3.6 14.4 14.4.1 14.4.2 14.4.3 14.4.4 14.5 14.5.1 14.5.2 14.5.3 14.5.4 14.6 14.7
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363 Structure and Functioning of ANNs ............................ 365 Basic Structure and Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365 The Nature of Input-Output Transformation ..................... 367 Types of Artificial Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 370 Deterministic State Estimation of Bioreactors .................... 373 Penicillin G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Ethanol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376 Glucoamylase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379 Activated Sludge Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381 Recombinant Protein-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383 Recombinant Protein-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 Bioreactor Estimations in the Presence of Noise . . . . . . . . . . . . . . . . . 387 Industrial Mycelial Fermentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 Adaptive Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389 Intrusion of Noise in the Start-up Phase ........................ 392 Radial Basis Network Analysis of Penicillin Fermentation . . . . . . . . 394 Applications to Disturbances and Process Faults . . . . . . . . . . . . . . . . . 395 Supervised Control of Bacillus thuringiensis Fermentation . . . . . . . . 396 Fuzzy Neural Control of Ethanol Production .................... 399 Fuzzy Neural Control of Recombinant E . coli . . . . . . . . . . . . . . . . . . . 401 Diagnosis of Plasmid Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404 Hardware Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
XVI
Contents
15
Advances in Modeling for Bioprocess Supervision and Control. . . 411 Andreas Liibbert and Rimvydas Simutis
15.1 15.1.1 15.1.2 15.2 15.2.1 15.2.2 15.2.3 15.2.4 15.2.5 15.3 15.3.1 15.3.2 15.3.3 15.3.4 15.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 Conceptual Aspects of Practical Modeling ...................... 412 Current State of Process Modeling in Industry . . . . . . . . . . . . . . . . . . . 413 414 Process Models for Typical Applications . . . . . . . . . . . . . . . . . . . . . . . Practical Constraints on the Modeling Procedure . . . . . . . . . . . . . . . . . 414 416 Modeling for State Estimation ................................ Modeling for Process Fault Analysis ........................... 424 Modeling for Process Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433 439 Modeling for Closed-Loop Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . Software Tools for Data Analysis and Modeling . . . . . . . . . . . . . . . . . 445 HardwareBoftware Bases Required ............................ 446 446 Data Exploration Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Modem Modeling Software Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452 Front-End Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453 Conclusions and Recommendations ............................ References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455 457 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Part Four: Validation 16
..........
465
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Strategies for Viral Safety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cell Bank System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Biological Raw Material ..................................... Current Good Manufacturing Practice (cGMP) . . . . . . . . . . . . . . . . . . Virus Clearance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Virus Inactivation ........................................... Virus Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Regulatory Acceptability of Methods for Virus Clearance . . . . . . . . . Calculation of the Clearance Factor ............................ Determination of the Virus Titer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reduction Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Evaluation/Assessment of Methods for Virus Inactivation . . . . . . . . . Virus Inactivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Evaluation/Assessment of Methods for Virus Removal . . . . . . . . . . . . Chromatography ............................................
465 465 465 468 469 469 470 470 470 474 475 476 477 477 485 485
Validation of Viral Safety for Pharmaceutical Proteins Joachim K . Walter, Franz Nothelfer, William Werz
16.1 16.2 16.2.1 16.2.2 16.2.3 16.3 16.3.1 16.3.2 16.3.3 16.4 16.4.1 16.4.2 16.5 16.5.1 16.6 16.6.1
Contents
XVII
16.6.2 16.7 16.8
Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Design of Downstream Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
Validation Issues in Chromatographic Processes . . . . . . . . . . . . . . . . 497
488 494 494 495
Gail Sofer 17.1 17.2 17.2.1 17.2.2 17.3 17.3.1 17.3.2 17.3.3 17.3.4 17.4 17.4.1 17.4.2 17.4.3 17.4.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Small-Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Process Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Small-Scale Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pilot and Full-scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Equipment and Automation ................................... Process Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Validation of Sanitization and Cleaning . . . . . . . . . . . . . . . . . . . . . . . . . Revalidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Process Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chromatography Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Addition. Deletion. or Order of Purification Steps . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
497 498 498 502 505 506 507 508 509 509 509 509 510 510 511 513
Contents
Part One: Processing 1
Strategies in Downstream Processing ..........................
3
Yusuf Chisti
1.1 1.2 1.3 1.3.1 1.3.2 1.3.3 1.3.4 1.4 1.4.1 I .4.2 1.4.3 1.5 1.6 1.7 1.7.1 1.7.2 1.7.3 1.7.4 1.7.5 1.7.6 1.7.7 1.8 1.8.1 1.9
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of Process Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . Product Quality and Purity Specifications . . . . . . . . . . . . . . . . . . . . . . . Endotoxins . . . . . . . . . . . . . . ................ Residual DNA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Microorganisms and Viruses . . . . . . . . . . . . . ................ Other Contaminants . . . . . . . . . . . . . . . . . . . . ................ Impact of Fermentation on Recovery . . . . . . . ................ Characteristics of Broth and Microorganism ..................... Product Concentration . ............................... Combined Fermentationry Schemes . . . . . . . . . . . . . . . . . . . . . Initial Separations and Concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . Intracellular Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Some Specific Bioseparations . . . . . . . . . . . . ................ Precipitation . . . . . . . . . . . ............................. Foam Fractionation . . . . . .................... ....... Solvent Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aqueous Liquid-Liquid Extraction and its Variants . . . . . . . . . . . . . . Membrane Separations . . . . . . . . . . . . . . . . Electrically Enhanced Bioseparations . . . . Chromatographic Separations . . . . . . . . . . . ................ Recombinant and other Proteins . . . . . . . . . . ............. Inclusion Body Proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 4 7 8 8 9 9 10 10 12 13 13 16 17 17 18 18 19 20 21 22 23 25 27 27 28
VIII
Contents
.
2
Protein Stability in Downstream Processing . . . . . . . . . . . . . . . . . . .
31
Kim Hejnaes. Finn Matthiesen and Lars Skriver 2.1 2.2 2.2.1 2.2.2 2.2.3 2.3 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5 2.3.6 2.3.7 2.3.8 2.3.9 2.3.10 2.3.11 2.3.12 2.4 2.4.1 2.4.2 2.4.3 2.4.4 2.4.5 2.4.6 2.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Protein Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Native State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Molten Globule State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Unfolded State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chemical and Physical Instability ....................... Proteolytic Degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N-terminal Degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Non-Enzymatic Hydrolysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Deamidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P-Elimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Racemization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conversion of Arginine to Ornithine ........................... Oxidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cysteinyl and Cystinyl Residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Denaturation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Precipitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Essential Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Effect of pH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Effect of Temperature .................................... The Effect of Redox Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Effect of Co-solvents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Effect of Protein Concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Effect of Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31 32 32 33 34 34 34 36 36 36 39 39 40 41 42 44 45 46 47 47 49 50 55 57 58 58 60 60 61
3
Production of Transgenic Protein . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
Gordon Wright and John Noble 3.1 3.2 3.3 3.3.1 3.3.2 3.3.3 3.3.4 3.3.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of Transgenic Technology ........................... Process Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Animal Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Milking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Milk Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Primary Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Polishing Steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67 67 69 70 70 71 71 72
Contents
IX
3.3.6 3.3.7 3.3.8 3.3.8. I 3.3.8.2 3.3.8.3 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 3.4.6 3.4.7 3.5 3.5.1 3.5.2
Formulation and Filling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viral-Specific Steps . . . . . . . . . . . . . . . . . . ................. Utilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ................. Water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cleaning and Sanitization .............................. Sterilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Facility Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of Facility Design ............................... Animal Housing and Milking ............................... Milk Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Primary Recovery and Polishing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Formulation and Finishing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of Building Finishes and HVAC . . . . . . . . . . . . . . . . . . . . . . Site Planning Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . cGMP and Regulatory Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Good Manufacturing Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Containment of Genetically Modified Organisms (GMO) . . . . . . . . . . References . . . . . . . . . . . ...................................
72 72 72 73
4
Harvesting Recombinant Protein Inclusion Bodies . . . . . . . . . . . . . .
81
74 74 75 75 75 76 76 76 77 77 78 78
Anton P. J . Middelberg and Brian K. O’Neill 4.1 4.2 4.2. I 4.2.2 4.2.3 4.3 4.4 4.4.1 4.4.2 4.4.2.1 4.4.2.2 4.5 4.5.1 4.5.2 4.5.3 4.5.4 4.6 4.6.1 4.6.2 4.6.3 4.6.4 4.6.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . What is an Inclusion Body? . . . ............................ Inclusion Body Formation . . . . . ............................ Inclusion Body Composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Size and Density of Inclusion Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . Properties of Cellular Debris .................................. Process Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Laboratory-scale Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Considerations in Synthesizing a Large-scale Process . . . . . . . . . . . . . Protein Refolding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Inclusion Body Recover .................................. Filtration . . . . . . . . . . . . . .................................. Modes of Filtration . . . . .................................. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Commercial Equipment and Operating Parameters . . . . . . . . . . . . . . . Inclusion Body Recovery by Filtration . . . . . . . . . . . . . . . . . . . . . . . . . Centrifugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modes of Centrifugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Commercial Centrifugation Equipment ......................... Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scale-up and Scale-down of Centrifuges . . . . . . . . . . . . . . . . . . . . . . . . Inclusion Body Recovery by Centrifugation . . . . . . . . . . . . . . . . . . . . .
81 81 82 82 83 84 85 85 86 87 88 90 90 91 92 93 94 94 95 96 98 99
X
Contents
4.7
Alternatives to Inclusion Body Recovery ....................... Abbreviations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
102 104 105
5
The Application of Glycobiology for the Generation of Recombinant Glycoprotein Therapeutics ....................
107
.
Jan B . L. Damm Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Structure and Function of Glycoproteins ........................ Difficulties in Establishing Carbohydrate Structure-Function Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Glycosylation-associated Effects on the Properties of 5.2.2 Glycoprotein Drugs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Glyco-engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Glyco-engineering at the DNA level . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 5.3.1.1 Carbohydrate Structure and Protein Backbone . . . . . . . . . . . . . . . . . . . 5.3.1.2 Choice of the Host Cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1.3 Selection of a Glycosylation Mutant . . . . . . . . . . . . . . . . . . . . . . . . . . . Glyco-engineering at the Biosynthesis Level .................... 5.3.2 Glyco-engineering at the Product Level . . . . . . . . . . . . . . . . . . . . . . . . 5.3.3 Conclusion and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1 5.2 5.2.1
6
The Release of Intracellular Bioproducts ......................
107 110 115 116 118 120 120 121 123 124 125 125 126 127 131
Anton P. J . Middelberg
6.1 6.2 6.2.1 6.2.2 6.2.3 6.2.3.1 6.2.3.2 6.2.3.3 6.3 6.3.1 6.3.2 6.3.3 6.3.4 6.3.5 6.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cell Wall Destruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cell Wall Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Strategies for Cell Disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Quantifying Cell Disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Direct Measurement of Disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Indirect Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting a Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chemical Disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Antibiotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chelating Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chaotropic Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Detergents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Alkaline Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Enzymatic Disruption ........................................
131 131 131 134 135 135 135 136 137 137 138 139 139
.
6.5 6.6 6.6.1 6.6.2 6.6.3 6.6.4 6.6.5 6.7 6.7.1 6.7.2 6.7.3 6.7.4 6.8 6.9 6.9.1 6.9.2
7
Contents
x1
Physical Methods of Cell Disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . High-pressure Homogeni ............... .......... Operational Parameters . ............. ................ Commercial Equipment ...................................... Cell Treatments Before Homogenization ........................ Predicting Disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Importance of Homogenizer Valve Design . . . . . . . . . . . . . . . . . . Bead Milling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Operational Parameters . . . ............................... Commercial Equipment . . ............................... ............................ Predicting Disruption . . . . The Importance of Agitato gn . . . . . . . . . . . . . . . . . . . . . . . . . . . . Other Methods of Mechanical Disruptio .................... .................... Downstream Impacts of Cell Disruption Debris Size Analysis . . . . . . . . . . . . . . . . .................... Predicting Debris Size .............................. Abbreviations and Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
140 141 141 142 142 144 145 147 148 149 149 153 155 157 157
Microcarriers in Cell Culture Produktion . . . . . . . . . . . . . . . . . . . . .
165
162
Bjiirn Lundgren and Geruld Bluml
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Production Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Production Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 Consumable Cost Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.2 ............................. 7.2.2.1 Culture Surface . . . . . . . . . . . . . . . . . . ............................... Serum and Additives 7.2.2.2 Important Developments ......................... 7.2.3 . . . . . . ........................... Microcarrier Background 7.3 Adhesion (Cell-Cell, Cell-Surface) ............................ 7.3.1 Immobilization Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.2 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.3 ...... Size Shape Diffusion Limits ....................... 7.3.4 Specific Density and Sedimentation Velocity . . . . . . . . . . . . . . . . . . . . 7.3.5 Rigidity and Shear Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.6 Porosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.7 Cell Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.8 Microcarrier Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Microcarrier History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.1 Advantages of Microcarriers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.2 Disadvantages of Microcarriers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.3 ............................................. 7.4.4 Scale-up Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.5
165 167 167 167 167 168 169 169 169 170 172 172 175 176 176 178 180 180 181 183
XI1
Contents
Choice of Supplier .......................................... The ‘Ideal’ Microcarrier ...................................... Microcarrier Culture Equipment ............................... Unit Process Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Small-scale Equipment ....................................... Large-scale Equipment - Stirred Tanks (Low Density) . . . . . . . . . . . . Packed Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fluidized Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fluidized Bed With External Circulation . . . . . . . . . . . . . . . . . . . . . . . . Cytopilot - Fluidized Bed With Internal Circulation . . . . . . . . . . . . . . Fluidization and Fluidization Velocity .......................... Culture Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Media and Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . pH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dissolved Oxygen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Redox Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stirring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .............. Control and Feeding Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Microcarriers in Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preparation of Carriers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Microcarrier Concentrations .................................. Inoculum .................................................. Cell Quantification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scale-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Colonization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Suspension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Documentation .............................................. Optimizing Culture Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Trouble-shooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stirred Microcarrier Cultures .................................. Fluidized Bed Trouble-shooting ............................... Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Vaccines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Natural and Recombinant Proteins . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of Carriers in Different Reactors (Packed Bed or Fluidized Bed Reactor) ........................ 7.10.3 Monoclonal Antibodies ...................................... 7.10.3.1 Comparison of Anti-HIV Monoclonal Antibody Productivity with Different Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.10.3.2 Comparison of a Hollow Fiber Reactor with a Fluidized Bed Reactor Potential Future Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 1 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.4.6 7.4.7 7.5 7.5.1 7.5.2 7.5.3 7.5.4 7.5.5 7.5.5. I 7.5.5.2 7 S.5.3 7.6 7.6.1 7.6.2 7.6.3 7.6.4 7.6.5 7.6.6 7.7 7.7.1 7.7.2 7.7.3 7.7.4 7.7.5 7.7.5.1 7.7.5.2 7.7.5.3 7.7.5.4 7.8 7.9 7.9.1 7.9.2 7.10 7.10.1 7.10.2 7.10.2.1
186 186 187 187 188 190 191 193 193 194 195 196 196 197 197 198 198 199 200 200 200 201 203 204 205 207 208 208 209 210 210 212 212 215 216 217 218 218 219 219 220
Contents
8
XI11
Purification and Characterization of Monoclonal Antibodies . . . . . 223 Paul Matejtschuk. Rose M . Baker and George E . Chapman
8.1 8.2 8.3 8.3.1 8.3.1.1 8.3.1.2 8.3.1.3 8.3.1.4 8.3.1.5 8.3.2 8.3.2.1 8.3.2.2 8.3.2.3 8.3.3 8.3.3.1 8.3.3.2 8.3.3.3 8.3.4 8.4 8.4.1 8.4.2 8.4.2.1 8.4.2.2 8.4.2.3 8.4.2.4 8.4.2.5 8.4.2.6 8.4.2.7 8.4.2.8 8.4.2.9 8.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Methods of MAb Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Purification . . . . . . . . . . . ................................. Initial Considerations . . ................................. Intended Use . . . . . . . . . ................................. Culture Method . . . . . . . ................................. Contaminants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Purification Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Clarification/Concentration ................................... Chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Other Purification Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Large Scale ......................................... Scale-up . . . . . . . . . . . . . ................................. GMP and Validation . . . ................................. Control and Automation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future DeVdQpPAetltS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Need for Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aspects of MAb Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Primary Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sequencing Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Peptide Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mass Spectrometry of MAbs . . . . . . . . . . . . . . . . . . . . . . . . . . . C-Terminal Sequencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Secondary and Higher Structure ............................... Functional Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Glycosylation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
223 224 225 225 225 226 226 227 228 228 230 231 233 233 235 235 236 236 236 237 237 238 239 241 242 242 243 246 247 248
XIV
Contents
.
Part Two: Quality and Characterization 9
Biological Standardization of Interferons and Other Cytokines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
255
Anthony Meager
9.1 9.2 9.2.1 9.2.2 9.3 9.3.1 9.3.2 9.3.3 9.4 9.5 9.6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interferons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Background Information: Definitions. Designations and Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Characteristics of IFNs for Clinical Use ........................ Interferon Standardization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic Principles of Biological Standardization . . . . . . . . . . . . . . . . . . . Bioassays for IFNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Design of Bioassays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interferon Standards . . ................................... Cytokine Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
257 257 258 260 260 262 265 266 270 272 272 272
The Strategic Role of Assays in Process Development: A Case Study of Matrix-Assisted Laser Desorption Ionization Mass Spectroscopy as a Tool for Biopharmaceutical Development 275
T. J . Meyers. P. G. Varley. A . Binieda. J . A . Puwis and N . R . Burns 10.1 10.2 10.2.1 10.2.2 10.2.3 10.3 10.3.1 10.3.2 10.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Materials and Methods . . . . . . . . . . . . . . . . .................... Validation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . In-process Analysis . ...................................... Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Identity Test - Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications In-Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
275 276 277 278 278 278 278 280 290
Contents
11
XV
Quality Control of Protein Primary Structure by Automated Sequencing and Mass Spectrometry . . . . . . . . . . . . . 291 Philip J . Jackson and Stephen J . Bayne
11.1 11.2 11.2.1 11.2.2 11.2.2.1 11.2.2.2 11.2.2.3 1 1.2.2.4 11.2.3 11.3 11.3.1 1 1.3.2 11.3.3 11.4 11.4.1 11.4.2 11.4.3 11.4.4 11.5 11.6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Automated Edman Degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chemistry .................... .................... Sample Preparation . . . . . . . . . . . . ......................... Compatibility with Edman Chemistry . . . . . . . . . . . . . . . . . . . . . . . . . . Manipulation of Samples in Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . Polyacrylamide Gel Electrophoresis (PAGE) . . . . . . . . . . . . . . . . . . . . Covalent Immobilization . ................................ Data Analysis . . . . . . . . . . ................................ Carboxy Terminal Analysis ........................... Automated Chemical Degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Enzymatic Degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mass Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FAB-MS . . . . . . . . . . . . . . . . . . . . . . . . . . ..................... PDMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... ESMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............ MALDI-MS . . . . . . . . . . . . . . . . . . . . . . . Protein Fragmentation . . . . . . . . . . . . . . Summary and Future Prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . ..................................
12
General Strategies for the Characterization of Carbohydrates from Recombinant Glycoprotein Therapeutics . . . . . . . . . . . . . . . . . 325
291 291 292 296 296 298 299 300 301 306 307 308 309 310 313 314 315
321 321 322
Gerrit J . Gerwig and Jan B. L. Damm 12.1 12.2 12.2.1 12.2.2 12.2.3 12.2.4 12.2.5 12.2.6 12.3 12.3.1 12.3.2 12.3.3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Functions of Glycoprotein Glycans . . . . . . . . . . . . . . . . . ....... Transport. Stabilizing. Protecting and Structural Functions . . . . . . . . Storage Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Masking Functions . . . . ................................... Receptor Functions . . . . . . . . ....................... Regulation of Clearance ......................... Tuning of Biological Activity . ....................... Types of Carbohydrate Chains in Glycoproteins . . . . . . . . . Structure of N-linked Carbohydrate Chains . . Structure of 0-linked Carbohydrate Chains . . . . . . . . . . . . . . . . . . . . . Structure of Glycosylphosphatidylinositol Anchors . . . . . . . . . . . . . . .
325 325 326 327 327 328 328 328
333 334
XVI
Contents
12.4 12.4.1 12.4.2 12.4.3 12.5 12.5.1 12.5.2 12.5.3 12.5.4 12.6 12.7
Biosynthesis of Glycoprotein Glycans .......................... N-linked Carbohydrate Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0-linked Carbohydrate Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Glycosylphophatidylinositol Anchors . . . . . . . . . . . . . . . . . . . . . . . . . . . Analysis of Glycoprotein Glycans ............................. General Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Release of Carbohydrate Chains from Proteins . . . . . . . . . . . . . . . . . . Isolation and Fractionation of Oligosaccharides . . . . . . . . . . . . . . . . . . Structural Characterization of Oligosaccharides . . . . . . . . . . . . . . . . . . Oligosaccharide ProfilingMapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . Structural Analysis of N- and 0-linked Glycans of Recombinant Human Erythropoietin (rhEPO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Liberation of the N-linked Carbohydrate Chains . . . . . . . . . . . . . . . . . Liberation of the 0-linked Carbohydrate Chains . . . . . . . . . . . . . . . . . Fractionation and Structural Determination of the 0-Glycans . . . . . . Fractionation and Structural Determination of the N-Glycans . . . . . . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.7.1 12.7.2 12.7.3 12.7.4 12.7.5
.
334 335 337 338 339 339 342 346 348 356 359 359 359 361 361 366 367 367 369
Part Three: Economics. Safety and Hygiene 13
Biosafety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
379
Yusuf Chisti
13.1 13.2 13.2.1 13.2.2 13.3 13.4 13.4.1 13.4.2 13.4.3 13.4.3.1 13.4.3.2 13.4.3.3 13.4.4 13.4.4.1 13.4.4.2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Risk Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Recombinant Microorganisms ................................. Animal Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Containment Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Risk Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Biological Safety Cabinets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Spill Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Buildings and Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Air Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Construction, Finishes and Practices ........................... Process Equipment .......................................... Fermentation Plant .......................................... Downstream Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
379 379 381 382 383 388 389 393 394 395 395 397 399 400 406
Contents
XVII
13.4.4.3 13.4.5 13.4.6 13.4.7 13.4.8 13.5
Other Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Personnel Protective Equipment ............................... Personnel Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Medical Surveillance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Biowaste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
408 408 409 409 410 411 411 412
14
Process Hygiene in Production Chromatography and Bioseparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
417
Glenwyn D. Kemp 14.1 14.2 14.3 14.4 14.4.1 14.4.2 14.4.2.1 14.4.2.2 14.4.3 14.4.4 14.4.5 14.4.6 14.4.7 14.4.8 14.5 14.5.1 14.5.2 14.5.3 14.5.4 14.5.5 14.5.6 14.5.7 14.5.8 14.6 14.6. I 146.2 14.6.3 14.6.4 14.7 14.7.1 14.7.2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . General Principles . . . . . ................................... ............................... Definitions . . . . . . . . . . . Possible Contaminants . . . . . . . . . . . . . . . . . . . . . . . . . . . Active Ingredients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bacteria (Vegetative) ...................................... External Infection . . ...................................... Internal Infection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bacteria (Spores) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FungiIAlgae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............... Viruses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... Endotoxins (Pyrogens) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............................ Cleaning Reagent . . . . . Non-specifically Bound Protein ............................... Design Considerations in System Hygiene ...................... General Materials of Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stainless Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rorosilicate Glass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .............................. Polypropylene . . . . . . . . . . . . .............................. Acrylic (Plexiglass) . . . . . . . . TPX (Polymethyl Pentane, PMP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PVC (Tygon) . . . . . . . . . . ...................... Fluoropolymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Elastomeric Materials (Seals) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . EPDM (Ethylene Polypropylene) . . . . . . . . . . . . . . . . . . . Santoprene (Norprene, Marprene) . . . . . . . . . . . . . . . . . . . Silicone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Elastomeric PTFE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mechanical Construction .................................. Connections and Seals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pipework SpoolsNalves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
417 417 419 420 420 420 421 421 421 422 422 422 423 423 424 425 426 426 426 427
428 428 428 428 429
XVIII
Contents
Column Seal Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............ Space-filling Seals ............................ ............ Distribution Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . Single Port . . . .......................................... Multi-Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Expanded Bed Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Closed System Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bubble Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . CIP Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Challenge Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . CIP Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........................................... CIP Cocktails . Concentrated Salts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sodium Hydroxide (NaOH) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Organic Solvents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thiomersal (Thimerosal, Merthiolate, Ethyl-mercurithiosalicylate) . . Chlorhexidine (Chlorhexidine Digluconate, Hibitane, (1,6-Di(4 -chlorophenyl-diguanido)hexane) ...................... 14.8.3.6 Detergents/Solvents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.8.3.7 Acidic Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.8.3.8 Oxidizing Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.8.4 Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.7.3 14.7.3.1 14.7.4 14.7.4.1 14.7.4.2 14.7.5 14.7.6 14.7.7 14.8 14.8.1 14.8.2 14.8.3 14.8.3.1 14.8.3.2 14.8.3.3 14.8.3.4 14.8.3.5
15
431 431 434 434 434 435 436 437 441 441 441 441 442 442 443 443 443 443 444 444 445
Strategies and Considerations for Advanced Economy in Downstream Processing of Biopharmaceutical Proteins . . . . . . . . . . 447 Joachim K . Walter
15.1 15.2 15.3 15.3.1 15.3.2 15.3.3 15.3.4 15.4
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Economic Potential in Downstream Processing . . . . . . . . . . . . . . . . . . Strategic Development of Unit Operations ...................... Technical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Raw Materials and Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Process Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Process Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Economy in Process Completion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
447 447 449 449 450 451 454 457 460
Part One Biochromatography
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology Roger M. Nicoud
1.1 Introduction Many different products are now purified by chromatographic processes, from the laboratory scale (several grams) up to the industrial pharmaceutical scale (a few tons per year) or even up to the ‘petrochemical scale’ (100000 T/year). Among the possible technologies, elution HPLC (sometimes with recycling) has become a very important part of the small-scale (1 0 T/year) market during the previous decade. Meanwhile, Simulated Moving Bed (SMB) technology has been used extensively for the very large-scale fractionations of sugars and xylenes during the past 30 years [1,21. At present, there is considerable interest in the preparative applications of liquid chromatography, even if chromatography is still too often considered as expensive. In order to make chromatographic process more attractive, attention is focused on the choice of the operating mode [3], in order to minimize eluent consumption and to maximize the productivity which is of key importance when expensive packings are used. Among the alternatives to the classical process (elution chromatography), much attention is paid to the SMB. SMB technology was introduced in the late 1950s [4] and has mainly been applied to very large-scale productions in the petrochemical and sugar industries [2]. Although SMB was recognized as a very efficient technology leading to a lower eluent consumption, it was strictly ignored in the field of fine chemistry and pharmaceuticals. This was probably due to the patent situation and the complexity of the concept. Only a short time ago, were separations of pharmaceutical compounds first performed using SMB technology [5,6] and other application areas have now been opened, such as fine chemistry, and the cosmetics and perfume industries [7]. SMB is now considered as a genuine production tool (for example the Belgium pharmaceutical company UCB Pharma recently announced its use for performing multi ton scale separation of optical isomers). More over, in addition to the huge commercial plants offered for petrochemistry (IFP, Rueil-Malmaison, France; UOP, Des Plaines, IL), smaller plants are now commercially available for fine chemistry, pharmaceutical industry or biotechnological processes. As described in the following text, even if SMB is still not used extensively in biotechnologies, it may lead quickly to very attractive applications.
4
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
1.2 Basic Principle In contrast to normal (elution) chromatography, SMB is a continuous process and is thus much more adapted to large-scale production. Moreover, SMB generally leads to a much lower eluent consumption - a second key advantage. The reason why elution chromatography leads to dilution can be explained through a simple example. Let us assume that a binary mixture A+B is injected periodically at the inlet of a chromatographic column. This implementation called elution chromatography is often used and leads to chromatograms similar to those given in Fig. 1-1. When looking at the outlet concentration variations, one can think that the system is really well optimized because no dead time ocurs between successive peaks. However, when looking at the internal profiles at a given time (T = 5 for instance), things appear completely different. First, some significant zones in the column are completely free of solutes or contain only product (A or B), and are thus completely unefficient for separation. Finally, a limited fraction (10% for the example of Fig. 1-1) of the column contains both products and is thus efficiently used. Consequently, it appears clearly that the packing utilization is to be improved. A possibility consists in using a counter-current contact between the solid and the liquid, in a system called True Moving Bed (TMB) as described in Fig. 1-2. The solid phase goes down in the column due to gravity; when it exits the system (zone I), it does not contain adsorbed products and is thus recycled at the top of the system (Zone IV). The liquid (eluent) stream follows exactly the opposite direction: it goes up and is recycled from zone IV to zone I. Feed, containing components A and B is injected at the middle of the column, and the fresh eluent at the buttom. Provided that the affinity of A and B for the solid are different (B being more retained than A), it is possible to choose flow rates in order to make A move upward and B move downward, thus leading to a spatial separation. This system requires two inlet lines (one for the feed and one for the eluent) and two outlet lines (one for the raffinate A and one for the extract B).
3
u
8.00
12,oo
~
4,OO
6,OO
2,oo
3,OO
r a
:
%
0,oo
0,oo
5,OO
10,oo
Dimensionless time
15,OO
0
100
200
300
400
Number of plates
Fig. 1-1. Chromatograms (left) and internal profiles (right) calculated for a linear system. The retention factors are RA = 1.5 and i ? = ~2. The column is equivalent to 400 theoretical plates.
1.2 Basic Principle Solid
Liquid
5
Raffnate (A)
A
Extract (B)
Fig. 1-2. True Moving Bed: two equivalent representations.
The classical moving bed comprises four different zones (warning: the zones definition varies according to the authors), in which different constraints must be fulfilled [8]:
I (between the eluent make-up and the extract points): the more retained product (B) must be completely desorbed. - Zone I1 (between the extract and the feed points): the less retained product (A) must be completely desorbed. - Zone I11 (between the feed and the raffinate points): the more retained product (B) must be completely adsorbed. - Zone IV (between the raffinate and the eluent make-up points): the less retained product (A) must be completely adsorbed. - Zone
All the internal flow rates are related to the inledoutlet flow rates by simple mass balances: QII
QIV
= QI - Q E ~ ~
= QIII -
QIII
QRaff
QI
= QII iQFeed
= QIV + QEI
And the inledoutlet flow rates are related by:
In fact, it is extremely difficult to operate a TMB because it involves circulation of a solid adsorbent: this interesting idea must consequently be implemented in a different way. As it will be shown, most of the benefit of a true counter-current operation can be achieved by using several fixed-bed columns in series and an appropriate shift of the injection and collection points: this is the SMB concept.
6
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
To understand the SMB concept, an easy way is to consider a modified moving bed system (Fig. 1-2) for which the solid and the solid are counter-currently contacted in a circular column. The two systems presented in Fig. 1-2 are clearly equivalent. A third equivalent implementation is given in Fig. 1-3: the solid included in the circular system is now kept fixed, but the inletloutlet lines are continuously moving. With this implementation, the solid is still moving with respect to the inlet/outlet lines. As it is extremely difficult to move continuously the inletloutlet lines, they are moved step by step, between a given number of fixed columns (Fig. 1-3). This mode is called Simulated Moving Bed. The solid is no longer moving; its flow is only simulated by shifting inlet and outlet lines. In fact, this simulated solid flow rate downward is directly linked to the shift period. The key is the proper selection of the flow rates: they must be chosen in order to stabilize the B front between zones I and 11, the A front between zones I1 and IV, and to allow separation between zones I1 and 111. The adequate choice of the flow rates requires a minimum knowledge of the physico-chemical properties of the system (adsorption isotherms, plate number) as presented later. Table 1-1 gives the relation between a SMB and its corresponding TMB [9]. V, is the volume of one SMB column. The inletloutlet flow rates of a SMB and its corresponding TMB are identical. A SMB does not exactly work in steady state but in periodic steady state: during a given period, the internal concentration profiles vary, but the internal profiles examined at the same time of two successive periods are identical (except for a one-column translation). The rules used to calculate the SMB flow rates from the TMB flow rates simply mean that the velocity of the liquid relatively to the solid is kept constant.
\L
Extract (B)
\L
Extract(B)
Fig. 1-3. Principle of the Simulated Moving Bed Left: concept (inlet'outlet lines are shifted continuously), Right: implementation (inlet/outlet lines are shifted step by step).
1.3 How Many Zones?
7
Table 1-1. Relation between a SMB and its corresponding TMB. TMB
SMB
Steady state
Periodic steady state
Solid flow rate
Periodic shift of the injection/collection lines (1 - E ) ' v, AT =
ni
~
Internal flow rates
QPb
ni
Internal flow rates
k = I, 11, 111or IV
Eluent, extract, feed, raffinate flow rates QLYb I QLb:
I Qbmb/
Qk$
Eluent, extract, feed, raffinate flow rates QkTb I Qkb:
I Q;mb I Qk:'
1.3 How Many Zones? A part of this paragraph has been taken from Hotier [lo]. Although the most widely used scheme among SMB separation processes is the four-zone scheme as described earlier, there are alternative schemes which are more suited to particular cases [10,11]. Those cases are related to binary or ternary separations. We shall not describe in this paper complex SMB processes where six zones (or even more) are involved, but will restrict our discussion to the threezone and to the five-zone SMB processes applied to binary separations or pseudobinary separations. In order to begin the discussion, let us consider the minimum number of zones required to possibly achieve a separation. In fact, only two zones are necessary to achieve separation of a binary mixture; if B is the most adsorbed component and A the least adsorbed one, we need (Fig. 1-4): 1) Before the feed injection point (with reference to the fluid flow direction) a zone where component A is desorbed, i. e. at the beginning of this zone the component A is completely removed from the solid and, as a consequence, from the liquid in equilibrium with that solid; this zone is known as zone 2 in the four-zone SMB process. 2) After the feed injection point (with reference to the fluid flow direction) a zone where the component B is adsorbed, i. e. at the end of this zone no trace of component B may be found in the liquid and as a consequence in the solid in equilibrium with that liquid; this zone is known as zone 3 in the four-zone SMB process.
With such a system A can be recovered in the liquid stream at the outlet of zone 111, but it may be considered how component B is recovered and how the solid sorbent is regenerated? Obviously, we have to remove component B adsorbed on the sorbent by
8
1 Simulated Moving Bed (SMB):Some Possible Applications for Biotechnology Solid
Fresh solid
Zone 111
B is adsorbed (goes down)
Zone I11
Rafllnate (A)
A+B
A is desorbed (goes up)
Zone I1
Zone I1
Zone I
Solid containing H
I
i t 4
cltie,ic
Fig. 1-4. Equivalent true counter-current. Left: the origin of the separation (two-zone process). Right: the simplest possible process (the three-zone process).
any known means: for example, by increasing temperature or partial pressure, or by concentration decrease or displacement. As there is no reason to perform this operation in the batch mode, it is preferred to continue with the counter-current mode. Thus, a third zone in which component A is desorbed from the sorbent is necessary; this is known as zone 1 in the four-zone SMB process.
1.3.1 Three-Zone Scheme In the three-zone SMB process (Fig. 1-4), zone IV no longer exists, and the flux exiting zone I11 is sent to the concentration unit. The three-zone SMB process is an economic alternative to the four-zone variety when eluent consumption is not an important concern and/or when the raffinate is a zero- or a low-value by-product. With an example taken from the literature [12], we will identify the main drawback of this three-zone SMB separation process. The authors have carried out experiments on SMB separation of three carbohydrate mixtures (i. e. fructose dextran, raffinose dextran, and fructose-raffinose) with both the usual four-section scheme and the minimum three-section scheme. Their main conclusion is that the three-zone scheme demands necessarily a higher eluent consumption than the conventional four-zone scheme; more precisely, they have found that the three-zone scheme demands 2.5 to 9 times more eluent than the four-zone scheme. In order to analyze these two SMB schemes, they consider an equivalent true counter-current system between the liquid and the solid. This latter creates an equivalent liquid counter-flow, that they deduce from the actual flow rates. This representation, which is fully valid for the four-zone scheme, has been challenged by [lo] and appears to be only partially valid in the three-zone scheme. Finally, for a given solid flow rate (thus roughly speaking for a given system size), the internal flow rates are simply related by:
1.3 How Many Zones?
Q, (3-zones) = Q, (4-zones) Q,, (3-zones) = QII (4-zones) QrIr(3-zones) = QIII(4-zones)
9
(3)
the feed and extract concentrations are identical between the three-zone and the fourzone schemes, the raffinate flow rates being related between the two processes by:
QRaf(3-zones) = QRaf (4-zones)
+ QIv (4-zones)
(4)
Consequently, the raffinate obtained with the 3 -zone process is more diluted:
Craf(3-zone) =
Qraf
Qraf(4-zone)
(4-zone)
+ Q I V (4-zone)
. Craf(4 -zone)
(4bis)
In fact, this dilution phenomenon is partially avoidable in a three-zone SMB separation process when a simulated counter-current is employed instead of a true countercurrent; when port motion occurs a fully rinsed column (thus full of eluent) is added to the end of zone 111: indead, during a first part of the switching period, pure eluent will emit from the raffinate outlet. This may of course be discarded or recycled to the eluent tank. During a second part of the switching period, component A diluted by eluent will be recovered. Even with such product splitting at the raffinate outlet, it is clear that for dispersion reasons a dilution effect will occur, which can be avoided in the four-zone scheme. From this discussion the role played by zone IV in the conventional SMB scheme becomes obvious: it is possible to recover fully the eluent contained in the bed porosity of the column that is switched from zone I to zone IV; it may thus be considered as an eluent saver. Nevertheless, there are cases where the advantages of this three-zone scheme (i. e. decreasing the overall length of the system, avoiding a recycling pump operated with four different flow rates), overcome the additional eluent consumption. For example, when the eluent is water (and is thus inexpensive), or when the packing is very expensive (to save one zone can lead to significant cost savings). These situations may happen in real cases, for example when removing salts from biological products, or sucrose from molasses.
1.3.2 Five-Zone Schemes There are basically three main reasons for adding a fifth zone in an SMB: 1. Recovery of a third fraction (option 1). 2. Removal of strongly retained products (option 2). 3. Improvement of one outlet purity (usually the extract) (option 3).
10
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
Solid
Solid
Liquid
Liquid
Zone V Zone IV Zone I11
Zone I1
Zone I
t
Eluent 2
Fig. 1-5. Different 5-Zones schemes. Left: option 1; Middle: option 2; Right: option 3.
The different equivalent moving beds associated to the previous objectives are presented in Fig. 1-5.
1.3.2.1 Option 1 On the basis of its principles, SMB is basically a binary separator. The addition of a fifth zone allows to obtain three fractions, two being pure, one containing two products. For instance, starting from a feed containing three products A, B and C (A is less retained than B, B less retained than C), the flowsheet given in Fig. 1-5 would allow us to obtain: two raffinate lines, one containing pure A, the second one a mixture A+B; and one extract line containing pure C.
1.3.2.2 Option 2 It appears sometimes that the feed contains essentially two products, A and B, and some strongly retained impurities. By adding a fifth zone which could be eluted by a second eluent, the process described in Fig. 1-5 (option 2) allows simultaneous performare of the separation between A and B and removal of the strongly retained impurities. A similar process has been described by Ganetsos and Barker [13].
1.3 How Many Zones?
11
1.3.2.3 Option 3
Let us now examine what is the advantage of introducing a fifth zone in an SMB process for purely binary separations. Theoretically, there is no need for such addition: when the flow rates in zone I1 and zone 111, as well as the switching period (the solid flow rates) are properly tuned, the separation can be theoretically perfect. In practice we have to deal with real systems and two effects may prevent the system to reach extreme purities (i. e. greater than 99.5 %). The first one is related to competitive adsorption: in some cases for low concentration of one of the two species compared with the other, the selectivity disappears or may even be reversed: this is equivalent to a ‘pinch’ or an azeotrope in distillation. The second one is tied to the technology of the SMB separation equipment: in between each pair of column four lines are connected to the intercolumn line. When the extract line is opened there may be some flooding that cannot be neglected from the three other lines (i.e. raffinate, solvent, feed) even if they are tightly closed due to the flow in the intercolumn line. For these two reasons when extreme purity and very high throughput are required at the same time, it may be necessary to add a fifth stream and consequently a fifth zone in the system. We will explain how this fifth zone operates when it is introduced between zone I and zone I1 in order to obtain extreme purity of the extract (or component B), but it must be understood that this could also be done for the raffinate. Part of the product obtained at the outlet of the extract stream, where component B is separated from the eluent (for example by distillation), is reinjected after the extract withdrawal point (with reference to the flow direction). The consequence is a drastic modification of the concentration profile: the concentration of component B between the extract withdrawal point and the feedstock withdrawal point rises to very high values (for example about 90%) when the flow rates in zone I and zone IT are properly tuned. The effects are the following: 1) the impurity (component A) is more diluted with regard to component B; 2) component B elutes component A generally better than the eluent does; it may be considered as displacement chromatography. Discussions regarding the advanages of this implementation can be fund in [lo] and [ 2 ] .
1.3.3 Conclusion Regarding the Number of Zones The four-zone SMB process scheme is generally the best and must be investigated first for binary separations. However, in several instances alternative schemes offer a better technical solution. The three-zone SMB is to be considered when the eluent consumption is not a limiting factor, while the five-zone SMB is interesting for multicomponent separation or when extreme purity of one of the two effluents is required, with a maximum unit throughput.
12
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
1.4 Technical Aspects A Simulated Moving Bed (whatever the number of zones) comprises a given number of columns (usually between four and 24), pumps (usually three to five) and valves allowing to connect the different flux between the columns. In the following, we will restrict our analysis to the classic four-zone SMB. There are different ways to connect columns in order to build a SMB. One important option is linked to the presence or absence of a recycling pump as explained in Fig. 1-6. The most classic way is to use a recycling pump located between two columns (for instance 12 and 1) as explained in Fig. 1-6. As the recycling pump is fixed with respect to the columns, it moves with respect to the zones and is alternatively located in zones IV, 111, 11, and 1. As the flow rates in the different zones are different, the pump flow rate varies from cycle to cycle. It should be pointed out that this constraint is perfectly mastered and that this design is relatively simple and is used with small differences on all the large-scale units. A more serious limit to this implementation is due to the volume of the recycling pump and associated equipments (flowmeter, pressure sensors): as the pump moves with respect to the zones, its volume leads to a dyssymetry which can in turn lead to a decrease in purities. This decrease, which is especially important for short columns and/or low retention, can be significant, but can be perfectly and easily overcome by the use for example of a shorter column, asynchronous shift of the inlets/outlets [14]. This last solution is extremely efficient and does not induce extra costs because it is a purely software solution. In a different implementation, the recycling pump is fixed with respect to the zones, and always located between zones IV and I where no solutes are present. In order to implement this idea, additional valves are needed, which makes the system more complex than the previous one. Its main interest is found when physical modulation is used, as in the supercritical fluid SMB for which it can be shown that a great interest could be taken from a higher pressure in zone I [15]. The only way to obtain this results is to maintain the recycling pump immediately before zone I. A final possibility is to use the eluent pump instead of the recycling pump as explained in Fig. 1-5(c). Even if it appears simpler, this implementation requires more valves than option (a), and has the drawback to recycle one outlet to the eluent tank. Moreover, whichever design is retained (type a, b or c), there are always different options to control the outlet flow rates: they can be controlled via a pump or via an analogic valve, via a flowmeter, or pressure control. Our experience led us to advise a system using a recycling pump which is fixed with respect to the columns (reliability, minimum number of valves), to use pumps instead of valves to deliver the flow rates, and to counterbalance the effect of the recycling pump by using an asynchronic shift of the inlet/outlet lines. A simplified flowsheet of a 12-columns unit built according to this concept is given in Fig. 1-7. At the inlet of each column, four valves (F, E, Ex, R) allow the connection to Feed, Eluent, Extract, or raffinate pumps.
1.4 Technical Aspects
13
10
Extract
I > Raftinate
<
11
5
I 12
Feed 7
Extract
<
k3;;J RECYCLE PUMP
> Rafiinate
Eluent Feed 7
Extract
> Rafflnate
< 5
3
+
Fig. 1-6. Different SMB implementations. (a) With a recycling pump fixed with respect to the columns. (b) With a recycling pump wich is fixed with respect to the zones (between zone IV and zone I). (c) Without recycling pump.
14
I Simulated Moving Bed (SMB): Some Possible Applications .for Biotechnology
7
8
~
R7 Ex7 E7
F7
I
I
I
FlOElOExlORlO
12
11
10
9
I
I
I l l 1
FIZ E l 2 Ex12 R1Z
1 , , , F4v , FfEqfYRf
LJ F5 E5 Ex5 R5
F3 E3 Ex3 R3
m
RECYCLE PUMP
FI E l Ex1 RI
Fig. 1-7.Possible flowsheet of a 12-column SMB.
For example, on the commercially available LICOSEP-LAB (Fig. 1-8), which is a versatile pilot plant, the following technical choices have been made: -
Maximum pressure, temperature: 100 bar, 60 "C
- Columns: 12 columns of 26 mm I.D. or 10 columns of 50 mm I.D. (high versatility).
- Eluent, extract, feed, raffinate and recycle pumps: two heads high-pressure piston pumps allowing maximum flow rate of 200 mL min-' (800 mL min-' for the recycle pump). - Valves: 48 high-pressure two-way valves. - Extract flow rate: delivered by a pump but controlled via the pressure measured immediately before the recycling pump. As stated before, a Moving Bed or Simulated Moving Bed is usually made of four different zones designated as:
-
zone zone zone zone
I: 11: 111: IV
anything anything anything anything
between between between between
Eluent line and Extract line. Extract line and Feed line. Feed line and Raffinate line. Raffinate line and Eluent line.
For instance, let us assume that at a given time: feed is injected at position 3, eluent at position 9, extract recovered at position 1, and raffinate at position 5. The configuration of the system is thus: F3 R5 E9 Ex12 and the zones are given in Fig. 1-9. The zone definition is important in order to understand how the recycle pump flow rate is chosen. Let us define the recycle flow rate (QR)as the flow rate occuring in zone I (arbitrary definition). During the separation process, flow rates in zone I, 11,
15
1.4 Technical Aspects
Fig. 1-8. LICOSEP-LAB (NOVASEP).
111, and IV must be kept constant (analogy with the TMB). Consequently, the recycle pump flow rate must vary according to the recycle pump location. Recycle pump location Flow rate of recycle Pump
Zone I
Zone I1
QRec
QRec-
QExt
Zone 111 QRecQFeed
Qextf
Zone 1V QRec-QExt+ QFeed-
QRaf
The technical concept of a SMB is not easy as this technology requires a strong precision for all the flow rates (usually better than 1 %), and great care must be taken with all connections between the different lines in order to minimize dead volumes. Moreover, all the columns are to be nearly identical, and very stable. This can be achieved even with soft gels, provided that an adapted procedure is used [ 161. Finally, as SMB is now proposed for biochemical applications, great care must be taken for all GMP issues (cleaning, reproducibility, software validation). Moreover, as SMB is probably always linked with a device allowing product recovery in the extracdraffinate streams and/or solvent recycling, these solvent handling devices are to be considered as a part of the SMB. An example will be given later, in section 1.6, Main applications and developments.
16
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
c
c
e
_ _ - - - - - -- - Zone I
_ .. .
...
\ \ \
/
\
/
\
/
\
/
\
/
\
I
\
I
\
I
\
1
I
Zone~~
I
,
I
I
I
Zone IV
\
I I
\
I
\
/
\
/
\
/
\ \ \ \
.....
. ---
1'
/ /
Feed
/
/
,
/
- - _ _ _ _ - -- Zone 111
c
0 0
c
/
Fig. 1-9. Representation of the zones in a Licosep-Lab.
1.5 Operating Conditions The steps to be followed when designing a SMB which would allow to process a given amount of feed per month have been described in detail [9]. The procedure described is based on the modeling of non-linear chromatography, an experimental procedure being likely to fail unless the adsorption isotherms are linear (this is very rare in fact or means to work at very low concentrations). It is a rigorous, versatile and general procedure which mainly requires the determination of competitive adsorption isotherms; this is not tedious and requires less work than is often claimed. A few competitive adsorption data, measured using the mixture to be separated (the pure components are not necessary), are usually enough to find the operating conditions of a SMB [9]. The design of a SMB (or a TMB) mainly relies in the adequate choice of the different flow rates: recycle, feed, eluent, extract, raffinate and solid (equivalent to a shift period); other important parameters to be evaluated are the following: Feed concentrations Number of columns per zone Column length - Column diameter - Particle size. -
All these parameters can be determined and optimized if the following data are available, from a laboratory-scale study.
1.5 Operating Conditions
17
1.5.1 Equilibrium Adsorption Isotherms In the case of a single-component system, the adsorption isotherm gives the concentration in the stationary phase versus the mobile phase concentration C when equilibrium is reached, at a given temperature. Even if it can sometimes be linear in a wide concentration range, the relation versus C is usually not linear. In the case of a multicomponent mixture, there is usually a competition between the various compounds to access to the adsorption sites. Consequently, does not only depend but on all liquid phase concentrations. Each component adsorption isotherm is on a relation of the following type:
c
c
ci
ci
Which becomes linear at low concentrations: -
ci = Ki . ci
Notice that the previous initial slope of the adsorption isotherm can be easily obtained from the knowledge of the retention time associated to a small injection per-
( 'T" -)
.K formed on a column, as this retention time is given by: tR = t o . 1 + &.V where to = - is the 'zero-retention time' based on the external bed porosity ~
Q
E
(commonly, E is about 0.36-0.4). Many different isotherm equations have been described in the literature [17]. Even if it has been challenged because it does not agree with the Gibbs' adsorption isotherm unless all saturation capacities are identical [I 81, the competitive Langmuir isotherm is very often used:
where
is the saturation capacity assumed here to be equal for all components and
K, a numerical coefficient quantifying the affinity of the solute towards the solid. In fact, the Langmuir isotherm often fails to fit experimental data and more complex isotherms, such as modified competitive Langmuir [I71 or bi - Langmuir [56], are suitable. This is especially true with systems exhibiting concentration-dependent apparent selectivities
(~J~,),/(~+.J.
It must be pointed out that the physico-chemical mechanisms allowing chromatographic bioseparations to be performed are not always 'adsorption-like' but can involve ion-exchange, ion-exclusion or size exclusion. Even if it is generally possible to fit experimental data with a mathematical function derived from the adsorption theory, it is strongly advisable to refer to the proper physico-chemical process before modeling the separation. For instance, ion-exchange can be modeled with selectivity
18
1 Simulated Moving Bed (SMB): Some Possible Applications .for Biotechnology
coefficients (derived from the mass action law) that can be constant or not [19,20], ion-exclusion can be modeled on the basis of theories based on the Donan exclusion ....
Van-Deemter Equation We have now to quantify how the fluid velocity affects the columns efficiency. The height equivalent to a theoretical plate H is defined as:
L H= N where L is the column length. It can be related to the experimental system parameters through the Van-Deemter [21] or Knox [22] equations, which especially give H as a function of the interstitial mobile phase velocity, K . In the case of preparative chromatography, where relatively high velocities are used, these equations can very often be simplified into a linear relation [23,24]: H=a+b.u
(9)
The a und b parameters are related to the diffusion coefficient, prorosity, mass transfer parameters. If internal diffusion is the main resistance to mass transfer (which is quite usual in preparative chromatography), their dependance towards the particle diameter d p is given by [25]:
Experience shows that A is usually about 2-3.
Darcy Equation The Relation giving the pressure drop in the column (per unit length) A PA. versus the mobile phase velocity u is given by the Darcy equation. The Kozeny-Carman equation is suitable for the laminar flows met in chromatography:
where hk is the Kozeny coefficient (close to 4.5) and p the eluent viscosity.
1.5 Operating Conditions
19
1.5.2 TMB and SMB: Two Equivalent Processes The TMB and SMB concepts are similar. In fact, it has been shown that a SMB and its hypothetical corresponding TMB have very close performances [26]. Knowing that optimum operating conditions can be found directly for a TMB and that to simulate this kind of process leads to much shorter computation times, it is in our interest to take advantage of this similitude when designing a SMB. Consequently, a study of a hypothetical TMB is performed first. The general method allowing optimization of SMB processes has been presented by Charton and Nicoud [ 9 ] , and is briefly sumarized in Fig. 1-10. For a given feed composition, optimum flow rates, i. e. giving the highest productivity and the lowest eluent consumption, are estimated first for an ‘ideal system’; this mainly means that
(TMB)
Feed concentrations
-
Column diameter
(SMB)
Fig. 1-10, Overview of the method allowing determination of SMB parameters. (From [ 9 ] ) .
20
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
kinetic and hydrodynamic dispersive effects are assumed to be negligible. This procedure is viable because it has been proven that TMB or SMB performances are only slightly sensitive to the number of plates [26]. In most cases, the number of plates required can easily be achieved and optimum flow rates are then available.
Optimum Flow Rates (TMB) In order to give the ‘feeling’ of the method allowing the flow rates to be obtained, we will consider the particular case of linear adsorption isotherms. This means that the concentrations in the adsorbed phase are linearily related to the bulk concentrations according to equation (6). When two species are considered, one has:
The steady-state profile is obtained because different flux are in equilibrium. Anywhere in the system, the flux of product i (i = A or B) moving downward with the concentration of the solid is: F d = hl.ci where M is the solid flow rate and i on the solid; meanwhile the flux of i moving upward with the eluent is Fu= Q . C, where Q is the liquid flow rate in the zone considered. Consequently, if:
ci
Bi = F, / F d
< 1 the net flux of i is downward
Oi = FuI
> 1 the net flux of i is upward
Fd
A constraint is associated to each TMB zone as expressed in Table 1-2: If all the inequalities given in Table 1-2 are assumed to be satisfied through the same factor fi (0 > l), one obtains: Table 1-2. Constraints to be fulfilled by the internal TMB flow rates in order to ensure a separation between two products A and B (B more retained than A). Zone
Physical constraint
I
B must move upward
Mathematical expression (linear case)
h4 .KB I1
A must move upward
h4 . K A I11
B must move downward
QIII
h4
IV
> I
>1 <
.Kg
A must move downward
ni . K A
<1
1.5 Operating Conditions
21
and finally, taking into account the relations of system (1):
M=
-
QF
KB1P-KA.P
Q E =~b?~. (XB- FA). p
QTMB Rec
= QITMB = p . M . KB
Q R =~M ~' (RB- K A )I p
System (15) allows all the TMB flow rates to be chosen, for a given feed flow rate and for a given value of the parameter which is a safety factor: if p is close to 1, the SMB is operated under its maximum possibilities, and it becomes very sensitive with respect to the number of plates and to the flow rates. If the p value is increased, the system is less productive but more robust. Practically, typical values of P are located between 1.00 and 1.05. It must be noticed that according to system (15), this parameter must fulfil the constraint:
We stress the fact that these relations are only valid for linear adsorption isotherms: sugars on cationic resins, diluted feed on silica (usually less than 10 g 1-'). . . In preparative chromatography, high feed concentrations are suitable and lead to non-linear adsorption behaviors. The non-linear (and related competitive) effects must absolutely be taken into account when evaluating the flow rates. This issue has been barely adressed in the literature [27-291. It is beyond the scope of this paper to describe exactly the calculation procedure; the aim is simply to indicate that the method is exactly equivalent to that presented for linear systems, but that the mathematics are more complex.
Feed Concentrations (TMB) Feed concentrations have a strong influence on SMB performances and must be well chosen. The productivity and eluent consumption are two main economic criteria involved in chromatographic processes [30]. Their variations versus the feed concentrations can be checked in order to choose an appropriate feed composition. This study can be quickly carried out for an 'ideal' TMB as mentioned in the previous section. It has been reported [9] that productivity increases and eluent consumption
22
1 Simulated Moving Bed (SMBJ:Some Possible Applications for Biotechnology
decreases when feed concentration increases: the variations are usually rather steep in the low concentration range but very smooth otherwise. As a consequence, low injection concentrations will have to be avoided. However, even if achievable, very high concentrations will not be suitable because: as soon as concentrations are high enough, the performances of the TMB (or SMB) are almost constant, - very high concentrations can lead to a very low feed flow rate which might be difficult to control. -
Minimum Number of Plates (TMB) The flow rates given previously lead to 100 % purities in the case of an ‘ideal’ TMB, equivalent to an infinite number of plates. The approach used here is to keep these flow rates and to seek the minimum number of plates N,,, required to reach the required purities, as high as 99 %. For this, a model of the TMB which will permit assessment of the influence of the system efficiency on purities must be used. Many different models have been applied to the modeling of chromatographic processes (cf. for instance in Ganetsos and Barker [ll]). The equilibrium stage model has been proven suitable under the usual conditions of high performance preparative chromatography [3 11, and can also be applied to true moving bed adsorbers [32]. The column is considered as an association of cells in series. The adsorption equilibrium is supposed to be reached in each cell, called equilibrium stage or plate. The broadening effects, linked to the mass transfer kinetics and to the hydrodynamics, are lumped together and are quantified by the number of theoretical plates, N , which can be derived from an ‘analytical’ pulse injection. The mass balance equation of a component i over a plate k when steady state is reached is:
where Q and ni are the fluid and solid flow rates. Solving equation (17) for each plate in each zone (with the proper liquid flow rate), together with the proper boundary conditions, allows calculation of the internal concentration profiles, and thus the extract and raffinate purities. The steady state of a TMB is calculated for different numbers of theoretical plates, an identical number of plates in each zone being assumed. The extract and raffinate purities are derived from each numerical simulation. At this stage, everything that can be obtained using the simple TMB representation has been obtained. In order to get more information, numerical work has now to be performed with the real SMB.
1.5 Operating Conditions
23
Number of Columns per Zone (SMB) The SMB parameters are derived from the TMB flow rates according to the rules summarized in Table 1-1. by: I Especially, the shift period At is linked to the simulated solid flow rate &
SMB Q~ec
=
TMB Q~ec
+
E
.
-. M
1- &
The only way to estimate the number of columns per zone, N,, is to perform numerical simulations of a SMB, including the shift of the injection und collection points at regular time intervals. This type of calculation can be performed using dummy values of the column volume and TMB solid flow rate (to estimate the period At). In the case of fixed-bed operations (elution chromatography, SMB), the mass balance equation of a component i over a plate k is:
where to is the mean residence time of the mobile phase in a plate and E the external porosity which is usually in the range 0.35-0.45. System (19) is a system of first-order ordinary differential equations and can be solved with classical numerical methods [33] when associated with a set of boundary and initial conditions.
Column Length and Diameter A simple way to derive the column length and diameter is to consider the minimum number of plates required for obtaining the required purities and a maximum pressure drop. From equations (9) and (12), the pressure drop and the total number of plates in the SMB are linked to the system total length and mean velocity by:
Nmin --
L
-
1
a+b.um
which is a system of two equations and two unknowns, urn and L. Knowing L and urn, the column length and diameter are easily calculated by:
24
I Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
Lcol
where
@$b
L =Nc
is the SMB average internal flow rate, and SZ the column section.
Practical Example A small injection of a mixture containing 50 g 1-’ of a sugar (sgr) and 50 g 1-’ of salt (slt) is performed on a column of 25 x 1 cm packed with an ion-exchange resin (250 pm) and eluted by 1 ml min-’ of water. The results are the following: the retention times are: t R (slt) = 8 min tR (sgr) = 11 min. The number of plates associated to the two peaks are estimated to be about: N (slt) = 400 N (sgr) = 200 (in fact, the resolution between the two products was so poor that the two products were injected independently to allow a more precise estimation). The pressure drop was about 0.1 bar. Estimate the size of a SMB able to process about 1 kg h-’ of a solution containing 100 g 1-’ of the previously defined feed mixture.
The column volume being 19.62 mL, the zero retention time is (assuming an exter0.4. 19.62 nal porosity of 0.4) =7.85 min. Consequently the initial slopes of the 1 adsorption isotherms are: 11-7.85 0.4 8-7.85 0.4 .= 0.267. The salts appears Eslt= 7.85 . = 0.012 Ksgr= 1-0.4 7.85 1-0.4 to be excluded from the resin (very small initial slope of the adsorption isotherm). ~
The purification of 1 kg/h of a mixture at 100 g 1-’ imposes a feed flow rate of 10 1 h-’. From system (15), the other flow rates are calculated, provided that a value is selected. With 0 = 1.05, one obtains:
ni= Q’,:”
10 = 41.2 - 0.012 * 1.05 = 41.2 . (0.267-0.012) . 1.05 = 11.03
Qig”= 1.05 .41.2
’
0.26 = 11.55
QiZ” = 41.2 . (0.267-0.012)
/ 1.05 = 10.00
using equation (18), one obtains the recycle flow rate for the SMB (all other flow rates being identical for the SMB and the TMB): 4 QiE” = 11.55 00.6 2 41.2 = 39.0 1 h-l. At this point, all flow rates are known. The estimation of column length and diameter requires the required number of plates and pressure drop, to be taken into account.
+
1.6 Main Applications and Developments
25
Regarding the number of plates, we will only consider the sugar (limiting species). With equation (8), (9) and (lo), one can write, taking into account the experimental m s-I): results obtained on the laboratory scale column (u = 2.12 x 200 0.25 2 ,250
1
+ b .2.12 .
thus b = 3.538 s.
plates I m
Nmin 1 Consequently (equation 20), - plates I m L 5.10T4 + 3 . 5 4 . ~ ~
AP
The experimental pressure drop allows us to write: - = 1887 . urnbar I m L The separation being a medium difficult separations, we can guess that 200 plates are enough [9] for performing the work (this could be confirmed with numerical simulation). Moreover, if one assumes a maximum of 5 bars in the system, the two preceeding equations lead to: L = 1.37 m urn = 0.0019 m s-l. The average flow rate in the system being about 33 1 h-’ (39.0 in zone I, 27.70 in zone 11, 37.70 in zone 111, 27.70 in zone IV) the previous average fluid velocity is obtained for a column section of 4.8 X lop3 m2, leading to a column diameter of 7.8 cm. Assuming that this separation can be performed with a SMB made of 10 columns, the length of one column must be 13.7 cm.
1.6 Main Applications and Developments There are many bioseparation problems for which SMB can be used. Most of them are reviewed in the following and can be classified in: -
separation of sugars desalting purification of proteins separation of ionic molecules separation in organic solvents optical isomers separation
1.6.1 Separation of Sugars This application is the most known, the separation between fructose and glucose being one of the largest applications of chromatography. Since the pioneering work of Barker [34], this separation has been investigated by many workers [35-371. A review of the work done by Barker’s team is given by
26
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
Ganetsos and Barker [ 111. This separation is performed on ion-exchange resins, using warm water as eluent. The preferred implementation consists of using polystyrene cation exchange resins in the calcium form: the fructose forms a complex with the calcium ions and is retarded, the glucose and other oligosaccharides are eluted with the eluent. In order to improve the productivity, some work has been done with zeolites (calcium form) instead of resins [38]. Another possibility consists of using anion exchange resins in the bisulfite form, the glucose being retarded by complexing with the other oligosaccharides being eluted first. This option is not used at production scale because of the lower stability of anion-exchange resins. The Sarex process has been reported [39] to be used to separate continuously a 500 g 1-' inverted carbohydrate syrup containing 42% fructose giving 90-94% pure fructose at recovery of over 90%. The glucose-rich fraction is about 80% and both product concentration were about 200 g/l. This separation can be implemented on columns of a few meters internal diameter. SMB packed with cationic resins in the calcium form have also been used for obtaining other monosaccharides such as xylose or arabinose [2]. Experiments have been performed using a feed mixture containing 21 g 1-' of glucose, 155 g 1 - I of xylose and 20 g 1-' of arabinose, the retention order being glucose, then xylose, then arabinose. The goal was to obtain a raffinate with no detectable traces of arabinose with a maximum recovery of xylose. The goal was reached as the following extract and raffinate composition were obtained:
- raffinate: glucose: 8.26 g 1-'; xylose: 73.69 g 1-'; arabinose: below detection -
limit raffinate: glucose: 0.09 g 1-'; xylose: 0.47 g 1-'; arabinose: 14.16 g 1-'.
This exemplifies perfectly that SMB is a binary separator: it allows a mixture to be split into two fractions, even if the mixture contains more than two products. Another interesting application is devoted to the separation between mono and disaccharides, or between disaccharides. For instance, Kishihara [40] studied the separation between palatinose and trehalulose and results obtained by NOVASEP for the separation between fructose and trehalulose are given in the following paragraph. The separation has been performed on Dowex 99 monosphere (350 pm) in calcium form Ca form using water (65°C) as eluent. Under these conditions, the retention I,~ = 0.4. factors of the two sugars are: E T ~ =~0.17 The separation has been performed on a SMB made of 12 columns of 2.6 cm i. d. and cm length, working with a maximum pressure drop of 5 bars and at a temperature of 65 "C. The feed containing 120 g 1-' fructose and 120 g 1-' of trehalulose was injected at a flow rate QF = 10.83 mL min-' leading to a productivity of 3750 g of feed/day on this small pilot plant. Excellent results were obtained: both extract (position 3) and raffinate (position 9) were recovered at 98% purity as shown on the internal profile (internal concentrations normalized by the feed concentration) given in Fig. 1-11. Moreover it must be noted that the pure fractions are recovered at a very significant concentration: 79% of the feed concentration for the fructose, and 65 % for the trehalulose. The SMB technology has also been used for performing the fractionation of dextran (polyglucoside mainly used as a blood plasma volume expander) by size exclu-
x ; ~ ~ ~ ~ ~
1.6 Main Applications and Developments
27
A Raff.
1
2
3
4
5
6
7
8
9 1 0 1 1 12
Column Number
Fig. 1-11.Separation between fructose and trehalulose. Internal profile obtained on a Licosep-lab (NOVASEP).
sion [ 111. The columns were packed with Spherosil XOB075 of 200 to 400 pm porous silica beads, and the technology has been proven to be efficient allowing to obtain, according to the flow rates very different fractions (from 10000 Daltons to 125 000 Daltons).
1.6.2 Desalting Desalting is a second simple and interesting application of SMB in biotechnologies. Different mechanisms can be used like ion-exclusion, hydrophobic interaction, size exclusion or ion retardant effect [38]. Glucose and NaCl have been separated on Retardion 11 A-8 [41], very high purity products have been reported for feed mixtures containing 3 mol 1-' of glucose and 3 mol 1 - I of NaCl. In that case, the adsorption isotherms are favorable (Langmuirian type). Similar work has been performed in order to separate NaCl from glycerol on Amberlite HFS-471X (8 % DVB) which is a strongly acid cation-exchange resin of the sodium form. In that case, the mechanism is ion exclusion (the glycerol can enter the internal porosity of the resin when the salts are excluded). The adsorption isotherm of glycerol is linear (as expected because glycerol has no interaction with the resin: it just enters in the internal porosity) when the adsorption isotherm of the salt is non favourable (anti-langmuir) as expected from an ion exclusion process (Donnan behavior). Instead of ion-exclusion, the size exclusion process has been used for performing the separation between NHdS04 and a protein [42]. In that case, the adsorption isotherms were simply linear.
28
1 Simulated Moving Bed (SMB):Some Possible Applications for Biotechnology
Finally, a hydrophobic interaction has been used by Hashimoto et al. [42] for performing the separation between phenylalanine and NaC1. In that case, NaCl having almost no interaction with the packing had a linear adsorption isotherm, when the phenylalanine exhibited a classical Langmuirian adsorption isotherm.
1.6.3 Purification of Proteins Only a few references regarding the use of SMB for performing the separation between proteins have been reported. The first attempt can probably be credited to Huang et al. [43], who performed the purification of trypsin from extract of porcine pancreas. They used succesfully a SMB made of only six affinity columns, which shows that SMB with a very limited number of columns can be attractive. Even if the four-zone process had fairly high performances in separation efficiency, the addition of a fifth zone used as a washing section (cf. Fig. 1-4 option 2) was advised. A recent proposal consists performing the purification of human serum albumin (HSA) [44] on two SMB connected in series: the first one was used for removing the less strongly retained components, the second one for removing the more strongly retained components. Finally, some results concerning the separation between myoglobin and lysozyme have been presented recently [50]. This separation can be performed on a support such as ACA 54 (Biosepra, France) with an eluent containing NaCl 0.15 M in water [L. Guerrier, personal communication]. On a SMB made of eight columns of 2.6 cm i.d. and 0.1 m length, very pure extract (> 98%) and raffinate (> 98 %) can be obtained from a 50-50 mixture at 2 g 1-'. An internal profile is given in Fig. 1-12.
0.9
=0
c
2 8 1
4 e 0
0.8 0.7
0.8
+MyiDglobine
-
-
-
0.5. 0.4
0.3 0.2 0.1
-
-
. 1
2
3
4
5
6
7
8
Column Number
Fig. 1-12. Separation of myoglobin/lysozyme: Internal profile on eight column SMB for on ACA 54 (Biosepra).
1.6 Main Applications and Developments
29
1.6.4 Separation of Ionic Molecules SMB technology has also been used for performing the purification of different ionic molecules. For example, large-scale production of lysine can be performed by SMB [46]. Pure betaine can be obtained from molasses via a process involving two chromatographic steps [47]:
+ glycerol is separated from the rest of the feed (mineral, carboxylic acids) thanks to ion exclusion - step 2: glycerol is separated from betaine in a second ion exclusion step
- step 1: a fraction containing a mixture betaine
L-Glutathione [48] produced by yeast fermentation is a tripetpide used in cases of liver disease. High-purity L-glutathione (99 9%) is required in the final crystallization step. Obtaining this highly pure glutathione is difficult, especially because of the presence of amino/acids, one of the limiting species being glutamic acid. The separation between the L-glutathione and glutamic acid is performed on a cation-exchange resin (Rohm & Haas, Amberlite IR200C, 350-590 pm). The separation has been implemented in a SMB made of 16 columns of 1 cm i. d. and 10 cm or 20 cm length. The SMB parameters have been determined via a procedure which is similar in essence to the procedure given in this chapter (determination of the internal flow rates thanks to the knowledge of the adsorption isotherms). Operating at 0.05 mol 1-' of HC1 (the pH value controlling obviously the adsorption of both products on the resin) the glutathione was obtained in the raffinate stream at 1.62 X lop3 mol Lpl at 99% purity with 99% yield. The productivity was 4.52 X lop4 mol L-' of adsorbendmin. Moreover, in that case, the steady state was reached in about 4 h, which is a very reasonable time.
1.6.5 Separation in Organic Solvents Many separations of organic molecules have been performed with SMB technology, but according to our knowledge, only two of them belong to the biotechnology area. 1. Some work has been done in order to perform separation of fatty acids. The first results have been published by Szepy [49] and were associated to the separation of c16 to C22 methyl esters. An interesting patent has been granted to PRONOVA: EPA, 697034 (21/02/1996) which associates classical (batch) chromatography with an SMB with two inlets for obtaining pure EPA ((220) and pure DHA (C22) from fish oil: this example exemplifies nicely the need to associate different technologies in order to optimize a production process. 2. Recently, the separation of stereoisomers of phytol(3,7,1,15 -tetramethyl-2-hexadecen-1-01, C20H400) has been described [7] at a relatively large scale (about 20 kg/ day). The synthetic phytol is a mixture of cis (33 %) and trans (67 %) isomers, the latter being used in perfumery. This separation is presented in the following text.
30
1 Simulated Moving Bed (SMB): Some Possible Applications .for Biotechnology
The separation between the phytol isomers is performed on classical silica (Lichroprep Si 60, 25-40 pm from Merck KGaA, Darmstadt) with an eluent made of heptane-ethyl acetate (75/25, v/v) at 17 "C. This separation is implemented on a Licosep 8-200 (NOVASEP) which includes eight axial compression columns of 200 mm i. d. and up to 400 mm long. For this particular case, column were packed with 3 kg of silica leading to a column length of 17.7 cm ( 5 0.6 %>.The Licosep 8-200 (Fig. 1-13) is built according to the flowsheet given in Fig. 1-7 and 1-9. Each outlet (extract and rafinate) was processed using a falling film evaporator consisting of 1-12 tubes, each of 3-m length and 10 mm i,d. For this separation, the pressure was set to 180 mbar, and oil temperature to 80°C. The concentrated extract and raffinate were recovered in 50-L glass vessels. The eluent vaporized under reduced pressure in the falling films evaporators was separated from the cis and trans phytol in gas-liquid separators and condensed in a 50-L glass vessel. This vessel was periodically emptied using a pump, which was servocontrolled by two level probes located in the vessel. Before recycling to the SMB, the composition of the solvent was adjusted. The flowsheet of the complete unit is given in Fig. 1-14. For a feed concentration of 105 g l-l, the system has been operated night and day under the following conditions: - Q R =~226 ~ 1 h-' - QFeed = 7.8 1 h-'
- Q ~ l =~36.0 ~ 1~h-'t
- Q ~ ~= 32.4 t ~ 1~h-' ~ - Q ~ ~= f11.4 f 1 h-'
t
leading to an extract purity of 98.4 % and to a raffinate purity of 99.4 %. It must be noted that the pressure drop, being only 26 bar under these conditions, the productivity could have been increased of about 52 %.
Fig. 1-13. Licosep 8-200: SMB made of eight axial compression columns.
1.6 Main Applications and Developments
31
,A 1 EXTRACT 1 I
I ,-
i-’
F.F. 1
S.M.B. : Simulated Moving Bed E.R. : Eluent Recycling F.T. : Feed Tank
B.S.
F.F
E.T. F.P.
: Falling Film Evaporators : Eluent Tank
: Feed Preparation
B.S.
: Back-up Solvent
Fig. 1-14. Flow-diagram of the Licosep 8-200 including the solvent recycling unit.
1.6.6 Separation of Optical Isomers The interests of SMB for performing large-scale separations of optical isomers are now recognized (very short development time, extremely high probability of success, attractive purification cost). An increasing number of published results are available [50,51] among which are: prazinquatel [52], 0-Blockers [53], chiral epoxide [5], thiadiazin EMD5398 [9], hetrazepine [6]. Very recently, the Belgian company UCB-Pharma announced their decision to purchase a large scale SMB from NOVASEP in order to perform optical isomer separation at several tons per year. Almost all these separations are performed on cellulose-based packings using organic eluent. The two first references were published in 1992 by Negawa and Shoji [54] on phenyl-ethyl alcohol and Fuchs et al. on threonine [55]. Details of the second reference are described as follows. Among all the examples cited previously, the threonine example is by far the least productive. It is presented here because it involves amino acids and a ligand exchange chiral chromatography which is seldom used at preparative scale and is thus probably more ‘bio’ than the others. The separation of the two optical isomers of threonine has been performed on a SMB made of 12 columns of 2.6 cm i.d. and 1 m length. These columns were packed with Chirosolve L-Proline (200 pm) from J.P.S. Chime (Bevaix, CH2022). These supports involves ligand exchange chromatography, was introduced by Davankow in 1968, the mechanism being based on the formation of complexes between the grafted proline, copper and the optical isomers contained in the solution. The eluent used is a mixture between acetic acid (0.05 M) and copper acetate (0.000125 M); the operating temperature was 25 “C.
32
2 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
1
2
3
4
5
6
7
8
S
1
0
1
1
1
2
Column number
Fig. 1-15.Separation between fructose and trehalulose. Internal profiles obtained on a Licosep 1226 (NOVASEP).
The racemic mixture was injected continuously at 5 g 1-' the operating being as follows.
- Q R =~22~ mL min-' - Q
F = ~4.2 ~mL ~min-'
- Q E , =~6.2~ mL ~ ~min-' - Q E = 7.4 ~ mL~min-'~ -
~
~
~
Q ~ ~= f3.0 f mL min-' AT = 43 min
Under these conditions, about 30 g of feed were processed per day, and very pure extract (99 %) and raffnate (99 %) obtained. An internal profile obtained with the set of parameters given previously is given in Fig. 1-15. This profile is very interesting in that it demonstrates what is possible when the non linear behavior of the system can be adjusted: the extract is recovered at a concentration of 60 % - this is already an excellent result, but the raffinate is recovered in a pure form and at a concentration which is 25 % greater than the feed concentration.
1.7 Comparison with Batch Chromatography Different comparisons between batch chromatography and SMB have been published. They are sometimes extremely positive for the SMB: For the L-glutathione/glutamic acid separation, [48] reported that the SMB allows productivity to increase 19-fold and eluent consumption to decrease 14-fold. - For the separation between the two enantiomers of the phenyl ethyl alcohol [54], it was reported that productivity is 6 1-fold higher and eluent consumption 200fold lower with the SMB. -
1.7 Comparison with Batch Chromatography
33
- For the separation of the optical isomer of a chiral hetrazepine [6], productivity
was 50-fold higher and eluent consumption 2-fold lower with the SMB. In other situations, even if the comparisons are still in favor of SMB, the differences between batch chromatography and SMB are less significant: - For the separation of a chiral epoxide [ 5 ] , solvent consumption was about 7-fold
lower with the SMB, but productivities were similar.
- For the separation of the seteroisomers of phytol with 25 ym silica [7], the productivity was 1.7-fold greater and eluent consumption 2-fold lower with the SMB. It appears to be very difficult to draw some general conclusions! In order to make an attempt to clarify this situation it is important to underline three basic differences between SMB and batch chromatography.
1.7.1 Influence of Flow Rates When designing a batch chromatography process, flow rate selection is only made in order to work with an acceptable pressure drop and number of plates, the main parameters being associated with the amount of feed injected and to the positions of the cut points. For the SMB mode, as explained in this chapter, the possibility of separation relies on an adequate choice of flow rates, and consequently we can expect the SMB to be very sensitive with respect to the flow rates. In order to exemplify this behavior, let us consider the separation of the optical isomers of the threonine already presented [ 5 5 ] . In Fig. 1-16, the influence of the recycle flow rate on the extract and raffinate purities is given. It appears that both extract and raffinate are pure for a recycle flow rate of 22 mL min-' (all other flow rates being fixed). If the flow rate is modified by about 5 %, the purities are decreased by about 15 %. A 5 % variation in the eluent flow rate in batch chromatography would have very limited influence on the purities. A similar behavior is obviously expected with respect to the other flow rates and to the shift period.
50 6o
1
20
20.5
21
21.5
22
22.5
23
Recycle flowrate (rnllrnin)
23.5
24
Fig. 1-16. Influence of the recycle flow rate on the purities of D and L threonine [ 5 5 ] .
34
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
As explained earlier in this chapter, the flow rates must be determined according to the adsorption isotherms (thus chromatographic retention). The SMB, in being very sensitive with respect to the flow rates, means in fact that the SMB is very sensitive with respect to the product retention; that means that one must ensure a very good stability of the retention.
1.7.2 Influence of Number of Theoretical Plates The fact that SMB is less sensitive with respect to the number of theoretical plates than batch chromatography is probably due to the fact that SMB involves a counter-current contact between the liquid and the solid phases. This difference in behavior has been recognized by Nicoud and Bailly [3] and quantified by Tondeur and Bailly [26]. In order to explain this behavior, let us consider the curve given in Fig. 1-17 in which a dimensionless loadability is given versus the number of plates equivalent to the system considered. The dimensionless loadability is the ratio of the injected amount (flowrate or volume) to the amount that could be injected for a system equivalent to an infinite number of plates; as a consequence, both curves reach 1 for a system equivalent to an infinite number of plates. It can be seen that the loadability of the classic elution mode breaks down rapidly for low-efficiency columns and, below 300 plates, the specifications are impossible to meet. The loadability increases continuously with the number of plates even for very efficient systems: this explains the reason for the success of preparative HPLC. The behavior of the TMB is completely different: even at 400 plates (thus about 40 plates per column for a 10 columns SMB!) it retains about 90% of its theoretical efficiency.
+
0
ma
lw
VJOCI
2000
Number of plates
~@IO
xao
Fig. 1-17. Comparison of sensitivity of fixed-bed and T M B loadability with number of stages. Separation: chiral epoxide on CTA [ 5 ] . Imposed purity and yield: 95 %.
1.7 Comparison with Batch Chromatography
*-
0
I
SelectiL.irj=l.8
12
0
;-------i----c--i
250
0
500
750
loo0
35
Selectiiiw 1.3
7
I
T_
~
0
-
I
1
m
m
m
Number of plates
Number of plates
Fig. 1-18. Comparison of eluent consumption for batch chromatography and for SMB: influence of the selectivity factor (Langmuir behavior) and of the number of plates. Purities and yield are kept constant, eluent consumptions are calculated according to the theory explained in Section 1.5.
From this very important result, one can derive a simple rule of the thumb: the SMB is of special interest when batch chromatography requires a significant number of plates (thus for limited selectivity factors) and/or when it is difficult (expensive) to be obtain the theoretical plates. The practical effect on eluent consumption is given in Fig. 1-18. For a given separation (given selectivity), the eluent consumption decreases when the efficiency of the system is increased until to reach a plateau. The calculations associated with Figs. 1-18 and 1-19 have been performed according to the following procedure. Species are expected to fulfill a Langmuirian behavior (KA = 1 ) the feed being a 50-50 mixture at 2 g 1-* total concentration. The extract and raffinate purities are fixed to 98 %. Whatever the selectivity (1.3 or 1.8), the eluent consumption is lower for SMB, but the interest is maximized for low selectivity and low plate numbers.
10 -
,
.
------
0 I
0
[email protected]
Selactiiiw 1.8
I
503
Number of plates
loo0
+---
0
-
0
1033m3003 Number of plates
Fig. 1-19. Comparison of productivity for batch chromatography and for SMB: influence of the selectivity factor (Langmuirian behavior) and of the number of plates. Purities and yield are kept constant. System sizes are calculated according to the theory explained in Section 1.5
36
1 Simulated Moving Bed (SMB): Some Possible Applications f o r Biotechnology
In contrast to what is obtained with the eluent consumption, productivity can exhibit a maximum with respect to the efficiency of the system (Fig. 1-19). The result can be simply explained: when a given number of plates is reached, the system behavior is completely dependent on the adsorption thermodynamics, there is thus no need to add theoretical plates. To add such plates would lead to an increase in column length without the possibility of injecting more, and thus to a decrease in specific productivity. It appears that SMB reaches its optimum for a lower number of plates and that this optimum is higher than the batch optimum. Both results are clearly beneficial for SMB. The results presented in Figs. 1-18 and 1-19 show that the SMB is generally significantly more performant than batch chromatography, but that the expected benefit depends on a lot a parameters, among which are selectivity and equivalent number of plates of both systems. Moreover, this property allows SMB to retain its efficiency (purities, productivity) even if the columns are partly degraded, as shown with the phytol example [7].
1.7.3 SMB is Continuous Oviously, a very important difference between batch chromatography and SMB is due to the continuous behavior of the SMB. This continuous behavior allows easier connection with solvent processing units but can also lead to theoretical difficulties: replacing discontinuous by continuous process imposes the complete redefinition of a ‘batch’ which is at the origin of quality control systems. The continuous nature of SMB may also have advantages in situations where bacterial growth is possible. For example, with dilute solution of sugars, bacterial development may lead to plugging throughout the system. In being continuous, SMB makes is easier to prevent stagnant zones and thus to avoid such fermentation problems.
1.8 Conclusion Simulated Moving Beds have been used successfully during almost 30 years at very large scale in the petrochemistry. Clearly it appears that this technology has a large potential for fine the chemistry and pharmaceutical industries. More and more applications are described for the biochemical field (leading sometimes to 10-fold lower eluent consumption compared with normal chromatography). Small-scale units are already available, which means that SMB can be operated for productions of < 1 kg as well as for very large productions (100 000 T/year) and for very different products, including enzymes or sugars, desalting or optical isomer separation. The SMB is basically a binary separator that presents three main advantages with respect to batch chromatography.
Abbreviations and Definitions
37
1, It allows the saving of significant amounts of eluent (according to our experience, this situation is general provided that binary separations are involved, however, the situation is more complex with multicomponent systems). SMB is basically a binary separator: if the product to be recovered is the first or the last to be eluted in a multicomponent mixture, SMB can be used without technical modifications (only the way of determining the flow rates was to be changed). In other situations, two SMB systems must usually be used in series. 2. It allows maximization of productivity: the interest of SMB with respect to batch chromatography is maximized for low selectivity problems or low-efficiency systems. 3. It is a continuous process, which simplifies the operation and especially the connection with associated equipments (for example, evaporation). However, despite such advantages SMB utilization requires a strict procedure, an efficient simulation software, and is less versatile than normal elution chromatography. In fact, these two modes are more complementary than competitors: SMB is in reality a production tool, while batch chromatography due mainly to its flexibility, may provide interesting advantages during the early stages of development.
Abbreviations and Definitions coefficients relating the height equivalent to a theoretical plate to the mobile phase velocity mobile and solid phases concentrations Kozeny coefficient height equivalent to a theoretical plate isotherm parameters column length solid flow rate amount of feed processed per unit time number of columns per zone number of theoretical plates minimum number of plates required TMB and SMB internal flow rates in zone k feed flow rate extract flow rate raffinate flow rate internal flow rate in zone k maximum pressure acceptable time zero retention time linear mobile phase velocity SMB average mobile phase velocity SMB column volume
38
1 Simulated Moving Bed (SMB): Some Possible Applications for Biotechnology
AP AT
h@ E
P
R
pressure drop period parameter relating the pressure drop to the mobile phase velocity external porosity mobile phase viscosity column section
References [ l ] Nicoud, R. M., LC-GC Zntl, 1992, 5, 43. [2] Balannec, B., Hotier, G., From batch to countercurrent chromatography, in: Preparative and Production Scale Chromatography, Ganetsos, G., Barker, P. E. (Ed.), New York: Marcel Dekker, 1993. [3] Nicoud, R. M., M. Bailly, in: Proceedings of the 9th Symposium on Preparative and Zndustrial Chromatography ‘Prep 92’, INPL, Nancy, France, 1992; pp. 205-220. [4] Broughton, D. B., 1961 US Patent 2 985 589. [5] Nicoud, R.M., Fuchs, G., Adam, P., Bailly, M., Kiisters, E., Antia, F, D., Reuille, R., Schmid, E., Chirality, 1993, 5, 267. [6] Nicoud, R.M., Bailly, M., Kinkel, J.N., Devant, R., Hampe, T., Kusters, E., in: Nicoud, R.M. (Ed.), Simulated Moving Bed : Basics and applications, INPL, Nancy, France: 1993, pp. 65-88. [7] Blehaut, J., Charton, F., Nicoud, R. M., LC-GC Zntl, Vol. 9, No. 4, pp. 228-238, 1996. [8] Ruthven, D.M., Ching, C.B., Chem Eng Sci 1989 44, 1011. [9] Charton, F., Nicoud, R.M., J Chrom A, 1995, 702, 97-112. [lo] Hotier, G., in: Simulated Moving Bed : Basics and applications, Nicoud, R.M., (Ed.), Nancy, France: INPL, 1993, pp. 95-117. [ l l ] Ganetsos, G., Barker, P. E. (Eds.), Preparative and Production Scale Chromatography, New York: Marcel Dekker, 1993. [12] Ching, C. B., Chu, K. H., Hidayat, K., Uddin, M. S., AZChE Journal, 1992, Vol. 38, No. 11, 1744-1 750. [13] Ganetsos, G., Barker, P. E., Semi continuous chromatographic refiners, in: Preparative and Production Scale Chromatography; Ganetsos, G., Barker, P. E. (Eds.), New York: Marcel Dekker, 1993. [14] Hotier, G., et al., 1996 US patent 5,578,216. [15] Nicoud, R. M. et al., 1995 US patent 5,422,07. 1161 Nicoud, R. M., LC-GC Znt, October 1993 Vol. 6, Number 10. [17] Nicoud, R. M., Seidel-Morgenstern, A., Isolation and Purification, 1996 Vol. 2, 165-200. [18] Levan, M. D., Vermeulen, T., J Phys Chem, 1981, 85, 3247. [19] Nicoud, R.M., Schweich, D., Water Resources Research, June 1989 Vol. 25, No. 6, pp. 1070-1082. [20] Dye, et al., Znd Eng Chem Res, 1990, 29, 849-857. [21] Van Deemter, J. J., Zuiderweg, F. J., Klinkenberg, A., Chem Eng Sci, 1956, 5, 271. [22] Knox, J. H., J Chromatogr Sci, 1977, 15, 352. [23] Horvath, C. and Lin, H. J., J Chromatogr, 1978, 149, 43. [24] Endele, R., Halasz, R., and Unger, K. J Chromatogr, 1974, 99, 377. [25] Villermau, J., in: Rodrigues, A. E., Tondeur, D. (Ed.), Sijthoff & Noordhoff, Alphen aan den Rihn, the Netherlands, 198 1. Percolation Processes: Theory and Applications. [26] Tondeur, D., Bailly, M., in: Simulated Moving Bed : Basics and Applications, Nicoud, R.M. (Ed.). INPL: Nancy, France, 1993, pp. 95-117.
References
39
[27] Rhee, H., Aris, R., Amundson, N. R., Philos Trans R Soc Lond, 1971 269, 187. [28] Ching, C. B., Ho, C., Ruthven, D. M., Chem Eng Sci, 1985, 43, 703. [29] Storti, G., Masi, M., Carra, S., Mobidelli, M., Preparative Chromatography, 1988, Vol. 1, (1); pp. 1-27. [30] Nicoud, R.M. and Colin H., LC-GC Int, 1990, Vol. 3, No. 2. [31] Golshan-Shirazi, S . , Guiochon, G., J Chromatogl; A , 1994, 658, 149. [32] Ernst, U.P., Hsu, J.T., Ind Eng Chem Res, 1989, 28, 1211. [33] Finlayson, B. A., Non-Linear Analysis in Chemical Engineering, Mc Graw-Hill, New York, 1980. [34] Barker, P. E., Critcher, X., Chem Eng Sci, 1960, 13, 82. [35] Hashimoto, K., Adashi, S . , Noujima, H., Maruyama, H., J Chem Eng Jpn 1983 16, 400. [36] Ching, C. B., Ruthven, D. M., Chem Eng Sci, 1985, 40, 877. [37] Ching, C. B., Ruthven, D:M., Hidajat, K., Chem Eng Sci, 1985, 40, 1411. [38] Hashimoto, K., Adachi, S . , Shirai, Y, Mortshita, M., Operation and design of Simulated Moving Bed adsorbers, in: Preparative and Production scale Chromatography, Ganetsos, G., Barker, P.E., New York: Marcel Dekker, 1992. [39] Blezer, H. J., De Rosset, A. J., Die Starke, 1977, 29, 393. [40] Kishihara, S. et al., J Chem Eng Jpn. 1989 22(4), 434. [41] Maki, H., Fukuda H. and Morikawa H., J Ferment Technol, 1987, 65, 61. [42] Hashimoto, K., Adashi, S . , Shirai, S . , Agric Biol Chem, 1988, 52, 2161. [43] Huang, S. Y., Lin, C. K., Chang, W.H., Lee, W. S., 1986 Chem Eng Commun, 456, 291. [44] Houwing J. et al., Delft University, The Netherlands, Proceedings of the First European Symposium on Biochemical Engineering Science, ISBN 1872327 109, Dublin, September, 1996. [45] Nicoud, R. M., Proceedings of Chiral Europe '96, 14-15 October 1996, Strasbourg, Spring Innovation Ltd. Publisher, Cheshire, SK7 IBA, 1996. [46] Van Walsem, H. J., Thompson, M. C. (AECI Bioproducts, Durban, South Africa), First European Symposium on Biochemical Engineering Science, ISBN 1872327 109, Dublin September, 1996. [47] Kampen, W. H., European Patent application, 90307701.4. [48] Maki, H., Separation of gluthatione and glutamic acid using a SMB adsorber system, in: Preparative and Production Scale Chromatography, Ganetsos, G., Barker, P. E. (Eds.), New York: Marcel Dekker, 1992. [49] Szpepy et al., J Chrornatogr, 1975, 108, 285-297. [50] Nicoud, R. M., Recovery of Biological Products VIll, ACS, 20-25 October 1996, Tucson, Arizona. [51] Kinkel, J. N., Proceedings of Chiral Europe '95, 28-29 September, London, Published by Spring Innovation Ltd, Cheshire, SK7 IBA, England, 1995. [ 5 2 ] Ching, C. B. et al., J Chromatogr, 1993, 634, 215-219. [53] Ikeda, H. and Murata, K., 4th Chiral Symposium Montrkal, September 1993. [54] Negawa, M., Shoji, F., J Chromatogr, 1992, 590, 113-117. [55] Fuchs, G., Nicoud, R. M., Bailly, M., in: Proceedings of the 9th Symposium on Preparative and Industrial Chromatography 'Prep 92 ', INPL, Nancy, France, 1992, pp. 205-220. [56] Jacobson, S . , Golshan-Shirazi, S., Guiochon, G. J Am Chem SOC, 1990, 112, 6492.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
2 Systematic Development of Chromatographic Processes Using Perfusion Chromatography Technology Scott Fulton and Thomas Londo
2.1 Introduction Liquid chromatography provides a unique combination of capabilities for the production of biopharmaceuticals - the separation power needed to purify even subtle molecular variants from complex mixtures, combined with gentle chemical and physical conditions which enable recovery of biological activity for complex, biological macromolecules and scaleability for applications ranging from producing a few milligrams of a product to multi kilogram-scale production of bulk drugs in columns. Developing a chromatographic separation method can be extremely challenging, however. The enormous flexibility of the technique - one of its most powerful features - as well as the complexity of both the target biomolecules and the biological sample matrix make it quite difficult to optimize a separation. One significant problem that process developers have faced has always been the time required for a single chromatographic run, which can take from 30 minutes to a day or more using conventional chromatography media. In an attempt to circumvent time-consuming development, developers often rely on past experience, tips from colleagues, or previously published papers. Unfortunately, what worked in the past for one application is often not appropriate for a new problem, leaving the developer with a less-than-satisfactory separation and few ideas for what to do next. When faced with a totally new separation problem, there seems to be no alternative to a lengthy, ‘trial and error’ development approach. Many developers tend to view each run as a trial solution to the separation problem, and the results are judged by how well they meet the objectives. If a given run or trial solution does not work, it is very tempting (in view of the very limited number of runs one can practically perform) to change several variables at once in a ‘best guess’ at the answer based on the data obtained so far. Once a method that ‘works’ is found, development usually stops. ‘Works’ in this case means ‘gives acceptable results’, and not necessarily the best results achievable. The ‘trial and error’ approach violates a fundamental principle of experimental science - change only one variable at a time. Because of this, it can be difficult to understand why a given method works, since the amount of actual information gained about the separation system is limited. It is easy to miss solutions that may
42
2 Systematic Development of P e f i s i o n Chromatography Technology
be far more advantageous. In addition, if something changes later (in the sample or the requirements of the separation itself, for example), there is little information to use in making the necessary adjustments, and the ‘trial and error’ approach must be started over again. Note that this principle does not preclude the use of statistical or experimental design techniques, in which more than one variable may be changed in a controlled way. Statistical experimental design can be a very powerful and sophisticated way to limit the number of runs, conserving both sample and time. Although several parameters may be varied at once, the objective is far different than simple trial and error. In this technique, experiments are constructed to gain more information about the system behavior with respect to all of the parameters and to determine which are important drivers for the system. With the recent advent of high-speed chromatographic techniques, such as perfusion chromatography, individual run times (for both the chromatography itself and much of the analysis) have been sharply reduced [l-31. This makes it practical to use an approach called systematic development. In this systematic development, each of the critical parameters of the separation system are empirically and systematically examined, one at a time. The resulting information about the behavior of the system enables the developer actually to design an optimal solution based on a complete set of real data, as well as implement and maintain the method with a great deal of assurance. This chapter outlines the concept and practice of systematic development of processes using high-speed chromatography. The basic principles of the approach are presented along with a number of examples.
2.2 Perfusion Chromatography It is commonly understood that with most conventional chromatography materials, one must make trade-offs between speed, resolution and capacity. The reason for this trade-off lies in the nature of the particles that make up conventional media. A typical chromatography particle is highly porous (with pores in the 100-1000 A range) in order to maximize the internal surface area for binding. Solute molecules are carried to the perimeter of the particles by the liquid stream as it flows through the packed bed of the column. Transport of the solute molecules to the inside surfaces of the particles occurs by diffusion within the pores (intraparticle diffusion) as illustrated in the left part of Fig. 2-1. Diffusion is a slow process, especially for large macromolecules, and becomes the limiting factor in a conventional chromatography separation. As flow rate across the column increases, there is less time available for this intraparticle diffusion to occur, therefore less ability for solute molecules to interact with the surface area inside the particles. Bandspreading increases, and resolution and capacity are lost if the flow rate is too high. For first-generation biological chromatography packings (so-called soft gels) with 90-200 pm particle diameters, the typical time for a single separation run is on the
2.2 Perfusion Chromatography
Conventional Chromatography
43
Perfusion Chromatography
Fig. 2-1. Pore structure and mass transport in conventional and perfusive particles.
order of hours or even days [4,5].The introduction of modern high-performance liquid chromatography (HPLC) materials enabled higher speed separations by a reduction in the particle diameter to the range 3-30 pm. This reduces the distance for solute diffusion within the particles, allowing operation at higher flow rates [6,7]. However, diffusion is still limiting in HPLC. A typical 'analytical' size HPLC column usually cannot be run much faster than 1 mL min-' (30 minutes to 1 hour total run time) without an unacceptable loss of effieciency due to reduced resolution and capacity. An accepted measure of column efficiency is known as plate height. This is given by the vanDeemter equation [8]:
where H is the plate height and a, b, and c are lumped terms that account for bandspreading due to longitudinal diffusion, eddy diffusion, mobile phase mass transfer, stagnant phase mass transfer, and stationary phase mass transfer. In particular, the c term refers to band spreading caused by pore diffusion. Each of the terms in the equation can be plotted independently and summed to arrive at an estimate of plate height values for different flow rates as shown in Fig. 2-2. At high flow rates, the contribution due to the c term dominates and is the largest contributer to increased plate height under these conditions. POROS' Perfusion Chromatography@media are designed to speed access to the interior of the chromatography particles by overcoming the diffusional mass transfer limitations of conventional chromatographic media. Unlike conventional chromatography particles, perfusive particles have two distinct types of pores - large throughpores that transect the particle and short diflusive pores that branch off from the throughpores, providing a large internal surface area for solute/particle interactions to occur (see Fig. 2-1). Flow through the packed column produces a pressure differential across each particle that induces flow through the throughpores. Sample molecules are carried by this throughpore flow into the interior of the particle and into contact with the network of diffusive pores. This dramatically reduces the slope of
44
2 Systematic Development of PerjLsion Chromatography Technology
H
A
U
Fig. 2-2. vanDeemter plot where H is the height equivalent to a theoretical plate, u is linear velocity of the fluid flowing through the column, A is a measure of interparticle channels and other non-uniformities, B is molecular diffusion in the axial direction, and C represents mass transfer.
the c term of the vanDeemter equation and causes the increase in plate height with increasing flow rate to be much less pronounced. Since the length of the diffusive pores (typically less than 1 pm) is small in comparison with the total particle diameter, the time required for sample molecules to diffuse to and from internal binding sites is very short [1,9]. The combination of the intra-particle flow and short diffusive pores effectively serves to access the entire surface area inside the particles much more rapidly than would be possible with conventional media, which rely solely on diffusion to achieve the same effect. Separations may therefore be carried out much faster on perfusion chromatography media with little or no practical loss in resolution or capacity. Complete run times on the order of a few minutes or less are typical. The remainder of this chapter presents the principles of systematic methods development and illustrates its usefulness when used in conjunction with perfusion chromatography media.
2.3 Principle of Systematic Development Systematic method development strives to take advantage of high-speed chromatographic methods to enable a more complete picture of the behavior of a given separation system to be developed and to utilize that complete picture in a rational design process, replacing the older ‘trial and error’ approaches. Systematic development may be viewed as a process with five stages (Fig. 2-3), as follows: -
Define Delineate the problem to be solved, including the nature of the target molecule itself and the sample from which you are separating it, the analytical methods you will use, the overall goals of the separation, and the resources available.
2.3 Principle of Systematic Development
45
Fig. 2 -3. Stages of systematic chromatography method development.
Experiment Gather empirical information about the behavior of the system with respect to each of the key variables. Evaluate Continually evaluate the resolution, recovery, capacity and practicality of the method. The results of the evaluation are fed back into the design of the experiment. Implement Design the final method, optimize it and put it into practice. Troubleshoot Solve problems with or fine tune the performance of your method. The following sections discuss in more detail the considerations for each of these systematic method development stages.
2.3.1 Define 2.3.1.1 Molecular Characteristics The starting point for defining a separation problem is obviously the target molecule and the sample from which it is to be separated. From the point of view of chromatography, all molecules share certain common characteristics, whether they are simple organic compounds or complex, multi-subunit proteins. These characteristics (Fig. 2-4) may be summarized with the acronym CHASM, which stands for Charge, Hydrophobicity, Affinity, Solubility and stability, and Molecular weight. Charge is a measure of the number of ionic charges (positive and/or negative) which are accessible on the surface of the molecule for binding to the chromatographic packing. Hydrophobicity (the inverse of which is called polarity) is a measure of the ‘oiliness’ or ‘water-hating’ character of the molecule or of functional groups within the molecule. Affinity refers to the presence of sites on the molecule which can interact with other molecules in a biospecific or ‘lock and key’ binding interaction. Solubility and stability are the ranges of chemical conditions and concentrations in which the molecule can stay in solution and maintain its biological ac-
46
2 Systematic Development of Peifusion Chromatography Technology
Negative charges
Hydrophobic groups
Affinity binding sltes
L
Molecular welght (size)
I
Fig. 2-4. Molecular characteristics that control surface binding interactions in chromatography.
tivity. Molecular weight is a measure of the size of the molecule. The shape of the molecule may also be important in some cases. By analyzing the CHASM characteristics of the target molecule relative to other molecules in a mixture, you can begin to look for properties which differentiate the target and will allow its separation. You should try to utilize any unique characteristics as a ‘handle’ for chromatographic binding, since this will give the maximum degree of separation. Once these properties are identified, you can then select the appropriate modes of chromatography, each of which interacts selectively with a particular CHASM characteristic (i.e. ion exchange for charge, hydrophobic interaction or reversed-phase for hydrophobicity, etc.). Note that you should not only consider the characteristics of the target molecule itself, but also the key impurities that must be removed. Sometimes, (such as when you are in the polishing stage of a separation process) it is the impurities rather than the target that become the focus of your efforts. The other aspect of CHASM analysis is to understand the limitations on the separation imposed by the biochemistry of the target molecule and sample. Solubility and stability, in particular, restrict the range of chemical conditions that you may use during chromatography and even the mode of chromatography you may select. You should gather any information about the acceptable pH range, ionic strength or organic solvent concentration your target will withstand with good solubility and stability before you design your experiments. You should also consider any cofactors or other buffer additives that the target molecule may require for either solubility or stability. Some of these additives may be essential for the biomolecule, but may interfere with one or more modes of chromatography, restricting your development possibilities.
2.3 Principle of Systematic Development
47
2.3.1.2 Sample Source Although you may have detailed information about the target and its molecular characteristics, you will most often not have such information about every other molecule in the sample. In some cases, the sample may be extremely complex. However, you must remember that each and every molecule in the mixture will be participating in the separation and may be competing with the target for binding sites on the column. One approach you may use in dealing with this problem is to consider the most important classes of molecules in the sample and to identify the key impurities that are the focus of your separation efforts. Understanding your goal is important here. For example, if you are cleaning up an antibody for use in an assay, simply obtaining a reasonable general purity may be sufficient. If, however, you are attempting to do an X-ray crystallography structural study of that same antibody binding to its antigen, it may be critical to remove traces of all contaminating proteins, particularly all other, non-specific antibodies.
2.3.1.3 Analytical Methods The analytical methods to be used during development are an absolutely critical part of defining your separation problem. During chromatographic method development, you will generate a potentially very large number of samples from separate chromatographic runs as well as individual fractions within those runs. The more samples you can analyze, the more variables you can examine experimentally and the more likely you are to find an optimal solution. Analysis is likely to be the most significant bottleneck you will encounter in the development process. A key strategy in breaking this bottleneck is to differentiate between screening analysis and final analysis. You use screening analysis simply to identify which of a series of chromatographic conditions gives the best separation, or is the best direction to take for further development. Screening analytical methods are fast and must be suitable for a large number of samples, either in parallel or sequentially through automation. Semi-quantitative or even qualitative information is often sufficient. The key is that you can use the screening analysis in a practical way to differentiate between a large number of different chromatographic results. You use final analysis to fully characterize the results of your optimized or semioptimized separation. Final analysis must be precise, accurate, and quantitative, and need not be especially rapid or suitable for large numbers of samples. The key is that you can use the final analysis to determine reliably if your optimized method is fully viable. One useful technique to alleviate the analytical burden is peak tracking (Fig. 2-5). In this method, you identify the peak of interest in an initial chromatogram, either by assaying fractions via screening analysis or running purified target molecule under identical conditions and comparing with a chromatogram of the crude sample. Once the peak is identified, you can often make an incremental change in the chromatographic conditions (shifting the pH, for example), and still be able to identify
48
2 Systematic Development of Perfusion Chromatography Technology
Peak identification
LA-
Chromatognm d sampb
Chromatogram ofstand8rd
Fraction screening analysis
kk-+ li
PeakTracking
pH 8.5
I t
It
I '
pH 7.5
pH 6.5
Fig. 2-5. Peak tracking. After the peak of interest is identified in the first chromatogram, it can be tracked in subsequent runs as small changes in running conditions are made.
the peak of interest from the pattern of the chromatogram. Often the final chromatogram may be very different looking from the first, but because the changes have been incremental, you have been able to track your target peak throughout the process. If this is the case, you may not need to do anything else to analyze the later chromatograms. Another form of peak tracking is now practical due to the speed of perfusion chromatography and the introduction of a system capable of independently switching two columns in and out of the chromatographic flow path (the only commercially available instrument capable of this today is the BioCAD Workstation made by PerSeptive Biosystems). This form of peak tracking has the advatage in that it positively identifies and tracks your product throughout the development process. In this mode, the first column is designed to selectively bind the molecule of interest and allow all other sample components to flow through (e.g. an affinity column). The second is the one on which development is being performed. Two identical runs are executed with the exception that the first run places the first column out of line and the second column in line; while the second run places both columns in line. A comparison of the two chromatograms shows a depletion of one, or some, of the peaks for the run with both columns in line. The location of the depleted peak or peaks indicates where product elutes under those conditions.
2.3 Principle of Systematic Development
49
An increasing trend in analysis is to combine different analytical techniques together in a single, automated method. An example of this is automated peptide mapping techniques in which a protein sample can be digested on-line with subsequent chromatographic separation of the peptide fragments. Recent instrument designs enable multi-dimensional chromatographic assays in combination with on-line immunoassays [ 101. 2.3.1.4 Separation Goals
Understanding your goals and objectives in detail is vital in successfully developing a chromatographic method. Preparative methods fall into a number of general categories:
Sample prep Some preparative methods may be viewed as preparing a sample for some other operation. Desalting, concentration and buffer exchange are typical applications. Sometimes these operations can be carried out non-chromatographically, but with high-speed perfusive media, they may be faster or more practical on a column. - Capture The goal of a capture step is to extract the target molecule from a crude solution, concentrate it and remove some of the bulk impurities. Capture steps are often the first stage in a longer purification process. - Purification The goal of a purification step is to move the target molecule from low or intermediate purity to high purity, where it is typically 90-99% of the final mixture. - Polishing In polishing, the goal is to remove trace quantities of contaminants from the target product which has already been purified. -
The approaches used for different preparative separation categories may be very different. For example, while capture steps often employ large particle, high-capacity media to handle large quantities of crude solutions, final purification and polishing steps are often run with small particle media at lower loading to maximize resolution. Negative chromatography steps (in which the target does not bind to the column while contaminants do bind) are very useful for polishing, but rarely used for capture or sample preparation. Appropriate goals to consider for preparative chromatography include recovery, final product purity, and capacity. The ability to scale up may be important, as well as economic factors such as media costs and cycle life. Where large quantities of material must be processed, process throughput (amount of material produced per given time on a given column) can also be critical. 2.3.1.5 Resources
After you have considered the characteristics of the target molecule and sample, the analytical methods you will use and your separation goals, you should think about the resources you have available for developing the method. For example, you
50
2 Systematic Development of Perfusion Chromatography Technology
may have plenty of time and equipment available, but if you only have a small amount of precious or expensive sample to use, you must limit your experiments carefully. To maximize your chances of obtaining optimal results, determine first how many runs you might be able to perform with the sample you have, and plan your experiments so that the most important variables are covered first. Deadlines and other factors will determine the time available, but you should also consider how much time and effort should be put into a new method. For instance, if you are purifying a protein only once for a limited study, it will certainly not be worthwhile developing a robust preparative method in which you have examined every variable. On the other hand, if you are developing a critical production process that will be used for over a period of years, time spent in fully characterizing and optimizing all the key parameters will almost certainly be well spent.
2.3.2 Experiment Once your separation problem has been fully defined, you next conduct a program of experiments to find a solution. The basic idea is to first outline the key variables which may affect the performance of the separation, then run an experimental program designed to examine the effect of each variable, one at a time. An empirical approach is necessary because of the complexity of chromatography itself, the biomolecules being purified, and the sample matrices. Even though mathematical or computer models may provide useful insights into a development direction, there is no substitute for reasonably comprehensive experimental data. 2.3.2.1 General Experimental Framework One major issue in designing an experimental program is that many of the variables are interrelated, so that changing one will affect the results of changing the others. The surface chemistry of the column packing is the most critical variable in this regard, since it has the most complex effect on both the ultimate separation and on the effects of all the other variables. The pH has a similar effect in the case of ion exchange. It is important to take these dependencies into account so that you will not miss any important effects, but in such a way as not to waste time and sample with unnecessary runs. An example of this problem is the approach most commonly used for ‘screening’ different column packings, in which each is tested under a single, standard set of operating conditions, and the results compared to select the ‘best’ packing for further development. Sometimes this works, but very often the best performance of each packing will not be observed because the single set of conditions selected for screening is not even close to optimal for all the packings. It is easy to miss a unique characteristic of a particular packing because it was not tested under suitable conditions. Taking these dependencies into account, it is possible to utilize a general framework for designing experiments in the systematic approach, as illustrated in Fig.
2.3 Principle of Systematic Development
51
2 - 6 . Experimentation begins with selecting one or more column packings to be
tested, together with a starting mobile phase chemistry. You should then at least roughly optimize the elution gradient for each packing. In the case of ion exchange, this gradient optimization should be performed for at a number of different pH values. An example of this is shown in Fig. 2-5. Once you have completed this ‘mini-optimization’ for each packing, you then choose the best one for further development. With the selected column, you can then ‘fine-tune’ by modifying other variables in the mobile phase one at a time (such as organic solvents, eluent composition, buffer salt, etc.). If you discover a modification that has a large and significant impact on the selectivity or recovery characteristics, you should ideally return to the beginning
Loading Study (optional)
(
Implementation
Fig. 2-6. Basic experimental framework for complete systematic development.
\
52
2 Systematic Development of Perfusion Chromatography Technology
and check the effect of the new mobile phase on each packing in your test panel. This insures that your ‘discovery’ will be thoroughly evaluated. Once you have determined the effects of the chemistry of the packing and mobile phases, you can then perform a loading study. A detailed discussion of the number, nature and range of the variables you should examine in a systematic development cycle is beyond the scope of this article. In general, a number of factors should be considered:
- The class of biomolecule you are separating and its characteristics. -
The modes of chromatography you are using.
- The resources (time, system and sample) you have available. -
Your objectives for the separation (e.g. purity criterian, mass recovery, activity recovery).
Some basic considerations in designing systematic development experiments are given in the following sections. 2.3.2.2 Column The most important aspect of the column is the surface chemistry of the packing in a given chromatographic mode. Some differences in surface chemistry are obvious (such as between strong and weak ion exchange packings) while others are more subtle (such as between the different ligand density packings). In all cases, however, the effects on the separation of changing the surface chemistry are complex and usually unpredictable, so a systematic testing approach is necessary. Fig. 2-7 illustrates the effect that different column chemistries have on the separation of a cell culture supernatant. Other aspects of the column are less complex, and may be optimized for a particular application as part of the implementation stage of method development. For example, the packing particle size can be chosen to balance the required backpressure and resolution. The column volume is generally selected based on the amount of sample being run, and the column hardware and format are chosen based on cost and convenience. Column length affects a number of parameters and should be changed with some care. The number of theoretical plates the column can generate is proportional to length. Column volume (and therefore sample capacity) are also proportional with length, if the diameter is held constant. Pressure drop is proportional to length, which directly affects the maximum possible flow rate. The run cycle time tends to increase with increasing column length. The column length usually does not have a critical effect on the selectivity of the separation itself. 2.3.2.3 pH The pH of the mobile phase is often the most important single variable in a separation. This is almost always true for ion exchange, but is frequently the case with
2.3 Principle of Systematic Development
53
Fig. 2-7. Effect of different perfusion chromatography cation exchange column chemistries on the separation of 0.5 mL of concentrated cell harvest. (Data presented in Fig. 2-7, 2-8, and 2-11 are from work conducted by A. Moisidis, M. McNamara, I. Roberts, and P. Schoofs, CLS Limited, Australia, and reproduced with permission of Today's Life Science, April 1995).
other modes of chromatography such as hydrophobic interaction, reversed-phase or even affinity. The pH affects not only the overall charge and charge distribution of the biomolecules in solution, but also can cause significant changes in their conformation in solution, which can change the functional groups which are accessible to the binding surface. Fig. 2-8 demonstrates the effect of pH on the purification of the partially purified cell harvest obtained from the POROS 20SP column shown in Fig. 2-7. Notice the significant loss of resolution by increasing pH from 4.0 to 5.5 in this example. One convenient means for changing the pH is to utilize a four-solvent blending system. Two of the solvent channels are used for a concentrated buffer, adjusted to the high and low pHs of the buffering range, respectively. The ratio of these two channels sets the pH. The other channels are used for water and concentrated eluent, respectively. By blending these channels appropriately with the ratioed buffer channels, you can vary both the buffer and eluent concentrations at any given pH. Some instrument systems can perform these blending calculations automatically under software control.
54
2 Systematic Development of Perjksion Chromatography Technology
o im
5
E 0.m
0 03
0
3
2
C
aJ
cu
N
a
0.KQ
Minutes
Minutes
0020
:
0
co
N
3
a
Fig. 2-8. pH map of partially purified cell harvest using the POROS SP chemistry shown in Fig. 2-7. Minutes
2.3.2.4 Gradient Optimization Gradient optimization is, in principle, a rather simple matter of adjusting the starting and ending eluent concentrations and gradient duration in column volumes (which sets the slope) for optimal resolution of all the molecules of interest in the mixture. One would think that a steep gradient covering a wide range in eluent concentration could be used directly by setting the starting concentration at the elution point of the first peak of interest, the ending concentration at the elution point of the last peak of interest, and the duration based on the number of peaks in between. Unfortunately, due to the behavior of molecules on a column, such a simplistic approach rarely, if ever, actually works. When eluting in a gradient, molecules are always moving down the column at a rate determined by its isocratic retention at the instantaneous eluent concentration. At very low eluent concentration, the molecule may move very slowly, or imperceptibly. However, as the eluent concentration increases during the gradient, the molecule moves down the column at an ever-increasing rate. Eventually, the eluent concentration can become high enough that the molecule is effectively unretained, at which point it certainly will elute. However, elution may occur well before that point, if the gradient slope is low enough so that the molecule reaches the bottom of the column during the ‘slow movement’ stage. This same effect also serves to increase the peak width with decreasing gradient slope. This is illustrated in Fig. 2-9. The degree to which the elution point shifts with changing gradient slope depends upon the number of sites of interaction between the molecule and the surface, as well
2.3 Principle of Systematic Development
I
55
... causes peaks to elute ... but at lower eluent concentration... /
furtherapah..
Fig. 2-9. Effects of gradient slope on elution.
as the mode of chromatography. For smaller molecules, the isocratic retention often changes rather slowly with eluent concentration, whereas for larger molecules the change may be extremely sharp. For this reason, changing the gradient slope can actually change the selectivity and even cause a reversal in elution order (Fig. 2-10). Thus, the conclusions you may draw from a steep gradient may not be valid as the gradient is optimized. This effect also can make it difficult to design step elution protocols based on the results of gradient runs. At best, the gradient results provide a useful starting point, but the optimal elution steps may well be at a significantly different eluent concentration than the gradient results suggest. Caution should be observed when using steps to separate isocratically closely retained molecules, because even very small
1
2
Eluent Concentatlon
Fig. 2-10. The effect of eluent concentration on isocratic retention vanes with molecular weight. In some cases, a subtle change can not only shift retention dramatically, but can even reverse the elution order.
56
2 Systematic Development of Perfusion Chromatography Technology
0.0
a
2
4
Minutes
0
2
I
4
I
8
Minutes
Fig. 2-11. Example of step elution. For maximum reproducibility, this technique should only be used when selectivity is high as shown here.
changes in the eluent concentration can dramatically affect the retention, especially for larger proteins. Often a shallow gradient may provide a more reproducible separation. Step elution should be used when the selectivity is high as in Fig. 2-11, or when the molecule of interest can be eluted under non-binding rather than weaklybinding conditions.
2.3.2.5 Loading In most applications, you will need to perform a loading study. Method development is often performed at very low loading, primarily to reduce the amount of sample consumed. (If sample is not at all limiting, you should perform method development experiments at higher loads.) In order to process larger amounts of sample, you need to determine how much can practically be run. Unfortunately, the precise behavior of the separation with respect to sample load is not easy to predict. The retention and bandspreading change in a non-linear way with loading. Different molecules can behave very differently, and the mode of chromatography and even mobile phase chemistry can have a major effect. Solubility can also be an important limiting factor, since molecules can become extremely concentrated within a peak on a column, even exceeding their solubility in the mobile phase.
I
2.3 Principle of Systematic Development 1.5
,
IL
57
L ~~
0.00.0
I
MU!I 10
0.0 0.0
5
...
Fig. 2-12. Effect of load on column performance. Notice how the small peak before the main peak becomes unresolvable above 500 pg of sample. This effect is unpredictable and can only be determined by experiment.
The only way reasonably to determine the effect of loading is by experiment simply running the separation with increasing sample loads. The maximum load at which the separation still meets the original objectives for resolution and recovery is referred to as the loadability of the column. Loadability is a function of all the other chromatographic conditions, and should thus only be determined after a separation is optimized. Figure 2-12 shows an example of a loading study. In this example, the small peak preceeding the major peak must remain resolved to provide adequate purification for this step. This remains the situation as sample load is increased from 125 pg to 500 pg. However, as the load is increased further to 1 mg, the small peak becomes unresolvable. Therefore, sample amounts should not exceed 500 yg for this column.
2.3.3 Evaluate During the experimental process, you will constantly need to evaluate the performance of the chromatographic runs against your objectives. The primary criteria are the resolution of the separation and the recovery of the product. For process applications, the capacity of the column is also important. In addition, there are a number of practical aspects that should be considered in evaluating a separation.
58
2 Systematic Development of Perfusion Chromatography Technology
2.3.3.1 Resolution and Purity Resolution is the actual measurement of the degree of separation between the target molecule and other molecules in the mixture. Although resolution is a convenient single parameter for characterization, you must always remember that it is the result of two completely independent effects - selectivity (the difference in retention between two peaks) and bandspreading (the width of each peak). Selectivity is perhaps the most critical parameter in achieving a satisfactory result, whether the objective is an analysis or a kilogram-scale preparative separation. The retention of each molecule in a mixture is affected by almost every aspect of the chromatographic method, including the selection of packing material, the mobile phase chemistry and the elution profile. Because it is critically affected by so many parameters, selectivity can be challenging to optimize. Bandspreading becomes important when the selectivity has been optimized and the peaks are still not completely separated. Peak width is determined primarily by the particle size of the packing material, but is also affected by sample load, column length, elution profile and even the details of the surface and mobile phase chemistries. Poor resolution can be corrected by dealing with either one of these factors (Fig. 2-13), but the techniques employed to do so and the costs involved in time and money may be quite different. It is often a better strategy to focus on improving the selectivity through method development (even though this is more challenging) since improvements in efficiency almost always come at the expense of higher pressure drop, lower capacity, and higher media costs. However, in cases where a purification is being run a few times on a relatively small scale, using the ‘brute force’ approach of higher efficiency may take less time and be quite effective.
I
Poor Resolution
I
Improve Etnclency
Improve Selectlvlty
(Easy development, higher pressure, reduced capadty)
(Hard development, same pressure, increased capadty)
I
Fig. 2-13. Effects of improving selectivity versus efficiency.
2.3 Principle of Systematic Development
59
Formal measurements of resolution based on peak separation are not the only measurements of the separation itself. Purity or specific activity refers to the amount of the specific target molecule (or target molecule activity) divided by the amount of total material (such as protein) present in the sample. The purification factor refers to the specific activity of the final eluted product divided by the purity of the starting sample. In the case of preparative chromatography, the purification factor may be a better, more meaningful measurement of the performance of the chromatographic separation than more abstract parameters such as resolution. In the case of a polishing step, where the target product is already at high purity, performance is often better measured by determining the concentration of the contaminants being removed.
2.3.3.2 Recovery It is not sufficient for a chromatographic column simply to bind molecules from a sample and separate them. Those molecules must also be eluted completely from the column. Recovery of product or sample can be measured in two different ways. Mass recovery simply refers to the amount of material eluted from the column divided by the amount injected. This is usually the most important measurement for analytical chromatography. Activity recovery refers to the amount of biological or functional activity eluted from the column divided by the amount injected. This may be very different than the mass recovery for complex protein molecules, since the conditions used for elution of all the mass may, in some cases, give rise to denaturation of the proteins and loss of their activity. Often the term yield is used instead of recovery.
2.3.3.3 Capacity The formal measurements of saturation capacity (amount of material bound per unit of packing at equilibrium saturation) and dynamic capacities (amount bound at breakthrough as a function of flow rate in a packed column) are an important starting point for determining capacity. However, the full capacity of the column can almost never be utilized completely. This is because bandspreading always increases and resolution thus decreases as the load of sample on the column is increased. The rate of decrease is a function of the particle diameter, with smaller (more efficient) particles showing a much more rapid decrease in resolution with load. This is one reason that larger particle packings are acceptable for preparative applications. As stated in the discussion concerning loading, a more relevant parameter is the loadability, defined as maximum effective sample load at which the required resolution or purity and recovery can be obtained. Again, loadability must be determined empirically in a loading study and should be performed after the rest of the separation is fully optimized.
60
2 Systematic Development of Perfusion Chromatography Technology
2.3.3.4 Practicality A wide range of different factors determine the practicality of a separation method for a given application. Issues such as the pressure drop, throughput, scaleability, reproducibility, or cost can all play an important role. The relative importance of these factors depends upon the specific objectives and situation. Robustness is one practical factor that can be important in preparative and process separations. Robustness is defined as the ability of a separation to withstand small changes in operating conditions (such as sample composition or mobile phase pH) and still meet the objectives. While not critical for methods that will only be run a few times, robustness is a highly desirable characteristic for protocols that must be used for long periods or transferred to different individuals or groups. Systematic development is key in producing robust separations methods.
2.3.4 Implement Once the basic parameters of the separation system have been experimentally explored and evaluated, you then can actually design and test a separation protocol to meet your objectives. How you approach this implementation stage depends very much upon your original goals. The following are some implementation considerations for multi-step separations and scale-up.
2.3.4.1 Multi-step Separations Purification from complex samples almost always requires multiple steps to achieve the degree of purity and level of removal of critical contaminants required. In designing these separations, the particular goal of each step (e.g. capture, purification, or polishing) should be considered, and the most appropriate mode and packing for each step selected. This can be a complex exercise, since so many different factors must be considered. The ‘database’ of information developed during a systematic development cycle is invaluable in making these design choices. When integrating multiple chromatographic steps, one important consideration is the sample preparation required for each step. Wherever possible, steps should be compatible with each other (i.e. the conditions for elution from one column should be conducive for binding to the next column). Reversed-phase and ion exchange are a good example of compatible methods, since the eluents used for each generally do not affect the binding to the other. Where complete compatibility is not possible, minimizing the sample processing between steps should be considered. For example, samples for ion exchange must be introduced to the column in low salt and are eluted in medium-to-high salt concentrations. Samples for hydrophobic interaction chromatography (HIC) must be introduced in high salt and are eluted in medium-to-low salt concentrations. If an HIC step is placed before an ion-exchange step, the salt would have to be removed,
2.3 Principle of Systematic Development
61
which can involve an extra dialysis or desalting step. If ion exchange is run first, however, salt only must be added prior to the HIC step, which can be a much simpler process. Simple dilution is a sample preparation strategy with high-speed media that is not often practical with conventional packings. Because perfusion chromatography columns can be loaded at very high speeds, it may well be faster to simply dilute a sample until the conditions are suitable, and directly load the diluted sample onto the column at high flow rate. This approach can often eliminate or at least simplify sample preparation, even with 'non-compatible' steps.
2.3.4.2 Scale-up When developing preparative or process chromatography, it is usually necessary to develop the method on a small scale, then scale-up the method to process a larger amount of material. Chromatography generally scales linearly without too much difficulty, provided a few key rules are followed: -
Keep the linear velocity and column length constant.
- Keep the sample and mobile phase compositions constant. - Increase the bed volume in proportion to the sample volume. -
Increase the various elution volumes in proportion to the bed volume.
Figure 2-14 illustrates these principles for scale-up. Notice that the column length, linear flow rate, sample preparation, and mobile phase buffers were all kept constant going from the small column to the large one.
T ''
Fig. 2-14. Example of proper scale-up technique. Left panel: 3 mL crude ascites (14 mg antibody) diluted 1:3 with starting buffer and run at 2.7 mL min-' (1000 cm h-I) on a POROS 50HQ 4.6 mmID X 100 mmL column. Starting buffer was 20 mM Trishis-tris propane at pH 8.5. Gradient went to 20 mM Trishis-tris propane pH 8.5 with 500 mM NaCl in 18 column volumes. Right panel: 10 mL crude ascites (70 mg antibody) diluted 1:2 with starting buffer and run at 13 mL min-' (1000 cm h-I) on a POROS 50HQ 10 mmID X 100 mmL column. Starting buffer was 20 mM Trishis-tris propane at pH 8.5. Gradient went to 20 mM Trishis-tris propane pH 8.5 with 500 mM NaCl in 20 column volumes.
62
2 Systematic Development of Peifusion Chromatography Technology
Some practical problems may interfere with simple scale-up. It is critical that the column design maintain a uniform flow distribution as the diameter is increased. The solvent blending and pumping systems may require different technologies for operation at the different scales. Sometimes the pressure drop may be limiting at large scale. Cost factors may prevent a direct linear scale-up. One key concept when developing a large-scale process is to think about the constraints of operation at large-scale and run the small-scale development system within those constraints for testing and optimization. An example of this idea of ‘scale down’ is to test a longer column than would ordinarily be run in a laboratory environment. At large scale, adding to column diameter can be far more expensive than adding column length, so a long, narrow column may be more practical than a short, fat one. On the other hand, increasing the column length increases the backpressure and decreases the maximum flow rate, increasing the separation time. Since the effect of column length is so critical, the effect of column length should be tested at small-scale. The advent of high-speed, high-throughput chromatography media has opened up a number of different design options for purification processes. With conventional media, capture of dilute feed streams is often a two-step process, in which a preconcentration stage using something like ultrafiltration is used to remove the water and a slow chromatography column is used to do the crude purification. With high-speed media, these two can be combined in a single chromatographic dilute feed capture step (Fig. 2-15). A second interesting option is to use the high throughput of perfusion media to reduce the size of the columns needed through rapid cycling [ll]. If a high-speed column can be operated with a cycle time, say, one-fourth that of a conventional column, then a column one-fourth the volume could be run four times in a row to get the same throughput (Fig. 2-16).
Conventional
I-
Perfusion
I
Fig. 2-15. Dilute-feed capture using perfusive media can eliminate preconcentration steps.
2.3 Principle of Systematic Development
63
Conventional Process
Cycling Process
Fig. 2-16.Cycling purification with high throughput column using perfusion chromatography media.
Cycling operations are not always desirable, for reasons of validation, quality control, etc. However, in many cases there are significant benefits of reduced space requirements, reduced operating costs, and greater flexibility of operation. Cycling can confer a particular advantage when the column packing is quite expensive, as is the case for affinity chromatography media. In research, cycling can completely eliminate the need to scale-up the column itself, but allow large amounts of material to be produced with the same column used for method development. This not only eliminates the cost of a larger column, but also reduces the uncertainty involved in scaling up.
2.3.5 Troubleshoot Even when following a systematic approach, at some point you may need to simply solve problems that arise with the method. Problems can arise in a number of different areas - bandspreading, peak shape, selectivity, recovery, pressure, reproducibility, column life, etc. Sometimes a method that has worked perfectly well for some time can suddenly fail. Solutions can range from the very obvious or simple to the very subtle or complex. A complete guide to troubleshooting chromatographic methods is well beyond the scope of this article. However, one cardinal rule in troubleshooting is to change only one variable at a time. This is also the key to systematic development and, in fact, the data you obtain by using a systematic approach will prove to be invaluable when any problems do arise. In troubleshooting, the one-at-a-time approach is absolutely critical, because you not only need to solve a problem, but you also need to know why it happened so that it will not recur.
64
2 Systematic Development of Perfusion Chromatography Technology
2.4 Summary Although systematic techniques are far superior to trial-and-error approaches for methods development, they have not been efficient to employ due to the characteristically long run times of conventional liquid chromatographic media. This has changed dramatically in the past few years with the introduction of high-speed perfusion media. Now, using perfusion chormatography in conjunction with systematic development techniques such as those described here, it is possible to gather complete information, including failure analysis data, about a purification project in less time than it normally takes to perform trial-and-error-based development. This not only allows the method to reach a higher level of optimization, but, since so much is learned about the characteristics of the purfication, also greatly reduces the effort required to redevelop the method should the design goals change.
References [ l ] Afeyan, N. A., Gordon, N. F., Mazsaroff, I., Varaday L., Fulton, S . P., Yang, Y.B., Regnier, F. E., J Chrornatogr 1990, 519, 1-29. [2] Afeyan, N.B., Fulton, S . P., Regnier, F. E., J Chromatogr 1991, 544, 267-279. [3] Afeyan, N. B., Fulton, S . P., Regnier, F. E., LC-GC 1991, 9, 824-832. [4] Peterson, E. A., Sober, H. A,, J Am Chem Soc 1956, 78, 751. [5] Porath, J., Flodin, P., Nature (London) 1959, 183, 1657. [6] Chang, S . H., Gooding, K. M., Regnier, F. E., J Chromatogr 1976, 225, 103. [7] Fagerstam, L., Soderberg, L., Waahlstrom, L., Frednksson, U. B., Plith, K., Wallden, E., Protides Biol Fluids 1982, 30, 621. [8] vanDeemter, J. J., Zuiderweg, F. J., Klinkenberg, A., Chem Eng Sci 1956, 5, 271. [9] Afeyan, N. B., Fulton, S . P., Regnier, F. E., in: Applications of Enzyme Biotechnology: Kelly, J. W. and Baldwin, T. 0. (Eds.). New York: Plenum Press, 1991. [lo] Nadler, T., Blackbum, T., Mark, J., Gordon, N., Regnier, F. E., Vella, G., J Chromatogr 1996, 743, 91-98. [ l l ] Fulton, S. P., Shahidi, A. J., Gordon, N. F., Afeyan, N. B., Biomechnology 1992, 10, 635.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
3 Hydrophobic Interaction Chromatography of Proteins Eric Grund
3.1 Introduction Suitable media for hydrophobic interaction chromatography (HIC) were first introduced commercially about 20 years ago [l], and since then a large range of products has become available. The technique is now widely used for both preparative and analytical separations of proteins [2] and there is interest in other application areas such as carbohydrates [3]. Applications for the downstream purification of enzymes, antibodies or recombinant proteins are frequent in commercial processes, where HIC complements other chromatography techniques such as ion exchange and gel filtration. However, industrial applications are rarely reported in detail in the literature. This chapter will attempt to give an insight into the practical use of hydrophobic interaction chromatography in industrial downstream purification.
3.2 Theory Hydrophobic interaction describes the non-covalent association of non-polar moieties which occurs in aqueous solutions. The dissolution of non-polar moieties in water is thermodynamically unfavorable, so they associate with each other [4] (Fig. 3-1). The interaction is promoted by high concentrations of ‘salting-out’ salts such as ammonium sulfate, an effect which parallels the ability to precipitate proteins from aqueous solutions according to the Hofmeister or lyotropic series (Fig. 3-2). Several models have been proposed to account for hydrophobic interactions in a chromatographic context [5-81. Proteins and peptides contain hydrophobic (i.e. non-polar) aliphatic and aromatic side chains, which are largely hidden within the molecule and are important for maintaining native tertiary structure. However, all proteins expose some hydrophobic side chains, and these will associate with hydrophobic ligands attached to a supporting chromatographic matrix.
66
3 Hydrophobic Interaction Chromatography of Proteins
p:+:p-Jqs ::: ooooooo
ooooo
000000
oooooo
e
ooooooo
s + o o o o
000000
P
0
P=Polymer matrix S=Solute molecule L=Ligand attached to polymer matrix H=Hydrophobic patch on surface of solute molecule W=Water molecules in the bulk solution
O 0 0
w ..
Fig. 3-1.Close to the surface of the hydrophobic ligand and solute (L and H), the water molecules are more highly ordered than in the bulk water and appear to ‘shield’ the hydrophobic moieties. Added salt interacts strongly with the water molecules leaving less available for the ‘shielding’ effect, which is the driving force for L and H to interact with each other. (Copyright Pharmacia Biotech, reproduced with permission.)
+increasing precipitation ( ‘salting-out’) effect Anions: PO4,-, SO,’-,CH,. COO-, CI-, Br, NO,-, CLO, SCN-
I-,
Cations: NH,+, Rb+, K+, Na+Cs+, Li+, W+, Ca2+, Ba2+ Increasing chaotropic (‘salting-in’) effect
Fig. 3-2.The Hofmeister series on the effect of some ions on precipitating proteins.
3.2.1 Solvent Additives Early models emphasized the correlation between the effect of salts and other additives on the surface tension of water and their different abilities to promote binding of solutes to or elution from a HIC medium [9] (Fig. 3-3). In the complete picture, additives and changes in conditions (salts, pH, temperature, organic solvents, etc.) can influence three components of the chromatographic system: 1. The properties of the solvent (water). 2. The hydrophobic ligand. 3. The solute molecule containing hydrophobic groups. Strongly ‘salting-out’ salts mainly influence the properties of the solvent, increasing surface tension and leading to an increase in hydrophobic interaction between solute and adsorbent. ‘Salting-out’ salts tend to stabilize secondary and higher structures of protein [10,11]. In contrast, some salts, like magnesium chloride, have little effect on binding despite a large positive effect on surface tension of water. Chaotropes or ‘salting-in’ salts, e.g. potassium thiocyanate, have the opposite effect and are believed to react to varying degrees with solvent-accessible hydrophobic patches on the surface of proteins, leading to their solubilization. This can perturb the structure of proteins as well as dissociating them from an immobilized hydrophobic ligand. Urea, guanidine hydrochloride [12,13] and detergents [ 141 also decrease
3.2 Theory
67
REDUCTION IN FREE VOLUME VAN DER WAALS FORCES BETWEEN LIGAND AND PROTEIN
ELECTROSTATIC INTERACTION WITH SOLVENT
WITH SOLVENT
REDUCTION OF CAVITY IN SOLVENT
Fig. 3-3. Schematic illustration of the association between an amphiphilic solute and a hydrophobic stationary phase ligand. Water represented by open squares covers the molecular surface area by which the total cavity is reduced upon contact of the two species. The arrows symbolize the forces acting in the system. (From Horvath, C., Melander, W., American Laboratory 1978,10, 17-36; reproduced with permission.)
binding, presumably through their well-described effects on proteins, although detergents will presumably also bind to the hydrophobic ligand. These additives normally denature proteins and are never a first choice in large-scale applications. Ethylene glycol has been used widely as an eluent on a laboratory scale, which reflects its ability to decrease the surface tension of the solvent while not showing extensive interaction with either protein solutes or hydrophobic ligands [ 15,161. Glycerol probably falls into the same category [17]. Methanol, ethanol and isopropanol have effects both on the solubility of non-polar groups, influencing the structure of proteins, and on the properties of solvent, strongly decreasing surface tension. They are used to recover strongly bound proteins but may result in denaturation and irreversible loss of activity. For a comprehensive review of the effect of additives see Arakawa and Narhi [ 181.
3.2.2 pH The pH of the equilibration buffer or eluent can also have an influence on hydrophobic interactions [19,20], although the outcome in chromatography is often difficult to predict. Observed changes in elution positions tended to be greater below pH 5 and above pH 8.5 (Fig. 3-4). Theoretically, the effects could be due to titration of
68
3 Hydrophobic Interaction Chromatography of Proteins
2
4
6
8.5
8.5
10
pH
12
Fig. 3-4. The pH dependence of the interaction between proteins and an octyl agarose gel, expressed as V,Nt (V, is the elution volume of the different proteins and V, is the elution volume of a non-retarded solute). Elution was by a negative linear gradient of salt. The model proteins used were soy trypsin inhibitor (STI), human serum albumin (A), lysozyme (L), transfenin (T), ribonuclease (R), egg trypsin inhibitor (ETI), and cytochrome c enolase (E), ovalbumin (O), (C). (From [19]; reproduced with permission.)
charged groups in proteins, increasing their hydrophobicty, or to changes in protein structure with pH. Since ammonium sulfate is the favored salt additive, most HIC experiments are limited to neutral or acid pH to avoid volatilization of ammonia.
3.2.3 Temperature Increasing temperature normally increases retention of proteins on a HIC medium, which is predictable since hydrophobic interaction is entropy-driven. However, peak widths can also be affected [21] in opposite directions for different proteins, possibly due to variable effects on the tertiary structure of proteins and on their flexibility. Moreover, temperature has a profound influence on the viscosity of the solvent (which is already high due to high salt concentrations) which in turn influences diffusion processes as well as backpressures in chromatography. Consequently, control of temperature is very important in HIC.
3.2 Theory
69
3.2.4 HIC versus Reversed-Phase-Chromatography In HIC and reversed-phase chromatography (RPC) the underlying mechanism responsible for the binding proteins is fundamentally similar. However, HIC uses a hydrophilic matrix substituted with a relatively low concentration of hydrophobic ligands. Proteins normally require a high concentration of salt to encourage binding and their elution is achieved with a step-wise or linear gradient down to low salt concentration, typically at constant, neutral pH (Fig. 3-5). These conditions are gentle or stabilizing for most proteins (although precipitation of some proteins might occur at high concentration of salt) and usually biological activity is not affected.
SOURCE 1 5 PHE
SOURCE 15 I S 0
\
i
do
SOURCE 15 ETH
10.0
15.0
20.0 Time
Fig. 3-5. Separation of a protein mixture on three different HIC media with the same polymeric matrix (SOURCE' 15, a porous, polystyrene divinyl benzene matrix, uniform particle diameter 15 pm) and three different ligand types: phenyl (PHE), isopropyl (ISO) and ether (ETH). Sample: myoglobin, ribonuclease, lysozyme, alpha-chymotrypsinogen. Column: RESOURCE', 1 mL, 6.4 mm i.d. X 30 mm (bed height). Flow rate: 1 mL min-I. Eluent A: 2 M ammonium sulfate in 0.1 M potassium phosphate, pH 7.0. Eluent B: 0.1 M potassium phosphate, pH 7.0. Gradient: 0-100 % B in 20 column volumes. (Copyright Pharmacia Biotech; reproduced with permission.)
70
3 Hydrophobic Interaction Chromatography of Proteins
RPC on the other hand uses a highly hydrophobic matrix substituted with hydrophobic ligands to the extent that most proteins and peptides bind in pure water or in buffers containing low concentrations of salt. Elution of bound solutes is achieved with a gradient of a water-miscible organic solvent such as acetonitrile, methanol or isopropanol, sometimes at low pH using ion pairing agents such as trifluoroacetic acid (TFA). The main practical consequence of this seemingly artificial difference is that proteins will usually maintain their native structures in the conditions used in HIC, whereas they often do not during RPC. Consequently, for preparative applications, HIC is generally successful for a wide range of proteins at all stages of their purification, whereas RPC is useful in specific cases, where stability is not a problem, usually in the final polishing stage to remove structurally similar contaminants.
3.2.5 Selectivity The theoretical modeling of HIC does not obviously predict a key property of HIC, i.e. that selectivity can differ for different HIC media and for the same medium under different operating conditions, to the extent that proteins can even elute in reverse order [17] (Fig. 3-6). Different hydrophobic ligands or the same ligands at different degrees of substitution, will usually bind specific proteins with different affinities.
Ammonium Sulfate
Potassium Phosphate
2.00
1.50 1.25 1.00 C Molarity of salt during elution
1.75
Fig. 3 - 6 . Variations in selectivity between ammonium sulfate and potassium phosphate on SOURCE" 15ISO. Proteins were injected individually and the chromatograms overlain. Lysozyme (LYS), transferrin (TRF), chymotrypsinogen (CTG), alpha-chymotrypsinogen (aCT), R-phycoerythrin (RPE), monoclonal IgGl (Mab). (From [21]; Reproduced with permission.)
3.3 Purification Strategies
71
The same ligands on adsorbents constructed in different ways (different coupling chemistry, different hydrophilization or different matrix) can also have important differences in selectivity and therefore usefulness for a specific separation. Multivariate analysis was used in one study to characterize similarities and differences between certain hydrophobic ligands coupled to Sepharose [22], Aromatic, aliphatic, charged and thiol-containing hydrophobic ligands can be expected, intuitively, to offer potentially useful differences in selectivity, and for production applications careful screening is decisive. In addition, the use of salts and other additives to modulate separations has been widely studied [17,18]. However, care should be taken before exploiting complicated regimes for large-scale applications where keeping methods simple and robust is an important maxim.
3.3 Purification Strategies A downstream purification process can be conveniently divided into three stages [23] (see Fig. 3-7): -
Capture of the desired protein from the crude starting material. Intermediate purification from other proteins and contaminants. Polishing to remove traces of impurities, particularly those resembling the desired product.
Cell Separation
I Clarified Culture Medium
Cell Disruption
I Cell Debris Removal
Fig. 3-7. Stages in downstream processing. (Copyright Pharmacia Biotech, reproduced with permission.)
72
3 Hydrophobic Interaction Chromatography of Proteins
The goals of each stage are somewhat different, as listed in Fig. 3-8, 3-9 and 3-10, which leads to a different approach to using HIC depending upon the stage in the downstream purification process. Put another way, Capture must interface with a complex starting material; Polishing must interface with the pure final product; and in Intermediate purification chromatographic techniques are interfaced with one another.
Some typical Capture objectives Stabilization of product Removal of proteases, etc. Removal of solids Removal of water (concentration) Removal of bulk quantities of proteins, NA's, carbohydrates Preparation for further chromatography
Fig. 3-8. Typical objectives of the Capture stage. (Copyright Pharmacia Biotech, reproduced with permission.)
Some typical Intermediate Purification objectives Removal of most proteins Removal of most nucleic acids Removal of endotoxins Removal of viruses
Fig. 3-9. Typical objectives of during Intermediate purification. (Copyright Pharmacia Biotech; reproduced with permission.)
Some typical Polishing objectives Removal of trace amounts of
- Host proteins - Structural variants of the product - Reagents - Leachables - Endotoxins - Nucleic acids - Viruses
Adjustment of pH/salts/additives
Fig. 3-10. Typical objectives during Polishing. (Copyright Pharmacia Biotech, reproduced with permission.)
3.3 Purification Strategies
73
3.3.1 Capture HIC is a good candidate for Capture since it can often concentrate the product and remove large amounts of contaminants. Of course, a prerequisite is that the starting matrial must tolerate the salt conditions required for binding. Frequently, moderate concentrations of ammonium sulfate (the salt most widely used to promote binding) will stabilize activity of proteins such as enzymes, but nevertheless a potential hazard is precipitation. Not only must the target protein remain in solution in the absence of the hydrophobic matrix, but so must all the contaminants in the crude starting material. Once applied to the column, the proteins must not denature, aggregate or precipitate on the column. For this reason, and for simple cost considerations, it is normal to try to minimize the salt concentration during sample application. HIC is most effective for Capture when the target molecule binds at a moderate concentration of ammonium sulfate, whereas most of the contaminants do not bind (Fig. 3-11). In Capture, the chromatographic step has to handle a high throughput and remove bulk impurities and excess water. Preliminary selection of media takes this into consideration. High selectivity is desirable during adsorption to achieve as much purification as possible during this phase of the run. The binding affinity must be high enough to concentrate the product and allow impurities to be washed off the column, but not so high that poor recovery of the product or on-column denaturation is a consequence. Commonly, a series of washing steps is carefully optimised to remove
100
3.0
2.0 50 1 .o
0.0
50
"0
Volume (titer)
Fig. 3-11. Production-scale Capture of h-EGF on Phenyl Sepharose@6 Fast Flow (high sub), a highly cross-linked, porous agarose matrix highly substituted (high sub) with phenyl ligands (also available at low substitution: low sub). Column: BPG 300 mm (id.) X 100 mm (bed height). Sample: yeast supernatant, ammonium sulfate added to 0.5 M. Sample volume: 80 L. Sample load 0.36 mg h-EGF/mL media. Flow rate: 212 L h-' (300 cm h-', loading), 42 L h-' (60 cm h-I, elution). Buffer A: 20 mM sodium phosphate, pH 7.0 + 0.5 M ammonium sulfate. Buffer B 20 mM sodium phosphate, pH 7.0. Purification time 1.5 h. Sample application stops and wash with buffer A starts at the point where the A280 drops from its plateau level. Elution with buffer B starts at the point where the conductivity trace drops. (Copyright Pharmacia Biotech, reproduced with permission.)
74
3 Hydrophobic Interaction Chromatography of Proteins
desired impurities while the product is firmly bound to the column. Binding capacity, scalability, ability to handle crude samples without clogging, existence of effective cleaning methods, operation at high flow rates and ability to operate at low backpressures (i.e. through a relatively large particle size) are also important considerations for large-scale use. Occasionally, an opposite strategy can be exploited where HIC is used to bind and remove contaminants, rather than to bind the product of interest. Capture of the target molecule is left for subsequent steps. Since batch-to-batch variability in the sample is normally greatest at the upstream end of a purification process, using HIC as a first Capture step requires careful testing for robustness of the method over the full range of conditions and loads likely to be encountered.
3.3.2 Intermediate Purification HIC is a useful complement to ion exchange, gel filtration and affinity chromatography, separating, as it does, according to a different property. It often provides the key to removing critical contaminants. Consideration must be given to linking with earlier and later steps to minimize intermediate sample treatment and provide an efficient process flow. For example, it may be possible to elute from an ion exchanger at high salt concentration and go directly onto a HIC column. Alternatively, it may be possible to choose a strong HIC adsorbent and elute at very low ionic strength, to ensure that little sample adjustment is needed before a subsequent ion exchange step (Fig. 3-12). Linking techniques
HIC
L-3 L
1
Fig. 3-12.Ion exchange (IEX) and HIC can be linked without very much adjustment or the need for an intermediate step. In practice, some dilution or addition of salts is usually required. (Copyright Pharmacia Biotech; reproduced with permission.)
3.3.3 Polishing Gel filtration and RPC are the favored techniques for Polishing because of the nature of the separation problems most frequent at this stage, and because they can deliver the product in suitable conditions for storage or drying. However, HIC is occasionally used for Polishing (Fig. 3-13) typically with an optimized gradient elution, to
3.4 Methodology
0
20
40
75
60 Time (min)
Fig. 3-13. Polishing a partially purified recombinant Pseudomonas aeruginosa exotoxin using SOURCEB 15PHE. Column: FineLINETMPilot 35 mm (i.d,) X 100 mm (bed height), bed volume 100 mL. Sample: Recombinant P aeroginosa exotoxin (expressed in E. coli.), partially purified, adjusted to 1 M ammonium sulfate. Sample load: approximately 250 mg. Flow rate: 33 mL min-’ (200 cm h-I). Eluent A: 1.0 M ammonium sulfate, 50 mM sodium phosphate, pH 7.0. Eluent B : 20 mM sodium phosphate, pH 7.0. Gradient: A + 0-45 % B in 15 column volumes. (Copyright Pharmacia Biotech, reproduced with permission.)
exploit fully selectivity differences during desorption, and with a small bead size medium to obtain high efficiency. Since Polishing normally involves the most difficult separation, conditions may need fine tuning. It may be relevant to consider selectivity adjustments by using different salts or pH. However, strict control will be needed to ensure scalability and reproducibility. A sensitive separation may succeed at the laboratory scale but may fail in the production hall because of uncontrolled changes in temperature, pH, time in holding tanks, equipment dead volumes, accuracy of gradients, etc. during scale-up and transfer from the laboratory [24].
3.4 Methodology Arriving at a method suitable for large-scale HIC involves a number of decisions which are made on the basis of experience and systematic tests. Briefly, tests are used to scout amongst media and to study conditions for binding and for elution and, after inital choices have been made, further optimization is done to get the most from the separation. Figure 3 -14 shows the fundamental chromatographic parameters which are optimized and illustrates that the emphasis is on throughput (capacity and speed) for Capture steps and on recovery and resolution for Polishing steps. With an enormous number of choices, it is important to avoid being trapped by an excess of experimental runs and consequent analyses. Experience often allows one
76
3 Hydrophobic Interaction Chromatography of Proteins CAPTURE
Rs
POLISHING
Rs
C= Capacity Rs= Resolution S= Speed Rec= Recovery (yield)
Fig. 3-14. Optimization of chromatographic parameters is always a balance. Typically, priorities differ in early Capture steps compared with late Polishing steps. In Capture, loading capacity and speed (flow rate) are important determinants of productivity. In Polishing, focus is usually on resolution and recovery. Removal of defined impurities is a goal at all stages. (Copyright Pharmacia Biotech, reproduced with permission.)
to avoid unnecessary experiments and statistical design helps to cut down still further.
3.4.1 Stability Window Even though moderate concentrations of ‘salting-out’ salts usually stabilize protein structure and activity, there are limits, and precipitation often occurs at some level. It is important to characterize the activity of the product and its stability in solution under conditions which may be used during chromatography (Fig. 3 -15). Aggregation or precipitation is sometimes quite slow [ 2 5 ] , allowing short exposure to high salt concentration. Online mixing is one way to achieve this in a controlled fashion, Without characterization, however, this time-dependence is a potential pitfall since a rapid experiment which works in the laboratory may unexpectedly fail in the production hall where hold-up times are usually longer and the sample may be left standing at high salt for a longer period of time. Subsequent to studies in free solution, other issues must also be addressed related to the interaction with a HIC matrix: On-column denaturation can occur [26-321, presumably through unfolding onto the hydrophobic ligand, even under conditions where the protein is stable in solution. This can be time-dependent and either reversible or irreversible, and is something to be aware of when choosing the medium (see below). - Denaturation during elution, especially when additives are used. As can be appreciated from the discussion under theory (see above), organic solvents, urea, guanidine hydrochloride and salting-in salts can all cause structural changes in the
-
3.4 Methodology
77
Alkaline Phosphatase stability
[Ammonium Sulphate] M Fig. 3-15.Activity of alkaline phosphatase from E. coli was studied at three different pHs over a range of ammonium sulfate concentrations, to determine the stability window for the application of HIC. (Copyright Pharmacia Biotech, reproduced with permission.)
desired product. The reversibility of these changes and recovery of activity must be established from case to case.
3.4.2 Chromatographic Medium The choice of appropriate chromatographic medium for a particular separation is influenced by both general considerations such as the reliability of a supplier, the suitability of a matrix for the stage in the separation, the envisaged final scale of operation, stability to cleaning methods, economics, etc. (Fig. 3 -16) and specific considerations such as the ability to separate product from an important contaminant, recovery of biological activity (Fig. 3-17), or loading capacity. Considering a specific separation, the extremes are set by the maximium salt concentration which the sample can withstand (stability window) - which will limit the choice of less-hydrophobic media, and the product recovery in a simple gradient down to low salt, which will limit the choice of more- hydrophobic media (Fig. 3-18).
78
3 Hydrophobic Interaction Chromatography of Proteins
Pre-selection of Separation Media
Compatibility with
requirements
Fig. 3-16.Pre-selection of media candidates eliminates many experiments. For example, only those which are truly compatible with the needs of the application, production-scale use and regulatory requirements, need to be tested. (Copyright Pharmacia Biotech, reproduced with permission.)
ETH
100%
6
8
d
IMass
Fig. 3-17. For sensitive proteins there may be loss of activity during HIC related to the hydrophobicity of the ligand or degree of substitution of the adsorbent. Here a specific monoclonal IgM was found to elute with good recovery from SOURCE' ISETH, but recoveries were poorer from more hydrophobic SOURCE 151SO and SOURCE 15PHE. (From [25]; reproduced with permission.)
3.4 Methodology
YoB
A280nrn
0.64
I
%B 100
A281
100 0.64 7
I
79
I I
,
Source 151SO
Source 15PHE
I I
,
I I
,
0.48
0.48
,
0.32
50
-
0.32 -
.50
, I
0.16
0.16
c 0 0
<
-
,
I
I
10
20
lime (rnin)
0
0 0
I 10
I 0 20 lime (rnin)
Fig. 3-18. In this example SOURCE" 15ISO is the preferred adsorbent because the majority of the monoclonal antibody (shaded peak) is recovered without resorting to the use of additives. With SOURCE ISPHE, although a large peak of IgG is eluted at the end of the gradient at low salt concentration, isopropanol is required for quantitative recovery. Column: RESOURCE' 1 mL, 6.4 mm (i.d.) x 30 mm (bed height). Sample: 50 p1 mouse ascites fluid containing IgG2b monoclonal antibody. Flow rate 1 mL min-' (200 cm h-I). Eluent A: 1.5 M ammonium sulfate, 50 mM sodium phosphate, pH 7.0. Eluent B: 50 mM sodium phosphate, pH 7.0. Gradient: A + 0-100 % B in 15 column volumes. Isopropanol (30 %) in eluent B was applied at the end of the gradient, indicated by the bar. (Copyright Pharmacia Biotech, reproduced with permission.)
For a Capture step, media can be screened by simple test-tube experiments at preferred binding salt concentration - does the medium bind the product and not the contaminants? Small pre-packed columns (containing media suitable for use in bioprocesses) are also useful to confirm that the binding is strong enough to allow application of large sample volumes, or long washes (i.e. product retention rather than retardation); they also allow elution and recovery to be studied (Fig. 3-19). For Polishing steps, fine beads (average diameter less than 35 pm) may be required to achieve resolution. Pre-packed short columns are useful for initial screening, but
80
3 Hydrophobic Interaction Chromatography of Proteins
mAl
Conductivity (r Ycm)
lOnm
400
150
300 100
200
50
100
0 I
0.0
0
I
10.0
Time (min)
Fig. 3-19. Small pre-packed columns, HiTrap@1 mL, were used for screening a range of media suitable for large-scale HIC. Five media were tested under identical conditions for their ability to purify a Fab fragment from a sample which had already undergone a capture step by expanded bed adsorption on STREAMLINE" SP. Phenyl Sepharose' 6 Fast Flow (high sub) gave the best result. Columns: HiTrap HIC test kit containing (UV traces from bottom upwards) Octyl Sepharose 4 Fast Flow, Butyl Sepharose 4 Fast Flow, Phenyl Sepharose 6 Fast Flow (high sub), Phenyl Sepharose 6 Fast Flow (low sub), Phenyl Sepharose High Performance. System: AKTATM explorer. Flow rate 2 mL min-' (300 cm h-'). Eluent A: 1 M ammonium sulfate, 50 mM sodium acetate, pH 5. Eluent B: 50 mM sodium acetate, pH 5 . Gradient: A + 0-100 % B over 20 column volumes. (Copyright Pharmacia Biotech; reproduced with permission.)
the need for resolution will steer towards greater bed heights and the use of shallow elution gradients. Scalability is often overlooked. Fine resolution achieved in the lab can easily be lost by lack of a compatible large scale column and poorly designed equipment. If the method is to be scaled-up from a laboratory to a production scale, the following considerations are important when choosing a medium:
- the existence of a column with performance which can match the media perfor-
mance; and the existence of packing methods for the medium which can give high efficiency during large-scale operation.
3.4 Methodology
81
3.4.3 Binding and Wash Conditions Optimizing the binding conditions is critical to achieve the separation. In Capture, the normal strategy is to separate product from impurities during sample application, wash away as many further impurities as possible under conditions where the product is still strongly retained, and use step-wise elution simply to recover the product as concentrated as possible. Purification is achieved by stripping off contaminants rather than resolving components during elution. Thus, different binding and washing conditions have to be screened for optimal selectivity, capacity, recovery of product, and reproducibility. As described above, the stability window should be defined and adhered to. Sample application conditions are usually also set such that contaminants less hydrophobic than the product are not bound, in order to increase the capacity of the adsorbent for the product of interest (Fig. 3-20). If the majority of contaminants are more hydrophobic than the product, it may be advisable to use HIC at a later stage when some of these impurities will have been removed by previous steps. It is important to ensure that the product is applied to the column under conditions which ensure its retention rather than retardation. If this is not achieved, the result can be low capacity and significant variations in product recovery and elution profile with different batches of starting material. The standard approach is to use up to 2.0 M ammonium sulfate buffered with 0.02-0.1 M phosphate, pH 7.0, for binding. The major limitation with ammonium sulfate is that its use is restricted to neutral and acidic pH, due to the titration of the ammonium ion and release of ammonia at alkaline pH. Other salts that can be useful are: - sodium sulfate, which is effective in promoting hydrophobic interaction and can
be used at alkaline pH but has low solubility, potassium phosphate, which can also be used at alkaline pH and - sodium chloride, which has relatively weak promoting effect on hydrophobic adsorption.
-
Monosodium glutamate and aspartate have also been reported to be useful at alkaline PH P31. Fig. 3 -20. Sample application conditions are often adjusted so that less hydrophobic contaminants are not bound, leaving greater capacity for the product of interest. Care is needed to ensure strong retention rather than just retardation. Column: Alkyl Superose@HR 5 mm (i.d) X 50 mm (bed height). Sample: 100 pl mouse ascites with IgGl monoclonal antibody in 0.8 M ammonium sulfate. Flow rate: 0.5 mL min-I. Buffer A: column equilibrated at different concentrations of ammonium sulfate in 0.1 M sodium phosphate, pH 7.0. Buffer B: 0.1 M sodium phosphate, pH 7.0. Gradient A + 0-100 % B over 20 column volumes. a) Sample applied at 2 M ammonium sulfate, both albumin and IgG are adsorbed. b) Sample applied at 1.5 M ammonium sulfate, albumin and IgG elute earlier. c) Sample applied at 1.0 M ammonium sulfate, albumin does not bind, IgG capacity is improved. d) Sample applied at 0.8 M ammonium sulfate, albumin does not bind, IgG is only retarded and elutes as a broad peak. (Copyright Pharmacia Biotech, reproduced with permission.)
82
3 Hydrophobic Interaction Chromatography of Proteins
$80
A280 nm
nm
2.0
2.0 f 0.5
0.5
Y
\
0"
%I
3 I
z
ri
ri
8
00
-I \
1.o
1.o
0 20
0
40 Time (min)
a)
20
b)
0 40 Time (min)
A280 nm
A280 nm
0.5
0.5
2.0
f Y
0"
% I z ri
C
8
1.o
1.o
0
c)
0
20 Time (min)
s
2
Y
0
f
20
0 Time (min)
3.4 Methodology
83
3.4.4 Elution Conditions Ideally, the product can be recovered by simply reducing the salt concentration of the eluent. As mentioned above, the choice of HIC medium is made with this as an important criterion, There are many additives which can be used to elute the product, as described in section 3.2.1, but these invariably complicate methodology and increase costs. The most frequently reported additive for elution is ethylene glycol which has the advantage that it rarely denatures proteins, but its high viscosity has limited its use at large scale. Others, such as urea, organic solvents, and detergents risk loss of biological activity. Detergents can also bind strongly to the hydrophobic matrix making, re-equilibration tricky. There are two main gradient elution strategies. When HIC is used at the Capture and Intermediate purification stages it is advisable to recover the product by stepwise gradient elution, in order to achieve product concentration and high throughput. During the Polishing stage, a linear gradient elution strategy is recommended in order to achieve fine resolution. Step-wise elution strategies are easy to handle in the production hall but require careful optimization to achieve robustness and reproducibility, and to avoid false peaks. Gradient elution requires more equipment and control, and is traditionally considered difficult at very large scale. However, it can be very robust in routine production. A well-optimized step will elute the product as a distinct, easily collected peak, in a small volume, whereas a linear gradient elution procedure will often result in significant dilution of the product and the need for multi-fraction collection to identify the start and finish of the desired product peak.
3.4.5 Sample Load Sample load is optimized for economic reasons. Orienting experiments are performed at small scale by applying sample, followed by washing, and then elution of product according to the established procedure. Applying the sample at a given flow velocity until the product just breaks through in the effluent from the column, allows determination of what is called the dynamic binding capacity or breakthrough capacity (often defined at 5 % of the frontal analysis curve). In practice it is rare to develop the chromatogram until it reaches a plateau or saturation point, due to the value of the sample. In Capture, a large proportion of the breakthrough capacity is often applied, to maximize throughput and product-concentrating effect. Such conditions differ very much from those experienced in analytical chromatographic separations, where low loads are usual. Displacement effects can occur. The product binding capacity can, therefore, be strongly influenced by the concentration of displacing impurities which may vary in different batches of feedstock [34]. In Polishing, the maximum sample load is usually a small fraction of the breakthrough capacity, limited by the resolution during gradient elution. Cutting the eluted
84
3 Hydrophobic Interaction Chromatography of Proteins
peak may be required and the maximum sample load is a careful balance between yield and acceptable purity of the product.
3.4.6 Cleaning Many of the additives mentioned above, although undesirable for eluting and recovering product, can be used to clean and regenerate a HIC adsorbent. In general, solutions of sodium hydroxide (0.5-1 .O M) are effective and popular, although care must be taken to ensure that the medium is stable at high pH. Manufacturers of media will normally recommend cleaning procedures where their products are stable, although some of these are made with a degree of optimism. The results in Fig. 3-21 indicate that the two media tested vary considerably in their stabilities upon exposure to sodium hydroxide. Cleaning-in-place is normally preferred in industrial applications to avoid unpacking and repacking a production-scale column. The cleaning procedure for HIC may need considerable method development, especially in steps used for Capture where loads of hydrophobic impurities, including lipids which are tightly bound to the medium, are greatest.
3.5 Scaling-up Just as in other forms of chromatography in packed beds, an optimized method is normally scaled-up by using a column with a greater diameter, while maintaining constant the column length, the sample load per volume of medium, the eluent flow velocity, and the gradient expressed in column volumes. Going from the laboratory scale to the production hall is not as straightforward as for other techniques such as gel filtration and ion exchange, however. The design of equipment with efficient gradient mixing and small dead volumes is also important, as well as the use of construction materials which are compatible with running and cleaning conditions. Potential pitfalls include: risk of slow development of aggregates and precipitates in solutions of salting-out salts; risk of locally high ammonium sulfate concentrations if solutions are made by adding solid ammonium sulfate to the sample material; problems due to high viscosities such as poor mixing and high backpressure; and effects of temperature differences. One consequence, for example, is to consider carefully the effect of hold-up times in the production hall (which are likely to be greatly different than at laboratory scale). Correct process development assumes an intimate knowledge of how things are done in the production hall, in order to do model experiments in the laboratory. Preparation of solutions and buffers in large volumes often cannot be done in the same way
3.5 Scaling-up
85
1.5 11.4
n
',
ll
1.2
I
1.o
(A) XXXXXXPhenyl
0.8
0.6
0.4
0.2
0.00
200.0 220
240
260
280
300
320
340
380 400 Wavelength nm
380
(B) Phenyl Sepharose 6 Fast Flow (high sub) 0.2
0.00 m.0 220
240
260
280
300
320
340
360 380 400 wavelenem M
Fig. 3-21. UV/visible spectra for samples taken after storage in 1 M sodium hydroxide at 40 "C showing ligand leakage at high pH from two widely used adsorbents. A is a commercially available polymeric HIC adsorbent; B is Phenyl Sepharose@6 Fast Flow (high sub). The spectra (from bottom to top) are from samples taken after 25, 49, 72, 144, 192 and 360 h. (Copyright Pharmacia Biotech, reproduced with permission.)
as in small-scale experiments. Working temperatures and liquid handling are often not the same in both environments. Figure 3.22 shows a rather dramatic example of what can happen if temperatures are not controlled carefully. Since solutions with widely differing viscosities and densities are frequently encountered during HIC, gradient development and mixing require special consideration.
86
3 Hydrophobic Interaction Chromatography of Proteins
4°C Sample, 23°C System
23°C Sample & System
2.0
3.0
4.0 Time(min)
5.0
2.0
3.0 4.0 Time(min)
5.0
Fig. 3-22.Effect of temperature on a HIC separation. In the left hand chromatogram, the sample and column were equilibrated at 23 "C. In the right hand chromatogram, the system was at 23 "C but the sample was straight from the cold-room at 4 "C. Hatched areas indicate elution position of the antibody. HIC medium: SOURCE" 15ISO. Column: RESOURCE" IS0 (1 mL) 6.4 mm (i.d.) X 30 mm (bed height). Sample: 1 mL mouse ascites containing monoclonal IgGl diluted 1 part ascites plus 4 parts 1.55 M ammonium sulfate (final concentration 1.24 M ammonium sulfate). Flow rate: 5 mL min-' (940 cm h-l). Buffer A: 1.55 M ammonium sulfate, 0.1 M sodium phosphate, pH 7.0. Buffer B: 0.1 M sodium phosphate, pH 7.0. Gradient: 2 column volumes wash with 100 % A then A + 0-100 % B over 10 column volumes. (From [24]; reproduced with permission.)
3.6 Conclusion HIC is a good complement to other separation techniques based on charge, molecular size or affinity. HIC is a well-established and versatile technique, and a wide range of media exists to cover a variety of different applications. HIC can be used at various stages in a downstream process, from initial Capture, through Intermediate purification to final Polishing, and its best position in a process depends upon the sample, and properties of the product versus the impurities which have to be removed. Such properties are best assessed chromatographically by orienting experiments at a small scale. Succesful scale-up assumes good insight into likely pitfalls with the technique, Pre-requisites are extensive study of the behavior of the sample components under the conditions used in HIC, a well-balanced, robust choice of chromatographic medium and running conditions, and the ability to model and down-scale conditions encountered in the production hall during optimization. Taking care over these points will reveal HIC as a useful and often powerful tool, worthy of a place in most purification schemes for proteins (Fig. 3 -23).
References
87
+ Mildktabilizing Wide range of media Potential unique selectivity Complements IEX, affinity and GF
. Unpredictable High salthiscosity Requires fine tuning Precipitation/denaturation
Fig. 3-23. HIC has a rightful place in many downstream purification schemes. The positive (+) and negative (-) features of the technique are listed. (Copyright Pharmacia Biotech, reproduced with permission.)
References [ l ] Janson, J.-C., Liis, T., in: Chromatography of Synthetic and Biological Macromolecules, Roger. E. (Ed.). Chichester, England; Ellis Horwood Ltd., 1978. [2] Hydrophobic Interaction Chromatography. Principles and Methods. Uppsala, Sweden: Pharmacia Biotech, 1993. [3] El Rassi, Z., J Chromatogr 1996, 720, 93-118. [4] Tanford, C., The Hydrophobic Effect. New York; John Wiley and Sons, 1973. [5] HjertCn, S . , in: Proceedings of the International Workshop on Technology for Protein Separation and Improvement of Blood Plasma Fractionation. Reston, Virginia, 1977, pp. 4 10-42 1. [6] Arakawa, T., Arch Biochem 1986, 248, 101-105. [7] Horvath, C., et al. in: Separation Processes in Biotechnology: Asenjo, J. (Ed.). New York: Marcel Dekker, 1990; Vol. 1; p 447. [8] Melander, W., Horvath, C., in: High Performance Liquid Chromatography: Advances and Perspectives. Melander, W., Horvath, C. (Eds.). New York: Academic Press, 1980; p. 114. [9] Melander, W., Horvath, C., Arch. Biochem Biophys 1977, 183, 200-215. [lo] Arakawa, T., Timasheff, S., Biochemistry 1982, 21, 6545-6552. [ I l l Arakawa, T., Timasheff, S., Biochemistry 1984, 23, 5912-5923. [12] Nozaki, Y., Tanford, C., J Biol Chem 1963, 238, 4074-4081. [13] Nozaki, Y., Tanford, C., J Biol Chem 1970, 245, 1648-1652. [14] Tanford, C., Nozaki, Y., Reynolds, J. A,, et al. Biochemistry 1974, 13, 2369-2376. [15] Tanford, C., Adv Prot Chem 1968, 23, 121-282. [16] Timasheff, S., Inoue, H., Biochemistry 1968, 7, 2501-2513. [17] Gagnon, P., Grund, E., Biopharm 1996, 9, 54-64. [18] Arakawa, T., Narhi, L. O., Biotechnol Appl Biochem 1991, 13, 151-172. [19] Fausnaugh, J.L., Kennedy, L. A,, Regnier, F.E., J Chromatogr 1984, 317, 141-155. [20] HjertCn, S., Yao, K., Eriksson, K.-O., Johansson, B., J Chromatogr 1986, 359, 99-109. [21] Berggrund, A. Drevin, I. Knuutila, K.-G., Wardhammar, J., Johansson, B.-L., Process Biochem 1994, 29, 455-463. [22] Kirsnas, P., Lindblom, T., J Chromatogr 1992, 599, 131-136. [23] Downstream 1994, 16, 14-16. [24] Gagnon, P., Grund, E., Lindback, T., Biopharm 1995, 8, 36-41. [25] Gagnon, P., Grund, E., Lindback, T., Biopharm 1995, 8, 21-27. [26] Rosengreen, J., Pihlman, S . , Glad, M., HjertCn, S., Biochim Biophys Acta 1975, 412, 51-61. [27] Kato, Y., Kitamura, T., Hashimoto, T., J Chromatogr 1984, 292, 418-426. [28] Kato, Y., Kitamura, T., Hashimoto, T., J Chromatogr 1984, 298, 407-418.
88 [29] [30] [31] [32] [33] [34]
3 Hydrophobic Interaction Chromatography of Proteins Ingraham, R., Lau, S., Taneja, A., Hodges, R., J Chromatogr 1985, 327, 77-92. Wu, S.-L., Figueroa, A., Karger, B.L., J Chromatogr 1986, 371, 3-27. Kato, Y., Kitamura, T., Hashimoto, T., J Liquid Chromatogr 1986, 9, 3209-3224. Regnier, F., Science 1987, 238, 319-323. Narhi, L. O., Kita, Y. A,, Arakawa, T., Anal Biochem 1989, 182, 266-270. Gagnon, P., Grund, E., Biopharm 1996, 9, 34-39.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
4 Displacement Chromatography : Application to Downstream Processing in Biotechnology Ruth Freitag
4.1 Introduction A large number of biologicals are of putative interest to the life and natural sciences. How to provide these substances for further research and finally application is a major concern of modern biotechnology. The answer to that question depends largely on the feasibility, the ease, and cost of their production. Sales of biotechnology products are expected to show double-digit annual growth rates over the next decades in several key sectors. The market for biopharmaceutical products, dominated by proteins (and peptides) for human therapeutics and diagnostics, looks especially promising. The product isolation, the ‘downstream process’, is most crucial in that area of the so-called ‘high value’ products, typically sensitive substances produced in low quantities by recombinant bacteria and increasingly also by the even more demanding but concomitantly more powerful recombinant mammalian cells.
4.2 Product Isolation in Biotechnology The Downstream Process In recent years bioprocess engineering has gradually shifted from concentrating on the well-characterized bioprocess towards the well-characterized product. In a well-characterized process, even minor details are standardized early on and then strictly maintained during process development and final production. Such a procedure becomes necessary when the product itself is beyond proper characterization, e.g. due to its complexity, heterogeneity, or lack of stability. When the integrity of the product can be established, on the other hand, the details of the production process become less deciding, This development has rendered the biotechnical industry much more flexible. An area that will perhaps benefit the most from this increased flexibility is the downstream process, currently a major bottle neck, where an innovative approach is considerably hindered by any early parameter fixation.
90
4 Displacement Chromatography: Downstream Processing in Biotechnology
During the downstream process the product is recovered from the complex and often dilute raw solution, the feed, and transformed into a stable marketable form. The degree of difficulty depends on two conditions :
1. The product concentration in the feed. 2. The desired final purity and composition (concentration, activity, heterogeneity, limits for certain impurities such as endotoxins, DNA, residual foreign proteins). At first, it may appear strange that the downstream process turns out to be a difficult part of bioproduction. But then, the challenge facing biotechnology in that area is not so much one of general feasibility - almost any substance may be procured in pure form some way or the other - but that of operating within a certain financial and dimensional framework. Feed product concentrations may vary between several g/L (antibiotics) and some pg mL-’ or even pg mL-’ (recombinant blood factors). In order to gain approval for, e.g. a pharmaceutical, it must be guaranteed that a sufficient amount can be provided to satisfy the foreseeable medical need. Even if this may mean only a few kilograms per year, the sheer size of the product stream may pose a problem in itself, given the typical product concentrations. Moreover, substances that are found in rather low feed concentration tend to require the very highest levels of final purity, e.g. parenteralia intended for use in humans. Concentration cum isolation are therefore the somewhat contradictory goals of many downstream processes in biotechnology. Conventionally, the downstream process is subdivided into four sections (Table 4-1). The high-resolution ‘separation’ stage TI1 of the downstream process is clearly Table 4 -1. Stages of the biotechnological downstream process. Downstream Processing in Biotechnology I Conditioning: Goal: Rendering the product stream suited for further (standard) processing Typical operations: Centrifugation, filtration (removal of producing organismsholids from extracellular products), cell lysis (intracellular products), renaturation (product as inclusion body) I1 Isolation: Goal: Removal of bulk impurities of considerable difference in physicochemical character Typical operations: Extraction (organic solvents, aqueous two-phase systems), precipitation (salt, alcohol), ultra-, diafiltration, adsorption (expanded, fluidized bed) 111 Separation: Goal: Removal of closely related impurities Typical operation: Chromatography
IV Polishing: Goal: Final Formulation Typical operations: Gel filtration, crystallization, lyophilization, addition of stabilizers, inert fillers, etc., operations for giving the final product shape
4.3 Modes of Chromatography
91
dominated by liquid chromatography. Most pertinent isolation schemes incorporate not one but several chromatographic steps. This has largely to do with the hitherto unsurpassed selectivity of chromatography. Concomitantly, however, chromatographic steps contribute drastically to the difficulties and costs of a given isolation process. It is perhaps significant, that truly large scale (bio-)production facilities dedicated, e.g. to the production of antibiotics or to the fractionation of complex materials such as blood and milk, rely on unit operations such as extraction, filtration and precipitation rather than chromatography to achieve their goals, even at the price of forsaking certain valuable substances, which are not recoverable by the latter procedures. Much of the reserve towards preparative chromatography originates in the predominant use of discontinuous elution chromatography in that context. The use of other (operation) modes of chromatography may help to resolve the dilemma between selectivity and productivity.
4.3 Modes of Chromatography Chromatographic separations are traditionally batch procedures. The sample mixture is introduced at one end of a column and resolved along the axis of that column due to a differential distribution between a stationary and a mobile phase. With the exception of size exclusion chromatography, differences in the free energy of the substances reaction (adsorption) with the stationary phase surface causes separation. Kinetic effects may overlay the thermodynamic ones. Hydrophobic, ionic or complex ‘biospecific’ interactions are mostly used in preparative biochromatography. Porous particles are used as stationary phase, in order to provide the highest possible surface area. The process can be mathematically described by the respective mass balance together with the appropriate initial and boundary conditions. The mass balance, the initial condition and the exit boundary condition apply to the chromatographic situation in general.
Mass balance: acilat
+ $dqi/at + u,aci/az
= Deff a2cilaz2
i = 1, 2 , . . , n
where ci is mobile phase concentration; qi is stationary phase concentration; 4 is phase ratio; uo is linear flow velocity; Deff is dispersion coefficient; t is time; and z is dimensionless column length.
Initial condition: Ci (0, Z) = 0 ; Exit boundary condition: (acilaz), = = 0 where L is column length.
O
i = 1, 2,.., n
Depending on the sample introduction, different inlet boundary conditions must be used. Following the treatment of Tiselius, three extremes are distinguishable [ 11. The sample mixture may enter the column as a Dirac or rectangular pulse (elution
92
4 Displacement Chromatography: Downstream Processing in Biotechnology
chromatography) or as a step function (frontal chromatography). The transition between these two modes is gradual. Highly overloaded elution chromatography, e.g. shows certain aspects of frontal chromatography. The third mode of chromatography is the so-called displacement mode. In this case a rectangular sample pulse is followed by a step function of another substance, dubbed ‘displacer’, which adsorbs strongly to the stationary face surface. The inlet boundary conditions of the three modes of chromatography are:
Elution: or :
ci(t, 0 ) = co, i O
i = 1, 2,.., n concentration in the feed, 8(t): Dirac function i = 1, 2 , . . , n co, i H(t) H(t): Step function O < t < T , Displacement: ci(t, 0 ) = c0, i i = 1, 2,.., n-1 (Sample) and : cn(t, 0) = co, n H(t-7) (Displacer) where T is duration of feed introduction. ci.0:
Frontal:
ci(t, 0 ) =
Elution chromatography is currently the preferred mode in both analytical and preparative biopolymer chromatography. It has been treated in several recent monographs [2-51 and will not be discussed further in this chapter. In frontal chromatography the feed is continuously infused into the column under conditions that favor the binding of all components but one. Only that least-retained component can be separated from the others. It is obtained in pure form at the column outlet until the other components break through. Frontal chromatography is the first step in many biopolymer purification schemes involving differential elution. The method is per se applicable when the product to be purified has much lower affinity for the stationary phase than the other feed components and therefore breaks through far ahead of the impurities.
4.3.1 Displacement Chromatography Displacement chromatography may be viewed as a special case of frontal chromatography or perhaps as its offspring. In fact, a method has been developed under the name of ‘sample displacement chromatography’ which exploits essentially a frontal effect for separation [ 6 ] . The advent of highly efficient HPLC instruments and columns together with an improved understanding of the theory of non-linear chromatography have recently provided new impetus for the displacement mode of chromatography. In displacement chromatography (Fig. 4 - l), the separation is based on an enforced competition of the feed molecules for the binding sites on the stationary phase surface. The substance mixture is loaded onto the column under conditions aiding strong binding. A considerable portion of the stationary phase capacity may be exploited during that phase. During loading some degree of separation is already
4.3 Modes of Chromatography
93
Fig. 4-1. Separation of a multicomponent mixture by displacement chromatography. (Adapted from [7].)
achieved due to a frontal chromatographic effect. Contrary to elution chromatography, the mobile phase serves mainly as an inert carrier and does not itself interact with the stationary phase. After the relevant feed components have been adsorbed, a solution containing the displacer is pumped through the column. A displacer should have a higher (dynamic) affinity to the stationary phase under chromatographic conditions than the sample components, so that it is able to compete successfully for the binding sites with any of them. Under these conditions, the displacer front pushes the sample components ahead of itself and thereby enforces a direct competition for the decreasing number of binding sites among the different compounds. The more strongly bound substances will themselves push the less strongly bound ones ahead. Finally, in a system governed by Langmuir-type isotherms, all sample components are focused into consecutive zones of the pure substances lined up according to stationary phase affinity. The so-called ‘displacement train’ or ‘scaled isotachic state’ has been formed. Under these - highly idealized - conditions, the borders between adjacent zones are self-sharpening. If for some reason, a molecule travels into the zone ahead of its own bulk zone, it will be among molecules with a lesser stationary phase affinity than itself, i.e. it will be strongly adsorbed and retained until overtaken by its own zone. The opposite is true for a molecule staying behind its zone for some reason. It will be immediately dislocated and moved ahead until rejoined to the proper zone. Following the breakthrough of the displacer front, the column needs to be regenerated and conditioned for further use. The velocity of the displacement train and thus the speed of the separation is dictated by that of the displacer front, U D , which in turn is in the case of ideal chromatographic conditions given by:
94
4 Displacement Chromatography: Downstream Processing in Biotechnology
As a result of the enforced isotachicity of the various substance zones, the individual ratios a q i l a c i for the various substances must equal a q D / a c D . An 'operating line' with the steepness qD/CD determines the entire system and the concentration in each sample zone depends solely on the relation between the respective component isotherms and the displacer isotherm (Fig. 4-2). The original concentration of the substance in the feed, on the other hand, is of hardly any consequence. In fact, since the column loading takes place under conditions favorable to adsorption, even highly diluted feeds can be given an excellent prognosis in terms of recovery, concentration factor, and purity, since as long as suitable conditions are chosen, the final product concentration can be considerably higher than in the feed. Just as important, however, in terms of preparative protein chromatography is the fact, that the concentration in the substance zone may easily be kept below some critical (aggregation, denaturation) level, since a Concentration plateau rather than a 'peak' is typical in displacement chromatography.
-
Concentration in the mobile phase
C
corresponding fully developed displacement train is shown below. The c D + - - - - - - - - _ _ _ _ - - _ _ _ _D_ _ _ _ _ _ _ _concentration ___ in the substance zones of
4.3 Modes of Chromatography
95
4.3.2 Use and Advantages of the Displacement Mode in Preparative Chromatography of Biopolymers The controllable final product concentration as well as the low waste production, i.e. the efficient use of the stationary and mobile phase capacity, are among the most obvious advantages of displacement chromatography over elution chromatography in preparative chromatography. Other than in affinity chromatography, on the other hand, several substances can be purified simultaneously, which is clearly an advantage whenever the feed contains more than one substance of value. This may be the case for certain natural sources such as blood/plasma, milk or plant extracts but also for recombinant proteins produced by mammalian cells. In this case the culture media contain substances such as bovine serum albumin (BSA), transferrin, insulin, etc. which are not consumed during the bioproduction and which, if recovered at sufficient concentration and purity, may be recycled. Such ‘media recycling schemes’ are popular, if rather theoretical, scenarios in the discussion of economic aspects of the costly mammalian cell cultures. Another major advantage of displacement chromatography, especially in the context of large-scale biopolymer chromatography, stems from the fact that the feed, the displacer, and the regenerant are introduced as simple step functions. Displacement chromatography is therefore much easier adapted to continuous separation of multicomponent mixtures than (gradient) elution chromatography. De Carli et al. successfully adapted a continuous annular chromatograph for that purpose [8]. In spite of these advantages only a few applications of displacement chromatography for protein isolation are known. The reason for this comparative neglect has to do with the ‘problem areas’ of displacement chromatography (DC), namely the monitoring of the displacement train, the determination of a suitable displacer, and the present difficulties in modeling and optimizing this exclusively non-linear chromatographic mode. Monitoring the displacement train, i.e. differentiating between the consecutive, highly concentrated substance zones, has in the past been difficult and time consuming. Typically fractions were collected and analyzed afterwards. While this has rather negative consequences for the process time, the potential for automation is also quite low in such a system. The use of a two-dimensional chromatographic system, where an analytical HPLC automatically provides an on line analysis of the effluent of the preparative column in short intervals, is a possible alternative to the off line analysis [9]. Modern analytical HPLC systems allow the required analysis to be carried out within seconds, while state-of-the-art process control software can use these data to control the automatic collection of the desired substances at a given quality, e.g. according to a pre-set purity and/or concentration threshold. The modeling of biopolymer displacement chromatography and suitable biopolymer displacers pose somewhat graver problems, however, which will be discussed below.
96
4 Displacement Chromatography: Downstream Processing in Biotechnology
4.4 Modeling and Theory of Displacement Chromatography It is perhaps symptomatic, that the first ‘practical’ application of displacement chromatography, predates the first theoretical treatment of such a system by more than 20 years. When preparations of pure rare earth oxides were needed in the Manhattan Project during World War 11, displacement chromatography was the way to prepare them, given the state-of-the-art separation techniques at that time. Even though Gluckauf provided an analysis of the displacement phenomenon and discussed effects of solute and displacer concentrations as early as 1935, it was only in the late 1960s that Helfferich et al. presented the first algorithm for a mathematical description of displacement chromatography based on the coherence model of ideal chromatography [ 101. Ideal chromatographic conditions prevail, when it is legitimate to assume plug flow of the mobile phase, i.e. the total absence of dispersion and infinite column efficiency, no mass transfer or kinetic effects and a system that is thermodynamically, consistently described by multicomponent Langmuir isotherms. Then the system of partial differential mass balance equations for the various compounds may be coupled via the multicomponent Langmuir isotherms and transformed using the so-called h-transformation, into a set of simple - and at that time already solvable - algebraic functions. In ideal displacement chromatography the isotachic state is always reached at some point of time and distance. In spite of the simplifications the model of coherence proved to be surprisingly suited for the description of displacement separations. In 1985 it was extended by Frenz and Horvath for the separation of proteins by high-performance displacement chromatography (HPDC) [ l l ] . The model is, however, only of limited worth in a practical situation and the development and optimization of a displacement separation is today still based almost exclusively on laboriously collected experimental data rather than on mathematical simulations. This is not a problem of displacement chromatography per se, however, but a general calamity of non-linear chromatography especially in the case of (bio-)polymers and their sometimes erratic adsorption behavior. In a more realistic model of non-linear chromatography, two aspects have to be considered. One is the physical transfer of the compounds through a fixed bed of porous particles; the other is the surface reaction between the compounds and the stationary phase. A realistic approach to the former aspect needs to consider phenomena like axial dispersion and mass transfer and kinetic resistance. These effects can be incorporated into the mass balance equation, e.g. by the introduction of dispersion coefficients (axial dispersion) and/or pore and overall column efficiency parameters (pore and film diffusion), and even radial velocity profiles [12,13]. The resulting numerical algorithms have proven to be powerful tools for the understanding and the design of non-linear chromatographic separations [14,15]. They form the basis of our current understanding of the effects of chromatographic parameters such as the column dimensions, the particle diameter, the mobile phase flow rate, the composition and concentration of the feed, and the concentration and heterogeneity of the displacer on the separation. A finite column efficiency may be automatically
4.4 Modeling and Theory of Displacement Chromatography
97
introduced into these presentations by the necessity of choosing finite number of steps in the calculations. In a separation that is primarily based on adsorption phenomena, the interaction of the solutes with the stationary phase deserves just as much attention - if not more so - as the transport phenomena, especially in the case of large molecules and their low diffusivity. The equilibrium relationship for the distribution of the components between the mobile and the stationary phase, c; and qi, is given by the respective adsorption isotherms. At low concentrations, Henry’s law usually holds and the relationship is linear. At higher surface concentrations, when the competition of the various substances for the adsorption sites can no longer be ignored, these interactions become exceedingly difficult to model over a sufficiently wide concentration range. This is especially the case for biopolymers, for whom adsorption is a complex process which may involve multipoint interaction, conformational changes, interaction of the components with each other (adsorbed and in solution), multiple retention mechanisms, etc. The slow diffusion rate of such large molecules makes the attainment of a true adsorption equilibrium within the practical time scale of chromatographic separations highly unlikely. Many of the recorded biopolymer single component isotherms can, however, be fitted to the Langmuir formula :
where ai and bi are substance-specific constants. Most theoretical treatments to date rely therefore on the multicomponent Langmuir formalism for the description of the non-linear multicomponent system, 4i =
aci 1
+
CbjCj’
which can easily be constructed for each component in question, once the individual Langmuir constants, bi and ai, have been derived from the respective single-component isotherms. However, the Langmuir multicomponent approach has been shown to be thermodynamically inconsistent whenever the saturation capacities of the various substances differ. The correlation with the experimental results is also often not satisfactory [12,16]. The separation factors will hardly ever be constant over the entire concentration range. The result may be a crossing of the multicomponent isotherms leading at worst to ‘elution azeotropes’, which no column despite its length will be able to resolve [ 16,171. The conformational changes and the interaction with already adsorbed molecules will lead to a change in interaction energy and thus to an Sshaped adsorption isotherm. The direct determination of multicomponent isotherms, on the other hand, is experimentally rather involved, if possible at all. Other models such as the LeVan and Vermeulen isotherm derived from the theory of the Ideal Adsorbed Solution (IAS) have been suggested to describe the multicom-
98
4 Displacement Chromatography: Downstream Processing in Biotechnology
ponent adsorption behavior of complex molecules under linear and non-linear chromatographic conditions [ 16,181. For ion-exchange displacement chromatography the steric mass action model (SMA) has been suggested by Brooks and Cramer [19]. The model describes overloaded ion-exchange chromatography of large molecules (proteins) simply by the law of mass action. In such it is kin to the multivalent ion exchange formalism proposed by Velayudhan and Horvath [20] and the stochiometric displacement model used by Kopaciewicz et al. [21]. Innovatively, however, it takes into account, that a large molecules will not only (actively) interact with certain adsorptive sites on the stationary phase surface, but will also (passively) cover other of these interaction sites simply due to its bulk. The authors claim good agreement between experimental results and the model’s prediction. The characteristic charge, oi, the steric factor, vi, and the adsorption equilibrium constant, Ki, the protein, all of which can be determined by a couple of gradient elution or frontal chromatographic experiments, are the only parameters required for the description of the non-linear adsorption and displacement of biopolymers. The model has since been extended to the description of immobilized metal affinity chromatography (IMAC) [22] and should in theory be applicable to any type of adsorption chromatography that relays on a single interaction mechanism. In spite of the fact that the numerical solutions always depend on a given set of assumption, a few general rules concerning the design of a displacement separation can be derived from the theoretical work. When the occurrence of an adsorption azeotrop can be ruled out, there should be an optimum column length for a given feed load. When the column length is increased beyond that required for the development of the displacement train, the ‘quality’ of the separation in terms of the sharpness of the displacer front or the border between two neighboring zones is not improved any further. On the other hand, the absolute amount of protein to be separated, but also the separation time and the column’s backpressure for a given flow rate increase with column length. Similarly, the displacer concentration needs to be optimized. While a higher concentration means a faster separation, it also diminishes the length of the zones and thus increases the amount of substance found in the shock layers. In praxis, however, the solubility of the displacer will usually be the limiting factor in that regard. Particle diameters and interstitial flow rates become important whenever intraparticular diffusion is a relevant phenomenon. So far, discussions of the effect of particle diameters in displacement chromatography are inconclusive. Smaller particles stand for higher efficiencies, i.e. lower theoretical plate heights. Theoretically derived arguments predict a minimum size however, below which the displacement kinetics become rate limiting. While Felinger and Guiochon [23] argue for an optimum ratio of the squared particle diameter over column length for each sample composition, Subramanian et al. [24] find their separation efficiencies unchanged for particle diameters of up to 90 pm. A similar discrepancy can be found in case of the carrier flow rate. Typically, flow rates in displacement chromatography are two- to ten-fold lower than in elution chromatography. In some cases, however, flow rates of up to 1 mL min-’ were successfully used in protein displacement chromatography [17]. Zhu and Guiochon convincingly argued that the optimum mobile phase linear
4.5 Displacers f o r Displacement Biochromatography
99
velocity for minimum shock layer thickness and maximum recovery yield does not only depend on the axial dispersion and mass transfer resistance as in elution chromatography, but also very much so on the retention factor and the concentration of the displacer [25]. Generally low retention factors (normalized retention times) for all involved components including the displacer support good displacement separations. The optimum values may well rest between 1.2 and 2.0 [23]. The direct comparison of overloaded elution and displacement chromatography based on the theoretical predictions shows that the optimum theoretical parameters are quite different in the two modes. It is therefore not advisable to transfer experience gained in elution biopolymer chromatography automatically to the displacement mode. The results of a direct comparison of the theoretical production rates and recovery yields of the two modes depend on the parameters and restrains used in the simulations [26,27]. The substance concentration ‘recovered’ in the fractions is much higher in displacement chromatography, however, which may well become decisive in practical applications of chromatographic separations, e.g. in biotechnology.
4.5 Displacers for Displacement Biochromatography The importance of the displacer for a successful displacement separation can hardly be overestimated. Choosing the right displacer is just as important in this case as choosing the right stationary phase. The ideal displacer is a substance which must met quite a number specifications. After all, it should:
non-toxic and biologically, chemically, mechanically, and thermally stable and show no interaction with the other solutes. - be highly uniform, since displacer impurities/heterogeneity may make column regeneration difficult or pollute the substance zones depending on their relative affinity. - have good carrier solubility, since high concentration is required for fast and productive separations, but also show a high (controllable) affinity for the stationary phase and allow a fast and easy column regeneration. - be
Detectability and costs, respectively the possibility to recycle the displacer, are other considerations that quickly enter the debate on the applicability of displacement chromatography in general. Ideally the displacer should also be easy to remove from the recovered substance fractions. For pharmaceutical applications it might even be necessary to sterilize the material. Displacement chromatography of comparatively small biologicals such as amino acids, peptides, and small proteins (antibiotics, insulin) is usually carried out in the reversed phase mode. Hydrophobic substances such as 2-(2-butoxyethoxy)ethano1 (BEE), decyltrimethylammonium bromide, cetyltrimethylammonium bromide (cetramide), benzyldimethyldodecylammonium bromide, dodecyloctyldimethylammonium chloride and palmitic acid are possible displacers in these cases [28].
100
4 Displacement Chromatography: Downstream Processing in Biotechnology
The situation is more difficult in case of biopolymer and protein displacement chromatography. It is, for example, often assumed, that a protein displacer needs to be a large molecule itself, all the more so, since little to nothing was known until recently on the physico-chemical basis of a ‘good’ biopolymer displacer [29]. The reasoning behind this concept was the idea that multipoint attachment is necessary to counteract multipoint attachment. Cases where the authors claimed to have used a comparatively low molecular weight substance as displacer, e.g. [30], were countered by the argument that extremely heterogeneous polymer preparations were used, which contained high-molecular weight substances as the ‘real’ displacers. Thus, the combination between a polyionic displacer and an ion exchange material has become a classic in protein displacement chromatography. Hydrophobic interaction chromatography, often a very useful alternative in preparative elution chromatography, is almost unknown in the displacement mode for the lack of suitable displacers, i.e. mildly hydrophobic, water-soluble polymers. Reversed-phase separations are less suited to preparative protein purification in general, since many proteins denature under these conditions. Other stationary phases have been used only occasionally. (Semi-)synthetic polyions such as chondroitin sulfate, dextrane sulfate, carboxymethyl starch, alginate, Eudragit, Nacolyte 7 105 and polyethyleneimine (PEI) have been used for protein displacement on ion exchange materials [31]. Since 1978, Torres and Peterson have promoted the use of (modified) carboxymethyldextranes (CM-D) for that purpose [32]. Their displacers are now commercially available from Bio-Fractionations (Logan, Utah, USA). Biopolymer displacements are also possible with proteineous displacers. With the possible exception of the latter, all of these substances show a broad molecular mass distribution and also tend to vary in structure. As a consequence, the affinity for the stationary phase varies and often less well retained displacer molecule fractions contaminate the displacement train. Proteineous displacers entail problems in terms of cost and stability. Lately it has been shown, however, that stationary phase affinity is much more important in a protein displacer than size, e.g. [9,33-351. While this will most often be the case for polymeric substances, there are numerous cheap, homogeneous oligomeric substances available, which constitute powerful protein displacers. Gadam and Cramer compared pentosan polysulphate (rel. mol. mass 3000) and dextrane sulfate (re1 mol. mass 50000) for the separation of whey proteins in anion exchange displacement chromatography [33], While the larger molecule was less sensitive to the experimental conditions, namely the salt concentration, the separation efficiency under optimized conditions was quite similar. The investigation of pentaerythritol-based dendritic polymers with relative molecular masses ranging from 480 to 5100 showed that the results could be interpreted by the SMA model and that charge densities and steric factors are more decisive in the separations, than the absolute size and number of interaction points of the displacer [34]. Jen and Pinto came to a similar conclusion in their investigation of various dextrane sulfates [35]. Breier and Freitag reported on the investigation of a series of homogeneous polyethylene glycol-based linear and dendritic displacers (rel. mol. mass : 1000-50000) that were modified to carry chelating groups at the ends [9]. In this case hydroxyapatite was used as stationary phase. The best results were observed
4.5 Displacers f o r Displacement Biochromatography
10 1
for the smallest of the investigated molecules. A further investigation on the applicability of small calcium chelaters as protein displacers showed, that both EGTA (ethylenglycol-bis(P-amhoethylether)-N,N,N', "-tetraacetic acid, rel. mol. mass: 380.4) and IDA (imminodiacetic acid, rel. mol. mass: 133.4) are possible protein displacer for hydroxyapatite [36].
4.5.1 The Rational Design of Protein Displacers A host of synthetic (high molecular mass) substances have been used for proteins displacement. Few if any of them have been synthesized with that explicit goal in mind, however. Torres and Peterson were among the first to chemically modify their (high molecular mass C-MD) displacers in order to gain control over the stationary phase affinity [37]. More recently, Vogt and Freitag have presented their building block system for target-orientated displacer synthesis by copolymerization [38]. The approach allows the production of a displacer for a given stationary phase and/or separation problem. The stationary phase determines the nature of a so-called specific monomer, such as ionic groups for ion exchange materials. If useful, a second monomer is used to add features to the displacer, which for example, facilitate recovery or detection. Especially the problem of displacer recovery and/or removal from the product zones deserves some attention. Here, the building block approach can become most useful. Certain water-soluble polymers, show an LCST (lower critical solution temperature). Upon reaching that temperature, the polymers show a rapid, usually reversible phase transition (precipitation). Below that temperature, they are soluble. The precipitation/resolution is a very fast reaction. The polymers can run through these cycle several hundred times, the tendency for unspecific coprecipitation, e.g. of the proteins, can be kept low [39]. In Table 4-2, the features of a set of polymers that show both a composition-dependent LCST and good displacer properties for proteins on anion exchanger phases are compiled. The corresponding displacements are shown in Fig. 4-3. A somewhat similar approach has been taken by Patrickios et al. [40]. They used group transfer polymerization to produce tri-block polymethacrylates bearing a sequence of positively charged groups at one end, a sequence of negatively charged ones at the other, and a neutral, hydrophobic middle block. Due to their dual-charge Table 4-2. Series of thermoreactive displacers with different LCST but comparable displacer qualities. ~
~~~
Displacer composition
LCST
Molecular mass
TPD TPD TPD TPD
20 "C 53 "C 60 "C >100"C
70 kg/mol 80 kglmol 85 kg/mol Unknown
10% 6% 5% 0%
102
4 Displacement Chromatography: Downstream Processing in Biotechnology
Poly- [(N,KDimethylaaylamid)rtat-(CEthenyl-7hydro~~m~h~-lindinon~
24 26 28 30 32 34 36 39 42 44 46 48 50 52
Fraction
-
10
ZD
E
Y
e
.-
L B
uE
Poly N,N-DimeUylocrylarnld)stat-(4-E~enyC7-hydroxy-3-methyl-~~~an~)l
.-e 8 e a
0
30
32
34
36
38
40 42
44
46 48 50 52 54
Fraction
Fig. 4-3. Displacement of whey proteins from an BioRad 'BioScale 42' column using two thermoreactive displacers under otherwise identical conditions. Upper: Displacer TPD 1 contains 5 mol% of the interactive monomer and has an LCST of 53 "C. Lower: Displacer TPD 2 contains 6 mol% of the interactive monomer and has an LCST of 60 "C.
nature, these polymers show isoelectric points much like proteins. The authors claim, that their polymers are suitable protein displacers for anion exchange displacement chromatography, while the amphophylic nature of the molecules facilitates recovery, e.g. by precipitation at the polymer's isolelectric point.
4.6 Special Forms of Displacement Chromatography Most applications of displacement chromatography are more or less straightforward preparative separations of two- or multicomponent mixtures and use an experimental set up as outlined above. However, apart from the standard scheme and application area, a number of derivations and highly specialized forms of displacement chroma-
4.6 Special Forms of Displacement Chromatography
103
tography exist, which may also be useful to the biotechnologist and workers in related fields.
4.6.1 Analytical Aspects of Displacement Chromatography Displacement chromatography will always be a predominantly preparative technique. As such it may become a useful tool in analytical (bio-)chemistry, however. The displacement process will focus even minor impurities into narrow, highly concentrated bands of the pure substance. At the same time no previous knowledge of the exact physical nature of the ‘impurity’ is necessary. Trace components, which may be difficult to isolate by conventional chromatographic methods, can be obtained in sufficient amounts to allow chemical characterization by established techniques. Ramsey et al. [41] demonstrated this most elegantly for P-naphthylamin containing an impurity at the parts-per-million level. Diethylphthalate was used as displacer. In the case of a biosynthetic human growth hormone produced at BioWest Research, trace components polluting the product at levels of as low as 0.1 % were made accessible to a characterization by mass spectrometry [42]. Displacement chromatography also played a key role in the discovery of two previously unknown amino acids, Amarine and Feline [42]. Displacement chromatography has repeatedly been used instead of conventional elution chromatography for concentrationheparation in hyphenated techniques. For example, Mhatre et al. used a LALLS (low angle laser light scattering) photometer for the on-line determination of the molecular weight of proteins separated by displacement chromatography [43]. Displacement chromatography was also used by Frenz et al. in connection to continuous flow fast atom bombardment (FAB)- and electrospray ionization (ES1)-mass spectrometry for the analysis of tryptic digest of a recombinant human growth hormone [44,45]. Cetramide served as displacer for the peptide mix on a C I S reversed-phase column. The exploitable capacity of the column could be increased by a factor of 50 to 100 compared with when elution chromatography was used in a similar set-up. The resolution was equal to that of the previously used chromatographic method. Thin-layer and forced flow thin-layer displacement chromatography (TLCD and FF-TLDC, respectively) can be viewed as analytical or semipreparative techniques. The sample mixture is applied as a spot to the normal planar stationary phases. The use of spacers is necessary to keep the substance zones apart and available for visualizing. Sudan black dye has successfully been used to that purpose. For example, TLDC has been used to isolate the metabolites of radiolabeled deprenyl [46] and its metabolites in rat urine and for the screening of ecdysteroids (a class of important steroid hormones) from plants [47].
104
4 Displacement Chromatography: Downstream Processing in Biotechnology
4.6.2 Separation of Isomers Displacement chromatography has been recognized as a method that is especially suited to the separation of closely related substances which differ only minimally in the stationary phase affinity. One area where this resolving power may benefit the biotechnical and pharmaceutical industry most directly, is that of the separation of optical and structural isomers at high throughputs and concentrations. Vigh and co-workers used cyclodextrin-silica columns for the separation of positional and geometrical isomers [48], Alpha- and P-cyclodextrin materials have been used in the normal and reversed-phase mode to separate substances of pharmaceutical relevance such as isobufen [49], a non-steroidal anti-inflammatory agent, and DDATHF (5,lO-dideazatetrahydrofolic acid) [50]. 4 -Tert-butylcyclohexanon was used to displace the isobufens, Cetramide to displace the DDATHF. Quantitative separation was possible even for isomers with retention factor ratios (a-values) as small as 1.08. For the less-retained isomer, purities and yields comparable with those achievable with overloaded elution chromatography on the same stationary phases were found, while the performance of the displacement mode was superior in regard to the more retained isomer. A series of homologous displacers of varied affinity for Cyclobond I1 columns (a-cyclodextrin silica) has been introduced by the same group in 1995 [51].
4.6.3 Miscellaneous Intraparticular mass transfer resistances have been identified as a major obstacle to the use of high mobile phase flow rates in biopolymer displacement chromatography. In that context, the use of perfusion chromatography has been suggested [52]. The stationary phases used in this type of chromatography (Perceptive Biosystems, Inc.) contain uncommonly large pores, so-called throughpores, through which a convective fluid flow is possible. Access of the inner particle surface becomes largely independent of the slow diffusion process and usually significantly less mass transfer resistance is observed. At a flow rate of 4 mL min-', a crude mixture of the genetic variants of P-lactoglobulin could be displaced within 90 seconds to yield 18 mg of the pure substances. Heparin was used as displacer. A packed-bed enzyme reactor with immobilized carboxypeptidase Y was used by Cramer et al. [53]in tandem with displacement chromatography for the preparation of N-benzoyl-L-arginyl-L-methioninamide from N-benzoyl-L-arginine and L-methioninamide. The enzyme cartridge was operated in the recirculation mode and the reaction mixture was separated using BEE as displacer on a Cis-column. The unreacted L-methioninamide was returned to the reactor. The system was operated in parallel and 460 mg of the product (purity >99 %) were obtained within 24 hours. A similar system was used by el Rassi and Horvath [54] for the separation of the reaction mixture in the ribonuclease TI-catalyzed synthesis of GpU from cyclic GMP in the pres-
4.7 Displacement Chromatography in Biotechnology
105
ence of a large excess of uridine. Nearly 100 mg of GpU with a purity of 99.7 % could be produced within 2.5 hours. N-Butanol was used as displacer. Torres and Peterson developed two interesting variants of the displacement mode. In the so-called spacer displacement chromatography, substances of varied affinity for the stationary phase are added to the mobile phase. These substances act as spacers and keep the substance zones apart. Due to their non-UV-activity they facilitate the monitoring of the protein zones and allow in theory at least to cut out a certain component of the complex feed [37]. It can be debated however, whether this is really an advantage or whether the spacer/displacer impurities often found in the substance zones are more of a disadvantage. Spacer displacement chromatography will obviously be helpful in trace analysis, i.e. whenever small amounts of a substance have to be isolated, which otherwise will be lost in the shock layer to the arraigning zones. The name of complex displacement chromatography was given to a variety, where instead of directly competing for the binding sites, the ‘displacer’ attaches itself to the bound target substances and thereby lowers its stationary phase affinity [55]. When a sufficient amount of displacer has been bound to the protein, the complex is released. Once more a separation of the displacer and the product becomes necessary.
4.7 Applications of Displacement Chromatography in Biotechnology Although further work is needed to exploit the potential of displacement chromatography, results accumulated over the last 14 years have demonstrated that the technique can be a powerful tool for the purification of antibiotics, peptides, and even proteins.
4.7.1 Separation and Isolation of Peptides and Antibiotics Modern peptide displacement chromatography started early in the 1980s and is closely connected to the group of Horvath and co-workers. Reversed-phase chromatography dominated that particular area of displacement separations. The separation of the product of a peptide synthesis from its closely related by-products remains one of the typical applications. In 1988, Cramer and Horvath used reversed-phase displacement chromatography for a preparative separation of the peptides synthesized by immobilized carboxypeptidase Y [ 5 6 ] . Displacers of varied hydrophobicity were used, including BEE and decyltrimethylammonium. The peptides were both isolated and concentrated from the diluted aqueous mixtures. Some 5.2 g were procured from a 500 ml feed in a single chromatographic run.
106
4 Displacement Chromatography: Downstream Processing in Biotechnology
In 1991, Viscomi et al. developed a large-scale multidimensional reversed phase and ion exchange displacement method to isolate a synthetic peptide fragment con-esponding to the fragment 163-171 of human interleucin-p. The efflux of the reversed phase column was transferred directly to the ion exchange column. From 100 mg to 35 g of the Merrifield synthesis-type product could be processed. In the reversedphase mode, benzyltributylammonium chloride was used as displacer; in the ion exchange mode ammonium citrate. Peptide purities were greater than 90%. The same group used reversed-phase displacement chromatography to isolated a synthetic peptide containing two epitopes of circumporozoite protein of Plasmodium fulciparum. Of this promising malaria vaccine, 50 mg were brought to more than 95 % purity using benzyldimethyldodecylammonium bromide as displacer [57]. Synthetic peptides intended for Plasmodium vivax malaria sero-epidemiology, were produced by solid-phase synthesis and purified by reversed-phase displacement chromatography [58]. Some 107 mg of the crude mixture were applied to the column and the product procured at a purity of 85 % using BEE as displacer. Controlling rather than maximizing the product concentration was a problem in the peptide isolation described by Jacobson [59], since the conventional elution chromatographic protocol was troubled by aggregation and precipitation effects. Melitin, a 26-amino acid residue peptide from honey bee venom, and respectively its synthetic variants, were isolated by Kalghatgi et al. [60] using benzyldimethylhexydecyl ammonium chloride as displacer. A non-porous c18 material (2 pm diameter) was used as stationary phase. Since intraparticular mass transfer is not possible in such materials, a very fast separation was possible. Viscomi et al. [61] used reversed-phase displacement chromatography to separate a- and 0-melanocyte-stimulating hormone from mixtures containing also derivatives. Benzyldimethyldodecylammonium bromide was the displacer. A semi-preparative protocol for the isolation of bovine and porcine insulin by reversed-phase displacement chromatography was proposed by Vigh et al. [62]. Up to 100 mg were purified using Cetramide as displacer. The level of proinsulin could be maintained below 100 ppm in the collected fractions. The separation of antibiotics is another common application of reversed-phase displacement chromatography. Kalasz and Horvath [63] separated more than 100 mg of commercial polymyxin B sulfate into its constituents using an aqueous solution of dodecyloctylammonium chloride as displacer. Product concentrations between 10 and 20 mg/mL were reached. Oligomyxcins A, B and C have been separated using palmitic acid as displacer [64]. Cephalosporin C has been isolated from a fermentation broth using BEE as displacer [65]. A 5-mL portion of the culture supernatant could be processed on a 4.6 x 350 mm column within 20 minutes.
4.7.2 Protein Separation Displacement chromatography should be the chromatographic mode most suited to the processing of the diluted product streams typically found in the case of biotechnical high-value products. Ion exchange displacement chromatography is most com-
4.7 Displacement Chromatography in Biorechnology
107
monly used in that area. Among the first applications was the purification of crude pgalactosidase from Aspergillus oiyzae on a weak anion exchanger with chondroitin sulfate as displacer [66]. A comparison with the elution mode demonstrated the superiority of displacement chromatography in terms of the utilization of the stationary and mobile phase capacity, throughput and waste production. Torres, Peterson, and co-workers have published repeatedly on applications of their CM-D displacers to practical protein separation. The separation of human serum proteins was among the earliest [32]. The low salt content of the fractions, unusual for ion exchange protein chromatography, was noted as a major advantage, since it allowed their direct analysis by electrophoresis. In 1985, the isolation of a protein, later identified Gc-2 globulin was achieved [67]. The protein was isolated from the blood serum of psoriasis patients, where it had previously been located only as a spot in a 2D-electrophoresis. Two anion exchange displacement chromatographic steps at different pH levels followed by elution chromatography on hydroxyapatite were used to process 6 mL of serum containing a total of ca. 400 mg protein and more than 100 different proteinaceous compounds. The separation resulted in the isolation of 0.5 mg of the pure protein. The solid-free serum sample could be applied directly to the first displacement column. Previously, 4 mg of that very Gc-2 protein had been isolated from 34 liters of serum using 13 chromatographic and electrophoretic steps. The CM-D displacers were also used in the isolation of alkaline phosphatase from E. coli periplasm and that of monoclonal antibodies (mAb) from ascites [37,55]. Since the mAb carry a positive net-charge under physiological conditions, complex displacement chromatography had to be used in the latter case. The method was scaled up from 1 to 450 mL, the recovery was 79 % in the largest scale. Spacer displacement chromatography, on the other hand was used to separate guinea pig serum and mouse liver cytosol proteins on medium to low-affinity adsorbents [37]. Ghose and Mattiasson have used displacement chromatography for the recovery of lactate dehydrogenase from beef heart [ 6 8 ] .The process scale-up in terms of column dimensions and protein load was investigated. Carboxymethyl starch was used as displacer. In a second paper [69] chondroitin sulfate C, alginate, and Eudragits were compared as alternative displacers and the displacement used in connection to a subsequent affinity chromatographic step on a Cibacron Blue Sepharose CL 4B column. Eudragits L and S are advertised as readily available, cheap, nontoxin protein displacers. Anion exchange high-performance displacement chromatography has also been the method of choice for purification/polishing of an industrial recombinant human growth hormone [42]. More recently, Kim has described the use of cation exchange and antibody exchange (Abx) displacement chromatography for the isolation of a proprietary thrombolytic protein from crude fermentation broth, containing among others albumin, insulin, transferrin, aprotinin, methotrexate, and bovine serum [70,71]. DEAE-dextrane was used as displacer in both cases. Freitag and co-workers have repeatedly reported on the use of displacement chromatography for protein isolation on hydroxyapatite and anion exchange phases [9,36,72]. The recovery of recombinant human Antithrombin I11 (rh-AT 111), an anticoagulant of high pharmaceutical interest, produced by adherent CHO-cells in a
108
4 Displacement Chromatography: Downstream Processing in Biotechnology
Transferrin: AT 111:
13.2 min 14.4 min 16.5 min
I
Minutes 25
,
I
25 26 27 28 29 30 31 32 33 34 35 36 37 38
Fraction
Fig. 4-4. Comparison of the separation of rh-AT 111 from CHO cell culture supernatant. Left: separation of a rh-AT 111-containing cell culture supernatant by standard hydroxyapatite elution chromatography. Column dimensions: 4.0 X 250 mm, stationary phase: 2 pm porous hydroxyapatite, buffer A: 0.05 M phosphate, pH 6.8 with 0.003 M CaC12 (to enhance binding of acidic proteins), buffer B: 0.4 M phosphate, pH 6.8, gradient: 0 % B to 80% B, flow rate 0.1 mL min-', sample size: 1 mL. Right: isolation of rh-AT III from a cell culture supernatant concentrate by displacement chromatography. Column: identical, camer: 0.02 M Tris/HCl, pH 9.0, flow rate 0.1 mL min-I, fraction size 0.05 mL, displacer concentration: 20 mg mL-]. Left: column run in the elution mode. Right: a supernatant concentrate was applied in case of the displacement column. (From [36], with permission.)
serum-free culture medium was possible by displacement chromatography on hydroxyapatite using EGTA as displacer (Fig. 4-4) [9,36]. The problem of crossing nonlinear single component isotherms and their influence on the chromatographic result was encountered during the optimization of that separation. The recovered rh-AT I11 still contained 10 % of other proteins, mostly transferrin. BSA is present only in trace amounts. Of the AT I11 activity originally detected in the feed, 84 % was recovered by the displacement separation. For comparison, the same column was also used in the elution mode (Fig. 4-4). One mL of the cell-free CHO culture supernatant was separated in a standard phosphate gradient. A quantitative separation of the rh-AT I11 is not possible under these circumstances. Column regeneration took only 2 minutes in the elution mode, compared with over 10 minutes in the displacement mode, however. Technical dairy whey containing all milk components save the casein fraction, was separated by anion exchange displacement chromatography using polyacrylic acid
4.7 Displacement Chromatography in Biotechnology 71)
109
,
L
c 8 30 c 0
.-2e,
PAA
20
Y
g 10 a 0
Fig. 4-5. Displacement of technical dairy whey on anion exchange column ,.oScale 4 2 using PAA as displacer. (From [72], with permission). Carrier: 0.02 M Tris/HCl, pH 8.0; flow rate 0.1 mL min-’.
(PAA) as displacer. The technical dairy whey contained 3.45 g/l a-lactalbumin and 12.65 g/l P-lactoglobulin, together with some other UV-active components. The whey (1 mL) could be directly applied to the Bio-Scale anion exchanger column (7 X 52 mm) without further preparation. Compared with the feed, both whey proteins were concentrated by a factor of 3 during the displacement separation (Fig. 4-5). Yields were 78 % for the a-lactalbumin and 92% for the P-lactoglobulin. The displacer front was sharp and little contamination of the preceding protein zones by the displacer took place. Some UV-active impurities were found throughout the fractions until deep into the displacer zone. The nature of the impurities was not investigated by the authors, other than to verify that they were not linked to other whey proteins such as BSA or immunoglobulins.
4.7.3 Miscellaneous Horvath et al. separated nucleotide and nucleoside mixtures on reversed-phase columns using butanol and benzyltributy lammonium chloride respectively as displacers [73]. Huang and Jin [74] reported on the use of displacement chromatography to purify methylesters of polyunsaturated fatty acids. Oleic acid was used as displacer. The dilemma of the carrier selection, i.e. enhancement of the adsorptive forces due to a low solubility for polar compounds versus high solubility for a polar compound, namely the displacer, during displacement, was solved by the use of a two-stage displacement process. Carrier 1 was acetonitrile/water 80:20, carrier 2 acetonitrile/ water 90: 10. Most recently, displacement chromatography has been suggested as
110
4 Displacement Chromatography: Downstream Processing in Biotechnology
an economically sound way for the large-scale purification of oligonucleotides 1751. An application note to that method has been published by PerSeptive Biosystems.
4.8 Conclusions Displacement chromatography stands an excellent chance to become a major chromatographic mode in large-scale biotechnological downstream processing. Due to the more efficient usage of the mobile and stationary phase capacities, the column dimensions can be much smaller for the same space time yield than in elution chromatography. Since the concentrations within the substance zones are controllable and independent of the respective concentration in the sample/feed, the recycling of valued substances becomes possible. By the same argument, waste substance are highly concentrated; therefore, the absolute waste volume is minimal. Since high concentration factors can be adjusted, the implementation of displacement chromatography early on in a downstream process may be most attractive. Due to the resulting high concentrations, the removal of the remaining impurities by other modes of chromatography should be facilitated. The comparative worth of overloaded elution and displacement chromatography is still open for discussion. The necessity of finding a suitable displacer and the time required for column regeneration will continue to be major aspects in this discussion. However, if these problems can be overcome, displacement chromatography does have an edge in terms of throughput, product concentration, and even product purity. Validation of displacement separations, i.e. by the FDA, is more currently more elaborate than that of well-known conventional chromatographic procedures. As the method becomes more common this can be expected to change in future.
References [ l ] Antia, F.D., Horvath, C., Ber Bunsenges Phys Chem, 1989, 93, 961-968. [2] Dechow, F. J., Separation and Purification Techniques in Biotechnology. Park Ridge, New Jersey: Noyes Publications, 1989. [3] Ganetsos, G., Barker, P. E., Preparative and Prodution Scale Chomatography. Chrom. Sci. Series, Vol. 61, New York: Marcel Dekker, 1993. [4] Fallon, A., Booth R. F. G., Bell, L. D., Applications of HPLC in biochemistry, in: Laboratory techniques in biochemistry and molecular biology. Amsterdam, New York, Oxford: Elsevier Science Publishers, 1987. [5] Subramanian, G., Process Scale Liquid Chromatography. Weinheim, New York, Basel, Cambridge, Tokyo: VCH Verlagsgesellschaft, 1995. [6] Hodges, R. S., Burke, T. W. L., Mant, C. T., J Chromatogr, 1991, 548, 267-280. [7] Horvath, C., Nahum, A., Frenz, J., J Chromatogr, 1981, 218, 365-393. [8] de Carli, J. P., Carta, G., Byers, Ch. H., AIChE Journal, 1990, 36(8), 1220-1228. [9] Freitag, R., Breier, J., J Chromatogr, 1995, 691, 101. [lo] Helfferich, F., James, D. B., J Chromatogr, 1970, 46, 1-28.
References
111
[ I l l Frenz, J., Horvath, C., AZChE Journal, 1985, 31(3), 400-409. [12] Bellot, J. C., Condoret, J. S., J Chromatogr, 1993, 657, 305-326. [I31 Phillips, M. W., Subramanian, G., Cramer, S . M., J Chromatogr, 1988, 454, 1-21. [I41 Gu, T., Tsai, G.-J., Tsao, G.T., Adv Biochem Eng, 1993, 49, 45-71. 1151 Katti, A.M., Guiochon, G., J Chromatogr, 1988, 449, 25-40. [I61 Antia, E D . , Horvath, C., J Chromatogr, 1991, 556, 119-143. [17] Subramanian, G., Cramer, S. M., Biotechnol Prog, 1989, 5(3), 92-97. [I81 Frey, D. D., J Chromatogr, 1987, 409, 1-13. [I91 Brooks, C. A., Cramer, S . M., AZChE Journal, 1992, 38, 1969-1978. [20] Velayudhan, A., Horvath, C., J Chromatogr, 1988, 443, 13-29. [21] Kopaciewicz, W., Rounds, M. A,, Fausnaugh, J., Regnier, F. E., J Chromatogr, 1983, 266, 3. [22] Kim, Y. J., Bioseparation, 1995, 5, 295. [23] Felinger, A,, Guiochon G., J Chromatogr, 1992, 609, 35-47. [241 Subramanian, G., Phillips, M. W., Jayaiman, G., Cramer, S. M., J Chrornatogr, 1989, 484, 225-236. [25] Zhu, J., Guiochon, G., J Chrornatogr, 1994, 659, 15-25. [26] Katti, A.M., Dose, E.V., Guiochon, G., J Chromatogr, 1991, 540, 1-20. [27] Felinger, A., Guiochon, G., Biotechnol Bioeng, 1993, 41, 134-147. 1281 Cramer, S. M., Subramanian, G., Sep Purif Methods, 1990, 19(1), 31-91. [29] Jen, S . C.D., Pinto, N.G., React Polym, 1993, 19, 145-161. [30] Jen, S. C. D., Pinto, N. G., J Chromatogr, 1990, 519, 87-98. [31] Freitag, R., Horvath, C., Adv Biochem Eng Biotechnol, 1995, 53, 17-59. [32] Peterson, E. A., Anal Biochem, 1978, 90, 767-784. 1331 Gadam, S . D., Cramer, S . M., Chromatographia, 1994, 39(7/8), 409-418. [34] Jayaraman, G., Li, Y.-F., Moore, J. A., Cramer, S.M., J Chromatogr, 1995, 702, 143-155. [35] Jen, S . C.D., Pinto, N. G., J Chromatogr Sci, 1991, 29, 478-484. [36] Kasper C., Breier J., Vogt S., Freitag R., Bioseparation, 1996, accepted for publication, BIOS, 348. [37] Torres, A.R., Peterson, E. A., J Chromatogr, 1992, 604, 39-46. [38] Vogt, S., Freitag, R. The Custom-Mode Protein Displacer - Synthesis Following the Building Block System. Presented at the 20th International symposium on High Performance Liquid Phase Separations and Related Techniques, June 16-21, 1996, San Francisco, USA, Poster NO: P-1503-M. [39] Galaev, Y., Mattiasson, B., Enzyme Microb Technol, 1993, 15, 354-366. [40] Patrickios, C. S.; Gadam, S. D., Cramer, S. M., Hertler, W. R., Hatton, T. A., Biotechnol Prog, 1995, 11, 33-38. 1411 Rarnsey, R., Katti, A.M., Guiochon, G., Anal Chem, 1990, 62, 2557-2565. 1421 Frenz, J., K / G C Zntl., 1992, 5(12), 18-21. [43] Mhatre, R., Qian, R., Krull, I. S., Gadam, S., Cramer, S., Chromatographia, 1994, 38(5/6), 349-354. [44] Frenz, J., Quan, C. P., Hancock, W. S . , Bourell, J., J Chromatogr, 1991, 557, 289-305. [45] Frenz, J., Bourell, J., Hancock, W. S . , J Chromatogr, 1990, 512, 299-314. [46] Kalasz, H., J. High Resol. Chrom Chrom Commun, 1983, 6, 49-50. [47] Kalasz, H., Bathori, M., Kerecsen, L., Toth, L., J Planar Chrom, 1993, 6, 38-42. [48] Vigh, G., Quintero, G., Farkas, G., in: Horvath, C., Nikelly, J. G. (Eds.). Washington DC: American Chemical Society (ACS) Symposium Series, 1990, Vol. 434, pp. 181-197. 1491 Farkas, G., Irgens, L.H., Quintero, G., Beeson, M., Al-Saeed, A., Vigh, G., J Chromatogr, 1993, 645, 67-74. [50] Irgens, L.H., Farkas, G., Vigh, G., J Chromatogr, 1994, 666, 603-609. [51] Quintero, G., Vo, M., Farkas, G., Vigh, G., J Chromatogr, 1995, 693, 1-5. [52] Gerstner, J. A., Morris, J., Hunt, T., Hamilton, R., Afeyai, N. B., J Chromatogr, 1995, 695, 195-204. [53] Cramer, S.M., el Rassi, Z., Horvath C., J Chromatogr, 1987, 394, 305-314.
112
4 Displacement Chromatography: Downstream Processing in Biotechnology
El Rassi, Z., Horvath, C., J Chromatogr, 1983, 266, 319-340. Tores, A. R., Peterson, E. A., J Chromatogr, 1990, 499, 47-54. Cramer, S. M., Horvath, C., Prep Chromatogr, 1988, 1(1), 29-49. Viscomi, G. C., Cardinali, C., Longobardi, M. G., Verdini, A. S., J Chrornatogr, 1991, 549, 175-184. [58] Bianchi, E., Del Guidice, G., Verdini, A. S., Pessi, A., Int J Peptide Protein Res, 1991, 37, 7-13. [59] Jakobson, J., in: Chromatography in Biotechnology: Horvath, C., Ettre, L. S. (Eds.). Washington DC, American Chemical Society, 1993; ACS Symposium Series Vol. 529, pp. 77-84. [60] Kalghatgi, K., Fellegvari, I., Horvath, C., J Chromatogr, 1992, 604, 47-53. [61] Viscomi, G. C., Lande, S., Horvat, C., J Chromatogr, 1988, 440, 157-164. [62] Vigh, G., Varga-Puchony, Z., Szepesi, G., Gadzag, M., J Chromatogr, 1987, 386, 353-362. [63] Kalasz, H., Horvath, C., J Chrornatogr, 1981, 215, 295-302. [64] Valko, K., Slegel, P., Bati, J., J Chromatogr, 1987, 386, 345-351. [65] Subramanian, G., Phillips, M. W., Cramer, S. M., J Chromatogr, 1988, 439, 341-351. [66] Liao, A., Horvath C., Ann NYAcad Sci, 1988, 589, 182-191. 1671 Torres, A. R., Peterson, E. A., in: Separations for Biotechnology: Verrall, M. S., Hudson, M. J. (Eds.). Published for the Society of Chemical Industry by: Chichester: Ellis Horwood Limited Publishers 1987, pp. 176-184. [68] Ghose, S . , Mattiasson, B., J Chromatogr, 1991, 547, 145-153. [69] Ghose, S . , Mattiasson, B., Biotechnol Tech, 1993, 7(8),615-620. [70] Kim, Y.J., Biotechnol Tech, 1995, 9(6), 417-422. 1711 Kim, Y.J., Biotechnol Tech, 1994, 8(7), 457-462. [72] Vogt, S., Freitag, R. J Chromatogr, 1996, accepted for publication. [73] Horvath, C., Frenz, J., el Rassi, Z., J Chromatogr, 1983, 255, 273. [74] Huang, S.-Y., Jin, J.-D., Bioseparation, 1994, 4 , 343-351. [75] Gerstner, J.A., BioPharm, 1996, 9, 30.
[54] [55] [56] [57]
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
5 Affinity Chromatography Jim Pearson
5.1 Introduction Conventional adsorbents act to separate proteins by exploiting a single physicochemical characteristic such as, charge (ionexchange), hydrophobicity (hydrophobic interaction chromatography), metal ion binding (metal chelate chromatography) or covalent bond formation (thiol exchange). Since any given protein in a cell extract is not likely to be unique with regard to any one of these properties, purification to homogeneity requires that a number of different adsorption steps be performed. However, the same protein is likely to be uniquely characterized with regard to the surface distribution of charges, hydrophobic and hydrophilic amino acid residues about the surface. An absorbent which interacts in a complementary manner with these features on a given complementary protein will therefore selectively absorb this molecule, enabling purification in a single step [l]. Such material may most easily be obtained by exploiting the small effector molecules, such as substrates, inhibitors, and coenzymes, for which most proteins possess surface binding sites. The forces contributing to the formation of a binary complex result from a combination of electrostatic, hydrogen bonding, hydrophobic and Van der Waal’s interactions [2,3]. The stability of the binary complex, reflected in the affinity of the small molecule for the protein binding site, is determined by the number of such interactions and the complementarity between them [4]. For instance, as ascertained by X-ray crystallographic studies [ 5 ] , the coenzyme NAD+ binds to a specific fold on the surface of dogfish muscle lactate dehydrogense in such a manner that the hydrophobic adenine residue is accommodated in a hydrophobic pocket, whereas the negative charge on the pyrophosphate is balanced by interaction with a positively charged arginine residue. Further hydrogen bond formation takes place between the ribose hydroxyl groups and juxtaposed carboxyl functions provided by aspartate and glutamate residues, while the nicotinamide ring is supported by two hydrophobic valine residues and by a hydrogen bond formed between the carboxamide function and the side-chain amine group of a lysine residue. By chemically coupling molecules such as NAD+, now termed the affinity Zigand, to an inert solid porous support, an affinity adsorbent is produced. Proteins in a crude extract possessing complementary binding sites for the affinity ligand may then be selectively adsorbed onto the affinity adsorbent. After washing off unbound material, the protein can be recovered by
114
5 Afinity Chromatography
decreasing the stability of the binary complex, for instance by increasing the ionic strength of the solution or a change in pH. Alternatively, the protein may be selectively removed from the solid support by adding a soluble competing ligand, such that binary complex formation preferentially takes place in solution, as opposed to the solid phase.
5.2 The Principle Like hydrophobic interaction and ion-exchange chromatography, affinity chromatography is an adsorption process in which differential binding and elution to a solid phase effects separation. This process is illustrated by the picture shown in Fig. 5-1, which shows the classical image of affinity chromatography. Here, a ligand that binds to a defined site on the target protein is attached to an inert chromatographic support. When the sample mixture passes through the column, the target protein binds to the solid-support via interaction of the binding site with the immobilized ligand. The specifically bound ligand can then be recovered by changing the environmental conditions to weaken the ligand/protein interaction. Similar diagrams can also be drawn up to illustrate other adsorption processes such as ion-exchange where the ligand would be replaced by a charge group and the protein-solid-phase interaction would be formed via ion-pair interactions. What sets affinity chromatography apart from these other adsorption processes? As set out 10 years ago, the answer was simple: affinity chromatography depends on the localized interaction of a defined region of the protein surface; typically a substrate binding site or immunological epitope, with a ligand derived from the naturally occurring compound. Thus, thyroxine binds specifically to the thyroxine receptor and an immobilized antibody will bind specifically to the protein against which it has been raised. However, the development of group-specific ligands, dye-ligands, mimetic ligands and most recently combinatorial techniques, have clouded this simple picture. Affinity chromatography must now include ligands which may be directed at any region of the protein surface. In addition, the ligands used may interact via a mixture of specific sites on the protein surface as well as including a component that interacts with the overall physico-chemical properties of the protein. As detailed by Hais [6], the concept of affinity chromatography may be traced back to the use of starch powder by Starkenstein [7] to adsorb a-amylase. However, the foundation of contemporary techniques lies first in the demonstration by AxCn et al. [8] of the cyanogen bromide activation procedure which enabled the facile coupling of nucleophiles to polysaccharides, and secondly the application of this method to agarose as an affinity matrix [9]. The ability to effect outstanding protein purification was exemplified early on by the 25 000 -fold purification of human transcobalamin I1 using Sepharose@4B-immobilized vitamin B12, which as the major part of a multistep procedure eventually led to isolation of homogeneous protein from human plasma with an overall purification factor of 2 X lo6 [lo].
5.3 The Ligand
115
Fig. 5 -1. Principles of affinity chromatography
5.3 The Ligand Of central importance is the choice of affinity ligand, which in turn affects the binding specificity of the affinity matrix and its cost. The general properties required of an affinity ligand may be summarized as follows [ l l ] ; 1. It must be easily derivatized to allow coupling to a support matrix in such a way as to leave the binding properties unaltered.
116
5 Affinity Chromatography
2. The ligand should be sufficiently stable to withstand the matrix coupling procedure and also any washing steps required during column operation. 3. The dissociation constant of the ligand/protein binary complex should be less than M so as to ensure significant binding to the adsorbent [12]. In addition, with a dissociation constant of less than M, severe conditions may be required to effect desorption [13]. Ligands suitable for immobilization may be divided into two groups, firstly ‘specific ligands’, secondly ‘group-specific’ or ‘general’ ligands. The first groups species such as hormones and monoclonal antibodies which bind complementary receptors and antigens respectively in a highly selective manner. However, the specificity achieved often entails very tight binding of the analyte to the column, and also that the adsorbent be dedicated to the purification of a single protein. A separate affinity adsorbent is therefore required for each protein and a separate immobilization chemistry has to be developed for each ligend. These problems may be partly alleviated with groupspecific adsorbents [14], in which an affinity ligand binding to an entire functional class of proteins is employed, enabling a single-affinity adsorbent to be applied to a range of proteins. Particularly useful are adenine nucleotide coenzyme-based materials since these compounds are found to bind over 30 % of recorded enzymes [ 151. Examples of such ligands include, NAD+ [14], AMP [I61 and coenzyme A [17]. Since protein binding to group-specific adsorbents may not achieve complete selectivity, additional resolution is normally obtained by a selective elution method, such as applying gradients of ionic strength, organic co-solvents, or competitive ligands. Selective elution may also occur through the formation of ternary complexes with free coenzyme in the eluant. Thus NAD+/pyruvate and NAD+kydroxylamine effected the specific elution of lactate dehydrogenase and yeast alcohol dehydrogenase respectively from an AMP-Sepharose@column. Conversely, ternary complex formation may also be used selectively to prevent adsorption [IS].
5.4 The Matrix The characteristics of the column support material listed in Table 5-1 play a major role in the performance of any affinity column. Table 5-1. Main requirements for affinity column packing materials. Rigid Spherical Hydrophilic Chemically resistant Large pore size Large surface area High density of chemically modifiable groups
5.4 The Matrix
117
5.4.1 Particle Size Before application to the affinity column many extracts may be quite viscous and contain large amounts of fines. Consequently, evenly packed beds of large spherical particles of around 100 ym diameter allow liquid to flow in the interstitial spaces without clogging up the bed.
5.4.2 Chemical Robustness Column fouling will inevitably occur, typically from cell debris, lipids, and protein aggregates and these must be removed before re-using the column if the performance is not to degrade. This produces a demand for chemically robust materials that can easily withstand the harsh environment of the recycling process. Cleaning, depyrogenation and sterilization of the column normally requires treatment with one of the following processes [19]:
- For sterilization: soaking in 70 % aqueous ethanol - this process will sterilize the column; however, the high amount of organic solvent may cause shrinkage of hydrophilic materials such as dextran and agarose. This in turn inevitably leads to cracks in the gel bed. - For sterilization and depyrogenation: pyrogens may be removed by soaking the media in 1 M sodium hydroxide for 5 h [20]. This process hydrolyses the pyrogens to produce soluble products which can then be washed off the column. Alternatively, a solution of 60 % ethanol containing 0.5 M acetic acid also produces a similar result; however, the high solvent content may shrink the gel bed (see above). For agarose supports, a mixture of 20% ethanol and 1 M acetic acid is equally effective. An additional requirement for materials to be used in affinity chromatography is that they withstand the harsh chemical treatments involved in the ligand coupling steps. This may involve high concentrations of organic solvent, and extremes of pH and temperature.
5.4.3 Particle Shape and Rigidity The recycling steps may constitute a significant portion of the total process cycle time. Therefore packings which allow for high linear flow rates are preferred. This requirement demands monodisperse spherical particles in well-packed beds that maximize the interstitial gaps in the bed [21]. In addition, the support particles must be sufficiently rigid so that they do not deform at high flow rates where the viscous drag of the solvent on the bed particles produces large compressive forces on the bed. Compressible particles packed into a bed typically show a critical point as the linear
118
5 Affinity Chromatography
-
Pressure Flow Curve of ACL Mimetic Blue 1 and Sepharose CLGB 1.4
1.2
1
B n 0.8 f!
a (d
ae! . 0.6 0.4
0.2
0 0
100
200
300
500
flow cmlh
Fig. 5-2. Comparison of pressure - linear velocity profites in affinity chromatography.
velocity of the solvent passing through the bed reaches the point at which the particles begin to compress. Figure 5-2 compares the pressure-linear velocity profiles of a 3.2 X 14.5 cm bed of agarose spheres (Sepharose@CL6B) and one of cross-linked agarose spheres (Mimetic" blue 1 6XL, Affinity Chromatography Ltd.). The more rigid cross-linked material fails at much higher flow velocities.
5.4.4 Pore Size and Accessible Internal Volume The bead must have a system of pores large enough to allow free diffusion of protein in and out of the bead. These pores must also form an interconnecting network throughout the bead to expose the entire internal volume of the bead to the protein.
5.4 The Matrix
119
5.4.5 Surface Chemistry Chemical modification of groups on the pore surface is required to immobilize ligands to a high density within the bead. Suitable groups include primary hydroxyls which can be activated to form epoxides or tosyl groups so that amine-containing ligands can be linked up straightforwardly.
5.4.6 Low Non-Specific Binding It is vital that separation occurs on the basis of interactions of the protein with the ligand alone. The matrix should form an inert scaffold to present the ligand to the protein. But no matrix can be completely inert. Hydrophobic patches on the matrix will adsorb many proteins at high salt levels - precisely the conditions often used to elute a protein from an affinity column. Ion-exchange sites similarly occur on many materials. Therefore the ideal matrix contains a completely hydrophilic, noncharged surface, typically containing large numbers of hydroxyl groups. These can also form the basis for chemical modification to introduce ligands onto the gel, for instance through reaction with cyanogen bromide, cyanuric chloride, or epichlorohydrin. However, the hydroxy groups in many polymers such as beaded dextrans and agarose perform multiple functions; they act as hydration sites to retain water within the gel bead, without which the bead collapses, and they present a hydrophilic surface to the bead. If many of these hydroxyl groups are removed in the process of activation and ligand coupling, then the gel bead may collapse through removal of water of hydration. In addition, the resulting beads may take on a markedly hydrophobic character. Table 5 - 2 lists some support materials used in affinity chromatography. Some have not found wide-scale application. Thus, beaded dextrans have small pores and the highly hydrated gel network is easily collapsed if water is removed from the gel in the process of either activating the gel or in the cleaning protocols outlined above. Materials such as control pore glass have small surface areas accessible to proteins and are not mechanically robust. Materials based on macro porous polymers show promise; for large-scale use they would have to demonstrate high internal volumes accessible to proteins. Beaded agarose continues to be the material of
Table 5-2. Affinity support materials. Beaded dextran Beaded agarose Coated silica Control pore glass Macroporous polymers [21] Perfluorocarbons
120
5 AfSinity Chromatography
choice for affinity applications as it combines a number of suitable properties. It is highly hydrophilic with few non-specific ion-exchange sites and possesses an abundance of immobilization sites. The bead structure does not depend upon a delicate framework of hydrated polysaccharide chains as beaded dextrans do, instead it is built up from a macro reticular scaffold of polysaccharide chains which form long helices, leaving very large pores throughout the bead. Thus, for the 6 % agarose beads typically used in affinity chromatography, the molecular weight cut-off is over 1 million and small proteins such as serum albumin can access over 90% of the internal bead volume. The hydrated framework presents many sites for chemical modification via primary hydroxyl groups, and up to 50 pmol per g of ligand may be introduced onto the matrix without affecting the integrity or internal volume of the bead. Lastly, agarose is chemically resistant to all the processes required in recycling affinity media, including extended exposure to 1 M NaOH for depyrogenation and sterilization. Agarose beads withstand polar solvents such as ethanol, isopropanol, acetone, dioxane, and dimethylfonnamide (DMF).
5.5 Ligand Selection and Development Several approaches can be identified by which ligands for affinity chromatography have been selected: Table 5-3.Methods for ligand selection. Naturally occuring ligand in vivo Naturally occuring ligand derivatives may not be stable chemical or biologically. Often no naturally occuring small ligands present ie antibodies, hormones. - Group-specific ligand - coenzyme, boronate, metal ion. - Only relavent to a small subset of proteins.
-
Ligand screening - dye-ligand, mimetic - Highly robust - General specificity applicable to wide range of proteins. - Specificity may be low. Ligand selection through combinatorial library - May offer methods applicable to all proteins to produce highly specific ligands which are chemically stable.
5.5.1 Ligand Choice Choose a ligand based on the known or suspected biological activity of the target protein by immobilizing the in vivo substrate, inhibitor or effector of the target protein. Two major drawbacks occur with this approach. First, the ligand may be subject to microbiological attack or be unstable. Second, the immobilization process may
5.5 Ligand Selection and Development
121
interfere with the protein-ligand interaction if the linkage site from the matrix occurs at a position on the ligand which is normally not accessible to solvent when the ligand is bound to the protein. For instance, in a study into the purification by affinity chromatography on AMP-agarose adsorbents [15], it was found that linkage of the AMP-affinity ligand via the adenine N6 position significantly favored binding compared with similar adsorbents prepared by linkage through the adenine C 8 position. This preference was found to reverse for dehydrogenases of a microbial origin. Choose a general ligand. Many coenzymes and cofactors bind a wide variety of proteins. For instance, NAD+ binds to NAD+-dependent dehydrogenases. Affinity ligands based upon this type of ligand may therefore bind a wide variety of proteins. Extra specificity may be introduced if ternary complexes form with the protein, ligand and a small substrate or inhibitor molecule, for instance the NAD+/pyruvatelliver alcohol dehydrogenase system. Other general ligands that have found widespread application in affinity purification include: Protein A - which is useful for the separation of immunoglobins; heparin - which binds a variety of serum proteins including antithrombin I11 in a specific ternary complex, as well as acting as a general ion-exchanger; and aprotonin, which binds a variety of proteases. As well as naturally occurring general ligands, a number of synthetic compounds are available which provide supports which are low cost and robust. These include phenylboronic acid, which is useful for the separation of glycoproteins [22,23], immobilized metal ions, and hydrophobic interaction supports comprising either long-chain alkyl groups or phenyl rings. These materials are often not classified as affinity supports as they do not display selectivity for any particular class of proteins. However, they do display site-selective binding based upon specific interactions with defined hydrophobic pockets on protein surfaces. Nonetheless, this class of compounds also separates according to the global properties of the protein and as such may also be classed under ion-exchange chromatography as a general purpose technique. Ligands based upon synthetic sulphonated polyarornatic chrornophores: dyeligand adsorbents. The principle involved here is very different from the approach outlined above. In many practical problems, little may be known of the three-dimensional structure of the target protein; therefore rational design of affinity ligands may be out of the question. Thus, to use an affinity technique, the ligand must be selected by a screening process. The most widely applied example of this approach lies in the exploitation of ligands derived from the chemistries used to manufacture commercial dyestuffs. The connection between the textile dyestuff industry and ligands for protein purification may at first sight seem surprising. However, there are good reasons for the connection, since both applications demand access to a wide variety of stable chemical structures, which can be easily immobilized into hydrophilic surfaces to yield a stable product that does not leak or fade. Group-selective affinity adsorbents based upon triazine dye ligands have been exploited in the purification of an extensive range of enzymes and proteins [24]. These materials are especially suited to large-scale applications. For instance, dyeligand chromatography was applied to the single-step purification of extracellular thermostable proteinases from thermophilic Bacillus and Therrnus cultures to give purification of 115- and 2195 -fold repectively [25]. Dye-ligand adsorbents often display high protein binding capacities, which may exceed that of biologically de-
122
5 Affinity Chromatography
rived affinity media by a factor of 10-100 [26]. In many cases, different triazine dyes show a marked variation in protein binding, for instance measurement of the dissociation constand, Kd, displayed by a selection of 19 dyes towards pig heart lactate dehydrogenase, yielded values ranging between 0.8 pM and 123 pM [27]. In order to exploit fully any differential affinity shown by triazine dyes in protein purification, a number of authors have carried out systematic small-scale screening tests to ascertain which derivative offered the best resolution and yield. These include studies of 14 dyes [28], 37 dyes [29] and 65 dyes [30]. Scopes [31] was able to divide the currently available dye-ligands into five groups, such that those found to bind least protein were placed in group 1, and those binding the most in group 5 . The ordering did not change dramatically between different sources or when adsorption conditions such as ionic strength and pH were varied. Thus, it was suggested that only one dye from each group need be selected for initial screening tests, greatly simplifying the procedure. However, the value of such screening tests is often diminished by insufficient characterization of the dye-matrix in terms of immobilized dye. The concentration of dye immobilized on agarose matrices reported in the literature can be quite variable and this factor should be taken into account when comparing different dye-adsorbents. Commercial screening kits are now available from Affinity Chromatography Ltd, which provide a selection of dye-ligand adsorbents manufactured according to defined conditions to provide consistent ligand purity and binding capacity. Screening tests may also unveil a dye which does no bind the protein of interest, but instead removes a substantial amount of contaminating material. If so, this may then be used as a preliminary ‘negative’ step in cleaning up the protein extract prior to a second ‘positive’ step. A number of processes ursing such tandem columns have been reported [32,33]. Since triazine dyes act as group-selective ligands, high selectivity is seldom realized in the adsorption stage and further resolution is dependent upon selective desorption using salt gradients, changes in pH or temperature, and affinity elution using competitive substrates, nucleotides or even free dye. Using such techniques, excellent purification is possible with dye-ligand affinity chromatography, as witnessed by the 12.600-fold purification of bovine lens aldose reductase [34]. There is now therefore no fixed dividing line between conventional adsorption process such as hydrophobic interaction and ion-exchange chromatography and affinity chromatography. At one end of the spectrum is ion-exchange chromatography, which separates proteins depending on their overall charge. At the other end is classical affinity chromatography, which separates entirely through localized interactions related to a small subset of the protein surface. In between lie many processes which do not fit into these two neat camps. For instance, phenyl agarose may separate proteins both through their overall partition coefficients, but also through interaction with hydrophobic pockets on the protein surface. Similarly dye-ligands possess arrays of aromatic groups and charged sulfonate residues and may behave as ion exchangers at low pH, while showing selective behavior based on complex interaction with the protein surface when conditions are chosen which balance the charge and hydrophobic interactions.
References
123
5.6 Combinatorial Ligand Selection This technique, at the moment very much in vogue [35-371 is a logical development from the ligand screening methods using triazine-based textile dyes. These screening methods based upon pre-existing ligands can only explore the interaction of a few dozen ligand structures with any protein. However, combinatorial methods offer ways to synthesize thousands of ligand structure. To enable this method the following conditions have to be met: The ligand must be built up from a collection of common precursors. The method must dictate defined combinations of the precursors to build up the final ligand on the solid phase. - Variation in the combinatorial ligand must allow for differential binding to target protein according to the combination of precursors used to build up the ligand. -- The ligand must be bound to an inert solid phase so that interaction with the protein reflects the ligand chemistry and not background adsorption to the solid phase. -
Among the first reports on combinatorial ligand selection was, ironically enough, the selection of RNA molecules to bind triazine-based textile dyes [38], normally used in affinity chromatography. Since then the technique has been extended to libraries of polypeptides [39], oligonucleotides [40], oligocarbamates [4 11, D-amino acid hexapeptides [42], and aminimides [43]. Most activity in this area has been concentrated on the building of vast libraries for the screening of potential pharmaceuticals. The screening of these vast arrays need only be carried out at one defined set of conditions - those expected in the physiological environment of the potential drug target. Pharmaceuticals must target specifically a single binding site on the target molecule. In the case of affinity ligand however there is no such restriction as any interaction with the target molecule surface serves as the basis for an affinity adsorbent. In addition, there is a much wider freedom in the environmental parameters such as ionic strength, pH, temperature, and even solvent. Therefore screening for a ligand suitable for affinity separation takes up a much greater part of the total effort. For this reason, libraries for screening affinity ligands can be comparatively small - in the region of a hundred or thousands of combinatorial structures.
References [ l ] Porath, J. J Chromatogr 1981, 218, 241-259. [2] Katchalski-Katzir, E., in: Affinity Chromatography and Biological Recognition, Chaiken, I.M., Wilchek, M., Parikh, I. (Eds.). London: Academic Press, 1983; pp. 7-28. [3] Mattos, C., Ringe, D. Nature Biotechnol 1996, 14, 595-599. [4] Lyklema, J., in: Affinity Chromatography and Related Techniques, Gribnau, T.C.J., Visser, J., Nivard, R.J.F. (Eds.). Amsterdam: Elsevier, 1982; pp. 11-27. [5] Holbrook, J.J., Liljas, A,, Steindel, S.J., Rossman, M.G., in: The Enzymes, vol. 11, Boyer, P.D. (Ed.). London: Academic Press, 1975; pp. 191-292.
124
5 Affinity Chromatography Hais, I.M. J Chromatogr 1986, 376, 5-9. Starkenstein, E. Biochem Z 1910, 24, 191-209. Axtn, R., Porath, J., Ernbach, S. Nature 1967, 214, 1302-1304. Cuatrecasas, P., Wilchek, M., Anfinsen, C.B. Proc Nut1 Acad Sci USA 1968, 61, 636-643. Allen, R.H., Majerus, P.W. J Biol Chem 1972, 247, 7709-7717. Lowe, C.R. Int J Biochem 1977, 8, 177-181. Graves, D.J., Wu, Y.T., in: Methods in Enzymology, vol. 34, Jackoby, W.B., Wilchek, M. (Eds.), New York: Academic Press, 1974; pp. 140-163. Yang, C., Tsao, G.T., in: Advances in Biochemical Engineering, vol. 25, Fiechter, A. (Ed.). Berlin: Springer-Verlag, 1982; pp. 19-42. Mosbach, K., Guilford, H., Ohlssen, R., Scott, M. Biochem J 1972, 127, 624-631. Lowe, C.R. Pure and Appl Chem 1979, 51, 1429-1441. Brodelius, P., Mosbach, K. FEBS Lett 1973, 35, 223-226. Rieke, E., Barry, S . , Mosbach, K. Eur J Biochem 1979, 100, 203-212. Svanas, G., Weiner, H. Anal Biochem 1982, 124, 314-319. Boschetti, E., Pouradier Duteil, X., Nguyen, C., Moroux, Y. Chimicaoggi 1993, MarcNApril, 29-35.
Sofer, G.K., Nystrom, L.E. Process Chromatography - A Practical Guide. London: Academic Press Ltd, 1989; pp. 51-85. Svec, F., FrCchet, J.M.J. New designs of macroporous polymers and supports: from separation to biocatalysis. Science 1996, 12, 205-211. Psotova, J. and Janiczek, 0. Chemicke Listy, 1995, 89, 641-648. Liu, X.C., Scouten, W.H. J Chromatogr 1994, 687, 61-69. Burton, S.J., in: Methods in Molecular Biology, vol. 11, Kenny, A,, Fowell, S . (Eds.), Totowa, New Jersey: The Humana Press Inc., 1992; pp. 91-103. Cowen, D.A., Daniel, R.M. J Biochem Biophys Methods 1996, 32, 1-6. Scawen, M.D., Darbyshire, J., Harvey, M.J., Atkinson, A. Anal Biochem 1982, 132, 413417.
Clonis, Y.D., Lowe, C.R. Biochern J 1980, 191, 247-251. Clonis, Y.D., Lowe, C.R. Biochim Biophys Acta 1981, 659, 86-98. Hammond, P.M., Atkinson, T., Scawen, M.D. J Chromatogr 1986, 366, 79-89. McFarthing, K.G., Angal, S . , Dean, P.D.G. Anal Biochem 1982, 122, 186-193. Scopes, R.K. J Chromatogr 1986, 376, 131-140. Miribel, L., Goldschmidt-Clermont, P., Galbraith, R.M., Amaud, P. J Chromatogr 1986, 363, 448-455.
Shelton, M.C., Toone, E.J. Tetrahedron-Asymmetry 1995, 6, 207-21 1. Inagaki, K., Miwa, I., Okuda, J. Arch Biochem Biophys 1982, 216, 337-344. Alper, J. Science 1994, 264, 1399-1401. Medynski, D. Bioflechnology 1994, 12, 709-710. Borman, S. Chemical Engineering News 1996, February 12. Ellington, A.D., Szostak, J.W. Nature 1990, 346, 818-822. Lam, K.S., Salmin, S.E., Hersch, E.M. Hruby, V.J., Kazmierski, W.M., Knapp, R.J. Nature 1991, 354, 82-84. Gold, L. Nature Biotechnol 1996, 14, 1080. Cho, Y.C., Moran, E.J., Cherry, S.R., Stephans, J.C., Fodor, S.P.A., Adams, C.L., Sundaram, A., Jacobs, J.W., Schultz, P.G. Science 1993, 261, 1303-1305. Dooley, C.T., Chung, N.N., Wilkes, B.C., Schiller, P.W., Bidlack, J.M., Pasternak, R.A., Houghten, R.A. Science 1994, 266, 2019-2022. Peisach, E., Cassebier, D., Gallion, S.L., Furth, P., Petsko, G.A., Hogan, J.C., Ringe, D. Science 1995, 269, 66-69.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
6 Large-scale Chromatography: Design and Operation C.J. A. Davis
6.1 Introduction In the past two to three decades significant advances have been made in the field of bioprocessing. These are largely due to the development of recombinant DNA techniques, which have allowed scientists to manipulate the genetic material within cells, leading to the development of processes to manufacture a range of biological molecules. The advances in genetic engineering have required, and contributed to, the development and refinement of protein purification techniques and in particular, chromatographic separations. These techniques are equally applicable to non-recombinant proteins, such as in blood fractionation, and other biological products. There is a considerable amount of literature [ I ] covering the separation and purification of biological molecules, but little on the issues to be addressed on the economic scale-up for commercial manufacture. The aim of this chapter is to provide an insight into the engineering design process, the issues involved, and some of the different options available. The issues raised are primarily concerned with the production of commodity-type products, such as some recombinant blood proteins. These commodities, unlike most recombinant products, are price-sensitive and therefore the process development and process engineering issues often differ from those in the high value, low volume biotechnology markets. The term ‘large-scale’ therefore refers to the production of multi-kilogram quantities of product with reusable columns and the associated engineering issues. Some of the issues raised are directed towards the biopharmaceutical industry and are not totally relevant to other industries. While designing and operating chromatographic processes, the process engineer should consider the aspects of current Good Manufacturing Practice (cGMP) required by the pharmaceutical industry. This includes validation, which ensures that a consistent, safe and, where appropriate, efficacious product is produced. It is also important that the safety of operators and maintenance engineers are considered. If the product is a potential health hazard then the appropriate levels of containment must be designed into the process. Environmental concerns should also be considered early in the design process. Therefore the key requirements for the process engineer when designing and operating a large-scale chromatography process are [ 2 ] :
126
6 Large-scale Chromatography: Design and Operation
1. The process must be reproducible, and the associated equipment must be robust. 2. The process, equipment and facility must be designed to ensure that the required levels of hygiene and safety are maintained during operation. 3. The process and equipment are designed such that data can be collected to demonstrate that requirements (1) and (2) can be met.
An important concept to be considered during the development of chromatographic processes is that of ‘scale-down’. Scale-down is the design of small-scale processes to mimic the final production scale in order to investigate key factors where appropriate, This will increase confidence in using the data generated to design a manufacturing facility. Scale-down is often difficult in the laboratory, but should be an important consideration in pilot-scale work. The operating philosophies to be followed at production scale, such as buffer preparation, should be mimicked in the laboratory and pilot plant. This will give valuable insights into potential operational issues. It is also important that process engineers are involved in the development process to minimize the scale-up issues.
6.2 Chromatography Systems 6.2.1 Introduction The development of chromatography systems is well defined and this section summarizes the concepts involved in developing such systems. It aims to provide an understanding of the concepts involved, where cost savings can be made and how systems are built up.
6.2.2 System Design The main design criteria to be satisfied before any design work starts are: -
What is the end product?
- What will the end product be used for? - What will be the approximate sales revenue? These questions should be answered in order that the initial financial viability of the project can be assessed. This will then allow the project scope to be accurately defined. The following five key factors should be considered for the process design of a chromatography purification system for the large-scale production of biopharmaceuticals :
6.2 Chromatography Systems
1. 2. 3. 4. 5.
127
Process control. Maintenance of sanitary conditions. Compliance with cGMPs. Validation of process and installation. Capital and operational costs.
The final design is often a compromise of these five factors. However, within the biopharmaceutical industry the compliance requirements can be an overriding factor. The levels of hygiene required and validation will be discussed elsewhere in this chapter. A typical column chromatography process consists of a series of steps. The control strategy to end each step should be kept as simple as possible and there are five common methods of doing this; absorbance, pH, conductivity, air sensors, and volume. Examples of the use of these methods are; pH to end equilibration, air sensors to end column loading, conductivity to control wash steps, absorbance to control product collection, and volume for fixed volume steps. Some designs also use pH and conductivity detectors upstream of columns to give confidence that the solution going onto the column is within the target specifications. While developing the process, the research and development scientists will have used various instruments to control the process. However, when designing largescale equipment it may be possible to exclude superfluous instrumentation. If, for example, conductivity has no control function then it could be omitted at the process scale. This will save capital and operational costs due to there being one less instrument to install and maintain. However, this approach will restrict the data available for monitoring column performance and the benefits may be outweighed by the advantages of including all instrumentation. A diagram of a typical large-scale column chromatography system is shown in Fig. 6-1. The system consists of a series of vessels connected to a chromatography column with a pump and the instruments required to run a process. The pump is fed with a series of buffer vessels holding the solutions used to perform chromatography, and the feed vessel containing the protein load material. The air sensor on the feed pipe work is designed to terminate loading when the feed vessel is empty. The reaction time of air sensors is such that the sensor should be positioned as close to the buffer manifold junction as practical to minimize protein losses, but is a compromise between minimizing protein losses and system reaction time. The pressure transducer after the pump is shown for use with positive displacement pumps where pressure cut-outs are required for safety, due to the limited pressure rating of equipment, and should be placed immediately after the pump. The guard filter and air trap protect the column and matrix from pariculates and air bubbles. The air sensor after the air trap is a fail safe to prevent air ingress onto the column and the consequent problems, including the possibility of column repacking. Immediately before and after the column are pressure indicators to monitor the pressure drop across the column (the pressure transducer before the filter can not be used for upstream column pressure measurement due to the pressure drops across the filter and air trap). The column pipework is designed for flow through the column in either direction, allowing column cleaning in the opposite direction to processing. This is
128
6 Large-scale Chrontatography: Design and Operation
Buffer vessel
Feed Vessel
Buffer vessel
Control feedback loop ..................................................
:
Air trap PT
rn
PI
+ Filter
AT Air sensor PI: Pressure gauge FT: Flow transducer
Product vessel
Waste
I
Fig. 6-1. Schematic diagram of a typical large-scale column chromatography system.
perceived to be a more efficient method of cleaning, as it involves running the cleaning fluids counter-current to the process flow direction. The flow meter controls the flow-rates through the column by modulating the pump output. The flow meter also totalizes liquid volume, allowing volumetric control. Subsequent to the flow meter there are the instruments used to control and monitor the purification process, normally pH, UV, and conductivity. Finally, the column outflow is directed either to waste or product collection. It is common practice to employ gradient elution in laboratory-scale purifications. For large-scale chromatography the use of gradients is not recommended, due to the added complexity of design and operation, and wherever possible step elutions and washes should be the method of choice. However, if the process requires gradient steps these can be designed and built into the system. The use of in-line dilution is however practical and can significantly reduce the tankage volume, and associated costs, of process plant. An example of a process strategy for in-line dilution is shown in Fig. 6 - 2 . In the example in Fig. 6-2 the process stream, load material or buffer, is being diluted with water to give a target conductivity at the conductivity probe (AIC) and a target flow rate through the column (FIC). The two pumps are started at
129
6.2 Chromatography Systems
I
I
I
I
Process
I I
($)
In-line mixer.
1
Q
0
U
I I
11
Diluent
Fig. 6-2. In-line dilution process strategy.
fixed outputs, which are known to give approximately the correct dilution, and then the system is switched to automatic control. The control system is set up so that if the conductivity drifts, the output to the diluent pump is altered. This affects the target flow rate through the column and so the output to the process pump is altered. While this type of system has been shown to provide accurate and stable control at a given setpoint, the tuning of the control parameters is critical. The design of the system should be appropriate to the operating flow rates in terms of pipe and pump sizes. It is important that at the large scale all steps are operated at similar flow rates, as the ability to control flow accurately over a wide range of flow rates is limited. Care should be taken to ensure that all process contacting materials of construction are compatible with the process conditions, including cleaning-in-place (CIP) and steaming-in-place (SIP), and approved by the Regulatory Authorities. Having developed the process pipework, the services have to be built around the system. These are compressed air, steam, CIP, steam traps, and drains. The detailed design of the pipework must ensure that the steam traps and drains are positioned correctly to ensure full drainage of systems during CIP and SIP. Figure 6-3 shows a typical arrangement for services on a system that can be cleaned and steamedin-place. The CIP supply and returns are numbered, indicating possible routes for cleaning fluids through the system. The steam supplies are set up to allow SIP of the key units independently of the rest of the system.
130
6 Large-scale Chromatography: Design and Operation
6.2 Chromatography Systems
131
Once the system has been designed and all the equipment sized, the detailed design and general arrangement (GA) drawings are developed. There are a number of important factors to consider during detailed design, including access for routine operation, cleaning, packing, and maintenance. The layout is dependent upon the specific system requirements along with the degree of automation and the available space. Chromatography processes are generally located in environmentally controlled areas and there will be pressure to keep these areas to a minimum. It is possible to locate buffer vessels and other large items of equipment outside of the controlled environment.
6.2.3 Automation Chromatography systems are generally automated to some degree and most of the process development will have been carried out with automated systems. The degree and type of automation specified will depend on the plant, the process and the economics of production. The potential advantages of automation are [3]: - Improved process reproducibility and control. -
Remote operation. Improved fail-safe capability. More reliable and complete historical records. Increased productivity through full-time equipment usage.
Process control of chromatography systems is well proven and the considerations that have to be taken into account are industry wide. They include the choice of hardware and software, and the validation considerations. The configuration should have the required processing power and be user friendly to facilitate development and implementation. It should include packages for sequencing (process control), trending (graphical representation of on-plant activities and derived data points), graphics (diagrammatic representation of the plant showing equipment status), security, alarm systems, and data back-up [3]. Increasing attention is being paid to electronic batch records and laboratory information management systems (LIMS) which can be integrated into the process management system, thereby moving towards the concept of paperless manufacturing. There are two main microprocessor-based control systems, DCS (distributed control systems) and PLC/SCADA (programmable logic controller/supervisory control and data acquisition) systems [4]. In both systems the operator interface is at a terminal remote from the software and the input, output, and control electronics. The user requirement specifications for any system should be clearly defined before vendors are approached as there are a multitude of different products on the market. DCS are fully integrated systems where all the required packages are included in one system, which has proven system integration, from one supplier. The advantages include the facility to maintain back-up electronics for control functions and critical instruments, improved reliability, system flexibility, and the use of information technology to store and present data.
132
6 Large-scale Chromatography: Design and Operation
PLC/SCADA systems differ from DCS in that they utilize components from a variety of different manufacturers which are then integrated into one system. The PLC hardware and programming packages are not normally obtained from the suppliers of the SCADA software packages. PLC/SCADA systems are generally cheaper than DCS and more readily available. However, it is necessary to program both the PLC and SCADA systems while ensuring the systems communicate with each other. Some system redundancy is lost and the use of multiple suppliers can complicate matters and may require the use of a systems integrator. The need for validation of computer systems used in a cGMP environment should be established at the start of a project. Validation will provide the documentary evidence that the system operates according to the user requirement specifications reliably and reproducibly over the system life cycle. This consists of hardware and software preparation, testing, and documentation. Guidance on this subject has been published by both the FDA and the Pharmaceutical Manufacturing Association’s computer systems validation committee [3].
6.3 Column Design 6.3.1 Introduction The scale-up guidelines for chromatography columns are summarized by Sofer and Nystrom [ 5 ] . One needs to maintain bed height, linear flow rate, sample concentration, and specific column loading while increasing load volumes, volumetric flow rates, and column diameters. While this is true for axial column configurations the development of radial flow chromatography has required different approaches to scale-up to be developed. An alternative approach to scale-up of columns was suggested by Naveh [6]. When columns are scaled-up on the basis of constant bed height, differences between laboratory and large-scale performance can be observed. In these cases appropriate scale-down experiments can be used to mimic the production scale process in the laboratory. The choice of column design can have a major impact on the design of systems and must be considered early in the design process. The design choice should be made after examining and analyzing the available options. The three standard materials of construction for chromatography columns are glass, acrylic, and stainless steel. At large scale there are no benefits from glass columns or columns with viewing ports. The portion of gel that is visible is a small fraction of the bed and the chances of observing potential problems are minimal. A significant disadvantage is the limited pressure rating of large glass columns, which decreases with increasing column diameter. It is important that the column materials of construction are stable to all column-contacting fluids. Similarly, acrylic columns offer no advantages over stainless steel and therefore stainless steel is recommended as the material of construction for large-scale chromatography columns.
6.3 Column DesiPn
133
6.3.2 Axial Flow Chromatography Axial flow chromatography is the traditional design where solutions flow perpendicular to the bed of matrix. The advantages of this are that it is a well-characterized system and the principles of scale-up are understood. The disadvantages are that the large columns can have flow distribution problems and they take up a lot of space. The packing of large axial columns can be difficult and often requires expert assistance. The majority of column designs are packed while open to the environment, raising issues concerning the ingress of contaminants during column packing. This is of particular importance to the pharmaceutical industry where environmental protection local to the column is required. There have been recent developments in the design of axial flow columns which can be pump packed. Axial column designs range in complexity from simple fixed bed columns to those that can be pump packed. Separations requiring long bed depths, such as gel permeation, can utilize columns in series that provide the required bed height. The three basic column designs are shown in Fig. 6-4. The three basic types of axial flow column are fixed bed, variable bed, and fixed bed with enclosed packing. Fixed bed axials are a simple and traditional column design and therefore well characterized. When designing production-scale systems, the bed height will be fixed and therefore the added complexity of variable bed height should not be necessary. However, packing fixed bed height columns is more complex than with variable bed height columns and this is often the driving force behind the choice of variable height columns. Care must be taken in ensuring that the column seal design, in variable bed columns, does not leave a dead zone, where the matrix is not flushed, around the outer edge where hygiene could be compromised. Recent designs of fixed bed axial columns include the facility to lower the top plate by a few centimeters to complete the packing process. This combines the packing advantages of variable bed axial columns with the economics of fixed bed columns. Enclosed packing overcomes some of the concerns about environmental exposure and provides a more robust and reproducible method of packing, as for radial flow columns (see section 6.3.3). The disadvantages are the expense and the mechanical complexity of the packing and unpacking nozzles. Expanded bed chromatography is a relatively new development in the field of industrial bioprocessing and is designed to combine clarification, concentration, and protein capture in one step, thereby reducing the number of unit operations required [7-91. The technique has been scaled up to 1.5 m diameter columns. The column is designed to operate in both the packed and fluidized mode by using a special flow distributor and moveable adapter. The column is equilibrated, loaded, and washed in the fluidized mode and eluted and cleaned in the packed mode. Cells and cell debris from fermentation, or cell disruption, are loaded onto the bottom of the column and pass through to waste, negating the requirement for unit operations to clarify the broth. In the fluidized mode, the properties of the matrix are such that a stable and homogeneous expanded bed is created and plug flow is achieved, producing characteristics similar to that of a packed bed. The potential advantages of the
134
6 Large-scale Chromatography: Design and Operation Flow distribution plate
-
-
7 1 1 Fixed Bed Axial Flow Column
-4 FTJ
Variable Bed Axial Flow Column
J L
Bed height adjuster
>
Flow distribution plate
Fixed Bed Axial Flow Column: Enclosed Packing Retractable unpacking nozzle Rocess idout
Fig. 6-4. Schematic representation of three axial column designs.
6.3 Column Design
135
technique center around the improved process economics of reduced capital equipment costs and overall reduced processing times. The disadvantages include the increased mechanical complexity, the increased space taken up by the system and, potentially the decreased matrix lifetime due to bead attrition.
6.3.3 Radial Flow Chromatography Radial flow chromatography is a relatively new development where the direction of flow is radially through a vertical bed of matrix. It is suitable for adsorptioddesorption-type chromatography, but is unsuitable for isocratic separations, such as gel permeation chromatography, due to the short bed depth [lo], and non-compressible resins where matrix settling can result in a poorly packed bed. A schematic representation of a radial flow chromatography column is shown in Fig. 6-5. During normal processing, solutions enter the column through the solumn entry (inlet) port, 1, and then flow across the top of the bed, bypassing the four column packing ports (two shown) and down the distribution channel, 8. Flow then passes radially through the bed contained within the outer and inner meshes, 5 and 6, before passing down around the central core, 7, a hollow cylinder which directs flow out of n
Key: 1. 2. 3.
4. 5.
6. 7. 8.
Inlet port. Top packing port. Bottom packing port. Outlet port. Outer mesh.. Innermesh. Central core. Distribution channel.
Fig. 6-5. Schematic representation of a radial flow chromatography column.
136
6 Large-scale Chromatography: Design and Operation
the column, and out of the outlet port, 4.The column is packed, in a contained manner, by pumping matrix into a flooded column through the packing ports, 2. Column unpacking is by pumping buffer into the column and the matrix exits through the unpacking ports, 3. The advantages of radial flow chromatography center on the large cross-sectional area perpendicular to the flow direction across a short flow path. Therefore, the pressure drop across the bed is reduced, in comparison with axial flow columns of equivalent size, permitting the use of higher flow rates. This allows greater potential for increasing throughput and concurrently reducing the quantity of matrix required without altering productivity [ 11,121. This makes the technology a financially attractive alternative in process design. Other advantages include the ability to pump pack the columns and the reduced floor space occupied by radial flow columns compared with similar axial flow columns. The disadvantages center on the increased costs and mechanical complexity. Pump packing of chromatography columns (both radial and axial) is perceived to have significant advantages over the packing methods used on some axial flow columns. The degree to which the matrix is exposed to the environment is reduced in that the matrix is transferred to a packing tank where the matrix can be equilibrated in a closed environment before pump packing. In addition, the packing process is more reproducible in that the columns are packed by pumping in the matrix until a predetermined pressure is reached. The manufacturers of radial flow columns claim that the beds can be rewetted and reused after air ingress without the requirement to repack the column. Personal experience of this phenomenon at the 10 L scale supports this observation; however, if the chromatography system is well designed, the correct procedures followed, and operators are well trained, columns will not be run dry. Scale-up of axial columns is achieved by maintaining a constant linear velocity and bed height (the vertical height of the matrix), while increasing bed diameter, and maintaining a constant linear velocity. The issue of variable linear velocities within a radial flow chromatography column is often raised as it is a distinct difference from axial flow chromatography. It has been shown that radial flow columns can be scaled up based on constant volumetric flow rate per unit bed volume and this is recommended [13]. If the variable linear flow rates are thought to be a concern, they are amenable to scale-down experiments. Radial flow columns are scaled up linearly, thereby simplifying the scale-up process [14]. The bed depth (the distance between the inner and outer meshes) is kept constant and the height of the column increased to give the required volume of matrix.
6.3.4 Column Operating Strategy There are increasing commercial pressures in the biopharmaceutical market which are driving companies to optimize the process and processing strategies to maximize efficiency and reduce production costs. These pressures are of particular importance to commodity-type biophamaceuticals.
6.4 Operating Philosophies
137
Table 6-1. Key issues in developing column operational philosophy. Issue
Concerns
Capital cost
Aim to minimize while ensuring capital employed is used to full potential. Process equipment, utilities and environmental issues to be considered. Aim to minimize investment in matrices installed in columns and spare matrix in stores. Review effect of cycle failure with respect to lost product. Ensure complexity does not compromise operability. Potential of lost product because unable to process due to being below column loading specification. Recovery and product quality often a function of loading. Contrast: increased use of equipment makes it more reliable, while decreased use of equipment reduces risk of failure. Ensure enough contingency to cope with potential failures.
Matrix inventory Column cycles Operability Column loading specifications Equipment reliability
Most products require several different chromatographic steps to provide the required quality and the number of columns of each type required and the way they will be operated should be defined early in the design process. The aim should be to develop a strategy that maximizes the use of capital equipment, while ensuring that the process is robust and reproducible. Sufficient time should be allowed for any specific cleaning, sanitization, and maintenance work. As part of the development of the operating strategy a mass balance should be produced detailing batch sizes and production times to enable an analysis of the available options to be performed. The options range from operating one column per process step and running it once per batch to operating a number of columns many times per batch. The key concerns to be considered are shown in Table 6-1. A detailed analysis of the options while addressing the concerns outlined in Table 6 - 1 will ensure that the most efficient column operating philosophy is developed.
6.4 Operating Philosophies 6.4.1 Introduction The development and approval of operating philosophies should be an early milestone in the design process. They will impact on the design and should be in place before design work starts. These philosophies should include a description of the process along with the required hygiene levels, the boundaries between sterile and sanitary operations, the CIP procedures and frequencies, water qualities, buffer and matrix management systems, and batch sizes, and any other key issues that affect operation.
138
6 Large-scale Chromatography: Design and Operation
6.4.2 Hygienic Factors The hygienic design of chromatographic systems is key to the production of quality product. Systems must be cleaned and operated in a manner appropriate to the product and to prevent the build-up of soil and ingress of contaminants. Process hygiene must be designed into the system and therefore the relevant philosophies should be defined early in the development program. It is standard procedure to clean all systems in place before use. The process pipework and vessels should be cleaned under standard CIP conditions, commonly 0.5 M sodium hydroxide at 60°C, although the use of additional acid washes may be required to remove specific process soils. There are three primary hygienic processes to be considered, CIP, SIP, and sanitization. While developing philosophies it is important to remember that sterility is defined as the 'the complete absence of living organisms' [15]. Sterility must be validated if it is claimed and this can be onerous. The hygienic quality of systems should be appropriate to their use. Chromatography columns are difficult to operate aseptically and hence the bioburden of all feed solutions must be controlled. If buffers are going to be stored they are generally filtered, immediately after make-up, into SIPed vessels. If required, they can be filtered onto the column. The product stream from the column is generally filtered as it enters the eluate vessel.
6.4.3 Buffers Philosophies concerning buffer preparation and storage should be developed. In the laboratory, buffers are commonly prepared with laboratory reagents. This may not be appropriate at a commercial scale due to high cost and availability. At the commercial scale, the use of different grade raw materials, the use of in-line diluted buffer concentrates, and the use of constituent acids and bases should be considered. If the buffer preparation philosophies developed are different from those used to develop the process then the new philosophies should be tested in scale-down experiments to ensure they do not have an adverse impact on the process. The raw material grades used to manufacture buffers should be challenged. Significant cost savings may be available if different grades, such as food-grade, can be used. The scale of operation, the ability to achieve buffer specification, the anticipated product selling price and raw material availability will all be factors in determining the appropriate grade of raw materials. These factors will also affect the outcome of an examination into the use of buffer concentrates and acids and bases. The use of acids and bases to manufacture salts has advantages and disadvantages. The disadvantages are associated with the safety aspects of materials handling and storage. Consideration must be given to heats of reaction and dilution. The advantages arise from handling fluids, reduced costs, and potentially, a reduced number of stock raw materials.
6.5 Validation
139
The use of buffer concentrates is where stocks of buffer concentrate are prepared and diluted before use. The advantages of this are reduced tankage volumes and capital costs. The disadvantages surround process complexity associated with the dilution process. The potential corrosion effects of concentrated salts should be considered and the appropriate materials of construction specified. There are many different ways that the above philosophies can be integrated into production use and only a careful analysis of the options will provide the most practical method for a particular process.
6.4.4 Column Qualification An important part of chromatography systems is column qualification where the efficiency of packing and pressure/flow characteristics are monitored. The tests should be carried out immediately after column packing to ensure the column is suitable for use. Repeat tests can be carried out at defined intervals, if required, to ensure there is no deterioration in column performance. Efficiency of packing and the degree of resolution within the system is generally analyzed by measuring the height equivalent to a theoretical plate (HETP). A small sample of acetone, or other substance, is applied and eluted. The substance used must not bind to the column, and should be a small molecule that will not be excluded from the void volume of beads, but will be detected by the downstream detectors. The peak is then analyzed and the HETP is determined. The HETP peak can also be used to determine the column asymmetry factor which is indicative of the homogeneity of packing. Pressure/flow curves are developed to characterize the relationship between flow and pressure drop across the column to ensure that stable process flow rates are achievable.
6.5 Validation 6.5.1 Introduction Validation is defined as the ‘action of proving that any material, process, procedure, activity, system, equipment, or mechanism used in the manufacture or control can, will, and does achieve the desired and intended result(s)’ [ 151. The subject of validation has been covered in many recent articles. Readers are directed towards a series of ten biotechnology validation articles which appeared in Pharmaceutical Technology starting in February 1994 [ 16-25]. These articles are an extensive review of the subject and a suitable starting point for anyone seeking detail in this area. Further detail can be found in the book, Process Chromatography: A Guide to Validation [26]. This section will examine the broad concepts
140
6 Large-scale Chromatography: Design and Operation
of validation that should be considered during the design and start-up of a chromatography system. The general areas of validation are design qualification (DQ), installation qualification (IQ), operational qualification (OQ), and process qualification (PQ). The qualification of systems is concerned with documented verification that specifications reflect the duty required by equipment, that equipment has been installed according to specifications, that it operates according to specification, and that it performs the process duty according to specification [28]. It is mandatory to have a validated process to produce therapeutic, and most diagnostic, biological products. Hence a well-controlled process is essential. In addition, a well-controlled production process will provide a number of benefits, including reduction of rejects, improved yields and processing time, enhanced employee quality awareness, and reduced in-process testing [27]. Validation can contribute to establishing and maintaining such a system while providing the confidence that a consistent and reproducible process is being operated.
6.5.2 Design Considerations The concepts of validation should be considered early in any project and must be integrated into the project as a whole. One approach to process validation is described by Bala [29] where the key elements to successful validation are summarked. Besides the general areas of DQ, IQ, OQ, and PQ, other factors must be included. These include a master validation plan, a validation team, defined boundary testing, i.e. clear scope of what is being tested, and clear definition of the expectations of internal customers, i.e. what others expect from the validation program. The disruption and expense of retrospective validation can be avoided by spending time at the front end of the project to design a robust validation program. Design qualification is carried out during detailed design and ensures that the design and specifications meet the requirements of the process. Wherever possible, the IQ should be performed with the supplier and the acceptance of equipment should be based on satisfactory IQ. This requires writing IQ documentation before orders are placed so that the conditions of the contract are clear. This should apply to all equipment, including services and process equipment.
6.5.3 Operational Considerations When IQ is complete the next validation phase is OQ. Generally it is necessary to commission equipment before it is passivated and operationally qualified. Passivation is the chemical treatment of stainless steels to ensure that an inert and passive layer, or surface film, is formed which acts a barrier to corrosion. Commissioning seeks verification of the as-built plant performance and allows any obvious problems
6.6 Economics
141
to be addressed before mechanical completion and before entering a formal OQ validation phase. This is part of good engineering practice. When writing OQ documentation it is important to ensure that only the important parameters are qualified and that the acceptance criteria in the protocols are relevant and realistic. There is a tendency to use OQ protocols as an extensive commissioning documentation system, but this involves unnecessary work and could result in OQ deficiencies that have no effect on the process and are undesirable as they need to be addressed before any Regulatory Authority submissions are made. A similar approach should be taken for the PQ protocols, which seek to document that the process operates in the way it was intended and that the key quality parameters are within specification. The critical quality attributes, measurable characteristics that affect product suitability for use, should be identified along with the critical process parameters that could affect the critical quality attributes. These should be identified on the basis of scientific judgement that may or may not require supporting experiments [27]. While many process parameters are controlled, not all are critical process parameters and the non-critical process parameters need not be validated.
6.6 Economics 6.6.1 Introduction The production of proteins by recombinant DNA technology is a multi-million pound industry and continues to grow. Purification costs are estimated to account for up to 80 % of manufacturing costs [3] and therefore are a target for close financial control. This is particularly true in the production of commodity-type biopharmaceuticals where a price premium is inappropriate. The requirements for quality, safety, and efficacy in biopharmaceuticals are a prerequisite and attention must be paid to the economics of production within the aforementioned considerations. As world markets increase and become more competitive, the need to control production costs becomes increasingly important. An increased understanding of the factors involved in purification processes combined with improvements and new developments in downstream processing equipment will all help to cut production costs. Upstream processing is relatively well understood and it is imperative that upstream and downstream processes are developed concurrently to ensure the whole process is as efficient and economic as possible, within the regulatory constraints of the product. The need to design and develop the process with economic considerations has been well documented [6,30,31]. Much of the work that affects the economics of production is in the process development stage which should seek to control costs without compromising quality. The input of process engineers in this phase of development can be valuable in minimizing issues of scale. The design of a commercial-scale process should ensure that costs are minimized within the constraints of the process and the regulatory considerations.
142
6 Large-scale Chromatography: Design and Operation
There are few economic data in the literature, and this is understandable given the competitive nature of the market. A review of the subject is provided by Sadana and Beelaram [32].
6.6.2 General Considerations The design of any manufacturing process will seek to provide a competitive advantage, the three main sources of which are first to market, high product quality, and low product cost [3 11. The cost of capital can have a significant contribution to product costs and many manufacturing processes are designed to maximize the utilization of capital equipment. This ensures that the most efficient use of capital is employed.
6.6.3 Bioprocessing Considerations Before starting the design process, the market size and estimated sales figures, volume, and price, need to be established to allow the financial viability of the project to be assessed. The costs of production can be broken into four major groups: expenses, capital equipment, labour, and facilities [3 11. The first step in this assessment is to perform a mass balance and size key items of equipment. This allows estimations of capital equipment and facilities. The batch size, predicted recoveries, annual production target, scale of processing, degree of automation, and labour requirements are all important factors that need to assessed within the requirements of the project. Another important consideration is the lifecycle of the product which will affect the pay-back time, and hence the economics of the project. The production costs can then be estimated: these include materials, utility costs, and labour. Matrix costs can be a significant item and the life-time of matrices should be optimized and validated. These variable costs are dependent upon the process and equipment used and are not always easy to define before more detailed design has taken place. However, these costs have to be assessed to ensure that the project is financially viable. Once all the costs have been estimated they should be set up in a cost model which allows the sensitivity of the project to changing costs to be analyzed. This sensitivity analysis combined with estimated market price, estimated sales figures and corporate financial policies will allow the overall financial viability of the project to be assessed.
References
143
References [I] Ersson B., Ryden, L., Janson, J.-C., in: Protein Purification. Principles, High Resolution Methods, and Applications: Janson, J.-C., Ryden, L. (Eds.). New York: VCH Publishers, 1989; pp. 3-32. [2] Bonnerjea, J., Terras, P., in: Bioprocessing Engineering: Systems, Equipment and Fac Lyderson, B.K., D’Elia, N.A., Nelson, K.L. (Eds.). New York: John Wiley, 1994; pp. 160-185. [3] Ransohoff, T.C., Murphy, M.K., Levine, H.L., Biopharm 1990, March, 20-26. [4] Hartnett, T., in: Bioprocessing Engineering: Systems, Equipment and Facilities: Lyderson, B.K., D’Elia, N.A., Nelson, K.L. (Eds.). New York: John Wiley, 1994; pp. 416-468. [ 5 ] Sofer, G.K., Nystrom, L.-E., Process Chromatography. A Practical Guide. London: Academic Press, 1989. [6] Naveh, D., Bio Pharm 1990, 3(5), 28-36. [7] Barnfield Frej, A.-K., Hjorth, R., Hammarstrom, A., Biotechnology and Bioengineering 1994, 44, 922-929. [8] Chang, Y.K., Chase, H.A., Biotechnology and Bioengineering 1996, 49, 5 12-526. [9] Chang, Y.K., McCreath, G.E., Chase, H.A., Biotechnology and Bioengineering 1996, 48, 355-366. [lo] Wallworth, D., in: Downstream Processing of Natural Products: Verall, M.S., (Ed.) New York: John Wiley, 1996; pp. 209-221. [ I l l Gu, T., Tsai, G.-J., Tsoa, G.T., Chemical Engineering Science 1991, 46, 1279-1288. [I21 Saxena, V., Wed, A.E., Kawahata, R.T., McGregor, W.C., Chandler, M. American Laboratory News October 1987, 112-1 20. [13] Lee, W.C., Tsai, G.J., Tsao, G.T., ACS Symp. Sex 1990, 427, 104-117. [I41 Saxena, V., Subramanian, K., Saxena, S., Dunn, M., Biopharm March 1989. 1151 Guide to Good Pharmaceutical Manufacturing Practice 1983: Sharp, J.R. (Ed.). London: HMSO, 1983. [ 161 Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: February 1994, 32-34. [I71 Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: March 1994, 38-41. [ 181 Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: April 1994, 22-28. (191 Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: May 1994, 29-32. [20] Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: October 1994, 48-55. [21] Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: November 1994, 40-46. [22] Akers, J., McEntire, J., Sofer, G., Biopharm: September 1994, 41-48. [23] Akers, J., McEntire, J., Sofer, G., Pharmaceutical Technology Europe: January 1995, 34-38. (241 Bozzo, T., Pharmaceutical Technology Europe: March 1995, 40-45. [25] Kennedy, C.M., Pharmaceutical Technology Europe: May 1995, 23-30. [26] Sofer, G.K., Nystrom, L.-E., Process Chromatography. A Guide to Validation. London: Academic Press, 1991, [27] PMA QC Section, Pharmaceutical Technology Europe: January 1994, 37-42. [28] Parenteral Drug Association, Journal of Parenteral Science and Technology: May-June 1992, 46(3), 87-97. [29] Bala, G., Pharmaceutical Engineering: MayIJune 1994, 57-64. [30] Spalding, B.J., Bioflechnology: March 1991, 9, 229-234. [3 I ] Wheelwright, S.M., Protein Purification: Design and Scale up of Downstream Processing. Munich: Hanser Publishers, 1991. [32] Sadana, A,, Beelaram, A.M., Bioseparation: 1994, 4, 221-235.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
7 Radial Flow Chromatography: Developments and Application in Bioseparations Denise M. Wallworth
Chromatography on a production scale is often unattractive from a cost and throughput standpoint. Resins used in process columns shrink and swell with changing buffer strength, causing a reduction in flow through the column: subsequent loss of bed integrity results in a low cycle rate before the column has to be repacked. Current resins that have been cross-linked for additional strength go some way to solving the problem, although at the expense of some ion exchange capacity. Nonetheless, there is an extensive need for chromatographic purification in the final stages of downstream processing and a consequential need to solve some of the inherent difficulties, and this has in recent years led to some new, novel, and beneficial technologies. Scale-up to the process level using axial columns frequently leads to problems with high column backpressures resulting in low throughput and gel compression. In addition, for axial columns, scale-up involves an increase in column diameter, with the aim of maintaining the same linear flow rate between laboratory and production columns. As a result, a compromise is frequently reached between increasing the column diameter and increasing column length, often simply because the physical size of the column cannot be accommodated in a process area. A further problem in the use of axial columns at the process level is the simple inconvenience of packing and unpacking very large columns manually. For powdered resins or high potency products, handling may also be potentially hazardous. The evolution of columns which may be slurry packed has proved a great advance and has removed some of the most costly aspects of chromatography in terms of time, wastage and safety. While axial columns which can be slurry packed have recently been developed, radial flow chromatography also provides for simple packing techniques, but with the addition of several other possible benefits arising from much lower pressure drops that are observed for these columns.
146
7 Radial Flow Chromatography: Developments and Application
7.1 Radial Flow Column Design Radial flow chromatography columns utilize two concentric cylindrical porous frits which hold the resin between them (Fig. 7-1). Eluent and sample flow from the outer cylinder to the inner cylinder, across the radius of the column, which is now the effective bed height. The center solid core houses the inner frit which collects eluent and product before exiting the column. The outer cylindrical frit in radial flow columns is therefore the inlet of the column, and as such has an extremely large surface area in contact with the resin, making radial flow chromatography ideally suited to adsorptive-type separations. The flow path of a radial flow column begins at the column inlet at the top of the column (Fig. 7-2). Capillary channels take the flow rapidly to the perimeter of the column where it fills the capillary space that lies just outside the outer frit. When sample is applied to a radial flow column, the entire surface area of the outer frit is utilized, giving even distribution. Flow then proceeds through the frit and across the resin bed. After passing through the inner frit (the equivalent of the bottom frit in an axial column), the flow path proceeds down the capillary channel to the exit port of the column. The design makes radial flow columns suitable for ion-exchange, affinity, hydrophobic interaction, reversedphase, and any other adsorption-desorption type of separation. However, they are generally unsuitable for bed depth-dependent isocratic separations such as sizeexclusion chromatography, except for low-resolution molecular weight separations. A consequence of the high inlet surface area (outer frit) and short bed depth is that column backpressures are typically very low, enabling exceptionally high flow rates which may often be as high as one or two column volumes per minute. It is almost equivalent to having an axial column with the ideal geometry of a wide diameter and shallow bed depth combined with a perfect distribution head. The use of radial flow
Fig. 7-1. Laboratory-scale radial flow column.
7.1 Radial Flow Column Design
147
Fig. 7-2. Diagram of a radial flow column
technology can provide access to new opportunities in downstream chromatography whereby some of the process rate-determining steps such as equilibration and washing may be speeded up considerably. The loading of large volumes of dilute feedstock from some bioprocesses can often be the slowest stage in a process: in radial flow columns this can be speeded up considerably, enabling the process to become economically viable. In addition, for those products which have a high binding constant, it is possible to load and adsorb in the forward direction and desorb by reversing the flow in a radial flow column, thereby reducing product dilution and speeding purification further. For labile products, this technique would potentially increase the yield considerably simply by reducing time spent on the column. Examples of pressure-flow rate curves, derived from process radial flow columns packed with Sepharose CL-4B resin are shown in Fig. 7-3: in general, these are found to be significantly less steep than those observed for equivalent axial flow columns.
148
7 Radial Flow Chromatography: Developments and Application
IPressureVI. Flow on Sepharose FF Resin on Supertlo Columns a0
-
70
f0L
00
-2oL
-
50
-C
?- 40
50L
75L
t G
lWC
30 20 10
0 0.2
0.4
0.6
0.8
1 Pressure (bar)
1.25
1.75
1.e
2.25
Fig. 7-3.Pressure-flow rate curves for radial flow columns.
7.2 Slurry Packing of Radial Flow Columns One key aspect of the design of the radial flow column is that it need not be taken apart either for packing or for unpacking. While not a critical factor for laboratory columns, this becomes very important at the process level. The hazardous aspects of removing a large and heavy inlet plate and the manual removal of spent resin, along with the handling of drying solids, can be avoided. Columns be packed faster and under highly sanitary conditions, since neither resin nor column need ever be exposed. The column can be subsequently unpacked also as a slurry straight into a disposal or regeneration tank. To pack a radial flow column, the column is placed in a closed loop system as shown in Fig. 7-4(a); the packing manifold is attached to the top of the column and to the pump - generally a peristaltic pump that has both forward and reverse flow capabilities. The first stage in the process is that the column is filled with packing buffer (Fig. 7-4(a)), a buffer that has the same (or higher) concentration than that of highest ionic strength buffer to be used in the process: a high buffer concentration will ensure that the resin is packed in its fully compressed form. For some resins that have a high shrinkage, this buffer can be increased to 0.5 M higher than the highest ionic strength that the column will experience in the process. This stage of the packing process appears to be easiest if the column is operated in reverse, pumping buffer
7.2 Slurry Packing of Radial Flow Columns
149
Fig. 7-4.(a,b) Packing procedures for radial flow columns.
into the column from the outlet port. Once all the air is expelled from the column, the flow path is reversed to allow the resin slurry to be pumped into the column through the packing ports (Fig. 7-4(b)). Generally, a 25-30 % slurry of the resin in packing buffer is used [ 11 at a flow rate of one-half column volume per minute. Excess buffer exits to the buffer tank through the inlet and outlet ports. Less compressible, or more rigid bead resins generally benefit from packing at a 10 % slurry concentration and a reduced flow rate.
150
7 Radial Flow Chromatography: Developments and Application
Table 7-1. Packing pressure data for radial flow columns.
Chromatographic resin
Column packing pressure increase
Sepharose 4Ba Cellulose Agarose Trisacryl Sepharose CL4B" Sepharose Fast Flowa Fractogelb, polydextran beads, Macroprep Silica
0.5 psi (0.03 bar) 1 .O- 3.0 psi (0.06-0.2 bar) 5.0- 8.0 psi (0.34-0.55 bar) 5.0- 8.0 psi (0.34-0.55 bar) 5.0- 8.0 psi (0.34-0.55 bar) 12.0-15.0 psi (0.83-1.03 bar) 12.0-15.0 psi (0.83-1.03 bar) 25.0 psi (1.72 bar)
Trade marks:
a
Pharmacia;
ToyoSoda.
When the column is almost packed, a rapid increase in backpressure is observed. Once the pressure has reached the specified pressure for the resin shown in Table 7-1 (lowering the flow rate towards the end of the packing process can assist in determining this end point), the flow of resin slurry is stopped. These pressures have been empirically derived [l], and allow for a highly reproducible packed bed and possible automation of the column packing process. The packing manifold is removed and flushed with fresh buffer to remove all excess resin. At the same time, packing port plugs are inserted into the packing manifold inlets: these press directly onto the column bed, removing any possible dead zones. In the final stage of column packing, the bed is conditioned by backflushing the column with packing buffer. Equilibration of the column ready for the first stage of the purification process can then be carried out, generally at a flow rate of between one-third to one column volume per minute. This packing process is similar whether the column is of a laboratory, pilot, or process size, and appears to provide for a highly stable column bed [l].
7.3 Scale-up Using Radial Flow Columns For axial columns, scale-up generally involves increasing the column diameter, preserving the same linear flow rate between small and large scale columns. More often, a compromise is reached between an increase in column diameter and an increase in column length. For radial flow columns, however, the bed height is kept constant and so larger columns are simply longer, giving exactly linear scale-up. In the 50-fold scale-up of an antibody purification from cell culture fluid [ 2 ] ,the scale-up was performed by increasing flow rate, sample size, and elution volume 50-fold. Two columns, 100 mL laboratory scale and 5 L pilot scale, were packed with DEAE cellulose, and 10 mL or 500 mL samples, respectively, applied. Flow rate was also proportionately scaled up from 10 mL min-I to 5000 mL min-'. A three-step
7.3 Scale-ua Usina Radial Flow Columns
151
gradient using 60 mM, 250 mM and 700 mM NaCl in 10 mM phosphate buffer, pH 8.5, was employed in both cases and the resulting chromatograms, shown in Fig. 7-5, exhibit an almost identical profile. In a further study [3], a 10-fold scale-up of a prorenin purification on QAE and ConA Sepharose packed into a radial flow column was shown to be linear, and activity and recovery levels were maintained. However, an additional result was that the yield increased 75-fold and it was concluded that a reduction in losses caused by on-column denaturation proteolysis or irreversible binding sometimes seen on axial columns had been reduced due to the short bed depth and shorter residence times. One of the most useful aspects of scale-up in radial flow columns is that the footprint of process columns is generally small in comparison with that in conventional columns (Fig. 7-6), often a critical factor in the costing of process plants.
I
Bed Vol
I
I
Linear Scale-Ua of MAb Purification from Cell Culture
Superflo-100 Superflo-5000 Scale-up Factor
100 mL 5000 mL 50X
Sample Mass (136 mg/ml) 10 mL 500 mL 50X
Flow Rate
Elution Vol
10 mUmin 500 mumin 50X
600 mL 30 liters 50X
Separation lime 55 min 55 min 1
SUPERFLO@-100COLUMN
I 0 ackinq: >ad: ,owrate: tart Buffer: tep Gradient
30
60
mil
DEAE Cellulose lOmM Cell Culture Fluid (Murine IgG) 10mUmin. lOmM Phosphate pH 8.5 60mM, 250mM, 700mM NaCl in start buffer
Packing: Load: Flowrate: Start Buffer: Step Gradient
Fig. 7-5. MAb scale-up using radial flow columns.
DEAE Cellulose
500mL Cell Culture Fluid (Murine IgG) 5OOmUmin. lOmM Phosphate pH 8.5 60mM, 250mM, 700mM NaCl in start buffei
I
152
7 Radial Flow Chromatography: Developments and Application
Fig. 7-6.Process-scale radial flow column.
7.4 Applications The initial step in the purification of the enzyme uridine phosphorylase, us d to catalyze the synthesis of a number of pyrimidine nucleosides, involves the break-up of the cells and removal of cell debris by diafiltration. This and other enzymes from E. coli were purified without this diafiltration step with an increased yield by utilizing a radial flow column packed with Q Sepharose [4]. Because of the high flow rates possible through the radial flow column, it was possible to force cell debris through the resin without clogging the column: the backpressure remained at 0.7 bar during a 3 h loading step on a 10 L column using a flow rate of 1.3 L min-'. Recycling the feedstock through the column ensured a high product adsorption of 95 % after 3 h. No channeling was evident under these conditions and, over some 60 cycles, no dimunition in resin binding capacity or stability was observed. A recent laboratory study on factor IX purification utilized a immunoaffinity resin packed into a 50 mL laboratory-scale radial flow column and also into a 4.8 cm diameter glass axial column [ 5 ] . The antibody was monoclonal and was coupled to
7.4 Applications
153
Sepharose CL4B. After equilibration with five column volumes of buffer (10 mM magnesium chloride, 100 mM sodium chloride, 20 mM phosphate, pH 7.0), the lyophilized coagulation factor IX was loaded. Antibody capacity appeared to be identical for both columns, suggesting that radial dispersion, mass transfer, and intraparticular diffusion do not have a significant impact on immunoaffinity chromatography. An alternative route to the purification of factor IX from human plasma has utilized an un-cross-linked cellulose (Whatman DE52). In the existing process, the cryogenic precipitate is adsorbed onto the resin in a batch process because the amount of precipitate present in the supernatant prevents the use of normal column techniques. In another study [6], the DE52 was alternatively packed into a 100 mL radial flow column and 5 L of the crude unfiltered cryoprecipitate loaded onto the column. Binding was achieved efficiently and the precipitate cleared the column without causing any increase in column backpressure. A comparison of radial and axial flow columns for the purification of a glycopeptide, secreted from Myxococcus xanthus [7], utilized a histidyl-Sepharose resin. The axial (80 mL) column was loaded at 1 column volume per hour and the product eluted at the same flow rate, producing a yield of active product of 49 %. It was found to be possible to use a fivefold larger radial flow column and run the loading and elution steps at three column volumes per hour: the yield in this case was 61.2 %. Radial flow columns have been used for the processing of a crude extract of cod meal in the production of DNase [8]. In the scale-up study, yields were found to remain constant when progressing from a 250 mL radial column to a 5 L size, while the scale up from a 60 mL to a 2.5-L axial column resulted in a reduction in yield from almost 100 96 to 76 %. Increased productivity (Table 7-2) in the case of the radial flow column enabled the chromatographic step to be cost effective. It was thought that the rapid removal of proteinases from the radial flow column resulted in less degradation of DNase, thereby improving enzyme recovery. Recombitant DNA technology has enabled the expression of large quantities of eukaryotic proteins in bacteria. Since they are often formed as insoluble inclusion bodies, renaturation has to be carried out in vitro. Optimal conditions for this are at low protein concentrations, generally below 100 mg mL-', generating large volumes of dilute proteins [9]. The use of radial flow chromatography was found to be critical to the success of the purification stage since a 4.0 L sample could be processed in less than 2 h by utilizing the high flow rates possible in radial flow
Table 7-2. Production of DNase from cod meal Column type
Column volume
Flow rate (L h-l)a
Cycle time (h)
Production rate
No. of cycles
Total process time (days)
Axial column, 37 cm x 16 cm Radial column, 3.5 cm bed depth
16 L
151 (9.5 cv h-l) 420 (84 cv h-l)
1.6
5.2
52
3.4
0.19
15.0
152
1.2
a
cv, column volume.
5L
154
7 Radial Flow Chromatography: Developments and Application
columns. Two 100 mL radial flow columns were packed with ion-exchange resins QSepharose FF and S-Sepharose FF, connected in tandem and equilibrated with buffer, pH 7.0. Refolded protein was applied to the columns at a flow rate of 50 mL, min-1 and the absorbance of the eluate monitored at 280 nm. The protein was eluted using the loading buffer with 1 M NaCl added and the peak eluted from the S-Sepharose column in about 60 mL of buffer. SDS-PAGE studies showed that the Q-Sepharose column removed most of the contaminating material, while the S-Sepharose column efficiently bound the recombitant protein - in this case providing a 67-fold concentration step in less than 2 h. A subsequent polishing step utilizing a hydrophobic interaction resin completed the entire purification in less than one day. Recent work on the scale-up of the purification of large recombitant and somewhat fragile adenoviruses intended for use in human gene therapies showed that radial flow chromatography could enable their purification on an industrial scale [lo]. It provided an efficient replacement for density gradient ultracentrifugation. Several different resins - ranging from anion exchange and hydrophobic interaction through to metal-chelating types - were evaluated and each were found to have some advantages and some disadvantages. The large-scale purification of self-complimentary oligonucleotides can prove to be quite complex. Problems arise from the tendency of the nucleotides to associate by base pairing, from the presence of bases and from the formation of secondary structures [ll]. It is often necessary to use modifiers, such as urea or formaldehyde,
F
3 Column Size: 500 ml. Superflo (Sepragen Corp.) Packing: IMPAQ@RG 2020-C18 (P Q Corp.) 46 rnl/min Flow Rate: Eluant A: 0.1 % TFA in water B: Acetonitrile Gradient: 20 to 50% in 120 minutes (Linear) Pressure: 10 psi
LYE
15
30
45
60
Fig. 7-7. Reversed-phase purification of proteins.
75
90
105 Minutes
120
135
7.4 Applications
155
to the buffer to prevent self-association, but their subsequent removal can prove costly: radial flow chromatography enabled the process to be simplified and speeded UP.
While many applications in bioprocessing utilize adsorption - desorption-type cellulose or agarose-based resins, studies have also been carried out using silica media. A reversed-phase separation [ 11 employing IMPAQ 200A, 20 pm C 18 silica in a 500 mL radial flow column provided high selectivity for a range of proteins (Fig. 7-7). The low bed height of radial flow columns means that they can also be used for rapid desalting, providing an alternative to diafiltration techniques. In one study [12], bovine serum albumin was desalted on a 100 L radial flow column using isocratic elution in tap water (Fig. 7-8). A 40 g sample of BSA was dissolved in 20 L of 1 M sodium chloride and loaded onto the radial flow column packed with BioRad P-6DG. The desalting process was completed in just under 1 hour.
II
CONDITIONS 40 grns BSA In 20L 1M NaCl 1 AUFS at 254 nrn lsocralic elution In lap water U1 = BSA; 1 2 =. NaCl 5 crn I hour Load 1.8LI rnin 3 psi 5.0LI rnln 16 psl Run BloRad P6
--
0
30
MINUTES
Fig. 7-8. Desalting technique using a radial flow column.
156
7 Radial Flow Chromatography: Developments and Application
7.5 Conclusion Although radial flow chromatography has been available for just over 8 years, it has in recent years become well accepted as an efficient and rapid method of chromatographic purification. Table 7-3 provides examples of several bioprocesses that currently utilize radial flow chromatography at the process scale: all employ either ion exchange, affinity, or hydrophobic interaction modes of separation. Table 7-3. Bioproducts currently in production using radial flow chromatography. Product Monoclonal and polyclonal antibodies Interleukins Colony stimulating factors (G-CSF, GM-CSF, EGF) Interferons Blood factors, VII, IX, etc. Recombinant albumin Diagnostic enzymes, restriction enzymes Vaccines - Hepatitis B, etc. Human growth hormones - biotropin, somatotropin Growth factors
Scale (where known) 10 L 20 L 350 L 100 L 5L 100 L 50 L
References [ 11 Sepragen Corpn, Hayward, California, unpublished data.
[2] Saxena, V., Weil, A. E., BioChromatography 1987, 2 , 90-97. [3] Saxena, V., Weil, A. E., Kawahata, R.T., McGregor, W. C., Chandler, M., Znternationnl Laboratory 1988, JadFeb, 50-51. [4] Weaver, K., Chen, D., Walton, L., Elwell, L., Ray, P., Biopharm 1990, July/August. [5] Tharakan, J., Belizaire, M., J. Chromatogr 1995, 702, 191-196. [6] Hellman, P., Sepragen Corpn., Application Bulletin. [7] Data courtesy of Akohm, E., University de Compeginas, France. [8] Data courtesy of Straetkvem, K. O., University of Bergen, Norway. [9] McCartney, J. E., BioTechniques 1991, 11.5, 648-649. [lo] Huyghe, B. G., Liu, X., Sutjipto, S., Sugarman, B. J., Horn, M.T., Shepard, M., Scandella, C. J., Shabram, P., Human Gene Therapy 1995, 6, 1403-1406. [ l l ] Banerjee, A., Bose, H. S., Roy, K. B., BioTechniques 1991, 11.5, 650-651. [12] Saxena, V., Subramanian, K., Saxena, S., Dunn, M., Biopharm 1989, March.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
8 Enhanced Diffusion Chromatography and Related Sorbents for Biopurification Egisto Boschetti and John L. Coffman
8.1 Introduction Chromatography is often used for the purification of biological macromolecules, but compared with other preparative techniques used in the chemical and pharmaceutical industry it exhibits low productivity and limited throughput. The large success of chromatographic separations of proteins is, however, related to its ability to achieve high degrees of purity starting from very crude protein mixtures. This purity can be reached because chromatography separation principles are based on various parameters which characterize the proteins and are discriminant enough to make an unambiguous differentiation between the components of a mixture [ 11. During more than 25 years of development, chromatographic packing materials for preparative protein separation have been continuously improved with respect to performance. Gels based on cross-linked dextran [ 2 ] , dilute polyacrylamide gels [3], or noncross-linked agarose [4], were media proposed at the early age of preparative biochromatography. They were then progressively replaced by more and more rigid and high-performance materials. The initial driving forces of this evolution were the difficulties associated with running large columns at the preparative scale, as a result of the compressibility of existing soft media. Improvements in resolving power of the media have also been an important field of investigation. The approaches chosen were the increase of the separation efficiency (mass transfer) and the enhancement of the selectivity (thermodynamics). Improvements in the separation efficiency had been obtained by decreasing the particle size of sorbents and, since very small beads of classical gel media were too soft, new HPLC rigid, silica-based media have been developed [ 5 ] . Protein sorption capacity has also been improved over the years, but its progression was not really spectacular. The oldest ion exchangers, such as fibrous cellulose sorbents, showed sorption capacities of about 80 mg mL-' of resin [ 6 ] .This was not far from the average of about 90 mg mL-' obtained with more modern preparative media for ion-exchange separation [7]. Among the preparative solid phase media are soft gels [8]. These gels have a relatively high capacity, but suffer from their flexible nature: at even moderate flow rates, these gels deform under the backpressure in the column, increasing the overall
158
8 Enhanced Diffusion Chromatography and Biopurification
backpressure throughout the column. This backpressure limits the flow rate, and thus the productivity through the column, to relatively modest amounts. In the past 10 or 15 years, new media concepts have been developed with the aim of increasing productivity at preparative scale. Homogeneous highly cross-linked polysaccharides, such as agars, have been successfully developed for large-scale use [9]. These gels showed good sorption capacity while preserving an acceptable pressure resistance, allowing their use at large scale, but still at moderate speed. Their ability to withstand alkaline treatments was also an important feature and justified their easy adoption for preparative and industrial chromatography. Macroporous materials based on fully synthetic polymers with a large number of hydroxyl groups were then developed in order to overcome some of the limitations of homogeneous networks. Pores of these media were extremely large and resulted from an agglomeration of nodules of polymers leaving interconnected pores of a size that could reach more than 6000 A. SeparonsTM and TrisacrylsTM [lo-121 were examples these media. Elimination of most of sieving effects, increased physical stability (mechanical resistance to medium pressure), very limited shrinking and swelling when changing pH or ionic strength, stability over a wide range of pH, and non-biodegradability, were major benefits obtained with such media. Later, the name ‘perfusion chromatography’ was coined to describe the mode of action of the existing macropore media, wherein intraparticle convective transport occurs in the macropores [ 131. Adsorption-desorption kinetics and availability of interacting chemical ligands for the proteins to separate have been enhanced by the development of so-called ‘tentacular gels’ [14]. Interacting chemical functions in these media are anchored to a linear flexible polymeric chain from the three-dimensional matrix, thereby increasing the mobility of the interacting groups. In the early 1980s, another concept was introduced that combined the very good sorption capacity of soft dextran gels and the rigidity of mineral oxides such as silica [15]. This concept was developed primarily to permit liquid chromatography to be applied at very large scale for the purification of large amounts of biologicals for human therapeutic use [16]. In spite of the presence of silica which occupied a significant part of a bead volume, the binding capacity was particularly high due to the nature and flexibility of the dextran gel. Operating flow rate reached values as high as 2-3 meters per hour with backpressures below a few bar in columns of about 1000 L of sorbent. SpherodexTM, which was the name of these materials, suffered from the disadvantage of limited stability in strong alkaline solutions due to the presence of a silica skeleton. Enhanced diffusion of macromolecules within very soft hydrogels has already reported in the past [17]. The media discussed in this chapter are, however, rigid and are called HyperD”. They are based on a mineral oxide porous skeleton, whose pores are filled with a synthetic hydrogel similar to that used in soft media mentioned above. Thus, the media allows high flow rates with low pressure drop across the bed for high throughput, in addition to a high binding capacity for a high productivity. These media also have higher mass transfer rates - characteristics that were not expected from the combination of a soft synthetic hydrogel and a rigid skeleton.
8.2 Overview of Intraparticle Difjiision
159
This mass transfer mechanism is identified here as enhanced diffusion and is known under the trade name of Hyperdiffusion'. This chapter reviews the information available to date on enhanced diffusion phenomena associated with HyperD. First discussed is what is meant by enhanced diffusion, after which the microstructure of HyperD is presented. Finally, experimental evidence with typical applications supporting enhanced diffusion mechanisms are illustrated and discussed.
8.2 Overview of Intraparticle Diffusion Mass transfer - why and how solutes move - has an important impact in liquid chromatography both on qualitative and quantitative levels. Qualitatively it indicates, for instance, how fast a solute can find a binding site, and therefore it dictates how fast a chromatographic column can be loaded and operated. The faster the solute binds, the faster the column can be operated, the higher the productivity of the column. Quantitatively, mass transfer is also important. Knowing that the mass transfer is dominated by slow intraparticle diffusion allows one to calculate that the height of a theoretical plate (which is a measure of the width of an eluted peak) decreases by a factor of four if one uses a chromatographic particle with half the radius. There are four fundamental mass transfer phenomena which occur in liquid chromatography, and influence the rate at which one may load or elute. The first is axial dispersion. Axial dispersion is a combination of two mass transfer phenomena: axial diffusion, which dominates at low flow rates and when Peclet number (Eqn 1) is <-2.6 and convective diffusion, which dominates at normal flow rates when Peclet number is >-2.6. Peclet number Pe (also called reduced velocity) is commonly defined by the following equation:
where &b is the interstitial void volume, v is the interstitial velocity, dp is the particle diameter and Dfis the free solution diffusivity of the solute. Axial diffusion causes slow broadening of any concentration gradient due to diffusion in the axial direction. Convective diffusion is caused by the random flow paths taken by solutes as they travel through a bed of randomly packed particles. After a number of particles N,, approached by the following calculation, Np
= 0.3 &b I (1-&b)2 Pe
(2)
where &b is the interstitial void volume, the random motions of the solutes can be modeled as an enhanced Fickian diffusion process [18]. Thus, axial and convective dispersion processes are lumped into a single Fickian diffusion term. The primary contribution to this term is, however, the convective diffusion at the most commonly used chromatographic speeds.
160
8 Enhanced Diffusion Chromatography and Biopurification
Boundary layer mass transfer is another mass transfer mechanism in liquid chromatography. When a solute moves from the flowing mobile phase toward the particle, it encounters a relatively stagnant layer of liquid. The primary mechanism of mass transfer through this layer is no longer convection, as it was in the flowing mobile phase: it is a pure diffusion mechanism. This stagnant layer through which the solute must diffuse is relatively thin; consequently, the mass transfer through boundary layer can be often ignored. Under certain conditions, however, it dominates the mass transfer, especially during the initial part of loading a chromatographic particle, or in the case of solid diffusion at low concentration of solute and/or with large particles. Once the solute is in the particle proper, it must move to find an available site. It does so by either intraparticle diffusion, or by intraparticle convection. Typically, intraparticle diffusive mass transfer often dominates over the other forms of mass transfer at velocities commonly used in liquid chromatography. The effect of this mass transfer can be minimized with a smaller particle diameter. The height of a theoretical plate decreases in fact by a factor of four for a factor of two decrease in particle size. Unfortunately, for process liquid chromatography, this factor of two decrease in particle size is accompanied by a factor of four increase in pressure drop, which is to some extent the limiting factor at the process scale. Inside the particle, rather than simply diffusing to find a site, the solute can move to a site as a result of convection in the particle. This convection can be caused by a large pressure drop across the particle, and typically can only occur in large pores of a very small-diameter particles. A practical result of this mechanism is that the chromatographic systems can be operated more quickly. Nevertheless, the pressure drop that causes the intraparticle convection also reduces the utility of the media for process chromatography. Further, the convection is induced at very high velocities, such that diffusion is still the limiting mass transfer resistance. Once the solute reaches an adsorption site, it may take time for the actual adsorption process to occur. This adsorption process may dominate the mass transfer in some affinity systems. In most systems however, such as hydrophobic interactions and ion exchange, the sorption kinetics are very fast compared with intraparticle diffusion.
8.2.1 Pore Diffusion versus Particle Diffusion Two different mechanisms of Fickian intraparticle mass transfer are known. The first is called pore diffusion; the second goes by a variety of names: solid diffusion, particle diffusion, or surface diffusion, depending upon the application [ 191. Pore difision is diffusion of a solute within the fluid filled pore of a porous media; the solute is stationary while adsorbed to the surface of the pore. Solid diffusion is the diffusion of a solute within a homogeneous particle without real pores: in this case, the solute when sorbed continues to move. ‘Homogeneous’ here means homogeneous on a microscopic scale, not on a molecular scale. There is technically no possibility for pore diffusion in the solid diffusion mechanism because
8.2 Overview of Intraparticle Diffusion
161
there are no real pores in which the solute is free to move. Surface difuusion is the diffusion of a solute in a heterogeneous particle while adsorbed to a pore wall. Surface diffusion and pore diffusion can occur in parallel, so modeling the flux into the particle must account for both phenomena. More particularly, we define here enhanced diffusion as the diffusion in a composite media made up of a homogeneous gel in which solid diffusion occurs, and a rigid, inert, skeleton in which no diffusion occurs.
8.2.2 Effect of Pore Size Intraparticle diffusion increases the frictional drag encountered by the solute. For pore diffusion, this increased drag occurs as hydrodynamic (viscous) interaction between the diffusing solute and the stationary pore ‘wall’. For surface diffusion, this increased drag occurs due to the increased friction between the contact of the solute and the pore ‘wall’. For homogeneous diffusion in a gel, this increased drag occurs due to the increased number of contacts between the solute and the flexible gel and other effects which increase the effective viscosity of the homogenous aqueous gel solution. In the situation of pore diffusion, the drag on the solute increases with the relative size of the solute to that of the pore. This increased drag can be quantitatively predicted to yield a diffusion coefficient of a protein in a porous particle of a given pore size [20] by the following equation:
where Dpis pore diffusion, Df is free diffusion of the solute and Rp is the size of the pores. Pore size relative to the solute size can affect solute mobility and therefore column performance very significantly. Under non-binding conditions, the reduced plate height (HETP/particle diameter) can increase over three times as the solute approaches 40 % of the pore size when intraparticle diffusion dominates the mass transfer of the column. This effect is shown in Fig. 8-1, where the reduced plate height is shown as a function of reduced velocity for various solute sizes relative to pore size (A). To increase the efficiency of the column, therefore, a large pore size media should be used. With increasing pore size, however, the specific surface area of the media decreases: more of the particle is devoted to empty space, and less to adsorptive sites (see for example [60]). This decrease in specific surface area directly decreases the sorption capacity of the media which is a serious limitation for a preparative chromatographic media.
162
8 Enhanced Di&sion Chromatography and Biopurification
h = 0.4 A
160 ‘
t
5
120-
h = 0.3
80
h = 0.2 h = 0.1
40
h=O
-
O J
Pa
Fig. 8-1. Effect of linear velocity on the total height of a theoretical plate. As the size of the protein approaches the size of the pores (A increasing), diffusion is hindered and the height of the plate increases. ‘H’ is the reduced value of HETP by the size of the particle diameter ‘dp’; ‘Pe’ is the Peclet number which depends on interstitial velocity and solute diffusivity (see Eqn (1)).
8.3 Enhanced Diffusion Situation The classical mode of intraparticle mass transfer is pore diffusion where, upon adsorption to a pore wall, a solute stops diffusing. When the solute continues to move even though interacting with a sorption site, higher mass transfer can be expected compared with the pore diffusion system, since the solute is diffusing while sorbed in addition to the typical pore diffusion that would occur in the particle even in the absence of the mobile sorbed solute. Thus, intraparticle diffusion is always enhanced in the situation where sorbed solute diffuses and can move actively from one site to another. Since ‘surface diffusion’, ‘particle diffusion’ or ‘solid diffusion’ is more efficient than the classical ‘pore diffusion’, the phenomena of enhanced diffusion has been studied in many areas of diffusion in porous material. It has been studied in catalysis, in ion-exchange of small ions, and even in more recent literature for macromolecules. A short overview of these studies is reported and discussed below. Discussed also are the modeling of diffusion in gel media, as well as the effect of diffusion on the dynamic binding capacity.
8.3.1 History of Enhanced Diffusion ‘Solid’ diffusion has long been known in catalysis (see for instance references [21-231); it always results in higher flux into the particle compared with pore diffusion alone. Surface diffusion occurs when molecules interact with a surface in such a
8.3 Enhanced Diffusion Situation
163
way that they continue diffusing. Surface diffusion most often contributes significantly to the overall flux for high surface area catalysts. The larger the surface area, the more molecules are bound per unit volume. This larger quantity of molecules diffusing on the surface translates into a higher relative flux on the surface compared to that in the pore. Thus, higher capacity sorbents tend to have higher contributions from surface diffusion. This type of diffusion has been noted mostly for small molecules. For macromolecules, such as proteins, continuous diffusion is less likely while bound due to the multiple point of contact between the macromolecule and the surface. Like surface diffusion in catalysis, it has long been recognized that ions in gel-type resinous ion exchangers are not chemically bonded to fixed sites [24,25]. This is defined as ‘solid diffusion’ rather than ‘surface diffusion’ noted for catalysis, since catalysis and these type of ion exchangers differ in that catalyst structure is heterogeneous on a microscopic lengthscale, while the gel-type ion exchanger structure is homogeneous on the microscopic lengthscale (about 0.2 pm). They are, however, similar in that molecules are physico-chemically interacting with a stationary phase, yet remain mobile. The ions in gel-type ion exchangers are freely moving like a gas in a confining vessel [26], or like electrons in the conduction band of a metal: the ions are not associated with any single site, but rather all the sites in the media. The reason for this mobility is that the ion exchange sites are close enough together in three-dimensional space to have the effective electric field from each charge overlapping the neighboring charge. Thus, the counter-ion does not need to hop out of a potential energy well to another site, but diffuses in a nearly homogeneous field. The flux of small ions into these media does not depend upon the external concentration of the ion when intraparticle diffusion dominates the sorption rate [25] and when the capacity of the media does not change with external concentration. There arises from solid diffusion that the flux into the particle does not decrease when the external concentration decreases. The flux is zero order with respect to external concentration (while remaining first-order with respect to sorbed concentration). To an outside observer, the rate of mass transfer does not decrease, as expected, for dilute solutions [27]. Performing mass transfer measurements at different concentrations of applied solute is critical to understanding solid diffusion, and its relative contribution to the overall flux compared with pore diffusion. Porous media, with very large pores to allow macromolecules to enter very easily, have been investigated historically for macromolecular solid or surface diffusion, both intentionally and accidentally. It has long been known that if surface diffusion occurred for macromolecules, higher fluxes would result in better performance in terms of efficiency. Several cases of surface diffusion or solid diffusion of proteins in chromatographic media have thus been investigated. Further, the difference between solid or surface diffusion and pore diffusion is subtle, and often mathematical treatments are used for solid diffusion without realizing that the physics of the model requires the solute to be mobile even when adsorbed. One case exhibiting solid diffusion is the adsorption of bovine serum albumin into a dextran-based ion exchanger [17]. The dextran has a soft, gel-like structure similar to those used in ion-exchange of metal ions. The distance between ion exchange sites
164
8 Enhanced DifSusion Chromatography and Biopurijkation
is presumably small, with a consequent overlapping of electric fields, allowing the possibility of solid diffusion. While the experimental data were interpreted with a solid diffusion model, experiments varying applied concentration to distinguish pore diffusion and solid diffusion were not performed. Thus, it was not clearly established whether solid diffusion occurred in these systems. Further, because the gels in this system were very soft, they could not be used at the high velocities necessary to demonstrate any effective increase in mass transfer due to solid diffusion. A similar case can be mentioned for diffusion in cellulosic ion exchangers [28]. Cellulose is less homogeneous on molecular lengthscales than dextran; thus, if ‘solid diffusion’ were to occur, it would be present on a surface, and thus would qualify as surface diffusion. Again, however, the experiments distinguishing pore diffusion control and solid diffusion control were not performed. In another case, Yoshida et al. [29] did perform the necessary experiments to show that surface diffusion occurred in parallel with pore diffusion in a porous chitosan network. Based on the exclusion limit of chitosan, any mobile sorbed macromolecule would be diffusing on a surface, and not in a homogeneous gel. The researchers found that though it appeared that surface diffusion was occurring on the porous chitosan, it did not dominate the mass transfer. Thus, in the history of investigating surface diffusion of macromolecules in chromatographic media, either not enough evidence was obtained to conclusively demonstrate surface diffusion, or surface diffusion did occur, but was not the dominant phenomenon of mass transfer.
8.3.2 Modeling Enhanced Diffusion Though the term ‘Hyperdiffusion”’ has been coined only recently, surface diffusion, particle diffusion, or solid diffusion have existed for decades in the catalysis and ionexchange fields for small molecules. These modes of diffusion are easily quantifiable, as are the nature of the relationships between the flux into a media exhibiting surface diffusion and concentration gradients of solutes. Further, the nature of macromolecular diffusion in a hydrogel has been extensively studied. As shown below, the description of protein penetration in polyacrylamide gels is shown to be well modeled by the Ogston model [30], which depends upon the size of the macromolecule and the geometric configuration of the gel polymer. The relationship of the friction between a macromolecule and the hydrogel filling the pores of HyperD media using the Brinkman model [31] is also described. The model presented relates the molecular weight of a globular protein to the diffusion coefficient of the globular protein in HyperD media. The results from the model for the diffusion coefficient are then used to describe the capture of solutes in chromatographic columns according to Yoshida’s solid diffusion model [29]. Combining the diffusivity model and the capture chromatography model, a relationship between molecular weight of a globular protein and the dynamic binding capacity is presented.
8.3 Enhanced Diffusion Situation
165
8.3.2.1 Steric Hindrance (Ogston's Model) The distribution of macromolecules in a non-interacting polymer network has been modeled in a variety of ways. The earliest and most useful model was presented by Ogston in 1958 [30], which describes the distribution of macromolecules in gels based on solute size, volume fraction of polymer, and polymer radius. The Ogston model has been used successfully in many gel systems, including dextrans [32-341, polyacrylamide gels [35], composites of polyacrylamide gels and agarose [36], and composites of polyacrylamide and porous membranes [37-391. The model assumes that the solute-polymer interaction is purely steric, that the polymers are infinitely long, and that there are no solute-solute interactions. The second assumption has been modified by West [40], and the third by Fanti and Glandt [41]. Of particular interest to the polyacrylamide/ceramic composite media discussed here is the incorporation of the model that was used by Fawcett and Morris [35]. The Ogston Model was used to predict and describe the partitioning of proteins in polyacrylamide gels of 30 different compositions. At constant cross-linking, the radius of the polymer (rod) was found to be independent on the volume fraction of polymer; however, at constant volume fraction polymer, the radius of the polymer appeared to increase linearly with cross-linking weight fraction from 5.3 A to 10.2 A up to a cross-linking fraction of about 15 %. This indicates that the polymer diameter is relatively independent of polymer concentration, but that 'bundling' of polymer can occur with increasing cross-linking concentration. The experimental measurements on the polyacrylamide media which most closely resemble the hydrogel used in HyperD media discussed here (5 % cross-linking and 10-17 % polymer), indicated that the effective gel pore size is smaller than typical porous media used in protein purification.
8.3.2.2 Brinkman's model Due to these small pore sizes, it is possible to model the diffusion of the macromolecule in the gel as diffusion in a continuum. For macromolecules of the order of 20-60 A, the gel appears effectively to be continuous and exerts a frictional force on a macromolecule that can be represented by a sphere with radius R,. The gel can be modeled as a Brinkman fluid [31]. This frictional force results in a diffusivity relationship analogous to the Stokes-Einstein equation for a sphere in a Newtonian fluid [42-441:
D,
1 (4)
where D, is the diffusivity in the gel, and Df is the diffusivity of the macromolecule in free solution [3 1,42,43]. Thus, by measuring or estimating the hydraulic permeability k, and by measuring or estimating the Stokes radius of the macromolecule, the diffusion coefficient of such a molecule in the Brinkman fluid can be calculated.
166
8 Enhanced Difision Chromatography and Biopurification
For globular proteins, the Stokes radius R, is well correlated [43,45,46] to the molecular weight of the protein by (5)
R, = 0.875(MW)”3
for proteins of molecular weight MW (Daltons) in water at 25 “C. Thus, the diffusivity of a globular protein in a gel of known permeability can be calculated from an easily measured parameter: the molecular weight of the protein. The intraparticle diffusivity was measured assuming that all proteins were mobile, whether they were sorbed or not. These intraparticle diffusion coefficients Ds,are related to the gel diffusion coefficient D, by
where cP is the void volume of the skeleton, and z is the tortuosity factor. The tortuosity factor z has been found to be 2.0 for many porous chromatographic media [20]. The predicted and experimental values for diffusivity of proteins agree within a factor of two, as shown in Table 8-1. Table 8-1. Comparison between calculated and experimental protein diffusivity on Q-HyperD 80 pm particle size. Protein used
Mol. weight Do (Dalton) (cm2 s-l)
R,
(A)
D, Eqn (4) D , Eqn (6) D , experim. (cm2 s-l) (cm2 s-l) (cm2 s-l)
Bovine serum albumin Ovalbumin Lysozyme a-Lactalbumin
67 000
6.8OE-07
35.2
9.37E-09
1.22E-08
9.20E-09
45 000 14 300 15 000
7.30E-07 1.12E-06 1.06E-06
31.1 21.2 21.6
1.22E-08 1S5E-08 3.00E-08
1.56E-08 2.09E-08 3.75E-08
1.50E-08 7.90E-09 1.60E-08
The diffusivities for proteins in HyperD appear predictable to within a factor of two based on the molecular weight of the globular protein and the measured Darcy permeability of the HyperD gel. In addition, the diffusivities appear to be predictable to nearly within a factor of two based only on the molecular weight of the globular protein and the volume fraction of polymer using empirical relationships to predict the Darcy permeability [37]. This degree of agreement based on just two well-characterized parameters is satisfactory in spite of possible refinements of the presented model. A priori predictions, such as the one above, for diffusivities in complex porous media are relatively difficult; however, the experimental results are in good correlation with the model.
8.3 Enhanced Diffusion Situation
167
8.3.2.3 Yoshida’s Model for Dynamic Capacity A diffusivity for a protein of a given molecular weight in a hydrogel can be estimated by the method outlined in the above section. With knowledge of other mass transfer phenomena and a chromatographic model, such a diffusivity factor can be used to predict protein uptake in a packed bed. A variety of more or less complex models can be used; however, for the quantitative understanding of the media, due to the approximations inherent in the estimation of diffusivity, an analytical model suffices for many applications. The model used here is described by Yoshida et al. [29]; it includes both boundary layer mass transfer and intraparticle solid diffusion for an adsorption system with irreversible (rectangular) isotherm. This model improves over the models by Vermeulen et al. [19], Hiester and Vermeulen [47], Cooper [48], Cooper and Liberman [49], and Weber and Chakravorti [50] by avoiding the assumption of the constant pattern valid only at sufficiently long times. The form of the analytical solution depends upon the residence time, the length of the column, the particle size, and other factors. The form valid for parameters typically found in process liquid chromatography is
c
1 (7)
where C is the concentration of the solute in the liquid phase and C, is the inlet concentration. 6 is defined in Eqn (8) below and represents the rate of mass transfer through the boundary layer divided by the rate of mass transfer in the particle:
where Nu is the Nusslet number, which represents a dimensionless concentration gradient averaged over the surface of the particle, D, is the effective intraparticle diffusion coefficient, using the solid diffusion model, C, is the concentration of the protein applied to the column, and qo is the static capacity for the media for the applied protein. The term E in Eqn (7) represents the dimensionless position of a solute at given position in the column and 0 is a dimensionless time that compares the residence time of a solute in a given point which has traveled down an ideal column with the response time of the boundary layer around the particle. This model is satisfied for given ranges of T and 4 and also for rectangular isotherms when 6 < 5. A correlation of Nusslet number and Peclet number for packed beds has been described by Athalye et al. [51]. 8.3.2.4 Predictability of Enhanced Diffusion Diffusivity of proteins in gel media can be reasonably well predicted using the Brinkman model [31]. This predicted diffusivity is then used in Yoshida’s model
168
8 Enhanced DifSusion Chromatography and Biopurijkation
(Ov), lysozyme (Lys) and a-lactalbumin (a-Lac). Media used were Q-HyperD (Q) and S-HyperD ( S ) .
O.SE-08
I
. 1000
.
. 3000
. 5000
7000
go00 Linear veloclty (cmlh)
Fig. 8-3. Dynamic binding capacity at 10% breakthrough (DBC) versus linear velocity of S HyperD for lysozyme. Two particle size of sorbent were used: 70 pm (open circles) and 200 pm (closed-circles). Experimental points correlate predicted values (straight lines) calculated as described in the text. Binding capacity were measured in 50 mM acetate buffer, pH 4.5 using breakthrough curves. Concentration of lysozyme was 1 mg mL-’.
[29] to predict the dynamic capacities of a variety of proteins. The predictions from these models are suitable for order-of-magnitude analysis, as well as observing general trends in the behavior of macromolecules in gel media. These order-of-magnitude analyses are as accurate as the models for diffusivity and mass transfer currently used in classical porous chromatographic material. Predictability is highlighted here with a concrete example of HyperD media where the intraparticle diffusivity measured assumed that all proteins were mobile, whether they were sorbed or not. In practice, the void volume of the gel-filled matrix is measured by mercury porosimetry after drying the media. The tortuosity factor used and found for many porous chromatographic media [20] including the porous mineral oxide is about 2.0. The model discussed above which relates the diffusion coefficient of a macromole-
8.4 Chromatographic Media f o r Enhanced Diffusion
169
cule to the hydraulic permeability [31] is used to estimate a diffusion coefficient based on the size of the protein. As a general rule, predicted and experimental values for diffusivity of proteins [87,88] agree within a factor of two, as shown in Fig. 8-2. Although a priori predictions for diffusivities in complex porous media are generally quite difficult, the degree of agreement based on just two well-characterized parameters seems satisfactory. Model predictions of dynamic binding capacity were also well correlated to experimental breakthrough curves, as for instance for lysozyme at a concentration of 1 mg mL-' obtained using S-HyperD 70 pm and 200 ym material (Fig. 8-3). For these experiments, 6 ranged from 1.5 for the 70 ym at 100 cm h-I to 5.8 for the 200 pm at 10000 cm h-l. This situation has also been demonstrated using BSA on Q-HyperD material.
8.4 Chromatographic Media for Enhanced Diffusion Dynamic binding capacity, which is related to diffusion phenomena, has been investigated in several chromatography media [52-541. The significance of experimental results has been progressively understood and will in the future be used to design more adapted networks. In this section, the nature and history of gel-like chromatographic media suitable for macromolecular purification is discussed first. Then the nature of the various parts, as well as general properties of a specific composite media, are presented including the mineral porous material used as a housing and the hydrogel used to fill the pores.
8.4.1 From a Theoretical Model to Practical Media Definition of polymeric tridimensional structures intended for protein interaction involves soft flexible hydrogels resulting from a diluted swellable body. Swelling and shrinking properties of such hydrogels can vary to a large extent as represented in Fig. 8-4. Ion exchangers from soft hydrogels were made in the past by cross-linking dextran and cross-linking polyacrylamide. Both had a high or very high sorption capacity for proteins when in swollen state, but when in the presence of high ionic strength or appropriate pHs to desorb the proteins, the soft gel volumes shrunk to such small values that it was not unusual that protein inside the gel network remained trapped. As long as the hydrogel remained in its swollen state, diffusion properties were considerably enhanced and the diffusion speed was high. Shrinking and swelling phenomena, however, rendered the soft hydrogel beaded sorbents unsuitable or very difficult to use for regular chromatographic separations in packed bed columns. Chromatographic media for macromolecule separation, introduced later, were addressed just to solve the problem of softness for an easier use in closed columns. Agarose [55], cross-linked agarose [9], macroporous agarose [56], cellulose [57],
170
8 Enhanced Diffusion Chromatography and Biopurification
I 1
3 0:l
6 012
7
013
9
0.4
11 0:s
13
016
PH i.e.
Fig. 8-4. Schematic representation of volume variations of a cationic dextran hydrogel as a function of pH and ionic strength (is.).
synthetic hydrophilic polymeric material [ 121, and coated polystyrene [58], were easily usable; binding capacity was, however, partially sacrificed. All these polymers were in fact much less flexible in aqueous systems because of their high cross-linking state or their partial hydrophobic character. Based on practicability problems at preparative and large-scale chromatography, high binding capacity soft hydrogels were combined with rigid mineral porous material by introducing soft hydrogels within the rigid structure of highly porous silica [15]. This material possesses in fact a pore volume (void volume inside the bead) as high as about 1 mL per 2 mL of settled beads, or 1 mL per g of dry beads. Moreover, this material displays pores as large as 3000 8, or more, so that linear soluble polymers may be easily introduced by diffusion and cross-linked in place. These media, called Spherodex', represented the first approach for a composite organominera1 media. In spite of the substantial volume occupied by the silica matrix (about 50 % of a bead volume) the protein sorption capacity per mL of resin was exceptionally high. The charged dextran hydrogel itself showed sorption properties estimated to around 260 mg proteins per mL of hydrogel which corresponded to 130 mg mL-' of resin, even though 50 % of the beads was occupied by an inert solid mineral material. On the other hand, a binding capacity of 130 mg mL-' of resin, is a very high value when considering that regular ion exchangers have sorption capacities of about 70-80 mg mL-'.These media involving cationic dextran and porous silica were used extensively both at laboratory scale [15] as well as at large industrial scale [16] for the production of biological macromolecules. Inside each individual bead, dextran gel showed microphenomena of swelling and shrinking as seen with regular ion-exchange dextran gels; the gel did not, however, modify the mechanical properties of the composite material, which were determined by the rigid silica structure. The swelling properties of soft hydrogels inside rigid structures generate substantial osmotic pressure when in the presence of low ionic
8.4 Chromatographic Media f o r Enhanced Diffusion
17 1
strength solutions. In some extreme cases the internal pressure due to swelling microphenomena is high enough to disintegrate the rigid porous network. In further developing soft gels distributed within the structure of rigid porous material, solutions of monomers were used to fill the pores instead of the linear soluble polymers [59]. This solution, composed of small co-polymerizable monomers, was defined in such a way as to obtain an appropriate tridimensional hydrogel suitable for protein adsorption. Once the pore volume was filled, the mixture of monomers was polymerized in place, giving rise to a cross-linked hydrogel with a geometrical tridimensional shape molded by the porous void structure. This synthetic hydrogel also displays properties of swelling as dextran gels creating similar phenomena inside the rigid bead structure. In the absence of the rigid skeleton, it swells from a factor of three to a factor of 14 larger than its original polymerized constrained state (Fig. 8-5). This is due to the charge repulsion of the same charges on the polymer backbone. The charges are heavily shielded by salts and solvents before polymerization, and can approach relatively closely to one another. However, after polymerization and when the shielding is removed, the electric fields from the stationary charges begin to overlap. They repulse one another, and the hydrogel swells if it is unconstrained. Even at the highest shielding at high salt concentrations typically used in ion-exchange chromatography, the unconstrained gel is three times larger than that as polymerized.
AS
polymerized
1.0 M NaCl 1.0 M NaCI 0.1 M NaCl + 20% ethanol
DI water
Fig. 8-5. Swelling properties of S-hydrogel (S) and Q-hydrogel (Q) used for the preparation of respectively S - and Q-HyperD, in various solutions.
172
8 Enhanced Dzrision Chromatography and Biopurification
8.4.2 The Nature of Porous Rigid Material Porous material is defined here as a solid incompressible structure with a high pore volume constituted of cavities and channels of different size. When the material is dried the pores contain air or an inert gas. A well-known example of such material, in spherical or irregular shape, is porous silica. Porous silicas can be differentiated by their pore volume and their pore size. The surface area is inversely proportional to the size of the pores and proportional to the pore volume [60]. Silicas are very popular in liquid chromatography [61]. Their most popular use is reversed-phase-chromatography in organic solvents for which the silica surface is grafted with hydrophobic hydrocarbon chains [62]. Controlled pore glass [63] is a peculiar aspect of silica material, where pores are generated by a solubilization of acid-sensitive components. Other examples of rigid porous materials are porous titania [64], alumina [65], and zirconia [66] obtained by special processes contributing to the final pore volume and pore size. All of these mineral structures can easily be filled either with linear soluble polymers cross-linked in place [67] or with monomers and polymerized in place [681. The choice of a particular porous material depends on the final properties of the resulting composite material; more particularly for preparative applications a high value of pore volume is preferred since it defines the amount of active hydrogel to be introduced per single bead and therefore the final binding capacity. Figure 8-6 shows electron microscopic pictures of typical mineral porous materials.
a
b
C
Fig. 8-6.: Scanning electron microscopy pictures of various porous mineral materials used for the preparation of HyperD material: porous ceramic material composed of zirconium and calcium silicate (a); porous titania (b); porous zirconia (c). Magnification for lower pictures is 1600; upper view pictures show the pore structure magnified 14 000 times.
8.4 Chromatographic Media for Enhanced Diffusion
173
8.4.3 The Gel-Filled Porous Structures and their Preparation Various methods of pore filling have been described; this review is restricted to the methods used to prepare HyperD ion exchangers whose properties are reported in the following sections. The porous dry mineral material is selected first according to its particle size, pore size, and pore volume. To achieve the highest binding capacity the volume of hydrogel trapped inside the mineral network is as large as the pore volume. Once the pore volume is known, a solution of selected monomers is introduced into the pores of the rigid material. When the filling operation is complete, the copolymerization of the components of the solution can start using classical initiation processes. If the monomers used have already the desired chemical functionality, the resulting polymerized composite material is just washed extensively and used. Ion exchangers, for instance, are obtained with hydrophilic ionic monomers carrying weak or strong ionic groups [69] (e.g. carboxyl, sulfo, phospho, tertiary or quaternary amines). When the co-polymer obtained inside the mineral beads does not yet contain the appropriate ligands for the suited chromatographic separation, chemical coupling reactions can be followed using well-known immobilization methods. Variations of this general recipe are easily possible, for instance, by mixing various monomers with different functionalities, by varying their respective concentrations and/or proportions, and by using various rigid porous material. Chromatographic properties are essentially dependent on the nature of the hydrogel inside the porous mineral material; this latter does not significantly influence the fractionation characteristics. Figure 8-7 shows chromatographic separations obtained with quaternary amino containing sorbents where the porous mineral moiety was a ceramic made of zirconiudcalcium silicate, titania, and zirconia. The choice of
Fig. 8-7. Chromatographic separations on various HyperD-type cationic sorbents (with quaternary aminogroups) obtained using different mineral porous materials: ceramic material based on zirconiudcalcium silicate (a), titania (b), and zirconia (c). Average bead size was for all sorbents was about 40 wm. Columns: 10 mm I.D. X 50 mm; sample: cytochrome c (first peak), human transferrin (central peak) and bovine albumin (third peak) (5 mg each component); adsorption buffer: 50 mM Tris-HC1, pH 8.6; elution linear gradient up to 0.5 M sodium chloride; flow rate: 150 cm h-'.
174
8 Enhanced Difision Chromatography and Biopurifcation
Table 8-2. Main binding characteristics of a Qhydrogel prepared using porous silica beads with different pore diameters. Characteristics of porous mineral material Surface area (m2g-l)
Particle size (pm)
Pore volume (cm3g-’)
Pore size
10 25 50 100
40-60 40-60 40-60 40-60
1 1
a
1 1
Properties of the final composite material
(A)
Ionic groups (peq mL-I)
BSA capacity” (mg mL-I)
Sorption efficacy (mg BSA peq-I)
3000 1250 600 300
220 230 230 250
110 90 90 80
0.50 0.39 0.39 0.32
Measured at 10 % breakthrough
the mineral moiety depends on the final chemical stability properties; additionally it induces variations on material density, which play a dominant role in expanded bed applications (see section 8.6.3.). The rigid porous mineral network that maintains the integrity of the soft gel during swelling and shrinking does not restrict diffusion nor influence very much the dynamic capacity as indicated in Table 8-2. Silica beads with a pore diameter of 300, 600, 1200, and 3000 A filled with the same gel, did not show in fact significant differences in bovine albumin adsorption under identical experimental conditions.
8.4.4 HyperD Microstructure The hydrogel moiety of HyperD sorbents closely resembles the early gel-type ion exchangers used for small and medium molecule separation, and is very different from the large-pore media used in liquid chromatography for separating macromolecules. The microstructure of this composite material has been recently studied and elucidated by transmission electron microscopy by Kessler and Agajanian [70], as shown in Fig. 8-8. To this end, Kessler and Agajanian first fixed the material in Spurr resin for sectioning and after slicing, exposed the sample to gold-labeled albumin to highlight adsorptive sites. In unfilled material (Fig. 8-8(a)), the gold-labeled albumin was not visible, indicating that it did not bind or it bound to the mineral surface where it did not provide enough contrast for visualization. Cationic hydrogel alone (Fig. 8(b)), which is used to fill the pores of mineral skeleton in HyperD, shows as expected, a homogeneous gold labeling due to the interaction of labeled albumin with the cationic charges of the hydrogel. The mineral porous material filled with tridimensional polymerized cationic hydrogel inside or HyperD (Fig. 8-8(c)) shows extensive colloidal gold labeling within the pores of the mineral skeleton seen as black dots of uniform size which is very similar to that in Fig. 8-8(b). This demonstrates that the pores of the mineral skeleton are substantially filled with the ion-exchange hydrogel on which molecules of BSA interact strongly.
8.4 Chromatographic Media for Enhanced Diffusion
a
b
C
175
d
Fig. 8-8. Transmission electron microscopy photomicrograph of structure stained with uranyl acetate and gold-labeled bovine serum albumin. (a) Porous mineral structure without hydrogel. (b) Cationic hydrogel used to make Q-HyperD. (c) Q-HyperD where the pore volume has been filled with the cationic hydrogel. (d) Classical porous polystyrene-based cationic sorbent (sold under the tradename Pores@), where the ionic interactions occur on the surface of the pore walls. Magnification, X 25 000. (Adapted from Kessler and Agajanian [70].)
In all pictures, holes appear in the section and are the result of the ultramicrotomy knife dislodging some solid material, as also evidenced by Tanaka et al. [71] where transmission electron microscopy was used to investigate similar mineral particles. Comparatively, a cationic polymeric polystyrene-divinyl benzene porous media, sold under the tradename Porosa, was also investigated. Here, the black dots of gold-labeled albumin bound only to the surface of the pores of this sorbent (Fig. 8-8(d)) and not as in the case of HyperD where the entire pore volume of the media was filled with albumin binding sites. This surface-dependent binding situation reduces significantly the sorption capacity of this type of media.
8.4.5 Main Properties of HyperD Ion Exchangers These composite sorbents are constituted of two distinct matrices, a very rigid passive ceramic mineral structure playing only a mechanical role, and a soft active hydrogel chosen for its physico-chemical properties to comply with the needs of preparative liquid chromatography for proteins and other biological macromolecules. Sorption capacity properties have been studied and some data are reported here. Recovery, separation of protein mixtures, resolution versus protein load, and resolution versus linear velocity are also discussed below. Binding capacity is one of the dominant features of a preparative sorbent, since it determines the highest applicable sample load and thus the productivity of a column. This information is even more useful when given with flow rate variation. Binding capacities in dynamic conditions are measured with the method of frontal analysis. Breakthrough curves performed with HyperD ion exchangers showed particularly
176
8 Enhanced Difiusion Chromatography and Biopurification
500
1000
1500
2500 3000 linear velocity (crnlh)
ZOO0
Fig. 8-9. Dynamic binding capacity (DBC) variation as a function of flow rate for Q-HyperD of different particle diameters. Binding capacities were measured by breakthrough curves using a solution of bovine serum albumin at a concentration of 5 mg mL-' in 50 mM Tris-HC1 buffer, pH 8.6. Particle size are 10 pm (a), 20 p (b), and 35 p (c).
high protein sorption capacity which does not diminish very rapidly when increasing the flow rate. Figure 8-9 shows some data using particles of HyperD of different size. For small particles, it has been reported [54]that dynamic binding capacity decreased about 25 % when increasing the flow rate from 150 to 900 cm h-*. Capture efficiency was also particularly high since the difference of dynamic binding capacity between 10 % breakthrough and 50 % breakthrough was less than 25 %. Dynamic binding capacity for pore diffusion media diminishes as a function of the linear separation velocity [41,52];this is roughly governed by the following equation:
where a is the slope of the curve, R is the radius of the spherical particle constituting the sorbent, q,, is the static sorption capacity at the equilibrium, K is a constant defined by the breakthrough point, D, is the effective diffusion of the protein, &b is the intraparticle porosity, and H is the column length. The slope a of the dynamic binding capacity versus flow rate curve is of high importance, but more importantly it has to be associated with a higher static binding capacity to obtain the highest level of productivity at preparative scale. Although binding capacity depends mostly on the chemical structure of a sorbent, a number of parameters affecting resolution and dynamic binding capacity are easily adjustable by the user; they can also be synergistically combined to optimize operational conditions (see Table 8 -3). However, the most important media characteristics impacting both resolution and dynamic binding capacity are dependent on the media properties originally defined. Mass transfer, selectivity, accessibility, and density of interacting ligands are of primary importance to enhance resolution properties. Small particle size symmetri-
8.4 Chromatographic Media for Enhanced Diffusion
177
Table 8-3. Solid media characteristics and operating parameters influencing liquid chromatographic separations. Selected main parameters
Modulation by the user
Fixed by the manufacturer
Efficiency
Flow rate Temperature Column length Loading volume Solute concentration Column packing
Particle size Particle size distribution Pore size and distribution Non-specific binding
Selectivity factor
Mobile phase composition pH, ionic strength Temperature Elution gradient (slope, shape) Column geometry
Polymer structure Ligand selection
Sorption capacity
Mobile phase composition Temperature Flow rate Solute concentration
Number of binding sites Accessibility of binding sites Diffusivity properties Pore size and distribution
cally and narrowly distributed, polymeric surfaces with reduced non-specific binding, are additional parameters to consider. Enhancing dynamic binding capacity is dependent on ligand availability, high polymer flexibility, and diffusive properties of the polymeric network. All HyperD ion exchangers were designed with respect to these parameters. Sorption kinetics of dilute proteins on various ion exchangers in suspension in a buffer showed that the phenomena are not only dependent on molecular weight of the solutes, but also on the nature and the structure of the sorbent. To elucidate this point, experiments were performed in batch mode measuring the optical density diminution of the protein solution under stirring, when a cationic sorbent was added. Three proteins of similar isoelectric point, but different molecular size, were used (thyroglobulin, MW 670 kDa, PI 4.7; BSA, MW 67 kDa, PI 4.9; betalactoglobulin, MW 36 kDa, PI .5.1), and the volume of the sorbent added corresponded to twice the total binding capacity for the protein in the buffered solution. Figure 8-10 shows that the decrease of the optical density of the solution is more rapid with Q-HyperD compared with other sorbents. According to this series of experiments it appeared that the ability to capture proteins from a diluted solution is enhanced with Q-HyperD. Recovery has also been investigated. Using the peak collection method with an amount of injected protein corresponding to 2 % of the total sorption capacity, the protein recovery is above 85-90% with HyperD ion exchangers. All recovery data compared favorably with the other regular high-performance ion exchangers for protein separation [54,72]. A number of chromatographic separations were performed using HyperD with various functionalities. Most of them compared favorably with modern ion-exchange
178
a
4
8 Enhanced Diffusion Chromatography and Biopurifcation
b
,1
100
I00
100
-
L
300 time (sec) 500
300 time (sec)
300 time (sec)
500
500
Fig. 8-10. Uptake of proteins of different molecular weight on three polycationic sorbents Q-Agarose beads M; Q-HyperD (Q-Polystyrene beads with very large pores M; M). Determinations were made in batch mode measuring the decrease of absorbance at 280 nm of the supernatant. Amount of sorbent was 1 mL; amount of protein was 50% of the total sorption capacity of each sorbent; all determinations were performed in 25 mL of 50 mM Tris-HC1 buffer, pH 8.6 at room temperature.
media of similar particle size. This work reported elsewhere [54,72,73] demonstrated the excellent separation capabilities of HyperD in classical physico-chemical conditions and using a linear salt gradient to elute the proteins. Figure 8-11 shows some chromatographic separation of HyperD ion exchangers. The resolution of a mixture of proteins is influenced not only by the ion-exchange performance of the solid media and its particle size, but also by the amount of injected proteins. Resolution versus loading is actually another important parameter
8.4 Chromatographic Media f o r Enhanced Dirusion
179
b
d
i
Fig. 8-11. Examples of separation of macromolecules using Q-HyperD columns. (a) Betalactoglobulin digest using trypsin for 36 h. (b) Cell culture supernatant containing a monoclonal antibody (mAb). (c) Protein mixture composed of cytochrome c (cyt c), myoglobin (myo), human transferrin (tr), ovalbumin (ov) and human albumin (hal). (d) Mixture of oligonucleotides obtained by chemical synthesis. All separations have been performed in the same conditions. Columns: 4.6 mm I.D. X 100 mm; adsorption buffer: 50 mM Tris-HC1, pH 8.6; elution linear gradient up to 0.5 M sodium chloride: flow rate: 150 cm h-’,
in preparative ion-exchange chromatography, since this ratio impacts to a large extent the limits of productivity of a preparative ion-exchange material. Experimental data from Q and S-HyperD demonstrated that this parameter is not influenced significantly by the column loading due to the large binding capacity of the HyperD media. It has been demonstrated that, for instance, when the loading was increased from 100 to 700 pg for a column of 1.7 mL the resolution remained almost constant [541. Another important point is the variation of the resolution when increasing the flow rate of a column. Here again, and within certain limits that are linked essentially to the diffusion time of the solute into the ion exchange gel, the resolution variation of HyperD compared favorably with more classical media (Fig. 8-12).
180
8 Enhanced Diffusion Chromatography and Biopurifcation resolution
I
3
'11 1
0.5
2
10
20
30 40 protein load (rnglmL)
100
300 400 500 flow rate flow rate (crnlh)
200
Fig. 8-12. Effect of load (a) and flow rate (b) on protein resolution on Q-HyperD. Determinations were made using human transferrin and bovine serum albumin as protein models. Columns: 4.6 I.D. X 100 mm, 50 mM Tris-HC1, pH 8.6. Elution gradients were obtained by increasing linearly sodium chloride concentration up to 0.5 M. Linear velocity for (a) was 150 cm h-I; load for (b) was 20 pg of protein mixture. Loading experiments were performed on 20 pm particles (open circles) and 35 pm particles (closed circles); flow rate experiments were performed using 10 pm particles.
8.5 Experimental Evidence of Enhanced Diffusion in
HyperD For practical preparative and large-scale applications, the qualification and quantification of the mechanisms of mass transfer in chromatography media is of particular help. In effect, a qualitatively different mass transfer mechanism may lead to specific and useful application processes, whereas quantification of the mass transfer mechanism is crucial for the design of large-scale chromatographic equipment. The study of mass transfer has been done for HyperD ion exchangers. This study demonstrated that these media exhibited solid diffusion. This assessment is shown by the analysis of batch uptake curves, shallow bed experiments, and breakthrough curves, which allow development of specific applications.
8.5.1 Macromolecular Conformation of Hydrogel Network In HyperD sorption, properties are dependent on the nature of the soft gel existing within the pore structure. If the gel were not constrained by the rigid skeleton it would swell significantly from the initial polymerized state in the conditions normally used in liquid chromatography (see Fig. 8 - 5 ) . In practice, however, the rigid mineral ceramic skeleton does not allow swelling and the charges in the ionexchange gel are therefore forced to interact strongly with one another.
8.5 Evidence of Enhanced Diffusion in HyperD
18 1
Thus, under the conditions normally found in ion-exchange chromatography, the charges in HyperD are closer together than they would otherwise prefer. The electric fields around the charges overlap and the possibility for solid diffusion occurs. This situation is similar to that in gel-like ion exchangers used for small molecules, where solid diffusion has been noted for many years [24, 251. As noted above, however, solid diffusion was not noted for macromolecules in these media, since the polymer chains were so close together so as to exclude macromolecules from the matrix. In HyperD ion exchangers, proteins must significantly penetrate the chromatographic particle, since binding capacity per unit volume is very high compared with the case of simple adsorption occurring only on the external available surface area. Interaction mechanism for the basis of binding capacity has been investigated. By examining the elution volume of macromolecules on HyperD under non-binding conditions, the distribution or the penetration of the macromolecule into the gel network has been determined. The extent of penetration is related to the molecular dimensions of the pores or spaces between the polymer chains in the gel. If the elution volume of the species of interest is simply the void volume of the column, then no significant penetration of the species into the particles has occurred, and the molecular dimensions of the ‘pore’ are similar to that of the molecular yardstick used. The void volume of the column &b was calculated by several methods and reported in Table 8 - 4. The gravimetrically determined extra-particle void volume compared favorably with the void volume determined by blue dextran also reported in Table 8-4, meaning that blue dextran does not appreciably enter the particle. The exclusion of blue dextran is not surprising: due to its large size, it is commonly used to measure the extra-particle void volume of chromatographic media under non-binding conditions. The void volume values for the smaller macromolecules such as 40 kDa dextran with a Stokes radius of 45 1$, and BSA, a 65 kDa globular protein with Stokes radius of 35 1$ [43], are very close to the elution volume of blue dextran and that of the gravimetrically determined extra-particle void volume, meaning they are also significantly excluded from the HyperD particle under non-binding conditions. This contrasts with typical porous media, which allow penetration of macromolecules into the pores even under non-binding conditions. This effect is shown in data taken by Horvath and Farnen [74], where the elution volume for variously Table 8-4. Comparative determination of void volumes of various molecules in non binding conditions. Solid media
Salt conc. (mM)
Void volume fraction (gravimetric)
Void volume fraction (Blue dextran)
Void volume fraction (Dextran T40)
Void volume fraction (BSA)
Q-HyperD F Q-HyperD M S-HyperD F S-HyperD M
175 185
0.49 0.48 0.49 0.50
ND ND 0.42
0.50 0.46 0.43 0.50
0.48 0.47 0.43
150 150
ND, not determined.
0.48
0.50
182
8 Enhanced Diffusion Chromatography and Biopurification
sized molecules are compared on two media: a gigaporous media and the gel-filled HyperD media. The HyperD particle significantly, but not completely, excludes relatively small proteins such as ribonuclease A, compared with other porous material, which does not significantly exclude even transfenin and polyvinylpyrrolidone. From these exclusion data, it can be concluded that the gel in HyperD has small effective pores, smaller than or equal to the size of a typical protein such as BSA. This is consistent with the gel concentration which is typically around 10-17% when compared with 5-7 % used for gel electrophoresis. These gel pores therefore sterically exclude proteins under non-binding conditions. From the transmission electron microscopy studies described above (section 8.4.4), the gel in the HyperD particle is pictured to be homogenous on at least the lengthscale of the resolution of the transmission electron microscope. These elution volume experiments, however, allow a much smaller yardstick to be used: the exclusion data above indicate a homogeneity on a lengthscale of 50 A or less. Further the hydrodynamic (Darcy) permeability [37,75,76], indicates that the gel is homogeneous to a much smaller lengthscale, on the order of 25 A. The distance therefore between polymer chains of the soft hydrogel is very small, of the order of 25 A, in other words, on the order of the lengthscale for proteins of interest, whose diameter is roughly 50 A. At any time, a protein diffusing into the HyperD hydrogel is in contact with more than one active polymer strand in any of three dimensions. In addition, based on a simple cubic distribution of charges, a typical HyperD particle with 200-300 pmol of ionic charges per mL of bed (which corresponds to 400600 pmol mL-’ of hydrogel), the distance between charges is roughly 15 A. A typical globular protein, upon penetrating the gel, would therefore be touching several ionexchange sites at one time. These ion-exchange sites are not arranged on a surface, as for typical porous media; they are arranged rather in three dimensions; the entire surface of the macromolecule is therefore likely to be interacting with many ionexchange sites.
Table 8-5. Determination of elution volumes as % of the column volume for various molecules of different size. Solute
Sodium chloride Leucinehaline Ribonuclease Myoglobin Bovine serum albumin Transferrin Pol yvin ylp yrrolidone a
MW (Da) 48 131/117 14000 17 800 67 000 80 000 40 000
Elution volume (% of column volume) Porous PST mediaa
S-HyperD
73 73 70 70 69 69 66
72 72 50 48 45 44 44
PST, polystyrene. This media had a pore size of about 4000 A and was from Polymer Labs Ltd
(UK).
8.5 Evidence of Enhanced Diffusion in HyperD
183
Table 8-6. Measured and calculated Darcy permeability for Q and S-HyperD ion exchangers. Sorbent type
Q-HyperD S-HyperD
k x 1014 (experimental)
k x 1014 (correlation)
0.342-0.679 0.150-0.169
0.5 18 0.200
Under non-binding conditions, macromolecules such as proteins are sterically excluded by repulsive interactions that result from the decrease in entropy of a polymer molecule that enters the gel [77]. Polymeric hydrogels constituting HyperD ion exchangers, however, are very soft, swellable bodies where proteins diffuse under binding conditions. All these facts and the high binding capacity of HyperD ion exchangers demonstrate that proteins clearly have an unambiguous access to the interior of the HyperD particle under binding conditions, even though they do not have significant access under non-binding conditions. It is believed that the repulsive steric interaction that keeps macromolecules out of the gel under non-binding conditions is overcome by the strong attractive forces between the ionic macromolecule and the ionic polymeric hydrogel. This force is believed to be strong enough to push the polymer chains aside and allow macromolecular penetration based on the high flexibility of polymer chains [78,79].
8.5.2 Batch Uptake Curves Intraparticle mass transfer is often quantified by the rate of mass taken up by the media in a stirred bath [78-861. If enhanced diffusion is the main mass transfer mechanism, then the boundary layer mass transfer resistance will dominate the uptake rate when the dimensionless group 6 (Eqn (lo), 6 is the rate of mass transfer through the boundary layer divided by the rate of mass transfer in the particle), is less than unity [29,87,88]:
where kf is the boundary layer mass transfer coefficient, R is the particle radius, C, represents the external protein concentration, D,is the intraparticle diffusion, and qo is the static capacity of the media for the investigated protein. At very low concentrations of protein where 6 is less than unity, the uptake rate depends upon the extraparticle boundary layer mass transfer. At high concentrations, when 6 is greater than unity, the uptake rate depends primarily upon the intraparticle diffusion. Thus, one can independently measure the boundary layer mass transfer coefficient and the intraparticle diffusivity by judicious choice of applied concentrations.
184
8 Enhanced Difision Chromatography and Biopurifcation
Table 8-7. Diffusivity and boundary layer mass transfer coefficient for bovine serum albumin (BSA), ovalbumin and a-lactalbumin in Q-HyperD and lysozyme in S-HyperD. ~~
Protein
Solid phase media
Intraparticle diffusion coefficient D , (cm2 s-l)
BSA BSA Ovalbumin a-Lactalbumin Lysozyme
Q-HyperD F Q-HyperD M Q-HyperD M Q-HyperD M S-HyperD M
9.2 9.2 15.0 16.0 1.0
x x x x x
10-9 10-9 10-9 10-9 10-9
Boundary layer mass transf. coeff. kf (cm s-l) 2.5 x 10-3 1.4 x 10-3 1.9 x 10-3 2.5 x 10-3 1.0 x 10-3
Fernadez and Carta [87] and Weaver and Carta [88] have recently measured the rate of mass uptake into HyperD media as a function of time in stirred bath for test proteins. The experimental results were analyzed to determine the boundary layer mass transfer coefficient kf at the lowest concentration and to determine the intraparticle diffusivity D, at the highest concentrations. These measurements (see Table 8-7) were then used to predict the behavior at intermediate concentrations, assuming that these mass transfer coefficients do not change as a function of concentration. Using a model including the mass action equilibrium model, boundary layer mass transfer, and particle diffusion, uptake rates of intermediate concentrations were predicted. The results from these predictions matched very well the experimental data for intermediate and extreme concentrations over a variation of 20-fold in concentration with a single pair of mass transfer parameters D, and kf , as shown in Fig. 8-13.
Co
.
200
~
--, 400
600
-
2 mglmL
. - . 800
1000 1200 time (sec)
Fig. 8-13. Batch uptake of ovalbumin solutions at different concentrations ‘Co’ on Q-HyperD media. Buffer used was 50 mM Tris-HC1, pH 8.6. Dots represent the experimental determinations, continuous lines are the numerical solutions of the kinetic model using an intraparticle diffusion coefficient D, of 1.5e-08 cm2 s-I and a boundary layer mass transfer coefficient kf of 1.9e-03 cm s-I. (Adapted from Fernandez and Carta [87] and Weaver and Carta [MI.)
8.5 Evidence of Enhanced Diffusion in HyperD
185
Alternatively, however, the pore diffusion model was used to fit the data (though not to predict the data). The results of such a fit also matched the uptake rates as a function of time as well as the above particle diffusion model. The intraparticle diffusivity so measured does, however, vary inversely with the applied concentration. The intraparticle diffusivity decreases by a factor of 20 as the concentration was increased by a factor of 20. The pore diffusion model does not therefore allow a single set of intraparticle diffusivities to describe the mass transfer as a function of concentration; the pore diffusion model requires a different diffusivity to describe the uptake at each different concentration. Moreover, the value of pore diffusivity required to fit the data at the lowest concentration, was significantly higher than the free diffusion coefficient. In the context of the experiments, enhanced diffusion phenomena were demonstrated to be the dominant mechanisms of mass transfer in HyperD ion exchangers.
8.5.3 Shallow Bed Experiments Shallow bed experiments allow one to determine the type of intraparticle mass transfer mechanism based on the pressure drop across each particle in the packed bed. The higher the pressure drop, the higher the hypothesized convective flow into the particle, the higher the rate of mass transfer into the particle. Fernandez and Carta [87] and Weaver and Carta [88] examined the uptake rates of the HyperD media by passing solutions of proteins through a bed of few particles thick for various lengths of time and at various velocities (and therefore various pressure drops across the particle). At the end of a duration of exposure, the protein was eluted from the column, and the amount of protein on the column was calculated by integrating the area of the eluted protein peak. As shown in Fig. 8-14, the authors [88] found the uptake rate of BSA into Q HyperD is independent of the velocity through the shallow bed, indicating that convective flow does not contribute to mass transfer at these velocities. They also found good agreement between the data and of the prediction from the numerical model for uptake into particles, including boundary layer mass transport and particle diffusion. The intraparticle diffusion coefficient was taken from the batch uptake experiment, and the boundary layer mass transfer coefficient was given by Carberry [89], which is valid for flow in packed beds only. The boundary layer mass transfer term in packed beds is in fact different than that in stirred baths, since the term depends largely on the velocity distribution around the particle. The mass transfer terms, therefore, were predicted completely a priori for the shallow bed experiment. This a priori predictions by Carta, Fernandez, and Weaver, shown in Fig. 8-14, agree very well with the experimental data at all velocities, concentrations, and for the two particle sizes. When the boundary layer mass transfer resistance dominates the uptake kinetics, the uptake rate increases with increasing velocity through the column. At higher velocities, the boundary layer resistance decreases, since the stagnant layer around the particle becomes smaller and the solute can diffuse more quickly from the
186
8 Enhanced Diffusion Chromatography and Biopurifcation
100
300
500
700
900
trme (sec)
Fig. 8-14. Shallow bed uptake of BSA solutions on Q-HyperD. Buffer used was 50 mM Tris-HC1, 4240 ), pH 8.6. Experiments were performed in three different linear velocities: 2550 (H (M and )5940 cm h-’ (M Continuous ). lines are the numerical solutions of the kinetic model using an intraparticle diffusion coefficient D, of 1.5e-08 cm2 s-l and a boundary layer mass transfer coefficient kfof 1.9e-03 cm s-l. (Adapted from Fernandez and Carta [87] and Weaver and Carta [88].)
well-mixed interstices through the stagnant layer, to the particle surface. This increase in uptake rate is also reflected in the correlation by Carberry [89], and is thus shown in the model predictions by Fernandez et al. [87]. The uptake rate in the shallow bed when intraparticle diffusion dominates (6 greater than unity) is invariant to velocity for the 2 mg mL-’ uptake rate. At all velocities studied by Carta, Fernandez, and Weaver, even up to 5940 cm h-I, the uptake rate remains independent of flow rate. Further, the uptake rate agrees with that predicted for a purely enhanced diffusion or solid diffusion phenomenon. The shallow bed can also be compared with the stirred batch to reach the same conclusions. In comparing the shallow bed and the stirred batch uptake experiments, care must be taken to match the conditions of uptake such that intraparticle diffusion dominates the separation (6 > l ) , If boundary layer mass transfer dominates the separation, when 6 is less than unity, then the uptake rate will depend on flow rate in the shallow bed, since the boundary layer resistance becomes smaller at higher flow rates. Since the protein uptake is similar and because the intraparticle convective flow is impossible in a stirred bath (no pressure drop can occur in stirred bath), Carta, Fernandez, and Weaver concluded that the mechanism of protein binding is not related to convective flow.
8.5 Evidence of Enhanced Dijfusion in HyperD
187
8.5.4 Breakthrough Curves The breakthrough curve of a protein loaded onto a column is also indicative of the mass transfer in the packed bed. The breakthrough curve is influenced by various factors such as axial dispersion, boundary layer mass transfer, and intraparticle diffusion. The poorer the mass transfer, the shallower the breakthrough curve, the less the effective dynamic capacity of the column.
8.5.4.1 Effect of Velocity and Particle Size Typically, the faster a column is operated, the poorer the efficiency performance of the column. When loading a column at high velocity, the quickly moving solute front does not have time to equilbrate with the freshly encountered media. As a result, the portion of the solute that does not equilibrate with the sorbent continues to move down the column, albeit at a reduced concentration. The faster-moving solute translates into a breakthrough of the solute at less than the static capacity of the media. Thus, the dynamic binding capacity is lower than the static binding capacity. Further, the particle size plays a significant role in the efficiency of solute capture. The smaller the particle size, typically the higher the dynamic binding capacity. The higher efficiency is due to the smaller diffusion lengths a solute must travel in the particle to reach a site, and hence to reach equilibrium. This can quantitatively be seen in the slope of dynamic binding capacity versus linear velocity (see Eqn (9)), in which the dynamic binding capacity versus the column velocity increases with roughly the square of the particle size. Thus, the dynamic binding capacity of a porous material is a strong inverse function of particle size. A decrease in dynamic binding capacity as a function of velocity is noted for HyperD media, even with enhanced diffusion. The dynamic binding capacity is not, however, as low as the case of typical pore diffusion. The dynamic binding capacities for HyperD 50 pm and 70 pm particles versus other porous media from 20 pm to 90 pm are shown in Fig. 8-15. That the capacity of the 70 pm particle exceeds that of the 20 pm particle at all velocities is indicative of a peculiar intraparticle mass transfer mechanism. Since the slope of these lines depends upon the square of the particle size (for the same intraparticle diffusion), the 70 pm particle should have a slope 12 times that of the 20 pm particle. This is not the case for HyperD, indicating a significantly higher intrinsic rate of mass transfer in the larger particle, a result of an enhanced diffusion mechanism.
188
8 Enhanced Diffusion Chromatography and Biopurification
2 140E
-
3 l 120.
a
E
(J
a
100-
I000
3000 4000 velocity (cmlh)
2000
Fig. 8-15.Effect of linear velocity on the dynamic binding capacity (DBC) for lysozyme of various polyanionic sorbents (sulfonate groups). ‘a’ and ‘b’ are S-HyperD of respectively 50 and 70 pm average bead size; ‘c’ and ‘d’ are two polystyrene-based sorbents of respectively 20 and 30 pm particle size; ‘e’ is an agarose-based sorbent of 90 pm average particle size. DBC were determined by frontal analysis using 50 mM acetate buffer, pH 4.5, at room temperature.
8.5.4.2 Effect of Solute Concentration Elucidation of the dominant form of mass transfer necessitates examining the particle uptake under varying applied solute concentrations. In a pore diffusion regime, the dynamic binding capacity may remain the same or decrease when the concentration of the solute decreases. Since HyperD ion exchangers do not appear to behave in a pore diffusion regime (see previous sections), the adsorption of dilute solutes can be enhanced particularly at high flow rates. A comparison study between S-HyperD and a porous media carrying similar sulfonate groups demonstrated that the binding capacity of S-HyperD increased with dilute solutions, especially at high flow rates for large (immunoglobulin G ) and small (lysozyme) proteins. This was not the case with classical media where pore diffusion dominates (Fig. 8-16).
600
1000
1500
2000
2600
Linear velocity (cdh)
Fig. 8-16. Effect of protein concentration on dynamic binding capacity (DBC) at high linear velocities on S-HyperD. When the concentration decreased from 10 (closed circles) to 1 mg mL-I (open circles), the DBC was increased particularly at very high linear velocities. All determinations were performed using breakthrough curves.
8.6 Benefits and Applications of Enhanced Diffusion.
189
These results indicate that the mass transfer mechanism in HyperD materials is different than that found in typical porous material, for which dynamic binding capacity remains the same or decreases when more dilute solutes are applied.
8.6 Benefits and Applications of Enhanced Diffusion. The mass transfer mechanism in HyperD media is relatively new and somewhat unusual for macromolecules. This phenomenon presents therefore some interesting features to exploit for the purification of biological macromolecules. For instance, because of enhanced diffusion, the dynamic binding capacity of these media can be greater for dilute solutions than for concentrated solutions. An application of this phenomena is the capture of proteins directly from very dilute feed stocks. Also, because of the increased efficiency in capture for highly diffusive particles, larger particles can be used for effective capture. Similarly, large dense particles can also be very advantageously used in a expanded bed mode.
8.6.1 Protein Capture with Dilute Solutions Chromatographic media displaying enhanced diffusion properties have unique advantages for protein capture from very dilute solutions, since the dynamic binding capacity can increase at low concentrations. The increased performance for dilute solutions has significant advantages in the bioprocessing industry. Most biological materials are in fact expressed at low concentration in cell supernatants. Current processes in monoclonal antibodies production, for instance, typically start with dilute amounts in fermentation broth; the concentration of immunoglobulins is typically between 40 and 1000 pg mL-'. In the case of antibodies, a classical capture approach is based on the use of immobilized Protein A achieving purities of about 75-95 %. Binding capacity of immobilized Protein A, however, decreases significantly when the monoclonal antibody concentration is below 1 mg mL-', therefore a preliminary treatment of the feed stock is required. Moreover, Protein A media are about 10 times more expensive than ionexchange media. Cation exchange CM-HyperD, for instance has a number of advantages over Protein A media: it adsorbs immunoglobulins G directly from diluted feed stocks without any ionic strength adjustment, suppressing preliminary operations of concentration and diafiltration; the load level is higher than for Protein A media, and the purity is frequently higher than 80 %. Polycationic HyperD with high ionic charge density are also very useful to deal with dilute solutions, eliminating preliminary steps such as concentration by ultrafiltration or precipitation. In the case of monoclonal separation, a Q-HyperD column is used first when pH adjustments of the feed stock induce some protein precipitation
190
8 Enhanced Diffusion Chromatography and Biopurifcation
mAb
/
f non adsorbed proteins
0
55
110
165
220
L 275
mL
Fig. 8-17.Separation of a monoclonal antibody directly from a crude cell culture supernatant on CM-HyperD. The pH of the cell culture was first adjusted to 4.5 by dropwise addition of 1 M HCl and then directly applied to a column equilibrated with a 50 mM acetate buffer, pH 4.5, containing 150 mM sodium chloride. Elution was performed using 1 M sodium chloride. Collected IgG fraction was then analyzed by high-performance gel filtration (see insert) where the purity of IgG was estimated higher than 85 % (TSK column SW 3000, 5 pm particles, 7.8 mm I.D. X 300 rnm long, equilibrated with a phosphate-buffered saline; loading was 50 p1 of IgG fraction and separation performed at 0.7 mL min-’). ‘a’ is the gel filtration analysis of the feed stock; ‘b’ is the eluted IgG fraction from CM-HyperD.
saving losses in yield. Antibodies are recovered in the flowthrough and impurities are bound on the solid phase. As shown in Fig. 8-17, CM HyperD selectively and reversibly binds the monoclonal antibody, while the other media tested bind also some impurities which, to some extent, co-elute under the same conditions. Therefore antibodies are separated in a single step and their purity can be as high as the one obtained using immobilized Protein A. This capture step is of particular interest since crude very dilute material can be used directly for protein capture, minimizing yield losses and improving process economics. Highly substituted CM-HyperD generally provides binding capacities for antibodies higher than 35 mg mL-’ at relatively high ionic strengths corresponding to the cell culture supernatant which prevent preliminary dilution.
8.6.2 Capture with Large Particles Due to high mass transfer in HyperD particles, especially evidenced with dilute solutions, flow velocities of thousands of centimeters per hour can be effectively adopted for loading. These speeds, which are generally too high for many process level applications due to the pressure drop in the chromatographic packing, can be advanta-
8.6 Benefits and Applications of Enhanced Diffusion.
500
1000
191
1500 2000 linear velocity (cmlh)
Fig. 8-18. Demonstration of the ability of large beads to capture efficiently immunoglobulins G. Black dots are experimental results and continuous lines are theoretical predictions (see section 8.3). ‘a’ is a curve relative to S-HyperD of 200 pm average bead size; ‘b’ represents the behavior of an agarose-based media of 90 pm particle size. All determinations were made using a 5 mg mL-I IgG solution in a low concentration acetate buffer, pH 4.5.
geously used with large particle HyperD ion-exchange media. Very low pressure drops (typically below 1.5 bar) are generated with large particles, and dynamic binding capacity is maintained at levels close to those for smaller particles. As an example, S-HyperD 200 pm particles were tested for the purpose of binding dilute solutions of 1 mg mL-l IgG and compared with an identical column of classical beads of cross-linked agarose. These columns were subjected to various flow rates to determine the dynamic capacity as a function of linear velocity of applied protein. The results are shown in Fig. 8-18, where the 200 pm S-HyperD particles show a higher dynamic capacity than the 90 pm S-Sepharose Fast FlowTMparticles up to 2000 cm h-l. The pressure drop through this column at 2000 cm h-’ is less than 1 bar, which is an acceptable pressure in most production columns. Also shown in the same figure is the comparison of the experimental data with the theoretical development shown in section 8.3.2.4. The good agreement between the prediction and the experimental data indicates the applicability of this model for predicting the behavior of various particle sizes.
8.6.3 Capture with Dense Particles Fluid bed or expanded bed processes have been described for the separation of biologicals from crude feed stocks [90-931. Contrary to classical packed beds, fluidiza-
192
8 Enhanced Diffusion Chromatography and Biopurification
tion of beads provides a practical option to process very crude material containing particles in suspension such as protein aggregates or cell debris. In this separation mode, microbeads are lifted inside a column by an upward liquid stream generated by buffers and sample solutions. In a fluidized bed, the particles leave larger empty zones between beads where the feed stock passes through, rather than packed beds where only small empty zones exist between beads. Various bead criteria are taken into consideration when designing the configuration of fluidized bed devices in which the bed expansion ratio is generally between two and five times that of packed bed. The size and the geometric shape of the sorbent beads influence the dynamic fluid bed performance to significant extent. Richardson and Zaki [94] demonstrated, for instance, that in a turbulent situation, if the sphericity of the particles decreases, the value of the exponent of the 'porosity' of the fluid bed increases. Moreover, dynamic binding capacity and mass transfer properties of sorbent beads impact the overall productivity. The behavior of the solid phase also depends on particle size distribution and its density [95] which strongly contribute to create a fluidized bed. Using beads with relatively low density the only possibility to use high flow rates, by maintaining constant the expansion factor, is to increase the bead diameter. This rule is illustrated on the following equations.
where U is the fluidization velocity, locity given by Stokes' law:
E
is the bed expansion and Ut is the terminal ve-
where p and pap are respectively the density of the liquid and the apparent density of the solid particles in the liquid phase, g is the gravity constant, d p is the particle diameter, and ,u is the dynamic viscosity of the liquid phase. The index n of Eqn ( 1 1 ) is a value obtained as follows:
n = (4.45+ 18 d p l D ) ( U f d p p / p ) 4 ' 1
(13)
where D is the bed diameter. This is valid only for:
Although this approach is logical as stated by the equations above, an increase of the particle size would unavoidably induce a higher mass transfer resistance. The time spent by the total diffusion inside the solid particle is in fact proportional to the square of the diameter of the bead. The use of smaller-diameter beads would decrease the intraparticle diffusion resistance, but to compensate the willingness of small beads to be lifted at low upward flow, increased intrinsic density must be
8.6 Benefits and Applications of Enhanced Diffusion.
193
used. With HyperD all fluid bed requirements described above are easily overcome by choosing dense mineral porous materials such as silicon oxides, titanium oxides, zirconium oxides, and any other mixture of intermediate density. Table 8 - 8 gives a general picture of some appropriate mineral oxides that could be chosen to obtain dense or very dense HyperD sorbents for fluid bed devices. With dense or very dense material the particle size can be maintained relatively small, even at high flow rates without the danger of generating an unacceptably high expansion factor. High flow rates can be utilized with relatively small, dense particles favorable for rapid diffusion times. HyperD ion exchange media are particularly suited for fluidized beds because of the enhanced diffusion mechanism which favors capture of macromolecules (see sections 8.3.2 and 8.5). Figure 8-19 represents the influence of bead diameter on the flow rate needed to obtain a fixed expansion of the bed for various materials of different densities. Speed related to particle size and to solid phase density are taken into consideration for all calculations of the productivity. For instance, a particle with a density of 1.2 and a particle diameter of about 190 pm produces a bed expanded by a factor of two at a linear velocity of 300 cm h-', To reach a linear flow velocity of 1000 cm h-' (for higher productivity) the diameter of such beads should be as large as 400 pm. From this representation it seems for example that zirconium-oxide based beads outperform lighter beads in term of productivity in all cases. The enhanced diffusion mechanism allows the capture of dilute solutions of material, which is essential to effective use of fluidized bed media. Since the HyperD gel can be placed in any porous material without loss of capacity or selectivity, the gel can be placed into dense porous materials, which allow smaller particles to be used at higher loading velocities in fluidized beds. These smaller particles consequently lead to higher efficiency in both loading and elution. Effective applications of such fluid beds with dense HyperD are those related to the selective capture of crude and dilute biologicals such as monoclonal antibodies.
Table 8-8. Mineral porous dense material suitable for the preparation of fluid bed solid phase sorbents. Mineral oxide
Molecular weight of metal ion
Density"
Silica Doped silicab Alumina Titania Zirconia Tantalum oxide Hafnium oxide
28.09 Not applicable 26.98 47.90 91.22 180.95 178.49
1.6-2.1 1.8-2.5 2.3-3.1 2.6-3.5 3.6-4.8 4.9-6.6 5.8-7.8
a
Density range is given for porous material where pores represent about 20-40 % of the total volume. Doping elements are zirconium, aluminum, titanium, calcium, magnesium.
194
8 Enhanced DifSusion Chromatography and Biopurification
so
I50
260
310
450
bead diameter (pm)
Fig. 8-19. Influence of the bead size on the linear velocity for a fluid bed expansion factor of two (HMO= 2). The apparent density and the particle size of a solid material (p) impacts very largely on the linear speed for a given bed expansion.
8.7 Conclusion Sorption capacity properties of solid phase media for liquid chromatography depend to a large extent on diffusion mechanisms of solutes. Diffusion is generally hindered by the porous network due to too small pores or repulsion forces. Enhanced diffusion properties of soft hydrogels cannot be exploited unless the hydrogels are constrained within the rigid structure of rigid porous materials. Enhanced diffusion on gels described in this chapter impacts the speed of chromatographic separation as a result of a more rapid interaction kinetics. Enhanced diffusion also permits application of the media where regular sorbents are of limited performance, such as adsorption on fluid beds with dense material to support rapid flow rates, and effective adsorption of very dilute protein solutions.
Acknowledgments The Authors thank Dr P. Girot, Dr L. Guerrier, Dr A. Schwarz and Dr N. Voute for providing experimental data on preparation of solid phase material and their extensive comparative evaluations. Many thanks also to Dr A. Schwarz for providing suggestions and experimental data on monoclonal antibody separation as well as for the very valuable comments during the preparation of this manuscript. We also thank Steve L. Kessler for his critical review and comments on the manuscript. Sepharose Fast Flow@is a trademark of Pharmacia, Uppsala, Sweden. Poros@is a trademark of PerSeptive Biosystems, Framingham, MA, USA. HyperD@, Spherodex@,Trisacryl@are trademarks of BioSepra Inc, Marlborough, MA, USA.
Abbreviations and Symbols
195
Abbreviations and Symbols A
Angstrom bovine serum albumin concentration of the solute in the liquid phase C CM carboxy methyl inlet concentration of the solute C0 diameter of column bed D Da Daltons dynamic binding capacity DBC effective diffusivity D, diffusivity in free solution Df diffusivity in a gel D, particle diameter dP pore diffusion DP intraparticle diffusion D* gravity constant g column length H height equivalent to a theoretical plate HETP boundary layer mass transfer coefficient kf MW molecular weight Nusslet number Nu = PeclCt number Pe = isoelectric point PI = static binding capacity 40 = radius of a bead particle R = size of the pores RP = Stokes, radius R, = fluidization velocity U = terminal velocity u, = interstitial velocity V = curve slope of dynamic binding capacity versus flow rate a = rate of mass transfer through the boundary layer over the rate of mass 6 transfer in the particle = bed expansion E = interstitial volume Eb = pore volume of empty rigid porous material EP = dynamic viscosity cc = size of the molecule relative to the size of the pore /I = relative residence time of the solute in a given point of a column B = density of the liquid phase P = apparent density of the solid particles in the liquid phase Pap = tortuosity factor 7 . = relative position of the solute in a column 5 BSA
196
8 Enhanced Diffusion Chromatography and Biopur9cation
References [ l ] Wisniewski, R., Boschetti, E., Jungbauer, A,, in: Biotechnology and Biopharmaceutical Manufacturing, Processing and Preservation, Avis, K. E., Wu, V. L. (Eds). Buffalo Grow, IL: Interphann Press Inc., 1996; Vol. 2, pp. 61-198. [2] Joustra, M. K., Prot Biol Fluid 1966, 14, 553-561. [3] Gressel, G., Robards, W., J Chromatogr 1975, 144, 455-467. [4] Boschetti, E., Tixier, R., Uriel, J., Biochimie 1972, 54, 439-444. [5] Hofstee, B.H. J., Biochem Biophys Res Commun 1975, 63, 618-627. [6] Peterson, E. A., Sober, H. A,, J Am Chem Soc 1956, 78, 751-756. [7] Boschetti, E., J Chromatogr 1994, 658, 207-235. [8] Male, C., Methods Med Res 1970, 12, 221-226. 191 Porath, J., Laas, T., Janson, J. C., J Chromatogr 1975, 103, 49-62. [lo] Mikes, O., Strop, P., Zbrozek, J., J Chromatogr 1976, 119, 339-354 [ l l ] Turkova, J., Seifertova, A., J Chromatogr 1978, 148, 293-297 [12] Girot, P., Boschetti, E., J Chromatogr 1981, 213, 389-396. [13] Afeyan, N., Gordon, M., Mazsarof, M., Varady, L., Fulton, S. P., Yang, Y. B., Regnier, F. E., J Chromatogr 1990, 519, 1-29. [14] Muller, W., J Chromatogr 1990, 510, 133-142. [15] Tayot, J. L., Tardy, M., Gattel, P., Plan, R., Roumiantzeff, M., in: Chromatography of Biological Polymers, Epton R. (Ed.), Chichester: Ellis Horwood, 1978; pp. 95-115. [16] Veron, J.L., Gattel, P., Foumier, P., Grandgeorges, M., in: Biotechnology of Blood Proteins, Rivat, C., Stoltz, J. F. (Eds.), London: John Lilbley, Eurotext, 1993; pp.183-188. [17] Tsou, H. S . Graham, E.E., AZChE J 1985, 31, 1959-1966. [18] Han, N. W., Bhakta, J., Carbonell, R.G., AIChE J 1985, 31, 277-288. [19] Vermeulen, T., LeVan, M. D., Hiester, N. K., Klein, G., in: Perry’s Chemical Engineering Handbook, 1984, Section 16; pp. 16-23. [20] Coffman, J. L., PhD Thesis, University of Wisconsin, 1994. [21] Carslaw, H. S . , Jaeger, J. C., in: Conduction of Heat in Solids, Oxford: Clarendon Press, 1959 pp. 23-24. [22] Satterfield M., in: Mass Transfer in Heterogeneous Catalysis, Cambridge, MA: MIT Press, 1970; pp. 47-54. [23] Smith, J.M., in: Chemical Engineering Kinetics, New York: McGraw-Hill, 1970; pp. 419423. [24] Heymann, E., O’Donnel, I. J., J Colloid Sci 1949, 4, 405-416. [25] Helfferich, F., in: Zon Exchange, New York: McGraw-Hill, 1962, reprint University Microfilm International, Ann Arbor, MI, pp. 255, 303, 305. [26] Tanaka, T., Sci Am 1981, 224, 124-137. [27] Aris, R., Ind Eng Chem Fund 1983, 22, 151-152. [28] Graham, E.E., Fook, C.F., AIChE J 1982, 28, 245-253. [29] Yoshida, H., Yoshikawa, M., Kataoka, T., AIChE J 1994, 40, 2034-2044. [30] Ogston, A. G., Trans Faraday Soc 1958, 54, 1754-1757. [31] Brinkman, H. C., Appl Sci Res 1947, 1 , 27-34. [32] Granath, K.A., Flodin, P., Makromol Chem 1961, 48, 160-168. [33] Laurent, T. C., Killander, J. A., J Chromatogr 1964, 14, 317-330. [34] Poitevin, E., Wahl, P., Biophys Chem 1988, 31, 247-258. [35] Fawcett, J. S., Morris, C. J. 0.R., Separation Sci 1966, I , 9-26. [36] Moussaoui, M., Benlyas, M., Wahl, P., J Chromatogr 1991, 558, 71-80. [37] Kapur, V., Transport in Polymer/Gel Modified Micropores, PhD Thesis, Camegie Mellon University; Pittsburg, 1995. [38] Tong, J., Anderson, J. L., Biophysical J 1996, 70, 1505-1513. [39] Tong, J., Anderson, J. L., in preparation.
References
197
[40] West, R., Biopolymers 1987, 26, 343-350. [41] Fanti, L. A., Glandt, E. E., J Colloid Interface Sci 1990, 135, 385-404. [42] Hwells, I. D., J Fluid Mec 1974, 64, 449-475. [43] Phillips, R. J., Deen, W. M., Brady, J. F., AIChE J 1989, 35, 1761-1769. [44] Phillips, R. J., Deen, W. M., Brady, J. F., J Colloid Interf Sci 1990, 139, 363-373. [45] Tyn, M. T., Gusek, T. W., Biotechnol Bioeng 1990, 35, 227-338. [46] Young, M. E., Carroad, P. A,, Bell, R. L., Biotecknol Bioeng 1980, 22, 947-955. [47] Hiester, N. K., Vermeulen, T., Chem Eng Prog 1952, 48, 505-516. [48] Cooper, R. S., Ind Eng Chem Fund 1965, 4 , 308-313. [49] Cooper, R. S., Liberman, D. A., Ind Eng Ckem Fund 1970, 9, 620-623. [50] Weber, T. W., Chakravorti, R. K., AICkE J 1974, 20, 228-238. [51] Athalye, A.M., Gibs, S. J., Lightfoot, E. N., J Ckromatogr 1992, 589, 71-85. (521 Janson, J. C., Hedman, P., Biotecknol Prog 1987, 3, 9-13. [53] Liapis, A. I., McCoy, M. A,, J Chromatogr 1992, 599, 87-103. [54] Horvath, J., Boschetti, E., Guerrier, L., Cooke, N., J Ckromatogr 1994, 679, 11-22. [55] Hjerten, S . , Biockirn Biophys Acta 1964, 79, 393-398. [56] Gustavsson, P. E., Larsson, P. 0..J Ckromatogr 1996, 734, 231-240. [57] Peska, J., Stamberg, J., Hradil, J., Ilawsky, M., J Ckrornatogr 1976, 125, 455-469. [58] Ugelstad, J., Mork, P.C., Adv Colloids lnterf Sci 1980, 13, 101-140. [59] Boschetti, E., Girot, P., US Patent, 5,268,097, 1993. [60] Wojicik, A,, Kwietniewski, L., J Ckromatogr 1988, 435, 55-65. [61] Unger, K. K., Packings and Stationary Phases. New York and Basel: M., Dekker Inc., 1990. [62] Vansant, E. F., Van der Voort, P., Vrancken, K. C., Characterization and Chemical Modification of the Silica Surface. Amsterdam: Elsevier, 1995. [63] Tarutani, T., J Ckromatogr 1984, 313, 33-44. [64] Trudinger, U., Muller, G., Unger, K. K., J Chrornatogr 1990, 535, 111-125. [65] Billiet, H., Laurent, C., de Galan, L., Trends Anal Chem 1985, 4 , 100-105. [66] Rigney, M.P., Weber, T. P., Cam, P. W., J Chromatogr 1989, 484, 273-287. [67] Zhou, F. L., Muller, D., Santarelli, X., Jozefonvicz, J., J Ckromatogr 1989, 476, 195-203. [68] Boschetti, E., Girot, P., US Patent, 5,268,097, 1993. [69] Boschetti, E., Girot, P., Guerrier, L., Santambien, P., J Biockem Biopkys Meth 1996, 31, 15-25. [70] Kessler, S. B., Agajanian, J., in preparation. [71] Tanaka, N., Hashidzume, K., Araki, M., Tsuchiya, H., Okuno, A., Iwaguchi, K., Ohnishi, S . , Takai, N., J Ckromatogr 1988, 448, 95-108. [72] Boschetti, E., Guerrier, L., Girot, P., Horvath, J., J Ckromatogr 1995, 664, 225-231. [73] Bouhallab, S . , Henry, G., Boschetti, E., J Ckrornatogr 1996, 724, 137-145. [74] Horvath, Cs., Farnen, D., HPLC95, San Diego, 1995. 1751 Kapur, V., Charkoudian, J. D., Kessler, S. B., Anderson, J. L., Ind Eng Chem Res 1996, 35, 3 179-3 185. [76] Ferry, J. D., Chem Revs 1936, 118, 373-378. [77] Haggerty, L., Sugarman, J. H., Prud’homme, R. K., Polymer 1988, 29, 1058-1063. [78] Sellen, D. B., British Polymer J 1986, 18, 28-31. [79] Sellen, B. D., J Polymer Sci 1987, 25, 699-716. [80] Colton, C. K., Satterfield, C. N., Lai, C. J., AICkE J 1973, 19, 628-635. [81] Awe, B. H., Liapis, A. I., AICkE J 1987, 33, 179-193. [82] Tsou, H.S., Graham, E.E., AICkE J 1985, 31, 1959-1966. [83] Graham, E. E., Dranoff, J. S . , I&EC 1982, 21, 360-365. [84] Pinto, N. G., Graham, E. E., Reactive Polymers 1987, 5 , 49-53. [85] McCoy, M. A., Liapis, A. I., J Ckromarogr 1991, 548, 25-60. [86] Chase, H.A., Chem Engr Sci 1984, 39, 1099-1125. [87] Fernandez, M.A., Carta, G., J Ckrornatogr 1996, 746, 185-198. [88] Weaver, L. E. Jr, Carta, G., Biotech Progr 1996, 12, 342-355.
198
8 Enhanced Diffusion Chromatography and Biopur$cation
[89] Carberry, J. J., AIChE J 1960, 6, 460-469. [90] Draeger, N. M., Chase, H. A., Bioseparation 1991, 2, 67-80. [91] Spence, C., Schaffer, C. A., Kessler, S . , Bailon, P., Biomed Chromatogr 1994, 8, 236-241. [92] Suding, A., Tomusiak, M., Paper No 61, ACS National Meeting, Denver, CO, 1993. [93] Wells, C. M., Lyddiatt, A., Patel, K., in Separations for Biotechnology; Verrall, M. S., Hudson, M. J. (Eds.), Chichester: Ellis Honvood, 1987; pp. 217-224. [94] Richardson, J. F., Zaki, W. N., Trans Instn Chem Eng 1954, 32, 35-53. [95] Foscolo, P. U., Gibilaro, L. G., Chem Eng Sci 1987, 4 2 , 1489-1500.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
9 Expanded Bed Adsorption Chromatography Rolf Hjorth, Patrik Leijon, Ann-Kristin Barnfield Frej and Christina Jagersten
9.1 Introduction Production of recombinant proteins is almost exclusively done in systems like bacteria, yeast, or mammalian cells, which in addition to the target protein, produce a large amount of biomass. The production level of protein in such systems is in some cases relatively low, resulting in large volumes necessary to produce the desired mass of protein. The procedures used to purify proteins for therapeutic or diagnostic use are based on different chromatographic techniques in packed beds that are not designed to handle cell containing feedstocks in large volumes. Unit operations that remove particulate material and reduce the feedstock volume are therefore very important stages in almost every production process for a recombinant protein.
9.1.1 Clarification and Recovery Techniques The traditional techniques used to remove cells and/or cell debris are centrifugation and microfiltration [ 11. To reduce the feedstock volume, these techniques have to be combined with or followed by an adsorption step. The centrifuges in current use are well suited to industrial processes and are able to handle volumetric flows exceeding 1000 L h-I in a completely contained environment, which is crucial when handling genetically modified organisms. Since the efficiency of a centrifugation process is dependent on particle size, density differences between the particles and the surrounding liquid, and viscosity of the feedstock [ 2 ] ,it is sometimes difficult to obtain a completely particle-free liquid. This is particularly the case when dealing with small cells like E. coli, or cell debris, where the clearance value for particles is expected to be 99 to 99.9 % [3]. It was recently shown that centrifugation was not efficient enough to clarify a cell culture broth for direct application onto a chromatographic column [4]. Another drawback of using centrifuges is that they are relatively expensive, resulting in high capital costs. An alternative to centrifugation for clarification of particle containing feedstocks is microfiltration [ 1,2]. Microfiltration is usually carried out in crossflow filtration
200
9 Expanded Bed Adsorption Chromatography
mode [5,6] which substantially improves the process. Fluxes in the range of 50 to 1000 L m-2 h-' have been reported [ 6 ] ;however, these fluxes are dependent on the composition of the feedstock. Fluxes in the range 25-85 L M-* h-I have been reported for cell harvesting of E. coli at different cell densities [ 5 ] , this is below the suggested economic limit for the crossflow filtration process, which is set to around 100 L m-* h-' [7]. One of the major advantages using crossflow filtration is the relative ease with which the process can be operated as a contained system. This is facilitated by the absence of moving parts, compared with a centrifuge. The fouling of the microfiltration membranes is still an unsolved problem, although some of the problems associated with fouling can be minimized by careful design of the operating parameters [6]. Neither centrifugation nor microfiltration has any significant ability to reduce the volume of the feedstock and thereby increase product concentration. A reduction in volume is usually obtained by including an adsorptive step in the recovery process. Adsorption in the presence of cells or cell debris has to be done in batch mode. Batch operations are rather inefficient processes and are consequently not often used for production of biopharmaceuticals. Attempts have been made to design continuous batch adsorption processes [8]; however, such processes have not to our knowledge been used at larger scales. Another approach to increasing the efficiency of batch adsorption is the use of submicron ion-exchange particles which are later removed from the cell homogenate by centrifugation [9].
9.1.2 Fluidized Beds A variety of fluidized bed systems have been introduced to facilitate the initial particulate removing steps while at the same time recovering the desired product. In a fluidized bed a particulate adsorbent in a column is forced to rise from its settled state by introducing an upward liquid flow into the column. In the fluidized state the bed voidage increases, thereby creating enough space in-between the adsorbent particles to let cells and cell debris pass unhindered through the bed as shown in Fig. 9-1. By choosing adsorbent particles that are able to bind the product, it is possible simultaneously to recover the protein of interest and to remove cells and cell debris. This should make the use of fluidized beds very attractive for the recovery of various biopharmaceuticals. However, until recently, few applications have been presented on the industrial use of fluidized beds, mainly due to the lack of suitable adsorbents. Early applications on fluidized beds describe the direct recovery of antibiotics from cell suspensions. In 1958, Bartel et al. [lo] described the recovery of streptomycin at large scale. The process used a cation exchanger which was scaled to handle up to 15 m3 h-' in a 1.2-m diameter column. A more recent example in the same field is the recovery of immunomycin from a culture of Streptomyces hygroscopicus [ 111. The immunomycin was adsorbed to a polystyrene-divinylbenzene particle in a methanol/water mixture. Although the initial product concentration was low, the product was recovered at a very high yield from feedstocks of 14000 L. Despite the successful use of fluidized beds for direct recovery of low-molecular weight compounds, the technique had not been used for recovery of proteins pro-
9. I Introduction
20 1
Fig. 9-1. The principle of expanded beds. The increased voidage between the adsorbent particles allows cells and cell debris to pass unhindered through the bed.
duced by recombinant organisms until very recently, the main reason being the lack of adsorbents that are compatible with proteins and can be used efficiently in fluidized beds. Traditional chromatographic adsorbents designed for use in packed beds are compatible with proteins but most of these adsorbents do not have the physical properties necessary to work effectively in fluidized beds. An additional factor that has limited the use of fluidized beds for protein recovery is the relatively high degree of mixing, which lowers the efficiency of the adsorption process. Several attempts have been made to develop fluidized beds so that mixing in the bed is reduced. One approach used is the introduction of segmented beds [12]. By sectioning a fluidized bed column into smaller compartments, back-mixing can be reduced and adsorption efficiency increased. A more commonly used approach is the magnetically stabilized fluidized bed (MSFB) [13-151. In a MSFB, back-mixing is reduced through the use of a magnetically susceptible adsorbent. By applying a magnetic field outside the fluidized bed column, the bed can be stabilized. Several types of magnetically susceptible adsorbents have been used in this kind of application, such as alginate/magnetite, polystyrene-divinylbenzene/magnetite, and agarose/ magnetite. In most applications, MSFBs are operated at laboratory scale only, and to our knowledge there is no such system in use for pilot or production scale. The most obvious reason is the relatively complicated equipment needed for MSFB. Furthermore, the magnetic fields necessary to keep large MSFBs in efficient operation generate a considerable amount of heat which has to be removed from the system.
9.1.3 Expanded Bed Adsorption Chase and Draeger [ 16-1 81 recently introduced the concept of expanded bed adsorption. (In general terms, an expanded bed describes a bed in which the bed voidage is larger than in a packed or sedimented bed. Expanded beds are sometimes considered
202
9 Expanded Bed Adsorption Chromatography
an integral part of the field of fluidized bed technology; some authors do not differ between the expressions. A detailed description of the characteristics of expanded beds versus fluidized beds will be given in section 9.2.) By generating a pressure drop over the inlet of a column using a sintered glass filter it was possible to obtain almost perfect plug-flow in the bed and thereby minimize the back-mixing. The bed voidage in the expanded mode increased to an extent that allowed cells to pass through unhindered, while the protein bound to the adsorbent. The results obtained clearly showed the usability of expanded bed adsorption in processes where cell containing feedstocks are handled. A review of the early developments of expanded beds for protein recovery has recently been published [19]. As mentioned earlier, the use of expanded beds for recovery of proteins has been hampered by the lack of suitable adsorbents. For recovery of low-molecular weight compounds, adsorbents with hydrophobic properties have been used [10,11]. Such materials are less useful for protein recovery. As a result, a large number of different types of adsorbents have been tested for use in expanded bed adsorption. Examples of materials tested include composites like cellulose/silica [20], dextradsilica [21], traditional chromatographic matrices made of agarose [ 161 or silica [22], adsorbents made of glass [23,24], and more recently hydrophilized perfluorocarbon material [25]. These adsorbents are of limited use in expanded bed adsorption for a variety of reasons. Conventional agarose-based adsorbents do not have the sedimentation properties for an efficient process in terms productivity, although they are well suited for protein recovery. Material containing amorphous silica cannot be used due to its limited stability under the harsh conditions used for cleaning and sanitization in downstream processing [26]. The use of porous glass material is hampered by the sometimes low protein binding capacity of such matrices [24]. The described perfluorocarbons show some promising properties, but protein binding capacity is still poor due to the low porosity of the beads [25]. The development of adsorbents has therefore been a crucial factor in the successful implementation of expanded bed adsorption in downstream processing. With the development of an agarose/crystalline quartz composite it was possible to obtain an adsorbent, STREAMLINE@, which has the sedimentation properties, the chemical stability, and the biocompability needed for an efficient expanded bed adsorption process [27-291.
9.2 Theoretical Background for Expanded Bed Adsorption 9.2.1 Hydrodynamics of Expanded Beds As was indicated earlier, the terms expanded bed and fluidized bed are used differently by various authors. Indeed, one may argue that all fluidized beds are expanded (relative to packed beds) and all expanded beds where particles are supported by frictional forces caused by fluid flow rather than mechanically by other particles resting beneath are, by definition, fluidized. The term ‘expanded beds’ was used to emphasize the fact that fluidized beds may behave very differently [17,30]. Below,
9.2 Theoretical Background f o r Expanded Bed Adsorption
203
we will use the term ‘expanded bed’ to describe fluidized beds exhibiting reduced back-mixing due to stratification, having an increased number of mass transfer units (plates). The increased number of mass transfer units achieved in an expanded bed results in better utilization of the adsorbent’s binding capacity. It should be noted however that reduced back-mixing is not only dependent on the design of the adsorbent to achieve its desired stratification, but also the flow distribution in the column. The flow from the distribution system should be uniform in magnitude as well as in direction, i.e. there must not be any eddies, jets, etc. It follows that columns for expanded bed adsorption have a different design from fluidized bed reactors, e.g. those used for mammalian cell culture.
9.2.2 Distribution System The need for a specially designed liquid distribution system for expanded beds derives from the low pressure drop over the bed. Usually, the flow through an evenly packed bed generates such a high pressure drop over the bed that it can assist the distributor in producing plug-flow through the column. Since the pressure drop over a fully expanded bed is much less than the pressure drop over the packed bed, the distributor in expanded bed columns must produce a plug-flow itself. Consequently, it is necessary to build in an additional pressure drop into the distribution system [3 11. One common approach to generate a pressure drop is to use perforated plates. Besides generating a pressure drop, a distributor has to direct the flow in a vertical direction only. Any flow in a radial direction inside the bed will cause turbulence that propagates throughout the column. Once a disturbance has entered the bed it is impossible to stop its propagation. Shear stresses associated with flow constrictions also require consideration when designing distributors. If any shear-sensitive molecules are subjected to high shear stresses in constricted flow, a degradation of the molecules can occur. The shear stresses should therefore be kept to a minimum in order to reduce the risk of molecular degradation. Practically, and from experience, the shear stresses in a distributor where laminar flow prevails should be in the range 1-30 N m-2. The shear stress for laminar flow is given by the expression:
where T is the shear stress, p is the fluid viscosity, and y is the shear rate (velocity gradient) which for a perforated plate is given by the relationship: U
y = -8 -
D
where u is the fluid velocity through the holes of the perforated plate, and D is the diameter of the holes.
204
9 Expanded Bed Adsorption Chromatography
Another function of the distribution system is to prevent the adsorbent from leaving the column. This is usually achieved by using a metal screednet mounted on the upper side of the distributor, facing the adsorbent. The choice of net is a critical issue. It must have a mesh size that allows particulate-containing material like fermentation broths to pass through it and yet at the same time confine the adsorbent to the column. It is also important that adsorbent beads are not trapped in the net, If this should happen, the beads will block the net and restrict particulate flowthrough. The distribution systems in both the bottom and the top of the column have essentially the same design, the only difference being that the adapter at the top should be movable and its distribution system used for elution in packed bed mode only. It therefore has a lower pressure drop than the distributor in the bottom. Finally, the distribution system must have a sanitary design. This means no dead zones, i.e. stagnant zones, where cellskell debris can accumulate. Once these accumulations of cells have formed it is difficult to get rid of them. Any cell residues that have not been washed out of the column must be considered as possible contaminants for the next production cycle.
9.2.3 Adsorbent The main concern when designing adsorbents for packed bed chromatography has always been to maximize the resolving power and the adsorption capacity, while maintaining good flow properties. Since absorbents in expanded beds are not in a fixed position as in packed beds, the adsorbent dispersion phase also needs consideration, in addition to the other parameters mentioned. The effect of the dispersion of the adsorbents together with the mass transport mechanisms described by the van Deemter equation [32] for packed beds constitute the parameters that affect the dispersion of the liquid phase in expanded beds, and hence the efficiency of the process. Adsorbent beads in expanded beds are in constant motion. Solid dispersion can be reduced to a minimum if bead movement is confined to smaller spaces. This can be achieved by a classified bed [33] where beads with a small variation in density with respect to size are classified according to size. When a polydispersed adsorbent is expanded, the larger beads are located at the bottom of the bed and the smaller ones at the top. If the larger beads have a higher density than the smaller beads, the classification will be enhanced further. It has been shown by Wnukowski [34] that the behavior of STREAMLINE adsorbents follows the model derived by Richardson and Zaki [35]:
where uo is the superficial velocity of the fluid, n is an index, E is bed voidage and ut is the terminal falling velocity of a particle in infinite dilution. Stokes’ law is used to express ut:
9.2 Theoretical Background for Expanded Bed Adsorption
205
where d, is particle diameter, g is acceleration due to gravity, p, is the particle density, p is the fluid density, and p is the fluid viscosity. The use of Stokes' equation is restricted to systems where Re,<0.2. Re, is the Reynold number for a particle in a fluid and is defined as udp P Re, = -
(5)
EP
Even though the Re, for the STREAMLINE adsorbent is in the interval 0.1-0.4, no significant difference has been observed between the data simulated by the Richardson-Zaki model and experimental data. The Richardson-Zaki model is only valid for monodispersed particles. The model thus has to be corrected for the particle distribution of the STREAMLINE adsorbents. This is done by using the Perfectly Classified Bed (PCB) model presented by Al-Dibouni and Garside [33]:
where HO is the sedimented bed height, EO is the sedimented bed voidage, AF, is the proportion of the total volume of particles having the diameter within the interval: d-Adl2 < d < d+Ad/2 and E is the bed voidage for the particles within that interval. The model states that the particle size group between d and (d-Ad) is contained within the bed between heights h and (h+Ah). Using this approach, each segment of height is thought of as a bed with a specific bed voidage that can be calculated with the Richardson-Zaki equation. The total bed height is obtained by simply adding up the heights of each segment. It is important to point out that the RichardsonZaki equation only describes expansion in the state of equilibrium, i.e. when the bed has stopped expanding. Figure 9-2 shows the predicted bed expansions for a STREAMLINE adsorbend at different superficial velocities and at different fluid viscosities. Another model which describes a segregated system is that set up by Kennedy and Bretton [36]. The mechanism of the model can be explained as follows. Particles in the superposed monocomponent beds are constantly moving in an essentially random manner. It is therefore to be expected that from time to time a particle from one component will cross the interface into another component where the voidage differs from the equilibrium state experienced before transfer occurred. The effect of this is a net force on the particle that tends to return it to the bed of its own species. In the steady state, the diffusive and convective fluxes of each component on a horizontal plane my be equated to yield the following statement of the Kennedy-Bretton model:
206
9 Expanded Bed Adsorption Chromatography
Degree of expansion
H
HO
1
o,i
:
:
100
200
:
0
300 400 Flow velocitv (cm/h)
Fig. 9-2. A computer calculation of the relationship between bed expansion and flow velocity at different viscosities of the feedstock. q is the relative viscosity where q is 1 for water.
where D,i is the particle dispersion of component i and C,i is the mass concentration of component i. The advantage of using this model is its ability to predict the variation of size distribution with height, in beds with narrow size ranges where the mixing is more pronounced. The preference for describing the segregation in STREAMLINE using the PCB model is its simplicity. The wide particle size distribution of the STREAMLINE adsorbent increases the stability of the bed which makes it possible to use the simple PCB model.
9.2.4 Evaluation of Bed Stability A common approach to evaluating the degree of mixing (dispersion) of the fluid in a system is the use of a tracer stimulus method (RTD), Residence Time Distribution). The tracer is injected into the system, either as a pulse or a step. The resulting concentration profile of the tracer at the outlet reflects the hydrodynamic conditions inside the system. It is important to choose a tracer that is inert to the investigated system and that does not affect the flow properties of the fluid. Otherwise, the tracer will not show the pure hydrodynamic response of the system. The RTD data obtained from the tracer are used to evaluate the flow characteristics by fitting mathematical models to them. The axial dispersion model and the tank-in-series model [37] are the mathematical models which are widely used to characterize hydrodynamics in flow systems. Both are simple models; each has a single parameter that quantifies the dispersion in the system. The axial dispersion model describes the mixing using a one-dimensional differential convective-dispersion model. The parameter that represents the degree of mixing is called the Peclet number and is defined as
9.2 Theoretical Background for Expanded Bed Adsorption
207
where D , is the axial dispersion coefficient. If Pe=O, the system is perfectly mixed, while Pe=m indicates a system with no dispersion, i.e. plug flow. The variance of the calculated response curve of a pulse stimulus is correlated to the Peclet number according to o2 = 2 l P e
(9)
In the tank-in-series model the system is represented as a series of equally sized, perfectly mixed tanks. The parameter describing the degree of mixing is the number of tanks, N. The variance of the calculated response curves of this model is given by
The larger the number of tanks, the less the mixing in the system. By comparing the variances of the two models, a useful relationship is obtained:
and by using Eqn (8)
It is not completely accurate to make this comparison, since the responses of the two different models are never identical [37]. However, for a system with a low degree of mixing, Eqn (12) provides accurate calculations of the axial dispersion coefficient. The axial dispersion coefficient for STREAMLINE has been calculated this way. When evaluating the flow characteristics in the expanded bed, one should take into account the dispersion contributed by the test system. From mathematical statistics it is known that the total residence time and its variance is made up of the sum of the residence times and the variances respectively, of the different parts of the process. By using this correlation the effect of the test system on the observed dispersion can be estimated according to the formula:
where a is the ratio of the test system volume to the expanded bed volume. Usually, the plate number in the test system, Nsys,is much higher than the plate number in the expanded bed, N b e d . This implies that the observed plate number, N&s, is equal to Nbed multiplied by the factor (l+a)2.In other words, the observed plate number is
208
9 Expanded Bed Adsorption Chromatography
an overestimation of the actual plate number in the expanded bed. Normally, the test systems used for evaluating bed stability have a fraction volume less than 0.05 which overestimates the plate number in the expanded bed by a factor less than 1.1. A protein breakthrough test is a valuable tool for verifying the stability of the expanded bed stability and adsorption performance. A protein solution is applied to the bed until the required breakthrough concentration of the protein is detected at the outlet. Protein adsorption capacity of the expanded bed can be calculated from these breakthrough curves, Ideally, the shape of the curve should be identical with the curves obtained from the same adsorbent in a packed bed. A comparison of adsorption efficiency between packed bed mode and expanded bed mode was made by Hjorth et al. [29]. The conclusion was that a properly designed expanded bed process can possess similar adsorption characteristics to a packed bed.
9.2.5 Scale-up of Expanded Beds Since expanded beds are intended for industrial use, a reliable scale-up of the technique is essential in order to maintain the same performance throughout the scales. The main concern when scaling up expanded bed processes is the column, especially the inlet and outlet liquid distribution systems. The most critical design parameters are the number of inlets and the extent of the pressure drop created. These two parameters have to be adjusted with the dimensions of the column. A large industrial column requires a higher pressure drop and a greater number of inlets than a small laboratory column. Table 9-1 compares STREAMLINE columns of different sizes with respect to plate number and the corresponding axial dispersion. The data lie within a very narrow interval which is necessary for a reliable scale up of an expanded bed process. Figure 9-3 shows protein breakthrough curves from experiments run in expanded bed columns with inner diameters ranging from 25 to 600 mm. These data clearly show that adsorption performance is maintained from laboratory scale to production scale. It has also been verified that there are no remaining living cells in these columns after the wash and the compulsory cleaning-in-place (CIP) procedure [38]. Table 9-1. Number of theoretical plates (N) and corresponding axial dispersion (Da) for expanded bed columns (from 25 mm to 600 mm inner diameter). Column
N
D, (m2 s-I)
STREAMLINE 25
50-7 0
3-4 x 10-6
STREAMLINE 200
35-50
5-6 x
STREAMLINE 600
35-50
5-6 x
~
~~
The experiments were performed using 15- cm sedimented bed height adsorbent run in expanded bed mode at a flow velocity of 300 cam h-'. Acetone was used as the tracer substance.
9.3 Development and Operation of Expanded Bed Processes
209
c/c,
0
10
30 40 50 60 70 80 90 100 ADDlied BSA Irno/rnl sedirnented adsorbent)
20
Fig. 9-3. Breakthrough capacities for expanded beds run at different scales using bovine serum albumin (BSA) as the test substance. C, is the BSA concentration at the inlet feedstream and C the BSA concentration in the flowthrough. (Reproduced from [38], with permission.)
An important consideration in the design of the columns is the selection of materials. All materials in a column that are in contact with liquid must display resistance to the chemicals used in the process, including the harsh CIP solutions. Different sizes of columns can have different materials. This is especially true for the column tubes. The small column tubes are usually made of glass while the larger tubes are made of stainless steel. The reason for this is the pressure rating required for large columns and also that these large column tubes cannot be made with the fine tolerances that are needed for a movable adapter. The disadvantage of using stainless steel tubes is their opacity. Process-scale stainless steel columns for expanded bed adsorption are therefore equipped with an adsorbent sensor which is positioned just below the adapter net, i.e. it follows the movement of the adapter. This sensor functions in the following manner. Ultrasonic waves transmitted from the sensor are reflected by the dense quartz cores of the beads. The reflected waves are then detected by a receiver in the sensor, making it possible to locate the surface of the expanded bed. Using this sensor the industrial expanded bed processes can be automated to meet the operating standards of today’s bioprocessing industry.
9.3 Development and ‘Operationof Expanded Bed Processes The working principle of the expanded bed is that it acts like a packed chromatography bed, yet it can handle particulate-containing feedstocks. When the feedstock is applied to the expanded bed column, the target protein is recovered by the adsorbent while particulates (e.g. cells and debris) and other contaminants pass unhindered through the void of the expanded bed (see Fig. 9-1).
210
9 Expanded Bed Adsorption Chromatography
9.3.1 Adsorbents and Equipment for Expanded Bed Adsorption Adsorbents intended for expanded beds must have certain properties to ensure stable fluidization. If an adsorbent designed for packed bed chromatography is used in an expanded bed, only an inconveniently low flow velocity (below 50 cm h-l) can be used since the bed expands too much at higher flow rates. Adsorbents for expanded beds must have a density which is high enough to ensure that the bed does not expand more than approximately three times the settled bed height at an appropriate process flow velocity (300 cm h-l). Another important feature of the adsorbent is the distribution of size and density of the particles (see section 9.2.3). Finally, the adsorbent must be robust enough to be suitable for use with crude feedstocks and rigorous cleaning solutions. STREAMLINE adsorbents (see Fig. 9-4) are specially designed for use in expanded beds. The matrix used for STREAMLINE DEAE, STREAMLINE SP, STREAMLINE Chelating and STREAMLINE Heparin consists of spherical macroporous, cross-linked 6 % agarose particles, containing crystalline quartz to increase the apparent density. The adsorbents can be cleaned with 1 M sodium hydroxide (NaOH) [39] and be autoclaved. The matrix used for STREAMLINE rProtein A consists of spherical macroporous, highly cross-linked 4 % agarose particles, containing an inert metal alloy to increase
Fig. 9-4. Photographs of different adsorbents for expanded bed adsorption. (a) STREAMLINE DEAE and (b) STREAMLINE rProtein A.
9.3 Development and Operation of Expanded Bed Processes
2 11
the density. The alloy is composed of 77 % Ni, 15.5 % Cr, and 7.5 % Fe. The stability of the alloy is high, with no detectable leakage of metals. With buffers commonly used in protein A chromatography, e.g. 0.1 M glycine, pH 3.0, metal leakage is below the detection limit of the analysis method (0.02 p mL-' eluent using induced coupled plasma atomic emission spectroscopy). The adsorbent is stable in all aqueous buffers commonly used in protein A chromatography and cleaning, e.g. 10 mM HC1 (pH 2), 1 mM NaOH (pH l l ) , 0.1 M sodium citrate (pH 3), 6 M guanidinium HC1, 20 % ethanol, and shows no significant change in chromatograpic performance after either 1 week's storage or 100 cycles normal use, at room temperature. A summary of the properties of the various STREAMLINE adsorbents is given in Table 9-2. The equipment used for expanded beds is similar to that used for packed beds and comprises column, pumps, valves, monitors, etc. The only major difference is in the design of the column (see section 9.2.2). Fig. 9-5 shows a typical experimental set-up for expanded bed adsorption and Fig. 9-6 shows different dimensions of STREAMLINE columns. Table 9-2. Some properties of commercially available adsorbents used in expanded bed adsorption chromatography. Adsorbent
Function- Particle ality size distribution (pm)
Mean Ligand density density" (g mL-I)
Breakthrough capacity (mg mL-l)-b
Total binding capacity (mg mL-')C
STREAMLINE DEAE
Anion exchange
100-300
1.15
0.13-0.21
35
55
STREAMLINE SP
Cation exchange
100-300
1.15
0.17-0.24
65
75
STREAMLINE Chelating
Metal affinity
100-300
1. I5
40
8
40
STREAMLINE Heparin
Affinity
100-300
1.15
4
ND
d
STREAMLINE rProtein A Affinity
80-1 65
1.3
6
20
50
" Ligand
density for STREAMLINE DEAE and SP were determined as mmol charged groups per mL of adsorbent and for STREAMLINE rProtein A as mg protein A per mL of adsorbent. For STREAMLINE Chelating ligand density is expressed as Cu2+binding capacity in pmole per mL adsorbent. Ligand density for STREAMLINE Heparin is mg of heparin bound per mL adsorbent. Breakthrough capacity was determined at 300 cm h-' flow velocity in expanded bed mode (15 cm settled bed height) at C/Co = 0.01 using bovine serum albumin (DEAE) and lysozyme (SP). For STREAMLINE rProtein A, breakthrough capacity was determined using human IgG under the same conditions. Breakthrough capacity for STREAMLINE Chelating was determined in a packed bed run at 300 cm h-' flow velocity (5 cm bed height) at C/Co = 0.01 using bovine serum albumin. Total binding capacity was determined in packed bed mode at 50 cm h-' flow velocity under the same conditions as in (b). Functionality of STREAMLINE Heparin was evaluated by its ability to bind antithrombin from bovine plasma. ND, not determined.
2 12
9 Expanded Bed Adsorption Chromatography
Hydraulic fluid
,v
Waste
Buffer
Buffer
I
I
-
Waste
I
\ /
I I
t i
I
Column
Feedstock
=D
UV-monitor Fraction collector
Fig. 9-5. Experimental set-up for expanded bed adsorption using STREAMLINE 25-200 columns (lab scale to pilot scale). The arrows indicate the direction of liquid flow when the column is operated in expanded bed mode. (Reproduced from [38], with permission.)
9.3.2 Experimental Strategy for Expanded Bed Adsorption To save time when a new purification process is being developed, it is suggested that the strategy shown in Fig. 9-7 be followed. Method Scouting This is performed in a small, packed bed column. When a smaller amount of adsorbent is sampled from a larger container, all of the adsorbent has to be resuspended (by gently shaking the container) and then poured into a sintered glass filter funnel. The liquid is removed (by water suction or by sucking with a pump) so that a 'dripdry' adsorbent cake remains on the filter. Care should be taken to ensure that air is
9.3 Development and Operation of Expanded Bed Processes
213
Fig. 9-6. Columns for expanded bed adsorption, Sizes ranging from 2.5 to 100 cm inner diameter.
not sucked into the adsorbent cake, as it can be difficult to completely wet the particles again. The desired amount of adsorbent is then sampled by cutting a slice of the adsorbent cake and then weighing it. When sampling adsorbent from a large container that holds too much adsorbent for the filter funnel, it can be poured into a
Method scouting Packed bed
Laboratory scale Expandedbed
Process verification Pilot scale Expandedbed
Scale Up Production scale Expanded bed
Fig. 9-7.Development stages for an expanded bed adsorption process from laboratory to production scale.
214
9 Expanded Bed Adsorption Chromatography
bucket and the liquid removed using a pump connected to a tube. The end of the tube is fitted with a screen (e.g. a buffer-filter-end) and gently moved over the surface of the settled adsorbent to suck away the excess liquid. If this procedure is not followed and the container is just shaken and some adsorbent is poured out, the sample will not be representative since the biggest and heaviest particles settle very quickly and only the smaller, lighter particles will be sampled. The column used for the method scouting should be at least 1.5 cm in diameter and with a bed height of 15 cm. If a smaller column is used, the protein capacity may appear (falsely) too low due to poor packing and too short a contact time with the protein feedstock. The appropriate amount of adsorbent is poured into the column as a 50% slurry. When the adsorbent particles have settled, the upper adapter is adjusted down into the bed to a level approximately 10% of the settled bed height. The performance of the column is tested by measuring the height equivalent to a theoretical plate (HETP) and peak symmetry as described by Sofer and Nystrom [40]. During method scouting the most suitable adsorbent is selected and conditions for binding (including feedstock preparation), washing, and elution are determined. The approximate binding capacity for the protein of interest is also defined. When the feedstock has been prepared (pH adjusted, etc.) it has to be clarified to avoid blocking the packed bed column. The starting flow velocity should be the flow velocity that will be used in the expanded bed (usually 300 cm h-l, see section 9.3.3).
Method Optimization This is performed with a lab-scale expanded bed column (e.g. STREAMLINE 25 or STREAMLINE 50). The method determined during method scouting is verified, but now the feedstock is unclarified. Here, the effects of cells and/or debris on bed expansion and binding of the protein of interest to the adsorbent are evaluated. Any changes in feedstock preparation are done here and the feedstock load per mL adsorbent is determined together with flow velocities, wash volumes, elution volume, CIP procedure, and storage conditions.
Process Verification This is performed with a pilot-scale expanded bed column (STREAMLINE 200). Here, the system factors (column area, adsorbent volume, feedstock volume, and volumetric flow) are proportionally increased according to the scale-up factor and the results are checked for consistency with the results obtained with the lab-scale expanded bed column.
Scale-up to the final production scale are usually performed in expanded bed columns like STREAMLINE 600 or STREAMLINE 1200. The system factors are proportionally increased as above and the final parameters are fine tuned. Scale-up to final production scale is usually a straightforward process due to the design of the
9.3 Development and Operation of Expanded Bed Processes
215
adsorbent particles and expanded bed columns. Fine tuning can mean adjustment of feedstock properties (such as viscosity, biomass dry weight); adjustment of wash and elution volumes; modification of CIP procedure, etc. Chromatographic variability is not usually a critical issue for scale-up. The ‘non-chromatographic’ factors are of greater concern. One such factor is the feedstock composition which often varies, not just between different scales of fermentation/culture, but also from batch to batch of the same process.
9.3.3 Operation of Expanded Beds The operation of an expanded bed can be divided into six stages which are pictured in Fig. 9-8. The bed height and linear flow rates mentioned below are the nominal values for STREAMLINE ion exchangers and STREAMLINE rProtein A. 1 . Before start up. STREAMLINE column is connected to the system to be used (pumps, valves, etc.). The column is filled approximately one-third full by pumping distilled water or equilibration buffer in the bottom inlet. Any air that is trapped in the distribution system is removed by moving a tube connected to suction over the whole surface of the bottom screen in the column. Approximately 5 cm of liquid is left in the column. The adsorbent slurry is poured or pumped into the column and the adapter is inserted. The bed is allowed to settle and the settled bed height is noted (usually 15 cm). 2. Expansion and equilibration. The column is levelled using a spirit level. The bed is expanded and equilibrated by pumping equilibration buffer upwards through the column at the flow velocity to be used during feedstock application (300 cm h-’ flow velocity). When the expanded bed is stable (usually after 20-30 minutes)
Before start
1
t
t
t
t
1
t
Expansion Feed and application equilibration
Fig. 9-8. Stages involved in the operation of an expanded bed adsorption process. Arrows indicate the flow direction.
216
9 Expanded Bed Adsorption Chromatography
and the pH and conductivity at the outlet are the same as at the inlet, the column is ready for feedstock application. Prior to feedstock application, the stability of the bed can be checked empirically by determining the axial dispersion in the expanded bed [38]. 3. Feedstock application. The feedstock is always stirred during application to avoid settling of particulates. If the behavior of the feedstock in expanded bed adsorption is not known, the adapter is left at its uppermost position to allow for any additional expansion that may be caused by the feedstock (it may have a higher density andor viscosity than the buffer). When the behavior of the feedstock is known, the adapter can be set 5-10 cm above the height at which the bed is known to expand with the feedstock being used. During feedstock application, the column is intermittently back-flushed (the flow direction is reversed) for approximately 5 seconds each time to avoid build-up of adsorbents and/or cells on the adapter. The interval between the back-flushes (usually every 10th to 30th minute) is determined from application to application depending on the properties of the feedstock. Another approach to feedstock application is to run the expanded bed at constant expanded bed height [41]. Under certain conditions, this gives a somewhat higher productivity compared with feestock application at constant flow velocity. 4. Wash. The bed is washed in expanded bed mode (300 cm h-' linear flow velocity) and the column is usually back-flushed as described above at the beginning of the wash. During the wash the adapter is lowered closer to the bed surface to speed up the wash. If the feedstock contains a lot of particulates, the adapter should not be lowered exactly at the start of the wash, but when the bulk of the particulates has been flushed out. If the adapter is moved down too early or too quickly, particulates may block the adapter screen. After the wash, the column is back-flushed a couple of times to remove any particulates that might still be trapped in the distribution system. The bed is allowed to settle, the adapter is moved down to the settled bed surface and the wash is continued in downwards mode (300 cm h-' linear flow velocity) for at least one settled bed volume or until the UV curve levels out. Normally, approximately 15 sedimented bed volumes are needed to wash out unbound proteins and cells [28,29]. 5 . Elution. Elution is performed in a downwards direction in settled bed mode (100 cm h-' linear flow velocity). Increased flow velocity during elution results in increased elution volumes. Elution in expanded bed mode increases the volume of the eluted fraction by about 40% [29]. 6. Cleaning. Cleaning is done in expanded bed mode to allow any trapped particulates to be washed out. It is performed immediately after each run. For some processes, standard cleaning protocols can be used without further development, for others, procedures are optimized on a case-by-case basis. Examples of cleaning procedures are described below under section 9.3.5. If the adsorbent is not to be used for some time, it should be transferred into a storage solution after cleaning. For STREAMLINE DEAE and STREAMLINE rProtein A, 20% ethanol (EtOH) should be used and for STREAMLINE SP, 20% EtOH containing 0.2 M sodium acetate. The ion exchangers can also be stored in 10 mM NaOH.
9.3 Development and Operation of Expanded Bed Processes
217
9.3.4 Feedstocks STREAMLINE expanded bed adsorption recovers protein products from a variety of particulate-containing feedstocks, e.g. cultures/homogenates of bacteria [27,28], yeast [38,42], mammalian cells [43-451, and other particulate-containing solutions such as renatured inclusion body preparations [46] and milk preparations [47]. For a successful recovery process there are a few requirements that must be met. The size of the particles in the feedstock may not exceed the mesh width (approximately 60 pm) of the retaining net in the column, otherwise they will block the column. For this reason, feedstocks containing large clumps of cells or cells forming mycelial pellets must not be applied directly onto the expanded bed. To remove the largest particles prior to application to the expanded bed a simple mesh screen can be used. Furthermore, the feedstock must always be stirred during application to avoid clumping and settling of cells and debris which may disturb the expanded bed. Application of feedstocks which are viscous and/or have a high biomass content may cause the expanded bed to overexpand, resulting in a build-up of adsorbent and/or cells on the adapter. If the expansion is not too great, the build-up can be prevented by intermittently reversing the flow direction. When the biomass content and/ or viscosity is very high, flow will become turbulent, resulting in channeling in the bed and poor adsorption characteristics. The upper limits for feedstock viscosity and biomass content will vary from case to case, but there are a few rules of thumb. For homogenates of E. coli, biomass dry weights up to 5 % work well. The optimum dry weight is 3-4 %, up to 7-8 % can be used. The viscosity of an E. coli homogenate should typically not exceed 10 mPas (measured at a shear rate of 1 s-l); however, up to 50 mPas (at 1 s-I) may work in some cases. In general, a homogenate prepared from a 'fresh' culture (directly from the fermenter) can have a higher dry weight than a homogenate prepared from a frozen cell paste, and still work well in the expanded bed. The only remedy for a feedstock with biomass dry weight that is too high is dilution (the drawback is the increased volume). Should the viscosity of the feedstock be too high there are other options besides dilution. If high viscosity is caused by the presence of a nucleic acid, the feedstock can be treated with a nuclease [28]. If it is caused by a nucleic acid and the feedstock is a homogenate, further homogenization can be used to shear the nucleic acid, resulting in lower viscosity. When a mammalian cell culture requires dilution because high conductivity in the growth media will not permit binding to an ion exchanger, a diluent with low conductivity and high osmolality should be chosen. The osmolality of a cell culture is approximately 300 mOsm kg-I and the conductivity is around 20-25 mS cm-'. To reduce the conductivity to allow for binding to an ion exchanger, usually a 1+2 to 1+3 dilution is necessary. An example of a suitable diluent is a solution of 200 mM D-glucose (approximately 200 mOsm kg-' and 0 mS cm-I). If water is used as diluent, the cells are likely to lyse and release the nucleic acid, which will cause an undesirable increase in viscosity of the feedstock as well as release of intracellular components. Temperature usually has a positive effect on both the binding characteristics [29] and the expanded bed process. A higher temperature will keep the viscosity of the
2 18
9 Expanded Bed Adsorption Chromatography
feedstock low, permitting the use of high flow velocities and shortening process time. The expanded bed process should therefore always be performed at room temperature unless there are valid reasons for running the process at a low temperature, such as a protease-sensitive target protein. For the same reasons, it may be advantageous to apply the feedstock directly from the fermenter without prior cooling. Adjustment of pH can sometimes cause aggregation of cells. This tends to occur with cation exchange since the pH has to be low to enable binding to the adsorbent. Aggregation is usually less severe if the pH adjustment is done immediately before application to the expanded bed, preferably directly in the fermenter, with good homogeneous mixing. The aggregation problem will not be observed during method scouting (packed bed with clarified feedstock) which is why it is important to check the effects of the chosen binding conditions on the unclarified feedstock early on during method development, before too much time and effort has been put into the process.
9.3.5 Cleaning Since reuse of the adsorbent is of importance to overall process economy, it is essential to perform a thorough cleaning procedure after each purification cycle. The adsorbents used for expanded bed adsorption are exposed to crude, unclarified feedstocks and therefore quite rigorous cleaning protocols are usually required. The cleaning procedure is facilitated if as many of the particulates as possible have been washed out during the wash-step prior to elution (during which the adsorbent is packed together). It is also necessary that the cleaning procedure is carried out directly following elution of the target protein. If adsorbent is allowed to remain in the column overnight before the cleaning is performed, it is usually more difficult to clean the adsorbent satisfactorily. The cleaning procedure should be developed according to the starting material and the contaminants present. If a cleaning protocol for a corresponding packed bed process exists, it can be used as a good starting point for cleaning a STREAMLINE adsorbent. Otherwise, the general cleaning method suggested for each type of adsorbent can be tried. The methods are described below and each method ends with re-equilibration buffer. If the adsorbent is to be stored before being reused, STREAMLINE DEAE and STREAMLINE rProtein A should be transferred to 20 % EtOH and STREAMLINE SP to 20 % EtOH containing 0.2 M sodium acetate. The ion exchangers can also be stored in 10 mM NaOH. General Cleaning Protocol for STREAMLINE Ion Exchangers
The adapter is set at twice the settled bed height. Upwards flow direction.
0.5 M NaOH in 1 M sodium chloride (NaCl), 30 cm h-l, contact time 4 hours. Distilled water, 100 cm h-l, three settled bed volumes. - 30% isopropanol, 100 cm h-', three settled bed volumes. -
9.3 Development and Operation of Expanded Bed Processes
219
- 25 % acetic acid (HAc), 100 cm h-l, three settled bed volumes. - Equilibration buffer, 100 cm h-I, 5-10 settled bed volumes (or until reequili-
brated).
General Cleaning Protocol for STREAMLINE rProtein A The adapter is set at twice the settled bed height. Upwards flow direction. - 20 % EtOH containing 2 M NaCl and 1 mM NaOH, 100 cm h-', contact time
2 hours. 20 mM sodium phosphate, pH 7.0 containing 5 % sarcosyl (sodium N-lauroylsarcosinate (weak anionic detergent)), 20 mM EDTA and 0.1 M NaC1, 100 cm h-l, approximately three settled bed volumes or until pH, conductivity and absorbance at the outlet are the same as in the inlet solution. The adsorbent is then left in this solution for 1.5 hours. - 20% EtOH containing 50 mM HAc, 100 cm h-l, 10 settled bed volumes. - Equilibration buffer, 100 cm h-l, 20 settled bed volumes.
-
The following are examples of efficient cleaning protocols for STREAMLINE SP which have been developed for different applications.
Cleaning After Recovery of Recombinant Interleukin-8 (IL- 8) from E. coli Inclusion bodies The adapter is set at twice the settled bed height. Upwards flow direction. - 1 % DARACLEAN 8471 (Grace Dearborn Ltd.) (commercially available clean-
ing agent which is a blend of caustic soda, alkaline salts and Triton@ CF lo), 30 cm h-l, contact time 4 hours. - Equilibration buffer, 100 cm h-I, four settled bed volumes. Cell discruption and solubilization of the inclusion bodies were performed by resuspending the frozen cell paste in 6 M guanidinium hydrochloride. The IL-8 was then renatured by stepwise dilution. The resulting solution, containing cell debris and large amounts of precipitates, was applied to STREAMLINE 50 column containing 300 mL STREAMLINE SP. Approximately 16 L feedstock were applied per cycle. The cleaning efficiency was evaluated by running 50 repetitive IL- 8 purification cycles with cleaning procedure between each cycle. The function of the adsorbent was tested by measuring the degree of bed expansion after each cycle and the breakthrough capacity of a model protein before the first cycle and after the 25th and 50th cycle. The results showed no significant changes in function over the 50 cycles [46]. When a cleaning protocol was developed, the first cleaning solution consisted of 0.5 M NaOH and 1 M NaC1. When the solution was applied onto the column, there were severe problems with precipitation, and large channels appeared throughout the bed. Attempts to clean the adsorbent in a packed bed with the same solution
220
9 Expanded Bed Adsorption Chromatography
also failed. When the cleaning solution was changed to 1 % DARACLEAN the problems disappeared and the new cleaning protocol proved to be very efficient.
Cleaning After Recovery of Recombinant anti-HIV Fab Fragment Expressed in E. coli The adapter is set at twice the settled bed height. Upwards flow direction.
0.5 M NaOH containing 1 M NaC1, 30 cm h-I, contact time 4 hours. - Distilled water, 100 cm h-l, three settled bed volumes. - Distilled water, inlet temperature 85-90 "C,100 cm h-l, 10 settled bed volumes. - 25 % acetic acid containing 20 % EtOH, 100 cm h-l, three settled bed volumes. - Equilibration buffer, 100 cm h-l, 10 settled bed volumes. -
The Fab fragment was produced in the periplasm of E. coli and osmotic shock was used to release the protein. The unclarified lysate was applied to STREAMLINE 50 column containing 300 mL STREAMLINE SP. Approximately 4 L feedstock were applied per cycle. The cleaning efficiency was evaluated by running 50 repetitive Fab purification cycles with a cleaning procedure between each cycle. The function of the adsorbent was tested by measuring the degree of expansion after each cycle and the dynamic capacity of a model protein before the first cycle and after the 30th and 50th cycle. The results showed very small variations in bed expansion and no change in dynamic capacity.
9.4 Applications of Expanded Bed Adsorption Here we would like to discuss a few examples where expanded bed adsorption chromatography has been used to capture proteins from crude feedstocks. A crude feedstock is in this sense defined as starting material containing particulates of any kind. These could be whole cells, cell debris, cell organelles, or other components that would normally be removed by centrifugation, flocculation/precipitation, and/or filtration steps from the feedstock prior to application to a packed bed (see section 9.1.1). The target protein may be a native protein or enzyme, or a recombinant one, where a host organism is expressing the foreign protein. The diversity of hosts for recombinant protein production has increased tremendously [48], ranging from bacteria and yeast fermentation to culturing of mammalian and insect cells to the (transgenic) production in intact plants or animals. The target protein is expressed intracellularly, extracellularly or in the periplasmic space (only in Gram-negative bacteria, e.g. E. coli) depending on a number of factors. Listed below are a few examples which use adsorbents and equipment specially designed for expanded bed adsorption (see section 9.3.1).
9.4 Applications of Expanded Bed Adsorption
22 1
9.4.1 Adsorption of Recombinant Protein Expressed in E. coli Periplasm using STREAMLINE DEAE [49] In this example, expanded bed chromatography was used to capture modified Pseudomonas aureginosa exotoxin A, required for the preparation of a conjugated vaccine. The protein, recombinantly produced in E. coli, accumulated in the periplasmic space of the bacterium and was released from the cells by osmotic shock. The resulting feedstock was then applied directly to the expanded bed of the anion exchanger (STREAMLINE DEAE). The exotoxin was adsorbed while particulates were washed out. Elution was performed in a downwards direction in settled bed mode. Figure 9-9 compares the capture of the exotoxin in packed bed with capture in expanded bed. Starting with 4.5 kg of E. coli cell paste in both cases, processing time was reduced from 8 to 2.5 h, elution volume was reduced almost three-fold, and recovery was slightly increased to 79 % by replacing the conventional clarification and first chromatography step with expanded bed adsorption. A nuclease (Benzonase') was added to the sample (75 units per gram of cells) to avoid fouling of the adsorbent due to the high viscosity caused by DNA.
Packed bed process
Expandedbed process
-I
4.5 kg Bacteria
Sucrose 50 mM Buffer
90 L Suspension
180 L Suspension
I
I STREAMLINE DEAE 4.7 L
-I
4.5 kg Bacteria
Sucrose 20 mM Buffer
I
2.5 h
I
1 1st Centrifugation
I
23h
I
13.5 L
2-3 h
I
4 DEAE Sepharose Fast Flow 25.5 L
I
3h
4
36 L Fig. 9-9. Capture of exotoxin using a packed bed procedure and an expanded bed procedure. The time for performing the different steps in the processes is given. (Adapted from [49].)
222
9 Expanded Bed Adsorption Chromatography
9.4.2 Adsorption of Recombinant Protein Expressed in E. coli Cells using STREAMLINE DEAE [28] In this example, human placental annexin V was recovered from an E.coZi homogenate. The annexins constitute a group of structurally related, calcium-dependent, phospholipid-binding proteins with a diversity in function [50].Annexin V is an anticoagulant protein and was selected as a model protein. It was cloned to be expressed intracellularly in E. coli and released from the harvested cells by high-pressure homogenization (three passages). This procedure effectively disrupted the cells and also reduced the viscosity of the sample caused by DNA. Due to the tendency of the annexins to associate to membraneous structures (phospholipids), the addition of a detergent (Triton X-100, 1 % (v/v) final concentration) increased the yield of annexin V. The results from the lab-scale and pilot-scale expanded bed capture of the recombinant annexin from crude feedstock are shown in Table 9-3.
9.4.3 Purification of Native gp 120 from HIV-1 Infected T Cells using STREAMLINE SP [51] The outer envelope glycoprotein gp 120 of human immunodeficiency virus 1 (HIV1) was purified in a two-step chromatographic procedure in which expanded bed adsorption (cation exchange) was the first. The feedstock, i.e. the virus cultured in human T cells, was treated with a detergent (Empigen@BB), to kill the viral particles and to solubilize gp 120, and a nuclease, (Benzonase) to reduce the viscosity of the sample before application to the expanded bed. The clarified and concentrated eluate was then further purified on an affinity column and pure gp 120 was recovered with retained specificity. Analyses for remaining DNA and RNA and nuclease were negative.
Table 9-3. Recovery of annexin V from a crude E. coli homogenate using STREAMLINE DEAE at laboratory and pilot scales. (Adapted from [28].) ~
Scale
Laboratory
Pilot
Column diameter (mm)
50
200
Adsorbent volume (L)
0.3
4.1
Feedstock volume (L)
1.7 141
26.5
Total cycle time excluding CIP (min) Yield of annexin V (%)
>95
>95
141
9.4 Applications of Expanded Bed Adsorption
223
9.4.4 Capture of Two Aprotinin Variants Produced in Hansenula polymorpha on STREAMLINE SP [52] Hansenula polymorpha is a methylotrophic yeast, which means it is able to use methanol as its sole energy and carbon source. This feature has been used to construct an expression system for heterologous proteins. In this work, two variants of the basic protease inhibitor aprotinin where produced. Clarification, concentration, and capture were performed on an expanded bed (cation exchanger). The only sample pretreatments were a 1: 1 dilution of the yeast culture (cell mass 100 g dry weight L-') with deionized water to reduce the conductivity (to 25 mS cm-') and adjustment of pH to 3.5. Totally, 6400 mL (5 % dry weight) were applied to 300 mL expanded STREAMLINE SP. A four-fold purified and seven-fold concentrated aprotinin was recovered (76 % yield).
9.4.5 Affinity Purification of G6PDH from Unclarified Yeast Homogenate using STREAMLINE Red H-7B [42] Affinity chromatography occupies a unique place in separation technology since it is the only technique which enables purification of almost any biomolecule on the basis of its biological function or individual chemical structure. The high selectivity of affinity separations derives from the natural specificities of the interacting molecules; affinity chromatography can therefore be used for purifying substances from complex biological mixtures. Selective extraction of the intracellular glycolytic enzyme glucose 6 -phosphate dehydrogenase (G6PDH) from disrupted yeast cells using a triazine dye (Procion" Red H-E7B) immobilized onto a STREAMLINE matrix was achieved by expanded bed chromatography. By using the enzyme's cofactor, NADP, as a selective eluent, G6PDH was recovered with a yield of 99 % and an average purification factor of 103. As a comparison with this affinity capture, a method for the purification of GBPDH on STREAMLINE DEAE was developed [53]. An ion-exchange adsorbent is cheaper and can be subjected to harsher cleaning procedures. Despite the relative non-specificity of the separation, it was possible to find conditions in which the adsorbent showed some degree of selectivity for the enzyme. When the method was fully optimized, G6PDH was recovered with a yield of 98 % and a purification factor of 12.
9.4.6 Capture of Humanized IgG4 Directly from the Fermenter using STREAMLINE rProtein A [54] Affinity purification of a humanized IgG4 was carried out by applying a myeloma cell culture to STREAMLINE rProtein A without any pretreatment of the sample. The fermenter outlet was connected directly to the inlet of STREAMLINE 200 col-
224
9 Expanded Bed Adsorption Chromatography
umn and approximately 100 L of the feedstock, at 37 "C, were loaded onto 4.7 L of the expanded adsorbent. After 1 h 93 L had been applied and after a further 1.5 h for washing and elution, more than 40 g of the monoclonal antibody was quantitatively recovered. The purity of the antibody recovered from the expanded bed column was comparable with that obtained by traditional packed bed protein A affinity purification of IgG from clarified cell culture supernatants. In parallel to the large-scale experiment, two consecutive small-scale experiments were performed with the same feedstock to test reproducibility and scale-up. The experimental conditions were linearly scaled down and 75 mL of the same batch of adsorbent were used in a STREAMLINE 25 column. A short cleaning-in place procedure was performed between the runs using 6 M guanidine-hydrochloride (two sedimented bed volumes) as cleaning agent.
9.4.7 Process for Purifying Recombinant Human Serum Albumin [55] Recombinant human serum albumin (rHSA) was produced in a yeast (Pichia pastoris). The culture was then heat-treated to deactivate the proteases originating from the host cells before applying it to an expanded bed (STREAMLINE SP). The introduction of an expanded bed process has enabled one filtration and two ultrafiltration steps to be omitted from a conventional process without affecting the quality of the product. The process was scaled up to a 1-m diameter STREAMLINE column with 150 L adsorbent (STREAMLINE SP) and 2000 L of the crude feedstock (diluted 1+1) was applied to the expanded bed. The total yield of rHSA from four different runs at large scale was 87 %, which agrees well with the yield obtained in small-scale experiments (5 -cm diameter column). A process scheme for the expanded bed process and the conventional process is shown in Fig. 9-10.
9.4.8 Capture of Native Alcohol Dehydrogenase from Baker's Yeast by Hydrophobic Interaction Expanded Bed Chromatography [56] In this work it was found to be advantageous to use hydrophobic interaction chromatography rather than ion-exchange chromatography to capture indigenous alcohol dehydrogenase from baker's yeast. Interestingly, a higher yield (94 Yo compared with 85 %) of the enzyme was obtained if the unclarified yeast homogenate was applied to an expanded bed (STREAMLINE Phenyl) than if the material was first clarified and then purified on a packed bed of Phenyl Sepharose@' 6 Fast Flow (low sub). Also, the packed bed fouled more rapidly than the expanded bed, indicating that even high-speed centrifugation was insufficient to clarify the feed stream.
References Expandedbed process
225
Conventional process
1
1
I
I
I
I
Fig. 9-10. Comparison of a conventional process and an expanded bed process for recovery of
Packed bed chromatography
Acknowledgments We would like to thank Drs L. Hagel, P. Hedman und P. Wnukowski for valuable comments on this manuscript. V. Sedgwick is acknowledged for linguistic revision.
References [l] [2] [3] [4] [5] [6]
Lee, S.-M., J Biotechnol 1989,II, 103-118. Wittington, P.N., Appl Biochem Biotechnol 1990, 23, 91-121. Datar, R. RosCn, C.-G., Chem Eng J 1987, 34, B49-B56. Berthold, W., Kempken, R., Cytotechnology 1994, 15, 229-242. Baily, F.J., Warf, T., Maigetter, R., Enzyme Microb Technol 1990, 12, 647-652. Fane, A.G., Radovich, J.M., in: Separation Processes in Biotechnology: Asenjo, J.A. (Ed.). New York: Marcel Dekker, Inc., 1990; pp. 209-262. [7] Kroner, K.H., Schutte, H., Hustedt, H., Kula, M.R., Process Biochem 1984, 19, 67-74. [8] Pungor, E., Jr., Afeyan, N.B., Gordon, N.F., Cooney, C.L., Bio/7"chnology 1987, 5, 604608. [9] Vorauer-Uhl, K., Skias, M., Trkola, A., Schonhofer, W., Unterluggauer, F., Schulz, P., Jungbauer, A., Bioseparation 1993, 3, 217-226. [lo] Bartel, C.B., Kleinman, G., Korzun, J.N., Irish, D.B., Chem Eng Prog 1958, 54, 49-51. [ l l ] Gailliot, F.P., Gleason, C., Wilson, J.J., Zwarick, J., Biotechnol Prog 1990, 6, 370-375. [12] Buijs, A., Wesseling, J.A., J Chrornatogr 1980, 201, 319-327. [13] Burns, M.A., Graves, D.J., Biotechnol Prog 1985, I , 95-103. [I41 Lochmuller, C.H., Wigman, L.S., Sep Sci Technol 1987, 22, 2111-2125. [15] Nixon, L., Koval, C.A., Xu, L., Noble, R.D., Slaff, G.S., Bioseparation 1991, 2, 217-230. [16] Draeger, N.M., Chase, H.A., Bioseparation 1991, 2, 67-80. [17] Chase, H.A., Draeger, N.M., Sep Sci Technol 1992, 27, 2021-2039. [18] Chase, H.A., Draeger, N.M., J Chromatogr 1992, 597, 129-145.
226
9 Expanded Bed Adsorption Chromatography
[19] Chase, H.A., Trends Biotechnol 1994, 12, 296-303. [20] Wells, C.M., Patel, K., Lyddiatt, A., in: Separations f o r Biotechnology: Verrall, M.S., Hudson, M. J. (Eds.). Chichester: Ellis-Horwood, 1987; pp. 217-224. [21] Morton. P.H., Lyddiatt, A., in: Ion Exchange Advances: Slater, M.J. (Ed.). London: Elsevier, 1992; pp. 237-244. [22] Dasari, G., Prince, I., Hearn, M.T.W., J Chromatogr 1993, 631, 115-124. [23] Spence, C., Schaffer, C.A., Kessler, S . , Bailon, P., Biomed Chromatogr 1994, 8, 236-241. [24] Thommes, J., Halfar, M., Lenz, S., Kula, M.-R., Biotechnol Bioeng 1995, 45, 205-211. [25] McCreath, G.E., Chase, H.A., Owen, R.O., Lowe, C.R., Biotechnol Bioeng 1995, 48, 341354. [26] Unger, K.K., J Chromatogr Library. Amsterdam: Elsevier Scientific Publishing Company, 1979. [27] Hansson, M., StBhl, S., Hjorth, R., UhlCn, M., Moks, T., BioRechnology 1994, 12, 285-288. [28] Barnfield Frej, A.-K., Hjorth, R., Hammarstrom, A., Biotechnol Bioeng 1994, 44, 922-929. [29] Hjorth, R., Kampe, S., Carlsson, M., Bioseparation 1995, 5, 217-223. [30] Hedman, P., Barnfield Frej, A.-K., in: Harnessing Biotechnology for the 21st Century: Ladish, M.R., Bose, A,, (Eds.). Washington DC: ACS Conference Proceedings Series, 1992; pp. 271-274. [31] Asif, M., Kalogerakis, N., Behie, L.A., AIChe J 1991, 37, 1825-1832. [32] Van Deemter, J.J., Zuiderweg, F.J., Klinkenberg, A,, Chem Eng Sci 1956, 5, 271-289. [33] Al-Dibouni, M. R., Garside, J., Trans I Chem E 1979, 57, 94-103. [34] Wnukowski, P., personal communication. [35] Richardson, J.F., Zaki, W.N., Trans I Chem E 1954, 32, 35-53. [36] Kennedy, S.C., Bretton, R.H., Am Znst Chem Engrs J 1966, 12, 24-30. [37] Levenspiel, O., Chemical Reaction Engineering, 2nd ed., Singapore: John Wiley & Sons, 1972. [38] Barnfield Frej, A.K., Johansson, H.J., Johansson, S . , Leijon, P., Bioprocess Engineering 1997, 16, 57-63. [39] Hansson, K.-A., Abstracrs, 7th European Congress on Biotechnology, Nice, France, February 1995. European Federation of Biotechnology, 1995. [40] Sofer, G.K., Nystrom, L.-E., Process Chromatography. A Guide to Validation. London: Academic Press, 1991. [41] Chang, Y.K., Chase, H.A., Biotechnol Bioeng 1996, 49, 512-526. [42] Chang, Y.K., McCreath, G.E., Chase, H.A., Biotechnol Bioeng 1995, 48, 355-366. [43] Thommes, J., Halfar, M., Lenz, S., Kula, M.-R., Biotechnol Bioeng 1995, 45, 205-211. [44] Batt, B.C., Yabannavar, V.M., Singh, V., Bioseparation 1995, 5, 41-52. [45] Thommes, J., Bader, A,, Halfar, M., Karau, A,, Kula, M.-R., JChromatogr 1996,752,111-122. [46] Barnfield Frej, A.-K., Bioseparation 1996, 6, 265-271. [47] Noppe, W., Hanssens, I., De Cuyper, M., J Chromatogr 1996, 719, 327-331. [48] Ogden, K.L., Bier, M., in: Molecular Biology and Biotechnology, A Comprehensive Desk Reference: Meyers, R.A. (Ed.). New York: VCH Publishers, Inc., 1995; pp. 106-110. 1491 Johansson, H.J., Jagersten, C., Shiloach, J., J Biotechnol 1996, 48, 9-14. [50] Creutz, C.E., in: Molecular Biology and Biotechnology, A Comprehensive Desk Reference: Meyers, R.A. (Ed.). New York: VCH Publishers, Inc., 1995; pp. 33-36. [5 I ] Gilljam, G., personal communication. [52] Zurek, C., Kubis, E., Keup, P., Hijrlein, D., Beunink, J., Thommes, J., Kula, M.-R., Hollenberg, C.P., Gellissen, G., Process Biochemistry 1996, 31, 7, 679-689. 1531 Chang, Y.K., Chase, H.A., Biotechnol Bioeng 1996, 49, 204-216. [54] Bonnerjea, J., personal communication. [55] Noda, M., Sumi, A., Ohmura, T., Yokoyama, K., 1996, European Patent Application, EP0699687A2. 1561 Smith, M.P., Bulmer, M., Hjorth, R., Titchener-Hooker, N.J., Proceedings: 5th World Congress of Chemical Engineering, July 1996, vol. 2, pp. 565-570.
Part Two Membrane Separations
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
10 Application of Membrane Bioseparation Processes in the Beverage and Food Industries Dan Donnelly, Joe Bergin, Tom Duane and Niall McNulty
10.1 Introduction Recent stringent standards imposed by regulatory authorities have increased the demand for membrane materials in the food and beverage sector, and this is predicted particularly to affect the dairy processing industry. An analysis of the expansion of the market in the United States for membrane materials has shown that demand will be $1.3 billion in 1998 [l]. Of this, 20% will be accounted for by the food and beverage sectors and the largest segment, for water and wastewater treatment, will be 44%. The pharmaceutical and medical applications are predicted to be 15 % of the total, whereas chemical processing will account for 8 %. Industrial gas production, the most rapid growth market for membrane materials, will make up 4.5 370 of the total market. The advantages of using a membrane process as opposed to other classical unit operations, e.g. evaporation, are improvements in product quality due to less severe heat treatment, low energy consumption, the small footprint and easy maintenance of the plant, and the varied applications of the technology. Drivers for the use of membrane technologies in industry include: - improvement in product quality; - reduction in production costs; and - use of hybrid processes which cannot function economically by either membrane
process or traditional process on a stand alone basis. Microfiltration, ultrafiltration, reverse osmosis, nanofiltration, electrodialysis, and gas separation by membranes are now established technologies in the food and beverage industries. In addition, exciting opportunities exist for the use of bioseparations which make use of catalytic membrane reactors. This chapter deals with the different types of membrane technologies available, the potential of bioseparations, and the established membrane technologies in the food and beverage sectors.
230
10 Membrane Bioseparation Processes in the Beverage and Food Industries
10.2 Membrane Technologies 10.2.1 Pressure-Driven Processes Microfiltration This process selectively removes particles in the size range of 0.01 to 10 pm from liquids or gases (Fig. 10-1). Two types of microfiltration (MF) technologies are in use: dead-end or crossflow MF. Dead-end MF is the most common type of MF encountered in the beverage industries, where it finds application in sterile filtration and clarification. It employs depth or surface membranes. In depth filtration, retained particles build up in the membrane void, either by a sieving action or by inertial impaction on the fibrous materials from which they are fabricated (Fig. 10-2(a)). In surface microfiltration, the filters are manufactured from synthetic polymers, and particles are retained on the upstream surface of the filter by a sieving mechanism (Fig. 10-2(b)). Build-up of particles during dead-end filtration necessitates the replacement or cleaning of the filter medium between filtration runs. For this reason, dead-end filtration is a batch process. Crossflow MF can be used as an alternative to dead-end MF. The crossflow configuration has the advantage that particles do not build up indefinitely on the membrane surface, rather they are sloughed off by the high shear imposed by the tangential flow of the bulk suspension relative to the surface of the membrane (Fig. 10-2(c)). For this reason high flux rates can be maintained for long periods of time. Inevitably, fouling of the membrane will occur over time and the flux rate will decline.
Fig. 10 -1. Classification of separation processes. (Courtesy of Memtech UK Ltd.)
10.2 Membrane Technologies
23 1
Fig. 10-2. (a) Depth filtration. (b) Surface filtration. (c) Crossflow filtration.
Ultrafiltration Ultrafiltration (UF) membranes have the ability to retain molecules in the molecular weight range 300 to 500000 Da (Fig.10-1). Membranes are classified based on the nominal molecular cut-off, which is defined as the smallest molecular weight species for which the membrane has more than a 90% rejection [ 2 ] . The driving force for transport across the membrane is the pressure applied to the feed liquid. A major advantage of UF employed in a crossflow configuration is that the retentate and the permeate are both recoverable. Hence, in the food and beverage industries UF is used for concentration and for fractionation purposes. Diufiltrution is a process that can be used in conjunction with UF to remove salts or low-molecular weight species from macromolecules, or for fractionation. It involves adding water to the process stream so that salts or low-molecular weight
232
10 Membrane Bioseparation Processes in the Beverage and Food Industries
species, which are freely permeable to the membrane, are decreased in concentration in the retentate. Reverse Osmosis Reverse osmosis (RO) is used to separate water from low-molecular weight solutes which have a high osmotic pressure, such as salts and sugars (Fig. 10-1). A high pressure (4000-8000 @a) is required to overcome the osmotic pressure exerted by such solutes - hence the term reverse osmosis. Reverse osmosis differs from UF in that UF membranes retain only large macromolecules, which exert only a low osmotic pressure with the smaller solutes being transported across the UF membrane. Ultrafiltration operates at lower pressures (20-2000 kPa) than RO. Reverse osmosis is normally used for concentration. However, it finds some applications in fractionation, such as in the removal of ethanol from alcoholic beers to produce non-alcoholic or low-alcoholic beer (NABLAB).
Nanofiltration Nanofiltration can be described as a type of filtration intermediate between RO and UF (Fig. 10-1). It allows partial demineralization as well as the concentration of macromolecules. This is accomplished by the use of membranes which are often negatively charged giving rise to selective salt rejection. Organic compounds of greater than 200 to 500 Da are also rejected. Such plants run with high flux rates and relatively low pressures.
10.2.2 Non-Pressure-Driven-Processes Electrodialysis Electrodialysis (ED) is a process in which an electrical potential is the driving force which effects separation of ions across ion-selective exchange membranes. Since the average velocity of ions in bulk solution is low (1 cm min-'), ED plant is arranged in ED stacks, composed of alternating anionic and cationic exchange membranes. When an electrical potential is applied across such a stack, the solutions between the anionic and cationic membranes are alternatively enhanced or depleted of electrolytes. The high and low salt solutions are withdrawn from their respective compartments. Non-electrolytes are unaffected by ED, therefore a high degree of demineralization can be achieved in the absence of product concentration (Fig. 10-3).
10.2 Membrane Technologies
233
Desalted product
Anionpermeable membrane
1
’
I
1
I
I
I
I
Cation-permeable
1
I
1
+ -L
H
I
I
Cathode
Anode
l
/
/
~
l
i
i
/
l
Feed
Fig. 10-3. Electrodialysis.
Gas Separation by Membranes Membranes can be used to separate gas mixtures. In such separations all streams are gases and no phase change occurs. Two types of gas separation exist. Porous membranes separate on the basis of the ratio of pore size to mean path size of the gas molecules. Non-porous (solution-diffusion) membranes separate on the basis of the gas molecules having to dissolve and diffuse across the membrane matrix. Many of the initial applications of gas separation have been in the petrochemical and chemical industries. Applications for such systems also exist within the brewing industry [3]. In the area of industrial gas production, membrane systems have numerous advantages over conventional cryogenic gas production methods, including reduced operating costs, lower installation costs, and ease of operation. A special case of gas separation is pervaporation (PV). Pervaporation is characterzid by the separation of a liquid phase from a gaseous phase by a third phase which is a membrane barrier. Mass transfer occurs selectively across the barrier to the gas side. A phase change is required for the solutes to diffuse across the membrane, permselective ‘evaporation’ of liquid molecules, and hence the process is termed pervaporation [4]. The driving force for the mass transport can be expressed, in a general way, in terms of an activity gradient [ 5 ] . Pervaporation is very suitable for the treatment of food and biological systems [6]. The application of pervaporation in the food processing area has recently been reviewed [7]. Hydrophobic membrane technology has potential applications for gas exchange in the brewing industry [3]. These membranes contain small pore sizes (0.05 pm) which allow relatively high pressures, up to 500 kPa, to be employed before the liquid will break through the pores and mix with the gas. This driving force for gas transfer is the difference in partial pressures between the gaseous phase and the liquid phase. If the partial pressure of a particular gas in the gaseous phase is higher than that in the
234
10 Membrane Bioseparation Processes in the Beverage and Food Industries
liquid phase, the gas will enter the liquid. Conversely, gas may be removed from the liquid phase if the partial pressure in the gas phase is lower.
10.2.3 Membrane Reactors There is no commonly accepted definition of a membrane reactor. The term is a generic one encompassing all processes that combine any biochemical reactor with a membrane process or a process in which the membrane itself acts as a reactor [8]. The former includes any system in which a membrane application, such as UF, MF, or ED, is combined with, for example, a fermentation process. The latter system allows continuous reaction with simulaneous separation of the product from the reaction mixture. This is accomplished by immobilization of the biological component in or on the membrane matrix. At the forefront of membrane separation technology is the development of catalytic membrane reactors (CMR). The acceleration in membrane technology development has led to an improved comprehension of membrane permeability and selectivity, Coupled with new developments in the field of enzyme immobilization and reactor design, this has led to the development of CMR. Catalytic membrane reactors and enzymatic membrane reactors are the terms used most frequently to describe this technology. The term enzymatic membrane reactor precludes the use of non-biological catalysts, both organic and inorganic, which have been used in the operation of CMR [ 9 ]. Catalytic membrane reactors have been likened to biological systems which involve selective transfer and a chemical modification. There is a field of research related to the application of CMR that mimics particular biological functions. In one particular case, islets of Langerhans, the cells responsible for glucose-dependent insulin secretion, were retained within a polyamide ultrafiltration membrane, thus creating a bioartificial pancreas [ 101.
10.3 Process 10.3.1 Process Conditions for Crossflow Filtration Crossflow filters can be used in a number of operating modes. Simple Batch (Figure 10-4) In this mode of operation, the process liquid is pumped from a feed tank to a membrane filter, from which the permeate is removed to another vessel, and the retentate is circulated back to the feed tank. As the permeate is removed, less liquid is returned to the feed tank. Consequently, the level in the feed tank drops until the
. 10.3 Process
rn I
I
Feed I concentrate Tank
I
235
Retentate
__ __
-.
...... Filtrate / . - - - ~
permeate
__
charge in the tank is too low for the process to continue. The filtration is then terminated.
Fed Batch (Fig. 10-5) The fed batch mode is similar to simple batch mode except that, as the level in the feed tank starts to drop, it is replenished with fresh process liquid. Operating in this mode, higher levels of concentration are attainable compared with the batch mode.
-Fresh
feed
1
Feed /concentrate Tank
. ....... Filtrate /
.--.,
permeate
l l I
L r
Pump
Fig. 10-5. Fed batch crossflow system.
236
10 Membrane Bioseparation Processes in the Beverage and Food Industries
Continuous (Fig. 10-6 and 10-7) This mode of operation allows continuous filtration and differs from the batch modes in that the retentate is not fed back to the feed tank. The feed stream is delivered to a recirculating membrane loop, which is connected to a series of such loops. In each stage, the concentration increases, until in the final stage, the requisite concentration is attained. The rate of product removal is balanced by the addition rate of fresh process liquid. Concentrate
collection
I
1
I
L - 4
Feed Tank
’
... F i h t e ............. L
Pump
Pump
--+T Pump
Fig. 10-6. Continuous crossflow system.
Fig. 10-7. Continuous crossflow system. (Courtesy of Memtech UK Ltd.)
permeate
10.3 Process
231
10.3.2 Membrane Materials Polymeric materials are the most widely employed in membrane separations. Cellulosics are the market leaders, although this dominance is being eroded to some extent by polymers of nylon, polysulphone, polycarbonate, polypropylene, polytetrafluoroethylene (PTFE), polyacrylonitrile, and polyvinylidene fluoride. Ceramic and inorganic materials are increasing their market share, especially in high temperature applications.
10.3.3 Membrane Configurations Four membrane configurations are shown in Fig. 10-8.
Fig. 10-8. (a) Tubular membrane. (b) Hollow-fiber membrane. (c) Spiral wound membrane. (d) Plate and frame membrane. (Illustrations courtesy of Memtech UK Ltd.)
238 -
10 Membrane Bioseuaration Processes in the Beverage and Food Industries
Tubular arrangements are employed when the liquids being processed are high in suspended solids content. The tubular membranes range from 4-25 mm internal diameter (Fig. lO-8(a)). Hollow fibers range in size 0.5-3 mm internal diameter and enable a large surface area to be housed in a small volume (Fig. 10-8(b)). Spiral wound arrangements can be operated at high pressures when employed with suitable spacer materials (Fig. 10-8(c)). This arrangement also provides large surface areas. Plate and frame systems with flat sheet membranes suitably arranged with flow and permeate channels, can be adapted for many applications (Fig. 10-8(d)).
10.3.4 Flux Rate Flux rate refers to the rate of permeate flow through the membrane. It is normally expressed as litres m-* h-' (LMH). Decreases in membrane flux are normally attributed to two particular effects.
- Concentration polarization, which occurs on all crossflow membranes. This is the polarized or gel layer that accumulates at the surface of the membrane causing a reduction in membrane flux. It is a reversible effect that can be reduced by backflushing and increasing recirculation rates of flow. If an effective cleaning regime is employed, original flux rates can be restored [ll]. - Membranefouling occurs when the polarized layer on the membrane builds up to such an extent that particles are adsorbed and deposited at the membrane pores. This leads to a large reduction in flux and a rejection of solutes that are intended to pass through. Membrane fouling can be distinguished from concentration polarization after shut down. When the polarized layer has dispersed it is concentration polarization, whereas, if it remains, fouling has occurred.
10.3.5 Membrane Cleaning and Disinfection A major disadvantage of membrane separations is fouling. Fouling adversely affects flux rates, reduces the operating time between the period of cleaning, and in some cases it can alter the membrane separation characteristics. Fouled membranes and auxiliary equipment are generally cleaned by cleaning-in-place (CIP) procedures. CIP regimes are usually based on various chemical or enzymatic treatments to restore the membrane to its original state, and it helps to maintain hygienic conditions within the plant. Many proprietary cleaning agents are available. Acids such as nitric acid (0.5 %) or ethylenediaminetetra-acetic acid (EDTA) are used to remove mineral deposits. Caustic-based detergents (0.5-1 % NaOH) are used to remove proteinaceous deposits. Enzyme cleaning agents containing hydrolytic enzymes such as amylases, proteases, or glucanases are sometimes used for specific applications, and
10.4 Catalytic Membrane Reactors for the Food Industry
239
are used at the pH optimum for the respective enzyme. Membranes used in the treatment of biological materials are usually cleaned daily [12]. Rinsing with water at high circulation rates and reduced pressure, or back-flushing from the permeate side of the membrane is also used to clean membranes. For sterile filtration and other aseptic processing, the filters and plant must be sterile. This may involve steaming in place, provided that such filters are capable of withstanding wet heat, i.e. they must be inorganic filters. Chemical sterilants such as hypochlorite, hydrogen peroxide, peracetic acid, and sodium bisulphite are also used for the disinfection of membranes. In most cases membranes are still chemically sterilized, while auxiliary plant is steam-sterilized. The use of membrane separations can be restricted by membrane fouling and customized cleaning regimes need to be developed for specific membranes and applications [ 131. Operating conditions can also be optimized to reduce fouling [14].
10.4 Catalytic Membrane Reactors for the Food Industry Catalytic membrane reactors can combine the use of a traditional chemical reactor with a membrane operation. Alternatively, by immobilizing the catalyst on the membrane, substrate transformation can take place in situ, (Fig. 10-9). It is these immobilized systems that allow both the catalytic reaction and the membrane separation to take place inside a single unit. There are a variety of membrane types employed in the construction of catalytic membrane reactors, which reflect the wide range of membrane separation options available.
Membrane Unit
Indicator Enzyme 0 Reactants A Product IB Pump
Product Collection Vessel
Shell
Fig. 10 -9. Membrane separation with enzyme immobilization.
240
10 Membrane Bioseparation Processes in the Beverage and Food Industries
10.4.1 Continuous Stirred Tank Membrane Reactors (CSTMR) The first type of catalytic membrane reactor to be developed consisted of a continuous stirred tank reactor (CSTR) coupled with a separation module, (Fig. 10-10). There are three different configurations of this reactor. The first consists of a CSTR linked to a dead-end separation unit from which the permeate is removed continuously. Such reactors require continuous agitation to prevent concentration polarization, and to accelerate the reaction rate. Operational inefficiency compared with other membranes reactors have made this type of reactor obsolete. The recycle CSTMR is more prevalent [ 151. The enzyme in this second type of CSTR system is circulated continuously through a membrane separation unit from which the enzyme-free permeate containing the product is removed. The retentate containing the free enzyme is then returned to the CSTR. The criteria for membrane selection in such systems are maximum enzyme retention and minimal product retention. The advantage of recycle CSTMRs over dead-end systems is that the membrane and reactor are separate entities, thus facilitating easier control of reaction conditions. The third type of CSTMR is the cascade reactor, which consists of a number of CSTRs in series. Such a serial arrangement can operate with each reactor having a dedicated membrane separation unit, or reactors can be coupled to a single-membrane separation unit. Operation of these units in series overcomes the low reaction rates generally associated with CSTMRs, and is used for multienzyme reactions such as the production of high-fructose corn syrup from starch [16]. The use of a serial array of membranes of different pore sizes facilitates the selective removal of individual products when a number of reaction products are present. The disadvantages with these reactors are the cost and the complicated process required.
r
Reactant Reservoir
U Reactant Product 8 Pump 0
A
Product Collection Vessel
Fig. 10-10. Stirred tank reactor with membrane separation.
10.4 Catalytic Membrane Reactors for the Food Industry
241
10.4.2 Membrane Reactor Configuration Membrane configurations compatible with enzyme immobilization include spiral wound [17,18], hollow-fiber [19-221, and tubular membrane arrangements [23]. A special case of a spiral wound format is the axial annular flow reactor, in which the membrane has no separation function and operates solely as an enzyme support 1241.
10.4.3 Enzyme Immobilization The term ‘immobilized enzymes’ means the physical localization of the enzyme molecules during a continuous catalytic process. The usual practice is to confine the enzyme to a water-insoluble matrix where it can be recovered for further use [25]. Enzyme immobilization techniques can be divided into the following groups: -
-
Covalent attachment of enzymes to solid supports. Adsorption of enzymes to solid supports. Entrapment of enzymes.
Covalent Attachment to Solid Supports Immobilization by covalent attachment involves fixing the enzyme molecule to the insoluble matrix by at least one covalent bond [25]. A variety of supports has been used, including porous glass, ceramics, stainless steel, sand, charcoal, cellulose, synthetic polymers, and metallic oxides [26]. Enzymes are usually immobilized through their amino or carboxyl groups. In most instances, the immobilization procedure consists of two steps: (i) activation of the support; and (ii) enzyme attachment. Numerous immobilization procedures are aimed at the suppression of protein unfolding. The multipoint attachment of the protein molecule to the surface of a support is the most promising [27]. This type of immobilization has made it possible to slow down by several thousand times both the reversible and irreversible thermoinactivation of enzymes, as well as their reversible unfolding caused by denaturants such as urea or the dissociation of oligomeric enzymes into subunits [28]. A combination of enzyme entrapment and covalent bonding has also been used [29]. This has been called the copolymerization method. It involves firstly modifying the functional groups of an enzyme with an analogue of monomer, and then copolymerizing the modified enzyme with a monomer and a bi-functional cross-linker is produced. A three-dimensional polymeric network is created, at the junctions of which are the enzymes molecules. The technique has been used to stabilize the enzyme Termitase [29].
242
10 Membrane Bioseparation Processes in the Beverage and Food Industries
Adsorption of Enzymes to Solid Supports The appealing feature of adsorption immobilization is its simplicity. An enzyme solution is added to a support, and the system is mixed to allow adsorption. Examples of the use of adsorbed enzymes in catalytic enzyme reactors are shown in Table 10-1.
Entrapment of Enzymes In this approach, pioneered by Chang in 1972 [30], enzymes are enveloped within various forms of membranes. These membranes are impermeable to enzymes and other macromolecules but are permeable to low-molecular weight substrates and products. A typical example is the entrapment of enzymes in hollow fibers. The
Table 10-1. Applications of catalytic membrane reactors. Reactor function
Membrane reactor type
Membrane/ support material
Enzyme
Method of immobilization
Hydrolysis of whey permeate
CSTMR
Chitin
Lactase
Covalent attachment
Hydrolysis of lactose in skim milk
Axial annular flow
PVC silica membrane
Lactase
Covalent attachment
Casein hydrolysis
CSTMR
Polysulfone hollow fibers
Alcalase
Enzyme in solution
Hydrolysis of butteroil
Hollow-fiber
Microporus polypropylene
Lipase
Adsorption
Lipolysis of butteroil
Hollow-fiber
Microporus polypropylene
Lipase
Adsorption
Hydrolysis of plant oils
Hollow-fiber
PTFE membrane
Lipase
Covalent attachment
Hydrolysis of olive oil
Hollow-fiber
Poly amide membrane
Lipase
Adsorptionlcovalent attachment
Hydrolysis of olive oil
Hollow-fiber
Polyamide membrane
Lipase
Adsorption
Concentration of pectin
Tubular
Titanium dioxide
Pectinase Adsorption
Concentration of pectin
Tubular
Titanium dioxide
Pectinase
Adsorption
Glucose production
Tubular
Zirconium hydrous oxide polyacrylate
Glucoamalyase
Adsorption
Sucrose inversion
Tubular
Ceramic membrane
Invertase
Covalent attachment
Ref.
10.4 Catalytic Membrane Reactors f o r the Food Industry
243
Snamprogetti Company in Italy has used penicillin acylase, lactase, and aminoacylase entrapped in hollow fibers [31]. Enzymes have also been entrapped within the porous matrix of hollow fiber ultrafiltration membranes [32,33].
10.4.4 Catalytic Membrane Reactors versus Conventional Bioreactor Systems The advantages of catalytic membrane reactors over conventional bioreactor systems include: The catalyst/enzyme is retained by immobilization and does not contaminate the product stream. It is available for continued operation of the system; immobilization can improve enzyme stability; the removal of products through membranes can improve yields by shifting reaction equilibria; modern membrane materials have been shown to be ideal supports for catalyst/ enzyme immobilization in relation to the vast surface area available for immobilization; the reaction and the physical separation of reactants from the products can be effected in the single membrane operation; membranes can also be seen to be advantageous in the retention of costly enzymatic cofactors [34,35]; removal of enzyme inhibitors as a reaction proceeds; advanced separation technology provides the opportunity to work in the gas or liquid phases with a variety of driving forces; aqueous and organic solvents can be used, or both in the case of biphasic catalytic membrane reactors [36]; better process control, higher productivity and better product specificity; and reduced production costs by savings in enzyme usage and process costs when compared with batch reaction systems. A number of potential problems may arise if operating parameters are not carefully selected. These problems lead to a reduction in reactor performance, and can be attributed to either a loss of catalytic activity or a reduction in membrane flux. The optimum conditions for enzymatic reactions are usually established by a series of experiments in a batch reaction system. When enzymes are immobilized, their conformation may be slightly altered, thus leading to a change in parameters such as optimum pH and temperature, or a change in the rate of reaction. These parameters must be carefully investigated in order to ensure optimal utilization of the reactor. Enzyme leakage from the unit may occur even if membrane size is designed to prevent loss. This leakage leads to a gradual reduction in activity. Leakage of cofactors may also occur due to their small size. In order to prevent a reduction in activity it may be necessary to add extra cofactor. The choice of membrane material may also affect free enzyme activity, as the membrane materials may poison the enzyme or unwanted adsorption may occur leading to a decrease in enzyme activity.
244
10 Membrane Bioseparation Processes in the Beverage and Food industries
Accumulation of substrates and products in a gel layer adjacent to the membrane can cause enzymatic inhibition.
10.4.5 Food-Based Applications of Catalytic Membrane Technology Most of the enzyme processes currently used in the food industry are based on either batch or CSTR systems. With a few exceptions, industry has been slow to investigate the potential applications of catalytic membrane reactors. It is expected that with the new developments in membrane technology and immobilization techniques, the potential of catalytic membrane reactors can be realized. A number of reactors have been developed on a pilot scale and in time these applications may be scaled up to a commercially viable industrial scale.
10.4.6 Catalytic Membrane Reactors in Dairy Processing Enzymes are in common use in dairy processing, for the production of milk protein products and cheeses. They are used in batch-type reaction systems, which can be costly in relation to enzyme usage. In the past 10 years, some potentially novel applications of catalytic membrane reactor technology in the dairy sector have emerged. The production of hydrolyzed milk protein is an area of considerable interest [37], as milk protein hydrolysates are employed as a constituent of baby foods, and in foodstuffs designed for protein intolerants or postoperative care patients. However, the main use for the products is as a food ingredient, as they have various functional properties such as mineral absorption, improving whipping and foaming properties, and as texturizing agents. The hydrolysis of casein in a CSTMR employing hollow-fiber ultrafiltration membranes has been described [15]. The enzyme utilized in this reactor was alcalase, a serine protease derived from Bacillus licheniformis. When an experimental CSTMR system was compared with an equivalent batch system, an improved yield of >90 % was observed compared with 50-60 % for the control batch system. The productivity of the CSTMR system was more than ten times greater than that of the batch reactor under equivalent conditions. It must be noted however that there was a significant reduction in enzyme activity over time in the CSTMR which was attributed to membrane poisoning. Of commercial interest is the enzymatic hydrolysis of lactose in both skim milk and whey solutions. Lactose intolerance is a medical condition whereby an individual cannot, or has a limited ability to, digest dietary lactose. This leads to associated medical problems such as diarrhoea and stomach cramps. Recent investigations have highlighted that this condition is quite prevalent, with more than 70 ‘70of the world’s population being lactose intolerant to some degree. Consequently, there has been increased interest in the production of hydrolyzed milk products for lactose intolerants. Hydrolysis of lactose in skim milk by immobilized P-galactosidase (lactase) has
10.4 Catalytic Membrane Reactors for the Food Industry
245
been carried out using an axial annular flow reactor [24]. The lactase enzyme immobilized by covalent attachment, originated from Bacillus circulans. Retention of enzyme activity over a prolonged period of operation was satisfactory within this system. The cheese manufacturing industry produces large amounts of whey as a by-product. For many years, whey was considered to have no commercial value, and its disposal was a problem since it ranks as a serious pollutant. However, the advent of new technologies described earlier have turned a potential pollution problem into a valuable resource for the dairy industry. The two main solid components of whey are lactose (75 %) and whey proteins (6 %). When the lactose is hydrolyzed it produces a sweet syrup containing mostly glucose and galactose. This syrup has a number of applications. It is used in bakery products, in confectionery products, in the production of ice cream, and in the manufacture of spreads, dressings, and soft drinks. Conventional acid hydrolysis has the disadvantage that it is accompanied by Maillard browning reactions, particularly if the substrate contains protein. The development of commercial lactase enzymes has led to enzymatic methods for whey hydrolysis. These methods are based on batch systems, with ultrafiltration for enzyme recovery after initial protein removal. Alternatively, immobilized enzyme systems can be employed. Whey has been ultrafiltered for protein recovery and the resulting permeate passed through a CSTR system that contained lactase immobilized by covalent attachment for the hydrolysis of the lactose [38]. This system produced interesting results, indicating that the reactor performed better when partial concentration of the whey feed was carried out before hydrolysis. An economic analysis of this system showed it to have potential. A commercial immobilized reactor system for skim milk and whey hydrolysis has been in use for a number of years in the Valio Dairy in Finland. The use of immobilized reactor systems coupled with the new separation methods for whey protein recovery have reduced dairy waste problems and added to the wide range of food ingredients produced by the dairy industry. The development and production of dairy flavor concentrates based on the wide range of fat components in milk and milk products is of great interest. Lingual lipases (pregastric esterases) are used in the dairy industry to impart flavor characteristics to dairy products (e.g. cheeses). They work by releasing branched-chain fatty acids from monoglycerides, diglycerides, and triglycerides. The action of this enzyme imparts desirable flavor notes to the products concerned. The lipase enzymes used are from both animal and microbial sources. Lipases from both sources have been used in catalytic membrane reactors for the production of hydrolyzed butter oil. A lipase from Aspergillus niger has been immobilized by adsorption in a hollow-fiber catalytic membrane reactor to produce hydrolyzed butter oil [20]. Time scales were adequate to justify commercial interest. The lipase was not stable over long periods, but it had a half-life (t%)greater than that of free lipase. Lingual lipases from a variety of sources were immobilized by physical adsorption in a hollow-fiber catalytic membrane reactor for the continuous hydrolysis of butter oil [ 191. Bovine, caprine, and ovine lipases were immobilized in separate experiments in the reactor. All three enzymes demonstrated no signs of decreased activity after 5 days of continuous operation at 40°C. The tlh of these isozymes were double those observed
246
10 Membrane Bioseparation Processes in the Beverage and Food Industries
for microbial enzymes, 14 days versus 7 days respectively. The other advantage of lingual lipases is that they preferentially liberate short-chain fatty acids, C ~ - C ~ O , from the target molecule. These short-chain fatty acids contribute desirable flavors. Microbial lipases tend to release long-chain fatty acids, producing soapy notes.
10.4.7 Catalytic Membrane Reactors in the Hydrolysis of Fats and Oils Fatty acids and glycerol produced by the hydrolysis of fats and oils are employed by the food and pharmaceutical industries for a variety of purposes. Traditional chemical hydrolysis is energy-intensive and this has led to the investigation of energy-efficient enzymatic hydrolysis by lipases. Furthermore, the use of lipases makes the process amenable to catalytic membrane technology. The efficacy of such technology in the hydrolysis of plant oils has been assessed using a lipase from Rhizopus nigricans which was immobilized by covalent attachment with glutaraldehyde on a PTFE membrane [39]. A lipase from Candida cylindracea was used in the evaluation of two catalytic membrane reactors for the production of fatty acids and glycerol [22]. The study compared a hollow-fiber membrane reactor where the olive oil substrate was in an oil-in-water emulsion, with a biphasic membrane reactor system with two separated liquid phases of oil and water. Of the two systems, the biphasic reactor showed a higher specific activity, and a constant specific rate over time. The fact that the two products were separated after the reaction was an added bonus. The study also compared adsorption and covalent attachment as immobilization techniques and found that covalent attachment was superior in terms of enzyme retention. In a separate study, the hydrolysis of olive oil in a biphasic membrane reactor was compared with that of the solubilized enzyme [36]. It was determined that the immobilized enzyme remained stable for a period 16 times greater than the enzyme in solution, and that the best mode of operation was a counter-current flow arrangement. It was observed that the quantity of lipase imnlobilized strongly affected the specific activity of the enzyme in the reactor. The use of catalytic membrane reactors in the hydrolysis of fats and oils has advantages in terms of lower energy consumption, greater process control, and the ability to hydrolyze oils with low water solubility in biphasic reactor systems.
10.4.8 Catalytic Membrane Reactors in Fruit Juice Processing Pectinase aids clarification of fruit juices by breaking down pectin which can cause haze, and it reduces viscosity, thus facilitating juice filtration [40]. Microfiltration and ultrafiltration are widely employed techniques in the commercial production of fruit juices. One of the most common problems encountered with these techniques is the fouling of membrane surfaces. A novel solution to this problem is to immobilize pectinase by adsorption on the membrane in order to prevent fouling, and to
10.4 Catalytic Membrane Reactors for the Food Industry
247
maintain high membrane flux rates. The efficacy of this technique in the concentration of pectin solutions has been examined [ 111. The immobilized pectinase led to an enhanced flux in the reactor system over a range of crossflow velocities. This study also indicated that if correct chemical cleaning procedures were followed, the membrane could be easily restored to its original permeability. A similar study indicated that substantially increased flux rates were achieved when the membrane was operated using a permeate recycle option [41].
10.4.9 Catalytic Membrane Reactors in Carbohydrate Processing One of the first areas where the food industry was quick to realize the potential of the use of enzymes was in the production of sugars and starches. Numerous enzymatic production methods are currently employed. Likewise this sector of the food industry was one of the first to investigate the use of catalytic membrane reactors. The potential applications in relation to carbohydrate hydrolysis have been reviewed [42]. These include polysaccharide hydrolysis, and the hydrolysis and conversion of oligosaccharides. Glucose production from dextrin has been effected using glucoamylase immobilized by adsorption on a zirconium hydrous oxide-polyacrylate nanofiltration membrane [41]. This study indicated that the optimum conditions for the immobilized enzyme were pH 4.0 and 50 "C, as opposed to pH 5.0 and 60 "C for the free enzyme in solution. The dextrose-equivalent (DE) for the permeate from the membrane reactor was ten times greater than that obtained with free glucoamylase over the same period of time. It was found that the rate of glucose production in the reactor increased with substrate concentration, whereas the rate of glucose production with free glucoamylase was limited by the enzyme concentration. The study proposed that glucoamylase immobilized on a membrane could be configured as a continuous hydrolysis reactor for the production of high-DE glucose syrups. A further application of catalytic membrane technology is sucrose inversion in molasses using immobilized invertase [23]. The enzyme was immobilized on a tubular ceramic membrane by covalent attachment with y-aminopropyl trimethoxy silane and glutaraldehyde. The study highlighted the suitability of this reactor for use with substances like molasses that contain impurities. The reactor converted more than 80% of the sucrose in the feedstream to glucose and fructose after 300 h of operation. The study did indicate however that the immobilized enzyme activity was reduced when molasses was used in the reactor instead of a sucrose solution. The enzyme deactivation time in the reactor was determined to be half that of the free enzyme in solution. A study into the production of sorbitol and gluconic acid using a charged membrane bioreactor with a coenzyme regeneration system has been carried out [34]. Sorbitol is produced from glucose using NADPH-dependent aldose reductase. The oxidized form of the coenzyme was enzymatically regenerated by conjugation with glucose dehydrogenase, along with the production of gluconic acid from glucose. The composite membrane used in this study was found to be effective in the retention of the NADPH coenzyme. This reactor system was considered promising with
248
10 Membrane Bioseparation Processes in the Beverage and Food Industries
regard to the production of sorbitol and gluconic acid which are valuable food ingredients, and also in demonstrating that reactions that are dependent on coenzymes can be carried out in membrane reactors. One of the major potential advantages of the use of catalytic membrane reactors in the sugar and starch industry is the options they can provide in relation to control of product composition. If a catalytic membrane reactor is set up for the production of glucose from dextrin on an industrial scale, careful control of the reactor conditions can facilitate the tailoring of product composition.
10.5 Applications in the Beer and Alcoholic Beverage Industries Membrane technology in the alcoholic drinks industry is used as an alternative to established technologies and as such has to compete in terms of both quality and cost of the process. Membranes can be expensive, they have a limited life, and can be difficult to clean. The industry is thus slow in adopting membrane technology. However, the use of membranes has the potential to provide not only novel solutions to processing problems but may also replace existing unit operations.
10.5.1 Removal of Alcohol using Reverse Osmosis The overall flavor impact of beer, or indeed any alcoholic beverage, can be attributed to the combined effect of the alcohol, the volatile flavor components (derived from yeast fermentation and hop addition), and the non-volatile, non-fermentable dissolved solids which are present at the end of fermentation (Fig. 10-11). Most prod-
u
\
Dissolved solids (non-volatile)
/
Fig. 10-11. Flavor balance.
10.5 Applications in the Beer and Alcoholic Beverage Industries
249
ucts have a flavor balance based on the interaction of these components. A change in any of the three components has a significant effect on the flavor of the final product. The popularity of low- and reduced-alcohol products has increased over the years and producers are constantly striving to improve the quality of their products. The main thrust of the work is directed towards the flavor matching of the reduced-alcohol product with the full-alcohol product. The definition of non-alcoholic and lowalcoholic drinks varies from market to market. Non-alcoholic beer (NAB) or alcohol-free beer (AFB) normally refers to products with less than 1 % alcohol by volume (ABV). Low-alcohol beers (LAB) tend to fall between AFB and full-alcohol products in terms of alcohol content. Some markets insist that the term ‘beer’ is replaced with ‘malt beverage’ if the ABV is below certain levels. The production of non-alcoholic or reduced-alcoholic products can be divided into two areas, namely preventing or restricting alcohol formation, or removing alcohol from the final product.
Prevention or Restriction of Alcohol Formation This is carried out in various ways: Use of wort (unfermented beer) in formulating final product. This is more a malt beverage rather than a NAB and has strong malty/worty flavors. - Reduction of the level of fermentable material available to the yeast by using - weak or diluted worts. - worts produced by high-temperature mashing, which gives more high-molecular weight, unfermentable dextrins. - selective membrane separation of the wort to remove some of the fermentables 1431. - Control of fermentation by: - stopping fermentation at the required alcohol level. - using yeasts with limited fermentation capability. - using cold contact of wort and yeast [44]; this produces little or no fermentation, but contact with the yeast is believed to reduce the worty flavors. - extending the cold contact process (based on immobilized yeast technology). In this, the wort is passed through a bioreactor where it is exposed to extremely high yeast concentrations [45,46]. By balancing temperature and contact time, the required alcohol content can be achieved.
-
Removal of Alcohol from a Final Fermented Product A common method of production of NABLAB is the removal of alcohol from a regular alcoholic beer. Although this appears wasteful and rather expensive, starting with a fully fermented product can result in a superior reduced-alcohol final product. The principal methods employed are:
250
10 Membrane Bioseparation Processes in the Beverage and Food Industries
- Distillation/evaporation.This is carried out under high vacuum conditions which allows alcohol to be removed at low temperatures (35-50 "C), thus minimizing any heat damage to the beer. Falling film and spinning cone evaporators are the most commonly used [47]. Evaporation removes both alcohol and water, resulting in concentration of the beer solids. The degree of beer concentration depends on the desired NABLAB alcohol content. The lower the required alcohol level, the greater the degree of beer concentration required on evaporation. This approach is very effective in removing alcohol, but it also removes the volatile flavor fractions, and can potentially cause heat damage to the remaining non-volatile fraction. Systems are available which minimize the degree of concentration while recovering some of the volatile flavor fraction [48]. This recovered fraction can then be incorporated back into the final product. - Membranes. Reverse osmosis and dialysis fall into this category [49,50]. Both of these systems can be operated at low temperatures which minimizes the risk of heat damage. However, it is difficult to remove alcohol selectively without removing a high proportion of the low-molecular weight volatile flavor components. It is also difficult to remove all the alcohol from the feed beer without a high degree of diafiltration. This leads to a need for large volumes of diafiltration water (up to four times the original beer volume). In addition, diafiltration water needs to be demineralized or processed through a reverse osmosis membrane in order to prevent a high concentration of minerals in the final product. For beer, the diafiltration water also needs to be de-aerated, to prevent oxidation and premature staling of the product. These factors make de-alcoholization by membrane technology expensive and improved quality is not always achieved. However, membrane technology is used in the wine industry for producing reduced-alcohol wines and for improving wine quality [51]. In this way reverse osmosis could be used to turn a poor quality, highly acidic wine into an acceptable reduced-alcohol wine by removing some alcohol and acidity. Other methods which can be used for alcohol removal include: adsorption, freeze concentration, and liquid C02 extraction. However, these methods are not currently commercially viable. An interesting development in the use of membranes for low-alcohol beers has been in the area of colorless beer production [ 5 2 ] . The alcohol-rich permeate from the reverse osmosis step, which is a by-product of LAB production, is used as a base to which various additions are made. The additions can include dextrins for body, hop constituents to give characteristic beer bitterness, and natural foamenhancers. Fruit flavors may also be incorporated to give a fruit-flavored malt beverage.
10.5.2 Removal of Microorganisms by Microfiltration Microfiltration is used extensively in the drinks industry for clarification, sterilization, and product recovery. It is the only form of membrane separation which can
25 1
10.5 Applications in the Beer and Alcoholic Beverage Industries
have both dead-end and crossflow applications. It can be used as a replacement for both clarification and sterilization.
Beer FiltratiodClarification Clarification of beer is more than just simple filtration. A major property of 'bright' beers, such as ale and lager, is the clarity, sparkle, and absence of haze in the final product. Brewers are not only concerned with the initial brightness or clarity of their beers but also with the long-term haze stability. Haze in beer is caused by the formation of protein/polyphenol complexes which occurs over time. A major function of beer clarification is the removal or reduction of these potential hazeforming precursors. This is tackled in a number of ways:
Cold filtering. Protein/polyhenol complexes are temporarily formed upon chilling beer. These complexes disappear upon warming. Therefore, filtering the beer while cold will remove some of the potential haze. Adsorption. Beer can be treated with various filter aids prior to filtering. Polyvinylpolypyrrolidone (PVPP) adsorbs polyphenols, and silica hydrogel adsorbs proteins. Both filter aids are then removed by filtration [53].Tannic acid can also be used to precipitate the haze-forming proteins. Proteolytic enzymes such as papain may be used to break down the proteinaceous haze precursors, though this is now less widely used. standard beer filtration system is shown in Fig. 10-12. Microfiltration of beer, as- a means of clarification, needs to address the problem of long-term haze stability and to incorporate some form of pre-treatment, such as those outlined above. This complication results in MF of beer not being commonly used as an alternative to normal filtration, but rather as an alternative to pasteurization. Bright Beer
Unfiltered Beer
b
PVPP or Silica Hydrogel treatment
Fig. 10-12. Standard beer filtration.
Pasteurization Pasteurization is used to stabilize beer microbiologically, by destroying any microorganisms which may spoil the beer. Beer for kegging is normally flash-pasteurized (typically 72 "C for 40 s) en route to the keg filler. The keg is washed and sterilized immediately prior to filling with beer. The beer is then filled into the keg by a closed
252
10 Membrane Bioseparation Processes in the Beverage and Food Industries
fill against an inert gaseous back-pressure. The keg filling process is essentially a sterile filling operation. Small pack beer (bottles and cans) is filled non-aseptically, sealed, and the container and beer are pasteurized in a tunnel pasteurizer with a lower temperature and a longer holding period (typically 60 "C for 20 min). If pasteurization is carried out properly, beer flavor should be unaffected. Microfiltration through a filter with a pore size not greater than 0.4 ym can be used as an alternative to pasteurization. For keg beer, the microfilter directly replaces the flash-pasteurizer and packaging is unchanged. However, for canned and bottled products the use of microfiltration requires that the beer be aseptically filled, necessitating upgrading of the standard filling machines. The use of MF versus pasteurization needs to be closely examined and Table 10-2 illustrates a comparison between these processes. Table 10-2. Comparison of pasteurization versus microfiltration for microbial stabilization of beer. ~~
Pasteurization
Microfiltration
Temperature
High temperature. May cause heat/ flavor damage if done incorrectly
Low temperature
Running costs
Requirement for heating and cooling
Low, but membrane life is limited and membranes are expensive
Cleaning
Daily to prevent scale build-up and sanitization
Daily to regenerate filter and for sanitization
Flow rate
Constant
Will vary throughout run due to debris build-up on filter membrane. This may be a problem if supplying a packaging line
Integrity
Once temperature is achieved, it is effective
Minor leaks may remain undetected. Integrity checks need to be carried out on filters
For economic reasons, MF for microbial beer stabilization is operated mainly dead-end rather than in crossflow mode. Beer for MF is generally processed through the normal beer filtration cycle. This guarantees haze stability, prolongs the life of the MF membrane, and maintains high flux rates through the membrane. By contrast, crossflow MF is used extensively in the wine and cider industries for clarification, as protein/polyphenol haze complexes do not arise in these beverages. Crossflow MF is also used in brewing, cider, and wine making for cell or product recovery as an alternative to centrifugation or settling [54-561. Tun or vat bottoms (lees) contain sediment and yeast material, and also a significant amount of product which is difficult to recover. Crossflow MF is used to recover a clear usable product from what would normally be considered a waste product. Because of the high solids content of the lees, the MF has to be crossflow rather than dead-end. Alternatives to MF are:
10.5 Applications in the Beer and Alcoholic Beverage Industries
253
Centrifugation or settling. The recovered product does not have the clarity of the microfiltered product. - Evaporation and recovery of the alcohol. The alcohol must be either sold as a byproduct or blended back at very low levels into the original or similar products. -
10.5.3 Gas Exchange using Crossflow Filtration A recent development in membrane technology has been the use of membranes for the addition and removal of gases to or from liquids [57,58]. All alcoholic fermentation is accompanied by carbon dioxide evolution. Thus, the fermented product will contain a high level of C02 in solution. If the final product is to be carbonated, then the presence of C02 will be advantageous. For a final product which is a not carbonated (e.g. still wine), the C02 needs to be removed. Whatever the case, the level of C02 will more than likely need to be adjusted (either up or down) in the final product. Addition of C02 is relatively simple and easy to control. It involves injecting the gas into the liquid at the required rate and ensuring adequate mixing to aid dissolution. The rate of addition can be monitored using gas flowmeters and feedback control is achieved using on-line C02 meters. The fact that the gas is physically bubbled into the liquid can lead to foam formation. In beer, foam formation can lead to loss of head-forming potential and bittering compounds. Removal of C02 is less controllable, and various methods used to accomplish this goal include: physical agitation and displacement using another gas (e.g. bubbling with nitrogen). Both these methods are rather inelegant and can generate large amounts of foam. Membranes can simplify both addition and removal of C02, and also offer greater control without foaming. The basic principle has already been described in section 10.2.2. The membranes used are microporous and hydrophobic, and are usually hollow-fiber with the liquid stream on one side and the gas stream on the other. Under normal operating conditions, the membrane should not allow liquid to pass through to the gas side. Gas is either transferred to or removed from the liquid by diffusion and is driven by partial pressure differences between both phases (Fig.10-13). Because the gas transfer is by diffusion, there is no bubbling and thus no foam formation. The addition or removal of gas can be controlled by feedback from a C02 meter to vary liquid flow or gas pressure. This technology can be applied to any gas exchange application, e.g. aeratioddeaeration, or nitrogenation.
254
10 Membrane Bioseparation Processes in the Beverage and Food Industries
L Liquid Phase
4 Liquid Flow
-I+-
I
Gas out
0.05 Micron pores
-I-
Gas in
TMembrane (Hydrophobic)
Gas Phase
Fig. 10-13. Gas exchange membrane.
10.6 Applications of Membrane Separations in Dairy Processing Dairy processing is a low-margin and energy-intensive business. There is considerable interest in replacement of energy-intensive unit operations, such as evaporation, with more energy-efficient techniques. This has been a significant driving force in the adoption of membrane-based processes in the dairy industry. High throughput is essential because of large volumes of milk, an unstable raw material, which have to be processed quickly [59].
10.6.1 Historical Background Membrane processing was introduced in the dairy industry in the late 1960s [60]. The early technologies were applications of both UF and RO to the processing of whey [61]. Over time, the range of applications has expanded considerably and current applications include [62]: waste stream concentration fractionation of milk concentration of protein concentration of UF permeate fat recovery from skim and whey streams diafiltration of whey partial demineralization of whey removal of bacteria from milk
10.6 Applications of Membrane Separations in Dairy Processing
255
Milk Pretreatment using UF To produce cheese to a consistent specification, standardization (adjustment of milk composition) may be necessary. The advantages of using UF to achieve standardization, as opposed to conventional methods, include: -
improved consistency; faster acidification and shorter process time; and more complete and rapid expulsion of whey [61].
Increases in yield may also result and this is thought to be due to the concentration factor achieved during UF. However, retention of the whey proteins normally lost in cheesemaking may account for the 10-12% increase in yield estimated for soft cheese manufacture [62]. Cost savings in rennet and starter culture are additional advantages [60], as is the decrease in day-to-day variation in process parameters [63].
Ultrafiltration in Production of Cheese Conventional cheesemaking equipment may be used with UF-standardized milk, provided that a two-fold volumetric concentration is not exceeded. This gives a very useful increase in output without incurring any additional capital expenditure [6 11. If this concentration factor is exceeded, special cheesemaking equipment is required. It has been claimed that UF-based cheeses are softer but that the water-binding capacity of whey protein causes such cheeses to be more resistant to drying out than traditional cheese [64]. There are a number of plants in operation producing the high moisture Ricotta cheese [63]. APV-Pasilac have installed some 50 plants in the period 1981-1995. Thus, UF technology is well established in the manufacture of fresh cheese [65].
Production of Cheese Base using UF A process which generates a product that could partially replace cheese in some applications is commercially attractive. Cheese base production is similar to cheese production, but without the lengthy and expensive flavor maturation step. Ultrafiltration and diafiltration have been proposed for the production of a cheese base [60]. Diafiltered retentate is pasteurized and salt and starter culture are added. When acidification is complete, the mixture is evaporated to produce cheese base. This process eliminates the need for a specialized cheesevat and minimizes whey production [61].
10.6.2 Whey Processing In the past, whey was seen as a troublesome by-product of cheese and casein processing with a high biological oxygen demand (BOD) and requires disposal. Whey
256
10 Membrane BioseDaration Processes in the Beveraae and Food Industries
Table 10-3. Typical rennet whey composition Component
%
Water Total solids Lactose Protein Ash Fat Lactic acid
93.6 6.4 4.6 0,9 0,7 0.1 0.1
volume is 86-88 % of the original milk volume used in cheese manufacture, thus the total volume of whey is large [62]. A typical rennet whey composition is shown in Table 10-3. Transport costs for such a dilute pollutant make disposal or further off-site processing expensive. Whey processing represents one of the early applications of membrane processes in the dairy industry [61]. European plants typically concentrate from 6.4 up to 24% total solids, while U.S. plants generally do not exceed 12% total solids [66]. Production of Whey Protein Concentrate by Microfiltration Whey protein concentrates are well established as nutritional supplements in formulated foods. With the application of membrane technologies, increased interest is being shown in the functional properties of such whey proteins, functional properties of interest being: emulsifying; foaming; gelling; solubility; heat stability and water uptake [61,66]. When whey is fractionated, there is an increase in the concentration of the functional protein components while most of the lactose is removed [67]. As lipids are known to impair the functional properties of whey proteins, several methods for lipid removal have been developed based on membrane separations [63]. Aggregation of lipoproteins by mild heat treatment followed by removal of these aggregates by microfiltration is but one of these methods [68]. If this process is applied to UFtreated whey retentate rather than whey, results are even better. The preferred pore size is 0.2 pm, as the estimated minimum diameter of lipoprotein complexes is 4 pm [63]. Direct microfiltration of whey using a 0.8 pm ceramic membrane results in an 80% removal of whey lipid. Such pretreatment is especially desirable for whey destined for baby food formulations [63]. Careful control of flow rates and trans-membrane pressures is required to minimize clogging during the process [69]. The major whey proteins, a-lactalbumin and P-lactoglobulin, differ in their functional properties. Fractionation of whey (Fig. 10-14), results in preparations having different functional properties. Admixtures of these protein fractions result in ingre-
10.6 Applications of Membrane Separations in Dairy Processing
257
Defatted whey protein
.1 UF
.1
Acidification and heating
.1
MF
L
Permeate
L
Neutralization
.1 UF
.1
P-Lactoglobulin
Fig. 10 -14. Production of P-lactoglobulin from defatted whey protein.
dients with customized functional properties. As both components are valuable, a combination of processes is used to maximize recovery of functional protein [697 11. Recovery of pure a-lactalbumin is more difficult, and contamination with K-casein glycomacropeptide, which has a number of physiologic and pharmacologic characteristics, is a recognized problem [61]. Whey protein concentrates have been the major application of UF technology in the dairy industry for some time [60]. There is interest in high protein whey protein concentrates as an egg white replacer. Functional protein production is a major growth area in the membrane processing of whey [71]. Such treatment results in the loss of lactose as well as loss of water. The effect is to produce a whey protein concentrate containing 35-80 % protein [67]. So-called 35 % whey protein concentrate has a protein/total solids ratio of 35, while 80 % whey protein concentrate has a proteidtotal solids ratio of 80. Such products are currently available from several processors [69].
Electrodialysis and Nanofiltration of Whey Whey which is high in salt is produced during manufacture of cottage cheese, quarg, and casein. It is also produced after the curd salting stage in Cheddar cheese production. Salt impairs functionality of whey proteins [61], and nanofiltration has been used to redress this. Salt removal from 84% to 95 % has been reported [69]. Large ions such as calcium and magnesium are retained to a greater extent than smaller ions such as sodium [62]. High sodium levels are particularly undesirable in baby food formulations, as babies are susceptible to natremia, and to ameliorate this a process for demineralizing and neutralizing acid-casein whey was developed [72]. Specific rejection of particular ions is variable for different membranes. Rejection rates of between 10% and 67% for sodium ions have been quoted [73]. Generally 2045 % demineralization may be anticipated. Using electrodialysis, demineralization rates of 50-80 % are attained, with the preferential removal of monovalent ions [73]. An economic analysis of demineralization of whey by nanofiltration versus electrodialysis has shown that for 5 3 2 % demineralization, nanofiltration is the most
258
10 Membrane Bioseparation Processes in the Beverage and Food Industries
cost-effective operation, whereas for >32 % demineralization electrodialysis is the operation of choice [74]. Where 290 9% demineralization is required, a combination of nanofiltration and electrodialysis plant is the most economically attractive option. Nanofiltration of whey UF permeate to remove minerals prior to lactose production results in considerable cost savings, as well as increased yield [63]. A mass-transfer model applicable to nanofiltration of whey has been developed [75]. Investigations show that it is acceptable to regard a complex multi-component salt system such as UF-whey as a three-component system with monovalent and divalent cations, and equivalent anion charges, for the purposes of evaluating membrane performance and predicting salt transport using the extended Nernst-Planck equation.
10.6.3 Ultrafiltration of Milk for Ice Cream Manufacture A typical ice cream formulation is shown in Table 10-4 [76]. The dissolved solids, derived principally from the milk solids non-fat (MSNF) and added sugar largely determine the freezing point of the ice cream and thus the hardness at any particular temperature. Standard sources of MSNF, e.g. skim milk powder, provide protein and lactose. If ultrafiltered milk retentate is used, it is possible to vary protein and lactose independently in the final product. The water-binding capacity of the extra protein reduces the amount of stabilizer to be added. Such ice creams are harder [76]. Lactose has long been known to cause ‘sandiness’ by formation of hard crystals [77]. Use of ultrafiltered permeate reduces this problem. In addition, it ought to be helpful to lactose-intolerant consumers. Table 10-4. Typical ice cream formulation. Component
%
Fat Milk solids non-fat (MSNF)
10 11 13 0.2 0.5 65.3
sugar Stabilizer Emulsifier Water
10.6.4 Microfiltration of Raw Milk In the past, liquid milk has been heat-treated to reduce the number of viable bacteria present. Early work with microfiltration of milk showed that fat separation prior to membrane processing was desirable. The Alfa-Lava1 Bactocatch system separates raw milk into skim and cream, subjects the cream plus the retentate from UF of
10.7 Applications of Membrane Separations to the Fruit Juice Industry
259
the skim to a very high-temperature - short-time heat-treatment, before recombining these with the permeate from the microfiltration of the skim [78]. The heat-treated retentate is used for animal feed when raw milk cheesemaking is employed. Pasteurization is still necessary with this system as most of the milk has not been heated to normal pasteurization temperature and thus would fail the phosphatase test which is the standard check for pasteurization [62]. The process cannot guarantee 100 % removal of pathogenic bacteria; hence the need to pasteurize the end-product [79]. The process may also affect milk composition [80,81]. However, the process extends the storage life of milk at 5 "C to at least 4 weeks, thus minimizing shopping frequency for the consumer [81]. In addition to microflora, somatic cells are removed from milk by microfiltration [79]. Microbial contamination of the brine used in cheesemaking is of considerable concern. Heat treatment has been the usual remedy; however, microfiltration is an obvious alternative. Ultrafiltration at high temperature operation, e.g. 50 "C, results in membrane fouling due to precipitation of calcium phosphate crystals. However, UF at ambient may offer an alternative. Bacterial retentions of 2 9 9 % can be expected using such a system.
10.7 Applications of Membrane Separations to the Fruit Juice Industry Eight hundred million liters of fruit juice were sold in the UK alone in 1994. This indicates that fresh fruit juice concentration by reverse osmosis is potentially one of the largest applications of membrane technology [82]. Traditional methods of fruit juice processing tend to apply heat to destroy microorganisms, inactivate enzymes, and achieve concentration. The disadvantages of such processing methods include [83]: loss of cloudiness due to inactivation of pectinmethylesterase (for fruit juices, cloudiness is associated with 'natural' in consumers' minds); - loss of volatile aroma topnotes in evaporation; - oxidation and Maillard browning reactions impacting negatively on both color and flavor; and - the huge cost of energy to operate such processes -
Membrane processes can overcome all these diffculties.
10.7.1 Apple Juice Initial studies on the concentration of apple juice by reverse osmosis showed that the extent of concentration was limited to 20-25 "Brix, and pectins tended to foul the membranes [go]. The solution to these difficulties is to depectinize the juice using
260
10 Membrane Bioseparation Processes in the Beverage and Food Industries
enzymes, followed by clarification using UF, in which case up to 60 "Brix can be achieved. Apple juice production is the biggest single user of membrane technology in the juice industry [67]. In addition to eliminating heat damage, microorganisms are removed, enzyme usage is minimized since enzyme is retained in the juice processing system, and tannins are removed which enhances both color and flavor. The number of processing steps is reduced resulting in higher juice yields (95-98 9%) than are attainable with traditional methods (90-93 %). UF-processed juice tends to darken over time due to protein/polyphenol interaction [84] and product manufactured using ceramic microfiltration membranes is even darker in color, as the larger pore sizes remove even less material [85]. The consumer is likely to perceive this negatively. Retention of sugars, acids, and flavors is excellent with modem membranes [66]. Depectinization of apple juice is an ideal application for spiral wound membrane modules and 'sparkling clear' juice is achieved by the addition of finings such as diatomaceous earth. Significant savings are possible using UF [86]; however, membrane-based processes alone are incapable of producing turbidity-free juices as low-molecular weight substances such as polyphenols are not removed [87]. Use of polyvinylpolypyrrolidone (PVPP) is advocated rather than conventional fining agents, because of its effectiveness, safety record, and lack of impact on color. Crossflow membrane plants are used for their removal. Back-washing is effective at maintaining flux rates using ceramic membranes, but has little impact using carbon membranes. However, the permeate tends to be more turbid as a result [88]. Pulsating entry flow increases the flux rate, probably by reducing formation of the fouling layer on the membrane surface. These conclusions do not appear to be universally valid. Under optimal conditions, periodic back-flushing is found to have little impact on flux rates, particularly at higher velocities [89]. Quoted flux rates vary by a factor of four [88].
10.7.2 Citrus Juice Citrus (orange, lemon, grapefruit) flavor compounds reside in an oil phase present in the juice as an emulsion. As the compounds are sparingly soluble in water, flavor retention in membrane-processed citrus juices should be enhanced in comparison with apple juice. Early application of RO to production of citrus juices, with particular reference to orange juice, has been reviewed [80,90]. Initially, RO was used as a means of boosting evaporator capacity, and the quality of the resultant product was considered acceptable. Ion-exchange chromatography combined with membrane separation processes results in reduced-acid citrus juices [83]. There is clear evidence of an increase in the ratio of "Brix to titratable acidity. However, without organoleptic analysis, the full practical implications of this remain unclear. Concentrate from the 'Fresh Note' process, a process in which UF separates the solids from the flavor-containing serum, has achieved high flavor scores [80]. The UF retentate is pasteurized and the
10.8 Applications of Membrane Separations to Cereal Processing
261
permeate is further concentrated through a series of high- and low-rejection membranes. The unpasteurized, concentrated, permeate stream is then blended back with the pasteurized retentate stream, producing a high-quality product.
10.7.3 Tomato Juice Processing Tomato juice presents a major challenge to the use of reverse osmosis [90]. Tubular membranes overcome problems associated with the high fiber content of this juice. The use of high- and low-rejection membranes, as applied successfully to citrus juices, is not successful for tomato juice as the reblended pulp and serum separate on standing [80]. Nonetheless, there is an acceptable process in operation based on reverse osmosis with good retention of flavor-active components and an absence of the browning normally associated with evaporated tomato juice.
10.8 Applications of Membrane Separations to Cereal Processing Cereal grains are the principal agricultural output on a global scale. As well as beer and spirits, cereals may be processed into a wide range of other products; flour and flour products, syrups, etc. If cereals are wet-processed, there is considerable scope for the application of membrane technology.
10.8.1 Corn Syrups The standard technique for the production of high-fructose corn syrup from starch is to produce a starch slurry, saccharify it with amylase or glucoamylase, isomerize the glucose to fructose using glucose isomerase, and separate chromatographically to produce 90% fructose syrup. This may then be suitably blended to give the target fructose content. A series of potential applications of membrane techniques in the corn refining industry has been identified [91]: -
-
-
clarification of syrups recycling of enzymes separating oligosaccharides from glucose concentrating starch or modified starch solutions purification of solvent (water) oil recovery effluent treatment
262
10 Membrane BioseDaration Processes in the Beveraae and Food Industries
Mineral membranes are better adapted to the needs of corn refiners, being more tolerant of the processing conditions, with minimal sugar losses and resistance to steam sterilization. Very low sugar losses are claimed with such systems. The chromatographic separation step requires vast amounts of water. If a complexing agent is introduced into the system, it would form a large polymer-fructose complex which could be separated by membrane methods [92]. An additional step is required to dissociate the polymer from fructose, followed by separation. The so-called ‘sweetwater’ traditionally has been evaporated from 5 % sugar to 30 % sugar. Counter-current RO is capable of achieving this concentration. This membrane separation technique uses solutions of high osmotic pressure on both sides of the membrane, thus allowing the system to run at lower differential pressure.
10.8.2 Protein Recovery While carbohydrate is the predominant component of cereals, they may also have significant protein levels. Barley grain contains up to 1 2 % protein by weight. Wheat flour typically contains 12-13 9% protein. Wheat gluten has a range of uses in food processing. It may be added to ‘weak’ flour to enhance its breadmaking properties. It is a suitable substitute for casein in some formulated foods, and when added to soya and oilseed flours it significantly improves their nutritional value. Unfortunately, gluten prepared by the more conventional processes can have undesirable cereal flavors, thus limiting its applications. A process based on UF technology for production of a dried gluten-product free of most of the undesirable flavors, colors and salts of earlier processes has been described [93]. Ultrafiltration/diafiltration using a 30 kDa molecular weight cut-off membrane was shown to be the key step. Previously there was little interest in corn gluten as its lysine content and thus its nutritional value was low. Recent improvements in both lysine and protein levels have changed this situation somewhat. A membrane-based process to recover this protein has been described [94]. This involves alkali extraction of the ground corn followed by pre-filtration and ultrafiltration. Protein in the retentate is precipitated and spray-dried, and permeate is passed to the RO system to concentrate salts, sugars, and other components, allowing the RO permeate (purified water) to be reused.
10.8.3 Waste Stream Treatment Typical waste streams from wet processing of cereals have a sufficiently high BOD to make disposal expensive. They are too dilute for use as animal feed. An example is production of ethanol (biofuel) from wheat. The ethanol, produced by fermentation, is stripped from the cereal mash. The liquid remaining, called thin stillage, is typically evaporated to produce syrups. A combination of UF and RO can be used
10.9 Future Trends
263
to produce a syrup-like concentrate and an aqueous permeate reusable as water since UF/RO is 67 % less expensive than the conventional evaporation process [92]. Wet milling of corn is the standard method of separating the protein from the starch in corn to produce cornflour. Vast amounts of steepwater are produced as a waste stream, requiring enormous amounts of energy for evaporation. This evaporation step could be replaced by RO [91]. Whereas 110 kJ kg-I water removed is required for RO, evaporation requires 700 kJ kg-* water removed, even in the more efficient plants. The RO permeate would be available for reuse as water without any added cost.
10.9 Future Trends In food and beverage markets use of membrane technologies will be sparked by tougher standards imposed by the Food and Drug Administration (FDA) and other regulatory bodies. In order to minimize waste, an increased emphasis will be placed on the recovery of energy and water from effluent. Along with the recycling of CIP solutions, such initiatives, initially designed to minimize waste, will ultimately lead to cost savings. Further cost saving to industry will result from the use of membrane technologies in conjunction with classical unit operations so as to extend the lifespan of existing plant. In the area of ingredient production UF-generated fractions will no doubt continue to be a growth area. In the beverage area, the use of gas exchange membranes for gassing and degassing applications will generate significant interest. Catalytic membrane reactors hold possibilities for the food and beverage industries and interest in this area should continue to grow.
Acknowledgments The authors wish to thank Dr. Eddie Collins for reviewing the text, and Eithne Mooney and Marie Mooney-Hynes (GBW Library and Information Services) for their help in the preparation of the manuscript. We gratefully acknowledge the assistance of Steve Birkin, Memtech UK Ltd. for helpful discussions and permission to reproduce illustrations.
264
10 Membrane Bioseparation Processes in the Beverage and Food Industries
References [ 13 Freedonia Group Inc., Membrane Separation Technologies to 1998, Ohio, USA. 1994. [2] Kulkami, S . S . , Funk, E.W., Li, N.N., in: Membrane Handbook: Ho, W.S.W., Sirkar, K.K. (Eds). New York: Van Nostrand Reinhold, 1992. [3] O’Shaughnessy, C., McKechnie, M., The Brewer 1996, 82 105-110. [4] Kritzsche, A.K., Kurz, J.E., in: Handbook of Industrial Membrane Technology: Porter, M.C. (Ed.). New Jersey, USA: Noyes Publications, 1992. NCel, J., in: Proc 4th lnt Conf Pervaporation Processes: Bakish, R. [5] Gref, R., Nguyen, Q.T., (Ed.). Englewood, USA: Bakish Materials Corporation, 1989; pp. 495-509. [6] Beaumelle, D., Marin, M., Gilbert, H., Trans I Chem E 1993, 71, 77-89. [7] Karlsson, H.O.E., Tragirdh, G., Trends Food Sci Technol 1996, 7, 78-83. [8] Drioli, E., Giorno, L., Chemistry & Industry 1996, I , 19-22. [9] Drioli, E., Natoli, M., Koter, I., Trotta, F., Biotechnol Bioeng 1995, 46, 415-420. [lo] Lombardi, C.R, Urso, A., Careddu, G., Ghirlanda G., Catapano G., Brisinda, G., Ceriati, F., Bellantone, R., Doglietto, G.B., Crucitti, F., Int J Artif Org 1992, 15, 126-130. [ l l ] Szaniwski, A.R., Spencer, H.G., Recent Progres des Genie Procedes 1992, 6 (22), 383-388. [12] TragArdh, G., in: Food Processing: Recent Developments: Gaonkar, A.G. (Ed.). Amsterdam: Elsevier Science B.V., 1995; pp. 87-111. [I31 Nwuha, V.O., Int J Food Sci Technol 1996, 31, 27-36. [14] Marshall, A.D., Munro, P.A., TrlgArdh, G., Food Bioprod Process 1996, 74, 92-100. [15] Mannheim, A., Cheryan, M., J Food Sci 1990, 55, 2, 381-390. 1161 Cheryan, M., Mehaia, M.A., in: Membrane Separation in Biotechnology: McGregor, W.C. (Ed.). New York & Basel: Marcel Dekker, 1986; Chapter 10. [17] Garcia, H., Malcata, EX., Hill, C.G., Amundson, C.H., Enzyme Microb Technol 1992, 14, 535-545. [18] Bouhallab, S . , Molle, D., Leonil, J., Biotechnol Lett 1993, 15, 697-702. [ 191 Garcia, H.S., Qureshi, A,, Lessard, L., Ghannouchi, S . , Hill, C.J. Jr., Lebensmittel- Wissenschaft & Technologie 1995, 28 (3), 253-258. [20] Malcata, F.X., Hill, C.J. Jr., Amundson, C.H., Biotechnol Bioeng 1992, 39, 984-1001. [21] Sims, K.A., Cheryan, M., Biotechnol Bioeng 1992, 39, 960-967. [22] Molinari, R., Santoro, M. E., Drioli, E., Ind Eng Chem Res 1994, 33, 2591-2599. [23] Nakajima, M., Nishizawa, K., Nabetani, H., Bioprocess Eng 1993, 9, 31-35. [24] Bakken, A X , Hill, C.J., Amundson, C.H., Biotechnol Bioeng 1992, 39, 408-417. [25] Zaborsky, G.R., in: Immobilized Enzymes, Cleveland: CRC Press, 1973. [26] Axen, R., in: Insolubilized Enzymes: Salmona, M., Saronio, C., Garrattini, S . (Eds.). New York: Raven Press, 1974. [27] Melik-Nubarov, N.S., Mozhaev, V.V., Siksnis, S . , Martinek, K., Biotech Lett 1987, 9, 725. [28] Gabel, D., Eur Biochem 1973, 33, 348. [29] Mozhaev, V.V., Biotech Techniques 1990, 4 , 255. [30] Chang, T.M.S., in: Artifcal Cells. Springfield, 111: Thomas, 1972. [3 11 Marconi, W., Morisi, F., in: Applied Biochemistry and Bioengineering: Wingard, L.B., Katchalski, E., Goldstein, L. (Eds.). New York: Academic Press, 1979, Vol 2; p. 219. [32] Sepracor (1978) Marlborough International Patent, WO 87 02381. [33] Sepracor (1989) Marlborough International Patent, WO 89 04784. [34] Ikemi, M., Koizumi, N., Ishimatsu, Y., Biotechnol Bioeng 1990, 36, 149-154. [35] Rothig, T.R., Kulbe, K.D., Buckmann, F,, Carrea, G., Biotechnol Lett 1990, 12, 353-356. [36] Giomo, L., Molinari, R., Drioli, E., Bianchi, D., Cesti, P., J Chem Tech Biotechnol 1995, 64, 345-352. [37] Savaiano, D.A., Levitt, M.D., J Dairy Sci 1987, 70, 397-406. [38] Illanes, A., Ruiz, A., Zuniga, M.E., Aguirre, C., O’Reilly, S . , Curotto, E., Bioprocess Eng 1990, 5, 257-262.
References
265
Rucka, M., Turkiewicz, B., Zuk, J.S., Krystynowicz, A., Galas, E., Bioprocess Eng 1991, 7, 133-135. Grassin, C., Fauquembergue, P., in: Industrial Enzymology. Godfrey, T., West, S . (Eds.). London: Macmillan Press Ltd., 1996; pp. 225-264. Wang, H.J., Thomas, R.L., Szaniawski, A.R., Spencer, H.G., J Food Process Eng 1994, 17, 365-381. Mehaia, M.A., Cheryan, M., in: Biotechnology and Food Process Engineering: Schwartzberg, H.G., Rao, M.A. (Eds.). New York, Basel: Marcel Dekker, 1990; pp. 92103. Gresch, W., US Patent 5510 125-A, 1996. Schur, F., UK Patent GB 2 112619B, 1993. Lommi, H., Brewing and Distilling International 1990, 21 (5), 22-23. Breitenbucher, K., Brewing and Beverage Industry International 1992, 1 , 28-3 1. Stein, W., MBAA Tech Quarterly 1993, 30 (2), 54-57. Kern, M., Brewing and Beverage Industry International 1994, 2, 17-21. Goldstein, H., Cronan, C., Chicoye, E., US Patent 4612 196, 1986. Von Hodenberg, G.W., Brauwelt International 1991, 2, 145-148. Vradis, I., Floros, J.D., in: Food Flavours, Ingredients and Composition: Charalambous, G. (Ed.). Amsterdam: Elsevier Science Publishers, 1993; pp. 501-520. Tripp et al., US Patent 005439699A, 1995. McMurrough, I., Kelly, R., Byrne, J., O’Brien, M., J Am SOC Brewing Chemists 1992, 50, 67-76. Petersen, B., Brewing and Distilling International 1996, 27, 20-21. Lenoel, M., EBC Monograph, XVI, Symposium Separation Processes, Leuven, 1990, 128-139. Bergin, J., personal commucation. Gill, C., The Brewer 1997, 83 (987), 77-84. Breitschopf, F., Dittrich, S., Koukol, R., Brauwelt 1996, 17/18, 819-821. Henis, J.M.S., J Cell Biochem 1993, 17, 40. Ostergaard, B., in: Concentration and Drying of Foods: McCarthy, D.A. (Ed.). Amsterdam: Elsevier, 1986; pp. 133-145. Rosenberg, M., Trends Food Sci Technol 1995, 6 , 12-19. Bird, J., J Soc Dairy Technol 1996, 49, 16-23. de Boer, R. (Ed.), New Applications of Membrane Processes, Brussels: Inc., International Dairy Federation, 1992. Minstry, V.V., Maubois, J.-L., in: Cheese: Chemistry, Physics and Microbiology: Fox, P.F. (Ed.). London: Chapman & Hall, 1993; Vol. 1, 2nd edn; pp. 439-522. Ottosen, N., International Dairy Federation Bulletin No. 311, 1996, Brussels, International Dairy Federation, pp. 18-20, Pepper, D., Desalination 1990, 77, 55-71. Anon., Food Technol 1990, September, 108-11 3. Maubois, J.-L., Ollivier, G., in: International Dairy Federation Special Issue No. 9201, 1992, IDF, Brussels; pp. 15-22. Anon., Dairy Foods 1992, February, 47-48. Kelly, P.M., Horton, B.S., Burling, H., in: International Dairy Federation Special Issue No. 9201 - New Applications of Membranes, 1992, Brussels, International Dairy Federation; pp. 130-140. [71] Short, J., in: Bioseparation Processes in Foods: Singh, R.K., Rizvi, S.S.H. (Eds.). New York: Marcel Dekker, 1995; pp. 333-350. [72] Giingerich, Ch., Hutson, G., in: International Dairy Federation Bulletin No. 311, 1996, Brussels, International Dairy Federation; pp. 11-13. [73] Jelen, P., J Inst Can Sci Technology Aliment 1991, 24, 200-201. [74] Kelly, J., Kelly, P., J Soc Dairy Technol 1995, 48, 20-25.
266
10 Membrane Bioseparation Processes in the Beverage and Food Industries
[75] van der Horst, H.C., Trimmer, J.M.K., Robbertsen, T., Leenders, J.J., J Membrane Science 1995, 104, 205-218. [76] Garcia-Nevarez, H., Wade V.N., Dairy Food Intl 1993, 58, 35-37. [77] Rothwell, J., Ice Cream Making: A Practical Booklet, Reading College of Estate Management, Published by the author at Reading, U.K., 1985. [78] Pedersen. P.J., in: International Dairy Federation Special Issue No. 9201, 1992, IDF, Brussels, pp. 33-50. [79] Maubois, J.-L., Proc XXIII International Dairy Congress, 1990, Montreal, International Dairy Federation; pp. 1775. [80] Cheryan, M., Act Rep R & D Assoc 1992 44 (l), 164-181. [81] Hoffmann, W., Klobes, H., Kiesner, Chr., Suhren, G . , Krusch, U., Clawin-Radecker, I., Larsen. P.H., in: International Dairy Federation Bulletin No. 311, 1996, Brussels, International Dairy Federation; pp. 45-46. [82] BSDA Annual Directory, 1996197, 1996. [83] Koseglu, S . S . , Lawhon, J.T., Lusus, E.W., Food Techno1 1990, December, 90-97. [84] Hohn, A., in: Processed Fruits: Science and Technology: Somogyi, L.P., Ramaswamy, H.S., Hui, Y.H. (Eds.). Lancaster, PA: Technamic, 1996 Vol. 1; pp. 95-116. [85] McLellan, M.R., in: Processed Fruits: Science and Technology: Somogyi, L.P., Ramaswamy, H.S., Hui, Y.H. (Eds.). Lancaster, PA: Technamic, 1996, Vol. 1 ; pp. 97-94. [86] Hackert, R., Sweintek, R.J., Food Proc 1986, 47, 80. [87] Schneider, T., Czech, B., Fruit Processing 1994, 4 , 302-306. [88] Ben Amar, R., Gupta, B.B., Jaffrin, M.Y., J Food Sci 1990, 55, 1620-1625. [89] Padilla-Zakour, O., McLellan, M.R., J Food Sci 1993, 58, 369-374. [90] Pepper, D., Orchard, A.C.J., Merry, A.J., Desalination 1985, 33, 157-166. [91] Lancrenan, X., Theoleyre, M.A., Kientz, G., Int Sugar Journal 1994, 96, 365-367. [92] Koseglu, S . S . , Rhee, K.C., Lusas, E.W., Cereal Foods World 1991, 36, 376-383. [93] Lawhon, J.T., US Patent 4 645 831, 1987. [94] Lawhon, J.T., US Patent 4 624 805, 1986.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
11 Recovery of Biological Products by Liquid Emulsion Membranes P.R. Patnaik
11.1 Introduction The efficient separation and concentration of biological molecules from fermentation broths has been a continuing challenge in the industrial implementation of biotechnological processes. Product separation often forms a substantial fraction (25 % to 75 %) of the total cost of production [l], and the phenomenal growth in investment in separation technologies [2] attests to the importance of downstream processing. Conventional methods such as distillation, centrifugation, crystallization, absorption, and electrophoresis are based on differences between compounds with respect to a single physical property. These methods therefore work efficiently when the property differences between solutes are large and/or high selecivity in a single stage is not required [3]. Their applicability to biological molecules is further limited by the sensitivity of these molecules to heat, pH, shear, etc. Thus, distillation may have to be carried out under subatmospheric pressure (to reduce the boiling points) or the speed of centrifugation may be limited by the possibility of cell lysis. Other methods such as whole cell extraction and ultracentrifugation require prederivatization of the solute; the more efficient chromatographic methods involve pretreatment of the broth, extensive capital and labor investment, and are difficult to scale up [4]. When the solute of interest is present in low concentrations, its physical properties are not very different from those of other solutes, and it is sensitive to the operating conditions, conventional methods offer poor recoveries. In such cases enhanced recovery under mild conditions is possible through chemical complexation with a molecule that binds selectively to the desired solute and transports it. This transport is usually across a thin liquid film that separates the broth from the receiving phase; hence the name liquid membrane. More accurately, the term liquid emulsion membrane (LEM) is used because a surface-active agent is added to create a threephase emulsion in which the donor and receiver phases are separated by the liquid membrane. Since they were first developed by Li [5], liquid emulsion membranes have been applied to a variety of processes: fractionation of hydrocarbons [ 6 ] , recovery of heavy metal ions [7,8], replacement of catalytic processes in solids by liquidphase catalysis [9,10], treatment of disorders in the blood stream [ l l ] , removal of contaminants from wastewater [ 12,131, and extraction of fermentation products
268
I 1 Recovery of Biological Products by Liquid Emulsion Membranes
[ 14-16]. The last two classes of applications involve biological or biochemical molecules and will be the focus of this chapter.
11.2 Principle and Advantages of LEMs The principle of an LEM may be understood from the schematic representation shown in Fig. 11-1. An LEM system contains three phases, which form a fine dispersion. Microdroplets of the innermost phase are contained in larger globules of the membrane phase; these globules are dispersed in the outer (continuous) phase. If the inner and outer phases are organic, the membrane phase is aqueous to ensure immiscibility. This is called an oil-water-oil emulsion and is an emerging technique for cellular and enzymatic reactions carried in non-aqueous media [17,18]. Because biological reactions are commonly carried out in aqueous media, water-oil-water emulsions are more widely used; here two aqueous phases are separated by an organic membrane such as paraffin or kerosene. The principle of solute transport is the same for both. Since the aqueous and organic phases are immiscible, an isotropic, stable emulsion can be produced only by decreasing the surface tension between the liquids. This is one major difference between LEMs and liquid-liquid extraction. Surface-active agents (or surfactants) are added to achieve reductions in the interfacial tensions. Most surfactants in common use are commercial products which are mixtures of several compounds whose exact composition is not revealed. Nevertheless, their main components and relevant properties are specified, and this guides the choice of a surfactant for a particular application. As shown in Fig. 11-1, a surfactant molecule has a hydrophilic head group and a lipophilic tail. The ratio of these two groups, called the hydrophilic-lipophilic balance (HLB), determines the kind of emulsion formed.
MODEL DROPLET
LIQUID
REAL
* /
S SUBSTRATE
P PRODUCT
\SURF ANY ACT DISPERSED PHASE
Fig. 11-1. Schematic diagram of a liquid emulsion membrane. (Reproduced from [22] with permission of Elsevier Science, Inc., New York 0 1987.)
11.2 Principle and Advantages of LEMs
269
If the HLB is less than about 10 (e.g. Span-80, Paranox-loo), a water-oil-water emulsion is formed, while surfactants with larger HLB values (e.g. Tween-SO, Triton X-100) promote oil-water-oil emulsions [ 141. The solute being separated may be transported either inward or outward. For example, in the controlled release of drugs from implanted capsules, the drug is transported outward from the inner microdroplets through a liquid-impregnated polymer membrane to the site where it is delivered in the body [19,20]. In most separation processes, however, the biological solute has to be recovered from a bulk phase; the transport is therefore inward. In either case it is important that only the species of interest be transported from a fluid containing many components. Sometimes good selectivity is possible simply by virtue of the preferential solubility of the desired species [ 12,16,21]. In many applications, however, good selectivity is achieved by using carrier molecules that bind to the desired molecules and ferry them across the membrane phase [15,22-241. The two methods are not mutually exclusive and may sometimes operate simultaneously [25]. At the inner interface, between the membrane and the receiving microdroplets, the solute is released and retained by reaction either with an acid or alkali (in physical transport) or with a counter-ion (in carrier-mediated transport). The fine dispersion of droplets creates a very large surface per unit volume (10003000 m2 m-3 [3]), which allows large fluxes and high productivity. In addition, since the inner aqueous phase has a much smaller volume (about one-tenth [23,26]) than the outer phase, both separation and concentration are achieved simultaneously. In addition, by encapsulating enzymes or cells inside the microdroplets, the transported molecules can be reacted further to other products [22,27]. The rapid conversion of the transported solute also enhances its flux; a 70-fold increase for certain organic acids has been reported [28]. LEM processes are readily scaled from laboratory to pilot or industrial levels. LEM technology has ben commercialized for the clean-up of waste streams from chemical plants in order to recover metals such as zinc, chromium and uranium, and useful organic compounds such as phenol, nitrophenol, and phosphates [29,30]. While the complexities and special requirements of biological processes have made the application of LEMs to fermentations for pharmaceutical products more difficult to scale-up, results at the pilot scale have shown considerable promise [14,3 1,321. This scalability has been demonstrated in different kinds of equipment [4,31,33,34] and for both free LEMs as well as those imobilized in porous solid supports [31-361. Unlike whole cell extraction, ultracentrifugation and chromatographic separation, LEM systems require little pretreatment of the broth. They are robust and their performance is not impaired by the presence of particulates [37] or live bacterial cells. This enables LEMs to be used directly with fermentation broth without prior removal of the biomass [33,38,39].
270
I 1 Recovery of Biological Products by Liquid Emulsion Membranes
11.3 Qpes of LEMs and their Preparation 11.3.1 Emulsion Liquid Membranes Until now we have used the term ‘liquid emulsion membrane’ in a broad sense covering both emulsions and impregnated liquid membranes. Strictly, an LEM is a threephase emulsion of the water-oil-water or oil-water-oil type, and it is prepared in two steps (Fig. 11-2). To make a water-oil-water emulsion, for example, first a two-phase emulsion of the inner aqueous phase in the organic phase is prepared by slowly adding the aqueous solution to a fixed volume of the organic liquid contained in a homogenizer; a high stirring speed of 8000-10000 r.p.m. [27] is maintained so as to generate adequate shear and create a fine dispersion of the aqueous phase. Surfactants and carrier molecules are present in the organic phase and, if further reaction of the recovered solute is desired, the appropriate enzymes and cells should be in the aqueous phase so that they are encapsulated. The two-phase emulsion is then dispersed in the continuous aqueous phase, which may be a fermentation broth. Since a coarser dispersion of water-containing organic globules is desired, the homogenization speed is lower (300-500 r.p.m.). The choice of this speed is important: too low a speed will produce large globules with thick membranes and low specific surface area, and a high speed will rupture the membrane. Normal diameters are 20 to 40 pm for the aqueous droplets and 200 pm to 2 mm for the two-phase globules [14,40]. A proper choice of the diameters is important. If the aqueous droplets are too small, too many of them will be present in each organic globule and the liquid membrane becomes thin and tends to rupture. On the other hand, large droplets result in low surface-to-volume ratios and low fluxes. For
INNER PHASE
MEMBRANE PHASE
EMULSIFICATION
EMULSION
OUTER PHASE
DISPERSION
Fig. 11-2.Method of preparation of an LEM. (Reproduced from [22] with permission of Eisevier Science, Inc., New York 0 1987.)
11.3 Types of LEMs and their Preparation
271
similar reasons, the globules also should be of an optimal size. The sizes of both the aqueous and the organic droplets depend on many factors - stirring speed, membrane composition, surfactants and carriers, and emulsion preparation conditions [4,14,16,22]. The choice of these factors is still as much an art as a science, partly because the compositions of the commercial surfactants and carriers used are often not known exactly and also because the source phase may have many interacting solutes. LEMs of the type described above have drawbacks such as swelling and rupture of the membrane, coalescence of the droplets, limited selectivity, and sensitivity to process disturbances. These features and their remedies will be discussed later. To alleviate some of the problems, alternate configurations of LEMs have been devised.
11.3.2 Immobilized Liquid Membranes The membrane phase is impregnated within the pores of a thin solid support, which is usually a flat sheet or a hollow fiber (Fig. 11-3). Since there is no emulsification, no surfactant is needed; as in LEMs, the carrier is contained in the membrane phase. The raffinate (donor) phase and the extract (receiver) phase flow on either side of the membrane and the mechanism of transport is the same as in LEMs. However, an additional feature governing the rate of transport is the diffusion rate of the carrier-solute complex across the thickness of the porous support. Although this additional resistance might slow down the rate of diffusion, it is more than compensated by reduced rupture and swelling of the membrane and improved selectivity, scalability, and life of the membrane.
Feed solution ,
solution
Fiat porou<solid matrix with immobilized liquid membrane
J
Annular porous cylinder with immobilized liquid membrane
Fig. 11-3. Two configurations of immobilized liquid membranes.
272
I 1 Recovery of Biological Products by Liquid Emulsion Membranes
Details of the immobilization procedure vary with the application. Two representative methods have been described by Bryjack and coworkers. In one application [25], polyacrylonitrile films were swollen in methanol and 1-decanol and then soaked in a M solution of the carrier in 1-decanol. Hollow-fiber membranes were washed with water, swollen by repeated washing with methanol, ethanol, and M solution of the carrier in 1-decanol, and rewashed 1-decanol, soaked in a 5x with water. These membranes were used to study the extraction of phenylalanine hydrochloride. In the second application [41], L- and D-isomers of amino acid hydrochloride were separated by a sulfonated polyethylene/poly(styrene-co-divinylbenzene) membrane. This membrane was washed alternately with 1 M HC1 and 1 M NaOH and left overnight in water. Then it was washed with methanol and ethanol, immersed overnight in 99.8 % ethanol, dried, and soaked in a chiral alcohol.
11.3.3 Contained Liquid Membranes In this configuration the membrane phase is separated from the two aqueous phases by a porous wall at each interface (Fig. 11-4). The walls may be flat sheets or hollow fibers; both configurations are available as modules. Teramoto and coworkers [42] used modules with spirally wound porous ribbons to separate ethylene from ethane by using silver nitrate as a carrier. Hollow-fiber modules are more common, perhaps because of easy availability and operation, and lower pressure drop than in spiral modules. Sirkar et al. [36,43] describe two applications of these modules: (i) separation of phenol from an aqueous solution of phenol and acetic acid; and (ii) recovery of citric acid. In a module containing a bundle of hollow fibers, the feed solution and the strip solution (receiving phase) flow counter-currently through adjacent rows of fibers. The inter-fiber space is filled with the membrane liquid, including any carrier. As usual, the solute of interest is transported from the feed solution, across the membrane, and into the strip solution. Because there is no emulsification and no immobilization, contained liquid membrane systems do not require as much preparatory work as LEMs and immobilized liquid membranes. Another advantage is that, since there are a large number of fibers (typically 200 to 300 [36,43]) for each of the two aqueous phases, small leaks of the membrane into some of the fibers do not seriously affect the performance. Likewise, leakage of the aqueous phase into the membrane phase is easily remedied by replenishing with fresh organic liquid. Contained liquid membrane processes allow stable operation over a long period and independent control of membrane phase pressure, and they do not require pre-equilibration. Compared with LEMs and immobilized systems, there is greater resistance to mass transfer, but this can be overcome by using a dilute acid or alkali in the strip solution [43,44].
11.4 Types of Separations
273
HYDROPHILIC
LIQUID
CARRYING THE FEED SOLUTION
Fig. 11-4. Structure of a hollowfiber liquid membrane module. (Reprinted from [44] by courtesy of Marcel Dekker, Inc. 0 1988.)
11.4 Types of Separations The transport of a solute across a liquid membrane may occur either by virtue of its preferential solubility or partition coefficient between two phases or with the aid of carrier molecules which bind selectively to the solute molecules.
11.4.1 l j p e I. Physical Separation Transport across the liquid membrane is driven by the ability of the desired species to partition into the membrane phase and its diffusion rate through the membrane. In order to maintain a large concentration gradient and to prevent the transported molecules from diffusing back into the source phase, the solute is converted in the receiving phase into a form insoluble in the membrane.
274
I 1 Recoveiy of Biological Products by Liquid Emulsion Membranes
HAc
HAc
IT
AKC
O H .
AC
Outer Phase
Membrane Phase
Interior Phase
Fig. 11-5. Illustration of physical separation: recovery of acetic acid. (Reprinted from [16] by courtesy of Marcel Dekker, Inc. 0 1988.)
A simple change in pH in the receiving (aqueous) phase is often sufficient to change the ionic character and prevent reverse diffusion of the separated solute. Dilute NaOH is commonly used to achieve this. In the separation of acetic acid from waste water [12], the diffusing acid is rapidly converted to the acetate ion, CH3COO-, whose charge prevents it from partitioning back into the membrane phase (Fig. 11-5). Sometimes, the solute may undergo enzymatic reaction and it may be necessary to prevent the product from being transported outward into the source phase (of the solute). In the extraction of phenol derivatives from blood, Volkel et al. [ 111 used encapsulated uridine diphosphoglucoronyl transferase and its cofactor to detoxify the phenolic compounds into their conjugated forms, whose reverse transfer was inhibited by NaOH. The use of L-malic acid to arrest the reverse diffusion of diltiazem, a cardiac drug, recovered from dilute aqueous solution has also been reported [45]; other biological applications are for the controlled release of drugs [19,20], the removal of cholesterol from the blood [46], and emergency removal of an overdose of drug intake [47,48]. Processes based on the Type-I mechanism are simple, inexpensive, and relatively easy to implement, but their selectivity and efficiency are not always good. Moreover, as the separation progresses, the neutralizing or anchoring compound in the inner aqueous droplets of the receiving phase gets depleted earlier in the droplets close to the surface of the organic globule than in those near the center. Hence, during the course of extraction the incoming solute has to diffuse further and further inward before being released and neutralized; consequently, the process becomes diffusion limited and slows down. These limitations are overcome by the use of carriers in the membrane, which gives rise to the Type-I1 mechanism.
11.4.2 Type 11. Facilitated Transport In facilitated transport a carrier compound is included in the intermediate phase; the carrier should be so chosen that it binds only to the desired species even in the presence of other compounds (as in fermentation broth or fluids in the body). Apart from
11.4 Types of Separations
275
improving selectivity, the carrier enables the pumping of solutes against their concentration gradients. The carrier molecules latch on to the solute molecules, traverse across the membrane, and release the solute at the opposite interface. The appropriate choice of the carrier is crucial to the separation process. A good carrier must satisfy certain requirements [3]: (i) the complexation and dissociation rates should be high; useful bond energies are 10-50 kJ mol-l; (ii) there should be no side reactions with other compounds and no irreversible or degradation reactions; (iii) there should be no co-extraction of solvent from the feed phase; and (iv) both the carrier and its complex with the solute should diffuse rapidly through the membrane phase; however, the diffusion should not be so fast that there is lack of sufficient time for complexation. Facilitated transport may be of either of two types: co-transport and counter-transport (Fig. 11-6). In both cases the carrier is a charged species. In co-transport, a selected solute in its ionic from is complexed with the camer and ferried across the membrane. Since the solute is anionic in many systems of interest [15,22], the carrier molecules have to be protonated to enable complexation. In counter-transport, the carrier exchanges one species in the source phase for another species (the counter-ion) in the receiving phase. By this shuttle system the desired solute accumulates in the receiving phase and the counter-ion in the source phase. For example, the extraction of penicillin G from fermentation broth has been carried out by both methods [ 15,49-5 11. For co-transport Amberlite LA-2 (N-lauryl-N-trialkylmethyl amine) and Hoe F2562 (diisotridecyl amine), both from Hoechst AG, were the most suitable; and for counter-transport (also called reactive transport) Adogen 464 (trioctylmethylammonium chloride, MW 404, SERVA) was the carrier and p-toluene sulfonic acid (pTS03-) the counter-ion. Normally, secondary amines are preferred for cotransport and quaternary ammonium salts for counter-transport. counter-tranwort
co-transport
combined tranwort
Fig. 11-6. Facilitated transport mechanisms. (Reprinted from [32] by permission of John Wiley & Sons, Inc. 0 1993.)
276
11 Recovery of Biological Products by Liquid Emulsion Membranes
Sometimes both kinds of transport operate together. In Bryjak et aZ.'s [25] experiments on the transport of phenylalanine hydrochloride, the carriers used contributed to 0-64% of the total flux, the balance being physical transport. Maximizing this fraction is important since it promotes selectivity. In the enzymatic production of L-leucine [40], leucine dehydrogenase catalyses the reductive amination of a-ketoisocaproate. During this process NADH becomes oxidized and has to be regenerated by reaction with formate. Two of the substrates, formate and a-ketoisocaproate, and L-leucine are transported by the carrier, Adogen 464, whereas the third substrate, ammonia, is soluble in the membrane phase (paraffin or kerosene) and passes through by simple diffusion.
11.5 Factors Affecting LEM and SLM Performance 11.5.1 Stability of Membranes Since liquid membranes are thin, mobile, and subject to shear, they may rupture during their preparation and use. To have good separation efficiencies, the membrane should be stable over a range of operating conditions and a long period of time. Many studies [23,52,53] indicate that the composition of the membrane is crucial in determining its robustness. LEM systems have large osmotic gradients, which promote the transport of water along with the solute; this causes swelling and rupture of the membrane [32,54]. Membrane composition and viscosity play a strong role in the degree of swelling. Even though swelling and rupture are less in immobilized and contained liquid membranes - collectively known as supported liquid membranes (SLM) - osmotic gradients are still large and cause leakage of the extractant through the porous supports [55,56]. A high tendency to solubilize water, low interfacial tensions, and high wettability of the polymeric membrane (in immobilized systems) reduce the stability of SLMs [32]. Unfortunately, some of these features are desirable in order to obtain a large specific surface area, integrity of transport and good selectivity [ 16,44,58]. Therefore the composition and thickness of the membrane may have to compromise between these features. Not all studies support the view that osmotic pressure gradients and solubilization of water in the membrane phase cause loss of membrane stability. During the removal of nitrate ions by trioctylmethyl ammonium chloride carrier in organic membranes (decanol or dibutyl phthalate) immobilized in microporous polypropelene membranes, Neplenbrock et al. [56] observed that rupture of the membrane was not dependent on the osmotic pressure difference but on the molecular structure of the carrier and the type of solvent. They suggested that liquid membrane failure was caused by the loss of membrane phase from the support, which depended on the membrane composition but not on water-related swelling. Rupture could be reduced substantially by having a NaCl concentration of 4.0 M in the stripping phase.
11.5 Factors AfSecting LEM and SLM Performance
277
It is important to quantify the rate of rupture. One method is to encapsulate an enzyme in the inner aqueous droplets and detect its activity in the outer aqueous phase as time progresses. The lower the activity in the outer phase, the less is the rupture of the membrane. For the production of L-leucine from a-ketoisocaproate, Makryaleas and coworkers [40] tested the stability of their paraffin and kerosene membranes by encapsulating NADH and detecting its leakage photometrically. Similar studies have been reported for penicillin-G-acylase with kerosene and for a-chymotrypsin with kerosene and cyclohexane [22]. A disadvantage of this method is that the enzyme can become inactivated while diffusing through the membrane, and hence the observed leakage may be lower than the actual leakage [14]. This limitation may be overcome by encapsulating metal ions instead of enzymes. Potassium was used by Meyer et al. [59] in the study of the hydrolysis of 4-acetoxy-cyclopentanone by pig liver esterase, and also by Thien et al. [4] for the transport of L-phenylalanine across a paraffinic membrane. Both studies reported that a proper composition of the membrane could reduce leakage rates to less than 1 % per hour. In some situations a metal ion may permeate through an intact membrane; then the observed concentrations in the outer phase indicate rupture rates greater than the true rates. Gadekar et al. [21] proposed encapsulating two metal ions such that one prevents permeation of the other. In their experiments on the extraction of nitrophenol, Ni and Cu individually permeated kerosene membranes but, when encapsulated together, Cu prevented the permeation of Ni. The rupture rate decreased with temperature, but so did the rate of extraction; this suggests that a judicious choice of temperature is also important. In addition to choosing the membrane phase, its composition and the temperature, methods to improve stability include: (i) maintaining an interface-immobilizing pressure difference opposing the direction of leakage [60]; (ii) modification of the organic and aqueous phases to reduce surface activity and miscibility [61]; and (iii) blocking the leakage by applying a cross-linked gel network in the pores of the support [57].
11.5.2 Membrane Swelling Because the large osmotic gradients across liquid membranes make it impossible to prevent fully the influx of some water along with the desired solute, swelling is an unavoidable phenomenon. Hydration is, however, not the only cause of swelling. Shear or occlusion or any suface active agent, including the carrier, may cause swelling. Swelling increases with the concentration of the surfactant or the carrier [21,23]. As the concentration is increased, the enhanced driving force initially overcomes the negative effects of swelling. Eventually however, swelling becomes significant and causes dilution of the product in the interior (aqueous) droplets (Fig. 11-7). Swelling of the membrane undermines the system performance in many ways: (i) the transported water dilutes the solute accumulated in the inner droplets; (ii) the driving force is reduced and the diffusion path across the membrane is increased;
278
11 Recovery of Biological Products by Liquid Emulsion Membranes 100
I
\’
I
so - 80
- 70 2m +
t1’ .Y
CI
-
60
150
A
3
$9
I
I
I
I
I
1
2
3
4
5
4o
6
Carrier concentration (%v/v)
Fig. 11-7.Effect of carrier concentration on the transport of phenylalanine and swelling of the membrane. (Reprinted from [23] by permission of John Wiley & Sons, Inc. 0 1988.)
(iii) if there is an enzymatic reaction inside the water droplets and the products have to diffuse out [22,58], the inward diffusion of water hinders this; (iv) since the swollen membrane occupies a larger fraction of the emulsion volume than the original membrane, volumetric productivity is reduced; and (v) swelling weakens the membrane, eventually causing rupture. Two mechanisms have been suggested for swelling. In the first [54], swelling occurs via a hydrated surfactant which travels back and forth across the membrane. The hydrophilic part of the surfactant is hydrated at the interface between the membrane and the phase with the highest water activity (usually the exterior phase). The surfactant molecule then diffuses to the other interface, releases water molecules into the inner phase (of low water activity and high salt concentration), and travels back to begin the process again. The second mechanism of swelling is via reversed micelles. Unlike hydrated surfactants, reversed micelles can accommodate some solute molecules along with water [62,63]; this inference is also supported by data [64,65] that charged solutes can be transported across the liquid membrane without carriers. There is also indirect evidence that surfactants such as sorbitan monooleate and commercial chelating agents such as D2EPHA favor reversed micelle formation [66,67]. The obvious ways to minimize swelling would be to reduce carrier concentration and emulsifier concentration, and to increase the viscosity of the membrane. However, there are limits to these methods because (a) very low concentrations result in poor emulsification and low rates of extraction, and (b) high viscosity slows down the diffusion processes. Since the permeation of water is a prime reason for membrane swelling, reducing this process through additives has also been tried. The hydration of Span-80 may be reduced by adding Aliquat-336, an anion exchanger [68], or cyclohexanone [21]; the latter study also showed that addition of 0-20 % paraffin to a kerosene membrane reduced swell as well as the degree of extraction (of p-nitrophenol). To reduce the osmotic gradient, and consequently the permeation of water, Scholler et al. [69] recommended adding an inert species to the external phase; in the extraction of lactic acid the use of 1.09 M glucose reduced the osmotic
11.5 Factors AfSecting LEM and SLM Performance
279
pressure difference from 3.60 MPa to 1.08 MPa, and the water transport rate from ~ 0.14 cm3 ~ m s-’ - ~to 0.004 cm3 ~ r n -s-l.
11.5.3 Carrier Concentration and Selectivity Selectivity is a key consideration in any separation process. Fermentation broths usually contain many components; for an LEM process to be feasible in practical applications, it should be possible to extract the desired species selectively from the broth. In Type-I processes, good selectivity requires that the membrane composition be so chosen that only the desired species is soluble in the membrane and there is a sufficiently large concentration gradient as a driving force. To maintain this driving force, the transported solute should not be able to diffuse back, and this is ensured by converting it into another form in the receiving phase. Examples of how this is achieved have been described earlier for phenolic derivatives in the blood [ 1I], acetic acid in wastewater [12], and an aqueous solution of diltiazem [45]. Since the transmembrane concentration gradients are initially large and decrease as extraction progresses, selectivity at short times is determined largely by the solubility of the solute in the membrane, while over a longer period of time the maintenance of the concentration gradient (by arresting the transported solute) becomes more important. This is well illustrated in the case of a phenol-acetic acid mixture [12] (Fig. 11-8). Initially, phenol is taken up faster because of its higher solubility in the membrane. As time progresses, the stronger acid ‘pushes’ the phenol out of the membrane. Selectivity in Type-I1 separations is governed by the carrier, which should from a complex only with the species to be extracted. Equilibrium of complexation is critical; both association and dissociation must be fast enough. The equilibrium constant for the reaction
u i.,
70
zb
6o
-
$ 50 k
*3
5
.3
40
4
v 30
$
wx
20
-0
3
I
Fig. 11-8. Uptake of phenol and acetic acid from a two-component solution as a function
280
11 Recovery of Biological Products by Liquid Emulsion Membranes
solute
+ carrier @ complex
is between lo3 and lo7 M-’ [70]. The enthalpy change is between -17 and -40 kJ mol-’, which corresponds to acid-base reactions, coordinate covalent bonds, hydrogen bonds, and van der Waals interactions; of these, only Lewis acid-base and coordinate covalent bonds produce good selectivity [3]. The choice of the best carrier for a given separation requirement is based on both science and experience. This is so because (a) the carrier affects not only selectivity but also stability of the membrane and (b) many useful carriers are commercial formulations whose exact composition is not known. Nevertheless, different classes of carriers may be prescribed for different kinds of applications. These include: (i) crown ethers for the chiral separation of amino acids and their esters [70,71]; (ii) cyclodextrins for isomeric enrichments, especially of drugs [72]; (iii) reversed micelles for the separation of proteins [63]; (iv) tertiary amines and quaternary ammonium salts for amino acids [22,23] and organic acids [32]; and (v) special carriers such as prostaglandins and steroid hormones for medical applications [73]. The specificity of a carrier is also affected by the presence of other ions which compete with the desired solute and by the hydrophobicity and polarity of the solute itself. Thien et al. [23] observed in the recovery of L-phenylalanine that the initial flux increased with the hydrophobicity of the amino acid. To simulate fermentation media, they added sodium sulfate, which competes with the carrier Aliquat-336. The percentage extraction decreased from 70 to 40 as the sulfate concentration increased from 0 to 0.2 M. In the light of a later study, Thien et al.’s inference about hydrophobicity may have to be modified. Bryjak et al. [41] reported that when immobilized chiral alcohols were used to separate L- and D-amino acid hydrochlorides, the stereoselectivity initially decreased as the hydrophobicity increased, reached a minimum, and then increased. They explained that when the amino acid does not posses n-electrons, selectivity varies inversely with hydrophobicity; the presence of n-electrons reverses this trend. It may also be noted that Thien et al. [23] and Bryjak et al. [4 11 used different scales for hydrophobicity; however, since both are relative scales, the qualitative deductions are not affected. From Bryjak and coworkers it is also seen that the effect of amino acid polarity on stereoselectivity is opposite to that of hydrophobicity, i.e. the selectivity passes through a maximum as the polarity increases.
11.5.4 Rate Controlling Process Figure 11-9 is a schematic representation of the transport resistances encountered by a species traveling from the external phase to the internal droplets. There are three main processes: (i) transport across the liquid film surrounding the emulsion globules (external mass transfer); (ii) transport across the liquid membrane; and (iii) reaction inside the droplets receiving the solute. Their relative importance varies with the extraction system, the operating condition, and time. This is illustrated in Halwachs et d ’ s experiments [74]. Their system had phenol in bovine blood as the exterior phase, while the internal aqueous phase contained UDPGA (urine diphos-
11.5 Factors Affecting LEM and SLM Performance
281
Liquid film resistance
(a)Well-mixed liquid emulsion membrane
1
1
I
I
Liquid film resistance
Diffuhonal resistance
,fi\\,
(b)lmmobilized flat sheet membrane
Liquid film resistance
I I
---
' \ \ U D i H u sresistance i o n a i
(c)Cross-sectional view of a hollow fiber contained liquid membrane
Fig. 11-9. Main transport resistances in different configurations of liquid membrane systems.
phoglucuronic acid) and UDPGT (urine diphosphoglucuronyl transferase). There was two-way flux through the membrane (Fig. 11-10): (i) phenol diffuses into the internal phase, where it complexes with UDPGA under the catalytic action of UDPGT; and (ii) the complex then diffuses in the reverse direction into the exterior phase. The system mimics the liver, and the underlying rationale is that the hydrophilic nature of UDPGA makes it easy to excrete its phenolic complex through the kidneys. Halwachs et al. observed that the fluxes were between those for pure mass transfer control and pure kinetic control. As they increased the enzyme
Phenol
Phenol
I
UDPGT
UDPGA-Ph
1
UDPGA-Ph
To liver
Exterior phase
Membrane phase
Interior phase
Fig. 11-10. Mechanism of phenol removal from hepatic blood by an LEM-based system. (Reprinted from [16] by courtesy of Marcel Dekker, Inc. 0 1988.)
282
11 Recovery of Biological Products by Liquid Emulsion Membranes
(UDPGT) concentration in the interior phase, the system approached mass transfer limitation. While it is thus important to have a reasonable concentration of enzyme (or a neutralizing compound) in Type-I separations, beyond a critical concentration the performance becomes controlled by the rate of diffusive transport. As the extraction proceeds, the aqueous droplets slowly lose their activity, by neutralization or deactivation or product inhibition. This loss of activity progresses inward from the outer surfaces of the droplets (i.e. their interfaces inside the membrane globules; see Fig. 11-1). Hence, in the later stages of extraction the solute has to diffuse across larger distances through the membrane, and the process then becomes controlled by this diffusion rather than by external mass transfer. This was observed by Chaudhuri and Pyle [75] for lactic acid and by Lee et al. [24] for penicillin G. In a once-through continuous process, this diffusion resistance is not encountered; however, a single pass does not extract all of the solute and hence recycling is usually adopted [24,44,76].
11.6 Modeling of LEM Systems To model the transport and reaction processes in LEM systems, we recognize there are three phases and two interfaces; complexation reactions occur in the membrane phase and enzymatic or neutralization reactions in the innermost droplets. A good emulsion may be assumed to be macroscopically homogeneous. This means there is no concentration gradient in the outer continuous phase and all two-phase organic globules are identical. Beyond this it is not reasonable to assume that every globule is a€sointernally homogeneous. However, the first model [20] assumed that the aqueous droplets contained in each globule were identical and could be coalesced (for mathematical analysis) into one large droplet (Fig. 11-11). The latter assumption was justified by considering that the liquid membrane was infinitesimally thin and therefore offered negligible resistance to mass transport.
Q Orginic
0-@
\ Aqueous
1
Fig. 11-11. Lumped two-phase model of an LEM globule. (Reprinted from [20] with permission of VCH Publishers 0 1989.)
This kind of model has been applied to the slow release of drugs from permeable capsules in the body [19,20]. The rate of release is:
11.6 Modelins? of LEM Svstems
283
where M is the rate of release at any instant of time t, MO is the initial amount of drug, A is the surface area of the droplet, K is the partition coefficient in the membrane phase, L is the thickness of the liquid membrane, V1 is the internal volume of the emulsion phase, and V, is the volume of the continuous phase. Usually V2 >> VI and hence both selective recovery and enrichment are achieved together [23,26]; then Eqn. (1) may be simplified to:
Equations (1) and (2) predict an exponentially decreasing rate of release. Some therapies, however, require a constant rate of drug release; the design of capsules and the models for them differ from those applied to LEMs used for fermentation product recovery [19,77]. This simple model has a number of weaknesses. Coalescing the droplets prevents analysis of the effects of droplet size and the number of droplets per globule, whose significance has been discussed in section 11.5. The membrane thickness also cannot be neglected; this implies a radial variation of solute concentration inside the organic phase, as a result of which the encapsulated droplets do not perform identically. Moreover, the model does not take account of reactions promoted by enzymes or cells in the droplets, a feature which has been usefully exploited to combine product recovery from one reaction system with further reaction to a final product [ 14,22,26,27]. These limitations were removed by Ho et al. [65] in the advancing front model (AFM) shown in Fig. 11-12. The AFM preserves the separate identities of individual droplets inside an organic globule, and it allows finite diffusion rates through the organic membrane. The encapsulated droplets (containing reagents) are much smaller than the parent globules, and so are their time constants. Therefore, the AFM GLOBULE OF
REACTION FRONT INTERNAL DROPLET DEPLETED OF
INTERNAL DROPLET
Fig. 11-12. The advancing front model. (Reproduced from [65] by permission of the American Institute of Chemical Engineers 0 1982 AIChE. All rights reserved.)
284
11 Recovery of Biological Products by Liquid Emulsion Membranes
assumes local equilibrium between the dispersed and continuous phases, and describes the concentration field within the globules in terms of a continuum based on the local average concentration. External mass transfer resistance, swelling, and breakage of the membrane are neglected. A further crucial assumption is that the rate of reaction of the solute upon its release into the inner aqueous droplets is infinitely faster than the diffusion rate in the membrane phase. Hence the solute reacts instantaneously, beginning with the droplets closest to the surface. As a result, there develops a sharp spherical reaction front which shrinks inward as the outer droplets become depleted of their reagent. Externally to the front there is only inward diffusion of solute (and possibly outward diffusion of products), which is trapped at the front and progresses no further inward. The inner core thus contains droplets with fresh reagent. Because this core reduces in size as the extraction progresses, the AFM is also called the ‘shrinking core model’ [75]. The governing equations of the AFM are presented below [66]. Continuous phase:
(3) t = 0:
C,
= ce0
(4)
Emulsion globules:
Deff
a
r2
ar
ac - - -(r2 at
ac
-);
ar
Rf S r % R
(5)
t=O: c=O(rSR) r = R : c=Kc,(t?O) r = Rf ( t ): c = 0 ( t 2 0)
Reaction front:
In Eqns.(3) to (lo), c is the solute concentration in the membrane phase, C, is its concentration in the continuous phase, R is the radius of the globule, Rf is the radius of the reaction front, V, is the total volume of all the aqueous droplets in the membrane globules, Vm is the total volume of the membrane phase, V, is the volume of the continuous phase, K is the partition coefficient for the solute, and Deffits effective diffusivity in the membrane phase. Ho et al. [65] showed that the AFM provided a good
11.6 Modeling of LEM Systems
285
simulation of the experimental extraction of phenol from an aqueous solution through a membrane composed of 1 % Span-80 (ICI), 3 % ENJ3029 (a non-ionic polyamine from Exxon) and 96 % S lOON (an isoparaffin from Exxon). The inner aqueous droplets had 0.375 M NaOH. This model also predicted satisfactorily the extraction of lactic acid from a fermentation broth [78]. Lorbach and Hatton [79] extended the AFM by removing some of its simplifying assumptions. They allowed the membrane globules to be of different sizes, in effect allowing coalescence and breakage, and included axial dispersion and mass transfer resistance between the membrane phase and the external continuous phase. These features are significant in column contactors, whose advantages over stirred vessels have been demonstrated for penicillin G [34,35]. In the modified AFM, Eqn.(3) becomes :
t = 0 : c, = co,
(2 2 0)
The boundary conditions depend on the type of operation. For co-current operation:
For counter-current flow these conditions get interchanged. L is the length of the column, z is the axial distance along the column, ue is the superficial velocity of the continuous phase and De is its effective diffusivity. f(R) is a distribution function for the emulsion globule size, R being its radius. Lorbach and Hatton chose the Mugele-Evans distribution [80]:
[
6dm 1 f(R) = exp -(61n 26(dm - 2R) ~ T R(dm-2R) where 6, d, and a are parameters. To compare with the work of Ho et al. [65], they studied the phenol-water system. They observed that in batch and continuous mixer-settlers, back-mixing in the continuous phase was significant only at low Peclet numbers (smaller than about 5) and globule size distribution had a noticeable effect only for long residence times. However, polydispersity of the globules was important in column contactors - it affected the dispersed phase hold-up, globule size velocities, inter-phase transport, and nonidealities such as segregation and bypassing. Although applicable to many LEM systems, the AFM and its extended version have one major limitation. The assumption that the rate of diffusive transport of
286
11 Recovery of Biological Products by Liquid Emulsion Membranes
the solute through the membrane is much slower than its rate of reaction in the inner droplets is not always valid, as in the removal of phenol from hepatic blood [ll], separation of a racemic mixture of D- and L-phenylalanine [23,41], and the reduction of nitrate to nitrite using cell homogenate from Micrococcus denitrifcans [38]. When diffusion and reaction occur at comparable rates, the incoming solute is not consumed instantly, and hence the reacting region is spread over a finite interval of the radius of the globule (Fig. 11-13). This generates a reaction zone, i.e. an annular shell within which the reaction starts at the outer surface and is completed at the inner surface. Like the reaction front, the reaction zone advances inward and, if there is sufficient solute and reagent and the globules are not very large, eventually collapses at the center, For isotropic globules the zone remains of constant thickness. Janakiraman [81] modeled this problem and showed that the reaction zone can extend from less than 10 % (for fast reactions) to almost 100 % of the emulsion globule radius. The mathematical models described above may be modified to include other features such as swelling of the membrane [75], decomposition of the product [82], and incorporation of porous solid supports [36,43]. SURFACE OF EMULSION GLOBULE
REACTION FRONT REACTION ZONE
DEPLETED ZONE
Fig. 11-13. The reaction zone model. (Adapted from [81] by courtesy of Marcel Dekker, Inc. 0 1985.)
11.7 Selected Applications 11.7.1 Separation of L-Phenylalanine and its Derivatives The early work of Scheper et al. [83] has been followed by many other studies of the separation of L-phenylalanine and its derivatives from a racemic mixture. Before considering their work, it will be instructive to discuss a few later studies. Thien and coworkers [23] approached the direct problem of separating the L-isomer from a feedstock containing both D- and L-phenylalanine. The membrane was composed of Solvent 100 Neutral (a parafinnic solvent) and Paranox 100 (a nonionic surfactant), both commercial products of Exxon; it also contained decyl alcohol
11.7 Selected Auulications
Outer Phase Membrane Phase Interior Phase
287
Fig. 11-14. LEM separation of L-phenylalanine from a racemic mixture. (Reprinted from [23] by permission of John Wiley & Sons, Inc. 0 1988.)
as a cosurfactant. Phenylalanine is a zwitterion; below pH 3 it is positively charged, between pH 3 and pH 9 both positive and negative species coexist, and at high pH values only the negative charge remains. Thien et al. adopted counter-current facilitated transport by Aliquat 336, a tri-capryl quaternary ammonium salt, as the carrier. To ensure complexation with the unipositive hydrophobic heads of the carrier molecules, the pH was maintained at 11 so that the amino acid had a negative charge. Chloride was the counter-ion (Fig. 11-14). An interesting feature of their study was that the optimal process conditions depended on whether the phenylalanine was only to be separated or to be concentrated also. For separation alone, a high concentration of C1- was desirable; however, this also caused considerable swelling of the membrane, and hence a lower concentration of C1- improved the concentration of the recovered L-phenylalanine at the cost of reduced separation efficiency. Ha and Hong [26] had a somewhat different approach and objective. Like Thien et al., they had the source phase as the outer continuous phase and used an organic membrane and a carrier. However, the membrane was of kerosene, the emulsifier was Span 80, and Adogen 464, a cationic surfactant from Fluka, was the carrier. The continuous phase had a mixture D- and L-phenylalanine methyl esters, both of which are soluble in the membrane and therefore diffused into the inner aqueous phase. The transport of the L-isomer was also facilitated by the carrier. The inner droplets contained a-chymotrypsin, which selectively hydrolyzed the L-ester into Lamino acid, and the unconverted D-ester was recycled. Because of the coupling of transport selectivity and enzymatic conversion, pH was a critical factor. A high pH enabled rapid recovery of L-phenylalanine as in Thien et al.’s experiments but it killed the enzyme activity. So, a pH of 7 was maintained to ensure good activity of a-chymotrypsin and complete conversion of the L-ester. At the neutral pH the zwitterion L-phenylalanine exists in both negative and positive forms. The negatively charged species could be transported outward by the carrier while the positive component was (largely) retained in the inner droplets. A schematic diagram of the mechanism is shown in Fig. 11-15. Ha and Hong’s work echoes that of Scheper et al. [83], who had earlier used the same technique to obtain L-phenylalanine from a racemic mixture of esters. While both groups of authors employed the same membrane composition and temperature
288
Outer Phase
11 Recovery of Biological Products by Liquid Emulsion Membranes
Membrane Phase
Inner Phase
Fig. 11-15. Schematic of an LEM-based system for the production of L-amino acids from a racemic mixture of esters. LE-H+: protonated L-ester; LE: neutral ester, LA+-: zwitterionic L-amino acid; Q+:Adogen 464. (Reprinted from [26] by permission of John Wiley & Sons, Inc. 0 1992.)
(25 "C), Scheper et al. used a pH of 6 while in Ha and Hong's experiments the pH was 7. This modest difference in pH made a significant difference to the enzymatic conversion of the L-ester and illustrates the importance of optimizing the process parameters; while the a-chymotrypsin lost 70% of its native activity in Scheper et al.'s study, Ha and Hong could retain 60 9% of the native activity. To avoid the limitations of agitated LEMs, supported liquid membranes (SLMs) are often preferred. Bryjak and coworkers, for example, immobilized the membrane in both flat porous sheets and hollow-fiber modules for the chiral separation of phenylalanine hydrochloride (PAH). Their first study [25] with pure L-amino acid hydrochloride revealed some interesting features: (i) hollow fibers had lower fluxes of PAH than flat sheets but had superior reusability; (ii) although the carriers used helped the transport of PAH, a substantial fraction of the PAH was transported through self-diffusion; (iii) contrary to the observations of Thien et al. [23], the fluxes of several amino acid hydrochlorides did not depend on the hydrophobicity of the amino acid; and (iv) the SLM had to be activated in order to enable the transport of solute; this was done by aqueous ethanol and propanol. Later work by these authors [41] focused on the use of two hydrophobic chiral alcohols, nopol and (2s)(-)-methyl-l-butanol, as the membrane phase immobilized in flat porous sheets. No additional carrier was used. The degree of separation of L- and D-isomers of PAH was measured in terms of the ratio of their fluxes, called the stereoselectivity. While their previous work [25] showed that fluxes did not depend on hydrophobicity, the stereoselectivity did [41], and it varied from 0.39 to 1.52 according to the type of chiral membrane phase and the properties of the amino acid. Supported liquid membranes for L-phenylalanine extraction have also been studied by Molinari et al. [84] and Shinbo et al. [ 8 5 ] . The former work applied the system of Thien et al. [23] to a flat sheet microporous hydrophobic polypropelene membrane. Shinbo et al. used a chiral crown ether in different membrane solvents immobilized in a similar polypropelene film; they recommended o-nitrophenyl octyl ether and p nitrophenyl heptyl ether as solvents.
11.7 Selected Applications
289
11.7.2 Extraction of Lactic Acid Pyle and associates studied the extraction of lactic acid (LA) from prepared aqueous solutions as well as from fermentation broth. Their first work [78] addressed the controlling parameters in a batch extraction. LA was in the outer continuous phase and the inner aqueous droplets had sodium carbonate to neutralize the acid and prevent its reverse transfer. The membrane phase solvent was a blend of n-heptane and light paraffin in the volumetric ratio 7:3; it contained Span 80 as the emulsifier and Alamine 336, a tertiary amine, as the carrier. Extractions were carried out at 20-22°C. Initially, the controlling resistance was external phase mass transfer but, as extraction progressed, diffusion through the membrane became the rate limiting step. The initial rate of extraction was sensibly independent of Span 80 concentration but the rate of swelling of the membrane increased almost three-fold as the emulsifier concentration was increased from 1 % to 6 %. Since swelling did not significantly affect the rate of decrease of LA from the continuous phase, it might be advisable to have a fairly high concentration of Span 80 so as to stabilize the emulsion. Similarly, the rate of extraction increased with Alamine 336 concentration but stabilized at about 10 %. Figure 11-16 shows the mechanism of extraction. Scholler et al. [69] extended this work to a fermentation broth obtained from a continuous culture of Lactobacillus delbreuckii (NICB-9282, NRRL-B445). The results were, however, inferior to those for aqueous solutions; whereas 80 to 90 % of the LA could be extracted from an aqueous solution, no more than 45 9% extraction was possible from the broth. Scholler et al. provided a possible explanation. Experiments with pure LA showed that a pH 2.4 was the best and that extractions were better if the pH was below the isoelectric point (3.86). So, the pH of the cell-free broth was reduced from 4.4 to 2.15 by adding HCl; the Alamine 336 then extracted HC1 in preference to LA because this carrier has a strong affinity for the anions of mineral
2HLA 2H+
I+t2 M + 2R,NH+LA'
2Na+
I(
+ CO,2-
( R s N H + ) ~ C ~ ~ 2NaLA ~. H,O+ COP
Outer Phase
(
2R,N
Fig. 11-16.Mechanism of facilitated extraction of lactic acid with Alamine 336 as carrier.
290
11 Recovery of Biological Products b y Liquid Emulsion Membranes
acids. While this competition was eliminated by extracting directly from the untreated broth (pH 4.4), the efficiency was only 15-20 %. Lazarova and Peeva [86] achieved better results by employing a different carrier and different conditions. Their fermentation was carried out by Lactobacillus casei and the medium contained phosphate, sulfate, nitrate, lactose, peptone, and agar. The membrane was composed of Aliquat 336 dissolved in n-octane, and a sodium chloride solution was used in the receiving phase. Increasing the concentration of Aliquat 336 favored removal of LA from the continuous phase, but its concentration in the receiving phase reached a peak at a carrier concentration of 5 g 1-' and then declined; this was attributed to increased viscosity of the organic phase. Increasing the initial concentration of LA lowered the extraction efficiency, but had little effect on the final amount transferred, i.e. the equilibrium was unaffected. Two aspects of this study provide an interesting comparison with Scholler et al. [69]. First, while the neutralizing reagent concentration (NaZC03 in Scholler et al. and NaCl in Lazarova and Peeva) did not have much effect on the percentage of LA extracted in Scholler et al. 's work, Lazarova and Peeva observed that large concentrations had a detrimental effect. Secondly, Scholler et al. reported that extractions at pH above the isoelectric point (3.86) were poor and lowering the pH by adding HC1 did not help improve the efficiency. On the contrary, in Lazarova and Peeva's experiments the donor phase pH varied from 5.64 to 4.68, the receiver phase pH from 5.37 to 5.94, and 70% LA recovery was achieved. These observations indicate that careful tailoring of the emulsion composition is a crucial aspect of LA extraction.
11.7.3 Citric Acid Recovery Citric acid is produced commercially by fermentation with Aspergillus or Candida fungi. In conventional processes, the acid is precipitated as calcium citrate from the filtered broth by the addition of lime. Sulfuric acid is then added to convert the precipitate to dilute citric acid, which is purified, decolorized, and crystallized [87]. More recently, liquid-liquid extraction followed by precipitation or backextraction has been recommended [88], and this suggests the use of LEMs to obtain citric acid directly from the fermentation broth. Two recent studies of citric acid recovery show a preference for SLMs. Friesen et al. [89] used a flat sheet membrane to immobilize the organic liquid membrane; the material was Celgard 400, a microporous hydrophobic polypropelene film from Celanese Plastics. The membrane phase contained Shell Sol 71, an aliphatic hydrocarbon solvent from Shell Oil, or Alkyl Aromatic Oil (Alkane 56), an aromatic hydrocarbon solvent from Chevron Research. Trilauryl amine, a tertiary amine, was the carrier and n-octyl or higher alcohols were also dissolved in the solvent to increas the partition coefficient (and consequently the flux) of citric acid. The mechanism of transport is shown in Fig. 11-17. Friesen et al. observed optimum values of temperature (60 "C), trilauryl amine concentration in Alkane 56 (38 %, v/v) and modifier (n-tridecanol) concentration in Shell Sol 71 (20%, v/v) for which the citric acid flux was maximum. Whereas
11.7 Selected Applications Feed Solution
I
Supported Liquid Membrane
C,H,OH(COOH)l
Citric Acid
29 1
Product Solution
-
C,H,OH(COOH), N+R,H.C,H,OH
(COOH),COO'
-
Fie. 11-17.Carrier-mediated transoort of citric tcid across an LEM. (Reprinted from [89] with kind permission of Elsevier Science-NL, Amsterdam, The Netherlands 0 1991.)
the flux passed through a maximum when Alkane 56 was the solvent, it increased continuously with carrier concentration in Shell Sol 71. Although the maximum flux with Alkane 56 was higher than with Shell Sol 71 and occurred at lower carrier concentrations, the authors had to prefer the latter solvent because the U.S. Food and Drug Administration had not approved aromatic solvents for use in the production of food-grade acids. For the same reason, n-decanol was chosen as the modifier even though n-tridecanol, nonylphenol, and isohexadecanol were superior. In principle, almost complete recovery was possible but this required very low citric acid flux and low concentration in the stripping phase; thus, the choice of operating conditions has to balance the rate of separation and the extent of concentration desired against the degree of recovery. A different kind of apparatus was utilized by Basu and Sirkar [36]. They used hollow-fiber modules. Each module had two sets of fibers; one set carried the feed aqueous solution of citric acid while the stripping solution (receiving phase) flowed through the other set (see Fig. 11-4). The liquid membrane was contained in the inter-fiber spaces; it was composed of tri-n-octyl amine (the carrier) dissolved in xylenes, heptane, or methyl isobutyl ketone (MIBK). Either pure water or aqueous sodium hydroxide was used as the stripping liquid; in the latter case citric acid was extracted as mono-sodium citrate. Their main objective was to assess the effect of key variables on the mass transfer rate of citric acid. As also observed by Friesen and coworkers [89], the mass transfer coefficient and mass transfer rate showed bellshaped profiles with respect to the carrier concentration, but the peaks were less sharp. The occurrence of maxima was explained by the opposing effects of greater complexation (implying faster transport) and higher viscosity (which slowed the diffusion rate) in the membrane phase as the carrier concentration was increased. About 99 % of the citric acid fed could be extracted and the rate of extraction varied linearly with the membrane area, implying easy scale-up. The results of both Friesen et al. and Basu and Sirkar confirm the long-term stability of SLMs. Friesen et al. could operate a flat sheet membrane module continuously for 400 h with fermentation broth as the feed solution. Basu and Sirkar's stable operation lasted 1500 h; because osmotic gradients cause some water to permeate through the liquid membrane, thus creating instability, they discharged a small amount of the membrane liquid once each day during the extraction process and
292
11 Recovery of Biological Products by Liquid Emulsion Membranes
replenished it with fresh liquid. Removal of even this small quantity of water-contaminated membrane fluid doubled the mass transfer rate.
11.7.4 Extraction (and Subsequent Reaction) of Penicillin G Schugerl and associates originally highlighted the usefulness of LEMs for the extraction of penicillin G from fermentation broth. The need for an alternative to the presently used recovery process may be briefly explained as follows. Since penicillin G is a weak acid (pK, = 2.75), the pH has to be reduced to below this value so that the acid exists in its free (and not anionic) form because only the free acid is soluble in an organic solvent, n-butyl acetate being commonly used. At such a low pH, penicillin G is unstable at ambient temperatures, and therefore the extraction is carried out at 0 "C. Even in this expensive low-temperature operation there is considerable loss of penicillin G. Schugerl and coworkers therefore applied LEMs to extract penicillin G at ambient conditions. There are two kinds of extraction. In ion-pair extraction, an aliphatic amine A dissolved in the organic phase reacts with the penicillin acid anion P and a proton H+ in the aqueous phase: A(org)
+ P-(aq) + H+(aq) @ AHP(org)
In ion-exchange or reactive extraction the penicillin anion is exchanged with another anion, usually C1-, carried by a quaternary ammonium salt: NR;Cl-(org)
+ P-(aq) a NR+4P(org) + Cl-(aq)
Likidis and Schugerl [ 151 investigated both methods. For ion-pair extraction they recommended Amberlite LA-2, a complex secondary amine from Rohm and Haas, and Hoe F2562, a diisotridecylamine from Hoechst, as good carriers. The quaternary ammonium salt Adogen 464, a trioctylmethyl ammonium chloride from Serva, was chosen for ion-exchange extraction. Both types of carriers were dissolved in an nbutyl acetate membrane, and both types of extraction were operable at ambient temperature and a much milder pH than in the conventional method. In ion-pair extraction the donor aqueous phase was at pH 5.0 and the receiving phase at pH 7.5, with 95 % extraction. The ion-exchange method used p-toluene sulfonic acid as the counter-ion, pH 6, and achieved 92 % extraction without buffers. While these experiments were carried out in a stirred vessel, Schugerl and coworkers have also studied the perfomance of different kinds of equipment - centrifugal extractor [3 11, extraction decanter [33], and three types of extraction columns [34]. The extraction decanter could use the mycel-containing broth directly and achieve 72 to 96 % extraction at pH 4.6 to 5.1. Between a pulsed perforated column, a Karr column, and a Kuhni column, the last design performed the best [34], with almost complete recovery being possible under certain conditions. This SLM system was also able to separate penicillin from a solution containing phenylacetic acid (PAA)
11.7 Selected Applications
293
outer phase
/
/
membrane
phase-
'-/C lCHPl
\
W
Fig. 11-18. Penicillin G extraction and synthesis of 6-APA. (Reprinted from [I41 with kind permission of Elsevier Science-NL, Amsterdam, The Netherlands 0 1990.)
[92]. Despite strong transport competition between penicillin G and PAA, the ratio of their degrees of extraction was consistently above 1.0, with a maximum of 1.8. Interest in recovering penicillin G from aqueous solutions also containing 6 -aminopenicillinic acid (6-APA) or PAA stems from the possibility of having a continuous process whereby penicillin G is extracted from fermenter broth and converted directly to 6-APA. Figure 11-18 is a schematic representation of this [14,22]. Penicillin G from the continuous phase is transported inward through a kerosene membrane containing Span 80 and Amberlite LA-2. The inner dispersed aqueous phase has penicillin amidase to convert penicillin into 6-APA and PAA; the membrane restricts the transport of 6-APA but PAA diffuses out into the continuous phase. Thus, it becomes important to allow only penicillin to be transported inward and prevent 6-APA from diffusing out; Hano et al. [90] and Lee et al. [91,92] considered these problems. In principle, it is possible to have an integrated process in which a fermenter is connected to an extraction column, de-emulsifying unit and a reactor so that 6-APA is the final product from a fermenter broth (Fig. 11-19). The work by Schugerl's group spawned many similar studies. Hano et al. [90] opted for a mixture of n-butyl acetate and kerosene as the membrane solvent, dioctyl amine as the carrier, and ECA 4360J (Exxon Chemical Co.) as the emulsifier. In an agitated system, the internal droplets contained 0.5 mol dm-3 Na2C03. Almost 100 % recovery of penicillin G could be achieved from pure solutions as well as a mixture of penicillin and 6-APA. Similar findings were also reported by Lee et al. [91], whose membrane phase differed in that it contained 20% (w/w) Span 80 in ECA 43605 as the surfactant. SLMs have also been employed for penicillin G extraction. Lee et al. [24] used a flat hydrophobic polypropylene membrane to impregnate the organic phase, which was 1-decanol with Amberlite LA-2 as the carrier. More than 95 % of the penicillin could be separated at a pH between 5.0 and 5.5 and a carrier concentration of 20 mM.
294
11 Recovery of Biological Products by Liquid Emulsion Membranes KUHNI EXTRACTION COLUMN CELL-FREE BROTH
/
EMULSION CELLS
00 INNER PHASE
I
PHASE
-.
c - l : ; p A / L I C - C t R E C " R ( I O R
' I
FERMENTER
CE LL- F R EE EMULSION PREPARATION
DROPLET
ENZYME PHASE
ENZYME PHASE
PRODUCT
v
v Fig. 11-19.Integrated process for penicillin production, extraction, and conversion. (Reprinted from [14] with kind permission of Elsevier Science-NL, Amsterdam, The Netherlands 0 1990.)
The concentration of the carrier was important in determining the controlling resistance: below 20 mM, diffusion through the liquid membrane was the controlling step, its resistance decreased as the concentration was increased, while beyond 200 mM, external mass transfer was the rate-limiting step. Since the permeability of the liquid membrane decreased with increasing pH of the feed solution and increased for the pH of the stripping solution, the authors recommended pH 6.0-6.5 for the former and pH 7.0 for the latter.
11.8 Separation of Proteins via Reversed Micelles in LEM Systems The economical separation of a selected protein from a fermentation broth with minimum loss of activity continues to be a challenging problem. Chromatographic meth-
11.8 Separation of Proteins via Reversed Micelles in LEM Systems
I
295
Micelle
Micelle
Outer phase
Membrane phase
Interior phase
Fig. 11-20. Diagrammatic representation of a reversed micellar LEM system for protein extraction. (Reprinted from [16] by courtesy of Marcel Dekker, Inc. 0 1988.)
ods are expensive; liquid-liquid extraction has low selectivity and yield, and causes considerable denaturation of the protein. LEMs using reversed micelles offer a viable alternative. A reversed micellar LEM contains aqueous reversed micelles in the organic membrane phase. The desired protein is trapped inside the micelle at the interface between the membrane phase and the (continuous) source phase. Like any carrier, the micelle travels across the membrane, releases the protein at the interface with the receiving aqueous phase, and traverses back (Fig. 11-20), Since the protein is protected by the micellar cage, there is little risk of denaturation by the membrane solvent. Reversed micelles thus provide good selectivity and retention of activity [93]. The surfactant used plays a key role in the process because it determines the micellar dispersion and stability of the emulsion, and it provides a protective cover for the protein entrapped in the micelle. Two surfactants are commonly used. Sodium bis(2ethylphenyl) sulfosuccinate (commercially known as AOT) is an anionic surfactant. Its suitability may be explained through the theory of Mitchell and Ninham [94]. If ‘v’ is the volume of the organic membrane molecule, ‘1’ its length and ‘u’ the crosssectional area of the polar head of a surfactant molecule at the interface, then a reversed micelle is possible only if vl(lu) > 1. Since AOT is a double-chain surfactant with a small head, this condition is easily satisfied. The second type of surfactants used in reversed micellar systems are quaternary ammonium salts, mainly trialkylmethyl ammonium salts such as Aliquat 336. They are cationic amphiphiles.
11.8.1 Factors Affecting Protein Recovery pH of the Aqueous Phase The pH determines the rate of dissociation of the charged residues which constitute the primary structure of the protein molecule and therefore the net charge on the protein. When the pH equals the isoelectric point, pK, the protein is neutral, and it is
296
11 Recovery of Biological Products by Liquid Emulsion Membranes
positively or negatively charged according to whether pH < pK or pH > pK. The role of pH in determining the degree of extraction has been confirmed for large proteins as well as amino acids and small peptides. Luisi et al. [95] showed that the dependence of the percentage transfer of tryptophan and L-tryptophyl glycine (a dipeptide) into an organic phase composed of TOMAC in cyclohexane was determined by the ionization of the amino acid group. Goklen and Hatton [96] studied the solubilization of three proetins, cytochrome c, ribonuclease, and lysozyme, all of similar size but different pK values. The organic phase had AOT in iso-octane. They observed that solubilization does not take place if pH > pK, i.e. the protein is negatively charged. When the pH is reduced to less than the pK, there is a sudden increase in solubilization because the surfactant and the protein then bear opposite charges and therefore attract each other. With a cationic surfactant, the opposite kind of behavior between pH and pK is expected, and Hatton’s [97] results on the solubilization of catalase using dodecyl-trimethyl-ammonium bromide in n-octane with hexanol as cosurfactant confirm this. Not all protein extractions, however, can be explained so clearly. In the solubilization of a-chymotrypsin in Aliquat 336 with isotridecanol as a cosurfactant, Jolivalt et al. [98] observed that while there was good separation at pH above the pK, the yield of a-chymotrypsin decreased very slowly as the pH was reduced and became negligible only at 4 units below pK. This means that a-chymotrypsin is significantly extracted by a cationic surfactant even when it is positively charged. Another anomalous system is a-amylase (pK = 5 . 5 ) ; van’t Riet and Dekker [99] studied its transport with TOMAC dissolved with octanol in iso-octane. According to theory, there should be good separation for any pH > 5 . 5 , but experiments showed that solubilization of a-amylase was strong only in a narrow band around pH 10 and dropped sharply on either side.
Ionic Strength and Type of Salt The presence of salts, mainly KC1 and NaC1, in the aqueous micelles affects the equilibrium size of the surfactant aggregates. The greater the ionic strength, the smaller will be the micelles, and the effect is stronger for larger ions, which have large range electrostatic interactions. As a result, extraction yield decreases as ionic strength increases 11961. Since the ionic effect depends on the concentration of the salt as well as the ionizability of the protein, the salt concentration may be so chosen as to have a micellar size which allows only a particular protein to be encapsulated and excludes others [loo]. Goklen and Hatton’s [96] study provides a lucid demonstration of this principle. In an AOT/iso-octane system applied to an aqueous solution containing cytochrome c, lysozyme, and ribonuclease-A, they observed that at a low concentration of KCl (less than 0.1 M) all three proteins were solubilized; at salt concentrations between 0.3 and 0.7 M, however, their solubilizations differed widely, cytochrome being the least solubilized and lysozyme the most. Beyond 0.8 M, the micelles were too small to accommodate the proteins, thus providing a method to recover the entrapped protein from the micelle; this is called back-extraction.
11.8 Seuaration of Proteins via Reversed Micelles in LEM Svstems
297
As with pH, there are systems which do not follow a common trend with respect to the effect of salt. The a-amylase case, referred to previously, is an example. For reversed micelles formed by TOMAC (0.4 %, w/v) in iso-octane containing 0.1 % (v/v) octanol, the maximum recovery of a-amylase remained practically constant at about 70 %, but the pH values corresponding to the maxima increased as the concentration of dissolved NaCl increased in the range 0 to 50 mM [ l o l l . Even at a fixed pH and salt concentration, micelle formation and solubilization of a protein depend on the type of salt; the solubilizations of ribonuclease, lysozyme, and trypsin are higher with CaClz than with MgC12 [102]. These observations indicate that our understanding of the phenomena associated with reversed micellar extractions is still incomplete. Factors such as the protein structure, conformation, and hydrophobicity should also be considered in addition to the isoelectric point, molecular weight, and salt concentration. As this might make a quantitative analysis complex and difficult, alternate methods such as neural networks [lo31 are now being employed to provide generalized predictions of protein partitioning. Mass Transfer Limitations When performed on a large scale, it is important to take account of the rates of transport of the proteins from the continuous source phase to the interface with the membrane and the rates of diffusion of the micelles through the membrane. Even with fast solubilization, transport limitations can reduce the efficiency of extraction. With a reversed micellar phase of trioctylmethyl ammonium chloride in isooctane, Dekker et al. [lo41 found that the mass transfer rate of a-amylase was controlled by diffusion through the interfacial film at the boundary between the continuous and the membrane phases during forward extraction. The reverse operation, however, was controlled by the rate of enzyme release at the opposite interface [lo51 and not by diffusion. In the extractions of a-chymotrypsin and a-amylase using porous hollow fibers. Dahuron and Cussler [ 1061 noted that three mass transfer resistances were significant - through the liquid membrane and through the boundary layers on both sides of the membrane. It is, however, possible that boundary layer effects were weaker, but not insignificant, in Dekker et al. 's experiments because a stirred vessel was used. Other Factors The organic solvent of the membrane phase and the concentration of the surfactant have a significant effect on the protein transfer efficiency. Increasing the surfactant concentration favors protein solubilization, resulting in a broader range of pH for good solubilization [95,96]. However, large concentrations of the surfactant also make it difficult to release the protein from the micelles into the receiving aqueous phase. Thus, an optimum concentration is chosen to balance these two effects. The organic solvent affects the size of the reversed micelles and consequently the
298
11 Recovery of Biological Products by Liquid Emulsion Membranes
water solubilization capacity. Luisi et al. [95] observed that the percentage solubilization of a-chymotrypsin varied with the membrane phase solvent and, for a given solvent (cyclohexane), the solubilization differed from one protein to another. Significant improvements in selectivity are sometimes possible by introducing affinity ligands in the organic phase. Woll et al. [lo71 enhanced the partitioning of concavalin A into reversed micelles by including the biosurfactant octyl-P-D-glycopyran in the organic phase. The transfer of the control protein, ribonuclease-A, was not affected by the biosurfactant. In general, high values of the protein-ligand binding constants in the organic phase and low binding constants in the aqueous phase favor solubilization [lOS].
11.8.2 Applications The conventional method of recovering proteins with the aid of reversed micelles involves two stages: (i) transfer of the protein from a source phase into the micelles; and (ii) back-extraction from the micelles into a receiving phase. Figure 11-21 is a flowsheet of this scheme. While the applications of this method are many and vaned [96,98,105], it would be desirable to combine the two stages into one by encapsulating reversed micelles inside liquid membranes. Although a number of workers have suggested the use of reversed micelles to shuttle proteins across an immiscible liquid barrier [3,16,95], operational difficulties have hindered development of the technology. This may be appreciated by recognizing that a reversed micellar LME is a four-phase system, thus increasing the difficulty of maintaining good selectivity and rapid transport along with stability of both the membrane and the micelles. Recent work however, promises the likelihood of reverse micellar LEMs being feasible for the selective separation of proteins. Armstrong and Li [lo91 studied the extractions of cytochrome c, lysozyme, and bovine serum albumin through an LME containing AOT reversed micelles in a small glass apparatus. The feed solution contained 0.1 M NH40Ac and different concentrations of KC1. Protein concentrations on the receiving side were measured spectrophotometrically. While the extraction of all three proteins increased with increasing KC1 concentration (i.e. with ionic strength), there were optimum values for pH and surfactant concentration, similar to reversed micellar phase
Extraction
Extraction
Fig. 11-21. Combined forward and backward extraction for two mixedsettler units with recirculation of the reversed micellar phase. (Reproduced from Trends in Biotechnology, June 1986, 153-161, with permission from Elsevier Trends Journals Cambridge, U.K. 0 1986.)
11.9 Economic Asnects
299
some earlier observations [95,98,99]. It is noteworthy that no significant conformational changes occurred in the proteins, and even a large protein such as bovine serum albumin could be separated, which was not possible by liquid-liquid extraction [96]. To obviate the limitations of a stirred LEM, Tsai and coworkers [110] used supported liquid membranes (SLM) for the extraction of a-chymotrypsin. The membrane phase was iso-octane with AOT reversed micelles, NaCl was used to control the ionic strength, and the pH was adjusted by a combination of citric acid, disodium hydrogen phosphate dihydrate, potassium dihydrogen phosphate, and sodium hydroxide. Polyvinyl ledene fluoride and polysulfone membranes were employed for immobilization. Kinetic studies in a stirred cell showed that interfacial transport between the membrane and enzyme solution was the rate-controlling step at low surfactant concentration, while diffusion through the membrane became rate limiting at high concentrations (above 200 mM). At very low concentrations (below 25 mM) the adsorption of AOT to the interface was the rate-determining step. Continuous recovery up to 100 h was demonstrated in an SLM, thereby attesting its feasibility.
11.9 Economic Aspects To be commercially feasible, LEM applications have to be technically and economically sustainable. The important technical problems have been discussed in this article and, with the availability of good surfactants and carriers and a variety of porous solid modules, most of these have been overcome or minimized. The economic viability of LEMs on a large scale still remains a weakness; this is partly due to inadequate understanding of the applicability (or otherwise) of LEM systems and the consequent inappropriate choices for their applications. It will be difficult, for example, for LEMs to displace liquid-liquid extraction or aqueous two-phase extraction when the concentration of the desired species in the feed is reasonably high (15-150 g l-l), the separation is possible under ambient conditions, the losses are affordable, and there are no competing species. Therefore, LEM technology will be economically suitable for the separation of biochemical zwitterions (e.g. amino acids and P-lactam antibiotics), the extraction of stereospecific compounds (e.g. L-phenylalanine) from racemic mixtures using recoverable enzymes, one-step recovery of a product (such as citric acid) which requires many steps in a conventional process, and direct separation and conversion of a fermentation product into another desired endproduct (e.g. recovery of penicillin G and its conversion to 6-APA). Two other applications of LEMs are promising for commercial exploitation. One is the recovery of sensitive products in very pure form from dilute solutions. Many products from recombinant strains are generated in low concentrations in the broth but have to recovered at high yields and purity in order to be less expensive and usable for human consumption through foods and pharmaceuticals. Products such as human insulin, interferons, streptokinase, and P-lactams qualify in this category;
300
I 1 Recovery of Biological Products by Liquid Emulsion Membranes
present methods of separation and purification involve many expensive steps and account for more than half the total cost of production [1,11]. The second area is the utilization of reversed micelles as carriers in a one-step LEM separation of biological molecules. This has two distinct advantages. By being encapsulated inside the aqueous micelles, the molecules are protected from loss of activity due to contact with the organic phase of the liquid membrane. And, while conventional processes employing reversed micelles require an extraction stage followed by back-extraction, the two stages can be combined into one by incorporating the micelles in a liquid membrane. One could even extend the use of reversed micelles to processes where the micelles transport the solute across the membrane and release it into a receiving phase where it reacts further, as in the penicillin example cited. It is in such novel processes and advancing technologies that LEMs and SLMs will find useful and commercially viable applications.
References [l] van Brakel, J., Kleizen, H.H., in: Chemical Engineering Problems in Biotechnology: Wink-
ler, M.A. (Ed.). London: Elsevier Applied Science, 1990; pp. 95-165. [2] Sadana, A,, Beelaram, A.M., Bioseparation 1994, 4 , 221-235. [3] Pellegrino, J.J., Noble, R.D., Trends Biotechnol 1990, 8, 216-224. [4] Thien, M.T., Hatton, T.A., Wang, D.I.C., Biotechnol Bioeng 1988, 32, 604-615. [5] Li, N.N., U.S. Patent 3410794 (1968). [6] Li, N.N., Znd Eng Chem Proc Des Dev 1971, 10, 215-221. [7] Melzner, D., Tilkowski, J., Mohrmann, A., Poppe, W., Halwachs, W., Schugerl, K., Hydrometallurgy 1984, 13, 105-1 23. [8] Volkel, W., Halwachs, W., Schugerl, K., J Membr Sci 1980, 6, 1-5. [9] Ollis, D.F., Thompson, J.B., Wolynic, E.T., A Z Ch E J 1972, 18, 457-458. [lo] Cussler, E.L., Evans, D.F., JMembr Sci 1980, 6, 113-117. [ l l ] Volkel, W., Bosse, J., Poppe, W., Halwachs, W., Schugerl, K., Chem Eng Commun 1984, 30, 55-66. [12] Terry, R.E., Li, N.N., Ho, W.S., J Membr Sci 1982, 10, 305-323. [13] Boyadzhiev, L., Benzenshek, E., Lazarova, Z., J Membr Sci 1984, 21, 137-144. [14] Scheper, T., Adv Drug Deley Revs 1990, 4 , 209-231. [15] Likidis, Z., Schugerl, K., J Biotechnol 1987, 5, 293-303. [16] Thien, M.P., Hatton, T.A., Sep Sci Techno1 1988, 23, 819-853. [17] Arnold, F.H., Trends Biotechnol 1990, 8, 244-249. [18] Halling, P.J., Biotechnol Adv 1987, 5, 47-84. [19] Langer, R., Chem Eng Commun 1980, 6 , 1-48. [20] Florence, A.T., Omotosho, J., Whately, T.L., in: Controlled Release of Drugs: Rosoff, M. (Ed.). New York: VCH Publishers, 1989; pp. 163-184. [21] Gadekar, P.T., Mukkolath, A.V., Tiwari, K.K., Sep Sci Techno1 1992, 27, 427-445. [22] Scheper, T., Likidis, Z., Makryaleas, K., Nowottny, Ch., Schugerl, K., Enzyme Microb Technol 1987, 9, 625-631. 1231 Thien, M.P., Hatton, T.A., Wang, D.I.C., Biotechnol Bioeng 1988, 32, 604-615. [24] Lee, C.J., Yeh, H.-J., Yang, W.-J., Kan, C.-R., Biotechnol Bioeng 1993, 42, 527-534. [25] Bryjak, M., Wieczorak, P., Kafarski, P., Lejczak, B., J Membr Sci 1991, 56, 167-180. [26] Ha, H.Y., Hong, S.-A,, Biotechnol Bioeng 1992, 39, 125-131.
References
30 1
[27] Schugerl, K., Scheper, T., 4th lnt Conf Chemistry Biotechnol Biologically Active Products, Budapest, 1987, pp. 133-152. [28] Rethwisch, D.G., Subramanian, A., Yi, G., Dordick, J.S., J A m Chem SOC 1990, 112, 16491650. [29] Protsch, M., Marr, R., Proc lnt Solvent Extr Conf, Denver, Colorado, 1983, p. 66. [30] Frankelfeld, J.W., Li, N.N., in: Handbook of Separation Process Technology: Rousson, R.W. (Ed.). New York: John Wiley, 1987; pp. 840-861. [31] Likidis, Z., Schugerl, K., Biotechnol Bioeng 1987, 30, 1032-1040. [32] Eyal, A.M., Bressler, E., Biotechnol Bioeng 1993, 41, 287-295. [33] Likidis, Z., Schlichting, E., Bischoff, L., Schugerl, K., Biotechnol Bioeng 1989, 33, 13851392. [34] Likidis, Z., Schugerl, K., Chem Eng Sci 1988, 43, 27-32. [35] Reschke, M., Schugerl, K., Chem Eng Sci 1985, 31, B19-B26. [36] Basu, R., Sirkar, K.K., A 1 Ch E J 1991, 37, 383-393. [37] Hayworth, H.C., Ho, W.S., Bums, W.A., Sep Sci Technol 1983, 18, 493-498. [38] Mohan, R.R., Li, N.N., Biotechnol Bioeng 1975, 17, 1137-1156. [39] Thien, M.P., Hatton, T.A., Wang, D.I.C., A 1 Ch E National Meeting, Miami Beach, FL, 3-7 Nov. 1986. [40] Makryaleas, K., Scheper, T., Schugerl, K., Ger Chem Eng 1985, 8, 345-350. [41] Bryjak, M., Kozlowski, J., Wieczorek, P., Kafarski, P., J Membr Sci 1993, 85, 221-228. [42] Teramoto, M., Matsuyama, H., Yamashiro, T., J Membr Sci 1989, 45, 115-136. [43] Sengupta, A,, Basu, R., Sirkar, K.K., A 1 Ch E J 1988, 34, 1698-1708. [44] Sengupta, A., Basu, R., Prasad, R., Sirkar, K.K., Sep Sci Technol 1988, 23, 1735-1751. [45] Basu, R., Sirkar, K.K., J Membr Sci 1992, 75, 131-149. [46] Yagodin, G., Lopukhin, Y., Yurtov, E., Guseva, T., Sergienko, V., Proc lnt Solvent Extr ConA Denver, Colorado, 1983, p. 385. [47] Chilamkurti, R.N., Rhodes, C.T., J Appl Biochem 1980, 2, 7-16. [48] Rhodes, C.T., Frankerfeld, J.W., Fuller, G.C., Symp Sep Encapsulation by Liquid Membranes, ACS Centennial Meeting, New York, 6 April 1976. [49] Reschke, M., Schugerl, K., Chem Eng J 1984, 28, Bl-B9. [50] Reschke, M., Schugerl, K., Chem Eng J 1984, 29, B25-B29. [51] Muller, B., Schlichting, E., Bischoff, L., Schugerl, K., Appl Microbiol Biotechnol 1987, 26, 206-2 10. [52] Ho, W.S., Proc lntl Congress on Membranes and Membrane Processes, Chicago, IL, Vol. I, pp. 692-694. [53] Kinugasa, T., Watanabe, K., Takeuchi, H., Proc Intl Congress on Membranes and Membrane Processes, Chicago, IL, Vol. I, pp. 706-708. [54] Colinart, P., Delepine, S., Trouve, G., Renon, H., J Membr Sci 1984, 20, 168-187. [55] Danesi, P.R., Richley-Yinger, L., Rickert, P.G., J Membr Sci 1987, 31, 117-145. [56] Neplenbrock, A.M., Bergeman, D., Smolders, C.A., J Membr Sci 1992, 67, 121-132. [57] Neplenbrock, A.M., Bergeman, D., Smolders, C.A., J Membr Sci 1992, 67, 149-165. [58] Simmons, D.K., May, S.W., Agrawal, P.K., in: Downstream Processing and Bioseparation: Hamel, J.F.P., Hunter, J.B., Sikdar, S.K. (Eds.). Washington D.C.: Am. Chem. SOC.,1990; pp. 108-129. [59] Meyer, E.R., Scheper, T., Hitzmann, B., Schugerl, K., Biotechnol Techniques 1988, 2, 127132. [60] Sirkar, K.K., U.S. Patent 4921 612 (1991). [61] Danesi, P.R., Proc lntl Solvent Extraction Conf, Munchen, Germany, Vol I; pp. 527-536. [62] Wiencek, J. W., Qutubuddin, S., A I Ch E National Meeting, New Orleans, Louisiana, 1986. [63] Tsai, S.-W., Wen, C.-L., Chen, J.-L., Wu, C.-S., J Membr Sci 1995, 100, 87-97. [64] Chilamkurti, R.N., Rhodes, C.T., J Appl Biochem 1980, 2, 17-24. [65] Ho, W.S., Hatton, T.A., Lightfoot, E.N., Li, N.N., A 1 Ch E J 1982, 28, 662-670. [66] Matsumoto, S., Inoue, T., Kohda, M., Ikura, K., J Colloid l n t e ~ a c eSci 1980, 77, 555-563.
302
I 1 Recovery of Biological Products by Liquid Emulsion Membranes
[67] Magadasi, S . , Frenkel, M., Garti, N., Kassan, R., J Colloid Interface Sci 1984, 97, 374379. [68] Draxler, J., Man, R., Ber Bunsenges Physik Chem 1982, 86, 64-69. [69] Scholler, C., Chandhuri, J.B., Pyle, D.L., Biotechnol Bioeng 1993, 42, 50-58. [70] Kemena, L.L., Kemp, N.J., Noble, R.D., J Membr Sci 1983, 15, 259-274. [71] Shinbo, T., Yamaguchi, T., Yanagishita, H., Sakaki, K., Kitamoto, D., Sugiura, M., JMembr Sci 1993, 84, 241-248. [72] Armstrong, D.W., Leu, H.L., Anal Chem 1987, 59, 2237-2241. [73] Nagappa, A.N., Bhaskar, K.U., Shanker, S . , Mitra, S . , Srivastava, R.C., Indian J Biochem Biophys 1988, 25, 350-355. [74] Halwachs, W., Volkel, W., Schugerl, K., Proc Zntl Solvent Extraction Conf, Liege, Belgium, 1984; p. 80. [75] Chaudhuri, J.B., Pyle, D.L., Chem Eng Sci 1992, 47, 41-48. [76] Christen, P., Minier, M., Renon, H., Biotechnol Bioeng 1990, 36, 116-123. [77] Baker, R., Controlled Release ofBiologically Active Agents. New York: John Wiley, 1987. [78] Chaudhuri, J.B., Pyle, D.L., Chem Eng Sci 1992, 47, 49-56. [79] Lorbach, D.N., Hatton, T.A., Chem Eng Sci 1988, 43, 405-418. [80] Mugele, R.A., Evans, H.D., Znd Eng Chem 1951, 43 1317-1324. [Sl] Janakiraman, B., Sep Sci Technol 1985, 20, 423-443. [82] Lee, K.H., Lee, S.C., Lee, W.K., J Chem Technol Biotechnol 1994, 59, 365-370. [83] Scheper, T., Halwachs, W., Schugerl, K., Chem Eng J 1984, 29, B31-B37. [84] Molinari, R., De Bartolo, L., Drioli, E., J Membr Sci 1992, 73, 203-215. [85] Shinbo, T., Yamaguchi, T., Yanagishita, H., Sakaki, K., Kitamoto, D., Sugima, M., J Membr Sci 1993, 84, 241-248. [86] Lazarova, Z . , Peeva, L., Biotechnol Bioeng 1994, 43, 907-912. [87] Atkinson, B., Mavituna, F., Biochemical Engineering and Biotechnology Handbook. Basingstoke, U.K.: Macmillan, 1991. [88] Baniel, A.M., European Parent 1049429 (1982). [89] Friesen, D.T., Babcock, W.C., Brose, D.J., Chambers, A.R., J Membr Sci 1991, 56, 127147. [90] Hano, T., Ohtake, T., Matsumoto, M., Ogawa, S.-I., J Membr Sci 1993, 84, 271-278. [91] Lee, K.W., Lee, S.C., Lee, W.K., J Chem Technol Biotechnol 1994, 59, 371-376. [92] Lee, C.-J., Yeh, H.-J., Yang, W.-Y., Kan, C.-R., Biotechnol Bioeng 1994, 43, 309-313. [93] Hatton, T.A., in: Surface-Based Separation Processes: Scametom, J.F., Hanuell, J.H. (Eds.). New York: Marcell Dekker, 1989; pp. 55-90. [94] Mitchell, D.J., Ninham, B.W.J., J Chem Soc Furaday Trans 1981, 77, 601-629. [95] Luisi, P.L., Bonner, F.J., Pellegrini, A . , Wiget, P., Wolf, R., Helv Chim Acta 1979, 62, 740753. [96] Goklen, K.E., Hatton, T.A., Sep Sci TechnoE 1987, 22, 831-841. [97] Hatton, T.A., A C S Symp Ser 1987, Ch. 9. [98] Jolivalt, C., Minier, M., Renon, H., J Colloid ZnterfSci 1990, 135, 85-96. [99] van’t Riet, K., Dejjer, M., 3rd European Congress of Biochemistry, 1984, p. 540. Cited by Jolivalt, C., Minier, M., Renon, H., in: Downstream Processing and Bioseparation: Hamel, J.F.P., Hunter, J.B., Sirdar, S.K. (Eds.). Washington D.C.: Am. Chem. SOC.,1990; Ch. 5, pp. 87-107. [loo] Castro, M.J.M., Cabral, J.M.S., Biotechnol Adv 1988, 6, 151-167. [ l o l l Dekker, M., van’t Riet, K., Baltussen, J.W.A., Bysterboch, B.H., Hilhorst, R., Laane, C., Proc 4th European Congress of Biotechnology 1987, p. 507. [lo21 Leser, M.E., Wei, G., Luisi, P.L., Maestro, M., Biochim Biophys Res Commun 1986, 135, 629- 635. [lo31 Banghman, D.R., Lin, Y.A., Znd Eng Chem Res 1994, 33, 2668-2687. [I041 Dekker, M., van’t Riet, K., Bijsterbosch, B.H., Wolbert, R.G., Hilhorst, R., A I Ch E J 1989, 35, 321-324.
References
303
[lo51 Dekker, M., van’t Riet, J., Bijsterbosch, B.H., Fijneman, P., Hillhorst, R., Chem Eng Sci 1990, 45, 2949-2954. [lo61 Dahuron, L., Cussler, E.L., A I Ch E J 1988, 34, 130-136. [lo71 Woll, J.M., Hatton, T.A., Yarmush, M.L., Biotechnol Prog 1989, 5, 57-62. [ 1081 Cabral, J.M.S., Aires-Barros, M.R., in: Recovery Processes for Biological Materials: Kennedy, J.F., Cabral, J.M.S. (Eds.). Chichester, U.K.: John Wiley, 1993; Ch. 9, pp. 247-271. [lo91 Armstrong, D.W., Li, W., Anal Chem 1988, 60, 86-88. [I101 Tsai, S.-W., Wen, C.-L., Chen, J.-L., Wu, C.-S. J Membr Sci 1995, 100, 87-97. [ 1113 Sadana, A., Beelaram, A.M., Bioseparation 1994, 4, 221-235.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
12 Membranes Modified for Biochromatography Egbert Muller and Elias Klein
12.1 Introduction This chapter reviews the current status of affinity membranes with an emphasis on ion-exchange technology. Strictly speaking, ‘affinity chromatography’ and other affinity separations have generally referred to selective adsorption processes based on steric interactions between biologically important molecules. However, the common usage of ion-exchange columns for separating proteins has often led to the description of these separations under the general term of affinity separations. The difference, of course, is that ion-exchange separations are generally not as specific as true affinity-based separations, although manipulation of loading and elution buffers does provide for an adequate degree of separation in many of the cases to be described here. Both ion exchange and affinity chromatographic columns rely on macroporous beads, having pore diameters generally larger than 15 nm, or 0.015 ym. The physical dimensions of the beads themselves range widely, depending on their material of construction. The earlier gel beads were generally greater than 100 ym in diameter, but more modem materials, such as silica are available at less than 15 pm diameters. The decrease in particle sizes has led to great increases in column operating pressures. Avoidance of high operating pressures is one of the reasons for the recent development of membranes to accomplish the same separations. Another reason is the easier scaleability of membrane-based devices. Membranes used for biological separations, whether relying on true affinity interactions or on ion exchange, have several common characteristics: they are made to have pore sizes in the range of 0.1 to 3 pm to provide high hydraulic fluxes at low transmembrane pressures. They must have relatively large internal surface areas, approaching 5-10 m2g-’ in order to provide adequate binding surfaces. They are generally less than 150 pm thick. And they must provide some chemical functionality to allow modification for the production of active sites. In the sections that follow we will describe a selection of membranes, both commercially available and in development, that are based on several polymeric matrices. The starting membrane is generally one produced for microfiltration applications and then modified to perform in these applications.
306
12 Membranes Modified f o r Biochromatography
12.2 Membrane Types Microporous membranes made from the following materials are available as starting matrices for affinity separation: -
Polyethylene Polysulfone Polyethersulfone Nylon(s) PVC composite Microporous regenerated cellulose
Only the polyethylene and PVC have no readily reactive functional groups for production of ion-exchange sites. However, as described later, this problem is resolved by some investigators using polyethylene through the use of ionizing radiation [l] to produce surface radicals, which in turn serve as initiators for graft polymerization. Both of the polysulfone types can be modified by reaction of their terminal phenolic groups, or by nuclear substitution in the aromatic rings. The nylons can be modified via reactions on their terminal amine groups, and the cellulosics are readily modified via their hydroxyl groups. The PVC composites contain dispersed modified silica particles. Except for the nylon and the cellulosics, these membrane materials are very hydrophobic and require surface modification that will allow wetting. Although the introduction of ion-exchange groups facilitates wetting, they are sometimes further modified with a carbohydrate polymer in order to reduce non-specific adsorption [2]. The physical configuration of these membranes generally falls into two categories; namely sheet membranes and hollow fibers.
12.2.1 Sheet Membranes These are generally produced as rolls, either by solution casting or by melt extrusion. They are packaged in housings suitable for chromatographic applications either as stacked sheets or as pleated sheets in single or multiple layers. The packaging technology is generally derived from microfiltration applications with similar requirements for low pressure drops at high flow rates. Sheet stocks have been used by Sartorius (Germany) [3] and Millipore ( MA) [4] to produce both ion-exchange and protein-modified affinity membranes
12.2.2 Hollow Fibers Hollow fibers have recently gained prominence in the field of microfiltration because they can be packaged in crossflow geometries that lower particulate fouling. This has led to their application in a number of ion-exchange and true affinity separations.
12.3 Binding Chemistries
307
The earliest version of hollow fibers in protein purification applications were those produced by Monsanto [5] for use as ion-exchange media. These were soon followed by r-Pr A modified fibers from Sepracor [6]. Both companies relied on polysulfone matrices. Development programs based on reacting radiation modified polyethylene hollow fibers were reported by Saito [7-91 and more recently a number of affinity modifications have been reported by Akzo and Merck [lo] based on modified nylons. All of these will be described in more detail below, as we discuss chemical modifications used in these applications.
12.3 Binding Chemistries 12.3.1 Differences Between Gel and Membrane Derivatization Chemistry Beaded materials have long been the preferred matrix materials for the chromatographic purification of proteins. One can find interesting similarities when comparing the derivatization procedure for beads and membranes. Basic materials for protein separations are generally hydroxyl group-containing polymers. The first materials for gel filtrations of proteins were dextran and agarose gels [l 11. These natural carbohydrate polymers are hydrophilic and exhibit low non-specific interactions. The glycosylic bonds are alkaline-stable, so that the gels can be purified with sodium hydroxide. Derivatives of these materials are generally based on modification of their hydroxyl groups, although the reactivity of these groups is not high. Consequently, the derivatization chemistry has to be drastic. Ion-exchanger materials can be produced by reaction of alkylarninohalogenalkanes or sulfonic acid halogenalkanes with hydroxyl groups to produce alkaline-stable ether bonds. Alternatively, the hydroxyl groups can be activated with cyanogen bromide [ 121 or diepoxides [ 131 to produce linkages with amines or other hydroxy-terminated molecules. The ligand densities of the hydroxyl groups in the polycarbohydrate gels are generally greater than 1 mmol g-' gel. The dextran and agarose gels were the most common materials in the market until today. A disadvantage is their low mechanical stabilty which limits the linear flow rates that can be used. In recent years many efforts have been made to develop particle gels with higher mechanical stability. Mechanically more stable gels can be made from vinylpolymers (Fractogel@,Toyopearl@)[ 141 and styrene gels (POROS') [15]. However, it must be recognized that these crosslinked polymers are not suitable for membrane fabrication because of their lack of solubility or melting points. These gels have pores between 60 nm and 2000 nm. The size of the pores has to be so large in order to increase the diffusion velocity of the large proteins into the beads. However, for purposes of derivatization chemistry, such increased pore sizes are undesirable; an increase of pore size can decrease accessible surface area and so limit ligand density. The derivatized vinyl and styrene gels have a lower
308
12 Membranes Modified for Biochromatography
hydroxyl group density than the carbohydrate gels, so their ligand site concentration must be increased by subsequent reactions. The surface of the styrene gel is hydrophobic and has to be shielded. Consequently, the characteristics of the mechanically stable materials are poorer for derivatization than for carbohydrate materials. Consequently, different methods of polymer coatings and derivatization were developed. Polymer coatings have diffusion advantages for proteins because of the permeable structure of the coated polymer: these coatings are generally hydrophilic and have high acessible ligand densities [16]. Polymer coatings can be produced by several routes (Fig. 12-1): 1. Reaction of the surface with tailor-made polymers and cross-linking. 2. By physical adsorption and cross-linking. 3. By graft polymerization.
Preferred membrane materials for bioseparations are cellulose, polyamide, poyethylene, and polyethersulfone. From these polymers microporous membranes with pore sizes larger then 100 nm can be produced. Although the accessible surface area with these membranes is also low, their derivatization chemistries do not differ widely from the chemistry for particulate beads. However, these polymers are materials which are not generally utilized as bead materials (polyamide, polyethersulfone, and polyethylene). Non-polymeric derivatization procedures for polyamide and polyethersulfone starts with modification of the end groups of the polymers, a procedure not commonly used in particle chemistry. With these derivatization procedures the ligand densities are low and the membranes are only suitable for affinity adsorption. For ion-exchange applications, the ligand density has to be higher: this can be achieved by polymeric coatings. Tables 12-1 and 12-2 show some properties of the basic materials and a selection of different derivatization procedures.
reaction of reactive ligands at the surface with polymers and cross-linking
Fig. 12-1.Different coating procedures.
physical adsorption of polymers and cross-linking
graft- and blockpolymers
12.3 Binding Chemistries
309
Table 12-1. Properties of membrane materials. Ligands
Basic material
Ligand densities
Derivatization techniques for affinity membranes
Hydroxy groups
Cellulose
> 1 mmolg-'
Same as in particle chemistry (CNBr or epoxy)
Carboxy, amino (end groups)
Pol yamide Nylon 6,6 and nylon 6
< 20 pmolg-'
Same as in particle chemistry, covalent reaction with other polymers
Aromatic (main chain) Hydroxy groups (end groups)
Polyethersulfone
> 1 mmolg-' > 1 mmol g-' < 20 pmolg-'
No chemical to modify
Polyethylene
-
Amide (main chain)
Adsorption of polymers, covalent bonding of polymers Plasma grafting of hydrophilic polymers polymers
Table 12-2. Coating procedures. authorlcompany
method
ligands
Pall corp. [17]
Activation of nylon membrane with cyanurchloride and dicylohex ylcarbodiimide Reaction of nylon with an polyamide-polyamine epichlorhydrine polymer Adsorption of hydroxyalkylcellulose on a polyethersulfone membrane Cross-linking of hydroxypropylacrylate and tetraethylenglykoldiacrylate (Polyvinyldifluoride membrane) Reaction of polyethersulfone or polyamide with acid chlorides Reaction of polyethersulfone membranes with bisoxiranes and cellulose Graft polymerisation on to amide bond of nylon, adsorption of nylon on to cellulose and subsequent graft polymerization Cross-linking of vinylmonomers on polyethersulfone (acrylamidopropylmethylsulfonic acid)
Activated membrane for affinity
AMF, Inc. [18]
Brunswick Corp. [19]
Millipore Corp. [20]
Pall Corp. [21] Sepracore Corp. [6]
Sartorius GmbH [22]
Monsanto Corp. [5]
Anion exchanger
Hydroxy groups for affinity membranes Hydroxy groups for affinity membranes
Affinity membrane Hydroxy groups for affinity membranes Epoxy groups for affinity and ion exchanger
Cation exchanger
3 10
12 Membranes Modified for Biochromatography
Table 12-2. (continued). author/company
method
Block polymerization to the end Merck KGaA and AKZO-NOBEL [10,23] groups of nylon Saito et al. [7,8,9] Graft polymerization to polyethylene Anspach, B. [24] E. Klein [25-271
Gelman Science [28] Gelman Science [29] Gelman Science I301
Formyl reaction with nylon and subsequent grafting of cellulose Reaction of polyamide with carbonyldiimidazole, cyanurchloride Reaction of bisoxiranes with polyamide and subsequent grafting of cellulose Cross-linking of vinylmonomers Reaction of polyamide membranes with polyacrolein Reaction of polyethersulfone with bisoxiranes and subsequent reaction with polyethylenimine
Epoxy groups for affinity and ion exchanger Epoxy groups for affinity, hydrophobic interactions and ion exchanger Hydroxy groups Affinity membranes
Cation exchanger Affinity membranes
12.3.2 Different Ligands Ion-Exchanger Membranes In the past the protein binding capacities of ion-exchanger membranes were not satisfactory. The binding capacities were too low compared with that of the particles. However, with the development of polymer coatings the situation has been changed. In particle chemistry, the most widely used ion exchangers are the DEAE-type (N,Ndiethylaminoethyl-) - a weak anion exchanger, TMAE-type (trimethylammoniumethyl) - a strong anion exchanger, and the S-type (sulfonic acid) - a strong cation exchanger. Most of the separation problems can be solved with these kinds of exchanger [31]. All these modifications are also possible to produce for membrane exchangers. As mentioned previously, the derivatization chemistries are very different. In the following text only some selected examples of different derivatization procedures and properties for membrane ion exchangers will be shown. Saito et al. [7,8] describe how electron beam-irradiated porous polyethylene hollow fibers were reacted with glycidylmethacrylate. By subsequent reaction with diethylamine or sulfite salts, a weak anion exchanger and a strong cation exchanger can be produced. The DEA-exchanger has a binding capacity of up to 400 g (bovine serum albumin, BSA) per kg hollow fiber. The dynamic binding capacity from a weak anion exchanger (reaction of the oxirane groups with ethanolamine) was 2530 mgmL-' of hollow fiber. Saito determined the protein binding capacity for a
12.3 Binding Chemistries
31 1
monolayer and pointed out that for such protein-binding capacities, the proteins are arranged in a multilayer arrangement. The sulfo exchangers showed binding capacities of 200 g kg-' of hollow fibers, the recoveries of the desorbed proteins being in the range of 90-100%. Contrary to what is seen with bead columns, the binding capacities of these membranes were totally independent of linear flow rate. Different types of ion exchangers were developed by Sartorius [22], and these are now commercially available. The membranes consist of cellulose and are flat sheets with binding capacities in the range of 30 mg protein binding capacity per mL membrane volume (anion and cation exchanger). These high binding capacities are also achieved with a polymeric coatings. Merck KGaA and AKZO-NOBEL [ 10,231 developed hollow-fiber ion-exchanger modules with protein binding capacities of 40-50 mg of bovine serum albumine per mL membrane volume for a DEA-exchanger and 50 mg lysozyme for a strong cation exchanger. The binding capacities are also independent of linear flow rate. The derivatization procedure for these hollow fibers is a block polymerization. It is interesting to remark, that the binding capacities for different basic materials with different polymer grafting procedures have similar protein binding capacities. Several other ion-exchanger membranes are available commercially, but these are not generally suitable for protein separations.
Membranes for Hydrophobic Interaction Chromatography Hydrophobic interaction chromatography for the separation of proteins is another chromatographic method [32]. The ligands at the surface of a support are mostly alkyl chains and aromatic rings. The proteins have to be bound at a high salt concentration and can be eluted by decreasing the salt concentration. The ligand densities for the hydrophobic ligands are in the range of 20-40 pmol mL-'. A major problem with hydrophobic interaction chromatography is the recovery of proteins, with better results being obtained from short alkyl chains such as propyl and ethyl. Most of the membrane materials have the disadvantage that they have unmodified surfaces with strong non-specific binding and thus are not very suitable for hydrophobic chromatography. By coating the suface with a much more hydrophilic polymer it is possible to make membranes for hydrophobic interaction chromatography. Saito et al. [9] have described the preparation and chromatographic properties of a phenyl group-containing membrane for protein separations. A microporous polyethylene membrane was radiation-modified to induce graft polymerization with glycidyl methacrylate and subsequent ring opening reaction with phenol. The residual epoxy groups were opened by hydrolysis in acid. Because of the balance of phenyl and diol groups, the membrane is suitable as hydrophobic interaction material. The binding capacities (BSA) of the phenyl derivative fiber was in the range of 50% higher than for the hollow fiber free of any phenyl groups. The binding capacity was 30 mgg-' of membrane. The recovery for bovine serum albumin was 87 % for the phenyl fiber, whereas the the original hollow-fiber membrane made of polyethylene exhibited a low recovery of 17 %. This was due to non-selective adsorption, i.e. irreversible adsorption of BSA on to polyethylene.
3 12
12 Membranes Modified for Biochromatography
Affinity Ligands In order for affinity ligands to bind covalently to the surface of a membrane they must first be activated. This can be done with a bifunctional activating reagent, or by polymer coating.
Low-molecular weight activating reagents There are many different activating methods which can be used for the activation of membranes. In the case of cellulose membrane, all procedures normally used with particles can be used in the same manner on the membrane. For uncoated polyamide and polyethersulfone, the selection of these methods is rather poor because the ligand densities of the available end groups are low. The most commonly used activation methods for hydroxy group containing, or hydroxy group coated membranes are shown in Table 12-3. For polyamide 6 and polyethersulfone membrane materials the most successful activation method is to use a three-step procedure [6]:
1. Activation of the end groups of nylon or polyethersulfon with one of these previously described methods. 2. Coating with a polycarbohydrate (most hydroxyethylcellulose). 3. Using the low-molecular weight leash chemistry again to immobilize the ligands. Linker or leashes, activating by polymer coating Especially for purposes of affinity chromatography, the spatial availabiliy of immobilized ligands is very often substantially improved by coupling of the ligands to ‘spacers’ in order to keep them at a distinct distance from the surface of the matrix. However, the influence of the particular spacer has sometimes been underestimated. In contrast to the situation in ion-exchange media, the ligand density in affinity matrices is usually low. Therefore the surface of the matrix is not effectively shielded
Table 12-3. Low-molecular weight activating methods. Activation method
Reactive ligands after reaction with OH-groups
Binding type after reaction with an amino group
Cyanogen bromide [12]
Cyanate ester
Iso-urea
Tosyl chloride [33]
Trifluormethanesulfonylester
Secondary-amine
N-hydroxy succinimide [34]
N-hydroxy succinimidester
Amide
Carbonyldiimidazole [35]
Imidazoly lcarbonate
Urethane
1,4 Butanediol diglycidol ether [I31 2-Fluoro-1-methylpyridine[36]
Oxirane
Secondary-amine
2-Alkoxy-1-methylpyridinium
Secondary-amine
12.3 Bindina Chemistries
3 13
from interactions with the analyte. Matrix, spacer, ligand, and solute form a very complex system preventing prediction of the precise role of the particular spacer arm. Many investigators advocate the routine use of a spacer. When comparing spacer and non-spacer options, we find situations where spacers can help, and situations where they do not [37]. Improvements of this situation could be expected by construction of affinity media with an effectively shielded matrix, high ligand density, hydrophilic spacer arms, and good spatial accessibility of the ligands. A recent attempt to develop such a type of affinity support for particles took advantage of the so-called ‘activated tentacle chemistry’ developed by Merck KgaA [38]. Figure 12-2 shows the differences between the three immobilization procedures. At first, it is possible to immobilize an ligand without spacer to the surface. The ligand densities can be high, but the availibility is low. The next possiblity is to use a conventional spacer. The major chemical leash species are [39]:
- Alkylchains (hexamethylendiamine) Amino acids (lysine) Diepoxides (1,4-butanediol diglycidol ether) Dialdehydes (glutaraldehyde) - Polyethyleneglycol -
Access to the ligands is much better, but the ligand densities are decreasing because it is impossible to obtain complete reaction, as with the activation chemistry without a spacer,
direct coupling to the surface (no spacer)
coupling to the free end of normal spacers (alkylamine, ether, amino acids n=4-12)
L coupling of ligands to grafted polymer chains (free rotating) tentaclearrangement m-20-50 Fig. 12-2. Spacer technologies.
L
L
L
1.
3 14
12 Membranes Modified for Biochromatography
A totaly different situation is encountered when a polymer coating is used. By reacting a surface with glycidyl methacrylate or with hydroxyethylcellulose, it is possible to obtain a surface with a high ligand density and good a c e s to ligands. Such a situation appears similar to that described for ion-exchanger membranes.
Group-Specific Ligands Group-specific ligands are those based on biological recognition parameters but not targeted to a very specific conformation or sequence of the ligate. They may be large macromolecules such as protein A, or they may be small nucleotides, dyes, or peptides.
Protein A Protein A is a component of the cell walls of most strains of Staphylococcus aureus. It shows high binding avidity for many immunoglobulins, but principally for the Fc region of IgG [40]. This property has allowed rapid isolation of human immunoglobulins, as well as their subclasses, and is used extensively in the recovery of immunoglobulins from a variety of sources, including production of monoclonal antibodies from cell cultures. Protein A has also been used as ligand in medical devices to reduce immune complex, as well as IgG concentrations. The most prominent example of bound protein A is the hollow-fiber cartridge device produced by Sepracore [41]. The microporous membrane consists of a blend of polyethersulfone (PES) and polyethylene oxide (PEO) coated on all surfaces with a covalently bound layer of hydroxyethylcellulose (HEC). The PES backbone provided physical strength, and the HEC coating provided activatable surface hydroxyl groups to which ligands could be covalently attached. Both the PEO and HEC conferred low non-specific protein binding characteristics. The modules contained recombinant protein A as received. The hydroxyl groups in the membrane-bound HEC had been activated by reaction with 2-fluoro-1-methylpyridinium toluene-4 mlfonate (FMP). The protein A was reacted with the activated membrane to form a stable secondary amine. The protein binding capacity for IgG was in the range from 3 to 25 mg/mL-' (based on total membrane volume). Klein et al. described a Nylon-based bound protein A [27] which was prepared according to the following procedure:
1. 2. 3. 4. 5.
Reaction of the end groups with glutaraldehyde. Reduction of Schiff bases to secondary amines. Reaction with hexanediamine. Reduction. Binding of protein A by carbodiimides.
The protein A was immobilized on hairpinned hollow fibers and showed IgG binding capacities from 25-30 mg mL-I membrane.
12.3 Binding Chemistries
3 15
Thiophilic ligands Thiophilic ligands are a good alternative to the described protein-based immunosorbent. The ligands are a series of sulfone and thioether bridges. The ligands exhibit affinity for many kinds of IgG and also to the IgY from hen eggs [42]. The reason for this behavior is a specific interaction between sequences of aromatic amino acids at the surface of the immunoglobulins and the sulfur-containing ligands. As reported by Wilchek [43] a Trp-Trp sequence conserved in all IgG at the hinge region is the explanation for binding of all IgG types to thiophilic sorbents. The chromatographic conditions are straightforward. The technique is similar to that of smooth, specific hydrophobic interaction chromatography. The antibodies are loaded in 0.6-0.8 M ammonium sulfate buffer, pH 7 , and can be eluted by decreasing the ammonium sulfate comncentration in the buffer. The advantage of the technique is elution at neutral pH and hence the danger of denaturation is diminished. Unfortunately, these sorbent conditions are not suitable for clinical applications. Finger et al. [44] described a thiophilic flat sheet membrane for the purification of monoclonal anibodies. The dead-end configuration showed a protein binding capacity of 0.02 mgcm-2 and, by arrangement in a spiral membrane, a capacity of 0.03 mg cm-2. Internal clogging was observed with the dead-end modules, caused mainly by the presence of BSA. The purification could be repeated only with the spiral module.
Biomimetic dyes Immobilized dyes have been found to act as pseudoaffinty sorbents for a large number of biological molecules. Triazine-linked dyes have been used extensively to mimic coenzymes that bind a number of dehydrogenases, hexokinases, alkaline phosphatase, carboxypeptidase G, and ribonuclease A. The dyes are chemically stable, relatively inexpensive and have binding coefficients in the range needed for ready elution of ligates. The Procion Blue [45] binding of albumin is a wellknown method for purifying that protein from contaminating transferrin, a protein of similar physico-chemical properties that often contaminates albumin. Champluvier and Kula [46] described a nylon membrane which contained Cibaron blue F3G-A. The dye-ligand was coupled to the nylon-based membrane either directly or using 1,6-diarninohexane or polyethylenimine as spacer. The dye capacity was 7 pmolmL-' membrane, and the protein binding capacities were in the range of 10-20 mg mL-' membrane (pig heart mitochondria1 malate dehydrogenase). Blue Sepharose showed a binding capacity of 12 mg per mL packed gel.
Metal chelates Several terminal amino acid sequences in proteins have been found to act as metalion ligands. These bind strongly with partially coordinated metal ions, generally
3 16
12 Membranes Modifed.for Biochromatography
Cu2+,Zn2+,and Ni2+.Chromatography based on separations effected by immobilized chelates to which metal ions have been coordinated is called immobilized metal-ion affinity chromatography (IMAC). The principles of the separations and many applications have been described in recent articles, as well as in the original introduction of the method by Porath et al. [47]. In practice, an organic grouping known to chelate metal ions, such as iminodiacetate (IDA), is covalently linked to a microporous support. The support can be a hydrogel such as Sepharose, or for high-pressure operations a chemically modified porous silica. The IDA atoms not involved in covalent linkage to the support participate in ligand-ligate formation with a metal cation. Since the ligand generally does not occupy more than three coordination sites in the ion’s coordination sphere, residual coordination sites remain. These can be filled with imidazole or thiol side chains on proteins or with certain sequences of protein terminal amino acids. For the purification of proteins produced by genetically modified bacteria, this latter approach offers offers the possibility of building into the product its own purification system. Smith et al. [48]showed that the peptide sequences His-Gly-His, HisTyr-NH2, and His-Trp have a high affinity for immobilized transition metals, More recently, these same authors showed that the ‘ability of a chelating peptide, such as His-Trp to bind immobilized transition metals can be transferred to a larger peptide and ultimately a recombinant protein’. The introduction of a terminal peptide useful for purification requires that the extra peptide can be removed to generate the desired product. Depending on the protein structure, the His-Trp peptide can be removed by tryptophan oxidative cleavage or by CNBr cleavage. Chelating proteins can be desorbed from immobilized metal ions by use of mass action displacement, or by changing the milieu to alter the metal-protein dissociation constant. Aliphatic mono- and diamines are useful displacing agents, since they compete for the metal-ion coordination sites. Combined with shifts in pH, the use of site-competing amines provides a rapid means of displacing metal-ion-coordinated proteins. Anspach et al. [24] described a nylon flat membrane with immobilized iminodiacetic acid. The membranes were obtained by a three-step procedure: (i) reaction of microfiltration nylon membranes (N66) with formaldehyde or with bisoxirane; (ii) coating with hydroxyethylcellulose; and (iii) immobilization of iminodiacetic acid by standard oxirane chemistry. The membranes showed copper binding capacities up to 100 nmolcm-2. The protein binding capacities were nearly 300 pgcm-2 for lysozyme (hen egg white). These values can be converted to unit volume of membrane using the manufacturer’s value of thickness. Values in the range of 30 mgmL-’ were found; these are in the same range as observed for particulate beads.
12.4 Kinetic Advantages of Membranes The preceding discussions described the basic membrane materials and types of modification. The chemistries of modifications differ less between bead and membranes. The membranes are also porous bodies with pore diameters in the range of
12.4 Kinetic Advantages of Membranes
317
0.1-2 pm. Newer bead materials also have large pores with the same size as those in the membranes. However, there is one major difference in operating conditions, namely the time required for the ligate to reach its binding site is much less when membranes are used compared with the use of bead columns. This is a consequence of the fact that two different mechanisms are employed. With membranes, hydrostatic pressure is used to convect the ligate to the binding site by causing the feed solution to flow through the membrane. In contrast, with beads, the ligate must diffuse into the bead to reach the binding site. Moreover, the distances involved are much smaller for the membrane process (i.e. less than one pore diameter of approx 1.0 pm), whereas the diffusion path for the bead may be as high as 50 pm [27]. Although some attempts have been made in recent years to produce beads with macroporous channels passing through them, in order to facilitate convective transport, the applicability of this mechanism to enhanced transport with beads has been debated [ 151. The analysis of residence times in porous materials can be made according to a Peclet analysis. The Peclet number is defined as: Pe = tD/tC = diffusion time/column residence time
(1)
or
where tD is the time required for diffusion to the ligate site; tc is the column residence time; LD is the diffusion path length in the material; D is the ligate diffusion coefficient; VOis the interstitial volume; and Q is the volume flow rate. The residence time needed for an efficient use of affinity sites is clearly a function of the diffusion path length and hence the particle size. The shorter the diffusion distance, the more rapid can be the flow of the feed solution. This was recognized long ago and led to the development of smaller-diameter chromatography support. However, smaller particles pack more densely and create higher pressure drops. Conceptually, the ultimate effort in this direction is to arrange the column material in an extremely broad, very shallow configuration - that is, like a membrane. In such a configuration even very slow perfusion rates, such as predicted by the Peclet analysis and by experience, could produce suitable filtrate flows. But even when affinity beads are arranged in sheet form, the diffusion distance from bead surface to bead core is not decreased, and the linear velocity through the shallow column is limited. By contrast, when a solution permeates through a microfiltration membrane, each solution element is convected to within 0.05 to 1.5 pm of the active site. Membrane diffusion distances are markedly smaller than those of beads. For Sepharose, Klein et al. [27] obtained a critical Peclet number (value that must not be exceeded to obtain efficient utilization of the resin capacity) in the range of 0.01. The diffusion path length of the ligate for Sepharose is set to 50 pm. The diffusion path length for membranes are much shorter. By assuming a path length for membranes of 1 ym, the flow rate can be 2500 times higher, for the
3 18
12 Membranes Modified for Biochromatography
same efficency. However, the intraparticle Peclet analysis alone is inadequate to describe fully intraparticle mass transfer when adsorption occurs. Here, the Damkohler number (second kind) comes into play: DaII = diffusion time/association time (ligate-ligand)
(3)
For the Interaction of IgG with immobilized protein A on to hollow-fiber membranes, Colton et al. 1411 determined a value of DaII in the range of 0.7. The diffusion time is 0.004 sec and the reaction time 0.06 sec. Thus, the process is reaction-limited and not determined by diffusion alone, thereby leading to a homogeneous-like reaction behavior in the fluid phase between IgG and immobilized protein A. The results can be summarized in the following relationships:
convective flow > flow by diffusion (membranes) convective flow c flow by diffusion (particles) The flow in membranes is thus a convective flow and does not depend on diffusion. This is in direct contrast to the situation in particles.
12.5 Packaging of Membranes Microporous membranes are produced in both sheet and hollow-fiber forms. These two geometries permit a wide range of packaging formats to be used, depending on the intended application. Hollow fibers have the advantage that packaging designs can be developed to permit binding of soluble ligates directly from suspension, using crossflow filtration. On the other hand, sheet membranes are relatively simple to package in conventional filter designs so that special centrifugation, potting and cutting tools are not required, as is the case for production of hollow-fiber modules. Figure 12-3(a) shows a traditional packaging of sheet membranes in stacked form. As the number of layers increases, the pressure required to perfuse the stack also increases. A benefit of this design is that there is essentially plug-flow through the package; only a small void volume - other than within the membrane - exists because of the header and collection geometry, and possible spacers used between the membranes. One of the difficulties in this stacked sheet design is the problem of feed distribution as the area is increased. This can be minimized by the use of pleated sheets in the module, as shown in Fig. 12-3(b). Pleating increases the available cross-sectional area but does not increase the volume of membrane in the housing. Generally, only three to four layers of membrane can be pleated simultaneously. While the pressure drop in this design is low, so is the packing density and some loss of ‘plug-flow’ characteristics occurs. Figs. 12-4(a,b) show how follow fibers can be arranged to provide either a deadend or a crossflow geometry, respectively. In the dead-end geometry, all of the solution entering the module must pass through the fiber walls - the site of ligand
12.5 Packaging of Membranes
3 19
Fig. 12-3. (a) Schematics of a dead-end stacked sheet module. (b) Schematics of dead-end pleated sheet module.
Fig. 12-4. Schematics of (a) dead-end hollow-fiber module and (b) crossflow hollow-fiber module.
binding - before exiting the module. When high fiber packing densities are achieved this design yields essentially plug flow. The crossflow geometry shown in Fig. 12-4(b) allows for a variable fraction of the fluid entering the fiber lumens to pass through the fiber walls into the module shell space. The advantage of this design is that it is capable of handling dispersion as well as clarified solutions. The flow in the fiber lumen is laminar and the behavior of particles in such a channel has been the object of much study [51]. A fraction of the fluid can be conducted through the fiber walls without fouling the pores of the fiber because of lift forces that tend to move particles in a direction opposite to the transmembrane flow. The fraction of the clear fluid which can be convected through the fiber walls depends on a number of variables, including the velocity within the lumen, the particle concentration, the particle size and density, and the opening pore size of the fiber wall. When preclarified solutions are used then these same modules can be used in the dead-end mode. However, the extra-luminal volume for these modules is higher than in the dead-end geometry, and some tailing of solutions (caused by mixing in the shell) occurs.
320
12 Membranes Modified .for Biochromatography
A major objective for using affinity membranes is, of course, the ability to recover purified and more concentrated forms of the target ligate. Methods of achieving high binding densities have already been discussed. Here, we consider how the module geometry can affect concentration of the eluted ligate. At the end of the loading cycle the membranes are generally washed to removed non-bound species. At some point the eluate is switched to provide conditions for the release of the ligate. When this happens the module is still filled with inert loadinglwashing buffer. Ideally, the eluate solvent will move through the module in a single front (i.e. plugflow) to sweep the ligate ahead of it and out in a single bolus. To characterize the behavior of the modules described, the Residence Distribution Times of the designs shown in Figs. 12-3 and 12-4 are shown in Fig. 12-5. The determination is made by producing a step change in the inlet concentrations of a fluid being pumped at constant speed through the modules. The outlet concentration is monitored from the instance of the change and expressed as CtIC,, where C, is the feed concentration of a non-binding solute, and Ct is the effluent concentration at time t. As flow begins, only the equilibrium buffer exits the module and Ct = 0. At some point the non-binding solute begins to appear at the exit point of the module. To determine how this is related to the interstitial volume of the module, we plot t l T , where t is the elapsed time since the step change and T is the time required to sweep out one interstitial volume of the module. When the curve rises at tlr = 1.0, the process is plug-flow [49]. If some of the non-binding solute exits the module prior to this value it indicates that part of the initial buffer has been retarded in its exit from the module; i.e. some mixing has occurred. As shown in Fig. 12-5, only the pleated modules and the crossflow modules show any significant signs of mixing. A more recent design for handling dispersions is an internal crossflow module developed by Akzo Nobel. The basic concept is illustrated in Fig. 12-6. Here, hollow fibers bent into a hairpin configuration is placed in a small-diameter tube. The packing density is deliberately made to approach 50-70 % so that dispersion flowing external to the fiber will experience a high pressure drop. At the inlet end of the Residence Distribution Times 1.2
Dead-end HF
1
-.-
Stacked Sheets
0.8 "0
+
0.6
Cross-flow 70 HFa
0.4
-+-
-
Crossflow 64 HFa
0.2
0
Pleated
t/c
Fig. 12-5. Residence distribution times for stacked sheet, pleated sheet, dead-end hollow-fiber and two crossflow hollow-fiber modules (64 and 70 fibers, respectively).
12.5 Packaging of Membranes
321
Fig. 12-6. Schematics of the Internal crossflow concept.
module there is pressure difference between the external and lumen sides of each fiber, causing a flow of the solution through the fiber walls. Since the pore sizes are 0.2-0.45 pm, none of the particles can permeate the fiber wall, and all are swept downstream to the exit of the tube. The pressure difference between outside and lumen of the fiber decreases as the solution travels toward the exit of the tube. At the point where the fibers end, the pressure difference is zero. The result of this process is that a fraction of the inlet stream is caused to permeate the fiber walls to contact the ligand sites and then exits through the fiber lumens to mix with the external stream at the open ends of the fibers. The fraction of the inlet stream which is shunted through the fiber walls is a function of the relative resistance of the shell flow path and the pressure drop required to permeate the fiber walls and flow out through the lumen. The resistance in the shell side of this device depends on the packing densities and spacing of the fibers and on the viscosity of the dispersion. Since the latter changes as the solution flows toward the exit, modeling of this process is extremely difficult. For purification of ligates from clear solutions it is desirable that as much of the affinity membrane's equilibrium capacity is loaded before any significant concentration of the ligate breaks through into the outflow from the cartridge. In column chromatography, this is achieved by using slow column perfusion rates and ensuring that no channeling exists. With membrane affinity devices, this problem is much less important since no significant diffusion time is required, as described earlier in the discussions of binding kinetics. We illustrate this using a dead-end module made with nylon hollow fibers modified with covalently linked protein A. The target IgG was a commercial immunoglobulin mixture used clinically. The breakthrough curve for both BSA (a non-binding species) and h-IgG at a feed velocity of XX bed volumes min-' are shown in Fig. 12-7. The residence time is in seconds. The capacity of the module, at the point when 10 % of the feed concentration breaks through, as a function of increased pumping rates, expressed as BVs min-' is
322
12 Membranes Modified f o r Biochromatography 5.0 Bed volumeslmin;
1.04 ml Dead-end HF Module 50
40 P
e
f 30 0
g20
E"
wO
IF
30 40 Bed Volume8
20
50
10
60
n
wO
10
20 30 40 rng IgG Offered
50
Fig. 12-7. IgG breakthrough and capture efficiency by recombinant protein A modified nylon hollow-fiber module.
30
f 25
2-
20
1:
'C
15
0
$) - 10 5
5
10
15
20
25
Linear Flow Velocity (BV/min) HF Nylon Module
High Perf. Beads
Fig. 12-8. Effect of superficial velocity on capacity at C/Co = 0.1 breakthrough.
shown in Fig. 12-8. On the same graph is a plot for a modem high efficiency resin bed. It is clear that the dynamic range of the membrane is immense compared with that of the bead column. The perfusion of a protein A-modified internal crossflow module with an IgG solution prepared with 5 % ( v h ) baker's yeast is shown in Fig. 12-9. The fall in IgG in the reservoir is plotted as a function of the number of bed volumes pumped through the module.
12.6 Ion Exchanger Hollow Fibers
323
pH = 7.4, flow: 50 mL/min, 145 mL reservoir, BV = 14.9 mL Fraction of Capacity bound
IgG mass adsorbed
O L ' 0
I
"
'
"
100 150 Bed Volumes perfused 50
0
50 100 150 Bed Volumes perfused
u No cells
IgG capacities: 5 % yeast No cells
5% yeast vlv
16.8 mg/mL membrane 20.6 mg/mL membrane
Fig. 12-9. IgG adsorption from 5 % yeast in physiological buffer solution (PBS).
12.6 Ion Exchanger Hollow Fibers As described in Chapter 6, the dead-end geometry of hollow-fiber bundles exhibits plug-flow without back-mixing for an unbound species. This is the same flow characteristic observed for a dead-end flat sheet membrane configuration or for particle beds. By using modified hollow fibers it should be possible to separate ligates in the same way as described for the other types of separating arrangements. From Merck KGaA and AKZO-Nobel, dead-end hollow-fiber ion exchangers were described [50] for fast separations of proteins. Table 12-4 shows some characteristic data of these ion exchangers. The hollow fibers are produced from high-molecular weight polyamide 6. They are macroporous with a mean pore diameter of 0.6 pm. The outer diameter of the fiber is 0.5 mm and the inner diameter is 0.3 mm. The hollow fibers in a hairpin configuration are embedded with both ends in potting material. At the surface of the hollow fiber is a covalently bound, solvent stable polymer coating. The membrane is stable against purification with alkaline solutions. The hollow-fiber ion exchangers are able to act as fast separation units, one advantage being the low pressure drop compared with that of beads. Fast gradient separations of standard proteins, are illustrated in Fig. 12-10.
324
12 Membranes Modified .for Biochromatography
Table 12-4. Characteristics of hollow-fiber ion exchangers. Module dimensions
Type of ion exchanger
Number of fibers
Membrane volume
Recommended flow rates (ml min-')
Dynamic protein binding capacity
80-6 mm
DEA, weak anion
64
0.7
0-100
80-6 mm
Sulfo, strong cation
64
0.7
0-100
35 mg, bovine serum albumin per module 40 mg, lysozyme per module
anion exchanger
cation exchanger absorption units
NaCl [molfl]
KCI [mol/l]
absorption units
L
1
1
2
3 4
t[min]
sample: 2.5 mg chymotrypsinogeneA and lysozyme buffer A: 20 mM sodium hydrogenphosphatepH 7 buffer 8: 20 mM sodium hydrogenphosphatepH 7 and 1 M sodium chloride flow: 20 mlfmin monitor: 280 nm column: 80 - 6 mm dead end hollow-fiber ion exchanger (Sulfo-type)
1
2
3
4
5
6
t [min]
sample: 1.3 mg transferrin, 3 mg ovalbumin and 2.8 mg beta-lactoglobulin buffer A: 20 mM Tris-buffer pH 8.5 buffer 8:20 mM Tris and 0.4 M potassium chloride 10 mlfmin flow: monitor: 280 nm column: 80 - 6 mm dead end hollow-fiber ion exchanger (DEA-type)
Fig. 12-10. Protein separations with dead end hollow-fiber ion exchanger modules.
References
325
References Saito, K., Ito, M., Yamagishi, H., Furusako, S . , Sugo, T., Okamoto, J., Ind Eng Chem Res 1989, 28, 1808-1812. Klein, E., Feldhoff, P.A,, US Patent 5 035 133, 1991. Laboratory Separations Products Reference Handbook, Sartorius Corporation, Edgwood, NY, 1994. Kracke, C., BioForum 1993, 12, 480. Henis, J. M. S . , Tripodi, M. K., Stimpson, D. I., European Patent 0221046 BI, 1986. Azad, A. R. M., Goffe, R. A,, Patent PCT WO 90/04609, 1989. Kobayashi, K., Tsuneda, S . , Saito, K., Yamagishi, H., Furusaki, S . , Sugo, T., J Membr Sci 1993, 76, 209-218. Tsuneda, S . , Shinano, H., Saito, K., Furusaki, S . , Sugo, T., Biotechnol Prog 1994, 10, 76-81. Kubota, N., Kounosu, M., Saito, K., Sugita, K., Watanabe, K., Sugo, T., J Chromatogr 1995, 718, 21-34. Miiller, E., Baurmeister, U., Abstracts, I 1 th Meeting of ISPPP, Luxembourg, November, 1996. Porath, J., Flodin, P., Nature 1959, 193, 1657-1659. Axen, R., Porath, J., Ernback, S . , Nature 1967, 214, 1302-1303. Sundberg, L., Porath, J., J Chromatogr 1974, 90, 87. Sasaki, H., Komiya, K., Kato, Y., European Patent 0006199 B1, 1979. Afeyan, N. B., Gordon, N. F., Mazsaroff, I., Vrady, L., Fulton, S. P., Yang, Y. B., Regnier, F. E., J Chromatogr 1990, 519, I . [16] Ratanayake, C. K.,Regnier, F.E., J Chromatogr 1996, 743, 15-23. [17] Degen, P. J., Martin, J., Schriefer, J., Shirley, B., US Patent 4693985, 1987. [18] Barnes, R. G., Chu, C., Emond, G. T., US Patent 4473475, 1984. [19] Wrasidlo, W. J., Mysels, K. J., US Patent, 1983. [20] Steuck, M. J., US Patent, 1986. [21] Troth, H. G., PCT 94/17903, 1994. [22] Demmer, W., Horl, H.-H., Nussbaumer, D., Weiss, A. R., Wiinn, E., PCT 92/15637, 1992. [23] Miiller, E., Gensert, R., PCT WO 9622316. [24] Beeskow, T.C., Kusharyoto, W., Anspach, F. B., Kroner, K. H., Deckwer, W. D., J Chromatogr 1995, 715, 49-65. [25] Klein, E., European Patent 0441660 A ] , 1991. [26] Klein, E., Eichholz, E., Yeager, D. H., J Membr Sci 1994, 90, 69-80. [27] Klein, E., Affinity Membranes. New York: John Wiley & Sons, Inc., 1991. [28] Wolpert, S . , PCT 91/10498, 1991. [29] Hu, H., Hou, C.-J., US Patent, 5531893, 1996. [30] Pernawan, K. P.W., Heisler, M. D., Saline, M., Kraus, M. A,, European Patent 0308206 BI, 1988. [31] Janson, J.-C., Ryden, L., (Eds.): Protein Purification New York: VCH Publishers, 1989; p. 116. [32] janson, J.-C., Ryden, L., (Eds.): Protein Purification New York: VCH Publishers, 1989; D. 207. [33] hilsson, K., Norrlow, O., Mosbach, K., Acta Chem Scand 1981, 35, 19-27. [34] Wilchek, M., Miron, T., Appl Biochem Biotechnol 1985, 11, 191-193. [35] Hearn, M. T., W., Bethel, G. S . , Ayers, J. S . , Hancock, W. S . , J Chromatogr 1979, 185, 463470. [36] Ngo, T.T., BioA’echology 1986, 4 , 134-137. [37] Hermanson, G. T., Malia, A. K., Smith, P. K., Immobilized Affinity Ligand Techniques. Academic Press, San Diego 1992. [38] Miiller, E., Miiller, A , , Badel, K., Seiler, A., European Patent 0565978, 1994.
326
12 Membranes Modified for Biochromatography
[39] Lowe, C.R., Harvey, M. J., Craven, D.B., Dean, P. D. G., Biochem J 1973, 133, 499. [40] Langone, J. J., Adv Immunol 1982, 32, 157-252. [41] Charcosset, C., Su, Z., Karoor, S . , Daun, G., Colton, C. K., Biotech Bioeng 1995, 48, 415427. [42] Bridonneau, P., Lederer, F., J Chromatogr 1993, 616, 197. [43] Wilchek, M., Abstracts, ESBC - Meeting, Nancy, 1995. [44] Finger, U. B., Thommes, J., Kinzelt, D., Kula, M.-R., J Chromatogr 1994, 664, 69-78. [45] Small, D. A. P., Atkinson, T., J Chromatogr 1981, 216, 175-190. [46] Champluvier, B., Kula, M.-R., J Chromatogr 1991, 539 (Z), 315-325. [47] Porath, J., Carlsson, J., Olsson, I., Belfrage, G., Nature 1975, 258, 598-599. [48] Smith, M. C., Furman, T. C., Pidgeon, C., Inorg Chem 1987, 26, 1965-1969. [49] Danckwerts, P. V., Chem Eng Sci 1953, 2, 1-13. [50] Miiller, E., Baurmeister, U., Abstracts, 11th International Symposium on AfJinity Chromatography and Biological Recognition, Kalmar, 1997. [Sl] Belfort, G., Davis, R.H., and Zydney, A.L., Membr Sci 1994, 96, 1-58.
Part Three Modeling
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
13 Computer Modeling of Chromatographic Bioseparation Andreas Spieker, Ernst Kloppenburg and Ernst-Dieter Gilles
13.1 Introduction Large-scale chromatography is increasingly finding new applications in biotechnology due to its high separating power, selectivity, versatility, relatively low costs, and ‘mild’ operating conditions. Products separated by process-scale chromatography are for example amino acids, proteins, hormones, monoclonal antibodies, vaccines, and antibiotics [ 11. In contrast to analytical chromatography, in preparative chromatography high feed concentrations are used. This leads to solute competition for sorbent sites resulting in non-linear effects. For this reason, chromatography of biological molecules is often quite complex, and many of the dynamic aspects are not well understood. In the development of chromatographic systems for large-scale biochemical separations it is important not only to get the most efficient separation system, but also to design the most efficient system at minimum capital and operating costs and maximum throughput. For the purpose of a better understanding of the process dynamics, for optimization and for scaling up, the process must be accurately modeled. The range of validity of the model must be large enough for the specific application. Experimental studies using biological solutes are expensive and complex. Simulation of these systems using computer programs (simulators) can be an efficient and economical method for optimization and scale-up. Although some experiments will be required, feedback from computer modeling and numerical simulation can greatly reduce the number of experiments. Here, the modeling of chromatographic systems based on balance equations and phenomenological equations describing the individual effects in the column is presented. These models are represented by systems of partial differential and algebraic equations, or systems of differential algebraic equations. Finally, the numeric solution of these models is discussed.
330
13 Computer Modeling of Chromatographic Bioseparation
13.2 Modeling 13.2.1 Modeling Assumptions In order to model the different processes occurring in a chromatographic column, first an abstract view of the internal relationships has to be developed and second some simplifying assumptions about the physico-chemical effects in the column have to be made. 13.2.1.1 Abstract View of the Column In the upper part of Fig. 13-1, a partially opened column filled with a packing of randomly distributed particles is outlined. The lower part of Fig. 13-1 shows a small magnified section of the column. In this abstract view of the column internals, distinction will be made between a bulk liquid phase containing the interstitial liquid, and a pore liquid phase inside the particles. Each particle is surrounded by a small liquid film, which represents the mass transfer resistance between the bulk liquid phase and the pore liquid phase. In the bulk liquid phase, mass is transported by convective flow in the void volume between the particles and by dispersive effects caused by the packed bed. In the particle pores, mass transport takes place by liquid diffusion. It can be very slow and rate-limiting for the separation of large molecules. The complex internal structure of the particle matrix is simplified by assuming the existence of a network of interconnected pores inside the solid phase. The adsorbed phase is viewed as a thin layer on the solid phase. It is assumed to be in local thermodynamical equilibrium with the pore liquid at each point of the particle. But, since it is difficult to assign a volume to a quantity of molecules or ions adsorbed on a surface, the adsorbent concentration C A , ~is related to the solid volume. The volume of the adsorbed phase is rather small and is therefore neglected. The interparticle void fraction, also called external porosity, represents the amount of space left for the bulk liquid phase between the particles and is denoted by EB. For the intra-particle void fraction or internal porosity, EP is used. The column void fraction or total column porosity, Etot, denotes the volume fraction of the entire liquid in a column section. It can be expressed as Etot = EB + (1 - E B ) EP. One class of models is characterized by the fact that fast mass transfer between bulk and pore is assumed. The film is then neglected and the bulk concentration cB,i and the pore liquid concentration cp,i are assumed to be equal. The volume of the combined pore and bulk liquid phase is represented by the column void fraction, Etot.
13.2 Modeling
331
view into the column
inlet
outlet
1
. - . . adsorbed phase
pore liquid phase
I
Fig. 13-1. Geometry.
Additional Modeling Assumptions The following modeling assumptions are made to further reduce the complexity of the model and to focus on the important physical and phenomenological effects. Naturally, these assumptions are not justified in all possible cases and must therefore always be treated with caution. One important assumption is that the activity coefficients of all components are assumed unity and non-idealities are neglected. 1. The packed bed. The column is modeled as a one-dimensional, spatially distributed system consisting of a homogenous bed of spherical porous particles with the bed porosity being constant throughout the column. All model parameters and boundary conditions are assumed to be constant in an intersection of the column. The bulk concentrations are only functions of time t and axial coordinate z. Axial dispersion [2] represents deviations from purely convective flow. It has much more influence than diffusive effects in the bulk, which are therefore neglected.
332
13 Computer Modeling of Chromatographic Bioseparation
Thefluidphase. The compressibility of the fluid is neglected. Since the solvent is in excess, the molar volume and the density of the mixture can be taken to be constant throughout the column. Isothermal operation is assumed. The chromatographic particles. The particles are assumed to be spherical, of equal size and small compared with the column diameter and length. Then spherical symmetry of the concentrations in the particles can be assumed. The porous particles may be suitably represented by a constant porosity. When the particles are modeled as distributed systems, the assumption is made that mass transport inside the particles occurs only through liquid diffusion and not through convection or surface diffusion. The adsorption process. The solvent is not adsorbed. Therefore, the mass balance of the solvent is not considered. The adsorption kinetics are very fast and can be represented by an equilibrium isotherm.
13.2.2 Model Structures The most complex model commonly used considers a bulk liquid phase, a pore liquid phase, and an adsorbed phase. Film mass transfer between the bulk and the particle surface, as well as diffusion inside the particles is also part of the model (Table 13-1). This model will be called the l + l d mass transfer model. Here, l + l d means one-dimensional axial distribution in the column, and one-dimensional radial distribution in each particle. If the particles or the molecules are small, diffusion inside the particle pores is fast and each single particle can be treated as a spatially concentrated system. This assumption leads to the Id mass transfer model. If, in addition, the mass transfer between the bulk fluid and the pore fluid is fast, the film can be eliminated and the two corresponding concentrations become equal. In the following, first the most complex model will be derived. Then the simpler models are derived and the relations between the models are shown. Table 13-1. Model overview. axial dispersion
film mass transfer
diffusion in particle
1+ 1d mass transfer model
0
0
0
1 d mass transfer model
0
0
1 d equilibrium model
0
13.2 Modeling
333
13.2.3 The l + l d Mass Transfer Model In the l + l d mass transfer model, the mass transport by diffusion inside the particle is considered as a possible rate-limiting factor and is therefore modeled in detail. This is especially true for large molecules like proteins and for large adsorbent particles. In the following, first the bulk mass balance equations and then the equations for the particles are derived. Mass Balance for the Bulk Phase In Fig. 13-2, a differential volume element of the bulk liquid phase with axial length dz is shown. Convective and dispersive mass fluxes Cc,i and j,,i) enter the element at location z and leave it at location z + dz. Additionally, film mass transfer between the bulk liquid phase and the pore liquid phase at the particle surface occurs. This results in an exchange flux j , i . A mass balance for component i with concentration C B , ~in the differential volume element yields
Y
mass accumulation
convection
In Eqn. (l), the cross-sectional area A of the column and the particle surface per unit volume w of the column is introduced. As all particles are assumed to be equally sized spheres of radius R, w is given by
pore liquid phase a t particle surface
h
liauid film
bulk
z
z+dz
Fig. 13-2.Differential volume element of the bulk.
334
13 Computer Modeling of Chromatographic Bioseparation
After division by A dz and expanding j,,i the limit dz + 0:
I z +dz
and j , i
I z +dz
by Taylor series, we get in
The exact meaning of the different flux terms is given in the following.
Convection The liquid phase enters the column with a volumetric flow the effective interparticle fluid velocity v can be calculated as t~ = V / ( AQ). Then the convective flux of the component i is described by
Dispersion The various complex effects arising from fluid flow through a packed bed such as back-mixing and dead volumes are summarized as dispersion [3,4]. Dispersion mainly depends on the velocity of the bulk phase and the Reynolds number (see section 13.3). In contrast to diffusive effects, dispersion acts equally on all species. The dispersion coefficient E is therefore the same for all components. The structure of the dispersion term is analogous to Fick’s law for diffusion
Empirical correlations for E are given in section 13.3.1.
Film Mass Transfer The liquid film linear driving force model, which is most often used, is based on the assumption that the entire resistance to mass transfer resides in a hypothetical stagnant film next to the surface. The diffusion flux in this stagnant film is assumed to be quasi-stationary. This means that the mass storage of the film is neglected. The mass flux in the film for each component is determined by the transfer resistance coefficient kf;iand the concentration difference between the bulk liquid phase and the pore liquid phase at the particle surface:
The film mass transfer resistance coefficient for the component i is equal to the diffusivity Di divided by the hypothetical film thickness 6i. Therefore, this coefficient or
13.2 Modelinrr
335
the film thickness can be used to describe the mass transfer resistance in the film. The film mass transfer resistance coefficient is commonly calculated using empirical correlations, see Section 8.3.2.
Boundary Conditions and Initial Conditions of the Bulk Phase The boundary conditions are derived from the assumption that mass transport outside the column is caused by convection only. This leads to the Danckwerts boundary conditions [ 5 ] . Mass balances for the two differential volume elements shown in Fig. 13-3 yield for dz + 0:
pore liquid phase a t particle surface
m
pore liquid phase at particle surface
A
liquid film
liquid film
phase
I 0
/ Otdt
phase
I L-dt
1
L
Fig. 13-3. Differential volume elements at the column boundaries.
If dispersion is negligible and only plug-flow occurs in the column, the dispersive term in Eqn. ( 3 ) and the outlet boundary condition are canceled. The inlet boundary condition then becomes
Mass Balance for the Particle The model of the particle takes into account diffusion that leads to radial distributions within each particle. A model for one representative particle at axial position
336
13 Computer Modeling of Chromatographic Bioseparation particle
solid phase with adsorbed phase /
1
r
1
r+dr
Fig. 13-4. Differential volume element of the particle.
z is derived. In Fig.13-4, a differential volume element in form of a spherical shell is outlined. Its volume dV is given by 4 dV = - 9z ((r+dr)3-r3) 3
and its surfaces Air and
are given by
In step a,) of Fig. 13-4, a part of this volume element is magnified. Then, in step b.), the solid phase together with the adsorbed phase and the pore liquid phase are viewed as distinct homogenous phases within the element. The diffusive mass flux is denoted by j d , i . A mass balance for component i with concentrations cp,i and C A , ~in the differential volume element yields dV
(
&p-+(l-&p) i:a
")
at
=
Since the pore fluid and the adsorbed phase are coupled by the equilibrium isotherm they can be viewed as one quasihomogeneous phase. This means that there is actually only one storage accumulating mass. This can cause problems in the computa-
13.2 Modeling
337
tion of consistent initial conditions for the resulting system of differential-algebraic equations (DAE) (see section 13.4.2). One possibility to avoid this problem is to introduce a fictitious concentration, C K , ~ ,which combines the two variables according to their volume fraction and represents the total concentration in the particle: cK,i = &P cP,i f (1 -&PI cA,i
-
(13)
Inserting Eqn. (13) into Eqn. (12) gives
After a Taylor expansion of j d , i l r + d r and dr
-+
0, the result is
The concentrations C A , ~and cpj are related by the equilibrium isotherm (equilibrium isotherms are treated in section 13.2.8).
Diffusion in the Pore Liquid Diffusion in liquids is in general more complicated than diffusion in gases, due to the strong intermolecular forces between the atoms and molecules [6]. In electrolyte solutions, the diffusion process of the ions is influenced by the electrostatic potential in the solution. These properties make it difficult to predict the transport in these systems. The complex internal structure of the chromatographic matrix has great influence on diffusion [7]. The increase of the effective path length can be accounted for by using a tortuosity factor [8]. If the pore radius of the particle is large compared with the mean free diffusion path of the molecules, the collisions between the molecules and the pore wall have little influence on the diffusion process in comparison with collisions between molecules. This mechanism is called ‘molecular diffusion’ In the other case, the diffusion rate is determined by the collisions with the pore walls. Diffusion in multicomponent systems can be adequately represented by the generalized Maxwell-Stefan equations, and by the thermodynamics of irreversible processes [9,10]. Usually a detailed modeling of the multicomponent-diffusion in liquids is too complex and time-consuming, so the usual approach is to neglect the multicomponent effects and to work only with binary diffusion coefficients so that the flux expression for binary diffusion is given by Fick’s law:
338
13 Computer Modeling of Chromatographic Bioseparation
For electrolyte solutions, the Nernst-Planck equation [lo] can be used. Since it is difficult to obtain diffusion coefficients for an arbitrary substance from literature, one has to use empirical correlations like those given in 13.3.3. Boundary Conditions and Initial Conditions The boundary conditions for the particle mass balance are derived using differential volume elements at the particle center and the particle surface, developed from Fig. 13-4. These volume elements are shown in Fig. 13-5. Because of the radial symmetry of the particle, no diffusion flux occurs at the particle center, the concentration gradient at r = 0 must be equal zero. At the particle surface, the internal and the external flux must be equal. The boundary conditions for the particle are (for t 2 0):
The initial conditions are (for 0 5 r 5 R and 0 5 z 5 L):
and
element at particle surface
element at particle center
solid phase with adsorbed phase
R-dr
I
I
0
O+dr
I R-dr
liquid phase
/ R
Fig. 13-5.Differential volume elements at the particle center and at the particle surface.
13.2 Modeling
339
The resulting l + l d Mass Transfer Model Equations Bulk mass balance Eqn. (3) with insertion of Eqns. (4), ( 5 ) , and (6):
Boundary conditions bulk (for t 2 0):
The particle model is valid at each axial position z. Therefore the concentrations C K , ~ , c p , i , and C A , ~are functions of the axial coordinate z, radial coordinate r and time t.
Boundary conditions particle (for t 2 0):
In order to reduce the number of equations and variables it is possible to eliminate C A , ~or cp,i from this equation system using Eqn. (24). Alternatively, if the equilibrium isotherm Eqn. (25) is explicit in C A , ~or C P , ~ ,it can be used for elimination.
340
13 Computer Modeling of Chromatographic Bioseparation
13.2.4 The Id Mass Transfer Model The main difference between the Id mass transfer model and the l + l d mass transfer model is the simplified modeling of the particle. In the Id mass transfer model, it is assumed that the diffusion is fast enough so that there are practically no concentration gradients within the particle. Therefore, the particle concentrations are no longer dependent on the radial coordinate r .
Mass Balance for the Bulk Phase The mass balance equation and the boundary conditions are the same as in the l + l d mass transfer model, with the exception of the film mass transfer. It is changed since cp,iIR is identical to c p l : jf,i(zjt)
= kf,i (cB,i - cP,i).
(29)
Mass Balance for the Particle The mass balance for a single particle at axial position z is:
4nR3 3
(Q?%+ at
(1 - cp) at
With o defined in Eqn. (2), this can be rewritten as:
Because the radial distribution in the particle has been neglected, the entire pore liquid and the entire solid can now be interpreted as two homogeneous axially distributed phases with no axial transport. Then Eqn. (31) can be directly derived from a mass balance for all particles in a differential volume element of the column (Fig. 13-6). With c K , ~defined in Eqn. (13) and insertion of Eqn. (29) the particle mass balance becomes:
13.2 Modeling solid phase with pore adsorbed liquq-,,,,,, phase
z
341
z+dz
liquid film
3s,i ... ..
... ...
Fig. 13-6.Differential volume element of the particle phase.
bulk liquid phase
Derivation of the Id Mass Transfer Model from the l + l d Mass Transfer Model Instead of deriving the particle mass balance Eqn. (32) as shown above, it also can be obtained from the mathematical reduction of the l + l d mass transfer model starting from the detailed particle mass balance in section 13.2.3. In the first step, the average value of the variable C K , ~over the particle radius is taken and named i?~,i: R
R
Taking the average value of Eqn. (23) in section 13.2.3 and inserting the boundary conditions Eqn. (26) yields
The assumption to be made now is that cp,i is approximately constant over the particle radius. Therefore, cp,iIR is approximately equal to Cp,i. Then the particle mass balance of the Id mass transfer model is obtained:
342
13 Computer Modeling of Chromatographic Bioseparation
This is equal to Eqn. (32) when
Cp,i
and ?K,i are substituted by
Cp,i
and
CK,~.
The Resulting Id Mass Transfer Model Equations The remarks about negligible dispersion (section 13.2.3) and the reduction of the number of equations (section 13.2.3) also apply here.
Boundary conditions (for t 2 0):
13.2.5 The Id Equilibrium Model The Id equilibrium model, visualized by its differential volume element in Fig. 13-7, can be derived from the Id mass transfer model using the assumption of equal concentrations in bulk and pore fluid. The summation of the bulk mass balance Eqn. (3) and the mass balance for the particle Eqn. (31) and the insertion of Eqns. (4) and ( 5 ) leads to:
13.2 Modeling
343
adsorbed phase
/
liquid phase
r
\L--4
i
1- €tot
&tot
Fig. 13-7.Differential volume element of the column.
Now, the assumption is made that the mass transfer between bulk and particle is very fast and the concentrations C B , ~and cp,i are therefore approximately equal and now are denoted by c&. Then, due to this quasi-stationarity, only one mass balance for each component is left. With the abbreviation
it follows that
In this partial differential equation, the fluid phase takes the volume Etot A dz of a differential volume element, but only the part EB A d z is moving by convective transport. The isotherm must be written as:
The boundary conditions and initial conditions are identical to the Id mass transfer model.
13.2.6 Simulation Results In this section, simulation results of the l + l d mass transfer model and the Id mass transfer model are compared using experimental data from a separation of two pro-
344
13 Computer Modeling of Chromatographic Bioseparation
Id Mass Transfer Model
1+1d Mass Transfer Model
Protein 1
Protein 1
Time Protein 2
Time Protein 2
Time
Time
C
Be
c
!
0
Fig. 13-8. Comparison of simulation results.
teins. In Fig. 13-8, the concentrations at the column outlet are shown; measured data are represented by circles. The model parameters were chosen in a way to get the best fit of the data. On the left side of Fig. 13-8, it can be seen that the Id mass transfer model is able to represent the data of the first protein, but fails to represent the data of the second protein. In contrast to that, the l + l d mass transfer model (results on the right side of Fig. 8-8) easily fits the data. This means that the diffusion process in the particles had to be considered in order to get an appropriate model for this process. The movement of a concentration profile traveling in the column is shown in Fig. 13-9. In this simulation, dispersive effects lead to zone broadening.
C
0 .-c
e C 8 C 3
-Id
axial coordinate Fig. 13-9. Moving concentration profiles in a column.
13.2 Modeling
345
13.2.7 Simulated Moving Bed Chromatography In the simulated moving bed chromatographic process (SMB) [ l l ] , a number of columns is cyclically connected (Fig. 3-10). A liquid flow is circulating through the columns. The process is divided into four zones by the input and withdrawal ports for extract, feed, raffinate, and solvent. The ports are moved one column ahead after each switching interval A t s as indicated by the dotted arrows in Fig. 13-10. This leads to an apparent counter-current between bulk liquid and adsorbent. When the process parameters are chosen correctly, moving concentration profiles with approximately constant shape develop. One or more of the solutes binding to the adsorbate more strongly can be withdrawn in the extract, the remaining weakly binding solutes appear in the raffinate. The SMB process can be modeled by connecting models for the single columns using mixer and split units. Additionally, a facility for modeling the periodical switching of the ports is needed. The resulting model consists of both continuous and discrete parts. It can be used to study the dynamical behavior of the plant in detail. Alternatively, as a simplified model, moving bed chromatography (MB) with a real counter-current of bulk liquid and adsorbent may be used. This leads to a continuous model. It can be used, e.g. for parameter identification, stationary optimization, and control design. In the moving bed model, the four SMB zones are represented by columns with the same length as the respective SMB zone (Fig. 13-11). No port switching occurs. Instead, the adsorbent is assumed to move opposite to the liquid (dashed arrows). V E ( t + At,)
M-Hl
extract V E ( t )
A
feed V F ( t )
...
zone I I
direction of port switching and of fluid flow
Vs(t
+ At,)
n
I
,-J+ .. . zorH-x Vrec
solvent Vqft)
Fig. 13-10. Simulated moving bed process.
VVRft
+ At,',
raffinate V~int)
346
13 Computer Modeling of Chromatographic Bioseparation liquid recycle flow
I
I
-I zone I I
zone I
zone Ill
zone IV I
I
1 feed
ji extract
solvent
ji raffinate
Fig. 13-11. Moving bed model. Dashed arrows indicate hypothetical adsorbent flow.
These zones can be modeled similar to usual chromatographic columns. The only difference is, that now the hypothetical movement of the adsorbent including pore liquid opposite to the bulk liquid phase is included in the mass balance. The MB model can be derived with or without including the effect of radial diffusion in the adsorbent particles. Here, only the first case is treated. The movement of the adsorbent with velocity u causes dispersion-less axial transport of pore fluid and adsorbate defined as
Considering a differential volume element (Fig. 13-12) the mass balance for adsorbate and pore fluid can be derived:
This equation replaces Eqn. (37) in the Id mass transfer model for batch chromatography. In contrast to the batch chromatography model, a boundary condition for the particle phase mass balance is now required. Since adsorbent flow is assumed to be purely convective, only an inlet boundary condition exists. A mass balance for a differential element at the right boundary z = L leads to solid phase with
f--
liquid film /
bulk liquid phase
Fig. 13-12. Differential volume element for moving bed model.
341
13.2 M o d e l i n ~ cK,i
IL
= CK,i,in.
(49)
For equivalence of the MB model to an SMB process with columns of length L ~ M B and switching interval Ats, the adsorbent velocity u has to be chosen as
The liquid velocity u in each zone must be set to a value resulting in the same relative velocity u u in the MB and the SMB process. Fig. 13-13 shows typical stationary concentrations profiles obtained with an MB model for the separation of three components. In the SMB process the concentration profiles move over the distance of one column length during a switching interval Ats (Fig. 13-14).
+
I
k--zone
+
zone II -+zone
1
IV
111 +-zone
4
I
0.4
/
\ \
I
axial coordinate
Fig. 13-13.Stationary concentration profiles of the moving bed model.
5- 0.5 0.4 -
E
0
1
2
3
4
5
6
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 axial coordinate (number of SMB column)
Fig. 13-14.Moving concentration profiles obtained with the simulated moving bed model.
348
13 Computer Modeling of Chromatographic Bioseparation
13.2.8 Equilibrium Isotherms Equilibrium isotherms describe the binding of liquid phase components to the solid phase. They couple the mass balance equations of all competing solutes in the particles and determine the retention behavior of the solutes. Only multi-component (competitive) isotherms will be discussed. It will be distinguished between adsorption isotherms and ion-exchange isotherms.
Adsorption Isotherms The adsorption isotherms shown here can be used to describe the equilibrium in normal and reversed-phase adsorption chromatography, affinity chromatography, and hydrophobic interaction chromatography. The competitive Langmuir isotherm is the most often used isotherm. It was developed to describe the adsorption from the gas phase on homogenous surfaces without interaction between the adsorbed molecules. The coverage of the surface is at maximum monomolecular. The transition from gas-solid to liquid-solid equilibrium is done by assuming ideal phases and replacing the partial pressures by concentrations [12]. If the saturation capacity Q is equal for all components, the isotherm is thermodynamically consistent [ 131. The fraction of an adsorbed component is proportional to the surface covered by its molecules. The solute affinity for the adsorbent is described by the parameter bi. The competition between the different components for interaction with the stationary phase depends on the affinity b i and on the actual concentration of the solute in the liquid phase C P , ~ :
1
+ C bj cpj j= 1
The resulting separation factors a i j are independent of the concentrations. The Langmuir isotherm is therefore not able to represent systems with variable selectivities. ai,j
CPi CAj := cA,i @,j
. .. =
9= const.
(52)
bi
If unequal saturation capacities Qi are used, the isotherm is no longer thermodynamically consistent, but in some cases gives a better representation of experimental data:
23.2 Modeling
349
j= 1
If only small amounts are adsorbed, e.g. in analytical range, the second term of the denominator in Eqn. (51) becomes negligible compared with the first and the result is the linear isotherm. This simplest isotherm can be seen as a linear approximation of the non-linear Langmuir isotherm:
If a surface is covered by two different types of binding sites, the behavior can be described using a bi-Langmuir isotherm [ 141:
j= 1
j= 1
The advantage is that the resulting separation factors ai,are no longer independent of the concentrations. The next isotherm to be discussed is the constant separation factor (CSF) isotherm [15]. The assumption of constant separation factors is maintained but, in contrast to the Langmuir isotherm, all binding sites are permanently occupied. The total concentration on the adsorbent is fixed and equal to Q. This means that the maximum capacity is equal for all components and that stoichiometric exchange occurs.
With the separation factor mi,, defined in Eqn. (52) the CSF isotherm can be derived after some calculation:
Effectively, only the capacity Q and the separation factors ai,l with respect to component 1 are needed to describe this isotherm. This isotherm is equal to the stoichiometric displacement model in the case of equal valencies (see section 13.2.8). Other isotherm models are for example the Langmuir-Freundlich isotherm (used by [16] for adsorption of proteins on an basic anion exchanger), the ideal adsorbed solution isotherm [17], and the real adsorbed solution isotherm [18,19]. Some other isotherm models are given in [ 151.
350
13 Computer Modeling of Chromatographic Bioseparation
Ion-Exchange Isotherms In ion exchange, isotherms must maintain the condition of electroneutrality. Therefore, the Langmuir isotherm is not appropriate here. In the stoichiometric displacement model (SDM), the equilibrium constant is derived using the mass action law formalism. The binary ion exchange of counter-ions I1 and I2 on an ion-exchange resin is modeled as a reaction [20,21]. In the following equation, ml and m2 denote the valencies of I1 and 12. & denotes the components bound to the resin:
Electroneutrality requires Q = ml cA,1 the equilibrium constant K is
+ m z c ~ , 2Neglecting .
the activity coefficients,
This isotherm can be easily extended to multi-component systems. Shallcross and Mehablia [22] use a mass-action law with activity models for the solution (Debye-Hiickel or Pitzer equation) and for the resin phase (Wilson equation). They calculate multi-component equilibria from binary equilibrium data. The steric mass action ion-exchange formalism (SMA) [23] explicitly accounts for the steric hindrance of salt counter-ions upon protein binding in multicomponent equilibria.
Other Isotherm Models Myers and B yington [24] found a thermodynamically consistent ion-exchange isotherm for multi-component systems with variable selectivities using binary data. Saunders [25] uses a modified Myers and Byington model for calculating equilibria for the exchange of amino acids. Melis [26] calculates thermodynamically consistent equilibria for mono- and divalent amino acids with variable selectivities, and accounts also for energetic heterogeneity of the exchanger.
13.3 Empiric Correlations For practical calculations respectively simulations, empiric correlations are needed to estimate unknown parameters describing diffusion, dispersion, and mass transfer resistances. These correlations are usually expressed using dimensionless numbers.
13.3 Empiric Correlations
35 1
13.3.1 Correlations for the Dispersion Coefficient As already mentioned in section 13.2.3, the various complex effects leading to axial mixing are combined into a single axial dispersion coefficient. This axial dispersion coefficient has to be determined from experiments or can approximately be calculated using empirical correlations like those presented by Chung and Wen [27]. This widely used equation is applicable for both fixed bed and fluidized bed. The Peclet number, Pe is calculated to give VL
P e = - = -(0.2 E ~ R E B
+ 0.01 1 Re0.48)
with the Reynolds number (Re)
Other correlations are given by Hejtmhek [2], Koch and Brady [28], and Gunn [29].
13.3.2 Correlations for the Mass Transfer CoeMicient The mass transfer resistance represented by a hypothetical film around the particle (see section 13.2.3) needs a film mass transfer resistance coefficient kf,i for each component. Usually, the correlations give an expression for the coefficient kf,i or the Sherwood number, Sh, as in the equation used.by Truei et al. [30]:
where the Schmidt number, Sc, is given by sc=
-.r
P Di
Other correlations are given by Foo and Rice [31] and Ohashi et al. [32]. Kataoka et al. [33] presented a correlation for the mass transfer to ion-exchange resins.
13.3.3 Correlations for Diffusivities For models taking into account diffusion in the chromatographic matrix, like the l + l d mass transfer model, for each component an individual diffusion coefficient
352
13 Computer Modeling of Chromatographic Bioseparation
is needed. Diffusion coefficients for free diffusion in liquids are in the order of cm2s-l, diffusion coefficients of proteins in chromatographic matrices are in the order of cm2 s-l. If neither literature data nor own experimental data are available, correlations are needed. The empirical correlation of Wilke and Chang [34] is satisfactory for the estimation of diffusion coefficients in dilute solutions of binary non-electrolytes with sufficient precision for most engineering purposes, i.e., about 10 % average error. The association parameter X is 2.6 for H20.
Other correlations are given by Mackie and Meares [8], Reid and Prausnitz [35], and Tyn and Calus [36]. The prediction of diffusivities for multi-component systems is difficult, especially for non-ideal concentrated solutions. First attempts of calculating coefficients for multi-component approaches have been made by Wesselingh [37] and others [10,38]. Q n and Gusek [39] propose a correlation for predicting the diffusion coefficients of macromolecules and proteins by an adaptation of the Stokes-Einstein equation to a model for the equivalent hydrodynamic sphere. The size and shape of protein molecules in solutionis accounted for by a so-called radius of gyration R,. For 86 different proteins and macromolecules, 87.4 % of the diffusivity predictions lie in a range of max. 20% error:
Di = 5.78.10-15[cm3PaK-1]T/(R,,i 7).
(65)
Other correlations are given by Phillies [40], Boyer 411, and Skidmore [42]. Gaigalas [43] presents measurements of BSA diffusivity in aqueous solutions in dependence of temperature, pH, BSA concentration, and ionic strength and provides some data for other globular proteins.
13.4 From Modeling to Numeric Simulation The main reason to create models for chromatographic processes is to be able to perform simulation experiments instead of real experiments or to use the models in an optimization procedure. These applications require a solution of the models. The models derived in the previous sections are always systems of coupled partial differential and algebraic equations (PDAE). For this type of non-linear equation system generally an analytical solution is not available. In order to get a numeric solution the method of lines (MOL) can be used. It is applicable to a wide range of problems and leads to reliable results with reasonable effort. With the MOL, the spatial derivatives appearing in the PDEs are discretized and the solution of the PDAE system is computed in terms of the values on a spatial grid. The spatial discretization can either be performed explicitly prior to any numeric computation (semi-discretization)
13.4 From Modeling to Numeric Simulation
353
or implicitly by an algorithm for the solution of PDAE systems. Both of these methods are discussed in the following.
13.4.1 Solution via Semi-discretization When the spatial derivatives are explicitly approximated using a spatial discretization, the PDAE system is transformed to a DAE system (differential and algebraic equations). The resulting initial value problem can be solved using reliable algorithms for integration of DAEs, e.g. DASSL [44], LIMEX [45]. This may be done within the framework of a flow-sheet-type dynamical simulator (section 13.4.2). For large systems, sparse matrix techniques should be used. The integration algorithms supply an error control for the time integration. Solution error due to the spatial discretization has to be controlled by the user. This means, that the discretization method and the number and position of the grid-points have to be chosen with care. It is possible to simultaneously solve additional differential and algebraic equations; they are just appended to the original DAE system. In this way, an arbitrary coupling of different columns and subsystems such as controllers, valves, and tanks can be simulated.
Spatial Discretization using the Finite Volume Method Currently, the most popular and widely used method for spatial discretization of onedimensional PDEs arising in chemical engineering and biotechnology is the finite difference method (FDM). However, this method has a main disadvantage that makes its application difficult when steep fronts at the boundaries are encountered. In this case, the approximation error in the boundary conditions can lead to a significant violation of the overall mass balance. To a certain extent this problem can be met by refining the grid near the boundaries, but this leads to a significant increase in the number of grid-points required. This difficulty can be overcome by using the finite volume method (FVM) [46]. The FVM is a more physically motivated formulation than the FDM - all material balances are fulfilled inherently. So even in the discretized system the basic physical principle of mass conservation is never violated. This desirable property is also shared by stage models, but these do not have the same versatility and strong mathematical background as the FDM or FVM. The finite volume method is now explained in detail using the PDE Eqn. (3) as an example. The component index is dropped for simplicity. Figure 13-15 shows the subdivision of the spatial coordinate into N sections or finite volumes with boundaries l k , l k + l and center points zk. First, the PDE is integrated over a section k:
354
13 Computer Modeling of Chromatographic Bioseparution
Fig. 13-15.FVM, subdivision of spatial coordinate.
This equation represents a mass balance for the volume element corresponding to section k. The remaining integrals in Eqn. (67) are approximated by the values in the center of the section:
Convective and diffusive flow over the section boundaries l k are approximated by interpolation using CB at neighboring section centers. In the case of constant coefficients in j , and j e as in Eqns. (4) and ( 5 ) , this is achieved by an approximation of C B ( Z , t ) and a c B ( z t ) l a z on the section boundaries. As an example, the results using a first-order approximation are shown. In [47], a second-order parabola through Cg,k-2, C g , k - l , C B , ~is used. However, this is only recommended with sections of equal length Az. From Fig. 13-16 the following expressions are easily derived:
13.4 From Modeling to Numeric Simulation
355
lk-1
‘k-1
Azk-1
li,
‘k
/k+1
Fig. 13-16. FVM, approximation of C B ( Z , t) and &(z, t ) I az on section boundary lk.
Azk
Inserting Eqns. (4) and (5) into Eqn. (67) with the approximations (68), (69), (70), and (71) for the sections k = 2 . . . N - 1 leads to differential equations for the values of CB at the section centers. In the case of sections of equal length Az this is:
-
W
-&jjf(Zk,t),
k = 2 . . .N - 1 .
(72)
EB
The boundary sections ( k = 1 and k = N ) need to be treated specially in order to include the boundary conditions Eqn. (7). These are formulated differently, again with the component index dropped:
In Eqn. (74), c& can be approximated by a straight line through c&-,
and cgl,:
With this approximation, the boundary conditions in the form of Eqns. (73) and (74) can be directly inserted into Eqn. (67) for k = 1 and k = N . The differential equations for the boundary sections become:
356
13 Computer Modeling of Chromatographic Bioseparation
Az
The mass balances for the particle are discretized analogously. For the one-dimensional model this is quite simple because there are no terms for axial transport and no boundary conditions. The application of the FVM to the l + l d mass transfer model yields a more complex equation system because there are two space dimensions. For the discretization of the radial particle coordinate, a non-equidistant grid with shorter elements near the particle surface should be used.
Other Discretization Methods For systems with very steep gradients, the quite recently developed ENO-ROE method [48,49] is a good alternative to the moving grid methods mentioned below. The approximation of the derivatives is continuously adapted to the solution. So, steep gradients can be handled without the need for excessively large numbers of grid points and with less overhead than moving grid methods introduce. With orthogonal collocation the solution is approximated by a set of orthogonal polynomials [50,5 11. The grid- or collocation-points are chosen as roots of an orthogonal polynomial. Orthogonal collocation for batch chromatography only makes sense in the form of orthogonal collocation on finite elements [52,53]. For global collocation, too many points are needed to be able to approximate the steep moving fronts encountered.
Moving Grid Methods In moving grid methods the locations of the grid points or the finite element boundaries in orthogonal collocation on finite elements are modified during time integration. The grid points are either calculated as part of the differential algebraic equation system or - which is the more common case - are redistributed between two adjacent
13.4 From Modelim to Numeric Simulation
357
integration steps when it seems necessary. In both cases, heuristic approaches are used to refine the grid where steep gradients are encountered. In this way the discretization is adapted to the solution. Obviously, moving grid methods have advantages when - as in chromatography steep fronts are present only on short intervals of the spatial coordinate. Only there a fine discretization is needed so that the overall number of grid-points can be much lower than with a fixed grid. This makes integration much faster. On the other hand, some time-consuming computations have to be made with each movement of the grid. The solution on the new grid has to be interpolated from the previous grid and the integration algorithm has to be restarted after each redistribution. Additionally, moving grid methods need some programming effort. For these reasons it is not generally advisable to use such methods.
13.4.2 Numeric Solution of the Discretized System Simulation Tools Computations involving the solution of DAE-systems should be performed within the framework of a suitable software tool. The tool should provide algorithms for the computation of consistent initial conditions, different integration algorithms, and an environment that allows an easy handling of the simulation task and analysis of the results. Flow-sheet-type dynamic simulators like DIVA, [54,55], SPEEDUP [56], and gPROMS [57] provide this functionality. Additionally, the interconnection of different units to a plant model is supported. Sometimes a facility to perform the semi-discretization automatically is provided. These simulators also support sparse matrix techniques which are very important for the effective solution of semi-discretized PDAE systems. All matrix operations like the computation of the Jacobian matrix are only performed with the non-zero matrix elements. This saves a lot of computation time and memory, as for a typical equation system arising from the discretization of a spatially distributed model, less then 5 % of the entries of the Jacobian are non-zeros. When tasks like numerical optimization or parameter identification are to be performed, the simulation tool should either provide this functionality or it should have an easy to use interface to external software. As an alternative to using a simulation tool, own programs may be written usingstandard non-linear solvers and integration algorithms, e.g. from the NAG and Harwell subroutine libraries.
Consistent Initial Conditions for DAE Systems For the computation of the solution of a DAE system, the initial conditions must be consistent, i.e. the algebraic equations must be fulfilled. This means that the number of degrees of freedom for choosing the initial conditions is less than the number of equations.
358
13 Computer Modeling of Chromatographic Bioseparation
In a DAE system, the variables with their time derivative appearing in the equations are called differential variables; the remaining variables are called algebraic variables. When the model contains only differential variables for distinct physical storages (as shown in section 13.2.3), then the number of degrees of freedom is equal to the number of differential variables. The initial conditions can be chosen in terms of the values of the physical storages. Then, from the algebraic equations the initial values of the algebraic variables can be calculated as the solution of a non-linear equation system [ 5 8 ] .
13.4.3 Solution using a Specialized PDAE Solver An algorithm dedicated to the solution of PDAE systems allows the user to find a solution of the equation system without an explicit discretization of the equations. The main advantage compared with semi-discretization is that an error control both with respect to spatial discretization and time discretization is provided by the algorithm. An example for the implementation of such an algorithm is PDEX [ 5 9 ] , which uses extrapolation methods both with respect to time and to the spatial coordinate. Discretization in all coordinates is done by the algorithm and is automatically adapted to the solution in order to match predefined accuracy criteria. This method is much faster in terms of computation time than methods using explicit semi-discretization, and is probably the most reliable and easiest in application. Its main disadvantage is that coupling of chromatographic columns with subsystems or other columns can not be simulated because the simultaneous solution of ordinary differential equations is not possible. Since it depends largely on the software used, this alternative can not be described in more detail here.
Abbreviations and Symbols
b21
[cm3mol-'1 [molcm-31 [molcm-31 [mol~ m - ~ ] [molcmp3] [molcm-31 [molcm-31 [molcmp3] [cm2s-l] [cm2s-'1 [molcmP2 s - ~ ] [molcm-* s-l] [molcm-* s-'1
cross-sectional area of the column coefficient in isotherm equations bulk fluid concentration fluid concentration in Id equilibrium model pore fluid concentration adsorbate concentration with respect to solid volume fictitious concentration averaged fictitious concentration in particle averaged pore fluid concentration in particle diffusivity axial dispersion coefficient mass flux due to convection of the adsorbent mass flux due to convection mass flux due to diffusion
Abbreviations and Symbols
Jf J k.,i Kl lk L LSMB
[molcm-2s-l] [molcmP2s-l] [cms-l]
PI
N mi MI
MS Pe Qi
r R R, Re sc Sh t At, T U V vm.1
v X Z
zk hzk
mass flux due to film mass transfer mass flux due to dispersion film mass transfer resistance equilibrium constant left boundary of element k in FVM column length length of single column in SMB-process number of components valency of component i molar mass of component i molar mass of the solvent Peclet number capacity of the solid phase radial coordinate in the particle particle radius radius of gyration Reynolds number Schmidt number Sherwood number time switching interval in SMB process temperature linear velocity of particles in a countercurrent system interstitial velocity of fluid molar volume of solute i at normal boiling point flow rate association parameter axial coordinate center of element k in FVM length of element k in FVM
Greek Letters separation factor of two components activity coefficient of component i film thickness intra-particle void volume (pore) interparticle void volume (bulk) total void volume axial coordinate dynamic viscosity density of the mobile phase radial coordinate external surface area of the particles per unit volume
359
360
13 Computer Modeling of Chromatographic Bioseparation
References [ 11 Jagschies, G., Process-Scale Chromatography, in: Ullmann 's Encyclopedia of Industrial
Chemistry, VCH, 1989. Volume B3, pp. 10.1-10.44. [2] Hejtminek, V., Schneider, P., Axial Dispersion under Liquid-Chromatography Conditions. Chem Eng Sci 1993, 48, 1163-1168. [3] Bird, R. B., Stewart, W. E., Lightfoot, E. N., Transport Phenomena. John Wiley & Sons, Inc., 1960. [4] Ergun, W., Fluid Flow through Packed Colums. Chem Eng. Progress 1952, 48, 89-94. [ 5 ] Danckwert, P. V., Continuous flow systems: distribution of residence times. Chem Eng Sci 1953, 2(1), 1-18. [6] Hines, A. L., Maddox, R. N., Mass Transfer - Fundamentals and Applications. Prentice Hall, 1985. [7] Deen, W. M., Hindered Transport of Large Molecules in Liquid-Filled Pores. AlChE J. 1987, 33, 1409-1425. [8] Mackie, J. S., Meares, P., The Diffusion of Electrolytes in a Cation-Exchange Resin Membrane - I. Theoretical. Proc R SOCLondon 1955, 232, 498-509. [9] Krishna, R., Problems and pitfalls in the use of the Fick formulation for intraparticle diffusion. Chem Eng. Sci 1993, 48(5), 845-861. [lo] Taylor, R., Krishna, R., Multicomponent Mass Transfer Wiley, New York, 1993. [ll] Nicoud, R. M., Simulated Moving Bed Chromatography: Principles and Application to Biological Molecules, in: Bioseparation and Bioprocessing. VCH, 1997, pp. 00-00. [I21 Guiochon, G., Golshan Shirazi S., Katti, A.M., Fundamentals of Preparative and Nonlinear Chromatography. Academic Press, 1994. [13] Kemball, C., Rideal, E. K., Guggenheim, E. A., Thermodynamics of Monolayers. trans Faraday SOC1948, 44, 948. [ 141 Jacobson, S . , Golshan-Shirazi, S . , Guiochon, G., Isotherm selection for band profile simulations in preparative chromatography. AlChE J 1991, 37, 836. [15] Berninger, J. A,, Whitley, R. D., Zhang, X., Wang, N.-H. L., A versatile model for simulation of reaction and nonequilibrium dynamics in multicomponent fixed-bed adsorption processes. Comp Chem Eng 1991, 15, 749-768. [I61 James, E. A., Do, D. D., Equilibria of biomolecules on ion-exchange adsorbents. J Chromatogr 1991, 542, 19-28. [17] Myers, A. L., Prausnitz, J. M., Thermodynamics of mixed-gas adsorption. AIChE J 1965, 11(1), 121-127. [18] Radke, C. J., Prausnitz, J. M., Thermodynamics of multi-solute adsorption from dilute liquid solutions. Chem Eng Sci 1972, 18, 761-768. [I91 Gamba, G., Rota, R., Carri, S . , Morbidelli, M., Adsorbed solution theory models for multicomponent adsorption equilibria. AIChE J 1989, 35, 959. [20] Velayudhan, A,, Horvhth, C., On the stoichiometric model of electrostatic interaction chromatography for biopolymers. J Chromatogr 1986, 367, 160-162. [21] Cysewski, P., Jaulmes, A,, Lemque, R., SCbille, B., Vidal-Madjar, C., Jilge, G., Multivalent ion-exchange model of biopolymer chromatography for mass overload conditions. J Chromatogr 1991, 548, 61-79. [22] Mehablia, M. A., Shallcross, D. C., Stevens, T. W., Prediction of multicomponent ion exchange equilibria. Chem Eng Sci 1994, 49(14), 2277-2286. [23] Brooks, C. A., Cramer, S . M., Solute affinity in ion-exchange displacement chromatography. Chem Eng Sci 1996, 51(15), 3847-3860. [24] Myers, A. L., Byington, S., Thermodynamics of ion exchange: prediction of multicomponent equilibria from binary data, in: Ion Exchange Science and Technology, Nijhoff, Dordrecht, 1986; p. 119.
References
361
[25] Saunders, M. S . , Vierow, J. B., Carta, G., Uptake of phenylalanine and tyrosine by a strongacid cation exchanger. ATChE J 1989, 35(1), 53-68. [26] Melis, S . , Markos, J., Cao, G., Morbidelli, M., Ion-exchange equilibria of amino acids on a strong acid resin. Ind Eng Chem Res 1996, 35, 1912-1920. [27] Chung, S . F., Wen, C. Y., Longitudinal dispersion of liquid flowing through fixed and fluidized beds. AIChE J 1968, 14, 857. [28] Koch, D. L., Brady, J. F., Dispersion in fixed beds. J Fluid Mech 1985, 154, 399-427. [29] Gunn, D. J., Axial and radial dispersion in fixed beds. Chem Eng Sci 1987, 42, 363. [30] Truei, Y.-H., Gu, T., Tsai, G.-J., Tsao, G. T., Large-scale gradient elution chromatography. Adv Biochem Eng/Biotechnology 1992, 47, 1-43. [31] Foo, S . C., Rice, R. G., On the prediction of ultimate separation in parametric punps. AIChE J 1975, 21, 1149-1158. [32] Ohashi, H., Sugawara, T., Kikuchi, K.-I., Honno, H., Correlation of liquid-side mass transfer coefficient for single particles and fixed beds. J Chem Eng Japan 1981, I4(6), 433. [33] Kataoka, T., Yoshida, H., Yamada, T., Liquid phase mass transfer in ion exchange based on the hydraulic radius model. J Chem Eng Japan 1973, 6, 172. [34] Wilke, C. R., Chang, P., Correlation of diffusion coefficients in dilute solutions. AIChE J 1955, 1, 264-270. [35] Reid, R.C., Prausnitz, J.M., Sherwood, T.K., The Properties of Gases and Liquids. McGraw Hill Book Company, 1988. [36] Tyn, M. T., Calus, W. F., Diffusion coefficients in dilute binary liquid mixtures. J. Chem Eng Data 1975, 20(1), 106-109. [37] Wesselingh, J. A , , Krishna, R., Mass Transfer Ellis Horwood, 1990. [38] Kooijman, H. A., Taylor, R., Estimation of diffusion coefficients in multicomponent liquid systems. Ind Eng Chem Res 1991, 30(6), 1217-1222. [39] Tyn, M. T., Gusek, T. W., Prediction of diffusion coefficients of proteins. Biotechnol Bioeng 1990, 35, 327-328. [40] Phillies, G . D. J., Benedek, G. B., Mazer, N.A., Diffusion in protein solutions at high concentrations: a study by quasielastic light scattering spectroscopy. J Chem Phys 1976, 65(5), 1883. [41] Boyer, P.M., Hsu, J.T., Experimental studies of restricted protein diffusion in an agarose matrix. AIChE J 1992, 38(2), 259. [42] Skidmore, G. L., Horstmann, B. J., Chase, H. A . , Modelling single-component protein adsorption to the cation exchanger S Sepharose FF. J Chromatogr 1990, 498, 113-128. [43] Gaigalas, A. K., Hubbard, J. B., McCurley, M., Woo, S . , Diffusion of bovine serum albumin in aqueous solutions. J Phys Chem 1992, 96(5), 2355-2359. [44] Brenan, K. E., Campbell, S . L., Petzold, L. R., Numerical Solution of Initial Value Problems in Differential-Algebraic-Equations. North Holland, 1989. [45] Deuflhard, U., Nowak, U., Wulkow, M., Recent developments in chemical computing. Comp Chem Eng 1990, I4(11), 1249-1258. [46] Patankar, S . V., Numerical Heat Transfer and Fluid Flow. Hemisphere Publ. Corp., 1980. [47] Leonhard, B. P., A stable and accurate convective modelling procedure based on quadratic upstream interpolation. Comp Meth Appl Mech Eng 1979, 19, 59-98. [48] Shu, C.-W., Osher, S . , Efficient implementation of essentially non-oscillatory shock-capturing schemes, I. J Comp Phys 1989, 83, 32-78. [49] Le Veque, R. J., Numerical Methods f o r Conservation Laws. Birkhauser Verlag, Basel-Boston-Berlin, 1992. [50] Finlayson, B. A., The Method of Weighted Residuals and Variational Principles. Academic Press, 1972. [51] Villadsen, J., Michelsen, M. L., Solution of Differential Equation Models by Polynomial Approximation. Prentice-Hall, 1978. [52] Carey, G. F., Finlayson, B. A,, Orthogonal collocation on finite elements. Chem Eng Sci 1974, 30, 587-596.
362
13 Computer Modeling of Chromatographic Bioseparation
[53] Yu,Q., Wang, N.-H. L., Computer simulations of the dynamics of Multicomponent ion exchange and adsorption in fixed beds - Gradient-Directed Moving Finite Element Method. Comp Chem Eng 1989, 13(8), 915-926. [54] Helget, A., Gilles, E. D., Dynamische ProzeD- und Anlagensimulation, in: ProzeJsimulation Schuler, H. (Ed.), VCH, 1994, pp. 109-148. [55] Kroner, A,, Holl, P., Marquardt, W., Gilles, E.D., DIVA - an open architecture for dynamic simulation. Comp Chem Eng 1990, 14, 1289-1295. [56] Aspen Technology. Speedup User Manual. Aspen Technology, Inc., Cambridge, Massachusetts, 1994. [57] Oh, M., Pantelides, C. C., A modeling and simulation language for combined lumped and distributed parameter systems, in: International Conference on Process Systems Engineering PSE’94, Kyongju, Korea, 1994; p. 37-44. [58] Kroner, A., Marquardt, W., Gilles, E. D., Computing consistent initial conditions for differential-algebraic equations. European Symposium on Computer aided Process Engineering 1, ESCAPE-1, 24-28 May 1992, Elsinore, Denmark. Comp Chem Eng (Suppl.) 1992, 16, 131-138. [59] Nowak, U., Frauhammer, J., Nieken, U., A fully adaptive algorithm for parabolic partial differential equations in one space dimension. Camp Chem Eng 1996, 20(5), 547-561.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
14 Neural Network Applications to Fermentation Processes P.R. Patnaik
14.1 Introduction Neural computing has emerged in recent years as a robust and versatile area of artificial intelligence. Its applications cover widely different disciplines such as electrocardiographs, optical character recognition, aircraft control, financial forecasting, assembly line processes, the structure of macromolecules, and chemical reactors t1,21. An artificial neural network (ANN) is an array of interconnected processing elements called neurons. The rationale behind ANN development and the term neuron stems from the original objective of devising mathematical structures that will (at least on a low level) mimic the functionality of the brain. While this objective might still be far-fetched, ANNs have grown in variety and applications; this is due both to the ability of ANNs to learn from their applications and of research workers to learn from the performances of ANNs. Before discussing the structure of neural networks, it is worth considering why they are used at all. ANNs receive information, process it, and generate output information. While the processing method is a crucial aspect of the performance of an ANN, it does not require a model of the system being studied. This is a significant advantage because good models based on fundamental principles of real processes are difficult to construct and tend to be too complex for automation [3]. ANN also have other advantages [l]: -
-
They are adaptive; they can learn from the data and infer subtle relationships which are useful but not obvious. They can generalize and function effectively even with incomplete or noisy data. This is useful because real-world data are rarely ‘clean’ and well-organized. Being arrays of non-linear neurons, they can be adapted to complex problems for which model development is prohibitively complex. Their parallel architecture enables fast computing and learning. Owing to their robustness, lack of the requirement of a process model and the ability to improve with usage, they outperform traditional techniques such as ARIMA and Kalman filters in predicting future trends from time-series data [4].
364
14 Neural Network Applications to Fermentation Processes
Despite many advantages over algorithmic and model-based techniques, it is important to be judicious in the construction, training and usage of ANNs. The structure (also called topology or architecture) of a network should have some connection with the process being studied. Unfortunately, the absence of a model, which is an advantage of neural simulation, makes it difficult to devise rules to choose a suitable topology. Nevertheless, because ANNs employ statistical methods for their data, these techniques can be used to guide the choice of a suitable network [ 5 ] . Almost every serious work with ANNs emphasizes the pivotal role of proper training. It is important to remember that developing a neural network is different from writing software according to an algorithm. How well a network performs depends on its training, which in turn depends on the choice of data and the training methodology. Data which are too few or redundant or badly scaled can impair the learning process. So can the extent of training. A network should not be overtrained or undertrained. An ANN which is trained insufficiently has obviously not learned enough about the process, and therefore it will be poor in making decisions. Even more dangerous is an overtrained network, because after having learnt the relevant features it has been forced to learn redundant features that conflict with and untermine its earlier learning. Again, there are guidelines to decide when to stop training [4-61. A properly trained neural network should be able to generalize accurately, i.e. it should be able to recognize and infer patterns or relationships from data which were not used for training. Although generalizability is what ultimately makes ANNs practically useful, they should not be applied in situations far removed from those of the training data. Consider, for example, a fermentation with recombinant bacteria containing a temperature-inducible plasmid [7]. This process functions altogether differently on either side of a critical temperature, T,. Thus, an ANN trained with data below (or above) T, may not be able to predict the performance above (or below) T,. Real applications, however, may require an ANN to function in wide ranges of parameters, either because the process itself is operated that way (as in the example just cited) or because the process has a number of steps and the parameters differ considerably among the steps. In such situations a hierarchy of networks is constructed, with each subnetwork representing one step or one aspect of the process [8,91. Compared with some other disciplines, applications of ANNs to chemical reactors are of recent origin, perhaps because their models are less complex than, for example, the flight of space craft, offshore structures, and weather forecasting. Nevertheless, dynamic optimization and on-line control of chemical reactors requires continual updating of process information and decision making on the basis of current information. This is especially true of biological processes because cellular metabolism is difficult to quantify in a time-varying environment, it is strongly influenced by disturbances and nonideal mixing conditions in a bioreactor, and many variables (such as intra-cellular concentrations) cannot be measured directly [ 10-121. Neural networks help to estimate variables on-line and in the optimization and control of bioreactors.
14.2 Structure and Functioning of ANNs
365
14.2 Structure and Functioning of ANNs 14.2.1 Basic Structure and Principles An artificial neural network (ANN) is a two-dimensional array of processing elements called neurons (or neurodes in some references). Figure 14-1 depicts a typical ANN. Notice that the neurons are arranged in layers. The first layer (the input layer) receives information and passes it on with little or no processing to the next (intermediate) layer, called the hidden layer because it does not deal directly with either the inputs or the outputs. There can be more than one hidden layer but, unless the outputs have steep ridges or discontinuities, one layer usually suffices. In fact, Cybenko [13] showed that any mapping from an n-dimensional real space W’ to 9Irncan be achieved with at most two hidden layers. Hornik et al. [14] later showed that any real continuous function can be approximated to an arbitrary degree of accuracy by one hidden layer with a sufficiently large number of neurons. The hidden layers are followed by an output layer, which does the final processing to generate the required results. Each input neuron receives a single signal or a string of signals representing the time-series data of one variable. For example, in a fed-batch fermentation with a time-dependent substrate feed rate, one input neuron may be assigned to receive sampled values of the feed rate. Likewise, each output neuron generates one output variable. This of course means there must be as many neurons in the input layer as the number of input variables, and similarly for the output layer. The real decisions thus involve the number of hidden layers (one or two), the number of neurons in each hidden layer and the flow of information (represented by the arrows in Fig. 14-1). As for the training procedure, there are no rigorous rules to enable these decisions, but there are good heuristics which will be presently shortly.
n ul
al
am .6%
> c)
3
-ca Bias
Bias= 1
Input layer
Hidden layer
Output layer
Fig. 14-1. Structure of a typical neural network.
366
14 Neural Network Applications to Fermentation Processes
I
Bias
Fig. 14-2. Schematic representation of internal processing by an artificial neuron.
For the moment let us focus on one neuron to understand how it acts on the information it receives. Figure 14-2 summarizes its functioning. The incoming signals undergo two or three stages of processing. First, a weighted sum of all signals at a given instant of time is calculated. If x j denotes the output signal from the j-th neuron and w i j the weight on the connection (or signal flow) from neuron j to neuron i, then the weighted input received by neuron i is:
c n
zi
=
w i j xj
+ Wi,"+]
j=1
where n is the number of neurons which feed information to neuron i. The second step is a transformation of Zi to an activation level. The activation (or threshold) level of an artificial neuron is equivalent to the level of excitement of a biological neuron. If F(Zi) denotes this (non-linear) transformation, then the result is :
Sometimes O i can be used directly. However, physical restrictions or mathematical expedience (such as rapid convergence) often require the output to be limited to a particular interval of values. Then one sets upper and/or lower thresholds, T" and TL, and defines the output as:
oj =
0, for TL ( Oj ( Tu 0 other wise
(3)
For the network of Fig. 14-1, each neuron processes information according to Eqns. (1) to (3) and transmits the output to the neurons in the next layer. This continues until the final outputs are generated. These are compared with the expected outputs from the training data and the differences (or errors) are used to correct the parameters of the network. This is called training a network. There are different methods of training, and indeed different configurations of ANNs themselves. Recognize at this stage that, as stated in Section 14.1, the ANN has used data of the process but no information about the physical principles associated with the process. In other words, an ANN simply provides a mapping between system inputs and outputs. Because physical principles are not used, it is essential that all important
14.2 Structure and Functioning of ANNs
367
features be reflected in the data used for training. The importance of proper data analysis can hardly be overemphasized, and an improperly trained network can produce misleading results [ 1,4]. The accuracy of input-output mapping, and thus of an ANN'S performance, depends on the weights, wij, and the transformation function, F(Zi), which is also known as the squashing function since it compresses the outputs to a prescribed range. The weights represent the relative contributions of the signals received by a n+ 1
neuron. Therefore
wij
= 1 for the i-th neuron. Initially random values are
j=1
assigned to the weights and they are updated as training progresses. The squashing function should generate a physically meaningful result. For example, if the output 0, represents cell mass concentration, a squashing function which produces negative results is unacceptable. There is one more aspect of the functioning of an ANN. Notice that Eqn. (1) con+ I no associated input. This is the weight of the bias neuron, which is tains w ~ , ~ with an artifact introduced in the input and output layers. Since this neuron receives a unit input, its effect is to shift the weighted sum, Zi, linearly. The main purpose of the biases is to prevent the network from getting stuck in a local minimum and aid its convergence to the global minimum [15,16]. This is a generic problem of optimization methods. In the neural network context, methods other than biases include momentum, adaptive learning [ 171, simulated annealing [18] and genetic optimization [ 191.
14.2.2 The Nature of Input-Output Transformation Artifical neural networks are, in essence, transformation systems that map a set of inputs to a set of outputs without phenomenological details. Let E(t) and L(t) be time-domain vectors of input and output variables, respectively. Then the functional transformation performed by each layer of a network K can be represented by the compositon of an affine transformation (weighted sum of the inputs) with a non-linear mapping (through the squashing function) [3]. Let i, h and o denote the input, hidden and output layers of an ANN, and let Ri(t), ?7h(t) and L(t) be their respective output vectors. Then,
_ _
w,
In Eqn. (4), F is the squashing function and w h and Toare the weight vectors. It may also be seen that the outputs at any instant of time depend on their values at a time (t-.r) and the current values of the input. Alternatively, it may be said that a network can predict the state of a process up to an interval of time T ahead of
360
14 Neural Network Applications to Fermentation Processes
its current state. In this sense it is desirable to have a large T so as to overcome the limitations of measurement delays [20]. Several choices are available for the transformation function F. Among the most commonly used for biochemical reactors is the sigmoidal function
As seen in Fig. 14-3, this function has a minimum of 0 and a maximum of 1, and is everywhere differentiable with a positive slope. It is therefore suitable for normalized physical variables such as concentrations. A variation of the sigmoidal function is the hyperbolic tangential function
which has a similar shape but varies between -1 and +1. Suppose a reaction may be operated within a temperature range [ T I ,T21. Expressed in normalized form around the mean, the range becomes [-1, +1] and Eqn.(6) is then a suitable candidate in a neural simulation. Another application might be for a bioreactor subject to input disturbances; the fluctuations of the output variables may be similarly normalized around their mean values.
Radial Basis
Saturating L i l n a r
Symmetric Satnreting Linear
Tan Signoidal
Fig. 14-3. Some transformation functions commonly used for bioreactors. In all plots the origin is (0,O) and the limits of the y-axes are (-l,+l). (Reproduced from Demuth, H. and Beale, M., Neural Network Toolbox, with permission from The Math Works, Natick, MA 0 1993.)
14.2 Structure and Functioning of ANNs
369
If the inflow disturbances have a Gaussian distribution, the bell-shaped radial basis transformation (Fig. 14 -3) mimics the response faithfully. This transformation is expressed as
ZT zi
F(Zi) = exp[ - -] 20;
(7)
where the superscipt T denotes a transpose and oi is the standard deviation of Zi. A recent application has been to a continuous fermentation for glyceraldehyde-3 -phosphate dehydrogenase with a recombinant Escherichia coli strain [21]. Figure 14-3 also shows some discrete transformations. Their equations are presented below. (a) Hardlimit function
1 for Zi 1 0 0 for Zi < 0
F(Zi) =
(b) Symmetric hardlimit function F(Zi) =
0 - 1 for Zi < 0 + l forZi
P
(9)
(c) Saturating linear transformation, s(t)
(d) Symmetric s(t)
While discrete transformation functions are not common in the application of ANNs to biochemical reactors, the linear continuous transformation is often used either for the output layer in a network with nonlinear transformations in the preceding layers or wholly by itself in adaptive linear networks (ADALINE) [5,22]. It has the form F(Zi) = aZi i.e. a straight line of slope ‘a’.
(12)
370
14 Neural Network Applications to Fermentation Processes
14.2.3 Qpes of Artificial Neural Networks Because ANNs have been developed for many purposes, they cover different algorithms and structures. There are more than 50 different types of networks and a comparably large number of transformation functions for individual neurons [2]. However, only some of these are of practical interest and a few of them apply to bioreactor problems. Structurally, ANNs differ in the number of layers of neurons, their layout and the types of connections between them. Although in Fig. 14-1 every neuron is connected to every other neuron in the next layer, this need not be the case always, i.e. some connections may be absent. Moreover, the information flow (the arrows in Fig. 14-1) need not always be forward, i.e. signals may also be fed back to the network. One of the simplest useful networks is the perceptron, created by Rosenblatt [23]. It is based on the hardlimit transfer function, Eqn. (8) or Eqn. (9), and follows a simple training law. The vector of weights W at each stage of iteration (called an epoch) is determined from the weight vector Wold of the previous iteration according to:
where X is the vector of input signals to a neuron, y is its output and
P=
[+
1 if the neuron’s answer is correct - 1 if the neuron’s is answer wrong.
This algorithm indicates the perceptron is most useful in classifying data into either of two categories, i.e. in applications which depend on yesfno answers. Not all problems are amenable to a two-way classification. Often, more than two choices are available or even a continuous range of choices, and the simplest ANN which provides this is the ADALINE, an acronym for adaptive linear network, and its extension, the MADALINE (many ADALINEs) [22] (Fig. 14-4). The ADALINE utilizes the linear transformation Eqns. (lo), (11) or (12) and the learning rule
where E is the error (or the difference between the actual output and the desired output). The quadratic dependence of the weights on the input signals promotes faster convergence than a linear dependence. The ADALINE can solve any linearly separable problem and it works quite well even for problems which are partially separable linearly, as in weather prediction [17]. Linearity may be a strong restriction because many real problems are non-linear. Moreover, ANNs such as the perceptron, the MADALINE, and attractor networks, which have either one or two layers of neurons, cannot solve problems that are not linearly separable (at least to a substantial degree for an approximate solution)
*
14.2 Structure and FunctioninP of ANNs -
37 1
Adaptive linear network
Recurrent network; R=recurrent neurons
Backpropagation network
Radial basis network; Squares represent bias neurons
Fig. 14-4.Topologies of neural networks used for chemical reactor problems.
[24]. This difficulty was overcome by the mathematical result that any non-linear function can be approximated to a specified accuracy by no more than two hidden layers of neurons [13,14]. Hence, the use of non-linear ANNs for many practical problems achieves good simulation without much increase in complexity. The most widely used is the backpropagation network, shown in Figs. 14-1 and 14-4. A backpropagation network contains one or two hidden layers between the input and output layers. The input layer merely distributes information as specified to the first hidden layer, whose neurons do the real processing. They calculate outputs according to Eqns. (1) and (2), and transmit them to the next layer. If this is another hidden layer, further processing is done and the results pass on to the output neurons, which generate the final (observed) results. The name backpropagation arises because each output neuron calculates the difference between the predicted result and the actual result (from the training data), and the derivatives of these errors with respect to the weights are passed back to the preceding layer. Each hidden layer neuron calculates the weighted sum of the error derivatives to find its contribution to the output error, corrects the weights and feeds this information back to the input layer. Because the primary flow of information is forward, and the recycling of erors is for iterative convergence, some workers [2,3] prefer the name feedforward ANN. Like the ADALINE, most backpropagation ANNs are trained by a gradient technique; this, however, differs somewhat and follows the rule:
372
14 Neural Network Applications to Fermentation Processes
Here, Awij is the change in the weight wi, (see Eqn. (l)), p is the learning constant (between 0 and l), E is the error in prediction and F(Zi) is usually the sigmoidal squashing function. Another significant difference from the ADALINE is in the computation of the errors E. The ADALINE’s method of subtracting the predicted output from the expected output applies only to the output layer of a backpropagation ANN. Neurons in all other layers calculate errors through the weighted backward summation described before. In simple form, the error of the i-th neuron in a hidden or input layer is:
where the asterisk signifies that the Ej’s refer to the fed-back errors of the next layer. Equation (16) indicates that it is useful to choose a squashing function that is differentiable. An alternative to the backpropagation network is the radial basis network. This is topologically similar but functionally different. The first difference is the choice of a radial basis transfer function, Eqn. (7), in place of a sigmoidal function. A Gaussian radial basis function has some useful mathematical properties that enable confidence intervals for each calculated estimate to be generated [2]. Secondly, unlike the backpropagation ANN, the weights on the arrows from the input layer to the hidden layer are all set to unity and do not change during training. The third difference is in the training methodology. A radial basis ANN is solved by initially clustering the measured data [25,26] into a training set and a test set. After the network has been trained and tested, the data are partitioned differently and the process is repeated until all exemplars in the original unpartitioned set have been used for testing exactly once [2]. Since radial basis networks cannot ignore certain inputs selectively, it is important to ensure that spurious data are removed and those used for training should have a strong correlation between the input and output variables. ANNs of the kinds described so far perform a static mapping between input and outputs. Since dynamics are not inherently included in their structures, one way of using such ANNs for time-dependent problems is to train with time-series data. Training is imparted in either of two ways. One method is to use data at, say, a time point, t, as inputs and values at the next point, t+l, as outputs. This is essentially an autoregressive approach, which is easy to implement but can lead to large cumulative errors. So a second approach is to minimize the final output error at time ‘t’ and all other intermediate predictions up to a specified prediction horizon. This method is called backpropagation in time [6]. Although the latter method has been successful in many real process applications, it results in extremely large networks with attendant difficulties in training. So it is preferable to have dynamics built into the network structure itself. Recurrent ANNs accomplish this. A recurrent ANN may be defined as one in which each input activity pattern passes through the network more than once before it generates
14.3 Deterministic State Estimation of Bioreactors
373
an output pattern [17]. Many ANNs can be modified to be recurrent; for fermentation applications the recurrent backpropagation network is the most common. Consider a typical network with built-in recurrence (Fig. 14 -4).Although the network receives three inputs and generates two outputs, the input and hidden layers have five and four neurons respectively. The extra neurons (denoted by R) are recurrence neurons, and they are present in order to handle feedback from the next layer. This topology allows output patterns in time to be recycled within the network, thus creating an inherent dynamic feature. It might seem that a recurrent ANN has many more neurons than a conventional backpropagation ANN. However, for satisfactory representation of time-series data a backpropagation ANN might well have more hidden layer neurons and/or be more difficult to train for practical nonlinear problems such as runaway recations, inflow disturbances and reactor failure. Besides, there are other types of ANNs less common for chemical engineering problems. These are: adaptive resonance [27], bidirectional associative memory [28], self-organizing networks [29], fuzzy neural networks [30], and learning vector quantization [29]. They are described in detail in the references cited.
14.3 Deterministic State Estimation of Bioreactors Most of the applications of ANNs to fermentation processes have been carried out since about 1990. However, there has been a considerable spurt of interest during this short span of time, mainly because the theoretical foundation had already been tested on non-biological reactors and early applications displayed significant successes. A major portion of ANN applications to bioreactors has been for the estimations of variables which are difficult to measure through on-line instruments. Some important applications are described in this section.
14.3.1 Pencillin G Industrial fermentation for pencillin G is a long-duration (180-200 h), fed-batch process which follows complex kinetics [31]. Two operating regimes can be identified during the fermentation. In the early stages, the system is operated to produce large quantities of biomass (the mycelium of Penicillium crysogenum), utilizing mainly the substrate in the initial batched media. Towards the end of this phase, the feed additions are increased as the initial substrate becomes exhausted. The addition of substrate is so regulated that its concentration remains low at all times, resulting in low growth rate and high penicillin yield in the second phase. However, a certain minimum rate of growth should be maintained in order to avoid cell lysis and, unfortunately, maximum yield of penicillin is obtained close to this constraint [32]. Since maximum productivity requires the biomass concentration to be regulated closely, the normal practice of analyzing off-line samples does not optimize the feed strategy. At best it avoids lysis conditions at the expense of improved produc-
374
14 Neural Network Applications to Fermentation Processes
tivity. Since direct in situ measurement of biomass concentration is difficult on an industrial scale, Di Massimo and coworkers [33] adopted a neural network approach to provide on-line estimates from measurable variables. Previous studies [34] had shown that the process variables which provided the most relevant information about the state of the fermentation were: (i) carbon dioxide evolution rate (CER); (ii) age of the batch; and (iii) concentrations of the key substrates. Since there are two key substrates [31,351, they applied a feed-forward (backpropagation) ANN with four input neurons, two hidden layers and one output (the biomass concentration). By applying the method of Wang et al. [36], the number of neurons in each hidden layer was determined to be six. Figure 14-5 shows that a trained ANN provided good esimates of biomass concentration for two batches whose data had not been used for training. A more complex case is shown in Fig. 14-6, which is for a batch contaminated by a yeast. The network’s prediction capability has deteriorated but it still follows the trend of the experimental data. Figure 14-7 is for a fermentation suffering from both con.tamination and false calibration of the C02 analyser. At about 90 h into the fermentation, an uncharacteristic drift in calibration resulted in deterioration of network performance; this was restored upon readjustment but again drifted after 140 h because of contamination. This shows that neural networks can be used to predict potential or imminent problems. Apart from being an indicator of malfunctions, biomass concentration provides the growth rate, which, together with batch age, can be used to estimate the amount of penicillin produced per unit of biomass. Di Massimo et al. [37] therefore fed the out-
. X
0
50
100 IIrno h
-
__,---* +
X
X
150
200
Fig. 14-5. Artificial neural network for biomass estimation during penicillin fermentation. _ - - - - Run 1 estimates; Run 1 measured; -Run 2 estimates; x Run 2 measured. (Reprinted from [lo] with kind permission of Elsevier Science-NL, Amsterdam, The Netherlands 0 1992.)
+
14.3 Deterministic State Estimation of Bioreactors
1
I
I
I
0
50
100
150
375
200
lime h
Fig. 14-6. Artificial neural network performance for contaminated fermentation of penicillin G. -Estimated biomass; Measured biomass. (Reprinted from [lo] with kind permission of
+
Elsevier Science-NL, Amsterdam, The Netherlands Q 1992.)
-
**
-
qP
0
**
I
50
1
1
100 lime h
150
2 0
Fig. 14-7. Neural network estimation of biomass concentration in the presence of false calibration Measured biomass. (Reprinted from [lo] with and contamination. -Estimated biomass; kind permission of Elsevier Science-NL, Amsterdam, The Netherlands 0 1992.)
+
376
14 Neural Network Applications to Fermentation Processes
put of the first network, along with the other two measurements, into a second ANN whose output was penicillin concentration. This had a 3 -4-4-1 topology, i.e. there were two hidden layers with four neurons each. The second network also provided accurate estimates for unseen data over the entire duration of fermentation. While these results demonstrate the usefulness of ANNs on an industrial scale, they have two shortcomings. One is the absence of quantitative indexes of network performance; computed values of mean squared error, cost functions, and receiver operating characteristic curves [4] would have strengthened the authors' claims. Secondly, Figs. 14-6 and 14-7 do not tell how in the absence of off-line measurements, the cause and nature of a problem (contamination or improper calibration) can be inferred from the estimated values. Studies described later in this section have addressed the question of quantitative measures of fault detection.
14.3.2 Ethanol Karim and Rivera [38] studied the unsteady state cultivation of Zymomonas mobilis to produce ethanol in a batch fermentation as well as a continuous fermentation with cell recycle. This fermentation is sensitive to temperature [39] and is inhibited by high concentrations of the substrate (glucose) and ethanol [40}. During batch fermentation, off-line samples were collected once every 15 minutes and analyzed for biomass, glucose, and ethanol concentrations. The COz concentration in the efflux gas was measured on-line to obtain the C02 evolution rate (CER). This was used along with on-line measurements of broth temperature, redox pontential, and optical density as inputs to a feed-forward network, whose outputs were the biomass, ethanol, and (unreacted) glucose concentrations. The optical density is a measure of the cell growth and the redox potential indicates the concentration of viable cells. The number of neurons in the single hidden layer was increased stepwise from two to ten and, from the prograss of the sum of the squares of the errors, three neurons were found to be optimal. Thus, the ANN selected had a 4-3-3 configuration. Five sets of data, at 30, 33, 35, 37, and 39.5 "C, were available. However, because of lack of common sampling times and differing durations of off-line analyses, it was not possible to combine the sets directly. Rather than interpolate to generate data points for a set of common times, Karim and Rivera [38] preferred to present the actual measurements sequentially. The ANN was first trained with data at one temperature; the converged weights and biases were used as initial conditions for the next temperature, and this was continued through all temperatures. Obviously, the order of presentation of the data is important; Karim and Rivera's order was from 39.5"C downwards to 30°C. The final sum of the squares of the errors (SSE) (after 10 h fermentation) is revealing. While network performance is the best for the final data set (Fig. 14-8) as expected, the SSE does not decrease monotonically (Fig. 14-9). This observation, and the fact that, for a given temperature, the SSE values did not change appreciably after about 25 epochs but the weights did, suggests a highly convoluted error surface [38].
14.3 Deterministic State Estimation of Bioreactors
377
-
h
-
3
60
2.5 i .8 2
50
40
1.5
30
1
20
p
0.5
10
;
h
L M v
26 i
G
'3
2
1 0
o a
0 0
2
4
6
a
Tirnc, h
10
Y B
Fig. 14-8.Neural network test data and estimates for batch fermentation of ethanol at 30 "C after sequential training to a sum of squares of errors (SSE) of 0.007. The continuous lines are the measured profiles and the broken lines are the estimates. C, biomass; 0,glucose; 0 , ethanol. (Reproduced from [38] with permission from Springer-Verlag GmbH & Co. KG, Berlin 0 1992.)
v1
aJ
0.15
+A !
E
;0.10 0-
L
0
0.00 Data set temperature,
OC
Fig. 14-9.Variation of final SSE during sequential training with data at decreasing temperatures in batch fermentation for ethanol. (Drawn from data taken from [38].)
To test the neural network, data at 37 "C were excluded from the training. Excellent estimates of all three output variables were obtained throughout the fermentation period (Fig. 14-10). The use of a prefiltering algorithm to minimize the noise in the raw data improved the accuracy of the estimates considerably, the SSE being reduced from 0.18 to 0.08. The ANN also acquired good generalization capability. An ANN trained sequentially with isothermal data sets was employed to obtain estimates of output variables for a non-isothermal fermentation. Good results were obtained for biomass and ethanol, but glucose concentrations diverged as fermentation progressed. Karim and Rivera [38] also studied ANN performance for a chemostat culture with recycle of cells. Input variables were the biomass concentration and the
378
*n
14 Neural Network Applications to Fermentation Processes
2
60 50
> g
' . . I I
40
30 20 v)
$
3
0.4
0
s o
0
0
2
4
6
8
10
8
4 ,. M
U
Time, h
Fig. 14-10. Neural network test data and estimations for ethanol fermentation at 37 "C with batch data run upto SSE of 0.08. Other details are the same as in Figure 14-8. (Reproduced from [38] with permission from Springer-Verlag GmbH & Co. KG, Berlin 0 1992.)
dilution rate, and the output variables were glucose and ethanol concentrations. To simulate industrial operation on a laboratory scale, the dilution rate was varied in a square wave fashion with a random amplitude between 10 h-' and 14 h-l. Furthermore, a 5 % amplitude random noise was superimposed to observe the effect of measurement noise on the ANN. In spite of these non-idealities, a simple network with two neurons each in the input, hidden and output layers was adequate for test data quite different from the training data. The comparisons are shown for a case where the dilution rate variation was triangular (Fig. 14-11) and another where the
30 25
-
20
29 80 6
-L
160
>
8
a
40
5
20
0
0 0
10
20
30
40
50
Time, h
Fig. 14-11. Test data for the neural network by varying the dilution rate in a triangular pattern. Inputs are 8biomass concentration and f dilution rate. Outputs are 8 glucose concentration and ++ ethanol concentration. Feed substrate + is provided as a disturbance in the process. (Reproduced from [38] with permission from Springer-Verlag GmbH & Co. KG, Berlin 0 1992.)
14.3 Deterministic State Estimation of Bioreactors
379
20
16
4
0
0
10
20 30 Time, h
40
50
Fig. 14-12. Neural estimations for batch ethanol fermentations with a network having two hidden nodes. The dilution rate was varied triangularly and there was a 20 % variation in the maximum specific growth rate around a mean value of 0.45 h-l. Other details are the same as for Figure 14-11. (Reproduced from [38] with permission from Springer-Verlag GmbH & Co. KG, Berlin 0 1992.)
input variations caused a 20% random variation in the maximum specific growth rate around a mean value of 0.45 h-' (Fig. 14-12). Karim and Rivera emphasize that in all training sessions the networks converged very fast, reaching close to the final SSE within 10 to 20 epochs. This an important consideration for on-line implementation.
14.3.3 Glucoamylase Glucoamylase is produced by Aspergillus niger in a starchy medium. Unlike ethanol fermentation, glucose does not inhibit the formation of the product. Batch operation is the preferred method, with a continuous supply of air. Although the duration of fermentation is shorter than for penicillin, it is long enough (50-60 h) to make it difficult to have direct sterile measurements. Moreover, as for penicillin and ethanol, the product activity and biomass concentration are the principal output variables of interest. In spite of their practical importance and use in the control of the process [41], both these variables are measured through off-line samples. Linko and Zhu [42] therefore attempted to provide continual on-line estimes through neural networks. Previous studies [41,43] had shown that on-line measurements of the oxygen uptake rate (OUR), the CER, pH, speed of agitation, and the accumulated volunies of COz released and nitrogen utilized were the most relevant variables to predict the time-dependent biomass concentration and glucoamylase activity. These variables were therefore the inputs to a feed-forward ANN. Following both therory [5,13] and previous examples [6,15,27], a single hidden layer was adopted and its
380
14 Neural Network Applications to Fermentation Processes
number of neurons was determined by trial and error. The optimal architecture was 6-10-2. Linko and Zhu wished to provide output estimates for two time steps ahead of input data; to do this the enzyme activity and biomass concentration at the additional times (t+l and t+2 for input data at time 't') had to be included among the output variables. This, however, did not increase the number of hidden neurons, implying that multi-step ahead predictions were possible without complicating the signal flow structure. Although this might suggest that a 6-10-(2M+2) network will work for M time steps, good accuracies of predictions far ahead in time may require either more hidden neurons or one more hidden layer. An alternate solution to the multistep prediction problem is to use dynamic neural networks or networks with additional transfer functions expressing time delays [3,6,33]. The ANN was trained with six sets of data covering pH between 4.4 and 5.6, and agitation rate from 760 to 1200 rpm. The temperature was held constant at 35 "C and the dissolved oxygen at 30 % of saturation through all experiments. After about 2000 epochs of training by backpropagation with momentum to speed up convergence and overcome local minima, the ANN could faithfully portray the performance of a fermentation run whose data had not been used for traning (Fig. 14-13). The efficiency of the ANN may be gauged from the calculated coefficients of determination (R). With six data sets for training, R2 was 0.995 for glucoamylase activity and 0.977 for biomass concentration in the test data set. Even when just one data set was employed for training, the R2 values were comparably good, being 0.984 and 0.985 respectively. It is worth noting that no filtering was done to reduce noise in the data, thus illustrating the robustness of a neural network simulation. The ability of ANNs to detect and learn key characteristics of a process under realistic conditions is one of their major advantages.
0
1
0
2
0
3 0 Time (h)
4
0
5
0
~
Fig. 14-13.Real-time estimation of enzyme activity and biomass concentration in a batch glucoamylase fermentation. Open symbols denote actual off-line measurements of glucoamylase activity (0) and biomass concentration (0)respectively, while filled symbols are the estimates. (Reprinted from [42] with kind permission from Elsevier Science Ltd., Kidlington, U.K. 0 1992.)
14.3 Deterministic State Estimation of Bioreactors
38 1
14.3.4 Activated Sludge Treatment The activated sludge process is a key step in the treatment of industrial wastewater. Even though it has been in use for many years, the complexity of the rheology and the fermentation have precluded the development of good simple mathematical models for design and control. So, either simple models are supplemented by operating experience [44] or complicated structured models have been proposed. The scheme of the activated sludge process is deceptively simple. Wastewater from a primary sedimentation tank is fed into an aerobic biological reactor (the sludge tank). The fermented broth goes to a settling tank, from which clarified water is drawn out at the top and the settled sludge is recycled to the first tank. Two complexities make modeling difficult. One is the water itself, which is a poorly characterized slurry with a complex rheology. The other problem is the presence of several microorganisms, whose simultaneous and interactive metabolisms are difficult to characterize in detail. While lumped models are satisfactory for steady-state design, they are not suitable for the changes in waste composition that an activated sludge reactor has to deal with. So structured models have been proposed [45,46] but their utility for on-line implementation is limited by their complexity and the difficulty of estimating and adjusting the parameters. This process is thus a good candidate for the application of a neural network. In seeking a suitable ANN, Tyagi and coworkers [47] recognized that the available experimental data did not permit a full-scale ANN to be trained. So they trained and tested the ANN first with simulated data and then with data from a pilot plant. The simulated data were generated by an unstructured model for the sludge vessel and a similar model based on the limiting flux theroy for the settling tank. The model was solved at steady state and critical loading. The objective of the ANN was to predict the recycle ratio, a, and the sludge outflow rate, F,, from data of the inlet (Si) and outlet (S) concentrations of substrate for the activated sludge reactor and the flowrate per unit cross-sectional area, FIA, of the settling tank (i.e. the velocity). A simple 3-2-2 backpropagation network was suprisingly adequate; the root mean square errors for a and F , were 0.0121 and 0.0717 respectively, compared with 0.1198 and 0.4493 for the best-fit non-linear equation. In the second part of the study, this network was expanded to simulate plant data spanning different recycle ratios, applied settling tank mass loading, overflow velocity from the settling tank, and biomass concentrations in the sludge tank and the recycle stream. The last two variables were the outputs and the first three the inputs of the neural network, which means the number of input and output nodes remained unchanged. However, as many of 14 nodes in the hidden layer and 3000 epochs of training were required. This underlines the complexity of an actual process. The inadequacy of deterministic models to describe the complexities of activated sludge treatment is further seen from Tyagi et d ’ s [47] result that the root mean square error for the biomass concentration in the sludge tank was 3599 (for data
382
14 Neural Network Applications to Fermentation Processes
0 Training data fitting
V
d
Ncural model prediction V
I
I
I
I
0.5
1 .o
1.5
2,o
/
0.0 I
o,o
Observed, F, (m3h-')
4
5
6
7
8
0
Observed, x (g 1.')
1
0
1
1
Fig. 14-14.Comparisons of neural predictions with model-generated training data for the activated sludge process. F, = sludge outflow rate; x = biomass concentration. (Reprinted from [47] with kind permission from Elsevier Science Ltd., Kidlington, U.K. 0 1993.)
in the range 4570 to 11300, with the largest possibly being an outlier) and that for the recycle stream was 1.001 (data range = 1.43-3.52) Figure 14-14 compares of the network's performance for simulated data and actual data.
14.3 Deterministic State Estimation of Bioreactors
383
14.3.5 Recombinant Protein-1 Fermentations based on genetically altered microorganisms have special problems which make in vitro instrumental measurements very difficult. The organisms are prone to structural and segregational instabilities [48] and are strongly sensitive to disturbances and to the mixing characteristics of the broth [12,49]. In addition, their stringent sterility requirement makes it difficult to insert sensors without the risk of contamination and interference with the mixing of the broth. If the recombinant protein is not secreted out by the cells, its on-line measurement becomes impossible. Even for an extra-cellular product, its rate of formation depends on both extracellular concentrations and the concentration of DNA molecules which contain the gene coding for the product [50]. The latter, as mentioned above, change dynamically owing to stablity problems [48] and the interaction of cellular metabolism with transport processes [49,50]. Given these limitations, it is important to have continuous non-invasive monitoring during the fermentation. This consideration motivated Glassey et al. [51] to study the possibility of applying neural networks. The experimental system used an E. coli K-12 strain transformed with a vector directing cytoplasmic accumulation of the recombinant protein upto 30-50 % of the total protein. Fermentations were run in batch mode until the oxygen uptake rate (OUR) reached the maximum oxygen transfer rate (OTR). The growth medium was so designed that it became carbon limited at this time, and a fixed level of dissolved oxygen was maintained thereafter by controlled supply of substrate. Glassey et al.’s [51] interest was in predicting the time-varying concentrations of the biomass, the genetically nonviable cells, and the recombinant protein. For the biomass, a 5 -2-3 -1 backpropagation network was optimal, with the input variables shown in Fig. 14-15. The ANN performance plots (Fig. 14-16) reveal useful information. For a fermentation with 20 g 1-’ yeast extract (which was added to the medium) the average prediction error was 13.3 % while for 5 g 1-’ yeast extract it was 25.4 %. The authors attribute the difference to insufficient training data in the second case. Prediction of the recombinant protein concentration and its accumulation (as a percentage of total microbial protein) required ANNs of topologies 5-6-1 and 5-4-1
Fig. 14-15.Artificial neural network employed by Glassey et al. [51] to estimate biomass concentration in a batch recombinant fermentation. a = feed rate; b = batch age; c = batch concentration; d = feed initiation time; e = temperature; f = biomass concentration.
384
14 Neural Network Applications to Fermentation Processes
1.2 [
I
0.6 -
a 0.2 0
5
0
15
10
time (hours)
-estimate(20gA)
,
. . . estlmate(5g/l) x
A
actual(?Og/i)
actual(5g/l)
Fig. 14-16. ANN model predictions and measured data for biomass concentration in batch recombinant fermentation. The figures in parentheses are yeast extract concentrations. (Reprinted from [51] by permission of John Wiley & Sons, Inc. 0 1994.)
respectively, with the same input variables as for biomass. Even with one hidden layer and fewer hidden neurons, the ANN could faithfully follow the strongly non-linear character of the accumulation profile which passed through a minimum and a maximum during the fermentation period. An even simpler network (4-3-1) was adequate for the concentration of cells which contained recombinant DNA but could not synthesize the protein because of structural rearrangement of the vector. Glassey et al. ’s study demonstrates that relatively simple ANNs can, with appropriate filtering of the data and judicious training, simulate recombinant fermentations even with the limited information available under industrial conditions.
14.3.6 Recombinant Protein-2 In an extension of their previous work, Glassey et al. [52] considered E. coli fermentations producing one of three proteins, depending on the whether there was no induction, chemical induction, or thermal induction. As before, batch operation until the OUR was maximum was followed by fed-batch operation, and there was a continuous supply of yeast extract. Backpropagation ANNs were employed for on-line estimates of the biomass concentration and the recombinant proteins. The results pose interesting questions about network architecture and training. While a 3-3-1 network was the ‘best’ for the biomass concentration, the specific growth (which, mathematically, is a function of only the biomass concentration) required a 3 -2-1 network. Considering that the biomass profiles do not have strong non-linearities (see their Figs. 2 and 3), it is difficult to understand why the topologies should differ. Moreover, Glassey et al. [52] do not provide data of how the sums of squares of the errors varied with the number of hid-
14.3 Deterministic State Estimation of Bioreactors
385
den nodes, or how the error distributions for 3-2-1 and 3-3-1 topologies compared with experimental errors. While most of the ANN predictions were satisfactory, an atypical case is shown in Fig. 14-17. This case highlights the importance of proper data analysis and training. The relatively inaccurate predictions occurred despite the introduction of a firstorder, low-pass filter and smoothening of off-line data by cubic splines. The authors say that the filter improved performance by 10.5 to 36.8 %, and suggest that difficulties of the kind seen in Fig. 14-17 may be overcome by developing an accurate ANN for 'typical' fermentation(s) and simultaneously employing a fault detector to eliminate non-typical behavior. The constitutive recombinant protein required a 3 -4- 4-1 network while its specific rate of production was adequately expressed by just one hidden layer with five neurons. Again, one may ask why the topologies differ and whether the sum-ofsquares error is a sufficiently good indicator [4]. In this context, Glassey et al. [52] mention that initially they explored the genetic algorithm method [19] to select the input variables and the best ANN topology through a software package Beagle (reference not given). Although Beagle could select inputs and topologies correctly in about 80% of the cases studied, this was said to be only marginally better than what was possible through trial-and-error by an experienced worker. Nevertheless, their study re-emphasizes previous attempts [33,53,54] to devise systematic algorithms to derive a suitable network configuration. As for the biomass, the ANN could mimic normal operation satisfactorily but was less successful in detecting vector instabilities. So an autoassociative ANN (6-3 -23 - 6) was chosen. It may be seen that there are three hidden layers, an unusual situation which indicates the complexity of structural instability problems in large scale fermentations that inevitably have other non-ideal effects. Glassey et d ' s [52] approach to the plasmid instability problem was based on Kramer's [55] observation that a plot of the output of one of the hidden neurons 0.8 I
0.7 -
3E
hp
0.5
-
0.3
-
0.2
-
0.4
..."
I
...'
..*'
,..'
-.
n i, L
I 35
30
lo l5 2otime (h205ura) .. . network estimate smoothed
-
A
40
45
raw
Fig. 14 -17. Experimental and ANN estimated values of biomass concentration in recombinant protein production by an E. coli strain containing a constitutive vector. This represents an atypical operation. (Reprinted from [ 5 2 ] with kind permission from Elsevier Science Ltd., Kidlington, U.K. 0 1994.)
386
14 Neural Network Applications to Fermentation Processes
against that of another may indicate deviations from normal process behavior. Such a set of plots is hown in Fig. 14-18. All plots begin from the lower center part of the graphed area. The plots indicated by (x) are for fermentations in which vector instability was observed while those marked (+) relate to stable operation. The two long plots on the left (0)are for fermentations run under controlled conditions to minimize vector instability. It is seen that the plots gradually move into the region of instability, suggesting that fermentation beyond a certain time is undesirable. It should not be inferred that Fig. 14-18 is only a qualitative indicator. All three axes were quantitated but, because the results pertain to an industrial application, neither the scales of the axes nor information about the proteins could be revealed. This study, however, demonstrates the power of an ANN to predict incipient problems from estimates of outputs which do not even need to have physical relevance.
0 0 0 inrt u u winst
A A A nom V V V
lnrt
+x +x +x
nom opt
0 0 0 nom
Fig. 14-18. Detection of plasmid structural instability during a recombinant fermentation by means of the outputs of the central hidden layer in an auto-associative ANN. (Reprinted from [ 5 2 ] with kind permission from Elsevier Science Ltd., Kidlington, U.K. 0 1994.)
14.4 Bioreactor Estimations in the Presence of Noise
387
14.4 Bioreactor Estimations in the Presence of Noise Biological reactions carried out on an industrial scale are prone to disturbances in the inflow streams, the mixing pattern in the broth, the measuring devices, etc. While the interactions between metabolic reactions and hydrodynamics are complex enough in a ‘smooth’ operation [ 5 6 ] , the influx of disturbances can alter the reaction pattern significantly. For instance, fluctuations in the flow rates of feed streams to a bioreactor can continually change the spatial and temporal distributions of substrate, products, and cells in the broth. This obviously changes the productivity. In a recombinant fermentation, the plasmid-retaining ability of recombinant cells depends on the variations sensed by them in the extra-cellular concentration [49,57]. By affecting the variations, process disturbances influence plasmid stability and consequently the rate of protein formation. It is diffcult to portray dynamics of these kinds through models which are simple enough to permit application in an automated system. Although adaptive observers such as the extended Kalman filter have been used, their limitations [58] have motivated increased attention toward neural networks.
14.4.1 Industrial Mycelial Fermentation Di Massimo et al. [37] studied the growth and fermentation by Fusarium graminearurn for the production of mycoprotein. The process operates in a continuous mode in bioreactors of about 40 m3 capacity. It is operated for very long durations (= 1000 h) and is therefore susceptible to process disturbances. In addition, the volume of the reactor and the growth of mycelial biomass make the broth heterogeneous and difficult to characterize. Adequate accumulation of the product takes place only in the presence of excess carbohydrate, so biomass control through straightforward nutrient limitation is not possible. Close regulation of biomass concentration is important for product quality. However, off-line samples are typically drawn every 4 h and each analysis takes 2 to 3 h. Thus, conventional monitoring provides an insufficient basis for process optimization and control. Di Massimo et al.’s approach to the problem was to employ a feedforward ANN with a 6 - 4 - 4 - 1 topology. The input variables were the dilution rate, the COz evolution rate, the oxygen uptake rate, the alkali addition rate, and two unspecified variables. Biomass concentration was the output variable. Figure 14-19 shows the performance of the network for a step change in the dilution rate. Despite the large ‘disturbance’ the ANN continued to follow the process accurately without the need for model adaptation (a result of good training) whereas an adaptive linear estimator had to undergo severe adaptation of the parameters to the change in dilution rate. The ANN was equally effective in tracking the biomass concentration in the presence of random noise, even when some distrubances caused a sharp change in fermenter performance (Fig. 14-20).
388
14 Neural Network Applications to Fermentation Processes
6 -
-
4.5
\
3 -
0
100
200
300
400
5
to
Time (Hours)
Fig. 14-19. Neural network estimation of biomass concentration following a step change in dilu, measured biomass; - - - -, estition rate during continuous production of mycoprotein. mated biomass. (Reprinted from [37] with the permission of SCI, London 0 1992.)
The authors, however, caution that an ANN may not function properly in the presence of sensory problems such as faulty calibration of the C02 analyzer in this example. 6 ,
t
I
I
I
0
100
200
300
4
0
Time (houre)
Fig. 14 -20. Neural tracking of biomass variations after process disturbances during continuous , measured biomass; - - - -, estimated biomass. (Reprinted fermentation for mycoprotein. from [37] with the permission of SCI, London 0 1992.)
14.4 Bioreactor Estimations in the Presence of Noise
389
14.4.2 Adaptive Neural Networks Two studies by Cheruy and coworkers provide insight into how ANNs can be designed to perform dynamic predictions in a noisy environment. In the first work [59], an unsteady state, continuous stirred tank bioreactor was simulated by an extended Kalman filter (EKF) and by a feed forward ANN. Data were generated by solving the material balance equations for biomass and substrate with the specific growth being inhibited by the substrate according to P=
S K , +S+ SZIK;
where p is the specific growth rate, S is the substrate concentration and Ki, K, are inhibition constants. The ANN was of the feedforward type with a 3-2-2 architecture and was trained by the backpropagation algorithm. As in the later work of Linko and Zhu [42] for glucoamylase and Karim and Rivera [38] for ethanol, the feed rate and concentrations of substrate and biomass at discrete times were used as input data, and the two concentrations one time step ahead were the outputs of the network. To generate the initial data for training the ANN, a bioreactor operating under optimal conditions for a dilution rate of 0.5 h-’ was subjected to pseudo-random step changes in the feed rate at each sampling interval. An interesting finding of the training procedure was that the time required per iteration (or epoch) increased linearly with the number of hidden nodes, and the sum of the squares of the errors after 5000 epochs increased hyperbolically. The latter observation is particularly important because it indicates that a network which is too large can learn the discrepancies and irrelevent features of the data in addition to the desired features. To predict the same data by a discretized EKF, it was assumed that all parameters of the model were known perfectly except K , and Ki, which were considered to be 0.3 and 0.9 instead of the ‘true’ values 0.25 and 1.0. The EKF and the ANN provided good predictions both without and with 30% Gaussian noise (Fig. 14-21), despite the large and sudden variations in the inflow rate. However, Thibault et al. [59] point out that the EKF performed well because many parameters (except K , and Ki) had precisely known values and the gain values could be intialized properly because the nature of the noise was also well known. In practice, neither a fermentation model nor a well-characterized noise is available, and predictions still have to be made with corrupt data. Thus, while an EKF may work in a relatively ‘clean’ situation, the ANN scores in real applications. Other limitations of the EKF for bioreactors have been mentioned before [2,58]. The second study by Cheruy and associates [60] extended their work to on-line prediction when kinetic changes appear during the fermentation. The previous model [59] was extended by including a kinetic rate in the biomass balance to account for decay of the biomass, X . When Kd = 0 there is of course no death (or decay) of cells. As before, data for the ANN were generated by solving
&x
390
14 Neural Network Applications to Fermentation Processes
Extended Kalman tllter ’
0.00
-0.01 -0.02
-
Neural model a
1st
order model
2nd order model
0.00
-0.01
-0.02 0.02
Noise
G -0 01 -0 02 Time 0.25
--
Extended Kdman filter
j;. 0.00
-
‘5 -0.25 -0.50 Neural model
2
0.00
‘3 -0.25
il
-0.50 0.25
R
0.00
P
E
-0.25
-0.50 1 0
1 8
10
2b
sb
4b
I
50
,
I
60
*
l
70
*
i
80
Time
Fig. 14-21. Comparison of the performance of an ANN and extended Kalman filter for substrate and biomass estimations in a substrate-inhibited reaction. The upper figure is for substrate concentration in the absence of measurement noise, while the lower figure is for biomass concentration with 30% Gaussian noise. (Reprinted from [59] by permission of John Wiley & Sons, Inc. 0 1990.)
14.4 Bioreactor Estimations in the Presence of Noise
391
the modified model under random fluctuations in the flow rate in the range [0.4, 0.51. The ANN was robust to the noise if there was no decay of cells but not so if Kd f 0. To improve performance in the presence of kinetic changes with noise, Van Breusegem et al. [60] adopted a moving learning window. The approach is briefly outlined below. 1. Select an ANN topology from experience or simulation. 2. Choose the length, L, of the learning window. 3. For each sampling instant T 2 L form a new learning data set with the last L successive pairs of input and output data vectors. 4. Retrain the ANN with the updated data set. The moving learning window method was also employed by Bhat and McAvoy [6] to predict and control the pH in a continuous flow, stirred reactor five time steps ahead of the input data. By shifting the data set in time, the adaptation scheme helps the network to learn new features and reduces the influence of obsolete or improper data whose sustained presence during training would have undermined the network’s learning ability [5,6,60]. The improvement resulting from this method is clearly seen in Fig. 14-22, where, apart from the disturbances in flow, Kd was changed abruptly from 0 to 0.05 at the dimensionless times T = 25 and 125 and returned to 0 at T = 75 and 175 respectively. Whereas the initial ANN (trained with data for & = 0) overpredicted the biomass concentration when Kd was changed, the adaptive ANN could adjust to the changes. Even this ANN shows sudden increases in the prediction errors at the transition
0.05 n
vr
I -0.00
-
(S -0.05
-
0.10
-
0.05
-
n
X
I
5-0.00
:
-0.05
r
0
l
c
q
l
50
l
l
c
l
l
r
r
l
l
100
140
260
2
Time
Fig. 14-22. Neural network tracking of noisy process variables in a substrate-inhibited reaction by a moving learning window. Thick lines represent an adaptive neural model and thin lines an initial neural model. (Reproduced from [60] with the permission of Canadian Society for Chemical Engineering 0 1991 .)
392
14 Neural Network Applications to Fermentation Processes
points of & because a one-step ahead prediction scheme was used. A multi-time step adaptive scheme [42] would have smoothed the errors, but at the cost of increasing the size of the network. The moving window method with adaptive hearing was successful even when there were delays in the availability of off-line data. Suppose the results of a sample taken at the instant T are available at the time (T+D).Then, in the previous scheme, a new learning data set is generated with the last L successive pairs of input and output vectors corresponding to the sampling instants ( T - G D + ~ )to (T-D). The ability of the ANN to function effectively even with measurement delays is very useful in fermentations where off-line analysis of intra-cellular components is a lengthy procedure but the process requires tight control in view of its sensitivity to disturbances, as in the case of recombinant strains [51,52].
14.4.3 Intrusion of Noise in the Start-up Phase An area of neural network application that has not yet attracted much attention is the start-up phase of a fermentation subject to disturbances. It is important to differentiate the start-up phase from the remainder of the fermentation because the inoculated cells are still growing to their optimal age and concentration, and have not fully adapted to the medium. Therefore a disturbance in this early period is likely to have an altogether different effect than at later times. This author’s studies [ 11,56,61] have shown that, in a recombinant fermentation, sensitivities of key variables to parametric disturbances can vary by several orders of magnitude and exhibit temporal variations quite different from their long-term patterns. While mechanistic models are inadequate for such complex dynamics, even the moving window methods [6,60] will be difficult to apply because very small learning windows will have to be shifted through small steps in time. The root of the problem lies in the imposition of dynamic pattern recognition on a static network, which the feedforward ANN is. Recognizing this limitation, Patnaik [21] adopted a radial basis ANN for the startup phase of chemostat in which glyceraldehyde-3 -phosphate dehydrogenase was produced by a recombinant E. coli harboring the plasmid pBR Eco gap. Timedomain data were generated through a model in which the probability of loss of plasmid upon cell division was a function of the specific growth rate [62]. Noise was introduced in the feed stream by 10% random fluctuations in the concentration and the dilution rate. In view of the weaknesses of feedforwarded ANNs described above, a radial basis ANN was chosen. Although such networks tend to require more neurons, they can often be trained faster [2]. Since previous studies [63,64] had shown that the presence of plasmid-free cells and the recombinant product in the inoculum may affect the course of fermentation, these variables were included in the input vector along with the dilution rate, the concentration of substrate in the feed, and the concentration of recombinant cells at the start of fermentation.
14.4 Bioreactor Estimations in the Presence of Noise
393
The variations of the prediction errors with the number of radial basis neurons indicated that the network was able to learn faster from data not inhibited by plasmid-free cells or product. This difference was reflected in the training data as well as the test data. The error distributions with test data (Fig. 14-23) show that a network trained with data corresponding to a starting culture without plasmid-free cells or product was able to predict better the performance of a fermentation with these cells and/or product in the inoculum than was possible in the reverse order. By analogy with similar observations by Thibault et al. [ 5 9 ] , Patnaik [21] attributed this to the greater dominance of noise in inhibited fermentations because of the slower rates of growth and product formation; as a result, the ANN tends to fit the trends in the noise in addition to those of the main process. The best radial basis ANNs had five input neurons for the variables mentioned earlier and four output neurons for the concentrations of the plasmid-free cells, plasmidbearing cells, substrate, and product. The number of radial neurons dependend on the training data set and the convergence criterion; for a 10 9% sum-of-squares error, four to eight neurons were sufficient, which is comparable with the number of hidden neurons in feedforward ANNs for similar problems. It is also seen in Fig. 14-23 that the ANN provides better predictions of the concentration of recombinant cells and their product than of the other two variables. This is so because the magnitudes of the former two were one to three orders larger than those of plasmid-free cells and the substrate; hence the disturbances had a more dominating effect on the latter pair of concentrations. While the errors are smaller than the parametric sensitivity changes [61], they point out the importance of proper scaling of all input and output vectors before training and testing a network.
v Ll
&
0 '
-50
-
"
W I
Y
g
o -30
-600,001
-10
0
0.01
-20 6
A
7 P
8
Fig. 14-23. Variation of the network prediction errors with concentrations for a radial basis network applied to glyceraldehyde-3-phosphate dehydrogensase fermentation. (Reproduced from [21] with permission from International Thomson Services, U.K. 0 1995.)
394
14 Neural Network Applications to Fermentation Processes
14.4.4 Radial Basis Network Analysis of Penicillin Fermentation Apart from their robustness and fast learning rates, radial basis functions enable the calculation of useful statistics of network performance This aspect and robustness in the presence of noisy data was investigated by Montague and Morris [2] for fedbatch fermentation in the production of penicillin G. While detailed information about the process is lacking because of commercial confidentiality, the results displayed in Fig. 14-24 provide useful information.
Fig. 14-24. Application of a radial basis network to an industrial penicillin fermentation. The central plots are the estimated biomass concentrations, and the upper and lower plots are the 95 % confidence limits. (a) and (b) show the ANN performance with the training data; (c) is a test of the network with ‘unseen’ data. Owing to commercial confidentiality, the numerical values of the concentrations could not be shown. (Reproduced from [ 2 ] with permission from Elsevier Trends Journals, U.K. 0 1994.)
14.5 Applications to Disturbances and Process Faults
395
The network had four inputs - fermentation age, oxygen uptake rate, COZ evolution rate, and substrate feed rate - from on-line measurements. The only output variable was the biomass concentration. The radial basis layer had eight neurons. The ANN was trained with two different sets of data. Figure 14-24(a) shows its performance for the first set, along with 95 % confidence limits calculated by the radial basis function. In spite of the visible effects of noise in the laboratory assays, the ANN has learnt the underlying fermentation behavior. Data set 2 (Fig. 14-24(b)) also shows a similar feature except with one difference. Beyond about 150 h into the fermentation, data set 1 shows a rising trend and data set 2 a falling trend. These changes occurred due to the intrusion of bacterial contamination, which a network trained with ‘good’ data would not be able to simulate. In Fig. 14-24(c) are plotted the network’s predicted biomass profile and the actual data for a run which had not been used for training. The confidence limits indicate that the estimates of biomass are good until the point where the two training sets diverge. There are several possible methods to overcome this problem. One method is to train the ANN with data from the corrupted range only (beyond 150 h in Montague and Morris’ example) and combine it with an adaptive linear estimator for the relatively ‘good’ data, i.e. data colored by noise but free from process failures. This is similar to the inferential control scheme proposed by Willis et al. [3]. The second method is to use two or more networks in a hierarchy so that each subnetwork learns a particular aspect of the process [63]. A third approach is the moving window method described in section 14.4.2; by updating the weights in each time slice, the ANN follows any aberrations in a process more closely than it would if it were trained with all data together.
14.5 Applications to Disturbances and Process Faults It was mentioned in section 14.1 that artificial neural networks are able to discern key features of a process even if the data are corrupted by noise [1,2,4]. Since most real data, especially from industrial operations, are affected by disturbances, this is a singular advantage. This aspect of ANN performance has led to two corollary areas of application. One is their use for the on-line detection of malfunctions during a fermentation. The other is neural simulation when quantitative specifications cannot be provided for certain variables. Timely diagnosis of faults is of paramount importance since they affect the subsequent course of fermentation and can spoil an entire batch of product. Fermentations based on genetically engineered organisms generate expensive products and are extremely sensitive to deviations from the prescribed optimal conditions [56,6 11. Insuch cases, prediction of a problem before its occurrence allows corrective action in advance. Fault detection through ANNs is based on the concept of pattern recognition. ANNs with unsupervised learning can be the equivalent of k-nearest neighbor or k-nn classifiers [64]. This idea is applied in fault diagnosis by recognizing that a
396
14 Neural Network Andications to Fermentation Processes
defect which has occurred or is imminent changes the normal profiles of the variables of interest. These changes may be temporal or spatial (e.g. in a packed bed or an air-lift reactor). It is often possible to identify different types of faults according to the profiles of key variables, which serve as a signature of the particular problem. Hoskins et al. [651, for example, could distinguish 19 different kinds of problems in a complex chemical plant by training a backpropagation network to recognize the output pattern for each case. The basic idea here and for bioreactors is to train the network with data from good experiments and faulty experiments until the ANN has learned to identify each problem. Although pattern recoginition was not used explicitly, Montague and Morris’ [2] examples of neural recognition of false calibration and contamination illustrate the usefulness of ANNs in detecting fermentation problems in an industrial environment. Sometimes it may not be possible or necessary to quantify the information in order to understand or analyze a process. We may, for example, be interested in knowing whether or not the temperature or the concentration of a poison has exceeded a threshold value that is fatal to the microbes, but not in the exact value of the variable. Such situations are said to be ‘fuzzy’, and recent developments in ANNs have made them applicable to fermentations based on fuzzy information. Linko [66] has provided an excellent introuction to bioengineering applications and to fuzzy logic on which these applications are based. Fuzzy reasoning is beneficial because microbial processes have many uncertainties. Many microbial systems are non-homogeneous, shear dependent, non-Newtonian, and sensitive to changes in operating conditions. Owing to complex interactions between metabolism, kinetics, and transport processes, an exact model or a scale-up procedure is difficult to formulate, and therefore most methods of design and analysis are based on ‘initial intelligent guesses’ [67]. Since both neural networks and fuzzy logic operate satisfactorily with uncertainties in data, it might appear natural to combine the two methods. The fact, however, is that they have grown separately and only in the past 3 or 4 years have fuzzy neural networks been applied to biotechnological processes. Nevertheless, the few publications that have appeared are important indicators of the potential of fuzzy ANNs for bioreactor optimization and control.
14.5.1 Supervised Control of Bacillus thuringiensis Fermentation Zhang et al. [68] have described an interactive study of the control of a batch fermentation to maximize the growth rate of Bacillus thuringiensis. Experimental data were obtained from a 2-L bioreactor operated at 26 to 3OoC, pH 6.0 to 8.0, and 2.5 to 30 9% dissolved oxygen (DO) concentration. The growth environment for each experiment was constant. Inputs to the ANN were: type of inoculum, the accumulated process time, temperature, pH, DO concentration, and optical density of the growth medium. Since the objective was to maximize the biomass growth rate, the optimal density was the only output variable. Thus, the feed forward ANN preferred by them had six input nodes and one output node. Zhang et al. tried different numbers of hidden
14.5 Applications to Disturbances and Process Faults
397
layers and varied the number of nodes in each layer. They complain that owing to the absence of a standard technique for choosing the network architecture, they had to decide the number of hidden layers and their sizes by trial-and-error. It may be recalled from section 14.3.6 that Glassey et al. [52] also had a similar problem: they, however, pointed to the possibility of genetic algorithms [19] being a useful guide to the initial design of an ANN and the choice of input variables. Masters [4] has also provided empirical rules to choose the starting number(s) of neurons for feedforward ANNs with one and two hidden layers. Thus, Zhang et al.’s design methodology is surprising because their ranges of the numbers of hidden neurons varied considerably; so even a well-established heuristic method might have speeded up the approach to the ‘right’ topology. It is also difficult to understand why a neural network with three hidden layers (6-30-15-8-1 topology) was tried since it has been proved [13] that two layers are sufficient to represent any continuous function. Not surprisingly, this ANN was inferior to two-layer networks, thus providing another example of spurious learning. The best configuration was 6-30-15 -1; a three-layer ANN overestimated the optical densities while other topologies underestimated them (Fig. 14-25). One limitation of all these ANN configurations was their inability to replicate the decline in optical density during the stationary phase. So Zhang and coworkers incorporated fuzzy logic by qualitatively grading the predicted optical density into 6 to 12 categories, corresponding to certain ranges of the actual values. Beyond nine grades there was no significant improvement. Since the outputs of the fuzzy ANN were normalized values within [0,1], they had to be converted back to actual optimal densities for comparison with experimental data. Zhang et al. [68] described three methods of doing this, and the normalized
network: 630.1 cai
network: 6.201 5-1
network:6-40.15-1
OO
1
2
3
4
s
6
7
8
9
10
Fermentation Time (h) Fig. 14-25. Approach to an optimal topology of a feed-forward ANN for biomass estimation and control in Bacillus thuringiensis fermentation. (Reprinted from [ 6 8 ] by permission of John Wiley & Sons, Inc. 0 1994.)
398
14 Neural Network Applications to Fermentation Processes
sum method followed the declining stationary phase behavior best of all. However, there was still the problem of optimizing the width of ranges of the qualitative grades so as to have the best overall representation of the optical density profile. This problem became apparent because a choice of grades which satisfied the stationary phase behavior was not good enough for the initial lag phase of fermentation. So, instead of having fixed widths for the qualitative grades, Zhang and associates adopted a triangular fuzzy membership function. Again, nine grades were sufficient but the ANN could now portray the entire range of optical density variation accurately. The trained ANN was implemented in a supervisory control system to maximize the optical density, which was an index of microbial growth, within the shortest possible time. The tuned control set points were the temperature and pH of the broth. At each sampling instant, the ANN predicted the next optical density and the possible set-points for the manipulated variables. The supervisory controller selected the set-point values that provided the highest increase in optical density over the latest value and then turned the slave controller. Figure 14-26 shows that a supervisory controller linked to a fuzzy neural network estimator provided a higher final optical density (at 1 h into the stationary phase) than a PID controller alone. Neural control also maintained a constant maximum optical density whereas in PID control the growth declined after the maximum was attained.
l2
g
t
101
PID.30OC.7.0~H (1-
PID, 26 'c,7.0 pH
42-
2
4
I
I
I
6
8
10
12
FermentationTme (h) Fig. 14-26. Comparison of biomass growth of Bacillus thuringiensis under PID control and under fuzzy neural control. (Reprinted from [68] by permission of John Wiley & Sons, Inc. 0 1994.)
14.5 Applications to Disturbances and Process Faults
399
14.5.2 Fuzzy Neural Control of Ethanol Production The preceding example shows how it is possible to combine a fuzzy ANN with a PID controller to have the benefits of both. This idea has been explored in more detail by Shimizu and coworkers. Their first study was of the production of ethanol by Saccharomyces cerevesiae in a fed-batch fermentation. The objective was to maximize the formation of biomass during the cultivation period (12 h). This is not a straightforward exercise because of the mechanism of the reaction, which is briefly explained here. Excess of the substrate glucose (which is supplied to the bioreactor) causes the Crabtree effect and increases ethanol formation, while in the absence of glucose the available ethanol is consumed as the substrate. If the substrate in the bioreactor is depleted, cell growth stops and consequently the DO concentration rises. In this situation, restoration of glucose supply again promotes cell growth and, when the biomass concentration is sufficiently high, there is again rapid consumption of glucose, leading to its shortage in the broth. This cycle of glucose consumption causes the DO concentration to oscillate. The approach to this problem is to regulate the supply of glucose in order to control the ethanol concentration. In the absence of an on-line glucose sensor suitable for long duration industrial application, Shi and Shimizu [69] utilized measurements of the DO and ethanol concentrations to develop a control policy for the glucose feed rate. According to the kinetics, the DO concentration either oscillates or does not, and the ethanol concentration either increases or decreases or stays constant with time. So the neural network was trained to recognize these patterns. Separate feedforward ANNs were employed for the two variables. Each network had one hidden layer with four neurons but the number of output neurons was obviously different - two for DO and three for ethanol. Neuron-fuzzy control was applied to the feed rate of glucose as follows. Let F(t) be the pre-determined time-dependent inflow rate based on the nominal operating condition. Since metabolic and environment variables change with time, F(t) has to be continually updated by an on-line correction, AF; this was done by the fuzzy control algorithm, based on the DO and ethanol concentrations. Shi and Shimizu [69] proposed three sets of fuzzy membership functions, one each for DO, ethanol, and AF. Depending on the choices from the first two sets, a particular membership function was chosen for A F according to prescribed rules. The set of functions for A F were bounded within Ifmin, fmax], and the strategy was to adjust the bounds according to the neural predictions based on the patterns of ethanol and DO concentrations. The results of conventional fuzzy control and neural fuzzy control are displayed in Figs. 14-27 and 14-28. For fuzzy control, fmin was set as -F andf,,, as F. Figure 14-27 shows significant oscillations in the DO concentration and the feed rate, which are suppressed by neural fuzzy control (Fig. 14 -28). Biomass concentration at the end of the batch was greater by 15 % in the latter method.
400
14 Neural Network Applications
to
Fermentation Processes
2 a
.c
2 .06 E
'c1
t
o
f2 Cultivation time (h)
Fig. 14-27. Conventional fuzzy control of baker's yeast fed-batch fermentation. (Reprinted from [69] with permission from Society for Fermentation & Bioengineering, Japan 0 1992.)
Shi and Shimizu, however, recognized that this improvement was achieved by suppressing ethanol formation. A practically useful fermentation will produce ethanol in addition to biomass. But this poses no additional difficulty for the ANN-linked controller because it merely requires altering the values of two binary parameters in the according to the same fuzzy rules as are applicable to cell variations offmi, andf,,, mass cultivation.
14.5 Applications to Disturbances and Process Faults
-s
401
.!2.
Y
Cultivation time (h)
Fig. 14-28. Neuro-fuzzy control of baker’s yeast fed-batch fermentation. (Reprinted from [69] with permission from Society for Fermentation & Bioengineering, Japan 0 1992.)
14.5.3 Fuzzy Neural Control of Recombinant E. coli In their study of S . cerevisiae, Shi and Shimizu [69] pointed out that the proposed fuzzy neural control could also be applied to the cultivation of E. coli because a similar relationship existed between glucose concentration and the changing patterns of acetate and DO concentrations. This thesis was tested with a recombinant E. coli generating 0-galactosidase in fed-batch operation [70]. A high productivity of recombinant P-galactosidase requires a good expression systems and high cell density. In fed-batch cultivation, specific growth rate is the most important variable as it affects both cell growth and protein expression. Similar to ethanol in the previous example, a lot of acetate is produced at high growth rates and this inhibits further growth. A lowering of the growth rate causes build-up of glucose in the bioreactor, which represses the expression of 0-galactosidase. These complexities make it difficult to devise a good control strategy. Nevertheless, fed-batch fermentation by conventional methods of control has been shown to be superior to continuous cultivation in terms of productivity and plasmid stability
402
14 Neural Network Applications to Fermentation Processes
[71]. Drawing on their own experience [69] and those of others [72], Ye et al. employed a fuzzy ANN to correct the experimental feed rate determined by a feedforward controller. The combined system was therefore of the feedforward-feedback tY Pe. The idea of using a fuzzy ANN to correct a pre-determined feed rate (of glucose) is similar to that employed in the cultivation of S. cerevisiae [69]. From the mathematical model of the fermentation, it can be shown that an exponential feed rate maximizes P-galactosidase production [7 11. This deduction is, however, based on the idealization of a constant yield factor for the cell mass, whereas in practice it does vary with time. So the calculated ‘optimal’ rate of glucose inflow should be corrected on-line according to: F(t) = F(t) (1+A) where F is the corrected flow rate, F the rate calculated from the model, and A is a compensation factor. The inputs to the ANN were the changes in specific growth rate, Ap, and pH (i.e. ApH) at each sampling point. These data were utilized by the ANN according to fuzzy rules such as: a) ZjAp is small and ApH is large, then the present feed rate of glucose should be increased. b) If Ap is large and ApH is small, then the feeding rate of glucose should be decreased. After training, the ANN utilized pH and specific growth rate data calculated from the on-line biomass concentration to compute the value of A required to modify the instantaneous feedrate determined by the PID controller. To determine the efficiency of incorporating a fuzzy ANN, Ye and associates studied the fed-batch fermentation both without and with the ANN. With feedforward control alone (Fig. 14-29), a consistently low glucose concentration did not result in a high cell mass concentration. A low residual concentration of glucose may avoid acetate formation, but it also results in starvation of the limited carbon source for cell growth. It may also be mentioned that the fermentation was run in batch mode for 6 h, at the end of which the initial glucose was almost exhausted and therefore glucose feed was started. After another 6 h, about 3 mM of IPTG (final concentration in the broth) was added to induce expression of 0-galactosidase. The inclusion of feedback compensation through the fuzzy ANN improved the final cell mass concentration almost three-fold (Fig. 14-30), but the final relative activity of P-galactosidase was lower and the unconverted glucose was more than before. Since most of the glucose supplied before the addition of IPTG had been utilized by the E. coli in both modes of control, Ye et al. [70] inferred that the system dynamics were different before and after induction. Therefore two fuzzy ANNs were employed in the control circuit, one for the pre-induction phase and one during the expression of P-galactosidase. This enhanced the protein activity considerably, the maximum and final values being 22000 IU/OD600 and 10000 IU/OD600, respectively. The final cell mass concentration reduced from 84 g 1-’ to 47 g I-’ but this
14.5 Applications to Disturbances and Process Faults
403
14000 12000
loo00
8ooo 6ooo 4ooo
2ooo 0
2
0
4
6
8
10
12
14
16
t
Time [h) Fig. 14 -29. Fed-batch cultivation of recombinant Escherichia coli by feed-forward control with exponential feeding policy. (Reprinted from [70] with permission from Society for Fermentation & Bioengineering, Japan 0 1994.)
Goo0
So00 4000
3000 2000
tI
1000 0 0
4
8
12
16
20
24
Time [h] Fig. 14 -30. Fed-batch cultivation of recombinant Escherichia coli with feed-forward control of exponential feed rate regulated by a fuzzy neural feed-back compensation. (Reprinted from [70] with permission from Society for Fermentation & Bioengineering, Japan 0 1994.)
need not be a disadvantage because the P-galactosidase concentration per gram cell mass was three-and-a-half times that with a single neural network. The fall in protein activity after attaining a peak is possibly due to decomposition by proteases [70]. If the fermentations were stopped at the peak values, the P-galactosidase activities would be 12 000 without any feedback ANN, 5600 with one fuzzy ANN and 22000 with two ANNs, thus again favoring fuzzy neural control.
404
14 Neural Network ADDliCatiOnS to Fermentation Processes
14.5.4 Diagnosis of Plasmid Instability Glassey et d ’ s [52] study of plasmid instability in a fed-batch bioreactor was briefly referred to in section 14.3.6. A significant and novel feature of their method was the use of three-dimensional plots of the time-dependent outputs from two hidden neurons. Fermentation data for a recombinant protein were simulated by an autoassociative ANN [2,4]. The results from three different experiments are plotted in Fig. 14-18. Even though commercial confidentiality precluded quantification of the axes and identification of the protein, a number of inferences can be drawn from the plots. According to Glassey et al., the distance and direction of the movements of the feature-tracks are indicators of plasmid-instability. The set of plots on the right ( X) pertain to fermentations in which vector instability was observed. The somewhat flat tracks denoted by (+) are for fermentations under normal conditions without instability problems. The two long tracks (0)on the left are for stable fermentations operated to investigate whether there was any long-term instability. The plots indicate that although the fermentations remained stable, they moved into a region of potential instability, implying that disturbances at this stage could easily destabilize the fermentation. The fourth set of short plots (+) moving vertically upward in Fig. 14-18 remain in the initial region; this was for batch fermentation and, although the operation showed no plasmid instability, it produced very little biomass and recombinant protein.
14.6 Hardware Implementation In addition to their functional advantages, ANN algorithms can be implemented with simple and standard hardware. Two configurations are shown in Figs. 14-31 and 14-32. The arrangement shown in Fig. 14-31 was used by Montague et al. [lo] for an industrial fermenter producing penicillin G. As mentioned before, this fed-batch operation may be controlled through the batch age, the C02 evolution rate, and the substrate feed rate. The batch age is, of course, known at any instant of time. The off-gas C02 concentration was measured by an infra-red analyzer. Together with air flow rate, this provided the C02 evolution rate. These data were passed to an IBM PS2/70 computer via a TCS signal processor which, according to the authors, provides a degree of local data conditioning. Substrate inflow rate was measured from a balance by linking it through an RS232 cable to the computer’s serial communication port. Note that the signal processor also has a two-way serial communication with the computer. It provided digitized time-series data to the neural network software, which estimated biomass and penicillin concentrations and computed the on-line adjustments required of the peristaltic pump supplying the substrate. The latter information was again communicated to the controller of the pump via the TCS signal processor.
405
14.6 Hardware Irnalementation
t
off-gas
Alr flow measurement
f * .. *
* .
Air
t
)c
I I
signal
COZ analyser
sample off -gas
I I
Serial link 4--+
L,, Serial link
Y
I Balance
-I
Fig. 14-31. Hardware implementation of neural network estimator for an industrial fed-batch fermentation to produce penicillin G. (Reprinted from [ l o ] with kind permission of Elsevier ScienceNL, Amsterdam, The Netherlands @ 1992.)
A similar experimental set-up was used by Ye et al. [70] for the high density cultivation of recombinant E. coli (Fig. 14-32). On-line data of the biomass concentration, temperature, DO concentration, and pH of the broth were transmitted via an analog-to-digital converter to an NEC 9801 personal computer. In the light of previous observations [2,33,37] that biomass is experimentally estimated from off-line samples and an ANN provides on-line estimates, it may be stated that Ye et al. [70] used the turbidity of the broth as an index of biomass concentration. It was stated in the description of Ye et d ’ s work (section 14.5.3) that the fuzzy ANN provided a feedback compensation to a feedforward PID controller. Thus, estimates of the changes in specific growth rate and protein concentration were combined with measurements of pH changes to manipulate the feed rates of glucose and ammonia solution. Note that the digital output of the computer is reconverted to an analog signal to enable it to act on the control valves. In both these studies, the essential equipment, besides the normal instrumentation of a bioreactor, are just a personal computer and a few signal processors and relays. It is the neural software that plays a pivotal role. As neural networks are applied to more demanding problems, greater speed and sophistication are required. This has led to the development of special microprocessor chips and even neural computers. Yamakawa [73] has described a rule chip and a
406
14 Neural Network Applications to Fermentation Processes
Air
4 i
j j
.............., d
..................... ..............................................................................
,;
i
i
i
U
Fig. 14-32. Experimental arrangement for the computer-controlled fed batch cultivation of recombinant Escherichia coli. (Reprinted from [70] with the permission of Society for Fermentation & Bioengineering, Japan 0 1994.)
defuzzifier chip for fuzzy logic controllers. The rule chip implements fuzzy rules. It has about 600 transistors and 800 resistors, an inference time of 1 ps, and can achieve a fuzzy inference from three input variables and one output variable. Defuzzification is needed to translate the fuzzy inferences into signals which can be implemented through a hardware device such as a solenoid valve. Typical defuzzification time is 5 ps. A fuzzy inference of more than three input variables can be achieved by connecting two or more rule chips in parallel. The number of defuzzifier chips needed to construct a fuzzy logic controller equals the number of output variables of the controller.
14.7 Concluding Remarks Artificial neural networks provide a viable method to simulate and predict the performance of fermentation processes under realistic conditions. When carried out on a practically useful scale, the complex interactions between kinetics and transport processes make it difficult to formulate from fundamental principles mathematical models which are simple but sufficiently accurate to permit on-line usage. ANNs do not have this problem because they are empirical structures that are designed according to the process data.
14.7 Concluding Remarks
407
However, ANNs are different from conventional black-box models. They have flexible architectures. they learn from the data and thus improve with usage, and they can function in situations which they have not experienced before. Their ability to function effectively without a fermentation model and even with data corrupted by noise and process faults makes ANNs especially useful in realistic problems. Time-dependent problems also are handled well by ANNs. While static fermentation problems have generally been described by feedforward neural networks, dynamic problems (e.g. fed-batch fermentation or continuous fermentation with a varying flow rate) require either an additional component (e.g. a delay function) or a different network topology. Auto-associative neural networks have been particularly useful for time-dependent fermentations. They provide a method to extract key features of a process under different operating conditions. These features serve as signatures of the process for different kinds of malfunctions. Another class of networks whose topology lends itself to unsteady-state problems is recurrent networks, of which Elman and Hopfield networks are the most common. Being software devices with ‘artificial intelligence’, ANNs provide rapid and frequent on-line estimates of variables which require slow off-line analysis or cannot be monitored reliably over the full duration of fermentation. They thus reduce process instrumentation while enhancing monitoring and control capability. Despite their demonstrated capabilities even for industrial-scale fermentations, certain difficulties remain in the applications of ANNs. For a given problem, there is yet no systematic method to choose the best type of network, the number of hidden layers, and the tranformation functions. Key features of network design are therefore based on experience and heuristics. Genetic algorithms offer useful guidance but in many real situations, where process disturbances make fermentation dynamics rather complex, designs based on heuristics have been as good as, and simpler than, those obtained through genetic algorithms. Development of a better theoretical framework for neural network design will enhance the applicability and efficiency of ANNs. Two immediate benefits of this are apparent. One is the development of general purpose algorithms for optimization of neural network configurations. Secondly, some problems (such as recombinant fermentations with strains containing temperature-sensitive plasmids) require more than one ANN in a hierarchy. Improvements in our theoretical understanding of ANNs will make it possible to automate the design of hierarchical networks. Looking further ahead, one can visualize that with improved software it will be possible to adapt neural networks dynamically to a process. This goes beyond the current practice of selecting a ‘good’ topology and updating the weights dynamically. An adaptive neural network will also alter its topology according to the process dynamics so that at every point in time the best ANN is in operation. Then artificial neural networks will have truly realized their potential as intelligent estimators and controllers [74].
408
14 Neural Network Applications to Fermentation Processes
References [l] Hammerstrom, D., IEEE Spectrum June 1993, 26-32. [2] Montague, G.A., Morris, A.J. Trends Biotechnol 1994, 12, 312-324. [3] Willis, M.J., Montague, G.A., DiMassimo, C., Tham, M.T., Morris, A.J., Autornatica 1992, 28, 1181-1187. [4] Masters, T., Practical Neural Network Recipes in C++. San Diego, CA: Academic Press, 1993. [5] Hush, D.R., Home, B.G., IEEE Signal Proc Mag January 1993, 8-39. [6] Bhat, N., McAvoy, T.J., Comput Chem Eng 1990, 14, 573-583. [7] Betenbaugh, M.J., Dhurjati, P., Biotechnol Bioeng 1990, 36, 124-134. [8] Mavrovouniotis, M.L., Chang, S., Comput Chem Eng 1992, 16, 347-369. [9] Baughman, D.R., Liu, Y.A., Ind Eng Chem Res 1994, 33, 2668-2687. [lo] Montague, G.A., Morris, A.J., Tham, M.T., J Biotechnol 1992, 25, 183-201. [ l l ] Patnaik, P.R., Appl Math Model 1994, 18, 620-627. [12] Patnaik, P.R., J Chem Technol Biotechnol 1994, 61, 337-342. [13] Cybenko, G., Math Control Signal Syst 1989, 2, 303-314. [14] Homik, K., Stinchcombe, M., White, H., Neural Networks 1990, 2, 359-366. [15] Linko, P., Zhu, Y., J Biotechnol 1991, 21, 253-270. [16] Hammerstrom, D., ZEEE Sprectrum July 1993, 46-53. [ 171 Caudill, M., Butler, C., Understanding Neural Networks: Computer Explorations, Vol. 1 and 2. Cambridge: MIT Press, 1994. [18] Aarts, E., van Laarhoven, P., Simulated Annealing: Theory and Practice. Newark: John Wiley, 1987. [19] Davis, L., Handbook of Genetic Algorithms. New York: van Nostrand Reinhold, 1991. [20] Tham, M.T., Moms, A.J., Montague, G.A., Chem Eng Res Des 1989, 67, 547-554. [21] Patnaik, P.R., Biotechnol Techniques 1995, 9, 691-696. [22] Widrow, B., Stems, S.D., Adaptive Signal Processing. New York: Prentice-Hall, 1985. [23] Rosenblatt, F., Principles of Neurodynamics. Washington D.C.: Spartan Press, 1961. [24] Minsky, M., Papert, S., Perceptrons. Cambridge: MIT Press, 1969. [25] Hofland, A., Montague, G.A., Morris, A.J., Proc Am Control Confr, Chicago, USA, 1992, pp. 480-484. [26] Hunt, K.J., Sbarabaro, D., Proc IEE Part D Confrol Theory Appl 1991, 138, 431-438. [27] Carpenter, G., Grossberg, S., Computer Vision Graphics Image Process 1987, 37, 54-115. [28] Kosko, B., IEEE Trans System Man Cybernetics 1988, 18, 1-6. [29] Kohonen, T., Selforganisation and Associative Memory. New York: Springer-Verlag, 1989. [30] Kosko, B., Neural Networks and Fuzzy Systems. Englewood Cliffs, N.J.: Prentice-Hall, 1992. [31] Tiller, V., Meyerhoff, J., Sziele, D., Schugerl, K., Bellgardt, K.-H., J Biotechnol 1994, 34, 119-131. [32] Pirt, S.J., Righelato, R.C., Appl Microbiol 1967, 15, 1284-1290. [33] DiMassimo, C., Montague, G.A., Willis, M.J., Tham, M.T., Moms, A.J., Comput Chem Eng 1992, 16, 283-291. [34] Montague, G.A., Morris, A.J., Wright, A.R., Aynsley, M., Ward, A.C., Proc IEE Part D Control Theory Appl 1986, 133, 240-246. [35] Montague, G.A., Morris, A.J., Wright, A.R., Aynsley, M., Ward, A.C., Can J Chem Eng 1986, 64, 567-580. [36] Wang, Z., Tham, M.T., Morris, A.J., Int J Control 1990, 56, 655-672. [37] DiMassimo, C., Lant, P.A., Sanders, A., Montague, G.A., Tham, M.T., Moms, A.J., J Chem Technol Biotechnol 1992, 53, 265-267. [38] Karim, M.N., Rivera, S.L., Adv Biochem Eng Biotechnol 1992, 46, 1-33. [39] Rivera, S.L., Karim, M.N., Proc Am Control Confr, San Diego, CA, 1990, p. 2144.
References
409
Erickson, L.E., in: Handbook of Anaerobic Fermentations: Erickson, L.E., Fung, D.Y. (Eds.). New York: Marcel Dekker, 1988; Ch. 5 , pp. 119-146. Linko, P., Aarts, R.J., Ann N YAcad Sci 1990, 613, 786-790. Linko, P., Zhu, Y.-H., Process Biochem 1992, 27, 275-283. Aarts, R.J., Suviranta, A., Rauman-Alto, P., Linko, P., Food Biotechnol 1990, 4, 301-305. Olsson, G., Anderson, B., Hellstrom, B.G., Holstrom, H., Reiniust, L.G., Vopatek, P., Water Sci Techno1 1989, 21, 1333-1345. Andrews, J.F., Water Res 1974, 8, 261-289. Curds, C.R., Water Res 1973, 7, 1269-1284. Tyagi, R.D., Du, Y.G., Sreeksrishnan, T.R., Villeneuve, J.P., Process Biochem 1993, 28, 259-267. Kumar, P.K.R., Maschke, H.-E., Friehs, K., Schugerl, K., Trends Biotechnol 1991, 9, 279284. Patnaik, P.R., Indian Chem Engr 1994, 36, 85-88. Nielsen, J., Villadsen, J., Chem Eng Sci 1992, 47, 4225-4270. Glassey, J., Montague, G.A., Ward, A.C., Kara, B.V., Biotechnol Bioeng 1994, 44, 397-405. Glassey, J., Montague, G.A., Ward, A.C., Kara, B.V., Process Biochern 1994, 29, 387-398. Bhat, N.V., McAvoy, T.J., Comput Chem Eng 1992, 16, 271-281. Schenker, B., Agarwal, M., Comput Chem Eng 1996, 20, 175-186. Kramer, M.A., Comput Chem Eng 1992, 16, 313-328. Patnaik, P.R., Chem Eng Commun 1995, 131, 125-140. Stephens, M.L., Lyberatos, G., Biotechnol Bioeng 1988, 31, 464-469. Ljung, L., System Identification Theory for the User. Englewood Cliffs, N.J.: Prentice-Hall, 1987. Thibault, J., van Breusegem, V., Cheruy, A., Biotechnol Bioeng 1990, 36, 1041-1048. van Breusegem, V., Thibault, J., Cheruy, A., Can J Chem Eng 1991, 69, 481-487. Patnaik, P.R., Biotechnol Techniques 1993, 7, 137-142. Mosrati, R., Nancid, N., Boudrant, J., Biotechnol Bioeng 1993, 41, 395-404. Mavrovouniotis, M.L., Chang, S . , Comput Chem Eng 1992, 16, 347-369. Devijver, P.A., Kittler, J., Pattern Recognition: A Statistical Approach. Englewood Cliffs, N.J.: Prentice-Hall, 1982. Hoskins, J.C., Kaliyur, K.M., Himmelblau, D.M., A I Ch E J 1991, 37, 137-141. Linko, P., Ann N YAcad Sci 1988, 542, 83-101. Modak, J., Lim, H., Tayeb, Y., Biotechnol Bioeng 1986, 28, 1396-1407. Zhang, Q., Reid, J.F., Litchfield, J.B., Ren, J., Chang, S.-W., Biotechnol Bioeng 1994, 43, 483-489. Shi, Z., Shimizu, K., J Ferment Bioeng 1992, 74, 39-45. Ye, K., Jin, S . , Shimizu, K., J Ferment Bioeng 1994, 77, 663-673. Yong, S.P., Kai, K., Iijima, S., Kobayashi, T., Biotechnol Bioeng 1992, 40, 686-696. Park, Y.S., Shi, Z.P., Shiba, S., Chantal, C., Iijima, S., Kobayashi, T., AppZ Microbiol Biotechnol 1993, 38, 649-655. Yamakawa, T., J Biotechnol 1992, 24, 1-32. Stephanopoulos, G., Han, C., Comput Chem Eng 1996, 20, 743-791.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
15 Advances in Modeling for Bioprocess Supervision and Control Andreas Liibbert and Rimvydas Simutis
15.1 Introduction When we are speaking about modeling for bioprocess supervision and control, we are primarily interested in improvements of industrial production processes. The main objective in this type of industrial activities is to improve the economics of the production processes in terms of the benefithost ratio [1,2]. There are several degrees of freedom for improving the process economics. The main technical aspects are process design and control. However, all other influences on the economics of production must never be disregarded. Particularly essential are the choices of appropriate raw materials and their prices. Here, we are primarily dealing with process supervision and control. The main aspects discussed are to find the best control strategy and to optimize real-time process control at the running plant. What, in this respect, is a process model good for? Process models are considered as vehicles for a systematic improvement the process performance. Activities aiming in process improvements can only be justified if, in any case where there are two process variants possible, one is able quickly and reliably to decide which of both is the better. Hence, we first need a clearly defined and simple-to-apply (!) performance criterion. There are different possibilities to improve a process. One approach is to choose the process control variables by trial-and-error. Then, however, success requires much intuition, and, in particular, good luck. Engineers prefer the systematic way, namely to derive relationships between the variables or parameters which can be manipulated and those which determine the process performance. These relationships are referred to as the performance relations or process models, since they are used to optimize the performance of the process numerically. A process model can be considered as a representation of the relevant process knowledge. In particular, it represents the knowledge about the ways in which the performance of the process can be influenced by the various process quantities that can be manipulated at the actual process by process engineers. In order to be really useful, these process representations (models) must be chosen in such a way that they can be processed in numerical optimization procedures, that are able to
412
15 Advances in Modeling f o r Bioprocess Supervision and Control
determine the optimal set point profiles of the control variables with respect to the performance criterion defined beforehand. The main attractiveness of this concept lies in avoiding unnecessary trials which without any question often appear with all trial-and-error approaches. By using good models one is thus able to avoid experiments that cost a lot but do not bring sufficient benefits. In this sense we understand the sentence attributed to Ludwig Boltzmann: ‘Nichts ist praktischer als eine gute Theorie’ (From the practical point of view, nothing is of more value than a good theoretical representation of the process, i.e. a good model). The immediate consequence of this objective of modeling is that the quality of the model must be measured by the advantage it brings for optimization and control at the concrete process under consideration. Since we are primarily interested in improving the process performance, the value of a model must be measured by the improvement in the process performance its utilization will bring. This definition is by no means obvious. In literature the model quality is usually defined by a modeling error, which is usually defined by the mean square deviation between the model output and the process data available to test the model. Both criteria however, can lead to different consequences as discussed by Simutis et al. ~31. The performance measures can be product quality, productivity, selectivity, etc., or any combinations of these quantities. Although the global task is simply to produce the particular product within given specification intervals of the parameters determining its quality at the lowest costs possible, the practical result is essentially dependent of the constraints by which the possible solutions are bound in real practice. These constraints might be very complex in practice. Hence, industrial optimization problem is to cope with the large number of constraints. Nevertheless, the main emphasis is first to obtaining the required product quality (benefit) and then minimizing the cost of obtaining it.
15.1.1 Conceptual Aspects of Practical Modeling Modeling must be considered a production process, the product being the model. The model must have some quality in order to be of real benefit. Hence, it must be ensured that the quality will be obtained during the production process. As in other production process, there are several conditions to be fulfilled before it makes sense to start the production:
- The personnel must be sufficiently skilled to produce a practically useful, i.e. valuable model. Modeling needs investments in computer facilities, i.e. hardware and software expenditures. - Modeling needs time. -
The process will be rated by the benefit-cost ratio and, finally, by the acceptance of the model in industry.
15.1 Introduction
4 13
Since the value of the product model - as we understand it here - is purely outputoriented, i.e. judged by the advantage it brings for process improvements, there should be no a priori preference to any representation of the input/output relationships. In this respect, a remark seems worthwhile concerning mathematical process representations. They appear to be highly overrated, which is understandable, since they are usually based on physical reasoning of a mechanistic understanding of the process. However, although understanding the process usually helps in process optimization, it is not always necessary. Experience showed that the large part of the available process knowledge cannot be formulated by physically reasoned mathematical models. In short, it is necessary to know how to improve a process, but it is not always necessary to exactly know the mechanisms. Engineers already responded to that situations with correlations, by which they quantified their experimentally obtained data, and with rules-of-thumb, by which they formulate their heuristic experiences linguistically. It is important not to disregard this information or knowledge, but to provide advanced techniques to activate an as large part as possible for improving the process performance in terms of its benefit-cost ratio. An important boundary condition influencing the true advantage of such an approach in industrial practice is: The different methods must be combined with each other and thus they must match concerning their levels of sophistication, accuracy, cost (expenditure), and last but not least simplicity with respect to the ease of its application by bioprocess engineers. In particular the accuracy of the models must match with the precision by which the optimization goals are formulated and performance of the optimization algorithms applied. A well-balanced choice of techniques is required. There is not much room for philosophical arguments concerning the advantage of classical mathematical process models - the benefit-cost ratio is the iron hand that rules modeling in industrial practice.
15.1.2 Current State of Process Modeling in Industry Currently, advanced model-supported process supervision and control techniques are not widely applied in industrial production processes. The main items, which prevented or at least limited the application of model-supported industrial process supervision, optimization, an control techniques are: Problem definition often has not been made in a sufficiently clear, quantitative way; thus, the task to be solved with the aid of the model was not defined well enough. - Modeling has not sufficiently been performed in a problem oriented way; hence, the models were not optimally adapted to the problems to be solved. - Missing programming tools that can be used during the entire procedure without changing computers and software environments reduced the efficiency of the process engineers in modeling. The available tools were too complicated to use and not really compatible with each other. -
414 -
15 Advances in Modeling f o r Bioprocess Supervision and Control
Models are usually too complicated and do require too much information that is not readily available.
In this article, the main problems are discussed wing some examples. The aim is to show what is new and has some perspective in model-supported process supervision, control, and optimization. It will be shown that more effective results in bioprocess improvement can be achieved by making use of modern modeling and software tools.
15.2 Process Models for Typical Applications Four classes of problems typical for industrial biotechnological processes are discussed here, in order to illustrate the possible advantage of new modeling technique:
1. 2. 3. 4.
Modeling Modeling Modeling Modeling
for state estimation. for process fault analysis. for open loop control (off-line optimization). for closed loop control (on-line control).
The accompanied examples chosen have been taken from real cultivation processes.
15.2.1 Practical Constraints on the Modeling Procedure Since we consider models as tools for solving concrete questions, in modeling an obvious trade-off must be made between constructing general purpose tools and special-purpose tools. Generally, it can be said that the more demanding the targets in process improvements are, the more specialized the models, i.e. the tools to be used to reach the goal, must be. However, in practice it transpires that we do not have enough knowledge about biochemical production processes as to generate very general models that, at the same time, could lead to sufficient benefits in a particular production process. Hence, we must mostly restrict ourselves to models that are particularly tailored to the concrete special tasks to be solved. Available models must be adapted to the particular questions to be answered. Obviously, only limited expenditure can be justified in modeling. The required expenditures can be divided into:
1. Manpower required to solve the modeling task - Time, the personnel can spend on the modeling project - Skill of the personnel employed on the project - Number of people working on the task
15.2 Process Models for Typical Applications
4 15
2 . Computer resources available - Computing power - Computer time - Software tools 3. Data situation - Data available to model the process - Data required to validate the models - Data that can be generated on the plant during the modeling project
In practice, manpower resources are usually most restricting. Hence, there is a strong need to improve the software tools which are considered to enhance the effectiveness of the process engineers, thus reducing the demands in manpower. The immediate consequence is that software tools become a very important issue. Moreover, powerful tools allow the personnel to more efficiently make use of already available data. Often, some costly experiments can be avoided in this way. Figure 15-1 summarizes the requirements. Models of technical processes are not as completely closed as models in physical science. They are more or less approximating reality only. Consequently, validation is an essential issue. This requires a lot of reliable data, that allow the process to be illustrated in all relevant situations and from different points-of-view. Without a minimum of data, there is no sense in starting any modeling task at all.
Analysia of Knowlcdge Sources and Resources
Definition of Problem
Choice of
Methods and Software
Modeling for Faults Analysis
Literature Data
Math4O&dS Expert System
Knowledge of Experts
Modeling for State Estimation
Experimental Data
--.
Modeling for Openloop Control
Development Time
I -
Modeling for C l o d Loop conhl
ManPower Budget
I
Dedgn and Implamentstion of MDdclr and Ngorithms
Fuuy Expert Systems
--
ANN Hybrid
Front-End Software systems
-Modeling Nlalpb Scilab, W-Wave 02, HybNet
I
ModeCBasedProcess Improvement
Fig. 15-1.General schema for process improvement using process modeling aproach.
41 6
15 Advances in Modeling .for Bioprocess Supervision and Control
When there are no data to validate the model, then modeling is merely like playing with ideas. The importance of reliable data, thus, cannot be stressed enough.
15.2.2 Modeling for State Estimation Tasks to be Solved in State Estimation
In state estimation, we are faced with the problem of monitoring in real-time the actual state of the running process. Without an information of the state variables it is impossible reliably to supervise the process. Several papers have been published showing the state-of-the-art in model-supported bioprocess monitoring (e.g. [4-71). Preferentially, measurement devices are taken to determine the values of the quantities characterizing the state of the process. However, in biotechnological production processes, the situation is more complicated, since most of the key quantities cannot be measured online, at least not with justifiable expenditures and/or sufficiently short response times with respect to changes in the process. Moreover, when they can be determined, then this is often not at the desired accuracy and with the desired sampling frequency. Hence they must be measured indirectly, i.e. they must be determined from easily measurable quanities by means of expressions relating the corresponding quantities with each other. Even if they can be measured it often makes sense to use the indirect methods additionally in order to reduce the estimation error. This is possible by extracting the additional information about the particular process quantity hidden in all the different measured signals that are related to the quantity under consideration. Obviously, the models relating the target and the measured quantities, depend on the measurement signals available in the concrete situation. Consequently, it is not possible to use one and the same model after the set of measurement quantities was exchanged. Thus, the models to be used in state estimation depend on the particular problem at hand. General process models are of minor practical advantage in process supervision. It should be stressed, however, that state estimation indispensably requires sufficiently informative measurement signals from the actual process, signals that carry enough information to determine the process variable under consideration. Needless to say, it is not possible to extract more information than is hidden in the signals. Hence, the quality of the measurement signals ultimately determines the amount of information which can be obtained about the actual process variable or the process state. In addition, the relationships between the measured and the desired quantities must perform well enough to be of sufficient value with respect to improving accuracy of the estimated state vector. Not until a careful validation of the model has been performed, can any serious decision in the sense of process control be based on the state estimations and predictions obtained with this model.
15.2 Process Models for Typical Applications
417
An excellent overview on classical and modern modeling methods for state estimation can be found in the review of Stephanopoulos et al. [8]. Here, our aim is to bring to attention some of the new developments in process modeling which allow to make state estimation more efficient.
Extended Classical Techniques Macroscopic balance equations If the process is well understood and the cultivation medium is well defined, then there is a good chance to solve the state estimation problem using models based on macroscopic mass balances [9-111. However, it proved to be of advantage to extend the classical procedure of using mass balances by also using available electrons or reductance balances [9]. In many cases, it is sufficient to restrict the efforts of constructing a real-time process model on mass, carbon, nitrogen, and available electron balances. The available electrons balance is merely a linear transformation of a simple material balance. Its advantage finally results from the fact that the hydrogen balance cannot be closed in practice. The main reason is that the rate by which water is produced in biochemical pathways, cannot be measured accurately enough, as water is generally present in large excess. Theoretically, only a few experiments are required to obtain the measurement data necessary to calculate non-measurable quantities from macroscopic models. However, a considerable problem faced in practice is that we are dealing with noisy data, and this requires additional measurements to be appropriately utilized. In the literature, some trends dealing with noisy data can be recognized. The first trend is to process additional measurement data in a least square approach utilizing an appropriately weighted combination of balances. Chattaway et al. [ 121 presented an excellent example for this case. A second trend to be mentioned is the simultaneous use of macroscopic balancing and simple kinetic models as demonstrated by Schalien et al. [13]. The general message from the mentioned examples on applications of macroscopic balances in practice is, that it is worthwhile to remember that the classical macroscopic reaction equations are the basic relations reflecting the net conversion of the substrates, which are converted into the products. Only those components need to be considered, that are consumed or produced in significant amounts. It is generally recommended to use this stoichiometric information first to interrelate the rate equations for the different key components (state variables), and then to compensate for the remaining uncertainty by simple kinetic models. As stated initially, the technique is more or less restricted to production systems in which well-defined substrates are used. Particularly in today’s industrial biotechnological practice, this is not merely a theoretical case, since an increasing number of companies are trying to produce high-value products on synthetic media, in order to become independent from the uncertainties in quality and cost of the natural substrates they used formerly. Usually, with this measure they obtain a higher repro-
418
15 Advances in Modeling .for Bioprocess Supervision and Control
ducibility as possible using natural substrates like molasses, corn steep liquor, etc. This at least partly compensates for the fact that defined media are most often more expensive. In all the cases where synthetic media are used, there is a good chance of finding macroscopic reaction equations and using them in state estimation problems. Unfortunately, for complex media and multiple products, the application of these methods is complicated and, thus, the construction of significantly reliable stoichiometric models requires major efforts.
Extended Kalman Filters When the time available is not limited and when a sufficiently accurate mathematical process model is already available or when there are also well-educated process specialists, then the classical Extended Kalman Filter (EKF) technique can be applied [8,14]. Due to the considerable number of restrictions, this technique has not widely been used in industry. Since the performance of EKF is sensitively dependent of inaccuracies in the process model, high costs for design and implementation as well as for maintenance are to be expected in most of their applications. Adapting the EKF technique to the everchanging process conditions in a production plant thus often requires more time for tuning work than is available up to the next process change. In order to improve the accuracy of the models for an EFK, and the ease by which they can be adapted to changing processes, it is of advantage to use hybrid process models. These activate a larger part of the available process knowledge and can be adapted to new conditions more easily than classical mathematical models. However, they require more process data. As an example, the state estimation in industrial beer fermentation can be considered where an extended Kalman filtering was applied at an industrial production reactor [ 151. In this example, considerable advantages were obtained with a distributed modeling concept. By distributed modeling, a breakdown of the comprehensive process model in several modules is meant. In the particular example, each module corresponds to particular fermentation phase. There are two essential advantages of this approach. 1. That the modules can more closely related to the process dynamics in the corresponding phases and are thus more accurate models. 2. That the models for the individual process phases can be kept smaller and more transparent than the comprehensive model. The immediate advantage is, that it becomes significantly easier to identify the parameters of these partial models.
Obviously this advantage must be paid for by the expense of deciding which part of the model to use in an actual situation. In this particular example, this decision can be made using heuristic knowledge about the process. In reference [15], fuzzy reasoning was used to bring the different model modules into action. The example showed, that the state estimation using this approach is more accurate than using the usual EKF technique.
15.2 Process Models f o r Typical Applications
41 9
Advanced Techniques Combination of balance equations with artificial neural networks If the process is already successfully running and we thus have many data but not the time to utilize them in a thorough investigation of process details, then some fine tuning can be obtained in the following way. Easy-to-use black-box estimators can be taken to represent the kinetic relationships between the variables characterizing the environment of the organisms and the bioprocess reactions rates. As such, ANNs turned out to be very efficient representations of biotechnological process kinetics. They can provide more accurate estimates of the specific reaction rates than classical rate expressions, and may replace the classical correlations like the Monod model in the mass balance equations, which can be set up on first principles. Among recently published literature, several interesting examples demonstrate that such hybrid combinations of artificial neural networks and basic balance equations lead to considerable advantages [16-181. When it comes to concrete application, the artificial neural network component in this kind of hybrid model requires some attention concerning the appropriate training procedures. The type of network most often used in this connection is the recurrent artificial neural network. This powerful technique can also be found in commercial software packages. Its decisive characteristics are that it requires input data taken with a constant time increment. Where such data are available the recurrent neural network algorithms perform quite well. However, in biotechnology, where we must deal with many off-line or quasi-off-line data, such regular data records must be generated using interpolation techniques. Unfortunately, with such interpolation we not only lose accuracy but also may incorporate some artificial disturbances. Furthermore, in biotechnology we are most often interested in reaction rates, as they are known to be the key quantities allowing judgment of the behavior of conversion processes. Rates must be determined from measured or estimated concentrations by differentiation. In this respect, noisy signals lead to even more distorted estimates of the specific rates. All these problems can be reduced significantly with the sensitivity equation technique [17]. Using this approach the identification of even very complex hybrid models can be performed very efficiently using the classical back backpropagation algorithms [ 191. Since these modeling and identification procedures are relative simple, they additionally simplify the quick adaptation of the model to changing process conditions. This is why such techniques are so attractive for industrial applications. Figure 15-2 depicts the structure of such a state estimator based on a combination of an ANN with a set of balance equations. It was constructed to estimate biomass and glucose concentration in a fed batch S. cerevisiae cultivation process. Figure 15-3 shows a typical result of such a state estimation. Here, and in all other applications of modeling - in particular in the cases where artificial neural networks are involved - we must validate the models appropriately before applying them in industrial practice. Although all models must be validated,
420
15 Advances in Modeling for Bioprocess Supervision and Control
1
I
OUR CPR
EPR Base
Volume
Glucose Feed
Fig. 15-2. Structure of the ANN-based estimator used for state estimation in fed-batch S. cerevisiae cultivation process.
the problem is of particular importance with all non-mechanistic process models like those represented by artificial nerual networks. The main reason for this need is that not every set of input/output measurement data carries enough information about the process behavior. Consequently, in order to ensure that the model works correctly, we must test it with independent data from process situations which might be of importance to the production process. Of course, the data must not already have been used during the model parameter identification procedure.
Time [h] Fig. 15-3.Typical state estimation results obtained during the fed-batch S. cerevisiae cultivation process.
15.2 Process Models f o r Typical Applications
421
As an example, consider the representation of the time-development during the beer fermentation of diacetyl, the key component determining the taste of beer. With the data available from many fermentation runs, a neural network was trained. Since the networks are very flexible representations, the result shows quite a good agreement with the data (Fig. 15-4(a)). However, during the validation procedure, A
Measureulent
-EstMion /ANN
0
a)
P
43
R
96
Tme [h]
Fig. 15 -4. The quality of the diacetyl-estimator during the industrial beer fermentation process. (a) For one of the training sets; (b) for one of the validation sets.
422
15 Advances in Modeling for Bioprocess Supervision and Control
the same neural network applied to additional data records from the same bioreactor showed a quite unsatisfactory result (Fig. 15-4(b)). The process variables used as input variables for the network obviously did not carry enough information about diacetyl production and degradation. If only the diacetyl degradation phase were modeled using the same data, it would be possible to obtain satisfactorily validated results. The same problem also appears with all other black-box approaches. Hence, it must clearly be stated that it does not suffice to perform a fit of a given model to some available data. The representation obtained must be validated. The recommended procedure is known by the name cross-validation [ 19,201.
Fuzzy rule systems When the production process is very complex with many non-linear, interacting components and there are available highly skilled process engineers who are able to describe the process sufficiently accurate by means of rules-of-thumb or other linguistic expressions, then fuzzy modeling (see Appendix) can be applied successfully for process supervision and control [21-231. Fuzzy-set based modeling is particularly effective, since [24]:
- There are well-performing possibilities for adapting these gray box models to available process data records. As shown in the Appendix, it is possible to convert the fuzzy rule base representing the process model into a neurofuzzy network, that can be trained on the process data with standard error backpropagation procedures. Hence, it becomes possible, not only to store the heuristic knowledge of the process engineers, but also to improve it using the process data. - The system parameters in this process representation are physical quantities. Hence, it becomes much easier for human process experts to examine the values obtained after such tuning procedures and to convince themselves about whether or not the tuning led to reasonable and meaningful results. - Such neurofuzzy networks can as easily be combined with classical balance equations as the simple neural networks descibed before. They also can be trained with the sensitivity equation technique.
Hybrid process models
If there is more time and the demands are higher, then a hybrid model should be constructed. By hybrid modeling, we mean modeling using combinations of different representations of knowledge and information. The components could be differential equations, artificial neural networks, fuzzy expert systems, different types of engineering correlations, etc. They can be used to activate knowledge from different areas such as mechanistic knowledge in terms of physical or biochemical theories or heuristic knowledge about the dynamic behavior of the process obtained from experience with operating the process. Since the pieces of knowledge from these sources are usually complementary to each other, a combination of them usually
15.2 Process Models for Typical Applications
423
means utilizing more a priori knowledge or constructing a more comprehensive process model. Figure 15-5 depicts some efficient combinations for constructing hybrid models.
,
I
i c)
Fig. 15-5. Constructions of hybrid models which combine knowledge from different sources of information.
424
15 Advances in Modeling for Bioprocess Supervision and Control
Examples for such hybrid models, successfully being used in bioengineering include:
1. Combinations of ANNs, classical kinetic expresssions and systems of ordinary differential equations. The components are weighted using the extrapolation measure (EM) concept [24] 2. Combinations of a classical kinetic expressions with radial basis function ANNs and systems of ordinary differential equations, where the ANNs acts as an socalled error compensator, correcting the errors of the classical model [18] 3. Combinations using so-called gating networks and systems of ordinary differential equations [19]. In this case an additional neural network dynamically determines the statistical weights which are necessary to determine the contributions of the different model alternatives at the final estimate of the kinetic wates. This network optimizes the rates in such a way that the performance function chosen becomes a maximum. All these variants of hybrid models can be trained with the sensitivity equation approach described by Schubert et al. [17]. Figure 15-6 shows the profile of acetate concentration values estimated during a production of recombinant proteins in an E. coli fermentation using a hybrid model constructed according the combination scheme 2 above.
15.2.3 Modeling for Process Fault Analysis Problem Statement Process quality assurance is a key issue on the agenda of every process engineer responsible for a production plant. Quality assurance means avoidance of process faults. Process faults have been divided into two categories (e.g. Royce [2]):
1. Equipment faults, e.g., - sensor faults, e.g., problems with pH and p02 probes or problems with pipe lines, e.g. blockage of the off-gas filter or cooling problems, problem with the heat exchanger system; and 2. Process faults, e.g., - contamination, by means of infections, or genetic drift, e.g., plasmid stability problems, or problems with the feedstock. The most important faults are those which tend to destabilize the process. The major problem is to avoid severe process faults by detecting them in their early beginning, i.e. so early that there is sufficient time to counteract before performance losses become unavoidable. Combating equipment faults, e.g. the fault of the signal from a pH electrode, is of importance since faulty signals would not only lead to a wrong estimate of the pro-
15.2 Process Models for Typical Applications
425
-Ferm. Nr. 1 8
Nr. 2
7
-
7
-I
6
5
Q)
Y
c a.
4
8 $ 3
a
2 1
0
0 a)
2
4
6
8
10
Time [h]
Fig. 15-6. Estimation of acetate concentration in recombinant E. coli cultivation process, using Monod-type kinetic model and a radial basis function ANN compensator. (a) Results using Monod-type kinetics only; (b) results using hybrid model.
426
15 Advances in Modeling f o r Bioprocess Supervision and Control
cess state but also to wrong control actions in the pH control and thus to a destabilization of the process. Obviously a failure of a physical device cannot be removed by a software action but its influence on the process can be minimized.
Recently Derived New Methods Traditionally, fault analysis is performed by state estimation and comparison of the current state with the a priori assumptions of the trajectories of the state variables. However, recently, some conceptually simpler additional techniques such as rulebased expert systems, principal component analysis, and autoassociative neural networks [2,21,25-271 have been shown to be more effective. As compared with the conventionally used state estimation techniques, the emphasis of the new methods in the field of fault analysis, is slightly different because of the particular problem statement.
Rule-based (expert-)systems An interesting approach developed in recent years is to make use of the heuristic knowledge of the process engineers and to formulate it linguistically by rules-ofthumb about how to detect and counteract in typical cases of process disturbances. Such rules can be processed using the techniques developed under the headline knowledge-based software systems or expert systems. Unfortunately, the cost of development and testing such rule-based systems is considerable. A particular problem with these systems is that it is extremely difficult to generate a sound rule base without conflicting rules, covering at least the essential aspects of the entire process. Experience showed that it is extremely difficult even to highly skilled process engineers to formulate their experience clearly enough. Thus, the number of process fault analysis systems based on such expert systems which already reached an application in industrial practice is low ~71.
Models based on fuzzy sets and fuzzy rules Fuzzy set theory can easily cope with the problem of handling conflicts in rule-based systems. Hence, the classical expert systems are being replaced by fuzzy expert sytems, which are much easier to develop, maintain, and update. The way that fuzzy expert systems work is shown in the Appendix. Recently, developments of techniques using fuzzy expert systems have been merged with developments of techniques in the field of artificial neural networks. The results are referred to as neurofuzzy networks, which are also briefly described in the Appendix. These techniques combine the transparent way of representing a priori knowledge by means of rules of thumb with the tuning and learning features of artificial neural networks [22,23].
15.2 Process Models for Typical Applications
427
Principal component analysis Process and sensors fault detection requires efficient algorithms which can handle large amounts of process data, that might be corrupted by measurement erro’rs. Strong correlations between the various process signals lead to further complications in the process fault analysis. In order to enhance transparency of the data with respect to process supervision, it is desirable to reduce the large number of measurement signals to a small number of highly informative signals such that the operator can keep track of them more easily. Since it became widely recognized that an appropriate reduction in dimension would increase transparency of multivariable data processing problems considerably, many activities have been invested to derive efficient techniques. In mathematical terms, the essential task is to transform the process variable space into a space of lower dimension without losing information hidden in the data. Simply neglecting some of the variables does not work generally. Standard statistical techniques for dimensional compression of process data is the principal component analysis (PCA) [ 7 ] . It performs a linear mapping from the original (m-dimensional) data space to a lower (f-dimensional) new space, i.e. f<m, such that all variables in the new space are independent of each other and contain the main body of information carried by the original data block. There are two variants of utilizing the features, as the variables in the new space are referred to as: 1. direct use of these quantities by the operator as new characteristic quantities of the process 2. forming correlations between these artifical variables by means of (2a) ‘Principal Component Regression (PRC)’ [29] and (2b) ‘Partial Least Square (PLS)’ techniques [30,31]. In principle, this is what engineers often did when they tried to correlate functions of the different process variables, which they often referred to as groups, instead of the original variables directly. It is well known that by a lucky choice of such functions, the correlation can often be made much simpler. With the partial least square techniques the intuition of the engineers is replaced by a systematic procedure. As principal component analysis techniques can easily be implemented and used in practice, this technique has been discussed in the literature as an effective tool for process fault analysis [25]. Unfortunately, this technique is restricted to linear transformations. The more interesting non-linear case will be considered in more detail next.
Autoassociative artificial neural networks Artificial neural networks are known to be powerful tools in representing the nonlinear behavior of a complex process by mapping the state vectors of its multidimensional input variable space onto the process output variable space.
428
15 Advances in Modeling for Bioprocess Supervision and Control
An Autoassociative Artificial Neural Network (aANN) is a special type of neural network that performs a non-linear coordinate transformation of the input variable space into some artificial variable-space of lower dimension with the aim of maintaining - in the transformed data - the features of the process one is interested in [lo]. The aANN technique is described in a more detail in the Appendix. The general idea of utilizing these artificial neural nets is the same as that behind using the PCA technique, the difference being that aANNs allow non-linear features to extract from the process data. Depending on the original set of input variables, the reduction of the dimension of the new data space can be made larger or smaller. The reduction possible without a significant loss of the performance of the network in representing the process behavior, essentially signals the redundancy of the original data. Thus, it depends on the set of measurement variables available. A structure of an autoassociative neural net is shown in Fig. 15-7. The forward transformation to the set of features is shown, together with the inverse transformation leading to the original set of variables. The layer containing the features is called ‘bottleneck layer’ in the figure, since it represents a bottleneck in the signal transmission between the set of original variables and its reconstructed version. The visualization of the features’ space can be very valuable in the process interpretation. When one feature is plotted against another, one often gets paths in that feature subspace of the process. The paths characterize the process; in particular, there are characteristic paths for good and characteristic paths for bad process realizations. One can distinguish between both visually, but also quantitatively by some distance measures.
Y1‘
Y2
... Ym’
INPUT LAYER
MAPPING LAYER
BOlTLENECK LAYER
DEMAPPING LAYER
OUTPUT LAYER
Fig. 15-7. Structure of an autoassociative neural network.
15.2 Process Models for Typical Applications
429
The other possibility for using the aANN in process supervision and control is to provide a relationship between the features and the performance measure of the process under consideration. This can be achieved by an additional artificial neural network or by a classical regression methods.
Examples for Fault Analysis Sensor faults Associative networks (cf. Appendix) are simple-to-use procedures that can be directly used in two kinds of problems associated with process sensor supervision and fault analysis. They can firstly reduce the noise in real signals and they can secondly often be used to reconstruct - to a sufficient degree - a missing sensor signal in case of a sensor breakdown. In this respect, the term ‘sufficient’ means that the reconstructed signal is so accurate that it is not necessary to stop the fermentation. As a first example of an autoassociative neural network we discuss a noise filter based on an autoassociative neural network. Such a filter is of considerable value when biochemical conversion rates are to be estimated from original process data. In such tasks, the main source of errors in numerical computations is signal noise. The noise-reduced reconstruction of the input data conceptually is the most simple application of an associative artificial neural network. Since such a network can only store and thus reconstruct the systematic information about the behavior of the corresponding process, but not the randomly appearing noise superimposed on the input signal, the reconstructed signal is essentially the input signal freed from the randomly appearing noise. Hence associative neural networks can be used for signalfiltering in a simple way without using much knowledge about signal and control theory. As an example, consider the signal obtained from a fluorescence probe which was used indirectly to estimate the biomass and the specific growth rate in a S. cerevisiae cultivation. A feedforward autoassociative artificial neural network with five layers (cf., Appendix) and the node sequence { 3 , 4, 2, 4, 3) has been used in the example. The three input quantities are: the signal from the (Ingold) fluorescence probe, the base consumption, and the cumulative oxygen uptake rate (OUR). The associative neural network was trained on data records from three fermentations and then used for the filtering the fluctuating fluorescence signal. Typical examples are shown in Fig. 15-8 for the raw data and the smoothed signal of the culture fluorescence as well as for the corresponding estimate for the specific growth rate p. From time to time during the fermentation process, some sensors signal may become unavailable or distorted due to sensor failures. In particular, when the sensor is involved in a process control chain, such failures can provide significant problems. In these cases, a reconstruction of the missing signal would help to put forward the process until the disturbed component has been repaired or replaced. Autoassociative networks can be used to replace unreliable or missing process signals, once they have been identified as such. The idea behind the estimation of the missing sensor signal with well-trained and validated autoassociative neural net-
430
15 Advances in Modeling for Bioprocess Supervision and Control
-Florescence
-P 2
7000 6ooo
-.-g =
1
3
g m 8
-
0 7
5
+cQo
A
v)
Y
i i gm
-1
loo0
0
6
4
8
10
12
14
-2
16
Time [h] 7000
-a>l "m E
-
- Florescence
-
aA"
6ooo5ooo-
I
8c - a,
M E0
=I
ii
-3ooo-
0
4
I
I
I
I
I
6
8
10
12
14
0,o
16
Time [h] Fig. 15-8. Use of an aANN for the filtenng of fluorescence signal in S. cerevisiae cultivation.
works is basically the same as with indirect measurements discussed above. The information essentially must be drawn from the dynamics by which all other variables are changing with time. As the autoassociative network can only reconstruct the input data vector Y when its components consistently describe a accessible input state, a missing or wrong component of the input vector, immediately leads to a deviation A between measured input vector Y and reconstructed vector Y', which is considerably higher than the mean error
15.2 Process Models for Typical Applications
A = mean: [Y - Y'],
431
expected for the aANN mapping.
{YI An efficient alternative is a fuzzy rule system that monitors the data in the sense of a watch-dog. In the case this rule system detects a sensor failure, it gives a signal to an autoassociative network to replace the measurement value under consideration. In order to compensate for that deficiency one can proceed in the following way. Assume that the single component Ymissof the input vector is missing, i.e. a wrong value appears in the input vector Y. Then it is possible to search for the best estimate of its value YmisseStby systematically varying it within the interval [Ymissmin,Ymi,smax] of its normal variations until the minimum error A is approached closely. This 1-D search within the known interval can be solved very rapidly, since only forward evaluations of the trained autoassociative network are required. Hence, this technique of replacing a wrong or missing signal is conceptionally simple, does not require much computing power, and can easily be incorporated into a process control software. Figure 15-9 shows an example for the estimation of a dissolved oxygen concentration signal during a laboratory-scale E. coli cultivation. In this example, four input signals, p02, OUR, Airflow, and Stirrer Speed were measured. During two time intervals, the measured p02 signal was assumed to be distorted and an artificial one was determined with an aANN containing two features (the layer structure was [4,5 , 2 , 5 , 4);cf. Appendix). Since the pO2 signal was only assumed to be missing, in Fig. 15-9 we are now able to compare the measured signal with the estimated one.
------. _ - - - - - PO, - real -PO, - estimated with an
90 -
-
autoassociative ANN
70 80
-E
60 50
-
-
.
0" 40
-
Q
-
30
-
20
-
10
-
0
\ d
.,
,,
,+I, ,,d#';--h.
--
knitation of sensor fauits
I
'
1
'
"
'
'
)
'
I
'
I
'
I
'.) '
I
432
15 Advances in Modeling for Bioprocess Supervision and Control
Process faults In the cases of animal or plant cells the processes must be monitored over very long cultivation times. In particular in these cases it is of significant advantage to recognize process faults as early as possible either to get a chance to correct the process by changing control variables, or to stop the process in order to save time and substrate costs. Associative neural networks can help in such situations since, generally, it is easier to keep track of a dynamic system if its behavior can be represented in a lower-dimensional representation. If the associative neural network has been trained on data from production runs, one observes typical narrow paths of the process in the feature space for correctly running processes. Experiences with these feature trajec-
Fig. 15-10.Application of an aANN to determine process faults during a recombinant CHO cultivation in an industrial erythropoietin production process.
15.2 Process Models for Typical Applications
433
tories show that they leave the beaten tracks produced by data from correctly running processes very sensitively upon process failures. A typical example is shown in Fig. 15-10 where a feature trajectory is depicted for a protein production process with recombinant CHO cells. This result was obtained with an autoassociative neural network with a node vector {7, 8, 2, 8, 7}, trained on six data records. The input vector was composed of the online measured signals of CPR, OUR, temperature, glucose, glutamine, pH, and fermentation time. By means of cross-validation techniques it was shown that two bottleneck nodes are sufficient to capture the principal dynamics of the process. The two signals appearing as feature signals of the nodes in the bottleneck layer are plotted against each other resulting in the trajectory depicted in Fig. 15-10. In one of the fermentations, an extreme low concentration of recombinant protein was observed. This led to a significantly different trajectory. Figure 15-10 thus indicates that the autoassociative neural network has been able to extract important features from the process data and that these feature variables can indicate problems with protein production during fermentation. The analysis of the features allows early detection of the process deviation by sensitively signaling deviations of the input variables from their normal behavior.
15.2.4 Modeling for Process Optimization In general, optimization of production processes means determining optimal set point profiles for the manipulatable process quantities over the entire course of the process. The term optimal beforehand requires the definition of a well-defined performance criterion. Several requirements must be met in optimization. First, the model relating the manipulatable process quantities with the variables that determine the process performance must be accurate enough for all relevant times, from the beginning of the process to its very end. Hence, the model’s prediction and extrapolation capacity must be considerable. Since production processes in biotechnology are complex in nature, it is not likely that this can be achieved with a simple process model. The consequence is that as much knowledge, particularly global knowledge, as is available, must be activated in order to obtain a model with a sufficiently high extrapolation capacity. As such models are rather complicated, process optimization make the highest demands on process modeling. The development of such well-performing models can only be done when sufficient measurement data, from previous process realizations are available. These data must contain enough information about all the dynamic effects of the process which might significantly influence process performance. In a later phase of modeling, such data are required to validate and retune the process models. Initially, however, when the process dynamics is not yet sufficiently known, the data records must be used to explore the process’ behavior, i.e. to discover the variables which are most important in describing the process’ dynamics. The first step on the way to an appropriate process model, is thus an exploratory data analysis [33]. In later development phases, partial models from different resources must be combined in order to refine the model by adding on more and more details about process behav-
434
15 Advances in Modeling for Bioprocess Supervision and Control
ior. Then, it is straightforward to make use of hybrid models (e.g. [17]), as they allow combination of modules from many sources of knowledge. Finally, once a model has been obtained that is sufficiently accurate, appropriate optimization procedures must be found that can take advantage of the complicated process models.
Initial Process Data Exploration and Analysis Before investing time in a detailed model of any process, it is clearly advantageous to analyse coarsely the process’ dynamic behavior by examining all the data available. The first task is to categorize the influence of the various manipulatable quantities on the process performance. In order to save time, it is advisable initially to restrict the number of input variables to those which are well known to have an essential influence on process performance. Since a clear model is not available at the beginning of such a development, simple black-box relationships are generally used to represent the basic relationships between the different process quantities. Traditionally, process engineers use power-law approaches referred to as engineering correlations to extract such relationships from their experimental data. Today, however, a much more powerful variant of such black-box representations, the ANNs are being used with major success. In particular, complex non-linear relationships can be represented in a much more accurate and flexible way with such networks. Even dynamic processes can be represented by neural nets [ 191. Neural nets can be viewed as process representations which incorporate learning from experience, the latter being provided in the form of process data. However, it obviously does not make much sense to learn what is already firmly known. Hence, applications of networks must be restricted to those aspects which cannot be described easily. As an example, consider process modeling for a concrete biochemical conversion process. When the reactants which are converted to significant amounts are known, a general mass balance can usually be formulated in a straightforward way, while the rate expressions describing the biochemical conversion rates are less easy to describe quantitatively. Hence, in the process data exploration phase it is straightforward to use so-called gray-box models. These are composed of simple ordinary differential equations to consider the precise knowledge about the mass balances and of artificial neural networks to represent the biochemical process kinetics which is less firmly known. Such a construct can be considered a first hybrid relationship to describe the process dynamics. In a first naive approach to establish such a hybrid relationship, one is led to use all the process variables from which measurement data are available as inputs to the artificial neural network describing the conversion rates. However, as in human education, it is ultimately much more efficient first to teach the basics and then to refine the knowledge step by step. In this way, the networks first learn the basic structure of the process and are then trained to learn details. The structure of the such exploration procedure is show in Fig. 15-11. This procedure is faster and ultimately does not require so much process data. In order to find the process quantities which are most important with respect to its influ-
15.2 Process Models for Typical Applications
Variables
-----
'
Reaction Rates A" *' /'
Balance
Predicted Outputs
435
Measured outputs
' Equations
#'
ence on the process performance, it is proposed to start the classification procedure with a hybrid system that, in its network component contains only those variables as inputs which are firmly known to be of undoubted importance. Then, one after another, the process variables are tested as to whether their inclusion into the input variables will lead to a reduction in the model error, i.e. the mean deviation between model predictions and the process data used for its validation. If the model error could be significantly reduced after addition of one of the process variables tested, then those quantities that upon addition to the basic input variables led to the least error, can be classified as the first candidate to extend the basic network. By means of the reduction in the model error that the individual additional process variables can provide, these variables are classified. When the reduction in the model error is larger than a critical limit of e.g. lo%, then it makes sense to extend the basic model describing the process kinetics by adding the variable that performed best to its input nodes. Then, it makes sense to repeat the classification procedure described before, since cooperative effects between the different process quantities could have changed the order of the sequence. By repeatedly applying this additional classification procedure, all the process quantities which have a significant influence on the process performance, can be incorporated into the process model in the order of decreasing importance. After these data exploration steps we not only gained information about the variables which influence the key quantities of the process, we also obtained a data-driven, gray-box model, which can be used to determine the set-points or profiles for the control variables. Moreover, the information obtained about the influence of the different variables considered, can be used as a first step in establishing a comprehensive process model.
Construction of Complicated Hybrid Models It should be kept in mind that the data exploration described in the last paragraph can only extract the information clearly and unequivocally contained in the available data. The corresponding process representations can only be regarded as a first approach to a well-performing model since there are more basic resources of process knowledge, e.g. models described in the scientific literature and the knowledge of experienced process engineers working with similar processes in practice. We have
436
15 Advances in Modeling for Bioprocess Supervision and Confrol
already discussed the possible combinations of a knowledge sources for design of a hybrid process models. Here, we will consider this question in more detail. In a straightforward hybrid process model the knowledge from the different resources as far as it can be used to improve the process performance - should be used. Hence, the problem to be solved is to represent this knowledge in such a way that it can be exploited numerically by means of convenient computer software. In the literature, mechanistic models are conventionally published in the form of mathematical process models. These can easily be converted into software. The situation is somewhat more complicated with the heuristic knowledge gained by process engineers from their experience with the process under consideration or similar processes. This heuristics can most often be formulated by means of rulesof-thumb, e.g. in the form of ‘if ... then ... else’-rules. Numerical processing methods of such rules were developed long ago in the area of ‘Artificial Intelligence’ under the headline expert systems. Two essential deficiencies restrict such expert systems most heavily. The first problem is that it is not easy to obtain a comprehensive set of rules which consistently describe the partial process under consideration, and the second is to get a conflict-free description of the process details. These two aspects are by orders of magnitude easier to solve with the modern variant of expert systems - the so-called fuzzy expert systems which also work with rule-based knowledge bases - but take advantage of a non-crisp representation of the variables considered and the fuzzy logic to replace the classical Boolean logic. Fuzzy expert systems (cf. Appendix) work with free parameters - the parameters of the membership functions of the fuzzy variables - that can be fitted to the process data and so far the rule system has been regarded as a basic function system for a non-linear blackbox representation of the process [34]. Fuzzy logic has been successfully applied in many technical applications. Mathematical equation systems, fuzzy expert systems, and artificial neural networks are the basic components to be combined for a well-performing process model. They can be used to describe the dynamics of the process under consideration at the level at which the various process components are actually understood. These representations should be used directly in the modeling software in order to avoid transformations into other representations which inevitably will lead to transformation losses. There is no doubt that with growing process knowledge, the representations of the different parts of the process will become more sophisticated. The aim of hybrid process modeling is to combine the different components effectively. A modular modeling concept, where the modules are arranged in some network structure might help significantly to increase the performance of the model as well as to keep its transparency high enough [35]. The advantages of such a network structure are manifold. A transparent structure helps to avoid mistakes and simplifies adaptation of the model to changes in production processes. Different modules for the same part of the process can be taken from different resources simply making use of complementary experience and process knowledge. These modules can be considered as different votes for effective process descriptions, which can simultaneously be used to form a weighted average. The performance losses which are usually observed upon transformation of model representations from one (e.g. heu-
15.2 Process Models f o r Typical Applications
431
ristic knowledge) into another (e.g. knowledge about physical mechanisms) can be very large. Performance enhancements can be obtained by tuning the hybrid networks by means of an error backpropagation technique similar to the training of artificial neural networks. An example of such a hybrid model for an industrial penicillin production process is shown in Fig. 15-12 [36]. HvbNet
Viscosity
OUR Model Classical Model
Input i
Balance Eouations
+ RBFN
I
Compensator)
r
131 Tuning
............................... .............. ......
4
-
time
1
sugar Fuzzy; Controller
Constraints
--1
+
I
I
Profit Index
Hybrid Model
Optimization Procedure
b) i.0,
C)
Fig. 15-12.Hybrid model used in the optimization of a penicillin production process. (a) Hybrid model structure; (b) optimization scheme; (c) optimization results.
438
15 Advances in Modeling for Bioprocess Supervision and Control
Optimization Procedures with Complicated Models From their experience, process engineers can usually propose fairly good profiles of the control parameters. It is straightforward to use this knowledge as a first guess for a systematic optimization of the profiles. In order to make the proposed profile accessible to computer software, it can be represented by an expansion after some appropriate system of basis functions. The most well-known of such representations is the expansion after powers of time, which is well known to everybody as a polynomial approximation. Optimization of the profile then means finding a set of polynomial coefficients, which provide a maximum of the quantitatively manifested objective function. This objective function must be chosen in a realistic way and must consider all practical constraints on the control of the process. It is well known that non-linear functions can most effectively be represented by artificial neural networks; thus, it is straightforward to use an ANN to represent also the control profiles. The conventional way is to use neural networks with sigmoid basis functions. However, as shown in [26], neural networks with special basis functions can led to better mapping properties.
Opportunities of Random Search Methods Classical optimization schemes like Pontryagin’s maximum principle or Bellman’s dynamic programming are powerful methods but suffer from several disadvantages. They can only be used by extraordinarily experienced personnel, they are significantly limited concerning the complexity of process constraints which can be handled and, finally, they require significant time for tuning the algorithms to the available process data. When the process, or the constraints under which it is operated, or the process objectives have changed, much additional time is required for adaptation work. For models, exceeding some level of complexity, together with the complex constraints under which realistic solutions must be found in practice, conventional optimization procedures like Pontryagin’s continuous maximum technique also usually require a large expenditure. For these cases, it was shown that optimization can be significantly simplified using random search procedures (e.g. [37]). Random search algorithms like chemotaxis, simulated annealing, or evolutionary programming are conceptually simple and thus easy to understand as well as simple to implement and to apply. Their disadvantages of requiring much computing power is no longer a severe restriction. Different approaches of random search techniques can be used. All of them are easy to understand, to codify and finally to operate on today’s workstations and even on PCs. When the expected control profiles are very complicated, the evolutionary programming approach is superior and, thus, recommended. Random search algorithms exhibit some further advantages as compared with the classical methods [38]. Using random search:
15.2 Process Models for Typical Applications -
439
it is less difficult to cope with local minima; and the algorithms are highly universal.
Thus, we propose to put most of the available time into improvements of the process models and to use the simple and reliable random search procedures for determining their optimal solutions. Figure 15-12 shows an example of a process optimization based on a hybrid process model for the particular case of the optimization of the substrate feed in the industrial penicillin fermentation process.
15.2.5 Modeling for Closed-Loop Control Closed-loop control is used to keep, despite all of the distortions appearing in reality, the running process on track, found desirable beforehand. The actions which become necessary upon a deviation of the process from its predefined trajectory can be determined more efficiently when the relevant dynamic features of the process are considered. In this sense, advanced high-performance controllers work in a modelaided way. As controllers need an accurate estimate of the state in order to determine accurately whether or not, and if, how much the process deviates from the desired path, state estimation is also an essential part of advanced controllers. In the following we focus our attention to model-based controllers.
Requirements Concerning the Process Model In closed-loop control, the requirements concerning the process model are different as compared with those of process optimization. The most important difference is that here, the emphasis is on predictions with a short-time horizon, since in closed-loop control quick reactions of the controllers are required, and within short time periods the controller must bring the process back onto its normal path. However, this advantage is compensated by the requirement that more accurate local predictions become necessary. Here, the details of the process dynamics must be known in order to compute precise reactions of the controller on disturbances. Another point is that the short-range predictions can make use of the measurement data acquired and preprocessed up to the actual time instant. The process state can then be determined in some kind of an extrapolation procedure. Consequently, the models must be formulated accordingly. Hence, they are quite different to the models used in optimization. In this sense, the different advanced control techniques may be distinguished by the amount of a priori knowledge they use about the dynamics and the process measurement data they use. Figure 15-13 shows some possible variants of controllers, the applicability of which depend on the knowledge about the process. In the following, the main control possibilities are shortly reviewed.
440
15 Advances in Modeling for Bioprocess Supervision and Control
Control Performance Model-
Necessary Knowledge
Fig. 15-13. The performance of closed-loop controllers depend on the amount of knowledge about the process used.
Adaptive Control Classical PID controllers do not make much use of the knowledge about the process, at least not in terms of a process model. Since the optimal parameters of a PID controller are dependent on the dynamic behavior of the process, they must be changed whenever the process dynamics changes. Changes of process dynamics is commonplace in biotechnology. More advanced variants of PID controllers automatically adapt their parameters according to the changes in the process’ dynamics. They are thus referred to as ‘Adaptive Controllers’ on a PID basis. In order to adapt their parameters, they must be able to recognize and classify changes in the dynamic behavior of the process. Additionally, they need an algorithm relating this information with the necessary corrections of the PID controller parameters. Unfortunately, the classic adaptation algorithms are not always user-friendly and robust, which complicates the application of these controllers to industrial fermentations. An efficient alternative is a fuzzy rule system, that supervises the process and tells whether or not an adaptation of the controller parameters is necessary. Based on heuristic knowledge it can also decide how to tune the PID controller [39]. Levisauskas et al. [40] developed a method which recognizes changes in the dynamic behavior of the process in the available online measurement data in a simple way, a technique to quantify these changes and a relationship to directly utilize this information in a determination of the necessary changes of the PID controller parameters. Figure 15-14 depicts a block diagram of this adaptive PID controller for the specific growth rate of recombinant E. coli bacteria in a fed-batch cultivation process. This case can be regarded as an example, where a minor amount of a priori knowledge is utilized to build an advanced controller. While in the usual model-supported adaptive control, the input/output-relations of the process are used to determine the controller parameters, this method [40] only needs the measurement data of the oxygen uptake rate and does not need an extended process model.
15.2 Process Models f o r Typical Applications
441
OUR
Exponential Filter
Cumulative Glucose Feed
Velocity
Form of PID Controller
W
I '
Set Point Growth Rate
Fig. 15-14. Block diagram of the adaptive PID controller for the specific growth rate of recombinant E. coli in a fed-batch cultivation process.
The concept behind this controller is referred to as the Transfer-Function Concept. In order the obtain the advantage with respect to conventional adaptive control, it is necessary to utilize exactly those measurements information which are most sensitive to the process dynamics. Figure 15 -15 shows experimental results from a fed-batch cultivation of recombinant E. coli where the specific growth rate was controlled with this controller. Neural Network Controllers
In recent years, many different applications of neural nets have been published, as have those of neural network controllers (cf. e.g. [41]). Different control schemes can be distinguished: -
Direct inverse control [42]; Model reference control [41,42]; Internal model control [17,41]; and Model predictive control [23,41].
In these model-aided controllers, the mathematical process model as well as its inverse are represented by means of an artificial neural network. Because of the better approximation capacity of the neural networks in the strongly non-linear processes we are dealing with in biotechnology, these controllers have some potential for industrial applications. However, problems with controller stability, which appear with all non-linear controllers do also appear with neural network controllers. Presently, successful implementations of neural network controllers in biochemical production processes are not known to the authors. Thus, it is to early to judge this development.
442
15 Advances in Modeling f o r Bioprocess Supervision and Control
lime [hl Fig. 15-15. Experimental results from a fed-batch cultivation of recombinant E. coli where the specific growth rate was controlled with adaptive PID controller.
Fuzzy Controllers In principle, the most simple and time-efficient method to construct a process controller is to use fuzzy controllers. The condition is that there is a skilled process operator, who is able to formulate how he or she themselves control the process. However, in practice it is not simple to extract relevant knowledge from experts and there are only a few experts who themselves are able to formulate their own knowledge about control both comprehensively and simply enough.
15.2 Process Models for Typical Applications
443
With regard to the techniques required to represent the rules in a computer and to process them, there is not shortage. There are many papers on the subject in the open literature and several ready-to-use software packages on the market. However, for the front-end-control systems used in biotechnology such software tools are not offered as integral parts. Hopefully, this situation will change in near future. More advanced versions of fuzzy controllers can be optimized on-line making use of process data. Appropriate optimization schemes can be based on genetic algorithms [43].
Cloning of Process Experts Experienced skilled process operators are able to control their process such that it produces products with properties within their desired specifications. This competence is particularly desirable after severe distortions of the process, i.e., when significant deviations from the predefined control path appeared, and it cannot be justified to go back to the old path. Then, the operator will look for the most costeffective way to the desired end products. For such situations, it would be of advantage to have an automatic process control that could perform on-line optimization of the control profiles and combine that with closed-loop control. A first step in the direction of a solution to that problem is to build software systems which are capable of learning to imitate the skilled human operator. This idea was named ‘cloning of experts’ [42]. Figure 15-16 sketches the structure of such a software for on-line bioprocess control. While in robotics, the aim was to copy the ability of humans to perform mechanical manipulations, the focus here is on imitating, by means of computer software, their ability to recognize process situations and to respond properly with a suitable control action. Such a software must makes use of a priori knowledge about the process (formulated by a model) and objective functions clearly defined by process experts beforehand. Essentially, this means automation of tasks that engineers use to perform. Fuzzy reasoning and fuzzy neural networks are tools that can be taken for such duties. Obviously, the software imitating operators must be developed by process engineers, who know how to operate the process. Concrete software modules may be generated in the following way. Two main software modules are required, one represents the process under consideration, the other imitates the operator actions. Consider that the process model has already been established and validated. It is essential that the simulator has the possibility to add to the model a non-stationary behavior and process noises in order to demonstrate process situations under different sources of distortions. Thus, we primarily consider the development of the imitator, i.e. the program that simulates the operator behavior. In the beginning, the operator is being observed while he or she is controlling the process. The actions, they perform in different situations, are converted into fuzzy rules. This fuzzy rule system can be considered the first approach to an imitating software. Since the operators do not decide on single measurement values but on shorter or longer trends in or shapes of the signals, a very important feature of such software it to recognize and classify trends in the signals.
444
15 Advances in Modeling for Bioprocess Supervision and Control
Expert Action
Process Expert
~
-
47
Trend Recognition
Two-step Expert
Fig. 15-16. Use of fuzzy Ann-based procedures for the bioprocess expert cloning.
In the next step this first fuzzy system will be tuned. This is made by running the software imitator in parallel to the controlling actions of the operator. From the deviations between the action of the human operator and the actions proposed by the software, the rule system will be improved. This can be done by extending the numer of rules, by improving existing rules, or by changing the membership functions of the fuzzy variables used. Further improvements of the imitator can then, in a next step, be obtained by means of process simulations using extended models that include additional non-stationary behavior and different noises. In this way it is possible to simulate the process for some more extreme process situations. Once again, the human operator must try to control the (now simulated) process. His or her reactions and the actions determined by the software are compared and again the differences are used to improve the imitator. After some iterations and applications in practice, the controller software cannot learn much more from the operator. Then, there is another way to improve the imitator, by using evolutionary programming techniques. The approach is first to define
15.3 Software Tools for Data Analysis and Modeling
445
a quantitative performance measure for the control and evolutionarily to improve the imitator. Finally, after the rule base is considered to be essentially complete, the controller software can be used directly to control the process or to assist the operator. In the case of operator assistance, all of his or her actions are compared with the actions that they beforehand taught the imitator. In this way, unusual actions will be detected by the software and the operator will be asked whether or not they would like to deviate from their usual procedure. In order to avoid misunderstandings it should be noted here that only human beings are considered intelligent enough to understand a complex process with noises and non-stationary behavior and to react directly on their insights. Software modules are not creative in this respect, but can perform many supervision tasks. Their most important advantages are that in what they can do, they are by far less expensive than human controllers, and they can work 24-hour a day, 7 days a week. Moreover, they usually perform tasks they are built for in a much more reproducible way.
15.3 Software Tools for Data Analysis and Modeling The very task of bioprocess engineers is steadily to improve the performance of their processes with a minimum of expenditure. Since the most critical issue in this business is efficiently making use of a priori knowledge about the process under consideration, they cannot delegate process optimization to control engineers. They must solve most optimization and control problems essentially by themselves, since they are expected to possess most of the required process knowledge. In order to cope with this highly demanding task, they are forced to use as effective tools as are available. Many powerful software tools became available on the market recently [44,45]. Nevertheless, we detected a considerable deficit in software that can be used in biotechnological laboratories as well as in pilot and production plants. The most restraining deficits of the currently accessible software are particularly due to missing compartibility between all the packages necessary to meet the demands. Here, we would like to mention some tools which we consider should be available at a well-equipped working place, and cite some examples of readily available software packages that meet the requirements to different degrees. Obviously the software must be able to support the process engineer in processing the large amount of measurement data appearing in an industrial laboratory, pilot plant, or production system. Statistical data analysis, pattern recognition, and extraction of the characteristic features of the signals are the first procedures that must be done with process data. Thus, these operations must mainly be supported by efficient data processing tools.
446
15 Advances in Modeling f o r Bioprocess Supervision and Conrrol
15.3.1 HardwardSoftware Bases Required As known to everybody who reads this article, the hardware which can be used to run the software described, can be broadly divided in two classes, the PCs and the Workstations. PCs, as are commonplace, become stronger with regard to their computing power each year. However, the workstations do not stagnate in this respect. Besides computational performance, workstations and PCs mainly differ by their operating systems. Most of today’s workstations run under UNIX, while most PCs are running under Windows (95). However, there is another competing operating system, Windows NT, emerging in the past years, which interestingly seems to unify both computer classes by building a platform on which both PC as well as workstation software can be run. It is not the aim of this article to discuss operating systems; however, the choice of hardware and the corresponding operating system has an immense influence on the software available to solving problems in process engineering. The higher power of UNIX workstation systems with regard to computing power often justifies their higher prices. However, user-friendliness is a decisive selection criterion for software. As with any other tool, the first requirement is that it provides a sufficient functionality, but when there are more than one alternative, then the convenience by which it can be used becomes decisive. Software tools must be accessible intuitively and their operation must be easy to learn. On PC systems much has been done since Microsoft Windows software came onto the market. Good examples of user-friendliness are the currently available text systems. These are significantly more userfriendly than most UNIX-based software available for workstations. Many scientific software tools running on workstation systems under UNIX must be improved in this respect, otherwise many people will be move to Windows NT. On the other hand, UNIX is the base for the rapidly advancing Internet activities.
15.3.2 Data Exploration Tools Requirements More and more data are being acquired during biochemical production processes. There are several reasons for this. Process quality assurance measures are rated higher with the years, and thus more has been invested into measurement devices and computers to acquire and store the data. In many production systems, FDA requirements led to a significant increase of measurement data recorded during production processes. Also, with the availability of complex measurement devices at reasonable prices, many more samples are taken from the reactors and complex off-line analyses are performed in the analytic laboratories. It is necessary quickly to select data from extended records or data-bases for further processing like integration, differentiation, transformation, filtering, etc. To
15.3 Software Tools f o r Data Analysis and Modeling
447
judge interactively the success of these operations, a quick visualization of the raw as well as the processed data is necessary. Indeed, the value of a software tool is much dependent on its visualization capacity, as graphs are predominantly used by engineers to discuss scientific results. The decisive question is how quick complex and extended data records can be represented by easy-to-comprehend 2D or 3D graphical representations. For convenience, most users prefer software with a so-called Graphical User Interface (GUI). The entire functionality of this software is controlled by mouse actions and only simple numerical as well as short text inputs are done via the keyboard. All different options of the software can be accessed by means of pull-down menus, of course via simple mouse clicks. In multidimensional state spaces it is of great help to visualize two-dimensional hypersurfaces. In order to grasp the structural information of a multidimensional graph it is necessary flexibly to change the perspective. This must be possible by simple mouse operations. Interactive control of the representations is required.
Examples of Data Exploration Tools
PV-WAVE As the volume of data in technical industries continues to expand, understanding the information hidden in that data becomes more challenging. PV-WAVE Advantage is an advanced visual data analysis software, designed to perform time-critical tasks in data comprehension and communication. The software is recommended for off-line data analysis. Via its subsystem ‘Interactive Language’, PV-WAVE provides a compact and efficient array-oriented, fourth-generation language. Its interactive structure reduces coding and eliminates compiling and linking. It supports single variables, collections of variables and all the language constructs of FORTRAN and C, which are essential in putting together different software tools and user-written algorithms. PV-WAVE’S open system features works with industry standard I/Os, it possesses a symbolic GUI-based debugger and many flexible data management functions. Most types of data can be accommodated, allowing to fit specific application needs. PVWAVE has excellent graphical capabilities that make it easy to identify trends and to find hidden regularities. It thus leads to a better understandig of the information hidden in the data. PV-WAVE includes many useful graphical routines, that allow display of 2D and 3D plots and volume visualization.
Iris Explorer Iris Explorer is an object-oriented visual programming environment for scientific and engineering applications. It is particularly suited for displaying and analyzing complete multi-dimensional data sets interactively, thereby keeping the required programming expenditures low. Applications can be customized within short time pe-
448
15 Advances in Modeling for Bioprocess Supervision and Control
riods, without the need for writing computer code, just using point-and-click selections. Iris Explorer is modularly designed; it includes a scripts interface, widgets for module control panels, modules for vector data analysis, and, in particular, modules from the well-known NAG library. Iris Explorer is designed for distributed processing environments as well as individual workstations. In its latest release (Version 3.0), Iris Explorer is available on a wide range of computing platforms. The software can be recommended for off-line as well as on-line data analysis. Prophet Prophet is a software package which includes data management, statistical analysis, and simple mathematical modeling. Its most prominent feature is a very strong and user-friendly graphical interface. Prophet allows analysis and presentation of data tables, graphs with a high resolution graphics and multiple windows. It is very easy to learn and use. Prophet is a free software available on the Internet and runs on UNIX workstations. Unfortunately, the system turns may be a little too slow when large data records are to be processed. It is suggested for bioprocess engineers who need a user-friendly software for statistical analysis of their data.
15.3.3 Modern Modeling Software Tools Requirements Today, process engineers need not spend much time in solving simple differential equation systems. Different solvers for classical mathematical models are available commercially (e.g. MAPLE, MATLAB, SCILAB). There are also many software tools that support the application of simple fuzzy rule systems, e.g. fuzzyTECH (Inform), Data Engine (MIT), FuzzyCLIPS (NASA), ECANSE (Siemens) [30]. In order to establish and test simple neural networks, a large variety of software tools is available. It begins with Neuralworks (Neuralware), BrainMaker (California Scientific Software) up to NeuroShell (Ward Systems) and Neuro Solutions (NeuroDimension). Most of the currently appearing software tools use the connected object simulation technique where different parts or components of the calculations, which make up the model evaluation, are summarized in modules which are represented graphically in the form of blocks, that can be selected and interconnected by simple mouse-based operations. The user needs only to change their parameters interactively on the screen. In flexible systems, the users can build their own blocks for repeatedly used software modules. Excellent examples of such software features can be found in SIMULINK and SCICOS. Figure 15-17 shows typical graphs in a SCICOS environment for a fed-batch fermentation.
15.3 Software Tools f o r Data Analysis and Modeling
449
Fed-Batch Bioprocess
U
Monod Kinetics
Balance
X&S
r
Volume
Time
Feed
Rx, Rs,Feed
Fig. 15-17.Realization of fed-batch process modeling task using the SCILAB/SCICOS modeling environment.
Examples of Modern Modeling Software
MATLABKIMULINK MATLAB is a computing environment for high-performance numeric computations and visualizations with a wide applicability in engineering science. MATLAB integrates numerical analysis, matrix computation, signal processing, and graphics in an easy-to-use software evironment. Problems and solutions are expressed just as they are written in the usual mathematical representation, i.e. without traditional programming languages. MATLAB also features a family of application-specific sets of functions that are called toolboxes. Toolboxes are available, e.g. for signal processing, control system design, dynamic system simulation, system identification, neural networks, fuzzy systems, and others. SIMULINK is a tool for modeling, analyzing, and simulating an extraordinarily wide variety of physical and mathematical systems. With SIMULINK it is possible to model processes graphically, sidestepping much of the nuisance associated with conventional programming. MATLAB and SIMULINK are available for a wide variety of PC-systems and workstations and is running under most of the commonly used operating systems.
450
15 Advances in Modeling for Bioprocess Supervision and Control
These general-purpose softwares package are recommended for off-line modeling and simulation in the field of process control.
SCILAB/SCICOS Scilab is well-elaborated software package, which has been developed for system control and signal processing applications and running in UNIX/X-Windows environments on workstations. Scilab’s source code is freely distributed. Its libraries and most of the interpreter are written in FORTRAN and are compatible with most scientific numerical libraries. The graphic facilities and the UNIX interface are written in C. Scilab is composed of three distinct parts: an interpreter, function libraries (Scilab procedures), and libraries of FORTRAN and C routines. These routines are of general interest to the modeling community and most of them are available through Netlib. A few of them have been slightly modified for a better compatibility with Scilab’s interpreter. A useful tool distributed with Scilab is Intersci, which is a set of routines that allow users to easily add on new primitives to Scilab, i.e. to add new modules, written in FORTRAN or C. In particular, Scilab provides a variety of powerful primitives for the analysis of non-linear systems. Integration of explicit and implicit systems can be accomplished numerically. There exist numerical optimization facilities for non-linear optimization (including non-differentiable systems optimization), quadratic optimization, and linear optimization. Scilab is an open programming environment, which puts the creation of functions and libraries of functions into the hands of the user. Functions are recognized as data objects by Scilab and, thus, they can be manipulated or created just as other data objects. Scilab is easily interfaced with FORTRAN or C subprograms. This allows the use of standardized packages and libraries in the interpreted evironment of Scilab. Scicos (Scilab’s Connected Object Simulator) is a Scilab package for modeling and simulation of hybrid dynamical systems. Scicos includes a graphical editor, which can be used to build complex models by interconnecting blocks, which represent either predefined basic functions already stored in Scicos libraries (palettes), or user-defined functions. Scicos can thus be used in a large class of complex applications. Scilab is strongly recommended for modeling simulation and optimization in bioprocess control applications. It is a very flexible tool which has about the same performance as Matlab, but is available on the Internet. Hybrid Tools
Combination of several software tools exist wich represent the different modeling approaches such as mechanistic mathematical models, heuristic rule-of-thumb systems, data-driven black box approaches, etc. The following examples should provide an idea of such hybrid modeling tools.
15.3 Software Tools f o r Data Analysis and Modeling
45 1
Gensym G2 Gensym’s G2 software offers a graphical, object-oriented environment for creating intelligent applications that monitor, diagnose, and control dynamic events in online and simulated environments. G2 can also be viewed at as a graphical, objectoriented environment for building and deploying intelligent real-time systems. Its language editor allows users to enter rules, models, and procedures that describe real-time operations. G2 includes concurrent execution of neural networks, rulebased systems, and mathematical procedures. The subsystem G2 GUIDE allows users to create graphical end user interfaces and real-time displays. It can be connected to off-the-shelf real-time data systems. A non-traditional extension is the subsystem Neuron-Line, which allows users easily to create neural network applications. G2 provides real-time connectivity to databases, PLCs, DCSs, and other real-time data systems. It can be run on workstations from Digital Equipment, IBM, Hewlett Packard, and Sun Microsystems, and on PCs under Windows NT, and Windows. It allows off-line as well as on-line applications but is more suited for larger application systems.
HybNet HybNet [30] is a user-friendly hybrid software package, which contains modules for mathematical models based on differential equation systems, heuristic models represented by fuzzy rule-based systems, as well as different non-linear black-box models which can be formulated by various types of artificial neural networks. There are also various other possibilities like neurofuzzy process representations. Many different modules representing different process components (sub-processes) for different process phases can be arranged in such a way that they can be processed quasi simultaneously. Its essential feature is its network structure in which the different nodes are software modules of different levels of sophistication. The network structure makes tuning of the model parameters to given process data much easier and efficient, since the ideas behind the error-backpropagation mechanism developed in the field of neural networks can be used. HybNet can easily be interfaced with open software components wirtten in general purpose languages like C, FORTRAN, etc. Also, it can work together with complex packages like Scilab. Due to its high flexibility, it allows an easy link to front-end processors installed at the real process. In this way it can be used on-line for process monitoring and advanced control. Figure 15-18 outlines the construction of the HybNet Software System.
452
15 Advances in Modeling f o r Bioprocess Supervision and Control
Fig. 15-18. Construction of the HybNet Software System.
15.3.4 Front-End Systems Definition and Requirements Front-end systems are hardwarehoftware systems that make the connection between the computer on which the state estimation and the high-level control procedures run, and the fermenter hardware including its activators and measurement devices. The front-end systems are used for data acquisition and to perform the low-level control like pH and temperature control. Most systems installed in the past years contain more or less extensively developed software packages for low-level data analysis. The reason why we discuss this is that these systems must be regarded as an integral part of the overall process optimization and control system. When advanced control procedures are to be applied, then the front-end systems must be compatible with the systems on the next level. Front-end systems must automatically deliver to the higher end process supervisory and control computers all the information about the process which are relevant to the process performance. This must be done using a strictly defined protocol. One
1.5.4 Conclusions and Recommendations
453
reason for the poorly developed process control in biotechnology is that the data exchange between front-end systems and supervisory computers does not perform properly.
Examples of Front-End Systems Currently there are many front-end systems available commercially. Only a few of the smaller ones are customized with respect to biotechnology. Most of the latter are offered by the vendors of fermenters. There are very large systems which are primarily developed for plant automation in the chemical and process industries; examples are systems from Foxboro, Siemens, ABB, etc. There are also medium-range software packages, e.g. those which are predominantly designed for smaller industrial production plants. Examples of smaller front-end systems are MFCS (Braun), Iris (Infors), and New Brunswick, (Applicon). Although there are a number of systems available in biotechnology, none of these could manage to become a standard in this field. In our experience, MFCSWin, the latest version of MFCS is currently the most advanced version of systems in this category. MFCS/win is based on the currently used 32-bit Windows technology (WindowsNT/Windows95) and offers the following features. It is tailored to fermentation applications. The software can be run on standard PC hardware and supports full network operation without an expensive network server, making use of clientherver architectures. In WindowsNT environments it supports multitasking. Dynamic data exchange (DDE) can be used for data transfer to the commonly used Microsoft products like MS ExcelTMNSAccessTM,etc. Special emphasis was placed on a software validation according the cGMP requirements. MFCS /win provides some flexibility for fermenter control and data acquisition.
15.4 Conclusions and Recommendations There are some recent developements in model-supported process supervision, control, and optimization that can be viewed as significant advances. With the extension of modeling techniques, particularly in the area of non-linear gray-box models, it becomes possible to take advantage of the wide variety of heuristic process knowledge accumulated by skilled process engineers. Additionally, the advent of artificial neural networks extended the value of black-box models significantly. At present, there are many well-performing techniques available, but there is shortage of easyto-use software tools which process engineers can apply easily at their plants; however, a few powerful tools are already available. In the view of the many possibilities of designing a well-performing process model, modeling can be viewed as a multidimensional optimization problem. The degrees of freedom are indeed manifold. Selection of software tools, for example,
454
15 Advances in Modeling f o r Bioprocess Supervision and Control
must consider the skills of the process engineers engaged in modeling, the available process data, and further process knowledge available. The information content of the data which can be acquired for the model development is of decisive importance in choosing the right modeling technique. For example, when the data base is scarce, it does not make sense to use artificial neural networks, which are the most powerful tools in modeling production systems, where extended data records exist, which sufficiently cover the relevant part of the state space. The extended spectrum of modeling methods allows to compensate - to some extent - detailed process knowledge with comprehensive data records from the process under investigation. Probably the most important criterion for the choice of modeling methods is manpower. The skills of the process engineers concerning knowledge about the process dynamics and experience with respect its control must not be underestimated. No less important is that even highly skilled process engineers need time enough to solve process modeling tasks. In these respects the choice of the targets of process improvements must be carefully chosen. The goal should be set properly, but not too ambiguously. A model that by which the goal cannot be reached in due time has cost time, but has not brought any benefit. Thus, the general recommendation is: 1. As the dominating cost factor in modeling is the cost for personnel, companies should invest in software tools that allow engineers to work more efficiently. The modeling methods and the corresponding software tools must be chosen according to the engineers’ skills, and with respect to their process knowledge and experience with modeling tasks. Time and manpower requirements must be estimated realistically. Concerning the optimization problem discussed, there is no doubt that an investment into efficient software is less expensive than into additional manpower. 2. The next important cost factor is process data; during process development, this means experimental investigation of the process. Here, the obvious goal must be to reduce the number of costly experiments. The consequences are that the available data must be exploited as much as possible and experiments must be prepared properly so as to avoid unnecessary investigations. New experiments must be systematically planned, making optimal use of the already available knowledge and data in order to get an optimum of information from the experiment. Hence, a model-supported experimental design is required using the models available. 3. An important practical point, which has often been underestimated and considered as a boring unattractive task, is data preprocessing in the sense that the data easily become accessible to many different data processing activities. Very often, we observed that the data necessary for process improvements were already available, but in a form that would take the engineers too much time to transform them into a representation in which they could be used conveniently. In such cases - which appear quite often - the data were generated in expensive experiments and then buried in data ‘cemeteries’, simply because insufficient effort was investigated in storing them in such a way that the activation barrier to use them later on is minimized.
References
455
4. In order to support the application of the new techniques described above in industry, they must be translated into efficient algorithms and made available to the process engineers by means of user-friendly software implementations. In order to bring the new developments into industrial practice, a paradigm change is required in the modelers’ laboratories: it is better to spend time on transparent robust algorithms than to develop too-detailed ones which are then too complicated and will not be understood by the persons that should apply them. The consequence is that they will be discarded. To state this more clearly in a slightly over-subtle way: Then, the real value of the techniques will never be validated in practice and, hence, the corresponding developments are degraded to ‘playing with software’. Thus, we need a set of software tools by which the different techniques on different levels of sophistication are implemented so that the needs of different departments can easily be met.
Acknowledgments Fruitful discussions with many colleagues are gratefully acknowledged. In particular, we would like to thank Dr. Hans Preusting and Dr. Henk Noorman from Gist-Brocades, Dr. M. Dors and Dr. I. Havlik from Hannover University, and Dr. Norbert Volk and Dip1.-Ing. Rui Oliveira from our laboratory.
References [ l ] Royce, P.N., TZBTECH 1992, 10, 232-234. [2] Royce, P.N., Crit Rev Biotechnol 1993, 13, 117-149. [3] Simutis, R., Oliveira, R., Manikowski, M., Fey0 de Azevedo, S . , Lubbert, A., Proceedings of the 1st European Symposium on Biochemical Engineering Science, Dublin, Ireland, 1996. [4] Christensen, L.H., Marcher, J., Schulze, U., Carlsen, M., Min, R.W., Nielsen, J., Villadsen, J., Biotechnol Bioeng 1996, 52, 237-247. [5] Konstaninov, K.B., Biotechnol Bioeng 1996, 52, 271-289. [6] Kell, D., Sonnleitner, B., TZBTECH 1995, 13, 481-492. [7] Locher, G., Sonnleitner, B., Fiechter, A,, J Biotechnol 1992, 25, 23-73. [8] Stephanopoulos, G., Konstantinov, K., Saner, U., Yoshida, T., in: Biotechnology: Rehm, H.J., Reed, G. (Eds.)., Weinheim: VCH, 1993; Vol. 3, pp. 355-400. [9] Roels, J.A., Energetics and Kinetics in Biotechnology, Amsterdam: Elsevier Biomedical Press, 1983. [lo] San, K.Y., Stephanopoulos, G., Biotechnol Bioeng 1984, 26, 1176-1188. [I I] San, K.Y., Stephanopoulos, G., Biotechnol Bioeng 1984, 26, 1189-1197. [I21 Chattaway, T., Demain, A.L., Stephanopoulos, G., Biotechnol Prog 1992, 8, 81-84. [13] Schalien, R., Fagervik, K., Saxen, B., Ringbom, K., Rydstrom, M., Biotechnol Bioeng 1995, 48, 631-638. [14] Stephanopoulos, G., Park, S . , in: Biotechnology: Rehm, H.J., Reed, G. (Eds.). Weinheim: VCH, 1991; Vol. 4, pp. 225-250. [l5] Simutis, R., Havlik, I., Lubbert, A., J Biotechnol 1992, 24, 211-234.
456
15 Advances in Modeling for Bioprocess Supervision and Control
[16] Psichogios, D.C., Ungar, L.H., AIChE J 1992, 38, 1499-1511. [17] Schubert, J., Simutis, R., Dors, M., Havlik, I., Lubbert, A,, J Biotechnol 1994, 35, 51-68. [18] Thomson, M.L., Kramer, M.A., AIChE J 1994, 40, 1328-1340. [ 191 Haykin, S . , Neural networks. A Comprehensive Foundation, New York: Macmillan College Publishing Company, 1994. [20] Pollard, J.F., Broussard, M.R., Garrison, D.B., San, K.Y., Comp Chem Eng 1992, 16, 253270. [21] Lu, Y.Z., Industrial Intelligent Control. New York: John Wiley & Sons, 1996. [22] Wang, L.X., Adaptive Fuzzy Systems and Control, Englewood Cliffs: Prentice Hall, 1994. [23] Brown, M., Harris, C., Neurofuzzy Adaptive Modeling and Control, New York: Prentice Hall, 1994. [24] Simutis, R., Havlik, I., Dors, M., Liibbert, A., in: Preprints of proceedings of the 6th international conference on computer applications in biotechnology. Munak, A., Schiigerl, K., (Eds.). Garmisch-Partenkirchen, 1995, pp. 59-65. [25] Moms, A.J., Montague, G.A., Willis, M.J., Trans IChemE 1994, 72A, 3-19. [26] Stephanopoulos, G., Han, C., Comp Chem Eng 1996, 20, 743-791. [27] Intelligent Systems in Process Engineering: Paradigms for design and operations. Stephanopoulos, G., Han, C., (Eds.). New York: Academic Press, 1996. [28] Wold, S . , Esbensen, K., Geladi, P., Chemometrics and Intelligent Laboratory Systems 1987, 2 , 37-52. [29] Esbensen, K., Schonkopf, S . , Midtgaard, T., Multivariate Analysis in Practice, Cam0 AS., 1995. [30] Geladi, P., Kowalski, B., Anal Chim Acta 1986, 185, 1-32. [31] Warnes, M.R., Glassey, J., Montague, G.A., Kara, B., Process Biochem 1996, 31, 147-155. [32] Kramer, M.A., AlChE J 1991, 37, 233-243. [33] Simutis, R., Lubbert, A., Biotechnol Prog 1996, 13, 479-487. [34] Sjoberg, J., Zhang, Q., Ljung, L., Benveniste, A., Delyon, B., Glorennec, P.Y., Hjalmarsson, H., Juditsky, A., Automatica 1995, 31, 1691-1724. [35] Oliveira, R., Simutis, R., Fey0 de Azevedo, S . , Lubbert, A., in: Proceedings of the 1st European Symposium on Biochemical Engineering Science: Glennon, B., Kieran, P.M., Luyben, K., (Eds.). Dublin, 1996. [36] Preusting, H., Noordover, J., Simutis, R., Liibbert, A., Chimia 1996, 50(9), 416-418. [37] Simutis, R., Lubbert, A. (1996), J Biotechnol 1997, 52, 245-256. [38] Fogel, D., Evolutionary computation, IEEE Press, New York, 1995. [39] Hang, C.C., Ho, W.K., Lee, T.H., in: Intelligent Control Systems: Gupta, M.M., Sinha, N.K., (Eds.). Saskatchewan: IEEE Press, 1996, pp. 345-382. [40] Levisauskas, D., Simutis, R., Borvitz, D., Lubbert, A,, Bioprocess Engineering 1996, 15, 145-150. [41] Hunt, K.J., Sbarbaro, D., Zbikowski, R., Gawthrop, P.J., Automatica 1992, 28, 1083-1112. [42] Werbos, P.J., in: Intelligent Control Systems: Gupta, M.M., Sinha, N.K., (Eds.). Saskatchewan: IEEE Press, 1996, pp. 791-803. [43] Moriyama, H., Shimizu, K., J Chem Tech Biotechnol 1996, 66, 217-222. [44] Kheir, N.A., System Modeling and Computer Simulation, New York: Marcel Dekker, 1995. [45] Braharn, R., IEEE Spectrum 1995, 32, 19-36. [46] Rurnelhart, D.E., Hinton, G.E., Williams, R.J., in: Parallel Distributed Processing: Rumelhart, D.E., McClelland, J.L. (Eds.). MIT, Cambridge, MA, 1986; Vol. 1, pp. 318-362. [47] Leonard, J., Kramer, M., Comp Chem Eng 1990, 14, 337-341. [48] Kramer, M.A., Comp Chem Eng 1992, 16, 313-328.
Aaaendix
457
Appendix Fuzzy Logic-Based Process Models Dynamic black-box models are used to represent approximately process responses on changes in its vector of input variables. In mathematical terms, we are speaking about a mappings between two vector spaces %i and $loof usually different dimensions. Such a mapping can often be represented by linguistically formulated rules. If these rules are based on fuzzy sets and are evaluated by a reasoning based on fuzzy logic, then we refer to an entire collection of fuzzy rules as a fuzzy process model. Since we are normally dealing with fuzzy rules reflecting the heuristic experiences which bioprocess engineers gained from observed input/output data, we classify these fuzzy models as colored (gray) black-box models. The advantage over simple black-box models like polynomial representations or other regression models is that it is easier to make use of a priori knowledge about the input-output relationships of the process. In the following text, the essential aspects of fuzzy models are shortly reviewed. For details refer the special literature, e.g. references [22] and [34].
Fuzzy Sets, Operators and Implications Assume we are dealing with a real process variable x, e.g. the oxygen uptake rate x = OUR. In modeling, we are often describing process situations by means of variables. Often, it is not possible to describe a particular process situation by a clearly defined vector of the state variables. Instead, it may be possible, as is often done in the laboratory, to classify values of one or more process variables, e.g. the variable x as belonging to a more or less vaguely defined range h of x , referred to as, e.g. ‘high’ values. In fuzzy theory, the attribute ‘high’ is thought to be more or less precisely fulfilled. The extent to which a concrete real value, e.g. a measured value a of OUR, can be regarded as a ‘high’ oxygen uptake rate is clearly context-dependent. It may be defined by a concrete function, called a membership function m&). In the fuzzy set theory, the membership function mh(X) is a real-valued function with values in the range 0 Imh(X) I1. Essentially, it reflects the knowledge about how fine different values of the quantity x can or must be distinguished in a particular process or in a particular task to be solved at the process. In order to be able to classify all possible values of x appropriately, a set of membership functions (mi}, covering the entire range of x-values which could be accessed in the process under consideration, should be defined. In such a case it is possible to specify x by a statement
‘x is h’
or in our concrete example
‘OUR is high’
Such a statement is sometimes called a fuzzy statement and the membership function, characterizing the attribute h (‘high’) is called a fuzzy set. If a measurement
458
15 Advances in Modeling f o r Bioprocess Supervision and Control
value a was recorded for OUR, than it can be attributed to the set of ‘high oxygen uptake rate values’ to an extent mh(a) and, may be, to another attribute m (e.g. medium OUR values) to an extent m,(a). Through this clearly defined membership function, the vague statement ‘OUR is high’ becomes accessible by computer software. The essential feature of fuzzy statement in the light of technical applications is that they can be used to form the implications i of the type
i:
If ‘x is h’ then ‘y is s’
e.g.9
i:
If ‘OUR is high’ then ‘PDR is small’
which are called fuzzy rules. The implication i can be used to more or less coarsely describe a relationship between the quantities x and y in the process. It is evaluated in a straightforward way, which may qualitatively be described in the following way. If a particular value a of x belongs to the class of h-values to an extent m,(a), then the corresponding value f~of y will belong to the s-class of y to at most the same extent only, i.e. the membership of the value p to the s-class of y will be reduced as compared with its original membership defined by the fuzzy set (specified by the membership function) m,(y). There are several possibilities to define quantitatively the concrete implication. One common example is to limit the membership mi&) of to the class s of y as a consequence of the implication i to a maximum value mh(a), i.e. to the range
0
5
mis(y> 5 mh(a),
in short: the more a belongs to the class h of x, the more the corresponding value p of y will belong to the class s of y. Fuzzy statements can be combined to new fuzzy statements using fuzzy operators and, or, not, in order to be able to describe process situations in more detail. For example it may be that we would like to characterize a particular situation in a biochemical production process by high oxygen uptake rate and medium product development rate PDR, then we could state: ‘OUR is high’
and
‘PDR is medium’
These fuzzy operators (here and) are logical operators which are defined in terms of the fuzzy logic. If the example mentioned appears in the condition part of the fuzzy rule and the corresponding membership functions are mh(x) and mm(Z),then the fuzzy rule j reads If ‘OUR is high’ and ‘PDR is medium’ then ‘BPR is small’ In a situation where the values a of x and x of z appeared, the membership function mi,(y) in the conclusion part will be limited to
Appendix
459
In order to describe all situations, which must be distinguished in a particular process, a collection of rules might be required. Such collections are called fuzzy rule bases. These rule bases might be considered particular fuzzy models of the process. In a practical application, all rules within a given rule base are evaluated quasisimultaneously. Then, at each time the rules have been evaluated, the actually resulting membership functions rn'k(y) for each output variable y are summed up over all i and all k to form a total function M b ) . In an act of defuzzification, an estimate of a concrete estimation 6 of y will be calculated by determining the abscissa 6 of the tenter of gravity of M(y). Thus, all steps of processing a fuzzy rule base can be clearly evaluated by simple algorithms. These algorithms form a procedure that relates some measured values a, p, ... of the process variables x , z, ... to some value 6 of the output variables, y, ... of the process. In this way it can easily be seen that we are dealing with a process model. Since the fuzzy rules relating the different process variables to each other can be formulated according to the experience of the bioprocess engineers with the process, the fuzzy process model provides some way to express a priori knowledge about the process. Tuning of the Fuzzy Model
The formulation of the fuzzy rules by experienced process engineers is dependent on their process knowledge. The same applies to their definitions of the membership functions of the different variables. Usually, the definition of the rules is given a higher credibility than the definitions of membership functions. Hence, the parameters specifying the latter can to some extent be regarded as free parameters which might be improved by fitting the rule set to available process data. This fitting can be performed by arranging the set of fuzzy rules in form of a neural network and then using the conventional error backpropagation technique to determine optimal parameter sets, just in the same way as during the training of conventional neural networks. With such a neurofuzzy network, the heuristic model formulation by the process engineers can be improved making use of the available process measurement data. The advantage of the neurofuzzy as compared with conventional neural network black-box models is, that the results of such improvements can be examined and judged by the experienced engineers, since they are represented in a physically meaningful form. Figure 15-19 shows the general scheme of the procedure of the fuzzy artificial neural network design from the rules formulated by the expert.
460
I S Advances in Modeling for Bioprocess Supervision and Control
Expert ~
F u u y Knowledge Base
p
X
>-
Neural Network 3
Balance Equation
'
D
Fig. 15-19.General scheme showing the procedure of designing a fuzzy ANN from the rules formulated by expert.
Autoassociative Artificial Neural Networks (aANNs) An Autoassociative Artificial Neural Network (aAnn) is a special type of artificial neural network that performs a non-linear coordinate transformation of the input variable space into some artificial variable-space of lower dimension with the aim of maintaining - in the transformed data - the main features of the process in which one is interested [32]. The general idea is the same as the idea behind the PCA technique; however, the applications of aANNs allow a non-linear feature extraction from the process data. This is being performed in a two-step procedure by first mapping the process data from the input variable space onto some intermediate representation in an artificial lower-dimensional variable space. Then, in a second step, the inverse mapping step, the intermediate representation is mapped back into the original input variable space. Autoassociative neural networks use a feedforward artificial neural network architecture most often composed of five layers: an input layer, three hidden layers, and an output layer. The structure of the autoassociative neural network is shown in Fig.
Appendix
461
15-7. The first of the hidden layers is called the mapping layer. The transfer functions of its nodes are sigmoidal functions. The second hidden layer is called the bottleneck layer. The transfer functions of its nodes are linear or sigmoidal functions. Finally, the third hidden layer, referred to as the demapping layer uses nodes with sigmoids as transfer functions. The bottleneck layer, which must be composed of a number of nodes that is smaller as the number of input variables, forces the network to develop a compact representation of the input data. During the training procedure the network weights are tuned so that the output vector Y’ matches the input vector Y as closely as possible with the set of training examples (identity mapping). This is the decisive objective for the optimization of the weights of the nodes of this network. For the actual training of the autoassociative network, any classical optimization algorithm, e.g. the backpropagation [46] or the conjugate gradient algorithm [47] can be used. Cross-validation should be used in order to select an appropriate, i.e. not too large but sufficiently small number of nodes in the bottleneck layer [48]. The variables of the new dimensionally reduced state space, represented by the values of responses of the nodes in the bottleneck layer, do not have the property to be measurable physical quantities. Since they must be able to describe the main features of the process, they are called ‘features’. Hence, they can be used to represent the process dynamics in a more compact way.
Part Four Validation
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
16 Validation of Viral Safety for Pharmaceutical Proteins Joachim K. Walter, Franz Nothelfer, William Werz
16.1 Introduction Pharmaceutical products, especially proteins, derived from biological sources, have been shown in the past to bear a risk potential to patients regarding drug safety. A number of incidents have occurred with respect to vaccination and human tissue, blood or body fluid-derived proteins: all of them have been related to virus contamination and infection [ 11. A serious concern for rDNA-derived proteins from recombinant mammalian cell cultures and monoclonal antibodies derived from hybridoma cell cultures is the potential risk of contaminating retroviral particles or adventitious virus [ 2 , 3 ] .Although any cell line considered to become a master cell bank (MCB) for manufacturing is extensively tested for any potential infective virus, the presence of an unknown but potentially harmful virus cannot be excluded, even despite the result that no viral activity is detectable by various assays. However, the presence of virus-like-particles (VLPs) detected by electon microscopy can be demonstrated for hybridoma cells and many recombinant mammalian cells [4]. Hence, prevention from virus infection at the stage of cell culture is mandatory, but in addition relevant purification steps in downstream processing have to be investigated and finally validated with respect to their capability to clear a potential virus contamination.
16.2 Strategies for Viral Safety 16.2.1 Cell Bank System The use of virus-free cell lines and the prevention of virus contamination during the manufacturing process are important strategies for drug safety. The dominant source of virus is the origin of the biopharmaceutical product. Human blood or blood- and tissue-derived products carry the highest risk for transmission of human pathogens. This is the most important reason to use non-human cell lines as origin for biopharmaceutical production. The concept of cell banking (Fig. 16-1) [ 5 ] is to have a
466
16 Validation of Viral Safety for Pharmaceutical Proteins
+selection of host cell line
I 1
expansion characterization
manipulation (transfection,cloning) expansion characterization production cell line
-I/-:-------. expansion
~-
production runs
'
1
runs
1
l I
runs
!
t extended Cell Bank
characterization
Fig. 16-1. Cell banking strategies for the production of biopharmaceuticals.
homogeneous pool of characterized cells from which each production run will start. The favorable concept of cell banking is to have different cell banks: a first choice is the host cell line. This host cell line should be a continuous non-human cell line that is suitable to the desired fermentation process (i.e. adapted to serum-free conditions and suspension growth). The host cell line can be tested for contaminations, like mycoplasma, microbiological contaminations, adventitious virus and retrovirus. The majority of actually used host cell banks are murine hybridomas and recombinant Chinese hamster ovary (CHO) cell lines. Nearly all the murine hybridomas were shown to carry retrovirus or retrovirus-like particles, whereas only a small por-
16.2 Strategies f o r viral
Safety
467
tion of the recombinant CHO cell lines harbour non-infective, retrovirus-like particles. Under no circumstances it is acceptable that designated production cell lines are contaminated with microbiological and infective adventitious agents. The host Cell Bank is the source for further manipulations (i.e. fusion with antibody production cells, transfection for the generation of recombinant products, cloning). After the creation of a production cell line by different kinds of manipulations, it is a further milestone in the development of a biophamaceutical to generate a Master Cell Bank (MCB). This MCB is a homogenate pool of the production cell line, frozen and stored in aliquots in liquid nitrogen by which each aliquot is representative for each other vial. The MCBs have to be characterized extensively for virus contaminations (Tables 16-1 and 16-2). Because of the large diversity of viruses and the lack of a test system or marker to detect all viruses, various assays have to be performed to demonstrate the absence of virus contaminations in the MCB (and, though not discussed here, in the final product or samples of fermentation harvest). Once the MCB has been found suitable for the production of biopharmaceuricals Working Cell Banks (WCBs) can be generated directly from the MCB. The WCB system guarantees the availability of the production cell line for the prospective life span of the product on the market. Again, the WCB should be tested for virus contamination according to the tests performed for the MCB or to a lesser extent, since the WCB originated directly from a fully characterized MCB. The extend of the characterization for WCB should be decided specifically according to the results of testing the MCB and the nature of the cell line. Table 16-1. Spectrum of tests for the characterization of Cell Banks. Adventitious virus test (infectivity in target cell lines) 0 human diploid cells 0 murine embryo cell line 0 human cell line 0 bovine cell line 0 production cells In vivo test (infectivity in animals) 0 suckling mice 0 adult mice 0 guinea pigs 0 fertilized eggs Specific virus tests 0 MAP (HAP; RAP) 0 different PCRs Retrovirus tests 0 co-cultivation 0 R.T.ase assay 0 PCR Electron microscopic examination
468
I6 Validation of Viral Safety for Pharmaceutical Proteins
Table 16-2.Specific virus detection by antibody production tests. Virus
Human pathogen
Hantaan Reo 3a Lactic dehydrogenase (LDHV) Extromelia (murine pox) Minute virus of mice (MVM) Mouse adeno Mouse hepatitis (MHV) Pneumonia virus of mice (PVM)a Sendaia Epizootic diarrhea of infant mice (EDIM) Mouse cytomegalovirus (MCMV) K-Virus Mouse thymic Murine encephalomyelitis (GDV 11) Lymphocytic choriomeningits (LMC)a Polyoma
0 0
Replication in human cell lines
0 0 0 0
0 0 0 0 0 0 0
0 0 0
Detection by mouse antibody production (MAP) test except for adetection by hamster antibody production (HAP) test.
Finally, the next cell banks that should be generated are the Extended Cell Banks (ECBs): at the end of a production run or beyond the maximal cell generations needed for the production such a new cell bank has to be generated. This ECB again needs to be characterized. The aim of this characterization is to prove whether a non-detectable virus was induced under the production conditions. The results of the MCB and WCB testing will give guidance for the range of tests for the ECB.
16.2.2 Biological Raw Material The remaining sources of a potential viral contamination (besides the production cells) could find their way into the production by the raw material for the cell culture media and by handling of the production cells during the manufacturing process itself. The prevention of such adventitious virus contamination could be prevented by the selection of the raw material and by the control of the manufacturing process according to the current Good Manufacturing Practice (cGMP). It is state-of-the-art to use serum-free media for the generation of cell banks and for the manufacturing of biopharmaceuticals. For many processes, developed before the emergence of the presently available techniques, serum (i.e. FCS) has to be used. It is not easily feasible to reformulate processes on the basis of new technologies, since this can imply additional preclinical or clinical testing and a new registration of the product. However, the suppliers of this serum batches impose strong selection
16.3 Virus Clearance
469
criteria, and each batch of serum has been tested for the presence of bovine viruses (i.e. BVD, P1-3, Parvo, leukemia). This is also important for other biological raw materials to be used (i.e. insulin, transferrin, serum albumin). In addition, the manufacturing process for such biological raw materials should be validated for its ability to remove efficiently and to inactivate potential virus contaminations. In case of biological raw material from bovine, sheep, or goat origin, where no test for the presence of TS-agent (transmissible agents) is available, the source of the material is of prime importance. No material from countries with a high prevalence of TS-agents should be used for the production of pharmaceuticals. Also, the tissues of infected animals are not equally infectious. The potential infectivity of the tissue used is, therefore, another important factor in risk assessment.
16.2.3 Current Good Manufacturing Practice (cGMP) The biotechnical production processes and operations have to be performed in compliance with cGMP [6-81. One of the major functions of cGMP is to prevent the introduction of adventitious virus contamination during the generation of cell banks and the manufacturing of biopharmaceutical product. The employment of trained and skilled personnel, the use of appropriate equipment as well as defined buildings and production areas are important prerequisites to prevent virus contamination. In all production areas a clean environment and a high level of hygiene by the operators is requested. The personnel and material flow should be restricted and regulated, and the process should be performed in closed systems. Only ultraclean water systems should be used, and where applicable the airflow should be HEPA-filtered. The environment of the manufacturing areas should be classified and monitored accordingly. All operations including testing should be performed according to documented Standard Operation Procedures (SOPS) in order to guarantee the quality, safety and efficacy of each individual biopharmaceutical product. Although the Cell Bank has been characterized, the biological raw material used for maintaining the production cells in culture must be qualified and the manufacturing processes of biopharmaceuticals must be in compliance with cGMP. The purification process has to be validated for the capability and capacity to remove and/or inactivate potential virus contaminations [9-181.
16.3 Virus Clearance A clear distinction must be made between virus inactivation and virus removal.
470
16 Validation o f Viral Safety for Pharmaceutical Proteins
16.3.1 Virus Inactivation The goal of virus inactivation is the irreversible loss of any viral infectivity. The loss of infectivity does not necessarily mean complete destruction or disintegration of the virion - which might be most desirable - but is already given by the irreversible denaturation of distinctive viral components that are essentially required by the virion in order to be able to infect a host cell. A prominent example is the solvenddetergent treatment of enveloped virus: the lipoid membrane-type envelope, which is essential for host cell docking, is dissolved, but the viral nucleocapsid is left intact, i.e. the viral protein core as well as the viral genome is still fully functional only the capability to attach to and penetrate a host cell membrane (i.e. the infection) is lacking. By introducing the nucleocapsid into a host cell artificially by microinjection, the virus reproduction can be initiated.
16.3.2 Virus Removal The goal of virus removal is the (mechanical) reduction of viral particles in number. Complete viral particles are functional and infective. There are several methodologies widely used for the mechanical removal of such viruses. Separation of virus from product might be achieved by adsorption of the protein product to a chromatographic matrix, while leaving the virus in the flowthrough, or vice-versa. Another approach is the retention of viral particles by means of filtration, either depth filtration (20-40 nm pore size) in a dead-end mode or ultrafiltration (cut off 5 300 kD) in a tangential-flow mode.
16.3.3 Regulatory Acceptability of Methods for Virus Clearance Validation of Virus Clearance The concept of validation [19-211 applies to all operations that are devoted to remove or inactivate virus. It must demonstrate the reliability of the respective unit operation in order to achieve the required product quality in a reproducible manner. The validation procedure is a complex of different routines which are complementary to each other (Table 16-3). The initial procedure for validation of process hardware is an installation qualification (IQ) which verifies the equipment installation, the inspection of utility connections, the compliance with given specifications and standards, and compliance of the actual design with the purchase specifications. Operational qualification
16.3 Virus Clearance
471
Table 16-3. Components of the validation procedure.
Standard operating procedures (SOPS)
Standard test procedures (STPs)
Process qualification: Installation qualification (IQ) Operational qualification (OQ) Performance qualification (PQ)
Process validation: Process description Risk analysis Data evaluation and documentation
(OQ) includes purpose and design characteristics of the installed equipment, acceptance criteria for system function and performance, as well as the determination of process parameters and process monitoring. Performance qualification (PQ) has to demonstrate the reliability of the equipment with respect to an effective and reproducible process leading to a (intermediate) product which meets all set release criteria. Standard operating procedures (SOPs) have to be defined for any handling of the equipment, including set-up, operation, regeneration, cleaning, and sanitization or sterilization, as well as maintenance and calibration. Standard test procedures (STPs) describe respective inprocess controls ; accordingly, specifications and rejection criteria have to be defined. The validation of the unit operation itself consists of a detailed process description, a risk analysis of the process, and evaluation and documentation of process data. The rationale for a validation of virus removal and inactivation is based upon possible hazardous side effects of potential viral contamination of a biotechnically manufactured protein product derived from mammalian cell culture, intended for use as a therapeutical drug or an in vivo diagnostic reagent. Protein biosynthesis in mammalian cells features glycosylated proteins of high molecular weight with no disulfide bridge limitations and secretion of the protein as a singular, native molecule, which is folded correctly to a complex tertiary structure; however, a potential retrovirus or adventitious virus contamination might lead to concerns on product safety. Even if no virus infectivity or reverse transcriptase activity can be detected for the MCB or ECB, the presence of virus-like particles (VLPs) can often be demonstrated by electron microscopy [22,23]. The microscopic evaluation gives no answer to the biological relevance of suspicious particles, especially regarding their infectivity. A prominent example is the presence of high numbers of A-type particles in hybridoma cells, where infectivity is extremely low or not detectable at all. Despite this discrepancy between the number of virus-like particles and infectivity, it is the general recommendation to calculate the overall reduction factor based on the particle number. Such an imbalance might induce the implementation of additional expensive but possibly unnecessary process steps. Nevertheless, the presence of a virus of unknown origin cannot be excluded. An unidentified virus might have unknown and potentially harmful physiological effects, and it is the unknown nature of the virus contaminant which complicates the development of a specific assay. Without a specific and sensitive assay it is impossible to monitor the presence as well as the removal or inactivation of the virus along the downstream process of the protein drug. With regard to a potential
472
16 Validation o f Viral Safety for Pharmaceutical Proteins
contamination by an unknown virus, a number of preventive measures must be established around and implemented into the complex manufacturing process of the protein in order to exclude the presence of any viral particle in the drug substance. Preventive measures at the level of cell biology include extensive testing of the producer cells for specific viruses and testing for adventitious virus at a number of stages during fermentation. The potency and efficacy of downstream operations for the removal or inactivation of virus must be demonstrated for the entire downstream process by the individual validation to those process steps that are considered to contribute to virus safety.
Model Viruses The selection of model viruses for the purpose of validation is critical and must take into account the nature and origin of the producer cell line. The model virus should be close to identical to a virus suspected in the cell line, or closely related to viruses that might infect the cell, e.g. retroviruses for recombinant or hybridoma cells. In order to achieve a maximum reduction factor for virus, the model virus should be grown to high titers and should be detectable in a simple but sensitive assay. Care has to be taken when concentrating a virus solution in order to increase the volumetric liter: the aggregation of viral particles might lead to an increased but not relevant mechanical removal by means of filtration, or a decrease in inactivation due to protection of viral particles in the core of the aggregate. Typically, the removal or inactivation of virus is described as ‘log reduction’, i.e. the decadic logarithm of the titer reduction. With regard to a potential infection by an unknown virus, the model viruses for the validation of virus clearance have to cover a broad range of virus features (Table 16-4 and 16-5) [12,14]. As a rule, process validation has to be performed at the manufacturing scale, using equipment of identical type and size. The contamination of manufacturing equipment with virus is evidently not desirable and, depending on the manufacturing scale, even not technically feasible [27]. Therefore, two strategies are accepted for the validation work with virus: the use of a replica for a small-scale production, or a linear down-scaled model of the manufacturing equipment, on condition that all relevant process parameters (Table 16-6) are fully representative for the production scale.
Table 16-4. Selection criteria for the choice of model viruses. Virus features
Feasibility
Size and shape of virus Envelopednon-enveloped Genome structure DNA/RNA Strandedness of genome Resistance to inactivation
Relevance for production Achievable high titer High sensitivity of detection Ease of detection
16.3 Virus Clearance
473
Table 16-5.Selection of model viruses for validation studies. Virus
Family
Genus
Natural host
Genome
Enve- Size lope
Shape
Resistamea
Vesicular stomatitis virus
Rhabdo
Vesiculo- Equine virus Bovine
RNA
yes
70x175 nm
Bullet
Low
Parainfluenza virus
Paramyxo
Paramyxovirus
Various
RNA
yes
100-200+nm
Pleomorph Spherical
Low
MuLV
Retro
Type C oncovirus
Mouse
RNA
yes
80-110 nm
Spherical
Low
Sindbis virus
Toga
Alphavirus
Human
RNA
yes
60-70 nm
Spherical
Low
BVDV
Flavi
Pestivirus
Bovine
RNA
yes
50-70 nm
Pleomorph Spherical
Low
Pseudorabies
Herpes
Poikilovirus
Swine
DNA
yes
120-200 nm
Spherical
Medium
Poliovirus Picorna Sabin Type 1
Enterovirus
Human
RNA
no
25-30 nm
Icosahedral Medium
Encepha- Picorna lomyocarditis virus (EMC)
Cardiovirus
Mouse
RNA
no
25-30 nm
Icosahedral Medium
Reovirus 3 Reo
Orthoreovirus
Various
RNA
no
60-80 nm
Spherical
SV 40
Polyoma- Monkey DNA virus
no
40-50 nm
Icosahedral Very high
Parvovirus
no
18-24 nm
Icosahedral Very high
Papova
Parvovirus Parvo (canine, porcine) a
Canine Porcine
DNA
Medium
Resistance to physico-chemical treatments based on studies of production processes. Resistance is relative to the specific treatment and it is used in the context of the understanding of the biology of the virus and the nature of the manufacturing process. Actual results will vary according to the treatment. These viruses are examples only and their use is not mandatory. (From [12].)
474
16 Validation of Viral Safety f o r Pharmaceutical Proteins
Table 16-6. Evaluation of down-scale models for validation purposes. Capacity and life time for membranes and chromatographic matrices Product quality before and after each unit operation Reliability of process parameters Chromatography:
Filtration:
buffer conditions, pH, conductivity linear flow rate back-pressure product load dynamic capacity selectivity resolution yield OD profile
buffer conditions, pH, conductivity flow rate, crossflow rate filtratehetentate flux recirculation rate inledoutlet pressure transmembrane pressure product load yield
16.4 Calculation of the Clearance Factor The downstream processing of biotechnically manufactured proteins derived from mammalian cell culture and other pharmaceutical products of biological, i.e. animal or human origin, provides a clearance factor for the removal and inactivation of potential viruses. The measures which are applied in order to demonstrate the freedom from infective virus are conceptionally comparable with the proof of sterility [24,25]. Absolute sterility of a product solution can only be demonstrated for the case that the entire production lot undergoes sterility testing, and that the test method is capable to detect all potential contaminants - but then no product would be left. Therefore, it is accepted practice to test fractions of the production lot based on a statistical approach, where the result will not be absolute but will provide a reasonable probability of freedom from a contaminant. According to the well-established methodology for sterility testing, the demonstration of an absolute exclusion of any potential virus is not pursued (and, of course, is technically even not feasible), but viral clearance provides for an acceptable limitation of the probability for a potential virus contamination of a single product dose. Hence, the manufacturing of pharmaceutical proteins is accepted, even though the occurrence of a virus contamination cannot be excluded absolutely for an unlimited time. Much more, it addresses the judgment on the acceptability for a level of drug safety and on a statistically evaluated probability for potential but extremely rare events.
16.4 Calculation of the Clearance Factor
415
16.4.1 Determination of the Virus Titer Effectivity of virus clearance is determined by spiking experiments for the respective unit operations. Virus distribution is monitored and balanced for the individual intermediates of such an operation: the virus titer of the load is measured and compared with the (residual) virus titer of the product containing fraction after processing, e.g. the flowthrough or eluate of a chromatographic process or the permeate of a filtration process. Prior to the titration of process samples the potential effect of the applied buffer solutions has to be investigated regarding interference with the detector cells or reduction of infectivity of the model virus. Typical detector cells for the titration of viruses are SC-1 cells (retrovirus), CV-1 cells (SV 40), L 929 cells (reovirus) and Vero cells (PI3). For an extended detection for retroviruses the XC plaques assay can be applied, where SC-1 cells are inoculated with the respective sample; after a defined period of cultivation the cell layer is UV-irradiated and overloaded with XC cells. After plaque formation, the cell layer is fixed, stained, and plaques are counted. With reference to the morphological shape of a retrovirusinfected cell monolayer, the titer is expressed in plaque- or focus-forming units (pfuFFU):
where X,, is the number of plaques or foci, and d is the dilution factor. Viruses not leading to the formation of plaques of foci are measured by their cytopathic effect (CPE) on the detector cells, and the titer is expressed as TCID50 (tissue culture infectious dose for 50 % of the entire cell number):
where
XO is the positive exponent of the highest dilution showing CPE; d ist the logarithm of the dilution step; n is the number of repeated assays; and X,i is the number of all virus-positive samples up to and including XO.
The balance of virus distribution throughout the complete process step, i.e including corresponding fractions such as washing and regeneration steps of chromatography or the retentate of filtration, is unlikely for most processes due to denaturation of virus by caustic solutions typically used for regeneration or due to capture of virus particles within the membrane matrix of a filter. The demonstration of virus clearance for a single validation experiment is limited mainly by two factors: the maximum available titer for the accepted viruses is in the range of lo7 to lo9 mL-'; it is further reduced by 1 log by the required spike of 1:10-20.
476
16 Validation of Viral Safety for Pharmaceutical Proteins
Another limitation is the technical difficulty to titrate the entire process fluids: the volume of a sample is dependent on the detector cell. Typically the titration is performed using a sample volume of 0.1-1.0 mL. A statistical approach [24] has to be taken in order to calculate the probability p that the investigated samples do not contain infective virus:
p = ( V - Vlv)n where Vis the entire volume of the process fluid; v is the volume of the sample; and n is the number of infective virus statistically distributed in v. Hence, samples taken from process fluids which contain a potential virus burden below 1 infective particle per mL may not contain any virus. Thus, an additional statistical evaluation might become necessary for the case that no virus can be detected in any of the samples. The assessment of the Posson distribution at its 95 % upper confidence limits is widely used as an appropriate methodology. The virus titer is determined for V > v, the probability p is calculated as
where c is the concentration of infective virus.
16.4.2 Reduction Factor The reduction factor of a designated unit operation is calculated upon the volumes of the process fluids and the virus titers measured for the load and the product-containing fraction after processing. The reduction factor is typically expressed in log 10 units, the 'individual reduction factor' R, for each unit operation is calculated as:
where Ri is the individual reduction factor; V1 is the volume of the load; los is the virus titer of the load; V2 is the volume of the processed product solution; and l o p is the virus titer of the processed product solution. The overall reduction factor for a virus within the entire purification process is the cumulation of the individual reduction factors. However, the cumulation of virus clearance can only be claimed for process steps which represent different physicochemical measures. Based on a cell assay variability which is measured in a logarith-
16.5 Evaluation/Assessrnent of Methods f o r Virus Inactivation
477
mic scale, a logarithmic reduction factor in the order of 1, i.e. a 90 % reduction in titer, is considered to be not significant for virus clearance. In order to set a numerical figure for the virus burden of a cell culture fluid at the time of harvest, electron microscopy (EM) is applied for counting viral particles in a distinctive volume. However, this approach raises additional questions: how representative can a sample of a few mL be for a few hundred liter- to several thousand liter-scale of cell culture fermentation? Furthermore, the cells in culture are grown to densities between lo6 to lo7 mL-'; the number of cells investigated by EM is reduced by several logs and is about lo3 mL-'. The identification of virus particles and their differentiation from particulate matter due to the preparation procedure requires extensive experience, and a confirmation of the viral nature is impossible. Examples of artefacts which originate from sample preparation include high-speed centrifugation of cell culture supernatant; the pellets derived from centrifugation typically habour complex aggregates which often conceal those details necessary to identify virus structures.
16.5 EvaluatiodAssessment of Methods for Virus Inactivation The course of virus inactivation can be measured and determined precisely in a respective kinetic study: applying the designated inactivation procedure, samples are taken at time intervals during the course of inactivation. Subsequently, sample aliquots are titrated on designated sensitive receptor cells in order to determine the residual infectivity. This clear and accurate experimental approach is fully accepted; virus inactivation procedures are first choice and are accordingly an essential integrative part of any downstream processing as well as for recombinant proteins derived from eukaryotic cells, especially mammalian cells or any human blood-derived products. The number/types of virus inactivation methods with respect to their compatibility to high-quality protein molecules for pharmaceutical use is limited. This fact, in combination with the need to demonstrate - and validate - a reduction of infective virus in a range of 12-15 log makes the additional use and application of reliable procedures for virus removal necessary - and is a serious reason why regulatory authorities as well as industry have to accept and rely on procedures for mechanical removal of virus.
16.5.1 Virus Inactivation Virus inactivation is preferably performed using harsh physico-chemical measures: acid treatment (destruction of nucleocapsid and genome), urea treatment (disintegration of nucleocapsid), solvent-detergent treatment (limited to dissolution of the
478
16 Validation of Viral Safety for Pharmaceutical Proteins
Table 16-7.Virus inactivationa by acid treatment and urea treatment. Incubation pH pH 3.0 pH 3.5 pH 4.0
SV-40 <1.0 <1.0 <1.0
Re03 <1.0 <1.0 <1.0
P13 >3.8 6.5 1.2
MuLV >3.0 >6.0 5.0
Urea 3.0 M 4.0 M
n.d.
2.6 4.6
5.3 5.1
>5.2 4.9
a
<1.0
All values given in log 10 units. n.d., not determined.
envelope of enveloped viruses [26], and heat treatment [27,28]. The addition of toxic or potentially harmful chemicals e.g. 0-propiolactone is not appreciated, as the agent has to be removed again, and expensive analytical testing must be done. Acid treatment as well as urea treatment works well with larger viruses, but virus particles <40 nm are inactivated only at very harsh conditions (pH < 3.0, 2 3 M urea) (Table 16-7), which are typically harmful to most complex protein molecules, i.e. typically proteins of a molecular weight >20 kDa. In addition to the determination of virus inactivation after a 2 -h incubation, kinetic studies provide a more expressive information on the course of virus inactivation (Figs. 16-2-1 6 - 4).
Fig. 16-2.Kinetic study on the inactivation of Murine Leukemia Virus at low pH.
16.5 Evaluation/Assessment of Methods for Virus Inactivation
Fig. 16-3. Kinetic study on the inactivation of P1-3 at low pH.
Fig. 16-4. Kinetic study on the inactivation of different viruses at pH 12.0.
479
480
16 Validation of Viral Safety for Pharmaceutical Proteins
.-t>
Time (h)
Fig. 16-5. Kinetic study on the inactivation of different viruses at 60 "C
Conventional heat treatment, originally being applied for human blood derived proteins, is an incubation of the protein solution at 60°C for 10 h. Again, small viruses of <40 nm are typically not significantly affected (Fig. 16-5). HTST heat treatment using microwave energy has dramatically changed the 'heat option'; due to radiation microwave heating allows for an instantaneous heat transfer in a continuous flow, there is no temperature gradient as with heat exchangers. Sensitive molecules like pharmaceutical proteins are exposed for 100-1000 ms to high temperatures, hold times at peak temperatures are reduced to 1-3 ms. HTST microwave heat treatment is performed using the commercially available 5 kW continuous flow UltraThermTMUTSAB 1, which was developed and brought to large scale cGMP manufacturing in a close collaboration between Charm Bioengineering, Malden MA, USA and Boehringer Ingelheim Pharma KG, Biberach, Germany. The UltraThermTM features a temperature range of 60-165°C at flow rates of 35-80 Lh-' and includes full data acquisition. An injection loop allows for validation work under full scale operating conditions, the flow path is disposable thus enabling convenient product change over and validation. Heat denaturation of the protein molecules does not occur even at temperatures far beyond the melting temperature determined by DSC (differential scanning calorimetry) (Fig. 16-6). Protein precipitation is the first denaturing effect which can be observed: an appropriate selection of a respective carrier fluid (buffer) is demanding and will contribute significantly to expand the operational temperature range. Smaller and more robust protein molecules are feasible to temperatures which inactivate high resistant parvovirus. As an example, rt-PA (recombinant tissue plasminogen activator) could be processed at temperatures up to 120°C (Fig. 16-7) while main-
48 1
16.5 EvaluatiodAssessment of Methods for Virus Inactivation
100
4
TM 1 84.6%
60 70 Temperature ( aC )
50
80
Fig. 16- 6. Differential scanning calorimetry (DSC)for the determination of the melting temperature of a humanized IgG-type monoclonal antibody (huMAbl). The huMAb was applied to DSC at a concentration of 1.74 mg mL-' in PBS at a scan rate of 45 "Ch-l. TM1 represents the melting temperature of the Fab portion, TM2 the melting temperature of the Fc portion of the antibody.
.*.
-
- r - rt-PA vadrnf 4.6~glmI
-.
*%.*
*.l)y...
rL-PA variant intermediate,0.8rg/mI -rt-PA
-
20
*.
f o w l a t e d bulk, 2.5rrq/nl
rt-PA Intefmediate, 2.3mglnJ
80
85
. *X..
._....
90
......_
95
.....
100
..
-.,
.+
...
105
peak temperature ("C) Fig. 16-7. Aggregation formation of pharmaceutical proteins during HTST microwave heating. rtPA and a rt-PA variant of pharmaceutical grade have been processed at a constant flowrate of 60 Lh-' through a 50" coil. The exposure time to elevated temperature was about 340 ms, hold time at peak temperature was about 2 ms. The data show that a careful screening both for appropriate buffer conditions (i.e. different buffer salts, salt molarities, conductivity and pH-value) and protein concentration is crucial for HTST processing.
482
16 Validation of Viral Safeiy for Pharmaceutical Proteins
taining is monomeric state and full biological activity, i.e. fibrin binding and plas- . minogen activation. The melting temperature for rt-PA in DSC was determined with 69.87 "C, at 2.5 mg mL-' and a scan rate of 60 "C. Even large protein molecules like immunoglobulins can be treated up to temperatures, which are already destructive for small non-enveloped viruses. Humanized monoclonal antibodies (huMAb1, IEP 7.0-7.6 and huMAb2, IEP 8.8-9.3) of pharmaceutical grade have been processed at a constant flowrate of 60 Lh-' through a 75" coil. The exposure time to elevated temperature was about 509 ms, hold time at peak temperature was about 2 ms. DSC resulted in a melting temperature TM1 for huMAb1 at about 64.6"C(Fig. 16-6) and for huMAb2 at about 71.6"C. The data show the significant influence of salt molarity and pH-value on the heat resistance of proteins during HTST processing. Organic buffers such as citrate buffer were identified to be superior to phosphate buffered saline. Due to the basic IEP's of both proteins, a slightly acidic pH was favorable. (Figs. 16-8 - 16-10). The technical performance of the UltraThermTM was demonstrated by processing clinical grade huMAb2 derived from 2000 L as well as 10000 L mammalian cell culture under cGMP (Fig. 16-11). The total residence time of the virus at elevated temperatures contributes to the efficacy. Experimental work with SV-40 showed that a two-fold increase in exposure time reduced the required inactivation temperature by about 5 "C (Fig. 16-12).
95
e Q)
8
2
85
L
Q)
E
80
5
75
K
- 0
0.25 M Citrate
- *- 0.1 M Citrate
8
.
65 20
75
7a
80
Peak Temperature ( "C) Fig. 16-8. Aggregation formation of pharmaceutical proteins during HTST microwave heating. Processing of huMAb1, influence of buffer molarity at pH 5.5.
16.5 Evaluation/Assessment o f Methods for Virus Inactivation
483
100
95
2 $
90
2
85
b E
80
z
75
8
s
0.1 M Citrate pH 6.0 -0.1
M Citate pH 5.5
70
65 75
80
85
95
90
100
Peak Temperature ( "C ) Fig. 16-9. Aggregation formation of pharmaceutical proteins during HTST microwave heating. Processing of huMAb2, influence of buffer pH.
6
5
c 4
6
s3
f!
CI)
0
2
1
0 60
65
70
75
80
85
90
95
100
peak temperature ("C) Fig. 16-10. HTST heat inactivation of virus. Different viruses have been investigated for inactivation at a constant flow rate of 60 L h-I through a 125 cm (50") coil. The exposure time to elevated temperatures was about 340 ms, hold time at peak temperature was about 2 ms. Interestingly, Mycoplasma shows a heat resistance superior to that of enveloped viruses.
484
16 Validation of Viral Safety for Pharmaceutical Proteins system start equilibration processing I
end of process I
I
63
62
I
I
II
iE
W
s
~
flow
61 b= 60
59 0
80
120
180
240
300
360
420
540
480
800
880
720
time (min) Fig. 16-11. HTST processing. The UltraThermTM was challenged under full scale continuous processing conditions at a set peak temperature of 90°C and a constant flow at 60 Lh-'. The system realized an extremely narrow range of the set process parameter. For HTST virus inactivation, the UltraThermTM is equilibrated for about 30 min with buffer, then switched to product processing.
7,
04 60
65
70
75
M
m
65
peak temperature [deg C]
..*. .22"coll
-75"Coll
$5
100
105
110
16.6 EvaluatiodAssessment of Methods for Virus Removal
485
16.6 EvaluatiodAssessment of Methods for Virus Removal As already mentioned above, two principal methodologies within the downstream processing of proteins are capable to remove virus; namely chromatography and filtration.
16.6.1 Chromatography Chromatographic procedures are highly effective for the separation of proteins where the physico-chemical properties of the proteins to be separated determine the type of chromatography to be applied. The virus capsid is a shell consisting of different proteins; also the envelope of enveloped viruses contains a significant number of different proteins. Due to this proteinaceous nature, virus might behave during chromatography as any other 'contaminating' protein - the virus may or may not bind to the chromatographic matrix under the given physico-chemical environment. In consequence, a respective study on virus removal by means of chromatography might demonstrate that - with respect to an arbitrary selection of model viruses - there will be always a certain virus which binds to the matrix while under identical conditions another virus will pass the matrix. As the experimental work for the validation of virus removal is limited to a reduced number of viruses, the proof - not even the evidence - is lacking that the chromatographic unit operation will significantly contribute to the removal of an unknown but potentially harmful virus.
Data on Virus Removal using Chromatography As already mentioned, the chromatographic approach regarding virus removal must be dealt with critically and carefully. The major concern - namely the vast variety of virus species and the heterogeneous protein composition of their capsid and/or envelope - is even worse in light of chromatographic theory: separation and resolution are not only dependent on the physico-chemical conditions of the buffer solutions, which have been determined as the result of development of this unit operation, but are also dependent on the composition of the product solution with respect to its contaminants - proteins and lipids/lipoid substances.
Fig. 16-12. HTST heat inactivation of SV 40. rt-PA solution (5 mg mL-' - spiked with SV 40 to a resulting titer (log 10 mL-') of 7.89 i: 0.56 - was processed at a constant flow rate of 60 1 h-' through coils of two different lengths: 5 5 cm ( 2 2 " ) and 190 cm (75"), resulting in exposure times of about 150 ms and 500 ms respectively, hold time at peak temperature was about 2 ms (Ultra T h e,m' T UTSAB 1, Charm Bioengineering, Malden, MA, USA). In all cases, no residual SV 40 infectivity could be detected for temperatures beyond 90 "C; the decrease in slope of the curves is a result of the statistical calculation according to Poisson distribution.
486
16 Validation of Viral Safety for Pharmaceutical Proteins
Table 16-8. Data on virus removal using affinity chromatography. Matrix
SV-40
Re03
P13
MuLV
Protein A Protein G Lysin
2.1 2.0->6.0 4.0
<1.0
3.2 3.0->6.0 6.5
4.0 3.0->6.0 >6.3
~
2.0a6.0 1.1
~~
Affinity chromatography as a first step to purify the product is most attractive, as the product might result in a purity >95 %. The basis of affinity chromatography is the biospecific interaction between the product and a dedicated ligand. However, even this biospecific interaction does not exclude non-specific interactions of the chromatographic matrix with impurities, i.e. also with virus (Table 16-8). Evidently, affinity chromatography still is the exception in the downstram processes due to the lack of appropriate ligands; Protein A and Protein G are wellknown ligands dedicated to antibodies (immunoglobulins). However, a potential disadvantage is the problem of ligand bleeding as well as a limited resistance to harsh regeneration procedures using 1.0 M NaOH. A rare example for robust, small molecules effective as affinity ligands are the amino acids arginine and lysine, both binding plasminogen activator selectively [29]. This situation is actually changing by the approach and application of combinatorial chemistry or phage display, both allowing for a rapid and reasonable cost effective screening of thousands of potential ligand molecules [30,311. With a reduced specificity - and this is the case with all the other chromatographic techniques, of which ion exchange chromatography is by far the most important the probability of protein interaction with other than the product protein increases significantly (Table 16- 9). The broad variety within the data of virus removal by chromatographic methods illuminates the concern: the potency of a single chromatographic separation to remove a complete set of model virus is neither evident, nor does it seem to be feasible. Of course, certain viruses are effectively removed, but others are not. This is exactly in serious contradiction to the principal concept why ‘model’ viruses have been selected - namely to cover a maximum range of different virus features. It is not even possible to relate the chromatographic separation to distinctive classes of viruses as it is possible for filtration processes (limitation by particle sizes) or inactivation processes. Chromatographic separation of viruses from protein products appears to be unpredictable random. Nevertheless, chromatographic processes are of value if the concern of virus contamination can be affected onto a distinct virus. Midstream of the downstream process, i.e. already at the second and potential following chromatographic steps, the product solution is typically 2 9 0 % pure - and separation is focused on this purity, residual contaminants are becoming ‘trace’ impurities. Spiking of such a pure product with a virus solution for the purpose of validation by 1 : l O to 1:20 definitely means changing the starting conditions dramatically. Typically the spiked virus solution is derived from a lysed cell culture, centrifuged at low speed to remove cell membranes, but to avoid the aggregation of viral
16.6 EvaluatioidAssessrnent of Methods f o r Virus Removal
487
Table 16-9. Data on virus removal using ion exchange, hydrophobic interaction, and gel permeation chromatography.
Ion exchange
SV-40
Re03
P13
MuLV ~~
<1.0 - 2.0 <1.0 - 4.8 <1.0 - 2.7 3.5 - >6.7
<1.0 - >6.6 3.0 - >6.0 3.0 - >6.0 5.3 - 6.1
<1.0 - 3.2 <1.0 - >6.0 1.5 - >6.8 3.0 - 5.3
<1.0 - >3.0 3.0 - >6.0 3.0 - >6.0 >6.3 - >6.7
Hydrophobic
SV-40
Re03
P13
MuLV
Phenyl Octyl Pyridyl
<1.0 - 1.8 1.1 1.8
<1.0
2.4 <1.0
<1.0 - 2.3 1.9 2.6
4.5 2.3
Gel permeation
SV-40
Re03
PI3
MuLV
Superdex 200 pg Superdex 75 pg
1.8 - 2.0 2.6
1.3 - 2.5 1.0 - 2.6
1.4 - 1.6 2.9 - >3.9
3.3 - 3.4 3.0 - 3.3
Cation Anion Anion Anion
(-SO3)
(-DEAE) (-N(CH3)3) (-TMAE)
~
a
<1.0 - 2.3
All values given in log 10 units.
particles. Due to the moderate g-forces applied, the supernatant still contains a huge amount of proteins and lipids. Hence, such a spike is of significant impact on the behavior of the original product solution, leaving sometimes a part of the product in those fractions which under manufacturing conditions are typically free of product, e.g. by displacement effects. As virus itself is loaded at particle numbers of lo6-los mL-', exhibiting a total different mass balance regarding the impurities and an amount that is far beyond any physiological sense but absolutely necessary in order to obtain appropriate log numbers for removal. The question is left open on how a single or a few viral particles would behave in the original product solution. These contradictions and difficult assessment of the chromatographic efficacy are well recognized both by the process scientist and the authorities. Chromatography as a reliable measure for the removal of virus is dedicated to some few individual techniques. The widely used ion exchange chromatography and hydrophobic interaction chromatography will be probably not a competitive methodology for the efficient and reliable removal of virus particles as individual unit operations. Taking an orthogonal approach, where the different selectivities of various chromatographic modes are assumed to contribute to an average removal of the broad variety of viruses, chromatography might be considered to represent an additional safety aspect, exceeding the required safety level of a 9-12 log virus clearance given by other, more reliable methodologies for viral clearance.
488
16 Validation o f Viral Safety for Pharmaceutical Proteins
16.6.2 Filtration Filtration of virus requires membranes with pores in a size range which is appropriate to retain viral particles. The viruses of interest cover two distinct size ranges: retrovirus is a spherical particle of about 100-140 nm in diameter and represents the most relevant virus type for mammalian cell cultures, whereas small viruses such as hepatitis and polio are about 30 nm in diameter. These types of viruses, together with retrovirus (HIV!), are the most relevant viruses for human blood derived-proteins. The two modes of filtration, i.e. nanofiltration and ultrafiltration, not only differ with regard to their filtration principle, but in addition they represent different ranges of pore sizes or molecular weight cut-offs (MWCO). Nanofiltration applied as depth-filtration is typically performed using a conventional cartridge design for microfiltration, and is operated in a dead-end mode of filtration. Nanofilters featuring pore sizes in the range of 20-70 nm utilize the classical depth structures, which are typical for commercial microfiltration membranes. Tighter pore sizes become difficult to develop and use in practise, because the back pressures would exceed dramatically a value of >0.3 Mpa. Additionally, the filtrate flux declines as the depth structure gets blocked by aggregates and fine particles. Due to this mechanism, the application of nanofiltration is preferred for virus with apparent diameters >40 nm. Typical harmful viruses which are smaller, and are found in contaminated blood (Hepatitis, Polio, etc.) are most challenging, hence require the definite lower range of pore sizes in order to remove small viruses to a reasonable extent (i.e. >3 log) in a reliable manner. Clearance of viruses > > 4 0 nm has been demonstrated to be very effective. Nanofiltration is inexpensive in practical use, and can be placed at any stage of the downstream process as long as the loaded solution is 0.1 ym prefiltered. A significant feature of such nanofilters is the possibility of measuring their integrity. Applying ultrafiltration (UF), an extensive range of different cut-offs is available, limited mainly by the size of the protein mocule of desire (‘Stoke’s radius’), as the protein must pass the membrane, while the virus is retained. A maximum MWCO for the ultrafiltration of virus is in the range of about 300 kDa, a cut-off that allows for the passage of nearly all relevant proteins, including antibodies, while virus particles >>40 nm are retained significantly; however, clearance of small virus <40 nm might be limited. A MWCO of about 200 kDa increases the clearance dramatically, still allowing for the passage of IgG-type antibodies (MW 150 kDa). A MWCO below 200 kDa features a significant retention of small viruses <40 nm. For its Omega 100K VR ultrafiltration membrane, Pall-Filtron reports a 4.76 log reduction for porcine Parvovirus. However, this cut-off limits the size of the applied proteins. A nominal diameter for such proteins cannot be given, as their geometry might vary (spherical, longitudinal) and size of the protein molecules (Stoke’s radius) is very much dependent on the physico-chemical conditions of the buffer solution. Depending on the process parameters, even of proteins such as IgG-type antibodies with a MW of 150 kDa can be filtered through a nominal 100 kDa ultra-
-
16.6 Evaluation/Assessment of Methods for Virus Removal
489
Table 16-10. Distribution of BSA and IgG2 after ultrafiltration using a 100 kDa and a 150 kDa UF-membrane. BSA
Membrane Omega 100 kDa Omega 150 kDa
IgG
Retentate
Filtrate
Retentate
Filtrate
4% n.d.
96 % n.d.
22 % <1%
78 %
>99 %
~
n.d.. not determined.
filter (Table 16-10). Yet, a MWCO of 100 kDa or perhaps a 150 kDa may be applicable to plasma protein products having a range of approximately 30-80 kDa MW for sufficient removal of viruses <40 nm. It should be noted that the determination of the MWCO is no absolute measure. Hence membranes of nominal identical MWCO from different suppliers might result in significantly different separation: A/G Technology’s hollow fiber VirA/Gard 500 viral removal membrane is of nominal 500 kDa cut-off, but compares most likely to Filtron’s Omega VR 150 kDa regarding filtration behavior in immunoglobulin processing. The technical operation of filtration processes and especially their validation, is not always simple as it may appear at first glance. There is on one side the concern of filtedmembrane integrity: extremely narrow pores which are necessary to obtain a respective virus rejection do not allw for a ‘bubble point test’ as used for microfilters: the pressure which is required to perform such a test for pore sizes of UFmembranes would be destructive for the membrane itself. Moreover, a bubble point test is only applicable to symmetric membranes due to physical reasons - it requires a minimum ‘tube’ length of the pores where air is entrapped. Most ultrafiltration membranes are asymmetric: the discriminating membrane surface layer determining the cut-off is limited to 1-5 pm, hence it is not accessible to the bubble point testing. Therefore, reputable membrane manufacturers provide testing procedures which are not ideal because of phyical limitations, but which are appropriate in combination with a validated and strictly controlled production process [32]. GMP audits by clients on a regular basis definitely contribute to the high manufacturing quality. A more serious concern for the validation of filtration processes is (partial) clogging of the membrane as well as the phenomenon of a boundary layer formation [33351. The virus spike of 1:lO-20 of a product solution is a significant burden of particulate matter. Not only are viral particles present at unphysiological high numbers of > lo7 mL-’, but also membrane residues and cell-derived vesicles represent a severe lipid/lipoid burden which might interact with the membrane surface. Thus, in a dead-end mode of filtration, this type of particulate might enhance the retention of viral particles to a degree that the experimental data on virus removal appear to be significantly higher than can be expected from the pore sizehirion diameter relationship. Evaluation studies using preparations of non-enveloped viruses which have been gently washed using low-concentration organic solvents such as 1- or 2-propano1 might be helpful to rate a filtration process with that regard; though of course the infectivity of the virus is essential and must be proven.
490
16 Validation of Viral Safety f o r Pharmaceutical Proteins
The mode of operation for UF is technically exact using expensive hardware. Process performance and validity can become a critical issue if UF is not operated under full control, or - much more so - if the theoretical background of the operation is not fully understood, inducing failure, especially with respect to a validation work in a downscaled model of the process. It is therefore imperative that any viral removal validation utilizing UF should follow a protocol that demonstrates viral removal as a function of classical operating parameters such as crossflow, TMP, and conversion ratio (%filtrate rate / %feed rate). This must also be incorporated as part of the overall validation of the physico-chemical nature of the macromolecular species. Tangential flow filtration (syn. crossflow filtration) is the basic principle for UF. The transmembrane pressure (TMP) in a UF system is the resulting force of differences between inlet pressure (Pi)and outlet pressure of the filtrate (PF) as well as of the retentate (PR)of the system and can be calculated accordingly:
Pi - PR
T M P = -- P F 2 The transmembrane pressure forces fluids and particles which are below the cut-off through the membrane pores, but forces molecules larger than the cut-off onto the membrane surface. Additionally, TMP induces a concentration polarization on top of the membrane surface. The shear forces created by tangential flow across the membrane surface support the removal of attached molecules, but the removal is not complete - and cannot be complete for physical reasons. The membrane surface is ‘covered’ with a limited region of static fluid, and the expansion of that region is determined by the applied crossflow velocity. The crossflow velocity again is limited by the system (pump) and protein product, which is typically not applicable to extreme shear forces. A system-conditioned pressure drop across the membrane might even lead to an unequal boundary layer. Depending on the material origin of the boundary layer, this layer might exhibit ‘pore sizes’ which are significantly smaller than the anticipated cup-off: the apparent virus removal is enhanced, or the ultrafiltration membrane even clogs completely (Fig. 16-13). Hence, it is indispensable that the filtration process has to be carefully monitored; it is mandatory that the product load/membrane area relation, crossflow velocity, pressures, ratio of flow rates for filtrate and retentate, and temperature must be identical to the designated manufacturing process. Data on Virus Removal using Filtration
Filtration is well accepted for the mechanical removal of particles. Quality and reliability have been demonstrated for sterile filtration using microporous membranes for decades. However, the removal of viral particles requires membranes with much smaller pore sizes. The acceptance is based upon the technical proof of the filter integrity prior to use and after use. Filtration techniques which are not feasible for integrity testing require additional features in order to obtain a warranty for the validity of results which have been elaborated in validation experiments.
16.6 Evaluation/Assessment o f Methods for Virus Removal
49 1
Fig. 16-13. (a) OD280 profile of a 200 kDa UF-filtrate showing an optimal permeation; (b) evidence of fouling at the end of filtration; and (c) membrane clogging. The processed fluid contained a monoclonal IgG2 at a concentration of 20 mg mL-'.
With regard to virus removal there are two principal filtration techniques: depth filtration through symmetric membranes, and ultrafiltration using asymmetric membranes. Both techniques are competitive, but could also be regarded as synergistic as the filtration principle is different and both filtrations can be claimed to remove virus additively; this approach has been accepted by the authorities for a number of products.
Nanofiltration There are actually a few different filter products for depth filtration on the market, claiming for the removal of virus particles: 1. 2. 3. 4. 5.
Pall Ultipor DV 20 Pall Ultipor DV 50, being the successor of the 40 nm filter Ultipor 40 Gelman Supor 30 Asahi Planova 15 Asahi Planova 35
So far, these filters are different in use and in acceptance: Ultipor DV 20, DV 50, and Supor 30 are available as 25 cm (10") filter cartridges and are easily placed at any step in the downstram process simply as an in-line filter, e.g. as the protective filter of a chromatographic column. The only prerequisite is that the applied solution is 0.1-0.2 ym prefiltered.
492
16 Validation o f Viral Safety for Pharmaceutical Proteins
With regard to virus reduction the respective numbers on log reduction given in the literature have to be judged upon the experimental conditions, e.g. for Pall’s DV 50 the clearance for polio in pure water is 3.2 X lo4 (as virus and protein tend to agglomerate/precipitate, forming significantly larger particles), in medium MEM + 10% FBS the clearance is low as 1.7 x lo2 [36]. A dreawback for this cartridge device might be the moderate filtrate flux (0.7-1 .O 1 min-’) at considerably high pressure (0.15 MPa), even for solutions with very low viscosity. With the DV 20, the filtrate flux is further decreased. Asahi Chemicals provide BMM-Membranes with a minimum pore size of 15 nm (Planova 15), featuring an effective virus removal for polio, SV 40 and Parvovirus [37]. The glass housings which harbour the hollow fibers are technically low standard with regard to a proper scale-up architecture. However, they can be operated both in a dead-end or tangential-flow mode. Asahi has developed an integrity test based on gold particles [38]; in addition, due to the hollow-fiber structure, Asahi has an appropriate down-scale model available. The numbers given in the literature indicate that nanofilters are highly effective if the virus size is significantly larger than the pore size, i.e. virus particles 2 4 0 nm can be cleared by several log numbers. As a result of the ease of use, such nanofilters are and will be widely used in downstream processes. The quotation for nanofilters differs significantly : while a 10” cartridge (effective filter area 1.63 m2) of Pall’s DV 50 filter is available at a list price of about 800 US $, which is still economically acceptable, Asahi’s Planova cartridge (1 m2) costs about 4000 US$. With respect to an acceptable process - the basis of any reasonable manufacturing process to be run for some years - such extreme prices are not necessarily encouraging to establish a manufacture processing which is based upon the validated use of such a filter device. Whether or not an ultrafilter - featuring a much tighter cut-off - will be used in addition, depends strongly on the risk evaluation on the product solution to be processed. For recombinant proteins and monoclonal antibodies, the virus of concern is a retrovirus, typically featuring a particle size of 100-140 nm. This virus size is definitely covered by nanofilters (Table 16-11). The virus problem regarding small virus particles is substantial for the blood industry: polio, hepatitis, and comparable small viruses in the range of 25-40 nm are a risk. These viruses will be retained reasonably by nanofilters featuring a respecTable 16-11.Data on virus removal using nanofiltrationa. Virus Filter
SV-40
Re03
P13
MuLV
Ultipor 40 Ultipor DV 50 Supor 30
c1.0 >5.8 <1.0
2.9 >l.O <1.0
>3.1 - >6.6 >6.8 1.5
>5.2 - >6.3
~
a
>5.9
>5.2 ~
_
_
_
All values given in log 10 units. As a process solution, PBS containing 2-10 mg mL-’protein was spiked 1:20 by a virus solution at a titer of lo6 to lo8 mL-’.
_
~
16.6 EvaluatiodAssessment of Methods for Virus Removal
493
tive pore size. The more 'popular' concerned retrovirus HIV (120-140 mm) will be filtered effectively by nanofilters. Ultrafiltration With regard to the (theoretical) cut-off, UF membranes could be first choice regarding virus particle removal. However, UF is facing two disadvantages: to date, there is no integrity test sensu structu available, and UF has to be performed as an independent unit operation requiring dedicated equipment. Typically the throughput/turnover of membranes is low, as they are operated repeatedly (almost unlimited), but of course are dedicated to one product. Only a few membrane types are used: Filtron's Omega VR Series, featuring cut-offs of 100 kDa, 150 kDa, 200 kDa, and 300 kDa (Table 16-12), Millipore's 300 kDa, Viresolve 70, and Viresolve 180 as well as A/ G Technologies' VirA/Gard 100, VirA/Gard 300 and VirA/Gard 500. The Omega membranes appear superior to Viresolve due to a poor filtration behavior of the Viresolve membranes, resulting often in high volumes by the need for diafiltration in order to save product, and related poor product yield (Table 16-13). All conventional UF modules as they are presently positioned and available in the marketplace will not be fully utilized for viral filtration until a non-destructive integrity test is developed that can be correlated to viral removal. Table 16-12. Data on virus removal using ultrafiltration". Virus Membrane
SV-40
Re03
PI3
MuLV
Omega 200 kDa Omega 300 kDa
>5.6 <1.0
>6.8 2.8
>7.0 >5.2
>6.8 4.0
a
All values given in log 10 units. As a process solution, PBS containing 2-10 mg mL-' protein was spiked 1:20 by a virus solution at a titer of lo6 to lo8 mL-'. Product recovery was found to be in the range of 97-100% for monoclonal antibodies without any significant increase in volume (<3 %).
Table 16-13. Product recovery using Viresolve 180 for the filtration of a monoclonal antibody.
after ultrafiltration 1. Diafiltration 2. Diafiltration 3. Diafiltration 6. Diafiltration
Product Recovery %
Volume increase %
80.0 84.4 87.3 89.5 97.2
no increase no increase 3.0 7.5 39.5
IgG2 (4 rng mL-') was filtered through a 300 crn2 Viresolve 180 module at a flux rate of 0.028 mL min-' cm-2.
494
16 Widation of Viral Safety for Pharmaceutical Proteins
16.7 Design of Downstream Processing The scientific and practical limitations described so far are of influence on the concept and design of appropriate downstream processes. The arrangement of downstream processing into different stages reflecting a defined product quality might indicate already the positioning of viral clearance steps (Fig. 16-14). Due to a potential adverse effect on product quality, those steps should be placed - if possible upstream within the entire process. Final purification should be dedicated to polishing steps regarding protein chemical criteria and bulk formulation.
Stage Capture
Downstream Processing
Purpose
TFF Microfiltration
Cell harvest
UF I DF
Concentration; adjustment of physical conditions
.1
NanofMraNon L filtrate Virus Inactivation
4
Block virus removal I inactivation
Ultrafiltration Final Purification
Fig. 16-14. Downstream processing of a recombinant protein. The presented scheme is applied to large-scale manufacturing of slCAM, a genetically engineered and modified version of JCAM-1, representing a protein member of the immunglobulin superfamily.
16.8 Conclusion There is no doubt that the most stringent need regarding methodologies for virus removal is related with the blood industry as well as with the biotechnology industry. Small viruses such as hepatitis, polio and Parvovirus are of severe concern, as they withstand to some extent even harsh inactivation procedures. In this sense, retro-
References
495
viruses such as HIV are less critical as they are highly sensitive to inactivation procedures. The biotech market, with regard to safety aspects, is actually changing. In contrast to blood-derived products, there is no case in public reporting a safety issue that is suspect or even related to virus. However, the reported case of an MVM infection of a production fermenter [3] is a stringent example that even a well-characterized cell line in a well-controlled environment is accessible for an adventitious infection. What is the outcome when an inapparent infection of an unknown small virus occurs? How realistic is such a scenario? In the present light, effective procedures which contribute to the removal or inactivation of a complete set of model viruses represent the methods of choice; however, the reliable methodologies that are applicable are currently rather limited.
References [ 11 Marcus-Sekura, C.J., Validation of removal of human retroviruses, Bethedsa, MD, USA:
Center for Biologics Evaluation and Research, FDA, 1991. [2] Office of Biologics Research and Review, Center for Drugs and Bilogics, Points to consider in the production and testing of new drugs and biological produced by recombination DNA technology, Bethesda, MD, USA: FDA, April 10, 1985. [3] Gamick, R.L., in: Viral Safety and Evaluation of Viral Clearance from Biopharmaceutical Products, Brown, F., Lubiniecki, A S . , Karger, Basel: Dev Biol Stand, 1996; Vol. 88, pp. 4956. [4] Anderson, K.P., Low, M.-A.L., Lie, Y.S., Lazar, R., Keller, G., Dinowitz, M., in: Production of Biologicals from Animal Cells in Culture, Spier, R.E., Griffiths, J.B., Meignier, B. (Eds.) ESACT, The 10th Meeting, Avignon, France. [5] Office of Biologics Research and Review, Points to consider in the characterization of cell lines used to produce biologicals, Bethesda, MD, 20892 USA: FDA, 1993. [6] Willig, S.H., Tuckerman, M.M., Hitchings IV, W.S., Good Manufacturing Practices f o r Pharmaceuticals, New York and Basel: Marcel Dekker Inc., 1982. [7] Sharp, J., Good Manufacturing Practice, Buffalo Grove: Interpharm Press, 1991. [8] Kommission der Europaischen Gemeinschaften, Die Regelung der Arzneimittel in der Europuischen Gemeinsckaft, Band ZV, Luxemburg: Amt fur amtliche Veroffentlichungen der Europaischen Gemeinschaften, 1990. [9] EG Guidelines: Validation of virus removal and inactivation procedures, EC DG III/8115/ 89-EN, 1991. [lo] Ministry of Health and Welfare, Documents necessary f o r application f o r approval of drugs manufactured using cell culture technology, Japan, 1988. [ 111 Health and Welfare, Drugs Directorate Guidelines. Utilization of continuous cell lines in the manufacture of biologics, Canada, 1990. [ 121 Kozak, R., Abstracts, Workshop on preclinical safety testing on monoclonal antibodies, Bethesda, MD, USA, 1992. Bethesda, MD, USA: FDA, 1992. [ 131 American Society for Testing and Materials (ASTU). Draft standard guide f o r determination of purity, impurities, and contaminants in biological drug products, Philadelphia, 1988. [14] ICH, Viral Safe3 Evaluation of Biotechnology Products Derived from Cell Lines of Human or Animal Origin, 1996. [15] Walter, J., Werz, W., McGoff, P., Werner, R.G., Berthold, W., in: Animal Cell Technology: Developments, Processes and Products, Spier, R.E., Griffith, J.B., MacDonald, C., (Eds.), Oxford, UK: Butterworth-Heinemann Ltd., 1991, pp. 624-634.
496
16 Validation o f Viral Safety for Pharmaceutical Proteins
1161 Walter, J., Werz, W., Berthold, W., Biotechn Forum Europe 1992, 9, 560-564. [17] Berthold, W., Walter, J., Werz, W., Cytotechnology 1992, 9, 189-201. [18] Lees, G., Onions, D., Technical Bulletin 4, Quality Biotech, Glasgow, 1990. [19] Walter, J., Allgaier, H., in: Manipulation of Mammalian Cells, Wagner, R., Hauser, H.J., (Eds.), GFB Braunschweig, Berlin-New York: Walter de Gruyter, 1997. [20] Loftus, B.T., Nash, R.A., Pharmaceutical Process Validation, New York and Basel: Marcel Dekker Inc., 1984. [21] Calreton, F.J., Agalloco, J.P., Validation of Aseptic Pharmaceutical Processes, New York and Basel: Marcel Dekker Inc., 1986. [22] Liptrot, C., Gull, K., in: Animal Cell Technology: Developments, Processes and Products, Spier, R.E., Griffiths, J.B., MacDonald, C., (Eds.), Oxford, UK: Butterworth-Heinemann Ltd., 1991, pp. 653-656. [23] Alain, R., Nadon, F., SCquin, C., Payment, P., Trudel, M., J Virol Meth, 1987, 16, 209-216. [24] Lower, J., in: Symposium on Virological Aspects of the Safety of Biological Products, London, England, 1990. Dev Biol Stand, 1991, Vol. 75, pp. 221-226. [25] Code of Federal Regulations 21 CFR, US.Government Printing Office, Washington, DC, April, 1991, pp. 210-211. [26] Horowitz, B., Curr Stud Hematol Blood Transfus 1989, 56, 83-96. [27] Walter, J.K., Werz, W., Berthold, W., in: Viral Safety and Evaluation of Viral Clearance from Biopharmaceutical Products, Brown, F., Lubiniecki, A S . , Dev Biol Stand, Karger, Basel, 1996, Vol. 88, pp. 99-108. [28] Werz, W., Hoffmann, H., Haberer, K., Walter, J., Arch Virol 1997 [Suppl] 13, 245-256. [29] Werner, R.G., Berthold, W., Arzneim-ForscWDrug Res 1988, 38 (I), 3, 422-428. [30] Kramer, A., Vakalopoulou, E., Schleuning, W., Schneider-Mergener, J., Mol Zmmun 1995,32 (7), 459-465. [31] MacLennan, J., Bioflechnology 1995, 13, 1180-1 183. [32] Oshima, K.H., Evans-Strickfaden, T.T., Highsmith, K., Ades, E.W., Biologicals 1996, 24, 137-145, [33] Howell, J.A., Sanchez, V., Field, R.W., Membranes in Bioprocessing, Glasgow: Blackie Academic & Professional, 1993. [34] McGregor, W.C., Membrane Seperations in Biotechnology, New York and Basel: Marcel Dekker Inc., 1986. [35] Cheryan, M., Ultrafiltration Handbook, Lancaster: Technomic Publishing AG, 1986. [36] Mixed Dextran UF Test Procedure, Pall Scientific and Technical Report, Pall Corporation, 1997. [37] Ohsawa, N., Sato, T., Virus removability of Planova TM 15 and its application, Technical Report, BMM Development and Business Promotion Department, Asahi Chemical Industry Co., Ltd., 1995. [38] BioEast '94, Integrity Test For Virus Removal Filter, Technical Report, BMM Development and Business Promotion Department, Asahi Chemical Industry Co., Ltd., 1994.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
17 Validation Issues in Chromatographic Processes Gail Sofer
17.1 Introduction Process validation has been defined by many regulatory agencies and noted experts [ 1,2]. The definitions all have in common the requirement for documenting that the process consistently produces a product that meets predetermined specifications. Meeting this requirement in the most efficient way requires that validation is designed into a process. Today, there are many efforts aimed at making it easier for firms to market products worldwide. Guidelines from regulatory agencies and the International Conference on Harmonization (ICH) are intended, among other things, to make it clearer for manufacturers what to validate and when. As these harmonization efforts continue and with increased experience, validation requirements should be more uniform and will be easier to design into a process. Today in the biotechnology industry, the validation topics which demand the most attention are viral clearance, cleaning validation, and software validation. The issues of comparability and validating process changes are also current topics of considerable interest. Validated chromatographic processes are employed for the manufacture of both therapeutic and diagnostic products that are regulated by agencies such as the U.S. FDA, EMEA, MHW, WHO, etc. (see Table 17-1). The intent of these agencies in requiring validation is to ensure that safe and efficacious products are manufactured. For both regulators and manufacturers, validation requires good science; and from the manufacturers’ perspective, validation is good business since when applied properly, it results in the production of consistent product and prevents failed batches. Chromatography removes unwanted substances from the desired product, providing the end-product quality required for a given application. Usually, several chroma-
Table 17-1.Key worldwide regulatory agencies ~
United States Food and Drug Administration European Medicines Evaluation Agency Japanese Ministry of Health and Welfare World Health Organization
FDA EMEA MHW WHO
498
I7 Validation Issues in Chromatographic Processes
tographic steps are employed in a purification process, and in order to validate the process, it is essential to qualify and quantify the substances that are being removed in each step and understand what the ultimate product purity must be. Control parameters for each chromatographic step must be developed, implemented, and validated so that the unwanted substances are consistently removed. Validation of chromatographic processes requires both small-scale and full- or pilot-scale work. At small-scale, the process is designed, cleaning routines developed, virus and nucleic acid clearance studies performed, and column lifetime studies begun. The process is usually further optimized at a pilot-scale and scaleup is validated once the final commercial scale is determined. At full- or pilotscale, equipment is formally qualified, consistency batches are made, and column lifetime is continually monitored. Processes may also need to be changed over time to incorporate new technologies, for example to achieve higher throughput and hence better economy, and process changes must also be validated.
17.2 Small-Scale 17.2.1 Process Design Principles In the design of a chromatographic process, one should first consider what the goals are, and a risk assessment should be made. For example, will the process be used to remove or inactivate viruses from human plasma? Will there be a high endotoxin load from the feedstream such as when E. coli or other Gram-negative bacteria are employed as the host to produce products of recombinant DNA technology? If the product to be used as a diagnostic, what impurities will interfere with the test? For a therapeutic, will DNA be removed to a level accepted by regulators? Will the product and process meet worldwide requirements or will they only be suitable for local marketing? Are the analytical methods for the process intermediates and final product sufficiently sensitive and are they validatable? Are the processing additives and processing tools, such as the chromatographic media, acceptable considering the final product’s intended use? All of these questions should be considered during process design.
Raw Materials Chromatography media, buffers, stabilizing agents, and detergents are just some of the raw materials used in a typical chromatographic process. To comply with good manufactoring practices (GMPs), all of these substances must be quarantined and tested for identity before release into the manufacturing environment. As described
17.2 Small-Scale
499
by Del Tito et al., a graded approach can be taken for testing raw materials used to manufacture clinical biopharmaceuticals [31. They present a ‘road map’ describing what testing is required during Phase I, 11, and I11 clinical trials and after a product license is obtained. The authors further note that some lot testing can be reduced once vendors are qualified. Vendor audits are a crucial element used to ensure raw materials meet their required specifications. The FDA’s discussion draft on Active Pharmaceutical Ingredients [4] states that ‘A supplier’s certificate of analysis may be used in lieu of performing other testing. In such cases the manufacturer should have a vendor qualification program in place and should periodically verify tests performed by the vendor’. A guideline for certifying vendors has been written by a PDA task force [ 5 ] . Selecting raw materials is critical for a chromatographic process. The most suitable chromatography media will tolerate rather harsh cleaning and sanitizing conditions, be compatible with high flow rates used for cleaning, sanitizing, and regeneration procedures, and provide high capacity and selectivity. In some cases, however, a compromise must be made. For example, immunoadsorbents provide exquisite selectivity and may reduce the number of chromatographic steps required to achieve the requisite product quality. Feldman et al. described the use of an immunoadsorbent for purification of Factor VIII, but the column is maintained in a sterile fashion, requiring validation of sterile filters [6]. Buffers and additives should be of the highest quality. When selecting detergents, it is recommended to select those with existing validatable assays that the end-user can validate at their facility or at a contract testing laboratory [7]. All too often, firms find they have added a processing agent for which no assay exists. In some cases, it is not even possible to obtain the chemical composition so that a test can be developed, leading to a regulatory delay. When selecting raw materials, it is important to evaluate more than one lot. Especially with critical raw materials that affect product quality, it is important to review the vendor’s specifications and try at least three lots that reflect the range of product characteristics. For ion-exchange chromatography media, for example, it is useful to test three lots that reflect the vendors’ specified ranges of capacity and particle size distribution.
Process Parameters During the design of a chromatographic process, many parameters need to be optimized. Today, fractional factorial analysis is used to minimize the number of experiments that must be performed [8]. In determining the critical parameters that affect product purity and recovery, it is always important to keep in mind what the final product and process requirements will be. This means considering the process economics, the final product dose, the intended use of the product, and the product source. Clearly, the costs of a process used to produce food-grade material must be lower than those for a therapeutic product. For products that will be delivered in large doses, for example human hemoglobin, the ability to produce large quantities will dictate the sizes of the chromatography columns and systems, which in turn
500
17 Validation Issues in Chromatoxraphic Processes
have a major impact on the facility design, requirements for utilities such as water for injection (WFI), and disposal issues. These factors may limit the selection of chromatography media to those that can be cleaned-in-place (see below) and provide high throughputs. The product source will have a major impact on the type of validation work required. For example, products produced by mammalian cell culture and those derived from blood require virus validation. Unique products produced to expedite clinical trials for terminally ill patients may not initially require as much validation of the purification process, but these products must be safe, and ultimately will require extensive validation. On the other hand, a process used to create a product intended for use in healthy children will be extensively validated as soon as possible. The chromatography-related parameters that must be optimized and specified include sample load, flow rates, elution scheme, regeneration, cleaning, sanitization, storage, start-up, re-equilibration, and holding times. For all parameters, the establishment of ranges in which the process delivers the requisite product is essential. When specification are too narrow, failed batches result. It is wise to establish reasonable, rather broad ranges initially, and then as more knowledge is gained about the process, the ranges can be tightened. Sample load will impact the throughput and, therefore, the economics of a process. It is important, however, to keep in mind that overloading a column does not improve process economics. For example, in the case of cell culture products, there may be considerable variability in the feedstream product concentration and impurity profile. If the column is overloaded, the purity requirement may not be met, resulting in a failed batch or a batch that must be reprocessed. Reprocessing requires extensive validation, documentation, and approval procedures and in most cases creates significant time delays. Failed batches are clearly quite costly, and it is much easier to load less material onto a column. For cell culture products, until the parameters are well established and a firm has a good grasp of the variability, a load of 50% of the dynamic capacity is a good starting point. With feedstreams from fermentation, such as those from E. coli, it is usually acceptable to start with loads up to 70% of the total capacity. Flow rates for sample loading, washing away contaminants, sample elution, regeneration, cleaning, and sanitizing should be correlated with pressure, keeping in mind that larger columns employed at scale-up and sample viscosity can also affect the back-pressure. If the chromatography media cannot tolerate the pressure generated in large scale, a company may be forced to go back to the design stage, delaying entry onto the market and losing market share. The elution scheme should be specified. If gradients are used, then their design will impact the chromatography system configuration and the software required to operate the system. For products requiring high doses, most firms today employ gradient elution for designing a purification scheme, but then switch to step gradients for manufacturing. The criteria to judge whether the step gradient will provide the same product and impurity profiles should be determined during early design stages. Cleaning, sanitization, and regeneration can often be accomplished in one step. During development of the cleaning procedures, it is important to take into account the types of impurities that one is trying to remove from the packed column and the
17.2 Small-Scale
501
stability of the chromatography media and nature of the functional group. For example, when cleaning hydrophobic interaction media, it is important not to add salt which causes increased binding of hydrophobic impurities; whereas for ion exchangers, a salt solution may be the most effective cleaning agent. In some cases, both hydrophobic and ionic interactions are causing binding of impurities. In this case, the order in which the cleaning agents is used can be critical to ensuring an effective, consistent cleaning routine. Cleaning and sanitization are often scrutinized by regulators. The demonstration of the absence of microorganisms, endotoxin, and carryover from previous runs is necessary. An interesting study in which multiple monoclonal antibodies were purified by a core facility illustrated that by implementing the use of a more effective cleaning agent, cross-contamination could be prevented [ 9 ] . Most firms begin monitoring both bioburden and endotoxin as soon as a process is designed. Cleaning validation takes place once the process is established, usually during the manufacture of clinical trial material. In chromatography, cleaning validation usually requires both large scale and small scale (see below) evaluations. Criteria for verifying re-equilibration need to be determined, and are dependent on the type of chromatography and buffers employed. For commonly used techniques such as ion exchange, conductivity and pH measurements are usually used to assess if a column is re-equilibrated prior to the next run. The acceptable range of values should be specified as soon as possible. Standard operating procedures (SOPS) for preparing buffers should be implemented from the beginning, since slight variability in buffers can cause significant fluctuations [ 101. Temperature and holding times must be specified, since they can have a major impact on product quality. Aggregation, proteolytic digestion, and microbial growth are just three of the many factors that may be affected by temperature and holding times. Processing time in chromatography is also critical. For example, in hydrophobic interaction, it is not uncommon to find a protein denaturated when it is allowed to stay in contact with a hydrophobic ligand for an extended time. Therefore, the processing times established during development must be maintained during scale-up. The experimental work to establish all of the parameters discussed above should be traceable back to paginated laboratory notebooks that are bound and have been signed by properly authorized personnel. The results should be summarized in a development report that is suitable for review by regulators and that also serves as part of the technology transfer package for pilot scale. Analysis of Product, Intermediates, and Impurities
Purity is assessed by evaluating the product with analytical methods that measure different properties. These orthogonal methods should also be used to evaluate the impurity profile, since it is necessary to demonstrate what impurities are removed during each chromatographic step. The sophisticated methods available today allow products to be thoroughly analyzed. The concept of a well-characterized product has recently been introduced. As stated by the U.S. FDA, .....'technical advances over the last 15 years have greatly increased the ability of manufacturers to control
502
17 Validation Issues in Chromatographic Processes
and analyze the manufacture of many biotechnology-derived products’ [ 111. This concept has resulted in favorable policy changes that will allow firms to simplify their applications to market well-characterized products, make process changes, and employ contract manufacturers. These products that are now called ‘specified’ include therapeutic DNA plasmid products, therapeutic synthetic peptides of 40 or fewer amino acids, monoclonal antibody products for in vivo use, and therapeutic DNA-derived products. To grasp what is accomplished in each chromatographic step, purification design specialists typically employ extensive analytical methods. During development, those test methods that have been shown to provide the most useful information are selected and the results are used to make in-process and product release decisions. It is important to keep in mind that the analytical methods will also require validation. And while it is essential to employ up-to-date methods, it is also necessary to recognize that in-process controls for chromatography should generally be relatively simple, but correlated with more sensitive and specific analyses. In-process controls are often elution time or volume, pH, conductivity, and UV profiles. The correlation of these parameters with more sophisticated, sensitive, and specific test methods allows the chromatographer to proceed without delays. Alternatively, inline sophisticated analytical tools, such as biosensors, may be selected by some firms for process monitoring in cases where more information is required to make in-process decisions.
17.2.2 Small-Scale Models Small-scale models of the process are used to analyze the efficacy of cleaning methods, perform column lifetime studies, evaluate the ability to change a chromatographic medium, and demonstrate removal of key impurities such as viruses and DNA. Regulators have expressed concern that small-scale experiments do not reflect manufacturing, and it is essential to validate that the small-scale model column truly reflects the manufacturing operation. Some of the more important parameters that must be considered are chromatography system design, column dimensions and design; buffer and chromatography media preparation; temperature; pH; flow properties; gradients; impurities; and product concentration [ 121.
Cleaning Cleaning validation is performed both at full scale and also by employing small-scale models. Frequently, the small-scale studies used to evaluate cleaning effectiveness are also used for column lifetime studies. After validating the scale-down of the column, sample is cycled as it would be in full scale. The effectiveness of cleaning routines is measured by assaying the rinse fluids for specific materials such as protein residues, specific product, endotoxin, and DNA, or generally for any carbon-contain-
17.2 Small-Scale
503
Table 17-2.Some analytical methods use in cleaning validation. Total protein UV spectroscopy ELISA HPLC Electrophoresis Total organic carbon (TOC) PH Conductivity Nucleic acid hybridization Limulus amebocyte lysate (LAL)
ing products. Protein assays, UV analysis, HPLC, ELISAs, and SDS-PAGE are some of the methods used to detect specific molecules. Conductivity, pH, and total organic carbon (TOC) are commonly used general detection methods [13]. Not all analytical methods need be used; those which make common sense for the application should be selected (see Table 17-2). However, the choice of cleaning methods and levels of cleaning should be defined and justified (see [4]). A small-scale study was used to design a validatable cleaning protocol for removal of negatively charged nucleic acids which bind tightly to anion exchangers. The study employed a sample of calf thymus DNA, and it showed that a solution of 1 M NaCl in 1 M NaOH was sufficient to remove all the bound DNA. But when a sample of a monoclonal antibody heavily contaminated with nucleic acids was used, the sodium chloride/sodium hydroxide solution was relatively ineffective and DNase was required to remove all the DNA [14]. However, it is important to keep these data in perspective. If DNA does not elute with the product or with sodium hydroxide/sodium chloride solutions, it is unlikely that one would have to implement the use of DNase as a cleaning agent, especially since this would require the addition of an assay for DNase. If, on the other hand, DNA was continually or sporadically eluting with the product after stringent cleaning, then such treatment might be needed. The ICH guideline on viral safety states that ‘Assurance should be provided that any virus potentially retained by the production system would be adequately destroyed or removed prior to reuse of the system. For example, such evidence may be provided by demonstrating that the cleaning and regeneration procedures do inactivate or remove virus’ [ 151. Q-One Biotech performed a study o n the ability of 0.1 M and 0.5 M sodium hydroxide to inactivate eight different viruses. The kinetics of inactivation were also reported - a factor which is essential for defining the contact time required for cleaning and sanitizing solutions (see [12]). Further cleaning evaluation can be made by collecting effluent from a blank run, and analyzing it for residual materials not removed by the cleaning routine. Additionally, the efficacy of the cleaning routine can be challenged by increasing the exposure time to the cleaning agent and analyzing effluent and by extending the holding time of a used column prior to performing cleaning-in-place. A cleaning effect is quite often obtained by storage solutions, and to evaluate this possibility,
504
17 Validation Issues in Chromatographic Processes
small-scale experiments on used chromatography media can be performed by allowing the media to remain in the storage solution and then assaying for residual material in the storage solution. An excellent study on cleaning S Sepharose” Fast Flow has been performed by Seely et al. [16]. This study was also used to evaluate media lifetime.
Lifetime Small-scale studies are combined with concurrent validation at pilot and full scale to determine column lifetime. It is obviously cost effective to maximize the life of chromatography media. Some data are usually required to be included in the dossier submitted to obtain market approval. These data can be obtained from small-scale models in which the media is used for repeated cycles of sample application, elution, cleaning and sanitization, and storage. It is essential to demonstrate that the chromatography media provide consistent product quality after repeated use. In the study by Seely et al. 1200 cycles were ultimately obtained on the laboratory model column and up to 240 cycles obtained at production scale. In addition to assaying the product quality, other indicators of media deterioration include changes in chromatographic profiles, decreasing product recovery, changes in pressure or flow rate, and changes in the re-equilibration profile. If a packed column is used for viral clearance, then it is essential to show that used media will give the same clearance as new media. As stated in the ICH guideline on viral safety evaluation, ‘Over time and after repeated use, the ability of chromatography columns and other devices used in the purification scheme to clear virus many vary. Some estimate of the stability of the viral clearance after several uses may provide support for repeated use of such columns’. Recent work by Pharmacia Biotech showed that ion exchangers taken from a manufacturing plant after 440 cycles of purification of IgG from human plasma provided the same viral clearance as new media [ 171.
Clearance of Impurities For evaluating clearance of some impurities it is necessary to use small-scale models so that sufficient quantities are employed for analysis and to enhance worker safety. Such is the case for virus clearance. Virus removal by chromatographic steps is not considered a ‘robust’ technique, but chromatography does enhance the overall viral clearance [18]. The ICH guideline states that for viral clearance studies, ‘The validity of the scaling-down should be demonstrated. The level of purification of the scaleddown version should represent as closely as possible the production procedure. For chromatographic equipment, column bed-height, linear flow-rate, flow-rate-to-bedvolume ratio (Lea,contact time), buffer and gel types, pH, temperature, and concentration of protein, salt, and product should all be shown to be representative of commercial-scale manufactoring’.
17.3 Pilot and Full-Scale
505
Small scale experiments are also used to validate removal of DNA and host cell proteins. For well-characterized products, process validation may replace routine QC lot release for these two impurities [19,20]. The assays employed must be sufficiently sensitive, and may have an influence on whether both small-scale and largescale runs are needed for validation. The feedstream used in the small-scale experiments should be fromproduction. Host cell protein assays have been addressed by Eaton [21]. For DNA assays, both hybridization with specific probes and the Threshold system for detecting total DNA may be used [22,23]. Small-scale experiments in which production feedstream is used can also be employed to evaluate leachables from chromatography media [24]. Small-scale studies are not, however, usually employed for demonstrating removal of endotoxin and bioburden. Rather, the removal of these impurities/contaminants is validated by employing in-process and final product testing at pilot and full-scale.
17.3 Pilot and Full-scale Usually, clinical trial material is made at pilot scale. This may, in fact, be the scale used to manufacture licensed product, or a further scale-up may take place prior to Phase I11 clinical trials or after product licensure is obtained. Most firms prefer not to make scale changed during Phase I11 trials, since to do so can lead to additional uncertainties at a time when clinical trial results are being interpreted. Whenever the scale is changed, it is necessary to demonstrate that the product is comparable with product made at the previous scale. This is done by employing the analytical methods that have been shown to evaluate thoroughly both the product and its impurities, often requiring biological and stability assays. It is quite possible that not all the analytical methods will be completely validated during the initial pilot-scale manufacturing runs, but a validation master plan that describes the firm’s intended schedule for performing validation activities should be prepared to keep a reasonable time line. FDA and other agencies accept a so-called ‘graded approach’ to validation. Recognizing that it is neither common sense nor economically feasible to validate every piece of equipment and every step of a process until it is firmly established, regulators have stated that full validation of process and equipment may not be complete until Phase I11 [25]. However, they do require that any equipment and/or processes which impact the safety of any clinical trial materials are validated. This includes, for example, sterilization cycles and autoclaves, but it is highly unlikely that a chromatographic process will be finalized, much less fully validated, during early (Phase I and 11) clinical trials. Once the process is finalized, formal validation activities, including equipment qualification, take place.
506
17 Validation Issues in Chromatographic Processes
17.3.1 Equipment and Automation The establishment of the design criteria for equipment is part of the validation effort. A larger-scale chromatography system may cause aberrations in performance that can adversely affect product quality. As discussed by Gagnon, system lag, clearance, and dispersion factors; volumetric accuracy of pumping systems: and proportioning accuracy of gradient mixing systems can all have an effect on the separation when changing equipment during scale-up [26]. When a robust process with realistic ranges has been established, these effects are not as detrimental, but they should be avoided as much as possible by selecting suitable equipment and following established scale-up guidelines. Scale-up guidelines and equipment selection for chromatography processes have been addressed by Hagel and Sofer (see [S]). Today’s chromatography systems are automated, which enhances the reproducibility of the chromatography unit operations, the ease of obtaining documentation of critical parameters, and minimizes worker contact with product. Prior to formal equipment qualification, a vendor will commission the system and ensure that it is functional. Each firm has its own routines for equipment qualification, and a protocol for performing equipment qualification is usually maintained by a Quality Assurance function. Software, hardware, and equipment used to manufacture pivotal clinical trial (i.e. Phase 111) material is formally qualified by performing an installation qualification (IQ) and an operational qualification (OQ). In the IQ, the components are listed and checked against the specifications. Auxiliary systems are also verified at this time. The work performed in this phase is tedious, but ensures that the correct system is installed. It is also a good time to check that all system documentation, including SOPS and calibration routines, are in place. In the OQ, the assembled system is tested to ensure that it performs as it should. The OQ has been called the ‘water test’ in chromatography, since no sample is used -just water, buffers, and some processing solutions such as those used for cleaning. During this phase, some of the testing includes evaluating accuracy of gradient formation; quality of column packing; pressure/flow over entire system; leak checks; and functioning of valves, monitors, sensors, and alarms. All of the phases of automated system qualification require that the personnel performing the work are qualified to do so, that they sign the documentation, and that an authorized person in the firm approves the completed document. Validation of the software is the most extensive part of the qualification. Much has been written about validation of control systems, and the reader is referred to these references for more specific details [27-301. Basically, there are two phases to software testing. The first phase is structural testing and requires input from the developer of the software to describe the standards used in programming, how the code was written and tested, programming language, qualifications of the programmers, internal organization and quality policy, arrangements for access to the source code (sometimes asked for by regulators), disaster plans, change control procedures, error notification policy, etc. This type of information is usually gathered by performing a vendor audit or by obtaining confidential files that contain this specific, often proprietary, information.
17.3 Pilot and Full-Scale
507
The second phase in software testing is functional and is performed at the user’s site. During this phase the IQ and OQ are also performed for the computer hardware as well as the software. It is essential to ensure the proper environment (e.g. relative humidity), emergency backup (e.g. uninterruptable power supply, UPS), and wiring and grounding for the computer. For the software, the proper version should be confirmed and documented. Finally, the installed equipment and software are tested together. When planned for and when experienced personnel perform the validation of an automated chromatography system, it takes approximately 2-5 days, depending on the complexity of the system. Vendors can assist in the qualification. Extensive knowledge of their systems allows them to prepare documentation packages such as forms for the IQ and OQ. They can even perform the work, saving a firm considerable time and resources, but is essential to keep in mind that the end-user is always responsible for validation and must, therefore, participate and sign off the appropriate documentation for the formal automated system qualification.
17.3.2 Process Validation Once the equipment is installed and qualified, the process can be validated. ‘In-process specifications should be derived from research or pilot scale batches or process variability estimates until sufficient process data is collected on full scale production batches’ [4]. After evaluating whether the scale-up has indeed worked, there may a need to do some fine tuning or even tightening of specifications. Minor modifications in chromatography might have to be made to accommodate variability that arises due to changes in scale in fermentation or cell culture and pre-chromatography isolation steps. Chromatography scale-up is unlikely to be a problem, but changes in cell culture scale sometimes lead to a feedstream that has a somewhat different impurity profile and different concentration of product. In the worst situation, the product is changed, e.g. glycosylation pattern, requiring major changes in the purification. This scenario should be avoided. Isolation techniques that are scalable do not usually cause a change in feedstream characteristics, but when, for example, a different mechanism of cell disruption is employed, there can be release of more endotoxin or nucleic acid or a higher temperature that causes product denaturation. An increase in nucleic acids can cause the feedstream to be more viscous, which can lead to backpressure problems in chromatography. The chromatographer needs, therefore, to work with other departments if process validation is to proceed smoothly. Process validation is performed by manufacturing three to five consecutive pilotor full-scale batches that meet all specifications for process parameters, process intermediates, and product. This phase is where the most extensive validation efforts take place, using both highly sophisticated analytical methods and routine monitoring of bioburden and endotoxin. Validated analytical methods are employed at this phase.
508
17 Validation Issues in Chromatographic Processes
17.3.3 Validation of Sanitization and Cleaning Sanitization Routine bioburden and endotoxin measurements help ensure that the process meets the established specifications. For chromatography processes, once the consistency batches are made and there is some further experience in making the product, routine endotoxin testing may suffice. Prior to eliminating a test method, it is advisable to discuss the data that support the intent to reduce testing with regulators. For chromatography systems and columns, bioburden and endotoxin challenges and concomitant sanitization data are usually available from the vendor [31]. Challenge experiments which identify areas of a system where microorganisms are likely to reside and the contact time, temperature, and sanitizing agent concentration that will eliminate them should not be performed at a user’s facility. Data on effective sanitizing agents are also available for most chromatography media. The user should ensure that the sanitizing agents selected are compatible with the combination of media, column, and system that are chosen to ensure there are no unexpected leachables or deterioration of packed column or equipment components.
Cleaning In chromatography processes, cleaning of both equipment and packed columns must be addressed. Because of their large surface areas, it is the chromatography media which present the greatest cleaning challenge. During the preparation of consistency batches, blank runs can be performed to validate the cleaning of packed columns. However, it is really the analysis of product consistency that demonstrates effective cleaning. The small-scale cleaning studies (see above) in which manufacturing-scale feedstream is used provide the data which give both the manufacturer and regulators confidence to implement a given cleaning protocol. Should the product purity change, a blank run, among other things, could be used for troubleshooting to determine if the column was properly cleaned. Cleaning validation of packed columns can only be achieved by evaluating the effluent. With equipment, such as tanks, valves, piping, etc., however, both swabbing and rinse fluid are evaluated. Swabbing is also employed for column hardware that will be used for more than one product (media are always dedicated to one product). Swabbing allows one to evaluate adherence of substances which are insoluble, but swabbing validation alone is not sufficient [32]. Other techniques for equipment validation include the use of coupons [33] and visual inspection should always be performed.
17.4 Process Changes
509
17.3.4 Revalidation Once a process is established and validated, it is necessary periodically to ensure that it is still being operated in a validated state. Quality reviews, including checks of adherence to SOPS, are usually performed more frequently when the process is new. A change control policy ensures that no unauthorized changes are made. When planned changes are made, however, the process must be revalidated, the extent of which is dependent upon the nature of the change.
17.4 Process Changes A recently published guidance on comparability describes the FDA's position on making changes [34]. An earlier attempt to categorize changes into those which require only annual reporting, those which require notification prior to implementation, and those which require preapproval has been discarded, but the lists produced in the initial proposal give some indication of the thinking about the criticality of specific changes [35]. Some changes relevant to chromatography processes are those in software versions; equipment; chromatography media; and addition, deletion, or order change for purification steps.
17.4.1 Software When a software version is changed on a validated system, a complete operational qualification and even a process performance qualification of the system may need to be repeated. Version improvements should be significant and change planned for, i.e. not during a manufacturing campaign. If the software change is a simple bug fix, then it may only be necessary to simply test a few functions.
17.4.2 Equipment Equipment changes in chromatography include changes to pumps, valves, system configuration, wetted components, etc. A pump may be replaced by one with the same mechanism and same wetted components, and the change simply documented. On the other hand, if the wetted components are changed, then there must, at a minimum, be some data that show the new wetted material will not leach and will tolerate the physical and chemical process conditions, e.g. cleaning procedures. If a sample delivery pump is changed to one with a new mechanism, then it might be necessary to demonstrate that the feedstream components are not sheared. Replumbing the system can cause major changes in dilution factors that can have an adverse effect on
5 10
17 Validation Issues in Chromatographic Processes
the product purification. Decisions for making chromatography equipment changes should be made by a team of personnel who can assess all the risk factors and design suitable test methods to ensure consistent product quality will be achieved.
17.4.3 Chromatography Media A change in chromatography media can require a major revalidation effort. If, for example, a chromatography column has been used to claim virus removal, a seemingly minor change can have a major effect on the level of virus removed. However, to improve product throughput, recovery, and purity, changes are made, but they must be planned for. In one case, a vendor was required, due to environmental considerations, to change a chromatography media. The change involved the addition of one hydroxyl group in the spacer arm of an ion exchanger. As seemingly insignificant as this may appear, it did have an impact on some unique separations. For one firm, a program was undertaken to ensure that changing the media would not affect the product or the process. Comparability was evaluated by examining pressure - flow curves for the two media; ionic capacity; peak shape, elution position, height, and width. Product purity was evaluated by multiple analytical methods [ 3 6 ] .
17.4.4 Addition, Deletion, or Order of Purification Steps Adding, deleting, or changing the order of purification steps can have a major impact on the end product. With the advancement in analytical techniques, it may be possible to make such changes and demonstrate product comparability without entering new clinical trials. Such changes must be planned for and the validation strategy thoroughly evaluated. The repeat of clinical trials is a costly undertaking that no firm wishes to do unnecessarily. In summary, validation of chromatography requires the use of good design which allows firms to use chromatography to purify the desired product with the desired characteristics in an economical process which can be validated. Special attention should be paid to validation of cleaning, scale-down studies, and automated systems. It is essential to recognize that improvements in technology will occur, and that it will often be necessary to make process changes to provide safer, more efficaceous products. Planning for process changes and their validation is essential to avoid regulatory delays.
References
5 11
References [ 11 U. S . FDA, Guideline on Gene?alPrinciples ofprocess Validation.Rockville, MD, USA, 1987. [2] Carleton, F. J., Agalloco, J. P., Validation of Aseptic Pharmaceutical Processes. New York: Marcel Dekker, Inc. 1986. [3] Del Tito, B. J., Tremblay, M. A,, Shadle, P. J., BioPharm 1996, 9, 45-49. [4] U. S . FDA Discussion Draft-Not for Implementation, Manufacture, Processing or Holding of Active Pharmaceutical Ingredients, August 1996. [5] PDA Supplier Certification Task Force, J Parenteral Sci Technol 1989, 43, 151-157. [6] Feldman, F., Chandra, S., Hrinda, M., Schrieber, A,; in: Quality Assurance in Transfusion Medicine. Boca Raton FL: CRC Press, Vol 11, 1993. [7] McEntire, J., BioPharm 1996, 9, 50-52. [8] Hagel, L., Sofer, G., Handbook of Process Chromatography: A Guide to Optimization, Scale up, and Validation. London: Academic Press, 1997. [9] Hale, G. J., Drumm, A., Harrison, P., Phillips, J., J Immunol Methods 1994, 171, 15-21. [lo] Gagnon, P., Purification Tools f o r Monoclonal Antibodies. Tucson, AZ: Validated Biosystems, Inc., 1996. [ l l ] Federal Register, May 14, 1996, 61, 24227. [12] Sofer, G., BioPharm 1996, 9, 51-54. [13] Baffi, R., Dolch, G. Garnick, R., et al., J Parenteral Sci Technol 1991, 45, 13-19. [I41 Dasarathy, Y., BioPharm 1996, 9, 41-44. [ 151 International Conference on Harmonization, Draft Guideline on Viral Safety Evaluation of Biotechnology Products Derived from Cell Lines of Human or Animal Origin. Federal Register 1996, 61, 21882-21891. [16] Seely, R. J., Wight, H.D., Fry, H. H., Rudge, S.R., Slaff, G.F., BioPharm 1994, 7, 41-48. [17] Anderson, I., Conner, S . E., Lindquist, L.-O., Watson, E. A., Oral Presentation. 24th Conference of the International Society of Blood Transfusion. Makuhari, Japan, March 31-April 5, 1996. [I81 Darling, A., BioPharm 1996, 9, 42-50. [I91 Riggin, A., Davis, G.C., Coprnann, T.L., BioPharm 1996, 9, 36-41. [20] Geigert, J., PDA Letter 1996, X X X I I , 1,9. [21] Eaton, L., J Chromatogr A 1995, 705, 105-114. [22] Ferre, F., Marchese, A. Griffin, S . Saidgle, A,, Richieri, S . , Jensen, F., Caarlo, D., AIDS 1993, 7, S21-S27. [23] Per, S . , Aversa, C., Johnston, P., Kopasz, T., Sito, A,, BioPharm 1993, 6 , 34-40. [24] Johansson, B.-L., BioPharm 1992, 4 , 34-37. [25] Scott, A., Division of Establishment Licensing, CBER, Oral Presentation, May 1995. [26] Gagnon, P., BioPharm 1996, in press. [27] Agalloco, J., Pharm Technol 1995, 19, 114-122. [28] PMA Computer Systems Validation Committee. Pharm Technol 1986, 10, 24-34. [29] EC Commission Directive 91/356/EEC: Computerized Systems, 1991. [30] U. S. FDA Division of Field Investigators. Office of Regional Operations. Software Development Activities, Reference Materials and Training Aids f o r Investigators. July 1987. [31] Pharmacia Biotech Technical Notes 209 and 214. Uppsala Sweden, 1991 and 1994. [32] Lombardo, S., Inampudi, P., Scotton, A., Ruezinsky, G., Rupp, R., Nigam, S., Biotechnol Bioeng 1995, 48, 513-519. [33] PDA Biotechnology Cleaning Validation Subcommittee, Cleaning and Cleaning Validation: A Biotechnology Perspective. Bethesda, MD, USA: PDA, 1996. [34] U.S. FDA, FDA Guidance Concerning Demonstration of Comparability of Human Biological Products, Including Therapeutic Biotechnology-Derived Products. April, 1996. [35] U. S . FDA, Federal Register 1996, 61, 2733-2750. [36] Seely, R., Fry, H., Oral Presentation, Well-Characterized Products, December, 1996.
Part One Processing
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
1 Strategies in Downstream Processing Yusuf Chisti
1.1 Introduction Biological products come from many sources: human and animal tissue (e.g., blood, pancreas, pituitary) and body fluids (e.g., milk of transgenics), plant material (e.g., Taxol@from the bark of Tuxus species, oils), microbial fermentations, cultures of higher eukaryotes, and raw broths from enzyme bioreactors. Irrespective of the source, crude extracts, fluids and broths invariably undergo separation and purification to recover the product in the desired form, concentration, and purity. Processing beyond the bioreaction step is termed downstream processing. Here the ‘bioreaction step’ includes producing plants and animals. A recovery process consists of physicochemical operations such as those listed in Table 1-1. The steps of a properly engineered downstream process are integrated with each other and with the bioreaction stage to yield an optimal recovery scheme [19]. This discussion is limited to factors which must be considered in developing any economically viable product purification and concentration scheme based on a Table 1-1. Bioseparation operations. Solid-liquid separations [ 1-31 Centrifugation, filtration, flocculation, flotation, sedimentation [4,5]. Membrane separations [l-3,6] Diafiltration and dialysis, microfiltration, pervaporation [7], reverse osmosis, ultrafiltration. Extractions Aqueous liquid-liquid extraction [ 1,8,9], extraction and leaching of solids [lo], reversed miceller extraction, liquid membrane extraction [ 11,121, solvent extraction [ 1,131, supercritical extraction. Chromatographic methods [ 1-3,141 Affinity, gel permeation, hydrophobic interaction, ion exchange. See also Volume 1. Thermal operations Distillation, drying [l], evaporation, freeze drying or lyophilization [ 151. Miscellaneous Adsorption, cell disruption [ 16-18], crystallization, electrophoresis and other electrokinetic methods, precipitation [ 11.
4
I Strategies in Downstream Processing
small selection of the many available processing operations. Individual operations are detailed in other sources [l-31 including some chapters of this series.
1.2 Overview of Process Considerations Anyone faced with designing a bioseparation scheme can take comfort in the variety of available separation processes (Table 1-1); however, the same variety can be a source of much distress. The number Nf of possible recovery flowsheets that can be theoretically devised to completely separate a mixture of C components by using S number of separation operations is given by Nf =
[2(c-1 ] !S("') c !(C- 1) !
Thus, complete separation of a five-component mixture using two separation methods would generate 224 possible flowsheets! Although only one component, the product, is usually wanted, several 'components' need separating. Examples of such components are cells, water, cell debris, nucleic acid polymers, added salts, and the remainder of the proteins. Often, a process must include additional nonseparating steps such as cell disruption, heating, and mixing. Not all possible flowsheets can be exhaustively evaluated; instead, experience and thorough knowledge of individual bioseparations and relevant fermentation must be relied upon to narrow the choices to a few practicable options for detailed evaluations and experimental testing. Factors that must be considered in designing a downstream processing scheme include the nature, concentration and stability of the product, the desired purity and end use. Because of contamination and supply considerations, there is a distinct trend to move away from direct extraction of human and animal sources to recombinant cells. Thus, for example, microbially produced recombinant human insulin and growth hormone are now available. For vaccines, too, attempts are underway to engineer safer organisms to produce the antigenic material that would otherwise be obtained from pathogens. The end use of the product may vary - research, in vitro diagnostic, food, animal feeds, soil inoculants, pesticides, medicinals, medical device, cosmetics, etc. The specific form of the product may include live human cells for medical purposes; live microorganisms, viruses (e.g., for vaccines), spores (e.g., for biotransformations, insecticides, and solid state culture), and higher organisms (e.g., nematodes); bioactive polymers, proteins and enzymes; inactive polymers (e.g., food protein, xanthan, PHB); smaller organics (e.g., streptomycin, amino acids, citric acid, ethanol, TaxoP); polypeptides (e.g., cyclosporine); and cellular organelles (e.g., nuclei, mitochondria, chloroplasts). Location of the product, whether extracellular, intracellular, or periplasmic, affects how it is recovered. Physical and chemical properties of the product and contaminants need addressing, and biosafety issues must be given attention (see Chapter 13).
1.2 Overview of Process Considerations
5
A further consideration is price relative to existing sources and other competing products. When no competing products or alternative sources can be identified, the estimated production costs would need to be compared with what the market can reasonably be expected to pay. As far as possible, the requisite purification and concentration should be achieved with the fewest processing steps; generally, no more than six to seven steps are used, a situation quite different from that in chemistry and biochemistry laboratories, where the number of individual steps is often not a major consideration, and purity of the product is usually more important than overall yield or costs [19]. The overall yield of an n-step process with step yield of x percent is (x/lOO)". Therefore, IZ must be minimized for a high overall yield. For example, a train of only five steps, each with 90% step yield, would reduce the overall recovery to less than 60% [20]. To minimize reduction of the overall yield, high-resolution separations such as chromatography should be utilized as early as possible in the purification scheme in keeping with the processing constraints that these steps require (e.g., clean process streams free of debris, particulates, lipids, etc.). Separation schemes incorporating unit operations which utilize different physicalchemical interactions as the bases of separation are likely to achieve the greatest performance for a given number of steps. Combining two separation stages based on the same separation principle may not be an effective approach. As an example, when two chromatographic steps in series are selected, e.g., gel filtration which separates based on molecular size, and ion exchange chromatography which separates based on difference in charge on the molecules, may be a suitable combination. Speed of processing is another factor that significantly affects the design of a recovery scheme. The size of the bioreaction step and the frequency of harvest usually determine the turnaround time for the downstream process train. Sometimes during processing, exposure of material to relatively severe environmental conditions is unavoidable. Very many factors affect stability, including temperature, pH, proteases and other degrading enzymes, mechanical forces, microbial contamination, oxidants, and other denaturing chemicals. In severe environments, the duration of exposure must be minimized and especial precautions (e.g., low temperature: addition of chemicals to reduce oxidation, etc.) are necessary to reduce the impact of exposure. The need for speedy processing constrains equipment choice and capacity. For example, the low pH necessary during extraction of penicillins affects stability, hence rapid extraction is essential, thus mixer-settler type extraction is contraindicated. Typically, a separation process must operate within the physiological ranges of pH and temperature (pH - 7.0; temperature 5 37 "C),but differences from the norms are not unusual. For example, enzymes such as lysozyme, ribonuclease, and acid proteases are quite stable at low pH values [19]. Some biologically active molecules, particularly proteins, may be sensitive to excessive agitation: however, enzymes, with the exception of multienzyme complexes and membrane-associated enzymes, are not damaged by shear in the absence of gas-liquid interfaces [1,21]. Except for the final few finishing operations, downstream processing is usually conducted under non-sterile, but bioburden-controlled conditions; however, prevention of unwanted contamination and cleaning and sanitization considerations require
6
1 Strategies in Downstream Processing
that the processing machinery be designed to the same high standards as have been described for sterile bioreactors [22,23]. Containment and hygienic processing requirements may severely affect equipment choice (see also Chapters 13 and 14). A commercial recovery scheme must be reliable and consistent. Process robustness is essential to economic production, process validation, and product quality. Automation assures consistency and rapid turnaround of the process equipment. Operations such as in-place cleaning are often automated [24]. Additional considerations include biosafety and containment. Bioproducts may be potentially allergenic, and they may produce activity associated reactions in process personnel [ 101. In addition, process material may be pathogenic, cytotoxic, oncogenic, or otherwise hazardous. Processing of such material requires attention to containment and biosafety both during design and in operation of the bioseparation scheme [10,25]. Certain processing operations are difficult to contain, and may pose peculiar operational problems. For example, gasket failures during high-pressure homogenization could create high-pressure sprays [ 161 and, unless designed with containment features, operations such as centrifugation may generate aerosols (see also Chapter 13). Small quantities of multiple products are sometimes produced in the same plant: a series of runs or campaigns of one product is followed by another. The risk of crosscontamination is high and adequate safeguards are essential. Experience suggests that cross-contamination with penicillins and penicillin-containing substances cannot be reasonably prevented in a multi-product facility. Because penicillins may produce adverse reactions in some patients, Good Manufacturing Practices (GMP) regulations demand dedicated penicillin processing facilities that are segregated from non-penicillin products. Separate air handling systems are necessary if a building processes penicillins as well as non-penicillin products. GMP regulations including the validation requirements [26], affect all aspects of downstream processing. Requirements depend on the kind of product (e.g., food, bulk pharmaceutical, final dosage form, etc.) and the jurisdiction. Willig and Stoker [27] should be consulted for specific guidance. The final few downstream processing steps include formulation which is highly product specific. How a product is formulated may critically affect its stability, efficacy, and bioavailability. Formulation may involve addition of fillers (e.g., starch, cellulose, sugar, flour), diluants, preservatives, sunlight protectants (e.g., carbon black, dyes, titanium oxide), dispersal aids, emulsifiers, buffers, moisture retainers, adjuvants (e.g., mineral oils and aluminum hydroxide added to improve antigenicity of certain vaccines), flavors, colors, and fragrances. Additional finishing operations may include sterile filtration, vialing, granulation, agglomeration, size reduction, coating, encapsulation, tableting, labeling, and packaging.
1.3 Product Quality and Purity Specifications
7
1.3 Product Quality and Purity Specifications The specifications on product purity and concentration should be carefully considered in developing a purification protocol. Concentration or purification to levels beyond those dictated by needs is wasteful. The acceptable level of contamination in a particular bioproduct depends on the dosage, the frequency of use, and the method of application (e.g., food, drug, oral, parenteral), as well as on the nature and toxicity (or perceived risk) associated with the contaminants [ 191. Products such as vaccines, which are used only a few times in a lifetime, may be acceptable with relatively high levels of other than the desired biomolecule. In some cases, contaminating protein levels of about 100 ppm may be acceptable. In vitro diagnostic proteins (enzymes, monoclonal antibodies) may tolerate greater levels of contaminants so long as the contaminants do not interfere with the analytical performance of the product. With certain diagnostic proteins, such as the blood typing monoclonal antibodies, cross-contamination causing misdiagnosis is an extreme concern because of possibly fatal consequences of mis-typing. Such concerns influence the design and operation of the downstream process, particularly for multi-product plants. Parenteral therapeutics usually must be purer than 99.99%. A variety of approaches are used to assure quality. Methods typically used with protein therapeutics are summarized in Table 1-2; Anicetti et al. [29] provide additional details. Requirements relating to some specific contaminants are discussed below. Table 1-2. Methods for quality assurance of protein therapeutics [28]. ~
Impurity or contaminant
Analytical technique
Protein contaminants (e.g., host cell proteins)
SDS-PAGE electrophoresis, HPLC, immunoassays (ELISA, etc.)
Endotoxin
Rabbit pyrogen test, LALa
DNA
DNA dot-blot hybridization
Proteolytic degradation products
IEFb, SDS-PAGE, HPLC, N- and C-terminus analysis
Presence of mutants and other residues
Tryptic mapping, amino acid analysis
Deamidated forms
IEF
Microbial contamination
Sterility testing
Virus
Viral susceptibility tests
Mycoplasma
21 CFRC method
General safety
As per 21 CFR 610.11
a
Limulus amoebocyte lysate;
Isoelectric focusing;
Code of Federal Regulations.
8
1 Strategies in Downstream Processing
1.3.1 Endotoxins Products derived from bacteria such as Escherichia coli will invariably be contaminated with bacterial cell wall endotoxins which can cause adverse reactions (headaches, vomiting, diarrhea, fevers, etc.) in patients unless reduced to very low levels kg per kg body weight). Endotoxins are extremely heat(e.g., less than 5 X stable lipopolysaccharides that are not easily removed from solutions of macromolecules. Ultrafiltration and reverse osmosis are effective for depyrogenation of water and small solutes. Other pyrogen removal methods are adsorption on activated carbon and barium sulfate, hydrophobic interaction chromatography, and affinity chromatography. Endotoxins bind to polymixin B affinity columns, but this method must be combined with detergent treatment for effectively removing protein-bound endotoxins [28]. Chromatography using LAL affinity matrix also removes endotoxins. As a guiding principle, processing must aim to minimize endotoxin contamination by controls on process water and other additives. In addition, aseptic and bioburden controlled operation, and frequent cleaning of equipment help to reduce contamination. The equipment cleaning protocol must include procedures proven for depyrogenation. Standard alkali-based cleaning procedures [24] are quite effective in depyrogenation of stainless steel equipment, but other methods are necessary for cleaning chromatographic columns and membrane filters. The depyrogenation step employed during cleaning of membrane filters usually involves a 30-minute, 30-50 "C treatment with sodium hydroxide (0.1 M), hydrochloric acid (0.1 M), phosphoric acid, or hypochlorite (300 ppm free chlorine). Thorough rinsing with pyrogen-free water follows. Similar procedures are used for chromatographic columns. An endotoxin-free product should be validated using the LAL test. This test is based on endotoxin-induced coagulation of amoebocyte lysate of horseshoe crab (Limulus polyphemus) at 37"C, pH 7.0. Less than 0.3 ng mL-' endotoxin levels are easily detected. Scrupulously clean glassware and water are necessary to prevent false positives. Some known interferences are EDTA, sodium dodecyl sulfate, urea, heparin, and benzyl penicillin.
1.3.2 Residual DNA Residual DNA from producing cells can potentially contaminate the product. DNA fragments from established animal cells were once believed to be potentially oncogenic, which prompted the U.S. Food and Drug Administration to recommend a contamination level of no more than 10 pg DNA per dose [30]. Less restrictive limits are now accepted because no oncogenic events were observed following injections of large doses of DNA into animals. Nonetheless, DNA is a contaminant and demonstration of its satisfactory clearance is essential to quality assurance of the product [30]. Residual DNA is removed usually by adsorption on strong anion-exchange resins at pH 2 4. Hydrophobic interaction chromatography is also effective and so is affinity chromatography under conditions that bind the desired protein but not the DNA.
1.3 Product Quality and Purity Specifications
9
1.3.3 Microorganisms and Viruses Parenteral products, other than certain vaccines, must be free of microorganisms and viruses [ 191. Products derived from potentially contaminated sources such as human donors, animals, and some cell lines, can be especially problematic. For such products, the purification scheme must demonstrate viral inactivation or removal unless the product is terminally sterilized by validated means (see also Volume 1, Chapter 16). Usually, in-series processing with at least two steps, each capable of six log virus removal or deactivation, would be necessary. Viruses can be removed by ultrafiltration, or deactivated by methods such as heating, treatment with chemicals (e.g., 0-propiolactone), solvents and detergents, and ultraviolet or gamma irradiation. In one study with plasma derived human serum albumin, heat treatment at 60 "C for 10 h in the final container produced more than five log reduction of vaccinia, polio1, vesicular stomatitis, Sindbis and HIV-1 within 10 minutes [31]. In another case, freeze-dried coagulation factors were treated at 80 "C for 72 h in the final vial. For Factor VIII, inactivation of HIV-1 occurred within 24 h, without significant deterioration of the product [31]. For a Factor IX preparation, treatment with solventl detergent combination of tri-(n-butyl) phosphate and Tween- 80 for 5 h inactivated a range of typical enveloped viruses within an hour [31]. Up to six log reduction of some typical enveloped viruses such as herpes simplex-] and Sindbis could be achieved in spiked samples using protein G column chromatography with acid elution; however, only three log reduction was observed for acid tolerant non-enveloped polio virus [31].
1.3.4 Other Contaminants For many biological products, particularly pharmaceuticals, seemingly minor alterations in downstream processing can have important implications on the performance of the product. For example, penicillins may be recovered by liquid-liquid extraction of either the whole fermentation broth or solids-free broth. The latter scheme requires an additional solid-liquid separation step than the whole broth process. However, the whole-broth extracted product has been known to cause more frequent cases of allergenic reactions in comparison with the other processing alternative. In fact, some pharmaceutical companies now demand of contract suppliers that, in addition to meeting product specifications in terms of measurable contamination, the product they supply must conform to a certain production method, in this case extraction after removal of fungal solids. When raw penicillin is for bulk conversion to semisynthetic penicillins, whole-broth extraction may be acceptable in view of the security afforded by the additional steps involved in making and purifying 6-aminopenicillanic acid from raw penicillin [19].
10
1 Strategies in Downstream Processing
1.4 Impact of Fermentation on Recovery Downstream processing should not be considered in isolation with the bioreaction step. Development of biocatalyst by natural selection, mutation, and recombinant DNA technology is a powerful means of influencing downstream processing [32]. Similarly, modification of fermentation feeding strategies, culture media and conditions profoundly affect the downstream process [32].
1.4.1 Characteristics of Broth and Microorganism Composition of the fermentation medium affects downstream recovery. Relatively poorly defined complex media components are often acceptable for producing commodity chemicals and bulk antibiotics, but usually not for parentral proteins. Lowserum and protein-free media are commonly employed in animal cell culture to greatly simplify recovery of sparing amounts of proteins produced. Similarly, the type of antifoam and its concentration must accommodate the recovery constraints. For some processes, alternative microorganisms may be a viable option. Preference should be given to faster growing, easy to process organisms. Selection of a producer must consider the overall productivity of the process, not just that of the fermentation step. Production of recombinant proteins in Saccharomyces cerevisiae may have important advantages relative to production in genetically modified bacteria such as Escherichia coli [33]. S. cerevisiae is generally recognized as safe for food and pharmaceutical use. In addition, unlike bacteria, the yeast does not produce endotoxins, and its broths are much easier to process than those of mycelial fungi and filamentous bacteria [33]. Unlike the DNA-laden homogenates of bacteria such as E. coli, yeast lysates are not excessively viscous. In yet other cases, it may be possible to naturally select autoflocculating strains, as has been done with certain brewing yeasts and bacteria. Cells may also be genetically modified into flocculating ones. Genetic engineering of producing organisms and products provides new opportunities for influencing downstream bioseparations. For example, recombinant fusion proteins with added polypeptide ‘affinity tags’ have been produced to facilitate purification [34,35]. Affinity tags have been developed for ion exchange, hydrophobic interaction, affinity, immunoaffinity and immobilized metal ion chromatography. Specific cleavage sites between the tag and the protein allow removal of the tag after purification [34]. Some of the available affinity tags and the chromatographic methods applied with those tags are listed in Table 1-3. Reagents and enzymes that have been used to cleave the tags, and the specific cleavage sites, are noted in Table 1-4. Another strategy for simplifying downstream recovery is genetic manipulation to enable extracellular secretion of the recombinant protein. Failing outright secretion, it may be possible to achieve secretion into the periplasm of microorganisms such as E. coli. Relatively mild disruption or extraction conditions can then be used for
1.4 Impact of Fermentation on Recovery
11
Table 1-3. Affinity tags and corresponding chromatographic separations [34]. Affinity tag
Chromatography scheme
Polyarginine
Ion exchange
Polypheny lalanine
Hydrophobic interaction
p-Galactosidase
Affinity
Protein A
Affinity
Antigenic peptides
Immunoaffinity
Polyhistidine
Metal ion chelate
recovery in comparison with products produced in the cytoplasm. Periplasmic secretion has additional advantages: periplasm of E. coli contains only seven of the 25 cellular proteases [36], hence, the likelihood of proteolysis is reduced. Moreover, periplasm contains only 100-200 proteins [36], therefore, selective extraction of periplasm yields a less complex, easier to purify mixture. In addition, the oxidative environment of periplasm is more favorable to formation of disulfide bonds than the environment of cytoplasm. Disulfide linkages determine the correct folding of the polypeptide chain and, therefore, its biological activity. Chemicals such as chloroform, Triton X-100, and combinations of lysozyme and EDTA [36] facilitate release of periplasmic proteins. Extraction chemicals should be tested for possible effects on protein stability. In one study, Garrido et al. [33] observed loss of 0-galactosidase activity even at 4°C when the enzyme was extracted with a mixture of chloroform and sparing amounts of sodium dodecyl sulfate (SDS). In larger quantities, SDS is a well-known protein denaturant [37].
Table 1-4. Chemicals and enzymes for specific cleavage of fusion proteins [34]. Cleavage reagent
Cleavage site
Cyanogen bromide
Met 1
Formic acid
Asp
Hy droxylamine
L Pro Asn L Gly
Collagenase
Pro-Val 1 Gly-Pro
Factor Xa
Ile-Glu-Gly-Arg
Enterokinase
Asp-Asp-Asp-Lys
Rennin
His-Pro-Phe-His-Leu-Leu
Carboxypeptidase A
C-terminal aromatic amino acids
Carboxypeptidase B
C-terminal basic amino acids
5
1
12
1 Strategies in Downstream Processing
Secretion or extracellular leakage of an otherwise intracellular product is sometimes achieved simply by modifying the fermentation conditions. For example, addition of penicillin during growth in certain amino acid fermentations produces cells that leak the amino acid which is recovered by isoelectric precipitation from the extracellular fluid.
1.4.2 Product Concentration Concentration of the product in the source material affects the cost of recovery. Concentrations are usually quite low; some values typically seen in culture broths are noted in Table 1-5. In addition to the product, the broth contains many contaminants - proteins, lipids, surfactants, carbohydrates, nucleic acid polymers, salts, components of the culture medium, pigments, organic acids, alcohols, aldehydes, esters, amino acids, and other metabolic products - some of which may be quite similar to the desired product. Some of the contaminants may be toxic or otherwise hazardous (e.g., endotoxins, mycotoxins). Downstream processing typically represents 60-80 % of the cost of production of fermentation products. Thus, superficially it may appear that process improvement should focus on downstream. This is not so. Even small improvements in the yield or purity of the product in the bioreaction step can have a significant effect on downstream recovery costs. As a rough guide, the selling price P (US$ kg-l) of a product (i.e., a reflection of cost of production) depends on its concentration Ci in the broth or the starting material. This dependence can be described by the equation
which is based on data compiled by Dwyer [38]. The potential for yield improvement at the bioreaction stage is usually high. Major yield enhancements have been fairly commonly achieved by strain selection, medium development, optimization of feeding strategies, and environmental controls. Table 1-5. Typical concentration of various products in raw fermentation broth. Final concentration (kg m-3)
Product Vitamin
B12
0.06
Monoclonal antibodies
0.1-0.5
Riboflavin
0.1-7
Antibiotics
0.2-35
Gibberelic acid
1-2
Amino acids
2-100
Yeast
30-60
1.5 Initial Seoarations and Concentration
13
1.4.3 Combined Fermentation-Recovery Schemes In keeping with a global approach to process improvement or intensification, schemes that combine the bioreaction stage and parts of downstream processing are potentially attractive [32]. Such schemes include extractive fermentations, fermentationa-distillation, perfusion culture using membranes, inclined settlers or ‘spinfilters’ to retain the cells in the bioreactor, fermentation-adsorption using chromatographic media, as well as other methods. Combining fermentation and recovery not only reduces the number of individual processing steps, but the productivity of the fermentation may also be substantially enhanced by eliminating or reducing the inhibitory effects of certain products. A novel scheme for retaining particles, particularly animal cells, in perfusion bioreactors relies on standing sound waves applied perpendicular to a vertically aligned harvest flow channel [39-411. The sound waves concentrate the suspended cells in bands aligned with the flow [42,43]. Gravity sediments such aggregated particles against the flow once the sound is switched off; hence, a clarified liquor leaves the flow channel whereas the solids are concentrated in the feed vessel. This type of separation in ultrasonic flow fields provides an effective means of retaining cells in continuous flow bioreactors. This technique allows easy maintenance of sterility as no mechanical items penetrate the sedimentation chamber. Moreover, there is nothing to clog, foul, or breakdown. Process-scale implementation of this method is being developed.
1.5 Initial Separations and Concentration The first few processing operations in a purification train are aimed at volume reduction to minimize processing costs by reducing the size of the downstream machinery. Removal of suspended material and substances which might interfere with further downstream operations are additional requirements of some of the early separation steps. Further, because viscous broths are difficult to handle, viscosity reduction should be achieved as early as possible to simplify pumping, mixing, filtration, sedimentation, etc. Removal of suspended solids, digestion of carbohydrates, or removal of nucleic acids are some of the operations that may be needed to improve broth handling. Typically, solid-liquid separation would be among the first processing steps for extracellular as well as intracellular products. For the latter, solid-liquid separations are usually a means of concentration of the biomass, or removal of the suspending culture fluid prior to disruption or other downstream treatment. Cell or other solid product washing operations often employ solid-liquid separation steps. The commonly used methods of solid-liquid separation are filtration and centrifugation. Centrifuges are used also to separate difficult to break emulsions and other liquid-liquid systems. Some examples are recovery of cream from milk, recovery of oil drops, fats (e.g. in rendering and meat processing plants) and waxes, and liquid-liquid extraction.
14
1 Strategies in Downstream Processing
Table 1-6. Types and applications of centrifuges [1,19]. Tubular bowl. Tubular bowl machines are capable of high g-forces, usually up to 20 000 g in industrial devices. Solids accumulate in the bowl and must be removed manually at the end of operation. Bowl capacity limits solids-holding capability. To ensure sufficient interval between bowl cleaning, the solids concentration in the feed should usually be 5 1 % volumelvolume; higher concentrations can be processed with smaller batches, for example, in production of certain vaccines. Good dewatering of solids is obtained. Multichamber bowl. Similar to tubular bowl machines. Division of bowl into multiple chambers increases solids-holding capacity. Solids must be discharged manually; hence, economic operation is feasible only with feeds with low concentration of solids. Good for polishing of otherwise clarified liquors. Capable of high g-forces. Gradation of g-forces from inner to outer chamber. Smallest particles sediment in the outermost chamber. Good dewatering of solids. Disc-stack. Lower g-forces than tubular bowl machines. Solids may be retained, or discharged intermittently or continuously by various mechanisms (e.g., periodic ejection of solids by hydraulic separation of upper and lower parts of the bowl; nozzle discharge under pressure; valves; etc.). Not all discharge methods are suitable for all solids. Solids must flow. Poor dewatering. Not suited for mycelial solids; good for slurries of yeasts and certain bacteria. Depending on the mechanism of solids discharge, may handle feeds with up to 30 % (v/v) solids. Scroll discharge. Scroll discharge decanter centrifuges are suitable for slurries with high concentration of relatively large, dense solids. Feed solids concentrations of 5-80 % (vlv) can be handled. Solids are discharged continuously. The g-forces are low. Suitable for fungal broths and dewatering of sewage sludge. Perforated bowl or basket centrifuges. Also known as filtering centrifuges. Useful for low-g recovery of relatively large, mostly crystalline solids. The perforated bowl is lined with filter cloth to retain solids, whereas the liquid passes through. Sedimented cake may be washed and recovered as fairly dry material. Not effective for particles below 5 pm, and loadings < 5 % (vlv) [#I.
Solid-liquid separations can be implemented in a variety of ways that are best suited to particular applications. Thus, as detailed in Table 1-6, many different designs of centrifuges are available [ 1,441. Similarly, filtration may be performed in conventional filter presses, horizontal and vertical leaf-type pressure filters, rotary drum pressure or vacuum filters with or without filter aid (or body feed or admix) and using different means of solids discharge. Production scale rotary drum filters tend to be quite large: 0.9-4.3 m drum diameter and up to 6 m drum width. Sterile operation is usually not feasible, and containment is difficult. Alternatively, solids may be recovered by membrane filtration either in dead end (e.g., in many filter sterilizations) or cross-flow modes; the latter may be implemented in flat plate, hollow fiber or spiral wound static membrane cartridges, as well as in dynamic modes [l]. While the variety of available options helps to ensure that specific needs are met, careful consideration of the problem at hand is required for selection of the optimal processing method. Alternatives should be considered whenever possible. For example, rotary drum filters with string discharge usually perform well in separating mycelial solids from penicillin broths, but this discharge mechanism, without filter aids, causes problems with broths of Streptomycetes and other bacteria [19]. Precoat drum filtration may be used with bacterial broths when biomass is not the desired
1.5 Initial Separations and Concentration
15
product. A knife blade (or doctor blade) discharge mechanism is used to continuously remove the deposited solids along with a thin layer of the precoat. Knife discharge without precoat or filter aids is suitable for recovering yeast from the filter cloth on drum filters; however, knife blades are not suited to cleanly cutting away a layer of deposited mycelial fungi because of the stringy nature of solids. Similarly, because of the concentration and the morphology of the solids, the disc stack centrifuge is not suitable for fungal fermentation broths, but properly selected scroll discharge machines are effective. Leaf filters are generally batch devices that are inexpensive to install, but labor-intensive to operate. Leaf filters are suitable for broths with little solids, e.g., in polishing of beer [19]. Gravity sedimentation may be employed as a volume reduction step prior to removal of solids by other means, but sedimentation by itself is not common for biomass removal in processing of high value products. Gravity sedimentation in thickeners and clarifiers [4,5] is encountered widely in sludge recovery in biological wastewater treatment. Certain solids may be recovered using hydrocyclones, but this method is little used in bioprocessing. When more than one processing option its technically feasible, evaluations of the economics of use in terms of capital expenditure on equipment and its operating costs (processing time, yields, labor, cleaning, maintenance, analytical support) is necessary for optimal process selection. Economic evaluations should be performed over the expected lifetime of the equipment [19]. For example, for separation of solids from fermentation broth, centrifugation and microfiltration may be two competing alternatives [ 11. In still other applications, for example when very fragile cells are to be separated from suspending liquid, centrifugation may not be an option. Some other concentration steps, applicable to products in solution, are precipitation [ 1,451, adsorption, chromatography [ 141, evaporation, pervaporation [7] and ultrafiltration [ 11. Some of these operations are equally capable purification steps (e.g., chromatographic separations). Certain steps (e.g., some chromatographic separations ; membrane separations) may require a relatively clean process stream, free of debris, lipids or micelles which may cause fouling of the equipment. Such steps are often used downstream of steps which can handle cruder material [19]. Sometimes the characteristics of fermentation broth or process liquor may be modified by pretreatment to enable processing by a certain method. Major changes in processing characteristics may be achieved by pH and/or temperature treatment, use of additives such as polyelectrolytes, other flocculants and enzymes, and changes in ionic strength [19]. Flocculants (e.g., alum, calcium and iron salts, tannic acid, quaternary ammonium salts, polyacrylamide) can enhance sedimentation rates by thousands of fold relative to unflocculated suspension. Aging of protein precipitates and crystals can substantially improve filtration and sedimentation. Addition of salts is sometimes helpful in dewatering difficult to dewater solids such as protein precipitates. Water is drawn out of the pores of the solid into the salt containing liquid film on the outside. Osmosis or chemical potential difference drives the flow. Among other factors, time of harvest can beneficially alter processing behavior of the broth as well as the stability of the labile product. Culture conditions and methodology influence microbial morphology, product formation and downstream recovery. For example, cells grown in defined media are generally easier to disrupt
16
I Strategies in Downstream Processing
than ones cultured in complex media [ 161. Also, high specific growth rates produce less robust cells.
1.6 Intracellular Products In general, a biological product is either secreted into the extracellular environment, or it is retained intracellularly. In comparison with the total amount of biochemicals produced by the cell, very little material is usually secreted to the outside; however, this selective secretion is itself a purification step which simplifies the task of the biochemical engineer. Extracellular products, being in a less complex mixture, are relatively easy to recover. On the other hand, because a greater quantity and variety of biochemicals are retained within cells, intracellular substances are bound to eventually become a major source of bioproducts [16]. Among some of the newer intracellular products are recombinant proteins produced as dense inclusion bodies in bacteria and yeasts. Recovery of intracellular products is more expensive as it requires such additional processing as cell disruption [ 16-18], lysis [ 161, permeabilization [46], or extraction. Intracellular polymers such as poly-P-hydroxybutyrate (PHB) may be recovered either by cell disruption [17,37] or solvent extraction. In principle, selective release of the desired intracellular products is possible, but in practice it is neither easily achieved nor sufficiently selective. Hence, the desired product must be purified from a relatively complex mixture, complicating processing and adding to the cost [19]. Nevertheless, an increasing number of intracellular products are in production. Economics of production may be improved by recovering several products (intracellular and extracellular) from the same fermentation batch P11. As for other separations, many options exist for the disruption of cells (Table 1-7). Of these, high-pressure homogenization is apparently the most suitable for bacterial broths, whereas bead mills are more widely used for fungal cultures [1,16]. For dissolved products, cell disruption conditions (e.g., pressure, number of passes) must be selected to prevent excessive micronization of debris because micronization complicates solid-liquid separation further downstream [ 161. However, when the product is an intracellular solid that is undamaged by homogenization, micronization of debris actually favors product recovery. This strategy is useful with protein inclusion bodies, certain cellular organelles, and sometimes with granules of bioplastics such as polyhydroxyalkanoates. Nonetheless, overzealous disruption conditions should be avoided in view of the recently published evidence that suggests loss of intracellular solids by micronization [37]. Disruption of bacterial cells releases large amounts of nucleic acids which increase the viscosity of the broth, often producing viscoelastic behavior. To ease further purification, the nucleic acids are usually removed by precipitation (e.g., with manganous sulfate, streptomycin or polyethyleneimine) [11; alternatively, viscosity may be reduced by enzymatic digestion of nucleic acids or high-shear processing in high-pressure homogenizers [ 191. Another alternative for eliminating nucleic acid polymers is heat shock treatment prior to disrupting the cells. Heat shock treatment
1.7 Some Specific Bioseparations
17
Table 1-7.Cell disruption options [16-181. High-pressure homogenization. Frequently used for large-scale disruption of yeasts and nonfilamentous bacteria. Generally not suitable for mycelial broths. Broth must be free of large suspended solids, tight cell clumps and flocs. Maximum acceptable particle size is about 20 pm, but a lower size is preferred. Slurry viscosity should not normally exceed 1 Pas [1,16]. Optimal viscosity and solids concentration ranges are narrower than for bead mills. Bead milling. Bead mills come in vertical and horizontal configurations with different mechanisms for retention of grinding media, and different types of agitators. Agitators that reduce back-mixing are preferred. Vertical mills are susceptible to fluidization and accompanying loss in performance. Typically three to six passes should achieve complete disruption. Useful for yeasts, mycelial fungi, algae; less efficient with bacteria. Grinding bead size affects disruption. Smaller the microbial cell, smaller the optimal bead size [16,33]. Autolysis. Under suitable conditions certain cultures would autolyse in the stationary phase upon completion of fermentation. Baker’s yeast can autolyse. Osmotic shock. Useful for animal cells and in specific cases for bacteria. Large dilutions may be necessary. Thermolysis. Sufficiently heat-stable products may be released by heat shocking the cells. Microbial susceptibility to heat shock treatment varies widely. Monvalent metal ions such as Na+ and K+ may aid thermolysis. Suited to specific cases. Enzymes and chemicals. Detergents, EDTA, solvents (e.g., toluene), antibiotics, and lytic enzymes may be used. Sometimes enzymes and chemical additives are used in combination with homogenization or bead milling to reduce the severity of mechanical treatment. Treatment with acids and alkalis may be useful in specific cases. Especially useful for extraction from periplasm. Others. Ultrasonication, desiccation, freeze-thaw, extrusion of frozen paste. Applicable only to laboratory scale.
would typically require rapid heating to at least 64°C and a holding time of 20-30 minutes. This treatment should digest almost all DNA/RNA. Shorter holding times may be satisfactory if complete degradation is not necessary for processability. Rapid temperature rise preferentially destroys proteases relative to RNA-hydrolyzing enzymes. Thermal treatment may be feasible for heat stable products [37] as well as for those produced as denatured inclusion bodies. Processing considerations relevant to some specific bioseparations are discussed in the following section.
1.7 Some Specific Bioseparations 1.7.1 Precipitation Proteins are easily concentrated by precipitation with organic solvents (e.g., ethanol, acetone), polymers (e.g., poly(ethy1ene glycol), poly(propy1ene glycol), dextran),
18
1 Strategies in Downstream Processing
and salts. Fractional precipitation allows for a degree of separation [ 11. Fractionation with ammonium sulfate is commonly used. Organic solvents produce a denaturing environment making low temperature processing necessary [ 11. Alcohol precipitation is frequently used in recovering biologically inactive dissolved polymers such as polysaccharides. Examples include precipitation of xanthan and gellan with isopropanol. Precipitation methods can handle large amounts of crude material, are easily scaled up, and can be implemented in continuous processing modes [ 1,471. However, precipitation is generally not useful for recovery from very dilute animal cell culture fluids. Ammonium sulfate precipitation for recovery of recombinant P-galactosidase from S. cerevisiae has been detailed by Zhang et al. [47].
1.7.2 Foam Fractionation Foam fractionation, microflotation or froth flotation is potentially useful for concentrating particles (cells, organelles, other small solids such as granules of PHB) and proteins into a foam phase for further recovery. The technique involves gentle bubbling of air (or other inert gas) at the base of a column of broth or solution. Hydrophobic solids and surface active molecules accumulate at the gas-liquid interface and rise with the bubbles. Collector surfactants and other promotors are often added to improve attachment. Additives such frothing agents and stabilizers may be necessary. Enrichment in the foam depends on physical collection efficiency of bubbles (i.e., on bubble size, hydrodynamics, bubbling rate, concentration of particles) and adsorption chemistry. Empirical investigation is essential for selecting suitable additives, concentrations, hydrodynamic regimes, and for assessing performance, including recovery from the foam phase. Culture conditions may be used to influence adsorption behavior. Froth flotation is encountered only occasionally in bioprocessing. Potentially, fermenters used in batch cultivation could subsequently be employed for froth flotation. Airlift bioreactors with gas-liquid separators [48] and added means of skimming the gas-floated biomass are used in activated sludge treatment of wastewater. Part of the harvested sludge is returned to the reactor as inoculum.
1.7.3 Solvent Extraction Rapid solvent extraction can be carried out in centrifugal extractors such as the Podbielniak and the Alfa Lava1 machines that are commonly used in antibiotics processing [l, 131. These devices were originally designed to handle solids-free liquids, but have been adapted to media containing limited amounts of small particles. Other more conventional extractors are banks of mixer-settlers, York-Scheibel column (suitable for solids-free liquids), and the reciprocating plate Karr column (suitable for whole broths). Supercritical extraction of solids and liquids with carbon dioxide or other solvents (e.g., pentane) may be useful for small organic solutes. In these
1.7 Some Specific Bioseparations
19
cases a concentrated solute is obtained easily by boiling off the solvent. Recently, serum albumin has been extracted into aqueous reverse micelles formed in carbon dioxide using a perfluoropolyether surfactant [49]. This opens up new opportunities for purification of proteins and other large molecules.
1.7.4 Aqueous Liquid-Liquid Extraction and its Variants Conventional liquid-liquid extraction based on partitioning between an aqueous phase and a water-immiscible organic solvent is not suitable for proteins and protein-based cellular organelles because of low protein stability in organic solvents. A suitable alternative is partitioning between two immiscible aqueous phases [1,8,9]. Such phases are obtained by adding two incompatible polymers - for example, poly(ethy1ene glycol) and dextran - to water, or by mixing a relatively hydrophobic polymer solution with salts. Examples of such systems are aqueous mixtures of PEG-PVA, PPG-dextran, PPG-potassium phosphate, PEG-ammonium sulfate, as well as others. Partitioning of solutes is brought about by differences in net charge and hydrophobicity. Higher-polarity molecules solubilize preferentially in the saltrich phase, whereas the relatively hydrophobic molecules concentrate in the polymer-rich phase. Polymers with attached affinity ligands - hydrophobic and ionizable functional groups - can improve partitioning behavior. Partitioning is strongly affected by pH, composition and type of phases (e.g., molecular weight of polymer, ionic strength, salt, polymer). In addition, the volume ratio of the phase mixture to that of the protein solution should be such that neither phase approaches saturation with protein. Aqueous two-phase systems have been successfully employed for enrichment of proteins, cells, organelles, and small molecules. Proteins that extract into the polymer phase are back extracted into the salt phase for recovery. Phase separation can be slow because of high viscosity and small density differences. Gravity separation is generally satisfactory for PEG-salt systems, but centrifugal separation may be necessary for PEG-dextran. Aqueous two-phase extraction is commercially employed, but it is relatively uncommon. Among relatively new developments in liquid-liquid extraction is reversed miceller extraction [ 121 also known as liquid membrane emulsion extraction. Reversed micelles are surfactant stabilized microdroplets of an aqueous phase suspended in a water-immiscible solvent. Contacting the reversed micelle-laden organic phase with an aqueous mixture of proteins or other solutes results in preferential transfer of one or more species from the aqueous phase to the organic phase, and from there to the aqueous core of the reversed micelles. The intervening organic phase constitutes a liquid ‘membrane.’ Extraction is influenced by pH and ionic strength of the bulk aqueous phase, and the nature of the reversed miceller core. Usually, a protein solubilizes in the reverse miceller phase at pH values below its isoelectric pH when the ionic strength is low. Once a component has been extracted, reversed micelles can be back-extracted with buffers to yield a solution rich in the desired substance. Back-extraction is favored by altering the pH and ionic strength. Factors such as hydrophobicity of the protein also contribute to partitioning behavior.
20
1 Strategies in Downstream Processing
A variation of the liquid membrane emulsion extraction is the supported liquid membrane extraction [ l l ; see also Volume 1, Chapter 111. No stabilizing surfactant is necessary in this case; instead, the liquid membrane-forming organic phase is supported in the pores of a porous solid that separates the two aqueous phases. Additives may be employed to enhance mass transfer through the organic phase [ 111. Reversed micelles and liquid membranes are not widely used at present.
1.7.5 Membrane Separations Cross-flow membrane filtration flux typically ranges over 10-1 20 L.m-*.h-'; the exact value depends on the membrane pore size and the viscosity of the suspending fluid. Microfiltration of animal cells and microbial homogenates is done best at transmembrane pressures less than 1.38 X lo4 Pa. Higher pressures, typically 6.934.5 x lo4 Pa, are used in recovering microbial cells. Because of the small pore size, ultrafiltration membranes invariably require high transmembrane pressures (13.8-27.6 x lo4 Pa) for reasonable flux. Polymer membranes predominate in bioprocessing, but ceramic and sintered metal membranes are used occasionally. Hydrophilic membranes are preferred for liquids. Hydrophobic polymer membranes are easily fouled by silicone antifoams which may cause as much as 50 % decline in flux. Low-molecular weight poly(propy1ene glycol) or poly(ethy1ene glycol) based antifoams are usually better. Mechanical foam control [24,50] during fermentation is sometimes helpful in eliminating or reducing antifoam consumption. Even without antifoams, membrane performance deteriorates over time, making periodic replacement necessary. Prior experience or experimentation are the only reliable predictors of membrane life [ 6 ] .Membranes are not easily cleaned; detectable residues of bioactive material may remain after any reasonable cleaning. Such situations require product-dedicated filters to prevent cross-contamination. Furthermore, polymer-based membrane filters cannot usually be heat sterilized; chemical sanitization and atmospheric steaming are the only options. Chemical cleaning, sanitization, and steaming lower membrane life; hence the choice of chemicals and cleaning conditions need to be carefully assessed. The major costs associated with ultrafiltration and microfiltration are the initial capital expense and the cost of membrane replacement; energy is not a major expense. The frequency of membrane replacement determines feasibility of membrane separations. In contrast, in reverse osmosis where the high transmembrane pressure is unavoidable, pumping expense and membrane replacement costs are major contributors to operating costs. As with centrifuges, membrane filter selection requires experimental evaluations [ 1,6]. Even in cross-flow operation, membrane filters experience performance loss due to concentration polarization or accumulation of a solute layer at the surface of the membrane. Small amounts of relatively large, dense inert solids such as cellulose fibers or polymer beads added to the feed are known to reduce concentration polarization by disturbing the fluid boundary layer on the membrane surface. Cross-flow
1.7 Some Specific Bioseparations
21
channels are sometimes also inserted with static turbulence enhancers such as wire screens, but such filter modules are not suitable for mycelial or filamentous biomass especially at high concentration of solids. Mechanical methods of increasing turbulence are employed in dynamic filters, but few such devices have gained any commercial acceptance. One dynamic configuration utilized two porous concentric cylinders with microfiltration membranes supported on the surfaces of the annulus. The inner cylinder rotated at high speed; differences in angular velocities of the fluid elements along the width of the annular gap produced Taylor vortices that substantially enhanced filtrate flux relative to static cross-flow operation [5 11. Nonetheless, limited scale-up potential prevented further development. A variation on the concentric cylinder theme has recently been introduced by Pall Filters. This design consists of a stack of supported circular microfilter membranes with mechanically agitated circular steel discs mounted inbetween. Rotation of discs dramatically enhances filtrate flux [52]. The stack supports up to 1.5 m2 membrane surface, but this may be substantially increased in future designs simply by increasing the overall height of the stack. The device is suited to recovering yeasts and non-filamentous bacteria from relatively less viscous broths. Membrane filters are used also in the diafiltration mode for buffer exchange, washing of solids, desalting, and removing other small molecules from solution of macromolecules. Pervaporation is another membrane separation that is particularly useful for low energy recovery of relatively volatile liquids (e.g., ethanol) from fermentation broths [ 7 ] . Permselective membranes separating the broth from a vapor phase allow only selective permeation of the desired solvent to the other side, where hot air or heat supplied to the membrane continuously evaporates the solvent, hence maintaining a mass transfer driving force. Membrane chemistry determines permselectivity.
1.7.6 Electrically Enhanced Bioseparations Electric fields may be used to enhance bioseparations [53,54], but commercial use is limited at present because of the damaging effects of ohmic heating that accompanies current flow. Electrolysis can be another problem. Nevertheless, electrokinetic forces on charged particles have been demonstrated to reduce concentration polarization and membrane fouling during microfiltration and ultrafiltration, thereby enhancing filtration rates [53]. Up to sevenfold enhancement of transmembrane flow has been recorded during microfiltration with direct current (DC) electric field strengths of 100-120 V cm-' [53]. Some of the problems associated with electric fields may be reduced by replacing the steady DC fields with pulsed direct current fields [53]. Electric discharges have been used also to break foams instantaneously during processing. The separation potential of electric fields is best illustrated by electrophoresis, which is a well-established extremely high resolution method for separation of proteins. Differences in molecular charge and weight are the bases of separation. However, despite attempts to scale-up [%], electrophoresis remains confined mostly to
22
I Strategies in Downstream Processing
laboratory use. Except for small volumes, rapid removal of heat generated has proven difficult without convective mixing that would destroy any separation.
1.7.7 Chromatographic Separations Enhancing speed has been a major preoccupation with chromatographic processing. Except for bed height dependent gel permeation, the speed of most chromatographic processes can be enhanced by replacing the usual high-resistance packed vertical columns with radial flow devices [ 5 5 ] . Adsorption media used in conventional columns can still be utilized, but the medium is packed in the annulus between two porous concentric cylinders. Radial flow columns attain 10- to 50-fold greater flow rates than conventional columns [ 5 5 ] . Industrial-scale simulated moving bed chromatographic systems are now available [56]. Among other improvements, better, more rigid yet porous chromatographic media that are less susceptible to bed compression have been developed [14]. Other novel media have enabled extremely high speed or perfusion chromatography. Unlike conventional media, perfusion media contain throughpores for bulk flow of fluid through the particle. Diffusional pores as in conventional media are also present. Throughpores allow high flow rates - up to 100-fold greater than in diffusive media [ 5 5 ] . Resulting convection within the particles reduces diffusive transport limitations. Another high-rate chromatographic system is expanded or fluidized bed chromatography. The medium bed is expanded or fluidized during loading by upflow of unclarified fermentation fluid or cell homogenate [ 5 5 ] . There is little pressure drop through the expanded bed. Plug flow of fluid is desired and easily attained. After adsorption, the microbial solids are washed away by upflow of water or buffer. The adsorbed product is recovered as in conventional chromatography by downward elution of settled, packed bed. Because this method handles unclarified fluids, some solid-liquid separation steps are eliminated. Fluid bed chromatography has been demonstrated with numerous fluids including broths E. coli, yeast, mammalian cells [ 5 5 ] , autolysed yeast, and blood plasma. A further rapid chromatographic method that may potentially handle solids-laden fluids is membrane chromatography. This technique employs ion exchange groups or other high-specificity adsorption ligands attached to inner surfaces of pores of conventional microfiltration membranes. Rapid flow through pores reduces diffusion limitations, hence speeding adsorption, and, later, desorption. Hollow fiber membrane modules that allow compact packing of large membrane areas have been used for membrane chromatography [57]. Some especially high-resolution chromatographic separations include HPLC and bioaffinity-based methods. Process-scale HPLC continues to be useful for small batches [38], but this method is expensive, slow, and the high-pressure columns appear to have reached an upper limit of about 0.3 m diameter and 2.4-3.0 m height. Bioaffinity chromatography with affinity ligands - receptors, antibodies, enzymes, and other active proteins - immobilized onto the support media has been used for
1.8 Recombinant and other Proteins
23
quite some time, but it remains expensive. Other problems are often poor stability of the affinity matrix, and ligand leakage into the product (see also Volume 1, Chapter 17). With few exceptions (e.g., protein A affinity columns can be cleaned with the strong denaturant guanidine hydrochloride (6 M) which solubilizes adsorbed proteins without affecting the ligand), ligand stability limits the column cleaning regimen. Because of those factors, a trend toward replacing labile bioaffinity ligands with inexpensive and robust alternatives (e.g., dyes, metal ions) is apparent. Note that some of the speed-enhancing techniques used with chromatography are equally applicable to non-chromatographic adsorptions. Adsorption using columns or slurries of activated carbon is commonly encountered in bioprocessing, particularly for removing pigments.
1.8 Recombinant and other Proteins Many of the newer recombinant biotechnology products are proteins [30,58]. While the general features of a bioseparation scheme for these products are the same as for other proteins, there are some unique constraints. Genetically modified microorganisms and cells of higher life forms are often more fragile than the corresponding wild strains [59,60]. This has implications for the design of cell-liquid separation stages. Also, recombinant proteins formed in bacteria and yeasts frequently precipitate inside the cell as dense, insoluble, denatured inclusion bodies. In this form proteins which may otherwise be toxic to the cell may be overproduced and remain protected against proteolytic activity within the cell. Most bacteria and fungi used in producing recombinant proteins also produce a variety of proteases that may degrade some of the desired protein within the cell and during recovery, soluble, non-inclusion body proteins being particularly susceptible to degradation. Degradation by acid proteases with a pH optimum of 2-4 may be minimized by processing at higher pH and low temperatures. Neutral proteases are not particularly thermostable and may be inactivated by heating to 60-70 "C for 10 minutes [19]. Many proteases are metalloproteins and require a divalent metal ion for proteolytic activity; chelating agents such as ethylenediaminetetraacetic acid (EDTA) or citric acid may be used to inactivate such proteases by binding the metal ions. Alkaline proteases of Bacillus sp., such as subtilisin, contain serine at the active site and are not affected by EDTA, but are inhibited by diisopropylfluorophosphate. The short-lived reagent phenylmethylsulfonyl fluoride protects against serine proteases. Antioxidants such as vitamin E and ascorbic acid protect against oxidation [ 191. Proteins tend to be more stable in concentrated solutions. Addition of poly(ethylene glycols) and other proteins such as albumins may have a stabilizing effect. Glycerol, sucrose, glucose, lactose, and sorbitol are often used as stabilizers in concentrations of 1-30 %. Enzyme substrates usually have an stabilizing effect, as do high concentration of salts such as ammonium sulfate and potassium phosphate. Metalloproteins may be stabilized by addition of metal salts. Divalent metal ions such as Ca2+, Cd2+, Mn2+ and Zn2+ stabilize various enzymes [19].
24
1 Strategies in Downstream Processing
Some commonly used sequences of protein purification methods have been outlined by Bonnerjea et al. [45] and by Wheelwright [3]. Chromatographic procedures are indispensable to producing high-purity proteins. Typically, the mean recovery or yield of separation steps such as those listed in Table 1-1 is 60-80 % [45]. Average and high values of purification factors associated with some protein purification operations are shown in Table 1-8 which is based on data compiled by Bonnerjea et al. [45]. Clearly, affinity chromatography far outperforms other methods, but compared with operations such as ion-exchange chromatography, the scope for further improving performance is small because many affinity separations already operate close to theoretical maximum [45]. Changes in processing volume, product yield, and total and specific activities occur during processing as illustrated in Table 1-9 for a relatively simple purification of brain tumor plasminogen activator (PA) from supernatants of cultured, anchorage dependent rat cells [61]. The purification in Table 1-9 was done at 4 "C. The serumfree conditioned medium used for recovery had an initial plasminogen activator activity of only 9 IU mL-' [61]. Zinc chelate-agarose chromatography was used as the first concentration/purification step. The culture fluid (6 L) was applied to the column ( 5 x 8 cm) at a flow rate of 200 mL h-l. The column was washed with Tris-HCI buffer (0.02 M, pH 7.5, 1 L) that contained 1 M sodium chloride, aprotinin and Tween-80 (0.01 %, vol/vol). Aprotinin, a protease inhibitor, and Tween-80 (poly(oxyethy1ene sorbitane monooleate)), a surfactant, are generally added at all stages of PA processing to, respectively, suppress proteolysis and overcome the surface adherent tendency of plasminogen activators [30]. After the wash, the column was eluted with a linear gradient of imidazole (0-0.05 M) in the wash buffer (1 L, 120 mL h-l). Pooled PA fractions were further purified on a concanavalin A-agarose affinity chromatography column. Dialysis was used to concentrate the pooled fractions, and a final gel filtration step (Sephadex G-150 superfine) was employed. The overall yield was 39 % [61]. This figure is fairly typical of large-scale protein
-
Table 1-8. Approximate values of purification factors observed during protein purifications. Based on Bonnerjea et al. t4.51. Operation
Affinity chromatography
Purification factor Average
High
100
3000
Dye-ligand affinity
17
-
Inorganic adsorption
12
100
6
100
15
60
Ion-exchange chromatography
8
50
Detergent extraction
4
12
Precipitation
3
12
Size-exclusion chromatography Hydrophobic interaction chromatography
25
1.8 Recombinant and other Proteins Table 1-9. Purification of tumor plasminogen activator [61].
Volumetric Specific Yield Purification Volume Total Total activity factor protein activity activity (IU mL-') (IU mg-') (%) (mL) (mg) (IU) Clarified medium Zinc chelate-agarose
100
500
362
94
19400
373
22750
37
116
0.53 20 800
2773
39000
39
199
270
53 000
100
138
50 000
Concanavalin A-agarose Gel filtration
52 7.5
1
196
6000
2.4
8.8
1.9
recovery. For example, overall recoveries of 23-47% were noted for a variety of processes (e.g., recombinant BST, recombinant human a-interferon, L-leucine dehydrogenase for use in chiral syntheses) reviewed by Wheelwright [3]. One exception was a somewhat impractical process for tissue-type plasminogen activator (tPA) for which the overall yield was only 6 % [3]. Other methods for large-scale tPA recovery have been presented by Rouf et al. [30].
1.8.1 Inclusion Body Proteins When possible, production of recombinant proteins as inclusion bodies has important advantages. Some proteins that form inclusion bodies are listed in Table 1-10. Inclusion bodies are easy to isolate, highly concentrated forms of the desired recombinant protein. Typically, inclusion bodies are spheroidal particles, 0.2-2.0 X m in diameter and 1100-1300 kg mP3density. The sequence of steps in recovery of inclusion body proteins is cell disruption, centrifugal separation of the inclusion body, washing, solubilization of the protein, and renaturation [ 191. Cell disruption by homogenization is the preferred technique in large-scale processing. Disruption by high-pressure homogenization has been detailed by Chisti and Moo-Young [ 161. Inclusion bodies are not affected by homogenization. Cell homogenates are centrifuged to sediment the dense inclusion body fraction. Centrifugation at 1000-12 000 g for 3-5 minutes is sufficient. Sedimentation of cell debris can be minimized by Table 1-10. Some proteins produced as inclusion bodies [19]. Bovine pancreatic ribonuclease
Human interleukin-2
Lysozyme
Bovine somatotropin (BST)
Human interleukin-4
Porcine phospholipase
Epidermal growth factor
Human macrophage-colony stimulating factor
Prochymosin
Human insulin Human y-interferon
Human serum albumin Immunoglobulins
Pro-urokinase Tissue-type plasminogen activator
26
1 Strategies in Downstream Processing
increasing the density and viscosity of the homogenate with additives such as 30 % sucrose or 50 % glycerol. The inclusion body fraction is washed with buffers containing 1 M sucrose, 1-5 % Triton X-100 surfactant [62] and, in some cases, low concentrations of proteolytic enzymes and denaturants. The wash steps remove soluble contaminants, membrane proteins, lipids and nucleic acids. At this stage the remaining solids fraction is > 90 % recombinant protein. The protein solids are solubilized in highly denaturing chaotropic media. Typically, 6-8 M guanidine hydrochloride or 8 M urea are used for solubilization at pH 8-9, 25-37 "C for 1-2 h [62]. Reducing agents are added to the solubilization media to break any inter- and intra-molecular disulfide bonds to fully solubilize the protein. Some reducing agents are 2-mercaptoethanol, dithiothreitol, dithioerythritol, glutathione, and 3-mercaptopropionate. Some typical concentrations are 0.1 M 2-mercaptoethanol, or 10 mM dithiothreitol [62]. The latter has a shorter half-life than 2-mercaptoethanol, but does not have the odor of 2-mercaptoethanol. Stability of thiol compounds in solution is dependent on pH, temperature, and the presence of metal ions such as Cu2+,which lower stability, and of stability enhancers such as EDTA. Good yields of some proteins can be obtained by solubilization without the reducing reagents, but for others reducing agents are essential. Of the denaturants, guanidine hydrochloride is preferable to urea, which may contain cyanate causing carbamylation of the free amino groups on the protein, particularly during long incubation periods in alkaline environments. Note though, that for some proteins, one denaturant may produce significantly higher overall yield than if solubilization with the other is used [19]. Performance has to be empirically evaluated. For refolding of solubilized protein into active entities, concentration of the denaturant and the reducing agent are reduced by dilution with a refolding buffer. Denaturants can be completely removed by ultrafiltration with addition of renaturing buffer, dialysis, or gel filtration. Renaturation from concentrated protein solutions produces lower yields of the active protein because of intermolecular aggregation in these solutions. Thus, renaturation is done at low protein concentrations, typically kg m-3 protein [62]. Yield of the active protein is enhanced by refold1-20 x ing in the presence of small, non-denaturing amounts (1-2 M) of urea or guanidine hydrochloride [62]. Presence of high-molecular weight polymers such as poly(ethylene glycol) may also improve yield [ 191. During refolding, formation of the disulfide bonds is achieved in one of three ways. The air oxidation method uses dissolved oxygen for oxidation of the cystine residues. The refolding buffer containing solubilized protein is aerated or exposed to atmosphere. Oxidation is accelerated by Cu2+ions at approximately M. Typical reaction conditions are pH 8-9, 4-37°C for up to 24 h [62]. Traces of 2-mercaptoethanol may enhance yield. Air oxidation is difficult to control [19]. The glutathione reoxidation method typically uses a 1O:l mixture of reduced and oxidized forms of glutathione at a concentration of M reduced glutathione [62]. Air oxidation is suppressed by using deaerated buffers held under a nitrogen atmosphere. The ratio of the reduced and oxidized forms of glutathione, the ratio of the glutathione and the cystine residues on the protein, the reoxidation temperature (4-37 "C) and time (1-150 h) provide flexibility to this method [62]. Low-molecular weight thiols other than glutathione may also be used. The third method of disulfide
Abbreviations and Symbols
27
bond formation, the mixed disulfide interchange technique, has been detailed by Fischer [62]. The inclusion body production stage should be optimized to rapidly form relatively pure, large and dense inclusion bodies which are easy to recover and solubilize. Production of proteolytic activity should be suppressed as far as possible. Purification and concentration are greatly simplified because of the already high starting protein concentration and purity in the inclusion bodies which are easy to separate from the bulk of the soluble proteins by centrifugation. The recovery of active protein from inclusion bodies is variable, but can approach 100%. In general smaller polypeptides are easier to refold into active forms [19]. Because of added processing, and the need to refold in dilute solutions, inclusion body-produced proteins tend to be expensive. With certain proteins such as tPA production as an inclusion body in bacteria is technically feasible but is not competitive with animal cell culture-derived product [30], even though the latter is a fairly expensive production method.
1.9 Conclusions The variety of bioseparations is vast, but usually a small selection of the available methods is sufficient to achieve the requisite purity. The aim always is to employ the fewest possible process steps consistent with the product quality specifications. In-depth knowledge of individual separations must be combined with insight into the bioreaction step to design an efficient, consistent and integrated overall production process. Whereas the scientific understanding of bioseparations continues to improve and several new capable separations have been introduced, downstream processing of biologicals remains an empirical art. Invariably, experimentation must be relied upon to aid process selection, implementation, and scale-up.
Abbreviations and Symbols BST C Ci DC DNA EDTA g
GMP HIV LAL n Nf P
bovine somatotropin number of components concentration of product in broth or starting material, kg mP3 direct current deoxyribonucleic acid ethylenediaminetetra-acetic acid gravitational acceleration, m s - ~ Good Manufacturing Practice human immunodeficiency virus Limulus amoebocyte lysate number of steps number of possible flowsheets selling price, U.S. $1984 kg-'
28
I Strategies in Downstream Processing
PA PEG PHB PPG PVA RNA S SDS x
plasminogen activator poly(ethy1ene glycol) poly-P-hydroxybutyrate poly(propy1ene glycol) poly(viny1 alcohol) ribonucleic acids number of separation operations sodium dodecyl sulfate step yield
References [l] Chisti, Y., Moo-Young, M., in: Biotechnology: The Science and the Business: Moses, V., Cape, R.E., (Eds.), New York: Harwood Academic Publishers, 1991, pp. 167-209. Fermentation technology, bioprocessing, scale-up and manufacture [2] Belter, P.A., Cussler, E.L., Hu, W.-S., Bioseparations: Downstream Processing f o r Biotechnology. New York: John Wiley, 1988. [3] Wheelwright, S.M., Protein Purification: Design and Scale up of Downstream Processing. New York: Hanser Publishers, 1991. [4] Christian, J.B., Chem Eng Prog, 1994, 90(7), 50-56. Improve clarifier and thickener design and operation. [5] Tiller, EM., Tamg, D., Chem Eng Prog, 1995, 91 (3), 75-80. Try deep thickeners and clarifiers. [6] Gyure, D.C., Chem Eng Prog, 1992, 88(11), 60-66. Set realistic goals for cross-flow filtration. [7] Fleming, H.L., Chem Eng Prog, 1992, 88(7), 46-52. Consider membrane pervaporation. [8] Abbott, N.L., Hatton, T.A., Chem Eng Prog, 1988, 84(8), 31-41. Liquid-liquid extraction for protein separations. [9] Raghavarao, K.S.M.S., Rastogi, N.K., Gowthaman, M.K., Karanth, N.G., Adv Appl Microbiol, 1995, 41, 97-171. Aqueous two-phase extraction for downstream processing of enzymes/proteins. [lo] Chisti, Y., in: Encyclopedia of Bioprocess Technology; Flickinger, M.C., Drew, S.W., (Eds.), New York: John Wiley, 1999; in press. Solid substrate fermentations, enzyme production, food enrichment. [ I l l Patnaik, P.R., Biotechnol Adv, 1995, 13, 175-208. Liquid emulsion membranes: principles, problems and applications in fermentation processes. [12] Pyle, D.L., J Chem Technol Biotechnol, 1994, 59, 107-108. Protein separation using reverse micelles. [ 131 Schugerl, K., Solvent Extraction in Biotechnology, New York: Springer-Verlag, 1994. [I41 Chisti, Y., Moo-Young, M., Biotechnol Adv, 1990, 8, 699-708. Large scale protein separations: engineering aspects of chromatography. [I51 Snowman, J.W., Adv Biotechnol Process, 1988, 8, 315-351. Lyophilization techniques, equipment, and practice. [16] Chisti, Y., Moo-Young, M., Enzyme Microb Technol, 1986, 8, 194-204. Disruption of microbial cells for intracellular products. [17] Harrison, S.T.L., Biotechnol Adv, 1991, 9, 217-240. Bacterial cell disruption: A key unit operation in the recovery of intracellular products. [18] Middelberg, A.P.J., Biotechnol Adv, 1995, I S , 491-551. Process-scale disruption of microorganisms.
References
29
[19] Chisti, Y., Moo-Young, M., I Chem E Symp Sec 1994,137, 135-146. Separation techniques in industrial bioprocessing. [20] Fish, N.M., Lilly, M.D., Biotechnology, 1984,2, 623-627. The interactions between fermentation and protein recovery. [21] Dunnill, P., Process Brochem, 1983, 18(5), 9-13. Trends in downstream processing of proteins and enzymes. [22] Chisti, Y., Chem Eng Prog, 1992, 88(1), 55-58. Build better industrial bioreactors. [23] Chisti, Y., Chem Eng Prog, 1992, 88(9), 80-85 Assure bioreactor sterility. [24] Chisti, Y., Moo-Young, M., J Znd Microbiol, 1994, 13, 201-207. Clean-in-place systems for industrial bioreactors: design, validation and operation. Flickinger, M.C., Sansone, E.B., Biotechnol Bioeng, 1984, 26, 860-870. Pilot- and production-scale containment of cytotoxic and oncogenic fermentation processes. Lubiniecki, AS., Wiebe, M.E., Builder, S.E., in: Large-Scale Mammalian Cell Culture Technology; Lubiniecki, A.S., (Ed.) New York: Marcel Dekker, 1990; pp. 515-541 . Process validation for cell culture-derived pharmaceutical proteins. Willig, S.H., Stoker, J.R., Good Manufacturing Practices for Pharmaceuticals: A Plan for Total Quality Control, 3rd edition, New York: Marcel Dekker, 1992. Garg, V.K., Costello, M.A.C., Czuba, B.A., in: Purification and Analysis of Recombinant Proteins; Seetharam, S., Sharma, S.K., (Eds.), New York: Marcel Dekker, 1991; pp. 2954. Purification and production of therapeutic grade proteins. Anicetti, V.R., Keyt, B.A., Hancock, W.S., Trends Biotechnol, 1989, 7(12), 342-349. Purity analysis of protein pharmaceuticals produced by recombinant DNA technology. Rouf, S.A., Moo-Young, M., Chisti, Y., Biotechnol Adv, 1996, 14, 239-266. Tissue-type plasminogen activator: characteristics, applications and production technology. Roberts, P., J Chem Technol Biotechnol, 1994, 59, 110-111. Virus safety in bioproducts. Chisti, Y., Moo-Young, M., Trans I Chem E, 1996, 74A, 575-583. Bioprocess intensification through bioreactor engineering. Garrido, F., Banerjee, U.C., Chisti, Y., Moo-Young, M., Bioseparation, 1994, 4, 319-328. Disruption of a recombinant yeast for the release of p-galactosidase. Hochuli, E., Pure Appl Chem, 1992, 64, 169-184. Purification techniques for biological products. Beitle, R.R., Ataai, M.M., Biotechnol Progress, 1993, 9, 64-69. One-step purification of a model periplasmic protein from inclusion bodies by its fusion to an effective metal-binding peptide. [36] French, C., Ward, J.M., J Chem Technol Biotechnol, 1992, 54, 301. Production and release of recombinant periplasmic enzymes from Escherichia coli fermentations. 1371 Tamer, I.M., Moo-Young, M., Chisti, Y., Znd Eng Chem Res, 1998, 37, 1807-1814. Disruption of Alcaligenes latus for recovery of poly-o(hydroxybutyric acid): comparison of highpressure homogenization, bead milling, and chemically induced lysis. Dwyer, J.L., Biotechnology, 1984, 2, 957-964. Scaling up bioproduct separation with high performance liquid chromatography. Baker, N.V., Nature, 1972, 239, 398-399. Segregation and sedimentation of red blood cells in ultrasonic standing waves. Kilbum, D.G., Clarke, D.J., Coakley, W.T., Bardsley, D.W., Biotechnol Bioeng, 1989, 34, 559-562. Enhanced sedimentation of mammalian cells following acoustic aggregation. Whitworth, G., Grundy, M.A., Coakley, W.T., Ultrasonics, 1991, 29, 439-444. Transport and harvesting of suspended particles using modulated ultrasound. Mandralis, Z.I., Feke, D.L., AIChEJ, 1993, 39, 197-206. Fractionation of suspensions using synchronized ultrasonic and flow fields. Weiser, M.A.H., Apfel, R.E., Acustica, 1984, 56, 114-119. Interparticle forces on red cells in a standing wave field. De Loggio, T., Letki, A,, Chemical Engineering, 1994, IO1(1), 70-76. New directions in centrifuging.
30
I Strategies in Downstream Processing
Bonnerjea, J., Oh, S., Hoare, M., Dunnill, P., Biotechnology, 1986, 4, 954-958. Protein purification: the right step at the right time. Dornenburg, H., Knorr, D., Process Biochem, 1992, 27, 161-166. Release of intracellularly stored anthraquinones by enzymatic permeabilization of viable plant cells. Zhang, Z., Chisti, Y., Moo-Young, M., Bioseparation, 1995, 5, 329-337. Isolation of a recombinant intracellular 0-galactosidase by ammonium sulfate fractionation of cell homogenates. Chisti, Y., Moo-Young, M., Chem Eng Prog, 1993, 89(6), 38-45. Improve the performance of airlift reactors. Brennecke, J.F., Chem Znd (Lond.), 1995, 21, 831-834. New applications of supercritical fluids. Chisti, Y., Bioproc Eng, 1993, 9, 191-196. Animal cell culture in stirred bioreactors: observations on scale-up. Kroner, K.H., Nissinen, V., Ziegler, H., Biotechnology, 1987, 5,921-926. Improved dynamic filtration of microbial suspensions. Lee, S.S., Burt, A., Russotti, G., Buckland, B., Biotechnol Bioeng, 1995, 48, 386-400. Microfiltration of recombinant yeast cells using a rotating disk dynamic filtration system. Brors, A., Kroner, K.H., in: Harnessing Biotechnology f o r the 2lS*Century: Ladisch, M.R., Bose, A., (Eds.), Washington, DC: American Chemical Society, 1992; pp. 254-257. Electrically enhanced cross-flow filtration of biosuspensions. Rudge, S.R., Todd, P., ACS Symp Sel; 1990,427, 244-270. Applied electric fields for downstream processing. Shanley, A., Parkinson, G., Fouhy, K., Chemical Engineering, 1993, IOO(l), 28-33. Biotech in the scaleup era. [56] Kim, I., Chemical Engineering, 1997, 104(1), 28-33. Biotech’s new mandate: more, cheaper, and faster. [57] Brandt, S., Goffe, R.A., Kessler, S.B., O’Connor, J.L., Zale, S.E., Biotechnology, 1988, 6, 779-782. Membrane-based affinity technology for commercial scale purifications. [58] Zhang, Z., Moo-Young, M., Chisti, Y., Biotechnol Adv, 1996, 14,401-435. Plasmid stability in recombinant Saccharomyces cerevisiae. [59] Dunnill, P., Chem Eng Res Des, 1987, 65, 211-217. Biochemical engineering and biotechno1ogy . [60] Moo-Young, M., Chisti, Y., Biotechnology, 1988, 6, 1291-1296. Considerations for designing bioreactors for shear-sensitive culture. [61] Bykowska, K., Rijken, D.C., Collen, D., Thromb Haemost, 1981, 46, 642-644. Purification and characterization of the plasminogen activator secreted by a rat brain tumor cell line in culture. [62] Fischer, B.E., Biotechnol Adv, 1994, 12, 89-101. Renaturation of recombinant proteins produced as inclusion bodies.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
2 Protein Stability in Downstream Processing Kim Hejnaes, Finn Matthiesen and Lars Shiver
2.1 Introduction To the protein chemist two issues are of major concern in downstream processing, protein purity and protein stability. This chapter deals with the latter subject, trying to answer some of the commonly asked questions: Is there any measurable property that would predict the stability of a protein during downstream processing? Which factors influence protein stability? Can extreme conditions be accepted during purification? Can proteins be stabilized by choosing optimal conditions during downstream processing to ensure a correct tertiary structure? With the introduction of recombinant technology and thereby expression of proteins in micro-organisms, the issue of protein stability in downstream processing has gained much attention. First of all, the protein of interest is often expressed in a foreign host organism to levels far exceeding the in vivo concentrations, resulting in chemical and physical instability illustrated by the formation of inclusion bodies in Escherichia coli. Secondly, a great number of proteins have found their way to industrial applications such as in biopharmaceuticals with strict demands for purity, stability, and process reproducibility. Thirdly, when purified, many of these proteins have been shown to unfold rapidly under formation of aggregates. Several parameters including protein concentration, pH, temperature, redox potential, and co-solvents influence the stability of proteins. These factors must be taken seriously into consideration in the design of the downstream process, in which the proteins are subject to environments far from what can be found in viva This chapter focuses on the physical and chemical factors affecting the stability of small globular proteins in solution. It is our aim to describe the effect of the abovementioned parameters on protein stability, and to suggest improved strategies to ensure a correct specific biological activity of the final bulk material.
32
2 Protein Stability in Downstream Processing
2.2 Protein Stability The term protein stability often refers to the preservation of the unique three-dimensional structure of a given protein. The globular proteins in aqueous solutions are not very stable. The difference in Gibbs' free energy between the native and unfolded state is in the range of 20-60 kJ mol-' under physiological conditions [1,2], and the stability of even very different proteins does not differ greatly [2]. The native three-dimensional structure of a globular protein is maintained by the sum of interacting forces, which observed isolated contribute very little to protein stability.
2.2.1 The Native State One of the major contributions to protein stability in aqueous solutions arises from burial of non-polar amino acid residues into the hydrophobic core of the globule in order to avoid the solvation entropy decrease of the hydrophobic residue [3]. The effect contributes 3.4-7.6 kJ mol-' to the conformational stability for each -CH2- group buried into the interior of the molecule [4]. The packing of amino acid residues into the hydrophobic interior results in both formation and optimization of a large number of interactions including ion pairs, hydrogen bonds, weakly polar interactions, and short-range repulsive electron cloud overlap [5]. Potentially stabilizing features include the tight packing of amino acid residues in the protein interior, the a-helix dipole, helix caps, and weakly polar interactions between aromatic groups [6]. The van der Waals forces are weak interactions having a bond energy of 4 kJ mol-I per atom pair. The contact distance between two atoms are from 0.3-0.4 nm, and the attractive force decreases with distance. Though weak, the van der Waals force is of significance for protein stabilization given the large number of interactions in the folded state. Electrostatic forces arise from the interaction between positively charged amino or guanidinium groups and negatively charged carboxyl groups. The effect is strongly pH-dependent. At low or high pH the protein becomes unstable, as the overall charge increases, leading to increased charge repulsion. Interactions among solvent-exposed amino acids on the protein surface are usually weak and contribute little to protein stability [7,8]. In contrast, interactions with a-helix dipoles are consistently seen to contribute to protein stability [9]. Intra-molecular hydrogen bonds are essential to the structure and stability of globular proteins. A hydrogen bond is an attractive interaction between two or more neighboring atoms mediated by a hydrogen atom. The hydrogen atom is shared between two generally electronegative atoms, the hydrogen donor (-NH- and -OH) and the hydrogen acceptor (nitrogen and oxygen atoms). The >C=O.....HN< hydrogen bond is the most prevalent with >C=O- ...side-chain, >HN.....side-chain, and side-chain....-side-chain hydrogen bonds accounting for the remainder [lo]. Despite the fact that donor and acceptor atoms can form hydrogen bonds with water in the unfolded protein, a net contribution to stability of 2.9-8.0 kJ mol-' is
2.2 Protein Stability
33
observed in the folded state [ll]. Hydrogen bonds are believed to play a major role in protein stability due to the dipolar nature of amino acids, and to the fact that the helix-coil transition is largely driven by hydrogen bonding [ 121. Other energetically favorable interactions involving phenylalanine, tyrosine, and tryptophan residues are suggested to comprise an important class (aromatic interactions) of stabilization with energies between 2.5 and 5.5 kJ mol-I [13]. Thus, it is the effective co-operation of different stabilizing forces that make the total contribution to protein stability much greater than the sum of the individual interactions [ 14-16]. In its native state, a globular protein comprises a tight packed hydrophobic core of density comparable with that of amino acid crystals [17] or small organic molecules [18]. An extensive formation of hydrogen bonds is observed in the interior [19], while a large majority of charged side-chains are located at the surface in contact with water [20]. The solvent-exposed residues, whether polar or non-polar, contribute little to the stability of the native state [21]. Within the interior, charged or polar groups not paired in hydrogen bonds are rarely observed [20,22]. Studies of point-mutated proteins reveal that the most deleterious effects on protein stability usually occur when buried or rigid sites are modified, suggesting that these sites are most important for the stability of the macromolecule [23]. A globular protein should only to some extent be regarded as a rigid molecule. More likely, the native state of a protein is a dynamic system fluctuating around a limited number of preferred conformations. Within limits, these conformational changes are in dynamic equilibrium, creating an intricate balance between rigidity and flexibility [24]. Protein-water interactions play an important role in maintaining the native structure and biological function of the protein. The dense packing of the hydrophobic core of the protein is a result of the polar nature of the water molecule, and although the influence of water on van der Waals forces, hydrogen bonds, and salt bridges is very complex, it is apparent that a change of the water concentration affects protein stability. Therefore, the stability of the protein into solution depends on parameters like pH, redox potential, ionic strength, and co-solvents [25,26].
2.2.2 The Molten Globule State Recent data strongly indicate that under mild denaturing conditions, protein unfolding is a first-order phase transition between the native and the molten globule state of the protein. The protein molecule undergoes this transition in a highly co-operative manner, exhibiting the all-or-none behavior typical of a two-state denaturation mechanism within a narrow range of parameters [ll].The molten globule state of proteins was recently claimed to be a third thermodynamic state in addition to the native and unfolded state [27]. The molten globule maintains much of the architecture of the native state, although the detailed tertiary structure of the native state is lost in the transition. However, the folding pattern is similar to the native state, and a high content of secondary
34
2 Protein Stability in Downstream Processinn
structure is maintained, as is the compactness of the molecule. The core remains packed while the outer shell is more expanded with increased side-chain fluctuations. Non-polar surfaces can be exposed to the solvent, increasing the protein’s susceptibility to proteases and its ability to aggregate. Even small changes of pH, temperature and buffer composition can induce denaturation [ 2 8 ] ,and it is fair to assume that the protein occasionally will exhibit properties of a molten globule state in downstream processing.
2.2.3 The Unfolded State The unfolded state of a protein is defined as a random coil representing an ensemble of readily inter-convertible conformers with equal or closely similar energies. Within certain limits each bond can freely rotate along the covalent chain independently of the rotation of all its neighbors [29]. No changes in covalent structure can be observed between the native and the denatured state with notable exceptions such as bovine pancreatic trypsin inhibitor, where folding is driven by the formation of disulfide bonds [30]. For a significant majority of proteins, large amounts of residual structure can persist under denaturing conditions depending on the solution parameters imposed on the polypeptide chain [3 1,321. However, the co-operative interactions stabilizing folded proteins would not be expected to persist in the unfolded state [ 3 3 ] .
2.3 Chemical and Physical Instability Changes in proteins can be divided into two distinct classes involving chemical and physical instability. The first class comprises reactions involving cleavage or formation of a covalent bond resulting in a new chemical entity. The derivatives formed as a result of chemical instability may have properties almost similar to those of the product. Separation is possible, but highly specialized methods are often needed. The second class refers to changes of secondary and tertiary structures, and does not involve covalent bond modification. Physical instability may lead to polymerization of the more or less unfolded molecules in a rapid aggregation reaction [34]. Irreversible aggregation results in great losses of product, and the initiation of the reaction can be very difficult to predict.
2.3.1 Proteolytic Degradation Enzymatic proteolysis within the cell is closely controlled due to a combination of factors including compartmentalization, inhibition, and structural stabilization. When the cell dies or is disrupted by mechanical or chemical means, the endogenous enzymes are more or less released to the medium. This event dramatically changes
2.3 Chemical and Physical Instability
35
the turn-over rate of the cellular proteins. The initial part of the purification process is therefore very much affected by the presence of proteolytic enzymes resulting in partial or total loss of product. The result of the proteolytic attack may vary from complete hydrolysis, single breaks within the peptide chain, or loss of a few N- or C-terminal amino acid residues [35]. The fact that the biological activity may be maintained following partial proteolysis emphasizes the need for detailed analysis of the isolated protein. Several symptoms can be indicative of a proteolytic problem. In the early purification phase, one-dimensional Sodium dodecyl Sulfalte (SDS) electrophoresis can be of much help as poor resolution, a high background staining, and loss of molecular weight bands are often indicative of enzymatic degradation, which also can be visualized by Western blotting techniques. Absence of biological or immunological activity may be due to proteolysis, although the decreased activity could also be related to structural changes of the molecule. During intermediary purification and polishing a variety of analytical tools (mass spectroscopy, capillary electrophoresis, peptide mapping, isoelectric focusing) are available to identify even minor changes of the primary structure. Heterogeneity in partly purified preparations constitutes a major problem in the polishing steps, as only the most sophisticated chromatographic methods will be able to separate the closely related compounds. The optimization of such steps is often very time-consuming and much effort should be made to avoid proteolytic cleavage early in the process. Most enzymes have a pH at which the proteolytic activity is optimal. At a slightly alkaline pH, the highly active enzymes of the vacuoles and lysozomes will be minimally active, and relatively strong buffers in the pH interval of 7-9 are recommended for extraction to ensure that the correct pH is maintained, when the cell content is released. Low temperature decreases the enzymatic activity. In combination with short processing times this is the most effective procedure in large-scale downstream processing, where the use of enzyme inhibitors is often omitted for economical and bio-safety reasons. The addition of protein-stabilizing agents such as glycerol or dimethylsulfoxide may lower the proteolytic damage by stabilization of the tertiary structure. Low-molecular weight substances such as co-factors may also stabilize the native structure. In many small-scale applications the use of proteinase inhibitors is widely accepted. For obvious reasons most inhibitors are toxic (some even highly toxic), and great care should be taken when handling these agents. Recommended inhibitor cocktails are given in Table 2-1. Finally, it should be mentioned that careful disruption followed by subcellular fractionation is an efficient method of avoiding proteolytic cleavage. Followed by selective removal of specific proteinases by immobilized inhibitors or substrates (such as benzamidine-SepharoseR or aprotinin-SepharoseR), the proteolytic activity can be substantially reduced. This subject has been reviewed by North [35].
36
2 Protein Stability in Downstream Processing
Table 2-1. Suggestions for inhibitor cocktails. The inhibitors mentioned are potentially harmful. Care should be taken in handling, and if the protein is to be used for biological experiments. Some reagents are only slightly soluble in aqueous media, and stock solutions must be prepared in organic solvents. Tissue or microorganism
Inhibitor cocktail
Animal tissue
1 mM PMSF 1 mM EDTA 1 mM benzamidine 10 pg/ml leupeptin 10 pg/ml pepstatin 1 pg/ml apronitin
Yeast
1 mM PMSF 5 mM phenanthroline
15 pg/ml pepstatin Bacteria
1 mM PMSF 1 mM EDTA
2.3.2 N-terminal Degradation N-terminal degradation of recombinant human growth hormone (rhGH) via diketopiperazine formation to des-Phe-des-Pro-rhGH has recently been reported [36]. The mechanism occurs without enzymatic catalysis. The degradation will probably take place in other proteins having proline as the second N-terminal residue.
2.3.3 Non-Enzymatic Hydrolysis In dilute acid, where the carboxyl group of aspartic acid residue is not dissociated, the aspartic acid peptide bonds are cleaved 100 times faster than other bonds [37]. Under conditions where other aspartyl bonds are stable, the Asp-Pro peptide bond is unusually labile at low pH values [38]. At alkaline pH, the peptide bond is difficult to hydrolyse [39]. Thus, little evidence of peptide bond hydrolysis was found when treating ribonuclease A and lysozyme with 0.2 M NaOH at 40°C for up to 48 h [40].
2.3.4 Deamidation One of the major events leading to structural deterioration in globular proteins is the deamidation reaction of asparagyl residues and to some extent, glutamyl residues. As the deamidized protein may exhibit reduced biological activity, the maximal content
2.3 Chemical and Physical Instability
37
of des-amido forms in bulk materials and in biopharmaceutical preparations is constantly being debated. The deamidation may also affect the stability of the protein by altering its tertiary structure. The list of proteins undergoing in vitro deamidation is comprehensive and includes well-known proteins such as insulin [41,42], human growth hormone [43], and cytochrome c [44]. The asparagyl and glutamyl residues are involved in the deamidation reaction. In model peptides the deamidation rate is more rapid with asparagyl residues than with glutamyl residues [45,46]. The instability of the groups in proteins correlates strongly with polypeptide chain flexibility [47], and the deamidation reaction depends on the character of the neighboring side-chains. In insulin, six amino acid residues are prone to deamidation: GlnA5, GlnA15,AsnA18,AsnA2I,A d 3 , and GlnB4 of which the three Asp residues are the most labile sites. Extensive deamidation of residue AsnA21 is observed in acidic solution, a reaction catalyzed by the terminal carboxyl group. At neutral pH deamidation takes place at residue AmB3 under formation of a cyclic succinimide intermediary, resulting in a mixture of iso-Asp and Asp derivatives. Increasing formation of Asp relative to iso-Asp derivative was observed with decreasing flexibility of the insulin molecule [42,48]. The deamidation reaction is strongly sequence-specific. The Asn-Pro sequence has a half-life 100 times greater than that of the Asn-Gly [46,49]. To some extent these observations can also be used on proteins taking the structural steric factors into consideration [45,46]. Deamidation in the two Am-Gly sequences in triosephosphate isomerase was found to be tenfold slower than in Asn-Gly model peptides [SO]. In the pH interval from 5-12 the non-enzymatic deamidation of asparagyl (and glutamyl) residues proceeds via a succinimide intermediary as outlined in Figure 2-1. The peptide bond nitrogen attacks the side-chain carbonyl carbon atom, resulting in the formation of a succinimidyl derivative. This reaction is relatively slow, and is followed by a rapid hydrolysis at either the a- or P-carbonyl group to generate isoaspartyl or aspartyl residues [49]. Alternate reactions can occur, resulting in fragmentation of asparagyl residues of the peptide chain or by deamidation as a result of formation of an isoimidyl intermediate similar to the succinimidyl intermediate [49]. At acidic pH (1-2) direct hydrolysis of the side-chain amide generates aspartic acid as the sole product. These reaction patterns have been confirmed by studies of the deamidation of ACTH, which under neutral and alkaline conditions deamidates through the formation of a cyclic imide intermediate and at acidic pH hydrolyses to form AspACTH [51]. Short peptides have been used to study the deamidation reaction as a function of temperature, pH, ionic strength, solvent, and effect of solvent dielectric [49,52]. The deamidation of the asparagyl residue through succinimide intermediates increases with temperature, elevated pH, and high ionic strength. In general the buffer species will influence the rate of deamidation as will the buffer concentration [49,53]. Protein deamidation is influenced by the same parameters as above, in a similar pattern, and by the identity of nearby amino acid residues [54,55]. Proteins are, of course, much more complex in structure, and some asparagyl residues may be buried into the interior of the molecule or be part of rigid domains. In biosynthetic human growth hormone nine Asn and 13 Gln residues are prone for deamidation
38
2 Protein Stability in Downstream Processing 0 I\
CH,-C-
0 II
CH,-CI
X
0
-
L aspartyl residue
N-
0
..
II
/c\
CH2
/ NH
I
CH
'Ni( X
0
NH, for asparginyl residue
X = OH for aspartyl residue
-
\N<
0 II CH,-C-
I1
c - NH,
\=/
CH
NH
\
II
0
L-asparaginylresidue
-
L succinimide residue
0 CH, --
0'
\C-
I
--
L isoaspartyl residue
0NH
CH
+ NH,'
\/\/\ NH
I/
0
L-aspartyl residue
Fig. 2-1. Deamidation. (a) Degradation of aspargyl and aspartyl residues in peptides and proteins at neutral or alkaline pH via intermediate succinimide formation, resulting in a mixture of aspartyl and iso-aspartyl residues. A similar reaction may occur for glutamyl residues. Adapted from [49]. (b) Direct hydrolysis of aspargyl residues in peptides and proteins under acidic conditions. In contrast to the intra-molecular reaction shown in (a), the rate of this reaction is much slower at neutral or alkaline pH. Adapted from [46].
[56]. High ionic and a minor site at with a major deamidation site at strength was found to increase the deamidation of cytochrome c 144,491. Addition of organic solvent decreased the rate of deamidation by lowering the dielectric constant of the medium [49]. In conclusion, the rate of deamidation of asparagyl and glutamyl residues in proteins can be minimized by working at neutral or slightly acidic pH [55,57], in buffers of low ionic strength, and at low temperatures.
2.3 Chemical and Physical Instability
39
2.3.5 p-Elimination One of the most frequent degradation reactions of proteins in alkaline solution is caused by the abstraction of a 0-hydrogen from cystinyl, seryl, and threonyl residues under formation of a carbanion. Depending on the nature of the side-chain, the carbanion can rearrange to form an unsaturated derivative (dehydroalanine or P-methyldehydroalanine) or add a proton to give the L- and D- amino acid residue (racemization). The reaction is outlined in Figure 2-2. Several studies indicate that the rate of 0-elimination is proportional to the hydroxide ion concentration, where the formation of the carbanion by removal of the 0hydrogen is the rate-determining step [39]. Consequently the OH- concentration should be kept low to prevent the 0-elimination reaction. Further, the reaction is influenced by temperature and presence of divalent metal ions [Y-601. The dehydroalanine and 0-methyldehydroalanine formed are quite reactive with a number of nucleophilic groups of proteins [39]. Heat-induced @-eliminationof cystinyl residues has been investigated at 100 "C in the pH range of 4-8. It was concluded that the first-order rate constant of @-elimination for different proteins was remarkably similar, and the reaction, therefore, was independent of the primary structure [61]. The sulfides formed in the degradation of cystinyl residues are themselves reactive, and they strongly influence the redox potential of the solution. Threonyl and seryl residues are lost through 0-elimination at about 3-7 % of the rate of loss of cystinyl residues [39], making the protein disulphide bond the far most sensitive with respect to P-elimination. However, derivatization of the hydroxyl groups of threonyl and seryl residues markedly enhanced the rate of loss [39]. Rz
I
CHR,
I
CH \ / \c/ NH
NH
\
-
I1
R, 1
CHR,
+
--
N 'H
0
Lamino acid residue
Fig. 2-2. The p-elimination reaction. cystinyl, seryl, and threonyl residues to form an unsaturated derivative or zation, see Fig. 2-3). Adapted from
'.
C
,"H
I1
0
' \
C / \ N , Hx, NH C
+ R;
I
0
Carbanion
The reaction is initiated by abstraction of a e-hydrogen from under formation of a carbanion. The carbanion can rearrange add a proton to give the L- or o-amino acid residue (racemi[39].
2.3.6 Racemization All amino acid residues except glycine are subject to racemization at alkaline pH resulting in formation of the D-enantiomers of the residue.
40
2 Protein Stability in Downstream Processing
R2 I
R*
I
CHR,
I
CH
NH
LN.i( \c/
\
II
0
Lamino acid residue
___,
0
CHR,
H‘
I
C
NH
\
‘\N<.\c/ 11
0 Carbanion
II
HN / ,
\cH/c\ I
/
NH
CHR,
I R2 Damino acid residue
Fig. 2-3.The racemization reaction. The initial step of the reaction is abstraction of the P-hydrogen by the hydroxide ion present in alkaline solution. by uptake of a proton this will result in either the L- or o-amino acid residue. The carbanion formed may also undergo @-elimination (see Fig. 2-2). Adapted from [39].
The initial step of the racemization reaction of amino acid residues is abstraction of the 0-hydrogen by the hydroxide ion present in alkaline solutions. By uptake of a proton this will result in either the L- or D-amino acid residue. The rate of racemization will depend upon the ease of abstraction of the 0-hydrogen, the stability of the carbanion, and whether the @-eliminationreaction is favored [39]. At pH 5-12 Asn, Asp, Gln, and Glu amino acid residues undergo a non-enzymatic modification into a succinimidyl intermediate. The five-membered ring rapidly hydrolyses to Asp or Glu or to the corresponding iso-forms in a variety of reactions resulting in racemization, generating a mixture of D- and L-derivatives [49]. The reaction is outlined in Fig. 2-3. Several reports have dealt with the racemization of proteins in alkaline solutions [62,63]. In one study the average extent of racemization of the amino acid residues of ribonuclease A, lysozyme, soybean protein, and casein was about 6 % in 0.2 M NaOH after 4 h at 40°C [63]. Thus, utmost care must be taken at pH above 10 (low temperature, selected buffers, short exposure time), as racemization is inevitably associated with conformational changes and thereby loss of function.
2.3.7 Conversion of Arginine to Ornithine The guanidinium group of arginine is hydrolyzed by OH- to give ornithine and possibly some citrulline, depending on the nature of the protein. The rate of arginine degradation is generally smaller than for cystine, lysine, threonine, and serine under the same conditions. The reaction is outlined in Fig. 2-4. The extent of loss of the amino acids affected by alkali treatment was proportional to the hydroxide concentration [39].
2.3 Chemical and Physical Instability
41
C-NH
Arginine residue
Ornithine residue
urea
Fig. 2-4. Conversion of arginyl residues to ornithyl residues. The hydroxide ion-catalyzed hydrolysis of arginyl to ornithyl residues generates omithyl and urea. Some citrullinyl residues may be formed as well. Adapted from [39].
2.3.8 Oxidation Oxidation has been observed in many peptides and proteins during their isolation, often resulting in loss of biological and immunological activity [64-661. At neutral or slightly alkaline pH conditions, the histidyl, methionyl, cysteinyl, tryptophanyl, and tyrosinyl residues are potential oxidation sites, while under acidic conditions the primary reaction is the oxidation of methionine to methionine sulfoxide [67]. Under mild oxidizing conditions the methionyl residue is oxidized to the corresponding sulfoxide as outlined in Fig. 2-5. Oxidation to methionine sulfone requires strongly oxidizing conditions rarely met in downstream processing. Methionine is very susceptible to auto-oxidation and photo-oxidation, the latter being an example of the influence of light rather than oxygen [53]. Cysteine is also easily oxidized to cystine at alkaline pH (where the thiol group is de-protoneated) in the presence of a mercapto-reagent. In another set of reactions, oxidation of cysteine to cystine is catalyzed by divalent metal ions, especially Cu2+. The reaction can be inhibited by addition of EDTA. In the absence of a thiol reagent or a nearby thiol, the cysteine residue may instead oxidize to sulfenic acid. A comprehensive summary of protein oxidation is given in [68].
42
2 Protein Stability in Downstream Processing
Oxidation of cysteine
RSH
mild
RS-OH
___*
cysteine
-
sulfenic acid
-
0 I1
R-SOH
0 I1
RS-OH
0 sulfinic acid
sulfonic acid
Oxidation of rnethionine
RS-CH,
methionine
mild
0 I1
RS-CH,
methionine sulfoxide
strong
0 I1
RS-CH, I1
0 methionine sulfone
Fig. 2-5. Oxidation of cysteine and methionine residues.
2.3.9 Cysteinyl and Cystinyl Residues The ability of the cysteinyl residue to form intra- and inter-molecular disulfide bonds makes this amino acid residue unique in terms of protein stability. A large number of extracellular proteins are stabilized via disulfide bonds, and in many cases the covalent cross-link is essential for maintaining their tertiary structure and their biological activity. Ribonuclease, for example, loses almost all activity when the four disulfide bonds are reduced [69]. Insulin and chymotrypsin are completely inactivated by the reduction of a single disulfide bond [70,71]. In addition to the static function described, reactive disulfide bonds sometimes perform a dynamic function in proteins [72]. A disulfide bond remains stable indefinitely in an isolated protein, its stability being entirely dependent on the environment. Major factors are redox potential, pH, and temperature, which must be carefully controlled to prevent bond cleavage. The disulfide bond is degraded in a variety of mechanisms of which some of the most important will be summarized below, (see Fig. 2-6). In neutral or alkaline media a nucleophile attack on a sulfur atom of the disulfide initiates the cleavage. The reactivity of some nucleophile reagents is HS-> RS-> CN>S032-> OH-. The thiol-disulfide reaction is perhaps of the most practical importance. This reaction consists of two steps of nucleophile substitution with the formation of a mixed disulfide at the intermediate stage. Both rate and equilibrium constants for the thiol-disulfide exchange exhibits common variation [73], and the reaction rate is influenced by the pK, of the proton sulfhydryl group (typical value is 8.5), the leaving group, steric hindrance, charge, and entropy [69]. The interaction of -SH groups with aliphatic disulfides proceeds mainly in alkaline media (usually at pH 9 . 3 , although shuffling reactions at pH 7 to 8 have been reported [74].
2.3 Chemical and Phvsical Instabilitv R,S'
+
-
R2S-SR2
4--~
R,S SR, + R2S'
-
+---F
R,S SR,
a) R,S' + R,S SR,
43
-
+ R2S'
0 - HN - CH - C
I1
-
0
CHZ
F) HN-CH-C
I CH2
S
1
SO'
CH2
I
-HN-CH--C-
0
b)
c)
cystinyl residue
cystinyl residue
cysteinyl residue
sulfenic acid
dehydroalanyl residue
Fig. 2-6. Reactions involving cysteinyl and cystinyl residues. (a) The two-step nucleophilic substitution reaction between the peptide or protein (RS-) and a thiol reagent (RzS-SRz) in alkaline media. (b) The hydrolysis of the cystinyl residue in alkaline media to the cysteinyl residue and a sulfenic acid residue. Adapted from [39]. (c) The proposed reaction for the 0-elimination of protein-bound cystine in alkaline solution. Adapted from [39].
In strongly acidic media the exchange reaction is proposed to take place via a sulfenium cation, which is formed by an attack of the disulfide bond by an electrophil (H+, Hg2+, Ag+) displacement on a sulfur atom of the disulfide [75]. OH- is a poor nucleophile towards the disulfide bond, but solvent-accessible bonds were shown to dissociate in 0.2 M NaOH [76]. Three mechanisms were investigated: hydrolysis, a-elimination, and p-elimination. The result of the study points to elimination of 0-hydrogen and formation of a persulfide intermediate as being the first step of the degradation by dilute alkali. Only rarely may the a-elimination
44
2 Protein Stability in Downstream Processing
reaction occur in polypeptides [61,76]. However, in strongly alkaline media the acarbon proton may also be removed [39,77]. One should be aware that catalytic quantities of HS- may arise from the p-elimination reaction thus initiating disulfide bond rearrangement. Presence of divalent metal ions (typically Cu2+) in combination with OH- and 0 2 may result in oxidation of cysteine residues by an ill-defined reaction mechanism [78]. Therefore, stock solutions are often deareated and made 1-2 mM with metal chelators such as EDTA. The lability of the disulfide bond in proteins may result in the formation of structurally altered forms of the protein with reduced biological activity and/or immunogenic properties. In proteins comprising both free -SH groups and disulfide bonds, rearrangement can happen without the presence of low-molecular weight thiol reagent. However, the number of proteins containing both -SH groups and disulfide bonds are relatively small. Among these proteins are serum albumin, papain, ficin, and P-lactoglobulin [72]. A novel protein derivative has been found during downstream development of biosynthetic human growth hormone. In this derivative a C y ~ ' ~ ~ - C trisulfide y s ' ~ ~ bridge was identified using electrospray mass spectroscopy [79].
2.3.10 Denaturation The term denaturation is used to denote the process in which the tertiary structure of the molecule is changed from the one typical of the native structure to a more disordered arrangement but without alteration of the amino acid sequence. In the process, the hydrogen bonds and intra-molecular interactions resulting in the co-operative stabilization of the native state are disrupted. The precise orientations are lost and regions of 'the inner core' are exposed to solvent water and co-solvent molecules [3,80,81]. For small globular proteins denaturation is an almost all-or-none process approximated rather well by the two-state transition [28]. The co-operativity of the denaturation process in many proteins results in an abrupt transition from the native to the unfolded state within a narrow range of temperature, pH, redox potential, or denaturant concentration. The different phases cannot be transformed gradually into each other, and there is no critical point at which they are indistinguishable [15]. Thermodynamically, the denaturation process can be observed by an increase of molar heat capacity, and a rapid enthalpy increase with increasing temperature [2,82]. In many cases denaturation is a reversible process in which the native structure can be re-established by careful adjustment of solvent composition or of pH, temperature, or redox potential. However, proteins may undergo irreversible denaturation in which the native structure cannot easily be obtained. An example is the group of proteins in which disulfide bonds have been destroyed, and where in vitro folding under very restricted conditions must be applied to regain activity. Another example is the hydrophobic aggregation observed among intermediates or unfolded molecules 134,831. A third example is the formation of stable intermediates as observed during renaturation of AE-IGF-1 [84].
2.3 Chemical and Physical Instability
45
Denaturation is a function of pH, temperature, redox potential, solvent composition, ionic strength, and other parameters of downstream processing. The marginal stability of the native state, and the loss of tertiary structure, make denaturation very difficult to predict. Within milli-seconds the entire protein solution can be transformed into an insoluble gel or precipitated aggregates, resulting in major loss of product.
2.3.11 Aggregation One of the serious consequences of exposing the protein to denaturing conditions is the exposure of hydrophobic residues to the aqueous solvent. This leads to disorganization of the water molecules, thus increasing the entropy of the system. In order to avoid this change, the water molecules try to maintain organization by aggregation, thereby decreasing the solubility of the protein. The aggregation reaction can be very fast, and will, in severe cases, lead to the formation of insoluble polymers. Folding of denatured or partially folded intermediates is a first-order reaction resulting in a constant folding rate at increasing concentration of the denatured or partially folded forms. Aggregation is assumed to be controlled by the initial dimerization step in a second-order reaction dominating over the first-order folding reaction. Consequently, an increase of concentration of denatured or partially denatured polypeptide will favor the aggregation reaction and thus form biologically inactive precipitates [85]. It has been stated that aggregates derive from partially folded intermediates rather than from unfolded or native protein [24,86] exemplified in the formation of insulin fibrils from partially unfolded insulin molecules [34] and the aggregation/association behavior of the protonated barnase intermediary [87] (see Fig. 2-7). 1irr
N-MG-U
Fig. 2-7. Unfolding and aggregation. The transition from the native state (N) to the molten globule state (MG) and from MG to the unfolded state (U) are reactions of first order (i.e., the reaction rate is independent of the protein concentration). Some proteins do not readily fold back to the native state, either because of stabilization of the intermediary compound I;, or due to the slow and complicated re-establishment of disulfide bonds. The aggregation reaction, in contrast, is a reaction of second order (i.e., the reaction rate depends on the protein concentration), resulting in aggregation of the molten globule to Iagg.
46
2 Protein Stability in Downstream Processing
Recent data strongly indicate that aggregation occurs by specific interaction of certain conformations of intermediates rather than by non-specific co-aggregation [88]. However, insulin molecules with nearly complete native-like structure, aggregate in a linear fashion [34]. Studies on the aggregation of apomyoglobulin in aqueous urea solutions showed that aggregation also may involve the association of unfolded molecules [89]. Proteins comprising free -SH groups may form inter-molecular disulfide bonds, leading to aggregation of the protein. Purification under reducing conditions erases this problem as demonstrated in the purification of recombinant glutamic acid decarboxylase [90]. Inactivation via aggregation has been reported for several proteins [89,91-98]. Expression of proteins in Escherichia coli often results in the formation of insoluble aggregates called inclusion bodies, probably comprising fully or partially denatured protein [99].
2.3.12 Precipitation One of the most widely used procedures in protein purification is precipitation by altering the solvent properties and thereby lowering the solubility of the protein. Being part of a purification scheme, mild precipitation procedures have been used to limit irreversible aggregation. In iso-precipitation, molecules are precipitated near their isoelectric point, where the electrostatic repulsion is minimal due to the zero net charge of the protein. Inclusion of a solute such as ethanol, which reduces the protein solubility further, will often improve the precipitation process. At low salt concentrations favorable interactions between salt ions and charged groups on the protein surface will result in an increase of the solubility of the protein (saltingin). The effect is identical for all proteins. High concentration of salts will in most cases lead to precipitation of the protein (salting-out). One of the most commonly used procedures is addition of ammonium sulfate to a 2-3 M concentration. The process is largely dependent on the hydrophobicity of the protein, and the optimal salts are those favoring dehydration of the non-polar regions without binding to the protein [loo]. The salting-out effectiveness of cations and anions is shown in Table 2-2 Table 2-2. The Hoffmeister series. The first ions in the series are known for their stabilizing effect on proteins. They markedly increase the surface tension of water and are preferentially excluded from the protein surface. The solubility of non-polar molecules is decreased in solvents like ammonium sulfate (salting-out). The last ions of each series may bind to the protein. They have little effect on the surface tension of water, and increase the solubility on non-polar molecules (salting-in). Cations NH4+ > K+ z Na+ > Li+ > Mg2+ > Ca2+> Gdn+ Anions S042- > HP042- > CH3COO- > C1- > N03- > SCN-
2.4 Essential Parameters
47
reflecting the order discovered by Hoffmeister [ l o l l . The effects of the individual ions are additive. Thus GdnCl is a strong denaturant, while GdnzS04 normally stabilizes proteins. Organic solvents such as acetone and ethanol have been widely used to precipitate hydrophilic proteins, which are less soluble in non-polar solvents. The water molecules are partially immobilized by hydration of the solvent reducing the water activity. The water molecules around hydrophobic areas will be displaced by the organic solvent reducing the hydrophobic attraction. The principal forces leading to precipitation are, therefore, likely to be electrostatic forces and dipolar van der Waals forces. The technique has its limitations, however, as many proteins tend to denature in the presence of organic solvents. Therefore, it is important to keep the temperature low in order to decrease the conformational flexibility and to prevent the organic solvent from penetrating the internal structure of the protein. A similar effect can be obtained with water-soluble organic polymers, provided they are not too viscous. Polyethylene glycol of a molecular weight between 6000 and 20000 Da is commonly being used.
2.4 Essential Parameters The term downstream processing refers to isolation of a given protein from a complex, heterogeneous mixture comprising other proteins, lipids, cell debris, DNA, RNA, Malliard compounds, etc., which results from the fermentation and/or extraction procedure. The ultimate goal is to isolate the protein in its purest form while retaining the specific biological activity of the molecule. Suppliers of protein biopharmaceuticals have met this challenge for years, facing increasing demands for higher purity and specific analytical methods for characterization. It is, therefore, of interest to focus on the various parameters characterizing each unit operation and to determine their influence on protein stability. These parameters include pH, temperature, redox potential, presence of co-solvents, protein concentration, and pressure.
2.4.1 The Effect of pH Conventional wisdom states that protein purification is best carried out near the physiological pH, at which the protein is believed to obtain maximal stability. The view can be justified for two reasons. The isoelectric point of many proteins is in the range of pH 5-8 [102], not far from the physiological pH. Further, many proteins are stable around pH 6 regardless of their isoelectric point, reflecting the ionization of histidine in the unfolded state [ 1031. However, the optimal pH can be very different from neutral. Pepsin is stable at pH 1-2, but denatures rapidly at pH 7 or more [loo]. The structural stability of globular proteins depends very much on protonation and deprotonation of potentially titrable groups like carboxyl (pK, 3 .O-4.7), imidazo-
48
2 Protein Stability in Downstream Processing
lium (pK, 8.0-8.5), sulfhydryl (pK, 8.0-8.5), amino (pK, 7.6-10.6), and phenolic hydroxyl (pK, 9.4-10.4) [39]. The effect of pH on secondary structure is reflected in the structural changes of two synthetic polypeptides, poly-L-glutamic acid and poly- L-lysine. At a low pH poly- L-glutamic acid spontaneously forms a helical structure, whereas poly- L-lysine is a random coil. As pH increases, the situation reverses, and at pH 10, poly- L-lysine spontaneously folds into a helix, whereas poly- L-glutamic acid forms a random coil [104]. The same pH dependency is observed in proteins. As pH is increased or decreased from PI, titrable groups are protonated or deprotonated and electrostatic interactions are broken, resulting in structural changes of the protein due to electrostatic repulsion [105]. By a change in pH, interior hydrophilic residues (Asp, Glu, Lys, Arg, His) may be positively or negatively charged, making the residue more solvent-accessible, again resulting in a structural change of the molecule. Although a non-buffered protein solution should automatically adopt a pH very close to PI, the buffer capacity is low. Therefore, even minor changes of the environment may move the system towards pH-extremes. In the pH range 4.5-8.0, the melting temperature, T , is independent of the pH provided PI of the protein is in the same range [106]. A variety of chemical protein modifications takes place in alkaline solutions, the reaction rate often being directly proportional to the hydroxide ion concentration. In alkaline solutions proteins may undergo degradation by deamidation, f3-elimination, hydrolysis, racemization, or breakage of disulfide bonds. Deamidation, hydrolysis, or breakage of disulfide bonds may also take place in acidic environment. Thus, the extent of chemical protein modification in downstream processing as a function of pH can be difficult to predict. Even in mildly acidic or alkaline solutions, deamidation of proteins is often observed, as is degradation of sulfur-rich proteins. Physical and chemical protein modification is, therefore, not only a phenomenon observed at extremes of pH; it should rather be regarded as a potential risk and possibility anywhere on the pH scale. Protein destabilization as a function of pH is illustrated by the following examples. At acidic pH insulin deamidates in position A21 [42], and between pH 1-2 quickly loses six amide groups by hydrolysis [107]. Around neutral pH, insulin deamidation is much slower and takes place exclusively at residue AmB3 [42]. Interferon-y was reported unstable upon acidic treatment at pH 4 [108,109]. At pH 3.0 L-asparaginase loses 98 % of its biological activity within 50 minutes; the antibody-protected enzyme retained 40 % activity under the same conditions [110]. Degradation of lysozyme and a-lactalbumin at the cystinyl residues was observed by treatment with 0.1 M NaOH for 24 h at 50 "C [60]. A 30 % racemization of the seryl residues was observed after treating lysozyme with 0.5 M NaOH for 2.5 h at 22 "C [lll].The average content of racemization of the amino acid residues of lysozyme, soyabean protein, ribonuclease A, and casein was approximately 6 % in 0.2 M NaOH after 4 h at 40 "C [63]. However, little evidence of hydrolysis of peptide bonds in ribonuclease A and lysozyme was found when the proteins were treated with 0.2 M NaOH at 40 "C for up to 48 h [40]. Cytochrome c unfolds in alkaline solution [112]. Degradation of chymotrypsin, trypsinogen, ribonuclease S, and lysozyme in 0.2 M NaOH was reported [76]. In contrast, the activity of low-molecular weight urokinase was nearly independent of pH in the range 2-11 [113].
2.4 Essential Parameters
49
2.4.2 The Effect of Temperature The difference in Gibbs’ free energy between the native and unfolded state (AG,) of a protein results from a balance between large and opposing entropic and enthalpic effects, AS, and AH,. The AH, and AS, are highly temperature-dependent [82,114] being linearly increasing functions of temperature [115], with a slope equal to the unfolding heat capacity of the protein 121. The heat capacity (Cp) of both the folded and unfolded states of a protein, is substantial and ACp between the folded and unfolded state is in the order of 8 kJ/(degree Kelvin x mol) [116]. The significant heat capacity difference results in a non-linear temperature dependence of AG as illustrated in Fig. 2- 8. Expressions taking the temperature dependence of the thermodynamic parameters of the unfolding into account have been derived [106,117]. AG, is a function of temperature with a maximum in the temperature range of 10 to 30°C for all globular proteins investigated to date. At this temperature the denaturation entropy change is zero and AG(Ts) = AH(T,). Thus, the point of maximum thermodynamic stability of the native structure is entirely due to enthalpic contribution. AG, decreases to zero both at higher and lower temperatures as illustrated in Fig. 2 - 8 [ 1151. This results in heat- and cold-denaturation of proteins, respectively. On the assumption of the two-state model for small globular proteins, the native state will collapse in an ‘all-or-none’ transition upon denaturation. AG, = -RTlnK AG, = -RTln[N]/[UI AGu = 0 when [NJ= [Ul The temperature at which AG, = 0 is called ‘the melting temperature’, T,, where T, = AHm/ASm,This shows that the heat denaturation occurs when AS, (chain entropy difference) exceeds the enthalpy from inter-atomic interactions of the native state. Denaturation at low temperature can be explained by the decrease in the strength of hydrophobic interactions as the temperature is lowered [ 1181. The hydro-
AG
T
Fig. 2-8. The free energy, AG, as a function of temperature. T, is the transition temperature at which the Gibbs’ free energy difference between the native and unfolded state is equal to zero. T,,, is the temperature at which the stability of the native state is at maximum and where AS(Tmax)= 0. Heat denaturation proceeds with an increase of enthalpy and entropy. Cold denaturation proceeds with a decrease in enthalpy and entropy. Adapted from [106].
50
2 Protein Stability in Downstream Processing
phobic interactions are strengthened with increasing temperature, while hydrogen bonds are weakened, resulting in loss of co-operativity [3]. The stability of the native state is therefore limited to the temperature interval where hydrophobic interactions are sufficiently strong to counteract the weaker dissipative forces [118]. The stability of a native protein in aqueous solution is determined not only by the interactions discussed so far, but also by the interactions between residues and the surrounding water molecules. The factors contributing to destabilization of the native state are the increase of the unfolding chain entropy and the decrease of the unfolding hydration enthalpy. The critical dependence of T, on solution conditions is illustrated by the pH dependency of T , for ribonuclease. T, increases from 46 to 65"C, as the pH of the solution varies from 3 to 7 [ 1191. Thermal denaturation is also a function of presence and type of neutral salts. The T, of ribonuclease decreases from 64 "C to 40 "C as the guanidinium salt is varied from sulfate to thiocyanate [120]. Measurement of T,(near PI) and AC, leads to evaluation of AH, TAS and AG for the unfolding process of a given protein. T , can be determined by circular dichroism, fluorescence spectroscopy, or UV-difference spectra, which will provide information about the structural content at a given temperature. From these data the transition curve for thermal unfolding in a given solvent can be constructed [121]. For small globular proteins denaturing according to the two-state model, the analysis is straightforward; for large proteins the two-state model may be tested by measuring the transition by at least two different methods [121]. A number of destructive covalent reactions of thermolabile amino acid residues have been observed upon heating. They include deamidation of asparagyl and glutamy1 residues in peptides at 100°C at pH 7.4 [45], cysteine oxidation at pH 8 in aamylase [122], 0-elimination of cystine residues at 100°C at different pH [61], hydrolysis of peptide bonds at aspartyl residues in proteins [ 1231, and thiol-catalyzed disulfide interchange at 100 "C [61]. The temperature of maximal stability is in the range of 10-30°C for all proteins investigated, a temperature range well suited for downstream processing. Therefore, thermal inactivation during purification should not be expected provided solvent conditions are kept optimal, i.e. close to PI, low content of buffer additives, low protein concentration, and correct redox potential. However, these optimal conditions are rarely met in purification of proteins, and solvent conditions can suddenly and unexpectedly change, resulting in a shift of T, into the temperature interval of the downstream process. Storage of protein solutions at low temperature may constitute a problem, as well, either because of the freezing-thawing procedure, or because of lowered freezing point due to additional co-solvents, at the risk of cold denaturation.
2.4.3 The Effect of Redox Potential The formation and cleavage of disulfide bonds in proteins is a function of the redox potential of the solution. A decrease in the redox potential towards more reducing
51
2.4 Essential Parameters
conditions will ultimately result in cleavage of the disulfide bond to free cysteinyl residues. Alternatively, an increase of the redox potential towards oxidizing conditions will result in disulfide bond formation, as seen in the in v i m renaturation of recombinantly expressed proteins. Most buffers in downstream processing are designed for maintaining pH in a narrow, well-controlled interval. However, little or no attention has been paid to the very low redox buffer capacity of these same buffers. Consequently, even minor changes of the environment may lead to significant changes of the redox potential of the solution. For a complete reaction, oxidation as well as reduction must occur: Redl
+ 0x2 = 0x1 + Red2
From the free energy of the reaction AG = AGO
+ RT
x lnK,,
and the Gibbs' free energy
E2 -
AG = -nF(E2 - E l ) El = EzmO- Elrno- RT/nF x InK,,,
where E is the redox potential, Emo is the redox potential at [Ox] = [Red] at which [Ox] and [Red] are maintained at unit activities at pH = 0, R is the gas constant, T is the absolute temperature in degree Kelvin, n is the number of electrons transferred, and F is the Faraday constant, and K,, is the equilibrium constant. Since there is no absolute value for redox potentials, the standard hydrogen half cell is given the value 0 V at any temperature. Thus, for the reaction
Eh
Ox2 + ne- = Red2 = Emo - RT/nF X ln([Red2l/[Ox~l)
which is the well-known Nernst equation, ignoring transfer of protons. redox potential relative to the standard hydrogen electrode. At 24°C Eh
= Po- 0.059/n
X
Eh
is the
log([Red2]/[0x2]).
Obviously the proton in redox reactions involving amino acids, peptides, and proteins cannot be ignored: 0x2 + e- + H+ = Red2H (n = 1) Eh = Em' - 0.059 X log([Red2H]/(Ox~ X [H+])) ( n = l ) Eh = Em' - 0.059 X log([Red2]/[0x2]) - 0.059 X pH ( n = l )
52
2 Protein Stability in Downstream Processing
At [Ox]/[Red] = 1, the redox potential (Eh) will change 0.059 V for every pH unit increase. Thus, raising pH from 5 to 12 decreases Eh with 0.42 V creating a much more reducing environment. Let us consider the two reactions Red, = 0x1 + ne0x2 + ne-= Red2 with the redox potentials El and Ez.If E2 > El, system 2 will oxidize system 1. In most cases the term RTInF x ln([Red2]/[Ox2]) contributes little to E, and the standard redox potentials can be used to compare the relative reducing or oxidizing power (an exception is Elrno The standard redox potential of selected half-cell reactions is given in Table 2-3. From the table it is seen that DTT, cysteine, and 2-mercaptoethanol are excellent reducing agents for proteins and that H202 or potassium ferricyanide are well-suited oxidants. In practice, a Pt-standard calomel electrode system is used for measuring the redox potential. The potential of the calomel electrode relative to the standard hydrogen electrode is 0.244 Vat 25 "C [127]. The system is standardized against a quinhydrone redox buffer solution using E, = 0.219 V at pH 4.00 and E, = 0.042 V at pH 7.00 at 25 "C, where E, is the redox potential measured with the calomel electrode system. In order to investigate the correlation between the redox potential of the solution and the stability of the disulfide bonds in proteins, the stability of human recombinant insulin was measured in solutions of increasing DTT concentration. The Aand B-chains of human insulin are held together by the two disulfide bonds only, as outlined in Fig. 2-9. Upon reduction the two chains will separate when the cystinyl residues are reduced to cysteinyl residues. The degradation can be monitored by RP-HPLC analysis of samples taken at different redox potentials. The result is shown in Fig. 2-10.
-
Table 2-3.Standard redox potentials. The standard redox potentials at [Ox] = [Red] at pH 7 (I?"') of selected half-cell reactions is shown for reducing agents (DTT, cysteine, glutathione) and for oxidizing agents ([Fe(CN)6]3-, 0 2 (alkaline solution), H2Oz). A mixture of reduced and oxidized glutathione is often used in refolding experiments to ensure correct redox potential and mercapto reagent concentration. Agent
I?"
Reagents
DTT
-0.33 V
Cleland [124]
Cysteine
-0.21 V
Cleland [124]
Glutathione
-0.23 V
Scott et a]. [125]
[Fe(CN)d3-
0.36 V
CRC Handbook of Chemistry and Physics [126]
Oxygen (alkaline solution)
0.40 V
CRC Handbook of Chemistly and Physics 11261
Hydrogen peroxide
1.78 V
CRC Handbook of Chemistry and Physics [126]
2.4 Essential Parameters
S
S
A chain I
I
5
1
10
I
15
53
21
S
S
/ B chain 5
S
S
I
I 10
15
20
25
30
Fig. 2-9. The primary structure of human insulin.
Human insulin remains stable at E h > 0.1 V. At lower redox potential the disulfide bonds are reduced, resulting in loss of intact molecules. At approximately 0 V, about 1 % of the insulin molecules remains intact (corresponding to 4 mM DTT in the said solution). The low buffer capacity is illustrated by the minute amounts of DTT needed to reduce the two disulfide bonds. A protein comprising cysteinyl residues may be kinetically or thermodynamically trapped in various ways. In the presence of thiols or other nucleophiles (other cysteinyl residues), rearranged disulfide bonds may result in irreversible formation of scrambled forms [84]. Under reducing conditions cysteinyl residues in the reduced form could be trapped, so that an unfavorable structural rearrangement must precede chemical oxidation. The presence of thiol reagents could result in formation of mixed
54
2 Protein Stability in Downstream Processing
disulfides with available cysteine residues preventing re-establishment of the disulfide bond(s). Further, the cleavage of one or several disulfide bonds may alter the structure of the protein resulting in other covalent modifications such as proteolysis, deamidation, or aggregation. Mercapto reagents have been used carefully in the renaturation of IGF-1, rhodanese, lysozyme, and in stabilization of glutamic acid decarboxylase (GAD65) during downstream processing as outlined in Table 2-4. The intra- and inter-cellular reactions between cysteinyl residues depend on the solution pH. The pK, of the typical cysteinyl sulfhydryl group is approximately pH 8.6. This value varies from protein to protein and among different positioned cysteinyl residues. The presence of a thiolate anion is essential for the thioVdisulfide reaction to occur, and the reaction is strongly inhibited at pH below 7. Maintaining a mildly acidic environment thus stabilizes proteins with cysteinyl residues. It appears that the redox potential of the solution is an important parameter in terms of stability for proteins containing cysteinyl residues. Aqueous buffers will normally exhibit a redox potential between 0.2 V and 0.5 V ensuring the stability of the disulfide bond. However, changes in environment (exemplified by shift of pH) may result in dramatic shifts of the redox potential towards destabilizing conditions. It is recommended that the redox potential be monitored using a reliable and Table 2-4. Examples of co-solvents used for stabilization of proteins. The buffers indicated for IGF-1, rhodanese, and lysozyme were used in refolding of the molecules. GAD65 were purified under reducing conditions to prevent inter-molecular disulfide bond formation and aggregation. Protein
Buffer
Purpose
Reference
IGF- 1
50 mM Tris, 2 mM cysteine, 2 mM EDTA, 25 % (v/v) ethanol pH 9.0
Ren aturation
Hejnaes et al. ~341
Rhodanese
10 mM Na-phosphate, 200 mM 2-mercaptoethanol, 0.3 M GdnC1, 0.5 % (w/v) lauryl maltoside pH 7.4
Renaturation
Tandon and Horowitz t981
Lysozyme
0.1 M Tris-HC1 1 mM EDTA 0.695 mM oxidized glutathione 4 M sarcosine pH 8.0
Renaturation
Maeda et al. [I281
GAD65
50 mM glycine, 10 mM DTT, 1 % n-octylglycoside, 1 mM glutamate, 1 mM PLP pH 9.5
Stabilization
Moody et al. 1901
2.4 Essential Parameters
55
stable system such as the Pt-calomel electrode system, and that redox buffers such as mixtures of thiol reagents are introduced. One should, however, be aware of the limitations of such buffers as illustrated by the aggregation of serum albumin in the presence of low levels of thiol reagent [129].
2.4.4 The Effect of Co-solvents A great number of co-solvents are used in downstream processing to ensure optimal purification conditions. Any of these additives will influence the stability of the native conformation. The choice of co-solvent at a given set of parameters (pH, temperature, ionic strength, protein concentration, and redox potential) is, therefore, of great importance for the net stability of the protein. The mechanism of stabilization is now well understood following a series of careful interaction measurements between proteins and the complex solvent environment, carried out by Serge Timasheff and co-workers. Intensive studies for more than three decades have shown that if a protein is dissolved into a mixture of two solvents, generally one of them will form more favorable interactions with the protein than the other. The preferred solvent is most often water and hence the expression ‘preferential hydration’. Consequently, the co-solvent will be excluded from the protein/solvent interaction layer. The effect of the co-solvent will be stabilization of the native state of the protein since the system tends to move towards minimizing the area of the water-protein interface. Destabilizing agents are able to bind to exposed groups on the protein surface, either by charge-charge or hydrophobic interactions. They will preferentially interact with the unfolded protein and favor its unfolding [ 1301. Small molecules and ions also contribute to protein stabilization by raising the water surface tension. The mechanism, based on the pertubation of the water surface tension, is a function of the physical chemistry of the water-small molecule interaction [130]. The different mechanisms of exclusion and binding have allowed Timasheff and co-workers [130] to suggest the existence of three classes of co-solvents. The first class comprises sugars, some amino acids, some salts, and certain polyols. Co-solvents belonging to this class (sucrose, glucose, mannose, glycine, alanine, glutamine, glutamic acid, and NazS04) increase the surface tension with no or weak binding to the protein. Glycerol, mannitol, and sorbitol also belong to this class; they have, however, affinity for polar regions. In general the stabilizing effect is concentrationdependent. The effect becomes apparent only at relatively high co-solvent concentration [131]. The second group comprises weakly interacting salts (MgS04, (Gdn)~S04,NaC1, Arg-HC1, and Lys-HC1). They increase the surface tension and bind to charged groups or to peptide bonds. The stabilizing effect will depend on protein charge and salt concentration. Valine binding to hydrophobic regions belongs to this group. Polyethylene glycol (PEG) and 2-methyl-2,4 -pentanediol (MPD) belong to the third class, where stabilization is caused by steric exclusion and repulsion from charged groups. Solvents able to bind to the surface of the protein make protein sta-
56
2 Protein Stability in Downstream Processing
bilization a balance between co-solvent exclusion and binding. Although PEG is strongly excluded from the protein surface domain, its ability to bind to non-polar amino acid residues [ 1321 will lower the unfolding transition temperature [ 1331 and thus destabilize the native conformation. MPD, another strong binder to hydrophobic regions is strongly repelled from charged groups [133] and is, therefore, excluded from the protein domain, stabilizing the native protein. However, a partially unfolded protein will expose non-polar groups to the surface, where MPD will bind and the net result will be destabilization. Co-solvents able to bind to exposed residues of proteins have been widely used in downstream processing to prevent aggregation of intermediary compounds. Aggregation has been suppressed at well-defined concentrations of denaturants [ 134,1351 or other additives [136]. Use of denaturants in downstream processing is a balance between the denaturing effect of the co-solvent and its power to prevent aggregation. Examples of additives used to reduce aggregation are 0.2 M arginine [137], PEG [97], and lauryl maltoside [138]. The effect of neutral salts (defined as strong electrolytes which are significantly soluble in water without bringing about a major change of solution pH [lOS]) depends on the individual contribution from the cation and the anion as outlined in the Hoffmeister series (Table 2-2). Direct interaction between the salt and protein can be electrostatic in nature, and this effect may be dominating at low salt concentrations. Ions also react with dipolar groups such as amino, carboxyl, and hydroxyl groups. Other salts have particular non-polar or hydrophobic character and exhibit a solubilizing effect on non-polar residues or surfaces, leading to destabilization or denaturation. Once the electrostatic effects have been saturated, the transition temperature is generally found to be a linear function of the salt concentration. The anion and cation affects T, in a roughly additive fashion [139]. At higher salt concentrations, protein stability depends on the ability of the salts to interfere with the hydration sphere of the protein. Salts excluded from the hydration sphere generally tend to stabilize the protein by preferential hydration and to decrease the solubility (salting-out). In contrast, salts that tend to disorganize the water structure destabilize the protein and increase the solubility (salting-in). In general, these effects follow the Hoffmeister series [ 1391 making ions like NH4+ and S042- excellent stabilizers, while SCN2- and c104- are de-stabilizers. The same destabilizing ions are also effective in dissociating non-covalent aggregates of globular proteins [ 1051. The impact of neutral salts on protein stability requires a high concentration (1-8 M) of the added salt, indicating that in general protein-salt interactions can be neither strong nor specific [ 1401. Detergents are amphiphilic surface active agents (surfactants) which are soluble in aqueous solvents on account of their hydrophilic matrices. The widespread use of detergents in protein chemistry arises from their extraordinary solubilizing effect on membrane proteins and inclusion bodies. The detergent intercedes between the hydrophobic surface of the protein and the bulk aqueous medium. The most effective detergents tend to be the most effective denaturants and the least likely to maintain biological activity. However, many detergents can be used without decreasing the biological activity. A recent interesting development is the synthesis of sodium oligooxyethylene dodecyl ether sulfates.
2.4 Essential Parameters
57
Protein aggregation may be inhibited by surfactants (1 % SDS, 0.1 % dodecylpoly(oxyethyleneglycolether),, 0.01 % poly(oxyethylene),-sorbitane-monooleate, or 0.0 1 % octylphenolpoly(ethyleneglycolether),) as shown for insulin [ 1411. Complete reversibility of rhodanese folding can be achieved with the use of the non-ionic detergent dodecyl-0-D-maltoside (lauryl maltoside). The detergent stabilizes an intermediate with exposed hydrophobic surfaces and prevents aggregation [ 1421. Not only this detergent, but other non-ionic, as well as zwitterionic, detergents also have favorable effects in activating the denatured state [ 1431. Anionic detergents such as SDS are known to increase the a-helix content of several proteins at low protein concentrations [144,145]. Increase of helix content to levels above that of the native protein has been shown for lysozyme, Bence Jones protein, W floribunda lectin, and Histone H2B in presence of SDS. The newly formed helices were stabilized by hydrophobic shielding 11461. Detergents bind to proteins and their removal may be extremely difficult. Another drawback is the restriction of purification techniques available when detergents are used in downstream processing. Ionic detergents will rule out ion-exchange chromatography, and most detergents will interfere with hydrophobic interaction and affinity chromatography. Another specific group of compounds binding to proteins is molecular chaperones that mediate the correct assembly of other polypeptides. Chaperones bind to aggregation-prone conformations of intermediates, thereby stabilizing the unfavorable conformation. The positive effect of protein disulfide isomerase and chaperones were shown in the production of a functional scFv fragment in a E. coli cell-free translation system [ 1471. Preferential interaction measurements have been carried out on a variety of co-solvents [ 1481. Without exception, all are preferentially excluded from the native globular protein, and preferential hydration always induces salting-out, regardless of the mechanism of the preferential interaction [149]. Thus, a strong precipitant is not necessarily a good stabilizer because of the possible interaction with the protein surface, while a good stabilizer will also be a good precipitant. It is, a priori, impossible to state whether a co-solvent will stabilize or destabilize a protein. The effect will depend on possible interactions with the protein surface. The subject is reviewed by Timasheff [130,148,150].
2.4.5 The Effect of Protein Concentration Protein denaturation is a reaction of first order [86] and therefore independent of protein concentration. In contrast, protein aggregation is a reaction of second order where the rate of reaction increases with protein concentration [85]. Therefore, formation of aggregates can be suppressed by working at low protein concentrations. The rate of insulin fibrillation was shown to be a function of the insulin concentration. In acidic solution (pH 2.5, 21 "C) the increase of solution viscosity was much faster in solutions of relatively high insulin concentration [34].
58
2 Protein Stability in Downstream Processing
Glucagon was shown to aggregate at concentrations above 1.5 mg/ml in 0.01 M hydrochloric acid at 30 "C. At 1 mg/ml the reaction proceeds very slowly, confirming the strong concentration dependence of the fibrillation reaction [ 1511.
2,4.6 The Effect of Pressure Pressure does affect protein stability, but at levels far exceeding those met in downstream processing [2,152-1551. At pressures above 570 MPa at 30"C, secondary structure elements of ribonuclease A co-operatively disrupted to a fully unfolded state without any residual structure [ 1561.
2.5 Summary Proteins are only marginally stable in aqueous solutions, the predominant stabilizing forces being the hydrophobic effect and hydrogen bonding. Mutual environmental factors affect the chemical and physical stability of proteins, ranging from presence of stabilizing proteins to parameters such as temperature and pH. A protein undergoing unfolding may not easily regain its native structure, either because of covalent modifications, formation of stable intermediates, or because of aggregation. It is, therefore, an important part of the downstream processing strategy to maintain the native structure of the protein throughout purification. In other words, to maintain the specific biological activity at a level comparable with the in vivo activity, assuming that the native structure is that of highest biological activity. Two issues are of concern before a purification strategy can be laid out. The physical and chemical properties of the protein are generally unknown prior to purification, leaving few data available to the protein chemist. Secondly, the influence of pH, redox potential, and co-solvents on protein stability is virtually unknown, increasing the possibility of unintended destabilization of the protein during purification. A protein's stability in the initial extract can be investigated provided that a bioassay is at hand. Under such circumstances, the bioactivity can be measured as a function of pH, temperature, redox potential, concentration of co-solvents, etc. which can give an idea of the parameter intervals at which the protein is stable, and which cosolvent will be of use during purification. It will also be possible to investigate whether the native structure can be re-established after destabilization, and thereby one can partly characterize the possible upper an lower limits of the said parameters. A downstream process comprises a set of different unit operations with the purpose of increasing the purity of the protein. In order to obtain the highest possible purification factor of each step, parameters like pH, temperature, ionic strength, and protein concentration are often optimized at the expense of protein stability and recovery. In chromatography, each unit operation consists of an application, a binding procedure, and an elution procedure characterized by a set of (different) parameters. Thus, the application sample, the immobilized protein, or the eluted frac-
2.5 Summary
59
tion makes up systems well defined by the parameter intervals. Fortunately, most of the parameters influencing protein stability can be measured in order to determine the lower and upper limits of each parameter which influence the protein stability in a given system. Once the system is defined, the protein stability can be tested by investigating the transition from the native to the unfolded state as a function of temperature and thereby making it possible to determine T,. The robustness of the system and the identification of major interactions between parameters can be studied by employing fractional factorial statistical designs allowing efficient means of testing many variables using a reduced number of data [157]. A well-defined operating space is of course a prerequisite in the production of bulk materials meant for biopharmaceuticals. The notorious instability of many proteins, leading to uncontrolled aggregation and degradation during purification, seriously raises the question whether protein stability should be paid as much attention as protein purity during downstream processing. A downstream process is normally divided into three parts: capture, intermediary purification, and polishing. The purpose of the capture procedure is to separate the protein from host cell proteins, lipoproteins, lipids, DNA or RNA, cell debris, Maillard compounds, and to remove excess water. In this early purification phase, proteolytic degradation may constitute a major problem, as could expression of misfolded forms of the molecule due to the foreign host used, and to over-expression of the protein. The incorrectly folded forms may initiate aggregation, resulting in severe loss of product. Therefore, very gentle methods are recommended for purification at that stage. The most brilliant solution to the problem is to make use of highly selective affinity matrices with ligands recognizing only the native structure. Not only does this concept assure a high purification factor, but the mild application and elution conditions characterizing affinity chromatography favor the stability of the native protein. The lack of effective cleaning procedures for immobilized antibodies and their high production costs have severely restricted the use of affinity chromatography in large-scale downstream processing. However, with the introduction of mimetic triazine-based organic ligands [ 158,1591 costs can be reduced to levels acceptable for large-scale operations. The triazine-based ligands can even be cleaned in 0.5 to 1.0 M NaOH without disrupting the matrix. The polishing steps are mainly introduced to isolate the product from derivatives having close to similar physical and chemical properties. Proteins with perhaps a stabilizing effect on the protein have been removed, and the protein is surely present in much higher concentrations than in the initial extract. At this stage of the downstream process, the choice of co-solvents and combinations of parameters may be of primary importance for protein stability, while contaminants like proteolytic enzymes no longer constitute a problem. With this, hopefully, the questions raised in the introduction to this chapter have been at least partly answered. Factors influencing protein stability have been identified, extreme conditions can be dealt with provided that the parameters influencing protein stability have been carefully adjusted, and a strategy has been laid out to assure the stability of the native structure during downstream processing. It is our belief that improved knowledge of protein stability is an important tool in the design
60
2 Protein Stability in Downstream Processing
of the downstream process. By mastering the different unit operations even denatured proteins can be brought back to their stabilized native form by in vitro folding, reflecting the great potential for manipulation of stability during purification.
Acknowledgements We thank Drs Jens Brange, Daniel Otzen, James Flink, and Benny Wellinder for valuable discussions and comments. We also thank Jytte Jarsholt and Dorte Worm for their help in preparing the manuscript.
Abbreviations and Symbols concentration heat capacity change in heat capacity dithiothreitol redox potential of the solution redox potential relative to the standard hydrogen electrode redox potential at [Ox] = [Red] at pH 0 at unit activity redox potential at [Ox] = [Red] at pH 7 at unit activity redox potential relative to the calomel electrode electron ethylenediaminetetra-acetic acid Faraday's constant change in Gibbs' free energy change in Gibbs' free energy under standard conditions difference in Gibbs' free energy for unfolding glutamic acid decarboxylase EC 4.1.1.15 guanidinium enthalpy difference for unfolding AHH,at the melting temperature high-performance liquid chromatography insulin-like growth factor 1 aggregated intermediary compound stable intermediary compound acid constant equilibrium constant molten globule state 2 -methyl-2,4 -pentadiol native state number of electrons transferred oxidized state
References
PEG PI PKa PLP PMSF Pt R Red RP Asm hSu
T Tm TS U
uv
61
polyethylene glycol isoelectric point -1ogKa pyridoxal 5’-phosphate phenylmethanesulfonyl fluoride platinum gas constant reduced state reversed-phase ASu at the melting temperature entropy difference for unfolding absolute temperature (in Kelvin) melting temperature temperature where ASU = 0 unfolded state ultraviolet
References [ l ] Pace, C.N., Crit Rev Biochem, 1975, 3, 1-43. [2] Privalov, P. L., Adv Protein Chem, 1979, 33, 167-241. [3] Kauzmann, W., Adv Protein Chem, 1959, 14, 1-63. [4] Pace, C.N., J M o l Biol, 1992, 226, 29-35. [5] Burley, S. K. and Petsko, G. A., Trends Biotechnol, 1989, 7, 354-359. [6] Alber, T., Annu Rev Biochem, 1989, 58, 765-798. [7] Sali, D., Bycroft, M., Ferscht, A.R., J Mol Biol, 1991, 220, 779-788. [8] Serrano, L., Kellis, J. T. Jr., Cam, P., Matouschek, A., Ferscht, A. R., J Mol Biol, 1992, 224, 783-804. [9] Matthews, B. W., Curr Opin Struct Biol, 1993, 3, 589-593. [lo] Stickle, D.F., Presta, L. G., Dill, K. A., Rose, G. D., J Mol Biol, 1992, 226, 1143-1159. [ l l ] Shirley, B.A., Stanssens, P., Hahn, U., Pace, C. N., Biochemistry, 1992, 31, 725-732. 1121 Dill, K. A,, Biochemistry, 1990, 29, 7133-7155. 1131 Burley, S.K., Petsko, G.A., Science, 1985, 229, 23-28. [14] Creighton, T. E., Biopolymers, 1983, 22, 49-58. [15] Privalov, P. L., Annu Rev Biophys Chem, 1989, 18, 47-69. [16] Creighton, T.E., Curr Opin Struct Biol, 1991, I , 5-16. [17] Richards, E M . , Annu Rev Biophys Bioeng, 1977, 6, 151-176. [18] Rose, G.D., Annu Rev Biophys Biomol Struct, 1993, 22, 381-415. 1191 Baker, E. N., Hubbard, R. E., Prog Biophys Molec Biol, 1984, 44, 97-179. [20] Chothia, C., J Mol Biol, 1976, 105, 1-14. [21] Matthews, B. W., Nut Struct Biol, 1995, 2, 85-86. [22] Creighton, T. E., J Phys Chem, 1985, 89, 2452-2459. [23] Matthews, B. W., Curr Opin Struct Biol, 1991, 1, 17-21. [24] Jaenicke, R., Eur J Biochem, 1991, 202, 715-728. [25] Seckler, R., Jaenicke, R., FASEB J, 1992, 6, 2545-2552. 1261 Oobatake, M., Tatsuo, O., Prog Biophys Mol Biol, 1993, 59, 237-284. (271 Ptitsyn, 0.B., Uversky, V.N., FEBS Lett, 1994, 341, 15-18.
62
2 Protein Stability in Downstream Processing
Privalov, P.L., Adv Protein Chem, 1979, 33, 167-241. Jaenicke, R., Prog Biophys Mol Biol, 1987, 49, 117-237. Creighton, T.E., Biochem J, 1990, 270, 1-16. Shortle, D., FASEB J, 1996, 10, 27-34. Neri, D., Billeter, M., Wider, G., Wiithrich, K., Science, 1992, 257, 1559-1563. Creighton, T. E., in: Mechnisms of Protein Folding: Pain, R. H. (Ed.), Oxford: IRL Press, 1994; pp. 1-25. Brange, J., Andersen, K., Laursen, E. D., Meyn, G., Rasmussen, E., J Pharm Sci, 1997, in press. North, M. J., in: Proteolytic Enzymes - a practical approach: Beynon, R . J., Bond, J. S. (Eds.), Oxford: IRL Press, 1989; pp. 105-124. Battersby, J. E., Hancock, W. S., Canova-Davis, E., Oeswein, J., O’Connor, B., Int Peptide Protein Res, 1994, 44, 215-222. Schultz, J., Methods Enzymol, 1967, 11, 255-263. Piszkiewicz, D., Landon, M., Smith, E. L., Biochem Biophys Res Commun, 1970, 40, 11731178. Whitaker, J. R., Feeney, R. E., Crit Rev Food Sci Nutl; 1983, 19, 173-212. Hayashi, R., Kameda, I., Agric Biol Chem, 1980, 44, 175-181. Berson, S. A., Yalow, R. S . , Diabetes, 1966, 15, 875-879. Brange, J., Langkjaer, L., Havelund, S . , Voelund, A., Pharm Res, 1992, 9, 715-726. Lewis, U. J., Cheever, E. V., Hopkins, W. C., Biochim Biophys Acta, 1970, 214, 498-508. Flatmark, T., Acta Chem Scand, 1966, 20, 1487-1496. Geiger, T., Clarke, S . , J Biol Chem, 1987, 262, 785-794. [46] Clarke, S., Stephenson, R. C., Lowenson, J. D., in: Stability of Protein Pharmaceuticals: Ahem, T. J., Manning, M. C., (Eds.), New York: Plenum Press, 1992; Part A, pp. 1-29. [47] Chazin, W. J., Kossiakoff, A. A., in: Deamidation and Isoaspartate Formation in Peptides and Proteins: Aswad, D. W. (Ed.), London: CRC Press, 1995; pp. 194-206. [48] Slobin, L. I., Carpenter, F.H., Biochemistry, 1963, 2, 22-28. (491 Brennan, T.V., Clarke, S . , in: Deamidation and Isoaspartate Formation in Peptides and Proteins: Aswad, D. W. (Ed.), London: CRC Press, 1995, pp. 66-90. [50] Yuksel, K. U., Gracy, R. W., Arch Biochem Biophys, 1986, 248, 452-459. [51] Bhatt, N.P., Patel, K., Borchardt, R. T., Pharm Res, 1990, 7, 593-599. [52] Robinson, A. B., Rudd, C. J., in: Current Topics in Cellular Regulation: Horecker, B. L. and Stadtmann, E. R., (Eds.), New York: Academic Press, 1974; pp. 247-295. 1531 Manning, M. C., Patel, K., Borchardt, R. T., Pharm Res, 1989, 6, 903-918. [54] Wright, H.T., Protein Eng, 1991, 4, 283-294. [55] Johnson, B. A., Aswad, D. W., in: Deamidation and Isoaspartate Formation in Peptides and Proteins: Aswad, D. W. (Ed.), London: CRC Press, 1995; pp. 91-113. [56] Becker, G. W., Tackitt, P. M., Bromer, W. W., Lefeber, D. S., Riggin, R. M., Biotech Appl Biochem, 1988, 10, 326-337. [57] Scotchler, J. W., Robinson, A. B., Anal Biochem, 1974, 59, 319-322. [58] Sen, L. C., Gonzalez-Flores, E., Feeney, R. E., Whitaker, J. R., JAgric Food Chem, 1977,25, 632-638. [59] Lee, H. S., Osuga, D. T., Nashef, A. S., Ahmed, A. I., Whitaker, J. R., Feeney, R. E., J Agric Food Chem, 1977, 25, 1153-1158. [60] Nashef, A. S., Osuga, D. T., Lee, H. S., Ahmed, A. I., Whitaker, J. R., Feeney, R. E., J Agric Food Chem, 1977, 25, 245-251. [61] Volkin, D. B., Klibanov, A.M., J Biol Chem, 1987, 262, 2945-2950. [62] Masters, P. M., Friedman, M., ACS Symp Sel; 1980, 123, 165-194. [63] Hayashi, R., Kameda, I., Agric Biol Chem, 1980, 44, 891-895. [64] Dedman, M. L., Farmer, T. H., Morris, C. J. O.R., Biochem J, 1961, 78, 348-352. I651 Tashijan, A.H. J., Ontjes, D. A., Munson, P.L., Biochemistry, 1964, 3, 1175-1182. [66] Morley, J. S., Tracy, H. J., Gregory, R.A., Nature, 1965, 207, 1356-1359.
References
63
[67] Brot, N., Weissbach, H., Trends Biochem Sci, 1982, 7, 137-139. [68] Cleland, J. L., Powell, M. F., Shire, S. J., Crit Rev Ther Drug Carrier Sys, 1993, 10, 307377. [69] Gilbert, H. F., Adv Enzymol, 1990, 63, 69-172. [70] Wold, F., Macromolecules: Structure and Function. London: Prentice-Hall, Inc. 197 1; pp. 53-89. [71] Martin, B. M., Viswanatha, T., Biochem Biophys Res Commun, 1975, 63, 247-253. [72] Torchinsky, Y. M., in: Sulfur in Proteins: Metzler, D. (Ed.), Oxford: Pergamon Press, 1995; pp. 199-217. [73] Shaked, Z., Szajewski, R. P., Whitesides, G. M., Biochemistry, 1980, 19, 4156-4166. [74] Soerensen, H. H., Thomsen, J., Bayne, S., Hoejrup, P., Roepstorff, P., Biomed Environ Mass Spec, 1990, 19, 713-720. [75] Benesch, R. E., Benesch, R., J Am Chem Soc, 1958, 80, 1666-1669. [76] Florence, T. M., Biochem J, 1980, 189, 507-520. [77] Torchinsky, Y. M., in: Sulfurin Proteins: Metzler, D. (Ed.), Oxford: Pergamon Press, 1981; pp. 57-65. [78] Misra, H. P., J Biol Chem, 1974, 249, 2151-2155. [79] Jespersen, A.M., Christensen, T., Klausen, N. K., Nielsen, P. F., Soerensen, H.H., Eur J Biochem, 1994, 219, 365-373. [80] Tanford, C., Adv Protein Chem, 1968, 23, 218-281. [81] O’FagBin, C., Sheehan, H., O’Kennedy, R., Kilty, C., Process Biochem, 1988, X X , 166171. [82] Kunth, I. D. J., in: The Protein Folding Problem: Wetlaufer, D. B. (Ed.), AAAS Selected Symposium 89, 1984; pp. 65-81. [83] Matthiesen, F., Hejnaes, K.R., Shiver, L., Ann N YAcad Sci, 1996, 782, 413-421. [84] Hejnaes, K.R., Bayne, S., Noerskov, L., et al., Protein Eng, 1992, 5, 797-806. [85] Kiefhaber, T., Rudolph, R., Kohler, H.-H., Buchner, J., Biotechnology, 1991, 9, 825-829. [86] Jaenicke, R., in: Protein Conformation: New York: John Wiley and Sons, 1991, pp. 206217. [87] Oliveberg, M., Fersht, A. R., Biochemistry, 1996, 35, 2738-2749. [88] Speed, M.A., Wang, D.I.C., King, J., Nat Biotech, 1996, 14, 1283-1287. [89] De Young, L.R., Dill, K.A., Fink, A. L., Biochemistry, 1993, 32, 3877-3886. [90] Moody, A. J., Hejnaes, K.R., Marshall, M. 0. et al., Diabetologia, 1995, 38, 14-23. [91] Chou, P. Y., Fasman, G.D., Biochemistry, 1975, 14, 2536-2541. [92] Zettlmeissl, G., Rudolph, R., Jaenicke, R., Biochemistry, 1979, 18, 5567-5571. [93] DeFelippis, M. R., Alter, L.A., Pekar, A. H., Havel, H. A., Brems, D.N., Biochemistry, 1993, 32, 1555-1562. [94] Sluzky, V., Klibanov, A.M., Langer, R., Biatech and Bioeng, 1992, 40, 895-903. [95] Mendoza, J.A., Rogers, E., Lorimer, G.H., Horowitz, P.M., J Biol Chem, 1991, 266, 13587-13591. [96] Bai, Y., Milne, J. S., Mayne, L., Englander, S. W., Proteins, 1994, 20, 4-14. [97] Cleland, J. L., Wang, I.C., Biochemistry, 1990, 29, 11072-11078. [98] Tandon, S., Horowitz, P., J Biol Chem, 1986, 261, 15615-15681. [99] Hartley, D.L., Kane, J.F., Biochem Soc Trans, 1988, 16, 101-102. [loo] Scopes, R. K., in: Protein Purification - Principles and Practice: New York: Springer-Verlag, 1987; pp. 41-71. [ l o l l Hoffmeister,. F., Naun-Schmiedelberg’s Archiv Pharmak. Exp. Patholog (Leipzig), 1988, 24, 249-260. [I021 Pace, C.N., Trends Biochem Sci, 1990, 15, 14-17. [lo31 Becktel, W. J., Schellman, J. A., Biopolymers, 1987, 26, 1859-1877. [lo41 Doty, P., Imahori, K., Klemperer, E., Proc Nut1 Acad Sci USA, 1958, 44, 424-431. [I051 Hippel, P. H., Schleich, T., in: Structure and Stability of Biological Macromolecules: Timasheff, S. N., Fasman, G. D. (Eds.). New York: Marcel Dekker Inc. 1969; pp. 417-557.
64
2 Protein Stability in Downstream Processing
I1061 Tombs, M.P., J Appl Biochem, 1985, 7, 3-24. 11071 Sundby, F., J Biol Chem, 1962, 237, 3406-3411. [lo81 Wheelock, E.F., Science, 1965, 149, 310-311. [lo91 Hoshino, T., Mikura, Y., Shimidzu, H., Kusurnoto, S., Kawai, J., Toguchi, H., Biochim Biophys Acta, 1987, 916, 245-250. [110] Shami, E. Y., Rothstein, A., Rarnjeesingh, M., Trends Biotechnol, 1989, 7, 186-190. [ l l l ] Whitaker, J.R., ACS Symp Ser; 1980, 123, 145-163. [112] Chalikian, T. V., Gindikin, V. S . , Breslauer, K. J., FASEB J, 1996, 10, 164-170. [113] Porter, W.R., Staack, H., Brand, K., Manning, M.C., Thromb Res, 1993, 71, 265-279. [114] Janin, J., Colloids and Surfaces, 1984, 10, 1-7. [115] Makhatadze, G. I., Privalov, P. L., J Mol Biol, 1993, 232, 639-659. [116] Shellmann, J. A,, Lindorfer, M., Hawkes, R., Grutter, M., Biopolymers, 1981, 20, 19891999. [117] Privalov, P.L., Khechinashvili, N.N., J Mol Biol, 1974, 86, 665-684. [118] Privalov, P. L., Griko, Y. V., Venyarninov, S. Y., Kutyshenko, V. P., J Mol Biol, 1986, 190, 487-498. (1191 Hermans, J. J., Scheraga, H.A., J Am Chem SOC, 1961, 83, 3283-3292. [120] Volkin, D. B., Middaugh, C. R., in: Stability of Protein Pharmaceuticals: [121] Baldwin, R. L., Eisenberg, D., in: Protein Engineering: Oxender, D. L., Fox, C. F. (Eds.), New York: Alan R. Liss, Inc, 1987; pp. 127-148. [122] Tomazic, S. J., Klibanov, A.M., J Biol Chem, 1988, 263, 3086-3091. [123] Zale, S. E., Klibanov, A.M., Biochemistry, 1986, 25, 5432-5444. [124] Cleland, W. W., Biochemistry, 1964, 3, 480-482. [125] Scott, E.M., Duncan, I. W., Ekstrand, V., J Biol Chem, 1963, 238, 3928-3933. 11261 CRC Handbook of Chemistry and Physics. 76thed. New York: 1996; pp. 8.21-8.33. [127] Dutton, P.L., Methods Enzymol, 1978, 54, 411-435. [128] Maeda, Y., Yarnada, H., Ueda, T., Irnoto, T., Protein Eng, 1996, 9, 461-465. [129] Hospelhorn, V.D., Cross, B., Jensen, E.V., J Am Chem SOC, 19854, 76, 2827-2829. [130] Timasheff, S. N., in: Stability of Protein Pharmaceuticals: Ahern, T. J., Manning, M. C. (Eds.), New York: Plenum Press, 1992; Part B, pp. 265-285. [131] Arakawa, T., Prestrelski, S. J., Kenney, W. C., Carpenter, J. F., Adv Drug Del Rev, 1993, 10, 1-28. [132] Arakawa, T. R., Timasheff, S. N., Biochemistry, 1985, 24, 6756-6762. [133] Pittz, E.P., Tirnasheff, S. N., Biochemistry, 1978, 17, 615-623. [134] London, J., Skrzynia, C., Goldberg, M.E., Eur J Biochem, 1974, 47, 409-415. [135] Brems, D.N., Havel, H.A., Proteins, 1990, 5, 93-95. [136] Buchner, J., Rudolph, R., Biotechnology, 1991, 9, 157-162. [137] Winkler, M. E., Blaber, M., Biochemistry, 1986, 25, 4041-4045. [138] Zardeneta, G., Horowitz, P.M., J Biol Chem, 1992, 267, 5811-5816. [139] Kristjansson, M.M., Kinsella, J. E., Adv Food Nutr Res, 1991, 35, 237-316. [140] Timasheff, S.N., Annu Rev Biomol Struct, 1993, 22, 67-97. [141] Lougheed, W. D., Albisser, A.M., Martindale, H. M., Chow, J. C., Clement, J. R., Diabetes, 1983, 32, 424-432. [142] Tandon, S., Horowitz, P. M., J Biol Chem, 1989, 264, 9859-9866. 11431 Tandon, S., Horowitz, P. M., J Biol Chem, 1987, 262, 4486-4491. [144] Jirgensons, B., J Biol Chem, 1967, 242, 912-918. [145] Hunt, A. H., Jirgensons, B., Biochemistry, 1973, 12, 4435-4441. [146] Jirgensons, B., J Protein Chem, 1982, I , 71-84. [147] Ryabova, L. A., Desplancq, D., Spirin, A. S . , Pliickthun, A,, Biochemistry, 1997, 15, 79-84. 11481 Timasheff, S. N., Annu Rev Biophys Biomol Struct, 1993, 22, 67-97. [149] Arakawa, T., Bhat, R., Timasheff, S. N., Biochemistry, 1990, 29, 1914-1923. [150] Timasheff, S. N., in: Protein-Solvent Interactions. Gregory, R. B. (Ed.), New York: Marcel Dekker, Inc., 1994; pp. 445-482.
References
65
[151] Beaven, G.H., Gratzer, W.B., Davies, H. G., Eur J Biochem, 1969, /2/, 37-42. [152] Heremans, K., Annu Rev Biophys Bioeng, 1982, I / , 1-21. [153] Silva, J.L., Annu Rev Phys Chem, 1993, 44, 90-113. [154] Gross, M., Jaenicke, R., Eur J Biochem, 1994, 221, 617-630. [155] Goossens, K., Smeller, L., Frank, J., Heremans, K., Eur J Biochem, 1996, 236, 254-262. [156] Takeda, N., Kato, M., Taniguchi, Y., Biochemistry, 1995, 34, 5980-5987. [157] Montgomery, D. C., in: Design and Analysis of Experiments: New York: John Wiley Sons, 1991; pp. 197-249. [158] Clonis, Y.D., Stead, C.V., Lowe, C.R., Biotech Bioeng, 1987, 30, 621-627. [159] Lowe, C. R., Burton, S. J., Burton, N. P., Alderton, W. K., Pitts, J. M., Thomas, J. A,, Trends Biotechnol, 1992, 10, 442-448.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
3 Production of Transgenic Protein Gordon Wright and John Noble
3.1 Introduction As the market penetration of recombinant DNA technologies increases, more and more pharmaceutical companies are looking to this area to strengthen their new product pipelines [l].Central to the success of this strategy is the development of production techniques that are price competitive with, or demonstrating a higher safety and efficacy than, conventiona) techniques such as surgery, chemotherapy, and existing drug products [2]. One of the most exciting advances within biotechnology in recent years has been transgenic technology, whereby the secretion of human proteins in the mammalian mammary gland could potentially revolutionize commercial protein production. The central strength of the technology is the ability to produce, in high volume, low-cost complex proteins for which current production methods are prohibitively expensive at large scale, or no successful production method is yet available. The following chapter overviews the technology and examines key issues that must be addressed in process development, facility design, and regulatory affairs as products reach commercial goals.
3.2 Overview of Transgenic Technology The core technology for the manufacture of proteins in the milk of transgenic animals is illustrated in Fig. 3-1. First, the DNA which encodes the therapeutic protein is linked with a milk gene promoter which ensures that the gene is switched on and the protein is made only in the mammary gland. Then the linked DNA is injected into the pronucleus of a fertilized egg which is placed in a foster mother and develops naturally through pregnancy to full term. The resultant offspring are screened for the presence of the transgene and are mated when mature. The human protein in the milk of the lactating animals is then measured and analysed. The animal which yields the highest levels of the required protein is used to produce a male which will become the founder animal of the production flock. This ensures that genetic varia-
68
3 Production of Transgenic Protein
Fig. 3-1. Core technology.
tion is kept to a minimum. Once a founder is established, all further work is carried out using conventional animal breeding. The levels and bioactivity of the protein in the milk of the transgenic animals are analysed and the purification processes and assays are then designed and developed. The protein is isolated from the milk using a combination of techniques adapted from the dairy industry and conventional pharmaceutical separation techniques such as column chromatography. This approach can be applied to a range of species including sheep, cows, goats, pigs, and rabbits. The choice is determined by issues such as time, cost, and the quantity of product required. Although the above describes how most transgenic animals are currently produced, the use of newer technologies would allow more precise modifications and be more efficient. For example, in the mouse one can make use of embryonic Table 3-1. Manufacturing routes for biopharmaceutical products. Method of Production
Transgenics
Blood fractionation
BacteriaUfungal fermentation
Mammalian cell culture
Secretion levels (g/L)
1 to 30
Product-specific
1 to 3"
0.1 to 0.8"
Purification yield (%)
50"
Product-specific
loa
60"
Product complexity
Wide-ranging
All blood proteins
No human glycosylation
Wide-ranging
Scale-up
straightforward
Straightforward
Complex
Very complex
Feedstock
Renewable
Limited supply
Defined
Defined and expensive
Safety issues
Species barrier to viral transfer
Same species viral transfer
Control of GMOs
Control of virus
a
From reference [ 2 ] .
3.3 Process Design
69
stem (ES) cells to target a specific region of the genome, adding or removing DNA sequences. This technology would be extremely beneficial in large animals, but to date no ES cells have been isolated for any of the livestock species. The recent advent of cloning [3] in livestock offers the possibility of many, if not all, of the same advantages. A comparison of transgenics with three well-established production methods: blood fractionation, bacterial fermentation, and mammalian cell culture, is given in Table 3-1. From Table 3-1 it is clear that transgenic technologies are capable of: (i) very high volume production; (ii) complex molecule production; and (iii) low cost production. The latter point is highlighted further by industry data which demonstrate up to 35 % reduction in production costs through the switch from fermentation to transgenic production routes [ 2 ] . It is thus easy to understand why companies such as PPL Therapeutics, Genzyme Transgenics, and Pharming have invested heavily in the technology 141.
3.3 Process Design A typical transgenic production process is shown in Fig. 3-2. At first sight, although the unit operations involved within the process seem like a standard biopharmaceutical process, closer examination reveals a unique combination of process operations. Progressing through the process we move gradually from an agricultural environment, through processing similar to that employed in dairies, into primary biopharmaceutical production, and then into a classical secondary pharmaceutical environment. The critical issue in process design is to facilitate this progression, through bands of differing technology, without compromising product quality. Central to this is the specification of relevant standards for each area of process operation. The process can be split into six areas: animal handling, milking, milk handling, primary recovery, polishing, and formulation/filling. A design philosophy
I
Primary 'Recovery
-
70
3 Production of Transgenic Protein
can be developed for each of these areas ensuring that systems are fit for purpose, readily adapted to scale-up, and display a realistic approach to the technologies involved.
3.3.1 Animal Handling In specifying the animal handling facilities one must have a clear vision of what the end point for a particular product will be. For example, the production of metric tonne quantities of protein in sheep, goats, or cows is likely to require vastly different strategies to the production of gramme quantities in rabbits. This is addressed in more detail in Section 3.4.6. To provide the necessary quality assurance, each animal within the production group should be effectively tagged, to provide unique identification and allow details of life history and health status to be recorded. Before admission to the milking flock each animal should comply with a minimum defined health status and produce milk consistent with predefined parameters.
3.3.2 Milking The milking area is a critical point within a transgenic production process as it is the point at which the product is ‘harvested’ from the production animal. It is thus the first time at which the milk can become contaminated with adventitious agents. It is not practical to milk animals under closed/aseptic conditions. Even in the cleanest dairy environment with pipework sterilization, levels of bacteria up to 10000 cfdml can still be detected [6]. Milking practices, equipment design, and process schedules must be designed to address this problem and ensure end-product quality is not compromised. Milking equipment is best selected from high-quality dairy systems with additional attention paid to issues of maximizing milk recovery, validation of system cleaning, and specification of control systems. Milk recovery will be critical during start-up when animal numbers are likely to be low. The integration of the parlour control system with the animal tagging system is essential to the maintenance of milk quality, ensuring that milk from risk animals does not enter the process stream. Typical practice at this stage would be to pool the milk from a single milking for shipping to the milk handling process. Any manipulation of individual milkings must take place within pre-defined acceptance criteria.
3.3 Process Design
71
3.3.3 Milk Handling To make best use of downstream processing equipment it is advantageous to be able to hold the milk after collection for subsequent poolinghb lotting. This allows the purification batch size to be tailored to fit the available equipment and decouples purification scale from animal numbers. Typically, during early production several milkings will be pooled in a single batch while as animal numbers grow milkings may be split into several batches. The key issue at this stage is thus the generation of a stable process stream which can be stored for significant periods. The levels of contamination outlined in Section 3.3.2 could, after only a few hours, lead to product loss. In a conventional dairy process this problem is addressed via pasteurization. This technique may not be acceptable in the production of certain proteins. Historical data indicate that when stored at 4"C, unpasteurized milk can begin to sour in under 20 h. It is thus unlikely that we would look to hold milk for use in biopharmaceutical production for more than 12 h without stabilization. For robust proteins, freezing and low-temperature storage, -50 "C, may be possible. For labile molecules, techniques such as chemical stabilization should be investigated. If no effective process is available, the holding of milk prior to purification will not be possible and downstream operations will have to be designed accordingly. Fat removal could also be a key step at this stage of the process as lipid micelles can have a serious effect upon the classical downstream processing operations that will be encountered later in the process. Disruption of the lipid micelles via shear can also lead to souring of the milk and the process stream should be handled with low-shear equipment where possible.
3.3.4 Primary Recovery The main aim at this stage in the purification is to separate the target protein from the bulk of the high concentration matrix of sugars, lipids, and other proteins which constitute the milk. A further key issue at this stage is that the process stream may well have a significant bioburden. It is thus essential that any purification strategy generates a process stream for which the bioburden can be controIled by, for example, filtration. Of prime importance is the removal of caseins (which form the bulk of the milk protein) from the process stream. One potential approach involves the solubilization of the caseins with the capture of the target protein on a chromatography column [7]. Typically, one would anticipate several purification steps resulting in bulk-purified target protein in a suitable buffer solution which can be filtered through a 0.2 micron filter. For proteins that are sensitive to protease attack the speed and temperature of these operations will be critical to product yield. Equipment encountered at this stage will be conventional biophannaceutical recovery systems [Sl such as chromatography and centrifuges. Given the special nature of the milk, real system trials are essential, even if milk reserves are scarce. Care
72
3 Production of Transgenic Protein
has to be taken in educating vendors as to the nature of the process, as few will be familiar with the protein levels encountered in product and waste streams. This is especially true where precipitation is used.
3.3.5 Polishing Steps Within this phase of the purification process the aim is to move from a bulk-purified product to a final product of the required purity. For human use, purities in excess of 99.9 % may be required [2]. Equipment and operations will be the same as for most modern therapeutic protein production systems [8], including chromatography and tangential flow filtration.
3.3.6 Formulation and Filling As with the final polishing steps the formulation and filling operation will most likely be identical to classical liquid dosage from products including buffer exchange, concentration, sterile filtration, filling, and freeze-drying.
3.3.7 Viral-Specific Steps The control of virallprion contaminants is a critical issue with mammalian-derived products. The removal of these components must be built into the purification strategy. Operations such as precipitation, chromatography, and freeze-drying all have a viral inactivation capacity and as such one can consider the whole process to act as a viral screen. Despite this, for less well-characterized feedstocks, of which milk is one, manufacturing processes will normally incorporate generic viral inactivation or removal steps [9]. Examples of such steps include holding at low pH, pasteurization, dry heat, solvenvdetergent, filtration, UV irradiation, and microwave irradiation. Typically one would expect to see an inactivation/removal step as the final stage of polishing. The choice of step and its integration into the process can be critical to product yield, facility design, and process economics and must be assessed in detail.
3.3.8 Utilities Appropriate choice of utilities is central to the design of safe, robust and cost-effective production processes and this is particularly true of transgenic products where there is such a change in production environment with increasing product quality.
3.3 Process Design
73
The larger the process scale, the more critical these issues will become. Leaving conventional building services aside [8] specialist utilities in transgenic facilities can be considered as : water, cleaning/sanitization, and sterilization.
3.3.8.1 Water All biopharmaceutical production processes utilize relatively large quantities of water in cleaning and production operations. To ensure that product integrity is not compromised and cost-effective design solutions are maintained the usage outlined in Table 3 - 2 is proposed. This clearly shows a gradual increase in water quality with increasing product purity. Table 3-2. Appropriate water grades within production process. Duty
Milking
Milk handling
Primary recovery
Polishing
Finishing
Cleaning
Towns water
Towns water
Process water
Process water
Process water
Final rinse
Towns watera Process water WFI
WFI
WFI
WFI
WFI
Product contact NIA
NIA
WFI
Process water represents a high-purity, low endotoxin, low bioburden water produced by ion exchange or reverse osmosis. a Plus sterilizing agent. NIA, not applicable; WFI, water for injection.
-
3.3.8.2 Cleaning and Sanitization The cleaning of biopharmaceutical production facilities is a critical activity and it must be integrated into the design from day one. Typically, one would anticipate a cleaning cycle involving a pre-rinse, clean, acid rinse, post-rinse and final rinse, and could be expected to last up to 45 minutes [lo]. In transgenic processes the need to control the scrapie agent can lead to the inclusion of an additional step utilizing the recirculation of 1 M NaOH for 1 h at 20 "C or above. The inclusion of such a step in the cleaning cycle will have significant impact upon the batch schedule and effluent production from any process and must be addressed in detail. In general the majority of the unit operations upstream of final filling are not suitable for steam sterilization and the bioburden within process equipment, especially in chromatography columns, is best controlled utilizing chemical sanitization and storage. Suitable agents include sodium hydroxide, ethanol/water, and hypochlorite.
74
3 Production of Transgenic Protein
3.3.8.3 Sterilization The final filling equipment will need to be sterilized prior to use. Equipment should be sterilised with clean steam, dry heat or a chemical sterilant. Occasional steam sterilization of upstream vessels and pipework may be advantageous.
3.4 Facility Design For any biopharmaceutical product the facility design is process-led. The building configuration and services must be specified such that the process can be operated in a manner which does not compromise product quality. The following sections highlight the key design features of the critical building areas.
3.4.1 Overview of Facility Design An overview of a typical transgenic facility design is given in Fig. 3-3. Here, the facility is presented as a central support core from which the various production modules hang as satellites. As one moves clockwise around the production module +
People & Materials Inlout
-Animal
*
...............
Husbandry support
Flow of people and materials Flow of product
Milking
............................................................................................................
:
Animal Handling Unit Production Unit
People & Materials Inlout
Fig. 3-3. Facility overview.
u
3.4 Facility Design
75
the product purity and the quality of the facility will increase accordingly. Key points on the design include: Complete segregation of personnel in contact with animals from those involved in purification. No personnel link between milking and purification only allowance for milk transfer. Segregation of primary recovery and polishing operations within a common zone to assist control of bioburden and cross-contamination. Segregation of formulation and filling activities from other purification operations to maintain viral/prion free status of final product. The support core handles the transfer of people from uncontrolled to the controlled production envelope, administration activities, quality control (QC) activities and maintenance activities.
3.4.2 Animal Housing and Milking It is essential that personnel working in the animal housing unit (AHU) are completely segregated from those producing the purified product. The AHU must be equipped with a dedicated entry and exit facility and enclosed such that the health status of the animals can be maintained. Visitors should be screened for disease and should not be allowed to enter the process building without suitable quarantine/decontamination.
3.4.3 Milk Handling Milk handling may form a segregated area within the production building with a dedicated entry/exit from the central support core. The transfer of milk from the milking area should, if possible, be via fixed lines. Where volumes or distances preclude this any containers must be suitably decontaminated before entering the production unit. Transfer of the batch to the recovery/purification module should be via fixed lines if possible.
3.4.4 Primary Recovery and Polishing Primary recovery and polishing are best designed as a suite with a dedicated, common, entry and exit from the central support core. Personnel, equipment, and product can move freely within the zone but there should be a general increase in purity and decrease in bioburden as one moves from recovery through polishing. Transfer to the formulation module should be via fixed lines to maintain the viral-free status of the product.
76
3 Production of Transgenic Protein
3.4.5 Formulation and Finishing Formulation and finishing should be designed as two separate modules each with dedicated entry and exit from the central support core.
3.4.6 Overview of Building Finishes and HVAC The finish and HVAC conditions within the various areas is heavily dependent upon scale and the nature of the process. At large scale, (> 500 L of milkbatch), it is possible that a closed process could be designed which would limit the need for specialist HVAC and building finishes to the formulation and filling areas. At smaller scales, the building finishes and HVAC in the production area are likely to be more critical to product quality. An overview of finishes and HVAC in the various building areas is given in Table 3-3. Where closed processing is not possible, the potential for contamination via HVAC systems must be addressed. Typically, each building module should have a dedicated system and within milk handling, recovery and polishing 100 % exhaust of the air may be a sensible option.
Table 3-3. Overview of building finishes and HVAC. Type of production
Animal housing
Milking parlour
Milk handling
Recovery1 polishing
Formulation1 filling
Small mammals
Laboratory
Laboratory
Class D Lab.
Class C Lab.
Class C/ Class A Lab.
Large mammals, small scale
Agricultural
High quality dairy
Class D Lab.
Class C Lab.
Class CI Class A Lab
Large mammals, intermediate scale
Agricultural
High quality dairy
Hygienic area
Class C clean rooms
Class CI Class A clean rooms
Large mammals, large scale
Agricultural
High quality dairy
Hygienic area
Hygienic area
Class CI Class A clean rooms
3.4.7 Site Planning Issues For large-volume production the location and number of production modules is a key decision. Typically, the adoption of two or more animal sites is recommended to ensure against production loss through disease. Given the potential instability of
3.5 cGMP and Regulatory Issues
77
the raw milk it will be beneficial to have milk handling operations local to these facilities. So long as the milk can be stabilized the use of a single, centralized, unit for purification, formulation and filling is possible.
3.5 cGMP and Regulatory Issues The following section highlights issues of product safety, efficacy and reliability. It also discusses containment requirements for genetically modified organisms.
3.5.1 Good Manufacturing Practice Within the pharmaceutical industry there is a high level of legislation to safeguard the safety and efficacy of drug products and there are well-established guidelines covering Good Manufacturing Practice (GMP) for biological products [ 11,121. In recent years the emergence of transgenic technology and its move towards commercialization has spurred the production of several regulatory documents detailing critical regulatory issues [ 13,141. Key issues arising between transgenic and classical rDNA products are as listed below with suggested actions:
The need to redefine concepts of the master and working cell banks. In many ways the first transgenic animal of the line (Genetic Founder) can be equated to the master cell bank and one or more male offspring (Production Founders) to a working cell bank. However, although this comparison can be useful, care should be taken since any new technology will best be dealt with by developing concepts which are customized to its needs. - The, potential variation of milk composition with lactation period, feed composition, individual animals and generations und the need to validate production processes to cater for this. Experience at PPL is that the composition of milk is less variable than might be suggested in the literature of this area. This is especially true if feeding, maintenance and housing of animals is carefully controlled. - Declaration of viral/prion status of animals and control of adventitious contamination. Animals can be serologically tested periodically for pathogens of specific concern. Unfortunately, control of prion-like diseases can only be controlled by clinical examination, although periodic culling of animals followed by post-mortem examination should be considered. - Control of sick animals and their segregation from the production flock. If an animal is identified which potentially could put the product at risk, it is important that segregation is well controlled and that policies on reintroduction of such animals back into the production herd are well considered. - Use of medicines during animal husbandry. This is a complex area since the number of chemicals that an animal may come into contact with may be very large, either given as medicines or through feeding. A great deal of information
-
78
3 Production of Transgenic Protein
is available in this area from the dairy industry and careful adaption of this information can go a long way to helping solve this problem. - Gene stability during breeding process: Not all transgenic lines ultimately prove to be stable. The number of copies of the transgene can become reduced for a variety of reasons and this normally results in a reduction in the level of the product found in the milk. Decreases in expression level of proteins has also been observed without an apparent loss of transgene copies [15]. The importance of using a stable transgenic line with a constant level of product in their milk cannot be overemphasized.
3.5.2 Containment of Genetically Modified Organisms (GMO) The manipulation of GMO is carried out under strict government legislation both in Europe and in the USA. For example, in the UK two Acts of Parliament, the contained use [16] and the deliberate release regulations [17,18] outline rules under which work must be undertaken. Once outside the animal housing the process stream does not constitute a biocontainment hazard.
References [I] Ernst Young, European Biotec 95 Gathering Momentum, Ernst and Young International. [2] Werner, R. G., Transgenic Technology a Challenge for Biotechnology, Biotechnica, 1995. [3] Wilmut, I., et al, Viable offspring derived from fetal and adult mammalian cells, Nature, 1997, 385. [4] Competition seeks to follow Genzyme Transgenics with a product into Clinic, Genetic, Engineering News, November I, 1996, p 5. [5] Harvesting Proteins, Biotechnology News, BMB Initiative, September 1996, EMAP Maclaren Ltd, p. 8. [6] Sheep, Dairy News, 1986, Vol. 3, No. 2, p. 7. [7] Cole, E. S., Production and Characterisation of Human Anti-Thrombin I l l Produced in the milk of Transgenic Goats, Biotechnica, 1995. [8] Laydersen, et al., Bioprocess Engineering, Systems, Equipment Utilities. John Wiley Sons, ISBN 0- 47 1-03544-0. [9] ICH, Quality of biotechnological products viral safety evaluation of biotechnological products derived from cell lines of human or animal origin. Step 2, Draft, 1/12/95. [lo] Sherwood, D., et al., Experiences with clean in place validation in a Multi product biopharmaceutical manufacturing facility. Eur J Parenteral Sci, 1996, 1(2), 35-41. [ 1I] FDA Good Manufacturing Practice Regulations for Biologics, Code of Federal Regulations 21, parts 600 to 799, U. S. Government Printing Office, 1990. [I21 The Rules Governing Medicinal Products in The European Community, Vol. 4, Good Manufacturing Practice for Medicinal Products, CEC 1992, ISBN 92-826-31 80-x. [I31 CBER, Points to consider in the manufacture and testing of therapeutic products for human use derived from transgenic animals. Food and Drug Administration Docket No. 95D-0131, 1995.
References
79
[ 141 Medicines Control Agency, Use of transgenic animals in the manufacture of biological and
medicinal products for human use. Eurodirect Publication No. 3612/93, 1993. (151 Caver, A., et al., Transgenic livestock as bioreactors: stable expression of human alpha-lantitrypsin by a flock of sheep. Biotechndogy, 1993, 2, XX-XX. [ 161 The Genetically Modified Organisms (contained use) Regulations, Health and Safety, SI 3217, 1992. [ 171 The Genetically Modified Organisms (deliberate release) Regulations, SI 3280, 1992. [ 181 The Deliberate Release of Genetically Modified Organisms to the Environment, Directive 90/220/EEC, May 1992.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
4 Harvesting Recombinant Protein Inclusion Bodies Anton P. J. Middelberg and Brian K. O’Neill
4.1 Introduction The overexpression of recombinant protein in Escherichia coli often leads to concentration of that protein as a solid granule called an inclusion body. Numerous proteins are reported to produce inclusion bodies in E. coli, including the somatotropins, growth factors, tissue plasminogen activator, and insulin. Techniques for processing inclusion bodies follow a highly conserved sequence of operations. As discussed in Section 4.4, these typically include cell disruption, recovery of the inclusion bodies, dissolution and renaturation to yield active protein, and high-resolution recovery of the final protein using conventional methods such as chromatography. In this chapter we will focus on the large-scale harvesting of recombinant protein inclusion bodies; a scale-up problem largely neglected in the current literature. In Sections 4.2 and 4.3 we review the properties of inclusion bodies and cellular debris. This knowledge is important in optimizing any collection strategy. In later Sections we then examine processes for dealing with inclusion bodies, before an in-depth examination of the alternatives for harvesting inclusion bodies, namely filtration, centrifugation and newer approaches involving in situ dissolution.
4.2 What is an Inclusion Body? Inclusion bodies (IBs) are solid, micron-scale protein particles contained within the cytoplasm of E. coli. They appear as electron-dense bodies under electron microscopy and can be observed under phase-contrast microscopy as refractile bodies. Most eukaryotic proteins form inclusion bodies when overexpressed in E. coli [l]. Cytoplasmic inclusions are not, however, unique to recombinant proteins. Many inclusions have been observed naturally in various procaryotic strains, including polyhydroxyalkanoate (PHA), polyglucoside, and polyphosphate. While this chapter focuses on the harvesting of recombinant protein inclusion bodies, the recovery techniques discussed here are generally applicable to other inclusions such as PHA. In this section we examine the factors influencing inclusion body formation before discussing the composition of inclusion bodies and their size and density. This pro-
82
4 Harvesting Recombinant Protein Inclusion Bodies
vides the framework necessary for devising optimal strategies for inclusion body recovery.
4.2.1 Inclusion Body Formation The likelihood of an inclusion body forming for a given protein expressed in E. coli is dependent on many factors, including the nature of the protein, the environmental conditions, and the host strain. However, no common characteristic that causes an inclusion body to form has been identified [2]. Wilkinson and Harrison [3] examined data for 8 1 proteins at 37 "C in E. coli that do and do not form inclusion bodies. They found that charge average and turn-forming residue fraction were correlated with the tendency to form inclusion bodies. Cysteine fraction, proline fraction, hydrophilicity, and the total number of residues were weakly correlated. The results were interpreted in terms of the inclusion body formation model proposed by Mitraki and King [4], shown in Eqn 1, translation
+ If tf Im + native If tf If* + IB
where I is an intermediate species (partially folded, f, that is capable of forming an aggregated inclusion body, or monomer-forming, m). They argued that a protein with many turn-forming residues would fold more slowly, hence giving a higher concentration of partially folded intermediate and an increased probability of inclusion body formation. It should be noted, however, that the study did not take into account other important factors which can have a major impact on IB formation, such as host-cell characteristics. Recently it has become clear that the in vivo folding pathway greatly influences the likelihood of inclusion body formation. For example, overproduction of folding chaperones such as GroEL and DnaK leads to a reduced propensity for inclusion body formation [5]. It is therefore clear that protein characteristics are only one determinant of inclusion body formation, and arguably a minor one. Differences in host-strain physiology, due to differences in chaperone and protease expression, will have a major impact.
4.2.2 Inclusion Body Composition The recombinant protein of interest typically comprises > 50 % of the inclusion body [2]. This is the key benefit of expression as an inclusion body. Relatively pure product can be simply obtained by physical fractionation, for example by filtration or centrifugation followed by dissolution and simple ion exchange. Such simple purification can easily give a product with > 90 % purity, compared with 1 % to 25 % purity after initial isolation for a soluble protein [2]. Of course, this advantage may be completely lost if protein refolding is inefficient (see Section 4.4).
4.2 What is an Inclusion Body?
83
Isolated IBs typically contain the protein of interest as well as various contaminants, including non-product polypeptides, nucleic acids, and cell-envelope components (e.g., outer-membrane proteins OmpA and OmpC/F). Hart et al. [6] examined Vitreoscilla hemoglobin (VHb) production in E. coli, and identified two cytoplasmic aggregates of different morphology. The granules differed in the relative fractions of VHb and pre-P-lactamase, the antibiotic-resistance protein encoded on the same plasmid. The inclusion bodies were also contaminated with the cytoplasmic elongation factor Tu, and by outer-membrane proteins OmpA and OmpF that were attributed to cell-envelope contamination following disruption. Valax and Georgiou [7] argue that many contaminants believed to be incorporated into IBs may, in fact, adhere to the IB following cell disruption. They found that the expression system and growth conditions have a pronounced effect on inclusion body composition for (3-lactamase. Inclusion bodies were purified to apparent homogeneity by differential centrifugation in sucrose gradients, maintaining the integrity and surface characteristics of the IBs. (3-lactamase lacking the native signal sequence (i.e., A(-20-1)-P-lactamase) formed cytoplasmic IBs that were virtually free of contaminating protein following sucrose-gradient purification. This indicates that IB formation in vivo can be a highly specific process. Periplasmic inclusion bodies had a consistently higher level of associated contaminants than cytoplasmic IBs, possibly due to different surface properties leading to enhanced adsorption. This belief was supported by the selective removal of contaminants through detergent washing. Overall, the level of contaminating polypeptides ranged from 5 % to 50 %, while phospholipids constituted 0.5 % to 13 % of the IBs. Nucleic acids were a minor contaminant. The study by Valax and Georgiou [7] concluded that most contaminants were not incorporated into the inclusion body during synthesis, but rather adsorbed to the IB following cell disruption. This suggests considerable scope for process optimisation by careful inclusion body washing prior to subsequent processing (see Section 4.4).
4.2.3 Size and Density of Inclusion Bodies Taylor et al. [8] were the first to examine the physical characteristics of inclusion bodies in any detail. They employed a combination of centrifugal disc photosedimentation (CDS) and electrical sensing zone measurement (ESZ). The mean diameters of y-interferon and calf prochymosin inclusion bodies were 0.81 and 1.28 pm, respectively. Sedimentation studies showed that the density of the IBs increased with the density of the suspending fluid, indicating an accessible voidage within the granules. In deionized water, the IBs had buoyant densities of 1124 kg m-3 for yinterferon and 1034 kg mP3 for prochymosin. Voidages of 70% for y-interferon and 85 % for prochymosin were determined by matching data from the two sizing methods. Olbrich [9] determined the size of prochymosin inclusion bodies to be 0.85 pm using ESZ. By matching ESZ and CDS data, an apparent buoyant IB density of 1140-1 160 kg m-3 was obtained after separating contaminant debris by low-speed
84
4 Harvesting Recombinant Protein Inclusion Bodies
centrifugation. In later experiments, Jin [ 101 measured prochymosin inclusion bodies by PCS and CDS and determined median diameters of 0.98 pm and 0.94 pm, respectively. Samples were again fractionated prior to analysis using low-speed centrifugation. The IB density determined by Olbrich [9] was used for CDS measurements. The physical characteristics of porcine somatotropin (pST) inclusion bodies have also been examined by CDS. A median IB size of 0.35-0.45 pm was obtained by high-density, fed-batch fermentation using minimal media [ 11,121. The reported size was based on an IB density of 1260 kg m-3, determined by cesium chloride gradients (Bresatec Ltd., Adelaide, Australia, personal communication), and compared well with estimates by electron microscopy. Unlike the inclusion bodies examined by Taylor et al. [8], the pST inclusion bodies were highly packed granules with a density approaching that of a crystalline protein precipitate. A range of insulin-like growth factor analogs expressed in E. coli has also been examined by CDS [13]. Typical mean inclusion body sizes were 0.3-0.5 pm, based on a density of 1260 kg m-3. Given the large variation in reported IB size, it is clear that independent determinations must be made for any specific protein. Also, it is often easier to report size as an apparent Stokes settling velocity rather than diameter, particularly if the aim is to recover centrifugally the IBs (see Section 4.6). This avoids the need to directly determine IB density, which can be difficult if the IBs have an accessible voidage.
-
4.3 Properties of Cellular Debris The properties of inclusion bodies were discussed in Section 4.2. It is clear that many contaminants adhere to the IB surface, and hence will be collected with the IB unless washing protocols are employed (Section 4.4). However, many more contaminants remain independent of the inclusion bodies in the homogenate following cell disruption. These include soluble components (e.g., nucleic acids, soluble cell proteins, lipid membrane vesicles) and insoluble fragments of the cell wall (e.g., peptidoglycan and associated cell-wall proteins and lipids). The soluble fragments may be readily separated by filtration or centrifugation (see Sections 4.5 and 4.6). However, fractionation of the insoluble IBs from the insoluble debris can be considerably more difficult, particularly if their physical sizes or settling velocities are similar. This fractionation is aided by an understanding of the size of the insoluble cell debris. A range of sizing techniques is available for characterizing cellular debris, including photon correlation spectroscopy (PCS), CDS, and ESZ. Each has significant disadvantages for debris sizing. PCS is a low-resolution technique that requires extensive sample preparation prior to size analysis. For example, Olbrich [9] centrifuged homogenate samples at 2000 g for 26 minutes, and took the supernatant as the cell debris sample and the pellet as the inclusion body sample. This is necessary as PCS cannot resolve the inclusion body and debris size distributions. A problem with this approach is preferential removal of large debris from the sample being analysed. Jin [ 101 demonstrated that up to 47 % of the cell debris co-sedimented with the inclusion bodies using Olbrich’s [9] fractionation scheme, while 14 % of inclusion body pro-
4.4 Process Synthesis
85
teins remained in the supernatant (i.e., the debris sample). It is clear that the apparent size distribution is affected by the pretreatment employed. Other methods also have disadvantages. ESZ is prone to orifice blocking, limiting the size range that this technique can analyze and generally making it unsuitable for sizing E. coli cell debris. CDS has low sensitivity below 0.2 pm [ 111 where much of the cell debris is located, and is prone to baseline drift and errors in extinction coefficient [ l l ] . Results from electron microscopy are also prone to error because of sample preparation techniques, including drying and plating, prior to analysis [ 141. A newly developed method for cell-debris size analysis [ 151 overcomes the limitations of traditional methods. It employs cumulative sedimentation analysis (CSA) in a laboratory centrifuge coupled with SDS-PAGE quantitation of sedimentable outermembrane proteins. The technique provides an accurate assessment of debris size even in the presence of inclusion bodies, without the need for sample pretreatment. However, it is rather laborious for routine quality control, and is more amenable to process optimization during developmental research. Key results from E. coli cell debris-sizing studies are presented in Chapter 6, and in particular Section 6.9 and Table 6-4. CSA indicates a median E. coli debris size of 0.5 pm after two homogenizer passes at 55 MPa, decreasing to 0.33 pm after 10 homogenizer passes [ 151. Models for the effect of repeated homogenization on debris size are presented in Section 6.9.2, and clearly show that debris size reduces with repeated homogenization. This dependence suggests that there will be a strong interaction between the selected disruption protocol and the ease of IB and debris fractionation. This is discussed further in Section 4.6.
4.4 Process Synthesis 4.4.1 Laboratory-scale Processes Figure 4-1 shows a summary of strategies employed to obtain active purified protein from inclusion bodies at laboratory scale. The key steps are cell breakage, IB sedimentation, and washing and solubilization [16]. The sedimentation step is often conducted at moderate centrifugal force in a laboratory centrifuge (e.g., 5 min at 10000 g ) [16] to effect some differential separation of the inclusion bodies and insoluble cell debris. Following protein renaturation (i.e., refolding), active product may be recovered by conventional methods such as chromatography. There are many variations to this standard approach, most of which are only minor deviations. For example, an additional cell debris-removal step is often added after solubilization to minimize the carry-over of debris to the chromatographic recovery system. At laboratory scale this usually consists of a high-speed centrifugation or depth-filtration step. Other variations are quite significant, and can include in situ solubilization of the inclusion bodies without prior cell disruption. These major variations are discussed further in Section 4.7.
86
4 Harvesting Recombinant Protein Inclusion Bodies E colt containing inclusion bodies
I T
Inclusion body release
I T
Differential centrifugation
=t
Cell Debris
/I T
Inclusion body washing
I T
Solubilise inclusion bodies and reduce disulphide bonds
II T
Purify
I/ I
Refold protein by
/I
v
Dilution
/I
I/
T
I
Dialysis
Diafiltration
/I I
Further purification
II v
Refolded meombinant protein
Fig. 4-1. Strategies for processing inclusion bodies at laboratory scale. (From Chaudhun [17]; reproduced with permission of the Annals of the New York Academy of Sciences.)
The area of protein refolding is a field of very active research. Some proteins are particularly difficult to refold with low consequent yields. An excellent review covering methods of dissolution and renaturation is available [181.
4.4.2 Considerations in Synthesizing a Large-scale Process Laboratory-scale processes for handling inclusion bodies may often be uneconomic if scaled directly. It is preferable to consider the scale of the final process, and then develop a laboratory process that is essentially a scaled-down version of the final anticipated design. This requires a knowledge of the key issues in synthesizing a large-scale process. Two key considerations are the efficiency of large-scale protein refolding, and problems associated with the collection of inclusion bodies.
4.4 Process Synthesis
87
4.4.2.1 Protein Refolding Poor refolding performance can destroy the economic viability of a given process, as demonstrated for tissue plasminogen activator (tPA) [19]. Two key factors must be considered in developing a laboratory-scale refolding process amenable to scaleup, namely the effect of protein refolding concentration, and the selected reactor operation mode. The choice of protein refolding concentration is perhaps the most critical. A simplified generic protein refolding scheme is give by Eqns (2) and (3),
21 -+ A
(3)
where D is the denatured protein obtained by protein solubilization, I is an aggregating intermediate, N is native protein, and A is a non-native aggregate. Eqn (2) is typically first order, while Eqn (3) is approximately second order. Clearly, with competing first- and second-order equations, the yield of N will be maximized by minimizing the protein refolding concentration (i.e., the concentration of D and hence I). Refolding is therefore normally conducted at extremely low protein concentrations in the laboratory. For process-scale work it may be desirable to increase the refolding concentration. For example, Kiefhaber et al. [20] showed that an increase M in the refolding concentration of porcine muscle lactic dehydrogenase from to lo-' pM led to a decrease in protein yield of only 25 %. At process scale this may represent an acceptable economic trade-off. For other proteins, an increase in concentration will decrease the yield to unacceptable levels. For example, tPA refolding at less than 2.5 mg L-' was necessary to achieve a reasonable renaturation yield of at least 25 % [19].This concentration was still too low for an economic process, as 75 5% of the process capital cost was associated with inordinately large refolding tanks. Clearly, the effect of protein concentration must therefore be examined at an early stage in process development. The choice of refolding reactor mode can also impact on process economics. Eqns (2) and (3) represent classical first- and second-order competition. The best conventional reactor mode for this scheme will be one that maintains a low concentration of reactant, which is either a continuous system or a fed-batch system with gradual addition of denatured protein [21]. Batch refolding will be the least successful, but continues to be extensively employed in laboratory studies. Cost-optimal reactor operating strategies have been developed for continuous protein refolding [21]. Recently, several novel methods have been proposed for increasing refolding yield. Vicik and DeBernadez-Clark [22] conducted a mathematical optimization of human carbonic anhydrase B refolding using diafiltration to reduce the denaturant concentration. The optimal diafiltration protocol and final urea concentration were found to be 0.088 min-' and < 0.4 M, respectively. Low diafiltration rates were required for high yields. Aggregation has also been reduced by conducting refolding in a gel-filtration column [23], although the cost of resin may preclude use at large scale. Reverse micelles have also been investigated [24], as they have the
88
4 Harvestina Recombinant Protein Inclusion Bodies
+
m
=,
Inclusion body
Solubilised protein
1-1 I
I
Refolded Drotein
Fig. 4-2. The refolding of proteins in reverse micelles, showing the protective isolation of the protein molecule by the micelle that leads to reduced protein aggregation. (From Chaudhuri [17]; reproduced with permission of the Annals of the New York Academy of Sciences.)
potential to isolate individual protein molecules, thus preventing the second-order aggregation reaction as shown in Fig. 4-2 [17]. Complete renaturation of pure, denatured ribonuclease-A has been obtained using this method [24]. General applicability to other proteins must still be demonstrated. 4.4.2.2 Inclusion Body Recovery The recovery of inclusion bodies inevitably leads to the co-recovery of some contaminants. These may be contaminants incorporated into the inclusion body during synthesis or through adherence to the IB surface during cell release (see Section 4.2.2), or may be insoluble cell debris collected with the inclusion bodies during their recovery by centrifugation or filtration. There is some evidence that contaminants associated with 0-lactamase inclusion bodies may reduce protein refolding yields [25]. The yield of active protein obtained by refolding from IBs was only 20-40% of that obtained from purified protein under the same conditions. Furthermore, virtually all contaminating protein present initially in the IBs appeared in aggregates that formed following removal of the denaturant. This suggests that the contaminating proteins somehow tended to promote aggregation. Contaminants may also affect refolding through proteolysis of the product. Babbitt et al. [26] have shown that proteases significantly reduced creatine kinase yield during the solubilization and refolding of IBs expressed in E. coli. The protease activity was presumably associated with cellular debris that co-sedimented with the IBs during batch centrifugation. Phenylmethanesulfonyl fluoride (a protease inhibitor) had no apparent affect on yield. However, significant yield improvements (ca. 100 X ) could be obtained by washing the IBs with the nonionic detergent octylglucoside prior to solubilization. Insoluble debris collected dur-
4.4 Process Synthesis
89
ing IB recovery may also foul chromatography columns downstream of the refolding operation. This will reduce the effective life of chromatographic resins and consequently increase process cost, and will also complicate process validation. Such debr i s may be easily removed by filtration or ultracentrifugation at laboratory scale following solubilization. Only depth filtration and ultrafiltration are practical at process scale, although both necessitate the inclusion of an additional process unit with its associated capital, operating, and validation costs. The three problems identified above, namely reduced refolding yields, proteolysis, and the fouling of chromatographic resins, suggests that inclusion body collection warrants close examination during process development. As demonstrated by Babbitt et al. [26], contaminants may be reduced by selective washing of the IBs. A variety of wash chemical are available [18]. The non-ionic detergent Triton X-100 (0.1 % to 4 % ) and low concentrations of denaturant (e.g., 2 M urea) are common choices. However, the use of wash chemicals may also cause problems at process scale. Washing with urea can significantly increase waste disposal costs because of its high biological oxygen demand. Detergents should be chosen with reference to their cost and ease of disposal. Triton X-100 has been employed in the industrialscale manufacture of porcine somatotropin (pST) [27] and insulin [ 2 8 ] . In the case of pST, Triton X-100 costs represented $2 M per annum, or 33 % of the total annual process chemicals and consumables cost. The cost for insulin manufacture was considerably less [28]. Other effects of adding wash chemicals during inclusion body collection also need to be considered. Removal of the added chemical will be necessary for validation purposes, and in some cases the chemical may have detrimental effects on downstream units. For example, incomplete removal of added detergent may lead to reduced membrane and chromatographic resin life. This is particularly problematic if the process is not robust enough to handle process disturbances that may lead to variable removal of the added chemical, again complicating process validation. An obvious but difficult method of reducing contaminant loads is to improve the removal of insoluble debris during the IB recovery stage. This is achieved at laboratory scale by low-speed centrifugation that sediments the majority of the dense IBs, while leaving the majority of debris in the supernatant. At process scale this benefit is often forgotten, and the aim is often to maximize the collection of inclusion bodies, and hence debris, without reference to the downstream problems. This represents the perceived difficulty of good fractionation at process scale. For example, Krueger et al. [I61 state that ‘differential centrifugation may be difficult or impossible on the process or commercial scale’. Wash chemicals are therefore extensively employed. However, recent biochemical engineering progress has been made in differential centrifugation, as discussed in Section 4.6. For some proteins, refolding yields are neither substantially affected by contaminants nor proteolysis. In such cases, filtration may be considered for IB collection, particularly if the capital cost of a centrifuge is considered prohibitive and an additional filtration module is available after solubilization to reduce insoluble debris contaminants. The collection of inclusion bodies by filtration is therefore considered further in the next section.
90
4 Harvesting Recombinant Protein Inclusion Bodies
4.5 Filtration Harvesting of inclusion bodies following fermentation and homogenization is traditionally conducted by centrifugation. This can be a costly exercise when scale-up is attempted. Pressure-driven membrane processes such as ultrafiltration (UF) and microfiltration (MF) have been demonstrated to be attractive processes for cell harvesting [29,30] and the recovery of enzymes [31,32]. Such membrane processes may also be applied to the harvesting of inclusion bodies [33,34], where the presence of cellular contaminants during refolding does not affect yield.
4.5.1 Modes of Filtration Two different modes of filtration are possible, namely surface filtration and depth filtration. Filtration for the recovery of inclusion bodies is exclusively by surface filtration at a membrane. The highly fouling nature of cell debris, particularly cell membrane-based components, means that cross-flow membrane filtration is the sole practical mode. Cross-flow filtration refers to a type of surface filtration wherein the main direction of the suspension flow is perpendicular to the flow direction of the recovered liquid. A wide variety of industrial processes may be classified as cross-flow filtrations. Examples include reverse osmosis, ultrafiltration, and microfiltration. However, an often accepted convention is that cross-flow filtration is normally considered as a process removing particles whose sizes range from 0.1 to 10 pm. This size range clearly covers the size of typical inclusion bodies and thus it is a potential technology for the fractionation of inclusion bodies from soluble proteins as an initial step in the purification of insoluble cellular proteins. In cross-flow filtration, the suspension flows under the applied pressure gradient along the porous membrane and liquid permeation occurs across the membrane. Consequently, as the liquid permeates the membrane, a fraction of the particles suspended in the feed will be deposited at the surface of the membrane and a solid cake will form. The thickness of this cake will increase with time as the processing proceeds. This is accompanied by a resultant decline in the rate of liquid permeation with time. Such time-dependent behaviour is characteristic of cross-flow filtration. The prime advantage of operation in the cross-flow mode is the rnininization of cake formation as a consequence of the scouring action of the flow. To ensure that the scouring action is effective, the flow conditions should be maintained in the turbulent regime. This may lead to large pressure losses around the retentate loop and consequent high energy costs. Occasionally, it may be necessary to cool the retentate. Membranes are normally cleaned in situ by pulsed backwashing. Pulses are usually of short duration and the filtered product is forced back through the membrane to detach the filter cake from the surface of the membrane. The backwashed liquor is simply mixed into the retentate flow. The flow’s scouring action ensures that the solids are resuspended. Chemical cleaning may be required to reverse the
4.5 Filtration
91
effects of long-term flux decline due to adsorption or precipitation leading to gross fouling or pore blockage. There are few available data on the extent of such fouling in the IB harvesting area. Inclusion body washing or diafiltration is readily accomplished by the addition of water and appropriate buffers. This enables the removal of low-molecular weight contaminating solutes.
4.5.2 Theory In cross-flow microfiltration or ultrafiltration, the filtration flux will depend primarily upon the transmembrane pressure, the bulk concentration of inclusion bodies, the gel layer concentration, and the mass transfer coefficient. The transmembrane pressure is readily defined as an average pressure drop across the filter according to Eqn (4).
Permeate flux normally declines over operating time as a consequence of the resistance produced by concentration polarization, cake formation, and other causes of membrane fouling. As stated earlier, such a decline is partially reversible by backflushing and/or pulsing of the membrane or in extreme cases by chemical cleaning. Clearly, cross-flow mode will limit the build-up of inclusion body particles and cellular debris on the membrane surface, and at steady state the particle layer is assumed to attain a thickness that is time-invariant but increases with distance from the filter entrance. The flow resistances of the cake and the flowing (polarized) layer of particles plus the intrinsic resistance of the membrane may be assumed to act in series. Hence, the permeate flux may be described by Darcy’s law. The transmembrane flux J is then,
where R , is the membrane resistance, R, is the gel layer resistance, and AP is the applied pressure drop. At high solute concentrations, solute precipitates at the membrane’s surface forming a thixotropic gel which is permeable to the solvent. If the concentration of the solute is moderate to high, the resistance of this gel layer will increasingly dominate the resistance contribution from the membrane. The permeate flux then becomes independent of the membrane’s permeability. As the transmembrane pressure rises, the resistance of the gel layer R, will increase as solid accumulates at the membrane/liquid interface. This resistance will continue rising until an equilibrium is attained. At this point, nett transport of the solute towards the membrane is balanced by back diffusion of the solute towards the solution bulk as a consequence of the applied pressure gradient. The thickness of the gel layer is thus determined by the convective and diffusive transport mechanisms controlling the concentration of particles near the membrane surface. Applying an elementary
92
4 Harvesting Recombinant Protein Inclusion Bodies
mass balance to the particle layer and assuming that steady state conditions prevail yields Eqn (6),
a ax
-(uC)
a + -((vC)
=
aY
where C is the concentration of the solute, u and v represent the axial and radial velocity components of the fluid, and D is the diffusivity of the solute. The diffusivity includes contributions from Brownian motion and diffusion induced by shear. The standard approach to simplifying this balance is to neglect axial convection [35,36]. Integrating the resultant reduced form of Eqn (6) yields the following ‘macroscopic’ mass balance: dC JC+D=O dY
(7)
The assumption of complete rejection at the membrane is implicit in this equation’s derivation. It may be integrated to suggest that the limiting value of the flux will decrease logarithmically with increasing concentration in the solution bulk,
where the subscripts b, c, p, and m denote the bulk, cake, permeate, and membrane, respectively. The mass transfer coefficient k is primarily a function of the flow geometry of the membrane, fluid properties, and temperature.
4.5.3 Commercial Equipment and Operating Parameters There are three types of commercial membrane configurations, namely capillary membranes, flat sheet modules, and spiral modules. In the first category, retentate flows inside tubes grouped into bundles and enclosed in cylindrical modules. These provide a large surface area in a moderately sized module, but incur high energy costs to maintain turbulent flow. Capillary types (< 2 mm) are only suitable for low concentration, non-fouling solids. Tubular membranes (2-25 mm diameter) have lower pumping costs and are suitable for high suspended solids loads (to about 20 % dry weight), including solids which may blind such as fermentation broths. Flat-sheet modules resemble plate and frame filters and the retentate and permeate are separated by sheets of the filter membrane. Typical spacings are 1-2 mm and turbulence is promoted by embossed dimples and screens. The narrow passages and presence of screens normally limits suspended solids to a maximum of approximately 10 %. Spiral modules are essentially a wound sandwich of flat-sheet-type membranes. The operating constraints are similar to those of flat sheets.
4.5 Filtration
93
The cross-flow concept is applicable over a wide spectrum of operations including ultrafiltration, microfiltration, and reverse osmosis. Cross-flow filtration is normally characterized by a particle cut-off size of 0.1 pm and greater. Membrane pore sizes are commonly 0.2, 0.45 and 1 pm with typical average filtration rates for cell harvesting and washing in the range 50-200 L mP2 h-'. In general, flux rates are below 100 L m-2 h-l, compared with a desired flux rate of approximately 150 L mP2 h-' for economic operation [37].
4.5.4 Inclusion Body Recovery by Filtration Two studies of cross-flow filtration for the separation of inclusion bodies from soluble proteins in recombinant E. coli have been reported. The initial study was undertaken by Forman et al. [33]. A recombinant strain of E. coli containing a gene encoding a portion of gp41, the transmembrane protein of the AIDS (HTLV-111) virus, was used. Their study examined the effects of key operating variables on soluble protein removal, namely cross-flow rate, transmembrane pressure, initial concentration, and ionic environment. Hydrophilic DuraporeTM membranes with mean pore sizes of 0.22 and 0.45 pm provided adequate retention for a feed material containing 5 % solids. The relationship between flux and transmembrane pressure corresponded qualitatively to predictions from the gel-polarization theory. A maximum flux of 8 L m-2 h-' was achieved at feed concentrations of 25 and 50 g L-*. As expected the removal of soluble proteins was enhanced at low ionic strengths as such conditions will minimize the tendency for protein aggregation. By contrast, increasing the cross-flow rate did not yield the predicted enchancement of solute passage through the membrane as a consequence of the scouring action of the increased tangential flow. A diafiltration experiment was also performed. A removal of 87% of the soluble protein was reported (flux = 1.7 L m-2 h-I at a constant average transmembrane pressure of 0.5 kPa). Operation in this mode provided an efficient method for continuously performing washing and concentration steps. Meagher et al. [34] developed a cross-flow filtration process for the purification of rIL-2 inclusion bodies from homogenized E. coli. Key elements of the process were a two-step diafiltration, an extraction using 7 M GuHC1, followed by a dilution of the solubilized inclusion bodies into an appropriate buffer. The process provided a threefold increase in the yield of rIL-2 product in the diluted extract when compared with an alternative centrifugation strategy. An additional benefit of the membrane process was the significant improvement in product purity quantified by high-performance liquid chromatography (HPLC). The second diafiltration step using 1.75 M GuHCl appears to have solubilized a significant proportion of the contaminants, allowing their easy removal. This improvement in purity may provide significant benefits and simplifications in subsequent processing steps such as refolding. Clearly, interest in membrane fractionation of IBs is in its infancy. However, crossflow filtration does offer significant potential advantages when compared with centrifugation for IB recovery from cell lysate. First, the energy requirements are significantly reduced and initial capital investment may be lower. Second, the ease of
94
4 Harvesting Recombinant Protein Inclusion Bodies
operation in the diafiltration mode simplifies washing and soluble-protein removal steps when compared with the repeated resuspension and sedimentation steps for centrifugation. However, significant improvements in the flux rates and percent transmissions to minimize buffer and membrane capital requirements may be necessary before adoption at commercial scales. Also, the working life of typical membranes is poorly quantified.
4.6 Centrifugation 4.6.1 Modes of Centrifugation A range of centrifuge designs is available. The most common are the chamber or multichamber bowl centrifuge, the decanter centrifuge, and the disc-stack centrifuge. Each is best suited to a particular application. For example, bowl centrifuges are best suited to streams with a low solids content, while decanter designs are capable of handling high solids throughputs, and are commonly employed in sludge-dewatering applications. Of the available designs, the disc-stack centrifuge is best suited to the recovery of protein inclusion bodies in bioprocessing. Figure 4-3 shows a cross-section of a disc-stack centrifuge. Material is fed through the center periphery to the outer disc edge. It then flows between the discs which
1 Feed 2 Discharge 3 Photocell 4 Discs 5 Sediment holding space 6 Solids ejection ports 7 Operating-water valve 8 Drain hole 9 Opening chamber 10 Closing chamber 11 Annular piston 12 Timing unit 13 Discharge pump
Fig. 4-3. Cross-section of a disc-stack centrifuge. (Reproduced from the Westfalia Separator booklet: Centrifugal Clarifiers and Decanters for Biotechnology with the kind permission of the GEA Process Technology Division and Westfalia Separator Australia Pty Ltd.)
4.6 Centrifugation
95
rotate at high velocity (ca. 10000 r.p.m.). Solids are collected on the bottom surface of each disc due to centrifugal force, before being flung outwards to the bowl periphery where a sludge is formed. The sludge may be collected from this region by opening the bowl. Clarified liquid flows from the top of the discs, forming a clarified supernatant stream. Several designs are available for solids discharge. Split-bowl designs periodically separate the upper and lower halves of the bowl, permitting solids discharge under extreme pressure. Other designs employ nozzles at the bowl periphery that permit continuous or periodic solids discharge. Regardless, essentially continuous solids discharge is possible for extended operational periods, enabling continuous operation without the need for centrifuge disassembly.
4.6.2 Commercial Centrifugation Equipment A variety of commercial centrifuges are available. The field is, however, dominated by two main manufacturers of disc-stack centrifuges, namely Westfalia Separator AG (Oelde, Germany) and Alfa Laval Separation AB (Tumba, Sweden). Both manufacturers offer contained, sterilizable disc-stack centrifuges suitable for inclusion body recovery, in a range of sizes from pilot scale (e.g., the CSA8 from Westfalia and the BTPX 205 from Alfa Laval) to production scale. Figures 4 - 4 and 4-5 provide
Fig. 4-4.Westfalia SC35 disc centrifuges. (Reproduced from the Westfalia Separator booklet: Centrifugal Clarifiers and Decanters for Biotechnology with the kind permission of the GEA Process Technology Division and Westfalia Separator Australia Pty Ltd.)
96
4 Harvesting Recombinant Protein Inclusion Bodies
Fig. 4-5. The Alfa-Lava1 BTUX 510 disc-stack centrifuge. (Reproduced from the Alfa Laval booklet: BTUX 510 Contained Separation System for Commercial Biotech Production, with the kind permission of Alfa Laval Pty Ltd Australia.)
examples of production-scale centrifuges suitable for the recovery of protein inclusion bodies.
4.6.3 Theory The aim of modeling centrifuge performance is to predict what fraction of material of a specific settling velocity will be collected. This may be used, with information on the properties of inclusion bodies and cell debris, to optimize centrifuge performance in a given application. The simplest description of centrifuge performance is provided by the so-called Sigma model. The fraction of particles collected in the centrifuge is given by equation (9).
4.6 Centrifugation
97
where Q is the centrifuge feedrate, Z is the equivalent settling area of the centrifuge, vg is the particle settling velocity, d is particle diameter, A p is the density difference between the particle and fluid, p is fluid viscosity, and g is gravitational acceleration. The equivalent settling area, Z,in Eqn (9) is a machine-specific parameter depending on the design of the centrifuge. For disc-stack designs, Z is given by equation (101,
where w is the centrifuge angular velocity, N is the number of channels between discs, 0 is the inclined angle between the axis and the disc surface, and r is the inner (2) or outer (1) radius of the discs. Empirical corrections to the exponents have been applied to Eqn (10) to take into account flow non-idealities. Sullivan and Erikson [38] defined the ‘KQ’ empirical correction by substituting w1.5 and 9,75 into equation (10) in place of w2 and r3. Equation (9) is based on several assumptions regarding centrifuge behavior. The most notable are that flow between discs is ideal and that particles are collected when they reach the centrifuge disc. The assumptions are so restrictive that the predictive behavior of (9) is extremely limited. It invariably overpredicts the fractional collection efficiency, f(d). As a result, collection efficiency is often determined empirically for a given suspension and centrifuge design using Eqn (1 l),
where C, is the concentration of particles of size d entering the centrifuge, and CLis the concentration of particles of size d leaving in the centrifuge supernatant. A plot off(d) versus d is termed the centrifuge grade efficiency, and is a critical determinant of the ability to separate solids with similar settling characteristics such as debris and inclusion bodies. Eqn (9) predicts that the grade efficiency curve will be given by Eqn (12),
with s = 2, where d, is the critical diameter for the given centrifuge (Eqn (13)). d, =
,/%
The critical diameter is the minimum sized particle that will be fully collected in the centrifuge (i.e., f(dc) = 1). Again, this equation will often provide a poor prediction of centrifuge performance as it is based on the assumptions underlying Eqn (9). Empirical grade-efficiency curves are therefore widely used. The simplest allows the
98
4 Harvesting Recombinant Protein Inclusion Bodies
exponent s in Eqn (12) to be determined by regression to experimental data. Alternatively, Eqn (14) has been widely employed to model centrifuge grade efficiency,
where k and n are parameters determined by regression to experimental data.
4.6.4 Scale-up and Scale-down of Centrifuges Several researchers have presented similarity criteria for centrifuge scaling, but none has found widespread use. In some respects this is not surprising considering the hydrodynamic complexity involved.
grade efficiency 1
0.8
0.6
0.4
0.2
0 0
0.5
1
1.5
2
2.5
dimensionless diameter d/dc Fig. 4-6. Grade efficiency curves for a Westfalia BSB-7 disc-stack with a reduced sedimentation and a full set of discs, A,determined by Mannweiler [39]. The Stokes line corresponds to area, 0, equation (9) (From Keshavarz-Moore et al. [41]; reproduced with permission of the Annals ofthe New York Academy of Sciences.)
4.6 Centrifunation
99
Generally, centrifuge scale-up is done using manufacturer's guidelines after determining performance on a pilot-scale machine. Direct scaling according to the equivalent settling area of the centrifuge is possible, provided that a sufficiently large machine was used for pilot-scale tests. Alfa-Lava1 typically scale results using the KQ correlation, while Westfalia applies the so-called Leistung, or power, factor (another effective area measure). Scale-up will generally be conservative because of the way in which scale-up factors are defined by the manufacturer. Considerable variation between machines with the same nominal equivalent area is also possible through minor variations in caulk thickness (i.e., the distance between parallel discs). A detailed experimental study of the scale-down of a Westfalia BSB-7 centrifuge has been undertaken using polyvinylacetate emulsion [39,40]. Grade-efficiency curves were determined experimentally and could be well described by Eqn (14). Significant departure from ideality was observed, as shown in Fig. 4-6. Mannweiler and Hoare [40] concluded that the Westfalia BSB-7 centrifuge may be scaled down to 10 % of its total available separation area by removing active discs. If the centrifuge is used to collect the majority of particles, as is normally the case for inclusionbody processing, then accurate prediction of full-scale throughput capacity for the production machine was possible. Such information is extremely important for bioprocessing applications, where there is often limited material available in initial pilot-scale testing.
4.6.5 Inclusion Body Recovery by Centrifugation Several studies into the collection of prochymosin inclusion bodies have been conducted. The grade-efficiency curves generated using PVA (Section 4.6.4) have been used to simulate the fractionation of prochymosin inclusion bodies and cell debris in a Westfalia SB-7 centrifuge [9]. High inclusion body collection was predicted to result in poor purity (i.e., a high collection of cell debris), while high centrifuge feedrates gave good paste purity but a reduced fractional inclusion body collection (Fig. 4 -7). The results were not confirmed experimentally. Experimental trials into the separation of prochymosin IBs and cellular debris using a disc-stack centrifuge followed the simulation studies. Simulated overall collection efficiencies were compared with experimental data [ 101. Deviations were observed although general trends were correct. A method for on-line control of inclusion body recovery using turbidity was also developed [ 10,421. The ratio of two absorbance measurements (OD600nm/OD40"nm) was shown to be a good correlator of the amount of inclusion body material in the centrifuge supernatant. The robustness of the method to changes in inclusion body size and other feed properties has not been defined, and a separate empirical correlating equation will clearly be required for each different feed stream. Nevertheless, the simplicity of the method suggests considerable potential for the on-line control of large-scale processing. Numerous centrifuge simulation studies have also been undertaken using porcine somatotropin (pST) as a model system. These inclusion bodies are considerably smaller than the prochymosin IBs (Section 4.2.3), thus complicating differential
100
4 Harvesting Recombinant Protein Inclusion Bodies
100 7 removal of cell debris
80 -
60 -
inclusion bodies
40 -
20 -
-0
200
400
600
800
1000
flowrate (L/h) Fig. 4-7. Simulated recovery curves for prochymosin inclusion bodies and cell debris, based on the work of Mannweiler [39]. (From Keshavarz-Moore et al. [41]; reproduced with permission of the Annals of the New York Academy of Sciences.)
separation. Approximate IB and debris size distributions were determined for pST using differential sedimentation [43]. The important effect of homogenate viscosity was demonstrated using the ideal Sigma centrifuge model. An experimental study of the recovery of pST inclusion bodies in a Westfalia SB-7 centrifuge was then undertaken, allowing the fractional recovery of inclusion bodies to be defined for a range of feedrates [44-461. An empirical curve for collection efficiency was determined and used to simulate the classification performance. The simulation results suggested that overall process cost is determined to a large extent by debris size reduction in the homogenizer. Multiple homogenizer passes were shown to be beneficial because of reduced debris size, enabling improved IB purification. The results also suggested that a given inclusion body purity could be attained at lower cost using multiple centrifuge passes. This is simply a consequence of the centrifuge grade-efficiency characteristics. Considerable experimental work into the recovery of IGF inclusion bodies in a disc-stack centrifuge has recently been completed [ 13,151. The impact that centrifugation has on overall product yield for a proteolytically sensitive analog of insulinlike growth factor has been defined. Significant yield improvements were obtained by optimizing the centrifugation protocol. Specifically, multiple centrifuge passes without detergent or chaotrope washing significantly improved protein yield following solubilization. Under standard solubilization conditions, a twofold increase in recoverable protein was achieved by introducing two buffer washing steps, even though this increased the inclusion body loss from 25 % to 42 %. The increased
yield resulted from improved cell-debris removal with multiple centrifuge passes, as suggested by the simulation studies discussed above. Rational optimization of the fractionation process requires accurate debris and IB size distributions, and centrifuge grade-efficiency curves that are determined directly for the type of material being separated. Accurate debris size distributions have been determined using CSA (see Section 4.3). Wong [ 151 has also used CSA to determine experimental grade-efficiency curves for cell debris from E. coli. The curves for a Veronesi KLE-160 disc-stack centrifuge could be modeled using Eqn (14). By combining the experimentally determined data and experimentally verified models, an extensive simulation study into optimizing the fractionation of IGF inclusion bodies from cell debris was completed [15]. Figure 4-8 shows the effect of centrifuge feedrate on the collection of inclusion bodies and the removal of cell debris for a single centrifuge pass. The benefit of repeated homogenization is clearly demonstrated, as it reduces debris size thus facilitating fractionation. Figure 4-9 shows the simulation results for multiple centrifuge passes. The benefit of repeated homogenization and centrifugation for improved purity are again clear. It is worth mentioning that the study by Wong [15] is the first to employ experimentally determined grade-efficiency curves for cell debris. Furthermore, debris size distributions were determined using a sizing method that does not require pretreatment of the sample that can compromise the results, and models were validated using experimental data. Consequently, a high degree of confidence can be placed
0
1
2
3
4
5
6
Norrnalised Centrifuge Feedrate, QE (rns-' x
lo9)
Fig. 4-8. Simulated curves for the recovery of inclusion bodies and the removal of cell debris in a disc-stack centrifuge. IB, inclusion bodies; CD, cell debris; N, number of homogenizer passes. (From Wong [15]; reproduced with the kind permission of Heng-Ho Wong).
102
Gp:
-
4 Harvesting Recombinant Protein Inclusion Bodies
1.o
&
0
m
0.8
r
0
5
p: 0.6 &
0
2. a2 0.4 U
-2 m
.-*g
0.2
u a
ct
0.0
0
2
4
6
8
Number of Centrifuge Passes Fig. 4-9. Simulated curves for the recovery of inclusion bodies and the removal of cell debris in a disc-stack centrifuge. IB, inclusion bodies; CD, cell debris; N, number of homogenizer passes; RCI, ratio of cell debris to inclusion bodies, and hence a measure of purity, and is normalized relative to the ratio in the homogenate before centrifugation. (From Wong [15]; reproduced with the kind permission of Heng-Ho Wong).
in the simulation results, which conclusively demonstrate that it is possible to optimize centrifuge operation through careful manipulation of process conditions. It is worth commenting briefly on the application of chemical washes during centrifugal recovery of inclusion bodies (see Section 4.4.2). The preceding results clearly show that the purity of the inclusion body paste can be controlled by optimizing the process conditions, and in particular the centrifuge feedrate, the number of homogenizer passes, and the number of centrifuge passes. While washing with detergents or chaotropes may be necessary in some cases, their use should only be investigated after the optimal centrifugation procedure has been defined without their use. In this way, it may often be possible to minimize downstream impacts by careful optimization of centrifugation conditions, and it may be possible to minimize or in some cases eliminate the use of wash chemicals.
4.7 Alternatives to Inclusion Body Recovery Traditional methods of dealing with inclusion bodies described in previous sections are well established. However, they have largely evolved from laboratory-scale approaches based around inclusion body release followed by in vitro solubilization. Recently, new approaches with improved potential for scale-up have been developed.
4.7 Alternatives to Inclusion Body Recovery
103
Work has shown that certain inclusion bodies can be partially solubilized in low concentrations of denaturant. Hart and Bailey [47] examined the solubilization of Vitreoscilla hemoglobin inclusion bodies in urea. Significant solubilization was achieved with low concentrations (< 3 M urea), suggesting that some inclusion body protein had a partially folded conformation and that a fractional dissolution and refolding process may be advantageous. Chang and Swartz [48] have shown that periplasmic inclusion bodies of insulin-like growth factor (IGF-I) can be partially solubilized and refolded, following cell disruption, in 2-3 M urea with < 10 mM dithiothreitol. Greenwood et al. [49] have also shown that an analog of IGF-I can be recovered in refolded form following dissolution in low concentrations of urea (< 4 M). Relatively low overall yields were obtained in all studies. Nevertheless, the work suggests that it is not necessary to fully denature the protein structure from inclusion bodies using highly concentrated denaturant. It may therefore be possible to develop large-scale processes that do not require denaturant concentrations close to saturation (e.g., 8 M urea), and that have the dissolution and refolding stages integrated into one unit. In situ dissolution methods for recombinant inclusion bodies have received interest recently. Effectively, these take intact cells containing recombinant inclusion bodies and selectively treat them with chemicals to effect solubilization of the inclusion body. The advantages of this approach are clear. First, culture may be taken directly from the fermenter and treated to release the protein of interest in a soluble form. This deletes several operations from the traditional approach, namely cell disruption and inclusion body collection and washing. Hart et al. [SO] have used this approach to recover IGF-I with an endogenous secretory signal sequence from E. coli. The protein was distributed equally between the soluble phase and periplasmic inclusion bodies, so a traditional IB recovery strategy gave low product yields. Treatment of alkaline fermentation broth (pH 10) with a chaotrope (2 M urea) and reductant (10 mM dithiothreitol) gave - 90 % of all IGF-I in an isolated supernatant. A two-phase extraction procedure was developed to provide direct extraction of the solubilized IGF-I [SO,Sl]. Urea tended to inhibit twophase formation and residual solids sedimentation. Highest recovery was obtained with a two-phase systems composed of 5 % sodium sulfate and 14 % PEG-8000. The partition coefficient in this system was approximately 8, indicating significant potential for process-scale application. This approach of in situ solubilization has been extended to cytoplasmic inclusion bodies of an analog of insulin-like growth factor (Long-R3-IGF-I, R. J. Falconer, Dissertation in preparation, The University of Adelaide). Effective dissolution was obtained by treatment of intact cells with 6 M urea, 3 mM ethylenediaminetetra-acetate, and 20 mM dithiothreitol. The treatment gave comparable solubilized IGF to the traditional approach of mechanical cell disruption followed by in vitro solubilization. Low protein release was achieved below pH 9. High cell concentrations led to a significant increase in viscosity, possibly limiting application at large scale. The technique was also extended to give selective release of the IGF protein. Replacement of the dithiothreitol in the treatment mixture with 2-hydroxyethyldisulfide inhibited IB solubilization but still permeabilized the host cells and released soluble contaminant proteins from the cytoplasm. This presumably acted by cross-linking proteins
104
4 Harvesting Recombinant Protein Inclusion Bodies
on the IB surface through disulfide bond formation, thus inhibiting urea solubilization. The soluble contaminating proteins were washed away prior to solubilization using the original treatment buffer. The purity of extracted IGF by this method was 50 % of the total protein, constituting a purification factor of greater than 2.5. Total IGF recovery exceeded 80 %. Methods involving in situ dissolution are limited to proteins that are not significantly affected by proteases and contaminants in the refolding mixture. Further development of these selective chemical release methods seems warranted given their potential for industrial-scale application. Other novel methods will undoubtedly emerge as further proteins expressed as inclusion bodies reach commercialization.
Abbreviations and Symbols C D
%
concentration (kg m-3) diffusion coefficent (m2 s-l) particle diameter (m) critical particle diameter (m) fractional collection of particles of size d acceleration due to gravity (m s - ~ ) distance between centrifuge discs (i.e., caulk height) (m) flux (kg m-2 s-l) parameter in equation (8) grade-efficiency exponent in equation (14) pressure (Pa) flowrate (m3 s-l) resistance (m2 kg-') grade-efficiency exponent in equation (12) axial fluid velocity (m s-') radial fluid velocity (m s-') particle Stokes velocity under gravitational force (m s-l)
Subscripts g m P
gel layer membrane permeate
d dc
f(4 g
h J
k n P
Q
R S
U V
Greek symbol V
b CI
r. w
fluid kinematic viscosity (m2 s-') density difference between particle and fluid (kg m-3) fluid dynamic viscosity (Pa s-I) centrifuge equivalent settling area (m2) centrifuge-disc angular velocity (rad s-')
References
105
Abbreviations CDS centrifugal disc photosedimentation CSA cumulative sedimentation analysis ESZ electrical sensing zone measurement IB inclusion body IGF insulin-like growth factor PCS photon correlation spectroscopy VHb Vitreoscilla hemoglobin
References [ l ] Marston, F.A. O., Biochem J, 1986, 240, 1-12. (21 Kane, J.F., Hartley, D. L., Trends Biotechnol, 1988, 6, 95-101. [3] Wilkinson, D. L., Harrison, R. G., Biotechnology, 1991, 9, 443-448. [4] Mitraki, A., King, J . , Biotechnology, 1989, 7, 690-697. [5] Lee, S. C., Olins, P. O., J Biol Chem, 1992, 267, 2849-2852. [6] Hart, R. A,, Rinas, U., Bailey, J.E., J Biol Chem, 1990, 265, 12728-12733. [7] Valax, P., Georgiou, G., Biotechnol Prog, 1993, 9, 539-547. [8] Taylor, G., Hoare, M., Gray, D. R., Marston, F. A. O., Biotechnology, 1986, 4, 553-557. Olbrich, R., Dissertation, University of London, 1989. Jin, K., Dissertation, University of London, 1992. Middelberg, A. P. J., Bogle, I. D. L., Snoswell, M . A,, Biotechnol Prog, 1990, 6, 255-261. Middelberg, A. P. J., O’Neill, B. K., Bogle, I. D. L., Snoswell, M., Biotechnol Bioeng, 1991, 38, 363-370. Wong, H. H., O’Neill, B. K., Middelberg, A. P. J., Bioseparation, 1996, 6, 185-192. Bailey, S . M., Blum, P. H., Meagher, M . M., Biotechnol Prog, 1995, 11, 53-539. Wong, H. H., Dissertation, University of Adelaide, 1997. Krueger, J.K., Kulke, M.H., Stock, J., BioPharm, 1989 March, 40-45. Chaudhuri, J . B., in: Annals of the New York Academy of Sciences: Recombinant DNA Technology 11; Bajpai, R.K., Prokop, A. (Eds.), New York: New York Academy of Sciences, 1994; Vol. 721, pp. 374-385. Fischer, B., Sumner, I., Goodenough, P., Biotechnol Bioeng, 1993, 41, 3-13. Datar, R.V., Cartwright, T., Rosen, C.-G., Biotechnology, 1993, 11, 349-357. Kiefhaber, T., Rudolfph, R., Kohler, H.-H., Bucher, J., Biotechnology, 1991, 9, 825-829. Middelberg, A. P. J.,Chem Eng J, 1996, 61, 41-52. Vicik, S., DeBernadez-Clark, E., in: Refolding. ACS Symposium Series 470: Georgiou, G . , DeBernadez-Clark, E. (Eds.), Washington: American Chemical Society, 1991; Vol. 470, pp. 180-196. Batas, B., Chaudhuri, J. B., Biotechnol Bioeng, 1996, 50, 16-23. Hagen, A. J., Hatton, T. A., Wang, D. I. C., Biotechnol Bioeng, 1990, 35, 955-965. Valax, P., Georgiou, G., in: Biocatalyst design for stabiliv and specificity: Himmel, M . E., Georgiou, G. (Eds.), Washington: American Chemical Society, 1993; Vol. 516, pp. 126-139. Babbitt, P. C., West, B. L., Buechter, D. D., Kuntz, I. D., Kenyon, G. L., Biotechnology, 1990, 8, 945-949. [27] Petrides, D., Cooney, C. L., Evans, L. B., Field, R. P., Snoswell, M., Computers chem Engng, 1989, 13, 553-561. 1281 Petrides, D., Sapidou, E., Calandranis, .I. Biotechnol , Bioeng, 1995, 48, 529-541. [29] Dostalek, M., Haggstrom, M., Biotechnol Bieong, 1982, 24, 2077-2084. [30] Haarstrick, A,, Rau, U., Wagner, F., Bioproc Eng, 1991, 6, 179-184.
106
4 Harvesting Recombinant Protein Inclusion Bodies
[31] Kroner, K.H., Biotech Forum, 1986, 3, 20-21. [32] Quirk, A. V., Woodrow, J. R., Enzyme Microb Technol, 1984, 6, 201-206. [33] Forman, S. M., DeBernardez, E. R., Feldberg, R. S., Swartz, R. W., J Membr Sci, 1990, 48, 263 -279. [34] Meagher, M., Barlett, R. T., Rai, V. R., Khan, F. R., Biotechnol Bioeng, 1994, 43, 969-977. [35] Trettin, D. R., Doshi, M. R., Chem Engng Commun, 1980, 4, 507-522. [36] Romero, C. A., Davis, R. H., Chem Eng Sci, 1990, 45, 13-25. [37] Kroner, K. H., Schutte, H., Hustedt, H., Kula, M.-R., Proc Biochem, 1984, 4, 67-74. [38] Sullivan, F. E., Erikson, R. A,, Ind Eng Chem, 1961, 53, 434-438. [39] Mannweiler, K., Dissertation, University of London, 1989. [40] Mannweiler, K., Hoare, M., Bioproc Eng, 1992, 8, 19-25. [41] Keshavarz-Moore, E., Olbrich, R., Hoare, M., Dunnill, P., in: Annals ofthe New York Academy of Sciences: Recombinant DNA Technology I: Prokop, A,, Bajpai, R. K. (Eds.), New York: New York Academy of Sciences, 1991; Vol. 646, pp. 307-314. [42] Jin, K., Thomas, 0.R. T., Dunnill, P., Biotechnol Bioeng, 1994, 43, 455-460. [43] Middelberg, A. P. J., Bogle, I. D. L., Snoswell, M., in: Proceedings of Chemeca 89 (Broadbeach, Qld). Australian Chemical Engineering Conference. Canberra: Institution of Engineers Australia, 1989, pp. 671-678. [44] Middelberg, A. P. J., O’Neill, B. K., Aust J Biotech, 1991, 5, 87-92. [45] Middelberg, A. P. J., O’Neill, B. K., Bogle, I. D. L., in: Proceedings of Chemeca 91 (Newcastle, NSW),Australian Chemical Engineering Conference. Canberra: Institution of Engineers Australia, 1991, pp. 706-712. [46] Middelberg, A. P. J., O’Neill, B. K., Bogle, I. D. L., Trans I Chem E part C, 1992, 70, 8-12. [47] Hart, R.A., Bailey, J.E., Biotechnol Bioeng, 1992, 39, 1112-1120. [48] Chang, J. Y., Swartz, J. R., in: Protein Folding: in vivo and in vitro. ACS Symposium Series 526; Cleland, J. L. (Ed.), Washington: American Chemical Society, 1993; Vol. 526, pp. 178188. [49] Greenwood, M., Kotlarksi, N., O’Neill, B. K., Falconer, R., Francis, G., Middelberg, A. P. J., in: Proceedings of the I994 IChemE Research Event. London: Institution of Chemical Engineers, 1994, pp. 250-252. [50] Hart, R. A,, Lester, P. M., Reifsnyder, D. H., Ogez, J. R., Builder, S. E., Biotechnology, 1994, 12, 1113-1117. [51] Hart, R.A., Ogez, J.R., Builder, S.E., Bioseparation, 1995, 5, 113-121.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
5 The Application of Glycobiology for the Generation of Recombinant Glycoprotein Therapeutics Jan B. L. Damm
5.1 Introduction This paper discusses the role that glycosylation plays in the development of recombinant glycoprotein therapeutics. It is generally conceived that the opportunities that recombinant DNA technology offers for the generation of new therapeutics can hardly be overestimated. However, this does not mean that there are no pitfalls and potential problems. In this chapter some of the problems that the pharmaceutical industry faces in producing recombinant glycoprotein drugs which are related to the glycosylation of the molecules, and some of the potential solutions and opportunities that glycobiology offers to circumvent these problems and ultimately to develop recombinant drugs with superior characteristics, are addressed. Recombinant DNA-based biotechnology emerged in the early 1970s as a challenging opportunity for the development of new glycoprotein pharmaceuticals. Although it took nearly a decade before recombinant DNA technology could be put into the day-to-day practice of the pharmaceutical industry, and it took until 1982 before the first biotechnology-based drug, human insulin, was marketed, it was clear from the beginning that biotechnology offered unseen possibilities and held great promise for breakthroughs in the development of new, complex glycoprotein drugs. Human insulin, which is a pure protein drug and not a glycoprotein, was followed by Somatrem for treatment of human growth hormone (hGH) deficiency in children in 1985. Thereafter, many recombinant drugs for treatment of various diseases followed. The approved recombinant therapeutics and vaccines till 1994 are listed in Table 5-1. Figure 5-1 shows the cumulative number of approved recombinant therapeutics and vaccines until 1994, and the US sales. Evidently, after a slow start the number of recombinant DNA therapeutics and vaccines, indicated by the bars in Fig. 5-1, has increased steadily to about 30 recombinant drugs in 1994. The line in Fig. 5-1 indicates U.S. sales of recombinant therapeutics in billion dollars. Thus, in 1990 recombinant therapeutics accounted for $1 billion sales in the United States, while in 1993 this figure already amounted to $3 billion. It is generally believed that the number of recombinant therapeutics and their sales will continue to grow in the near future. At present, about 150 biotech products have been evaluated in the clinic, and a far larger number is engaged in the research phase. A recent survey among the main American pharmaceutical companies showed that
108
5 The Application of Glycobiology
Table 5-1. Approved recombinant therapeutics and vaccines until 1994. Product
Indication
Company (Trade name)
1982 Human insulin
Diabetes
Eli Lilly/Genentech(Humulin)
1985 Somatrem for injection
hGH deficiency in children
Genentech (Protropin)
Hepatitis B prevention Hairy cell leukemia Hairy cell leukemia Reversal of acute kidney transplant rejection
Merck (Recombivax HB); Chiron Hoffmann-La Roche (Roferon) Schering-PloughiBiogen (Intron A) Ortho Biotech (Orthoclone OKT3)
Acute myocardial infarction hGH deficiency in children
Genentech (Activase) Eli Lilly (Humatrope)
AIDS-related Kaposi’s sarcoma AIDS-related Kaposi’s sarcoma Genital warts
Hoffmann-La Roche (Roferon)
Genital warts Hepatitis B prevention
Interferon Sciences (Alferon N inj.) SmithKline Beecham (Engerix-B); Biogen Amgen (EPOGEN); JohnsonLkJohnson: Kirin
1986 Hepatitis B vaccine MSD Interferon alfa-2a Interferon alfa-2b Muromonab-CD3 1987 Alteplase (tPA) Somatotropin for injection 1988 Interferon alpha-2a Interferon alpha-2b 1989 Interferon alfa-n3 Hepatitis B vaccine Erythropoietin 1990 Erythropoietin PEG-adenosine Interferon-gamma- 1b Alteplase (tPA) Erythropoietin CMV immune globulin 1991 Filgrastim (G-CSF) P-glucocerebrosidase Sargramostim (GM-CSF) Sagramostim (GM-CSF) Interferon-alpha-2b
anemia associated with chronic renal failure
Schering-Plough/Biogen (IntronA)
anemia associated with AIDS/AZT ADA-deficient SCID Management of chronic granulomatous disease Acute pulmonary embolism anemia associated with chronic renal failure CMV prevention in kidney transplant patients
Amgen (Procrit); Ortho Biotech
Chemotherapy-induced neutropenia Type I Gaucher’s disease Authologous bone marrow transplantation Neutrophil recovery following bone marrow transplantation Hepatitis C
Amgen (Neutrogen)
Enzon; Eastman Kodak Genentech (Actimmune) Genentech (Activase) Ortho Biotech (Procrit) MedImmune (Cyto-Gam)
Genzyme (Ceredase) Hoechst-Roussel (Prokine); Immunex Immunex (Leukine); Hoechst-Roussel Schering-PloughlBiogen (Intron A)
5.1 Introduction
109
Table 5-1. (continued). Product
Indication
1992 Antihemophilic factor Hemophilia B Aidesleukine (interleukin-2) Renal cell carcinoma Indium-I 11 labeled antibody Detection, staging, and follow-up of colorectal cancer Indium-1 11 labeled antibody Detection, staging, and follow-up of ovarian cancer Antihemophilic factor Hemophilia A Interferon-alpha-2b 1993 Erythropoietin Interferon-beta DNAse Factor VIII Ery thropoietin 1994 Filgrastim (G-CSF) Enzyme (PEG-L-asparaginases) Human growth hormone Glucocerebrosidase
Hepatitis B
Company (Trade name)
Armour (Mononine) Chiron (Proleukin) Cytogen (OncoScint CR103); Knoll Cytogen (OncoScint OV103); Knoll Genetics Inst.; Baxter Healthcare (Recombinate) Schering-Plough/Biogen (Intron A)
Chemotherapy -associated Amgen (Procrit); Ortho Biotech anemia in non-myloid malignancy patients Relapsinghemittkg multiple Chiron; Berlex (Betaseron) sclerosis Cystic fibrosis Genentech (Pulmozyme) Hemophilia A Genentech; Miles (Kogenate) Anemia associated with Ortho Biotech (Procrit) cancer and chemotherapy Bone marrow transplant Refractory childhood acute lymphoblastic leukemia Short stature caused by hGH deficiency Type I Gaucher’s disease
Amgen (Neupogen) Enzon (Oncaspar) Genentech (Nutropin) Genzyme (Cerezyme)
Source: Biotechnology in the U.S. (Institute for Biotechnology Information, Research Triangle Park, NC, 1995).
more than 30% of their research projects are biotechnology-based. These data are reflected in the expectations for sales and market share of recombinant therapeutics. According to the Institute of Biotechnology Information of the North Carolina Biotechnology Centre, by the turn of the century the U.S. market for recombinant therapeutics will amount to at least $9 billion. This forecast is based on the still increasing demand for complex and selective therapeutics, such as for the treatment of AIDS, autoimmune diseases, Alzheimer’s disease, and cancer. However, there is also reason for modesty: although impressive, the current U.S. sales of recombinant therapeutics accounts for (only) 5 % of total U.S. sales of therapeutics and undoubtedly investors and speculators had expected more. Why is the list of presently marketed recombinant therapeutics not as long as one might had anticipated some 10 years ago? Potential problems in the production of recombinant drugs include:
110
5 The Application of Glycobiology
40
35
v)
30
0 3
25 20
a > -0
&z $‘p
8 s 7
6
0
. I
m
c
5e
15
4 v ) 3 4 )
10
2 % 1 m
5 0
Fig. 5-1. Cumulative number of approved recombinant therapeutics and vaccines until 1994 and their sales in the U.S, 1995 + 1996: preliminary data.
Instability Insolubitily Disappointing production rates Instability of the selected cell line Difficulties in up-scaling of production and purification of the recombinant product Undesired biodistribution Too long or too short circulatory half-lives Complex registration procedures Patent issues addition to these problems, the fact that many recombinant drugs are glycoproteins rather than proteins is also partly responsible. This means that glycosylation, next to other post-transcriptional modifications of the protein, is a phenomenon which has to be dealt with in the right way in order to obtain suitable products. However, apart from being a complicating factor which sometimes jeopardizes successful development of recombinant drugs, glycosylation may also create new opportunities to overcome some of the aforementioned problems and in fact it may open the way to development of improved recombinant therapeutics.
5.2 Structure and Function of Glycoproteins For a proper understanding of the opportunities that glycosylation offers to circumvent some of the aforementioned problems in the generation of recombinant glycoprotein therapeutics, a brief discussion of some general structural and func-
5.2 Structure and Function of Glycoproteins
111
tional features of glycoproteins is necessary. For more details, the reader is referred to Chapter 12. Glycoproteins are ubiquitous in nature. Table 5-2 shows some examples of glycoproteins and their function. Glycoproteins may function for instance as enzymes, hormones, growth factors, lectins (glycoproteins that bind particular carbohydrate sequences), membrane constituents, serum constituents, or structural components. The impact of glycosylation on the physico-chemical and functional properties of a (recombinant) glycoprotein drug are best illustrated by an example. Chorionic gonadotropin (CG) and the pituitary hormones lutropin (LH), follitropin (FSH), and thyrotropin (TSH) are members of a glycoprotein hormone family which have different physiological functions. Each glycoprotein hormone is a heterodimer consisting of two non-covalently associated subunits, designated a and Within a given animal species the a-subunits of LH, FSH, TSH and CG arise from a single gene [l] and have identical amino acid sequence [2-41. However, the two homologously localized [5] N-linked oligosaccharide chains differ significantly in structure [6,7]. Although highly homologous (> 80 %) in protein, the 0-subunits of LH, FSH, TSH, and CG arise from separate genes and differ in amino acid sequenc[S], number of glycosylation sites (varying from 1 to 6), and type (only N-glycosidic versus. N-
s.
Table 5-2. Occurrence and functions of some representative glycoproteins. G1ycoprotein
Source
Molecular weight (kDa)
Carbohydrate content (%)
Enzymes Alkaline phosphatase Carboxy-peptidase Y
Mouse liver Yeast
130 51
18 17
Hormones Human chorionic gonadotropin Erythropoietin
Human urine Human urine
38 34
31 29
Potato Soybean
50 120
50 6
Membrane constituents Gly cophorin Rhodopsin GP- 120
Human erythrocytes Bovine retina HIV virus
31 40 120
60 7 50
Serum glycoproteins IgG Thyroglobulin
Human serum Calf thyroid
150 670
10 8
Structural glycoproteins Collagen
Rat skin
300
Other Interferon tPA Mucins
Human leukocytes Human serum Mucosal epithelium
Lectins
26 65 103-1 04
0.4 20 15
80
112
5 The Application of Glycobiology
and 0-glycosidic) and structure of carbohydrate chains (general aspects of N-and 0glycosylation are discussed in Chapter 12; the specific differences in glycosylation between CG, FSH, LH, and TSH have been reviewed previously [ 5 ] ) . Figure 5 -2 depicts the three-dimensional model of human chorionic gonadotropin (hCG), based on its recently determined crystal structure [9]. In humans, hCG is produced by the syncytiotrophoblasts of the placenta, a cell type which lacks secretory granules and releases the hormone constitutively following synthesis [lo]. Human chorionic gonadotropin plays an important role during pregnancy and is a typical example of a complex glycoprotein. hCG exhibits LH-like effects and is responsible
Fig. 5-2. (Colours please see front cover picture.) Three-dimensional model of human choriogonadotropin. The model is based on the crystal structure of deglycosylated hCG9 (PDB code lhrp). The protein part of the molecule (ribbon) and the four N-linked carbohydrate chains (spheres) are shown on the same scale. The oligosaccharides are attached to As1152 (top, right) and Am78 (bottom) of the a-subunit (green), and to Asnl3 and 30 (top, left) of the @-subunit (blue). The binding region is indicated in red. It should be noted that the spatial orientation of the carbohydrate chains is arbitrarily set as they are not present in the crystal structure. The carboxy-terminal peptide of the P-subunit (amino acid residues 131-145) is not depicted because its 3D-structure could not be deduced from the crystal [Figure reproduced by courtesy of Prof. Dr. P. D. J. Grootenhuis (Dept. of Computational Medicinal Chemistry, N.V. Organon, Oss)].
5.2 Structure and Function of Glycoproteins
113
for continued existence of the corpus luteum and stimulation of progesterone production during the first 6-8 weeks of gestation. The serum concentration of hCG in pregnancy can reach levels of up to several hundred nM during the first trimester [ 1I]. For any hormone, this is an extraordinarily high concentration, and indeed, extrapolations from in vitro studies suggest that this hormone level is much larger than that required for maximal progesterone production by the corpus luteum. The hormone also stimulates progesterone production by the placenta and probably exhibits other biological actions on the fetus. hCG can be detected in maternal serum about four days after fertilization. Following the peak at the end of the first trimester the serum level falls, but is still significant at term. Free subunits can also be detected in the maternal serum. It is noteworthy that the production of the a-subunit appears to be the limiting factor for hCG formation in early pregnancy; in contrast, it is the synthesis of 0-subunit that seems to be limiting late in gestation. The two non-covalently linked, crest-like-shaped protein subunits of hCG are represented by the ribbon in Fig. 5-2. The a- and @subunits are indicated in green and blue, respectively. The stretches in red indicate the receptor-binding domain of the hormone, involving parts of both the wand 0-subunits. The dimeric glycoprotein hormone carries four N-linked carbohydrate chains or oligosaccharides, indicated by the spheres, which are covalently linked to the protein backbone via the amide function of asparagine residues. The a-subunit contains two N-linked oligosaccharides at asparagine residues 52 and 78. The remaining two N-linked carbohydrate chains are present at asparagine residues 13 and 30 of the 0-subunit. In addition, four O-linked oligosaccharides occur, attached to serine residues 121, 127, 132 and 138 of the carboxy-terminal part of the 0-subunit. This portion of hCG-0, comprising about 30 amino acid residues, distinguishes hCG-0 from the 0-subunits of the pituitary glycoprotein hormones. The carboxy-terminal region has been suggested to have occurred via loss of a termination codon when hCG-0 evolved from LH-0, thus permitting the 3’-untranslated region to become incorporated into the coding sequence [12]. This unique region of hCG-0 may contribute to some of the properties of this glycoprotein, and it has been exploited for the development of highly specific immunoassays [13]. It should be noted, the hCG-0 carboxy terminal peptide is not shown in Fig. 5-2 because so far it could not be deduced from the crystallographic data, probably because it is too flexible. From the model it is apparent that the carbohydrate chains are relatively large entities - often the true proportion of the glyco-part of a glycoprotein is underestimated - and that the oligosaccharides occur on the outside of the molecule. The relatively large proportion of the carbohydrates, together with the location on the periphery of the molecule are the main reasons for the strong influence on the physico-chemical properties of the glycoprotein hormone. For instance, when the carbohydrate chains are (partly) removed, or when N-glycosylation is prevented by mutation of the glycosylation sites in the recombinant glycoprotein, the molecular mass and size of the molecule are obviously much lower (about 30 %). Also, the molecule is less hydrophilic, which is reflected by a dramatic drop in its solubility, leading to the formation of insoluble aggregates. Furthermore, the biochemical and physical stability of the molecule is affected: it is more prone to proteolytic breakdown in the circulation and the aI0-dimer tends to dissociate into (non-active) subunits.
114
5 The Application of Glycobiology
The effect on the physico-chemical properties in turn modifies the molecule’s biological function. For hCG, as for the related glycoprotein FSH [I41 and many other glycoproteins, it is known that the amount, type, and composition of the oligosaccharides influences the in vitro and in vivo bioactivity of the hormone. The role of N-linked oligosaccharides in glycoprotein hormone bioactivity has been extensively investigated and reviewed (see [4]). In general, the effects of chemical or enzymatic deglycosylation, the prevention of N-glycosylation by tunicamycin, or deletion of N-linked carbohydrates by mutations at the DNA level, on receptor binding, clearance, and bioactivity have been evaluated. Removal of sialic acid (a negatively charged monosaccharide which usually occurs in the terminal position of the carbohydrate chain, see Chapter 12) from hCG results in increased receptor binding by the hormone [ 151. However, (partly) desialylated hCG is quickly removed from the circulation via asialo glycoprotein receptor-mediated clearance in the liver. The efficiency of this clearing mechanism is illustrated by the t i of completely desialylated hCG in humans which is in the order of minutes as opposed to the t i of native hCG which is 70 h. Mainly due to the influence of the presence of sialic acid - in the terminal non-reducing position of the oligosaccharides - on hCG receptor binding and liver-mediated clearance, desialylation has an opposite effect on the in vitro and in vivo activities: the in vitro activity is slightly enhanced (because sialic acid is not required for receptor binding or bioactivity p e r se), whereas the in vivo activity is dramatically reduced. Complete enzymatic deglycosylation of hCG does not alter receptor binding, but abolishes cAMP production as well as biological responses such as spermatogenesis (see [ 161). Studies utilizing enzymatic deglycosylation have, however, yielded conflicting results in a number of instances. Chemical deglycosylation by HF solvolysis has also been used [17] to examine the role of the N-linked oligosaccharides in bioactivity [ 16,18,19]. With the exception of the Asn-linked N-acetylglucosamine residue, complete removal of the oligosaccharides is accomplished with HF solvolysis with little or no detectable damage to the protein subunits. Alternatively, hCG molecules missing one or more of the N-linked carbohydrates have been obtained by site-directed mutagenesis of the N-glycosylation sites [20-221. Both the studies with HF-deglycosylated hCG and N-deglycosylated hCG muteins have led to the following conclusions:
-
1. Post-synthesis removal of the carbohydrate moieties does not hinder cdp subunit recombination. 2. N-linked oligosaccharides are not required for receptor binding in vitro. 3. Deglycosylated hormones display significantly reduced abilities to stimulate cAMP production in target cells despite unaffected receptor binding. In essence, it appears that chemical deglycosylation has the effect of converting the hormone from an agonist into a competitive antagonist, since the deglycosylated product retains the ability to bind to the appropriate receptor, but not to induce the biological effects. Supporting evidence for the role of the oligosaccharides comes from the properties of hCG produced in patients with choriocarcinom [23]. This hCG, having an identical amino acid composition as native hCG but an altered glycosylation pattern, was found to exhibit a threefold increase in affinity for its receptor, but a much lower biological activity when compared with normal hCG.
5.2 Structure and Function of Glycoproteins
115
The various effects of glycosylation on the physico-chemical and biological properties of glycoproteins in general are dealt with in Section 5.2.2. The effects that glycosylation may exert on the physico-chemical and biological properties of glycoproteins is one of the main reasons for gathering information about the glycosylation when producing glycoprotein drugs by recombinant DNA technology. It should be noted however, that glycosylation does not always influence biological function; there are examples of recombinant glycoproteins with aberrant glycosylation or no glycosylation at all which seem to function perfectly normal in vitro [24-261. On the other hand there are also examples where a non-native glycosylation results in unwanted biological properties. Therefore, to be on the safe side, the glycosylation of the recombinant product should be identical to, or at least closely resemble, the glycosylation of the natural product. More interestingly, manipulation of the glycosylation may give an opportunity to improve the properties of the drug. Especially, when (on basis of sound analytical and biological data) structure-function relationships can be established, the fact that most protein drugs are glycosylated may be exploited to improve the therapeutic properties (see Section 5.3). The second reason for gathering information about glycosylation is that the recombinant product, including its carbohydrates, must be characterized to verify for instance identity and batch-to-batch reproducibility, also with respect to the carbohydrate moiety. This is important for the manufacturer to check as to whether the product meets required specifications. Moreover, at present a detailed characterization is included in the package of demands of the registration authorities. Questions to answer with respect to glycosylation are: - type of carbohydrate chains: -
composition and amount;
- Structure;
reproducibility of glycosylation: differences with natural glycoprotein; and - biological consequences.
-
-
To answer these questions considerable effort is put into the analysis of the carbohydrate moiety of new recombinant glycoprotein drugs. For a discussion of the strategies that can be adopted to carry out the characterization of the carbohydrate moiety of recombinant glycoproteins the reader is referred to Chapter 12.
5.2.1 Difficulties in Establishing Carbohydrate Structure-Function Relationships Unfortunately, the presently available data (for review, see [27]) show that the effects of glycosylation on the functioning of a (recombinant) therapeutic are not entirely predictable. There are several reasons for the difficulty in predicting specific rules for carbohydrate functions.
116
5 The Application of Glycobiology
1. There is an indirect genetic control over the structure of the protein-bound carbohydrates. This is due to the mode of biosynthesis of protein-linked carbohydrate chains which involves both co- and post-translational events, resulting in an inherent, yet characteristic, variability. As a consequence, a glycoprotein therapeutic generally is not a population of identical molecules, but rather is composed of discrete subsets of differently glycosylated molecules, the so-called glycoforms [28], that have different physico-chemical and biochemical properties which in turn may lead to functional diversity. Consequently, any glycoprotein drug that consists of different glycoforms will exhibit a composite activity, reflecting a weighted average of the activity and incidence of each glycoform. Therefore, structure-function relationships for glycoprotein-linked carbohydrate chains are not necessarily apparent at the level of the native glycoprotein (population). Rather, structure-function relationships should be studied on the level of the individual glycoforms of the glycoprotein therapeutic. However, the preparative isolation of individual glycoforms, if possible at all, may require quite some effort. 2 . Carbohydrate functions and protein functions are closely related or even intertwined. Therefore, it frequently is difficult to distinguish the properties imposed by the carbohydrate moiety from the functions that are intrinsic to the protein part of a glycoprotein drug. Obviously this precludes studies towards specific functions of the carbohydrate moiety. 3. Carbohydrate structure-function relationships are not static but dynamic. A specific carbohydrate structure may serve different purposes in different cells (of the same organism) or at different times in the life cycle or state of development of the same cell. So in fact, a particular structure may have different functions depending on its localization or the physiological status of the organism which of course may obscure structure-function relationships. 4. Multivalency seems to be a key-word in carbohydrate function. In many cases per definition it will not be possible to associate a certain biological function with a specific individual carbohydrate structure. Rather, structure-function relationships must be searched for in the context of glycosylation patterns and concentration of carbohydrate chains, both in time (which again refers to the physiological status of the organism) and space (which relates to multivalency as key factor in recognition events). For instanc, it is more or less established that during carbohydratemediated recognition events only the concerted interaction between a multitude of carbohydrate ligands and their receptors evokes a biological response. Obviously, this phenomenon complicates the study of structure-function relations.
5.2.2 Glycosylation-associated Effects on the Properties of Glycoprotein Drugs Nonetheless, some general glycosylation-associated effects on the properties of glycoprotein drugs, acting in- or interdependently of each other, are recognized. As outlined in Fig. 5 - 3 , two types of effect on the properties of glycoprotein drugs must be distinguished, namely the influence on the physico-chemical properties, and the
5.2 Structure and Function of Glycoproteins Property
Effecf
physico-chemical
size mass solubility viscosity charge
biological
antigenicity stability clearance intracell. routing biodistribution receptor binding cell-cell contacts
117
Fig. 5-3. Glycosylation-associated effects on the properties of glycoprotein drugs.
influence on the biological properties. As already discussed for hCG, the effects of the presence of carbohydrate chains on physico-chemical properties of the glycoprotein drug are quite trivial - but are nonetheless sometimes of decisive importance such as the effects on size, mass, tertiary structure, solubility, viscosity, and charge. The physico-chemical characteristics may in turn affect - in a more complex way - the biological functioning of the glycoprotein drug, involving antigenicity, stability, plasma half-life, intracellular routing, organ targeting and biodistribution, receptor binding and cell-cell recognition events. Some specific examples of how carbohydrate characteristics influence or even determine the biological properties of a glycoprotein drug include: 1. The influence of the sialic acid/galactose ratio of a glycoprotein therapeutic on its circulatory half-life via determination of the kinetics of liver-mediated clearance [29,30]. The over-riding influence of the presence of sialic acid in terminal position of the oligosaccharides of hCG, masking the penultimate galactose residue and thus preventing rapid hepatic clearance, has already been mentioned. Similar effects are observed for many other glycoprotein (drugs), e.g. recombinant FSH [311 (PuregonR) and erythropoietin [32,33]. 2. The increased resistance against proteolysis by shielding potential protease cleavage sites. It has been shown for instance, that the carbohydrate chains (in particular the sialic acid residues) of von Willebrand factor protect the protein against amino-terminal proteolytic cleavage and are essential for maintenance of its multimeric structure [34,35]. Similarly, the oligosaccharides of bovine pancreatic ribonuclease protect the molecule against protease degradation [36]. 3. The recognition of terminal sialic acids of glycoproteins and glycolipids in the cell membrane by various viruses and bacteria, mediating cellular infection [37-391. 4. Recognition of polylactosamines on erythrocytes [40] and/or platelets [4 11 by autoimmune antibodies, leading to autoimmune destruction of the cells causing autoimmune hemolytic anemia.
118
5 The Application of Glycobiology
5. Recognition of sialylated, fucosylated lactosaminoglycans on leucocytes by the Eselectin of endothelial cells [42-441, mediating extravasation and also inflammation. 6. The involvement of O-linked carbohydrate chains in mammalian fertilization. Bleil and Wassarman [45] described the essential role of galactose at the terminal non-reducing position of specific O-linked carbohydrate chains of the murine egg zona pellucida glycoprotein ZP3 for the species-specific recognition and primary binding of sperm cells. It is generally conceived that the ZP3 -linked oligosaccharides fulfil a similar function in various other species [46-481. Miller et al. [49] reported the involvement of p 1-4 galactosyltransferase in murine sperm-egg recognition. The authors suggested that the galactosyltransferase on the sperm membrane mediates fertilization by binding to the aforementioned O-linked carbohydrate chains located on the zona pellucida glycoprotein ZP3. Hence, the sperm surface galactosyltransferase and the egg coat glycoprotein ZP3 function as complementary adhesion molecules that enable the recognition and primary binding of murine gametes. An overview of the presently recognized effects of the carbohydrate moiety on the biological functioning of the parent (recombinant) glycoprotein therapeutic can be found in the compendious review paper of Varki [27]. Taken together, the various functions of carbohydrate chains can be summarized in that they either mediate specific recognition events, or that they modulate biological processes. Unfortunately, it is not possible to make general statements about which carbohydrate characteristics in general are best for a(ny) recombinant glycoprotein therapeutic. Rather, for each individual recombinant drug the optimal glycosylation must be unravelled in relation to its desired therapeutic profile.
5.3 Glyco-engineering To exploit the opportunities of glycosylation and to deal with its restrictions as good as possible, at least two requirements must be fulfilled. The first is the establishment of sound relationships between (protein-linked) carbohydrate structure(s) on the one hand and their influence on the biological functioning of the glycoprotein drug on the other hand. Such a structure-function relation should give an answer to the question ‘what should be made?’, or more precisely, ‘what should the glycosylation of the recombinant product look like in order to give the (recombinant) product its desired therapeutic properties?’ Once it is known what the glycosylation of the ideal product should look like, the second requirement is an answer to the question ‘how can that be accomplished?’. This is where glyco-engineering comes into the picture. Glyco-engineering may be defined as the ensemble of techniques that allow the controlled production and/or manipulation of the carbohydrate moiety of glycoproteins. Here, the relation between the protein backbone, the selected cell line and the culture conditions on the one hand, and the obtained glycosylation pattern on the other hand is essential.
5.3 Glyco-engineering
119
The importance of glycosylation for production levels of recombinant drugs is well known. Absence or incorrect glycosylation may lead to a dramatic drop in productivity due to decreased or impaired biosynthesis or secretion. In the absence of glycosylation, the newly formed proteins might get stuck in the cell lumen or in the membrane, resulting in reduced secretion or even complete loss of secretion. Furthermore, the earlier discussed complications with respect to (lack of) solubility, stability, biological activity, etc. of the products due to incorrect glycosylation or absence of glycosylation frequently occur. In addition to obtaining glyco-engineered recombinant therapeutics with increased bioactivity or specificity, controlled modification of the carbohydrate moiety may be used to eliminate, some potential problems, such as antigenicity of recombinant therapeutics that carry potentially immunogenic carbohydrate determinants, to enhance or reduce the clearance rate, or to eliminate unwanted carbohydrate heterogeneity. Reduction of the carbohydrate heterogeneity for instance may make life easier because it facilitates the purification, characterization, quality control, prediction of pharmacokinetic behavior and - if wanted - crystallization of the recombinant therapeutic, which in turn may lead to a shorter development time and registration procedure. In summary, it can be stated that because the carbohydrate chains influence the physico-chemical and biological properties of the glycoprotein drugs, glycosylation may be exploited to alleviate the mentioned problems and to improve the properties of the recombinant drug by applying glyco-engineering techniques. Engineering of glycosylation is conceivable at three levels: (i) the DNA-level; (ii) the biosynthesis level; and (iii) the product level (Fig. 5-4).
GLYCO-ENGINEERING
Fig. 5-4. The three levels at which glyco-engineering is possible.
120
5 The Application of Glycobiology
5.3.1 Glyco-engineering at the DNA level At the DNA level there are again three options to manipulate the glycosylation of the recombinant product, namely via the protein backbone of the recombinant glycoprotein, by choosing the host cell, and by selection of mutants and/or coexpression of glycosyltransferases and glycosidases, (Table 5 -3). Table 5-3. Glyco-engineering at the DNA level. Carbohydrate structure and protein backbone Sites for N- and 0-glycosylation, GPI anchors - Accessibility of glycans to glycosylation enzymes - Determination of carbohydrate fine structure -
Host cell - Glycosyltransferases - Glycosidases Selection of mutants, e.g. CHO-tetra a2-6 ST - CHO-LEC11 al-3 FT - CHO-LECS L SA/G
-
-
CHO-PIR.LEC1
1 SA/G/GN
sTri/Tetra-antennary (s)LeX/(s)Leaglycocon jugates terminal GN terminal mannose
Epo, in vivo bioactivity t Target ELAM-1IGMP140 Target reticulo-endothelial cells Target reticulo-endothelial cells
~~
GPI, glycosylphosphatidylinositol; CHO, Chinese hamster ovary; a2-6 ST, a2-6 sialyltransferase; al-3 FT, a l - 3 fucosyltransferase; SA, sialic acid (N-acetylneuraminic acid); G, galactose; GN, N-acetylglucosamine; sTri/Tetra, sialyted tri-/tri' or tetra-antennary oligosaccharides; (s)Le"'", (sialy1)Lewis X or A; Epo, erythropoietin; ELAM, endothelial leucocyte adhesion molecule; GMP, a-granule membrane protein; ?, increase; 1,decrease.
5.3.1.1 Carbohydrate Structure and Protein Backbone
First of all, when glyco-engineering at the DNA level is considered it is important to realize that the glycosylation of the protein is affected by the protein itself in at least three ways (this topic is covered in more detail by Cumming [ S O ] ) . First, and most obviously, the protein presents the potential sites for N-glycosylation, O-glycosylation, and the attachment of glycosylphosphatidylinositol anchors (GPI-anchors, see Chapter 12). For N-glycosylation a consensus amino acid sequence asparagine-X (i.e., any amino acid, except proline) -serine/threonine is required. This implies that by mutating or deleting the consensus sequence, N-glycosylation can be knocked out at specific sites. In contrast, by insertion of consensus sequences additional N-linked oligosaccharides might be attached (it should be noted, however, that presence of a consensus sequence is a necessary, but not sufficient requirement for N-glycosylation). For 0-glycosylation and GPI anchors a consensus sequence is not known and hence the situation is less clear.
5.3 Glyco-engineering
121
The second way in which the protein influences its own glycosylation is by determining the accessibility of its attached oligosaccharides to the enzymes involved in glycosylation. By limiting access, or by forcing its oligosaccharide chains into particular conformations, the conversion of the carbohydrate chains into substrates or non-substrates for the (de)glycosylating enzymes, can be determined. In fact, the over-riding influence of the protein backbone on its own glycosylation is exemplified by the establishment of consistent glycosylation patterns for a particular (recombinant) glycoprotein, irrespective of the cell line in which it is produced (there are many exceptions, however, e.g. [5 11). Finally, the specific amino acid sequence may determine the carbohydrate fine structure. This is illustrated for example by the gonadotropic hormone-specific Nacetylgalactosamine transferase which is able to attach N-acetylgalactosamine residues to the carbohydrate chains linked to luteinizing hormone, but not to the carbohydrate chains of the related follicle stimulating hormone [ 5 ] , in spite of the fact that both glycoproteins are produced in the same organ and that there are only small overall differences in the protein backbone. 5.3.1.2 Choice of the Host Cell The second option to manipulate the glycosylation of the recombinant product at the DNA level is by choice of the host cell. It is generally recognized that the glycosylation is more or less host cell-specific due to the repertoire, concentration, and compartmentalization of glycosidases and glycosyltransferases present. Glycosyltransferases add monosaccharides to the growing carbohydrate chain during its biosynthesis, whereas glycosidases are enzymes that remove monosaccharides during the final processing of the oligosaccharides (see [52]).The concerted action of both types of enzymes is needed in the biosynthesis of carbohydrates. Both types of enzymes exhibit a high degree of specificity towards the nucleotide sugar donor, the oligosaccharide acceptor, and the anomeric configuration and type of carbohydrate linkage formed. Since the type of host cell that is chosen for the production of the recombinant drug to a large extent predefines the ensemble of glycosyltransferases and glycosidases present, the right choice of host cell is extremely important for the generation of a ‘desired’ glycosylation pattern in recombinant glycoproteins. So far, various types of host cells have been used for the production of recombinant therapeutics. The most important are Escherichia coli, yeast, insect, and mammalian cells, (Table 5-4). E. coli has been the classical host because it is easy to grow and maintain in large quantities, and it gives high protein yields. However, it does not possess the glycosylation machinery of eukaryotes which means that products that normally occur as glycoproteins are produced as proteins. (It is worthy of note that Messner and Sleytr [53] reported the biosynthesis of (bacterial) surface layer glycoproteins in bacteria, but this seems to be an exception.) Because the unglycosylated proteins often are not properly folded, or are insoluble or form aggregates or give problems with respect to in vivo activity or antigenicity, prokaryotes like E. coli are in many
122
5 The Application of Glycobiology
Table 5-4. Frequently used host cells for the production of recombinant (g1yco)protein drugs.
E. coli Easy to grow and maintain in large quantities - No glycosylation machinery -+ consequences for production, conformation, (in)solubility, aggregation, in vivo activity, antigenicity? -
Yeast cells Easy to grow and maintain in large quantities - Yeast-type glycosylation + Immunogenic compatibility?, bioactivity? -
Insect cells - Relatively easy to grow - High production rates - Mammalian-like N- and 0-linked carbohydrates Mammalian cells - Culture and maintenance not without problems - Production levels relatively low - Mammalian glycosylation machinery
cases not suitable for the production of genetically engineered glycoprotein therapeutics. Yeast cells combine the major advantage of prokaryotes, namely that they are easy to grow and maintain in large quantities, with the main advantage of eukaryotes, namely glycosylation. However, yeast-specific carbohydrate chains differ dramatically from their mammalian counterparts, which may compromize the immunogenic compatibility and/or bioactivity of the glycoprotein drug. Future research will learn whether this problem can be tackled by the use of yeast glycosylation mutants which produce (N-linked) carbohydrate precursors that are also common to mammalian cells, which may then be processed to the desired glycosylation pattern by treatment with the appropriate glycosidases and/or glycosyltransferases. During the last couple of years the application of baculovirus-transfected insect cells has emerged as a promising possibility for the production of recombinant glycoproteins. Transfected insect cells are relatively easy to grow, allow high production rates of the recombinant glycoprotein and seem to be capable of synthesizing nearly the whole spectrum of N- and 0-linked mammalian carbohydrate chains [54,55]. However, there is still debate with respect to the similarity/equivalency of the insect and mammalian glycosylation machinery. Although the culture and maintenance of large quantities of mammalian cells is not without problems and production rates are frequently lower than in the aforementioned cells, mammalian cells are in general the preferred host for the generation of recombinant glycoprotein therapeutics. The main reason for this is the equivalence of the glycosylation machinery of all types of mammalian cells, including human cells. There may, however, still be considerable differences in the concentration or presence of the glycosidases and glycosyltransferases between the various host cells. Therefore, also in this case the choice of a proper cell line remains a key issue. If the ideal glycosylation pattern for a particular glycoprotein drug is known, or
5.3 Glyco-engineering
123
more realistically, if it is known which general characteristics the carbohydrate chains should have or not have, a proper cell line can be selected. Preferably this will be a cell line which is known to express the desired carbohydrate features in its natural products. Frequently used mammalian cell lines to date are BHK and Chinese hamster ovary (CHO) cells and their various mutants. 5.3.1.3 Selection of a Glycosylation Mutant
The third option to manipulate the glycosylation of the recombinant product at the DNA level is by selection of an appropriate glycosylation mutant. At present various (mutant) cell lines have been selected or engineered [56-621, aiming at the synthesis of recombinant glycoprotein therapeutics with increased effectiveness. Mutants can be obtained by screening or by genetic engineering. In the latter case the desired type of glycosylation activity is engineered into the cell line by co-expression of the appropriate glycosyltransferases or glycosidases. Among the mutant cell lines currently available is the so-called CHO-tetra mutant transfected with the a2 - 6 sialyltransferase gene for the production of erythropoietin (Epo) with modified carbohydrate chains 1321. For erythropoietin it is known that the biological activity in vivo is highly dependent on the structures of its carbohydrate chains [63]. Epo molecules that bear predominantly diantennary carbohydrate chains (carbohydrate chains that contain two branches, see Chapter 12) display a much lower bioactivity than those that bear predominantly tetra-antennary oligosaccharides (carbohydrates that have four branches). By choosing a CHO cell line tansfected with the a 2 - 6 sialyltansferase gene, recombinant erythropoietin with fully sialylated, primarily tetra-antennary carbohydrate chains could be produced with significantly increased bioactivity. Probably, the same result may be obtained by transfection of the host with N-acetylglucosaminyltransferase-IVand -V, the enzymes which are responsible for the formation of higher branched oligosaccharides, or by supplementing the culture medium with exogenous interleukin-6 (IL- 6), which has been shown to shift the N-acetylglucosaminyltransferase-111activity in myeloma cell line OPM- 1 towards N-acetylglucosaminyltransferase-IVand -V activity [64], resulting again in production of glycoproteins with tri-/tri’- and tetra-antennary carbohydrate chains. Another example is the CHO-LEC1 1 mutant for the production of glycoconjugates that carry the sialyl-Le x/a determinant [65]. The aim here is to target the drug carrying the sialyl-Le x/a determinant to endothelial cells or lymphocytes that express the cell adhesion molecules ELAM-1 or GMP-140. A third example of the potential use of mutant cell lines concerns the CHO-LEC8 [661 and CHO-PIR.LEC 1 [67] mutants which produce carbohydrates that terminate in N-acetylglucosamine and mannose, respectively, in stead of sialic acid (more precisely, N-acetylneuraminic acid) or galactose. As already suggested by Stanley [68], this mutant could be used to produce recombinant glycoprotein therapeutics that specifically target the reticuloendothelial cells. It has been shown [69] that administration of exogenous glucocerebrosidase is effective in the treatment of patients with Gaucher’s disease, one of the most common diseases of glycolipid metabolism leading to accumulation of glycolipids in various cells and organs,
124
5 The Application of Glycobiology
resulting in severe pathological effects. However the drug - glucocerebrosidase - is only effective when it is selectively targeted to the reticuloendothelial cells. This was realized by treatment of the natural enzyme with neuraminidase, P-galactosidase and P-hexosaminidase, respectively, resulting in glucocerbrosidase-carrying carbohydrate chains terminating in mannose residues. The processed glucocerebrosidase molecules are targeted specifically to the reticuloendothelial cells [70] via recognition by the mannose-binding lectin which is present on the reticuloendothelial cells [71]. Obviously, it would be much more straightforward to produce this enzyme in the CHO-LEC8. or CHO-PIR.LEC1 mutant which would directly yield the desired product. A more detailed description of these and other examples of mutant cell lines that are used or potentially can be used to produce glycoprotein drugs with designed carbohydrate chains can be found in the review by Stanley [68].
5.3.2 Glyco-engineering at the Biosynthesis Level After selection of a specific cell line for production, glyco-engineering can still be performed at the biosynthesis level, viz. during the production of the recombinant product in the roller bottle, fermentor, etc. by adapting the culture conditions in order to manipulate the glycosylation of the recombinant product. Presently, it is recognized that culture conditions, like density, age and growth rate of the cells [72], concentration and type of nutrients [73], presence of growth differentiation factors, cytokines [74], glycosylation inhibitors [75] and glycosylation enzymes [76], pH, C02,02, and ammonium concentration, all affect glycosylation somehow. Therefore, in theory each of these factors might be exploited to manipulate the glycosylation (for a review of the environmental and bioprocess effects on protein glycosylation the reader is referred to Goochee et al. [74]). Unfortunately however, at present the relation between the culture conditions and the carbohydrate structures - or more generally, the glycosylation pattern - of the recombinant product is poorly understood. Hence, when applying this strategy a word of warning is in place. Additional and/or irreproducible heterogeneity of the carbohydrate moiety of a recombinant therapeutic as a result of uncontrolled or unstable culture conditions is highly unwanted. Therefore, in general the primary goal is to control and monitor the culture conditions as good as possible in order to obtain a ‘reproducible carbohydrate heterogeneity’. However, if in time sound relations between culture conditions and synthesized carbohydrate structures can be established, the second goal will be to engineer the desired glycosylation type into the recombinant product by careful manipulation and control of the culture conditions. So far there are several examples of the feasibility of this approach, like the application of glycosylation inhibitors, the induced expression of silent glycosyltransferase genes by growth factors, cytokines or agents like retinoic acid and butyrate - which affect the differentiation of the cultured cells - and the production at high or low pH, oxygen, C02 or glucose levels (for reviews, see [77] and
ml.
5.4 Conclusion and Perspectives
125
5.3.3 Glyco-engineering at the Product Level Finally, glyco-engineering can be performed on the product level. After its biosynthesis and collection, the recombinant product generally has to be purified. During the purification process it may be possible - apart from removing contaminants - to selectively remove, or alternatively selectively concentrate, certain glycoforms. Provided that this can be done under controlled conditions, the purification process may be used to selectively obtain the glycoforms which exhibit the desired product profile. Another possibility is to modify the carbohydrate chains of the purified product, for instance by treatment with chemicals or glycosidases, in such a way that the desired glycosylation characteristics are obtained. An example of this approach is the preparation of the mannose-terminating glycoforms of glucocerebrosidase to treat Gaucher’s disease. In both cases - selective purification and modification of the carbohydrate chains of the purified glycoprotein - the desired glycosylation characteristics are engineered into the finished product afterwards. In general this will be a rather expensive approach, yet in specific cases it might be practically and economically feasible.
5.4 Conclusion and Perspectives Recombinant glycoprotein therapeutics capture an increasingly important part of the total market for therapeutics. In view of this, possibilities to facilitate and optimize the development of recombinant glycoprotein drugs is the subject of much contemporary research. Figure 5 -5 summarizes the main opportunities for the development of improved recombinant glycoprotein therapeutics as discussed in this chapter. The knowledge about the relationships between (protein-linked) carbohydrate structure(s) on the one hand and the protein backbone, selected cell line, and culture conditions on the other hand, and the relation between carbohydrate structure and its biological function or its effect on the biological functioning of the glycoprotein is growing. This fulfils the first requirement for the generation of new, improved glycoprotein drugs. The second requirement is the availability of glyco-engineering techniques and dedicated analytical facilities (Chapter 12). Together, this may ultimately result in rational design of glycosylated recombinant therapeutics with improved characteristics by controlled manipulation of the carbohydrate moiety.
126
5 The Application of Glycobiolopy
Structure - function relationships Proper5
Glyco-engineering recDNA / production / Product-technology
Effect
- host cell - selection mutants - co-expression glycosyltransferases, glycosidases
- glucose, ammonium
- inhibitor glycosylation enzymes
- PO2 - hormones
- growing rate, density and age of the cells
- selective purification glycoforms
I
- enzvmatidchernical modification glycosylation
Optimized
gly coprotein
Increase of
Reduction of
reproducibility production solubility stability biological activity tuning of activity
heterogeneity antigenicity clearing development time
Fig. 5 -5. Possibilities for the development of optimized recombinant glycoprotein drugs on basis of structure-function relationships and by using glyco-engineering techniques.
Acknowledgement Part of this material is adapted from articles originally published in Pharmaceutical, Technology Vol. 7, No. 8, 1995 and BioPharm, Vol. 8, No. 9, 1995.
References
127
References [ l ] Boothby, M., Ruddon, R. W., Anderson, C., McWilliams, D., Boime, I., J Biol Chem, 1981, 256, 5121-5127. [2] Pierce, J. G., Parsons, T. F. Ann Rev Biochem, 1981, 50, 465-495. [3] Norman, A. W., Litwack, G., in: Hormones, Orlando, x. et al. (Eds.), New York: Academic Press, 1987. [4] Ryan, R. J., Keutmann, H.T., Charlesworth, M. C., McCormic, D. J., Milius, R. P., Calvo, F. O., Vutyavanich, T., Rec Prog Horm Res, 1987, 43, 383-429. [5] Baenziger, J. U., Green, E. D., Biochim Biophys Acta, 1988, 947, 287-306. [6] Green, E.D., Baenziger, J. U., J B i d Chem, 1988, 263, 25-35. [7] Green, E. D., Baenziger, J.U., J Biol Chem, 1988, 263, 36-44. [8] Esch, F. S., Mason, A. J., Cooksey, K., Mercado, M., Shimasaki, S., Proc Nut1 Acad Sci USA, 1986, 83, 6618-6621. [9] Lapthorn, A. J., Harris, D. C., Littlejohn, A,, Lustbader, J. W., Canfield, R. E., Machin, K. J., Morgan, F. J., Isaacs, N. W., Nature, 1994, 134, 455-461. [lo] Hoshina, M., Boothby, M., Boime, I., J Cell Biol, 1982, 93, 190-198. [ l l ] Cole, L. A., Kroll, T. G., Ruddon, R. W., Hussa, R. O., J Clin Endocrinol Metab, 1984, 58, 1200-1 202. [I21 Fiddes, J.C., Talmadge, K., Rec Prog Horm Res, 1984, 40, 43-74. [13] Birken, S., Canfield, R., Agosto, G., Lewis, J., Endocrinology, 1982, 110, 1555-1563. [14] De Leeuw, R., Mulders, J. W. M., Voortman, G., Romhout, F., Damm, J. B. L., Kloosterboer, H. J., Excerpts on Human Reproduction, 1996, 1, 13-21. [15] Kalyan, N. K., Lippes, H.A., Bahl, 0.P., J Biol Chem, 1982, 257, 12624-12631. [16] Sairam, M. R., in: Hormonalproteins andpeptides, Vol. 11, Li, C. H. (Ed.), New York: Academic Press, 1993, pp. 1-79. [17] Manunath, P., Sairam, M.R., J Biol Chem, 1982, 257, 7109-7115. [18] Keutmann, H. T., McIllroy, P. J., Berger, E. R., Ryan, R. J. Biochemistry, 1983, 22, 30673072. [19] Sairam, M.R., Bhargaui, G.N., Science, 1985, 229, 65-67. [20] Matzuk, M. M., Boime, I., J Cell Bid , 1988, 206, 1049-1059. [21] Matzuk, M.M., Keene, J.L., Boime, I. J Biol Chem, 1989, 264, 2409-2414. [22] Matzuk, M.M., Boime, I., J Biol Chem, 1988, 263, 17106-17111. [23] Mizuochi, T., Nishimura, R., Derappe, C., Taniguchi, T., Hamamoto, T., Mochizuki, M., Kobata, A., J Biol Chem, 1983, 258, 14126-14129. [24] Moonen, P., Mermod, J. J., Emst, J. F., Hirschi, M., DeLamarter, J. F., Proc Nut1 Acad Sci USA, 1987, 84, 4428-4431. [25] Mackenzie, P.I., Biochem Biophys Res Commun, 1990, 166, 1293-1299. [26] Lace, D., Olavesen, A. H., Gacesa, P., Carbohydr Res, 1990, 208, 306-311. [27] Varki, A,, Glycobiology, 1993, 3, 97-130. [28] Rademacher, T. W., Parekh, R.B., Dwek, R.A., Annu Rev Biochem, 1988, 57, 785-838. [29] Ashwell, G., Harford, J., Annu Rev Biochem, 1982, 51, 531-554. [30] Smith, P.L., Kaetzel, D., Nilson, J., Baenziger, J. U., J Biol Chem, 1990, 265, 874-881. [31] De Leeuw, R., Mulders, J. W. M., Voortman, G., Rombout, F., Damm, J. B. L., Kloosterboer, H. J., Mol Hum Reprod, 1996, 2, 361-369. [32] Takeuchi, M., Inoue, N., Strickland, T. W., Kubota, M., Wada, M., Shimizu, R., Hoshi, S., Kozutsumi, H., Takasaki, S., Kobato, A., Proc Natl Acad Sci USA, 1989, 86, 7819-7822. [33] Delorme, E., Lorenzini, T., Giffin, J., Martin, F., Jacobsen, F., Boone, T., Elliott, S., Biochemistry, 1992, 31, 9871-9876. [34] Berkowitz, S. D., Federici, A. B., Blood, 1988, 72, 1790-1796. [35] Kessler, C.M., Floyd, C. M., Frantz, S. C., Orthner, C., Thromb Res, 1990, 57, 59-76. [36] Joao, H.C., Scragg, I.G., Dwek, R.A., FEBS Lett, 1992, 307, 343-346.
128
5 The Application of Glycobiology
[37] Varki, A., Glycobiology, 1992, 2, 25-40. [38] Schauer, R., Trends Biochem Sci, 1988, 10, 357-360. [39] Weis, W., Brown, J. H., Cusack, S., Paulson, J. C., Skehel, J. J., Wiley, D. C., Nature, 1988, 333, 426-43 1. [40] Roelcke, D., Hengge, U., Kirschfink, M., Vox Sang, 1990, 59, 235-239. [41] Koerner, T.A. W., Weinfeld, H. M., Bullard, L. S. B., Williams, L.C. J., Blood, 1989, 74, 274-284. [42] Berg, E. L., Robinson, M. K., Mansson, O., Butcher, E. C., Magnani, J.L., J Biol Chem, 1991, 266, 14869-14872. [43] Larsen, G.R., Sako, D., Ahern, T. J., Shaffer, M., Erban, J., Sajer, S.A., Gibson, R.M., Wagner, D. D., Furie, B. C., Furie, B., J Biol Chem, 1992, 267, 11104-11110. [44] Phillips, M. L., Nudelman, E., Gaeta, F. C. A., Perez, M., Singhal, A. K., Hakornori, S., Paulson, J.C., Science, 1990, 250, 1130-1132. [45] Bleil, J. D., Wassarman, P. M., Proc Natl Acud Sci USA, 1988, 85, 6778-6782. [46] Noguchi, S., Nakano, M., Eur J Biochem, 1992, 209, 883-894. [47] Wassarman, P. M., Development, 1990, 108, 1-17. [48] Hokke, C. H., Damm, J. B. L., Penninkhof, B., Aitken, R. H., Kamerling, J. P., Vliegenthart, J.F.G., Eur J Biochem, 1994, 221, 491-512. [49] Miller, D. J., Macek, M.B., Shur, B.D., Nature, 1992, 357, 589-593. [SO] Cumming, D. A., Glycobiology, 1991, I , 115-130. [Sl] Parekh, R.B., Dwek, R.A., Thomas, J.R., Opdenakker, G., Rademacher, T. W., Wittwer, A. J., Howard, S.C., Nelson, R., Siegel, N.R., Jennings, M.G., Biochemistry, 1989, 28, 7644-7662. [52] Montreuil, J., Vliegenthart, J. F. G., Schachter, H., in: Glycoproteins. New Comprehensive Biochernistry, Vol. 29a, Neuberg, A. and Van Deenen, L. L. M. (Eds.), Amsterdam: Elsevier, 1995. [53] Messner, P., Sleytr, U. B., Glycobiology, 1991, 1 , 545-551. [54] Luckow, V. A,, Summers, M.D., Bionechnology, 1988, 6, 47-55. [55] Davidson, D. J., Fraser, M. J., Castellino, F. J., Biochemistry, 1990, 29, 5584-5590. [56] Lowe, J. B., Kukowska-Latallo, J. F., Nair, R. P., Larsen, R. D., Marks, R. M., Macher, B. A., Kelly, R. J., Ernst, L. K., J Biol Chem, 1991, 266, 17467-17477. [57] Stanley, P., MoZ Cell Biol, 1989, 9, 377-383. Biol Chem, 1990, [58] Smith, D.F., Larsen, R.D., Mattox, S., Lowe, J. B., Cummings, R.D., .I 265, 6225-6234. [59] Stanley, P., Glycobiology, 1991, I , 307-314. [60] Stoll, J., Rosenwald, A,, Krag, S. S., Nuto AS1 Sel; 1990, H40, 151-166. [61] Lee, E.U., Roth, J., Paulson, J.C., J Biol Chem, 1989, 264, 13848-13855. [62] Potvin, B., Kumar, R., Howard, D. R., Stanley, P., J Biol Chem, 1990, 265, 1615-1622. [63] Takeuchi, M., Kobata, A., Glycobiology, 1991, 1, 337-346. [64] Nakao, H., Nishikawa, A,, Karashuno, T., Nishiura, T., Iida, M., Kanayama Y., Yonezawa, T., Tarui, S., Tanaguchi, N., Biochem Biophys Res Commun, 1990, 172, 1260-1266. [65] Howard, D. R., Fukuda, M., Fukuda, M.N., Stanley, P., J Biol Chem, 1987, 262, 168301687. [66] Stanley, P., Trends Genet, 1987, 3, 77-81. [67] Zeng, Y., Lehrman, M.A., Anal Biochem, 1991, 193, 266-271. [68] Stanley, P., Glycobiology, 1992, 2, 99-107. [69] Barton, N. W., Furbish, F. S., Murray, G. J., Garfield, M., Brady, R. O., Proc Natl Acad Sci USA, 1990, 87, 1913-1916. [70] Barton, N. W., Brady, R. O., Dambrosia, J. M., Di Bisceglie, A.M., Doppelt, S. H., Hill, S. C., Mankin, H. J., Murray, G. J., Parker, R. I., Argoff, C. E , N Engl J Ged, 1991, 324, 14641470. [71] Stahl, P. D., Rodman, J. S., Miller, M. J., Schlesinger, P. H., Proc Natl Acad Sci USA, 1978, 75, 1399-1403.
References
129
[72] Hahn, T. J., Goochee, C.F., J Biol Ckern, 1992, 267, 23982-23987. [73] Hayter, P.M., Curling, E. M. A., Baines, A. J., Jenkins, N., Salmon, I., Strange, P. G., Tong, J. M., Bull, A. T., Biotecknol Bioeng, 1992, 39, 327-335. [74] Goochee, C. F., Monica, T., Biotecknology, 1990, 8, 421-427. [75] Elbein, A.D., FASEB J, 1991, 5, 3055-3063. [76] Srivastava, G., Kaur, K. J., Hindsgaul, O., Palcic, M.M., J Biol Ckem, 1992, 267, 2235622361. [77] Warren, C. E., Curr Opin Biotecknol, 1993, 4, 596-602.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
6 The Release of Intracellular Bioproducts Anton P. J. Middelberg
6.1 Introduction Biological molecule recovery is usually done from the aqueous phase using techniques such as chromatography (see Volume 1). These techniques require that the molecule is available for recovery and consequently is not encased in a cell structure. For this reason, one of the first steps in many bioprocesses is the release of protein from the cell cytoplasm to the suspending medium. Protein excretion technology is developing rapidly for organisms such as Saccharomyces cerevisiae and Escherichia coli that have a limited natural capacity for transferring biomolecules from the cytoplasm to the suspending medium. However, the approach is often product-specific and is subject to disadvantages, including the need to fuse a signal sequence to the natural protein. Consequently, the usual approach to product release involves a method of cell-wall disruption prior to subsequent downstream processing. Available techniques range from chemical treatment to disorder the wall, to non-specific mechanical methods that physically tear the cell wall apart. This chapter focuses on the release of intracellular bioproducts using such methods.
6.2 Cell Wall Destruction Destruction of the cell wall is aided by a knowledge of wall structure (Section 6.2.1), allowing effective strategies for cell disruption to be defined (Section 6.2.2). The effectiveness of a particular strategy can then be gauged using a suitable measurement technique for cell disruption (Section 6.2.3).
6.2.1 Cell Wall Structure Gram-negative bacteria have a relatively simple wall structure. E. coli is typical of many Gram-negative bacteria, and is perhaps the best studied. Its wall consists of
132
6 The Release of lntracellular Bioproducts n
n
n
Common Antigen ( E C A ) Lipopolysaccharide (LPS, Endotoxin, O-Antigen )
Phospholipid Lipoprotem
Peptidoglycan ( Murein )
Fig. 6-1. Schematic representation of the Gram-negative bacterial cell wall. (From Rietschel, E. Th., Brade, H., Brade, L., Kawahara, K., Luderitz, Th., Schade, U., Tacken, A., Ziihringer, U. (1986), in: Biological Properties of Peptidoglycan. Berlin: Walter de Gruyter Co., 1986; p. 341 Reproduced with the kind permission of Walter de Gruyter Co.)
two basic layers: the outer membrane and a peptidoglycan layer. The outer membrane comprises mainly lipopolysaccharide (LPS), phospholipid (PL), and lipoprotein (LP), arranged as shown in Fig. 6-1. Note that divalent cations such as Ca2+ and Mg2+ stabilize the outer wall structure by cross-linking the LPS molecules. The outer membrane provides a primary physical barrier retaining molecules such as proteins within the periplasmic space. It is a major barrier to free diffusion and hence the excretion of proteins to the extracellular space. The peptidoglycan layer serves primarily as a stress-bearing layer within the wall: it provides the cell’s mechanical strength. It is considered permeable to protein molecules and salts and therefore is not a major diffusion barrier. Peptidoglycan consists of glycan chains cross-linked by peptide bonds [l]. Recent evidence suggests the E. coli wall has essentially one stress-bearing layer of peptidoglycan across 75-80 % of the cell’s surface and localized multi-layered patches [ 2 ] .A simplified representation of E. coli peptidoglycan structure is shown in Fig. 6-2. In addition to the wall, a thin lipid membrane separates the culture medium and the wall from the cell cytoplasm. This cytoplasmic membrane has little mechanical strength and disrupts when the wall is removed or compromised unless cells are osmotically stabilized. It is a major diffusion barrier separating the cytoplasmic contents from the periplasm and hence the extracellular fluid. Another Gram-negative host of interest for product release in bioprocessing is Alcaligenes eutrohpus. This host has been extensively studied for the production of polyhydroxyalkanoates, which have applications as biodegradable polymers. The wall structure is typical of Gram-negative bacteria, comprising an outer membrane, a peptidoglycan layer, and a cytoplasmic membrane.
6.2 Cell Wall Destruction
133
Fig. 6-2. Artist’s view of enzyme complex involved in E. coli peptidoglycan synthesis. Parallel glycan chains cross-linked by peptide bonds are clearly visible. (From Holtje, J.-V. (1993), in: Bacterial Growth and Lysis: de Pedro, M. A., Holtje, J.-V., Loffelhardt, W. (Eds.). New York: Plenum, 1993; p. 425. Reproduced with the permission of Plenum Publishing Corporation.)
Gram-positive bacteria have a wall that lacks the outer membrane, and hence a defined periplasmic space. The peptidoglycan layer is considerably thicker than in Gram-negative bacteria, constituting up to 90 % of the cell-wall mass (compared with 5-20% in Gram-negative bacteria). Minor components of the wall include the cytoplasmic membrane and acidic polysaccharides (teichoic acids) attached to the outer wall. The increased peptidoglycan content makes Gram-positive bacteria difficult to disrupt. Fortunately, many are natural protein secretors. Like bacteria, yeast such as S. cerevisiae have an inner membrane that separates the cell cytoplasm from the cell wall and the culture medium. However, the similarities end at this point. The wall of S. cerevisiae is more complex than that of E. coli, and considerably harder to break. It comprises up to 30% of the cell mass. Recent reviews summarize the current state of knowledge regarding yeast wall structure [3,4]. There appears to be three key layers. Mannan linked with protein (mannoprotein) forms an outer layer that is susceptible to proteolytic attack. It is associated with a central core of alkali-soluble glucan that has an amorphous appearance under electron microscopy, and may confer flexibility to the wall. Finally, an electron-dense layer borders the inner membrane; this is probably alkali-insoluble p1,3 glucan with a high degree of polymerization and some p1,6 links. This layer has a fibrillar appearance under electron microscopy, and has a direct role in maintaining wall rigidity and shape. It is this glucan layer that is believed to provide cell mechanical strength, possibly aided by the amorphous-like central glucan layer, while the external protein layer provides the main permeability barrier for macromolecules. It is worth noting that the structure of the cell wall is highly dynamic, and changes in response to a wide variety of growth and environmental conditions as well as the cell cycle [4]. Furthermore, the boundary between the mannoprotein and glucan layers is not clearly defined. There is evidence that the mannoproteins are interwoven into the glucan matrix, and in some species may penetrate the entire wall [4].
134
6 The Release of Intracellular Bioproducts
6.2.2 Strategies for Cell Disruption The release of intracellular bioproducts requires destruction of the barrier separating the product from the extracellular fluid. This process of compramising the cell wall is often called cell disruption. A knowledge of cell wall structure, as given in Section 6.2.1, suggests various selective methods of compromising cell-wall integrity. The primary diffusion barriers in Gram-negative bacteria such as E. coli are the inner and outer membranes. Chemicals that selectively destroy or disorder these membranes, such as detergents and chaotropes, have been used with varying success. Given their specificity, they often disrupt either the inner or outer membrane, but rarely both. Combinations of chemicals are therefore gaining increasing interest. Another approach is to destroy selectively the strength-bearing peptidoglycan in the wall through enzymatic treatment (e.g., lysozyme). Destruction of the peptide cross-links or the glycan chains in peptidoglycan will compromise cell integrity and release intracellular products due to the cell’s high osmotic pressure. However, access to the peptidoglycan is limited by the protective outer membrane, so this must often be destabilized prior to enzymatic treatment. The induction of endogenous enzymes that degrade the wall through chemical or physical treatments (e.g., mild temperature rise, mild osmotic shock) is also possible. Certain chemicals (e.g., antibiotics) also inhibit critical growth functions, thus leading to cell lysis with some release of intracellular species. Simple physical processes for cell disruption such as severe thermal treatment and explosive decompression are also possible. As for Gram-negative bacteria, a range of methods is available to break the yeast cell wall and thus release the intracellular contents. The primary diffusion barriers are the cytoplasmic membrane and the mannoprotein complex in the cell wall. These are difficult to compromise, and chemical treatments generally only permeabilize yeast cell walls. While these treatments facilitate the entry of small molecules or probes into the cell (e.g., for in situ assays of enzymatic activity), they are generally unsuitable for large-scale protein release. Selective enzymatic degradation of the yeast wall by successive protease and glucanase attack has, however, been successfully used. This approach uses knowledge of cell wall structure to establish a suitable combination of degrading enzymes. In general, selective methods of protein release have proved less than satisfactory for the release of intracellular bioproducts. The complexity of the cell wall limits the amount of protein recovered, and these methods are generally used only in specific cases. Combinations of specific treatments must often be established, and results usually vary with the state of the target cell wall. However, increasing research in this area is leading to new combinations of general applicability, particularly for E. coli (see Section 6.3). To overcome the limitations of selective methods, mechanical methods of cell disruption have received widespread use in bioprocessing. Available methods at process scale include high-pressure homogenization and its derivatives, and bead milling. These methods apply brute force to tear apart the wall components non-selectively, and are often capable of effecting complete product release in a contained way and without the need for added chemicals.
6.2 Cell Wall Destruction
135
6.2.3 Quantifying Cell Disruption It is clear from the preceding section that a variety of strategies are available for cell disruption, and each affects the cell-wall structure in a unique way. These different methods of release have lead to different methods of quantifying cell disruption, which may be conveniently categorized as direct or indirect.
6.2.3.1 Direct Measurement of Disruption Conceptually, the simplest method to determine cell disruption is to establish directly how many cells are destroyed during treatment. This can be done by directly counting the number- or volume-fraction of cells destroyed, for example by microscopy. Microscopic observation can be aided by stains that test cell-wall integrity. Nevertheless, it remains tedious even with automated image analysis software. Direct counting may be automated to overcome this. For example, methylenetblue dye exclusion and automatic cell counting using a hemocytometer can be used to quantitate cell disruption [ 5 ] . Several researchers have used particle size analyzers to monitor cellular disruption, and also to monitor the size of fragments following disruption (see Section 6.9.1). Elzone particle sizer analyzers (e.g., Coulter counters) are popular, and clearly show the shift to smaller particle size as cells are destroyed. However, they are prone to operational problems such as orifice blocking by cellular debris. Further, accurate quantitation of cellular disruption is difficult due to convolution of the cell and debris peaks. The analytical disc centrifuge has been successfully employed to quantitate E. coli disruption accurately, even in the presence of recombinant inclusion bodies [6]. This is a high-resolution technique that determines disruption by ratioing the cell-peak areas before and after disruption. It is not, however, useful for yeast because of excessive overlap of the cell and debris peaks. Another possibility is to use live cell counts, thus giving an indication of how many cells are inactivated by treatment. Of course, this only gives a measure of cell inactivation and does not guarantee that intracellular contents are actually released.
6.2.3.2 Indirect Methods Indirect methods estimate the volume fraction of cells destroyed by directly quantitating the release of specific intracellular ‘markers’, such as the protein of interest. Although these methods provide indirect measures of disruption, they do provide a direct measure of the protein of interest. The most common approach measures total cell protein in the sample supernatant using dye-binding assays that are readily available (e.g., the Bradford, Lowry, and BCA assays). The fractional release of protein, R,, is given by Eqn ( l ) ,
136
6 The Release of Intracellular Bioproducts
where C is the concentration of soluble protein in the supernatant of homogenate (h) or feed (0)samples, and C, is the maximum possible release of soluble protein, corresponding to complete cellular disruption. For highly concentrated suspensions, it may be necessary to use a dilution technique to correct for the change in liquidphase volume fraction as cells are destroyed [7]. Direct measurement of protein concentration using the dilution technique may be unsatisfactory when protein denaturation occurs. A mass-balance approach based on total Kjeldahl nitrogen has been developed to overcome this limitation [S]. Although less accurate than the dilution technique, it is particularly useful when protein is denatured. A range of other specific methods is available for inferring disruption. Several researchers assay for the release of specific enzymes from different cellular locations (e.g., periplasmic, cytoplasmic, cell-wall associated). Monitoring these enzymes gives some indication of the breakdown of cellular structure (e.g., the release of cell-wall-associated enzymes is normally slower, and high levels of release usually indicate an advanced level of cellular breakdown and hence disruption). Direct measurement of product release, for example by HPLC or ELISA, is also possible. In all cases, it is standard to calculate fractional release according to Eqn (1). Correction for changes in the volume fraction, as for dye-binding assays, will generally be necessary.
6.2.3.3 Selecting a Method No particular method of monitoring cell disruption is ideal for all cases, and choice is often made on the basis of simplicity, cost effectiveness, equipment availability, and personal choice. In general, indirect measurement using dye-binding is commonly employed as it is fast, cheap, and requires no specialized equipment beyond a laboratory centrifuge and a spectrophotometer. Direct and indirect measures generally give similar results when product degradation is minimal. It is worth mentioning that indirect measures of cellular disruption, such as dyebinding assays, have lowest accuracy at high levels of disruption [9]. The opposite is true of direct methods, which have highest accuracy as disruption approaches 100 %. Direct measurement is therefore preferred when accurate measurements are required, as in modeling and optimization. Also, product degradation is sometimes observed under harsh disruption regimes. Direct determination of cellular disruption will allow the independent processes of cellular disruption and product degradation to be deconvoluted. However, direct disruption measures are also prone to problems. Specifically, cells that appear intact under microscopy may have actually released some intracellular contents. Similarly, cell wall permeability to integrity-testing dyes does not guarantee that cellular contents have actually been released. In all cases, a reasonable degree of caution must be exercised.
6.3 Chemical Disruption
137
6.3 Chemical Disruption As mentioned in Section 6.2.2, chemical treatments have been most effective at releasing protein from Gram-negative bacteria such as E. coli. Chemical treatment of the yeast wall generally only leads to permeabilization without significant protein release, restricting their use to in situ enzymatic assays. Methods of permeabilizing the yeast cell wall have been detailed previously [lo], so will not be discussed further here.
6.3.1 Antibiotics Antibiotics are known for their ability to lyse bacteria and prevent further growth, primarily by interfering with the biosynthetic capacity of growing cells [ 111. Antibiotics interfere with peptidoglycan synthesis and assembly in Gram-negative bacteria such as E. coli, thus permitting enhanced action of endogenous enzymes that degrade peptidoglycan. Although lysis is rapid, the cost of antibiotics is prohibitive at large scale and results depend on the state of the culture. Most antibiotics, for example, do not lyse stationary phase bacteria.
6.3.2 Chelating Agents Section 6.2.1 shows that the cell wall of Gram-negative bacteria is stabilized by divalent cations such as Mg2+ and Ca2+ that cross-bridge adjacent LPS patches on the outer membrane. Chelating agents such as ethylenediamine tetra-acetic acid (EDTA) have a unique ability to bind these ions, causing rapid release of significant amounts of LPS [12]. Activity is highest in the presence of Tris buffer because destabilization requires replacement of the cations with large monovalent organic amines [13]. EDTA is primarily of use in releasing periplasmic proteins, or in combination with other treatments such as lysozyme (see Section 6.4), as it has no effect on the cytoplasmic membrane nor the cell-wall peptidoglycan. Periodic EDTA treatment (0.5 to 3.0 mM) has been employed to effect the in situ release of soluble periplasmic P-lactamase from immobilized recombinant E. coli without significant loss of
cell viability [14].
6.3.3 Chaotropic Agents Chaotropic agents such as guanidium salts and urea disrupt the structure of water, weakening solute-solute interactions. They lead to effective solubilization of integral membrane proteins. The requirement for high chaotrope concentrations generally limits their use at process scale.
138
6 The Release of Intracellular Bioproducts
Synergistic effects are seen when chaotropes are employed with other chemicals. For example, protein release from E. coli is possible with relatively low concentrations of guanidine-HCl (up to 4 M) in the presence of low concentrations (up to 2 %) of the non-ionic detergent Triton X-100 [15]. At very low gu-HC1 concentrations (ca. 0.1 M), a direct synergistic effect is seen. Protein release occurs due to molecular alteration of the outer wall and solubilization of the cytoplasmic membrane. However, some resistance to total protein release remains after treatment, limiting the total amount of protein that can be extracted. A subsequent study demonstrated that high protein release (> 75 %) from exponentially growing cells could be obtained using 0.4 M gu-HCI and 0.5 % Triton X-100 [161. Chaotrope (2 M urea) in the presence of a reducing agent (10 mM dithiothreitol or 50 mM cysteine) has also been used to recover insulin-like growth factor fusion protein expressed as periplasmic inclusive bodies [ 171. Under alkaline conditions (pH lo), approximately 90% protein recovery is achieved. The method was coupled with a direct two-phase extraction to provide initial purification from contaminant protein and insoluble cellular debris. Chaotrope action can also be enhanced through the synergistic effects of chelating agents. Urea combined with EDTA can release cytoplasmic protein from E. coli at the same level as mechanical disruption [ 181. Presumably, EDTA disrupts the outer membrane in conjunction with urea providing the chaotrope with access to the cytoplasmic membrane. This approach has been applied to achieve the in situ solubilization of recombinant inclusion bodies in E. coli, and has been extended to enable selective release of soluble cytoplasmic contaminants prior to in situ dissolution of the inclusion body, thus providing crude initial purification (R. J. Falconer, Dissertation in preparation, University of Adelaide). The treatments are effective against exponential and stationary phase E. coli, and achieve the same recovery as the traditional mechanical disruption and in vitro dissolution and refolding approach.
6.3.4 Detergents Detergents interact with lipid components in cell membranes, leading to their solubilization. They have been used with varying success to extract proteins from E. coli as outer membrane LPS provides partial protection, particularly by restricting detergent access to the cytoplasmic membrane. At low concentration, detergents generally only solubilize membrane proteins from prepared wall samples (e.g., by mechanical disruption) and are particularly useful when the desired protein product is associated with the cell wall. Protein release from whole E. coli cells using the anionic detergent sodium dodecyl sulfate (SDS) at high concentration (1 mg SDS per 5-11 OD600 absorbance units of cell mass) has been achieved [ 191. Fractional protein release was not reported. Complete SDS removal was achieved by adsorption onto zeolite Y. The SDS offered protection against proteases, presumably through selective denaturation. The synergistic use of the non-ionic detergent Triton X-100 and chaotrope has also had some success (see Section 6.3.3).
6.4 Enzymatic Disruption
139
6.3.5 Alkaline Treatment Alkaline lysis using hydroxide and hypochlorite is an effective and cheap method for releasing intracellular products and acts by saponifying the cell-wall lipids. The technique is extremely harsh and consequently the product must be resistant to degradation at high pH. This limits the applicability for protein recovery. However, alkaline treatment has received widespread interest for the recovery of polyhydroxyalkanoates (PHA, a biodegradable polymer) from Gram-negative bacteria such as E. coli and A. eutrophus. PHA recovery from A. eutrophus has been achieved using sodium hypochlorite [20], but with a 50% reduction in the molecular weight of the polymer. This has been overcome by combining alkaline lysis with chloroform extraction of the polymer, using dispersions of the two chemicals [21]. Under optimal conditions a purity in excess of 97 % and a recovery of 9 1 % was obtained. Minimal product degradation occurred because of the protective effect of chloroform. PHA recovery from A. eutrophus using alkaline treatment in the presence of detergents has also been attempted [22]. Treatment with 1 % Triton (pH 13) followed by a brief hypochlorite treatment gave a purity in excess of 98 %, but with significantly higher product degradation than the hypochlorite-chloroform dispersion method. Hypochlorite extraction of PHA has been extended to recombinant E. coli giving high product purity with minimal product degradation and no detectable change in polymer granule size [23,24]. Alkaline treatment prior to mechanical disruption by high-pressure homogenization has also been employed for PHA recovery from A. eutrophus (see Section 6.6.3).
6.4 Enzymatic Disruption The addition of foreign lytic enzymes that selectively degrade the cell wall is a gentle yet powerful method for effecting protein release from a range of organisms. For Gram-negative bacteria, the approach is to use enzymes such as lysozyme that attack the strength-providing peptidoglycan in the wall. Enzymatic access to the peptidoglycan is aided through disruption of the outer membrane, for example using EDTA (see Section 6.3.2). By removing the strength-bearing element, cells that are not osmotically stabilized will lyse and release their cytoplasmic contents. The enzymatic degradation of yeast is somewhat more complex, as a result of the layered design of the cell wall (see Section 6.2.1). Complex enzyme systems comprising proteases that degrade the mannoprotein complex and glucanases that degrade the strength-providing glucan network are required. Treatment with a simple glucanase leads to minimal disruption because of the protective action of the outer mannoprotein complex. Available enzyme systems for yeast have been reviewed [25] and include relatively inexpensive commercial preparations such as Zymolyase 20 -T from continuous culture of Oerskovia xanthineolytica (Seikagaku America Inc., Rockville, MD) and lytic systems from Cytophaga and Rhizoctonia. Cell-wall break-
140
6 The Release of lntracellular Bioproducts
down is aided by thiol reagents that activate endogenous proteases and break disulfide bonds in the cell wall. Release of cytoplasmic protein from stationary-phase E. coli has been achieved using 25-50 pg mL-' lysozyme in combination with 100-800 pg mL-' EDTA [26]. Enhanced release was obtained at 58 "C although the temperature optimum for lysozyme is 35 "C, presumably because of increased outer-membrane disruption at the higher temperature and hence improved lysozyme penetration. Combinations of 5-30 pg mL-' polymyxin and lysozyme also released protein from exponential cells. Polymyxin is a cationic polypeptide antibiotic with an aliphatic chain that disorganizes and penetrates the outer membrane, providing access for the lysozyme to the peptidoglycan. The disruption of yeast cells by enzymatic treatment has been modelled extensively [27]. Disruption occurs in a step-wise fashion with protease attack of the outer wall preceding glucanase attack. When a sufficient amount of glucan has been solubilized, the inner membrane ruptures, releasing the cytoplasmic contents. Breakdown of subcellular structures follows lysis. Lysozyme is available relatively cheaply from egg-white preparations, but enzyme cost can still be prohibitive for large-scale applications. This is also the case for commercial systems for yeast digestion. Immobilization of enzymes is a practical way of reducing costs, but is limited by a loss of enzyme activity and inaccessibility of the substrate to enzyme within the pores of solid support matrices. Lysozyme has been immobilized on smooth fibres to improve substrate access [28], but the efficacy of this approach was limited by a rapid loss of enzyme activity. Further work may lead to effective immobilization methods that are generally applicable and suitable for process-scale use.
6.5 Physical Methods of Cell Disruption Microorganism disruption can also be achieved through physical treatments including supercritical extraction, explosive decompression, and thermal treatment. Protein may be released from E. coli by exposing the outer membrane to elevated temperatures. Periplasmic proteins are released at 50-55 "C [29] while cytoplasmic proteins are released within 10 minutes at 90°C [30]. Results depend on the strain and in particular its outer membrane characteristics, the rate of heating, and the temperature at which the organisms are stored prior to thermolysis. Thermolysis gives large cell debris, but may also lead to unacceptable rises in viscosity due to DNA denaturation [30]. Mild thermal treatment designed to deactivate E. coli prior to homogenization can be detrimental to homogenization efficiency (see Section 6.6.3). Mild heat treatment can also lyse yeast and release intracellular components through autolysis rather than direct thermal destruction. Treatment at 45-50 "C and pH 5.5 leads to the production of glucanases, proteases, and mannanases endogenous to viable yeast cells. These enzymes act in concert to degrade the wall causing protein release within 24-36 h [31]. This rate of release may be too slow for recombinant products, particularly due to the higher levels of proteases resulting
6.6 High-pressure Homogenization
141
from mild heat treatment, but the technique is routinely employed in preparing yeast hydrolysates. Explosive decompression has been used to extract product from yeast [32]. Supercritical carbon dioxide is contacted with cell suspension for a specified time and expands rapidly when the pressure is released, causing disruption of the wall. The technique is extremely gentle and extracts off-flavors caused by lipid components, but has a relatively low efficiency.
6.6 High-pressure Homogenization High-pressure homogenization is a non-specific mechanical method of effecting cell breakage at large scale. It is the most common disruption method for bacteria such as E. coli. The usual design employs a positive displacement pump that forces cell suspension at high pressure through a spring-loaded valve. As shown in Fig. 6-3, the suspension accelerates to high velocity in the valve with a concomitant drop in pressure, before impinging on an impact ring and leaving the system at basically atmospheric pressure. The precise mechanism of disruption is unclear. However, it is likely that disruption occurs in response to shear, decompression, impingement, and possibly turbulence and cavitation. Recent evidence suggests that the pressure gradient experienced by the cells is a strong positive correlator of the extent of cell disruption [33,34].
I
I\\\\y
'IMPACTRING
I
Fig. 6-3. The basic process of cell disruption by high-pressure homogenization. (Reproduced from Pandolfe, W.D., Cell Disruption by Homogenization, APV information booklet, with the kind permission of APV.)
6.6.1 Operational Parameters Key operational parameters in high-pressure homogenization are operating pressure and the number of times that the suspension passes through the homogenizer. The valve design also has a critical effect on disruption efficiency (see Section 6.6.5). Disruption is generally improved by raising the inlet temperature, but at the expense
142
6 The Release of Intracellular Bioproducts
of an increased exit temperature due to heating during homogenization. In general, a temperature rise of approximately 2.4"C per 10 MPa of operating pressure can be expected. Denaturation of DNA at higher exit temperatures (e.g., > 40°C) can cause problems in subsequent processing, although DNA can be sheared using a second 'high-shear' homogenizing valve in series with the primary disruption valve. These two-stage machines are no longer widely employed because of higher capital and operating cost, especially as viscosity problems can be reduced through judicious choice of inlet temperature. Cell concentration also has a minor effect on disruption, with decreased disruption at higher feed concentrations [6,35]. The decreased disruption efficiency at high concentration is small compared with the increase in processing time caused by dilution. It was concluded that the optimal E. coli feed concentration is the maximum possible that does not lead to practical handling difficulties due to high viscosity.
6.6.2 Commercial Equipment There are two main suppliers of commercial homogenizer equipment. APV Baker homogenizer division comprises APV Rannie (Copenhagen, Denmark) and APV Gaulin (Massachusetts, USA), and produces homogenizers operating at pressures up to 1500 bar. Niro-Soavi (Parma, Italy) produces machines with operating pressures up to 1000 bar at flowrates up to 9000 L h-l. Higher pressures are possible for custom-manufactured machines (personal correspondence, GEA Australia). Figure 6-4 shows a typical cell-disruption homogenizer. Each manufacturer provides a range of valve designs in various materials, such as ceramic and tungsten carbide. Ceramic valves are particularly recommended when abrasive suspension such as homogenate containing recombinant inclusion bodies must be processed. Both manufacturers offer a range of laboratory machines available for process testing by the potential client. It is, however, worth noting that no reliable scale-up rules for high-pressure homogenization have yet been developed. The performance of a laboratory-scale machine will typically exceed that of a full-scale production machine as valve radii and lift distances (the distance between the valve and seat when open) will be higher in the production machine, giving lower impact velocities and pressure gradients.
6.6.3 Cell Treatments Before Homogenization Upstream influences can have a devastating impact on cell disruption during homogenization. For example, batch thermal deactivation of stationary-phase E. coli prior to homogenization reduced disruption efficiency at 55 MPa to 25 % from 80 % [36] because cells became smaller and tougher following the thermal treatment. This effect could be negated by charging glucose to the fermenter prior to initiating thermal deactivation.
6.6 High-pressure Homogenization
143
Fig. 6-4. The Niro Ariete NS 301 1 high-pressure homogenizer. (Reproduced with permission from a photograph kindly provided by GEA Process Technology Division.)
Beneficial effects can be achieved through selective treatment of cells prior to homogenization. Brief alkaline treatment of A. eutrophus containing PHA at pH 10.5 prior to homogenization increased soluble protein release by 37% [37]. Pretreatment with the detergent Sarkosyl at 1 % also improved soluble protein release by 38 % following homogenization through synergistic effects (release was minimal prior to homogenization). Enzymatic pretreatment can also aid mechanical disruption. For example, Vogels and Kula [38] have demonstrated that pretreatment of Bacillus cereus with lytic enzyme (0.5 mg cellosyl g-' wet cells) prior to homogenization in a Gaulin MC4-TBX homogenizer raises disruption from 40 % to 98 % after a single homogenizer pass at 70 MPa. The homogenization of E. coli is also aided by chemical pretreatment. For example, pretreatment of exponential-phase cells with 1.5 M guanidine HC1 and 1.5 % Triton X-100 has been shown to significantly improve disruption 1391.
144
6 The Release of Intracellular Bioproducts
6.6.4 Predicting Disruption High-pressure homogenization data may be described using Eqn (2), which was originally developed for yeast [7]. Soluble protein release, RR is a function of the number of homogenizer passes, N , the homogenizer pressure, P, and two empirical constants, a and k. log
(&)
=kNP
The parameters a and k are dependent on a range of factors, including the type of cell and its growth phase, the growth media composition, the design of the homogenizer and homogenizing valve, and operating parameters such as temperature and to a lesser extent cell concentration. Some studies also demonstrate a dependence of k on P, particularly at high levels of disruption. Table 6-1 gives values for parameters a and k from a range of studies. Despite the large variation in parameter values, Eqn (2) continues to be widely used because of its relative simplicity. The wall-strength model was developed to address the dependence of cell parameters on feed properties [9,40]. Disruption for multiple passes is calculated by Eqn (3), m
D = 1 - S(1 - f o ( m N f s ( s ) d S
(3)
0
where f,(S) and &(S) are the stress and strength distribution functions, respectively.
Table 6-1. Typical model parameters for equation ( 2 ) for common microorganisms from a range of homogenizer studies. Organism
k (MPaa)
a
Reference
S. cerevisiae coli (stationary) coli (exponential) coli (induced)b eutrophus
9.7 x 1.5 x 0.38 4.5 x 4.6 x
2.9 1.71 0.64 1.65 1.7
[I1 t401
E. E. E. A. a
10-6
10-3 10-3 104
All values based on natural logarithm in equation ( 2 ) . Cells containing recombinant protein inclusion bodies.
~401 161 t311
6.6 High-pressure Homogenization
145
Table 6-2.Model parameters for the wall-strength model (equations (3) to ( 5 ) )for E. coli [9] and yeasts [41] for P > 35 MPa. Parameter
E. coli
Brewer’s Yeast
Baker’s Yeast
d rn n
7.85 12.6 0.393 33X-8 .OL+48.8 3.82
10.6 12.2YO.18 0.38 51 4
10.6 12.2YO.18 0.38 61 to 64 10 to 14
S 0
There is a total of five parameters. Regression to single-pass disruption data for E. coli (182 data points, 21 different cultures) gave a good description of data when four parameters were constant (i.e., independent of the feed cell properties). Variation in disruption data was accounted for by a single parameter, termed the mean effective strength of the culture, which is correlated with measurable properties of the feed cells (mean cell length, L, and fractional peptidoglycan cross-linkage, X ) as in Table 6-2. The correlation predicts to within 6 %, thus allowing true a priori prediction of disruption for a fixed homogenizer system (APV-Gaulin 15M) independently of variations in the feed cells. The wall-strength model was originally developed for E. coli, but subsequent work demonstrated that it is able to describe the disruption of yeast [41]. Parameter values are provided in Table 6-2. Despite its usefulness in predicting disruption independently of feed cell variation, the wall-strength model remains dependent on assumed functional forms for the strength and stress distributions, and parameters determined by regression. To remove this limitation, fundamental work into defining the true homogenizer-stress and cell-strength distributions is currently being undertaken (A. R. Kleinig, Dissertation in preparation, University of Adelaide).
s,
6.6.5 The Importance of Homogenizer Valve Design The success of homogenization lies in the valve design. A variety of designs are available from APV Baker, as shown in Fig. 6-5. A comparison of these valves showed that the CD design is more efficient than either the standard or CR design. This is probably due to the efficient conversion of pressure to kinetic energy in the valve inlet, and consequent high impingement velocity at the impact ring. The NiroSoavi ‘Nanovalve’ (Fig. 6- 6) is designed to have a larger-than-standard valve radius, thus reducing valve lift at a fixed feedrate. It is claimed that this improves disruption above that achievable with a standard valve. One way of improving disruption efficiency is to reduce the distance between the valve exit and the impact ring. The lateral jet issuing from the valve spreads as it moves toward the impact ring, thus reducing its average velocity. Higher impact velocities are obtained by moving the impact ring closer to the valve. Keshavarz-
146
6 The Release
of Intracellular Bioproducts
Impact Ring
Standard
CR
CD
Fig. 6-5. Various homogenizer valve designs. CD, knife-edge configuration; CD, cell disruption configuration. (Reproduced from Pandolfe, W. D., Cell Disruption by Homogenization, APV information booklet, with the kind permission of APV.)
Fig. 6- 6. The Niro ‘Nanovalve’ high-pressure homogenizer valve. (Reproduced with permission from a photograph kindly provided by GEA Process Technology Division.)
6.7 Bead Milling
147
Moore et al. [42] studied the effect of impact distance for yeast disruption in a highpressure homogenizer. The group kP" in Eqn (2) was correlated with stagnation pressure at the impact ring, P,, which is related to valve gap, h, and impact distance, I: according to Eqn (6): P,
1
K
-
Y2h2
Kleinig et al. [43] examined the effect of impact distance on the disruption of E. coli by high-pressure homogenization. A modified form of Eqn (2) was developed, giving Eqn (71,
where Y* is the dimensionless impact distance (i.e., Y/Y,,d).Values for parameters k and a in Eqn (7) are given in Table 6-1. Recent work has suggested that disruption is favored by large pressure gradients [33]. By combining numerical fluid simulations with experimental data, it was shown that disruption is uniquely correlated with the maximum pressure gradient experienced at the valve inlet when the impact ring is removed. Subsequent work has also shown that disruption can be correlated with the pressure gradient experienced at the impact ring after suspension has left the valve [34], and provided an improved correlating equation superseding Eqn (6). Combined, these studies suggest that the 'optimal' homogenizer valve will be one that achieves a high pressure gradient at the valve inlet and a high impingement velocity at the impact ring. Of course, the valve must also have acceptable wear characteristics.
6.7 Bead Milling Bead mills are relatively simple devices originally developed for the wet grinding of pigments in the paint industry and for the comminution of ceramics and limestone. The basic design is shown in Fig. 6-7. A jacketed grinding chamber is fitted with a rotating shaft through its centre. The shaft is fitted with agitator discs of varying design that rotate at high speed (ca. 4000 r.p.m.) with the shaft. These discs violently agitate glass beads (0.5-1 mm diameter) within the chamber, causing them to collide. Cell suspension is pumped through this violently agitated bead phase, causing cells trapped between colliding beads to be disrupted. Beads are retained in the chamber by a sieve or axial slot at the outlet. Cooling is essential, as virtually all energy input is dissipated as heat. The precise mechanism of disruption in bead milling is not known, although compaction and shearing between the beads and energy transfer from the beads to the cells are believed to be important. Bead milling is normally employed for large organisms such as yeast cells. Smaller cells such as E. coli are more difficult to disrupt by bead milling as they are less easily trapped between the colliding beads.
148
6 The Release of Intracellular Bioproducts inlet
t
outlet
agitator
I
J ,’ drive
Fig. 6-7. Schematic drawing of the grinding chamber of the Netzsch LME 4 bead mill. (Reprinted from Chem. Eng. Sci., 47, Bunge, F., Pietzsch, M., Muller, R., Syldatk, C., Mechanical disruption of Arthrobacter sp. DSM 3747 in stirred ball mills for the release of hydantoin-cleaving enzymes, 225-232 [44], with permission from Elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, OX5 lGB, UK.)
6.7.1 Operational Parameters A large number of parameters affect the results obtained during bead milling. These include agitator design and velocity, bead loading and size, inlet temperature, suspension flowrate, and cell concentration. Table 6-3 summarizes the effect of bead mill operational parameters. In general, high agitator peripheral velocities (5-10 m s-l), high bead loadings (80-90% of free grinding chamber volume) and small beads (ca. 0.5 mm) give the highest levels of disruption during bead milling. Table 6-3. Qualitative observations reported in the literature on the effect of bead milling variables. Variable
Typical values
Qualitative effects on disruption
Bead size
0.2 to 0.8 mm
Smaller beads are better, but are more difficult to retain, and may tend to be fluidized. In practice, moderate sized beads (e.g., 0.5 mm) are preferred
Agitator peripheral velocity
4 to 12 m s-’
Higher velocities cause higher disruption, but may lead to inefficient energy use if stress probability is limiting (see section 6.6.3)
Cell concentration
5-50 % WJV
Only second-order effects reported
Flow rate
Machine specific
Disruption decreases with increasing flowrate
Bead volume fraction
70-90 %
Disruption increases at higher loadings, but agitator wear and heat generation may be unacceptable. 80-85 % provides best compromise
Temperature
5-40 “C
Slight decrease in disruption as temperature increases. Only second-order effects reported
6.7 Bead Milling
149
6.7.2 Commercial Equipment Two large suppliers of bead mill equipment for bioprocessing applications are Willy A. Bachofen AG Maschinenfabrik (Basel, Switzerland), marketing the Dyno-Mill range, and Netzsch Feinmahltechnik (Germany) who sell the LME range. The Dyno-Mill KDL unit is a laboratory prototype, shown in Fig. 6-8. Larger units (e.g., the KDS unit with a 5-L working volume) are also available. Netzsch also manufactures bead mills with a range of working volumes (e.g., the LME4 with a 4-L working volume and speeds of 200 to 2500 r.p.m.).
Fig. 6-8. Schematic diagram of a Dyno Mill KDL bead mill. (Reproduced from the DynoR-Mill KDL information booklet with the kind permission of Willy A. Bachofen AG, Basel, Switzerland.)
6.7.3 Predicting Disruption Limon-Lason et al. [4S] showed that the disruption of S. cerevisiae in batch bead mills could be described by Eqn (8),
150
6 The Release of Intracellular Bioproducts
where, t is the time of disruption and k is a rate constant correlated with agitator velocity. For continuous operation, the bead mill was modelled as a series of j continuous stirred-tank reactors (CSTRs) giving Eqn (9), 1 1-Rp
-= (1
+
5) j
(9)
where z is the mean residence time in the mill. Residence time distribution (RTD) studies showed that the Dyno-Mill KD5 mill was equivalent to 5 CSTRs in series, while the Dyno-Mill KDL unit varied between 1.2 and 2.0 CSTRs. For the KDL unit, k was correlated with the agitator peripheral velocity, v,, according to Eqn (10) independent of the type of agitator design for batch and continuous operation,
k = Kv,
(10)
where K = 0.0036 m-l. Eqn (10) gives k values varying from approximately 0.02 to 0.07 s-l for the KDL unit with typical agitator velocities of 6 to 20 m s-l. For the KD5 mill, k was approximately constant and equal to 0.02 s-l or 0.04 s-l depending on the impeller design but not the agitator velocity. Melendres et al. [46] studied disruption of S. cerevisiae in a Dyno Mill KDL unit and correlated the rate constant, k , with the effective volume between beads, B, and the frequency of bead collisions, 5 giving Eqn (11). The parameter is a constant of proportionality between the bead and agitator velocities. The frequency of bead collisions was related to bead loading, a, bead diameter, db, agitator peripheral velocity, v,, and voidage, F , using the kinetic theory of gases,
where the product
pf was determined by regression to data, giving Eqn (12):
of = 1.05 x
d:57
(12)
Eqn (11) indicates that the rate constant is proportional to agitator velocity for the KDL unit as also demonstrated by Eqn (10). However, this proportionality fails at low agitator velocities and Melendres et al. [ 5 ] concluded that disruption is minimal below some critical agitator velocity equal to 5 m s-' rpm for the KDL unit. An extensive study on the disruption of Arthrobacter sp. DSM 3747 in a Netzsch LME4 bead mill for varying agitator velocities (v,, 4 to 12 m s-'), bead diameters (db, 0.1 to 1.5 mm) and cell concentrations (c, 10 to 55 % w/v) at a bead filling fraction (y) of 0.80 has been conducted [44]. Disruption correlated with either specific energy input (Eqn 13) or stress probability (Eqn 14), depending on the operational regime.
6.7 Bead Milling
l n 1-Rp ( 2 )
15 1
= k E = k ( y )
In the above equations, M is the torque on the agitator shaft, V is the free grinding volume, c is cell concentration, w is the rotational speed of the agitator, dp is the cell diameter, and a is the bead filling rate (weight ratio of beads to yeast). Disruption correlated well with specific energy input, E, for small beads (db = 0.355 mm), moderate to high cell concentrations (c = 10 % to 55 %), and low to moderate agitator speeds (v, = 4 to 8 m s-l, as shown in Fig. 6-9. In this regime, an increase in specific energy input lead to a higher stress intensity during bead collision, and this caused higher disruption. However, the correlation was poorer at small (0.110 mm) and large (1.50 mm) bead sizes, and worsened at higher agitator velocities and lower concentrations. Under these conditions, disruption correlated with stress probability as shown in Fig. 6-10. It appears that at low concentrations and high velocities, stress intensity between colliding beads already exceeded cell strength. An increase in specific energy input and hence stress intensity did not increase disruption. The limiting factor in this regime was the need to bring cells into the active volume of the colliding beads. Disruption therefore correlated with stress probability.
0
100
200
Specific energy
300
(00
( kJ Ikg)
Fig. 6-9. Correlation of disruption with specific energy input to the bead mill for various agitator velocities (v,) and cell concentrations (c). DR is disintegration rate and is equivalent to Rp defined by equation (1). (Reprintedfrom Chem. Eng. Sci., 47, Bunge, F., Pietzsch, M., Muller, R., Syldatk, C., Mechanical disruption of Arthrobacter sp. DSM 3747 in stirred ball mills for the release of hydantoin-cleaving enzymes, 225-232 [44], with permission from Elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, OX5 lGB, UK.)
152
6 The Release of Intracellular Bioproducts 100
-
$ W
60
+
c
.-c0 +
G .-c 2
u
)
4J
4-
20
0 100
5
2
10'
2
Stress frequency
5
101
2
5
10'
( w , t ' ~ ' d , , .d-1 )
Fig. 6-10. Correlation of bead-mill disruption with stress frequency for various bead sizes (4.DR is disintegration rate and is equivalent to R,, defined by equation (1). Stress frequency is equivalent to stress probability, defined by equation (14). (Reprinted from Chern. Eng. Sci., 47, Bunge, F., Pietzsch, M., Muller, R., Syldatk, C., Mechanical disruption of Arthrobacter sp.,DSM 3747 in stirred ball mills for the release of hydantoin-cleaving enzymes, 225-232 [44], with permission from Elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, OX5 lGB, UK.) MECHANICAL SEAL HOUSING
7
DISRUPTEDSLURRY
E
JACKET
4
COOLING WATER
!-INLET FOR CELL SLURRY
Fig. 6-11. Schematic diagram of the Annu Mill 01. (Reproduced from Bzoseparation, 4, (1994), 319-328, Disruption of a recombinant yeast for the release of 8-galactosidase, Garrido, F., Banerjee, U. C., Chisti, Y., Moo-Young, M. Figure 1. 0 1994 Kluwer Academic Publishers [49], with kind permission from Kluwer Academic Publishers.)
6.7 Bead Milling
153
The work by Bunge et al. [44] is important from a practical perspective. Studies suggest high disruption is achieved at high agitator velocities, as described in Section 6.7.1. However, this may represent inefficient use of energy if the limiting factor is stress probability. The most efficient use of energy occurs when disruption correlates with stress intensity (i.e., specific energy input) as a sufficient number of cells are transported in zones of active bead collision. The disruption of S. cerevisiae in a novel bead mill, the Annu Mill 01 shown in Fig. 6-11 (Sulzer Brothers Ltd., Winterthur, Switzerland), has been examined [47]. Protein release was described by Eqn ( 1 5 ) ,
where N is the number of passes through the mill and Q is the flowrate. The rate constant k was empirically correlated with the bead filling fraction and flowrate.
6.7.4 The Importance of Agitator Design Agitator design in bead milling is a key determinant of disruption efficiency. There are two main considerations in selecting an agitator, namely the efficiency of kinetic energy transfer from the agitator to the beads, and the effect that the agitator has on the mill’s RTD in continuous operation. Ideally, kinetic energy transfer will be as high as possible to ensure maximum stress intensity between the colliding beads. This should be achieved while maintaining a narrow residence time distribution (i.e., while maximizing the number of CSTRs in series). A narrow RTD ensures that all cells spend approximately the same time in the bead mill, and thus have an equal probability of disruption. Short-circuiting of cells from the feed to the mill exit is thereby avoided. These two effects may often oppose one another, so an optimum trade-off must be identified. Agitator design has been extensively studied for Netzsch mills [48,49]. For the Netzsch LME 20 mill, a cooled pin agitator (LMJ 15) without stationary counterpins affixed to the mill wall gave higher S. cerevisiae disruption than either a double-disc design (LME 20 ‘D’) or the standard eccentric disc design (LME 20). These designs are shown in Fig. 6-12. Interestingly, the LME 20 ‘D’ design gave a better RTD, but this is clearly offset by improved kinetic energy transfer in the LMJ 15 design. When stationary counter pins are affixed to the mill wall, the RTD of the LMJ 15 design approaches that of an ideal mixer (i.e., a single CSTR) and disruption efficiency is consequently reduced. It is worth noting that although the LMJ 15 design gives higher disruption than the standard agitator, it is also prone to high wear that may increase overall operating costs. A comprehensive study of 12 different agitator designs has also been undertaken for a Netzsch LME 4 mill [49]. Changing the standard eccentric ring arrangement to a 8-notched, two-cone discs arrangement (Fig. 6-13) increased S. cerevisiae disrup-
154
6 The Release of Intracellular Bioproducts
1.o
0 w i 0.8
<.-
c
0.6
0 .-
3
r &, 0.4
a 0 P .-
0.2
w x
Reduced time 8
Netrsch LME 20 mill equipped with a double disc stirrer (LME 20 ‘D), 0
Netzsch LME 20 mill with eccentrically mounted impellers on the drlve shaft (LME 20), 0
Netzsch LMJ 15 mill with a pin agitator (LMJ 15), A
n
Fig. 6-12. Various agitator designs, showing the effect on RTD. (From Kula, M.-R., Schiitte, H. (1987), Biotech. Prog., 3, 311-312. Reproduced with permission of the American Institute of Chemical Engineers. Copyright 0 1987 AIChE. All rights reserved.)
6.8 Other Methods of Mechanical Disruption
155
.@Fig. 6-13. Designs of the eccentric ring agitator (top) and the notched two-cone discs agitator (bottom). (From Schutte et al. [49]; reproduced with permission of the Annals of the New York Academy of Sciences.)
tion substantially. The rate constant k in Eqn (8) increased from 0.41 min-' to 1.52 min-I, and this was attributed to improved energy transfer to the beads. The modified agitator also gave higher disruption for Brevibacterium ammoniagenes, although optimum disruption for this organism was obtained using a 6-notched disc arrangement with a 45" angular offset between discs. In all cases the rate constant for B. ammoniagenes was approximately one order of magnitude below that for S. cerevisiae. The above studies suggest several points that need to be considered when selecting a bead mill and agitator. First, the optimum agitator design is microorganism-specific, necessitating RTD tests and batch tests to assess k for any given application. The aim will be to select a design that gives high segmentation of the mill and a consequent high number of CSTRs while promoting efficient kinetic energy transfer. It will be desirable to operate in the region where disruption correlates with stress intensity to ensure efficient use of energy (see Section 6.7.3). Finally, maximizing disruption through careful agitator selection does not guarantee an optimal solution: the agitator must also have acceptable wear characteristics.
6.8 Other Methods of Mechanical Disruption Another form of mechanical disruption gaining increased interest is the Microfluidizer (Microfluidics Corp., MA, USA), which operates by impacting two streams of cell suspension at high velocity. Disruption of E. coli in a Microfluidizer has been described by Eqn (16),
156
6 The Release of Intracellular Bioproducts COOLING SAMPLE PRODUCT
COOLING JACKET
INLET OUTLET
E HlGl CYL
Fig. 6-14. Schematic diagram of the Constant Systems Ltd. cell disruptor, showing the principle of operation. (From the Constant Systems Ltd. ‘Cell Disruption’ information booklet. Reproduced with the kind permission of Constant Systems Ltd., Warwick, U. K.)
6.9 Downstream Imuacts of Cell Disruution
157
where R, is the fractional release of protein prior to mechanical disruption due to enzymatic pretreatment [SO]. The value of a varied between 0.6 and 1.77 depending on the properties of the feed cells and k varied from 0.27 X to 16 X lod3MPaa. Treatment of E. coli with EDTA-lysozyme prior to disruption in a Microfluidizer marginally improved disruption (from 76 % to 88 % for uninduced cells, and from 86 % to 98 % for induced cells) [51]. Disruption using EDTA-lysozyme treatment without Microfluidization was minimal. S. cerevisiae has also been disrupted in a Microfluidizer following Zymolyase pretreatment 1521. Four passes through the Microfluidizer at 95 MPa gave almost complete disruption, compared with only 32% disruption when enzymatic pretreatment was not employed. Only a very low level of disruption (5 %) was achieved using Zymolyase treatment without mechanical disruption, clearly indicating the synergistic effects of enzymatic pretreatment and Microfluidization. Parameters for Eqn (16) were a = 3.03, b = 1.30, and k = 2.77 x MPa-”. Figure 6-14 shows the general arrangement for a new mechanical disruptor available from Constant Systems Ltd (Warwick, U. K.). High-pressure cell suspension is rapidly accelerated to give a jet with high kinetic energy, that impacts a target before collection at atmospheric pressure. Cooling is provided by a jacket. Disruption presumably occurs in the regions of high pressure gradient (i.e., in forming the jet and impacting the target), as for conventional high-pressure homogenizers. A range of machines from a ‘single-shot’ laboratory unit (10 mL) to units capable of continuously processing 1.1 L min-’ at 1000 bar are available.
6.9 Downstream Impacts of Cell Disruption From the preceding sections it is clear that upstream units and operating conditions affect the extent of cell disruption by altering cell-wall properties. Similarly, disruption conditions can affect subsequent downstream operations, particularly chromatography. Insoluble cellular debris fouls chromatographic resins and complicates process validation. Further, cell-wall proteases associated with insoluble cellular debris may destroy the product in downstream operations such as dissolution and refolding. This has been demonstrated for analogs of insulin-like growth factor expressed as inclusion bodies in E. coli [53]. To minimize these problems it is common to separate cellular debris from the product prior to chromatography, usually by centrifugation or filtration. Optimization of these units is aided by knowledge of the debris size distributions generated during disruption.
6.9.1 Debris Size Analysis Several studies have addressed the issue of debris size following mechanical disruption. Key results are summarized in Table 6 - 4 for E. coli. All available methods suffer from disadvantages. A common problem is the need for extensive sample pre-
158
6 The Release of Intracellular Bioproducts
Table 6-4. Summary of results from studies on the effect of mechanical disruption on E. coli debris size. ~
Conditions
Sizing methoda
Homogenizer, N = 1, P = 55 MPa
PCS and CDS
0.43
Homogenizer, N = 5 , P = 55 MPa
PCS and CDS
0.28
Homogenizer, N = 3,
PCS
0.22
Microfluidizer, N = 3, P = 60 MPa
PCS
0.45
Bead Mill, mean residence time 4 min
PCS
0.53 (bimodal peaks at 0.47 and 0.78 pm)
[551
Homogenizer, N = 1, P = 62 MPa
SEM
0.17
[391
Homogenizer, N = 1, P = 62 MPa, Pretreated with 4 M guanidine HCl
SEM
Homogenizer, N = 2, P = 55 MPa
CSA
0.50
Homogenizer, N = 10, P = 55 MPa
CSA
0.33
Median debris-size (pm)
Reference
P = 60 MPa
a
75 % of particles < 0.19 pm
[39]
Centrifugal disc photosedimentation (CDS); Photon correlations spectroscopy (PCS); Scanning electron microscopy (SEM), or Cumulative sedimentation analysis (CSA).
paration that can destroy the meaning of a size distribution obtained using c o m e r cia1 instruments (e.g., photon correlation spectroscopy). Only one method, cumulative sedimentation analysis (CSA), does not require sample pretreatment. However, this method is rather laborious. The study by Agerkvist and Enfors [55] for the mechanical disruption of E. coli warrants closer inspection as it provides comparative results for different homogenizer systems. The Microfluidizer generated large debris (mode 450 nm), while the homogenizer generated smaller debris (mode 190 nm) after multiple passes. The size distributions from a Dyno-Mill KDL bead mill showed a bimodal distribution with peaks centred at 47 1 nm and 777 nm after 4 minutes of disruption. Large debris is beneficial where the product is soluble, but detrimental where the product is an insoluble inclusion body. Clearly, selection of a mechanical disruption system needs to be done with reference to the desired debris size, not just the efficiency of product release.
6.9 Downstream Imuacts of Cell Disruption
159
6.9.2 Predicting Debris Size Particle-size distribution data for S. cerevisiae following disruption in an APV-Gaulin LAB60 homogenizer have been determined using electrical sensing zone measurement, and have been fitted to a Boltzmann-type equation,
(17)
where F(d) is the cumulative undersize mass fraction, d is particle size, d50 is the median particle size, and w is a parameter related to the width of the distribution [56].Parameters d50 and w were correlated with experimental data as follows,
=
W*
w o- w ~
*
= -2.3d5, d;, < 0.33
WO
w*=
w, - w ~
*
= 5.5d5, - 2.4
*
d5, 2 0.33
WO
where P is operating pressure, Pt and k d are parameters (115 bar and 670 bar, respectively), d50* and w* are dimensionless forms of the Boltzmann distribution parameters, and subscript o denotes the parameter value before homogenization. Eqn (17) has been employed to describe debris-size reduction for E. coli during high-pressure homogenization in an APV-Gaulin homogenizer [53].Parameters dso and w for E. coli were correlated with the number of homogenizer passes, N , according to Eqns (21) and (22),
I:[
In - = k2No.l
where k l ranged from 0.48 to 0.66 pass^^.^^ and k2 ranged from 1.62 to 1.92 paw0.', depending on the characteristics of the feed cells. The homogenizer operating pressure was fixed at 55 MPa. In an attempt to provide a rational basis for modelling debris-size reduction, a model for E. coli debris-size reduction based on grinding theory has been developed [53]. The model provided an excellent prediction of experimental data, does not
160
6 The Release of Intracellular Bioproducts
require any assumed functional distribution of the debris size, and can be used given information on the initial cell size distribution and the disruption efficiency during homogenisation. The final equation for the cumulative undersize distribution of cell debris after N passes, F d , is given by Eqn (23),
where xi is the discretized size (arbitrarily chosen), DN is the total disruption after N homogenizer passes, F d c is the cumulative undersize distribution of debris generated from whole cells at pass N , and F d d is the cumulative undersize distribution of debris generated by the comminution of existing debris in the homogenizer feed at pass N . F d c and F d d are given by Eqns (24) and (25),
where F d and F, are the cumulative undersize distributions of debris and whole cells, respectively, and a’,a,, and a d are parameters whose values are given in Table 6-5. The total cumulative size distribution of any homogenate is then simply obtained by combining the debris and intact cell distributions according to their relative proportions in the homogenate, giving Eqn (26):
Note that the model accounts for the simultaneous processes of debris generation from intact cells and debris size reduction, which are physically quite distinct. Table 6-5. Parameter values for the E. coli debris-size reduction model based on grinding theory developed by Wong [53]. Stationaryphase recombinant E. coli (not induced)
E. coli containing recombinant inclusion bodies (induced)
c(
2.4 f 0.16
2.3
a,
1.37 2 0.11
1.5 k 0.10
ad
0.48 k 0.065
0.85 f 0.10
Parameter
&
0.13
Abbreviations and Svmbols
161
Abbreviations and Symbols a,
C C
D d d50
E h kd
k K L m
M n N P P*
Q RP S
-
S SP
t
V vu
W
Xi
X Y Y"
parameter in equations (24) and (25) concentration wet biomass concentration disruption (volume- or number-fraction cells destroyed) diameter or model parameter median particle size specific energy input frequency of bead collisions cumulative undersize mass fraction homogenizer valve lift parameter in equation (18) rate constant parameter in equation (10) average cell length wall-strength-model parameter torque on bead mill agitator shaft wall-strength-model parameter number of bead mill or homogenizer passes pressure threshold pressure in equation (6-18) flowrate fractional release of soluble protein effective strength mean effective strength stress probability time volume peripheral agitator velocity Boltzmann distribution parameter discretized size interval fractional peptidoglycan crosslinkage distance between valve exit and impact ring dimensionless impact distance (Y/Y,,d)
Greek symbols a' parameter in equations (24) and (25) a weight ratio of beads to organisms in bead milling effective disruption volume between beads B & voidage 7 proportionality constant in equation (1 1) d wall-strength-model parameter z mean bead mill residence time w angular velocity
pm-a' kg m3 kg L-' -
pm or m or Pm kJ kg-' m-3
s-l
-
pm or m bar MPa-a or s-' mPm MPa-" Nm -
MPa or bar bar L h-l or mL min-' -
s or min L or m3 m s-l Pm Pm
mm
-
kg kg-' mm3 -
s or min rad s-'
162
6 The Release of Intracellular Bioproducts
Subscripts b beads in bead mill C intact (i.e., undisrupted) cells d debris (total) dc debris generated by cell disruption dd debris generated by comminution of debris h homogenate H homogenate m maximum, corresponding to 100 % disruption N after N homogenizer passes 0 feed sample (i.e., before disruption) particle (i.e., cell) P S stagnation std standard homogenizer impact ring Abbreviations LPS lipopolysaccharide PL phospholipid LP lipoprotein PHA polyhydroxyalkanoate SDS sodium dodecyl sulfate EDTA ethylenediamine tetra-acetic acid RTD residence time distribution CSTR continuous stirred-tank reactor
References [l] Holtje, J..-V., Glauner, B., Inst Pasteur Res Microbiol, 1990, 141, 75-103. [2] Labischinski, H., Hochberg, M., Sidow, T., Maidhof, H., Henze, U., Berger-Bachi, B., Wecke, J., in: Bacterial Growth and Lysis: Metabolism and Structure of the Bacterial Sacculus: de Pedro, M. A., Holtje, J.-V., Loffelhardt, W. (Eds.), New York: Plenum Press, 1993; pp. 9-21. [3] Fleet, G. H., in: The Yeasts: Yeast Organelles: Rose, A. H., Harrison, J. S.,(Eds.), London: Academic Press, 1991; Vol. 4, 2ndEd., pp. 199-277. [4] Klis, EM., Yeast, 1994, 10, 851-869. [5] Melendres, A. V., Unno, H., Shiragami, N., Honda, H., J Chemical Eng Japan, 1992, 25, 354-356. [6] Middelberg, A. P. J., O'Neill, B. K., Bogle, I. D. L., Snoswell, M., Biotecknol Bioeng, 1991, 38, 363-310. [7] Hetherington, P. J., Follows, M., Dunnill, P., Lilly, M. D., Trans Inst Chem Eng, 1971, 49, 142-148. [8] Engler, C.R., Robinson, C. W., Biotecknol Bioeng, 1979, 21, 1861-1869. [9] Middelberg, A. P. J., Dissertation, University of Adelaide, 1992. [lo] Felix, H., Anal Biochem, 1982, 120, 211-234. [ l l ] Spratt, B.G., Philo Trans R SOC Lond B, 1980, 289, 273-283. [12] Marvin, H. J., ter Beest, M. B. A., Witholt, B., J Bacteriol, 1989, 171, 5262-5267.
References [13] [14] [15] [16] [17]
163
Nikaido, H., Vaara, M., Microbiol Rev, 1985, 49, 1-32. Ryan, W., Pamlekar, S. J., Biotechnol Prog, 1991, 7, 99-110. Hettwer, D., Wang, H., Biotechnol Bioeng, 1989, 33, 886-895. Naglak, T. J., Wang, H. Y., Biotechnol Bioeng, 1992, 39, 732-740. Hart, R. A., Lester, P. M., Relfsnyder, D. H., Ogez, J. R., Builder, S. E., Biotechnology, 1994, 12, 1113-1117. [18] Falconer, R. J., O’Neill, B. K., Middelberg, A. P. J., Biotechnol Bioeng, 1997, in press. [19] Eriksson, H., Green, P., Biotech Tech, 1992, 6, 239-244. [20] Berger, E., Ramsay, B. A., Ramsay, J . A,, Chavarie, C., Braunegg, G., Biotechnol Tech, 1989, 3, 227-232. [21] Hahn, S. K., Chang, Y. K., Kim, B. S., Chang, H.N., Biotechnol Bioeng, 1994,44, 256-261. [22] Ramsay, J. A., Berger, E., Ramsay, B. A., Chavarie, C., Biotechnol Tech, 1990, 4, 221-226. [23] Hahn, S. K., Chang, Y. K., Lee, S. Y., Appl Environ Microbiol, 1995, 61, 34-39. [24] Middelberg, A. P. J., Lee, S. Y., Martin, J., Williams, D. R. G., Chang, H. N., Biotechnol lett, 1995, 17, 205-210. [25] Andrews, B. A,, Asenjo, J. A,, Trends Biotechnol, 1987, 5, 273-277. [26] Dean, C.R., Ward, O.P., Biotechnol Tech, 1992, 6, 133-138. [27] Hunter, J. B., Asenjo, J. A., Biotechnol Bioeng, 1990, 35, 31-42. [28] Lee, C.-K., Ku, M.-C., Biotechnol Tech, 1994, 8, 193-198. [29] Tsuchido, T., Katsui, N., Takeuchi, A,, Takano, M., Shibasaki, I., Appl Environ Microbiol, 1985, SO, 298-303. [30] Watson, J. S., Cumrning, R. H., Street, G., Tuffnell, J. M., in: Separationsfor Biotechnology: Verrall, M. S., Hudson, M. J. (Eds.), London: Ellis Horwood, 1992; pp. 105-109. [31] Reed, G., Nagodawithana, T. W., Yeast Technology. New York: Van Nostrand Reinhold, 1991. [32] Nakarnura, K., Enomota, A., Fukushima, H., Nagai, K., Hakoda, M., Biosci Biotech Biochem, 1994, 58, 129771301, [33] Kleinig, A. R., Middelberg, A. P. J., Chem Eng Sci 1996, 51, 5103-5110. [34] Kleinig, A. R., Middelberg, A. P. J., A 1 Ch E J, 1997, in press. [35] Kleinig, A. R., Mansell, C. J., Nguyen, Q. D., Badalyan, A., Middelberg, A. P. J., Biotech Tech, 1995, 9, 759-762. [36] Collis, M .A,, O’Neill, B. K., Middelberg, A. P. J., Bioseparation, 1996, 6, 55-63. [37] Harrison, S.T.L., Dennis, J. S., Chase, H. A., Bioseparation, 1991, 2, 95-105. [38] Vogels, G., Kula, M.-R., Chem Eng Sci, 1992, 47, 123-131. [39] Bailey, S. M., Blum, P. H., Meagher, M. M., Biotechnol Prog, 1995, 11, 533-539. [40] Middelberg, A. P. J., O’Neill, B. K., Bogle, I. D. L., Trans Inst Chem Eng, 1992, 70, part C, 205-212. [41] Kleinig, A. R., Middelberg, A. P. J., in: Better living through innovative biochemical engineering: Teo, W. K., Yap, M. G. S., Oh, S. K. W. (Eds.), APBioChEC 94 Third Asia-Pacific Biochemical Engineering Conference, Singapore, June 1994, Proceedings, pp. 607-609. [42] Keshavarz Moore, E., Hoare, M., Dunnill, P., Enzyme Microb Technol, 1990, 12, 764-770. [43] Kleinig, A. R., O’Neill, B. K., Middelberg, A. P. J., Biotech Tech, 1996, 10, 199-204. [44] Bunge, F., Pietzsch, M., Miiller, R., Slydatk, C., Chem Eng Sci, 1992, 47, 225-232. [45] Limon-Lason, J., Hoare, M., Orsborn, C.B., Boyle, D. J., Dunnill, P., Biotechnol Bioeng, 1979, 21, 745-774. [46] Melendres, A. V., Honda, H., Shiragami, N., Unno, H., Bioseparation, 1991, 2, 231-236. [47] Garrido, F., Banerjee, U. C., Chisti, Y., Moo-Young, M., Bioseparation, 1994, 4, 319-328. [48] Schiitte, H., Kraume-Flugel, R., Kula, M. R., Ger Chem Eng, 1986, 9, 149-156. [49] Schiitte, H., Jiirging, B., Papamichael, N., Ott, K., Kula, M.-R., in: Annals ofthe New York Academy of Sciences: Enzyme Engineering IX: Blanch, H. W., Klibanov, A. M. (Eds.), New York: New York Academy of Sciences, 1988; Vol. 542, pp. 121-125. [50] Sauer, T., Robinson, C. W., Glick, B. R., Biotechnol Bioeng, 1989, 33, 1330-1342.
164
6 The Release of Intracellular Bioproducts
[51] Lutzer, R. G., Robinson, C. W., Glick, B. R., in Proceedings of the 6'h European Congress on Biotechnology: Alberghina, A., Frontali, L., Sensi, P. (Eds.), Amsterdam: Elsevier Science BV, 1994; pp. 909-916. [52] Baldwin, C., Ronbinson, C. W., Biotechnol Tech, 1990, 4, 329-334. [53] Wong, H. H., Dissertation, University of Adelaide, 1997. [54] Olbrich, R., Dissertation, University of London, 1989. [55] Agerkvist, I., Enfors, S.-O., Biotechnol Bioeng, 1990, 36, 1083-1089. [56] Siddiqi, S. F., Titchener-Hooker, N. J., Shamlou, P. A., Biotechnol Bioeng, 1996, 50, 145150.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
7 Microcarriers in Cell Culture Production Bjorn Lundgren and Gerald Bliiml
7.1 Introduction Animal cell culture is currently operated at 10000-L scale in suspension cultures and 4000 -L scale in microcarrier cultures of anchorage-dependent cells (Fig. 7-1). Both are large unit operations. As the size of such operations increases, as does investment in resources and personnel. Cell culture conditions are also more critical. Failures become far more costly. The pressure in biotechnology production today is for greater speed, lower costs, and more flexibility. Ideally, a production unit should be compact (requires less investment), and modular (for use in different production schemes) (Fig. 7-2). Designs that allow the same plant to be used for different cell types (bacteria, yeast, insect cells, plus suspension and anchorage-dependent animal cells) are thus preferable. Animal cell culture is constantly being developed to increase unit productivity and thus make the production costs of animal cell products more competitive.
Fig. 7-1. Polio virus vaccine production in 1000-L fermenters. (Courtesy of Institute Merieux, France.)
166
7 Microcarriers in Cell Culture Production
Fig. 7-2. Example of a multipurpose reactor design.
Because animal cells have a relatively low productivity, large amounts of culture supernatant are needed per clinical dose of final product. Extremely large volume cultures will be needed to produce kilogram quantities of a therapeutic monoclonal antibody, for example. The productivity of large-scale cell culture can be increased either by scaling-up to larger volumes with cell densities of 2-3 x lo6 mL-*, or by intensifying the process in smaller volumes but with higher cell densities (up to 2 x lo8 cells mL-'. When intensifying cell densities, more frequent media changes are needed and eventually perfusion has to be applied for which many competing technologies are available. Microcarrier culture of anchorage-dependent or entrapped cells lowers the volume to cell density ratio and thus belongs to the second of these alternatives. This technique has many advantages for the commercial manufacturer. It operates in batch or perfusion modes and is well suited to efficient process development and smooth scale-up (Figs. 7-3 and 7-4). In addition, the reactors can be modified to grow other organisms. This chapter reviews the principles and methods of using microcarriers in cell culture production.
Surface
Macroporous
High Macroporous
High Oenslty Macroporous
Stirred tank Batch culture
Stirred tank Perfusion culture
Fluidized bed Perfuslon culture
Packed bed Perfuslon culture
Fig. 7-3. Different microcarrier alternatives and applications.
7.2 Production Considerations
Surface
Macroporous
High Density Macropomus
Batah
Perfusion
hrhraion
167
High Density Macroporous pwtudon I
I
Stirred tank
Stirred tank
Fluidized bed
Packed bed
Fig. 7-4. Different microcarrier culture set-ups.
7.2 Production Considerations 7.2.1 Production Economy Calculating total production economy is complex. Many aspects need to be considered; the organism to be cultured, amount of product needed, medium volumes, equipment, staff, downstream costs, etc. The major part of the Cost Of Goods Sold is, however, fixed costs [I]. Investments in equipment and personnel thus determine whether a project will be profitable or not. Ideally, investments should be suitable for other projects if the one planned is not successful. Sometimes the choice of technology is restricted. For example, batch/fedbatch processes must be used when producing cytopathic viruses. To maximize the benefits of perfusion, immobilization is also needed to prevent the washout of cells. The cost ratios for perfusion, continuous-flow, and batch techniques to produce monoclonal antibodies are 1 : 2 : 3.5 [2].
7.2.2 Consumable Cost Comparison The three major consumables are the culture surface (for anchorage-dependent cells), serum (or protein additives), and medium. The medium cost can normally not be controlled, other than by negotiating price with different suppliers. 7.2.2.1 Culture Surface To make a true cost comparison of cell culture surfaces, grow a specific cell line on different supports and calculate the cost of each support /yield of lo6 cells. Start by calculating the price per m2 of material. Take care when comparing porous materials.
168
7 Microcarriers in Cell Culture Production
Some determinations may also measure surfaces inside pores that are not accessible for cells. The true cell culture surface will thus be smaller than that measured. Remember that the larger the diameter of cell, the smaller the accessible area. Large cells thus give much lower densities in macroporous carriers. As large a surface area as possible is desirable as it allows a higher number of cell doublings. With a large surface area and by optimizing the inoculation density, it is possible to maximize the cellular multiplication in each culture step. This reduces the number of scale-up steps and equipment necessary to reach final production volume (cell number). Comparing cost of surface aredyield of lo6 cells, reveals the following. Microcarriers and the Cell Cube (Costar) have the same order of magnitude and the lowest cost, roller bottles (1400 cm2) are five times more expensive, and hollow fiber reactors 200 times more expensive in generating the same number of cells. (The cell density used for the hollow fiber is that given by the supplier.)
7.2.2.2 Serum and Additives The most expensive additive to the medium is serum (proteins). Reducing serum in the medium by 1 % lowers costs by approximately 3 US $ L-I. Comparing the costs of a suspension culture with a fluidized bed run reveals the following. (It is assumed that serum concentration can be reduced by 1 % due to the higher cell density in the fluidized bed - this is a modest assumption.) Bed volume is 5 L. This is normally perfused with 10 volumes of mediahed volume/ day = 50 L per day. The saving is then 3 x 50 = 150 US $ per day. If run for 30 days, the total saving is 4500 US $. Subtracting the cost of the carriers gives a net saving of 3740 US $ per run. A 5-L fluidized bed reactor with perfusion thus produces the same amount of cells and products in 1 month as a 1500-L batch reactor does in one week at a consumable cost saving of 3740 US $. The investment cost is also much lower for the smaller system. Table 7-1 compares the medium utilization (mL spent medium per mg product) of a human hybridoma cell line in a continuous stirred tank reactor (CSTR) with a fluidized bed reactor. Table 7-1.Medium utilization in continuous stirred tank and fluidized bed reactors.
Culture volume Dilution-perfusion rate (L h-l) Medium L per day Product (mg) per day mL spent medium per mg product
CSTR
FBR
20 L 0.0125 6 50 120
20 L (10 L of carrier) 0.105 50 400 125
7.3 Microcarrier Background
169
7.2.3 Important Developments Continuous development of manufacturing technology is needed if animal cell culture processes are to be competitive. Better cell culture, higher productivity, and simpler media requirements are all important [l]. Yields should increase in relation to space used. Processes should be robust with high success rates and minimum down-time and the number of operations should be as few as possible. Plants should be multipurpose to fully utilize capacity. Remember that fixed costs make up most of the Cost Of Goods Sold, and that R&D normally focuses on work that affects variable costs !
7.3 Microcarrier Background 7.3.1 Adhesion (Cell-Cell, Cell-Surface) Cell adhesion is a multi-step process (Fig. 7-5). For an adherent cell to attach quickly to a surface, it needs to be round (have a disrupted cytoskeleton) and have exchanged its cell surface receptors for newly produced ones. It is therefore necessary to have trypsinized cells in a single cell suspension (trypsin also acts as a growth factor). Work has shown that when the surfaces on which cells grow are degraded, the cells remain flat and retain their cell-cell contacts. These cells did not attach for several hours when seeded onto a cell culture surface. When they did attach, the distribution over the surface was very uneven due to clumping. This resulted in an inhomogeneous cell distribution on the carriers and a slower growth rate. Normal and transformed cells seem to adhere in slightly different ways (see Section 7.4.1). In vitro, adherent cells bind to extracellular matrix components; type I and IV collagen, fibronectin, vitronectin, laminin, chondronectin, thrombospondin, heparan sulfate and chondroitin sulfate. The ‘normal’ in v i m adhesion process is considered to occur via integrin receptors binding to primarily fibronectin mol. wt. 220 KDa) and vitronectin (65 KDa). These proteins are abundant in serum-containing media (10 % FCS contains 2-3
Fig. 7-5.The adhesion process.
170
7 Microcarriers in Cell Culture Production
pg fibronectin mL-’) and quickly bind to the surface of the cell culture material. The cells in their turn bind to these proteins via the integrin receptors, with Ca2+andMg2+ ions as cofactors. A minimum of 15 ng cm-2 of adsorbed fibronectin is required for BHK cell attachment. Fibronectin binds easily to gelatin, which is why gelatin gels are used for the affinity purification of fibronectin. It is also the reason why gelatin is a good cell culture substrate and used to manufacture some microcarriers. Vitronectin is considered to be more potent and active at even lower concentrations than fibronectin. Vitronectin is also called serum spreading factor and has been shown necessary for the spreading out phase of cells. These proteins are also produced by a number of cell lines. Other integrin receptors present on the cell surface are the collagen and laminin receptors. These proteins are not usually present in serum, but are sometimes added to coat the cell culture surface. Other molecules have been reported to be involved in lymphocyte adhesion but will not be discussed here. The integrin receptors, together with cadherin receptors, are also responsible for cell-cell binding and the formation of tight cell junctions and sheets. At a later stage of attachment, the cells themselves produce multivalent heparan sulfate, which reinforces the binding. Finally, the cells assemble the cytoskeleton and spread out. Anchorage-dependent cells cannot proliferate without being spread out, so it is essential to enhance all steps involved in adhesion. Cells can adhere to a wide variety of materials; glass, various plastics, metals (stainless steel 3 16, titanium used in implants), dextran, cellulose, poly-lysine, collagen, gelatin, and numerous extracellular matrix proteins (see above). Borosilicate glass is normally negatively charged. Attachment to it can be increased by treating the surface with NaOH or by washing with 1 mM Mg acetate. Plastics used in cell culture include polystyrene, polyethylene, polycarbonate, Perspex, PVC, Teflon, cellophane, and cellulose acetate. These organic materials are made wettable by oxidizing, strong acids, high voltage, UV light, or high-energy electron bombardment rendering them negatively charged. One major drawback of plastics is that they do not normally withstand autoclaving. Poly-d(1)-lysine is an artificial molecule that can also induce attachment when coated onto surfaces. It appears that it is the charge density and not the type of charge that is important for attachment [3]. Cells that grow attached on plastic surfaces should readily attach and grow on carriers.
7.3.2 Immobilization Principles Immobilization was first described as early as 1923 by Carrel1 in the paper ‘Immobilisation of animal cells’ [4]. Today, biocatalysts (cells) are defined as immobilized when they are restricted in their motility by chemical or physical methods while their catalytic activity is conserved. Different immobilization techniques include binding biocatalysts to each other (aggregates) or on carriers, and physical entrappment in a polymeric matrix or through membrane separation (Fig. 7-6). Immobilization allows heterogeneous catalysis, in contrast to catalysis in which substrate and biocatalyst (cells or enzymes) are homogeneously distributed. An eco-
7.3 Microcarrier Background
171
Cell Retention
I
Centrifuganon
I
Cellnumb
Fig. 7-6. Different immobilization principles.
nomically viable separation of suspended iocatalysts ,.om product is hardly possible. However, the immobilization technique does enable preliminary separation of product and biocatalyst, even during upstream processes. Immobilization also creates and maintains a high density of cells in a small volume (Table 7-2). Another major advantage is that the medium feed rate is not dependent on the growth rate of the cells. The higher throughput of medium guarantees higher volumetric productivity. In fact, some cell lines cannot grow or produce the desired product without immobilization. Butler [ 5 ] calculated the surface-to-volume ratios of different immobilization systems. Roller bottles had a ratio of 1.25, packed beds (with spherical carriers) 10, and artificial capillaries 30. Microcarriers (25 g L-l) had 150, by far the best surface to volume ratio. Table 7-2. Cell densities in different immobilization systems. System
Cell density (m L-')
Suspension Cell retention Cell immobilization Ascites Tissue
106 107 108 108 2 x 109
172
7 Microcarriers in Cell Culture Production
7.3.3 Materials Materials are important because of their chemical, physical, and geometric effect on the carrier. For example, they influence toxicity, hydrophilicity, hydrophobicity, microporosity, mechanical stability, diffusion of oxygen or medium components, permeability, specific gravity, and shape (form, size, thickness, etc.). A wide variety of materials have been utilized to produce microcarriers: plastics, (polystyrene, polyethylene, polyester, polypropylene), glass, acrylamide, silica, silicone rubber, cellulose, dextran, collagen (Fig. 7-7) (gelatin), and glycoseaminoglycans. These materials can be formed into different shapes, with spherical the most common, though fibers, flat discs, woven discs, and cubes are also found. Carriers are usually positively or negatively charged, though non-charged carriers are also available. These are normally coated with collagen or gelatin or have fibronectin or fibronectin peptides coupled onto the surface. Glycoseaminoglycan microcarriers are slightly negatively charged. Cells bind directly to the collagen, fibronectin, or fibronectin peptides and to the glycoseaminoglycan microcarriers (see Section 7.3.1). Unfortunately, protein-coated and glycoseaminoglycan carriers cannot be autoclaved as the protein structures are destroyed, but gelatin (denatured collagen) may be autoclaved. Gelatin also has a very high affinity for fibronectin, which is why it is so suitable as a cell culture substrate.
Fig. 7-7.Collagen-based macroporous microcarrier (Verax).
7.34 Size Shape and Diffusion Limits The diameter of the different carriers varies from 10 pm to 5 mm, the smaller carriers being best suited for stirred tanks. The higher sedimentation rates of larger carriers
7.3 Microcarrier Background
173
make them suitable for fluidized and packed beds. The smaller the carriers, the larger the surface in the settled bed volume because of the smaller void volume between them. A diameter of 100-300 pm is the ideal size for smooth microcarriers, as a very narrow size distribution is most important for good mixing in the reactor. The emulsion and droplet techniques produce round carriers, though an exception is the DEAE cellulose carrier, which has a cylindrical shape (Fig. 7-8). Macroporous carriers are on average bigger because their pores may be up to 400 pm wide. However, a large pore size must be balanced against the disadvantages of bigger particles such as diffusion limits and higher shear stress on the outer surface. Mass transfer in the immobilized cell aggregate is a significant problem in immobilized cultures. The poor solubility of oxygen in the medium at 37 "C and the high consumption rate of the cells make it a marker for limitations in the cell aggregate.
b)
Fig. 7-8. Different microcarrier shapes. (a) Spherical (Biosilon Nunc). (b) Cylindrical, side view (Immobasil ASL).
174
7 Microcarriers in Cell Culture Production
Fig. 7-8.(continued). (c) Cylindrical, front view. (d) Woven discs (Fibracell, Bibby Sterilin).
Keller [6] reports that in cell cultures with high densities (up to 2 x m-3 in the mol per cell and second, and a cell layer), an oxygen consumption rate of 5 x medium volume which is tenfold the cell mass, consumes the oxygen within 3 minutes. The single oxygen molecule has to overcome three barriers before it reaches the cell in the middle of the carrier; firstly transport from the gas phase into the medium, then transfer from the medium to the cell mass, and finally diffusion and consumption through the cell layers. The OTR (oxygen transfer rate) can be increased by increasing the volume-specific surface or using pure oxygen instread of air. The cells themselves should not be exposed directly to oxygen because of its toxic effect. However, the protection afforded by macroporous structures allows the use of pure oxygen.
7.3 Microcarrier Background
175
Fig. 7-9. Diffusion limits in CHO aggregates (Hematoxylin staining, size of aggregate 3 mm; (note necrotic center).
Many different methods of oxygenation are described or used. Bubble-free aeration [ 7 ] , gassing with large bubbles [8] or using microbubbles [9,10] are used in high-density immobilized cultures (see Section 1.6.5). Depending on the size of the carrier, the oxygen concentration on its outside must be increased. Carrier particles up to 500 pm in diameter can be supported with an oxygen tension of about 10 % (Fig. 7-9); if the particles are larger than 900 pm, limitations occur if the oxygen tension is below 35 % [6]. Griffiths [ I l l found sufficient oxygen penetration into cell layers up to 500 pm thickness.
7.3.5 Specific Density and Sedimentation Velocity Smooth microcarriers in stirred tank reactors have a specific density just above the medium between 1.02 and 1.04 g cmP3, while Materials or material mixtures used to produce macroporous microcarriers have specific densities between 1.04-2.5 g ~ m - The ~ . sedimentation velocity is a better parameter for the suitability of a microcarrier in a specific reactor type. This is because not only the specific density but also size and shape influence the sedimentation velocity. Velocities lower than 30 cm / m i x 1 do not create enough circulation and mixing for efficient nutrient supply throughout the carrier bed [6]. Adherent cells also tend to form bridges between microcarriers in fluidized beds. Higher sedimentation rates (150-250 c d min-') will prevent the bridges from forming.
176
7 Microcarriers in Cell Culture Production
7.3.6 Rigidity and Shear Force The rigidity of microcaniers is important in long-term cultures. The materials used should withstand the organic acids and proteases found in culture supernatants. Abrasive carriers made of brittle materials such as glass or ceramics could harm cells, valves, and bearings and cause problems filtering culture supernatants. In turbulent fluids, particle/particle collisions and particlektirrer collisions are highly energetic [12]. In contrast to stirred tanks, the shear forces in fluidized beds are homogeneously distributed and impellerkanier collisions are not possible. Shear forces in fluidized beds correlate with particle sedimentation velocity and reactor type. Keller [6] measured shear tensions of 0.3-0.5 N cm-2; which are far below the damaging shear tension for kidney cells (10 N cm-2) [13]. Spier et al. [14] transferred shear forces into wind velocities to demonstrate the force affecting the cells. The linear velocity should not be more than 0.3 km h-' during attachment of anchoragedependent cells while a velocity of 95 km h-' is necessary for cell detachment.
-
7.3.7 Porosity Carriers can either be solid or microporous. Microporous carriers allow the cells to take up and to secrete material also on the basolateral side of the cell. Molecules up to mol. wt. 100 KDa can penetrate these carriers (Fig. 7-10). Note that when
Fig. 7-10. Microporous microcarrier (Cytodex Amersham Pharmacia Biotech). Molecular weight cut-off = 100000 Da.
7.3 Microcarrier Background
177
the microcarriers are entirely confluent, there can be a different environment inside the beads than on the outside! The latest development in microcarrier technology is macroporous carriers that allow cells to enter the carriers, the average pore size of the different types is being between 30 and 400 pm (Fig. 7-lla,b). As the mean cell diameter of single cells in suspension is about 10 pm, this allows cells easy access into the carriers. Macroporous carriers are also suitable for immobilizing non-adherent cell types (Fig. 7-12); in this case, the cells are forced into the matrix and entrapped. Macroporous carriers provide higher cell densities and are therefore normally used in perfusion culture. The porosity of macoporous carriers is defined as the percentage volume of pores compared with the total carrier volume and is normally between 60 and 99 %. The advantages and disadvantages of microporous and macroporous carriers are covered
Fig. 7-11. Macroporous microcarrier. (a) Cytopore (Amersham Pharmacia Biotech). (b) Microporous microcarrier (SoloHill).
178
7 Microcarriers in Cell Culture Production
Fig. 7-12. Scanning electron micrograph (SEM) of macroporous carrier with entrapped hybridoma cells. (Cytoline 2 Amersham Pharmacia Biotech)
later in this chapter. In spite of the large number of microcarrier designs and types, very few are still available commercially and even fewer fulfil industrial standards for large-scale manufacturing processes.
7.3.8 Cell Observations Transparency of a microcarrier is important for simple cell observations in a light microscope (Fig. 7-13a,b). In vaccine production especially, it is important to see the morphology of cells directly on the carrier to identfy the correct moment at which to infect the cells or to harvest the virus. Unfortunately, due to the size, three-dimensional structure, and component materials of a number of microcarriers, the cells cannot be observed clearly with a light microscope. Because of the long preparation time and the effect of dehydration on cell shape and morphology the scanning electron microscope is not suitable for such observations (Fig. 7-14). However, confocal laser scanning microscopy [15] is an excellent tool for making cells visible in the pores of macroporous beads. With this technique, it is possible to make optical sections through the carrier and create a three-dimensional reconstruction. A viability stain with two fluorescent dyes (FDA fluorescein diacetate for living cells and ethidium bromide for dead cells) allows the viability of cells in three-dimensional structures to be estimated [16,17] (Fig. 7-15). Other staining methods such as MTT [18] also make cells visible in macroporous structures.
7.3 Microcarrier Background
179
Fig. 7-13.Transparent microcarriers with cells (Cytodex). (a) Hematoxylin staining, magnification 5 0 ~ . (b) Hematoxylin staining, magnification 2 5 0 ~ .
Fig. 7-14.SEM of dehydrated pig kidney on microcarriers. (Courtesy G. Charlier, INVR, Brussels, Belgium).
180
7 Microcarriers in Cell Culture Production
Fig. 7-15. Fluorescein di-acetate viability stained CHO cells in a macroporous microcarrier (Cytopore).
7.4 Microcarrier Technology 7.4.1 Microcarrier History Microcarrier technology began with a paper by Professor Van Wezel in 1967: ‘Growth of cell-strains and primary cells on microcarriers in homogeneous culture’ [19], where he described the use of DEAE-Sephadex A-50, a positively charged ion exchanger, to grow human fibroblast-like cells. What he achieved was a homogeneous unit process. Microcarrier cultures then comprised cultivation of anchoragedependent animal cells on small spherical particles kept suspended in culture medium. In 1969, Van Hemert et al. published the paper ‘Homogeneous cultivation of animal cells for the production of virus and virus products’ [20], describing the use of human diploid cells grown on microcarriers. However, the original microcarriers had too high a charge of DEAE and were toxic to cells. Levine et al. [21] suggested lowering the level of DEAE substitution, and thus Cytodex microcarriers - launched in 1981 - were the result of collaboration between Van Wezel and Pharmacia Biotech AB. Early on, Van Wezel found that transformed cells detach from ‘smooth’ microcarriers and grow in the culture medium as aggregates. Many workers contributed to the research and development of other ‘smooth’ microcarriers. The next major development was macroporous gelatine microcarriers developed by Nilsson et al. [22] (Fig. 7-16), which allowed growth inside the beads, thereby increasing cell density and protecting the cells. The final development was presented in an article by Young and Dean [23] describing the use of microcarriers for animal cells in fluidized beds. This step allowed the immobilization of both anchorage and
7.4 Microcarrier Technology
181
Fig. 7-16. Macroporous gelatin microcarrier (Cultisphere).
Fig. 7-17. 125 pm microcarriers used to stimulate cell aggregation (SoloHill).
suspension cells in high cell density production systems. Recently, microcarriers with a smaller diameter, 125 pm, have been used to induce and stabilize cell aggregation [24] (Fig. 7-17).
7.4.2 Advantages of Microcarriers Microcarriers have many advantages: they are essential when surfaces are needed for anchorage dependent cells, and are also inexpensive (in terms of price per m2). Microcarrier technology results in a homogeneous culture system that is truly scale-
182
7 Microcarriers in Cell Culture Production
able. Because of their large surface area-to-volume ratio, they occupy less space in storage, production, and waste handling. The surface also allows cells to secrete and deposit an extracellular matrix. This helps to introduce certain growth factors to cells. The spherical microcarriers have short diffusion paths, which facilitates nutrient supply in general, while the extracellular matrix provides cells with support to build their cytoskeleton, and to organize organelles intracellularly. Sometimes, however, this leads to increased productivity of functional product. Microporous carriers (see Fig. 7-10) allow the cells to create a micro-environment inside the beads and also facilitate polarization and differentiation of cells. As the cells are immobilized on the microcarriers, it is easier to retain them in culture during perfusion, a process which separates cells from products. At high cell concentrations
Fig. 7-18.(a) Empty macroporous microcarrier (Cytoline 1). (b) Macroporous microcarrier with high cell density of CHO cells.
7.4 Microcarrier Technolozy
183
and perfusion rates, the residence time of the product at 37 "C is very short, and it can quickly be separated from the cells and cooled down. Macroporous microcarriers have additional advantages. They allow cells to grow in three dimensions at high densities (Fig. 7-18), which stabilizes the cell population and decreases the need for external growth factors. This makes it easier to use lowserum, serum-free, and even protein-free media which, of course, cuts costs (see Section 7.2.2). The high cell density confers more stability and improves the longevity of the culture, making macroporous microcarriers suitable for long-term cultures. The structure protects the cells from shear forces generated by the stirrer, spin filter and air / oxygen sparging [25], which facilitates oxygen supply. Macroporous microcarriers can be used both for suspension (entrapment) and anchorage-dependent cells. Various cell culture technologies can be applied, including stirred, fluidized, and packed bed reactors. The process advantage of macroporous carriers is that the high perfusion rates maintain a homogeneous environment which not only ensures a sufficient nutrient supply but also removes toxic metabolites. Vournakis and Runstadler [26] note that a regular distribution of oxygen and other nutrients in the pores is secured through a 'micropump'. The convection stream through the fluidized bed creates a pressure drop over the carrier surface that causes medium to flow in and out of the carrier. Changing media from fetal calf serum-containing to protein-free is sometimes only possible with immobilized suspension cells or has a shorter adaptation and is less time consuming than with non-retained cells [27]. The growth rate of adherent cells in macroporous carriers is reduced by 30-50 % compared with that in T-flasks and is independent of carrier concentration and stirrer speed [28].
7.4.3 Disadvantages of Microcarriers Some carriers have to be washed and prepared. Scale-up using cells harvested from microcarriers is more complex than suspension expansion and, it is even more difficult to harvest cells from macroporous carriers. Cell enumeration and harvesting are more difficult due to the higher cell density, the latter condition also making it more difficult effectively to infect all cells simultaneously, when especially using non-lytic viruses. Finally, large carriers with small pores may restrict the diffusion of nutrients to some cells.
7.4.4 Types Tables 7-3 and 7-4 list the major commercially available microcarriers. The lists are not complete and all carriers for research have been omitted.
184
7 Microcarriers in Cell Culture Production
Table 7-3. Surface microcarriers. Name
Material
Diameter Reactors (Pm)
Cytodex 1 (Amersham Pharmacia Biotech) DEAE-Dextran Cytodex 2 (Amersham Pharmacia Biotech) THMAP-Dextran Cytodex 3 (Cellex) Gelatin-Dextran Biosilon (NUNC) Polystyrene Bioglass (Solo Hill) Glass, collagen coated plastic FACT III (Solo Hill) Modified collagen DE 52/53 Whatman DEAE-Cellulose
147-248 135-200 141-211 160-300 90-500 90-500
STR STR STR STR STR STR
Table 7-4. Macroporous microcarriers. Name
Material
Diameter (Pm)
Pore size (Pm)
Porosity
Reactors
(%I
Cultispher-G, S, GL (Percell Biolytica)
Gelatin
170-500
-50
50
STR
In formatrix (Biomat. C o p )
Collagen-glycoseaminoglycan
500
40
99
STR
Microsphere (Cellex)
Collagen
500-600
20-40
75
FB, PB
Siran (Schott Glaswerke)
Glass
300-5000
10-400
60
FB, PB
Microporous MC (Solo Hill Labs Inc.)
Polystyrol
250-3000
20-150
90
STR
Cytopore 1, 2 (Amersham Pharmacia Biotech)
Cellulose
180-2 10
30
95
STR
Cytoline 1, 2 (Amersham Pharmacia Biotech)
Polyethylene
2000-2500
10400
65
FB, PB, STR
ImmobaSil ASL
Silicone rubber
1000
50-150
> 40
STR
FB, fluidized bed; PB, packed bed; STR, stirred tank reactor.
7.4.5 Scale-up Considerations Scale-up starts with the creation or choice of cell line! The characteristics of the cell line will then greatly influence the scale-up possibilities and technology choices. When choosing the cell line, it is essential to bear in mind the production technology available and the final scale of production - this is especially important if hardware investments have already been made. For example, the scale and technology used
7.4 Microcarrier Technology
185
affects the stability of expression as well as the adhesive properties required by the cell. The higher the degree of transformation, the more difficult it is to get the cells to attach and spread onto surfaces. It may thus prove difficult to find a suitable ‘industrial’ microcarrier if technology screening is left too late after designing the cell line. When evaluating different microcarriers to be used in a production process, the consequences of the choice at the final process scale must be considered. Normally, it is easier to handle a microcarrier that is autoclavable. It should also be possible to handle the material in the open, e.g. to subdivide lots more easily. If mistakes are made during process start-up, it helps if the microcarriers can be resterilized without loss of material. It is more difficult to sterilize the carriers at larger scale; as an autoclave is no longer viable due to the large volumes involved. Normally, the microcarriers are then sterilized inside the bioreactor. The type of microcarrier is intimately linked to the bioreactor design. Large, highdensity microcarriers always have to be prepared inside the bioreactor. The size of the carriers will influence their transfer through valves and tubing during sterilization and harvesting. The size and density will also influence how easy it is to keep carriers in suspension (i.e. impeller and reactor design, stirrer speed, etc.) and how quickly they sediment to allow large volume media changes (fed batch). These characteristics also affect how easy it is to retain the microcarriers in a perfusion system (settling zone, spin filters). It must also be considered if the microcarrier generates truly homogeneous cultures, or causes heterogeneity. Homogeneous cultures are normally easier to scale-up as they facilitate oxygen and nutrient supply, and make it easier to remove waste products. The choice of solid versus porous microcarriers will impact on how easy it is to wash the cells during media changes from serum to serum-free media, and to wash and harvest cells from the carriers. Solid carriers are normally more easy to handle in these respects. Macroporous carriers are the most difficult at harvest. Gelatin or collagen microcarriers are the easiest to harvest, as it is possible enzymatically to degrade the microcarriers and recover the cells. The simplicity of a low-density system should be weighed against the demands and complexity of high productivity, high cell density cultures. Another aspect to consider is inactivation of the carriers after production. This is especially important when producing hazardous agents. Here, it is once again advantageous if the carriers can be autoclaved prior to disposal. Positively charged carriers will adhere to nonsiliconized glass and stainless steel. This can be overcome during cleaning by washing with high-pressure water, automatically via spray balls or manually by using high-pressure washes. This ‘stickiness’ can be reduced by changing pH to decrease the charge, and increasing the ionic strength by adding salt to the washing solution. The material used to produce the carriers will influence waste disposal, as occasionally the material is degraded naturally. One further aspect to consider is whether the microcarrier is made of a material that is used in downstream chromatography processing; this will facilitate process validation if it is the case. Finally, it is important to determine whether the microcarrier is an established product that is already registered with regulatory authorities for similar production processes. If so, it should also facilitate in the registration of a new process.
186
7 Microcarriers in Cell Culture Production
7.4.6 Choice of Supplier Important aspects to consider when evaluating suppliers are their production capacity, i.e. the batch size available. This affects purchasing and testing during full-scale operation. It is important to evaluate what function and quality testing is performed by the supplier, and if a certificate of analysis is available for each batch. This helps to minimize batch-to-batch variation and ensure process consistency. Normally, at least three different lots should be evaluated before scale-up. The availability of product stability and leakage studies is also important. Regulatory issues to consider are if the company works according to IS0 9000, cGMP, and if it is possible to audit the supplier. Ask if they have Regulatory Support Files or Drug Master Files available for the product (Table 7-5). Support in form of application work, product information, trouble-shooting, and validation support is important. The financial stability of the manufacturer is important to ensure secure supply for the lifetime of the process. Finally, the price of the microcarriers could be an issue; however, as stated previously, the main part of Costs Of Goods Sold is fixed costs! Table 7-5. Regulatory Support File and Drug Master File. Regulatory Support File
Drug Master File
Information directly to customer
- Information directly to FDA
Updated automatically when new information available, information accessible to customer
- Updated annually, information not accessible to customer
World-wide support
+ Well accepted system
No recognized approval
- No recognized approval
No manufacturing information
+ Contains manufacturing information
7.4.7 The ‘Ideal’ Microcarrier There is no such thing as the ideal microcmier (Fig. 7-19); there are always compromises. However, the desirable features can nevertheless be described as: autoclavable for easy handling autoclavable attachment functions (for use in serum-protein-free medium) available with a Drug Master or Regulatory Support File available in large batches for industrial customers both non- and specific attachment functions for a large number of cell types (rapid attachment, spreading, and proliferation) certificate of analysis for each batch good long-term stability (minimal leakage)
7.5 Microcarrier Culture Equipment
187
Fig 7-19. The ‘ideal’ microcarrier.
- high batch-to-batch consistency - high surface-to-volume ratio (large multiplication steps) - material of non-biological origin (minimize viral risks), inactivation procedures
possible and available for material of biological origin - macroporous for shear force protection - no nutrient limitation at center of matrix (macroporous) -
-
-
non-toxic, non-immunogenic matrix (transplantation purposes, products used as pharmaceuticals) possibility to count cells porous to allow for cell polarization/differentiation thoroughly quality controlled and meet the standards for GMP transferable between vessels (ease of scale-up) transparent for ocular inspection uniform size (cell growth homogeneity)
7.5 Microcarrier Culture Equipment 7.5.1 Unit Process Systems A unit process system contains all cells used in production within the same vessel (Fig. 7-20). This means that all cells are grown under ‘identical’ conditions! Examples of truly homogeneous unit process systems for carriers are microcarrier cultures
188
7 Microcarriers in Cell Culture Production
Fig. 7-20. Surface-to-volume ratio comparison of a unit process system (macroporous microcarriers) and a multiple unit system (rollers).
in stirred tanks and fluidized bed cultures with sufficient oxygen supply to not limit the scaleability. The Verax system is not truly homogeneous as it creates an oxygen gradient throughout the bed as the medium passes through. This makes it difficult to scale up. Other examples of unit process systems are the glass bead reactor run as a packed bed and the Fibracel woven plastic discs, also run as a packed bed. In both cases, the carriers with cells are stationary, which results in channeling of media in the reactor and which causes necrotic zones to arise inside the carrier bed.
7.5.2 Small-scale Equipment Small-scale equipment for microcmier cultures often comprises plastic bacterial Petri-dishes for stationary cultures, roller bottles for mixed cultures, and glass spinner flasks for stirred cultures. Borosilicate glass is normally used in spinners. The spinners with the best performance for microcarriers are the Techne bottles equipped with bulb stirrers. The spinner flasks are placed on magnetic stirrers in incubators or warm rooms. Normally there are no controls attached. However, the Superspinner supplied by B. Braun Biotech contains a microporous polypropylene membrane for bubble free aeration in the medium. The aeration tubing is connected to a pump which delivers oxygen and C02 incubator (Fig. 7-21). Spinners are avail-
7.5 Microcarrier Culture Equipment
189
Fig. 7-21. B. Braun Superspinner.
able up to 20-L size, but it is easier to handle 4 X 5-L flasks in incubators and sterile work benches. This volume is sufficient to generate inoculate for 1- to 200-L cultures. For process development, a number of 1- to 10-L fermenters (at least two) are used (Fig. 7-22) which are usually made of borosilicate glass. However, to prevent cells and positively charged carriers attaching, the glass needs to be siliconized. The silicone solvent must be washed away thoroughly before use.) The temperature of the fermenter is controlled by circulating water between the double jacket and a waterbath, or by a heating jacket. There is also complete pH, dOz, and stirrer control. Often, the controllers are also equipped with pumps for adding alkali and/or medium perfusion. The fermenter bottom should be rounded to give better mixing, and a marine impeller used to minimize shear forces. A spin filter is needed to run perfusion and sparge gas into the culture; in smaller reactors, this can be mounted on the stirrer axis. The mesh size is normally between 60-120 pm.
190
7 Microcarriers in Cell Culture Production
Fig. 7-22. B. Braun laboratory fermenter.
7.5.3 Large-scale Equipment - Stirred Tanks (Low Density) Larger stirred tanks are normally made of electropolished pharmaceutical grade stainless steel. Before using a stainless steel vessel for the first time; it should be washed with a mixture of 10 % nitric acid, 3.5 % hydroflouric acid, and 86.5 % water. Both specialized sampling devices and spin filters are commercially available for larger reactors. A conical-shaped reactor offers a number of benefits (Fig. 7-23). As the probes are placed low in the reactor, it is possible to vary the volume in the reactor over a greater range and thus retain control. In addition, there is a large headspace-to-surface area ratio for gassing.
7.5 Microcarrier Culture Equipment
191
Fig. 7-23. Example of a conically shaped stirred tank reactor.
The large stirred tank reactor is the most widely used reactor in the fermentation industry and has already been scaled up to 10 000 L. Its major advantages are its conventional design, proven performance, homogeneous results, and good potential for volumetric scale-up. Its disadvantages are that it is a low-intensity system with high shear forces at large scale and oxygen supply limitations.
7.5.4 Packed Beds There are two different approaches to this technology. The first is to pack a cage with plastic woven discs (Fig. 7-24) or other microcarriers and then place it inside a stir-
192
7 Microcarriers in Cell Culture Production
Fig. 7-24. Packed cage reactor (New Brunswick, Fibracel).
red tank. Medium is circulated through the cage using a marine impeller. Gas is supplied via sparging. Reactors of this type are available from New Brunswick (USA) and Meredos (Germany). The second approach is to pack glass beads into a column. Siran beads of 3-5 mm diameter are frequently used [29] (Fig. 7-25), though 5 mmdiameter beads normally give higher cell yields[30]. An airlift-driven system usually gives better oxygenation.
Fig. 7-25. Glass macroporous microcarrier for packed beds (Siran).
7.5 Microcarrier Culture Equipment
193
Low surface shear, high unit cell density and productivity, and no particle/particle abrasion are some of the major advantages of packed beds. However, poor oxygen transfer, channel blockage and difficulties in recovering biomass from the bed have all limited their use.
7.5.5 Fluidized Beds The Verax system and the Cytopilot are examples of two different designs of fluidized beds. 7.5.5.1 Fluidized Bed With External Circulation
Verax introduced this technology for animal cells [28]. The fluidized bed is equipped with an external recirculation loop connected with a gas exchanger (hollow fiber cartridge), heating elements, pO2, pH and temperature sensors, and a recirculation pump (Fig. 7-26). Certain aspects of the external circulation loop can be problematic, such as shear stress created by the pump, oxygenator fouling, and sterilization procedures in large scale. A gas exchanger transfers oxygen to the culture medium in the external loop. A certain oxygen tension at the entrance of the bed is achieved. The supernatant is increasingly depleted of oxygen on its way up the fluidized bed from the bottom to the top of the reactor (Fig. 7-26). The requirement of the oxygen tension from the biological system therefore limits reactor height and scalability. This problem of an oxygen gradient along the reactor, height can be overcome by integrating a membrane module directly into the fluidized bed as described in [31] (see also Fig. 7-29b). Reactors of this type are available from B. Braun, Melsungen Germany) with a settled carrier volume ranging from 20 mL up to 5 L.
Base addition
Fig. 7-26. Schematic drawing of the Verax system.
194
7 Microcarriers in Cell Culture Production
7.5.5.2 Cytopilot - Fluidized Bed With Internal Circulation The fluidized bed with internal recirculation (Cytopilot) was developed at the Institute for Applied Microbiology in Vienna, Austria, in co-operation with the company Vogelbusch GmbH, Vienna, Austria [32]. Cytopilot comprises a lower and an upper cylindrical chamber (Fig. 7-27). The lower chamber has a bottom adapted to the special flow conditions and is equipped with the following: a heating circuit via a double jacket, sampling and discharge facilities, a magnetic stirrer (rotating in both directions), and probe nozzles for pH and d02. The liquid agitated by the magnetic stirrer is conveyed via the distributor plate to the microcarriers in the upper chamber of the vessel. The hydrodynamic pressure lifts the settled microcarriers to form a fluidized bed with a clear boundary between the top of the fluidized microcarriers
IpH.
Fig. 7-27.The Cytopilot concept, including oxygenation by microsparging.
7.5 Microcarrier Culture Equipment
195
and the uppermost part of the medium volume. The bed expands or contracts as a function of the stirrer speed. The medium then flows through a sieve to the internal recirculation loop and back to the stirrer in the lower chamber of the vessel. Microbubbles of oxygen are sparged homogeneously into the downflow in the draft tube and then uniformly distributed by the impeller (Fig. 7-27). This system provides oxygen gas hold up via the dispersed oxygen bubbles (continuous transfer of d02 from gas to liquid). It both minimizes do:! gradients in the system and greatly increases the theoretical height of the fluidized bed. 7.5.5.3 Fluidization and Fluidization Velocity
If a carrier bed laying on a distributor plate is streamed through with medium, it will have a pressure drop (Ap). This pressure drop reduces the pressure (p) of the carrier bed on the plate. If Ap equals p , the bed will expand (Fig. 7-28). This is the fluidization point. The maximum fluidization speed depends on the sedimentation rate of the microcarriers. When the upward flow rate is higher than the sedimentation rate of the microcarriers, they will be carried to the top, which is known as flush out. Flush out velocities are between 10-to 100-fold higher than the velocity at the fluidization point. The efficiency of mixing in a fluidized bed is much higher than in many other systems. This efficiency increases with the particle size of the carriers. Because there are no channels between carriers for cells to block, and because it allows cells to be retrieved from the reactor, the fluidized bed represents an improvement over the packed bed reactor. Its scale-up potential and very good mass transfer are additional benefits. Drawbacks include particle/particle abrasion and shear stress affecting growth on the bead surface (This affects only 10 % of the available area of macroporous beads.)
a, 0
c
a,
a,
E W
4 fluidization point
flush out point
a,
2 a, v)
Q
velocitv
Fig. 7-28. Fluidization principles.
196
7 Microcarriers in Cell Culture Production
7.6 Culture Conditions 7.6.1 Media and Components It is important to note that a certain amount of medium will only generate a finite amount of cells of a particular cell line. This is independent of the culture technology chosen! Basic media formulations were originally DME and DMEM with added F10/12 or medium 199 enrichments. RPMI 1640 was often the base for hybridomas. Today, there is a multitude of different media developed for specific cell types, many of them serum-free. A number of suppliers provide collapsible plastic bags with ready-made media to be hooked-up to the bioreactor. The most recent development is liquid concentrates that can be automatically mixed with water on-line, sterilized, and fed into the bioreactor, (this system is available from Gibco). The media composition during growth and during maintenance of the culture can be quite different. Often serum is reduced or even completely removed at later stages of high cell density cultures. For some cell lines, it is even possible to run protein-free media at later stages of the culture. The buffer system frequently used in cell culture media, carbon dioxide/sodium bicarbonate, is a weak buffer system. At times, it is beneficial to change to a better buffering system. HEPES (10-20 mM) is, therefore, used as an alternative (331, especially in serum-free media where the buffering capacity of the serum needs replacing. Nutrients that are often depleted quickly include glutamine and, when growing human diploid cells, cysteine. Amino acid analysis will help determine utilization rates for different cell types. Glucose is used in a wasteful manner by cells, so it is desirable to start at concentrations below 2 g L-l and then add more after two to three days. An alternative is to switch to other sugars as sources of carbon and energy, for example, fructose or galactose.
Additives Common additives to low-serum or serum-free media are insulin (IGF-1 )(5 mg L-l, transferrin (5-35 mg L-I), ethanolamine (20 pM), and selenium ( 5 pg L-l). Mixtures of these supplements are now commercially available. Media for adherent cells should always contain Ca2+ and Mg2+ ions as these act as cofactors for adhesion. Sometimes, sodium carboxymethyl cellulose (0.1 %) is added to prevent mechanical damage to cells. Pluronic F68 (0.1 %) is used to reduce foaming and to protect cells from bubble shear forces in sparged cultures, especially when low- or serum-free media are used. Cyclodextrin or dextrans are also used in serum-free media as albumin replacements.
7.6 Culture Conditions
197
7.6.2 pH pH is very important during inoculation. Cell attachment to carriers with an electrostatic surface is highly dependent on the right pH (e.g., a pH of 7.4 is recommended for Vero cells and Cytodex 1 microcarriers). However, the pH of the medium has little effect on cell attachment to coated microcarriers [34]. An alkaline pH prevents/ prolongs attachment, and higher pHs kill the cells. The lower setting for pH is normally 7.0, but below 6.8, it becomes growth inhibitory. An incubator set point of 5 % COz is normally used, together with sodium bicarbonate to stabilize pH. Autoclavable probes are normally used in fermenter systems and an upper and lower pH set point chosen. If alkali is needed to compensate for an acidic pH, addition of 5.5 % NaHC03 is preferable to 0.2 M NaOH. NaOH can, however, be used in very wellmixed systems where it is not delivered directly onto cells. Note that silicone tubing is gas-permeable and can cause changes in pH during media transport.
7.6.3 Dissolved Oxygen The solubility of oxygen in aqueous solutions is very low (7.6 pg mL-'). The mean oxygen utilization rate of cells has been determined as 6 pg per lo6 cells h-l. Oxygen supply depends on the oxygen transfer rate (OTR = Kla (C*-C). OTR is one of the main limiting factors when scaling-up cell culture technology. Oxygen supply is often via surface aeration, and can be increased further by using medium perfusion, increased oxygen pressure, membrane diffusion, or by direct sparging of air or oxygen into the culture medium (Fig. 7-29). OTR will increase in vessels having a large height-to-diameter ratio because of the higher hydrodynamic pressure at the base of the vessel. Membrane diffusion is inconvenient because a lot of tubing is needed, which also makes it expensive. Medium perfusion and oxy-
a)
Fig. 7-29. Examples of d 0 2 supply devices. (a) Microsparger.
198
7 Microcarriers in Cell Culture Production
b) Fig. 7-29. (continued). (b) bubble-free aeration (Fluidized bed system of KFA Julich).
genation in a separate vessel have been particularly effective in microcarrier cultures. Sterilizeable 0.22-pm non-wettable filters are used to supply gases continuously to cultures.
7.6.4 Redox Potential The Redox potential represents the charge of the medium. It is a balance of oxidative and reducing chemicals, pO2 concentration, and pH. An optimum level for many cells is +7S mV, which equals a p02 concentration of 8-10% (approximately SO% of air saturation). The Redox potential falls under logarithmic growth and is at its lowest 24 h before the onset of stationary phase [3S,36].
7.6.5 Stirring Avoid stirrers with moving parts as these will damage both cells and microcarriers. Top-driven reactors are best if the carriers/cells are not physically separated from the stirrer. The geometry and speed of the impeller greatly influence OTR. With certain vessel designs and a marine impeller, 150 r.p.m. has been achieved without being detrimental to the cells.
7.6 Culture Conditions
199
7.6.6 Control and Feeding Strategies For good final product quality, it is important to have as steady state conditions as possible. For example, variations in sugar concentration will invariably affect final product glycosylations. In simple cell culture set-ups, carrier samples are taken, cells are counted, and morphology is examined (photographs taken) to document growth. As cells grow, stirrer speed is increased to compensate for the increased weight of the carriers (empty carriers 1.03 g mL-' cells 1.015-1.070 g mL-' depending on cell type). This keeps them suspended and increases gas transfer. After about three days in culture, the medium turns acidic and needs to be changed. Media supply is determined empirically in relation to growth and productivity, or by amino acid, glucose, or lactic acid analysis. In bioreactors, there is continuous control of pH, dO2, and stirrer speed, usually via a programmable controller. pH is normally initially controlled by adding C02 to lower pH. Nitrogen will wash out COZ and increase pH. Adding alkali via sodium hydrogen carbonate or NaOH solutions will increase pH during very high lactic acid and CO2 production. dO2 is controlled by adding air or, at high cell concentrations, pure oxygen. Increasing stirrer speed can be coupled to the increased demand for oxygen supply and increases OTR. To use oxygen in sparging, the cells need to be protected from the gas bubbles to avoid toxic effects. Media utilization is normally determined by taking samples to measure glucose concentration. Sampling is either done manually or automatically via a flow injection analysis biosensor system connected on-line to the bioreactor [37,38] (Fig. 7-30). Sampling device
FIA biosensor system
ffer
purge
Fermentation
Retention pump
Substrate pump
Potentiostat 2
1Fig. 7-30. Flow injection analysis (FIA) set-up.
Q
ProcessControl
200
7 Microcarriers in Cell Culture Production
These results are then used to determine the rate of manual media change or to regulate automatically the perfusion rate via pump speed (Fig. 7-30). In addition to glucose, glutamine and even metabolites like lactate can be used to measure media utilization. Ammonia can be measured on-line with the FIA biosensor system. Oxygen consumption rate can also be used to determine cell growth and be used as a parameter for feeding strategies when optimized and correlated with the metabolic rates of the substrates. By switching off the supply and looking at the linear consumption of oxygen over time, the number of cells can be calculated.
7.7 Microcarriers in Practice 7.7.1 Preparation of Carriers It is a major advantage if the carriers are supplied dry, as the exact amount needed can be weighed and then prepared in situ by autoclaving. If a mistake is made during preparation, the carriers can be re-sterilized. Pre-sterilized microcarriers must always be handled under sterile conditions and can normally not be re-sterilized by autoclaving. At large scale, it is advantageous to prepare the carriers directly inside the reactor because of the problem of heat transfer into vessels standing in autoclaves. The correct amount of carrier is added to an already cleaned and sterilized reactor. Then a volume of distilled water large enough for the microcarriers to swell correctly (hydrated microporous microcarriers) and be covered after settling is added to the fermenter. (Note that for soft microcarriers, the swelling factor in water is much larger than in PBS !) The reactor is then closed and the vessel re-sterilized while the carriers are stirred. The culture medium is made up from powder, allowing for the water volume remaining inside the reactor. It is then sterile-filtered, added to the reactor and all parameters equilibrated. The reactor is then ready for inoculation. This method works well with serum-containing media but caution is required when working with serum-free media, where additional washing, sedimentation, and decanting of washing buffer may be necessary prior to sterilization. Carriers for packed or fluidized beds are also normally prepared inside the reactor. At small scale, the entire reactor can be sterilized by autoclaving, while at larger scale, carriers are sterilized in situ.
7.7.2 Microcarrier Concentrations With packed beds, the airlift or cage is filled with beads or discs. In homogeneous stirred tanks, the concentration of surface microcarriers in batch cultures is normally 2-3 g L-l. However, this can be increased up to 15-20 g L-' (surface area of
7.7 Microcarriers in Practice
201
60 000-80 000 cm2 L-' when perfusing the culture. In fluidized beds, 10-50 % of the column volume is filled with carriers. Fluidization is normally better at higher concentrations. In packed beds, the cage or column is normally filled with microcarriers, the amount being related to the size of the cage or column.
7.7.3 Inoculum It is essential to equilibrate and stabilize all culture parameters before adding the inoculum. Temperature and pH are especially important. When inoculating, actively growing cells in logarithmic growth should be used and cells in stationary phase avoided. If the cells are stationary, i.e., in GO, they must be returned to the cell cycle by addition of growth factors. This is normally a 20- to 24-h process, which leads to an initial lag phase in growth. Stationary phase cells can cause the loss of one day process time at each subculture step. The cells used for inoculation should be single cells (Fig. 7-31) and absolutely not aggregates, which will otherwise cause heterogeneity. Surface carriers are normally inoculated with 0.5-2 x lo5 cells mL-' or 0.5-2 X lo4 cells cm-2. Macroporous carriers are normally inoculated with about the same cell concentrations. Inoculate the cells in one-third to one-half of the final volume. Stir immediately at the lowest speed to keep the microcarriers in homogeneous suspension and, after the attachment phase, add media to the final volume. If the vessel is very well suited for microcarriers, it may be possible to start directly with the final volume. In some processes, large volumes of cells are generated on microcarriers, harvested, and frozen. These cells are then thawed and used immediately as inoculum. The process is then scaled up two more steps before final production, which makes production more flexible.
Fig. 7-31. Single-cell suspension.
202
7 Microcarriers in Cell Culture Production
Fluidized beds are inoculated with 1-2 X lo6 cells mL-' of packed bed carrier volume. The inoculation density is highly dependent on both cell line characteristics and media composition. The stirring rate is also related to the shear sensitivity of the cell line, with normal stirring rates of between 20-100 r.p.m. Stirring speed is initially low and increased as the cell concentration increases. During inoculation, the fludized bed is run as a packed bed, and after - 5 h the bed can be fluidized. Consider from the beginning how to create the large quantities of cells needed to inoculate a production-scale fluidized bed. Imagine 1 kg of product is needed and that the anchorage dependent cell line produces 100 mg L-' carrier per day. A fluidized bed with 100 L of carriers has to be operated over a period of 100 days. Thus, 2 x 10" cells would be needed to inoculate these 100 L of carriers (2 x lo6 per mL of carrier). This would require 2000 roller bottles, each of 850 cm2. The problem could be overcome by: -
Preparing the inoculum as aggregates in a stirred tank; scaling-up on the same carriers via trypsinization; scaling-up on different carriers (smooth carriers to macroporous); and Carrier-to-carrier transfer.
For suspension cells, the easiest way is always to create the inoculum in stirred tanks before transfer to macroporous carriers. Figure 7-32 shows that the attachment phase after inoculation is finished after 2 h on macroporous carriers in fluidized bed applications. Fluidization can thus be started very soon after introducing the cells into the reactor.
1,5E+OB
I.OE+OB
t
4 0
1
2
3
4
5
Time (h)
Fig. 7-32. Cell attachment kinetics in fluidized bed culture.
6
7
8
9
7.7 Microcarriers in Practice
203
7.7.4 Cell Quantification Cell numbers can either be determined directly or indirectly.
Direct methods These include cell enumeration by counting whole cells (attached cells trypsinized) or crystal violet-stained nuclei [39]. Others are determining total cell mass via protein or dry weight measurements. Cell viability is more difficult to quantitate, but is often done via trypan blue (0.4 %) or erythrosine B (0.4 %) staining. Another alternative is the MTT colorimetric method [18]. Some cell culture systems cannot be sampled, and therefore numbers cannot be determined directly, nor morphology studied via hematoxylin staining/fixation. Indirect methods These measure metabolic activities such as glucose or oxygen consumption, lactic or pyruvic acid production, COz production, and increase in product concentration. They are quite useful during logarithmic growth, but can be misleading later. Another possibility is to measure enzyme concentrations in the culture. One example is lactate dehydrogenase (LDH), which is less dependent on the different phases of growth. A relatively new technique is to measure capacitance (Aber instruments), the advantage of this technology being that it measures cell mass even within macroporous carriers, and so monitors cell growth continuously within the bioreactor. All these cell counts provide a good picture of cell growth if the same detection method is always used. It is nearly impossible to compare different culture runs (performed with different cell lines) measured with different cell-counting methods. This is demonstrated by Capiaumont et al. [40], who found up to a factor of 8 difference in between the various cell-counting methods (Table 7-6). Table 7-6. Cell contentration determination techniques. Parameters
Cell count at the end of culture (lo6 per rnl)
Glucose Lactate Alkaline phosphatase @-Glucuronidase Alanine Mean glucose/lactate Counted nuclei
2.5 4 1.14 0.9 0.9 3.2 3
204
7 Microcarriers in Cell Culture Production
7.7.5 Scale-up Scale-up generally means a lengthy development period to ensure that all parameters are under firm control. It is best, therefore, to attempt working with a limited number of cell lines. Process simulation is also an essential part of scale-up and determines that the cells cope with scale-up without alterations and with maintained productivity. The viability and productivity of the cells has to extend beyond the planned time of the production process. An essential part of scale-up is the cellular multiplication in each scale-up step. If it is possible to inoculate at a low cell density, i.e., lo4 mI-' and the final yield is lo6 mL-', the multiplication factor is 1OOX. The larger the multiplication factor in each step of scale-up to final production volume, the fewer steps are needed for its achievement. It is important to maximize this as it minimizes the number of steps/operations needed ! This also affects the investments required. If the above applies, it is possible to inoculate a 100-L fermenter from a 1-L spinner. In the case of colonization (see Section 7.7.5.2), a 1-L spinner could inoculate a 1000-L fermenter. Scaling-up by volume normally involves moving from glass to stainless steel vessels, from mobile to static system, and from autoclavable to in situ sterilization. Additional equipment is also needed at larger scale, such as seed vessels, medium hold tanks, and sophisticated control systems. Scale-up of microcarrier cultures can be done by increasing the size of the vessel or by increasing the microcarrier concentration. Production units up to 4000 L have been achieved by increasing size. Factors that influence this strategy include reactor configuration and the power supplied by stirring. The height-to-diameter ratio is one important factor. For surface aeration, the surface area-to-height ratio should be 1:1. Variables that affect impeller function include shape, ratio of impeller to vessel diameter, and impeller tip speed. Larger impellers at lower speeds generate less shear forces, with marine impellers having been found to be more effective for cells. Pneumatic energy supplied via air bubbles or hydraulic energy in perfusion can both be scaled-up without increasing power input. Desirable features of a stirred microcarrier reactor include no baffles, curved bottom for better mixing, double jacket for heating and cooling, top-driven stirrer, and a smooth surface finish (electropolished). Scale-up by increasing the microcarrier concentration requires perfusion, which makes it necessary to have a separation device to keep the carriers in the reactor. A settling zone or a spin filter are both suitable for this task. The limiting factor for higher cell concentrations is usually oxygen supply. It is difficult to use direct sparging with normal microcarriers in stirred tanks, as the carriers may accumulate and float in the foam created. It can however be achieved if large bubbles are used or by sparging inside a filter compartment. Other alternatives that increase oxygen supply include increasing surface aeration, perfusion via an external loop and oxygenation device/vessel [41]. It is easier to increase oxygen supply via sparging in packed or fluidized beds as the cells are protected inside the macroporous carriers. Scale-up of fluidized beds is
7.7 Microcarriers in Practice
a)
205
b)
Fig. 7-33.Scale-up of fluidized bed technology. (a) Cytopilot Mini (100-500 ml bed volume). (b) 100 L Cytopilot fluidized bed reactor (up to 40 L bed volume).
linear, i.e., the diameter-to-height ratio is the same, which keeps the fluidization velocity constant. As oxygen is supplied via direct sparging with microbubbles (fluidized bed with internal loop), there is a continuous gas to liquid transfer throughout the bed, which makes the technology easily scaleable (Fig. 7.33a,b). Airlifts have been scaled-up in this way to 10000-L scale! 7.7.5.1 Harvesting The aim of this step should be to have an inoculum consisting of a single cell suspension of highly viable cells that were in logarithmic growth prior to harvesting. To achieve this, it is necessary to develop and follow a strict harvesting protocol to apply after the carriers have sedimented and the supernatant been decanted. The main task is to break the cell-surface and cell-cell interactions (the cells must be harvested prior to confluence) and to round up the cells. As cell binding is dependent on divalent ions, the carriers must be washed with citric acid or EDTA (0.2 %, w/v) containing buffers (phosphate-buffered saline, PBS) to help detach the cells. If the media contains serum, a-1-anti-trypsin, a trypsin-blocking protein abundant in serum, has also to be removed by washing. Microporous carriers require more extensive washing compared with solid carriers, normally two to five washings (this
206
7 Microcarriers in Cell Culture Production
should be optimized depending on cell/microcarrier/medium), with a volume of washing solution equal to the sedimented microcarrier volume. As trypsin has a pH optimum close to pH 8, it may be necessary to increase the buffer capacity of the PBS used for harvesting to maintain a pH of 7.4 throughout the procedure. Note that the substrate concentration for trypsin increases drastically for sedimented beads compared with ordinary flask cultures as the cell density becomes very high! The same ratio of units of trypsidnumber of cells should be maintained, if possible. Normally, this means increasing trypsin concentration tenfold (i.e., 0.2 % w/v, but this may depend on the supplier) compared with harvesting flasks. To speed up harvesting, both the washing solution and the trypsin solution should be pre-warmed to 37°C. As trypsin activity is dependent on Ca2+ ions, it is advantageous to separate the EDTA washes and the final trypsinization. Different suppliers should be screened for a suitable trypsin for a particular cell line; ‘crystalline’ trypsin should not be used. It is normally better to use a trypsin contaminated with other proteases, as this is usually more effective! Some shear force may have to be applied to quickly detach the cells. In spinners, stirring speed can be increased during this step or flasks shaken. At larger scale, the bioreactor can be emptied into a special harvesting reactor developed by Van Wezel (available from B. Braun) (Fig. 7-34). This vessel is divided into two compartments by a stainless steel 60-120 pm mesh filter. The upper compartment contains a Vibromixer, a reciprocating plate with 0.1-0.3 mm holes moving at a frequency of 50 Hz. The microcarriers are collected on top of the mesh. Washing is done by adding washing buffer and draining it through the mesh. Pre-warmed trypsin is added so that it just covers the microcarriers. This is left for some minutes (depending on the cell line), after which the Vibromixer is used for a short period to help detach cells. After detaching, the cells separate from the used carriers by draining through the mesh. Additional washing is done to improve yield. The filters work most efficiently with hard carriers (plastic, glass) that do not readily block the mesh. It is also pos-
Fig. 7-34. Vibromixer harvesting vessel.
7.7 Microcarriers in Practice
207
sible to detach only the cells and to transfer the whole mixture of used beads and cells to the new reactor. The trypsin has to be inactivated before the cells can be used as an inoculate. Add either serum to the harvest or aprotinin, soybean trypsin inhibitor, if serum-free cultures are required. If serum is used, it can be added at the same time as the cells. 7.7.5.2 Colonization
Cultures of cells that do not attach well to carriers or that detach easily during mitosis, can be scaled-up just by diluting the microcarrier culture with fresh microcarriers and more media [42]. A 1:lOOO dilution has been achieved with Chinese hamster ovary (CHO) cells (Fig. 7-35). Microcarriers have been used for protease-free transr
t
0
C e l I
C ~ + ~ c e r + , - a t i o n
,
6 I SCRLE-UPEO CULTURE
SEED CULTURE
5
I
5
0 CULTURE
T!ME
10
15
20
4
OFiYS
b)
Fig. 7-35. Carrier to carrier transfer (Cytopore). (a) Kinetics. (b) Different phases of colonization.
208
7 Microcarriers in Cell Culture Production
d)
Fig. 7-35.(continued). (c, d) Different phases of colonization.
fer of cells by simply placing them into blood vessels or onto cell culture surfaces and allowing the cells to migrate onto the surface. The migration method normally works better for transformed cell lines such as CHO, hybridomas, etc. 7.7.5.3 Suspension
Cells that grow in suspension can first be scaled-up by increasing the volume of the suspension cultures. During final production, the cells can be immobilized at high densities in carriers and the product continuously harvested via perfusion. 7.7.5.4 Documentation The most critical variable in cell culture production is the cell line itself. Therefore, it is essential to have thorough documentation, process control, and very strict proto-
7.8 Optimizing Culture Conditions
209
cols, for all steps from thawing the cells, through small-scale culture, and up to final production. This naturally applies also to evaluation studies, scale-up, and process development/trouble-shooting. Without such documentation, the reproducibility and quality of the production process will be very poor and the chances of success minimal. During each phase and step, document: 1. Cells; type, viability, plating efficiency (if attached), doubling time, split ratio at passaging, saturation density (cells cm-2,cells mL-', yield after harvesting, passage number. Check and keep track of morphology (photographic documentation). 2. Preparation of microcarriers; type, weightholume, batch, hydrationlwashing solution and procedures, sterilization, medium equilibration, culture parameter equilibration (pH, dOz, temperature). 3. Culture vessel; type, preparation, media volume, amount of microcarriers used, stirrer speed, air lift or fluidization rate. 4. Medium; batch, composition, preparation, additives and their batch numbers, buffer system. Serum testing, type, treatment before addition. 5. Gas supply; gases added, batches, set points, flowrates, aids for supply, stirrer, sparging, membrane, filters used. 6. Culture program; inoculation density, preparation of inoculum, inoculation volume, stirring during inoculation, maintenance program, set points, sampling. 7. Results; cell culture results of each step. Parameter control. 8. Cell harvesting and recovery: washing, chelator treatment, enzyme treatment, viability, yield, enzyme inactivation. 9. Scale-up; separation of carriers, colonization, suspension. 10. Productivity: media change, virus infection, multiplicity of infection, product concentration.
To help document your work, detailed microcarrier cell culture record sheets and problem-solving check lists are available from the authors.
7.8 Optimizing Culture Conditions The quickest way to determine the key variables in a culture is to set up factorial design experiments. Feed a limited number of experimental results into mathematical models and evaluate the variables that require attention and further optimization. The fewer the cell lines, the more in-depth the optimization that can be done. However, each recombinant construct has different growth properties and requirements. To fully optimize media utilization, analyze the amino acids and determine which factors in the medium are limiting. These factors can be added selectively, or the medium composition changed to better suit the cell type. If there is no time or resources for this analysis, use a surplus of medium to maintain steady-state conditions and to maximize growth rate.
210
7 Microcarriers in Cell Culture Production
7.9 Trouble-shooting 7.9.1 Stirred Microcarrier Cultures Some problems may arise when working with stirred microcarrier cultures for the first time. The following list summarizes typical areas of difficulty and the most likely solutions. These points also form a useful check list when culturing new types of cells. 1. Medium turns acidic when microcarriers added. - Check that the microcarriers have been properly prepared and hydrated. 2. Medium turns alkaline when microcarriers added. - Gas the culture vessel and equilibrate with 95 % air, 5 % CO2. 3. Microcarriers lost on surface of culture vessel. - Check that the vessel has been properly siliconized. 4. Poor attachment of cells and slow initial growth. - Ensure that the culture vessel is non-toxic and well washed after siliconization. - Dilute culture in PBS remaining after sterilization and rinse microcarriers in growth medium. - Modify initial culture conditions, increase length of static attachment period, reduce initial culture volume, or increase the size of the inoculum. - Check the condition of the inoculum and ensure it has been harvested at the optimum time with an optimized procedure. - Eliminate vibration transmitted from the stirring unit. - Change to a more enriched medium for the initial culture phase. - Check the quality of the serum supplement. - If serum-free medium is used, increasing attachment protein concentration may be necessary (fibronectin, vitronectin, laminin). - Check for contamination by mycoplasma. 5. Microcarriers with no cells attached. - Modify initial culture conditions, increase length of static attachment period, reduce initial culture volume. - Improve circulation of the microcarriers to keep beads in suspension during stirring. - Check the condition of the inoculum, especially if it is a single-cell suspension. - Check that the inoculation density is correct (number of cellshead). 6. Aggregation of cells and microcarriers. - Modify initial culture conditions, reduce the time that the culture remains static. - Increase stirring speed during growth phase, improve circulation of microcarriers. - Reduce the concentration of serum supplement as the culture approaches confluence. - Reduce the concentration of Ca2+ and Mg2+ in the medium. - Prevent collagen production by the cells by adding proline analogs to the culture medium.
7.9 Trouble-shooting
21 1
7. Rounded morphology of cells and poor flattening during growth phase. - Replenish the medium. - Check the pH and osmolality of the culture medium. - Reduce the concentration of antibiotics if low concentrations of serum are used. - Check for contamination by mycoplasma. 8. Rounding of cells when culture medium is changed. - Check temperature, pH, and osmolality of replemshment medium. - Reduce the serum concentration. 9. Cessation of growth during culture cycle. - Replenish the medium or change to a different medium. - Check that pH is optimal for growth. - Re-gas the culture vessel or improve supply of gas. - Reduce stirring speed. - Check for contamination by mycoplasma. 10. Difficulties in controlling pH. - Check that the buffer system is appropriate. - Improve the supply of gas to the culture vessel, lower the concentration of COZ in the headspace, or increase the supply of oxygen. - Improve the supply of glutamine, supplement the medium with biotin, or use an alternative carbon source, e.g. galactose. 11. Difficulties in maintaining confluent monolayers. - Check that pH and osmolality are optimal. - Reduce the concentration of serum supplement. - Improve the schedule for medium replenishment. - Reduce the concentration of antibiotics. - When culturing cell lines that produce proteases in a serum-free medium, it may be necessary to add protease inhibitors to prevent the cells from detaching (CHO cells have been shown to secrete proteases!). 12. Broken microcarriers. - Ensure that dry microcarriers are handled carefully. - Check the design of the culture vessel/impeller and ensure in case of a bottomdrive agitation system that the double mechanical seal is designed properly, e.g., no existing graps for microcarriers. 13. Difficulty in harvesting cells from microcarriers. - Ensure that the carriers have been washed extensively together with mixing. - Check that approximately the same amount of protease (U per cell) is used as when harvesting from flasks. - Check that the trypsin has not been thawed for too long (loss of activity). - Check that sufficient shear force is used in addition to the trypsinization. 14. Flotation of microcarriers in foam due to sparging. - Reduce the serudprotein concentration as much as possible. - Add pluronic F68 to decrease foaming. - Add polymers to increase viscosity. - Aerate via silicone tubing (Diesel, bubble-free aeration), via spin filter (New Brunswick, Celligen), via external loop (vessel, hollow fiber).
212
7 Microcarriers in Cell Culture Production
7.9.2 Fluidized Bed Trouble-shooting 1. Culture medium too acidic. - Expel CO2 by using a sparger that creates large bubbles. - Add sodium hydroxide to titrate the pH, observe the osmolarity. - Try to increase buffer capacity. - Optimize the culture medium and oxygen support to avoid production of lactic acid. 2 . Bridging of microcarriers. - Use higher circulation rates. This shortens contact time between the microcarriers. Bed expansion should be between 150 and 200%. 3. No attachment. - Initial cell density too low. - pH level and do:! concentration were incorrect during inoculation. - Contact time between cells and the microcarriers was too short. More time is needed in packed bed mode. - Microcarriers were not washed properly. - Microcarriers were not equilibrated. - Cell inoculum was in stationary phase and not in the exponential growth phase. 4. Oxygen supply too low - Use microsparging technique with a sparger that creates small bubbles (pore size around 0.5 pm). - Use pure oxygen for gasing. The cells inside the pores of the macroporous microcarriers are protected against the toxicity of oxygen. - increase the circulation rate.
7.10 Applications Well over 500 publications reflect the many applications of microcarrier technology and the great number of different cell lines cultured (Table 7-7). Today, its main industrial use is to produce vaccines, natural and recombinant proteins and, increasingly, monoclonal antibodies. An interesting minor application is its use in artificial organs (livers). The number of applications run using microcarrier technology may be affected by the choice of producer cell and its glycosylation pattern. If more natural adherent cell lines are chosen, this will greatly increase the applications of the technology.
7.10 Applications
Table 7-7. List of cells grown on microcarriers Tissue
Cell line
Adrenal
Mouse cortex tumour - Y-1
Amnion
Human
Amniotic cells
Human amniotic fluid
Bone marrow
Human Human
-
Carcinoma
Human Human Human Human Human Human
nasal - RPMI 2650 larynx - HEp 2 oral - KB cervical - HeLa colon thuroid
Conjunctiva
Human
-
Cornea
Rabbit - SIRC
Endothelium
Rabbit coronary endothelium Human coronary endothelium Mouse brain capillary endothelium Bovine pulmonary artery endothelium
Epithelium
Human
Fibroblast
Human foreskin - FS-4 Human foreskin Detroit 532 Human - SV40 - transformed WI-38 Mouse - SC-1, 3T3, 3T6, L-cells, L-929, A9 Mouse - transformed Mouse - embryo Chicken - embryo Human - embryo Rat - embryo Rabbit - embryo Human - Xeroderma pigmentosum Muntjac - adult skin Human - HT 1080 Mouse
Fibrosarcoma
-
-
-
WISH
Detroit 6 Detroit 38
Chang D
NITC 2544
Fish
Rainbow trout gonad - RTG Fat head minnow - FHM Carp epithelioma - EPC
Glial
Rat
Glial tumor
Rat - C6
Glioma
Human
Heart
Human atrial appendage - Girardi heart
Hepatoma
Rat
-
HTC, Morris MHlCl
2 13
214
7 Microcarriers in Cell Culture Production
Table 7-7. (continued). Tissue
Cell line
Insect
Drosophila Spodoptera Trichoplusia
Kidney
Human embryo Human embryo - Flow 4000, L-132 Bovine embryo - MDBK Monkey - primary Dog - primary, MDCK, transformed Rabbit - primary, NZ White, LLC-RK RK- 13 Rat - NRK, transformed Pig - PK-15, IBR Syrian hamster - HaK, BHK, transformed Potroo - Pt-k-1 Rhesus monkey - LLC-MK2 African Green monkey - Vero, CV-1 BSC-1, BGM, GL-V3
Leukemia
Human monocytic
Liver
Human primary hepatocytes Rat primary hepatocytes Chimpanzee Human - Chang liver
Lung
Chinese hamster - Don Chimpanzee embryo - CR-1 Human embryo - L-132, MRC-5, MRC-9, WI-38, IMR-90, Flow 2002, HEL 299 Cat embryo Bat - Tb 1 Lu Mink - Mv 1 Lu
Lymphoid
Human - lymphoblastoid Human - lymphocytes
Macrophage
Mouse - peritoneal, peripheral blood Rat - peritoneal Human - peripheral blood Mouse - P388D1
Melanoma
Human Mouse
Muscle
Chicken myoblasts Rat muscle-derived fibroblasts
Neuroblastoma
Mouse - Neuro-2a
Osteosarcoma
Human
ovary
Chinese hamster
-
-
J 111
CHO
7.10 Applications
2 15
Table 7-7. (continued). Tissue
Cell line
Pancreas
Rat
Pituitary
Rat Bovine
Rhabdomyosarcoma
Human
Synovial fluid
Human - McCoy
Thyroid
Pig
-
RD
7.10.1 Vaccines A vast majority of vaccine producers in Europe, and many others world-wide, use surface microcarriers to produce live attenuated or inactivated vaccines for human and veterinary use (Fig. 7-36a;b) [43, 441. Recently, they have begun to produce viral vectors used in gene therapy, adenovirus, and murine retroviruses, with both lytic and non-lytic viruses being produced. As the cells are eventually killed, natural batch or fed-batch processes are used. Normally low cell densities are cultured in stirred tank cultures for this purpose. One novel application is to use diploid MDCK epithelial cells to produce influenza vaccine [45].
Fig. 7-36. (a) Vaccine production phases (4-45 h). Vero cell growth and virus production of Herpes simplex virus on Cytodex.
216
7 Microcarriers in Cell Culture Production
Fig. 7-36.(continued). (b) Virus production kinetics (reproduced from [43]).
7.10.2 Natural and Recombinant Proteins A number of processes for naturally produced proteins are based on the culture of diploid cell lines on surface microcarriers in stirred tanks. Most new recombinant proteins are expressed in CHO cells. They attach and grow intially on surface microcarriers, but after some days aggregate and begin to fall off. The majority of CHO cells have been adapted for suspension culture and grown at fairly low cell densities in stirred tanks in batch or fed-batch cultures. Lately, however, some processes utilizing maroporous microcarriers to increase cell density have been developed. (Fig. 7-37a,b).
a)
Fig. 7-37.(a) SEM micrograph from a cut of an empty Cytopore.
7.10 Auulications
2 17
Fig. 7-37.(continued). (b) of a confluent CHO culture.
This has in some cases led to increased productivity. In the report by Shirokaze et al. [46], a double productivity of r-I4 could be obtained with immobilized culture compared with suspension. Production was measured by ELISA over an ll-day period. The total productivity in suspension in 10% calf serum was 2 mg, serum-free 1.8. In immobilized culture it was 3.8 and 3.2 mg, respectively. These run as perfusion cultures utilizing spin filters in stirred tanks. In addition, some processes are being set up using r-CHO cells in fluidized bed cultures up to 100 L reactor volumes (60 L fluidized bed volume). 7.10.2.1 Comparison of Carriers in Different Reactors (Packed Bed or Fluidized Bed Reactor)
In this experiment, a r-CHO expressing a recombinant protein was cultivated in a packed bed and a fluidized bed reactor and the productivity and functionality of the product compared (Fig. 7-38) [47]. Due to the better mixing in the fluidized bed, the cells have a better nutrient and oxygen supply (no channeling). Because of this homogeneous environment, a 10-fold increase in productivity could be observed in the fluidized bed.
2 18
7 Microcarriers in Cell Culture Production
300
T
0 0
200
400
600
800
1000
1200
Time (h)
I -+Fluidized bed
+ Packed bed
Fig. 7-38. Productivity comparison of packed bed versus fluidized bed.
7.10.3 Monoclonal Antibodies The increased use of monoclonal antibodies in medical therapy requires cost-efficient production as the dose per patient is normally large. Packed bed technology was used for production in the past [48]. With the availability of scaleable fluidized bed technology, high productivity process can now be run over prolonged periods to produce active antibodies.
7.10.3.1 Comparison of Anti-HIV Monoclonal Antibody Productivity with Different Processes These experiments were performed with the Xenohybridoma cell line 3D6/LC4 expressing a human anti-HiV-1 gp 41 antibody and a r-CHO cell line constructed at the institute of Applied Mikrobiology, Vienna by co-transfecting CHO cells with two expression plasmids carrying the cloned cDNA of the heavy and light chains of the antibody. Cell number was determined by counting nuclei with a Coulter counter after treating the carrier with 0.1 M citric acid containing 1 % Triton X-100. Human IgG concentration was analyzed by anti-human IgG-ELiSA. The stirred tank cultures were run as batch, fed batch, and continuous cultures using the hybridomas. The fluidized bed used hybridomas and r-CHO with Cytoline 1 and Cytoline 2 as the macroporous matrix. In addition, a stirred system using Cytoline 2 with r-CHO was run. Table 7-8 shows two calculations of productivity: productivity per liter of carriers, and productivity per liter of reactor volume. The fluidized bed gave a 10-fold
7.11 Potential Future Applications
219
Table 7-8. Comparison of HIV monoclonal antibody productivity with different processes.
Reactor
Cell density (cell number/ mL carrier)
Productivity/ L carrier (mg L-I . day)
Productivity/ L reactor volume (mg L-' . day)
Stirred tank reactor batch Stirred tank reactor fed batch Stirred tank reactor continuous Fluidized bed Cytoline 1 Fluidized bed Cytoline 2 Fluidized bed r-CHO, Cytoline 1 Stirred system r-CHO, Cytoline 2
6 x lo5 4-5 x 105 4-5 x 105 4 x 106 5-7 x 106 1.4 X I O8 6 X lo7
2.5 4.5 5.5 40 110 30 150
2.5 4.5 5.5 20 55 150 37.5
increase over the stirred culture. Up to a 30-fold increase was obtained by combining CHO cells as the expression system for antibodies with fluidized bed technology and Cytoline microcarriers as the matrix for the cells.
7.10.3.2 Comparison of a Hollow Fiber Reactor with a Fluidized Bed Reactor Comparison of a hollow fiber reactor with a fluidized bed reactor [49] showed that the former enables the production of IgA at high level (> 1 g L-') but with a negative effect on the fraction of active material. The percentage of active fraction was between 30 and 77 %. In the fluidzed bed reactor cultivated on macroporous carriers, the active fraction was higher than 80 % (personal communication).
7.11 Potential Future Applications Applications currently being evaluated are cell expansion of blood cells (Fig. 7-39) via immobilization of hematopoietic stem cells in bioreactors, and the expansion of cytotoxic lymphocytes to generate sufficient cell numbers to be used in cell therapy. In some applications, cells are encapsulated inside a capsule (to protect them from the immune system) and then transplanted [50]. Macroporous microcarriers could easily be encapsulated when they are confluent with cells. Epithelial cells grown on Cytodex microcarriers are used for wound healing (burns) [51]. Growing cells on (degradable) microcarriers and using the entire cellkarrier complex in transplantation is also discussed in the report by Schugens et al. [52].
220
7 Microcarriers in Cell Culture Production
Plasma Cell
Neutrophil
Fig. 7-39. Hematopoietic stem cell (HSC) expansion [50].
References [ l ] Naveh, D., Japanese Association for Animal Cell Technology, 1995. [2] Griffiths, J.B. (1992) in: Animal Cell Culture, Freshney, R. I. (Ed.), 2"* ed., Oxford University Press, 1992, pp. 47-93. [3] Maroudas, N. G., J Theoret Biol, 1975, 49, 417-427. [4] Carrel, A,, J Exp Med, 1923, 38, 407. [5] Butler, M. (1988), in: Animal Cell Biotechnology, Vol. 3, Spier, R. E., Griffiths, J. B., (Eds.), Academic Press, Inc, pp. 284-300. [6] Keller, J., Dissertation, ETH, Zurich, 1991. [7] Lehmann, J. H., Piehl, G. W., Schulz, R., Dev Biol Stand, 1987, 66, 227-240. [8] Spier, R. E., Whiteside, J. P., in: Animal Cell Biotechnology, Vol. 4, Spier, R. E., Griffiths, J. B. (Eds.), Academic Press, pp. 123-132. [9] Reiter, M., Zach, N., Gaida, T., Bluml, G., Doblhoff-Dier, O., Unterluggauer, F., Katinger, H., in: Animal Cell Technology: Developments, Processes and Products, Spier, R. E., Griffiths, J. B., MacDonals, C. (Eds.), 1992, Buttenvorth-Heinemann Ltd, Oxford, pp. 386-390.
References
22 1
[lo] Zhan, S., Handa-Corrigan, X., Spier, R. E., Biotechnol Bioeng, 1993, 41, 685-692. [ I l l Griffiths, J.B., in: Animal Cell Biotechnology, Vol. 4, Spier, R.E., Griffiths, J. B., (Eds.), 1990, pp. 147-166. [I21 Cherry, R. S., Paputsakis, E. T., Biotechnol Bioeng, 1988, 32, 1001-1014. [13] Stathopoulos, N. A,, Hellums, J. D., Biotechnol Bioeng, 1985, 27, 1021-1026. [14] Spier, R.E., Crouch, C. E., Fowler, F., Dev Biol Stand, 1985, 66, 255-262. [15] Bancel, S., Hu, W. S., Biotechnol Prog, 1996, 12, 398-402. 1161 Nikolai, T. J., Peshwa, M. V., Goetghebeur, X., Hu, W. S., Cytotechnology, 1991, 5, 141-146. [17] Peshwa, M. V., Kyung, Y. S., McClure, D. B., Hu, W. S., Biotechnol Bioeng, 1993, 41, 179187. [18] Mosmann, T., J Immunol Meth, 1983, 65, 55. [19] van Wezel, A.L., Nature, 1967, 216, 65-65. [20] Van Hemert, P. A., Kilbum, D. G., van Wezel, A.-L., Biotechnol Bioeng, 1969, 11, 875-885. [21] Levine, D. W., Wong, J. S., Wang, D. I. C., Thilly, W. G., Somatic Cell Genetics 1977, 3 , 149-1 55. [22] Nilsson, K., Buzsaky, F., Mosbach, K., Biotechnology, 1986, 4, 989-990. [23] Young, M. W., Dean, R.C., Biotechnology, 1987, 5, 835-837. [24] Seifert, D., Klausmann, N., et al. Williamsburg Bioprocessing Conference, 1994. [25] Martens, D., Nollen, E. A. A., Hardeveld, M., Van Der Velden, X., De Groot C. A. M., De Gooijer, C., Beuvery, E.C., Tramper, J., Cytotechnology, 1996, 21, 45-59. [26] Vournakis, J. N., Runstadler, P. W., Biotechnology, 1989, 7, 143-145. [27] Reiter, M., Bluml, G., Zach, N., Gaida, T., Kral, G., Assadian, A., Schmatz, C., Strutzenberger, K., Hinger, S., Katinger, H., Ann NYAcad Sci, 1992, 665, 146-151. [28] Rundstadler, P. W., Cemek, S. R., in: Animal Cell Biotechnology, Vol. 3 , Spier, R. E., Griffiths, J. B. (Eds.), Academic Press Inc., London, pp. 306-320. [29] Whiteside, J. P., Spier, R. E., Biotechnol Bioeng, 1981, 23, 551-565. [30] Looby, D., Griffiths, J. B., Cytotechnology, 1988, I , 339-346. [31] Biselli, M., Born, C., Schroder, B., Wandrey, C., Gasification tube module and reactor for cell cultivation, US Patent 5,601,757, 1997. [32] Reiter, M., Bluml, G., Gaida, T., Zach, N., Doblhoff-Dier, O., Unterluggauer, F., Noe, M., Plail, R., Huss, S., Katinger, H., Biotechnology, 1991, 9, 1100-1102. [33] Good, N. E., Biochemistry, 1963, 5, 467. [34] Berry, J. M., Butler, M., Biotechnol Bioeng, 1996, 50, 627-635. [35] Toth, G. M., in: Cell Culture and its Applications, Acton, R. T., Lynn, J. D. (Eds.), Academic Press, New York, p. 617. [36] Griffiths, J.B., Dev Biol Stand, 1984, 55, 113. [37] Loibner, A. P., Zach, N., Doblhoff-Dier, O., Reiter, M., Bayer K., Katinger, H., in: Animal Cell Technology: Products of Today, Prospects for tomorrow, Spier, R. E., Griffiths, J. B., Bertold, W. (Eds.), Butterworth Heinemann, pp. 372-375. [38] van der Pol, J. J., Spohn, U., et al., J Biotechnol, 1994, 37, 253-264. [39] van Wezel, A. L., in: Animal Cell Biotechnology, Vol. 1, Spier, R. E., Griffiths, J. B. (Eds.), Academic Press Inc., London, pp. 266-281. [40] Capiaumont, J., Legrand, C., Gelot, M.A., Straczek, J., Belleville, F., Nabet, P., J Biotechnol, 1993, 31, 147-160. [41] Prokop, A., Rosenberg, M. Z., Adv Biochem Eng, 1989, 39, 29. [42] Kamiya, K., Yanagida, K., Shirokaze, J., in:Beuvery, E. C., Griffiths, J. B., Zeijlemaker . (Eds.), Developments towards the 21 Century. Kluwer Academic Publisher. pp. 759-763. [43] Meignier, B., Mangeot, A., Favre, H., Dev Biol Stand, 1981, 46, 249-256. [44] Aunins, J. G., Bibila, T. A., Gatchalian, S., Hunt, G. R., Junker, G. R., Lewis, J. A., Licari, P., Ramasubramanyan, K., Ranucci, C.S., Seifert, D.B., Zhou, W., Waterbury, J., Buckland, B. C., in: Animal Cell Technology, 1997, Carrondo, M. J. T., eta 1. (Eds.), pp. 175-183. [45] Brands, R., van Scharrenburg, G. J. M., Palache, A.M., in: Animal Cell Technology, 1997, Carrondo, M. J. T., et al. (Eds.), pp. 165-167.
222
7 Microcarriers in Cell Culture Production Shirokaze, J., Yanagida, K., Shudo, K., Konomoto, K., Kamiya, K., Sagara, K., in: Developments towards the 21"' Century, beuvery, E. C., Griffiths, .IB., . Zeijlemaker . +(Eds.), Kluwer Academici Publisher, pp. 877-881. Mudie, D., Karampetsos, L., Byrne, S., Animal Cell Technology: From Vaccines to Genetic Medicine. 14th ESACT Meeting, Vilamoura, Portugal, May 20-24, 1996. Bliem, R., Oakley, R., Matsuoka, K., Varecka, RT, Taiariol, V., Cytotechnology, 1990, 4, 279-283. Stoll, T.S., Ruffieux, P.A., Liillau, E., von Stockar, U., Marison, I.W., in: Immobilized Cells: Basic and Applications, 1996, Wijffels, R. H., Buitelaar, R. M., Bucke, C., Tramper, J. (Eds.), Elsevier, pp. 608-614. Edginton, S. M., Biotechnology, 1992, 10, 1099-1106. Dimoudis, N., Hartinger, A. (1994) PCT Int Appl96 12510, P:35pp., CL: A61L27/00, PRTY APPL: 4438015 (DE). Schugens, C., Grandfils, C., eta 1, J Biomed Muter Res, 1995, 29, 1349-1362.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
8 Purification and Characterization of Monoclonal Antibodies Paul Matejtschuk, Rose M. Baker and George E. Chapman
8.1 Introduction In the 20 years since the hybridoma technique for making monoclonal antibodies (MAbs) was reported [l], they have become immensely important in many diverse areas of biological and medical research. More recent in vitro methods of generating and selecting specific antibodies from the vast number of potential permutations of antigen-binding specificities [ 2 ] have contributed to their practical application. The ability to screen large combinatorial MAb libraries (typically expressed on the surface of bacteriophages) for the desired antigen-binding property without the need for in vivo immunization has a number of advantages over the traditional methods: (i) the time taken to assemble a panel of MAbs to a specific antigen is greatly reduced; (ii) the panel is likely to be considerably larger; and (iii) MAbs to highly toxic antigens can be produced. It is now possible to buy MAbs to a large number of protein, polysaccharide, and nucleic acid antigens ‘off the shelf’ for research use. Such uses may include immunoassays, immunoaffinity purification, blocking of enzymatic or biological function, and fluorescenthadioactive labeling for microscopic or autoradiographic visualization of tissue components. It is likely that we will see much wider industrial applications of MAbs in the future. The use of MAbs in highly specific biosensors is already being realized. MAbs to transition-state analogs frequently possess enzymatic activity, and this is completely independent of whether enzymes of that specificity exist in nature [3]. Thus, it is possible to generate completely novel enzymes (‘mabzymes’) which may have uses in chemical, biochemical, and pollution control applications. MAbs have played a significant and increasing role in medicine, firstly as reagents for diagnostic tests but now also as in vivo diagnostic (imaging) agents and as therapeutic agents, either by themselves or as the targeting moieties of conjugates with toxins, drugs, or enzymes. The emergence of techniques for engineering the MAb molecule, replacing the potentially antigenic species-specific parts of the antibody by their human equivalents, has been particularly important for development of MAbs as therapeutic agents [ 2 ] . An alternative approach to generating human MAbs has been to engineer the animal: transgenic mice with human immunoglobulin genes [4], which is an ingenious solution to the apparent inability to generate stable and productive human hybridoma cell lines.
224
8 Purification and Characterization of Monoclonal Antibodies
8.2 Methods of MAb Production The original rodent hybridoma method [ 11 involved the fusing of a rodent B lymphocyte with a non-producer mouse myeloma to give an immortalized cell line producing the MAb from the original B lymphocyte. Rodent-human heterohybridomas have been produced by fusing human B lymphocytes, (usually transformed with Epstein-Barr virus, EBV) with a non-producer mouse myeloma [5] and these will of course produce human MAbs. In the early days of hybridoma technology, the MAbs were frequently produced by in vivo ascites culture in mice, and the first licensed therapeutic MAb (OKT3) was made in this way. However, ascites culture is rapidly being consigned to history on considerations of quality, animal ethics and lack of scaleability, and in vitro cell culture has largely taken over. Whatever type of hybridoma cell line is created, they are transformed cell lines and are not anchorage-dependent. Hence they can be (and usually are) grown by suspension culture in stirred vessels or airlift fermenters. Under these conditions, the culture media often has to be supplemented with animal serum, or at least some protein. This places major demands on the downstream cell separation and purification, because the MAb is then a minor protein component of a large volume of dilute cell culture supernatant. However, high-density cell culture techniques are being used to an increasing degree for hybridoma production. This is exemplified by hollow fiber cell culture, consisting of a cartridge full of hollow semi-permeable fibers, where the cells are grown on the outside of the fibers. The media is pumped through the lumen of the hollow fibers, providing the packed cells outside the fibers with nutrients and oxygen via the walls of the fibers and taking away waste products. There is slow perfusion of protein-supplemented media through the cell mass and the MAb which is secreted into this media cannot, by virtue of its molecular size, pass back through the fiber walls. The net result is a much higher concentration of MAb and a higher ratio of product to high-molecular weight supplements than for low-density suspension culture. The lower harvest volume is also very attractive for scaling-up processes. Combinatorial libraries typically use Escherichia coli as the expression vehicle and the various DNA manipulations and graftings that are now becoming commonplace in MAb technology are also performed in E. coli. Variable domain fragments usually fold correctly and can be made to undergo periplasmic secretion, with the light and heavy chain domains either allowed to associate spontaneously or (more efficiently) by engineering them so that they are linked covalently by a linker peptide or disulfide bond. However, expression of intact MAbs in bacteria has been very disappointing. The lack of correct folding probably contributes to their tendency to form inclusion bodies in E. coli. It appears that correct folding and subsequent secretion of MAbs needs eukaryotic molecular chaperone systems. Another problem with bacterial expression of intact Mabs is the lack of glycosylation. IgGs are atypical glycoproteins: the major glycosylation sites, at the hinge ends of the CHZdomains, give rise to glycans which do not project outwards into the solvent but inwards to form a spacer between these domains, which is essential for many of the effector functions mediated by the Fc part of the molecule. Thus, it is unlikely that bacterial expression systems will ever find a major role in the production of therapeutic
8.3 Purification
225
MAbs. However, they are a relatively inexpensive (and hence attractive) option for making MAb fragments where the only required function is antigen binding, for example in enzymatic catalysis and immunoaffinity applications. Recombinant MAbs have been expressed in various mammalian cells. Myeloma cell lines appear to offer the best expression, but Chinese hamster ovary (CHO) is also widely used for production of therapeutic MAbs because it is such a microbiologically clean and well-characterized cell line. Expression in yeast has also been attempted [6] but the yields were low and glycosylation of MAbs in yeast will almost certainly be different from that which occurs in human cells. Glycosylation can be an important issue for any expression system used to make a MAb for therapeutic use. There are significant differences in Mab glycosylation patterns between mammalian expression systems, and even between the same cellular expression system under different cell culture conditions. These can give rise to rapid clearance or antigenicity in vivo.
8.3 Purification From a regulatory point of view, biological products and chemical entities have been approached somewhat differently from each other and this has had important consequences for the purification and characterization of biological products. Increasingly, as higher-purity protein products are developed and more discriminating protein characterization methods are perfected, this distinction is disappearing. Formerly it was accepted that it might not be possible to fully define a biological product and it followed from this that the consistency and control of the manufacturing process was critical to ensure product consistency. Now the emphasis is shifting to the concept of the well-characterized molecule which allows for demonstration of product consistency despite changes in purification process. This, in turn, allows greater flexibility of purification method during product development and diminishes the need to establish a suitable purification scheme early which can be scaled-up with identical performance.
8.3.1 Initial Considerations The following points must be considered when devising a purification scheme for a MAb.
8.3.1.1 Intended Use
The first consideration in the choice of purification strategy is the required specification: what purity is required for a given application? If the MAb is to be used as a therapeutic or in vivo diagnostic (imaging) agent, then not only must it be of high
226
8 Purification and Characterization of Monoclonal Antibodies
purity, but it must also be free of potentially harmful trace contaminants such as viruses, bacterial toxins, cytokines, and DNA (though regulatory requirements are relaxing on the last). These requirements also apply to MAbs used in the immunoaffinity purification of therapeutics. In these cases the concept of purification factor is largely inappropriate, and the performance of a purification step is measured in terms of clearance factors for defined contaminants. The required clearance factors are usually so large that they can only be achieved using chromatographic methods of purification. The performance of precipitation methods is inherently limited by interstitial supernatant in the precipitate. They are, however, often used in the early stages of purification schemes, followed by chromatographic steps. For in vitro diagnostic or research tools there may only need to be minimal [7-91 or no [lo] purification. Important components of the specification of the purified product in this case will be the affinity of the antibody and absence of cross-reactivity in an assay.
8.3.1.2 Culture Method The earliest MAb preparations were made in mouse ascitic fluid and this method of production produces a relatively concentrated feedstock which may be preferable for production of material on a small scale for diagnostic or research purposes. Larger amounts are now usually produced in a cell culture system which may give a very dilute product requiring concentration before further processing. The type of production system used has another profound implication for downstream processing as it determines the nature of contaminants present.
8.3.1.3 Contaminants Mouse ascitic fluid has a high protein concentration [ 111 of which - 40 % may be the MAb of interest [12]. It will contain other mouse IgG, high levels of lipids, and possibly also infectious agents from the mouse. A summary of the types of contaminant which might be encountered in MAb cell culture systems is shown in Table 8-1, although cell culture supernatants may vary widely in the type of contaminants they contain. It is an advantage if the cells can be grown in serum-free medium as this will mean that the problem of separating bovine IgG (introduced as a component of fetal or new-born calf serum) from the MAb does not have to be addressed. There are likely to be other proteins, such as albumin and transferrin, present, as well as many small molecular weight nutrients, waste products, and lipids. Many cell lines, even human ones, may have been fused with, or exposed to, other mammalian cell lines at some point in their development and infectious agents (particularly viruses) which may have been derived from this source must be considered. Some cell lines may have been transformed with a virus such as EBV. If a product is to be used therapeutically, extensive screening of the cell lines for adventitious agents is obligatory and this will give invaluable information for designing and vali-
8.3 Purification
227
Table 8-1. Potential contaminants in cell culture supernatants.
Component
Potential contaminants Serum-supplemented medium
Serum-free medium
Water
Trace elements Organics Endotoxins
Trace elements Organics Endotoxins
Powdered media
Minimal nutrients Phenol red Endotoxins
Full nutrients Phenol red Endotoxins
Proteins
From animal sera: Albumin (50-60 9%) Immunoglobulins (- 10% in whole serum, < 0.1 % in fetal calf serum) Protease inhibitors (- 10 9%) Transferrin (2-5 %) Lipoprotein (1-2 %) Peptide hormones (< 0.1 9%) Misc. proteins (20-35 9%) Proteases (Total 30-50 mg mL-')
Albumin (0.4-0.5 mg mL-') Transfernin (0.03 mg mL-')
Other components
Viruses from animal sera Lipids: Cholesterol Triglycerides Phospholipids Steroids Vitamins sugars Trace elements
Inorganic salts Glucose Vitamins Lipids: Lecithin Cholesterol 2-mercaptoethanol Amino acids
Hybridoma cells
Secreted proteins (50-200 pg mL-') Cellular proteinddebris Nucleic acids Viruses
Secreted proteins (50-200 pg m L ' ) Cellular proteinddebris Nucleic acids Viruses
dation of the purification process. As cell culture produces a relatively dilute MAb preparation, water may be regarded as a major contaminant. 8.3.1.4 Scale The scale at which the final purification is to be run is important in determining the purification system which may be used. There are very few, if any, purification methods which cannot be scaled-up. However, methods differ enormously in their ease of scale-up. For example, one would not normally consider slab gel electrophoresis cap-
228
8 Purification and Characterization of Monoclonal Antibodies
able of practical scale-up. Precipitation methods are simple to apply in the laboratory, since bucket centrifuges are a standard item of laboratory equipment, but on scale-up the only practical solution may be a flow-through centrifuge, which at large scale is a complex and very expensive item. On the plus side, it may also be used for a cell separationklarification process step. In contrast, the scale-up of chromatographic methods is relatively straightforward. The consideration of future process scale-up at the development stage should include an appraisal of GMP and engineering problems [ 131 at the final scale where appropriate.
8.3.1.5 Cost The cost consideration in choosing a MAb purification scheme has a number of components associated with it. First there are equipment costs, for example centrifuges, chromatography columns, pumps, monitors, and ultrafiltration equipment. Second, the cost of the consumables used in the purification process, which in the case of affinity chromatography media for example can be considerable. Third, the time and cost of developing a purification process. In the case of a MAb for therapeutic use, years of investment in process development and validation go into bringing a product to market, when the investment can be recouped. Anything which can shorten this lead time is worth considering, sometimes even at the expense of process equipment costs. The researcher who may be simultaneously developing several MAbs cannot afford to spend a lot of time developing purification methods for each one. These considerations tend to favor generic methods for MAb purification, methods which can be translated from one product to the other with a reasonable degree of confidence not only that they will work, but that they will need little optimization, and one can easily build on the experience of others. Examples would include ammonium sulfate precipitation and protein A affinity chromatography. Other chromatography techniques are likely to need significant modification for each MAb due to factors such as the isoelectric point, surface hydrophobicity, and pH stability of the individual MAb.
8.3.2 The Purification Scheme The purification scheme can be divided into three stages. First is removal of cell debris, which may also include removal of excess water. The second stage is the major purification from gross contaminants in the feedstock. Finally, the third stage may be a polishing step if this is necessary for the intended use of the product. Although a simple purification scheme is highly desirable, most therapeutic applications of MAbs require several steps to produce the necessary degree of freedom from contaminants to meet regulatory requirements. The type of purification step which may be used at each stage is summarized in Table 8-2. The early stages in a purification strategy will aim to achieve both concentration and purification. For some methods of MAb production, the first step will be clari-
8.3 Purification
229
Table 8-2. Purification techniques which may be used at different stages of a process scheme. Stage
Clarification
Primary purificationkapture
Further purification
Polishing
Purification technique
Precipitation
Affinity chromatography
Ion-exchange chromatography
Gel filtration chromatography
Ion-exchange chromatography
Hydrophobic interaction chromatography
Diafiltration
Centrifugation Filtration Aqueous twophase partition
Hydrox y lapatite chromatography Thiophilic chromatography
Ultrafiltration
Gel filtration chromatography
Immobilized metal ion chromatography Precipitation Hydrophobic interaction chromatography
fication by removal of cells and cell debris, although this may not be necessary, for example, if the culture supernatant has been produced in a hollow fiber cell culture system. The next stage will be the major purification step and will often also be a concentration step, unless a previous clarification step has achieved the required degree of concentration. The number of steps included in the third stage of further purification and final polishing will depend on the intended use of the MAb. These steps will be designed to reduce contaminants such as DNA, host cell protein, endotoxins, process chemicals, and others such as protein A which may have been introduced early in the purification scheme. Usually, different stages will incorporate different techniques to maximize discrimination between the Mab and the contaminants. For example, if protein A affinity chromatography is used at stage two, an ion-exchange column can be used subsequently to separate the MAb from leached protein A by exploiting differences of PI between the proteins [14]. Polishing steps might include a buffer exchange by gel filtration or diafiltration. Other steps may be introduced at any stage specifically to prevent virus contamination of the product. Any purification scheme for a therapeutic product will have to include more than one such step and again these must be different in virus reduction mode to maximize their effectiveness. Specific steps often used include solvent/ detergent inactivation, which is effective for lipid-enveloped viruses, and virus filtration which is most effective for medium-sized or large viruses [15]. Viral clearance may also be demonstrable during capture chromatographic steps and some elution conditions, such as low pH, may give useful levels of inactivation of some viruses [161.
230
8 Purification and Characterization of Monoclonal Antibodies
Most MAbs referred to in the literature are sourced either from human or rodent cell lines, or heterohybridomas of these, although other types of expression system are increasingly being used, particularly for immunoglobulin fragments. The majority of the cell lines secrete IgG antibodies, although a substantial minority produce IgM, and occasionally references to other classes are found. The purification methods discussed below are generally applicable to IgG molecules; where they are equally or more appropriate for other classes of antibody, this is specifically mentioned.
8.3.2.1 ClarificatiodConcentration Downstream purification of MAbs produced in mammalian cell lines is simpler than the purification of many other products of cell culture processes as the required protein is secreted from the cell in a soluble form. Thus, cellular disruption is not required. Centrifugation or filtration are the two most common methods of removal of cells and naturally occurring cell debris from tissue culture supernatant [ 17-19]. With continuous-harvest culture systems, such as hollow fiber cell culture, filtration can be performed in-line during supernatant collection. An aqueous two-phase extraction procedure has also been used [20]. Most of the commonly used cell culture systems for MAbs result in cell suspensions in which the MAb is in solution at very low concentrations. At industrial scale this can amount to thousands of liters; hence it is desirable on economic and practical grounds to reduce the process volume at the earliest possible stage. It may in fact be essential for the performance of a precipitation step. Concentration can be done by ultrafiltration, which at large scale can also be a very expensive piece of equipment, but if one uses a tangential-flow ultrafiltration system, the same equipment (if not the same membrane) can also be used for cell separation. The reduction of process volume can be achieved very efficiently using a chromatographic capture process. Ion-exchange capture is an obvious choice, though many MAbs will not bind to either cation or anion exchangers under physiological conditions of pH and ionic strength, so the cell culture supernatant has to be conditioned by a combination of dilution and pH adjustment, or by diafiltration. Conditioning is generally not required for protein A or protein G affinity capture, though the disadvantage here is the much greater cost of the media, which may have a very limited lifetime under these process conditions. Conventional column chromatographic capture is not the preferred method for handling solutions containing particulates, which will foul the column bed and greatly reduce the gel lifetime. Methods have now been devised which allow cell-containing feedstocks to be loaded directly onto chromatography columns, thus combining clarification and purification in one step. These include expanded bed adsorption [21] and fluidized bed adsorption [22], and such systems could be applied to the production of MAbs.
8.3 Purification
231
8.3.2.2 Chromatography
Chromatography is a widely used technique for the purification of MAbs [11,23]. Affinity chromatography using protein A and protein G media is widely used for purification of MAbs, since it can easily achieve clearance factors of > 1000. However, in the case of MAbs for therapeutic use, one must be sure that the affinity ligand which leaches off the matrix in small amounts does not end up in the product. This is usually achieved by adding a chromatography stage after the affinity step, typically ion exchange. A conventional biochemical high-purity standard is sufficient for most of the other applications for MAbs, which again leads one towards chromatographic methods. For many research reagent applications a clarified cell culture supernatant, or at most the product of a simple precipitation purification step, is all that is required. However, the cruder a preparation is, the more likely it is to be unstable due to protease activity, either from the parent cells or from bacterial contamination. The major advantages of chromatography are that it can be readily scaled-up and is one of the easier techniques to operate to GMP. Affinity Chromatography This is a very powerful purification technique which can give high single-step purifications [24]. The most common types of affinity ligands in use for MAb production are immobilized bacterial cell wall proteins, e.g., staphylococcal protein A [25] or streptococcal protein G [26]. These proteins are available commercially both as the free proteins and immobilized on a variety of chromatographic supports. One of the limitations of this technique may be the affinity of the ligand for immunoglobulins of different species or of different class within one species. For example, mouse IgGl MAbs have low affinity for protein A and, within human IgG subclasses, IgGl binds to protein A but IgG3 does not. This problem may be overcome in some cases by enhancing binding with increased ionic strength of the loading buffer and the MAbs can then be eluted at near-neutral pH. These bacterial proteins interact primarily through the Fc region of the IgG molecule, although both protein A and protein G have also been shown to interact with regions in the Fab part of the molecule and thus they are also used for purification of IgG fragments [27,29]. Both protein A and protein G are available in recombinant forms which possess advantages over the native protein, for example the deletion of the albumin binding site. These ligands give a good purification, especially if the feed stock contains no other IgG than the MAb of interest. However, there is always a small degree of leakage of the protein ligand and, for therapeutic products, further purification will be required. The other often-cited disadvantage of an immobilized protein ligand is its inability to withstand the kind of sanitization regimes frequently used in GMP applications. Nevertheless, these ligands can be used repeatedly with applications of a chaotropic agent such as 6 M guanidinium chloride with no loss of capacity [16]. Although proteins A and G are the most frequently used ligands for affinity chromatography of MAbs, the disadvantages mentioned above have led to the development of alternatives; for example the specific antigen for the desired antibody [30-321, a small molecular weight, non-protein affinity ligand marketed as Avid
232
8 Purification and Characterization of Monoclonal Antibodies
AL [33], which also interacts with the Fab and Fc regions of the IgG, and Protein L, another bacterial cell wall protein, which has been used for preparation of kappa light chain [34]. Each of these has their own set of advantages and disadvantages. Using the specific antigen neatly overcomes the problem of purifying a MAb from a mixture containing other IgGs. On the other hand, a new chromatography gel has to be created for each MAb to be purified, which will be costly, and the conditions used to elute the MAb from the gel may have to be as harsh as the elution conditions for proteins A or G. Also, the antigen may be toxic at very low concentrations requiring a high level of clearance of any leached ligand. Elution of the MAb from immobilized protein A or protein G is usually by a change in pH. Human MAbs require a low elution pH, although some murine MAbs can be eluted from protein A at near-neutral pH. A pH as near to neutral as possible should be used to minimize protein denaturation and, in any event, the pH of the recovered solution should be neutralized as quickly as possible. In this respect, Avid AL has an advantage as elution at neutral pH is possible, although it may have other disadvantages in binding charged substances from tissue culture medium such as phenol red [35]. A method for elution of MAb from protein G at alkaline pH has been reported [36]. Ion-exchange Chromatography This is another powerful, well-established technique which separates molecules on the basis of charge [37]. Proteins can acquire either a positive or a negative charge by manipulation of pH either side of their PI and will then bind to immobilized groups with the opposite charge. Desorption from an ion exchanger is usually by increasing ionic strength but may also be by changes in pH. Despite the structural commonality of all MAbs, they show a wide range of PIS. Thus, both cation- and anion-exchange chromatography can be used for the purification of individual Mabs [38]. When anion-exchange chromatography is used for IgG [12,39] or IgM [9] purification, a mixture of proteins are bound to the column and are selectively desorbed by gradient elution. This can be an inefficient use of the chromatography resin. In addition, phenol red, which is often present in cell culture supernatants, will bind to the anion exchange gel and will further decrease the capacity and could interfere with elution. Therefore it may have to be removed from the feedstock by pre-treatment. Cation-exchange chromatography for IgG [40,41] or IgM [42] can be manipulated so that only the MAb will bind, decreasing the unit volume of gel required. In this case elution will need less critical control, making scale-up easier. Ion exchange may also be used in a non-adsorptive mode in which the contaminants are bound to the gel while the MAb elutes in the flow-through [43]. A mixed mode ion-exchange resin [44] has been used for murine and rat IgG MAbs and good purity and recovery have been reported. This system has also been used for IgM [45] and for bispecific Mabs [46]. Although ion exchangers have advantages in GMP applications where they can be sanitized under extreme acid or alkaline conditions, when used as a primary capture step, this purification method may need separate optimization for each MAb and the cell culture supernatant may need manipulation (for example in ionic strength) before it can be applied to the gel. As molecules are separated on the basis of charge
8.3 Purification
233
and charge distribution, it may be difficult to completely eliminate contaminants with similar PI to the MAb. Other chromatography systems These systems, which have been used for purification of MAbs, include thiophilic chromatography [47-491, hydroxylapatite [50,5 11, immobilized histidine ligand [52], and immobilised metal affinity chromatography [3 1,531. Hydrophobic interaction chromatography is also used as a primary capture step [54] as well as a later step in purification schemes [12,55,56]. Mode On the larger scale, a fast chromatographic throughput is required due to the large volumes to be processed. Chromatography supports such as zirconium oxide [67] and PorosTM[58] have been developed to allow operation under increasingly high flow-rate without significantly increasing back-pressure. Factors to consider in setting an upper limit for flow-rate in any chromatography system include the rate of diffusion into the bead pores for a gel (often rate limiting), and the rate of the binding reaction for any capture process (usually relatively fast). With this in mind, chemistries on membrane surfaces have also been developed and applied [8,59,60].
8.3.2.3 Other Purification Systems Although chromatography is widely used as a purification tool, precipitation systems particularly ammonium sulfate [ 12,611 and polyethylene glycol (PEG) [62] precipitation or a combination of the two [63] are also used, often in the first stage of processing, but sometimes also as a major purification step. Precipitation is particularly applicable to IgM preparation and to purification of more concentrated ascites fluid as it is easier to perform on a small scale. Preparative isoelectric focusing is also used for some applications, but is not very amenable to scale-up [64].
8.3.3 Large Scale There is no easy definition of large scale in the production of MAbs. The defining issues may be the use of the antibody and the effective dose. 8.3.3.1 Scale-up
Factors which have to be taken into consideration when planning scale up include: - reproducible performance - operation of process equipment to required hygiene and safety levels - effect on processing time and buffer volumes
234
8 Purification and Characterization of Monoclonal Antibodies
Any purification scheme other than those for research purposes will be developed on a small scale but finally operated on a larger production scale. There must be reproducibility of process performance over this scale-up for three major reasons. First, it is important that investigations with purified product carried out early in the development are relevant to the final product. Second, it is usually desirable to carry out the majority of the validation work on a smaller scale than the final process scale. Table 8-3. Retention times of product peaks during purification of MAbs from two cell lines. (i) Cell line 1; small scale Peak retention time (min)
Mean SD n (ii) Cell line 1; 5
X
Cla
C2 Product
c3
610.2 2.0 14
85.1 2.3 14
45.7 2.5 12
scale-up; linear dimensions preserved Peak retention time (min)
Mean SD n
Column Ib
Column 2
Column 3
617.1 1.4 4
86.3 0.9 4
54.1 0.7 4
(iii) Cell line 2; small scale Peak retention time (min) Column 1"
Column 2
Column 3
612.9 2.5 8
154.5 5.3 8
52.0 2.8 8
~~
Mean SD n
(iv) Cell line 2; 5 x scale-up; linear dimensions preserved Peak retention time (min)
Mean SD n a
Column Ib
Column 2
Column 3
614.3 4.0 4
155.2 2.7 4
61.8 3.0 4.0
Retention time corrected for load time. Retention time corrected for load time and changes in pump wash.
8.3 Purification
235
Third, the financial viability of the project will need to be predicted from yields at the earlier, small scale and if the yield changes over scale-up this may adversely affect the project. Reproducibility of performance can be measured in several ways, including by yield or levels of contaminants in the product. With a chromatographic process, one measure of performance is the retention times of the peaks eluted from the columns. In a purification process that we have developed which consists of three chromatographic steps, these retention times have been very consistent over five-fold scale-up (Table 8-3), and it is this level of consistency which is a major argument for the use of chromatographic processes for large-scale MAb production. Discussion of the parameters which have to be taken into consideration when planning a largescale chromatography system can be found elsewhere [65-671. Factors which might affect the hygiene and safety of the product and which might change over scale-up include the design of liquid handling equipment, such as valves and pumps, and also the nature of the materials in contact with the process flow. Changes in processing time with scale-up may cause changes in the nature of the product if the length of exposure to denaturing conditions is changed. They may also affect the costing of the product through antibody losses and increased overhead expenses.
8.3.3.2 GMP and Validation GMP and validation issues would fill a chapter on their own. Suffice it to say that it is essential to follow the relevant regulatory guidelines such as FDA Points to Consider in the Manufacture and Testing of Monoclonal Antibody Products f o r Human Use [68] or CPMP Notes f o r Guidance: Production and Quality Control of Monoclonal Antibodies [69] and to resolve the points raised for each particular MAb. These will certainly include viral safety, residual levels of DNA, host-cell protein, process chemicals, and the presence of any other adventitious agents in the product. Sufficient control of the process to demonstrate reproducibility, including an evaluation of the effect of expected run-to-run changes in process parameters, will also be required. 8.3.3.3 Control and Automation Increasingly, the issue of control of the production process in all protein purification schemes - including those for MAbs - is being addressed by replacing human operators with an automated system 170-721. Provided that the systems can be shown to be reliable, a large amount of data can be collected for each purification to provide assurance that the process proceeded as expected and was within operating limits. Unattended operation is sometimes also possible. Chromatographic processes lend themselves particularly well to this level of automation and this may be one reason for their increasing adoption for production-scale purification of protein therapeutics.
236
8 Purification and Characterization of Monoclonal Antibodies
8.3.4 Future Developments It seems likely that future generations of MAbs will be produced via a route that makes increasing use of genetic manipulations. Already MAbs have been produced which have had affinity tails grafted onto the protein sequence to facilitate purification [73,74]. MAbs can now be produced by raising antibodies in an experimental animal and then grafting the gene sequence for the immunoglobulin hypervariable region onto human IgG frameworks. Human antibodies can also be generated from a non-immunized phage display library [75], which is a useful technique for antibodies to toxic antigens. Such MAbs or MAb fragments may be produced in a variety of cell lines, both mammalian and non-mammalian [31,761, although where glycosylation is important in determining functionality a mammalian cell line may be preferable. For some applications, production of immunoglobulin fragments may be sufficient or even desirable, and in these cases use of a genetic construct is the preferred route [28]. For example, it has been suggested that the use of F(ab)z fragments in an ELISA would reduce non-specific interactions [77]. Fab and F(ab’)2 fragments could be used to target radioisotopes to tumors; their small size in relation to intact immunoglobulin should allow greater tumor penetration and they are also cleared rapidly [29]. Bispecific MAbs could also have applications in therapeutics, for example forming a bridge between a tumor cell and a cytotoxic cell and thus inducing the destruction of the tumor cell [31]. Chromatography or similar adsorption processes will continue to be widely applied to the purification of such MAbs. Expanded bed adsorption chromatography is being used for MAb purification [78] and other new systems, such as displacement chromatography [79] and novel small molecular weight ligands [SO], may be developed in the future for the processing of MAbs. The increasing capability for full characterization of the MAb product will have important consequences for purification, and developments in this area are discussed in the next section.
8.4 Characterization 8.4.1 The Need for Characterization The level of characterization required for a monoclonal antibody will inevitably reflect the intended use. Whereas for a research application the basic information relating to the class, antigen specificity, and any cross-reactivity may be all that is needed, for commercial applications - and in particular where a therapeutic use is envisaged - a substantial amount of detailed molecular and functional data will be needed. We will concentrate in this discussion on the latter category. Biotherapeutics derived from natural sources are intrinsically heterogeneous (as for example with the immunoglobulin products used in passive immunotherapy
8.4 Characterization
237
which are derived from thousands of individual human plasma donations and where an absolute characterization of the product pool would be impossible ). Even where a single active agent was know, as for example with clotting factors or hormones, there was always a range of microheterogeneity due to the polymorphic expression of the product in a start pool derived from multiple individuals. Such products required, in order to ensure efficacy and safety, a lot-by-lot testing of batches within acceptable in vivo models, which was costly and which has become less and less ethically acceptable. In contrast, chemical drugs have been characterized predominantly upon their physico-chemical properties with modern and rigorous molecular methodologies being applied to the analysis of purity, homogeneity, and potential contaminants. With the advent of biotechnologically derived agents and advances in highresolution analytical tools it has become possible to define more accurately the efficacious agent in biological parenterals. As a result, in November 1995 the FDA issued revised guidelines in terms of what it described as ‘well-characterized’ or ‘well-specified products’. These products are those for which a range of robust validated physico-chemical and functional assays could be applied and where the structure and activity of the agent was well understood. Monoclonal antibodies were one such group of biotechnological products specified by the FDA and in the present discussion of the characterization of monoclonal antibodies this strategy will be referred to. In February 1997, a further revision of the Points To Consider for the Manufacture and Testing of Monoclonal Antibodies for Human Use was issued 1681.
8.4.2 Aspects of MAb Characterization 8.4.2.1 Primary Sequence Knowledge of the cDNA sequence and the corresponding post-translational amino acid sequence are crucial to the characterization of the MAb. The sequences of rodent and human myeloma-derived antibodies had been much studied and were available for comparison when assessing the sequences of new agents. Data on human MAbs were more limited and the means for their expression less amenable in general to large-scale production, though consensus human immunoglobulin sequences were also available for most of the subclasses and Gm allotypes [81]. Therefore, it was comparatively straightforward to assess whether or not the MAb under study conformed to the normal consensus sequence of its species immunoglobulin. However, with the advent firstly of chimeric and then humanized antibodies which have been specifically altered either at particularly immunogenic or functionally important sequences, the checking of the sequences at these crucial regions of the molecule became an important control in the assuring the successful exploitation of these MAbs. Indeed, the changes in three-dimensional structure and the effects on immunogenicity and efficacy were not always as predictable as had been at first
238
8 Purification and Characterization
of
Monoclonal Antibodies
expected and much work was required in comparing these humanized antibodies to their original non-human immunoglobulin counterparts [821. With the advent more recently of engineered antibodies and antibody fragments, the genes that code for the antigen-binding domains may have been expressed through several different host systems during their manipulation before expression of the intact MAb or engineered fragment has been arrived at in its productionscale host-cell line. For instance a pair of heavy and light chain variable region genes may have been initially selected from a bacterial host system using filamentous phage, then expressed at a larger scale after selection to ensure in vitro antigen recognition. Following this they will have been combined with constant region genes from the desired immunoglobulin class and subclass and eventually presented in a mammalian cell line for large-scale production and detailed study of in vivo efficacy. At each stage silent errors in transcription or translation may have occurred, only to show up as potential problems once the intact molecule is evaluated in its intended target or model system. It is thus essential to have the product of the Master Cell Bank characterized in detail in terms of both the cDNA sequence and the translated protein product. Any changes in the sequence of the MAb should be investigated during the development of a clinically relevant antibody, and tests should be performed to confirm that the products of each large-scale culture conforms to that of the Master Cell Bank on which the bulk of the characterization has been performed. The techniques to perform these analyses are well documented and known in the laboratory but require automation realistically to tackle the substantial task to the necessary level of assurance and within a short timescale. Where an immunoconjugate is the intended final product, whether it is a radionuclide complexed within a chelating ligand or a fusion protein with a larger entity such as a molecular toxin or hybrid effector molecule, the task of post-translational sequence analysis becomes even more important in order to check that the complexation has occurred, measure the proportion of unmodified antibody molecules or free ligand that remain, and identify any undesirable side reactions that may have occurred. These are particularly important checks if the complexed molecule is potentially toxic. Radio-labeled monoclonals and antibody fragments frequently use a small number of suitable radio-isotopes, indium-111, iodine-125 and -131, technetium- 99, and yttrium-90. These molecules are usually attached to antibodies by cross-linkers such as 2-iminothiolane [83] or chelators such as diethyl triamino penta-acetic acid.
8.4.2.2 Sequencing Strategies cDNA DNA sequencing technologies (such as the Sanger method) have been well defined and indeed automated systems for DNA sequencing have been available for several years. The assignment of primary protein sequence deduced from the cDNA sequence is now more easily performed than direct protein-sequencing methodologies. However, the gene sequence alone tells one nothing of the post-translational
8.4 Characterization
239
modifications that occur in proteins after transcription of the mRNA. Although some potential glycosylation sites are detectable from the cDNA sequence, there is no guarantee that these sites will actually be occupied in the mature protein. Hence it is necessary while initially identifying the primary sequence from DNA methodologies to supplement this with selective peptide sequencing methods in order to obtain the full picture of the protein structure. Amino Acid For antibodies, the size of the light and heavy chains precludes direct protein sequence analysis, such has been possible for the smaller protein biotherapeutics. Instead, sequencing must be preceded by a controlled hydrolytic process and the resultant peptides separated and collected prior to analysis. Due to the distribution of suitable target amino acid residues digestion using proteolytic enzymes, primarily trypsin, has been the strategy most commonly adopted [84]. Use of specific chemical digestions (such as with cyanogen bromide) have also been reported [85]. Limited digestion might reveal altered sequences as anomalously migrating peptides in electrophoretic profiles. N-terminal sequences blocked by cyclization of the glutamine residues can be treated with pyroglutamate aminopeptidase to remove the pyroglutamy1 residue, or the blocked peptide can be analyzed by mass spectrometry (MS) after cleavage from the protein and the sequence deduced from the mass [86], again taking into account the possible multiplicity of peptides that could arise from partial blockage, affecting only a subpopulation of the antibody molecules present. Initial studies have followed the work on the sequencing of normal immunoglobulin in requiring the separation of the heavy and light chains and their independent digestion and sequencing. More recently, probably due to the excellent separation capabilities of modern reverse-phase HPLC, the digestion of whole molecules of antibody have been demonstrated [87]. This can yield 90 or more peptides which must then be separated and sequenced. 8.4.2.3 Peptide Mapping
Whereas the complete primary structure confirmation should be made once for a ‘reference’ lot of MAb, there is a role for limited sequence analysis as a batch-tobatch consistency test and possibly as a stability-indicating method. Proteolysis can be achieved by enzymatic means usually using TPCK-treated HPLC-purified trypsin or other sequencing grade enzymes (Lys-C, Asp N, V8 protease). Alternatively, partial chemical degradation may be useful, although only CNBr cleavage at methionine or BNPS skatole cleavage at tryptophan are sufficiently specific to be of value. Rather than obtaining exhaustive sequences, these peptides can be assigned and changes resulting from chemical modification inferred often merely by amino acid analysis or most recently by MS. In a recent review of peptide mapping [88], the advantages of computer-assisted modeling and the use of photodiode array detection to provide full UV spectral information are discussed (Fig. 8-1) as well as the application of capillary electrophoresis for peptide separation and the
240
8 Purification and Characterization of Monoclonal Antibodies
8 0 8
E C
'1
?
Q v
8s $? C U P
a,
o >
8"
-N 3 0
9
0 N N 0
9
8 N ameqlosqij
.-u
h
0
8 Ln
0
8 d
h
0
.G
z.5
E
OF
8 N
0
9
9 ,
-
r='o
- t
__cJ
--__z
0
0
7
9
z
0
7
(UU
~
0
z
0
x
0
2
0
8
PLZ) a3ueqJosqv
Fig. 8-1. Value of spectral analysis in peptide mapping. The reverse-phase peptide map (i) of the tryptic digest of a human monoclonal antibody light chain is shown. The spectra of two peptide components are shown (ii and iii) and compared with their known sequences. The contribution at 280nm due to tryptophan is detected in (ii). (Sequences confirmed courtesy of Dr Will Mawby, Bristol University, UK.)
8.4 Characterization
24 1
identification of the peptides by MS. A paper has appeared from a consensus of pharmaceutical firms in the field [89] on the validation of peptide mapping methods. Among the important issues they highlighted were the use of spiking with synthetic peptides to establish quantitative recovery of peptides applied in the profile generated, and the use of a digest of reference antibody to be compared with the profile of the test sample. Additionally, more conventional validation issues such as establishing reagent quality and column consistency were discussed. A fully validated peptide map has been deemed to be an acceptable tool for the assessment of the genetic stability of biotechnological medicines. Precipitation of insoluble peptides out of solution may result in irreproducible peptide maps and the presence of denaturants such as guanidinium chloride or urea may help. By running peptide maps of digests pre- and post-treatment with disulfide-reducing agents the presence of the disulfide bonds can be inferred. Automated peptide mapping has been reported recently [90] with the automation helping to reduce inter-digest variability. 8.4.2.4 Mass Spectrometry of MAbs Mass spectrometry of the whole MAb molecule has been documented using electospray ionization [9 11 and also by matrix-assisted laser desorption ionization-time of flight ( MALDI-TOF) [92]. Mass spectrometry has been widely used to infer the nature of heterogeneity in peptides resulting from N-terminal blockages, amino acid oxidation, or degradation. Another area where MS has helped in recent years is in determination of glycosylation heterogeneity. Immunoglobulins contain a common Fc glycosylation in the C"2 domain which is a complex type N-glycan, usually asialylated. However, differences in the processing following the trimannose core has been documented on a number of ocassions and this topic will be discussed further later on. Mass spectrometry is unable to distinguish between isomeric forms of sugars, but can reveal heterogeneity due to the presence or absence of the respective saccharide subunits [93].
8.4.2.5 C-Terminal Sequencing The C-terminal sequence (of at least three residues) can be determined directly either by manual or automated C-terminal sequencing [94]. Several different chemistries are marketed for direct sequencing and each claim a specific capacity for determining a wide range of the amino acids that may potentially be present. Analysis has been either by reverse-phase HPLC of the cleaved amino acid or by MALDI-TOF mass spectroscopy. Alternatively, enzymatic release of the C-terminal amino acids using carboxypeptidase can be used [95]. Cases of the post-translational clipping of C-terminal lysine residues on antibody heavy chains have been documented for both plasma-derived immunoglobulin and monoclonals [96,97]. In these cases, components corresponding to the full-length and truncated C-terminal peptides are unlikely to co-migrate and the different forms have been separated by ion-exchange chromatography. Mass spectrometry
242
8 Purification and Characterization of Monoclonal Antibodies
may again be useful in assigning such components given the known C-terminal sequence, the theoretical peptide masses being calculated and compared with experimental results.
8.4.2.6 Secondary and Higher Structure Details of the secondary and higher structure can be obtained by using techniques such as circular dichroism [98], intrinsic fluorescence [99], and differential scanning calorimetry [loo], although immunoglobulins are still too large to be accessible to meaningful analysis by nuclear magnetic resonance. Such studies are useful not only to ensure a native molecule but also to detect changes induced on storage. Precise and consistent readings are required that will allow accurate comparison of runs performed under different conditions, necessitating sophisticated software. Rather than seelung absolute identities for the individual profile components the changes observed between samples under different conditions are indications of shifts in the degree of definite structural motifs, e.g., alpha-helix or beta-sheet. With molecules such as MAbs, where the generic structures are known it may be unnecessary to go too far down the road of absolute assignments of secondary and higher structure. More limited data along with accurate and specific functional and bioassays can be used to infer the integrity of the overall higher order structure.
8.4.2.7 Functional Activity Another important way in which MAbs can be characterized is in their specificity and cross-reactivity. Indeed, specificity may be the primary criterion for a research application. The specificity of MAbs is assessed by immunoassays, of various types including for example serological, ELISA and RIA. Methodologies for these techniques are well known [loll. Increasingly, popular are methods based upon surface plasmon resonance. This technique allows quantitative data on binding specificity and kinetics to be obtained rapidly [102]. A bioassay is needed which is specific for the natural antigen and which, if possible, mimics the in vivo mode of action [103]. Data from such functional assays combined with that from the physical characterization methods together provide information on the nature of the MAb, which is invaluable in demonstrating comparability during process changes, development and scale-up. Not only should MAbs be shown to be reactive with a specific antigen of interest but it is also necessary to establish the cross-reactivity, if any, of the MAb. Crossreactions need not be restricted to molecules with similar structures to the antigen of interest, but indeed may occur with composite antigens forming a similar binding site from the association of functional groups on unrelated molecules. It is required even at an early stage in clinical trials that the potential for cross-reactivity with a number of tissues be addressed. It is important when using histochemical studies to ensure that the reactivities observed are likely to be relevant in the in vivo state and are not merely artefacts of the cytological preparative methods.
8.4 Characterization
243
8.4.2.8 Stability There is a need for limited but less time-consuming methods for the detection of any changes in the antibody throughout cell culture, harvest, processing, and formulation and indeed, across shelf life. Such techniques can be a mixture of old and new; with traditional methods such as IEF and SDS gel electrophoresis being supplemented by their capillary electrophoretic counterparts [ 104,1051, and chromatographic methods similarly indicating structural integrity. The IEF profile or fingerprint of each MAb is characteristic and modifications to glycosylation, oxidation and deamidation of amino acid residues can all give rise to differences in the IEF profile [106,107], although some heterogeneity may be present from the beginning of culture (for instance, differences in the C-terminus due to clipping). However, not all changes in the structure of the MAb are necessarily detected by one method, and rather the approach taken is to adopt a battery of analytical techniques which are robust, quantitative and accurate in order to arrive at an overall assessment of the integrity of the antibody [108]. SDS-PAGE is a good method for showing differences in the composition of light and heavy chains as a result of differences in glycosylation, peptide integrity, and antibody class and subclass. The advantages of analyzing the antibody under both non-reducing and reducing conditions indicates changes that are both non-covalent or covalent in nature. For instance, published work on OKT3 revealed a covalent modification resulting in cross-linkage of the heavy and light chains during storage under an oxidizing atmosphere [ 1091. Most regulatory documents still require analysis of the MAb by traditional methods using high-sensitivity silver staining and Coomassie dyes; however, the newer methods have several advantages. The capillary form of these techniques allows more ready quantitation of components by monitoring of peptide backbone absorbancies which are less sensitive to differences in amino acid composition and alleviate differences based solely upon variation in the interaction of protein with dyes without resorting to time-consuming calibration of densitometric traces (Fig. 8-2). Most CE machines allow the automated processing under PC control of a number of samples and this again saves resources and simplifies data storage and retrieval. Free-flow electrophoresis and micellar electrophoresis utilizing the tendency for hydrophobic interaction with the detergent micelles have also been applied to the quality control of MAbs [lo51 being shown to be precise, quantitative, and reproducible. Well-proven techniques such as HPLC size exclusion chromatography (SEC HPLC), used in the standardization of plasma-derived immunoglobulins, are also of value for the analysis of MAbs. This technique, when run at near-neutral pH, gives a good indication of the level of aggregation and any fragmentation which may occur in the product over time. Several monoclonals have been documented to show anomalous behavior on some SEC HPLC matrices [110] and so a typical profile for a given MAb should be compared rather than absolute retention times for individual components. This technique is a valuable tool in stability studies and there is evidence from plasma immunoglobulins that relationships between degradation at elevated temperatures and the given storage temperatures are useful predictors of antibody instability [ l l l ] , although the regulations are clear that shelf life must be based upon real-time storage at the designated storage temperature.
244
8 Purification and Characterization of Monoclonal Antibodies
0.024
1-
1
intact immunoglobulin +
+ tracking dye
0.020-
0.024
’
- 0.020
h
E
scu
0.016-
-
0.016
-
0.012
-
0.008
v
a
4.1
2 0.012-
J
4
m
e0
(ii)
: :0.008a 0.0041
A
I/
II
1 0.004
0.000
0.000
I -0.004
-0.004 15
20
Time (min)
25
30
Fig. 8-2. Capillary electrophoresis to monitor antibody stability. Comparison of the SDS capillary CE absorbance profiles of frozen control antibody (i) and antibody stressed by storage at ambient temperature for 20 weeks (ii). Additional peaks in the stressed sample are indicated by arrows. Samples run on an eCAP SDS14-200 kD application kit using a Beckman PACE 5510 CE system fitted with a 47 cm neutral gel-filled capillary.
The use of elevated temperatures or other extreme conditions to provide degraded antibody material can be useful both to establish the detection limits and suitability of the techniques used as stability indicating methods, and may also assist in the prediction of the longer-term degradative changes which can occur in proteins [112]. The expression of recombinant human MAbs in SCID mice has been suggested as a means of testing the in vivo stability and half life of molecules intended for therapy [113]. Other powerful chromatographic techniques which have been demonstrated as useful in the characterization of MAbs include cation exchange chromatography [96], hydrophobic interaction chromatography [ 1141, mixed bed ion-exchange chromatography [44], and chromatofocusing [ 1151 or preparative IEF. These techniques, where applied successfully to differentiate between intact and degraded immunoglobulin molecules, have the advantage that they can yield mini-preparative scale quantities of the isoforms which greatly assists in their characterization. They also are - in principle - quantitative, rapid, and capable of automation. Some evidence for proteolytic activity in harvests from hybridoma cell culture has been reported [116,117]. Indeed, one study has indicated a degradation of intact MAb over the course of cell culture over several weeks as revealed by non-reducing SDS-PAGE [ 1181. Hence, it is important to monitor harvests, intermediate bulk solutions and purified final product for the presence of proteases by the most sensitive
8.4 Characterization
245
methods available. However, electrophoretic heterogeneity may also be due to incomplete antibodies resulting from disulfide exchange between chains [ 1191. Such instability has been reported previously and these profiles may be characteristic for each antibody (and in our experience do not show signs of increasing degradation, as would be indicative if proteases were active), though some authors have suggested that these forms may increase in intensity with protein modification due to deamidation/oxidation [107]. One report of the presence of half antibody in IgG [120] was addressed by engineering the susceptibility to self dissociate out of the MAbs by site-directed modification. Similarly, a tendency for IgG degradation via a copper ion-mediated cleavage was reported for both monoclonal and polyclonal IgG and this again could be combatted by engineering the site of modification [121]. Formulation may play an important role in preventing the degradation of MAbs. In their work on OKT3, Rao and colleagues [lo91 demonstrated the value of an inert Table 8-4. Some literature reports of degradative changes in monoclonal antibodies. MAb/Clone name
Degradation observed
HER-2 Humanized MAb
LC deamidation at Am30 C-terminal clipping of HC Lys450 Tyr to Gln at HC" cyclic imide formation at HC Asp102
TB/C3 Mouse MAb
Unspecified proteolysis
Unnamed
Deamidation monitored by IEF
IgG4 CB72.3
80 kDa HL component due to Ser241
17-1A
C-terminal clipping in HC
E-25 humanized anti-IgE
Isomerization at Asp32 in LC at room temperature
OKT3 mouse MAb
Deamidation Asn386, Asn423 in HC Oxidation at Met34 in HC, C-terminal clipping of HC Deamidation at Am156 Some oxidation at Met174 in LC Covalent cross-link of LC Tyr46 and HC CyslO5
OKT4a humanized
Cleavage at Asp270 HC, also some cleavage at Ser220, Thr250, Thr335, and Thr350. Trace of cleavage at Ser203 in LC
RSHZ 19 humanized anti RSV
Oxidation of a Met residue C-terminal clipping of HC N terminal pyroglutamate in HC
Campath 1H humanized
Copper-induced cleavage at Lys226-Thr227 in HC at elevated temperature
a
Reference
Rather than a degradative change this report describes the predominance of a variant sequence antibody over culture time due to differential expression of two clones.
246
8 Purification and Characterization of Monoclonal Antibodies
atmosphere in preventing covalent cross-linking between chains, whereas in the same paper they report that an inert atmosphere had little effect on preventing deamidation changes. The formulation of biotechnological therapeutics has been reviewed [122] in terms of the likely modifications which may occur including isoaspartate formation, deamidation of asparagine and glutamine, and oxidation of cysteine and methionine residues. These types of modification have been monitored in MAbs [109,123]. Table 8-4 outlines the types of modification which have been reported for monoclonals under development.
8.4.2.9 Glycosylation The topic of MAb glycosylation has occupied many reviews in its own right and the effects of variations in glycosylation in terms of functional efficacy have been discussed for a number of examples [127]. While it is clear that absence of glycosylation results in the loss of Fc-mediated effector functions [128] and reduced in vivo half-life [1291, the influence of differences in the glycosylation pattern applied to the tri-mannose core is less clear. For some MAbs studied by certain functional assays there appears to be a role for full galactosylation of the core glycan. For instance, Kumpel et al. [130] showed that MAb possessing a higher proportion of galactosylated (G1 and G2) glycans had higher potency in lymphocyte antibodydependent cell cytotoxicity (ADCC) than those with lower overall galactosylation. However, those antibodies with primarily monogalactosylated glycan appeared to have normal in vivo half life [131]. For another MAb there appears to be no effect on the functional assay results with galactose removal [132]. Certainly it is possible to influence glycosylation in terms of the species and type of host cell and the culture conditions [ 126,133,1341. For instance, initial studies of baculovirus-infected insect cells demonstrated non-mammalian-type terminal glycosylation processing, but later work with carefully selected strains seems to indicate that this can be overcome [135]. A number of technologies are available for analyzing glycans. Analysis without removal of the glycan from the protein can be possible by using lectin blotting [136], however this technique may not be as readily quantitative as other methods that are applied to glycans released either by enzymatic deglycosylation using PNGase F or chemical deglycosylation using hydrazine [ 1371. Once isolated, glycans can be analyzed by size-exclusion chromatography [ 1381, high-performance anionexchange chromatography [139], HPLC [140], or electrophoretic means [141]. Mass spectrometry has proven a useful tool in the characterization of glycans and glycopeptides [ 1251. Much has been published on the role of IgG glycosylation in disease and in addition to a conserved C p domain N-glycosylation site both variable region glycosylation and 0-glycosylation of MAbs has been reported for some mouse monoclonals [ 1421. Complete glycans (i.e., those possessing galactose and sometimes also sialic acid) can be produced by human lymphocytic cells lines. However, other cell lines popularly used in biotechnology may not be capable of this and so the glycosylation desired is an important consideration during the choice of host cell [135]. The gly-
8.5 Conclusions
247
cosylation found on most humanlmouse heterohybridomas has been shown to be of a rodent type [ 1431 and to lack bisecting N-acetylglucosamine. In comparative studies using hamster (CHO), mouse (NSO), and rat (YO) cells producing the humanized Campath-1H antibody differences in the level of galactosylation, fucosylation, and bisecting N-acetylglucosamine were found [ 1441 and, whereas monocyte-mediated killing was unaffected, variations were found in the potency as measured by ADCC. In addition, cell lines from some species may introduce immunogenic glycosylation such as the Gal 1-3 Gal disaccharide [145]. For one IgM monoclonal a change from ascitic fluid to stirred cell culturing conditions had a dramatic effect on the in vivo half-life [146]. For an antibody destined for human replacement or prophylactic therapy where a normal in vivo half-life is required, then it is advisable to use a host cell capable of producing glycans as similar as possible to those of the native B cell. However, where antibody fragments are used (as is often the case in immunoscintigraphy or cancer therapy) then a short halflife is not a problem and the glycosylation capability of the host cell line is less vital.
8.5 Conclusions We have reviewed the areas of the purification of monoclonal antibodies and their characterization, and given our perspectives on some of the matters discussed. The strategy applied for both downstream processing and analytical assessment of a MAb is dependent upon the use for which it is intended. Whereas for a research tool this may be the minimum to allow use of the immunoglobulin as a reagent in subsequent studies, for a MAb destined for a therapeutic application the workload will be considerable as the regulatory requirements are clearly defined. Purification protocols rely largely upon initial capture steps (mainly affinity methods), although for the clinical applications multi-stage chromatography is necessary. In such applications the processes must be carefully and thoroughly validated to show consistency. Modern high-performance characterization methods together with functional activity studies are also an important part of the development process for successful manufacture of such biotherapeutics [ 1471. The recent concept of a well-characterized product has provided both a challenge to the analytical methods required and an opportunity to reduce the levels of pre-clinical and clinical evaluation which is necessary in order to assure consistency when performing modifications to the process of manufacture.
248
8 Purification and Characterization of Monoclonal Antibodies
References [ l ] Kohler, G., Milstein, C., Nature, 1975, 256, 495-497. [2] Winter, G., Milstein, C., Nature, 1991, 349, 293-299. [3] Lerner, R. A., Benkovic, S. J., Schultz, P. G., Science, 1991, 252, 659-667. [4] Brugemann, M., Caskey, H. M., Teale, C., Waldmann, H., Williams, G . T., Surani, M. A., Neuberger, M. S., Proc Natl Acad Sci USA, 1989, 86, 6709-6713. [S] Teng, N.N., Lam, K.S., Proc Natl Acad Sci USA, 1983, 80, 7308-7312. [6] Wood, C. R., Boss, M. A., Kenten, J. H., Calvert, J. E., Roberts, N. A,, Emtage, J. C., Nature, 1985, 314, 446-449. [7] Inouye, K., Morimoto, K., J Biochem Biophys Methods, 1993, 26, 27-39. [8] Menozzi, F. D., Vanderpoorten, P., Dejaiffe C., Miller A. 0.A., J Immunol Methods, 1987, 99, 229-233. [9] Rothman, S. W., Gentry, M. K., Gawne, R.D., Dobeck, A. S., Ogert, R., Stone, M. J., Strickler, M. P., J Liq Chromatogr, 1989, 12, 1935-1947. [lo] Venktash, N., Murthy, G. S., J Immunol Methods, 1996, 199, 167-174. 1111 Manil, L., MottC, P., Pernas, P., Troalen, F., Bohuon, C., Bellet, D., J Immunol Methods, 1986, 90, 25-37. 1121 Guse, A. H., Milton, A.D., Schulze-Koops, H., Muller, B., Roth, E., Simmer, B., Wachter, H., Weiss, E., Emmrich, F., J Chromatogr A, 1994, 661, 13- 23. [13] Kenney, A. C., in: Monoclonal Antibodies: Production and Application: Mizrahi, A. (Ed.), New York: Alan R. Liss, Inc, 1989; pp. 143-160. [14] Hanna, L. S., Pine, P., Reuzinsky, G., Nigam, S., Omstead, D.R., BioPharm, 1991,4,33-37. 1151 Roberts, P. L., in: Separations for Biotechnology: Pyle, D. L. (Ed.), Cambridge: SCVThe Royal Society of Chemistry, 1994; Vol. 3, pp. 420-426. 1161 Baker, R. M., Brady, A.-M., Combridge, B. S., Ejim, L. J., Kingsland, S . L., Lloyd, D. A,, Roberts, P. L., in: Separations for Biotechnology: Pyle, D. L. (Ed.), Cambridge: SCYThe Royal Society of Chemistry, 1994; Vol. 3, pp. 53-59. [17] Ostlund, C., Trends Biotechnol, 1986, 6, 288-293. 1181 Birch, J. R., Thompson, P. W., Boraston, R., Biochem Soc Trans, 1985, 13, 10-12. 1191 Strobel, G.-J., in: BIOTEC 3: Therapeutics, Diagnostics, Cell Culture, Product Isolation: Hollenberg, C. P., Sahm, H. (Eds.), Stuttgart: Gustav Fischer, 1990; pp. 85-89. [20] Sulk, B., Birkenmeier, G., Kopperschlager, G., J Immunol Methods, 1992, 149, 165-171. [21] Chang, Y.K., Chase, H. A., in: Separations of Biotechnology: Pyle, D. L. (Ed.), Cambridge: SCYThe Royal Society of Chemistry, 1994; Vol. 3, pp. 106-112. [22] Morton, P., Lyddiatt, A,, in: Separations for Biotechnology: Pyle, D. L. (Ed.), Cambridge: SCI/The Royal Society of Chemistry, 1994; Vol. 3, pp. 329-335. 1231 Jungbauer, A., Wenisch, E., in: Monoclonal Antibodies: Production and Application: Mizrahi, A. (Ed.), New York: Alan R. Liss, Inc, 1989; Vol. 11, pp. 161-192. 1241 Odde, D. J., in: Handbook of Downstream Processing: Goldberg, E. (Ed.), London: Blackie Academic, 1996; pp. 70-89. [25] Lindmark, R., Thoren-Tolling, K., Sjoquist, J., J Immunol Methods, 1983, 62, 1-13. 1261 Bjorck, L., Kronvall, G., J Immunol, 1984, 133, 969-974. 1271 Carter, P., Kelley, R. F., Rodrigues, M. L., Snedecor, B., Covarmbias, M., Velligan, M. D., Wong, W.L. T., Rowland, A.M., Kotts, C. E., Carver, M. E., Yang, M., Bourell, J.H., Shepard, H.M., Henner, D., BioTechnology, 1992, 10, 163-167. [28] Zhu, Z., Zapata, G., Shalaby, R., Snedecor, B., Chen, H., Carter, P., Bio Technology, 1996, 14, 192-202. 1291 Proudfoot, K.A., Torrance, C., Lawson, A.D. G., King, D. J., Protein Expression and Purification, 1992, 3, 368-373. [30] Laffer, S., Vangelista, L., Steinberger, P., Kraft, D., Pastore, A., Valenta, R., J lmmunol, 1996, 157, 4953-4962.
References
249
[31] De Jonge, J., Brissinck, J., Heirman, C., Demanet, C., Leo, O., Moser, M., Thielemans, K., Mol Immunol, 1995, 32, 1405-1412. [32] Kazemier, B., De Haard, H., Boender, P., Van Gemen, B., Hoogenboom., J lmmunol Methods, 1996, 194, 201-209. [33] Ngo, T. T., Khatter, N., J Chromatogs 1992, 597, 101-109. [34] Tyutyulkova, S., Paul, S., in: Methods in Molecular Biology, Vol51: Antibody Engineering Protocols: Paul, S. (Ed.), Totowa, New Jersey: Humana Press Inc., 1995; pp. 395-401. [35] Shi., J.Y., Goffe, R. A,, J Chromatogr A, 1994, 686, 61-71. 1361 Croze, E. M., European Patent 0453 767 Al, 1991. [37] Aguilar, M. I., Hodder, A.N., Heam, M. T. W., in: HPLC of Polypeptides, Proteins, and Polynucleotides: Hearn, M. T. W. (Ed.), Weinheim: VCH, 1991; pp. 199-245. [38] Yang, Y., Harrison, K., J Chromatogr A, 1996, 743, 171-180. 1391 Burchiel, S. W., Billman, J.R., Alber, T. R., J Immunol Methods, 1984, 69, 33-42. [40] Carlsson, M., Hedin, A., Inganas, M., Harfast, B., Blomberg, F., J Immunol Methods, 1985, 79, 89-98. [41] Hakalahti, L., Vihko, P., J Immunol Methods, 1989, 117, 131-136. [42] Boonekamp, P. M., Pomp, R., Science Tools, 1986, 33, 5-8. [43] Nadler, T. K., Regnier, F. E., in: Chromatography in Biotechnology: Horvath, C., Ettre, L. S. (Eds.), Washington: American Chemical Society, 1993; pp. 14-26. 1441 Nau, D. R., in: Commercial Production of Monoclonal Antibodies: A Guide f o r Scale-Up: Seaver, S. S. (Ed.), New York: Marcel Dekker Inc., 1987; pp. 247-275. [45] Chen, F.-M., Naeve, G. S., Epstein, A. L., J Chromatogr, 1988, 444, 153-164. 1461 Allard, W. J., Moran, C. A., Nagel, E., Collins, G., Largen, M.T., Mol lmmunol, 1992, 29, 1219-1227. [47] Knudsen, K. L., Hansen, M. B., Henriksen, L. R., Andersen, B. K., Lihme, A., Anal Biochem, 1992, 201, 170-177. [48] Nopper, B., Kohen, F., Wilcheck, M., Anal Biochem, 1989, 180, 66-71. [49] Porath, J., Maisano, F., Belew, M., FEBS Lett, 1985, 185, 306-310. 1501 Smith, G. J., McFarland, R.D., Reisner, H.M., Hudson, G. S., Anal Biochem, 1984, 141, 432-436. [51] Stanker, L. H., Vanderlaan, M., Juarez-Salinas, H., J Irnmunol Methods, 1985, 76, 157-169. [52] El-Kak, A., Vijayalakshmi, M. A., J Chromatogr: Biomedical Applications, 1991, 570, 2941. [53] Hale, J.E., Beidler, D.E., Anal Biochem, 1994, 222, 29-33. 1541 Johansson, H. J., Daniels, I., Weitman. A., Soderstrom, L., Westin, G., in: Biologicals from Recombinant Microorganisms and Animal Cells: White M. D. (Ed.), Weinheim: VCH, 1991; pp. 409-414. [55] Shadle, P. J., Erickson, J. C., Scott, R. G., Smith, T. M., International Patent WO 95/22389, 1995. [56] Grunfeld, H., Moore, P., J lmmunol Methods, 1997, 201, 233-241. [57] Can; P. W., Blackwell, J. A., Weber, T. P., Schafer, W. A., Rigney, M. P., in: Chromatography in Biotechnology: Horvath, C., Ettre, L. S. (Eds.), Washington: American Chemical Society, 1993; pp. 146-164. [58] McCarthy, E., Vella, G., Mhatre, R., Lim. Y.-P., J Chromatogr A, 1996, 743, 163-170. [59] Jungbauer, A,, Unterluggauer, F., Uhl, K., Buchacher, A., Steindl, F., Pettauer, D., Wenisch, E., Biotechnology and Bioengineering, 1988, 32, 326-333. [60] Jungbauer, A,, Unterluggauer, F., Steindl, F., Ruker, F., Katinger, H., J Chromatogr, 1987, 397, 313-320. [61] Vallera, D. A,, Burns, L. J., Frankel, A. E., Sicheneder, A. R., Gunther, R., Gajl-Peczalska, K., Pennell, C. A., Kersey, J. H., J Immunol Methods, 1996, 197, 69-83. [62] Neoh, S. H., Gordon, C., Potter, A,, Zola, H., J Immunol Methods, 1986, 91, 231-235. 1631 Brooks, D. A., Bradford, T. M., Hopwood, J. J., J Immunol Methods, 1992, 155, 129-132.
250
8 Purijication and Characterization of Monoclonal Antibodies
Righetti, P. G., Wenisch, E., Jungbauer, A,, Katinger, H., Faupel, M., J Chromatogr, 1990, 500, 681-696. Chase, H. A., in: Discovery and Isolation of Microbial Products: Verrall, M. S. (Ed.), ChiChester: Ellis Honvood Ltd., 1985; pp. 129-147. Bonnerjea, J., Terras, P., in: Bioprocess Engineering: Systems, Equipment and Facilities: Lydersen, B. K., D’Elia, N. A., Nelson, K.L., (Eds.), New York: Wiley, 1994; pp. 159-186. Edwards, J., in: Handbook of Downstream Processing: Goldberg, E. (Ed.), London: Blackie Academic, 1996; pp. 167-184. FDA, Points to Consider in the Manufacture and Testing of Monoclonal Antibody Prodoucts for Human Use, 1997. CPMP, Note for Guidance: Production and Quality Control of Monoclonal Antibodies, 1994. Chase H. A., J Chem Tech Biotechnol, 1986, 36, 35 1-356. Chapman, G. E., Matejtschuk, P., More. J. E., Pilling, P., in: Separations for Biotechnology: Pyle, D. L. (Ed.), Cambridge: Society for Chemical IndustryElsevier, 1990; Vol. 2, pp. 601610. Kenney, A. C., Chase, H.A., J Chem Tech Biotechnol, 1987, 39, 173-182. Knappik, A., Pliickthun, A., BioTechniques, 1994, 17, 754-761. Pharmacia, Science Tools from Pharmacia Biotech, 1996, I , 10. [75] Vaughan, T. J., Williams, A. J., Pritchard, K., Osbourn, J.K., Pope, A.R., Eamshaw, J. C., McCafferty, J., Hodits, R.A., Wilton, J., Johnson, K. S., Nature Biotechnology, 1996, 14, 309-314. [76] Deramoudy, EX., Chaabihi, H., Poul, M. A., Margaritte, C., Cerutti, M., Devauchelle, G., Bernard, A,, Lefranc, M. P., Kaczorek, M., in: Animal Cell Technology: Developments towards the 21”‘ Century: Beuvery, E. C., Griffiths, J. B., Zeijlemaker, W. P., (Eds.), Dordrecht: Kluwer Academic Publishers, 1995; pp.469-473. [77] Yurov, G. K., Neugodova, G. L., Verkhovsky, 0.A,, Naroditsky, B. S., J Immunol Methods, 1994,177, 29-33. [78] Hjorth, R., Trends Biotechnol, 1997, 15, 230-235. [79] Jungbauer, A., Uhl, K., Schulz, P., Tauer, C., Gruber, G., Steindl, F., Buchacher, A., Schoenhofer, W., Unterluggauer, F., Biotechnol Bioeng, 1992, 39, 579-587. [80] Birkenmeier, G., Dietze, H., in: Biotechnology of Blood Proteins: Rivat, C., Stoltz, J.-F. (Eds.), Paris: Colloque INSERM/John Libbey Eurotext Ltd., 1993; Vol. 227, pp. 201-206. [81] Grubb, R. E., in: Immunochemistry: van Oss, C. J., Regenmortel, M. R. V. (Eds.), New York: Marcel Dekker, 1994; pp. 47-68. [82] Clark, M. R. (Ed.), Protein Engineering of Antibody Molecules for Prophylactic and Therapeutic Applications in Man. Nottingham, U.K.: Academic Titles, 1993. [83] Andersson, C. J., Schwarz, S. W., Connett, J.M. et al., J Nucl Med, 1995, 36, 850-858. [84] O’Connor, J. V. O., Keck, R. G., Harris, R. J., Field, M. J., Dev Biol Stand, 1994, 83, 165173. [85] Schenerman, M. A., Phillips, K., BioPharm, 1997, 10, 20-26. [86] Lewis, D. A,, Guzzetta, A. W., Hancock, W. S., Costello, M., A n d Chem, 1994, 66, 585-595. [87] Harris, R. J., Murnane, A.A., Utter, S.L., Wagner, K. L., Cox, E.T., Polastri, G. D., Helder, J. C., Sliwkowski, M.B., Biotechnology, 1993, 11, 1293-1297. [88] Hancock,W. S. (Ed.), New Methods in Peptide Mapping For The Characterisation of Proteins. New York: CRC Series in Analytical Biotechnology, 1996. [89] Allen, D., Baffi, R., Bausch, J., Bongers, J., Costello, M., Dougherty, J.,Jr., Federici, M., Gamick, P., Peterson, S., Riggino, R., Sewerin, K., Tuls, J., Biologicals, 1996, 24, 255-275. [90] Nadler, T., Blackburn, C., Mark, J., Gordon, N., Regnier, F. E., Vella, G., J Chromatogr, 1996, 743, 91-98. [91] Bennett, K. L., Smith, S. V., Lambrecht, R. M., Truscott, R. J. W., Sheil, M. M., Bioconjug Chem, 1996, 7, 16-22. 1921 Alexander, A. J., Hughes, D.E., Anal Chem, 1996, 67, 3626-3632.
References
25 1
[93] Kroon, D. J., Freedy, J., Burinsky, D. J., Sharma, B., J Pharm Biomed Appl, 1995, 13, 1049-1054. 1941 Bailey, J.M., Tu, O., Issai, G., Ha, A., Shively, J.E., Anal Biochem, 1995, 224, 588-598. [95] Pattersom, D. H., Tan; G.E., Regnier, F. E., Martin, S. A,, Anal Chem, 1995, 67, 39713978. [96] Harris, R. J., J Chromatogr A, 1995, 705, 129-134. [97] Powell, M. F., in: Characterization and Stability of Protein Drugs: Pearlman, R., Wang, Y.D. (Eds.), New York: Plenum Press, 1996; pp. 1-140. [98] Mulkerrin, M. G., in: Spectroscopic Methods For Determining Protein Structure: Have1 H. A. (Ed.), Weinheim, VCH Press, 1996; pp. 5-27. [99] Jiskoot, W., Hlady, V., Naleway, J. J., Herron, J. N., in: Physical Methods to Characterize Pharmaceutical Proteins: Herron, J. N., Jiskoot, W., Crommelin, D. J. A. (Eds.), New York: Plenum Press, 1995; pp. 1-63. [loo] Chowdhry, B.Z., Cole, S.C., Trends Biotechnol, 1989, 7 , 11-18. [ l o l l Channing-Rodgers, R. P., in: Basic and Clinical Immunology: Stites, D. P., Ten, A. I., Parslow, T. G. (Eds.), New York, Appleton & Lange, 1994; pp. 151-194. [lo21 Cooper, L. J.N., Robertson, D., Granzow, R., Greenspan, N. S., Mol Immunol, 1994, 31, 577-584. [lo31 Jeffcoate, S., Trends Biotechnol, 1996, 14, 121-124. [lo41 Guttman, A., Electrophoresis, 1996, 17, 1333-1341. 11051 Pritchett, T., in: Handbook of Capillary Electrophoresis Applications: Shintani, H., Polonsky, J. (Eds.), London: Blackie, 1997; pp. 240-254. [lo61 Gianazza, E., J Chromatogr A, 1995, 705, 67-87. [lo71 Hunt, G., Moorhouse, K.G., Chen, A. B., J Chromatogr A, 1996, 744, 295-301. [lo81 Roberts, G.D., Johnson, W. P., Burman, S., Anumula, K. R., Carr, S.A., Anal Chem, 1995, 67, 3613-3625. 11091 Rao, P. E., Kroon, D. J., in: Stability and Characterization of Proteins and Peptide Drugs; Case Histories: Wang, Y. J., Pearlman, R. (Eds.), New York: Plenum Press, 1993; pp. 135158. [110] Michaelson, T.E., LGfsgaard, M. F., Aase, A., Heyman, B., JZmmunol Methods, 1992, 146, 9-16. [ l l l ] Page, M., Ling, C., Dilger, P., Bentley, M., Forsey, T., Longstaff, C., Thorpe, R., Vox Sang, 1995, 69, 183-194. [112] Usami, A,, Ohtsu, A,, Takahama, S., Fujii, T., JPharm Biomed Anal, 1996, 14, 1133-1140. [113] Bazin, R., Boucher, G., Monier, G., Chevrier, M. C., Verrette, S., Broly, H., Lemieux, R., J lmmunol Methods, 1994, 172, 209-217. [114] Rinderknecht, E., Zapata, G. A,, World patent W096/33208, 1996. [115] Jungbauer, A,, Tauer, C., Wenisch, E., Uhl, K., Brunner, J., Purtscher, M., Steindl, F., Buchacher, A,, J Chromatogl; 1990, 512, 157-163. 11161 Karl, D. W., Donovan, M., Flickinger, M. C., Cytotechnology, 1990, 3, 157-169. [117] van Erp, R., Adorf, M., van Sommeren, A. P., Grinbau, T. C., J Biotechnol, 1991, 20, 249261. [118] Mohan, S.B., Chohan, S.R., Eade, J., Lyddiatt, A,, Biotech Bioengng, 1993,42, 974-986. [119] Li, L., Sun, M., Gao, Q.-S., Paul, S., Mol lmmunol, 1996, 33, 593-600. [120] Angal, A. S . , King, D. J., Bodmer, M. W., Turner, A., Lawson, A. D. G., Roberts, G . , Pedley, B., Adair, J. R., Mol Immunol, 1991, 30, 105-108. [121] Smith, M. A., Easton, M., Everett, P., Lewis, G., Payne, M., Riveros-Moreno, V., Allen, G., lnt J Peptide Prot Res, 1996, 48, 48-55. [122] Manning, M. C., Patel, K., Borchardt, R. T., Pharm Res, 1989, 6 , 903-918. 11231 Moellering, B. J., Tedesco, J. L., Townsend, R.R., Hardy, M. R., Scott, R. W., Prior, C. P., BioPharm, 1990, 3, 30-38. [124] Cacia, J., Keck, R., Presta, L. G., Frenz, J., Biochemistry, 1996, 35, 1897-1903. [125] G o o n D. J., Baldwin-Ferro, A., Lalan, P., Pharm Res, 1992, 9, 1386-1393.
252
8 Purification and Characterization of Monoclonal Antibodies
Rao, P., Makowski, M., Meyer, E., Williams, A., Baldwin, A., Ferro, A., Hanigan, E., Kroon, D., Numsuwan, V., Tran, A., Rubin, E., BioPham, 1991, 4 , 38-43. Wright, A,, Morrison, S. L., Springer Series in lmmunopathology, 1993, 15, 259-273. Lund, J., Tanako, T., Takahashi, N., Sarmay, G., Arata, Y., Jefferis, R., Mol lmmunol, 1990, 27, 1145-1153. Wawrzynczak, E. J., Cumber, A. J., Parnell, G.D., Jones, P.T., Winter, G., Mol lmmunol, 1992, 29, 213-220. Kumpel, B. M., Rademacher, T. W., Rook G. A. W., Williams, P. J., Wilson, I. B. H., Hum Antibod Hybrid, 1994, 5 , 143-151. Goodrick, J., Kumpel, B., Pamphillon, D., Fraser, I., Chapman, G., Dawes, B., Anstee, D., Clin Exp Immunol, 1994, 98, 17-20. Boyd, P.N., Lines, A. C., Patel, A.K., Mol Zmmunol, 1995, 32, 1311-1318. Wright, A., Morrison, S. L., Trends Biotechnol, 1997, 15, 26-32. Monica, T. J.,Goochee, C.F., Maiorella, B.L., Biotechnology, 1993, 11, 512-515. Jenkins, N., Parekh, R. B., James, D.C., Nature Biotechnol, 1996, 14, 975-981. Sumar, N., Bodman, K. B., Rademacher, T. W., Dwek, R. A,, Williams, P., Parekh, R. B., Edge, J., Rook, G.A. W., Isenberg, D.A., Hay, F.C., Roitt, I.M., J lmmunol Methods, 1990, 131, 127-136. Patel, T. P., Parekh, R.B. in: Methods in Enzymology: Lennarz, W. J., Hart, G. W. (Eds.), New York: Academic Press, 1994; Vol. 230, pp. 57-66. Kobata, A. in: Methods in Enzymology: Lennarz, W. J., Hart, G. W. (Eds.), New York: Academic Press, 1994; Vol. 230, pp. 200-208. McGuire, J. M., Douglas, M., Smith, K.D., Carbohydr Res,1996, 292, 1-9. Guile, G.R., Rudd, P.M., Wing, D.R., Prime, S. B., Dwek, R. A., Anal Biochem, 1996, 240, 210-226. Okafo, G., Burrow, L. M., Neville, W., Truneh, A., Smith, R. A. G., Reff, M., Camilleri, P., Anal Biochem, 1996, 240, 68-74. Coco-Martin, J. M., Brunink, F., van der Velden de Groot, T. A. M., Beuvery, E. C., J lmmuno1 Methods, 1992, 155, 241-248. Tandai, M., Endo, T., Sasaki, S., Masuho, Y., Kochibe, N., Kobata, A., Archiv Biochem Biophvs, 1991, 291, 339-348. [I441 Lifely; M. R., Hale, C., Boyce, S., Keep, M. J., Phillips, J., Glycobiology, 1995, 5, 813822. [145] Sheeley, D.M., Merrill, B. M., Taylor, L. C.E., Anal Biochem, 1997, 247, 102-110. [146] Maiorella B. L., Winkelhake, J., Young, J., Moyer, B., Bauer, R., Hora, M., Andya, J., Thomson, J., Patel, T., Parekh, R., Biotechnology, 1993, 11, 387-392. [147] Maiorella, B.L., Ferris, R., Thomson, J. et al., Biologicals, 1993, 21, 197-205.
Part Two Quality and Characterization
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
9 Biological Standardization of Interferons and Other Cytokines Anthony Meager
9.1 Introduction An increasing number of biologically active proteins have been discovered and characterized since the early 1970s. Many of these have biological activities that have encouraged their development on a large scale for clinical evaluation as biotherapeutic products. These include interferons (IFNs), interleukins (ILs), colony-stimulating factors (CSFs), and polypeptide growth factors (PGFs), which collectively are designated as ‘cytokines’. What they have in common are that they: (i) induce biological activities via specific cell surface receptors; (ii) are themselves inducible in most cases; (iii) act locally within the environs of producer cells in either an autocrineor paracrine-manner; and (iv) are active at very low concentrations both in vivo and in v i m . This last property has permitted the development of in vitro biological assays (bioassays) for the quantification of the biological potency of cytokines. Bioassays are the only means by which the potency of cytokines can be determined. Physico-chemical methods of analysis are essential to ensure certain aspects of the quality of cytokines, but, on their own, do not allow full characterization of these complex, biologically active proteins; neither do such methods predict or measure cytokine potency. It is therefore essential that the potency of cytokines is quantified in robust, well-designed bioassays. These must be monitored for sensitivity on an assay-to-assay basis by the inclusion, with test samples, of an appropriately defined and characterized biological standard, normally containing the homologous cytokine, in every bioassay. To this end, individual testing laboratories have developed inhouse biological standards or reference reagents for the particular cytokine(s) they are developing, or have developed. The subject of biological standardization is, however, a complex, and often difficult one to address, particularly where cytokines are concerned, principally due to the fact that most cytokines mediate a variety of biological activities in vitro. Thus, there are usually no ‘reference’ bioassays for individual cytokines. Bioassays conducted in different laboratories often depend on variable sources of somatic mammalian cells and culture reagents, and are subject to variations in both design and methodology. This leads to the definition of unitages of biological potency for cytokines being bioassay-dependent and thus, from different laboratories, the reporting of potency values in non-comparable units. To overcome this problem, which has
256
9 Biological Standardization of Interferons and Other Cytokines
grown in importance as cytokines have ‘entered’ the clinic, the World Health Organization (WHO) has instigated and coordinated international efforts to evaluate suitable, well-characterized, cytokine materials for the purpose of establishing international standards (IS) and reference reagents (RR) for individual cytokines. WHO IS and RR, which contain lyophilized cytokines, comply with WHO guidelines for the preparation, characterization, and establishment of international and other standards and reference reagents for biological substances [l], and are of proven stability on storage at -20°C. The National Institute for Biological Standards and Control (NIBSC), Blanche Lane, South Mimms, Herts., EN6 3QG, (Tel: +44 1707 654753, Fax: +44 1707 646730) has played the major role in the development and preparation of candidate IS and RR for individual cytokines. NIBSC has, on behalf of WHO, and in collaboration with the Center for Biologics Evaluation and Research (CBER) (The National Institutes of Health (NIH), Bethesda, Maryland 20205, USA), initiated and coordinated many international and other collaborative studies aimed primarily at the evaluation of candidate IS and RR and the identification of the most suitable one (among such candidate IS and RR) to serve as the WHO IS or RR for a particular cytokine. Priorities for cytokine standardization are set by the WHO Consultative Group on Cytokine Standardization (CGCS). Following full statistical analysis of raw data received from participants in any one collaborative study, the study’s organizers make recommendations to WHO CGCS and the WHO Expert Committee on Biological Standardization (ECBS) as to which they think is the most suitable candidate IS/RR, among those evaluated, to serve as the WHO IS/RR. Final adoption and establishment of WHO IS and RR is decided by WHO ECBS on an annual basis. Once an IS or RR for a particular cytokine has been established with an assigned ampoule content in international units (IU) or reference units (RU), ampoules are made available on a request basis to enable laboratories world-wide to calibrate their bioassays and to facilitate the assignment in IU or RU to ‘in-house’ working standarddreference reagents. The WHO IS or RR should not be used routinely to calibrate bioassays; this should be effected with the in-house working standardheference reagent. Such a strategy should help preserve stocks of the WHO IS or RR so that ampoules for distribution will last for many years. Eventually, ampoule stocks of a current WHO IS will be run-down to a point at which the IS will need replacing. The IS here plays an important role in that it is used as a basis for providing continuity over long periods of time for the IU of biological activity. The first WHO IS is used to calibrate accurately the second WHO IS of the same cytokine before the first IS is exhausted. This strategy requires the best achievable standards of accuracy, reproducibility, and stability. An impressive historic example of this approach is illustrated by the insulin standards where the first IS and IU of biological activity were established in 1926. Nonetheless, in 1997, over 70 years later, manufacturers, national control laboratories, physicians, and patients will use exclusively the same IU, although the current IS is the fifth IS for insulin. For the majority of cytokines, biological standardization has been a recent endeavor, but work was started towards this end in the 1960s with the then newly available IFNs from human, mammalian, and avian sources. The biological standardization of IFNs illustrates well the principles involved and the difficulties, both practical and theoretical, that are common to the biological standardization of cytokines in general.
9.2 Interferons
2.57
Therefore, this chapter will focus largely on the biological standardization of IFNs. However, an understanding of this process requires some familiarity with IFN designations and the molecular and biological characteristics of IFNs, and these are outlined in the next section.
9.2 Interferons 9.2.1 Background Information: Definitions, Designations and Characterization The word ‘interferon’ was coined in 1957 to describe a substance produced by virusinfected chick cell cultures which, on transfer to fresh uninfected chick cell cultures, could elicit a protective antiviral effect [2]. Subsequently, IFNs have been induced from many types of mammalian, including human, cells and demonstrated to be active against a broad spectrum of viruses [3]. Although the protein nature of IFNs was established at a relatively early stage following its discovery [4], it was only following the introduction of large-scale production methods in the 1970s [5] and the simultaneous development of efficient purification procedures [6,7] that sufficient amounts of IFN became available for molecular characterization and clinical evaluation. Further progress followed rapidly from the late 1970s with the advent of recombinant DNA (rDNA) technology which, together with the pharmaceutical industry’s desire to produce pharmacologically active proteins cheaply and growing evidence that IFN had antitumor activity [8] besides antiviral activity, led to the cloning and subsequent mass production of IFN for clinical evaluation. The successful cloning of the IFN-a type, the main IFN produced by virusinfected human leukocytes, revealed it to be a mixture of up to 13 distinct, but molecularly closely related proteins, now known as IFN-a subtypes. Each IFN-a subtype is expressed from a separate chromosomal gene and contains 16.5-166 amino acids [9]. In contrast, the IFN-0 type, the IFN produced chiefly by stimulated human diploid fibroblasts, was found to be a single molecular species, 166 amino acids long, which was evolutionally distantly related (ca. 28 % homology) to IFN-a subtypes, but antigenically distinct from them [lo]. Additionally, a third IFN type, known as IFN-o, was cloned in around 1985 and also shown to be a single molecular species of 172-174 amino acids. However, IFN-o is produced by human leukocytes and probably arose by divergence from an ancestral IFN-a subtype [ l l ] , but is now antigenically distinct from ‘modern-day’ IFN-a subtypes [ 121. Despite the divergence of primary amino acid sequences, these three IFN types, a , p, o,share a common three-dimensional structure of five a-helical regions arranged in a compact bundle [ 131. This characteristic, together with shared biophysical and biological properties, e.g., stability of activity at acid pH, and recognition of a common class of cell surface receptors [14,15], has led to IFN-a subtypes, IFN-0, and IFN-o, being designated collectively as type I IFN.
258
9 Biological Standardization of Interferons and Other Cytokines
A distinct IFN, originally known as ‘immune IFN’ because of its T-cell source, was cloned in 1982 [16]. Its amino acid sequence was found to be unrelated to any of those of the type I IFNs. Unlike the latter, it was subsequently shown to have a homodimeric structure, although each protomer contained five a-helical domains in an arrangement similar to that of IFN-f3 [13,17]. This IFN is now known as IFN-y and, by virtue of the acid-lability of its activity, has been classified as type I1 IFN. IFN-y has also been demonstrated to bind to a class of cell surface receptors distinct from that recognized by type I IFNs [18], although the two receptor classes show some homology in their extracellular domains [ 191. Biological responses to IFN are initiated by IFN binding to cell surface receptors and contingent activation of cytoplasmic signal transduction pathways, and manifested following expression of a number of ‘IFN-inducible’ genes [20]. Induction of antiviral action, which is dependent on such inducible protein synthesis, can now be seen as just one of many activities mediated by IFN; these activities include inhibition of cell proliferation, regulation of cell differentiation, regulation of functional cellular activities, and immunomodulation [2 11. Despite molecular differences between type I (a, (3, w) - and type I1 (y) -IFNs, and differences that exist in signaling pathways [22,23] and inducible genes [20], most of these activities are shared by type I and type 11 IFNs [21,23]. However, type I1 IFN appears to have some unique activities, e.g., macrophage activation [24], induction of de m v o class I1 MHC antigen (HLA-DR) expression [25], and, in some cases, to differ in the degree a shared activity is elicited.
9.2.2 Characteristics of IFNs for Clinical Use Until about 1978, partially purified leukocyte IFN, derived from the supernatants of Sendai virus-infected human buffy-coats, produced by the Finnish Red Cross Blood Transfusion Service (FRCBTS) on a non-commercial basis, was the only IFN available in sufficient amounts for clinical evaluation. Leukocyte IFN continues to be manufactured on a modest scale by FRCBTS, but is now also made commercially elsewhere. The composition of leukocyte IFN is approximately 90% IFN-a,a heterogeneous mixture of up to 14 IFN-a subtypes of which IFN-a1 and IFN-a2 subtypes are major components [9,26,27] , 2-3 % IFN-0 and 7-15 % IFN-w [28]. Purification methods often utilize IFN-a-specific antibody affinity chromatographic separations, which result in IFN preparations containing mixtures only of IFN-a subtypes, but in different proportions to those of the unpurified leukocyte IFN. To overcome the problem of limited supplies of leukocyte IFN that could be derived from human buffy-coats, the Wellcome Foundation set up production of IFN from Sendai virus-infected Namalwa cells, a B-lymphoblastoid cell line that could be grown in large suspension cultures [29]. The purified product is known as lymphoblastoid IFN, and consists of a heterogeneous mixture of IFN-a subtypes, but in different proportions to those found in leukocyte IFN [30]. The IFN-a2 subtype is a major component of lymphoblastoid IFN [30].
9.2 Interferons
259
IFN-a was cloned in 1980 [9] and the mass production of particular IFN-a subtypes by genetically engineered Escherichia coli cultures was rapidly developed. Initially, two allelic variants of the IFN-a2 subtype, designated IFN-a2, [26] and IFN-aZb [3 11 were manufactured and simultaneously made available for clinical trials in 1981. IFN-a~,differs from IFN-aZb in only one amino acid, Lys23 instead of Arg23. A third variant of IFN-a2, IFN-az,, with a substitution of arginine for histidine at position 34 has also been manufactured [32]. All three variants synthesized by E. coli lack the O-linked oligosaccharide side chain present at Thrl06 in human somatic cell-derived IFN-a2 [33]. Although the three recombinant IFN-azs are very closely related molecularly, in vitro biological and antigenic differences have been reported [34]. The development of other individual IFN-a subtypes for clinical use has been rather limited, because of patents curbing their exploitation and the dominant market share of IFN-a2 products . IFN-a subtypes, such as a1, a,and as, appear to have interesting properties, and may potentially be clinically useful. However, the revolution in rDNA technology has made possible the construction of many kinds of ‘non-natural’ hybrid IFNs [35,36] and idealized consensus sequence IFN-a molecules [37]. Thus, hybrid IFN-a molecules were produced with an N-terminal section of one IFN-a subtype and a C-terminal section of another subtype, and these have been designated, for example, IFN-allz, IFN-a1/8 [36]. These displayed interesting in vitro properties [35,36], but have not been, with the exception of one particular form of IFN-a1/8, tested clinically. A recombinant IFN-a homolog, IFN-conl, having a computer ‘idealized’ consensus amino acid sequence based upon known sequences of IFN-a subtypes has also been developed [37]. IFN-con1 has been reported to be more active in vitro in certain bioassays than IFN-a subtypes, including IFN-a2 [38]. Both IFN-a1/8 and IFN-con1 have been found to have more moderate side effects in phase I clinical trials than IFN-az and leukocyte IFN-a or lymphoblastoid IFN-a [36]. Fibroblast IFN-0 derived from stimulated human diploid fibroblasts is both difficult to produce in large quantities and to purify [6]. However, modest amounts have been and continue to be prepared from this source for clinical use since the late 1970s. Isolation of the IFN-0 cDNA in 1980 [lo] led to attempts to mass produce IFN-p in E. coli in a way similar to that was successful for the production of IFN-a2. Production by this means proved problematic as IFN-/3 lacks one of the four cysteine residues which form two intramolecular disulfide bridges in IFN-a subtypes, leading to mismatched pairing of its three cysteines in E. coli and the formation of inactive IFN-P molecules. Subsequently, one manufacturer substituted the cysteine at position 17, which is not involved in disulfide bond formation, for serine and has developed a recombinant IFN-P Serl7 product, Betaseron or Betaferon (also known as IFN-p-1-b) [39]. This product, and any other recombinant IFN-0 derived from E. coli, lacks the N-linked oligosaccharide side chains present in human fibroblast-derived IFN-0. Glycosylated, recombinant IFN-P can, however, be manufactured using transfected Chinese hamster ovary (CHO) cells as producer cells. The carbohydrate composition of CHO cell-derived recombinant IFN-P is non-identical to that of human fibroblast-derived IFN-P [40]. It is by comparison to E. coli-derived recombinant IFN-0 more soluble, probably on account of its hydrophilic carbohydrate side chain.
260
9 Biological Standardization of Interferons and Other Cytokines
Interferon omega (IFN-o) is a component of leukocyte IFN [28]. It has been cloned and expressed both in E. coli and CHO cells [41]. However, IFN-o is a glycoprotein containing an N-linked oligosaccharide side chain, and thus production of glycosylated IFN-o is only attainable in CHO cells [41,42]. Interferon gamma (IFN-y), like IFN-P and IFN-o, is an N-linked glycosylated protein when derived from human cells [43]. Production from human T lymphocytes or CHO cells is possible, but yields are low. Therefore, the major route of manufacture is by expression in E. coli and subsequent purification from bacterial lysate. Nonglycosylated recombinant IFN-y is biologically active [44].
9.3 Interferon Standardization The biological standardization of IFNs is of critical importance to both the pre-clinical testing and clinical development of IFNs. It is clear that the uniform reporting of potency values of IFN activity in an internationally defined unitage is highly desirable in research publications and is essential for clinically used IFNs to ensure corredprecise dosing. However, there are (as outlined in Section 9.2.2) several different natural IFNs, with the possibility of markedly extending this range by rDNA technological methods. This molecular heterogeneity is associated with variable biological capability, quantitatively and, in some cases, qualitatively, among different IFN molecules, and has presented an impressive challenge to those involved in the practical aspects of biological standardization.
9.3.1 Basic Principles of Biological Standardization The global aim of biological standardization is to permit the reliable comparison of the results of bioassays performed by different individuals with different reagents and at different times [45]. Such results must therefore be valid, reproducible, and as accurate as possible. The validity of bioassays is thus critical for achieving biological standardization. Section 9.3.2 will discuss the various types of bioassays and methodology. What follows here is a consideration of the basic principles of standardization. In terms of clinical chemistry, an assay is seen as a procedure to quantify an ‘analyte’. In the present discussion, for ‘analyte’ read IFN or cytokine. An IFN (or cytokine) can be quantified if an assay generates a measurable parameter that reproducibly increases with increasing concentration or dose of IFN (or cytokine), i.e. in a concentratioddose-related manner [46,47]. When two or more different IFN preparations are being compared, they must behave identically in the assay for the assay to be truly valid. Assay reproducibility is essential for biometric validity, and is best achieved by minimizing, where possible, known variables and ensuring that all variables (both known and unknown) are allocated treatments randomly (see Section 9.3.3). It helps greatly if the measured parameter is exclusively related to the IFN
9.3 Interferon Standardization
261
concentration, and not to the concentrations of active impurities, i.e. that the assay has appropriate specificity. A consequence of this is that the purer the IFN preparation, the less specific an assay system needs to be, and vice versa. It should be stressed, however, the use of purified IFN does not confer specificity on an assay system. In practice, it is often difficult completely to avoid 'interference' from substances that may contaminate IFN preparations, e.g., other cytokines, endotoxin (LPS). The contribution of such contaminants to assay responses should, where possible, be evaluated. Appropriate steps should, if necessary, be taken to minimize interference. It has been mentioned previously that when two or more IFN preparations are being quantified in the same assay, they must behave identically for the biometric validity of results. This is an essential tenet for assay calibration, where one IFN preparation serves as the standard to which other IFN preparations included in the assay must behave as if they were a more concentrated or more dilute solution of the standard preparation. Stated differently, if like is compared with like, in the same assay system, under the same conditions, then any difference in measured response from the assay system reflects only the differences in concentrations/doses. Evidence that a test IFN preparation and an IFN standard behave similarly in an assay system is best displayed by parallelism of the graphic plots of the (log) dose-response curves (Fig. 9.1). In theory, the dose-response curve of the test IFN should, if it con-
Relative potency of the two preparations
Log Dose Fig. 9-1. Graphical plots of sigmoidal log dose-response curve in a comparison of two preparations that behave similarly in a bioassay system. ABCDEFGH represents the shape of a typical log dose-response curve. The potency of one preparation relative to the other is represented by the amount of one preparation which gives the same response as a measured quantity of the other, i.e., the horizontal distance between the linear part of the curves B-F and b-f. Use of three dose levels of each preparation gives the minimum information necessary to allow assessment of linearity and parallelism of the preparations.
262
9 Biological Standardization of Interferons and Other Cytokines
tains IFN molecules identical to those in the IFN standard, be exactly parallel to the dose-response curve of the IFN standard [46,47]. However, it must be emphasized that although parallelism is an essential part of the evidence to prove validity of results, it is not, by itself, proof that the two IFNs compared are identical. Where parallelism of dose-response curves is evident, the potency of the test IFN relative to the standard IFN is represented by the amount of the test preparation which gives the same response as a known (measured) amount of the standard, i.e., the horizontal distance between the linear part of the curves A-H and a-h (Fig. 9.1). In practice, evidence of parallelism of dose-response curves may be difficult to demonstrate conclusively for reasons related to the numerous variables that are inherent in design and performance of bioassays (see below). Therefore, parallelism should be assessed statistically, both within and among assays performed on different occasions, to improve the biometric validity of results [47]. Another important aspect for standardization is the ‘precision’ with which results can be obtained from an assay. Precision in this context means the consistency of agreement of repeated measurements, and this may be considered as an aspect of validity. It is usual, when stating the estimate of potency of an IFN preparation, also to calculate the precision of the potency estimate, which is normally presented as fiducial limits at a given level of probability, e.g., 95 % ( P = 0.95). The 95 % limits encompass the range of values between which the assay potency estimates could be expected to fall 95 times if the assay was done 100 times with the same precision. For clinical IFN products where precise estimation of potency is important, fiducial limits are set at not less than 64% and not more than 156% of the stated potency. The goal should be to achieve an estimated potency that is not less than 80% and not more than 125 % of the stated potency. To obtain this level of precision requires careful consideration of available assay systems and designs. The range of bioassays for IFN potency estimations that are available is discussed below.
9.3.2 Bioassays for IFNs IFNs exert a wide spectrum of different biological activities in vitro [21], which include antiviral-, antiproliferative-, and differentiative-actions in many cultured somatic cells. Thus, there are potentially a large number of bioassay systems that are capable of providing dose-response data for IFN potency estimations. However, historically, IFNs are thought of as antiviral agents and, as a consequence, most laboratories interested in IFN research have developed antiviral assays [48,49]. The easy availability of human and mammalian cells, especially transformed, or tumor-derived cell lines, has provided a great variety of cell ‘substrates’ for antiviral assay development. Additionally, there are several human and mammalian viruses that are accessible, easy to propagate, and sensitive to the antiviral action of IFNs, which enables them to be used in antiviral assays 149). The number of cell-virus combinations is potentially vast. Nevertheless, nearly four decades of experimentation has demonstrated that only a small, select number of cell-virus combinations are really suitable and useful for antiviral assays [49]. Most human IFNs exhibit
9.3 Interferon Standardization
263
Table 9-1. Human and mammalian cell lines used in routine antiviral assays for the quantification of human IFN. Cell line
Origin
Morphology
Source
Hep2
Human larynx carcinoma
Epithelial
ATCC"ICCL231
WISH
Human amnion
Epithelial
ATCClCCL25
A549
Human lung carcinoma
Fibroblastic
ATCClCCL18.5
GM-2504
Human skin (trisomic chr 21)
Fibroblastic
HGMCRh/GM-2540E
GM-2767
Human skin (trisomic chr 2)
Fibroblastic
HGMCRIGM-2767B
2D9
Human glioblastoma
Fibroblastic
W. Daubener (Germany)
MDBK
Bovine kidney
Epithelial
ATCClCCL22
EBTr
Embryonic bovine trachea
Fibroblastic
ATCClCCL44
a
ATCC, American Type Culture Collection. HGMCR, Human Genetic Mutant Cell Repository.
some degree of species specificity and human cells appear the best choice for the assay of human IFNs. Alternatively, some cell lines, e.g., Vero, from closely related species such as monkeys may be usable for human IFN potency estimations. IFN-a, and to a lesser extent IFN-fl, are active in bovine and ovine cell lines; the bovine kidney cell line, MDBK, has been widely used. The most commonly used cell lines for antiviral assays for the titration of human IFNs are listed in Table 9-1. The viruses which are commonly used in antiviral assays include murine encephalomyocarditis virus (EMCV), Mengo virus, vesicular stomatitis virus (VSV), Semliki Forest virus (SFV), and Sindbis virus. These are all RNA viruses which replicate well in the absence of IFN, often causing a cytopathic effect in infected cells [49]. While they infect and replicate in human cell lines, these viruses are not known to be pathogenic, except in extreme circumstances, in humans, and are thus relatively safe to use. Antiviral assays follow the basic schedule of: (i) addition of serial dilutions of IFN preparations to cell monolayers in the wells of microtitre plates; (ii) incubation at 37°C for 20-24 h to establish the antiviral state; (iii) infection of cells with the 'challenge' virus; (iv)incubation at 37 "C for 20-24 h to produce viral manifestations, e.g., a cytopathic effect in unprotected and partially-protected cells; and (v) processing assays by addition of stain or other means to render quantifiable in a subjective/objective manner the antiviral effect of IFN [48,49]. Staining, for example, enables the dilution of IFN at which 50 % of cells in the monolayer are protected, and thus alive, to be identified visually. More objective analysis may be made by stain elution and the spectrophometric measurement of optical densities in individual wells included in the IFN titration. Optical density values (ordinate) are then plotted against the (log) dilution, or reciprocal dilution (abscissa), to generate dose-response curves. Comparison of dose-response curves of test preparations of IFN against the dose-response curve of the IFN standard of assigned potency permits calculation,
-
264
9 Biological Standardization of Interferons and Other Cytokines
normally by computer-assisted means, of the potencies of the test preparations of IFN (see Section 9.3.1 for discussion of biometric validity). Potency values should be expressed in international units (IU). Besides antiviral assays, the potency of IFN preparations can be assessed by a variety of other quantitative bioassays. For example, IFN-a, IFN-0, and IFN-o, but not IFN-y, have antiproliferative activity in the human Burkitt’s lymphoma cell line, Daudi [50, 511. The Daudi cell line has appeared the most sensitive among a variety of human tumor-derived cell lines, [3H]thymidine incorporation being the most common means of quantifying the effect of type I IFNs. Dose-response curves are generated by plotting c.p.m. incorporated versus reciprocal dilution and analysed in a similar way to those generated by antiviral assays. In addition, IFNs are cell-differentiating agents, inducing the expression of several cell surface- and intracellularproteins. Cell surface proteins include class I MHC antigens (all IFNs induce/augment expression) [52], class I1 MHC antigens (only IFN-y induces de ROW expression) [53], intercellular adhesion molecule-1 (ICAM-1, -1FN-y is the best inducer) [54] and the 9-27 (Leul3) antigen (type I IFNs are the best inducers) [55]. Usually, the up-regulated expression of cell surface antigens is detected using specific monoclonal antibodies. These bind to ‘fixed’ cells at the end of incubation with IFN, and are themselves quantified by the addition of an anti-murine IgG-enzyme conjugate and routine ELISA procedures. The amount of color developed is proportional to the expression of cell surface antigen and this is related to the dose of IFN. Again, dose-response curves are generated; this time absorbances or optical densities are plotted against the reciprocal dilutions. Such assays are often referred to as bio-ELISA or bioimmunoassays [54]. IFNs also induce a spectrum of intracellular proteins. Where these are enzymes, e.g., indolamine 2,3 -dioxygenase (IDO), the effect of IFN may be quantified by measurement of enzymatic activity, the conversion of L-tryptophan to N-formyl-kynurenine in the case of I D 0 [56]. Where they are inert, these IFN-induced proteins are detected by specific antibodies. One such example is the MxA protein, an antiviral protein involved in resistance to influenza virus, which is highly inducible in type I IFN-stimulated cells [57]. Alternatively, use can now be made of IFN-inducible gene promoters. For example, the IFN-inducible Mx gene promoter may be linked to firefly luciferase coding gene sequences, and this construct used to transfect cells. Selected transfected cells containing Mx-Luc then respond to IFN by producing luciferase which, with an appropriate luciferase assay reagent, produces fluorescence, that may be measured spectrophometrically [58]. Lastly, IFNs appear to counteract the actions of other cytokines, and the inhibitory effect can be the basis of quantitative assessment of IFN activity. For example, type I IFNs inhibit the stimulatory effect of interleukin-5 (IL-5) or granulocyte-macrophage colony-stimulating factor (GM-CSF) on the proliferation of the human leukemia cell line TF-1 [59]. In summary, there is a variety of bioassays, with different measurable parameters, that are suitable and useful for the quantification of IFN potency. Having such a largesse of methods and activities to test may help to understand aspects of IFN biology that are likely to be involved in disease processes on the one hand, and that underly beneficial/adverse effects in the IFN therapy of human diseases on the other. How-
9.3 Interteron Standardization
265
ever, this diversity of methods and IFN actions is also a cause of concern for the standardization of IFNs, since there are difficulties in establishing standards suitable for use in all types of bioassay. For instance, is an IFN standard whose potency is defined solely by antiviral activity really suitable for calibrating the antiproliferative activity of IFN?
9.3.3 Design of Bioassays Most bioassays are now carried out using 96-well microtiter plates. Such bioassays utilize a systematic design related to the row and column structure of these plates. This has a variable influence on cell and virus growth and thus the positions within the 96-well plate in which test preparations are placed in relation to the standard often influence the shape of dose-response curves and potency determinations. Where there is replication of dilution series of the same preparation of IFN, statistically significant differences are likely to be observed between the replicate series, even when these replicates are placed in neighboring rows or columns of the same microtiter plate and started with aliquots from the same stock solution [60]. Besides positional effects, non-random variability of responses in bioassays in general may result from a variety of reasons. These include: The variable accuracy and effectiveness of multi-channel pipets for serial dilution of test and standard preparations. These can introduce untoward biases which contribute to uneven responses. The inherent difficulty of addition of a constant number of cells to the individual wells of a microtiter plate . Homogeneous single-cell suspensions should be used, but this may not be easy to achieve with naturally adherent cells, and settling effects within a cell suspension can lead to greater numbers of cells being transferred to microtiter (plates which are last in the series to be filled than those which are first in the series. Point (1) may also apply if multi-channel micropipeters are used to transfer the cell suspension to microtiter plates. The diverse factors which affect the maintenance of cell monolayers or suspension in microtiter plate wells, and the replication of viruses (where used in antiviral assays). Variation in cell culture materials, especially calf serum, of which different batches can have markedly different properties, can have different effects on cells. Other inadvertant variations may also affect cell responsiveness. For example, the temperature and pH controls in the incubators in which the microtiter plates are placed may vary from time to time. Microbial contamination of the cell line, especially by mycoplasma. This may have profound effects on cell responsiveness and the degree of these effects may vary from occasion to occasion. ..I. Inadvertant variation introduced by staining procedures, or other procedures required to process the assays, both within an assay and among assays carried out on different occasions. These variations may reflect variable accuracy and effectiveness of multi-channel micropipets in the delivery of stain or other mate-
266
9 Biological Standardization of Interferons and Other Cytokines
rial to microtiter plate wells. The condition of cells at the end of the assay, or at the time of assay processing, can introduce further variation. This may reflect uneven distributions of cells among columns or rows of microtiter plate wells, or uneven growth or survival of cells in wells. For example, it has been observed in many assays that utilize microtiter plates that the growth and survival of cells may be abnormal in the outer wells compared with that in the inner wells. Thus, the design of assays and the control of factors affecting assay responses are of great importance for ensuring accurate and reproducible estimates of relative potency. Experience accumulated from the data obtained in international collaborative studies to evaluate candidate IFN and cytokine standards has shown that certain assay designs are to be preferred. It has been found that designs in which each preparation is tested on several plates in any assay with independent serial dilutions of the preparation on each plate have generally given more accurate and reproducible estimates of relative potency [60]. Moreover, where several preparations are tested on each of several plates, designs in which each preparation is placed at different positions on the different plates should be considered to counteract positional effects. A duplicate of one test preparation should be included in each assay to provide a check on the control of extraneous factors. Complete randomization over all factors other than dose of IFN (or cytokine) which affect assay responses may not be feasible. However, careful design can permit some measure of the influence of such factors and more valid use of classical methods of analysis.
9.4 Interferon Standards Attempts to prepare reference preparations or standards for IFNs began soon after their discovery in 1957. The main work in this area was carried out in what was then the Medical Research Council (MRC) laboratory sited in Hampstead, London. (This facility subsequently became The National Institute for Biological Standards and Control in 1975, and transferred to South Mimms, Hertfordshire, in 1987). Initially, lyophilized reference preparations for chicken IFN (British Research Standard, catalog number 62/4) and monkey (simian) IFN (catalog number 63/1), prepared from rather crude IFN preparations, were developed. The first two figures in the catalog number indicate the year in which the reference preparation was made, i.e., 62/4 = 1962. Later in that decade a further preparation of chicken IFN (67/18) and new reference preparations of human leukocyte IFN, including the MRC Research Standard B (69/19), were produced. Reference preparations for mouse IFN, rabbit IFN, and human fibroblast IFN were subsequently developed in the U.K. and the U.S. during the 1970s . With hindsight, it is now seen that the development of these human, mammalian, and avian IFN reference preparations was carried out in complete ignorance of the structure of IFN molecules and of their molecular heterogeneity. This undoubtedly has had an impact on more recent attempts to develop IFN reference preparations, leading to issues concerning the definition of international unitages and their continuity (as discussed more fully below).
9.4 Interferon Standards
267
In 1969, the International Association of Biological Standardization held an International Symposium on Interferon and Interferon Inducers at which it was recommended that specific preparations of IFNs for human, chick, mouse, and rabbit, to which an agreed unitage of activity had been assigned, be adopted as Research Standards. From then on, these MRC Research Standards, as well as others prepared by the NIH, Bethesda, were made available and used in assays of activity of IFNs. The next decade saw the first clinical trials with human IFN. Although the results of these trials were generally disappointing, interest in IFNs continued to increase with the demonstration of their antitumor activity. By 1977, the WHO recognized an urgent need to establish International Standards for IFNs [61]. On the recommendation of WHO ECBS, a study group to review the status of the then current research standards was convened in September 1978 (Woodstock, Illinois, U.S.A.) by the NIH and co-sponsored by WHO. The study group proposed that as Research Standards for human leukocyte, mouse, rabbit, and chick IFNs (adopted in 1969) [62] had been widely accepted internationally, and as the subsequent NIH standard for human fibroblast IFN had also been used extensively in many countries, that they were considered also as International Standards [63]. The findings of the study group were considered at the 30th ECBS meeting in November 1978 and the following five preparations were adopted as International Reference Preparations (IRP) of IFNs: 1. Preparation 69/19 (MRC Research Standard B) as the IRP of IFN, human leukocyte, 5000 IU of activity per ampoule. 2. The NIH preparation G023-902-527 as the IRP of IFN, human fibroblast, 10000 IU per ampoule. 3. The NIH preparation G002-904-511 as the IRP of IFN, mouse, 12000 IU per ampoule. 4. The NIH preparation G019-902-528 as the IRP of IFN, rabbit, 10000 IU per ampoule. 5 . Preparation 67/18 (proposed replacement British Research Standard B) as the IRP of IFN, chick, 80 IU per ampoule. It should be pointed out that the IU for each IFN type and species were non-equivalent, and non-correlatable. Therefore each IRP was to be used only to calibrate future national standards or laboratory standards for the same typehpecies of IFN contained in the IRP [63]. In addition to the IRP of IFN, human leukocyte, 69/19, a further NIH preparation of similar material designated G023-901-527 with assigned potency of 20 000 IU (against 69/19) was made available [63]. As the momentum of clinical research accelerated in the late 1970s and early 1980s, breakthroughs in basic research led to the cloning of several different molecular species of IFN, and an understanding of the heterogeneous nature of human leukocyte IFN, now designated IFN-a. In 1981, cloned and highly purified IFN-a2 subtype became available for clinical trials. Subsequently, other IFN-a subtypes, IFN-P (formerly fibroblast IFN), IFN-y (formerly immune IFN) , IFN-o, and a variety of hybrid and consensus sequence IFN molecules were produced (see Section 9.2.2). These developments led to the call for new candidate IFN standards to be prepared
268
9 Biological Standardization of Interferons and Other Cytokines
from the highly purified, cloned human IFNs which, following evaluation by international collaborative studies, would serve as International Standards of individual IFN types or subtypes. This call was pursued enthusiastically on behalf of WHO by the successors of the 1978 study group. Three international collaborative studies carried out during the 1980s [64-671 generated data that led to the establishment of International Standards of human IFN-al, human IFN-a2a, human IFN-azb, human lymphoblastoid IFN, human IFN-0, human IFN-P [ser17], and human IFN-y (Table 9.2). However, the IRP for IFN human leukocyte (69/19) was used, with one exception, as the primary standard to which all new IFN-a materials were calibrated. This broke a central principle in biological standardization in that like was not compared with like. The material in 69/19 was a heterogeneous mixture of up to 13 different Table 9-2. International standards for IFNs established by the WHO Expert Commitee on Biological Standardization. Preparation no.
Interferon
67/18
Chick IFN
69/19b Ga23-90 1-532
Defined activity (IU/ampoule)
Year established
Source
80
1978
NIBSC"
Human leukocyte IFN
5 000
1978
NIBSC
Human lymphoblastoid (Namalwa) IFN
25 000
1984
NIH
83/514
Human recombinant IFN-a1
8 000
1987
NIBSC
Gxa01-901-535
Human recombinant IFN-a2,
9 000
1984
NIH
82/576b
Human recombinant IFN-azb
17 000
1987
NIBSC
Gb23-902-53 1
Human IFN-P (fibroblast-derived)
15 000
1987
NIH
Gxb02-90 1-535
Human recombinant IFN-P [Ser17]
6 000
1987
NIH
GxgO 1-902-535
Human recombinant IFN-y
80 000
1995
NIH
Ga02-902-5 11
Mouse IFN-a
16000
1987
NIH
Gb02-902-5 11
Mouse IFN-P
15 000
1987
NIH
Gu02-901-5 11
Mouse IFN-a/p
10000
1987
NIH
Gg02-901-533
Mouse IFN-y
1000
1987
NIH
G-019-902-528
Rabbit IFN
10000
1978
NIH
a
NIBSC, The National Institute for Biological Standards and Control, Blanche Lane, South Mimms, Hertfordshire, EN6 3QG, U.K.; NIH, Research Branch, National Institute of Allergy and Infectious Diseases, National Institutes of Helath, Bethesda, MD 20205 U.S.A. At its 40th Meeting Report [67], the WHO ECBS recommended that all new preparations of human IFN-a should be determined relative to 69/19, the IRP of IFN, human leukocyte. They further recommended that this should also apply to existing preparations of human IFN-a, and that 82/576, IS of human recombinant IFN-aZb, should not be used for relative potency assays. It should be further noted that, because of inconsistent reporting of human IFN-a potency, the current IS of human IFN-a have been re-evaluated in a large International collaborative study organized by NIBSC and CBER [69].
9.4 Interferon Standards
269
IFN-a subtypes, plus small percentages of IFN-0 and IFN-w, whereas the material in a candidate IFN-az, standard was solely IFN-a2,. Additionally, although it was not recognized at the time, the material in 69/19 also contained a range of other cytokines, including IL-10, TNF-a, and IL-6, which had the potential to affect the biological activity of IFN. It was thought then that the establishment of International Standards of different, well-characterized, IFN-a-type materials would assist manufacturers of IFN-a to harmonize unitages of their products. Regrettably, complications have resulted for a number of potential reasons: -
-
-
-
the continuity of IU from 69/19 to other IFN-a standards remains in doubt because of the dissimilarity of materials compared; the assigned potencies of the new IFN-a standards are possibly incorrect due to the fact that collaborative studies were too small (seven to eight participants in most studies) to generate sufficient data from which accurate potency values could be derived; the availability of multiple IFN-a International Standards has the potential for leading to inconsistencies in reporting the results of activity assays; and the IU has been defined on the basis of the antiviral activity of IFN-a and it remains unclear whether the current International Standards of IFN-a materials are suitable for the calibration of potency assays based on biological properties of IFN-a other than antiviral activity, e.g., antiproliferation or immunomodulation. Such alternative potency assays may have more validity for the activity required for the treatment of target diseases, such as cancer. Similar issues have appeared to affect calibration of assays with new IFN-0 and IFN-y International Standards.
Recognition of these potential problems with the standardization of IFNs has become more apparent in the last decade (1987-1997). Several manufacturers of IFN found that they were unable to make a choice from the list of WHO International Standards that would meet their needs for calibrating their own IFN products. In some cases it was found that, by changing from one International Standard to another, a significant readjustment of potency to an IFN product was required. Potentially, this meant that manufacturers would present a variation to the specifications of their product(s) which might be unacceptable to the regulatory licensing authorities. Confusion could occur in the clinic if for example a patient was receiving a 45 MIU IFN product one day and then, due to a readjustment for assays calibrated with a newly available IFN International Standard, the same product had to be assigned an activity of 8 MIU. In theory, this should not be possible, but that it has occurred in practice [68], illustrates a complex situation that probably relates back to the prior use of poorly characterized and purified, often heterogeneous materials used to develop the original early International Standards of IFN. For the most part, the standardization of other cytokines has not suffered these problems, due to the fact that almost all were developed from highly purified recombinant materials (normally homogeneous preparations of single cytokines) from the middle to late 1980s onwards. With regard to the standardization of IFN-a, the WHO Consultative Group on Cytokine Standardization (CGCS) requested in 1995 that the U.K. National Institute for Biological Standards and Control (NIBSC) and the U.S.A. Centre for Biologics
270
9 Biological Standardization of Interferons and Other Cytokines
Evaluation and Research (CBER) organize an International collaborative study to compare the activities and relative potencies of the several available IFN-a preparations, including those derived from human cells containing mixtures of IFN-a subtypes and those derived by rDNA methods containing single IFN-a subtypes, in different assays [69]. This international collaborative study has been carried out and the analysis of data completed. Seventeen (or a defined subset thereof) ampouled preparations of IFN-a were evaluated by 92 laboratories in 29 countries for their suitability to serve as International Standards for these materials. The preparations were titrated in a wide range of in vitro bioassays, including antiviral and antiproliferative assays. A report of the study’s finding is in progress. Recommendations to the WHO CGCS and ECBS are expected to be made in 1998. It is hoped that past difficulties encountered in IFN-a standardization will be resolved by choosing the most suitable IFN-a preparations as International Standards. The WHO CGCS has recently discussed similar problems relating to the standardization of human IFN-P, as were apparent for IFN-a, and has recommended that NIBSC organize a collaborative study to address the pertinent issues.
9.5 Cytokine Standards A range of cytokine preparations has been developed at NIBSC. These, in many cases, have been evaluated in International collaborative studies in bioassays [70721. Analysis of the data of these studies has provided information on the suitability of these cytokine preparations to serve as WHO International Standards of different individual cytokines. Several have now (Table 9-3) been adopted by WHO ECBS and established as full WHO International Standards. However, the rate of discovery of new cytokines has been so rapid that it has been difficult to organize International collaborative studies for each and every one of them. The WHO ECBS therefore has instituted a new category of reference materials which have potency assigned on the basis of data provided by small collaborative studies, often involving two or three participating laboratories. Many different cytokine preparations have recently been accepted on this basis and serve as WHO Reference Reagents for particular cytokines (Table 9.3).
271
9.5 Cytokine Standards
Table 9-3. International standards and reference reagents of cytokines established by the WHO Expert Commitee on Biological Standardization. Preparation Cytokine no.a
Defined activity Status (IU or Ulampoule)
Year established
861632
Interleukin-la, rDNA, E. coli
117000
IS
1990
861680
Interleukin-lp, rDNA, E. coli
100 000
1s
1990
861504
Interleukin-2, Jurkat cell line-derived
202
IS
1988
911510
Interleukin-3, rDNA, E. coli
1700
IS
1995
881656
Interleukin-4, rDNA, CHO cell-derived
1 000
IS
1995
901586
Interleukin-5, rDNA, SW25 cell-derived
5 000
RR
1996
891548
Interleukin-6, rDNA, CHO cell-derived
100 000
1s
1994
901530
Interleukin-7, rDNA, E. coli
100 000
RR
1996
891520
Interleukin-8, rDNA, E. coli
1000
IS
1996
911678
Interleukin-9, rDNA, CHO cell-derived
1 000
RR
1996
921788
Interleukin-1 1, rDNA, E. coli
5 000
RR
1996
951544
Interleukin-12, rDNA, CHO cell-derived
10 000
RR
1996
941622
Interleukin-13, rDNA, E. coli
1000
RR
1996
951554
Interleukin-15, rDNA, E. coli
10 000
RR
1996
891512
Macrophage colony-stimulating factor, rDNA, CHO cell-derived
60 000
IS
1994
881502
Granulocyte colony-stimulating factor, rDNA, yeast-derived
10000
1s
1994
881646
Granulocyte macrophage colonystimulating factor, rDNA, E. coli
10000
IS
1995
931562
Leukemia inhibitory factor, rDNA, E. coli
10000
RR
1996
931564
Oncostatin M, rDNA, E. coli
25 000
RR
1996
871650
Tumor necrosis factor alpha, rDNA, E. coli
40 000
IS
1993
871640
Tumor necrosis factor beta, rDNA, E. coli
150000
RR
1996
a
All of these preparations are held at The National Institute for Biological Standards and Control (NIBSC), Blanche Lane, South Mimms, Hertfordshire, EN6 3QG, U.K. NIBSC also holds reference materials of no defined status for other cytokines, including Interleukin-10 (921516), stem cell factor (911682), flt-3 ligand (961532), transforming growth factor beta-1 (8915 14), transforning growth factor beta-2 (901696), bone morphogenetic protein-2 (931574), and for a number of chemokines.
272
9 Biolonical Standardization of Interferons and Other Cytokines
9.6 Conclusions Significant progress has been made in the biological standardization of cytokines, including IFNs. Problems, however, remain in some cases, particularly with some IFNs where a long history in the development of reference materials, initially from impure IFN preparations, and the heterogeneous nature of certain types of IFN, e.g., IFN-a, have combined to illustrate how difficult it can sometimes be to choose ideal biological standards and to ensure continuity of the international unitage. Nevertheless, it is essential to have International Standards of biologically active cytokines, since these are the only effective means of calibrating ‘in-house’ standards and biological assays used to measure the activity (potency) of cytokines. They are critical to the development of cytokines as biological medicinal products, both at the pre-clinical and clinical stages. It is imperative that the biological potency of such cytokine products is determined accurately in well-calibrated biological assays to ensure that patients receive safe, effective and consistent dosages. The very successful collaboration between NIBSC (U.K.), CBER (U. S.A.), WHO CGCS, and ECBS in bringing about the development and establishment of International Standards and reference reagents of the various and numerous human cytokines is expected to continue well into the next century.
Acknowledgements I am indebted to Mrs Deborah Richards for expert typing of this manuscript.
References [l] World Health Organization. Tech Rep Series, 1990, 800, 181-213. [2] Isaacs, A,, Lindemann, J., Proc Royal SOCB, 1957, 147, 258-267. [3] Stewart, W. E. I., in: The lnteiferon System: Vienna, New York: Springer-Verlag, 1979, pp. 134-183. [4] Fantes, K. H., in: Purification, concentration and physico-chemical properties of interferons, Finter, N. B. (Ed.), Amsterdam: North-Holland, 1966, pp.118-180. [5] Cantell, K., Hirvonen, S . , Tex Rep Biol Med, 1977, 35, 138-144. [6] Knight, E., Proc Natl Acad Sci USA, 1976, 73, 520-523. [7] Rubinstein, M., Rubinstein, S., Familetti, P. C., et al. Science, 1978, 202, 1289-1290. [8] Gresser, I., Maury, C. J., Brouty-Boy6 D., Nature, 1972, 239, 167-168. [9] Nagata, S., Mantei, N., Weissmann, C., Nature, 1980, 287, 401-408. [lo] Taniguchi, T., Mantei, N., Schwarzstein, M., et al. Nature, 1980, 285, 547-549. [ l l ] Capon, D. J., Shepard, H. M. and Goeddel, D.V., Mol Cell Biol, 1985, 5, 768-779. [12] Adolf, G.R., J Gen Virol, 1987, 68, 1669-1676. [13] Senda, T., Shimazu, T., Matsuda, S. et al. EMBO J, 1992, 11, 3193-3201. [14] Uze, G., Lutfalla, G. l., Gresser, I., Cell, 1990, 60, 225-234. [15] Novick, D., Cohen, B., Rubenstein, M., Cell, 1994, 77, 391-400.
References
273
[16] Gray, P.W., Leung, D.W., Pennica, D. et al., Nature, 1982, 295, 503-508. [17] Ealick, S.E., Cook, S.E., Vijay-Kumar, S. et al. Science, 1991, 252, 698-702. [I81 Aguet, M., Dembic, Z., Merlin, G., Cell, 1988, 55, 273-280. [19] Bazan, J.F., Cell, 1990, 61, 753-754. [20] Sen, G.C., Lengyel, P., J Biol Chem, 1992, 267, 5017-5020. [21] De Maeyer, E., De Maeyer-Guignard. J., in: The Interferon Gene Family, New York: Wiley Interscience, 1988, pp. 5-38. [22] Miiller, M., Briscoe, J., Laxton, C. et al., Nature, 1993, 366, 129-135. [23] Vilcek, J., Oliviera, I. C., Int Arch Allergy Immunol, 1994, 104, 311-316. [24] Le, J., Prensky, W., Yip, Y. K. et al., J Immunol, 1983, 131, 2821-2826. [25] Collins, T., Korman, A. J., Wake, C. T. et al., Proc Nut1 Acad Sci USA, 1984, 81,4917-4921. [26] Goeddel, D.V., Leung, D. W., Dull, T. J. et al., Mature, 1981, 290, 20-26. [27] Kauppinen, H.-L., Hirvonen, S., Cantell, K., Methods Enzymol, 1986, 119, 27-35. [28] Adolf, G.R. virology, 1990, 175, 410-417. [29] Phillips, A. W., Finter, N. B., Burman, C. J. et al., Methods Enzymol, 1986, 119, 35-38. [30] Zoon, K. C., Miller, D., Bekisz, J. et al., J Biol Chem, 1992, 267, 15210-15216. [31] Streuli, M., Nagata, S., Weissmann, C., Science, 1980, 209, 1343-1347. [32] Dworkin-Rastl, E., Dworkin, M.B., Swetly, P. J., Interferon Res, 1982, 2, 575-585. [33] Adolf, G.R., Kalsner, I., Ahorn, H. et al., Biochem J , 1991, 276, 511-518. [34] von Gabain, A., Lundgren, E., Ohlsson, M. et al., Eur J Biochem, 1990, 190, 257-261. [35] Alton, K., Stabinsky, Y., Richards, R. et al., in: Production, characterization and biological effects of recombinant DNA derived human IFN-a and IFN-y analogs: De Maeyer, E., Schellekens, H. (Eds.), Amsterdam: Elsevier, 1983, pp. 119-128. 1361 Horisberger, M., Di Marco, S., Pharmacol Ther, 1995, 66, 507-534. [37] Ozes, O.N., Reiter, Z., Klein, S. et al., J Interferon Res, 1992, 12, 55-59. [38] Blatt, L., Davis, J., Klein, S. et al., J Interferon Cytokine Res, 1996, 16, 489-499. [39] Mark, D. F., Lu, S. D., Creasey, A. A. et al., Proc Natl Acad Sci USA, 1981, 81, 5662-5666. [40] Utsumi, J., Mizuno, Y., Hosoi, K. et al., Eur J Biochem, 1989, 181, 545-553. [41] Adolf, G.R., Friihbeis, B., Hauptmann, R. et al., Biochim Biophys Acta, 1991, 1089, 167174. [42] Adolf, G. R., Maurer-Fogy, I., Kalsner, I. et al., J Biol Chem, 1990, 265, 9290-9295. [43] Rinderknecht, E., O’Connor, B.H., Rodriguez, H. et al., J Biol Chem, 1984, 259, 67906797. [44] Rinderknecht, E., Burton, L. E., in: Biochemical characterization of natural, and recombinant, IFM-gamma; Kirchner, H., Schellekens, H. (Eds.), Amsterdam: Elsevier, 1984, pp. 397- 402. [45] Bangham, D. R., in: Assays and Standards: Gray, C. C., James, V. H. T. (Eds.), London: Academic Press, 1983, pp. 256-297. [46] Bliss, C., in: The Statistics OfBioassays: New York: Academic Press, 1952, pp. 256-297. [47] Finney, D. J., in: Statistical methods in Biological Assay, 3rd edition, London: Charles Griffin Co Ltd, 1978. [48] Grossberg, S., Jameson, P., Sedmark, J., in: Came, P., Carter, W. (Eds.), Berlin: Springer, 1983, pp. 23-43. [49] Meager, A., in: Quantification of interferons, by anti-viral assays and their standardization; Clemens, M.J., Morris, A.G., Gearing, A.J.H. (Eds.), Oxford: IRL Press, 1987, pp. 129147. [50] Adams, A., Strander, H., Cantell, K., J Gen virol, 1975, 28, 207-217. [51] Nederman, T., Karlstrom, E., Sjodin, L., Biologicals, 1990, 18, 29-34. [52] Hermodsson, S., Strannegard, 0.J., Jeansson, S., Proc Soc Exp Biol Med, 1984, 175, 44. [53] Gibson, U. E.M., Kramer, S. M., J lmmunol Methods, 1989, 125, 105-113. [54] Meager, A., J Immunol Methods, 1996, 190, 235-244. [55] Deblandre, G., Marinx, O., Evans, S., et al., J Biol Chem, 1995, 270, 23860. [56] Daubener, W., Wanagat, N., Pilz, K., et al., J Immunol Methods, 1994, 168, 39-37.
274
9 Biological Standardization of Interferons and Other Cytokines
[57] Ronni, T., Melen, K., Malygin, A. et al., J Immunol, 1993, 150, 1715-1726. [58] Canosi, U., Mascia, M., Gazza, L. et al., J Immunol Methods, 1996, 195, 55-61. [59] Mire-Sluis, A. R., Page, L. A., Meager, A. et al., J Immunol Methods, 1996, 195, 55-61. [60] Gaines Das, R. E., Meager, A,, Biologicals, 1995, 23, 285-297. [61] Memorandum, Bulletin WHO,1978, 56, 229-240. [62] International Symposium on Standardization of Inteiferon and Inte feron Inducers: Basel, New York: Karger, 1969, p. 328. [63] Interferon Standards: a memorandum, J Biol Standardization, 1979, 7, 383-395. [64] WHO Report on the Standardization of Interferons. WHO Tech Rep Ser; 1983, 687, (Annex I), 35-60. [65] WHO Report on the Standardization of Interferons. WHO Tech Rep Ser; 1985, 725, 28-64. [66] WHO Report on the Standardization of Interferons. WHO Tech Rep Ser; 1988, 771, (Annex l), 37-87. [67] WHO Expert Committee on Biological Standardization 4Sh Report. WHO Tech Rep Series, 1995, 858, 6, 18. [68] Paty, D. W., Li, D. K. B., UBC MS MRI Study Group, et al., Neurology, 1993,43, 662-667. [69] Mire-Sluis, A., Gaines Das, R., Zoon, K. et al., J Interferon Cytokine Res, 1996, 16, 637643. [70] Meager, A., Gaines Das, R. E. J., J Immunol Methods, 1994, 170, 1-13. [71] Mire-Sluis, A., Gaines-Das, R., Thorpe, R., J Immunol Methods, 1996, 194, 1-12. [72] Mire-Sluis, A. R., Gaines-Das, R., Thorpe, R. et al., J Immunol Methods, 1995, 179, 117126.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
10 The Strategic Role of Assays in Process Development: A Case Study of MatrixAssisted Laser Desorption Ionization Mass Spectroscopy as a Tool for Biopharmaceutical Development T. J. Meyers, P. G. Varley, A. Binieda, J. A., Purvis and N. R. Burns
10.1 Introduction Without a robust, efficient production process the development of any protein as a pharmaceutical product is not feasible. To achieve such a process it is obviously necessary to be able to monitor the quality, quantity, and consistency of the product. Similarly it is widely appreciated that the finished product attributes of purity, potency, identity, and stability comprise the backbone of any specification. The measurement of these attributes requires appropriate assays. Perhaps what is less well appreciated is that the development of the assays that will be applied to the specification of a licensed, finished product is intimately linked to the development of the production process itself. Both process and assay development go through series of inter-dependent iterative cycles of development and that the quality of any manufacturing process and its resultant product is fundamentally linked to the quality of the analytical support during its development. The only certain feature of the purification process as it moves from research through clinical evaluation to full-scale, market production is that it will change. Without the analytical tools to provide steering through the effects of these process changes, there is little hope of successfully developing a product. It has often been said of biologicals that ‘the process is the product’, by which it is meant that the manufacturing process defines the physical attributes of the product; if this is so, then it is equally true that the process can be equally well defined as the analytical scrutiny to which it has been subjected. In this paper we illustrate these points through data on the development, validation, and application of a single assay, Matrix-assisted laser desorption ionization-time of flight spectroscopy (MALDI-TOF), in the context of the process development for a specific protein. The complexity of protein-based pharmaceuticals is such that no single analytical method is capable of describing all the quality attributes of a product of this class. This has resulted in the quality of these products being controlled by a combination of very tight regulations,on the manufacturing process and the establishment of an assay ‘portfolio’ in which a battery of assays are used to describe the quality of the final product. However, recent advances in analytical methodologies are moving the emphasis away from the regulation of the manufacturing process towards final
276
I0 The Stratenic Role of Assays in Process Development
product analysis [l]. Despite this shift in emphasis there are two main reasons why a well-characterized production process is required for protein pharmaceutical development. First, no process will deliver a truly homogeneous product as a variety of post-translational modifications introduce microheterogeneity. Second, as a product moves through clinical evaluation changes to the process will occur which will inevitably lead to changes in the composition of the product. This molecular variation may be reflected in the pharmacokinetics, toxicity, or clinical efficacy of the product. Therefore, not only is there a need to employ the increasingly sophisticated analytical methods to the final bulk and finished product, but also in process development and process characterization. At the forefront of the improvement in analytical methods has been mass spectroscopy, particularly electrospray (ES-MS) [2] and matrix-assisted laser desorptiod ionization time-of-flight mass spectroscopy (MALDI) [3]. Since the molecular mass of a protein is dictated by its amino acid composition, the accurate determination of molecular mass can act as a valuable confirmatory identity test. ES-MS is a higher-resolution method than MALDI (precision figures range from 0.01-0.005 % and 0.1-0.05 %, respectively) [4,5]. However, MALDI has a number of advantages over ES-MS : notably its high sensitivity, relative insensitivity to buffer components in the sample; the tendency not to be subject to compound specific suppression effects; the ease of sample preparation and analysis; and the relatively low cost of the instrument [6,7]. These factors all contribute to making MALDI particularly useful for process development and identity testing. This paper describes a formal validation study on MALDI and the application of this technique in the context of the process development and quality control of a recombinant protein intended for clinical use. The protein in question is BB10010, which is an engineered variant of Macrophage Inflammatory Protein-la (MIP-1 a). BB-10010 differs from wild-type MIP-1 a by a single amino acid which dramatically alters its association equilibrium, thereby rendering it soluble in physiological buffers [8]. BB-10010 is currently being clinically evaluated as an adjunct to chemotherapy in cancer. More recently the discovery that MIP-1 a prevents HIV proliferation [9] has heightened expectations for the molecule.
10.2 Materials and Methods Each of the variant proteins were purified from culture supernatant derived from recombinant yeast strains containing vectors with genes that encode a 69-amino acid protein based upon MIP-1 a [8]. The yeast were cultured and expression induced and the variants prepared as described previously [10,11]. The fidelity and homogeneity of the BB-10010 was confirmed independently by ES-MS, N-terminal sequencing, SDS-PAGE and reversed phase HPLC. All samples were stored in phosphate-buffered saline at a concentration of 2 mg mL-' prior to use. The molecular weights of all samples were determined by direct calibration with bovine insulin as the internal calibrant, molecular weight 5733.6 Da (Sigma Chemical Co., Poole, Dorset, U.K.). All solvents or other chemicals used in this work were of HPLC
10.2 Materials and Methods
277
Table 10-1. Summary of molecular weight data for analysis of protein BB-10010 and its variants. Parameter
Experimental mean (Da) Standard deviation (Da) No. of observations
cv (a) a
BB- 10010 (7668.6 Da)a
Variant A (7712.6 Da)=
Variant B (7685.5 Da)a
All data
Selected datab
All data
Selected datah
All data
Selected datah
7669.3 7.04 48 0.09
7668.5 5.55 39 0.07
7715.4 6.99 48 0.09
7713.8 4.4 43 0.06
7691.7 14.7 48 0.19
7678.0 4.78 39 0.06
Theoretical molecular weight. Contains only data selected using pre-defined criteria that the peak width at half-height is < 1.5%.
grade (BDH, Poole, Dorset, U.K.). Table 10-1 details the MIP-1 a variants used for the validation. MALDI was performed on a LaserMAT 2000 mass spectrometer (Thermo BioAnalysis, Heme1 Hempstead, Hertfordshire, U.K.) [ 121. All samples were analysed in 0.5 pL of a-cyano, 4-hydroxycinnamic acid matrix solution concentration 10 mg mL in 70 % acetonitrile, 30 % water, 0.1 % trifluoroacetic acid, and 2 pmol of bovine insulin internal calibrant. Purified proteins were analyzed at concentrations of 0.2 and 0.02 mg mL-' prepared by dilution of a 2 mg mL-' protein stock solution with water. A 0.5-pL aligout of each protein sample was applied to each sample slide, giving loadings of 13 pmol and 1.3 pmol, respectively.
10.2.1 Validation Study Parameters studied for validation were specificity and precision (repeatability). Specificity was validated by demonstrating the ability of MALDI to measure and distinguish accurately between the molecular weights of BB-10010 and single amino acid variants of BB-10010 (shown in Table 10-1). The repeatability was examined by assaying BB-10010 on 12 slides with bovine insulin as the internal calibrant. All four target sites on each slide, six containing 13 pmol of BB-10010 and six containing 1.3 pmol of BB-10010, were analyzed. A spectrum was deemed acceptable if the width of the peak at half-peak-height was < 1.5 % of the protein's molecular weight.
278
10 The Strategic Role of Assays in Process Development
10.2.2 Statistical Analysis SAS Version 6.08 was used to perform all the statistical analysis. A Wilcoxon rank sum test was performed between the mean molecular weight for the BB-10010 and each of the BB-10010 variants. From the Wilcoxon Rank Sum test, a significant difference at the 95 % confidence level between the BB-10010 and each BB-10010 variant was applied to show assay specificity.
10.2.3 In-process Analysis MALDI has been used to characterize and develop the BB-10010 production process. Samples from fermentation harvests and expanded bed adsorption chromatography (EBA) were 0.2 pm filtered to remove any cell debris that was present. The fermentation harvest and in-process samples were diluted with water, where required, to allow a minimum of -1.3 pmol to be loaded in 0.5 pL. Waste fractions, where necessary were concentrated using C18 Sep Pak cartridges (Waters, Milford, MA) by eluting the sample in 90 % acetonitrile, 10 % water, and 0.1 % trifluoroacetic acid and drying down the eluate by vacuum centrifugation.
10.3 Results 10.3.1 Identity Test - Validation Figure 10-1 shows typical spectra obtained for BB-10010 and the two BB-10010 variants. The results are summarized in Table 10-1. Data for each protein are presented in two columns, the first column summarizes all the data. The second column summarizes the results from the data that meet the pre-defined criteria that the width of the sample peak at half-weight is < 1.5 % of the protein’s molecular weight (115 Da for BB-10010). This is a specific, pre-defined rule which we have found to be useful in ensuring that an occasional poor-quality spectrum does not impact upon identity testing of purified BB-10010, and was included in the validation protocol. The advantage of applying this criterion to our data is illustrated in the study results. Some spectra are not included in the final analysis, resulting in an improvement in both the precision and the accuracy of the molecular weight determinations. The basis of the validation of the identity test is to demonstrate the ability of the technique to distinguish between BB-10010 and very closely related variants. None of the mean experimental molecular weights differed from the theoretical molecular weight by more than 3.4 Da. In the case of BB-10010, the mean experimental molecular weight was 7668.5 Da, a mere 0.1 Da difference from the theoretical molecular weight. The samples all had a coefficient of variation of less than 0.10 %. A Wil-
10.3 Results
279
7671 .O I
;;u ,
ic
5734.6
15 14
5500
6000
6500
7000
7500 8000 8500 9000 9500
a)
Mass ( d z )
16.0 15.5 15.0
1
7687.2
I
Width at M Peak Height = 57.9h
6
12.5 12.0
5500
6000
6500
7000 7500 8000 8500 9000 9500
b)
W Z )
14.2 14.0
77 3.9
n
BovineInsulin
57q4.6
13.8 13.6
.f ;::;
E
42
13.0 12.8 12.6 12.4 12.2 12.0 11.8 5500
6000
6500
7000 7500 8000 8500 9000 9500
h
r(dz)
c)
Fig. 10-1. Typical MALDI mass spectra for each of the proteins used in the validation study: (A) BB-10010, (B) Variant A, and (C) Variant B.
280
10 The Strategic Role
of
Assays in Process Development
Table 10-2. Effect of sample loading upon mass determination of BB-10010 protein and its variants. Loading
Experimental mean (Da) Standard deviation (Da) No. of observations
cv (%) a
BB-100 10 (7668.6 Da)a
Variant A (7712.6 D a y
Variant B (7685.5 Da)a
13 pmol
1.3 pmol
13 pmol
1.3 pmol
13 pmol
1.3 pmol
7668.5 5.40 17 0.07
7668.6 5.79 22 0.08
7714.5 4.84 20 0.06
7713.1 3.97 23 0.05
7687.7 5.58 20 0.07
7686.3 3.79 19 0.05
Theoretical molecular weight.
coxon rank sum test was used to compare mean experimental calibrated results of BB-10010 against each of the BB-10010 variants. The P value was 0.05 in each case. This indicates that the molecular weights of BB-10010 and its variants can easily be distinguished one from another. This is illustrated by the calculation of the 95 % confidence interval for the BB-10010 data. Using the number of determinations as n = 10, this was calculated as the mean molecular weight t 3.97 Da. This confidence interval implies that we can be 95 % confident that the population mean for BB-10010 is between the values 7664.5 to 7672.5 Da, when 10 measurements are made. On this basis we routinely perform 10 measurements at five different concentrations across the working range of the spectrometer (for this protein) on a single sample in a formal BB-10010 identity test. Table 10-2 compares the results from 1.3 pmol and 13 pmol loadings of BB-10010 and the two BB-10010 variants. These amounts approximate the extremes of the working range for the spectrometer for this particular protein. There are no significant differences in the consistency or absolute values of the molecular weights determined at each of the sample loadings. The molecular weight determination is therefore independent of sample loading across the working range of the instrument.
10.3.2 Applications In-Process A key advantage of the MALDI technique is its relative robustness in the presence of buffer salts and other contaminants. The technique is therefore able to be used to analyze proteins during their production without excessive sample preparation which may in itself affect the analysis. Figure 10-2 shows a spectrum of a typical fermentation harvest from a BB-10010 pilot-scale production run. Table 10-3 shows data concerning the repeatability of such measurements. The fermentation harvests had an overall mean experimental molecular weight of 7668.8 Da, differing from the theoretical molecular weight by 0.2 Da. The Wilcoxon rank sum test compared the mean experimental calibrated results for the BB-10010 used in the study against the overall mean for the fermentation harvests. The P value was 0.762.
10.3 Results
281
Table 10-3. Summary of the analysis of BB-10010 fermentation harvests using the original expression vector. Parameter
Data from eight production runs
Experimental mean (Da) Standard deviation (Da) No. of observations
7668.8 2.96 41 0.04
cv (%)
The samples had an overall coefficient of variation of 0.1 %. The mean and coefficient of variation were similar to the data obtained for BB-10010 during the validation study. Subsequent to the validation of MALDI as an identity test, the technique was applied to the development and characterization of the BB-10010 production process. This involved the characterization and optimization of the process to ensure the consistent production of optimal amounts of high-quality material. An understanding of the process on the molecular level also allows the impact of changes to the process upon product quality to be assessed. This information can then be used to ensure that any equivalence issues are minimized. One such change required to facilitate scale-up during BB-10010 development has been the use of an alternative expression vector. Figures 10 -2 and 10-3 show typical fermentation harvests from the production of BB-10010 using the different expression vectors. Two additional peaks of larger mass than BB-10010 can be seen in the spectrum in Fig. 10-3. These peaks were not observed with the original expression vector (Fig. 10-2). MALDI was used to confirm the clearance of these impurities from the BB-10010
Fig. 10-2. Typical MALDI spectrum of a fermentation harvest from the BB-10010 production process. This material was produced from the original expression vector.
282
10 The Strategic Role of Assays in Process Development
23.
7y9.7 IntemaICalibrant Bovine Insulin
22.
21.
1
5734.6 I
20.
-1:::
4 17. 9 .
g:. 14. 13.
Internal Calibrant Bovine Insulin 573f.6
I 5500
m o
6500
7wo
7500
Mass ( d z )
8000
a5m
cam
ssoo
b)
Fig. 10-3. MALDI mass spectra from the EBA stage of the BB-10010 production proces using the new expression vector. (A) Column load (fermentation broth). (B) column flow-through.
10.3 Results
20.
Bovine Insulin 573f.6
19. 18.
c
d 16.
c (
I ::: P
7675.8
7669.6 I
Internal Calibrant Bovine Insulin 5734.6 I
7512.4
Fig. 10-3. (continued). (C) column wash fraction. (D) BB-10010 product eluate.
283
284
10 The Strategic Role of Assays in Process Development
product stream (Fig. 10-3). Although the low quantities involved precluded an accurate mass-balance, the presence - and indeed removal - of these impurities was confirmed by examining the discarded fractions. This results in a product eluate from both processes which are indistinguishable. The MALDI spectrum of the EBA eluate in Fig. 10-3 has a main peak that corresponds to intact BB-10010 and small peaks that correspond in mass to N- or C-terminally degraded protein. This indicates that contaminating exo-proteases are present in the product stream. Figure 10-4 shows how MALDI was used to establish that these proteases are separated from the product during the course of the EBA chromatography. After incubation of the column load and wash fractions at 25°C for 10
18.
Internal Calibrant BavineInsulln
&'7.
I
573;4.6
l91
7670.9
I
Fig. 10-4. MALDI mass spectra from the EBA stage of the BB-10010 production process. (A) column load material; (B) column wash fraction.
10.3 Results
285
766;1.8
20 1
5500
6ooo
6500
7000
7500
8000
85w
Qmp
9500
Mass ( d z )
4
Fig. 10-4. (continued). (C) BB-10010 product eluate.
Internal Cahbrant Bovine Insulin 5734.6
75113.9 7399.81
I
I 6000
6500
7wO
7500
(dz)
8000
8500
9oM
9500
a)
Fig. 10-5. MALDI mass spectra of the samples shown in Fig. 10-4 following a 10-day incubation at 25 "C. (A) column load material.
days the formation of degraded BB-10010 demonstrates that proteases are present in these fractions. The degradation is most apparent in the column wash which no longer contains intact BB-10010. The largest peak in this fraction corresponds to the loss of three residues from the termini (N or C ) of the molecule. This indicates that the bulk of the contaminating proteases are contained in this fraction. Interest-
286
10 The Strategic Role of Assays in Process Development
7
7669.2
23
I
22.
I
21.
*b IS' c-(
!.
$!
18. 17.
16' 15.
Internal Calibrant Bovine Insulin 5734.6 I
7512.0
Fig. 10-5. (continued). (B) column wash fraction; (C) BB-10010 product eluate.
ingly, in both the flow-through and wash fractions, the larger contaminating peak (- 8560 Da) remains unchanged, indicating that the proteases do not affect this impurity. In the BB-100 10 product eluate there is no significant difference between the levels of degradation present before (Fig. 10-4) and after the 10-day incubation period (Fig. 10-5). The results therefore demonstrate the removal of proteases from the product stream by this stage of the process. The application of MALDI in BB-10010 process development is further illustrated by the spectra in Fig. 10-6. This shows analysis of samples from the hydrophobic interaction chromatography (HIC) stage of the production process. The spectrum
10.3 Results
40
287
7671.1
1
I
Internal Calibrant Bovine Insulin
5734.6 I
7512.3 7399.2I] 5500
6wo
65a)
7m0
'\
7500
BDOO
8500
gDoD
gglo
Mass ( d z )
766?.4
Fig. 10-6. MALDI mass spectra across the HIC stage of the BB-10010 production process. (A) column loading; (B) BB-10010 product fraction.
of the product eluate contains a single peak with a molecular weight that corresponds to intact BB-10010. Degraded or truncated BB-10010 which is present in the column load is eluted in the waste fraction. This illustrates how MALDI has been used to design and optimize the HIC stage of the process in order to separate full-length BB-10010 from any BE-10010 degradation products.
288 32
10 The Strategic Role of Assays in Process Development
I
1
30 .
28.
p4. 26.
A
22.
!.
20.
4
Intend Calibrant Bwire Insulin
5734.6
Fig. 10-6. (continued). (C) column waste fraction.
10.4 Discussion Mass spectroscopy has been at the forefront of recent progress in the analysis and characterization of proteins as pharmaceutical products. As these techniques have become more routine and increasingly accessible, their application in the field of analysis of protein-based pharmaceuticals has focused mainly upon the characterization and understanding of final bulk or finished product. This is illustrated here by the validation study which demonstrates the ability of MALDI to be used as a simple, but effective, identity test for BB-10010. The primary application of the technique is, however analysis of the BB-10010 production process. Here, the ability of the technique directly to analyze the product in crude mixtures, such as fermentation harvests, is utilized to make MALDI an effective tool for development and analysis of the production process. The data obtained from the validation study show the inherent assay variability which defines the accuracy of the technique. The results demonstrate that when 10 spectra are acquired with an average molecular weight of within 4 Da of 7668.6 Daltons, we can be greater than 95 % confident that the analyte has the molecular weight expected for BB-10010. This is the basis of a powerful identity test. During this work we have considered the limitations of the technique. For example, it is possible theoretically to have a completely unrelated compound with a similar molecular weight. Alternatively, the protein could be altered by either amino acid substitutions (in itself unlikely), or chemical modifications that result in a mass change of < 4 Da. Any such change would result in the incorrect identification of an analyte as BB-10010. However, most amino acid mutations or chemical modifications would result in a mass change of > 4 Da. For example, only 7.9 % of all the
10.4 Discussion
289
possible single amino acid mutations in BB-10010 would result in a mass change of less than 4 Da (analysis not shown). The combined probability of a change occurring and resulting in a mass change of < 4 Da is therefore very low. For this reason we believe that the MALDI identity test is a valuable and worthwhile addition to the portfolio of assays used to characterize BB-10010. The probability of incorrect identification could be reduced by using electrospray ionization mass spectroscopy or even tandem mass spectroscopy techniques. These techniques do not however display all the unique advantages of MALDI, namely a combination of sensitivity, high tolerance to buffers and salts, and speed and ease of analysis. For example, the use of electrospray would require the protein to be subjected to a desalting step to remove buffer contaminants prior to analysis. Moreover, the exclusive use of tandem mass spectroscopy would prohibit rapid analysis and data interpretation. This would cause problems with effective real-time analysis of BB-10010 production process intermediates, which was the primary goal of this work. The data obtained from the fermentation harvest are comparable in quality with those from purified material from the validation study. The BB-10010 fermentation harvest contains more contaminants than at any other point during the production process. The results therefore demonstrate how MALDI can provide accurate molecular weight determinations, despite the presence of contaminating buffers, salt, and other small molecules. This is the key advantage of MALDI over other types of mass spectroscopy. This work was subsequently extended to characterize and optimize each stage of the BB-10010 production process using MALDI as the principal analytical tool. The technique has been applied to monitor routinely all stages of the production process to ensure that it is functioning correctly and is under control. The turn-around time of typically 10 minutes, and speed and ease of data interpretation also contribute to make MALDI an extremely convenient and effective in-process monitoring tool. This work demonstrates how MALDI has been an important factor in determining the quality and consistency of the BB-10010 product and illustrates how product quality relates to, and is best addressed by, process development. The characterization of the final product, although becoming more important as analytical methodologies improve, does not make the detailed analysis of the production process redundant. In fact, it could be argued that the reverse is true. As ever more sensitive methods of analysis are used, any shortcomings in product quality will be more readily identified which in turn will require that more detailed in-process analysis is carried out. MALDI, with its ability to perform fast accurate analysis in relatively crude mixtures, will be at the forefront of this analytical strategy.
290
10 The Strategic Role of Assays in Process Development
References [l] Henry, C., Anal Chern, 1996, 68, 674A-677A. [2] Fenn, J. B., Mann, M., Meng, C. K., Wong, S.F., Whitehouse, C. M., Science, 1989, 246, 64-71. [3] Karas, M., Bhar, U., Geisseman, U., Mass Spectrom Rev, 1991, 10, 335-358. (41 Nguyen, D.N., Becker, G. W., Riggin, R. M., J Chromatogr A, 1995, 705, 21-45. [ 5 ] Geisow, M. J., Trends Biotechnol, 1992, 10, 432-441. [6] Stults, J. T., Curr Opin Struct B i d , 1995, 5, 691-698. [7] Kaufmann, R., J Biotechnol, 1995, 41, 155-175. [8] Hunter, M. G., Bawden, L., Brotherton, D. et al., Boold, 1995, 86, 4400-4408. [9] Cocchi, F., DeVico, A. L., Garzino-Demo, A,, Arya, S. K., Gallo, R. C., Lusso, P., Science, 1995, 270, 1811-1815. [lo] Clements, J.M., Craig, S., Gearing, A. J. H., Hunter, M.G., Heyworth, C. M., Dexter, T. M., Lord, B. I., Cytokine, 1995, 4, 76-82. [ l l ] Patel, S. R., Evans, S., Dunne, K., Knight, G. C., Morgan, P. J., Varley, P. G., Craig, S . , Biochemistry, 1993, 32, 5466-5471. [12] Mock, K. K., Sutton, C. W., Cottrell, J. S., Rapid Comrnun Mass Spectrometry, 1992, 6, 233238.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
11 Quality Control of Protein Primary Structure by Automated Sequencing and Mass Spectrometry Philip J. Jackson and Stephen J. Bayne
11.1 Introduction Protein primary structure can be defined in two ways. The simpler definition refers to the linear sequence of amino acid residues in a protein, as specified by the gene sequence which codes for it. This definition assumes that each position in the sequence is occupied by one of the 20 amino acid residues specified by the genetic code and is used to provide a unique description of a particular protein in the databases. The more complex definition of primary structure includes all of the covalent bonds within a protein, taking into consideration co- and post-translational modifications such as glycosylation, methylation, and disulfide bond formation. Excluded in both definitions are noncovalent interactions which, in addition to covalent bonding, govern the three-dimensional structure of a protein. This chapter deals with the analysis of protein primary structure at both levels, focusing on applications in the quality control of peptides and proteins, whether synthetic, native, or recombinant, for use in therapeutics, antibody production, and crystallography, for example. Linear sequence determination may be sufficient to verify a protein’s identity, characterize contaminants, or detect proteolytic degradation. However, since it may be structurally or functionally crucial for the desired product to have undergone accurate bioprocessing or synthetic modification, more complex characterization strategies, involving both chemical and mass analysis, are also covered.
11.2 Automated Edman Degradation The idea of analyzing a protein by the sequential chemical cleavage of amino acid residues from the N-terminus was first published in 1930 (for a historical review, see [l]). The major breakthrough in the chemistry occurred in 1950 when Pehr Edman (1916-1977), then of the University of Lund in Sweden, published a manual chemical process, popularly referred to as Edman degradation [ 2 ] . This chemistry has remained largely unaltered over the past 48 years, except for automation, adaptations to the original reagent formulations, and increased sensitivity [3].
292
11 Quality Control of Protein Primary Structure
At present, there are a number of different types of N-terminal sequencing instruments in laboratories around the world, supplied by several manufacturers. As would be expected, there are differences between these instruments, relating to hardware, reagent formulations employed, and means of sample presentation. However, as far as chemistry, operating principles, and limitations are concerned, all of the instruments currently in use (some are 14 years old) are similar enough to allow coverage in a single generic description.
11.2.1 Chemistry Before sequencing is started, the protein sample must be immobilized on to an inert support material, which is then placed inside a heated reaction cartridge on the instrument. Depending on the sample type or instrument, the target protein may be adsorbed on to a glass-fiber filter [4], a proprietary membrane disk [ 5 ] , a biphasic sorbant column [ 6 ] , or a poly(viny1idene difluoride) (PVDF) membrane [7,8]. In all of these cases the sample is retained on the support by noncovalent interactions. For such samples the instrument must incorporate chemicals and delivery regimes which are specifically designed not to cause solubilization and subsequent loss from the reaction cartridge by elution. Sample retention, particularly on glass-fiber supports, can be augmented by the addition of a polymeric quaternary amine carrier known as Polybrene [4]. It is also possible to immobilize proteins covalently (via their amino groups) to a PVDF-type membrane functionalized with isothiocyanate or via carboxyl groups after carbodiimide activation, to an amino-functionalized membrane (see Section 11.2.2). In essence, all automated protein sequencers (also referred to as sequenators) consist of three subsystems as summarized in Fig. 11-1. The first is the reaction cartridge
Fig. 11-1. The three subsystems of a modern, automated protein sequencer. Samples are placed in the reaction cartridges where they are subjected, one at a time, to the coupling and cleavage steps of Edman degradation. After cleavage, residues (as ATZ derivatives) are transferred to the conversion flask and subsequently injected (as FTH derivatives) on to a reverse-phase HPLC system for identification and quantitation.
11.2 Automated Edman Depradation
293
which typically operates in the temperature range 45-55 "C and holds the protein sample. Its construction is instrument-specific but mainly designed according to the particular means of sample presentation. A single instrument may in fact possess more than one reaction cartridge (the maximum at present is four) to increase sample throughput by allowing additional samples to be loaded and sequenced automatically overnight or over a weekend. The first chemical step in protein sequencing (Fig. 11-2) is termed coupling and occurs with the entry of both the Edman reagent, phenylisothiocyanate (PITC), and the coupling base into the reaction cartridge. The PITC is stored and delivered as a 0.5-10 % (v/v) solution in n-heptane or acetonitrile. Typically, only a few microliters are delivered at a time, just sufficient to wet the sample support. The solvent is then driven away by a stream of argon or nitrogen, leaving the PITC, which is volatile only under reduced pressure, in contact with the sample. The coupling base, a tertiary amine such as triethylamine, diisopropylethylamine, or N-methylpiperidine
1: Coupling
amine
U
2: Cleavage
3: Conversion
1 I
9
F3C-C-OH)
ATZ
7
-N-
Fig. 11-2. Edman degradation chemistry.
TFA
294
11 Quality Control of Protein Primary Structure
is stored in aqueous solution, sometimes with methanol, ethanol, or propanol as cosolvent (several formulations exist) but delivered in some instruments as a vapor over a period of 3-8 minutes. This delivery method is, as mentioned above, to minimize sample wash-out by what would be an effective protein solvent. Liquid base delivery is used in instruments where either the applied volume is precisely metered so that no flowthrough occurs or the sample is covalently immobilized. One function of the coupling base is to buffer the environment within the reaction cartridge at high pH (9-10) to promote the deprotonation of the protein’s terminal aamino group so that it can react, by nucleophilic addition, with the PITC. The other function is catalytic in that a PITC adduct can also form and subsequently react more rapidly with the terminal amino group by nucleophilic substitution, regenerating the coupling base in the process. Evidence for this mechanism is provided by the observation that coupling efficiency in Edman degradation increases with the leavinggroup potential of the tertiary amine [9]. In a typical sequencing cycle, the PITC and coupling base deliveries are repeated two or three more times to drive the reaction as near to completion as possible, then all remaining base is blown from the reaction cartridge by argon or nitrogen. The vast excess of PITC, together with any reaction by-products, are extracted by the delivery of nonpolar solvents such as ethyl acetate, n-heptane, l-chlorobutane, either singly or in combination, or acetonitrile/toluene mixtures. Again, the rationale is to maximize the extraction of everything except the sample, which is now derivatized at its N-terminus with a phenylthiocarbamyl (PTC) group. The next step to occur in the reaction cartridge is known as ‘cleavage’ whereby the N-terminal residue side chain forms part of a cyclic derivative after reaction with anhydrous trifluoroacetic acid (TFA) which is delivered either as vapor in a stream of argon or nitrogen, or as a precisely measured pulse of liquid. When vapor delivery is employed, the sequencing chemistry is referred to as ‘gas phase’ and the optimum duration of this step is 10-20 minutes. The ability to deliver liquid TFA allows this step to be speeded up to 5-10 minutes; however, the delivery volume must be accurately controlled to prevent sample wash-out. The cleavage step is complete when the derivative, an anilinothiazolinone (ATZ) undergoes a rearrangement and detaches from the rest of the polypeptide chain, leaving a new N-terminus. In the case of proline, the cleavage reaction is relatively slow and leads to the carry-over of a large proportion (about 30%) of the signal from this particular residue into the next cycle. This signal lag accumulates through subsequent cycles, essentially increasing the background and curtailing the potential length of sequence read. Therefore, if the protein sequence is already known, it is usual to incorporate doubled cleavage times into the appropriate cycles to enhance sequencing efficiency for proline-containing samples. Extended cleavage steps are not employed at every cycle because this would increase the total exposure time of the sample to TFA, leading to a rise in unwanted side reactions (see Section 11.2.3). After cleavage, the ATZ derivative, which possesses the original amino acid residue side chain, is extracted with nonpolar solvents, for example, ethyl acetate, 1chlorobutane, or acetonitrile/toluene and transferred to the conversion flask, the second subsystem of a protein sequencer. The nature of the side chain, however, modifies the extractability of the ATZ such that no single solvent is optimal for all resi-
11.2 Automated Edman Degradation
295
dues. In addition, the sample support medium exerts an influence with the consequence that certain residue signals may be slightly depressed and carried-over. This type of lag, seen more commonly with histidine and arginine, does not accumulate through subsequent cycles, in contrast to proline-associated lag. At this point, the new N-terminus is ready for the next round of coupling and cleavage. Meanwhile, in the conversion flask at 60-70°C the ATZ solution is blown down to a few microliters, then reconstituted in 25% (v/v) aqueous TFA, which drives a rearrangement to the more stable phenylthiohydantoin (PTH) derivative. Despite the overall stabilization, side reactions which essentially lead to signal reduction can occur. For this reason a scavenger, dithiothreitol at 0.01 % (w/v) is included in the conversion reagent. This additive has a positive effect on the final recoveries of all PTH derivatives, and is most apparent in the case of lysine. After conversion, which takes 10-20 minutes, the aqueous TFA is evaporated from the conversion flask to leave the PTH ready for reconstitution in a suitable solvent, for example 10-20 % (v/v) aqueous acetonitrile, for injection on to a high-performance liquid chromatograph (HPLC), the third protein sequencer subsystem. Like the ATZ, the PTH retains the original amino acid residue side chain and this determines its elution time from a monofiinctional C1g reverse-phase column. The mobile phase consists of an aqueous/organic solvent gradient system, with the PTH derivatives detected and quantitiated spectrophotometrically at 269 nm (Fig. 11-3). Each degradation cycle generates a separate chromatogram, therefore a polypeptide sequence is read by interpreting successive chromatograms as shown in Fig. 11-4.
4
10 -Retention Time in minutes
20
Fig. 11-3. Reverse-phase HPLC of PTH amino acid derivatives. Peaks are identified by the standard one-letter code for amino acids. PTH-Cys is missing because cysteine is only identifiable during sequencing after sample alkylation. In the chromatography system shown here, PTH-pyridylethyl-S-cysteine for example, elutes between PTH-Tyr and -Pro. PTH-Lys elutes at the hydrophobic end of the chromatogram because it is &-PTC-derivatizedby the Edman reagent and therefore carries an additional phenyl group.
296
11 Quality Control of Protein Primary Structure
II
I
Fig. 11-4. Sequence data for the first four residues of bovine o-lactoglobulin. These are Leu, Ile, Val, and Thr.
11.2.2 Sample Preparation 11.2.2.1 Compatibility with Edman Chemistry Once an appreciation of the chemistry of Edman degradation has been gained, the necessary criteria for protein or peptide sample preparation are easily rationalized. While it is crucial that the sample should not have undergone any deleterious chemical changes during production, media which are formulated to preserve conformational stability or biological activity are not necessarily compatible with sequence analysis by Edman degradation. One particular type of chemical modification renders a sample resistant to Edman degradation. As described in the previous section, after reactions involving the Edman reagent PITC and the coupling base, the N-terminal a-amino group becomes derivatized. Such derivatization is only possible by virtue of the amino group’s nucleophilicity. If this property is eliminated by prior derivatization, then Edman degradation cannot begin. In other words, the sample is N-terminally blocked. A large proportion, probably more than 50 % according to current experience [lo], of eukaryotic proteins are naturally blocked by co- or post-translational processing.
11.2 Automated Edman Degradation
297
If the blocked N-terminus is known to be the cyclic, glutamine-derived residue pyrrolidone carboxylic acid (pyroglutamic acid), de-blocking prior to sequencing can be achieved routinely by incubation with pyroglutamate aminopeptidase ( 5 -oxoprolyl peptidase, EC 3.4.19.3) which allows Edman degradation from residue 2 onwards [lo]. Unfortunately, for other commonly encountered N-modifications, for example formylation, acetylation, myristoylation, and methylation, there are no reliable, specific de-blocking protocols. An acylpeptide hydrolase (N-acylamino-acid releasing enzyme, EC 3.4.19.1) [ l l ] can be employed for the removal of a-N-acylated N-terminal residues, as the name suggests. Unfortunately, there are known to be problems with the stability of the enzyme and its low activity on polypeptides larger than 10-15 residues [ 111. Chemical methods using, for example, trifluoroacetic acid vapor [12] or concentrated hydrochloric acid in methanol [lo] carry the risk of sample fragmentation by nonspecific peptide bond cleavage. Sequencing of N-terminally blocked samples requires controlled proteolysis and subsequent isolation of the peptide fragments (Section 11-5). Even then, the N-terminal fragment is still refractory to Edman chemistry but may be verified by mass spectrometry (Section 11-4). Samples which are not naturally blocked at the N-terminus should not be exposed to chemicals or conditions which might lead to blockage. Amino-reactive compounds such as aldehydes, carboxylic acid anhydrides, active esters, and cyanates should not be present at any stage of sample preparation. Furthermore, components of buffers, solvents, etc. which are liable to contain amino-reactive contaminants should be used at sufficient levels of purity. For instance, it is common practice to pre-treat urea solutions with a mixed-bed ion exchange resin, e.g., Amberlite MB1, to remove ammonium cyanate. However, as cyanate is continually formed from urea in aqueous solution, sequencing efficiency of samples stored in the presence of urea may be compromised. Artefactual N-terminal blockage problems can often be managed by keeping protein concentrations as high as possible, or processing and storage at low temperature. Maintaining the pH well below the pK, of the a-amino group (i.e., below 9), to reduce its nucleophilic potential by protonation, may also be helpful. However, one should be aware that in this environment the protein may be subject to other problems like denaturation, aggregation, and precipitation. For the coupling reaction of Edman degradation, N-terminal protonation has to be inhibited by maintaining a high pH. In this situation, the terminal amino group is particularly susceptible to blockage by suitably reactive sample matrix components, especially since the temperature is now also elevated. Sequencing efficiency may similarly be compromised if the ability of the coupling base to deprotonate the N-terminus is inhibited by the presence of buffer compounds of pK, < 9 in the sample matrix. For this reason, matrix components such as acetate, phosphate, citrate, N[2-hydroxyethyl]piperazine-N'-[2-ethanesulfonic acid] (HEPES), 2-[N-morpholino]ethanesulfonic acid (MES), and other commonly used buffers of this type must be removed before sequencing. Fortunately, protein sequencer manufacturers have addressed this problem by providing support materials which allow routine sample desalting during immobilization [6,8]. Desalting is also required to remove PITC-reactive matrix components, e.g., thiols, ammonia, primary and secondary amines (such as Tris), amino acids, and ampho-
298
I1 Quality Control of Protein Primary Structure
lines. Apart from effectively reducing the PITC concentration and therefore the coupling efficiency, these compounds form stable PITC adducts which are extracted with the ATZ derivatives subsequently to appear on the PTH analysis chromatograms. In some cases, the adducts co-elute with particular PTHs, for example the ammonia adduct with PTH-Asp, causing potential difficulty in reading the sequence. The most common experience with these adducts is that they are present in such high concentrations, by protein sequencing standards, that their signals obscure those of the PTHs. It should also be noted that volatile ammonium salts cannot be completely removed by lyophilization, even if this is repeated several times. Tertiary amines for use as the coupling base in Edman degradation are extensively purified: nonsequencing grade products typically contain sufficient contaminating ammonia and primary or secondary amines to cause problems. Other nonvolatile matrix components which require removal prior to sequencing are multivalent metal salts, detergents, guanidinium salts, carbohydrates, and glycerol. These compounds are deleterious to Edman degradation by various mechanisms including coordination with reaction intermediates, promotion of sample wash-out, reaction with amino groups, and inhibition of sample penetration by the reagents. 11.2.2.2 Manipulation of Samples in Solution In terms of final sample application to a sequencing support matrix, with or without desalting as necessary, specific protocols are provided by instrument manufacturers for use by sequencer operators. The aim of this section is to provide guidelines for the manipulation of samples before presentation for sequencing so that unbiased data may be obtained. Sequencing for research purposes frequently involves sample amounts at low- or even sub-picomole levels. In such situations, sample manipulation strategies are designed to maximize recovery by minimizing losses by adsorption on to surfaces. One such strategy is to maintain protein solutions at high concentration which, as stated in the previous section, also reduces the extent of N-terminal blockage by matrix contaminants. However, when samples are concentrated by centrifugation either through an ultrafiltration membrane or in vacuo, it is important to not allow them to dry completely. If this happens, sample recovery tends to be reduced. For quality control of bulk and formulated protein or peptide products, sample amounts tend not to be limiting. However, issues relating to the reconstitution of freeze-dried material and further dilution still need to be addressed. Clearly, to assess product purity by any analytical method it is of paramount importance for the sample to be representative. For this reason the solvent selected to reconstitute a freeze-dried sample should dissolve all polypeptide components completely. The potential problem of sample component bias caused by the use of an inappropriate solvent is illustrated in Fig. 11-5. Furthermore, in a situation which may be more specific to sequencing, the solvent may also have a profound effect on the relative amounts of different polypeptide components within a sample which become immobilized on a sequencing support matrix.
11.2 Automated Edman Degradation
L
Y
I
A
V
E
T
G
L
y
I
A
V
E
T
G
299
Fig. 11-5. Sequence data for an equimolar mixture of bovine 0-lactoglobulin (sequence: Leu, Ile, Val, Thr) and a synthetic decapeptide (sequence: Tyr, Ala, Glu, Gly). Duplicate lyophilized samples were reconstituted in (A) 0.1 % (vlv) trifluoroacetic acid in water and (B) 5 % (vlv) acetic acid, 40 % (v/v) propan-2-01 in water before sequencing. (A) shows a significantly lower recovery of 0-lactoglobulin compared with (B), illustrating the importance of using appropriate solvents to obtain representative sequence data.
Typical quality control sample loadings on to a protein sequencer are in the range 200 pmol to 1 nmol, depending on the anticipated impurity levels. It is generally not worth exceeding 1 nmol because of the effect of background (see Section 11.2.3). Consequently, if a sample requires dilution before sequencing it is worth making a careful choice of both container and solvent, again to eliminate the possibility of component bias. Polystyrene and glass tubes are notorious for polypeptide adsorption which, at low sample concentrations, may become selective. It is therefore preferable to use high-quality polyethylene or polypropylene tubes which, if required, can be purchased with passivated internal surfaces. Preferential adsorption of hydrophobic polypeptides on to tube walls can also occur when the organic solvent concentration in the sample diluent is too low. For this reason, water alone is rarely appropriate as a diluent and detergents are not included for noncovalent sequencing because of potential problems with immobilization or sequencing efficiency, depending on the type of support. A frequently used diluent is 0.1 % (v/v) TFA/lO-20 % (v/v) acetonitrile in water. 11.2.2.3 Polyacrylamide Gel Electrophoresis (PAGE) In addition to its analytical applications, PAGE is widely used as a means of preparing protein samples for sequencing [7]. In a biotechnological context, the same gel which is used to monitor the isolation or verify the purity of a protein can be used
300
11 Quality Control of Protein Primary Structure
as the first step towards characterizing any impurities or confirming the identity of the target protein by Edman sequencing. When PAGE is used for preparative purposes, several precautions need to be taken. First, it is advisable to either use the best available acrylamide, methylenebisacrylamide and sodium dodecylsulfate (SDS) for gel production or purchase highquality, commercially made gels. This precaution not only minimizes potential Nterminal blockage by contaminants but also gives minimal staining backgrounds, tight protein bands, and optimal resolution. Freshly made gels should be aged at 4 "C for 16-24 h to allow time for all free radicals generated during polymerization to be consumed. In some laboratories it is customary to subject gels to pre-electrophoresis with reduced glutathione in the cathode buffer to scavenge free radicals and oxidants before the sample is applied [13]. It is preferable to disperse the sample in loading buffer at 40-50 "C rather than 90-100 "C as is usual for analytical PAGE. This precaution reduces the thermal degradation of tryptophan to improve its recoveiy during sequencing (see Section 11.2.3). After electrophoresis, the proteins within the polyacrylamide gel are electro-transferred to a PVDF membrane [7] before visualization by staining. Protein bands are then excised and can be placed directly into the reaction cartridge of a sequencer. In terms of sensitivity, as a general rule if a band is visible after staining with Coomassie Brilliant Blue R250 or Naphthol Blue-Black (Amido Black) there is sufficient protein for sequencing. To control background levels, staining solutions should only be used once.
11.2.2.4 Covalent Immobilization
Progressive sample extraction, leading to a reduction in the potential length of sequence read, can be prevented by covalent attachment to the sequencing support. This approach, also called 'solid phase' sequencing, is used routinely in instrument systems which are designed specifically for covalent sequencing but can also be valuable for particular applications on instruments which normally operate with noncovalently immobilized samples. Peptides shorter than about 30 residues which are largely hydrophobic or with an increasing hydrophobic bias towards the C-terminus tend to be more rapidly desorbed during sequencing from a noncovalent support than peptides with predominantly polar or ionic residues. Consequently, where there is a specific need to confirm the identity of residues further towards the C-terminus than is possible with noncovalent immobilization, covalent sequencing would be appropriate. A more frequent application of covalent sequencing, however, is the confirmation of residues that have undergone post-translational modification by, for example, phosphorylation or glycosylation. These modifications, being hydrophilic, prevent the extraction of their ATZ derivatives during sequencing with the normal nonpolar solvents and result in gaps in the sequence. Covalent immobilization enables the use of more polar solvents such as methanol to extract these modified residues without the risk of also desorbing the sample from its support. As mentioned in Section 11.2.1, covalent immobilization requires pre-derivatized membrane supports. These are currently available commercially in the form of kits
11.2 Automated Edman Degradation
301
which also contain the appropriate reagents and buffers [14]. A widely used protocol is to immobilize a sample via the C-terminal and side-chain carboxyl groups to an amino-functionalized support by inducing amide cross-links with 1-ethyl-3 -(3 dimethylaminopropy1)carbodiimide (EDC). The sample obviously needs to be free of any matrix components which might compete in the reaction such as ammonia, amines, and carboxylic acids. Appropriately for hydrophobic samples, however, is tolerance of high concentrations of strong detergents such as SDS which can be removed by extensive washing after immobilization. During sequencing aspartic acid, glutamic acid, and the C-terminal residues tend to be detectable at lower levels than the other residues because only those ATZ molecules from the total population which are not immobilized are extractable for conversion to PTHs. Provided that the amount of sample is not limiting, this problem would be outweighed by those factors which dictate the use of covalent immobilization. The PTH derivatives of phosphorylated residues, by virtue of their double negative charge, co-elute with the injection artefacts in the normal PTH chromatography systems. Confirmation or identification of phosphorylation sites by covalent sequencing is usually undertaken by using 32P-radiolabeled samples and collecting the methanolextracted ATZ derivatives for scintillation counting [ 101. For glycosylated residues, the PTH derivatives can be accommodated on a modified chromatography system [ 151. The identification of modified amino-acid residues in proteins and peptides typically requires the application of at least several complementary technologies, two of which are chemical degradation and mass spectrometry.
11.2.3 Data Analysis As will be apparent later in this section, protein sequencing is not generally regarded as a strictly quantitative method. Nevertheless, the first important point about sequence data analysis and interpretation is that quantitation should never be ignored. The amount, expressed in nanomoles or picomoles, of sample loaded into the instrument should ideally be known, having been determined by a validated assay or amino acid analysis. Once sequence data have been obtained and quantitated by comparison with PTH calibration standards, it is always observed that the amount of the N-terminal residue is less than the original amount of sample loaded. This amount, termed the ‘initial yield’ is typically in the 30-70 % range and dependent on several factors relating to the sample, its means of isolation and presentation for sequencing; also the instrument hardware, optimization, and chemicals. A major influence on initial yield is the degree of N-terminal blockage. Proteins with completely blocked N-termini will obviously give no sequencing signal whatsoever, only an amino acid background related to the amount loaded and the molecular weight, as will be discussed below. Partial blockage is probably unavoidable and occurs during both polypeptide isolation and sequencing. An N-terminal glutamine residue is susceptible to spontaneous cyclization to form pyrrolidone carboxylic acid, under both acidic and alkaline conditions [lo]. While such conditions can be
302
11 Quality Control
of Protein Primary Structure
avoided during sample preparation, especially if the N-terminus is known to be glutamine, they occur within a protein sequencer as part of Edman degradation. Assuming that all possible precautions to minimize blockage have been taken during sample preparation, side reactions in addition to glutamine cyclization occur during sequencing. Some are specific to serine, threonine, and cysteine which have reactive side chains, while others are caused by reagent contaminants and therefore minimized by the use of specially purified sequencing chemicals. In addition to chemical blockage, initial yield is determined by the amount of Nterminal PTH derivative eventually recovered for detection after all the reaction steps, solvent extractions, and transfers through the instrument subsystems. Not only is none of these steps 100% efficient but also the ATZ and PTH derivatives are subject to varying degrees of degradation. Serine and threonine undergo f3-elimination to form dehydro-derivatives which then partially react with dithiothreitol, the oxidant scavenger present in the conversion reagent. Consequently, the final recoveries of PTH-Ser and PTH-Thr are 30-40 % and 60-70 % respectively relative to PTH-Ala or PTH-Leu, for example, which are considerably less labile. PTH-Trp is recovered at about 10 % for sample loadings > 50 pmol, decreasing to zero for loadings < 5 pmol because of the lability of the indole ring system. Owing to the reactivity of the thiol group, PTH-Cys is only detected during sequencing in exceptional circumstances, giving rise to a blank chromatogram at a cysteine residue position. The only way to identify cysteine residues in a sequence is by prior S-alkylation of the sample by one of the many protocols available (e.g., see [10,16]). The remaining two residues to undergo significant degradation are asparagine and glutamine. These deamidate to form the corresponding acids so that PTH-Asp and PTH-Glu always accompany PTHAsn and PTH-Gln respectively. It is worth noting, however, that partial deamidation can also occur during bioprocessing, isolation, and storage. Furthermore, the extent of deamidation in all cases is highly dependent on the identity of neighboring residues, in terms of both linear sequence and three-dimensional structure [17]. As Edman degradation progresses there is a general downward trend in the recovery of each successive residue. The amount of each residue recovered can be plotted against its position in the sequence. On a semi-logarithmic scale a straight line may be fitted to the data by linear regression, the slope of which gives the average repetitive yield, expressed as a percentage as shown in Fig. ll-6A. This graph may be modified to exclude the labile residues, as in Fig. ll-6B, to obtain a more realistic impression of sequencing efficiency, especially if these residues bias the slope of the line. An alternative way of eliminating bias associated with the recovery characteristics of different residues is to calculate repetitive yields for individual residues which occur more than once in the sequence, using the equation
where RY is the repetitive yield, YA is the amount of a residue the first time it occurs in the sequence (position A), and YB is the amount of the same residue as it occurs further along the sequence (position B). Repetitive yield is not only dependent on protein sequencer performance but also polypeptide chain length and amino acid composition. A protein such as bovine
11.2 Automated Edrnan Degradation
303
--
a l l m
8
YO = 8.7 pmol R.Y. = 91.9 %
-$ -
W
21
f
1. Y0 =
7.3 pmol
R.Y. = 95.3 % 1
10
20
Residue Number Fig. 11-6. Repetitive yield determinations from sequence data for the first 21 residues of bovine plactoglobulin. Linear regression calculations were based on (A) all residues and (B) all residues except serine, threonine, and tryptophan which occur at lower recoveries and, in this case, bias the slope of the line.
0-lactoglobulin (mol.wt. 18.4 kDa) typically sequences over the first 20 residues with a repetitive yield of 93-96% when immobilized noncovalently (see Fig. 116). Under identical conditions a 20 -residue peptide typically sequences with a repetitive yield in the range 80-85 %. Therefore, as polypeptide chain length decreases the tendency for sample wash-out increases during noncovalent sequencing. This tendency is enhanced in relatively hydrophobic samples and, as described in Section 11.2.2.4, may be circumvented by covalent sequencing. Alternatively, for quality control of specific samples, custom-designed solvent delivery regimes which minimize wash-out, usually with increased chemical background as the trade-off, can be used. Other composition-related factors to impact negatively on repetitive yield concern the number of serine, threonine, nonalkylated cysteine, glutamine, and proline residues occurring in the sample over the stretch of sequence which is subjected to Edman degradation. The amino acid side chains of serine, threonine, and cysteine are susceptible to trifluoroacetylation during the cleavage reaction. As sequencing progresses, any residues which are derivatized in this way undergo O+N (serine, threonine) or S+N (cysteine) acyl shift on reaching the N-terminus, with a consequent increase in the population of blocked polypeptide molecules [ 3 ] . The other blockage event occurring in the presence of TFA is glutamine cyclization, as discussed above. These side reactions can all be minimized by optimizing the cleavage time, however, proline requires an extended exposure to TFA for efficient cleavage (see Section 11.2.1). Consequently, whenever a proline is encountered in the sequence, carry-over into the next cycle occurs with the result that the proline and
304
I 1 Quality Control of Protein Primary Structure
all subsequent signals are effectively reduced, lowering the overall repetitive yield. These observations mean that the sequencing efficiency of individual samples is idiosyncratic and that repetitive yield values cannot be interpreted without the relevent sequence data. A sample giving an initial yield of 100 pmol for an N-terminal alanine, for example, gives an alanine signal at residue 20 of 37.8 pmol at a 95 % repetitive yield. This residue 20 signal would be reduced to 4.6 pmol at a repetitive yield of 85 %; however, even this signal is 100-fold higher than the theoretical minimum detectable level for modem sequencing systems. The length of sequence read depends not only on the actual residue yield over a number of degradation cycles but also the amino acid background level. The sequence only becomes unreadable when the yield becomes indistinguishable from the background. As is the case for repetitive yield, background levels are sample-dependent and influenced by molecular weight and amino acid composition. In general, longer polypeptides give rise to higher backgrounds because there are more peptide bonds per mole to undergo spontaneous acidolysis during exposure to TFA in the reaction cartridge. Samples rich in serine, threonine, aspartic acid, and cysteine (nonalkylated) tend to fragment more readily as a result of cyclization reactions involving the side chains, followed by rearrangements which lead to peptide bond fission [9]. Unlike repetitive yield, the background level increases as a function of the amount of sample loaded and is independent of the actual amount giving rise to a sequencing signal, i.e., the initial yield. These factors have two consequences. First, the length of sequence read does not increase in proportion to the amount of sample loaded into the sequencer. The same protein sequenced with different amounts loaded should give initial yields in the same proportions and identical repetitive yields. However, higher loadings produce higher backgrounds which converge with the progressively falling sequence signal after fewer degradation cycles. Second, samples of the same protein sequenced with identical loadings but exhibiting different degrees of N-terminal blockage give rise to the same repetitive yield and background level. Lower initial yields mean that this background level is again reached after fewer cycles. These relationships between initial yield, repetitive yield, sample characteristics, and background are illustrated in Fig. 11-7. Apart from ensuring that instrumentation
Residue Number
Fig. 11-7. The relationship between theoretical initial yield (YO),repetitive yield, background, and length of sequence read. When two samples containing the same amount of a particular polypeptide are sequenced, the repetitive yields (i.e., the slope of the solid lines) and backgrounds (broken line) should be identical. If one sample gives a lower sequencing yield as a result of a higher level of N-terminal blockage, the length of sequence read is curtailed, stopping at residue i intead of j.
11.2 Automated Edman Degradation
305
is fully optimized and operated using the appropriate sequencing protocol for a particular sample, two further means of maximizing the length of sequence read are concerned with controlling background levels. The first is to minimize contamination which might otherwise produce a chemical or amino acid background (see Section 11.2.2.1). The second strategy is to sequence proteolytic fragments derived from proteins whose molecular weight or composition might otherwise generate excessive background levels (see Section 113. From the above discussion, it is clear that quantitation in Edman degradation is subject to the idiosyncracies of individual polypeptides and amino acid residues. Consequently, for the majority of sequencing applications, quantitation is secondary to qualitative residue identification. As stressed in the first paragraph of this section, quantitation is, however, still valuable. Comparison of the actual amount of sample loaded into the reaction cartridge with the initial yield can provide an indication of whether the sequence belongs to a major component of the sample or, if this is N-terminally blocked, a minor component. This information may be useful in the identification of contaminants during isolation process monitoring. For quality control, the initial yield expected for a particular protein product is established as part of method validation. As an alternative to quoting the N-terminal residue yield, extrapolation to the y-axis of a repetitive yield graph gives an estimate of the amount of sample available for sequencing before the start of Edman degradation, the theoretical initial yield,or YO.This parameter may provide a better indication of the original sample amount than the yield of the N-terminal residue when this forms one of the low recovery PTH derivatives. On the other hand, the theoretical initial yield can be biased by the means of repetitive yield calculation, as exemplified by comparing Fig. 11-6A and 11-6B. The method used to represent the data should be determined empirically according to which is more sensitive to deviations from the standard. Such deviations can result from either incorrect bioprocessing or degradation during isolation or storage. N-terminal blockage by, for example, reaction with a reducing sugar is detectable by a decreased initial yield. Proteolytic cleavage might produce a frayed N-terminus, with signals from residues 2 and possibly 3 appearing with residue 1 (sequence ‘preview’). Initial yield data may then be used to set acceptance criteria on levels of contaminants such as truncated species which are closely related to the desired product. Quantitating the relative amounts of different polypeptide components of a sample may not be as easy because these may not be reflected precisely by the relative initial yields (see Fig. 11-5). To apply sequence analysis to characterizing mixtures requires a validation strategy in which the relative proportions of the expected components are varied over an appropriate range. In this way, factors which relate initial yields to the amounts present can be determined. Where theoretical initial yields are employed, a sufficient number of cycles to provide a representative repetitive yield line should be performed. This number would be influenced by the positions of serine, threonine, and tryptophan in the sequence and is usually in the 10-15 range. Repetitive yield is a measure of sequencing system performance for a particular polypeptide sample. Its utility in quality control is to demonstrate that the complete analytical system including sample dissolution, immobilization on the sequencing
306
11 Quality Control of Protein Primary Structure
support, instrument programming and performance, and PTH quantitation is valid for its intended purpose. Protein sequencer manufacturers use a standard protein such as bovine 0-lactoglobulin to test the installation and performance of an instrument which should achieve minimum initial and repetitive yields under specified analysis conditions. Instrument verification should be carried out as part of quality control and in some laboratories this is done both before and after the product analyses. Whether or not a performance test is done using the manufacturer’s test protein, verification always involves sequencing a reference sample of the product.
11.3 Carboxy Terminal Analysis For many quality control applications, full verification of the identity and chemical structure of a polypeptide is attained without direct C-terminal analysis. In other words, determination of the C-terminal amino acid residue and adjacent residues in the C-terminal region does not necessarily provide information which cannot be deduced from a combination of Edman degradation, mass analysis, and peptide mapping. If the polypeptide is small, both complete Edman degradation from its N- to Cterminus, assuming that the N-terminus is not blocked, and unambiguous characterization by mass analysis within the accuracy limits of the instrumentation (see Section 11.4) is possible. For both types of analysis the upper mass limit for characterization of a polypeptide as a single molecule is, in practice, dependent on individual circumstances but typically about 3000-5000 Da, i.e., 25-40 residues. To overcome the limitations associated with the analysis of larger polypeptides, digestion and subsequent analysis of the isolated peptide fragments, i.e., peptide mapping or fingerprinting, is the preferred strategy. This approach to protein characterization will be covered in Section 11.5. Of relevance here is that, by cleaving a protein into more conveniently sized pieces, complete verification of its primary structure is possible without the need for specific C-terminal residue identification or sequencing from the C-terminus. Confirmation of correct C-terminal processing or determination of the extent of spurious proteolytic degradation at the C-terminus, for example, therefore requires several steps: the generation, identification, isolation, and analysis of the C-terminal peptide fragment (or fragments). Consequently, the principal reasons for direct C-terminal analysis are speed and economy. Like N-terminal sequencing, the idea of C-terminal analysis is not new. The first methods for N-terminal and C-terminal sequencing were published in 1930 [l] and 1926 [ 181 respectively. While N-terminal sequencing bas been in common use since Edman’s publication in 1950 [2], this has not been the case for C-terminal sequencing. The reason for this is the relative inefficiency of carboxyl group activation compared with that of the amino group [l]. Because of the potential utility of Cterminal sequencing, there has been considerable commitment to developing a viable chemical degradation method [ 18,193. The challenges associated with the development of a chemical C-terminal degradation method have provided an incentive for the use of enzymatic degradation with carboxypeptidases [3]. As will be discussed in the following sections, these two approaches currently have neither the sensitivity
11.3 Carboxy Terminal Analysis
307
nor length of sequence read achievable with N-terminal sequencing. However, for quality control applications in biotechnology, these limitations may not be disadvantageous.
11.3.1 Automated Chemical Degradation Two automated C-terminal sequencing systems have recently entered the protein analysis arena. The first was launched as a commercial product in 1994 [20]. The second system is, at the time of writing, undergoing its pre-launch testing phase in a number of protein sequencing laboratories and has been used in several research projects, e.g., [21]. Both systems are currently subject to progressive modifications in both chemistry and instrument programming with a view to enhancing their sensitivity and performance. Because specific details are liable to become rapidly outdated, only a general appreciation of C-terminal sequencing chemistry will be given here. As in Edman degradation, the first step in C-terminal sequencing is referred to as coupling and similarly involves derivatization to form an active moiety. Carboxyreactive compounds in current use are diphenylphosphoroisothiocyanate[20] and acetic anhydride [18]. After coupling, an intramolecular cyclization to form a thiohydantoin is induced at the C-terminus. One problem at this point is that, because aspartic and glutamic acid side chain carboxyl groups are also activated, deleterious side reactions involving amino, hydroxyl, and thiol groups can occur in the steps leading to thiohydantoin formation. Premature termination of sequencing by this mechanism, a major contributor to the inefficiencies associated with this chemistry, can be circumvented by prior derivatization of nucleophilic groups by for example N-phenylcarbamylation and 0-acetylation, and quenching activated side chain carboxyls by amidation [21]. The next step is cleavage of the C-terminal residue as its thiohydantoin derivative which, like the phenylthiohydantoin resulting from Nterminal degradation, is identified and quantitated by reverse-phase HPLC. Refinements to the cleavage chemistry have been introduced by the instrument manufacturers to improve efficiency. These include the use of aqueous TFA vapor to enable cleavage at proline, which is otherwise highly inefficient [20], and alkylation prior to cleavage to increase the reaction rate [ 191. Another refinement involves the inclusion of thiocyanate in the cleavage reagent formulation to drive the direct formation of the pre-cleavage derivative of the following residue. In this way, Cterminal activation of the polypeptide is only done at the first cycle with the consequence that spurious fragmentation of the sample during sequencing cannot generate background signals (cf. Edman degradation) [ 191. Key elements of the currently available C-terminal sequencing chemistries are shown in Fig. 11-8. As in the case of Edman degradation, overall sequencing efficiency is highly sample dependent but C-terminal sequencing is, at present, significantly less sensitive. A sequence read of 10 residues typically requires a sample loading of 1000 pmol and a realistic minimum loading to obtain three residues is about 100 pmol. Despite these current
308
11 Quality Control of Protein Primary Structure
A lCarboxylActivation] .)Formation
A
-1C-Termina;
Cleavagd
I
IC-Terminal Identification
1-
Formation
I
C - T e r m i n a ' ~ A s p / O l uAmidation 7
I
Alkylation
ISerlThr Acetylation
IC-Terminal Cleava
-Terminal Alkylationl
k-Terminal Identification
Fig. 11-8. Current C-terminal sequencing chemistries.
limitations, a significant number of both commercial and academic laboratories have invested in the available technologies.
11.3.2 Enzymatic Degradation In view of the difficulties encountered during the development of a robust chemical C-terminal degradation method, there has been an interest in exploiting carboxypeptidases for this purpose [ 2 2 ] . The original strategy, summarized in Fig. 11-9, involves amino acid analysis to identify the residues released from the C-terminus of the protein sample by carboxypeptidase action. To derive sequence information, released amino acids must be identified and quantitated over a time-course on the basis that the release of each amino acid can only occur once the previous one has
11.3 Carboxy Terminal Analysis
309
Carboxypeptidase
Remove aliquots at timed intervals
uwww
1
2
3
Add to sulfosalicylic acid to denature
4
Time
Fig. 11-9. C-terminal sequencing by using carboxypeptidase and amino acid analysis.
been removed. In practice, this method is complicated by the individual substrate specificities of carboxypeptidases. For example, carboxypeptidase A (EC 3.4.17.1) releases histidine, glutamine, threonine, and nonpolar residues rapidly from the Cterminus. Asparagine, serine, lysine, glycine, and acidic residues are released more slowly while proline and arginine are not released at all. Furthermore, the penultimate residue has an effect on the rate of C-terminal release [23]. Therefore method optimization for each polypeptide sample must be done on an individual basis by testing a range of carboxypeptidases and enzyme:substrate ratios to achieve reproducible and interpretable data. The sensitivity of this method is determined by that of the amino acid analysis method used (reviewed in [3]) and the background levels; however, a realistic minimum sample amount might be 500 pmol. A potential problem is the inability to detect and quantify ragged C-termini as a result of bioprocessing or spurious proteolytic degradation, both potentially important applications of C-terminal analysis in quality control.
11.3.3 Mass Analysis Since gaining wider accessibility to the protein characterization laboratory, mass spectrometry has largely replaced amino acid analysis for sequencing in conjunction with carboxypeptidase digestion [24]. Instead of identifying the released amino acids over a time-course, the masses of the truncated forms of the sample polypeptide are
3 10
11 Quality Control of Protein Primary Structure
Mass interval: 115.09 113.16 Resldue 1
ASP
Residue 2 Residue3
101.11 Da
Leullle Thr
Fig. 11-10. C-terminal ladder sequencing by using carboxypeptidase and mass spectrometry.
used to deduce the C-terminal sequence. This approach, outlined in Fig. 11-10, is referred to as ‘ladder sequencing’ because the mixture of truncated polypeptide chains are interpreted in decreasing order of mass, starting with the mass of the complete molecule and calculating each successive mass difference to derive the sequence of residues. The advantages of ladder sequencing are that it is more rapid and sensitive than both the chemical and carboxypeptidase-amino acid analysis methods, with potential minimum sample loadings in the 10-100 fmol range [24]. An obvious limitation is that some residues and residue combinations are isobaric (see Section 11.4). Another possible limitation is that, because of the idiosyncracies of carboxypeptidase action (see above) or ionization properties of polypeptides, some truncated forms may be missing from the mass profile. Therefore, while it may still be possible to deduce the combination of residues within a particular segment of the polypeptide, their sequence will be indeterminate. Ragged C-termini may be detected by mass analysis of the sample before the addition of carboxypeptidase, provided that all molecular species readily ionize. C-terminal sequencing can also be done by mass spectrometry without the use of carboxypeptidases. This approach takes advantage of the ability of some types of instrumentation to induce polypeptide fragmentation, producing ladder sequencing data. This and other applications of mass spectrometry will be covered in the next section.
11.4 Mass Spectrometry Figure 11-11 shows the basic components of a mass spectrometer. The different types of instrument differ in their means of sample introduction, ionization, and mass analysis; therefore specific details will be covered in relevent sections below (see also [25-281). All MS methods aim to generate gas phase ions with the subsequent determination of their mass-to-charge ( d z ) ratios and hence their masses. The more
11.4 Mass Spectrometry
introduction
3 11
P analyzer
Fig. 11-11. The basic components of a mass spectrometer.
easily ions are formed (e.g., by protonation), the better the probability of obtaining a mass measurement. Every ion produces its own characteristic m/z value(s) and fragmentation enables structural diagnosis, although interpretation of the resulting spectra may require considerable skill on the part of the analyst. Twenty years ago, the idea of using mass spectrometry (MS) for protein analysis would have been met with considerable resistance, with thoughts of huge instruments, incompatible solvents, complicated data programs, and specialized training. Many of the technologies available at that time were applicable only to relatively small molecules, as opposed to peptides and proteins, because of difficulties with volatilization and ionization. The utility of MS in protein primary structure determination has, however, long been recognized. It is clearly possible, in principle, to deduce the amino acid composition of a polypeptide from its molecular mass and, by various fragmentation strategies, to determine its sequence. Ambiguities, however, may arise because of the isobaric residue pairs, leucine/isoleucine and glutamine/lysine, and combinations which are close in mass, e.g., arginine/glycine + proline. Additionally, MS may allow the identification and localization of co- and post-translational modifications. In fact, a well-known scientist stated at a conference that he could not see any future for Edman degradation because of the ‘simplicity’ of sequencing proteins and peptides on a tandem mass spectrometer. One of the drawbacks not mentioned was the large amount of physical space that was required for such an instrument and its mass limit, at that time, of around 2500 Da. The impressive accuracy of this type of mass spectrometer, with its ability to resolve isotopes, illustrates the traditional analytical chemistry focus of MS before specific strategies and hardware were developed to enable protein and peptide analysis. An early and well-studied approach made use of peptide derivatization by N-acetylation and N,O-permethylation to enhance volatility for GC-MS [29]. This strategy enabled the analysis of peptides with up to ten residues and characteristic fragmentation patterns could be interpreted to derive sequence data, as shown in Fig. 11-12. Applicability to polypeptide MS continued to evolve in the early 1980s with the development of fast atom bombardment (FAB) ionization, which will be described in Section 11.4.1. In the mid 1980s, Biolon AB, a Swedish company based in Uppsala, introduced a plasma desorption mass spectrometer (PDMS). This was the first compact and simple (for the non-MS specialist) proteidpeptide mass analyzer,
312
11 Quality Control of Protein Primary Structure
I
I CH,CO---N---CH--C---N---CH+ I
II
CH,
0 CH,
I
Fig. 11-12. Ions formed during mass analysis by GC-MS of a permethylated peptide which has been dissociated with a collision gas (see Fig. 11-13 and 11-14).
capable of measuring molecular masses up to about 10-30 kDa (see Section 11.4.2). Continued development of ‘friendly’ mass analyzers resulted in the emergence of so-called hyphenated techniques, in addition to GC-MS, such as LC-MS and CEMS. The advent of atmospheric pressure interfaces to mass analyzers permitted the use of liquid junctions (see Section 11.4.3). Simultaneously, the attainment of increasingly higher mass measurements, particularily by the use of matrix-assisted laser desorption ionization (MALDI) instruments was the result of the pioneering work of Hillenkamp and Karas in 1988, who broke the 100 kDa barrier (see Section 11.4.4). As can be gathered from this introduction, there are several approaches through various analyzers to the measurement of mass; each has its own particular strengths. These will be addressed in the following sections describing four main types of instrument. The order of presentation should not imply any preference on our part. These four types of mass analyzer can be divided into two general categories. With respect to sample application, FAB- and electrospray (ES)-MS are congenial to hyphenated techniques such as online LC. They are, however, critical with respect to solvent composition, in that salts can disturb and often obliterate any analyte ion measurements, owing to the fact that the solvent ions ‘absorb’ ionization potential, inhibiting ionization of the analyte. PD- and MALDI-MS require, at present, that samples are dry when introduced into the high-vacuum region of the instrument. They are more accomodating with respect to salt content; the PDMS target can be washed after sample application to remove interfering salts, the presence of sodium ions in particular being detrimental to the quality of the spectrum obtained.
11.4 Mass Spectrometry
313
11.4.1 FAB-MS Barber and colleagues' introduction of FAB ionization technology provided a significant step forward in the analysis of labile and non-volatile compounds, including polypeptides [30,31]. A viscous, nonvolatile solvent such as glycerol is used as the sample matrix but other useful matrices include thioglycerol and 2, 2'-dithiodiethanol, which are thought to enhance sample protonation. The sample/matrix mixture is placed on to the tip of a probe which is introduced into the ion source and held at high potential. A beam of, for example, xenon or argon atoms bombards the sample, releasing ions which are then extracted, focused and accelerated into the mass analyzer. Ions generated by FAB tend to be the singly charged (M+H)+ species, but (M+Na)+, (M+K)+, and glycerol adduct ions can also occur. In most situations, the practical upper mass limit is 3-5 kDa. As mentioned above, FAB-MS is amenable to online coupling with HPLC by means of a continuous-flow probe [32] and tryptic digests (see Section 11.5) have been successfully analyzed in this way. In addition, when it is coupled to a tandem MS, as shown in Fig. 11-13, selection of a specific peptide ion prior to further collision enables sequence analysis [33]. This type of analysis also permits distinction of leucine and isoleucine in a peptide, owing to intra-sidechain fragmentation in addition to the series of main-chain fragment ions formed by the dissociation (Fig. 11-14) [25,34]. FAB-MS has been successfully used for the determination of disulfide bonds in proteins by comparing mass data before and after reduction. However, it is worth noting that the thiol-containing matrices referred to above may lead to unintentional disulfide bond reduction [33]. There may be advantages in using FAB-MS for the analysis of hydrophobic peptides since there appears to be a positive correlation
lixzziq
MS-1 B.
Lens El
*
Collision
Collector il I D
1
/
SLens
Fig. 11-13. A tandem (MS-MS) mass spectrometer. Ions formed from the sample are resolved by the first electric (El) and magnetic (B1) field regions. A selected ion is then exposed to gas at low pressure within a collision cell prior to separation of the dissociated fragments in the second electric (Ez) and magnetic (Bz) fields.
314
11 Quality Control of Protein Primaiy Structure x3
y3
23
x2
y2
22
xl
yl
zl
Fig. 11-14. Possible fragment ions formed by collision induced dissociation (CID, also referred to as collisionally activated dissociation, CAD) of a polypeptide during an MS-MS experiment. This type of data can be interpreted to obtain a polypeptide sequence.
between the hydrophobicity of a peptide and its ionization efficiency by FAB. This may be explained by the glycerol used to disperse the sample inducing the accumulation of hydrophobic peptides at its surface. An increased occurrence of fragment ions may be observed with more concentrated samples because of facilitated diffusion of the analyte to the surface, where collision-induced dissociation (CID) by the bombarding atoms occurs.
11.4.2 PDMS The basic principles of PDMS are described in Fig. 11-15 [34,35]. The sample, in solution, is loaded on to a nitrocellulose-coated Mylar foil target. For best results, the target should be attached to a horizontal spinning rotor so that the solvent is rapidly removed by the centifugal effect while the analyte is retained on the nitrocellulose. Any salts present can be washed off the target with water while it is spinning on the rotor. Figure 11-15 presents a typical mass spectrum obtained by PDMS, which in the absence of salts gives predominantly the (M+H)+ ion, with fragment ions (e.g., the des-carboxyl species) seen only at high sample loads. No sequence information can be derived by PDMS unless used in conjunction with, for example, carboxypeptidase digestion which can be done on the target (see Section 11.3.3). Although mostly used for samples of molecular mass under 10 kDa, the technique has been used for some proteins of higher mass, for example growth hormone and trypsin. For characterization of polypeptide mixtures, PDMS only provides qualitative information; signal levels reflect peptide-specific ionization efficiencies, not absolute amounts. In other words, peptides giving a low relative abundance in a spectrum may be the predominant components of a mixture and vice versa.
11.4 Mass Spectrometry
Start
3 IS
stop
miz
Fig. 11-15. Plasma desorption mass spectrometry (PDMS). The sample is placed on a target which is held at 10-20 kV. Ionization is induced by collisions with particles produced by nuclear fission of the radioactive californium source. Each fission event produces two particles which are ejected simultaneously in opposite directions. While one particle desorbs ions from the target, the other triggers a start signal in the time-to-digital converter (TDC). The desorbed ions are accelerated towards a grid at ground potential then travel in a field-free region before hitting a detector which simultaneously triggers a stop signal. In this type of instrument, m/z ratios are resolved by time-of-flight (TOF). The lower panel shows a PDMS spectrum of a decapeptide (1148 Da). The signal at -44 Da corresponds with the loss of a carboxyl group.
11.4.3 ESMS The use of ESMS for the mass analysis of proteins and peptides has expanded exponentially since the pioneering work of Fenn in the late 1980s [36]. Its adaptability to online analysis, particularily with HPLC, has been of tremendous benefit to the protein scientist and peptide mapping (see Section 11.5) of recombinant proteins for quality control can be performed routinely with LC-MS. The probable mechanism of ionization by electrospray, which is not fully understood [37], is shown in Fig. 11-16. Ions are resolved with a quadrupole mass analyzer, which scans across a specified m/z range, e.g., 100-3000 Da, filtering and allowing ions to reach the detector in 0.5 Da steps. Instruments can either be fitted with a single quadrupole or, for tandem MS, a triple quadrupole, as illustrated in Fig. 11-16 [38-401. The top m/z limit for quadrupole mass analyzers is 2500-3000 Da. Therefore the ability to analyze proteins of higher molecular mass depends on the sample’s acquisition of many charges to produce a series of ions with m/z values within the operating range of the quadrupole. A typical ESMS spectrum of a protein is shown in Fig. 11-17. The molecular mass of a protein analyzed by ESMS is cal-
3 16
11 Quality Control of Protein Primary Structure N, Curtain Gas
A: Ion Source
% %
& B e Sample Introduction
rn
Explosion
C P
Molecular Ions
Ion Evaporation
B: Mass Analyser
Ion Source
Q1
lonScanl Precursor
Q2
Q3
Detector
E l
Fig. 11-16.Triple quadrupole electrospray mass spectrometry (ESMS). The ion source (A) comprises a needle at high electrical potential through which the sample flows in solution. Charged droplets emerge from the needle and solvent evaporation occurs as they are accelerated towards the entrance to the mass analyzer. The flow of nitrogen (curtain gas) aids solvent evaporation and blows uncharged liquid away. As the droplets reduce in size, a point is reached when the charge density becomes too great and coulombic explosion occurs. Molecular ions are then released in the vapor phase. The mass analyzer (B) contains a first quadrupole (Ql) which scans the m/z ratios of the ions entering from the electrospray. The second quadrupole (42) holds selected precursor (parent) ions in contact with gas at low pressure while CAD gives rise to product (daughter) ions which are then scanned by the third quadrupole (Q3).
culated by using a computer algorithm supplied by the instrument manufacturers, with a typical accuracy of 0.01 %. Because mass resolution is by filtration, the number of ions reaching the detector compared with those actually generated by the ion source is typically reduced by five to six orders of magnitude. Despite this, sensitivity for protein analysis by ESMS can be around 1 picomole. The characterization of glycoproteins presents particular challenges owing to microheterogeneity and the presence of sialic acids, which both have a negative impact on sensitivity. This problem was largely overcome in a study of tissue plasminogen activator by neuraminidase treatment to remove sialic acids, and trypsin digestion followed by online HPLC to allow analysis of individual glycopeptide fragments by LC-MS [41]. HPLC also overcomes a potential problem with ESMS: suppression of sample ionization by the presence of salts which reversephase LC removes. In ESMS, a useful application is single ion monitoring to scan peptides as they enter the mass analyzer for specific structural features such as glycosylation or phos-
11.4 Mass Soectrometn,
8
100
-s
-
3 17
B
C
I
cn
C 0-
C ; 50-
.-5 -m d 0-
_ - A I
I
phorylation sites. It is possible rapidly to alternate the potential at the entrance to the mass analyzer between normal and increased settings so that, in the latter case, CID occurs. At the same time as measuring the total mass of a peptide it is possible to record the presence of specific ions released during dissociation. For example, positive ions at 162 or 203 Da are derived from carbohydrate moieties, while a negative ion at 79 Da is phosphate.
11.4.4 MALDI-MS This technology had its origins in the 1960s, but only with the development of suitable matrix compounds, particularly the pioneering work of Karas and Hillenkamp in the 1980s, did its forte as an analysis method for determining the molecular weights of proteins greater than 100 kDa emerge [42,43]. The principle of the method is similar to that of PDMS, as can be seen in Fig. 11-18 with a laser substituted for the radioactive source. Whereas in PDMS only one choice of surface is available, in MALDI the choice of matrix is determined to some extent by the analyte of interest. The most commonly used matrix compounds include nicotinic, dihydroxybenzoic,
318
11 Quality Control of Protein Primary Structure
30kV
15kV
Target
OkV
Detector
\\
\3
Laser Beam (337 nm)
I
1
mlz
Fig. 11-18. Matrix-assisted laser desorption ionization (MALDI) mass spectrometry. The polypeptide sample is mixed with a matrix compound, e.g., sinapinic acid, before application to the target. A laser beam activates the matrix which transfers energy and ions to the sample. Desorbed ions are then accelerated across a potential gradient towards a detector for resolution by TOF, as in PDMS (see Fig. 11-15). The lower panel shows a MALDI mass spectrum of a 45 kDa protein. Accompanying the singly charged (M+H)+ ion are the (M+6H)6+, (M+4H)4+, and (M+2H)*+ ions with m/z values of 7780, 11454, and 22763 respectively.
sinapinic, and a-cyano- 4 -hydroxycinnamic acids [44]. With MALDI analysis, the most predominant signal can be attributed to the singly charged species, in contrast to electrospray analysis, and the upper limit of detection is theoretically unlimited, but practically ca. 200 kDa. Multiply charged species can also be found, as can be seen in Fig. 11-18, but their abundance is dependent to some degree on the matrix used. Tryptic digests (without prior separation by HPLC) may be analyzed by MALDI-MS and a higher resolution may be achieved by use of a reflectron, which in effect doubles the flight length. The enhanced resolution provided by the reflectron is apparent in Fig. 11-19 and also enables sequence information to be obtained from fragmentation due to post source decay [45]. On the MALDI targets, as is the case on PDMS targets, the possibility exists for performing reactions such as enzymatic digestion of the protein or peptide of interest, reduction of disulfide bonds, and other manipulations. In this way, samples for MALDI and PDMS can, in principle, be used many times. A relatively new configuration of MALDI-MS is ‘delayed extraction’, in which the ionizing event is separated from the acceleration event [46]. This enables higher laser power to be used, with a consequent improvement in signal intensity and mass
11.5 Protein Fragmentation
L7
LaserBeam
319
Linear
ReflAron
(Ion Mirror)
m/z Fig. 11-19. MALDI mass spectometry incorporating a reflectron. Ions accelerated towards the linear detector can be deflected towards the reflector detector to effectively double the flight length. Panel (A) shows the signal given by neurotensin, using a-cyano-4-hydroxycinnamic acid as the matrix, in linear mode. The same sample analyzed in reflector mode (B) shows the presence of isotopes as a result of enhanced resolution.
accuracy. This configuration can function in both linear and reflector modes and is of particular benefit in the latter.
11.5 Protein Fragmentation In the previous sections, there have been several references to protein fragmentation, which can be achieved by chemical or enzymatic means or both. Commonly used methods are summarized in Table 11-1, with more detail provided in [3,10]. A widely used application in protein quality control is in peptide mapping or ‘fingerprinting’. Exploitation of the cleavage site specificities of enzymes, trypsin being the most popular, results in the generation of set of peptides which can be analyzed by HPLC (with UV detection), LC-MS, or MS without prior separation. By comparison of the pattern obtained from the test sample with that from a reference, deviations from the expected primary structure may be detectable. Owing to the progressive decrease in signal and increase in background experienced with Edman degradation, it may not be possible obtain sequence data as far away from the N-terminus as needed. Alternatively, the N-terminus of the sample may be blocked and therefore refractory to Edman degradation. In these situations,
320
11 Quality Control of Protein Primary Structure
Table 11-1. A summary of enzymes and reagents for protein fragmentation. Protease or reagent
Cleavage site
Conditions
endoproteinase Arg-C
-Arg-X-
ammonium bicarbonate PH 8 8 h at 37°C
endoproteinase Asp-N
-X-Asp-
ammonium bicarbonate PH 7 4-16 h at 40 "C
chymotrypsin
-Trp,Tyr,Phe,Leu-X-
ammonium bicarbonate, Tris-HC1 pH 8, 4 h at RT
endoproteinase Glu-C
-Glu-Xbut slow at -Glu-Glu-
ammonium bicarbonate, Tris-HC1 pH 8 4-16 h at RT
endoproteinase Lys-c
-Lys-x-
Tris-HC1 pH 8 > 2 h at RT
pepsin
-Glu/Phe/Tyr/L,eu-X-
0.1 % TFA, 0.1 % HC1 2 h at RT
thermoly sin
-X-LeuAle/PheNal/Ala/Met-
Tris-HC1 pH 7.6 (+ 2 mM CaC12) 1-16 h at RT
trypsin
-Arg/Ly s-Xbut not when -Arg/Ly s-Pro-
ammonium bicarbonate, Tris-HC1 pH 8 4-16 h at RT
cyanogen bromide
-Met-Xbut very slow -Met-Ser/Thr-
70% formic acid overnight at RT
h ydroxy lamine
-Asn-Gly-Asn-Asp-
1.8 M-hydroxylamine for 3 h at 45 "C
o-iodosobenzoic acid
-TT-X-Tyr-X-
o-iodosobenzoic acid (4 g L-') in 4 M guanidinium-HC1 16 h at RT
BNPS-skatole
-Trp-X-
BNPS-skatole (10 g L-l) in 50% acetic acid 48 h at RT
acid
-Asp-X-Asp-Pro-
70% formic acid for 24-86 h
the only way to obtain internal sequence data is by fragmentation followed by separation of the peptides. This procedure will enable identification of the majority of peptide fragments from a protein, but will not give any information as to the order in which they occur in the protein. Thus, the protein can be digested by a second, or third method to derive overlaps which enable the complete primary structure to be
Acknowledgements
321
deduced. This strategy clearly involves a large amount of sequencing; therefore, in the quality control environment, peptide mapping and mass analysis can provide complementary information to minimize the time and cost of Edman degradation. Fragmentation is the obvious means of preparing specific segments of a protein for further characterization. The task of verifying modifications such as glycosylation or disulfide bonds at specific sites on a protein is simplified or even made practicable by this strategy. The choice of cleavage method depends on the sequence of the protein, which, for quality control purposes will already be known. Proteolytic enzymes are generally used in preference to chemical methods because they tend to generate smaller fragments which are more amenable to isolation by reverse-phase HPLC. Because chemical cleavage sites (e.g., Met -X and Trp-X, see Table 11-1) are relatively few in number within typical proteins, fragments can be large and may not adequately simplify the subsequent analysis. In addition, the strongly denaturing conditions used in chemical cleavage protocols frequently cause aggregation to the extent that the fragments are poorly recovered and resolved by HPLC. The problem may be overcome in some cases by the use of C4-type reverse-phase media instead of C8 or c18 and by using propan-2-01 in acetonitrile instead of acetonitrile alone as the mobile phase organic modifier. Many proteins are resistant to enzymatic proteolysis unless placed under denaturing conditions. This resistance may be due to either tertiary structure or steric hindrance by modifications like glycosylation. The inclusion of denaturants such as urea, SDS, or acetonitrile must be tempered with the stability (activity) of the protease under those conditions. The best option is to consult data sheets provided by suppliers of proteases. It should be noted that many of the proteases in common use have their optimum activity at pH values above 7 (see Table 11-1). At this pH and above, disulfide bonds are liable to scramble. Therefore, for the determination of disulfide bond connectivities, fragments must be generated at pH < 6 using mild acid cleavage or pepsin, for example.
11.6 Summary and Future Prospects We have described two types of instrumentation used for the structural characterization of proteins. Mass spectrometry is rapidly gaining importance as a technology which complements the more traditional Edman sequencing. It is clear that no single technology can provide all the information which is demanded: an orthogonal approach is required.
Acknowledgements We would like to express our thanks to Jonathan MacBeath, Ian Blench, Kevin Howland, and Dorte Christensen for their valuable input.
322
I1 Quality Control of Protein Primary Structure
References Doolittle, R. F., in: Methods in Protein Sequence Analysis: Elzinga, M. (Ed.), Clifton: Humana Press, 1982; pp. 1-24. Edman, P., Acta Chem Scand, 1950, 4, 283-293. Allen, G., Sequencing of Proteins and Peptides. Amsterdam: Elsevier, 1989. Hewick, R.M., Hunkapillar, M. W., Hood, L. E., Dreyer, W. J., J Biol Chem, 1981, 256, 7990 -7997. Linse, K. D., Carson, W., Farnsworth, V., Technical Information Bulletin T-0105. Beckman Instruments, Inc. Horn, M. J., Miller, C. G., Harrsch, P. B., Woo, W., Wagner, G. W., Presentation Abstracts, 6th Symposium of The Protein Society, San Diego, CA, 1992. Cambridge: Cambridge University Press, 1992. Matsudaira, P., J Biol Chem, 1987, 262, 10035-10038. Werner, W. E., Hsi, K.-L., Grimley, C., Yuan, P.-M., Presentation Abstracts, 9th Symposium of The Protein Society, Boston, MA, July 1995. Cambridge: Cambridge University Press, 1995. Tam, G. E., in: Methods of Protein Microcharacterization: Shively, J. E. (Ed.), Clifton: Humana Press, 1986; pp. 155-194. Aitken, A,, Identification of Protein Consensus Sequences. Chichester: Ellis Honvood, 1990. Farries, T. C., Harris, A., Auffret, A.D., Aitken, A., Eur J Biochem, 1991, 196, 679-685. Wellner, D., Panneerselvam, C., Horecker, B. L., Proc Nat Acad Sci USA, 1990, 87, 19471949. [13] Dunbar, B., Wilson, S.B., Anal Biochem, 1994, 216, 227-228. 1141 Coull, J. M., Pappin, D. J. C., Mark, J., Aebersold, R., Koster, H., Anal Biochem, 1991, 194, 110-120. 1151 Gooley, A. A,, Packer, N. H., Pisano, A., Redmond, J. W., Williams, K. L., Jones, A., Loughnan, M., Alewood, P.F., in: Techniques in Protein Chemistry VI: New York: Academic Press, 1995; pp. 83-90. [16] Yarwood, A., in: Protein Sequencing: a Practical Approach. Findlay, J. B. C., Geisow, M. J. (Eds., Oxford: IRL Press, 1989, pp. 119-145. [17] Wright, T. W., Crit Rev Biochem Mol Biol, 1991, 26, 1-52. [18] Tarr, G. E., in: Methods in Protein Sequence Analysis: Wittman-Liebold, B. (Ed.). Berlin: Springer-Verlag, 1989; pp. 129-136. [19] Boyd, V. L., Bozzini, M. L., Zon, G, Noble, R. L., Mattaliano, R. J., Anal Biochem, 1992, 206, 344-352. [20] Miller, C. G., Bailey, J. M., Genetic Engineering News, 1994, 14, 16. [21] Nguyen, D. N., Becker, G. W., Riggin, R. M., Boyd, V. L., Bozzini, M. L., Yuan, P.-M., Loudon, G. M. Presentation Abstracts, gth Symposium of The Protein Society, Boston, MA, July 1995. Cambridge: Cambridge University Press, 1995. [22] Ambler, R. P., Methods Enzymol, 1972, 25, 262-272. [23] Ambler, R.P., Methods Enzymol, 1972, 25, 143-154. [24] Patterson, D. H., Tarr, G. E., Regnier, F. E., Martin, S . A., Anal Chem, 1995, 67, 3971-3978 [25] Biemann, K., Biomed Environ Mass Spectrom, 1988, 16, 99-111. [26] Burlett, O., Yang, C.-Y., Guyton, J.R., Gaskill, S . J., J Am SOC Mass Spectrom, 1995, 6, 242-247. [27] Naylor, S., Findeis, A. F., Gibson, B. W., Williams, D. H., J A m Chem Soc, 1986, 108, 63596363. [28] Siudzdak, G., Mass Spectrometry for Biotechnology. New York: Academic Press, 1996. 1291 Biemann, K., in: Biochemical Applications of Mass Spectrometry: Waller, G. R. (Ed.), New York: Wiley, 1971.
References
323
[30] Barber, M., Green, B. N., Rapid Cornmun Mass Spectrom, 1987, 1 , 80-83. [31] Barber, M., Bordoli, R.S., Elliott, G. J., Sedgewick, R. D., Taylor, A.N., Anal Chem, 1982, 54, 645A-657A. [32] Ashcroft, A. E., Chapman, J.R., Cottrell, J. S., J Chromatogl; 1987, 394, 15-20. [33] Biemann, K., in: Protein Sequencing: a Practical Approach. Findlay, J. B. C., Geisow, M. J. (Eds.), Oxford: IRL Press, 1989, pp. 99-118. [34] Cotter, R. J., Anal Chem, 1992, 64, 1027A-1039A. [35] NcNeal, C. J., The Analysis of Peptides and Proteins by Mass Spectrometry. Chichester: Wiley, 1988. [36] Fenn, J. B., Mann, M., Meng, C. K., Wong, S. F., Whitehouse, C. M., Science, 1989, 246, 64-71. [37] Kebarle, P., Tang, L., Anal Chem, 1993, 65, 972A-986A. [38] March, R. E., Hughes, R. J., Quadrupole Storage Mass Spectrometry. New York: Wiley, 1989. [39] Bruins, A.P., Covey, T. R., Henion, J. D., Anal Chem, 1987, 59, 2642-2646. [40] Whitehouse, C. M., Dreyer, R.N., Yamashita, M., Fenn, J. B., Anal Chem, 1985, 57, 675679. [41] Guzzetta, A. W., Basa, L. J., Hancock, W. S., Keyt, B. A., Bennett, W.F., Anal Chem, 1993, 65, 2953-2962. [42] Karas, M., Hillenkamp, F., Anal Chem, 1988, 60, 2299-2301. [43] Hillenkamp, F., Adv Muss Spectrom, 1989, 11, 354. [44] Cohen, S.L., Chait, B.T., Anal Chem, 1996, 68, 31-37. [45] Kaufmann, R., Spengler, B., Lutzenkirchen, F., Rapid Cornmun Mass Spectrom, 1993, 7, 902-910. [46] Vestal, M. L., Juhasz, P., Martin, S . A,, Rapid Commun Mass Spectrom, 1995, 9, 1044-1050.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
12 General Strategies for the Characterization of Carbohydrates from Recombinant Glycoprotein Therapeutics Gerrit J. Gerwig and Jan B.L. Damm
12.1 Introduction The past few years have seen an impressive increase of the number of biotechnologically produced glycoprotein-drugs approved for diagnostic andlor therapeutic use in humans (see Chapter 5). The carbohydrate moieties of these pharma-glycoproteins often play an essential role in their overall biological properties, e.g., pharmacokinetic and immunogenic behaviour in v i v a In first instance, it is important that the oligosaccharide moiety of recombinant glycoprotein does not differ significantly from its ‘natural’ human counterpart. However, the glycan structures are dependent on the expression system and cell culture conditions, commiting pharmaceutical companies to monitor batch-consistency and batch-uniformity of recombinant glycoproteins. On the other hand, by investigating the biological activities and the carbohydrate chain structures of recombinant glycoproteins with altered glycans, a new approach to elucidate the function of carbohydrate chains of glycoproteins has been opened. Understanding the influence of glycosylation on the structure and function of (recombinant) glycoprotein therapeutics, in most cases, still requires the full characterization of all oligosaccharides attached covalently to that glycoprotein. This characterization of glycosylation is generally seen as a relatively difficult problem, in particular because it demands for the use of highly sophisticated instrumentation. In this chapter, we describe the types and some functions of carbohydrate moieties occurring in glycoproteins, together with some biosynthetical aspects. Furthermore, we will discuss some general strategies for the isolation, fractionation and characterization of glycoprotein glycans.
12.2 Functions of Glycoprotein Glycans Many biological functions have been ascribed to oligosaccharides of the various classes of glycoconjugates (i.e., conjugates of carbohydrate and other biomolecules, such as proteins and lipids). From the literature data [l-51, it is evident that the functions of carbohydrate chains span the whole spectrum from effects on physico-chemical properties of the molecule to which they are linked to highly sophisticated func-
326
12 General Strategies for the Characterization of Carbohydrates
tions in cell-cell recognition, fine tuning of biological activity, and masking of functions. However, in spite of the accumulating data, it is not possible to denominate generalized functions for (specific types and/or structures of) carbohydrate chains. Rather, for each (recombinant) glycoprotein the effect(s) of glycosylation on all the properties of the molecule have to be determined. The various reasons why it is so difficult to unravel clear functions for oligosaccharides have been presented in Chapter 5. Bearing this warning in mind, several more or less, specific functions of carbohydrates can be mentioned. In general, the carbohydrate functions described below refer to, but are not necessarily limited to, protein-linked carbohydrate chains in higher animal species.
12.2.1 Transport, Stabilizing, Protecting and Structural Functions Already during the biosynthesis of glycoproteins the attached oligosaccharides can play a crucial role. A well-accepted function of oligosaccharide units of glycoproteins is the involvement in the initiation of correct polypeptide folding in the rough endoplasmic reticulum (rER), and in the subsequent maintenance of protein solubility and conformation. Many proteins that are incorrectly glycosylated fail to fold properly and, by consequence, fail to exit the ER and are degraded [6-81. The carbohydrate sequences also influence the further intracellular trafficking of the glycoprotein. The best known example is the Man-6 phosphate residue present in high-mannose-type N-glycans of newly synthesized lysosomal enzymes: the Man-6 phosphate residue is the label which targets the enzymes to their final destination in the lysosomes. Deficiency in phosphorylation impedes lysosomal targeting leading to I-cell disease and pseudo-Hurler polydystrophy [9,10]. Glycosylation of proteins, especially when multiple carbohydrate chains are attached, creates a ‘carbohydrate shell’ around the protein. For an impression of the dimensions of oligosaccharide chains versus the protein to which they are attached, the reader is referred to Chapter 5 (3D model of hCG). The carbohydrate cover may serve to protect the protein against recognition by proteases [ 1I] and/or antibodies [ 121. The stabilizing function (with respect to denaturation) of submaxillary gland mucin, and the protective function (against auto-digestion) of gastric mucus have already been described in Chapter 5 . It is widely recognized that carbohydrates are essential for the physical maintenance and functional integrity of various structural elements in the body, such as collagens and proteoglycans [ 131. Chondroitin sulfate and keratan sulfate chains, for instance, have a pronounced effect on the organization and tensile strength of cartilage [14]. Also well-known is the ‘glycocalyx’ covering the surface of whole cells, thus functioning as a structural and protective element. The porosity and functionality of ‘borders’ in the body formed by continuous membrane systems (e.g., the basement membrane and extracellular matrix of the vascular wall) is profoundly influenced by the presence and composition of glycoconjugates [15].
12.2 Functions of Glycoprotein Glycans
327
12.2.2 Storage Function Glycosaminoglycans, such as chondroitin sulfate, heparin, and heparan sulfate, localized in the extracellular matrix, tightly bind various growth factors, such as fibroblast growth factors and macrophage colony stimulating growth factor 1161. The glycosaminoglycan-mediated sequestration of growth factors warrants the concentration of growth factors at or near their target area and prevents diffusion of these biologically highly active molecules to ‘unwanted’ sites. Furthermore, after binding to the glycosaminoglycan chains the growth factors are less prone to proteolytic degradation. Several other storage functions have been reported, e.g., the binding of complement regulatory protein H to sialic acid residues on cell surfaces [ 171 (preventing the alternate pathway of complement activation), and the sequestration of calcium [ 181 and sodium [19] ions and water 1201.
12.2.3 Masking Functions Carbohydrates may serve to prevent infection by microorganisms or (unwanted) immune reactions by masking protein or other carbohydrate epitopes. Examples include: (i) the 4- and/or 9-0-acetylation of terminal sialic acid residues which prevent the recognition, binding and invasion of certain bacteria and viruses (influenza A and B) [21] that normally infect cells via sialidase-mediated recognition of terminal sialic acids of membrane bound glycoconjugates; (ii) the presence of heterogeneous carbohydrates on secreted mucins and in milk which inhibit the infection of the gut by microbial pathogens [22,23 1; (iii) terminal sialic acids on 0-linked oligosaccharides which mask the recognition of core 0-linked glycans by natural antibodies against T and Tn antigens [24]; and (iv) the presence of terminal sialic acid residues on glycoproteins and cell surfaces which prevent recognition by the asialoglycoprotein receptor and/or uptake by macrophages via the Gal/GalNAc lectin ~51. By contrast, carbohydrate sequences may also render the cells on which they occur vulnerable to infection, noxious agents, autoimmune responses or malignant disease: (i) terminal sialic acids and specific carbohydrate sequences of glycoconjugates on mucosal surfaces serve as recognition points for various viruses, protozoa, pathogenic bacteria, chlamydiae, mycoplasma and parasites (thus mediating infection) [26,27]; (ii) particular gangliosides on the cell surface can act as receptor for several bacterial toxins [28]; (iii) certain N-linked glycans on gastric parietal cell glycoproteins act as antigen for antibodies involved in autoimmune gastritis and pernicious anemia, causing mucosal atrophy and parietal cell loss [29]; and (iv) sialylated, fucosylated (po1y)N-acetyllactosamine-glycanson tumor cells act as ligands for selectin molecules on endothelial cells, thus probably enhancing the metastatic capability of the tumor cells 1301, to mention just a few examples.
328
12 General Strategies for the Characterization
of
Carbohydrates
12.2.4 Receptor Functions Several examples have been reported where oligosaccharides function as ligands for a receptor, thus mediating the interaction between biomolecules (as such or imbedded in larger cellular systems, like membranes) in the body. For instance, the interaction between oligosaccharide ligands and the selectin family of receptor proteins, mediating the adhesion of (stimulated) leukocytes or platelets to activated endothelial cells [31-331, and the interaction between B cells and activated T or B cells, mediated by the CD22P lectin of B lymphocytes [34]. Furthermore, cell-cell and cell-matrix interactions during development and tissue organization are proposed to be conferred by the mutual recognition of soluble or surface-bound pGal-binding lectins and N-acetyllactosamine units of the complementary cells [lo]. The role of oligosaccharides in the species-specific recognition of sperm and egg cell has already been discussed in Chapter 5. The interaction between heparan sulfate chains on cell surface proteoglycans and extracellular matrix proteins, such as fibronectin, laminin, and thrombospondin, has been implicated in cell adhesion, differentiation, spreading, or invasion [35,36].
12.2.5 Regulation of Clearance A long-known function of glycosylation is its effect on the circulatory half-life of soluble glycoconjugates and even whole cells. In general, the carbohydrate-mediated clearance of glycoproteins (drugs) from the blood is driven by lectins (carbohydrate receptors) that occur in cells/organs that are specialized in the removal of (partially degraded and/or non-functional or harmful) molecules or cells from the circulation. An example is the removal of pathogens by the macrophage Gal/GalNAc receptor that recognizes Gal/GalNAc-terminated glycans of endogenous and exogenous origin [37]. Several types of lectins occur in the liver: the asialoglycoprotein receptor which removes glycoproteins that bear exposed Gal residues [38], and the non parenchymal terminal 4-O-sulfate-@-GalNAcreceptor which removes certain glycoprotein hormones [39].
12.2.6 Tuning of Biological Activity One of the best documented funtions of protein-linked carbohydrate chains is the effect on the biological activity of the protein to which they are attached. These properties are defined as ‘tuning’ effects. Briefly, it can be stated that the oligosaccharides, depending on their structure and number, modulate the biological activity of the glycoprotein (enzymes, hormones, etc.) in subtle ways through effects on binding affinity (of the glycoprotein to its receptor), efficiency of signal transduction, circulatory half-life, and biodistribution. Many examples can be found in the literature
12.3 Types of Carbohydrate Chains in Glycoproteins
329
[3,40]. As a specific example, the influence of the carbohydrate chains on the biological activity of hCG has been described in Chapter 5 . The fact that certain terminal carbohydrate sequences on glycoconjugates act as blood-group determinants is commonly known. Easy access to the literature on the functions of glycans can be gained via the compendious review paper of A. Varki [41].
12.3 Types of Carbohydrate Chains in Glycoproteins Glycoproteins are biomacromolecules consisting of a polypeptide backbone with covalently attached carbohydrate side chains. These molecules may contain from 0.4% (by weight) to more than 80% carbohydrate (see Chapter 5). Three major classes of glycan chains can be distinguished: N-glycosidically linked carbohydrate chains, in which the reducing-end sugar residue (generally N-acetylglucosamine) is linked to the amide nitrogen of asparagine, - 0-glycosidically linked carbohydrate chains, in which the linkage is formed between the reducing sugar residue and an amino acid hydroxyl group,
-
Table 12-1. Covalent linkages between monosaccharides and proteins.
Class
Linkage
N-linked: GlcNAc+ Asn GalNAc- Asn Glc+Asn L-Rha+ Asn 0-linked: GalNAcjSedThr GlcNAc+Ser/Thr Man+Ser/Thr L-Fuc+Ser/Thr Gal+Ser/Thr Glc+Ser XylhSer ADP-ribmy lation: ADP-RibhNC- Arg ADP-Rib-tN6- Asn ADP-Rib-+N'-His ADP-Rib+O-C(0)Glu ADP-Rib+S-Cy s
glypiation: Man6+PO.+(CH2)2+NHK(O)+protein
Class
Linkage
other 0-linked: Glc+Tyr GalhOH-Ly s L-ArafhOH-Pro GalhOH-Pro GalhOH-His GlcA-OH-Trp GlcA+OH-Phe GlcA+OH-Ser S-linked:
GalhCys Glc-Xys glycation:
Glc+Lys Rib+Lys amide bond: GlcA/GalA(6+Na)Lys GlcA/GalA(6+Na)Thr/Ser GlcA/GalA(6-+Na)Ala MurNAc(3 +Na)Ala
330
12 General Strategies for the Characterization of Carbohydrates
Linkage
occwrence
StrLlrn 0
N-Glycosidic
II CWCOOH I
GlcNAoAsn
NY I
Widely distributed in animals, plants, and micro-organisms
NHAC
OGlycosidic GalNAc-Ser/Th
R 0-CH
Glycoproteinsfrom animal SoLlrCes
I NY
Protmglycans,
Xyl-ser
on 4 r
Y
-
{
I
O
O
H
human thymglobulin
OH
yN-CH-COOH
I
C b
I
Gal-Hyl
collagens w-NY
Plant and algal glycoproteins
Fig. 12-1. Some common carbohydrate-protein linkage types.
12.3 Types of Curhohydrute Chuins in Glycoproteins
33 1
- Glycosylphosphatidylinositol(GPI)-anchors (glypiation [42,43]). A (glyco)pro-
tein is commonly linked to the GPI anchor by its C-terminus through an amide bond involving the amine of an ethanolamine moiety. N- as well as 0-linked carbohydrate chains can occur at one protein backbone. In some glycoproteins, other, less common, covalent bonds involving N- or S-atoms have been found between the &-amino group of Lys and Glc or Rib (glycation [44]), between Rib and the side chains of Arg, Asn, or His residues (ADP-ribosylation [45]), and between Rib and Cys (S-glycosylation [46]). The established carbohydrate-protein linkage types are compiled in Table 12-1, and some common carbohydrate-protein linkage types are illustrated in Fig. 12-1.
12.3.1 Structure of N-linked Carbohydrate Chains In general, the N-glycans have a common pentasaccharide core, consisting of two GlcNAc residues and three Man residues, which can be extended with additional monosaccharide units giving rise to four types of N-glycosidic carbohydrate chains 140,471 (Fig. 12-2):
(or high-mannose) type, consisting of Man and GlcNAc. The number of Man residues linked to the pentasaccharide core generally varies between 0 and 6 in most mammalian glycoproteins, but in yeast over 100 Man residues can be attached. The oligomannose type includes also structures with up to three non-reducing-end terminal Glc extensions (see Section 12.4.1 ) and sometimes 0-phosphorylated Man residues. - N-acetyllactosamine (or complex) type, being composed of Man, GlcNAc, and NeuAc. The core is, in general, extended with two or more GalP1-4GlcNAc units (sometimes repeated) and terminated with NeuAc. In addition, Fuc, GalNAc, Xyl, and noncarbohydrate substituents like acetate, lactate, sulfate, phosphate, or methyl groups may occur [41]. In mammalian glycoproteins, di- to penta-antennary branching of the core structure is regularly found. - Hybrid type, combining the characteristics of the oligomannose and the N-acetyllactosamine type. In most hybrid type (and some complex type) structures, an extra GlcNAc residue (inter- or bisecting GlcNAc) is attached to the core 0-Man. - Xylose-containing type, in which the P-Man residue is substituted with 01-2 linked Xyl. Frequently, a l - 3 linked Fuc and occasionally a l - 6 linked Fuc is present at the Asn-bound GlcNAc residue. Small extensions on the a-Man residues may occur. These Xyl-containing carbohydrate chains are often found in higher plants. - Oligomannose
332
12 General StrateRies f o r the Characterization of Carbohydrates B
D3
Manal-2Manal-6
4,
Manal-6\
A
D2
Mana 1-2Mana 1-3
ManP l-4GlcNAcp l-4GlcNAc-Asn
Mana I-2Mana 1-2Manal-3 c
DI
/ 3
2
1
4
(oligomannose type)
8'
7,
Neu5Acn2-3Galp 1-4GlcNAcp1-6
~
I
tri' 1
Neu5Aca2-3Galp 1-4GlcNAcp 1 6'
5'
6
5
2
Neu5Aca2-3GalP 1-4GlcNAcP 1-2Mana 1-3
tri
NeuSAca2-3GalP I-4GlcNAcp 1-4 8
1
(N-acetyllactosamine type)
D3
B
Manal-2Mmal-6 DZ
A
d,
Mana1-6,
Manal-2Manal-3
Manp 1-4GlcNAcPl-4GlcNAc-Asn / 3
NeuSAca2-3Galp 1-4GlcNAcpI-4Mana 1-3 6
5
2
1
4
(hybrid type)
ManP I - ~ G I C N A C14GlcNAc-Asn !~ M a n a 1-3
1
4
XylP 1-2
(Xylose-containing type) Fig. 12-2.Examples of the four types of N-linked carbohydrate chains of glycoproteins. The common pentasaccharide core is shown in bold face. For the N-acetyilactosamine type the di-, tri-, tri'and tetra-antennary sub-types are indicated. The standard notation of the monosaccharide residues has been included.
12.3 Types of Carbohydrate Chains in Glyctiproteins
333
12.3.2 Structure of 0-linked Carbohydrate Chains The 0-glycosidic carbohydrate chains are classified according to the specific combination of the bond-forming amino acid and the sugar residues. In animal systems, the mucin (sub)type, in which the carbohydrate-protein linkage is formed between GalNAc and Ser or Thr, occurs frequently. The GalNAc residue linked to Ser/Thr may be extended with Gal, GlcNAc, or GalNAc, giving rise to at least eight established core structures [48,49] (Fig. 12-3). The peripheral sequences which are attached to the core structures are formed from GlcNAc, Gal, Fuc and NeuAc/NeuGc residues and often contain elements also occurring in N-linked oligosaccharides. Furthermore, the Gal and GalNAc residues may be sulfated. Apart from the mucin type, other types of 0-linked carbohydrates are known [50]. As indicated in Table 12-1, monosaccharides other than GalNAc may be linked to hydroxy amino acids. These sugar residues may also serve as core elements for further extension [51]. 0-glycosylation of glycoproteins with GlcNAc seems to be a dynamic modification of specific intraTY Pe
Core 1
5Pe
Structure GalNAc-Ser/Thr GalPl-3
core6
GIcNAcPl-6
Core 2
Core 5
Structure
GalNAc-Ser/Thr GalNAcal-3 GIcNAcPl-6
GalNAc-Ser/Tnr
GalNAc-Ser/Thr
GalPI-3 Core
GalNAcal-6
GalNAc-Ser/Thr ‘Ore
GalNAc-SedThr
GlcNAcPl-3 Core 8
GlcNAcP 1-6
GalNAc-SedThr Gala1 -3
GalNAc-Ser/Thr
Core 4
GlcNAcPl-3
Fig. 12-3. Eight core structures of mucin-type 0-linked carbohydrate chains of glycoproteins.
334
12 General Strategies for the Characterization of Carbohydrates
cellular proteins, in which addition and removal of 0-GlcNAc is a highly regulated modification [52].Based on recent observations for recombinant glycoproteins produced i n CHO cells, it can be assumed that 0-glycosylation is more widespread than hitherto recognized [53].
12.3.3 Structure of Glycosylphosphatidylinositol Anchors Glycosylphosphatidylinositol (GPI) anchors are widespread in eukaryotes, especially in parasitic protozoa. GPI anchors have been recognized as an important alternative mechanism for attaching various proteins to the cell membrane [54]. The GPI is anchored into the cell membrane by its lipid moiety and connected by its terminal ethanolamine phosphate, via a peptide linkage, to the C-terminus of a protein. Structural analysis of GPI anchors from different organisms has led to the proposal of an evolutionarily conserved core structure EtN-PO4-6Manal-2Manal-6Manal4GlcNH~a1-6PI[55,56] (Fig. 12-4). This conserved core can be modified by a wide variety of carbohydrate side chains and ethanolamine phosphate residues. In some GPI structures, acylation (palmitic acid) of the inositol ring occurs. Also the lipid moiety exhibits extensive variety, for instance, acylglycerols in Trypanosoma brucei, alkylacylglycerols in Leishmania, and ceramides in Dictiostelium, Saccharomyces and Paramecium [55].
Ri + 2Manal-2Manal
L6
R4 + 4Manal-4GlcNHzal -6myo-Inositol-l-PO~-LIPID
r3 2T R3
Rz
r"
Rs
Fig. 12-4. Glycosylphosphatidylinositol anchor structure with evolutionarily conserved glycan backbone. Some frequently occurring substituents are: R1, aMan, aGlc; Rz,ethanolamine phosphate; R3,aGal2.4; R4, [aGlc]+/.-flGalNAc, [pGal]+/.-flGalNAc; Rs, palmitate; R g , ethanolamine phosphate.
12.4 Biosynthesis of Glycoprotein Glycans The biosynthesis of protein-bound oligosaccharides differs in several aspects from the biosynthesis of nucleic acids and proteins, which are synthesized as linear molecules with one type of linkage through a template-directed mechanism. Oligosaccharide chains can be branched as well as linear, and their monomeric units are connected
12.4 Biosynthesis of Glycoprotein Glycans
335
to one another by many different linkage types. Oligosaccharides are not biosynthesized by a template-directed mechanism but indirectly by the concerted action of highly specific glycosyltransferases, enzymes which are under genetic control. A consequence of the lack of a template-directed mechanism for oligosaccharide synthesis is the existence of several different oligosaccharide chains at the same glycosylation position, designated as microheterogeneity. Therefore, glycosylation of a protein usually generates a set of glycoforms, all of which share an identical polypeptide backbone but are dissimilar in either the structure or disposition of their oligosaccharide units or both. The formation of glycoforms is not necessarily random but can be genetically regulated and is highly reproducible within one cell type under constant physiological conditions.
12.4.1 N-linked Carbohydrate Chains The biosynthesis of N-glycans begins in the rER and is similar in lower and higher animal species as well as in the plant and fungal kingdoms. However, the ultimate structures of the mature carbohydrate chains vary enormously for a certain glycoprotein and are mostly species-, tissue-, organ-, and even cell-type-specific [43,57,58]. A lipid-linked oligosaccharide intermediate (GlcSMangGlcNAcz-PPDol) is constructed by sequential addition of individual sugar residues to dolichol pyrophosphate (dolichol phosphate cycle). Seven sugar residues (underlined in Fig. 12-5) are added directly from UDP-GlcNAc and GDP-Man on the cytoplasmic side of the ER, whereas the other residues are added from Man-P-Do1 and Glc-P-Do1 GlcNAc2-PP-Dol
t Man ~I.GlcNAc2-PP-Dol
Man,a-zMaopl~GlcNA~-PP-Dol
Protein I
c
Manal-ZManal
,3-
,$&I pl-4G~Ac~l-4GlcJAcal-PP-Dol
Fig. 12-5. The dolichol phosphate pathway leading to N-glycosylation of proteins. The superscript letters indicate the probable order (alphabetical) of addition of a-Man residues.
336
12 General Strategies for the Characterization of Carbohydrates
in the lumen of the ER [59-611. The entire pre-assembled oligosaccharide is then transferred en bloc under the catalytic action of an oligosaccharyl transferase [62,63] to the acceptor amino acid (Asn) located in an Asn-X-Ser/Thr/(Cys) sequence within the polypeptide (X can be any amino acid except Pro) [64,65]. It has to be noted that not all consensus sequons are glycosylated and that probably additional secondary and tertiary structural elements of the protein are required to realize glycosylation [66-691. Once transferred to the protein, the carbohydrate chain undergoes a series of trimming (within the lumen of ER and cis-Golgi) and elongation steps (within the medial- and trans-Golgi) before obtaining its final structure. This processing takes place through the interactive, stepwise action of glycosyltransferases and exoglycosidases [6 1,70-721. A schematic representation is shown in Fig. 12- 6. Glycosyltransferases transfer monosaccharide residues from an activated donor, usually a nucleotide sugar, to the growing oligosaccharide chain. Exoglycosidases commonly remove monosaccharide units one by one. Processing of N-linked oligosaccharides is controlled by several factors such as the genetic control of
oligomannose type
I
hybrid type
1
v+yLy++J -Asn-
--.-.-Asnmulti-antennary
-Asn-di-antennary
-Asnmono-antennaw
~acIyIIactosaminetype
I
Fig. 12-6. Simplified scheme of the major pathway for the assembly of N-linked carbohydrate chains on newly synthesized glycoproteins. The figure indicates the topography of certain steps and the origin of oligomannose-, hybrid- and (mono-, di-, and multi-antennary) N-acetyllactosamine-type glycans. Involved enzymes are: a, a-glucosidases; b, a-mannosidases; c, N-acetylglucosaminyltransferase I; d, a-mannosidases; e, N-acetylglucosaminyltransferase 11; f, fucosyltransferase; g, N-acetylglucosaminyltransferase IV; h, !3-galactosyltransferase; i, sialyltransferases.
12.4 Biosynthesis of Glycoprotein Glycans
337
enzyme expression, the intracellular localization of processing enzymes, the availability of substrates, and the substrate and acceptor specificity of the enzymes [73-751. The exact control mechanism of all enzymes is not precisely understood and their action leads to different oligosaccharide structures synthesized at the same site on a protein, a phenomenon known as microheterogeneity. In some cases, sulfate, phosphate and/or Fuc residues may be added. Finally, the mature glycoprotein exits from the trans-Golgi network in membrane-bound vesicles. The contents of these vesicles are either secreted out of the cell, delivered to the plasma membranes as membrane glycoproteins (eventually linked to GPI anchors), or targetted to other organelles inside the cell [76].
12.4.2 0-linked Carbohydrate Chains The biosynthesis of 0-linked oligosaccharides is a post-translational process wherein carbohydrate is linked to hydroxyl groups of hydroxy amino acids within the polypeptide. Almost any amino acid bearing a hydroxyl group, including the less common hydroxyproline and 3 -hydroxylysine, can be 0-glycosylated. The biosynthetic data of 0-glycans are mainly collected from studies of mucin-type structures [77791. 0-glycan biosynthesis is initiated by GalNAc transfer to peptide, mainly in the cis-Golgi. Subsequently, the 0-linked carbohydrate chains are mainly synthesized in the medial- and trans-Golgi by direct transfer of single monosaccharide residues from sugar nucleotides to the growing oligosaccharide chain, following pathways defined by the sequential action of (competing) glycosyltransferases [2,80]. Hence, the heterogeneity found in 0-glycans is due to competition of glycosyltransferases for the same acceptor and the relative amounts of the various glycosyltransferases produced by a cell. Several core structures have been established which are substrates for further elongation [49] (Fig. 12-7). In contrast to the Asn-X-Ser/Thr
1A t \ Gal PI-3GalNAc-R
/:!
GlcNAc PI-6GalNAc-R
GlcNAc PI-3GalNAc-R (core 3)
-\
GlcNAc p l -
GlcNAc pl-6, /GalNAc-R Gal pl-3 (core 2)
1
"\ ,GalNAc-R
GlcNAc pl-3 (core 4)
Fig. 12-7. Biosynthetic pathways of some 0-glycan-core structures. Paths along solid lines are well established in mucin biosynthesis and paths along dotted lines can occur but are very slow. Some of the involved enzymes are: a, core 1 83-Gal-transferase; b, core 2 !36-GlcNAc-transferase; c, core 3 03 -GlcNAc-transferase; d, core 4 86-GlcNAc-transferase. R = Ser/Thr.
338
12 General Strategies for the Characterization of Carbohydrates
sequon for N-glycosylation, no specific amino acid sequence has been established for general O-glycosylation, although the substrate site is favored by the presence of relatively high concentrations of Pro, Ser, and Thr, and disfavored by strongly hydrophobic or hydrophylic residues. Notably, it has been found that Fuc can be linked to Ser/Thr within an apparently conserved sequence -Cys-X-X-Gly-Gly-Thr/ Ser-Cys- (X = any amino acid), occuring in epidermal growth factor domains of several proteins [81]. The synthetic pathways for O-glycans are usually classified according to the types of linkage of carbohydrate to protein: GalNAc-O-SeriThr in mucins and mucin-type 0-glycans, Xyl-O-Ser/Thr in proteoglycans, Gal-O-hydroxylysine in collagens, GlcNAc-O-Ser/Thr in nuclear and cytoplasmic proteins, and Man-O-Ser/Thr in yeast proteins [73]. The sugars commonly found in O-glycans are GalNAc, Gal, GlcNAc, NeuAc and Fuc. In mucins, the O-glycans are usually found in clusters on the polypeptide.
12.4.3 Glycosylphophatidylinositol Anchors The biosynthesis of the GPI precursor mainly takes place in the lumenal face of the endoplasmic reticulum of the cell and is broadly similar in protozoan parasites and mammalian cells 182-841. In general, it is assumed that the process involves the sequential transfer of monosaccharides. First, GlcNAc from UDP-GlcNAc is transferred by an a-GlcNAc transferase to phosphatidyl inositol (PI), followed by deN-acetylation to glucosamine-PI. Subsequently, three a-Man residues coming from Man-P-Do1 are added in single steps [85,86]. Ethanolamine phosphate from phosphatidyl ethanolamine is transferred to the terminal mannose. In mammalian cells, at least one extra ethanolamine phophate group is added and also, palmitoylation of inositol occurs early in the assembly of the GPI precursor [87]. Then, depending on the cell species, a complex series of fatty acid remodeling reactions can take place. The precise nature and composition of the lipid moiety (diacylglycerol, monoacylglycerol, alkylacylglycerol, ceramide) vary with protein and cell type. Additional a-Man residues, linked to the ethanolamine phosphate substituted terminal Man, are a common modification found in both lower and higher eukaryotes [88]. The final structure of the GPI anchor is depicted in Fig. 12.4. (Section 12.3.3). It is most probable that the glycosyltransferases involved in the GPI biosynthesis are different from those involved in N,O-glycoprotein synthesis [89,90]. The knowledge of the biosynthetic pathway still shows some major hiatus concerning the time and site of addition of extra sugar residues (e.g., Gal, Glc, Man, GalNAc, NeuAc) occurring in some protozoan species, and the precursors and enzymes involved. A (g1yco)protein intended for anchoring to GPI must contain an N-terminal signal sequence for entry into the lumen of the endoplasmic reticulum and a GPI-signal sequence at the C-terminus. The GPI-signal peptide is cleaved and the newly exposed a-carboxyl group is directly attached to the pre-assembled GPI precursor via an amide bond involving the amine of ethanolamine [54,9 11. However, several protozoa also synthesize free GPIs which are not covalently linked to protein and which appear to be metabolic end-products.
12.5 Analysis
of
Glycoprotein Glycans
339
Due to the dramatic increase in studies of glycoconjugates in biological processes (glycobiology), a growing number of carbohydrates are now being reported which differ from the ‘standard’ N,O-glycans, in terms of their structure and protein linkage as well as their location in cell compartments [50]. Novel carbohydrate chains 0linked via Fuc or Glc found in coagulationfactor glycoproteins [92,93], 0-linked GlcNAc found on a wide variety of cytoplasmic and nucleoplasmic proteins [5 1,52,94], and the discovery of mitochondria1 glycoproteins [95,96], are just a few examples. Studies on the biosynthesis of these ‘new’ glycoconjugates are still in a preliminary stage.
12.5 Analysis of Glycoprotein Glycans 12.5.1 General Aspects The primary structure analysis of glycoprotein glycans has still not reached the level of routine analysis and remains a highly specialized and laborious endeavour. No single technique is able to provide all information needed for the exact structural definition of a complex glycan. The structural analysis depends on the combined use of several physical, chemical, and biochemical techniques [97,98]. To date, several approaches based on H-nuclear magnetic resonance (NMR) spectroscopy, mass spectrometry, enzymatic procedures, and profiling techniques are available but these current approaches have to be applied with great care. Since, in general, (recombinant) glycoproteins contain multiple-glycosylation sites and the carbohydrate chains usually exhibit considerable structural heterogeneity, it is (almost) not possible to analyze complete glycan structures on the intact glycoproteins. However, the rapid progress of multi-dimensional techniques in NMR spectroscopy and mass spectrometry implicate prospects for the future. Nevertheless, at the moment, a more practical approach is to release the attached carbohydrate chains from the protein, purify the oligosaccharide pool from the deglycosylated protein and reaction additives, and fractionate to individual, pure oligosaccharides, which can be identified using different chemical and spectroscopic methods. Methods for the structural analysis of (recombinant) glycoprotein glycans have been reviewed regularly [99-1051. For detailed laboratory protocols of all the main methods currently used to elucidate the structure and biosynthesis of glycoproteins, the reader is referred to the book Glycobiology: A Practical Approach [ 1061. A general strategy for characterizing and analyzing (recombinant) glycoprotein glycans applicable on most glycoproteins is summarized in Fig. 12-8. A good approach to obtain separately N- and 0-glycans from N,O-glycoproteins is the enzymatic release of the N-glycans, followed by a chemical release of the 0-glycans [107]. But first, in order to get an overview of the carbohydrate portion of a protein the intact glycoprotein is subjected to monosaccharide analysis. The determination of the monosaccharide composition is currently established after methanolysis and analysis of volatile derivatives by gas chromatography(-mass spectrometry) [ 108,109].
’
340
12 General Strategies for the Characterization of Carbohydrates
monosaccharide
4
b,Ggbcoproteinb 1
4 PNGa'se-F i
analysis
-&
Analysis of intact N.Oekcoerotein lH-/i*-NMR (2D) ES-MS MALDI-MS
GPCBio-GeYSuperdex
I
1
FPLCResource Q
t
alkaline horohydride
ProfilingMapping
I
t
t
FPLCResource Q
AC/pon-A
i
'H-NMR, monosaccharide I
t
HPLCLichrosorb-w
4
+
H-NMR HPAEC/Carbo Pac
4
'H-NMR monosaccharide analysis methylation analysis MS exo/endo-glycosidases lectins
Fig. 12-8. A general strategy for the release, isolation, fractionation, and analysis of (recombinant) glycoprotein glycans.
This method allows simultaneous analysis of neutral monosaccharides, acetamido sugars, and sialic acids. A typical gas chromatogram is presented in Fig. 12-9. By using an internal standard, information about the carbohydrate content is also obtained. The occurrence of relatively large amounts of GalNAc and/or Man indicates the presence of 0- and/or N-linked carbohydrate chains, respectively. It is obvious that monosaccharide analysis can also be performed on isolated pure glycans to determine the molar ratios of the constituent monosaccharide residues of an individual
Fig. 12-9. Monosaccharide composition analysis of glycoproteins by gas chromatography-mass spectrometry. (A) Gas chromatogram of a standard mixture of trimethylsilylated (methyl ester) methyl glycosides on a CP-Sil 5 column (25 m X 0.32 mm). Oven temperature program: 130" to 230 "C, at 4 "C min-'. Each monosaccharide gives rise to multiple peaks due to dp- and pyranose@)/furanose(n forms. (B) Gas chromatogram of the constituent monosaccharides of recombinant human erythropoietin (rhEPO), after methanolysis/re-N-acetylationltrimethylsilylation.(C, D, E, and F) (GC-)EI-mass spectra of the trimethylsilyl (methyl ester) methyl glycosides of aFucp, aManp, aGlcNAcp, and pNeuAc, respectively. IS = internal standard (mannitol).
341
12.5 Analysis of Glycoprotein Glycans
f
1003
'"1 C
'p" ._1
50
100
'"1 D
,
24q'
IL
J1Y
1
300
200
133 147 103 , . I 1
50
1. 1
335
363379 350
m/z 400
450
500
73
.
%-
0
150
CH,
+*lo
'
'"1 E
100
',
305 1-
191 I,
.,., 150
173
TMSO
I
LVY
345
200
250
I
300
350
I
435
377
400
450
mlz 500
-"'"314
CH,OTMS
--"
: 200
*no
250
350
400
450
m/z
500
An,
342
12 General Strategies for the Characterization of Carbohydrates
carbohydrate chain. The absolute configuration (D or L) of the monosaccharides are eventually determined by gaschromatography of their (-)2-butyl glycosides [ 1lo]. The presence of a GPI membrane anchor on a protein can be inferred by analysis for the components ethanolamine and myoinositol using amino acid analysis and GC-MS, respectively [ 111,1121. Another indication for the presence of GPI anchor is to analyze selectively for fatty acid components by releasing them with alkali, acidify to protonate the fatty acids, partition into organic solvent, and derivatize to the fatty acid methylesters for GC(-MS) analysis. For a detailed structural characterization of the GPI glycan moiety the protein must first be split off from the GPI.
12.5.2 Release of Carbohydrate Chains from Proteins The liberation of the carbohydrate moieties from glycoproteins can be achieved in different ways depending on the type of linkage of the glycan to the protein. Commonly used procedures to generate carbohydrate chains are summarized in Table 12-2. Alternatively, proteolytic digestion of the glycoprotein can be used to obtain glycopeptides which, after subsequent separation, can also give information about the glycosylation sites of the protein. The use of this method however, can be limited by the distribution of the sugar chains, since steric hindrance imposed by the carbohydrates may impair complete degradation of the peptide backbone, leaving adjacent carbohydrate moieties unseparated. Table 12-2. Procedures to generate partial structures, representing one glycosylation position. Proteolytic and chemical cleavage of glycoproteins leads to the formation of glycopeptides, whereas the other procedures liberate carbohydrate chains from the protein. Conversion of reducing constituent monosaccharide residues into a fluorescent derivative or its corresponding alditol may follow the liberation of the carbohydrate chains. Type of linkage N-gl ycosidic
0-glycosidic
Proteolytic digestion Chemical protein degradation Hydrazinoly sis" Alkaline borohydride treatmenta Enzymatic hydrolysis
Proteolytic digestion Chemical protein degradation Hydrazinoly sis" Alkaline borohydride treatment" Enzymatic hydrolysisb
a
Different conditions are applied for the release of N- and 0-glycosidic carbohydrate chains. The enzyme generally used (endo-a-N-acetylgalatosaminidase), has a rather restricted specificity.
12.5 Analysis of Glycoprotein Glycans
343
Chemical Methods Hydrazinolysis [113] is a chemical method for releasing glycans. A fully automated hydrazinolysis procedure is achievable by the GlycoPrepTM 1000 (Oxford GlycoSciences) [ 114,1151. The chromatographic procedures necessary for recovery of the carbohydrate pool purified from amino acids, peptides (stemming from the degraded protein), and reactants, is integrated. Although different reaction conditions for N- and O-glycosidic linkage cleavages have been developed by Oxford GlycoSciences there is always mutual contamination of both types of carbohydrates. A further disadvantage of hydrazinolysis is the (limited) occurrence of side reactions leading to degradation or conversion of the carbohydrate chains [ 1161. This can complicate and compromise the structural analysis because of the introduction of additional, artificial heterogeneity. Notwithstanding the foregoing, a clear advantage of the automated hydrazinolysis procedure is its potential for relatively high throughput carbohydrate analysis, which makes it suitable for analysis of batch-to-batch consistency of (recombinant) glycoproteins. A chemical method, most often used for specific release of O-linked glycans is mild alkaline borohydride treatment (p-elimination reaction) [107,117]. To avoid peeling reaction at the reducing terminal, sodium borohydride is added to the reaction mixture to convert the oligosaccharides to the corresponding sugar alcohols (alditols) as soon as they are released from the polypeptide [ 1181. All chemical methods result however in various modifications of the released oligosaccharides. For the removal of glycans from the glycoprotein/peptide with the aim of recovering the intact proteinlpeptide for sequence analysis, the use of anhydrous trifluoromethanesulfonic acid (TFMS) has been found to be most successful [105]. N- and O-glycans are cleaved non-selectively leaving the primary structure of the protein/ peptide intact, but the Carbohydrate chains are destroyed.
Enzymatic Methods Alternatively, carbohydrate chains can be released from the glycoprotein by enzymatic procedures. Two important types of N-glycan-releasing enzymes are known, namely the endo-0-N-acetylglucosaminidases (further referred to as endoglycosidases), which hydrolyze the GlcNAcP 1- 4GlcNAc linkage in the N,N’-diacetylchitobiose unit of Asn-linked carbohydrate chains and the peptide-N4 -(N-acetyl-p-glucosaminyl) asparagine amidases (PNGases), which hydrolyze the p-aspartyl glycosylamine linkage. Specific studies [ 119-1231 have shown that the glycan specificity of the endoglycosidases is rather rigid, which renders these enzymes unsuitable for obtaining a complete profile of the Asn-linked oligosaccharides. For instance, endoglycosidase H (Endo-H) from Streptomyces plicatus cleaves only oligomannose- and hybrid-type N-glycans. Two different amidohydrolases, namely PNGase-A (from almond emulsin) and PNGase-F (from Flavobacterium meningosepticum) are frequently used to split off intact N-linked oligosaccharides. Before using these enzymes, in some cases, the protein must be denaturated to ensure that the N-glycosidic bond is accessible [ 124,1251. PNGases have however a unique
344
12 General Strategies for the Characterization of Carbohydrates
_ _ _ _ _ _ _ ~
resistance to denaturating substances like chaotropic agents, detergents, and disulfide reductants. The most important functional feature that PNGases-A and -F have in common, is the ability to release high-mannose, hybrid, as well as complex types of Asn-linked oligosaccharides provided that: (i) the glycosylation site is accessible to the enzyme; (ii) both the C - and N-terminal group of the Asn-residue bearing the carbohydrate chain are in peptide linkages; and, (iii) the carbohydrate chain attached to a glycoprotein comprises at least two residues [ 1261. Nonetheless, following digestion, it is always advisable to check (e.g., by SDS-PAGE or MALDI-MS) that no N-glycan remains associated with the protein since the enzyme frequently does not release all glycans with the same efficiency (time dependance). Furthermore, it is known that PNGase-F does not work when the Asn-linked GlcNAc bears a Fucal-3 residue [127,128]. Until recently, the enzymatic release of 0-linked glycans was restricted to only one type. Endo-a-N-acetylgalactosaminidase (from Streptococcus pneumoniae) releases only Galpl-3GalNAc from Ser/Thr in the polypeptide. Substitution of the disaccharide by NeuAc, Fuc, or GalNAc residues or the absence of the Gal residue
inositol +Iester-linked palmitic acid
de-acylation
@
a
glucosamine tri-mannose core +substitnents N-ncetylglncosnmine
HF dephosphorylation
nitrous acid deamination & borohydride reduction
N-acetylation
Fig. 12-10. A general procedure for generating neutral glycan fragments from GPI anchors.
345
12.5 Analvsis of Glvcourotein Glvcans
abolishes hydrolysis. In addition, an endo-N-acetylgalactosaminidasefrom Streptomyces has recently been shown to release more complex 0-glycans [129]. In order to analyze the GPI anchor, a combination of chemical and enzymatic methods is used depending on the structural information desired [42,82]. In general, the first step in the characterization of GPIs is the enzymatic removal of the protein which can be linked via ethanolamine phosphate to the GPI glycan. Proteases (e.g., pronase, trypsin, papain) are used to this end. Phosphate groups are removed from the glycans by treatment with aquous HF yielding the GPI core glycan free from protein and lipid. The lipid portion can also specifically be released by enzymatic cleavage with phosphatidylinositol-specific phospholipase C from Bacillus (PIPLC) or GPI-specific phospholipase D (GPI-PLD). Palmitoylation of inositol renders the GPI structure resistant to cleavage by these enzymes. As an alternative for the last step, nitrous acid deamination can be applied which cleaves the linkage between myoinositol and glucosamine (converting glucosamine into 2,5 -anhydromannose). A general procedure [54,111,130] for generating neutral glycan fragments from GPIs involves deacylation to iimprove solubility, dephosphorylation with ice-cold aquous HF and conversion to neutral species via N-acetylation or nitrous acid deamination/sodium borohydride reduction as depicted in Fig. 12-10. For isolation and fractionation of intermediates, several chromatographic procedures (described below for N,O-glycans) including HPAEC, are used. Subsequently, the resulting glycan fragments can be analyzed by 'H-NMR spectroscopy as demonstrated in Fig. 12-11 for the glycan core derived from Toxoplasma gondii GPI [130].
Mana 1-2Mana 1-6
lllll
I
NAc
Glca 1 -4GalNAcP 1-4Manal-4AHM-ol Anorneric protons (H-I) Man 2
I
3
AHM-ol
GalNAc
H4
5.2
5.0
4.6
4.4
4.2
IJ
ppm 4.0
x 'I,
3.8
3.6
3.4
2.10
2.05
Fig. 12-11.500-MHz 'H-NMR spectrum of the core glycan derived from GPI of Toxoplasrna gondii. The HOD signal (4.65-4.85 ppm) has been omitted. The relative scale of the NAc proton region differs from that of the rest of the spectrum. The spectrum was recorded in D20 at 300 K.
346
12 General Strategies for the Characterization of Curbohydrates
12.5.3 Isolation and Fractionation of Oligosaccharides Usually, after enzymatic release of the N-glycans from the protein, a gel filtration step (Bio-Gel P-100 or Superdex 75) is sufficient to separate the glycan pool from the residual high-molecular-mass protein. Direct structural analysis of the released oligosaccharides is mostly impeded by the heterogeneity of the sample and thus fractionation is essential. Separations are accomplished by exploiting differences in mass, charge, and hydrophilic properties of the oligosaccharides. Preparative and/ or analytical separation of carbohydrate chains can be carried out using a variety of chromatographic or electrophoretic techniques (Table 12-3), including gel permeation chromatography, several types of HPLC (anion-exchange, amine-adsorption, and reversed-phase), high-pH anion-exchange chromatography (HPAEC), and affnity chromatography on immobilized-lectin columns. An important feature during fractionation is the detection method. Although, most glycans have (low) UV activity, for analytical purposes and especially when limited amounts of material are available, it is sometimes required to label the glycans. Commonly used methods are reductive amination with a fluorescent compound such as 2 -aminobenzamide (2-AB) [131] or 2-aminopyridine (PA) [132], and reduction with alkaline sodium borotritide [ 1331 for highly sensitive detection. Today, special developed glycan labeling and detection kits, eventually already in an automatical application, are on the market (Oxford GlycoSciences, Beckman, Bio-Rad, Takara Shuzo). Table 12-3. Several fractionation methods and detection techniques used in the analysis of glycoproteins glycans. Separation methods Detection methods Chromatography
Electrophoresis
Affinity" Amine-adsorption Anion-exchange Gas-liquidc Gel-filtration High-pH anion exchange Paper Reversed-phase Thin-layer
Capillary-zone Paper Polyacrylamide slab gel
" Based
Fluorescenceb Mass spectrometric Pulsed amperometric Radiochemicald Refractive index Ultra-violet absorbance
on lectins or antibodies. Glycans are tagged with a fluorescent agent. Used in combination with mass spectrometric or flame-ionization detection; fractionated components are not recovered. A radioactive isotope is introduced into the glycans.
12.5 Anulysis o j Glycoprotein Glycans
347
Separation According to Charge Traditional methods like paper electrophoresis and low-pressure anion-exchange chromatography are more and more replaced by medium- and high-pressure ionexchange chromatography, using FPLC and HPLC systems, respectively. Chromatography on a Mono Q or Resource Q anion-exchange column (Pharmacia FPLC system) offers a fast and reproducible separation method for sialyl-oligosaccharides, based on the sialic acid content [ 134,1351. However, the molecular mass of the compounds having the same number of sialic acid residues has a minor influence on the elution position. Separation according to the number of acidic (e.g., sulfate or carboxyl) groups can also be achieved by HPLC on MicroPak AX-5 and AX-10 [ 1361 and Lichrosorb-NHz [ 1371 columns. More recently, high-performance anionexchange chromatography with pulsed amperometric detection (HPAEC-PAD) has been introduced for the separation and sensitive detection (10-100 pmol) of linkage and branch positional isomers of both neutral and acidic oligosaccharides (Dionex BioLC system with polymer-based pellicular column) [ 138,1391. In this application charge is introduced by ionization of hydroxyl groups of the oligosaccharide chains at alkaline pH (1 2-14). Correlation of the retention times with the structure of different oligosaccharides suggests that the accessibility of the readily ionizable hydroxyls to the quaternary amine stationary phase is a major determinant of the enhanced resolution [140,141].
Separation According to Mass For fractionation of oligosaccharides according to their molecular mass, size exclusion chromatography (gel filtration) on Bio-Gel can be used [142]. Bio-Gel P-4 is most suitable for separation of (preferably neutral) oligosaccharides ranging in size equivalent to up to 24 glucose residues [133]. More recently, high-performance size exclusion chromatography using large pore polymeric column materials was introduced.
Separation According to Hydrophilic Properties and Structure By HPLC on chemicaly modified silica, higher resolution can be obtained than by chromatography on soft gels. Reversed-phase HPLC, which depends on hydrophobic interaction between the sample and the C-18 stationary phase, has been used for the separation of a large array of neutral oligosaccharides [ 143,1441. Reversed-phase HPLC can also be applied to the fractionation of charged oligosaccharides by use of ion pairing reagents like triethylamine [145]. In normal-phase HPLC, using Lichrosorb-NHz, the retention of carbohydrates on the amino-bonded silica is probably based on hydrogen bonding between the hydroxyl groups of the sugar and the amino group of the stationary phase. Alkylamino bonded silica columns have, compared with HPAEC, low-resolution properties for neutral and charged N- and O-glycans [142,146]. It has to be noted that the spatial structure of the carbohydrate chain determines the accessibility of the carbohydrate submolecular components to the col-
348
12 General Strategies for the Characterization of Carbohydrates
umn stationary phase, thereby influencing its chromatographic behavior in both normal- and reversed-phase HPLC. Recently, some new semi-automated procedures based on chromatographic techniques are developed for (fast) structural analysis of glycoprotein glycans and these will be discussed in Section 12-6.
Separation According to Affinity Differences Towards Lectins Numerous lectins have been purified and in some cases their properties are well defined. Their specific affinity for distinct sites in carbohydrate chains can be exploited for the isolation and purification of a wide range of oligosaccharides (and glycoconjugates) by affinity chromatography. Many examples can be found in the literature [147-1511. However, the binding specificity of lectins is in general complex, being determined in the first instance by the nature of the monosaccharides and their glycosidic linkages, but furthermore by steric factors, and in case of glycoconjugates even by non-specific interactions between the lectin and the carbohydrate-containing macromolecule. The lectin binding specificity is constantly redefined as the knowledge in this field increases and however positive this may be in itself, it may influence the benefits of lectin affinity chromatography for oligosaccharide purification procedures.
12.5.4 Structural Characterization of Oligosaccharides The final goal in the analysis of carbohydrate chains is to determine the number, nature, order, and ringconformation (D/L and pyranose/furanose) of the constituent monosaccharide residues of the purified oligosaccharides. Furthermore, the anomericity ( a or p) and the substitution pattern of the individual monosaccharides must be determined. Finally, the nature and localization of chemical substituents (e.g., 0methyl, acetate, lactate, sulfate, phosphate) potentially present on a given monosaccharide residue must be determined. Ultimately, this all leads to the description of the complete carbohydrate chain in a structural formula. Since several biological functions are ascribed to carbohydrate chains as functional parts of glycoproteins (see Section 12.2), conformational analysis of these chains becomes more and more important [152]. In order to understand how carbohydrate chains are involved in biological processes, determination of the threedimensional structures of these biomolecules is a crucial step. In the last decade an increasing effort has been made to establish three-dimensional structures of oligosaccharides as part of glycoproteins. Experimental data of carbohydrate chain conformations in solution are mainly obtained by advanced (multi-dimensional) NMR spectroscopy methods (e.g., NOESY, ROESY), which are compared with computer models obtained by theoretical methods (e.g., Molecular Dynamics simulations and Molecular Mechanics calculations). A detailed description is beyond the scope of this chapter, which is focused on primary structure determination.
12.5 Analysis of Glycoprotein Glycans
349
Enzymatic and Chemical Analytical Methods
A powerful method for determining the monosaccharide sequence of carbohydrate chains is the sequential exoglycosidase digestion procedure [ 133,153,1541. Exoglycosidases are hydrolases that cleave monosaccharide residues from the non-reducing terminal of the carbohydrate chain. Most of the enzymes are highly specific toward their substrate, including the anomeric configuration. They are named according to their specificity, e.g., 0-galactosidase cleaves the 0-galactosyl linkage. Frequently used exoglycosidases are: a-mannosidase, P-galactosidase, 0-N-acetylhexosaminidase; a-fucosidase and sialidase [101,153]. It has to be noted that aglycon specificity also plays a role for certain exoglycosidases. The principle of sequencing involves the stepwise degradation of the carbohydrate chain. The unknown, usually tritium-labeled, pure glycan is digested with an exoglycosidase and the remaining glycan is analyzed to determine if monosaccharides have been cleaved. After recovering of the digested glycan, it is subjected to another exoglycosidase. The amount of monosaccharide released by each exoglycosidase digestion can be estimated by analyzing the change in effective size of the glycan by gel permeation chromatography (Bio-Gel P-4 column) before and after the enzymatic digestion [155]. This procedure is repeated until the entire glycan sequence is revealed. Enzymatic analyses can be performed on small amounts (150 pmol) of labeled sample. As the method relies on the substrate specificity of each enzyme, special care must be taken to avoid contamination of each enzyme with other exoglycosidases. Methylation analysis is the most reliable chemical method for the elucidation of the substitution pattern of the individual monosaccharide residues. It involves the methylation of all free hydroxyl groups of the oligosaccharide followed by the liberation of the methylated monosaccharides by hydrolysis and subsequent qualitative and quantitative analysis of volatile derivatives. The position of the free hydroxyl groups of the partially O-methylated monosaccharides indicate the positions in which the sugar residue was substituted. Effective and complete methylation is essential for obtaining reliable results. Many methylation methods have been reported (see review [156]), but Hakomori’s procedure [157] using a mixture of methylsulfinylcarbanion and methyl iodide as the methylation reagent turned out to be most used. Usually, the hydrolysis products are converted into partially 0methylated alditol acetates by reduction with sodium borohydride, followed by acetylation with acetic anhydride [ 1581. Separation, identification, and quantitation of the various partially O-methylated alditol acetates can be performed by gas chromatography, in most cases combined with mass spectrometry [109]. An example of methylation analysis is depicted in Fig. 12-12.
350
12 General Strategies for the Characterization of Carbohydrates D,
n
Manal-ZManal\ 4MLal\
Manal-ZManal m A D,
/3
“
$ : n ~ l J C l c ~ A c p I-4Glc~Ac-ol A
/
Manal-2Mana I-2Manal
1
pMAA
-1,Sdi-O-acetyl-2,3,4,6-
permethylation hydrolysis reduction(NaBD4) acetylation GC-MS
Mol. ratio
Structural feature
tetra-O-methylhexitol
3
1,2,Stri-Oacetyl-3,4,6tri-O-methylhexitol
4
+2)Man( I+
di-O-methylhexitol
2
+3,6)Man( I+
1,4-di-Oacetyl-3,6-di-Omethyl-2-N-methylacetamido2-deoxyhexitol
1
Man(l+
1,35,6-tetra-O-acetyl-2,4-
4-mono-O-acetyl-l,3,5,6-
tetra-O-methyl-2-N-methylI
+4)GlcNAc-ol
3
CHDOAc
I ; !
205 z ~
-
ll
‘I’
Fig. 12-12.Methylation analysis of an oligomannose-type oligosaccharide alditol. The El-mass spectra of two partial methylated alditol acetates (PMAAs) are shown as example.
12.5 Analysis of Glycovrotein Glycans
35 I
Mass Spectrometry Different new techniques of mass spectrometry have been introduced in recent years for the structural study of carbohydrate chains. Methods using soft ionization techniques have become very important. The progress achieved in this field has been reviewed by several authors [159-1611. Already for a long time electron impactmass spectrometry (El-MS) in combination with gas chromatography, is used for the identification of monosaccharide derivatives. For instance, methanolysis of an oligosaccharide yields a mixture of methyl glycosides of neutral monosaccharides, amino sugars, and sialic acid (as methyl ester), which can be determined jointly by GC-MS after re-N-acetylation/trimethylsilylation [ 109,1621 (Section 12.5.1, Fig. 12-9). GC-MS is also generally used for the identification of the partially methylated alditol acetates derived from neutral and N-acetylamino sugars in the methylation analysis of glycans. The positions of the 0-methyl and 0-acetyl groups can be deduced from the specific fragmentation observed in the highly characteristic EI-mass spectra (Section 12.4.1, Fig. 12-12). Ring size information can also be obtained. Fast atom bombardment mass spectrometry (FAB-MS) has proven to be a valuable technique for the structural analysis of intact carbohydrate chains, especially with respect to molecular mass determination and sequence analysis. Positive- as well as negative-ion FAB-MS data have been reported for oligosaccharides, oligosaccharide-alditols, and glycopeptides as underivatized, permethylated, and peracetylated derivatives [163]. During FAB-MS, samples are ionized and desorbed from a liquid phase (matrix) using a beam of accelerated atoms (Ar, Xe, Cs). High-field and highresolution magnetic sector mass spectrometers provide the possibility to investigate polar nonvolatile compounds or derivatives with relatively high molecular masses. In the positive mode the molecular mass can be deduced from the relatively abundant protonated molecular ion (M+H)+, and from the frequently present cationized molecular ions (M+Na)+ and (M+K)+. In the negative-ion mode, (M-H)- acts as quasi-molecular ion. In Fig. 12-13, a di-antennary asialoglycopeptide is used as an example of sequence information deduced from a negative-ion FAB mass spectrum [ 1091. The most important fragmentation reaction involves cleavage of the glycosidic bond between the anomeric carbon atom and the interglycosidic oxygen, accompanied by a hydrogen migration. In the case of positive-ion FAB-MS of permethylated and peracetylated derivatives, cleavages have been detected on either site of the glycosidic oxygen. Charged fragments can result from the nonreducing as well as from the reducing end of the molecule. Recently, another soft ionization technique, namely electrospray mass spectrometry (ESMS), has been introduced to the glycoprotein field [ 1641. ESMS is applicable to high-molecular mass molecules (- 100 kDa), including intact glycoproteins. The method is particularly valuable to obtain information whether a protein has been post-translationally modified by the observation of any mass difference between the observed signal and that calculated as the sum of the amino acids present in the sequence. Having available some beforehand information about expected glycan structures, the probable composition of the glycoforms may also be observed. In principle, a stream of liquid containing the sample is directly injected into the ion
352
12 General Strategies for the Characterization of Carbohydrates
Molecular mass 1754
rdz 1753 M-H m/z1735 pl-Hz01-H m/z 1619 [1591+HCO]-H m/z 1591 [1591+H]-H r d z 1573 [1591+H-H20]-H
r d z 1416 [1388+HCO]-H
rdz 1388 [1388+H]-H r d z 1370 [1388+H-HzOI-H m/z 1254 [1226+HCO]-H m/z 1226 [1226+H]-H
Fig. 12-13. Molecular mass and sequence information from the negative-ion FAB-mass spectrum N-glycan linked to asparagine. of an asialo-di-antennary-N-acetyllactosamine-type
source of a mass spetrometer. A spray of microdroplets is generated which, after passing a series of ‘skimmers’, encounters a drying gas. The net effect is the creation of charged molecular species, devoid of solvent, ready for analysis by (quatrapole) MS. Large molecules tend to carry multiple charges and the mass to charge (m/z) is measured. The data show a distribution of signals carrying varying numbers of net positive charges (working in the positive-ion mode) on protonatable basic sites (Lys, Arg) in the molecule. The peak top charge distribution data can be ‘transformed’ using a computer algorithm to produce a MS profile of intensity against mass. MALDI-MS is the acronym for matrix-assisted laser desorption and ionization mass spectrometry. In principle, desorption and ionization of molecules (in a matrix) is induced by means of electromagnetic radiation from a laser and the mass of an ion is measured by determination of its m/z value using, in general, a time-of-flight (TOF) mass spectrometer [165,166]. This technique can be used to analyze intact glycoproteins as well as pure glycans and is therefore highly suitable to determine the number of glycans and their individual mass. Basically, a particular molecular species (e.g., a carbohydrate chain or a particular glycoform of a protein) yields only one MS signal (in contrast to ESMS) and hence by correlating mass to primary structures data are obtained to support an assigned structure. The method can be used for (preferably deacidified) glycan pools to get a mass profile or on pure glycans for exact mass determination. Small sample amounts (0.1 pg pure glycan or 10 pg glycoprotein) are required. The application of MALDI-MS in the glycoprotein field is rapidly increasing and numerous studies have appeared in literature recently [97,167-1691. In Fig. 12-14, the MALDI mass spectrum of FSH is depicted as an example of the potential of the technique for the identification of intact glycoproteins. For batch-control procedures, several approaches, including HPLC in an on-or offline configuration with MS, have been described [ 170-1731. In these approaches
12.5 Analysis of Glycoprotein Glycans
D
B
353
galactose
Fig. 12-14. MALDI-mass spectrum of follicle stimulating hormone (FSH) glycoprotein
either free or derivatized oligosaccharides and glycopeptides were separated and subsequently analyzed by ES-, MALDI- or FAB (tandem)-MS.
'H-nuclear Magnetic Resonance Spectroscopy It is but 20 years ago that high-resolution 'H-NMR spectroscopy was introduced for the elucidation of the primary structure of glycans derived from glycoproteins [174,175]. Nowadays, it is one of the most commonly used methods because, in addition to determining the primary structure, NMR can provide information on the conformation and molecular dynamics of the molecule in solution. For interpretation of the 'H-NMR spectrum of a carbohydrate chain in terms of primary structural assignment use is made of structural-reporter group signals as listed in Table 12-4. In carbohydrates, most of the sugar-skeleton protons give resonance signals in a crowded region between 6 3.5 and 6 3.9 ppm. The chemical shifts of specific types of protons resonating at clearly distinguishable positions outside of this bulk region in the spectrum, together with their coupling constants and the signal line-widths bear information on the primary structure. As example, a 'H-NMR spectrum is depicted in Fig. 12-15. The conversion of the chemical shift values of the structural-reporter group signals to carbohydrate structures makes use of extensive libraries of reference compounds [ 176,1771. Recently, database-related computer programs have become available which contain tables of 'H-NMR chemical shift values and corresponding carbohydrate structures, and literature references [ 178,-
354
12 General Strategies for the Characterization of Carbohydrates
Table 12-4. 'H-NMR structural-reporter group signals for carbohydrate chains of glycoproteins. -
Anomeric protons H-l
-
Amide protons (in H20)
- Man H-2 and H-3 -
GalNAc-ol H-2, H-3, H-4 and H-5
-
Sialic acid H-3 protons (equatorial and axial)
-
Fuc H-5 and H-6 protons (CH3)
-
Gal H-3 and H-4
-
Protons shifted out of the bulk region due to glycosylation shifts
-
Protons shifted out of the bulk region due to the presence of non-carbohydrate substituents like acyl, sulfate, and phosphate groups
-
Protons belonging to substituents on carbohydrate residues like 0-methyl, N, 0-acetyl and N-glycolyl groups
2eu54c
NeuSAc
NraSAc'
NeuSAC
H-3e atoms NeuSAc: Man H-2 atoms
,NeuSAc
. .
b
i7
3
52
51
50
L9
L7
L6
L5
LL
L3
L2
41
I
27
5
1
15"l
21
20
18
17
6 lppml
Fig. 12-15. 500-MHz 'H-NMR spectrum showing the structural-reporter-group regions of a diantennary-N-acetyllactosamine-typeN-glycan. The relative scale of the NAc proton region differ? from that of the rest of the spectrum. The spectrum was recorded in D 2 0 at 300 K.
18I]. Significant improvement in the interpretation of NMR spectra of complex carbohydrates derives from the application of multi-dimensional (2D and 3D) NMR methods stimulated by dramatic advances in computer technology and in the construction of high-field, superconducting magnets over the past decade. A limitation
12.5 Anulysis of Glycoprotein Glycuizs
355
of the NMR spectroscopic analysis (although non-destructive) is still the amount of sample (at least 20 pg of pure glycan) required. However, using a micro-probe this may be lowered to about 5 pg. Several excellent books [182-1841 and reviews [ 185-1891 dealing with the application of NMR techniques for macromolecular structural analysis have appeared. As mentioned before, due to the multiple glycosylation sites and high heterogeneity, causing a multitude of (overlapping) protein- and carbohydrate-derived NMR signals, it was not possible to analyze carbohydrate structures directly on the intact glycoproteins by NMR spectroscopy. Recently, however, structural NMR studies of both carbohydrate and protein moieties of intact complex glycoproteins appeared in literature, showing that considerable progress has been made in this field [190-1921. The structure assessment by NMR spectroscopy of the a-subunit of human chorionic gonadotropin (hCG), an intact glycoprotein subunit, was probed by 'H and gradient-enhanced H-I3C heteronuclear correlation spectroscopy [ 193,1941. Also, the glycosylated amino-terminal adhesion domain of the human T-cell surface glycoprotein CD2 was solved in solution by NMR methods [195]. Other intact glycoproteins investigated by 'H and/or 13C NMR spectroscopy are: hen phosvitin [ 1961, IgG [ 1971, submaxillary mucins [ 1981, glycophorin A [199] and pineapple stem bromelain [200]. Another exciting development is the
'
Table 12-5. Methods to obtain information about specific carbohydrate features. Information
Methods
Carbohydrate content, composition, on-configuration
Colorimetric determinations; GC-monosaccharide analysis; GC-absolute configuration determination; NMR-spectroscopy
Molecular mass of gly coprotein/glycan (presence of glycosylation)
Gel filtration chromatography; Mass profile FAB/ES/ MALDI-mass spectrometry; SDS-PAGE (beforehfter enzyme treatment)
Nature of carbohydrate-peptide linkage (N/O)
Proteolytic digestion; Amino acid analysis; Examination of alkali lability; hydrazinolysis
Type of glycans (oligomannose, complex, hybrid), glycoforms
GC-monosaccharide analysis; Sizekharge profile analysis; Capillary electrophoresis
Number/proportions of glycans present
Size/charge profile analysis; Mapping by HPLC, HPAEC, FACE, MALDI-MS
Sequence of monosaccharide residues
Digestion by exoglycosidases; Partial hydrolysis; NMR-spectroscopy; mass spectrometry
Positions of glycosidic linkages
Methylation analysislGC-MS; FAB-MS; NMR-spectroscopy
Anomeric configuration
Digestion by exoglycosidases; NMR-spectroscopy
Certain structural determinants
Antibody responses; Endolexo-glycosidases; Affinity chromatography (lectins)
Type of charged substituents
Sizelcharge profile analysis; HPLC; HPAEC; NMR-spectroscopy
Spatial structure of glycoprotein/glycan
X-ray analysis; (2D/3D) NMR-spectroscopy; Molecular dynamics, mechanics and modeling
356
12 General Strategies f o r the Characterization of Carbohydrates
progress that has been made in expression and labeling of glycoproteins (e.g., hCG) in CHO cells with stable NMR isotopes like I3C and 15N [201]. This ‘friendly’ labeling procedure for mammalian post-translationally-modified proteins holds promise to solve the structure of intact glycoproteins in solution more routinely. Moreover, it opens the way to determine the conformation of (“C-, I5N-labeled) carbohydrate chains on the surface of glycoproteins, and furthermore, the assessment of interactions of the carbohydrate chains with the peptide backbone come into reach. Table 12-5 summarizes the various methods that can be applied to obtain information about specific carbohydrate features.
12.6 Oligosaccharide Profiling/Mapping Because the biological and physico-chemical properties of (recombinant) glycoproteins frequently are significantly influenced by the type of glycosylation, knowledge about the glycosylation pattern is of utmost importance. Although in many cases the exact function of the carbohydrate chains has not been established as yet, much attention should be paid to obtain the ‘right’ glycosylation pattern. Therefore, means of assessing batch-to-batch consistency during production of glycoprotein pharmaceuticals, in terms of protein glycosylation, is urgently required. The necessary information can be obtained by the procedures described earlier in this chapter, but for routine analysis of production batches a simpler, less labor-intensive procedure is needed. Currently, methods are being developed, based on oligosaccharideprofiling procedures, facilitating reliable and fast batch control of glycoproteins. Next to the analysis of the monosaccharide composition (discussed in Section 12.1; Fig. 12-9), the glycans are released from the protein backbone and directly scanned by a combination of different separation techniques, such as more-dimensional HPLC [202,203] and HPAEC [ 140,1411. Also, gel and capillary electrophoresis seem to be highly promising to obtain oligosaccharide profiles [204-2081. Identification of the glycans is commonly based upon the co-elution (chromatography) or co-migration (electrophoresis) with standard compounds. Comparison of the profiles (oligosaccharide-fingerprints) provides direct information on the (structural features of the) intact carbohydrate chains present on the (recombinant) glycoproteins of different batches. During profile/mapping analysis, fractionation of the different glycans in the pool of released carbohydrates is normally performed by either of three separation procedures using the chromatographic techniques described earlier in this chapter: 1 . Mono-Q in combination with Lichrosorb-NH2. Released carbohydrates are first fractionated by FPLC on a Mono-Q anion-exchange column on the basis of the sialic acid (or other acidic substituents) content, and then subfractionated by HPLC using a Lichrosorb-NHz colum [135]. Detection in both procedures is usually performed by UV. 2. High-pH anion-exchange chromatography with pulsed amperometric detection. HPAEC-PAD has been introduced for the separation and sensitve detection of car-
12.6 Oligosacchuride Profiling/Mupping
357
IIIIIIIII(IIIIII111(111111111(IIIIIIIII~I11111111~111111111~11111111
30
40
50
60
70
80
90
(min)
Fig. 12-16. Glycosylation profile of a glycoprotein. HPAEC elution pattern of the enzymatically released N-glycans of bovine fetuin. The column (CarboPac PA-1, 25 X 0.4 cm) was eluted with 0.1 M NaOH, using a NaOAc gradient (0-230 mM), at a flow rate of 1 ml min-I. Detection was performed by pulsed amperometry.
bohydrates [141]. Fractionation is mainly on basis of charge, leading to the formation of carbohydrate charge-clusters. Linkage and branch positional isomers of carbohydrates are separated within each cluster (Fig. 12-16). 3. GlycoSep C in combination with GlycoSep H . This relatively new procedure can be compared with the Mono-Q/Lichrosorb-NHz method. The main differences are that the carbohydrates are labeled prior to separation with 2-aminobenzamide (2AB) at the reducing terminus [131] and the use of volatile elution buffers, thus circumventing various desalting steps. The first separation step is performed on the GlycoSep C column (Oxford GlycoSciences) on the basis of the sialic acid content (or the overall negative charge). Thereupon subfractionation takes place on the GlycoSep H column (Oxford GlycoSciences) on basis of hydrophobicity/size of the 2-AB labeled glycans. Detection in both procedures is performed by fluorescence detection. Currently, attempts are made to develop reproducible and (partly) automated analysis instruments, of which some are already commercially available (Table 12- 6). Although, both the monosaccharide compositional data and the profiling data allow the fast comparison of different batches of a (recombinant) glycoprotein, for each specific purpose, criteria must be set to define, on the basis of the obtained chromatograms, whether two batches are identical with respect to glycosylation or
358
12 General Strategies for the Characterization of Carbohydrates
Table 12-6. Several commercially available semi-automatic carbohydrate analysis instruments. System
Company
Performance
GlycoPrepTM 1000
Oxford GlycoSciences”
Release (hydrazinolysis) and recovery of N- and 0-linked carbohydrate chains from glycoproteins”
GlycoMapTM 1000
Oxford GlycoSciences
Size exclusion chromatography of released (radiolabeled) carbohydrate chains with radiochemical or refractive index detection
RAAMTM2000 (Reagent Array Analysis Method)
Oxford GlycoSciences
Degradation of fractionated and released glycans in an array of defined multiple exoglycosidase mixturesc%d
G1ycoTAGTM
TaKaRae
Two-dimensional high performance liquid chromatography of released glycans which are labeled with 2-aminopyridinef.g
FACE (Fluorophore-assisted carbohydrate electrophoresis)
Glykoh
Polyacrylamide slab gel electrophoresis of released glycans which are labeled with 2-aminoacridone’
Glyco DocTM Imaging System
Bio-RadJ
Polyacrylamide slab gel electrophoresis of released fluorescent labeled glycans
-
Oxford GlycoSciences, Abingdon, U.K. Often used in combination with HPAEC. Performed with neutral carbohydrate chains only. Used in combination with Gl;coMapTM 1000 and Glycosequencer: a gel filtration profile of degradation products is indicative for a structure. TaKaRa, Kyoto, Japan. Best performed with neutral carbohydrate chains. g Carbohydrate samples are automatically derivatized on the Palstation 4000. Glyko, Novato, U.S.A. ’ Analysis is performed with a Glycoscan Fluorescence Electrophoresis System. Bio-Rad Laboratories, Hercules, California, U.S.A. a
“
J
not. In situations like the occurrence of undefined carbohydrate chains in an oligosaccharide-map, or the setting-up of a fingerprinting procedure for routine batch control, as a first step an approach involving a full delineation of the structure of each oligosaccharide component is still required.
12.7 Structural Analysis qf N - and O-Linked Glvcuns o j rhEPO
359
12.7 Structural Analysis of N- and O-linked Glycans of Recombinant Human Erythropoietin (rhEPO) To illustrate the general strategy for the characterization of the carbohydrate chains from a recombinant glycoprotein, the analytical procedure for recombinant human erythropoietin (rhEPO) expressed in CHO cells is described here [106,209]. The molecular mass of EPO is 34-39 kDa, and the polypeptide chain has one O-glycosylation site at Ser126 and three N-glycosylation sites at Asn24, Asn38, and Asn83, respectively [210]. The total carbohydrate content is 40% of the molecular mass. EPO has been expressed in various heterologous cell systems, e.g., CHO [211], BHK [212], Y2 cells derived from NIH-3T3 [213], insect cells [214], and cultured tobacco cells [215]. Several studies on the structure and function of the carbohydrate chains of (rh)EPO have been carried out [216,217]. The recombinant human glycoprotein is an important therapeutic agent for the treatment of anemia associated with renal failure, since it stimulates red cell proliferation and differentiation in the bone marrow [218,219]. Glycosylation (both the extent and the precise oligosaccharide structures attached) significantly influences the biological and physico-chemical properties of (rh)EPO [220-2231.
12.7.1 Liberation of the N-linked Carbohydate Chains rhEPO (50 mg) was dissolved in 5 ml 50 mM Tris-HC1, pH 8.4, containing 50 mM EDTA. Subsequently, 1 % (by volume) 2-mercaptoethanol and 1 % (massholume) SDS were added, and the mixture was kept for 3 minutes at 100°C. After cooling to room temperature, the sample was diluted twice with incubation buffer, and PNGase-F (from E meningosepticurn, Boehringer Mannheim) was added (0.2 U mg-' EPO). The mixture was incubated for 16 h in an 'end-over-end' mixer at ambient temperature, then the solution was heated for 3 minutes at 100 "C, and after cooling, another aliquot of PNGase-F (0.2 U mg-' EPO) was added. The incubation was continued for 24 h. The deglycosylation was checked by SDS-PAGE to be essentially complete as the molecular mass of rhEPO shifted from 35 kDa (native) to 22 kDa (N-deglycosylated). The N-glycan pool was isolated by gel filtration chromatography on a Bio-Gel P-100 column (47 x 2 cm), eluted with 25 mM NH4HC03, pH 7.0, at a flow rate of 22 ml h-I. The eluent was monitored at 206 nm. After desalting on a Bio-Gel P-2 column (45 X 1 cm) using water as eluent, the N-glycan pool was ready for fractionation.
12.7.2 Liberation of the O-linked Carbohydrate Chains The P-100 void volume fraction (containing the protein possessing the O-linked glycan) was lyophilized and suspended (approx. 10 mg ml-l) in 0.1 M NaOH, contain-
360
12 General Strategies for the Characterization of Carbohydrates
A
GalNAc-ol
/
NeuAcu2+3
Gal3 H-1
Galfil-13
Gal3 H-4
I
/
I
4 4
-
, I
4.6
4.4
6(ppm)
4.2
4.0
3.8
3.6
3.4
--1--
2.8
2.6
2.2
2.0
1.8
1.6
NeuAca2-+6,
GalNAc-ol
,Galp1-+3
/
NeuAca243
Ly
,
4.8 -4fm,
,
,
,
4.4
,
,
4.2
,
NAc's
,
4.0
'
3.8
3:6
'
3:4
'
u. . , . a 't--2.8 2.6 2:2 2.0 1.8 1.6
Fig. 12-17. 500-MHz 'H-NMR spectra of the released 0-glycans of recombinant human erythropoietin (rhEPO) from CHO cells. The relative scale of the NAc proton region differs from that of the rest of the spectrum. The spectra were recorded in DzO at 300 K.
12.7 Structural Analysis of N - and 0-Linked Glycans of rhEPO
361
ing 1 M NaBH4. The solution was kept for 24 h at 40 "C, then cooled on ice and neutralized with 4 M HOAc. Boric acid was removed by repetitive co-evaporation with MeOH, containing 1 % HOAc. Finally, the material was resuspended in water and centrifuged at 12000 g. After desalting of the supernatant on a Bio-Gel P-2 column (45 x 1 cm) using water as eluent, the 0-glycan pool was ready for fractionation.
12.7.3 Fractionation and Structural Determination of the 0-Glycans Separation according to charge on a Mono Q HR 515 anion-exchange column (Pharmacia) yielded two carbohydrate-containing fractions, denoted 0 1 and 0 2 , which were purified by HPLC on Lichrosorb-NHz, yielding the fractions 01.2 and 02.1, respectively. Both fractions were investigated by 500-MHz 'H-NMR spectroscopy and revealed the 'H-NMR spectra of a monosialylated trisaccharide alditol NeuSAca2-3Galp 1-3GalNAc-01 and a disialylated tetrasaccharide alditol Neu5Aca2-3Galpl-3(Neu5Aca26)GalNAc-ol, respectively, as depicted in Fig. 12-17. In a recent study comparing the 0-glycans from urinary EPO and rhEPO from CHO cells, in rhEPO also Gal~l-3(Neu5Aca2-6)GalNAc-ol was detected [211,224]. It has to be noted that the 0-glycans of rhEPO are completely different from those of urinary EPO containing only GalNAc and Neu5Aca2 - 6GalNAc [225].
12.7.4 Fractionation and Structural Determination of the N-Glycans The enzymatically released N-glycan pool was first fractionated according to charge on a Mono Q HR 515 anion-exchange column (Pharmacia), yielding four carbohydrate-containing fractions, denoted N1, N2, N3, and N4, having elution positions corresponding to mono-, di-, tri-, and tetrasialylated complex type N-glycans, respectively (Fig. 12-18). A further subfractionation of each Mono Q fraction was obtained by HPLC on Lichrosorb-NH2. Because a detailed discussion of the structure elucidation of all rhEPO-derived N-glycans is beyond the scope of this chapter, only the tetrasialylated fraction N4, which was separated into eight fractions (N4.1-N4.8) by HPLC (Fig. 12-19), will be discussed. For the same reason, we will only focus on the major fractions N4.4, N4.6, and N4.8. At this stage an investigation by 'HNMR spectroscopy of the fractions is already beneficial to get an impression of further heterogeneity. Fraction N4.4 contained a single compound which could be identified by 'H-NMR spectroscopy as being a tetrasialylated tetra-antennary oligosaccharide: 8'
7'
NeuSAcuZ-3Galp 1-4GlcNAcp 1-% Neu5Aca2-3Galp 1-4GlcNAcp 1-2Manul-6 6'
5'
Fucal-6 Manpl-4GlcNAcpl-4GlcNAc
NeuSAcu2-3G~lpl-4G1~A~p1-2Ma~al-3
NeuSAux2-3Galpl-4GlcNAcp1-4 8
1
2
1
-1 -...f N4.4
362
12 General Strategies for the Characterization of Carbohydrates
................
-3 -nz
0.5
N4
0.05
0
I
I
I
4
0
I
I
I
I
I
12
8
I
I
16
20
I
I
(ml)
Fig. 12-18. FPLC elution pattern of the enzymatically released N-glycans of recombinant human erythropoietin (rhEPO) from CHO cells. The column (HR S/S Mono Q) was eluted with a NaCl gradient in H20 as indicated, at a flow rate of 2 ml min-'. Detection was performed at 214 nm.
N4.4
1
N4.6
N4.: N4.1
0
60
30
90
120
(m1)
Fig. 12-19. HPLC elution pattern of FPLC fraction N4 derived from rhEPO. The column (Lichrosorb-NH2) was eluted with 30 mM K2HPOdKH2P04 pH 7/acetonitrile (37.5:62.5, v/v), at a flow rate of 2 ml min-'. Detection was performed at 205 nm.
This is the major constituent of all N-linked carbohydrate chains of rhEPO. Fraction N4.8 also contained a single compound and its structure was determined by 'H-NMR spectroscopy as being: ert
ert
8'
7'
NeuSAcu2-3Galp 1-4GlcNAcp l-3Galp 1-4GlcNAcP 1-6, \
:'
NeuSAca2-3Galp I-4GlcNAcp I-3GalP I-4GlcNAcP 1-2Manal-6 eit
elt
6'
5'
NeuSAca2-3G$3 I-4Glci&p
1-2!yanal-3
NeuSAca2-3Galp 1-4GlcNAcp 1-4 8
7
Fucal-6 Manpl-4GlcNAcp l-4GlcNAc 2
1
N4.8
363
12.7 Structural Analysis of N- and 0-Linked Glycans of rhEP0
The location of two extra N-acetyllactosamine units in the upper branches was proved by enzymatic digestion experiments as described below. From the 'HNMR spectrum of fraction N4.6 the presence of a mixture of tetrasialylated tetraantennary oligosaccharides, each containing one extra N-acetyllactosamine unit, was deduced. The components in fraction N4.6 were separated by HPAEC, affording the subfractions N4.6.1 and N4.6.2 (Fig. 12-20). 'H-NMR spectroscopy in conjunction with enzymatic digestion experiments identified the structures of the compounds to be as follows: exi
8'
ed
7'
NeuSAca2-3GalP 1-4GlcNAcP 1-3GalP1-4GlcNAcP 1-6 NeuSAca2-3Galp 1-4GlcNAcP 1-2Manal-6 6'
5'
t*H
Fucald
Manp 1-4GlcNAcP I-4GlcNAc
NeuSAca2-3G~I~I-4Glc~A 1 -c2pM d a l - 3
I
2
NeuSAca23Galp 1-4GlcNAcP 1-4 8
8'
N4.6.1
7
7'
NeuSAca2-3Galp 1-4GlcNAcP 1-6 NeuSAcaZ-3Galp I-4GlcNAcp I-3Galpl-4GlcNAcpl-2Manal-6 eit
6'
ext
'
5'
Fucald ManP 1-4GlcNAcP 1-4GlcNAc
NeuSAca2-3G:lp 1 -4GlcN?Acp I - 2 M d a l - 3
I
2
NeuSAca2-3GaQ31-4GlcNAcP 1-4 8
N4.6.2
7
N4.6.1
-
200
k
I
k 100
0 0
40
80
120
(ml)
Fig. 12-20. HPAEC elution pattern of HPLC fraction N4.6 derived from rhEPO. The column (CarboPac PA-1, 25 X 0.9 cm) was eluted with 0.1 M NaOH, using a NaOAc gradient as indicated, at a flow rate of 4 ml min-I. Detection was performed by pulsed amperometry.
364
12 General Strategies .for the Characterization o f Carbohydrates
To determine the branch location of the N-acetyllactosamine repeating units in the tetra-antennary oligosaccharides, the compounds were degraded with endo-0-galactosidase, followed by digestion with N-acetyl-0-glucosaminidase [226] (Fig. 1221). Endo-P-galactosidase (from Bacteroides frugilis, Boehringer Mannheim) hydrolyzes the 0-galactosidic linkage in the Gal0 1-4GlcNAc element in a poly-(N-acetyllactosamine) sequence when the galactose residue is not terminal [227]. After digestion, the trisialo compound (in case of one extra N-acetyllactosamine unit) and the
j Endo-0-galactosidase
I
ext
8'
ext
NeuSAca23GalP 1-4GlcNAcP1-3GalP1-4GlcNAcP1-6, \
Neu SAca2-3GalP 1-4GlcNAcP1-2Mana 1-6, 6' 6
5' 5
4'
Fucal-6 Manpl-4GlcNAcf3 1-4GlcNAc
NeuSAca2-3Galp 1-4GlcNAcP1-2Mana 1-3/'
3
2
1
NeuSAca2-3GalP 1-4GlcNAcP1-4' 8
7
I
7'
GlcNAcPl-6,
J
NeuSAca2-3GalP 1-4GlcNAcP1-2Manal-q 6'
5'
6
5
4' 4
Fucal-6 \ ManP 1-4GlcNAcP 1-4GlcNAc 2
NeuSAca2-3GalP 1-4GlcNAcP1-2Manal-3'
1
NeuSAca2-3GalP 1-4GlcNAcP1-4 8
7
NeuSAca2-3Galp 1-4GlcNAcP1-2Manal-6 6'
6
5' 5
4'
FUCU~-6 \ Manp 1-4GlcNAcPl-4GlcNAc
NeuSAcu2-3Galp 1-4GlcNAcp1-2Manctl-3 '
3
2
1
NeuSAcu2-3Galp 1-4GlcNAcp1-4 8
7
Fig. 12-21. Determination of the branch location of an N-acetyllactosamine repeating unit in a tetra-antennary oligosaccharide derived from rhEPO. Succesive incubation with endo-p-galactosidase and N-acetyl-p-glucosaminidaseyielded a tri-antennary oligosaccharide which could be identified by 'H-NMR spectroscopy.
365
12.7 Structural Analysis of N - and 0-Linked Glycans of rhEP0
disialo compound (in case of two extra N-acetyllactosamine units) were isolated by Mono Q anion-exchange chromatography. These compounds were then treated with N-acetyl-0-glucosaminidase(from jack beans, Sigma) to release the terminal GlcNAc residue. In this way oligosaccharides (N4.6.1D, N4.6.2D, and N4.8D, see Table 12-7) were obtained that could unambiguously be identified by high-resolution 'H-NMR spectroscopy, providing conclusive evidence for the structures of the intact compounds. The results obtained by 'H-NMR spectroscopy in terms of 'H-chemical shifts of the structural-reporter group protons are usually presented in the form of a table (see Table 12-7). Table 12-7. 'H-chemical shifts of the structural-reporter-group protons of the constituent monosaccharides of some N-linked oligosaccharides derived from recombinant human erythropoietin. For the short-hand symbolic notation, see text. Reporter group .
Residue
Chemical shifts (ppm) in
3
N4.4
H- 1
H-2 H-3
H-4
GlcNAc-laa 5.182 GlcNAc-lP 4.688 GlcNAc-2aa 4.659 G I c N A c - ~ ~ 4.664 ~ 5.131 Man-4 4.857 Man-4' 4.563 GIcNAc-5 4.593 GlcNAc-5' 4.542 Gal-6 4.545 Gal-6' 4.542 GIcNAc-7 GlcNAc-7' 4.545 4.542 Gal-8 4.559 Gal-8' GlcNAcext Galext Man-3 4.203 Man-4 4.220 Man-4' 4.090 Gal6 4.117 Ga16' 4.117 Gal8 4.117 Ga18' 4.117 Galext Ga16' n.d. Ga18' n.d.
s
BB
N4.6.1
N46.2
N4.8
N4.6.1D N4.6.2D
5.182 4.690 4.659 4.663 5.130 4.854 4.564 4.589 4.542 4.547 4.542 4.547 4.542 4.467 4.701 4.556 4.206 4.2 19 4.091 4.117 4.117 4.117 n.d.d 4.117 n.d. 4.159
5.182 4.689 4.659 4.664 5.130 4.858 4.563 4.602 4.542 4.455 4.542 4.545 4.542 4.557 4.699 4.557 4.205 4.219 4.08 1 4.1 16 n.d. 4.116 4.116 4.1 I6 4.161 n.d.
5.182 4.689 4.659 4.664 5.129 4.855 4.561 4.60 4.541 4.454 4.541 4.546 4.541 4.468 4.700b 4.556b 4.21' 4.22 4.080 4.1 16 n.d. 4.116 n.d. 4.116 4.161 4.161
5.182 4.692 4.664 4.667 5.114 4.905 4.560 4.574 4.547 4.547 4.547
5.181 4.690 4.663 4.668 5.115 4.884 4.557
N4.8D
4.542
5.181 4.69 1 4.663 4.668 5.114 4.915 4.559 4.543
-
-
4.543
-
4.542 4.546 4.542 4.566
-
-
-
4.212 4.212 4.11 4.117 4.117 4.117
-
4.2 13 4.2 13 3.96 4.115 4.115
-
4.547
-
4.21 4.21 3.96 4.117 -
-
4.543 -
-
4.117 4.117 -
n.d.
-
-
-
n.d.
-
-
-
366
12 General Strategies ,for the Characterization of Carbohydrates
Table 12-7. (Continued).
Reporter
H-3a H-3e NAc
H- 1
CH3
a
Residue
Neu5Ac NeuSAc GlcNAc-1 GlcNAc-2aa GlcNAc-20 GlcNAc-5 GlcNAc-5' GlcNAc-7 GlcNAc-7' GlcNAcext NeuSAc Fucaa Fucp Fucaa Fucp
Chemical shifts (ppm) in
N4.4
N4.6.1
N4.6.2
N4.8
N4.6.1D N4.6.2D N4.8D
1.804e 2.756" 2.038 2.094 2.090 2.047 2.038 2.075 2.038
1.802" 2.757' 2.037 2.094 2.090 2.047 2.037 2.075 2.037 2.037 2.03 1 4.900 4.908 1.211 1.223
1.803" 2.757e 2.038 2.092 2.088 2.047 2.038 2.074 2.038 2.038 2.031h 4.898 4.905 1.210 1.222
1.801e 2.757e 2.037 2.091 2.088 2.047 2.037 2.075 2.037 2.037g 2.03Ih 4.897 4.905 1.210 1.222
1.802' 2.756' 2.040 2.095 2.095 2.044 2.044 2.074 2.031' 4.893 4.901 1.211 1.223
-
2.031h 4.901 4.909 1.211 1.222
1,802' 2.756' 2.037 2.09 1 2.087 2.044
1 .800b 2.756h 2.038 2.092 2.092 2.043
-
-
2.073 2.052
2.073 2.03 1g 4.886 4.893 1.209 1.221
-
2.03 1 4.888 4.897 1.210 1.222
a and 0 stand for the a and 0 anomers of GlcNAc-1; Signal stemming from two protons; Values given with only two decimals because of spectral overlap; n.d.; not determined; e Signal stemming from four protons; Signal stemming from three protons; g Signal stemming from two NAc groups; hSignal stemming from four NAc groups; 'Signal stemming from three NAc groups.
12.7.5 Discussion The glycosylation of rhEPO is of great importance for its biological functioning. It has been demonstrated that removal or modification of the glycan chains, or prevention of glycosylation at specific sites, results in altered in vivo and in vitro activity [225,228]. Although rhEPO posesses only one 0- and three N-glycosylation sites, over 35 different N-linked oligosaccharides and two 0-linked oligosaccharides could be identified. The established N-glycan structures ranged from mono-sialylated di-antennary structures to fully-sialylated (containing extra N-acetyllactosamine units) tetra-antennary structures. Between these two extremes a great variety of di-, tri/tri'- and tetra-antennary structures, containing 1-4 sialic acid residues and/or 1-2 extra N-acetyllactosamine units was found [209]. These results illustrate nicely the formidable heterogeneity that can occur in a (recombinant) glycoprotein. Compared with natural EPO isolated from the urine of patients with aplastic anemia, the N-glycan heterogeniety was rather similar [228,229]. However, some differences in com-
Abbreviations
367
position and relative amounts could be noticed. Recombinant EPO seems to have a somewhat higher content of tetra-antennary oligosaccharides and a tendency to contain higher amounts of N-linked carbohydrate chains with Galfil-4GlcNAcp 1-3 repeats. Furthermore, the carbohydrate chains of natural EPO contain Neu5Ac in a2-3 and a2-6 linkage to Gal, while rhEPO from CHO cells contains three types of sialic acid, namely NeuSAc (95 %), Neu5Gc ( 2 %) and Neu5,9Ac2 (3 %), only in a2-3 linkage. The presence of Neu5Gc and Neu5,9Acz has also been observed in other recombinant glycoproteins expressed in CHO cells [230,23 11. With respect to the a l - 6 fucosylation of the Asn-bound GlcNAc residue in CHO and BHK EPOs, amounts between at least 80 % and nearly 100 % have been reported [216]. Glycosylation of recombinant glycoproteins is influenced by numerous factors as outlined in Chapter 5. With the above-described example, the extreme importance of an accurate and reliable analysis of recombinant glycoprotein therapeutics with respect to their carbohydrate content in terms of composition and structure is clearly demonstrated. In the context of therapeutic administration of recombinant-DNA glycoproteins, the glycosylation patterns of the engineered proteins demand thorough consideration with regard to applicability, tolerance, and patent position.
Acknowledgements The authors gratefully acknowledge Drs J. J. M. van Rooijen, R. GutiCrrez Gallego, and C.H. Hokke for their contributions. The rhEPO study was supported by the Netherlands Foundation for Chemical Research (SON/NWO) and Organon International BV (Oss, The Netherlands).
Abbreviations Man Fuc Gal Glc XYl Ara Rha Rib GlcNH2 GlcNAc GlcNAc-01 GalNAc GalNAc-ol NeuAc/NeuS Ac NeuGc/NeuSGc
D-mannose L-fucose D-galactose D-glucose D-xylose L-arabinose L-rhamnose D-ribose D-glucosamine N-acetyl- D-glucosamine N-acetyl- D-glucosaminitol N-acetyl- D-galatosamine N-acetyl- D-galatosaminitol N-acetylneuraminic acid (sialic acid) N-glycolylneuraminic acid
368
12 General Strategies .for the Characterization of Carbohydrates
GlcA GalA MurNAc rER Ala Arg Asn CYS Glu GlY His HYLYS HyPro LYS Pro Ser Thr ADP CMP GDP UDP Dol-P GPI AC FPLC GPC HPLC HPAEC PAD GC MS FAB EI ES MALDI TOF NMR 6 SDS-PAGE PNGase-F ConA FSH hCG hCG-a hCG-@ rhEPO
glucuronic acid galacturonic acid N-acetylmuramic acid rough Endoplasmic Reticulum L-alanine L-arginine L-asparagine L-cy steine L-glutamine L-glycine L-histidine 5 -hydroxy-L-1ysine 4 -hydroxy-L-proline L-ly sine L-proline L-serine L-threonine adenosine diphosphate cytidine 5’-monophosphate guanosine diphosphate uridine diphosphate dolichol phosphate glycosyl phosphatidyl inositol Affinity chromatography Fast protein liquid chromatography Gel permeation chromatography High performance liquid chromatography High pH anion exchange chromatography Pulsed amperometric detection Gas chromatography Mass spectrometry Fast atom bombardment Electron impact Electrospray Matrix assisted laser desorption ionization Time of flight Nuclear magnetic resonance chemical shift Sodium dodecyl sulfate polyacrylamide gel electrophoresis peptide-N4-(N-acetyl-P-glucosaminyl)asparagine amidase-F Concanavalin A Follicle stimulating hormone (follitropin) human chorionic gonadotropin a-subunit of human chorionic gonadotropin @-subunitof human chorionic gonadotropin recombinant human erythropoietin
References
cDNA CHO BHK EtN 2 -AB HF PA PIPLC TFMS TMS
369
complementary DNA Chinese hamster ovary Baby hamster kidney Ethanolamine 2 -aminobenzamide Hydrofluoric acid Pyridylamino Phosphatidyl inositol-phospholipase C Trifluoromethane sulfonic acid Trimethylsilyl
References [I] Paulson, J.C., Trends Biochem Sci, 1989, 14, 272-276. [2] Jentoft, N., Trends Biochem Sci, 1990, 15, 291-294. [3] Ginsberg, V., Robbins, P. (Eds.), Biology of Carbohydrates, Vol 3. New York: J. Wiley, 1991. [4] Cumming, D. A,, Glycobiology, 1991, 1, 115-130. [5] Elbein, A. D., Trends Biofechnol, 1991, 9, 346-352. [6] Kobata, A,, Eur J Biochem, 1992, 209, 483-501. [7] Rasmussen, J.R., Curr Opin Struct Biol, 1992, 2, 682-686. [8] Hart, G. W., Curr Opin Cell Biol, 1992, 4, 1017-1023. [9] Kornfeld, S., Mellman, I., Annu Rev Cell B i d , 1989, 5, 483-525. [lo] Varki, A,, in: Cell surface carbohydrates and cell development, Fukuda, M. (Ed.), Ann Arbor: CRC Press, 1992, pp. 25-69. [ l l ] Kretz, K. A., Carson, G. S., Morimoto, S., Kishimoto, Y., Fluharty, A. L., O’Brien, J. S., Proc Nut1 Acad Sci USA, 1990, 87, 2541-2544. [12] Wetzler, L. M., Barry, K., Blake, M. S., Gotschlich, E. C., Infect Immunol, 1992, 60, 39-43. [13] Kjellen, L., Lindahl, U., Annu Rev Biochem, 1991, 60, 443-475. [14] Hardingham, T.E., Fosang, J., FASEB J , 1992, 6, 861-870. [15] Ruoslahti, E., Annu Rev Cell Biol, 1988, 4, 229-255. [16] Ruoslahti, E., Yamaguchi, Y., Cell, 1991, 64, 867-869. [17] Men, S., Pangburn, M. K., Proc Natl Acad Sci USA, 1990, 87, 3982-3986. [18] Bremer, E., Hakomori, S., Biochem Biophys Res Commun, 1982, 106, 711-718. [I91 Zaleska, M.M., Erecinska, M., Proc Nut1 Acad Sci USA, 1987, 84, 1709-1712. [20] Bhaskar, K. R., Gank P., Turner, B. S., Bradley, J. D., B a n d , R., Stanley, H. E., LaMont, P., Nature, 1992, 360, 458-461. [21] Varki, A,, Glycobiology, 1992, 2, 25-40. [22] Rose, M. C., Am J Physiol Lung Cell, 1992, 263, L413-L429. [23] Yolken, R. H., Peterson, J. A., Vonderfecht, S.L., Fouts, E. T., Midthun, K., Newburg, J., J Clin Ivest, 1992, 90, 1984-1991. (241 Inoue, M., Ton, S., Ogawa, H., Tanizawa, O., Am J Clin Pathol, 1991, 96, 711-716. [25] Schwarz, A.L., Annu Rev Immunol, 1990, 8, 195-229. [26] Karlsson, K. A,, Curr Opin Struct B i d , 1995, 5, 622-635. [27] Schenkman, S., Jiang, M. S., Hart, G. W., Nussenzweig, V., Cell, 1991, 65, 1117-1125. [28] Ofek, I., Sharon, N., Curr Top Microbiol Immunol, 1990, 151, 91-114. [29] Goldkorn, I., Gleeson, P. A,, Toh, B.-H., J Biol Chem, 1989, 264, 18768-18774. [30] Dennis, J. W., Laferte, S., Waghorne, C., Breitman, M. L., Kerbel, R. S., Science, 1987, 236, 582-585.
370
12 General Stratezies for the Characterization o f Carbohydrates
[31] Lowe, J. B., Stoolman, L. M., Nair, R. P., Larsen, R. D., Berhend, T. L., Marks, R. M., Cell, 1990, 63, 475-484. [32] Walz, G., Aruffo, A., Kolanus, W., Bevilacqua, M., Seed, B., Science, 1990, 250, 11321135. [33] Foxall, C., Watson, S. R., Dowbenko, D., Cennie, C., Lasky, L. A,, Kiso, M., Hasegawa, A,, Asa, D., Brandley, B. K., J Cell Biol, 1992, 117, 895-902. [34] Starnencovic, I., Sgroi, D., Aruffo, A., Cell, 1992, 68, 1003-1004. [35] Carey, D. J., Crumbling, D. M., Stahl, R. C., Evans, D. M., J Biol Chem, 1990, 265, 2062720633. 1361 Drake, S. L., Klein, D. J., Mickelson, D. J., Oegema, T. R., Furcht, L.T., McCarthy, J. B., J Cell Biol, 1992, 117, 1331-1341. [37] Li, M., Kurata, H., Itoh, N., Yamashina, I., Kawasaki, T., J Biol Chem, 1990, 265, 1129511298. [38] Ashwell, G., Harford, J., Annu Rev Biochem, 1982, 51, 531-554. [39] Drickamer, K., Cell, 1991, 67, 1029-1032. [40] Allen, H. J., Kisailus, E. C. (Eds.), Glycoconjugates: Composition, Structure and Functions, New York: Marcel Dekker Inc, 1992. [41] Varki, A,, Clycobiology, 1993, 3, 97-130. [42] Ferguson, M. A. J., Williams, A. F., Annu Rev Biochem, 1988, 57, 285-320. [43] Stevens, V. L., Biochem J , 1995, 310, 361-370. [44] Bailey, A. J., Sims, T. J., Avery, N. C., Halligan, E. P., Biochem J, 1995, 308, 385-390. [45] Rosa, J. L., Perez, J. X., Ventura, F., Tauler, A., Gil, J., Shimoyama, M., Pilkis, S. J., Bartrons, R., Biochem J, 1995, 309, 119-125. [46] Hoshino, S., Kikkawa, S., Takahashi, K., Itoh, H., Kaziro, Y., Kawasaki, H., Suzuki, K., Katada, T., Ui, M., FEBS Lett, 1990, 276, 227-231. [47] Montreuil, J., Adv Carbohydr Chem Biochem, 1980, 37, 157-223. [48] Hanisch, F. G., Chai, W., Rosankiewicz, J. R., Lawson, A.M., Stoll, M. S., Feizi, T., Eur J Biochem, 1993, 217, 645-655. [49] Brockhausen, I., in: Glycoproteins. New Comprehensive Biochemistry; Montreuil, J., Vliegenhart, J. F. G., Schachter, H. (Eds.) Amsterdam: Elsevier, 1995, Vol. 29, pp. 201-259. [50] Hayes, B. K., Hart, G. W., Curr Opin Struct Bid, 1994, 4, 692-696. [51] Roquemore, E. P., Chou, T.-Y., Hart, G. W., Methods Enzymol, 1994, 230, 443-460. [52] Haltiwanger, R. S., Kelly, W. G., Roquemore, E. P., Blomberg, M. A,, Dong, L. Y., Kreppel, L., Chou, T. Y., Hart, G. W., Biochem SOC Trans, 1992, 20, 264-269. [53] Stults, N. L., Cummings, R. D., Glycobiology, 1993, 3, 589-596. [54] Mayor, S . , Menon, A. K., Cross, G.A.M., J Cell B i d , 1991, 114, 61-71. [55] Frankhauser, C., Homans, S. W., Thomas Oates, J. E., McConville, M. J., Desponds, C., Conzelrnann, A,, Ferguson, M. A. J., J Biol Chem, 1993, 268, 26365-26374. [56] Azzouz, N., Striepen, B., Gerold, P., Capdeville, Y., Schwarz, R. T., EMBO J, 1995, 14, 4422-4433. [57] Goochee, C. F., Gramer, M. J., Andersen, D. C., Bahr, J. B., Rasmussen, J. R., Biotechnology, 1991, 9, 1347-1355. [58] Lis, H., Sharon, N., Eur J Biochem, 1993, 218, 1-27. [59] Kornfeld, K., Kornfeld, S., Annu Rev Biochem, 1985, 54, 631-664. [60] Elbein, A. D., FASEB J , 1991, 5, 3055-3063. [61] Moremen, K. W., Trimble, R. B., Herscovics, A., Glycobiology, 1994, 4, 113-125. [62] Kean, E.L., J Biol Chem, 1991, 266, 942-946. [63] Gilmore, R., Cell, 1993, 75, 589-592. [64] Jenkins, N., Parekh, R. B., James, D. C., Nature Biotechnol, 1996, 14, 975-981. [65] Grinnell, B. W., Walls, J. D., Gerlitz, B., J Biol Chem, 1991, 266, 9778-9785. [66] Kaplan, H. A,, Welply, J. K., Lennarz, W. J., Biochim Biophys Acta, 1987, 906, 161-173. [67] Abeijon, C., Hirschberg, C.B., Trends Biochem Sci, 1992, 17, 32-36.
References
37 1
Cummings, R. D., in: Glycoconjugates: Composition, Structure and Function: Allen, H. J., Kisailus, E. C. (Eds.), New York: Marcel Dekker Inc. 1992, pp. 333-360. Shakin-Eshleman, S. H., Spitelnik, S. L., Lakshni, K., J Biol Chem, 1996,271, 6363-6366. Hirschberg, C.B., Snider, M.D., Annu Rev Biochem, 1987, 56, 63-87. Paulson, J.C., Colley, K. J., J Biol Chem, 1989, 264, 17615-17618. Kleene, R., Berger, E. G., Biochim Biophys Acta, 1993, 1154, 283-325. Schachter H., Curr Opin Struct Biol, 1991, 1, 755-765. Sharper, J. H., Sharper, N. L., Curr Opin Struct Biol, 1992, 2, 701-709. Van den Eijnden, D.H., Joziasse, D.H., Curr Opin Struct B i d , 1993, 3, 711-721. Hughes, R. C., in: Protein Glycosylation: Cellular; Biotechnological and Analytical Aspects: Conradt, H. S. (Ed.), Weinheim: GRF Monographs VCH, 1991, pp. 1-1 1. Carraway, K. L., Hull, S. R., BioEssays, 1989, 10, 117-121. Carraway, K. L., Hull, S. R., Glycobiology, 1991, 1, 131-1 38. Schachter, H., Brockhausen, I., in: Glycoconjugates: Composition, Structure and Function; Allen, H. J., Kisailus, E. C. (Eds.), New York: Marcel Dekker Inc., 1992, pp. 263332. Strous, G. J., Dekker, J., Crit Rev Biochem Molec Biol, 1992, 27, 57-92. Harris, R. J., Spellman, M. W., Glycohiology, 1993, 3, 219-224. Tomas, J. R., Dwek, R. A,, Rademacher, T. W., Biochemistry, 1990, 29, 5413-5422. Ferguson, M. A. J., Curr Opin Struc't Biol, 1991, 1, 522-529. Tartakoff, A.M., Singh, N., Trends Biochem Sci, 1992, 17, 470-473. Nenon, A.K., Mayor, S., Schwarz, R.T., EMBO J , 1991, 9, 4249-4258. Brewis, I. A,, Ferguson, M. A. J., Mehlert, A,, Turner, A. J., Hooper, N. M., J Biol Chem, 1995, 270, 22946-22952. Costello, L. C., Orlean, P., J B i d Chem, 1992, 267, 8599-8605. McConville, M. J., Ferguson, M. A. J., Biochem J, 1993, 294, 305-324. Ferguson, M. A. J., Biochem Soc Trans, 1992, 20, 243-256. Englund, P.T., Annu Rev Biochem, 1993, 62, 121-138. Amthauer, R., Kodukula, K., Brink, L., Udenfriend, S., Proc Nut1 Acad Sci USA; 1992, 89, 6 I 24- 6 128. I Nishimura, H., Takao, T., Hase, S., Shimonishi, Y., Iwanaga, S., J Biol Chem, 1992, 67, 17520-17525. I Kuraya, N., Omichi, K., Nishimura, H., Iwanaga, S., Hase, S., J Biochem, 1993, 114, 763765. [94] Chou, C.-F., Omary, M. B., J Biol Chem, 1993, 268, 4465-4472. [95] Levrat, C., Ardail, D., Louisot, P., lnt J Biochem, 1990, 22, 287-293. [96] Gamier, F., Rousson, R., Lerme, F., Vaganay, E., Louisot, P., Gateau-Roesch, O., Eur J Biochem, 1992, 206, 853-858. [97] Dwek, R. A., Edge, C. J., Harvey, D. J., Wormald, M. R., Annu Rev Biochem, 1993,62, 65100. [98] Lennarz, W. J., Hart, G. W. (Eds.), Methods Enzymol, 1994, 230. [99] Kamerling, J.P., H b d , K., Vliegenhart, J.F.G., in: From Clone to Clinic; Crommelin, D. J. A,, Schellekens, H. (Eds.), Dordrecht: Kluwer Acad. Publ., 1990, pp. 295-304. [loo] Parekh, R. B., Patel, T. P., Trends Biotechnol, 1992, 10, 276-280. [ I 011 Hounsell, E. F., in: Methods in Molecular Biology: Glycoprotein Analysis in Biomedicine: Hounsell, E. F. (Ed.), Totowa, NJ: Humana Press Inc., 1993, Vol. 14. [I021 Spellmann, M., Anal Chem, 1990, 62, 1714-1722. [I031 Kamerling, J. P., Pure Appl Chem, 1994, 66, 2235-2238. [ 1041 Verbert, A. (Ed.), Methods in Glycoconjugates. Reading: Hanvood Academic Publishers, 1995. [ 1051 Kamerling, J. P., Biotechnologia Aplicuda, 1996, 13, 167-180. [I061 Fuduka, M., Kobata, A. (Eds.), Glycobiology, A Practical Approach, Oxford: Oxford University Press IRL, 1993.
372
12 General Strategies for the Characterization of Carbohydrates
[lo71 Damm, J. B. L., Kamerling, J. P., Van Dedem, G. W. K., Vliegenthart, J. F. G., Clycoconjugate J, 1987, 4, 129-144. [lo81 Kamerling, J.P., Gerwig, G. J., Vliegenthart, J. F. G., Clamp, J. R., Biochem J, 1975, 1.51, 491-495. [lo91 Kamerling, J. P., Vliegenhart, J. F. G., in: Clinical Biochemistry: Principles, Methods, Applications. Lawson, A.M. (Ed.), Berlin: Walter de Gruyter, 1989, pp. 176-263. [IIO] Gerwig, G. J., Kamerling, J. P., Vliegenthart, J. F. G., Carbohydr Res, 1979, 77, 1-7. [ l l l ] Menon, A.K., Methods Enzymol, 1994, 230, 418-442. [112] Schneider, P., Ferguson, M. A. J., Methods Enzymol, 1995, 250, 614-640. [113] Takasaki, S., Mizuochi, T., Kobata, A., Methods Enzymol, 1982, 83, 263-268. [114] Fox, S., Genetic Engineering News, 1991, Nov/Dec. [115] Patel, T. P., Bruce, J. A,, Merry, A,, Bigge, C., Wormald, M., Jaques, A,, Parekh, R. B., Biochemistry, 1993, 32, 679-693. [I161 Bendiak, B., Cummings D.A., Carbohydr Res, 1986, 1.51, 89-103. [117] Zinn, A. B., Plantner, J. J., Carlson, D. M., in: The Clycoconjugates. Horowitz, M. I., Pigman, W. (Eds.), New York: Academic Press, 1977, pp. 69-85. [118] Whistler, R.L., BeMiller, J.N., Adv Carbohydr Chem, 1985, 13, 289-329. [119] Elder, J. H., Alexander, S., Proc Natl Acad Sci USA, 1982, 79, 4540-4544. [I201 Plummer, T. H., Jr, Elder, J. H., Alexander, S., Phelan, A. V., Tarentino, A. L., J Biol Chem, 1984, 2.59, 10700-10704. [121] Donald, A. S. R., Feeney, J., Biochem J, 1986, 236, 821-828. [122] Maley, F., Trimble, R. B., Tarentino, A.L., Plummer, T. H., Jr., Anal Biochem, 1989, 180, 195-204. [123] Stenbe, K., Gross, V., Hosel, W., Thran-thi, T. A., Decker, K., Heinrich, P.C., Glycoconjugate J, 1986, 3, 247-254. [124] Chu, F.K., J Biol Chem, 1986, 261, 172-177. [125] Langer, B.G., Hong, S. K., Schmelzer, C.H., Bell, W.R., Ana2 Biochem, 1987, 166, 2121257. [126] Tarentino, A.L., Plummer, T.H., Jr, Methods Enzymol, 1994, 230, 44-57. [127] Treffer, V., Altmann, F., Marz, L., Eur J Biochem, 1991, 199, 647-652. [I281 Staudacher, E., Altman, F., Marz, L., Hbrd, K., Kamerling, J. P., Vliegenthart, J. F. G., Glycoconjugate J , 1993, 9, 82-85. [129] Ishii-Karahasa, I., Iwasw, H., Hotta, K., Tanaka, Y., Omura, S., Biochem J , 1992, 288, 475482. [130] Striepen, B., Zinecker, C. F., Damm, J. L. B., Melgers, P. A. T., Gerwig, G. J., Koolen, M., Vliegenthart, J. F. G., Dubremetz, J.-F., Schwarz, R. T., J MoE Biol, 1997, 266, 797-813. [131] Bigge, J. C., Patel, T.P., Bruce, J.A., Goulding, P.N., Charles, S.M., Parekh, R. B., Anal Biochem, 1995, 230, 229-238. [132] Hase, S., in: Methods in Molecular Biology: Glycoprotein Analysis in Biomedicine: Hounsell, E.F. (Ed.), Totowa: Humana Press Inc., 1993, Vol. 14, pp. 69-80. [133] Yamashita, K., Mizuochi, T., Kobata, A., Methods Enzymol, 1982, 83, 105-126. [134] Van Pelt, J., Damm, J. B. L., Kamerling, J. P., Vliegenthart, J. F. G . , Carbohydr Res, 1987, 169, 43-51. [135] Damm, J.B. L., Voshol, H., H k d , K., Kamerling, J. P., Van Dedem, G. W. K., Vliegenthart, J. F. G., Glycoconjugate J , 1988, 5, 221-233. [136] Green, E.B., Baenziger, J.U., Anal Biochem, 1986, 1.58, 42-49. [137] Bergh, M.L., Koppen, P., Van den Eijnden, D.H., Carbohydr Res, 1981, 94, 225-229. [138] Townsend, R. R., Hardy, M., Oledino, J. D., Carter, S. R., Nature, 1988, 33.5, 379-380. [139] Manzi, A.E., Diaz, S., Varki, A,, Anal Biochem, 1990, 188, 20-32. [140] Townsend, R.R., Hardy, M.R., Glycobiology, 1991, 1, 139-147. [I411 Hermentin, P., Witzel, R., Doenges, R., Bauer, R., Haupt, H., Patel, T., Parekh, R. B., Brazel, D., Anal Biochem, 1992, 206, 419-429.
References
373
[142] Van Pelt, J., Kamerling, J. P., Vliegenthart, J.F. G., Verheijen, E W., Galjaard, H., Biochim Biophys Acta, 1988, 965, 36-45. [143] Dua, V.K., Dube, V.E., Li Y.-T., Bush, C. A,, Glycoconjugate J, 1985, 2, 17-30. 11441 Bendiak, B., On; J., Brockhausen, I., Velh, G., Phoebe, C., Anal Biochem, 1988, 17.5, 96105. [I451 Henderson, S. K., Henderson, D. E., J Chromatogr Sci, 1986, 24, 198-203. [146] Mutsaers, J. H. G.M., Van Halbeek, H., Vliegenthart, J. F. G., Tager, J. M., Reuser, A. J. J., Kroos, M., Galjaard, H., Biochim Biophys Acta, 1987, 911, 244-251. [147] Debray, H., Montreuil, J., Lis, H., Sharon, N., Carbohydr Res, 1986, 151, 359-370. [148] Narasimhan, S., Freed, J. C., Schachter, H., Carbohydr Res, 1986, 149, 65-83. [ 1491 Liener, I. E., Sharon, N., Goldstein, I. J. (Eds.), The Lectins: Properties, Function and Applications in Biology and Medicine. Orlando: Academic Press, 1986. Sharon, N., Lis, H., Lectins. London: Chapman & Hall, 1989. Kobata, A., Endo, T., J Chromatogr, 1992, 597, 111-122. Lis, H., Sharon, N., Annu Rev Biochem, 1986, 53, 35-67. Kobata, A,, Anal Biochem, 1979, 100, 1-14. Kobata, A,, Eur J Biochem, 1992, 209, 483-501. Mizuochi, T., in: Methods in Molecular Biology: Glycoprotein Analysis in Biomedicine: Hounsell, E.F. (Ed.), Totowa: Humana Press Inc., 1993, Vol. 14, pp. 55-68. Jay, A,, J Carbohydr Chem, 1996, 15, 897-923. Hakomori, S. I., J Biochem Tokyo, 1964, 5.5, 205-208. Stellner, K., Saito, H., Hakomori, S. I., Arch Biochem Biophys, 1973, 155, 464-469. Harrison, A. G., Cotter, R. J., Methods Enzymol, 1990, 193, 3-37. McCloskey, J. A., Methods Enzymol, 1991, 193, 539-768. Settineri, C. A,, Burlingame, A. L., in: Carbohydrate Analysis-HPLC and Capillary Electrophoresis. El Razzi, Z. (Ed.), Elsevier Amsterdam, Sciences, 1995, pp. 447-5 14. Dell, A,, Oates, J. E., in: Analysis of Carbohydrates by GLC and MS, Bierman, G. J., Mc Ginnis, G. (Eds.), Boca Raton, FI.: CRC Press, 1988. Puffing, K.L., Welpy, J.K., Huang, E., Henoin, J.D., Anal Chem, 1992, 64, 1440-1448. Linsley, K. B., Cham, S.-Y., Chan, S., Reinhold, B. D., Lisi, P. J., Reinhold, V. N., Anal Biochem, 1994, 219, 207-217. Hillenkamp, F., Karas, M., Methods Enzymol, 1991, 193, 283-298. Cotter, R. J., Anal Chem, 1992, , 1027-1031. Billeci, T. M., Stults, J.T., Anal Chem, 1993, 65, 1709-1716. Stahl, B., Klabunde, T., Witzel, H., Krebs, B., Steup, M., Karas, M., Hillenkamp, F., Eur J Biochem, 1994, 220, 321-330. Sutton, C. W., O’Neill, J. A., Cottrell, J. S., Anal Biochem, 1994, 218, 34-36. Smith, K.D., Harbin, A.M., Carruthers, R.A., Lawson, A.M., Hounsell, E.F., Biomed Chromatogr, 1990, 4, 261-266. Car, S. A., Barr, J. R., Roberts, G.D., Anumalu, K. R., Taylor, P.B., Methods Enzymol, 1990, 193, 501-518. Huberty, M.C., Vath, J.E., Yu, W., Martin, S.A., Anal Chem, 1993, 65, 2791-2800. Liu, J., Volk, K. .I. Kerns, , E. H., Klohr, S. E., Lee, M. S., Rosenberg, I. E., J Chromatogr, 1993, 632, 45-56. Dorland, L., Haverkamp, J., Schut, B. L., Vliegenthart, J. F. G., Spik, G., Strecker, G., Fournet, B., Montreuil, J., FEBS Lett, 1977, 77, 15-20. Vliegenthart, J.F. G., Van Halbeek, H., Dorland, L., Pure Appl Chem, 1981, 53, 45-77. Vliegenthart, J. F. G., Dorland, L., Van Halbeek, H., Adv Carbohydr Chem Biochem, 1983, 41, 209-374. Kamerling, J. P., Vliegenthart, J. F. G., Biol Magn Reson, 1992, 10, 1-194. Doubet, S., Bock, K., Smith, D., Darvill, A,, Albersheim, P., Trends Biochem Sci, 1989, 14, 475-477. Van Kuik, J. A,, HBrd, K., Vliegenthart, J.F. G., Carbohydr Res, 1992, 235, 53-68.
374
12 General Strategies for the Characterization o f Carbohydrates
Feizi, T., Bundle, D., Curr Opin Struct Biol, 1996, 6, 659-662. Van Kuik, J. A,, Vliegenthart, J. F. G., Trends Food Sci Technol, 1993, 4, 73-77. Dabrowski, J., in: Two-Dimensional NMR Spectroscopy: Application for Chemists and Biochemists; Croasmun, W.R., Carlson, R.M., (Eds.) New York: VCH, 1987, pp. 349386. Friebolin, H., Basic One- and Two-Dimensional NMR Spectroscopy. Weinheim: VCH, 1993. Van der Ven, F. J. M., Multidimensional NMR in Liquids. New York: VCH, 1995. H k d , K., Vliegenthart, J. F. G., in: Glycobiology: A Practical Approach. Fukuda, M., Kobata, A. (Eds.), Oxford: IRL Press, 1993, pp. 223-242. Serianni, A. S., in: Glycoconjugates: Composition, Structure and Function; Allen, H. J., Kisailus, E. C. (Eds.), New York: Marcel Dekker, 1992, pp. 71-102. I Van Halbeek, H., Curr Opin Struct Biol, 1994, 4, 697-709. I Van Halbeek, H., Methods Enzymol, 1994, 230, 132-168. I Peters, T., Pinto, B. M., Curr Opin Struct Biol, 1996, 6, 710-720. I Mer, G., Hietter, H., Lefevre, J.-F., Nature Struct B i d , 1996, 3, 45-53. I Wyss, D.F., Wagner, G., Curr Opin Biotech, 1996, 7, 409-416. [I921 De Beer, T., Van Zuylen, C. W. E. M., Leeflang, B. R., Hgrd, K., Boelens, R., Kaptein, R., Kamerling, J. P., Vliegenthart, J. F. G., Eur J Biochem, 1996, 241, 229-242. [193] De Beer, T., Van Zuylen, C. W. E. M., Hdrd, K., Boelens, R., Kaptein, R., Kamerling, J. P., Vliegenthart, J. F. G., FEBS Lett, 1994, 348, 1-6. [194] Van Zuylen, C. W. E. M., De Beer, T., Rademacher, G . J., Haverkamp, J., Thomas-Oates, J. E., Hdrd, K., Kamerling, J. P., Vliegenthart, J. F. G., Eur J Biochem, 1995, 231, 754-760. [195] Wyss, D. F., Choi, J. S., Wagner, G., Biochemistry, 1995, 34, 1622-1634. [196] Brockbank, R. L., Vogel, H. J., Biochemistry, 1990, 29, 5574-5583. [197] Gilhespy-Muskett, A.M., Partridge, J., Jefferis, R., Homans, S. W., Glycobiology, 1994, 4, 485-489. [198] Gerken, T. A,, Butenhof, X. J., Shogren, R., Biochemstry, 1989, 28, 5536-5543. [199] Dill, K., Hu, S., Berman, E., Paola, A. A., Lacombe, J. M., J Prot Chem, 1990, 9, 129-136. [200] Lommerse, J. P. M., Kroon-Batenburg, L. M. J., Kroon, J., Kamerling, J. P., Vliegenthart, J.F. G., J Biomol NMR, 1995, 5, 79-94. [201] Lustbader, J. W., Birken, S., Pilak, S., Pound, A,, Chait, B. T., Mirza, U. A,, Ramnarain, S., Canfield, R.E., Brown, J. M., J B i d NMR, 1996, 7, 295-304. [202] Fomiya, N., Awaya, J., Kuromo, M., Endo, S., Arata, Y., Takahashi, N., Anal Biochem, 1988, 171, 73-90. [203] Rice, K. G., Takahashi, N., Namiski, Y., Tran, A. D., Lisi, P. J., Lee, Y. C., Anal Biochem, 1992, 206, 278-287. [204] Edge, C. J., Rademacher, T. W., Wormald, M. R., Parekh, R. B., Butters, T. D., Wing, D. R., Dwek, R. A., Proc Nut1 Acad Sci USA, 1992, 89, 6338-6342. [205] Stan; C.M., Masada, I., Hagka, C., Seid, B., Glycobiology, 1993, 3, 511-520. [206] Hermentin, P., Doenges, R., Witzel, R., Hokke, R. C., Vliegenthart, J. F. G., Kamerling, J. P., Conradt, H. S., Nimtz, M., Brazel, B., Anal Biochem, 1994, 221, 29-41. [207] Susuki, S., Honda, S., TRAC, 1995, 14, 279-288. [208] Hu, G.F., J Chromatogr, 1995, 705, 89-103. [209] Hokke, C. H., Bergwerff, A. A,, Van Dedem, G. W. K., Kamerling, J. P., Vligenthart, J . F. G., Eur J Biochem, 1995, 228, 981-1008. [210] Krantz, S.B., Blood, 1991, 77, 419-434. [211] Inoue, N., Takeuchi, M., Asano, K., Shimizu, R., Takasaki, S., Kobata, A,, Arch Biochem Biophys, 1993, 301, 375-378. I2121 Nimtz, M., Wray, V., Riidiger, A,, Conradt, H. S., FEBS Lett, 1995, 365, 203-208. [213] Goto, M., Akai, K., Murakami, A., Hashimoto, C., Kawanishi, G., Takahashi, J., Ishimoto, A,, Chioba, H., Sasaki, R., Bioflechnology, 1988, 6, 67-71.
References
375
[214] Wojchowski, D. M., Orkin, S. H., Sytkowski, A. J., Biochim Biophys Actu, 1987, 910, 224232. [215] Matsumoto, S., Ikura, K., Ueda, M., Sasaki, R., Plant Mol Biol, 1995, 27, 1163-1172. [216] Watson, E., Bhide, A,, Van Halbeek, H., Glycobiology, 1994, 4, 227-237. [217] Storring, P. L., Gaines Das R. E., J Endocrinol, 1992, 134, 459-484. [218] Takeuchi, M., Kobata, A,, Glycobiology, 1991, I, 337-346. [219] Koury, M. J., Bondurant, M. C., Eur J Biochem, 1992, 210, 649-663. [220] Dube, S., Fischer, J. W., Powell, J. S . , J Biol Chem, 1988, 263, 17516-17521. [221] Fukuda, M.N., Sasaki, H., Lopez, L., Fukuda, M., Blood, 1989, 73, 84-89. [222] Delorme, E., Lorenzini, T., Griffin, J., Martin, F., Jacobsen, F., Boone, T., Elliott, S . , Biochemistry, 1992, 31, 9871-9876. [223] Higuchi, M., Oh-Eda, M., Kuboniwa, H., Tomonoh, K., Shimonaka, Y., Ochi, N., J Biol Chem, 1992, 267, 7703-7709. [224] Takeuchi, M., Takasaki, S., Miyazaki, H., Kato, T., Hoshi, S., Kochibe, N., Kobata, A,, J Biol Chem, 1988, 263, 3657-3663. [225] Tsuda, E., Kawanishi, G., Ueda, M., Masuda, S., Sasaki, R., Eur J Biochem, 1990, 188, 405 -4 11 . [226] Hokke, C. H., Kamerling, J. P., Van Dedem, G. W. K., Vliegenthart, J. F. G . , FEBS Lett, 1991, 286, 18-24. [227] Scudder, P., J Biol Chem, 1984, 259, 6586-6591. [228] Takeuchi, M., Takasaki, S., Shimada, M., Kobata, A,, J Biol Chem, 1990, 265, 1212712130. [229] Sasaki, H., Bothner, B., Dell, A,, Fukuda, M., J Biol Chem, 1987, 262, 12059-12076. [230] Hokke, C. H., Bergwerff, A. A,, Van Dedem, G. W. K., Van Oostrum, J., Kamerling, J. P., Vliegenthart, J. F. G., FEBS Lett, 1990, 275, 9-14. [231] Bergwerff, A. A,, Van Oostrum, J., Asselberg, F. A. M., Burgi, R., Hokke, C. H., Kamerling, J. P., Vliegenthart, J. F. G., Eur J Biochem, 1993, 212, 639-656.
Part Three Economics, Safety and Hygiene
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
13 Biosafety Yusuf Chisti
13.1 Introduction The bioprocessing industry has an undeniable record of safe operation. Yet, equally undeniable is the continuing public concern regarding the safety of biotechnology [ 11, including bioprocessing. Many hazards are associated with industrial bioprocessing: genetically modified organisms with real or perceived risks may have to be handled; highly pathogenic bacteria, viruses, and potentially contaminated substances such as blood may need to be processed; or the bioproduct may be so active that minute amounts may cause allergenic, toxic, or other activity-associated reactions in personnel exposed to it [I]. To assure safe processing, bioprocess engineers, operators, and managers must be intimately aware of the nature of the biohazard, the containment and regulatory issues, and how design and operation must satisfy the biosafety demands. This chapter examines risk assessment, biohazard containment and inactivation practices, and other biosafety issues relevant to industrial bioprocessing. Considerations relating to deliberate release of genetically modified organisms (GMOs) are not discussed as that subject is outside the scope of this chapter. Deliberate release has been reviewed elsewhere [2].
13.2 Risk Assessment Most micro-organisms and animal cells used in industrial processes pose little or no risk to human health and the environment; nevertheless, some high-risk human, animal, and plant pathogens are used (Table 13-1). By definition, a pathogen is any Table 13-1. Some commercially used hazardous microorganisms. Bordetella pertussis Clostridium tetani Corynebacterium diphtheriae Mycobacterium tuberculosis Salmonella typhi
Yellow fever virus Poliovirus Rabies virus Rubella Foot-and-mouth disease virus
380
13 Biosafety
micro-organism or virus that can cause disease in any other living organism, including other micro-organisms. Whenever a pathogen is used, it must be contained. In the long run, the trend is to move to safer processes by replacing harmful microbes with non-pathogenic recombinant producers. The nature of the viable agent - microorganisms, viruses, animal and plant cells has the greatest impact on the containment needs. Whenever possible, ‘generally recognized as safe’ or GRAS species should be used in commercial processes. Recombinant variants of GRAS organisms are preferable to non-GRAS microbes. When non-GRAS species must be used, known pathogens should be avoided, or variants not capable of producing disease should be preferred. For example, pathologically incompetent Escherichia coli K-12 strains are used in producing recombinant proteins. Many micro-organisms have a long history of safe use in food [3], and at least one previously unused species has been successfully commercialized as human food after exhaustive safety testing [4]. Use of new species or those with unknown risks must be preceded by assessment of risk [5,6] including pathogenicity testing [7]. The internal institutional biosafety committee, in consultation with the published guidelines and the Recombinant Advisory Committee of the U.S. National Institutes of Health (NIH), establishes the appropriate containment level for a strain. Understanding of how viable and bioactive substances spread and invade the body is essential to assessing risk of processing. Microbial entry into the body occurs through inhalation of aerosols and particles; transfer to mouth via contaminated hands; damaged skin; and eye-hand contact or splashes. In addition, bioactive substances may be absorbed through intact skin. Aerosols and airborne particles are particularly troublesome sources of contamination. Aerosols spread easily and widely. Particles smaller than 5 pm dry instantly and remain suspended in the environment for long periods while circulating with air currents [8]. Many operations generate aerosols, including centrifugation, homogenization, mixing, blending, aeration of liquids, leakage of liquids under pressure, and handling of solids. Laboratory procedures can contribute. For example, aerosols are generated by bursting bubbles, breakage of liquid film as in pipeting, drops falling on surfaces, splashes, ultrasonic vibrations, sampling with syringe and needle, and during pouring and siphoning [9,10]. Contaminated apparels, hands, equipment and process streams, and circulating air spread micro-organisms, as do insects, rodents, and other pests. Although relatively few infection episodes have been associated with industrial activity - the majority having occurred in research and diagnostic laboratories only about 20 % of the laboratory-acquired infections have been ascribed to specific causes [8]. Unknown or unrecognized causes for most of the events suggest a continuing insufficiency of knowledge on the links between operational practices and infection. Thus, continuing vigilance is advised. Potential sources of contamination include direct accidental inoculation (needles, sharps, cuts or abrasions, animal bites), inhalation of aerosols, ingestion, and contact of contaminated material (hands, spills, contaminated surfaces) with membranes [8]. In addition to microbes, most physiologically active fermentation products - antibiotics, mycotoxins, enzymes, steroids, hormones, vaccines, deactivated microorganisms, antibodies, and other proteins - can be disruptive to health, and certain prod-
13.2 Risk Assessment
381
ucts are highly toxic. Aflatoxins are potent carcinogens. The fermentation conditions - temperature, pH, type of substrate, agitation, metabolic energy source, dissolved oxygen and carbon dioxide, the nature and concentration of micronutrients, metal ions, and other chemicals - influence the spectrum of biochemicals synthesized by an organism [3]. Under certain environmental conditions organisms such as Aspergillusflavus and Aspergillus oryzae are known to produce lethal toxins [ l l ] . Species of the genus Claviceps and some members of other genera produce toxic ergot alkaloids. Mycotoxin production is widespread among fungi [ 121. Several types of micro-organisms are known to cause allergenic reactions when inhaled in large amounts. Implicated organisms include Actinomycetes, Aspergillus sp., Aspergillus niger, Aureobasidium pullulans, Bacillus subtilis, Baculoviruses, Candida tropicalis, Penicillium sp., and Penicillium citrinum [ 111. Allergenic reactions may be rapid, or the response may not occur until several hours after exposure, making connecting to the allergen difficult. Reactions may be extremely serious and, occasionally, fatal. Severe allergenic reactions to Bacillus subtilis proteases are well known [ l l ] . The hazard posed by nonviable bioactive material such as cytotoxic agents or endotoxins may not be eliminated by sterilization. In such cases, additional chemical decontamination of work areas, equipment and waste streams would be necessary using validated processes [13]. For example, solutions of sodium hydroxide (0.1 M) readily inactivate the botulinum toxin and are recommended for surface decontamination [ 141. Gram-negative bacteria produce thermostable endotoxins. Endotoxin-containing aerosols may be generated, for example, during cell disruption. Inhaled endotoxins are implicated in allergenic response; parenteral administration causes a pyrogenic reaction and other symptoms. Adverse reactions to endotoxins have been observed with Enterobacter agglomerans, Flavobacterium sp., Methylophilus methylotrophus, Methylomonas methanolica, Pseudomonas aeruginosa, Serratia marcescens and E. coli [ l l ] . Up to 4 % of the dry weight of E. coli K-12 has been estimated to be endotoxin [ l l ] . An action threshold value of 30 x lo-’* kg m-3 for airborne endotoxin has been recommended [ 111. Increasingly, industrial processes use recombinant micro-organisms and animal cells. Some specific issues relating to such use are discussed in Sections 13.2.1 and 13.2.2. Biosafety considerations for solid-state fermentation processes have been discussed by Chisti [3], and issues relating to composting have been treated by Stentiford and Dodds [15]. The commonly used bioprocessing schemes and individual unit operations have been detailed elsewhere [ 16-20].
13.2.1 Recombinant Microorganisms Development of containment requirements for recombinant microbes must consider environmental and ecological consequences of inadvertent release. Important considerations include survivability and colonization potential of the organism in the environment, and the organism’s ability to transfer any part of it’s genome to indigenous
382
13 Biosafety
populations [21]. Survival and persistence studies are carried out in ecosystems such as activated sludge, mammalian gastro-intestinal tract, soil, and river water [21]. Gene transfer studies may be combined with those of persistence. Such work should be ‘designed to show that the recombinant construct behaves similarly to the host in a representative ecosystem where the organism could be introduced inadvertently’ [21]. Assessments of potential biohazard should take into account characteristics of the unaltered parent, the unaltered plasmid vector, and the transposable elements. The U.S. Food and Drug Administration (FDA) has discussed these aspects in some detail [21]. As a general guide, a recombinant production strain should not have any known combination of pathogenicity, high colonization ability, and high genetic transfer competency [21]. In addition to the genes of interest, selectable marker genes are introduced into the host during transformation. Safety of such markers and the proteins they encode remains a subject of debate [22]. Specifically, the effects of any antibiotic resistance markers should be considered: such markers may facilitate colonization of gastro-intestinal tracts of fermentation process workers receiving antibiotic therapy [21]. Indeed, evidence is emerging that antibiotics in animal feeds alter the intestinal microflora in farm animals, and similar altered ecosystems become established in farm-workers that routinely contact the animals. Recombinant E. coli K-12 strains and their plasmidless hosts are unable to establish in environments consistent with various deliberate release scenarios [23]. Moreover, those strains are non conjugating and apparently incapable of transferring genes [23] to other organisms.
13.2.2 Animal Cells Cell cultures may be contaminated with pathogenic viruses (e.g., HIV, hepatitis B) and mycoplasma (e.g., MycopZasma pneumoniae). Again, immunosuppressed individuals are especially susceptible even to otherwise harmless viruses. Human cell lines are particularly high-risk and so are those derived from nonhuman primates; other mammalian cells are somewhat less risky, but may harbor agents capable of producing disease in humans (e.g., rabies, bovine spongiform encephalopathy agent, Hantaan virus in rodents). Avian and fish cells, and those from invertebrates may be lower risk. Previously uncontaminated cells can become infected during processing through human contact, or use of contaminated sera. Other than being contaminated, established animal cell lines are potentially tumorigenic. Immunocompromised individuals are particularly susceptible, but a healthy immune system may be effectively circumvented if the transformed cell culture is compatible with an individual [24]. In one case, an accidentally inoculated (needle puncture, human tumor cells) laboratory worker developed a tumor [24]. Clearly, as with any microbiological process, workers with temporarily damaged skin should not handle viable material. Because cell lines can harbor undetected viruses, and the lability of therapeutic proteins rules out the use of severe treatments that are capable of destroying viruses, a contamination-free product cannot be guaranteed; however, the risk of contamina-
13.3 Containment Levels
383
tion can be reduced to extremely low levels using a multifaceted approach including use of exhaustively characterized cells and in-process controls. Cells used in production originate in a manufacturer’s working cell bank (MWCB) that is derived from a master cell bank (MCB). Only well-characterized cell lines are used in production. Characterization must assess identity of the cell line, it’s microbial and viral contamination, genetic stability, and, for genetically modified cells, the genetic construct must be verified [24]. Guidelines for characterization of cell banks have been established by the FDA, and common practices have been described by Lubiniecki [25]. Use of a well-characterized cell in production is insufficient assurance of a safe product; additional in-process controls are necessary. Controls that need implementing include assessment of pre-harvest culture broth for relevant viruses and establishment of acceptanceh-ejection criteria for viral loads taking into account the validated capabilities of downstream virus removalhnactivation steps. Combinations of those approaches reduce risk to extremely low levels. In fact, so far not a single case of viral infection has been associated with the use of cell culture-derived biopharmaceuticals (see also [12]). Product purity and viral safety issues for cell culture-derived therapeutic proteins have been discussed further by others [ 12,25-291.
13.3 Containment Levels Containment requirements are based on assessment of potential biohazard of an agent as reflected in it’s risk classification [5]. Conventional pathogens have been categorized into four risk groups with increasingly stringent containment needs. The U. S. Centers for Disease Control and PreventionPJational Institutes of Health recognize biosafety containment levels BL1-4 for laboratory operations. A different biosafety level assignment is used for large-scale processing: GLSP, BL1-LS, BL2LS, and BL3-LS. The BLl-LS (LS = large scale) corresponds to BL1 of the laboratory scale, and so forth. The GLSP is a lower level than the BL1 designation. The BL4 has no equivalent at the large scale, and agents requiring BL4 containment are not used in commercial production. BL4 organisms require high-level containment; they are pathogenic and hazardous to laboratory personnel; and they produce transmissible diseases for which no prophylaxis or treatment exists. The biosafety level assignment considers whether an agent is pathogenic, poses a hazard to laboratory and plant personnel, is transmissible to the community, and availability of prophylaxis or treatment. Pathogenicity, or the ability to produce disease, depends on factors such as virulence, invasiveness, and infectivity. The specific risk categories for various organisms have been noted by Richardson and Barkley [14]. Table 13-2 lists the BLx containment requirements for several pathogens. In addition to the United States Public Health Service CDC-NIH biohazard classification system [14], other guidelines have been established by the World Health Organization (WHO), the European Federation of Biotechnology (EFB), and national agencies in the United Kingdom [18], Canada [30], as well as other countries. Here, the focus is on the U.S. practices that are similar to those of the other developed
384
13 Biosafety
Table 13-2.Biohazard level classification of some pathogens. Microorganism Bacteria Bordetella pertussis Clostridium tetani Corynebacterium diphtheriae Coxiella burnettii Mycobacterium tuberculosis Neisseria meningitidies Salmonella typhi Yersinia pestis Fungi Aspergillus flavus Viruses Ebola Hantaan Hepatitis B HIV Influenza Lassa Poliovirus Rabies virus Rubella Yellow fever
BL2
BL3
BL4
0 0
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0
0
The containment levels shown are for relatively small-scale operations. For large-scale work, use of the next higher containment level is recommended (except BL3). Consult Richardson and Barkley 1141 for further guidance.
countries. Practices and regulatory frameworks for other jurisdictions have been noted by Collins and Beale [20] and by Hambleton et al. [19]. Table 13-3 lists the four risk categories used by the WHO in classifying hazardous micro-organisms. Of the four biosafety levels that are relevant to large-scale work, the GLSP level is suitable for non-pathogenic and non-toxigenic strains (including genetically modified variants) that have an extended history of safe industrial use. The U.S. GLSP derives from the ‘good industrial large-scale practice’, or GILSP guidelines originally established by the Organization for Economic Cooperation and Development (OECD) in 1986 for use with suitable recombinant strains. Ideally, all industrial processes should comply with those minimal requirements. GLSP is appropriate to organisms satisfying the following criteria [311: 1. The host organism is non-pathogenic and free of adventitious agents. The host has an extended record of safe industrial use, or it has built-in incompetencies that limit it’s survival in the environment, and it has no adverse environmental consequences. 2. The genetically modified version is non-pathogenic and safe in an industrial setting, and without adverse environmental consequences.
13.3 Containment Levels
385
Table 13-3. Biohazard risk classification used by the World Health Organization. Risk group 1
(no or very low individual and community risk) A microorganism that is unlikely to cause human or animal disease.
Risk group 2
(moderate individual risk, low community risk) A pathogen that can cause human or animal disease but is unlikely to be a serious hazard to laboratory workers, the community, livestock, or the environment. Laboratory exposures may cause serious infection, but effective treatment and preventive measures are available and the risk of spread of infection is limited.
Risk group 3
(high individual risk, low community risk) A pathogen that usually causes serious human or animal disease but does not ordinarily spread from one individual to another, directly or indirectly. Effective treatment and preventive measures are available.
Risk group 4
(high individual and community risk) A pathogen that causes serious human or animal disease and that can be readily transmitted from one individual to another, directly or indirectly. Effective treatment and preventive measures are not usually available.
3. The DNA vector is well-characterized and free from known harmful sequences. To the extent possible, the size of the insert is limited to that necessary for the intended function, and the insert does not increase the stability of the construct in the environment unless required by the intended function; the insert is poorly mobilizable, and it does not transfer resistance markers to micro-organisms not known to naturally acquire resistance if such acquisition would compromise use of a drug for controlling disease in man, veterinary medicine, or agriculture. GLSP requirements further include: (i) implementation of a health and safety program; (ii) suitably trained personnel and written operational procedures; (iii) facilities, equipment, protective clothing and practices appropriate to risk; (iv) regulatory compliance with regard to discharges to the environment; (v) minimization of aerosol generation to prevent adverse risk to employee health; and (vi) a spill control plan within the emergency response plan [3 11. Often, bioprocesses must comply also with the Good Manufacturing Practices (GMP) regulations; hence, in certain areas an otherwise GLSP process may actually meet much higher standards. For example, GLSP does not require any special containment, but to meet product protection requirements, use of enclosed equipment is advisable for processing of biopharmaceuticals [32]. Genetically modified Saccharomyces cerevisiae, E. coli K-12, B. subtilis, Aspergillus oryzae, and Chinese hamster ovary (CHO) cells, for example, can be used under GLSP classification. However, it should be noted that the bioactivity and the nature of the specific recombinant product being produced could strongly affect the acceptable containment level. Thus, an otherwise GLSP species being used to make a toxic or unusually bioactive substance, may have to be contained not because of an intrinsic ‘biohazard’, but because of the nature of the product. GLSP-compliant
386
13 Biosafety
processing aims to attain the lowest practicable exposure of workplace and the environment to any physical, chemical, or biological agent [ 181. Processes requiring greater containment than GLSP should preferably be designed to one level higher than the minimum acceptable biosafety level. Typically, the cost of building to BL2-LS level is a minor increment over BL1-LS [33]. Indeed, many U.S. companies have designed production facilities to BL2-LS containment requirements where the BL1-LS measures would have sufficed [34]. Occasionally, the containment requirements for a given micro-organism differ slightly between jurisdictions. Again, the preferred practice is to err toward caution. The GLSP and BLxLS containment requirements are summarized in Table 13- 4. As noted earlier, depending on the risk, process support laboratories may have to comply with BL1-3 laboratory standards. Even a minimum containment GLSP process support laboratory should comply with good microbiological practices (Table 13-5) that are intended to protect both the operator and the product. Table 13-5 lists minimal requirements. Detailed guidelines on laboratory practices appear elsewhere [10,14,30]. In view of the different guidelines for laboratory- and large-scale operations, the question of demarcation between the two scales is important. In the U.S., processes larger than 10 L are assessed as ‘large scale’. In the U.K., there is no specific volume guideline to distinguish between large and small scale [ls]. For otherwise identical circumstances, the risk can be reasonably assumed to increase with the scale of operation, although the contrary has been argued [18]. The EFB Working Party on Safety in Biotechnology continues to produce ‘reports’ [9,24,35-391 that provide a useful perspective on biosafety issues. These reports have discussed plant pathogens [37,38], handling of micro-organisms of various risk classifications [35], assessing the impact on human health [9], hazard-based classification of micro-organisms [39], work with human and animal cell cultures [24], and general biosafety issues [36]. The regulatory approaches to biosafety issues in the European Community and the United States have been compared [311. Because the published guidelines specify only general requirements - not methods of compliance - ongoing consultations with experts and the internal biosafety committee [40] are essential. Future European standards are likely to be quite specific on ‘technical specifications, codes, methods of analysis and lists of organisms’ than the current U S . practices. In most countries, responsibility for biosafety issues is spread over several regulatory agencies. In the United States, biotechnology products and production facilities may come under the jurisdiction of the FDA, the U.S. Department of Agriculture (USDA), the Environmental Protection Agency (EPA), and the Occupational Safety and Health Administration (OSHA). The roles of the relevant agencies and the specific Acts under which they are empowered have been summarized elsewhere [34]. ~
v
Complete change. Minimize release using procedural controls. Minimize release using engineered controls. Prevent release. Comply with local environmental codes. Inactivate by validated means. g Required in all GMP-compliant processing. Also consult Richardson and Barkley [ 141 and the Canadian Medical Research Council laboratory biosafety guidelines [30].
a
GLSP
1 . Only authorized personnel allowed 2. Written procedures and training for good housekeeping and safety 0 3. Implementation and enforcement of institutional codes for hygiene and safety 0 4. Protective work wear and changing facilities 0 5. Hand washing facilities 0 6. No eating, drinking, smoking, mouth pipeting, or cosmetics application in work area 0 7. Institutional accident reporting system 0 8. Biosafety manual 9. Medical surveillance 10. Closed equipment or other primary containment for processing viable agent 11. Inactivation of culture by validated procedures before removal from closed systems 12. Enclosed sampling, material additions, and transfers to/from closed systems to prevent/minimize Ob aerosols, surface contamination, etc. 13. Treatement of exhaust gases from closed equipment 14. Inactivation of viable agent by validated means before opening of closed systems 15. Emergency plan, systems and procedures for handling large accidental spills 0 16. No leakage of viable agent from rotating seals or other penetrations and mechanical devices 17. Evaluation and monitoring of integrity of containment in closed systems 18. Evaluatiodvalidation of containment with host organism prior to using recombinant organism 19. Containment and treatment of effluent before discharge oe 20. Permanent identification of closed process equipment and use of identification on batch records 21. Display of universal biohazard sign on contained equipment when processing viable agent 22. Display of universal biohazard sign on doors to contained areas during operation 23. Low-pressure operation of process systems 24. Operations to be in a controlled area: Separate specified entry Air-locks at all entrances (including emergency exits) Readily cleaned and decontaminated finishesg Protection of utilities, services, process piping and wiring against contamination Separate gowning and washing facilities at each entrance; shower facilities in close proximity Personnel should shower before leaving controlled area Areas sealable for fumigation Sealed penetrations into area Ventilation (controlled negative pressure in area; HEPA filtered exhausts, once through ventilation)
Specification
Table 13-4. The biohazard containment requirements.
of
0 0 OC
OC
C .
0
0
0 0 0 0 0 0 0
BL1-LS
0 0 0 0 0 0 0 0
0
0 0 0 0
of 0
0 0 0
0 0 0
0
Od
d.
0
0
0
0 0 0 0
oa
0 0 0
BL3-LS
of
0
0 0 0 0
O d
.*
0 0 0 0
0 0
0 0 0
0
0
BL2-LS
388
13 Biosafety
Table 13-5. Good microbiological practices for safe handling of potentially risky microorganisms
WI. 1. Operators should have a basic knowledge of microbiology. All personnel should be aware of the risks of cultivated pathogens. Only essential personnel with the necessary training should be allowed into the work area. Practices that prevent spread of pathogens should be routinely followed. Suitable full-front laboratory apparels should be used. Work wear should remain within the work area. Hands must be washed with suitable disinfectant soap after removing latex gloves. 2. No eating, drinking, smoking, mouth pipeting, and application of cosmetics in the work area. No contact between work area materials or tools and the mouth of operators. Use of good aseptic technique. 3. Aerosol generating activities (e.g., filling of bottles and tubes, centrifugation) should be confined to biological safety cabinets. Any washing activities require special care.
4. Infected waste is sealed in containers the outside of which is disinfected prior to transfer to autoclave or incinerator. 5 . Use of validated thermal or chemical sterilization processes that assure the requisite kill.
6. Use of reliable equipment.
7. Disinfection of all work surfaces and hands after normal work.
8. Disinfection of work surfaces, floors and hands after spill of infectious material. 9. An emergency action plan with details of first aid, cleaning and disinfection. Staff trained to deal with emergencies.
10. Decontamination of laboratory clothing.
13.4 Risk Management Design and operation of a bioprocessing facility must assure safety of personnel within the facility and those in the surrounding community [41]. Protection is achieved through a combination of engineered facilities, processes, and equipment; worker training and education; use of personnel protective equipment; operational practices; validation of machinery and methodologies; controlled access to facilities; biosafety committee or subcommittee; and medical and environmental surveillance [8]. Typically, enclosed process equipment or biological safety cabinets provide primary containment to the viable material. In the event of inadvertent release from process equipment, further secondary containment is provided by the building. Environmental monitoring seeks to ascertain the satisfactory functioning of primary and secondary containment [42]. This section details the various aspects of risk management, including selection and use of biological safety cabinets which are invariably encountered at numerous stages of bioprocessing. Also included is a section on handling of biohazardous spills.
13.4 Risk Management
389
13.4.1 Biological Safety Cabinets Biological safety cabinets are the primary means of containment in process support laboratories and during early stages of culture development. Based on design and the protection afforded, biological safety cabinets are designated as Class I, 11, and 111. Capabilities of the various classes are summarized in Table 13-6 [43]. Note that laminar flow ‘clean benches’ are not biological safety cabinets and should not be used to handle potentially hazardous material. Class I cabinets (Fig. 13-1) do not protect the work area against microbial or particulate contamination. The operator is protected so long as a minimum linear air velocity of 0.4 m s-l is maintained through the front opening [43]. The cabinet is hardducted to the building exhaust system (Fig. 13-1). Class I1 biosafety cabinets (Fig. 13-2) protect the operator, the product, and the environment. The work area is bathed in downward laminar flow of particle-free, recirculated air. In addition, air from the room is drawn in through the front opening to prevent leakage of aerosols and contaminated air. The linear air flow rate at the opening should be 0.4 m s-l or greater. HEPA filtered air may be exhausted into Table 13-6. Protection capability and biohazard suitability of various classes of biological safety cabinets. Biohazard level
Cabinet class
Protection provided
BL 1-3 BL 1-3 BL 4
t
Personnel
Product
Environment
Yes Yes Yes
No Yes Yes
Yes Yes Yes
t
Fig. 13-1. Class I biological safety cabinet.
t
I I1 (A, B1-3) I11 (Bl, B2)
390
13 Biosafety
t PI
t PI
Fig.13-2. Class 11 biological safety cabinets Types A (top) and B.
13.4 Risk Management
391
the laboratory as in Class I1 Type A cabinets (Fig. 13-2), or the discharge may be hard-piped to the building exhaust system as in Class I1 Types B1-3. Type B1 cabinets recirculate part of the air over the work area, hence they may be used to process only minute amounts of volatiles. Type B2 cabinets are total exhaust devices that may be used also for some chemical containment, so long as the fumes are not susceptible to electrical ignition. As a general rule, no class of biological safety cabinets is suited to handling volatile toxic substances, but nonvolatile toxic chemicals can be handled in all classes of cabinets. Class I1 Type B3 cabinets are ducted Type A devices that like other Type B cabinets provide a minimum linear air velocity of 0.51 m s-l at the opening. All positivepressure contaminated plenums within a Type B3 cabinet are surrounded by negative-pressure chambers to prevent leakage to the environment. Class I11 biosafety cabinets (Fig. 13-3) are designed for handling BL4 biohazard agents. The cabinet is a fully sealed chamber with HEPA filtered air inlet and exhaust. The front end is provided with a sealed window and ports with heavyduty, arm-length rubber gloves. Access to the chamber is through a side-mounted, disinfectant-filled dunk tank or through a sterilizable double-door pass-through such as an autoclave. The operator, the environment and the work area are protected. Air from Class I11 cabinets must be exhausted through two HEPA filters in series, or one HEPA filter and an incinerator. A dedicated, independent exhaust system exterior to the cabinet is used to maintain air flow [43]. The enclosed work chamber is kept at a lower pressure than the laboratory. Usually a 0.5 inch (1.3 cm) water gage pressure differential is maintained [43]. Class I11 cabinets are usually installed only in maximum containment laboratories having other suitable safeguards. Biological safety cabinets rely on HEPA filters at air exhaust and/or intake to provide requisite protection to personnel, product and the environment [43,44]. Gener-
t
Fig. 13-3. Class I11 biological safety cabinet. Side-mounted dunk tank not shown.
392
13 Biosafety
ally, HEPA filters are rated to remove particles down to 0.3 pm with an efficiency of 99.97 %, but more expensive higher efficiency (99.99 % or higher) filters are available. The 0.3 pm particles are least easily filtered compared with larger or smaller particles; hence that size is used for HEPA filter performance specifications [44]. Filters are susceptible to shock-induced mechanical damage; therefore, performance and integrity of a biosafety cabinet must be certified after initial installation, after relocation, after repair, and at yearly intervals. The specific certification tests depend on the type of cabinet. Some essential tests include the downflow velocity and volume testing (Class I and 11); the inflow velocity test (Class I and 11); the negative-pressure testing (Class I1 and 111); air flow smoke patterns tests (Class I and 11); the HEPA filters leak tests (Class 1-111); the cabinet leak test (Class I1 and 111); and testing of a l m s and interlocks (Class 111). HEPA filters should be decontaminated prior to replacement. Decontamination is usually done with formaldehyde or hydrogen peroxide vapor, and decontamination provisions should be provided during installation. Class I1 cabinets usually have sensors for monitoring the pressure drop across the HEPA filter, and a low exhaust flow alarm is provided. Proper technique is essential for safe use of biological safety cabinets. The containment air curtain at the opening of Class I1 cabinets is easily disrupted by rapid sweeping arm movements into and out of the cabinet [43]. Other activities such as rapid movement of personnel around the cabinet and opening/closing of room doors also disrupt the air barrier [43]. Arm movements into/out of the cabinet should be slow and perpendicular to the front face of the cabinet [43]. The number of arm entries should be minimized by preparing a checklist of the required materials and placing them in the cabinet. The seating height should be adjusted so that the face of the operator is above the front opening [43]. Manipulations should be delayed for about 1 minute after armshands are placed inside the cabinet [43]. Arms, hands or other objects should not rest across the front grill or the room air may flow into the sterile work area. All manipulations inside the cabinet should be at least 10 cm (4 inches) from the front grill [43]. Any aerosol-generating equipment should be operated in the rear of the cabinet [43]. Used pipets should be discarded into a horizontal, disinfectant-containing discard tray kept within the cabinet [43]. Potentially contaminated material should not be brought out until after surface decontamination with a suitable disinfectant [43]. The cabinet air blower should be switched on at least 3-5 minutes before commencing work [43]. The work surfaces and the interior walls should be disinfected before and after use by wiping with 70 % ethanol or other suitable disinfectant. If chlorine bleach is used, a second wiping with sterile water is needed to remove residual chlorine, which is corrosive to stainless steel [43]. Also, any material and containers placed in the cabinet should be wiped with 70% ethanol to reduce risk of contamination. Consult Richmond and McKinney [43] for additional guidance on selection, installation, and proper use of biological safety cabinets. Management of spills within a biosafety cabinet is discussed in Section 13.4.2.
13.4 Risk Management
393
13.4.2 Spill Management Primary containment is provided by fully closed process equipment such as fermenters, centrifuges, heat exchangers, and pumps. However, even the best designed primary containment can fail; therefore, emergency response procedures must be inplace. Accidental release from process equipment poses a potential risk to employees and the local environment [45]. The extent of risk depends on the pathogenicity, virulence, invasiveness, and infectivity of the agent, and the volume being handled [45]. Release or spill may occur within the confines of a biological safety cabinet, in unconfined areas within a facility, or more broadly due to catastrophic failure of a process equipment [45]. For spills within a biosafety cabinet, immediate decontamination should be effected by a trained, suitably equipped (latex gloves, laboratory coat, safety glasses) technician while the cabinet air circulation system continues to operate [45]. Decontamination requires flooding the work tray with a disinfectant while minimizing aerosol generation. Thorough contact of the spill and the disinfectant is necessary for a preferred minimum of 30 minutes. The spill is then absorbed into disposable cloth or paper towels and discarded into autoclavable bags. The work surface, the cabinet walls, and any equipment inside is wiped with a disinfectant-soaked cloth. If the spill extends to the exhaust grills, the catch basin should be flooded (30 minutes) with disinfectant which is then drained into an autoclavable bag. A disinfectantsoaked cloth is used to wipe the grill and the catch basin [45]. The outside of the autoclavable container and bag should be wiped with a disinfectant-soaked cloth. Upon completion of cleanup, all solid material (including gloves, wiping cloth, lab coat, and any contaminated garments) that came into contact with the viable agent should be placed into an autoclavable bag. This material should be autoclaved at 121 "C for a minimum of 1 h or other suitable period that has been previously established and validated for a particular load size and distribution [45]. Once contaminated gloves and clothing have been removed, germicidal soap should be used to wash arms, hands, and face. Disinfectants suitable for most purposes are chlorine bleach (500 ppm available chlorine), iodine solution (25 -1 600 ppm available iodine), formaldehyde (0.28.0 %), and 2 % glutaraldehyde. Only chlorine bleach is satisfactory for treating liquids; other agents noted are suited for wiping surfaces, glassware, etc. Disinfectants such as quaternary ammonium compounds, ethyl alcohol, and isopropyl alcohol are not broadly effective. Chlorine bleach should not be used on stainless steel process equipment, and care should be taken to ascertain that bleach is compatible with the fluid being disinfected. The disinfection capability of chlorine bleaches declines with increasing pH. Readily accessible spill carts should be provided to deal with small unconfined spills [45]. A spill cart should have supplies of chlorine- and iodine-based disinfectants (e.g., 5 % Wescodyne, and 5 % Clorox), spill control supplies (autoclavable squeegee, autoclavable dust pan, autoclavable forceps, autoclavable biohazard bags, bucket, disposable wipes, and spill pillows), as well as protective clothing and equipment (disposable lab coats and jump suits, disposable latex gloves, dispo-
394
13 Biosufety
sable safety glasses, autoclavable boots, and half-face or full-face respirator with HEPA filter cartridges). The following treatment procedure, adapted from Van Houten [45] is recommended: Warn others of spill and leave the area holding your breath to avoid inhaling potentially hazardous aerosols; remove contaminated clothing, folding contaminated areas inwards and discard into an autoclavable bag; wash potentially contaminated body areas as well as face, arms, and hands with germicidal soap; shower if necessary; wear protective clothing (disposable lab coat, latex gloves, safety glasses, autoclavable boots) and, if necessary, HEPA filter-equipped half or full-face mask; enter the area with the spill cart; use spill pillows to isolate floor drains if present and connected directly to sewer (i.e., not connected to a biokill treatment facility); encircle the spill with disinfectant, ensuring adequate spill-disinfectant contact while minimizing aerosolization; allow a 30 -minute contact time; pick up broken glass and other sharp objects with the forceps, dust pan, squeegee, and place them in a leak-proof autoclavable container; use disposable wipes or spill pillows to mop up the liquid and discard it into an autoclavable bag; wipe the outside of the autoclavable bags with a disinfectant-soaked cloth. Use an uncontaminated, disinfectant-soaked cloth to wipe the area of the spill. Upon completing clean-up, all solid material that came in contact with the viable agent should be placed into autoclavable bags. Bags, containers, and contaminated clothing should be decontaminated in an autoclave at 121 "C for 1 hour or other prevalidated period. Arms, hands, and face should be washed with germicidal soap. Shower if necessary. The spill cart should also be disinfected. Exposed personnel may have to undergo prophylactic or other treatments and medical surveillance in accordance with preestablished policies. Larger, unconfined spills are usually handled by especially trained spill response personnel using procedures similar to the ones noted for smaller spills.
13.4.3 Buildings and Facilities The design of a facility determines it's ability to provide secondary containment. Facilities processing especially hazardous material should preferably be located away from heavily built-up areas. In extreme cases, dispersal patterns for any inadvertently released material should be considered for all the meteorological scenarios relevant to the location. Access to the facility should be restricted to authorized personnel, with certain areas being 'off limit' to all but the relevant personnel [41]. Access control may require security fencing, electronic card controlled entry, vandal-proof exterior windows, etc. In addition to containment, the design of the facility needs to assure protection of the product. Products need to be protected against contamination, particularly microbial contamination to which they are highly susceptible [ 3 2 ] . Protection must be provided throughout manufacturing and storage [32]. All facilities need an effective insect and rodent control program. Certain bioprocess facilities may require holding or handling of infected or potentially infected or suspect animals. Design requirements for such facilities are beyond the scope of this chapter; consult Richardson and Barkley [14] for guidance.
13.4 Risk Management
395
Table 13-7. Design concepts €or biohazard containmenta. 1 . Controlled access 2 . Work areas at negative pressure relative to surroundings 3 . HEPA filtered air exhaust 4. Additional Containment of aerosol-generating activities 5. Personnel training 6. Personnel protective equipment 7. Decontamination of bioactive process wastes 8. Medical surveillance of ‘at-risk’ personnel 9. Environmental monitoring a
Adapted from Flickinger and Sansone [ 131.
The principal concepts for design and operation of biohazard containment facilities are summarized in Table 13-7. Specific features are discussed in the following sections. The guidelines given comply with the U.S. FDA recommendations [32].
13.4.3.1 Layout The layout of the facility affects efficiency of operations, the potential for containment, and prevention of cross-contamination. Attention to layout is required by the GMPs [46]. Because the flow of personnel, equipment, materials and air in the facility must be controlled, the building and the process must be closely integrated by design [47]. ‘Contained’, ‘clean’, and ‘dirty’ areas should be identified on the process flow sheets and the building layout drawings. Movement of personnel, equipment, process streams, and air across containment boundary must assure integrity of containment through a combination of engineered systems and operational protocols. In general, flows must be unidirectional [32], from clean to dirty areas, and not vice versa. The dirty and clean paths should not cross [32], and there should be no back-tracking. The biohazard containment areas and the aseptic product filling areas should be located in different wings of the facility, with no sharing of common hallways or direct access [32]. Adequate space must be provided for various uses. There should be no overcrowding in work areas, especially the contained areas. 13.4.3.2 Air Handling Quality of the ventilation air and it’s flow in a facility are crucial to containment, biosafety, as well as protection of the product. Air in a facility is handled by the heating, ventilation and air conditioning (HVAC) system. Designing the HVAC system requires classification of process areas as ‘clean’, ‘dirty’, or ‘contained’. Air-locks are used to isolate the contained areas and those with critical cleanliness requirements, from other zones. Effective containment depends critically on management of air flow and pressure differentials [32]. Area pressure differentials help to prevent airborne contaminants
396
13 Biosafety
from intruding into contamination-free parts of a facility. Pressure differentials of 1.3 mm (0.05 inches) of water are typically used between adjacent areas. Areas containing infectious agents (e.g., viral vaccine) must be maintained under negative pressure relative to the surroundings [41,48]. Otherwise, the air flow is generally from ‘clean’ to ‘dirty’ areas. Access to a contained process area should be through an air-lock that is maintained at a lower pressure than the contained area and the outer access corridor [41]. Access should be restricted (e.g., card-controlled entry). The HVAC system design should assure that unwanted air pressure differentials do not develop in the event of mechanical failure. Visual and audible indication of ventilation failure should be provided. When feasible, once-through ventilation is preferred to prevent spread of contamination or the likelihood of cross-contamination [41]. Air from the facility is exhausted usually on the roof, away from any air intake. All air from contained areas in a BL3-LS facility should be exhausted through HEPA filters, and the area should be under negative pressure with respect to the surroundings (Table 13-4). The BL2-LS facility design guidelines do not specify secondary containment through air flow management, or HEPA filtering of the area’s air supply and exhaust. Primary containment must be achieved by using closed systems or appropriate biosafety cabinets (Table 13-4). The requirements notwithstanding, HEPA filtered air supply is recommended for minimizing the potential for contaminating the product [33]. Pipework should be minimized in contained areas, and wall penetrations and light fixtures should be sealed. Aerosol build-up in work areas can be reduced by HEPA filtered ventilation with a sufficient number of air changes per hour - 20-30 are not unusual in BL2-LS processing areas [33]. The mandated minimum number of air changes in various areas may have to be exceeded to account for factors such as humidity and heat rejection in the area, it’s typical function, personnel capacity, production of vapors and fumes, and generation of aerosols. The relative humidity in most processing areas is controlled at 40-50 .+ 5 %. Relative humidities exceeding 50 % promote corrosion, whereas values lower than 40 % lead to problems with static electricity. Ventilation is discussed further by First [44], del Valle [48], and Lee [49]. Air quality is of particular concern in the processing environment of the sterile final dosage forms of pharmaceuticals. Areas where the sterile product, containers, and closures are exposed to the environment are designated as ‘critical’ areas. Examples include ‘fill rooms’ and other aseptic processing areas. For a long time, the air quality in critical areas was required to be at least Class 100, that is, no more than 100 particles of 2 0.5 pm per cubic foot of air. Higher standards are now in demand. In critical areas, the number of colony forming units (CFU) should not exceed one per cubic foot (0.03 m3) of space, and a 0.05-inch (1.3 mm) water gage positivepressure differential must be maintained relative to adjacent areas. Class 100 areas are typically contained within Class 100,000 areas in which the particle count of 2 0.5 pm particles does not exceed lo5 per cubic foot (0.03 m3). In addition, the CFUs do not exceed 25 per 10 cubic feet (0.3 m3) of air. Moreover, the Class 100,000 surrounding space must equal or exceed 20 air changes per hour. A higher number, generally 60-75 air changes per hour, must be provided in critical aseptic fill rooms [48]. The Class 100 area itself should be designed for at least 600 air
13.4 Risk Management
397
changes per hour. HEPA filtered air is almost always supplied at the ceiling, and low wall-mounted returns are preferred. Aseptic filling areas need to be maintained at positive pressure [32]. Additional issues relating to HVAC system design for protection of the product have been discussed by del Valle [48] and Dobie [50].
13.4.3.3 Construction, Finishes and Practices How easily and well a facility may be cleaned, sanitized or decontaminated depends on it’s construction, and finish. The building and room finishes are subject to GMP and containment guidelines [46,47,30,5 11. Processing areas should have non-shedding, smooth, impervious, splash-resistant finishes that are capable of being cleaned and disinfected. Even a GLSP-level fermentation area should be capable of being hosed down. Contained areas should be capable of being sealed and decontaminated by disinfectant spray and by fumigation. Decontamination procedures should be established beforehand [41] and validated. Formaldehyde vapor is a commonly used fumigant, but it is carcinogenic. In addition, mixing of formaldehyde with chlorine-containing disinfectants can produce potent lung carcinogens. Paraformaldehyde gas may be used to surface-sterilize heat-sensitive equipment. Floors must slope to drains. Drains should have a slope of at least 2 cm per linear meter to assure complete drainage. Floor drains in areas that are susceptible to spills should not be connected directly to the municipal sewer [33]; instead, the drains should be piped to the biokill system. Drains should be provided with 20 cm water trap seals to prevent back-escape of vapor, gases, and aerosols from the containment sump into the work area. Sometimes, in addition to water traps, the contained drains are provided with check valves that prevent back-flow in case of a pressure build-up in the containment sump [33]; but usually suitable venting of sumps and tanks is sufficient to preclude pressure build-up. All traps should be decontaminated after a spill, and on a regular basis by pouring several trap volumes of a disinfectant solution down the drain. Catastrophic failure of pressure vessel fermenters is unlikely, but large leakages of fluid may occur from failed valves, ports, or gaskets. All material released up to the full volume of the largest fermenter should be contained within a diked area emptying into a sump, and treated through the biokill system. After the spill has been treated, the diked area should also be disinfected (see Section 13.4.2). The spill containment and treatment system should be capable of functioning on demand [32], and it should have sufficient capacity to handle the entire process volume. A minimum containment capacity of twice the production capacity has been recommended [32] which is quite reasonable considering that the process fluid as well as the subsequent wash effluent must be contained. Generally, concrete floors with troweled-on epoxy finish are preferred in processing areas, but in some low-traffic laboratory areas welded PVC flooring (not tiles) may be used. The epoxy finish (or vinyl)-should extend at least 10 cm up the walls, and the edges should be coved. Walls are generally concrete masonry units with epoxy paint. Ceilings are suspended dry wall, painted, or epoxy finished. In aseptic areas and those processing BL1-LS or higher agents, all internal corners
398
13 Biosafety
(including wall-to-ceiling) should be preferably coved and all finishes should be flush (base flush with wall, door and window frames flush with walls). Sealed windows are the norm. In addition, any exterior windows should be break-resistant whenever high-level containment is required. Light fixtures should be flush mounted, and, in high-level containment areas, they should be of a type that can be serviced from outside the contained area. The area furnishings should be of a sanitary design, resistant to water, process chemicals, and disinfectants. Usually, work surfaces are stainless steel or epoxy tops with baked epoxy painted casework [33]. Placement of equipment and furnishings should not interfere with cleaning and disinfection [33]. Equipment may be placed on housekeeping pads with radiused edges, or raised on legs that comply with hygienic design standards. Floor penetrations should be minimized and penetrations should be fully sealed to prevent seepage [33]. Supporting equipment from wallmounted brackets, or from overhead supports is preferable [33]. Hand-washing facilities should be provided near exits in the contained areas. Sink faucets (taps) should be automatic, or elbow- or foot-operated. Suitable disinfectant soap should be provided. Although hand air dryers have been recommended [33], paper towels are preferable. Electric dryers recirculate particles and aerosols that are deposited on hands. The BL2-LS facility design has been discussed by Miller and Bergmann [33] and the essential requirements are noted in Table 13-4. A BL3 -LS contained area should have separate facilities for gowning and washing at each entrance and showering facilities should be provided in close proximity (Table 13-4). Air-locks should be provided at all entrances and exits (including emergency exits) in a BL3-LS contained area. Only a minimal number of essential personnel should be allowed into contained areas. Monitoring should be from outside the contained area, through sealed windows, intercoms, and closed-circuit television. Containment features for pilot-scale fermentation facilities for producing cytotoxic agents and oncogenic viruses have been described [13]. Some of the methods noted are no longer state-of-the-art, but are effective nonetheless. For example, more elegant and reliable contained sampling methods are now available [52]. Specific construction details of animal cell culture facilities have been noted by Donnelly [53] and Lubiniecki [25]. The GMP regulations require provision of separate facilities for handling of sporeforming microorganisms. In addition, dedicated, segregated facilities are needed for processing penicillins (and other 0-lactams) because cross-contamination with penicillins and penicillin-containing substances cannot be reasonably prevented in a multiproduct facility [12]. Separate air handling systems are necessary if a building processes penicillins as well as non-penicillin products. Similarly, in facilities producing several viral vaccines, Damm [411 recommends complete isolation of areas dealing with different viruses. Containment and decontamination capabilities must function under normal conditions as well as during emergencies [54]. Provisions and practices must be in-place for evacuation during emergencies such as fire. Consequences of loss of power on containment should be evaluated during design, and emergency power should be provided to prevent loss of containment. In addition, some essential process equipment may require standby power and steam supply.
13.4 Risk Mununement
399
13.4.4 Process Equipment The design of process machinery determines it’s primary containment capability. Moreover, specific equipment may have specific hazards associated with it’s use. For example, high-pressure equipment such as cell disruptors may generate sprays of contaminated fluid in the event that a gasket fails [%I. Similarly, aerosols are produced during centrifugation and submerged aeration of culture broth in fermenters. Design and evaluation of process equipment require identifying areas where potential leakage of contained material could occur (Table 13-8). Points of possible leakage should be minimized, better contained, or, when feasible, eliminated. Containment capabilities should be assessed for all equipment, including fermenters, centrifuges, filters, solvent extraction units, cell disruption devices, spray and freeze dryers, sterilizers and autoclaves, pumps, valves, pipes, and heat exchangers, bottling and vialing machinery, downstream purification units, and waste treatment systems. In addition, the HVAC system and the utilities should be assessed for potential of becoming contaminated. Potentially contaminated lubricating fluids, steam condensate, wash fluids, etc., should be treated through the contaminated waste system. Process equipment, control cabinets, electrical housings, switches, etc., should be splash-resistant. Alternatively, control cabinets and instrumentation may be located outside the contained area and serviced through sealed cables [40]. Utility lines (e.g., steam, air, water, vacuum) that are connected directly to process equipment are at risk of becoming contaminated [40]. Protective measures include maintaining a positive pressure in the delivery lines relative to contaminated equipment, use of microbial-grade filters and back-flow preventers. To the extent feasible, equipment entering and leaving the contained area should be decontaminated by thermal sterilization in double-door autoclaves connecting the inside and the outside of the work area. Waste should be sterilized in a separate autoclave that is not used for process sterilizations 1331. Autoclaves should have interlocked doors, and fail-safe devices that prevent opening the autoclave until a complete sterilization cycle has been implemented. Pre-vacuum type autoclaves are preferred for sterilization of solid waste and other general process uses. Sterilization cycles and load configurations should be pre-validated. Principles of thermal inactivation have been described elsewhere [ 161. Table 13-8. Points of potential release of contained material. - Shaft seals (vessels, pumps) - Flanges (pipes, valves, vessels) - Pumps - Points of entry of probes and sensors - Valve packings - Ruptured diaphragms in valves - Sample points - Perforated metal bellows - Leaking seals and gaskets anywhere
- Perforated pipes, vessels, equipment - Leakage into cooling water - Tubings and hoses - Sanitary connections - Rupture discs - Pressure relief valves - Exhaust gases - HVAC ducts, seals, filters - Waste collection and treatment system
400
13 Biosafety
13.4.4.1 Fermentation Plant Fermenters usually contain large amounts of potentially hazardous viable material. Frequently, in addition to containing the viable agent used in production, entry of any other viable material into the fermenter must be prevented. Leakages from fermenters are not uncommon and considerations relating to containment and treatment of spills have already been discussed (Sections 13.4.2 and 13.4.3). In addition, all through operation, most fermenters need to be aerated, and, the fermenter exhaust gases must also be sterilized. Usually, two 0.2 Fm rated absolute hydrophobic filters in series are used to filter sterilize the exhaust air. Those filters are often heated to above dew point to prevent condensation, and, on larger or highly aerated vessels, filters are preceded by condensers [40]. In addition, cyclonic separators installed before the filters may be used to protect the filter against contamination by foam and spray. Mechanical foam breakers may also be used [56]. In some cases, one
SEAL FLUID/ LUBRICANT
INSIDE OF VESSEL
1SEAL FACE 1
VESSEL WALL
STERILE WATER/ STEAM INLET
Fig. 13-4. Double mechanical seal with water lubrication. A mechanical seal consists of two seal rings made of materials such as silicon carbide, tungsten carbide, or carbon. The flat faces of the rings press against each other to form the seal. The small gap between the faces is lubricated by a film of liquid (either culture broth or the sterile sealing fluid) or a film of gas (in dry-running seals). One of the seal rings forms a static seal with the shaft and rotates with the shaft. The other ring makes a static seal with the vessel or the stationary seal housing, and remains stationary. Two such seals in close proximity on the same shaft constitute a double mechanical seal. A film of fluid between the running and the static faces is essential to the sealing action, and any damage (e.g., scratching) to the seal faces would produce leakage. Seals must be periodically replaced as a part of the preventive maintenance program.
13.4 Risk ManaRement
401
exhaust gas filter may be followed by an incinerator. Integrity of the exhaust filters should be checked in-place using the forward flow diffusion method after the filter is sterilized, but before use. Fermenters are protected against overpressurization with a rupture disc [57] that should be piped to a HEPA-filter-vented containmenthreatment system. Sometimes, the rupture disc is followed by a pressure relief valve so that the vessel returns to a contained state once the pressure is released. This arrangement should have an attained pressure indicator for detecting disc rupture. In addition, use of overpressure sensors to shut off the sparger air supply in case of overpressurization has been recommended [13,401, but this practice is unusual: normally, the air supply regulator is set to a pressure significantly lower than the value required to open the rupture disc; thus, in case of a blocked exhaust filter, the vessel would attain a pressure equal to the air supply pressure that is still well within the vessel's capabilities. Double mechanical seals (Fig. 13-4) with sterile lubricating water between them are used to seal agitator shafts in fermenters requiring high-level containment. The pipework required for sterilizing and lubricating the seal with sterile clean steam condensate is shown in Fig. 13-5. The pressure in the lubricating water chamber
cs
I
I
outlet shown on the seal housing correspond to those shown on the seal assembly in Fig. 13-4. For sterilization, the clean steam (CS)
402
13 Biosafety
~
(Fig. 13-4) is kept higher than that in the fermenter, hence any leakage is into the vessel [40,57]. Low-pressure alarms are recommended for the sealing fluid chamber. Miller and Bergmann [33] recommend using a ‘collection tube’ in the seal fluid chamber drain for detecting seal failure. Presumably, debris or colored matter would accumulate in the tube if the seal on the culture side failed. A conductivity sensor in the water chamber should provide a better method of detecting leakage of culture fluid into the chamber. Most fermentation media are relatively conductive, whereas uncontaminated sterile water produced by condensing clean steam is a poor conductor of electricity. Rotating seals can be eliminated altogether by using magnetically coupled agitators, but torque consideration limit the vessel size to about 800 L with animal cell culture bioreactors [57], and only to about 80 L with microbial fermenters. Large airlift bioreactors that do not require mechanical agitation are particularly suited to containment [ 16,581. In addition, airlift reactors are more reliable than the mechanically stirred ones. Added protection against microleaks of viable agents can be provided by specifying double O-ring seals on all fermenter entry ports. In situ sterilizable probes that can be removed from a fermenter during cultivation, and decontaminated in-place before exposing the surroundings are available. The fermenter flanges may have double O-ring gaskets with a zone of live steam in between. Similarly, a live steam barCLEAN STEAM
c) FILTER
8
-M-
17 18
1 I
I
I
-
a EXHAUST
19
1
CLEAN CONDENSATE
iWASTE
Fig. 13-6. Steam barrier at the harvest valve on a fermenter [52]. During culture, steam is supplied to the outlet side of the closed harvest valve through valve 11. The condensate drains through valve 16 and the steam trap to the biokill system. Inlet and exhaust air is sterilized through submicron absolute filters. A double mechanical seal at the point of entry of the agitator shaft (Fig. 13-4) assures leak-free operation.
13.4 Risk ManaPement
403
rier can be maintained on the outside of valves that connect directly to the fermenter (Fig. 13-6) [ 5 2 ] . All contaminated or potentially contaminated condensate should drain to the biokill system. Closed systems should be sampled such that exposed surfaces are not contaminated, and no aerosols are released [40]. This requirement must be met at the BL2-LS and higher. Needle-and-syringe sampling through rubber septa does not meet containment requirements. Moreover, hypodermic needles have been frequently implicated in accidental inoculation of handling personnel. Contained sampling that releases no viable aerosols is detailed in Fig. 13-7 [57]. In contrast, an uncontained sampling system is depicted in Fig. 13-8. The contained sampling principles detailed in Fig. 13-7 can be incorporated in automated sampling devices one of which, available from Bioengineering AG, is shown in Fig. 13-9. The sample container should be opened within a suitable biosafety cabinet. The acid and alkali reservoirs in BL3-LS fermentation plants should be stainless steel pressure vessels that are hardpiped to the fermenter [40]. For lesser level containment, glass reservoirs connected using silicone rubber or other similar tubing and peristaltic pumps are acceptable for pH control. Hardpiping, as opposed to using rapid connection couplings, is preferred for minimizing aerosol generation when high-level containment is necessary [ 131. Culture transfer between fermenters should CLEAN STEAM
a
FILTER SAMPLE TUBE
Fig. 13-7.Sterile, contained, aerosol-free sampling [52]. (a) Prior to sampling, the fermenter sampling valve 4 is closed and a clean steam barrier is maintained on the outside of the valve (steam supply valve 1) and condensate is removed (valve 2) to the contained drain. For sampling, valves 2 and 1 are closed, and the system is allowed to cool. The steam trap and valve 2 assembly is disconnected at the sanitary quick coupling. A presterilized (autoclave) sample container (b) having a 0.2 pm breathing filter is attached to the fermenter as in (c). Valve 3 of the sampling device remains closed. The steam trap assembly is reconnected as shown in (d). Valves 1 and 2 are opened in sequence to sterilize the connection using steam at 121 "C for 25 minutes. Valves 2 and 1 are now closed in sequence. After the assembly has cooled, the sample is withdrawn by opening valves 3 and 4. After the sample has been collected, valves 4 and 3 are closed, and the connection is resterilized by opening valves 1 and 2. Upon cooling, the connection is returned to state (a), and the contained sample container is opened in a biological safety cabinet. The system shown is suitable for biohazard containment as well as sampling of bioactive substances that are inactivated by thermal sterilization.
404
13 Biosafeiy
0 FERMENTER WALL @ @ @
STEAMVALVE
@
SPRING
SOCKET FLUIDENTRY
INSIDE
a
00 0 SAMPLE/ CONDENSATE
Fig. 13-8. Assembly for uncontained sampling (shown closed). The sampling assembly is mounted in port 0 on the wall 0 of the fermenter. For sterilization, steam supplied through valve CC surrounds and enters the sample pipe at 0; condensate issues from the sample outlet. Once the assembly has sterilized, the steam supply is shut off and the system is allowed to cool. The sample is withdrawn by pushing the knob 8 to move the fluid inlet 0 into the fermenter. Releasing the knob causes the spring 0 to push the valve into closed position. (Diagram courtesy of Bioengineering AG.)
13.4 Risk Management
405
be through permanent hardpiped transfer lines in BL3 -LS facilities [40]. Pertinent transfer practices have been described by Chisti [52]. A transfer system meeting the BL2-LS criteria is shown in Fig. 13-10 [52]. Although the needle-and-diaphragm type of connections are still frequently used, particularly during inoculation offermenters, latest designs of safety connection devices (Fig. 13-11) have virtually eliminated the need for needle-type connectors. Fully contained processing within closed equipment is feasible and has indeed been implemented, for example, in production of recombinant human interferon using E. coli K-12 [59]. Facilities manufacturing vaccines such as mumps, measles, CLEAN STEAM
+-
INOCULUM
-
WASTE
Fig. 13-10. A transfer system meeting the BL2-LS standards [52]. For transferring inoculum from bioreactor 1 to bioreactor 2, a pipe section is connected (sanitary couplings) between points A and B on the transfer plate. The entire transfer pipe between the two fermenters is now sterilized by supplying steam through valves 1 and 3 while valves 6 and 7 are open. The condensate drains through valves 2 and 4. Valves 5 and 8 remain closed. Once the system has sterilized and cooled, valves 5 and 8 are opened, and contents of fermenter 1 are transferred to fermenter 2 by pressurizing (sterile air) vessel 1 relative to vessel 2. Upon completion of the transfer, valves 5 and 8 are closed. The entire transfer line is steam sterilized and cooled prior to removing pipe section A-B. As noted by Chisti [52],the correct sequencing of the various valves is important during sterilization and transfer. All contaminated condensate is piped to the biokill system.
7
Fig. 13-9. Contained sampling. Normally, valve 0 is closed by the action of spring 0 and the needle @ is retracted into the housing. For sampling, a closed sampling bottle with a rubber diaphragm cap 0 and a breathing filter @ is attached to the sampling device. Steam is now run through 0 to sterilize the sample path, including the outside of the rubber diaphragm on the sampling bottle. The condensate is withdrawn at @. Once sterilization is complete and the assembly has cooled, valve 0 is opened by pneumatic action of compressed air supplied to chamber 0. The spring is compressed, the needle moves through the rubber diaphragm into the bottle, and sample flows in. Releasing the air pressure in chamber 0 closes the valve by the action of spring 0.The assembly must be re-sterilized and cooled before the sample bottle is removed. The entire sampling operation, including installation and removal of bottles can be automated, and sampling can be done at pre-programmed intervals. (Diagram courtesy of Bioengineering AG.)
406
13 Biosafety
CLOSED
OPEN
FER M ENTER ~KNURLED NUT NOT SHOWN)
Fig. 13-11. A safety connection valve for coupling inoculum, acid, alkali, and antifoam reservoirs to the fermenter. The nipple 0 on the closed valve assembly is connected to transfer tubing attached to an empty reservoir equipped with a breathing filter. The entire set-up (safety connection valve, transfer tube, and reservoir with filter) is autoclaved and cooled. The reservoir is filled with culture fluid inside a biological safety cabinet. The valve body @ is now installed in a port on top of the fermenter. The fermenter is sterilized after installation of all reservoirs (e.g., inoculum bottle, acid and alkali containers), and cooled. For inoculation, the protective clamp 0 is removed and the valve 0 is pushed in to allow pumping of the fluid into the fermenter. The system shown is satisfactory for GLSP processing, but not for higher-level containment.
rubella, varicella, or hepatitis need particular attention to containment [411. Fermenter design practices noted by Chisti [52,57] are generally sufficient for GLSP and BL1-LS operations, and can be easily extended to BL3-LS level (e.g., use of double mechanical seals; two exhaust filters in series). More general design issues relating to equipment for submerged culture [16,52,57] and solid-state culture [3] have been addressed elsewhere. 13.4.4.2 Downstream Processing During processing, the viable agent should be removed or inactivated as early as feasible in the process sequence. This usually means inactivation upon completion of fermentation, prior to downstream processing. Inactivation of E. coli K-12 using sulfuric acid in commercial processing has been noted [59]. Alternatively, cells may be
13.4 Risk Management
407
removed by microfiltration or ultrafiltration; however, operations such as centrifugation and macroporous filtration (e.g., filter presses) do not remove all viable particles. Once the viable agent has been deactivated or removed, and the bioproduct poses no special risk, subsequent processing may proceed in open systems [18] with due regard to the GMP-dictated requirements for protecting the product. When inactivation or removal are not feasible (e.g., live vaccines), downstream processing must be contained. Containment of some downstream processing machinery - particularly centrifuges and cell-disruption devices - has been discussed by Deans and Stewart [60], and other commercially relevant mechanical cell-disruption equipment has been described [16,55]. In addition to the viable agent, the physiological and toxicity profiles of products and by-products constrain the choice of process equipment. Certain process schemes may be ruled out by the extent of containment needs and the amount and types of waste streams that would need to handled. In one case, a difficult-to-contain rotary drum filter that also generated difficult-to-dispose filter aid-mixed solids was replaced with fully contained ultrafiltration for recovery of cefoxitin [61]. Sometimes, primary containment within the process equipment may not be feasible, or contaminated machinery may have to be dismantled - for example, during harvesting of solids from tubular bowl batch centrifuges - and processing within enclosures would be necessary for primary containment. Isolation of aerosol-generating process equipment in HEPA-filter-exhausted, negative-pressure enclosures is a suitable means of containment [ 13,411. The containment room doors should be interlocked with the equipment so that the doors can be opened only when the equipment has stopped running, and sufficient time has elapsed for several space volumes of air to be exhausted from the contained area. Ideally, the equipment itself should be designed for primary containment, and isolation within enclosures should be an added safety measure to contain accidental release from high-pressure devices such as centrifuges and cell disruptors. Performance of the primary and the secondary containment should be validated and monitored regularly. Multiproduct facilities may require product-dedicated process equipment to eliminate the likelihood of cross-contamination. This is especially so for equipment that cannot be reasonably freed of all traces of a product. For example, chromatography media and membrane filters may have to be dedicated to specific products. The use status - clean, in use, dirty, washed, sterile, etc. - of equipment should be clearly identified at all times. Downstream purification and formulation areas should process only one product at a time. A label control program must be in place. Handling of a concentrated, bioactive product can be the most dangerous part of bioprocessing [ 111. Fine powders are easily aerosolized; hence, handling of freezedried cultures or toxins can be particularly hazardous [62]. Because of their aerosol-generating potential, spray-dryers and freeze-dryers require special attention to containment [ 111. Freeze-drying of biohazardous substances has been discussed by Adams [62]. As a general rule, automation and fully enclosed mechanized operation can significantly enhance containment while reducing the need for human-process contact.
408
13 Biosafety
13.4.4.3 Other Systems Heat exchange devices are frequently encountered in bioprocessing plants. Corroded heat exchangers (vessel jackets, plate heat exchangers, condensers, shell-and-tube exchangers, etc.) or those with leaking gaskets can contaminate the cooling fluid with the viable agent. Heat exchange equipment should be selected with regard to containment requirements, and cooling water may have to be monitored for contamination. Clean-in-place (CIP) systems are another common feature of bioprocessing facilities. Automated CIP systems reduce exposure of personnel to hazardous material and assure consistent cleaning. The CIP system design has been treated by Chisti and Moo-Young [56]. For contained facilities, fully closed CIP systems should be used. Attention should be given to cleanability and sterilizability of the CIP system itself, and generation of aerosols should be prevented. In view of the containment and crosscontamination considerations, certain process areas may require dedicated CIP systems.
13.4.5 Personnel Protective Equipment Certain process operations are diffcult to contain, and even the best designed primary containment can fail. Therefore, use of personnel protective clothing appropriate to risk is essential. Lab coats over street clothing are satisfactory for BL2-LS [33] and lower-rated containment areas. Correct gowning room practice is essential to preventing spread of contamination. Protective clothing - gowns, shoe coverings, head covering, face masks, and gloves - should be removed in the proper fashion before leaving the work area [41]. BL3 -LS operations, for example, in hepatitis vaccine production, require that personnel remove all street garments down to underwear and don ‘bunny suits’, face masks, and gloves, prior to entering the work area [41]. Protective clothing that completely isolates an individual from the environment is sometimes used [ 131. Depending on the characteristics of the product, such protective suites may be required even for BL2-LS or lower-rated processes. This type of isolation is provided by positivepressure suits with battery-operated, forced air supply drawn from the surrounding environment through HEPA filters. Head-and-shoulder half suites with HEPA filtered air supply can also be used over other disposable protective clothing. The suits are equipped with low-battery alarms. Use of showers is necessary prior to leaving the degowning area of a BL3-LS biohazard containment zone [32]. Lower level containment areas should be equipped with hand washing facilities near the exit from the area. Sinks with automatic or hand- or foot-operated faucets (taps) are used. Suitable germicidal soap should be provided. Awareness of the biohazard is important to risk reduction. Consequently, BL2-LS and higher containment areas must display the universal biohazard sign (Fig. 13-12) on the entrance to the contained area. Additional information should include the containment level, the specific biohazard agent, any special entry requirements, and emergency contact details of responsible personnel.
13.4 Risk Management
409
Fig. 13-12. Universal biohazard sign.
13.4.6 Personnel Training Personnel training is an essential part of safe bioprocessing. Even the best designed facilities, equipment, and practices will fail to provide the intended protection if the operators do not have the knowledge, the training, and the right attitude to personal safety, that of colleagues, the product, and the community. Training should be provided in specific processing methods, operation of equipment, use of personnel protective equipment, gowning practices, aseptic and good microbiological technique (see Table 13-3,containment and biosafety measures consistent with the hazard, emergency procedures, and authorized practices. Written operational protocols should identify the specific, actual or potential, hazards. Established practices should be strictly followed [32]. Personnel should be supervised to assure consistent use of prescribed practices [32,40]. Training in Good Manufacturing Practices is also required, and should be a continual process.
13.4.7. Medical Surveillance Routine medical surveillance appropriate to risk is recommended. For example, electrocardiograms monitoring of individuals working with cardiotoxic substances [ 131, and seroconversion of individuals handling antigens. Although medical surveillance by itself does not protect against exposure, surveillance is useful in early detection and treatment. Surveillance also helps in identifying procedural or mechanical lapses. Medical surveillance is especially necessary when pathogenic, potentially pathogenic, or new micro-organisms are being investigated, or when the nature of the hazard is unknown. In addition to the viable agent, the bioactivity of the product or any contaminants may pose a health hazard. Thus, for example, a nonpathogenic recombinant species may pose no risk beyond that associated with the corresponding wild strain, but the product of the inserted foreign gene may be highly bioactive, allergenic, or otherwise toxic.
410
13 Biosufety
When suitable vaccines are available, the process personnel as well as those providing support services but not directly working in the process areas (e.g., management, administration), should be immunized [41]. A ‘wait time’ is necessary for development of immunity. Process workers should show resistance to the infective agent before being allowed into the work area [41]. Furthermore, it should be recognized that vaccination may not guarantee protection against a high dose of the etiologic agent [8]. The workplace should be receptive to reporting of illnesses and accidents. Even apparently minor incidents - for example, being scratched while cleaning a process vessel - should be reported and recorded. Personnel who are ill, and those with open sores and cuts, should not be allowed into critical work areas. Immunocompromised individuals, those with diseases such as cancer and diabetes, those undergoing antimicrobial, steroid or immunosuppressive therapy [8] are especially at risk. Attention to protection of peripheral support staff, for example the maintenance and cleaning personnel, is especially important because they may not have the knowledge or training for the potential risks [3].
13.4.8 Biowaste All contaminated liquid, solid, and gaseous waste from a facility must be decontaminated prior to release. Solid waste is generally autoclaved or incinerated. Gases are filter sterilized and/or incinerated. Liquid effluent is collected into a containment sump and treated through the biokill system. All decontamination procedures should be validated, and the treated material should be examined for sufficiency of kill before being released. The regulations (OSHA, NIH/CDC, EPA, U.S. Postal Service, etc.) relating to disposal and shipping of biohazardous wastes in the United States have been discussed by Turnberg [63]. Effluent from the containment sump may be chemically disinfected or heat sterilized. Sterilization may be batchwise or continuous using direct steam injection or indirect heating. Good mixing of chemical additives and uniform sterilization temperature must be achieved for defined periods. Waste decontamination areas are generally held at negative pressure which is mandatory when decontaminating BL3 -LS effluent. There should be provisions for preventing accidental release of untreated material to sewer. For example, a locked effluent drain valve may be employed; the valve is opened only when a treated batch is released. Use of chemical disinfection prior to thermal sterilization is recommended to reduce the hazard in case of inadvertent effluent release (e.g., rupture disc failure). The decontamination process may be automated. The sump and the sterilization tanks should be vented only through HEPA filters that may have to be heated to prevent condensation. The filtered exhaust from the waste decontamination tank may have to be treated for odor control if odor is a nuisance [33]. Odor may be controlled by incineration, scrubbing, or absorption. Effluent decontamination has been discussed further by Wirt et al. [54]. Other relevant effluent management issues have been examined by Court [64]. Miller and
Abbreviations
4 11
Bergmann [33] have discussed treatment of BL2-LS effluent. Design features of batch thermal biokill systems have been described by Kossik and Miller [65]. In addition, the hygienic design practices noted by Chisti [52,57] for fermentation plant apply also to biokill machinery. After decontamination, the biohazardous waste is generally disposed of using the same practices that apply to other nonbiological waste. Sometimes special treatment is necessary to destroy nonviable bioactive or otherwise environmentally harmful substances [63,66,67].
13.5 Concluding Remarks Few bioprocess engineers have experienced a major facility design and construction project; consequently, there is little awareness of industrially relevant biosafety issues. Knowledge of the hazards posed by viable agents and biological materials is essential to safe practice of biotechnology. Industrial processing of any agent must be preceded by assessment of risk and evaluation of the containment needs. Design of the process, process machinery, buildings and the operational practices must be consistent with the biosafety requirements, GMP considerations and other regulatory demands. Process systems and practices should be validated to assure that the intended capability is attained. All this must be complemented with a trained and knowledgeable workforce. Safe bioprocessing rests on a triad of training, procedures, and engineered design [33]. Inadequacies in any of those aspects could lead to safety failures. The practice and the technology for safe bioprocessing is continually evolving. Guidelines given in some recent literature are already quite dated [5,40,68,69]; therefore, consultation with experts is advised for information on state-of the-art practices. The major containment concepts discussed here are summarized in Table 13-7.
Abbreviations BLx BLx-LS CDC CFU CHO CIP DNA EFB EPA FDA GILSP GLSP
Biosafety level x (x = 1-4) Biosafety level x (x = 1-3) large scale Centers for Disease Control and Prevention Colony forming unit Chinese hamster ovary Clean-in-place Deoxyribonucleic acid European Federation of Biotechnology Environmental Protection Agency Food and Drug Administration Good industrial large-scale practice Good large-scale practice
412
13 Biosafety
GMO GMP GRAS HEPA HIV HVAC MCB MWCB NIH OECD OSHA PVC USDA WHO
Genetically modified organisms Good manufacturing practices Generally recognized as safe High efficiency particulate air Human immunodeficiency virus Heating, ventilation and air conditioning Master cell bank Manufacturer's working cell bank National Institutes of Health Organization for Economic Cooperation and Development Occupational Safety and Health Administration Poly(viny1 chloride) United States Department of Agriculture World Health Organization
References [ 11 Chisti, Y., Trends Biotechnol, 1993, 11 (6), 265-266. Industrial bioprocess safety.
[2] Stephenson, J.R., Warner, A,, J Chem Techno1 Biotechnol, 1996, 65, 5-14. Release of genetically modified micro-organisms into the environment. [3] Chisti, Y., in: Encyclopedia of Bioprocess Technology: Flickinger, M. C., Drew, S. W. (Eds.), New York: John Wiley, 1999; in press. Solid substrate fermentations, enzyme production, food enrichment. [4] Angold, R., Beech, G., Taggart, J., Food Biotechnology, Cambridge: Cambridge University Press, 1989. [5] Kearns, M. J., Pharm Eng, 1989, 9 (4), 17-21. Containment of biological hazards: Effect of guidelines on the design of pharmaceutical facilities and process equipment. [6] Winkler, K. C., Parke, J. A. C., in: Safety in Industrial Microbiology and Biotechnology: Collins, C. H., Beale, A. J. (Eds.), Oxford: Butterworth-Heinemann, 1992; pp. 34-74. Assessment of risk. [7] Hacker, J., Ott, M., in: Safety in Industrial Microbiology and Biotechnology: Collins, C. H., Beale, A. J. (Eds.), Oxford: Butterworth-Heinemann, 1992; pp. 75-92. Pathogenicity testing. [8] Liberman, D. F., Developments in Industrial Microbiology, 1984, 25, 69-75. Biosafety in biotechnology: A risk assessment overview. [9] Lelieveld, H. L. M. et al., Appl Microbiol Biotechnol, 1995, 43, 389-393. Safe biotechnology. Part 6. Safety assessment, in respect of human health, of microorganisms used in biotechnology. [lo] Collins, C. H., Laboratory-Acquired Infections, 3rd edition, Oxford: Butterworth-Heinemann, 1993. [ l l ] Bennett, A.M., in: Biosafety in Industrial Biotechnology: Hambleton, P., Melling, J., Salusbury, T.T., (Eds.), London: Chapman Hall, 1994; pp. 109-128. Health hazards in biotechnology. [ 121 Chisti, Y., Strategies in downstream processing. In: Bioseparation and Bioprocessing: Processing Biomolecules and Cell Cultures, Vol. 11: Subramanian, G., (Ed.), Weinheim: WileyVCH, 1998; pp. 3-30. [13] Flickinger, M. C., Sansone, E. B., Biotechnol Bioeng, 1984, 26, 860-870. Pilot- and production-scale containment of cytotoxic and oncogenic fermentation processes.
References
413
[14] Richardson, J. H., Barkley, W. E., (Eds.), Biosafety in Microbiological and Biomedical Laboratories, Znd edition, U.S. Department of Health and Human Services, Washington: U.S. Government Printing Office, 1988. [15] Stentiford, E. I., Dodds, C. M., in: Solid Substrate Cultivation: Doelle, H. W., Mitchell, D. A., Rolz, C. E. (Eds.), London: Elsevier, 1992; pp. 21 1-246. Composting. [16] Chisti, Y., Moo-Young, M., in. Biotechnology: The Science and the Business: Moses, V., Cape, R. E., (Eds.), New York: Harwood Academic Publishers, 1991; pp. 167-209. Fermentation technology, bioprocessing, scale-up and manufacture. [ 171 Doran, P. M., Bioprocess Engineering Principles, London: Academic Press, 1995. [18] Vranch, S.P., in: Bioprocessing Safety: Worker and Community Safety and Health Considerations: Hyer, Jr., W. C., (Ed.), Philadelphia: American Society for Testing and Materials, 1990; pp. 39-57. Containment and regulations for safe biotechnology. [19] Hambleton, P., Melling, J., Salusbury, T. T. (Eds.), Biosafety in Industrial Biotechnology, London: Chapman Hall, 1994. [20] Collins, C. H., Beale, A. J., (Eds.), Safety in Industrial Microbiology and Biotechnology, Oxford: Butterworth-Heinemann, 1992. [21] Jones, R.A., Matheson, J. C., J. Znd. Microbiol. 1993, 11, 217-222. Relationship between safety data and biocontainment design in the environmental assessment of fermentation organisms - An FDA perspective. [22] Ozcan, S.,Firek, S . , Draper, J., Trends Biotechnol,l993, I 1 (6), 219. Can elimination of the protein products of selectable marker genes in transgenic plants allay public anxieties? [23] Kane, J. F., J Znd Microbiol, 1993, 11, 205-208. Environmental assessment of recombinant DNA fermentations. [24] Frommer, W. et al., AppZ Microbiol Biotechnol, 1993, 39, 141-147. Safe biotechnology (5). Recommendations for safe work with animal and human cell cultures concerning potential human pathogens. [25] Lubiniecki, A. S.,(Ed.), Large-Scale Mammalian Cell Culture Technology, New York: Marcel Dekker, 1990. [26] Roberts, P., J Chem Technol Biotechnol, 1994, 59, 110-111. Virus safety in bioproducts. [27] Rouf, S.A,, Moo-Young, M., Chisti, Y., Biotechnol Adv, 1996, 14, 239-266. Tissue-type plasminogen activator: Characteristics, applications and production technology. [28] Anicetti, V. R., Keyt, B. A,, Hancock, W. S . , Trends Biotechnol, 1989, 7(12), 342-349. Purity analysis of protein pharmaceuticals produced by recombinant DNA technology. [29] Garg, V. K.,Costello, M. A. C., Czuba, B. A., in: Purification and Analysis of Recombinant Proteins: Seetharam, S . , Sharma, S. K., (Eds.), New York: Marcel Dekker, 1991; pp. 29-54. Purification and production of therapeutic grade proteins. [30] Health and Welfare Canada and Medical Research Council of Canada, Laboratory Biosafety Guidelines, Ottawa: Minister of Supply and Services Canada, 1990. [31] Van Houten, J., Fleming, D.O., J Znd Microbiol, 1993, 11, 209-215. Comparative analysis of current US and EC biosafety regulations and their impact on the industry. [32] Hill, D., Beatrice, M., Pharm Eng, 1989, 9 (4), 35-41. Facility requirements for biotech plants. [33] Miller, S.R., Bergmann, D., J Znd Microbiol, 1993, 11, 223-234. Biocontainment design considerations for biopharmaceutical facilities. [34] Perkowski, C. A., in: Bioprocess Engineering: Systems, Equipment and Facilities: Lydersen, B. K., D’Elia, N.A., Nelson, K. L., (Eds.), New York: John Wiley, 1994; pp. 729743. Containment regulations affecting the design and operation of biopharmaceutical facilities. [35] Frommer, W. et al., Appl Microbiol Biotechnol, 1989, 30, 541-552. Safe biotechnology 111. Safety precautions for handling microorganisms of different risk classes. [36] Kiienzi, M. et al., Appl Microbiol Biotechnol, 1985, 21, 1-6. Safe biotechnology: General considerations.
414
13 Biosafetv
[37] Kuenzi, M. et al., Appl Microbiol Biotechnol, 1987,27,405. Safe biotechnology 2. The classification of microorganisms causing diseases in plants. [38] Frommer, W. et al., Appl Microbiol Biotechnol, 1992, 38, 139-140. Safe biotechnology (4). Recommendations for safety levels for biotechnological operations with microorganisms that cause diseases in plants. [39] Lelieveld, H. L. M. et al., Appl Microbiol Biotechnol, 1996, 45, 723-729. Safe biotechnology. 7. Classification of microorganisms on the basis of hazard. [40] East, D., Stinnett, T., Thoma, R. W., Developments in Industrial Microbiology 1984, 25, 89105. Reduction of biological risk in fermentation processes by physical containment. [41] D a m , P. G., in: Bioprocessing Safety: Worker and Community Safety and Health Considerations: Hyer, W. C., Jr., (Ed.), Philadelphia: American Society for Testing and Materials, 1990; pp. 58-64. The biological production facility - Design for protection of the worker and the community. Liberman, D. F., Ducatman, A.M., Fink, R., in: Bioprocessing Safety: Worker and Community Safety and Health Considerations: Hyer, W. C., Jr., (Ed.), Philadelphia: American Society for Testing and Materials, 1990; pp. 101-110. Biotechnology: Is there a role for medical surveillance? Richmond, J. Y., McKinney, R. W., (Eds.), Primary Containment for Biohazards: Selection, Installation and Use of Biological Safety Cabinets, U S . Department of Health and Human Services, Washington: U.S. Government Printing Office, 1995. [44] First, M. W., Developments in Industrial Microbiology, 1984, 25, 77-87. Ventilation for hazard control. [45] Van Houten, J., in: Bioprocessing Safety: Worker and Community Safety and Health Considerations: Hyer, W. C., Jr., (Ed.), Philadelphia: American Society for Testing and Materials, 1990; pp. 91-100. Safe and effective spill control within biotechnology plants. [46] Willig, S. H., Stoker, J. R., Good Manufacturing Practices for Pharmaceuticals: A Plan for Total Quality Control, 3rd edition, New York: Marcel Dekker, 1992. [47] Rao, A. K., Adey, H., Cherntech, 1989, October, 632-637. Designing a bioprocessing facility. [48] del Valle, M. A., BioPharm, 1989, April, 26-42. HVAC systems for biopharmaceutical manufacturing plants. [49] Lee, J. Y., BioPhann, 1989, February, 42-45. Environmental requirements for clean rooms. [50] Dobie, D., in: Bioprocess Engineering: Systems, Equipment and Facilities: Lydersen, B. K., D’Elia, N. A., Nelson, K. L., (Eds.), New York: John Wiley, 1994; pp. 641-668. Heating, ventilating, and air conditioning (HVAC). [5 I] Johnson, H. L., Stutzman, D.A., in: Bioprocess Engineering: Systems, Equipment and Facilities: Lydersen, B. K., D’Elia, N. A., Nelson, K. L., (Eds.), New York: John Wiley, 1994; pp. 67 1-708. Programming and facility design. [52] Chisti, Y., Chem Eng Prog, 1992, 88 (9), 80-85. Assure bioreactor sterility. [53] Donnelly, R. W., Pharm Eng, 1989, 9 (3), 9-12. Design and construction review of one of the first large scale mammalian cell culture facilities. [54] Wirt, G.D., Orichowskyj, S.T., Wu, J. J., Chem Eng Prog, 1991, 87 (l), 49-53. Decontaminate biotech wastes effectively. [55] Chisti, Y., Moo-Young, M., Enzyme Microb Technol, 1986, 8, 194-204. Disruption of microbial cells for intracellular products. [56] Chisti, Y., Moo-Young, M., J Ind Microbiol, 1994, 13, 201-207. Clean-in-place systems for industrial bioreactors: Design, validation and operation. [57] Chisti, Y., Chem Eng Prog, 1992, 88 (l), 55-58. Build better industrial bioreactors. [58] Chisti, Y., Airlifi Bioreactors, London: Elsevier, 1989. [59] Weibel, E. K., Chimia,1994, 48 (lo), 457-459. GMP and biosafety aspects in the production of recombinant IFN alpha-2a.
References
415
[60] Deans, J. S., Stewart, I. W., in: Biosafety in Industrial Biotechnology: Hambleton, P., Melling, J., Salusbury, T. T., (Eds.), London: Chapman Hall, 1994; pp. 149-177. Containment in downstream processing. [61] Paul, E. L., in: Bioprocessing Safety: Worker and Community Safety and Health Considerations: Hyer, W. C., Jr., (Ed.), Philadelphia: American Society for Testing and Materials, 1990; pp. 65-73. Design criteria for safety in the isolation and purification of antibiotics and biologically active compounds. [62] Adams, G. D. J., in: Biosafety in Industrial Biotechnology: Hambleton, P., Melling, J., Salusbury, T. T., (Eds.), London: Chapman Hall, 1994; pp. 178-212. Freeze-drying of biohazardous products. [63] Turnberg, W. L., Biohazardous Waste: Risk Assessment, Policy, and Management, New York: John Wiley, 1996. [64] Court, J. R., in: Biosafety in Industrial Biotechnology: Hambleton, P., Melling, J., Salusbury, T. T., (Eds.), London: Chapman Hall, 1994; pp. 240-267. Managing the effluent from bioindustrial processes. [65] Kossik, J.M., Miller, G., Chem Eng Prog, 1994, 90 (lo), 45-51. Optimize cycle times for batch biokill systems. [66] Watt, J. C., Wroniewicz, V. S., Ioli, D. F., in: Environmental Biotechnology: Omenn, G. S., (Ed.), New York: Plenum Press, 1988; pp. 307-322. Environmental concerns associated with the design of genetic engineering fac [67] Freeman, H. M., (Ed.), Standard Handbook of Hazardous Waste Treatment and Disposal, New York: McGraw-Hill, 1989. [68] Werner, R. G., in: Safety in Industrial Microbiology and Biotechnology: Collins, C. H., Beale, A. J., (Eds.), Oxford: Butterworth-Heinemann, 1992; pp. 190-213. Containment in the development and manufacture of recombinant DNA-derived products. [69] Tuijnenburg Muijs, G., in: Safety in Industrial Microbiology and Biotechnology: Collins, C. H., Beale, A. J., (Eds.), Oxford: Butterworth-Heinemann, 1992; pp. 214-238. Monitoring and validation in biotechnological processes.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
14 Process Hygiene in Production Chromatography and Bioseparation Glenwyn D. Kemp
14.1 Introduction The very nature of industrial bio-chromatography entails the potential for contamination of the equipment with biologically active materials. These materials must not be allowed to reach the final product of the process. An integral part of any chromatographic step in the production of biologically active materials is the removal of both contaminant and residual material from the chromatography system after each step. The following chapter gives an overview of potential sources of contamination and reviews methods used to remove such contaminants. The maintenance of cleanliness within a production chromatography plant is considered from the perspective of designing the equipment to both minimize initial infection and to allow the removal of any contamination which may occur. Consideration is also given to methods of sanitization which are applicable to process chromatography.
14.2 General Principles The preferred method of maintaining hygiene in process-scale chromatography systems is by cleaning-in-place (CIP), whereby the entire contact path is cleaned in situ. This is due in large part to the time and effort required to unpack and re-pack very large columns. There is also the additional risk of degeneration of the chromatography media by repeated handling and potential exposure of the operators to bioactive compounds during the packfunpack cycle, For this reason once a column has been packed and proved to have an acceptable performance it is more economical to clean the column in situ for as long as possible, i.e., until the performance is no longer acceptable (due, for example, to loss of capacity or disruption of the gel bed). In order to get the greatest benefit from effective CIP it is essential to take steps to eliminate infection or contamination of the equipment in the first place. This is best achieved by rigorous cleanliness in handling the chromatography buffers and samples and by strict adherence to current good manufacturing practices (cGMP). Al-
41 8
14 Process Hygiene in Production Chromatography and Bioseparation
though this adds time and costs to the overall process, the potential for lost time and product due to contamination makes any extra effort in pre-chromatography preparation a worthwhile investment. It should also be noted that giving adequate consideration to hygiene and CIP from the earliest possible stages of both process and system design will reap great rewards when the final production process is running. One of the best investments which can be made for process reliability is a well-designed chromatograph. It is frequently the case that the very nature of the feedstock being applied to the column will result in contamination. For example, bacterial cell lysates produced as an initial step in the recovery of recombinant protein will inevitable contain endotoxins and other proteins, DNA and RNA which must be removed from the system after each batch. In such a process there is also a high probability of some unbroken cells being present which can form a potential infection of the system resulting in increasing bio-burden over a period of time. The manufacture of recombinant proteins from the perspective of process validation was the subject of a review by Jungbauer and Boschetti [ 11. Some processes require a CIP step regardless of the presence of contaminants. In order to maintain biological activity, many blood products for example are purified chromatographically from unsterilized blood [2]. In accordance with general principles of microbiology, the pooled blood fractions must be assumed to contain pathogenic viruses. Therefore there is a pre-defined requirement to clean the media after each batch, regardless of the actual presence of viral contamination. Because of this it is often the case with very high-value blood products that the gel is simply sterilized and disposed of after each batch. However, the initial sterilization of the gel is usually carried out by an in situ step prior to removal of the gel from the column for disposal, to reduce risk to the operators. In cases such as these, where it is not possible to remove infective contaminants, the equipment being used must be designed for ease of cleaning, with maximum operator safety and the cleaning protocols used must be fully validated. It is important to note that although individual components within a system can be tested and assessed for cleanliness, a full CIP validation can only be made for a complete process under standard operating conditions as a part of the process qualification (PQ).
14.3 Definitions Sterilization Sterilization can be defined as ‘a method for removing (i.e., killing) all viable organisms’. It should be noted that sterilization may still leave pyrogens within the system, and indeed can be a source of pyrogen increase as the killed organisms break down. Sanitization Sanitization can be defined as ‘a method of lowering the total content of viable organisms to an acceptable level’. In this case it should be noted that sanitization does not necessarily entail the complete removal of viable organisms, some of
14.4 Possible Contaminants
41 9
which may remain within the system. The essence here is to keep levels of contamination under control and low enough to represent an acceptable statistical risk. A successful sanitization is usually defined as producing a significant log reduction in the number of viable organisms present. Although sanitization does not necessarily remove pyrogens, it is common for depyrogenation and sanitization to be combined in a single process step. CIP (Clean-in-place) CIP can be defined as ‘a method of sanitising a system in situ without dismantling it’. The aim of CIP is to reduce the amount of viable organisms to an acceptable level while causing the minimum disruption of the system and with the minimum exposure to the operators. SIP (SterilizeBteam-in-place) SIP can be defined as ‘a method of sterilising a system in situ without dismantling it’. The precise meaning of SIP is still somewhat variable, and can be taken to indicate either steaming in place, or sterilizing in place. It should be noted that steaming in place will usually result in a fully sterile system (provided that the system has been appropriately designed for steaming); however there are alternative methods of sterilization-in-place (such as irradiation) which do not utilize steam.
14.4 Possible Contaminants Some of the most common contaminants within a system are considered below. Although the list is by no means exhaustive, it will provide an indication of the wide range of potential contaminants present within a bioprocessing environment.
14.4.1 Active Ingredients By definition, the purification of pharmaceuticals and biologics involves the handling of biologically active materials. Within a given sample there will often be more than one biologically active ingredient. Since only one ‘product’ is usually being purified, any other active ingredient can be thought of as a contaminant. If any such biologically active materials remain within the system they can be carried over into subsequent batches and cause cross-contamination. The extremely high potency of many current and potential therapeutics (such as interferons and peptide hormones) provides an indication of the dramatic effect even trace amounts of biologically active contaminant may have on the safety of the final product. As analytical techniques improve so the limits of detection are increasingly pushed back, placing an ever-increasing onus on the manufacturer to provide purer and purer end products.
420
14 Process Hygiene in Production Chromatography and Bioseparation
14.4.2 Bacteria (Vegetative) Bacterial infection can be divided into two subgroups, external (adventitious) infections, caused by organisms outside the production process and internal infection by organisms introduces by the production process itself.
14.4.2.1 External Infection The presence of bacteria in the environment acts as a constant source of potential contamination. All healthy individuals have a microbial flora associated with their skin and mucous membranes. Often, chromatographic operations are not carried out in extreme clean room conditions or in containment areas (although this type of operation is increasingly becoming the norm), and infection from organisms carried by the operators must be avoided. Common commensal organisms include the pathogen Staphylococcus aureus and various Pseudomads such as Pseudomonas aeruginosa. These organisms can be introduced into the system during the column packing stage or via buffer tanks.
14.4.2.2 Internal Infection Bacteria are used extensively in the production of recombinant proteins, the proteins being recovered from bacterial growth broths. Usually there are pre-chromatographic processing steps to harvest and lyse the cells, for intracellular products, or to separate the cells from the growth medium, for extracelluar excretion products. In either case there remains the potential for some viable cells to reach the chromatography stage. Although the gross presence of cells would inevitably result in the gel bed becoming blocked and manifest as an unacceptable increase in back-pressure, the presence of a few bacterial cells may not cause any obvious problems with the column, but would clearly be enough to infect the system. A further complication is caused by the nature of the sample stream from recombinant products during initial capture steps. In this case, the feedstock is, by definition, a bacterial growth medium and is therefore formulated to encourage the proliferation of any organisms, even if they are in the form of an unwanted infection within the system. Expanded bed chromatography is a relatively new chromatographic approach to the capture of product from crude growth culture. This method allows the passage of entire growth cultures to be passed through a chromatography resin [3]. Although offering benefits from reduced processing steps, expanded bed chromatography is clearly highly susceptible to contamination by bacteria from the growth medium. Thus, careful consideration must be given to the cleaning and decontamination of expanded bed columns after exposure to crude cell suspensions. A further source of internal infection can arise from the deliberate infection of the column with bacteria during cleaning validation studies. This will be discussed in more detail below.
14.4 Possible Contaminants
421
14.4.3 Bacteria (Spores) Spore-forming bacteria pose a particular problem for system sanitation. The spores of bacteria are heavily dehydrated and effectively inert. The are therefore resistant to most methods of sanitation and sterilization. Once again, spores can be introduced accidentally from an external source (the physical nature of spores encourages airborne and aerosol distribution), an internal source (although the conditions for sporulation are normally avoided in the fermenter this is not always possible), or by deliberate infection of the system for validation studies. Clearance of spores forms one of the most rigorous challenges to a CIP protocol.
14.4.4 FungUAlgae There is a relatively small, but important, use of fungi in the production of recombinant proteins, and some fungi are used for the production of secondary metabolites as final end products. In general, fungi can be cleaned from a system using similar methods to those used for bacteria. There are, however, the added complications of spore formation and mycelial growth within a system. While Sacchoromyces species are only found as single cellular organisms, some other yeasts such as Cundida species are prone to dimorphism and can switch between single cell and pseudomycelial forms. If this occurs, then the dead organism will not be flushed out of the system after sanitization but will decompose in situ potentially giving rise to further problem with decomposition products. Thus, while their larger size make them easier to remove from the process stream before the chromatography step, fungi often prove to be a more formidable problem if infection of the system does occur. Algae can become a contaminant in poorly maintained process plants. These organisms can proliferate in poorly maintained buffer tanks under conditions of apparently no nutrition. They are also able to live in relatively high pH environments and can thus be particularly difficult to deal with using conventional NaOH solutions. As with fungi, the main principle is to avoid contamination in the first instance.
14.4.5 Viruses As discussed above, the production of blood products exposes the chromatographic system to viral contamination. The high pathogenicity of the viruses potentially present (hepatitis B E , HIV) [2] gives extra cause for rigorous cleaning. Chromatographic separation per se has been suggested as a virus reduction step [2,4], although this depends upon the nature of the chromatography matrix and separation protocol. It is worth bearing in mind that some degree of viral clearance will accompany most chromatography steps in a manufacturing process.
422
14 Process Hygiene in Production Chromatography and Bioseparation
Due to their parasitic nature, viruses are unable to replicate unless they are within a host cell. For this reason, the presence of a small number of virus particles within a system can be difficult to detect and enumerate. Fortunately, many of the commonly contaminating viruses are relatively fragile outside of the host cell and therefore easy to inactivate. The issue of indirect viral contamination by retention of host cells or microbial vectors should be considered primarily as an exercise in removing the host contaminant. As with bacteria, viruses may also be used deliberately to infect a column, although this is usually done in a smaller scale-down study due to the cost and hazards of exposing a full-scale production column to potentially pathogenic virus. Viral clearance studies are considered in more detail below.
14.4.6 Endotoxins (Pyrogens) Endotoxins (or pyrogens) are heat-stable lipopolysaccharides, associated with the outer membranes of Gram-negative bacteria. These molecules initiate a febrile reaction (fever) when injected. The degree of fever varies with different pyrogens and in different patients; in worst cases the reaction may be fatal. Pyrogens are in effect decomposition materials. Elimination of viable bacteria does not necessarily mean the elimination of pyrogens; indeed, the action of bactericides may even produce a novel source of pyrogens as the killed bacteria degrade. The complex nature of pyrogens means they will frequently interact with the chromatographic resin in some way [5] which may hinder their removal.
14.4.7 Cleaning Reagent The toxic nature of the cleaning reagents themselves is often overlooked as a potential contaminant within the system. It is essential that all traces of CIP solution are removed before the next batch of sample is processed.
14.4.8 Non-specifically Bound Protein Proteins will bind non-specifically to most surfaces. This is often an area of concern to biochemists, especially when they are dealing with very small amounts of protein. The data in Table 14-1 show that the level of non-specific protein binding is very small for all the materials commonly used in the manufacture of chromatography columns and systems (for example, in a 1000 mm diameter column with a bed height of 250 mm, 240 mg of protein will be bound non-specifically to the glass tube). Given that production systems are usually used in a preparative environment with a large loading of protein in the sample, there is no real problem with the loss of product
-
14.5 Design Considerations in System Hygiene
423
Table 14 -1. Nonspecific binding of bovine serum albumin to materials commonly used in the manufacture of bioseparations equipment [6]. Material
Protein (mg cm-2)
Borosilicate glass Polypropylene TPX (polymethyl pentane) Acetal Stainless steel Acrylic (perspex) Santoprene EPDM
30 27 9 30 < 1” 12 23 32
a Below
detectable limit.
due to protein binding. However, there is a potential risk of cross-batch contamination from the binding and displacement of proteins from surfaces within the system. A more serious problem can arise from non-specific binding of protein to the chromatography resin due to the intrinsically high surface area available. Thus, this is another area which needs to be addressed by the CIP process. Fortunately, CIP with NaOH solutions is very effective at removing non-specifically bound proteins. Washing the samples referred to in Table 14-1 with 0.1 M NaOH reduced the non-specifically bound protein to less than the level of detection (< 1 mg cm-*) in all cases.
14.5 Design Considerations in System Hygiene 14.5.1 General Materials of Construction The same materials are utilized almost universally by manufacturers in the construction of columns. The primary consideration in selecting the materials of construction for any system to be used in biotechnology or biopharmaceutical production must be the biocompatibility of the material [ 7 ] . However, in order to carry out efficient cleaning, the materials of construction should also be selected to withstand the chemicals used for the process CIP procedures. Since these chemicals are almost always more extreme than those used for the routine separation, the CIP cocktails will often have a strong influence on the materials of construction. In general, the surface finish is important for all materials; the rougher the surface the greater the area potentially available for contaminants to bind. In extreme cases poor surface finish can provide crevices and ‘caves’ for contaminants to accumulate. For example, it is important that there are no ‘bubbles’ within molded component which may be partially exposed during subsequent manufacturing steps. The relative benefits and shortcomings of the most frequently used materials are discussed below.
424
14 Process Hygiene in Production Chromatography and Bioseparation
14.5.2 Stainless Steel A variety of grades of stainless steel are available. The grades of stainless steel differ in their content of additive metals (Table 14-2). Although Grade 306 is commonly used for structural elements, such as frames and enclosures, the grade of stainless steel usually used for wetted surfaces is 316L. This has a good compromise between increased corrosion resistance and cost. However, it is important to note that 316L stainless steel is not entirely resistant to salt corrosion. The resistance of stainless steel to corrosion can be significantly decreased by welding, unless appropriate precautions are taken to blanket the metal in an inert gas during welding. The surface finish of stainless steel is critical to its cleanliness [8]. Finishes are usually quoted as having a certain RA or Grit value (Table 14-3). Typical finishes are illustrated in Fig. 14-1. It can be seen that the smoothest finish is obtained by electropolishing. This entails the electrolytic removal of the surface layer of steel. For the process to be fully effective it requires careful optimization of the polishing solution (electrolyte) used, the current applied and most importantly the design of the electrode ‘jig’ since the electrode should be positioned near the surface of the component to be electropolished. For small-diameter and complex pipe spools this is often extremely difficult. The corrosion resistance of stainless steel is due to the presence of an extremely thin oxide layer on its surface. This passive oxide film can only form on clean, uncontaminated surfaces. A typical final treatment for non-electropolished stainless Table 14-2. Chemical composition (%) of commonly used grades of stainless steel. Grade
304 304L 316 316L 316Ti 317L 321
0.08 0.03 0.08 0.03 0.08 0.03 0.08
2.0 2.0 2.0 2.0 2.0 2.0 2.0
Cr
Ni
18-20 18-20 16-18 16-18 16-18 18-20 17-19
8-10.5 8-12 10-14 10-14 10-14 11-15 9-12
Ti
Mo -
-
2-3 2-3
0.4-0.7
-
-
3-4
0.4-0.7
Table 14-3. Grit and RA (roughness average) equivalents for stainless steel surface finishes. Grit
Nominal RA (mm)
120 180 240 320 400 Bright polish
0.8 0.4 0.3 0.23 0.15 0.08
14.5 Design Considerations in System Hygiene
- 5
425
Typlcal Size Bacterium
-;i $-J+?& 1
&
- 2
- 3 - 4
Vacublast forging
Mirror polish Forging
steel is passivation or pickling. This entails the exposure of the clean (degreased) stainless steel components to a strongly oxidizing solution which will promote the formation of the protective oxide [9,10]. A typical passivation cocktail would be a solution of 20 % (v/v) nitric acid and 5 % (w/v) potassium dichromate. Electropolishing results in a surface which already has a protective oxide layer, therefore a further chemical passivation step is not required. However, for nonelectropolished components, passivation will increase corrosion resistance and is therefore recommended. Among the few disadvantages of stainless steel are its weight, making large columns more difficult to handle, and its susceptibility to corrosion by chloride solutions. However, perhaps the greatest disadvantage of stainless steel for the production of chromatography column tubes is the lack of visibility. The ability to observe the gel bed can provide great benefits; void formation, due to bed settling is immediately apparent, as is poor flow distribution if the feedstock is colored.
14.5.3 Borosilicate Glass Glass is commonly used in Iaboratory-scale columns. Its advantages are that it is transparent and bio-inert, but its disadvantages are its brittleness and low mechanical strength. While these factors are less important in small columns (where glass can be used successfully within systems operating at relatively high pressures), in production-scale columns the weakness of glass can become a serious disadvantage. Additionally, glass is less suited to a production floor environment where it is more prone to accidental damage. Small scratches or chips in glass tubes will seriously reduce the maximum pressure that the tube is capable of withstanding, and care must be taken when dismantling, cleaning, and assembling glass columns.
426
14 Process Hygiene in Production Chromatogruphy and Bioseparation
14.5.4 Polypropylene Polypropylene is a cost-effective and easy-to-machine polymer. It is strong and biocompatible; however, it is not transparent and is only translucent in thin sections. Polypropylene is commonly used to manufacture structural elements within a column or system (end cells, flow cells, etc.) and also for the manufacture of pipework, as it is relatively easy to bend and shape.
14.5.5 Acrylic (Plexiglass) Acrylic offers the transparency of glass, but with a higher strength. However, it is expensive in large precision-formed sheets or tubes, and has a poor chemical compatibility. The chemical resistance of acrylic to alcohols can often be a problem as ethanolic solutions are used both as part of a CIP procedure and also as a bactericidal storage solution. The relative resistance of acrylic to attack by alcohols depends mainly upon the ethanol concentration, temperature, exposure time, and the manufacturing process of the acrylic tube. In general, column tubes made from cast acrylic show better solvent resistance than tubes formed from annealed rolled sheets.
14.5.6 TPX (Polymethyl Pentane, PMP) Although TPX is more commonly found in laboratories in the form of beakers and measuring cylinders, the use of this material in chromatography columns has increased significantly in recent years. TPX has good transparency, strength, and chemical resistance, and is biocompatible. Older columns made from TPX had a tendency to turn yellow or cloudy over a period of time as a result of the action of incident UV light. However, modern grades of TPX are now more resistant to UV ageing. The main drawback to TPX is the difficulty in manufacturing large diameter or long column tubes using current moulding techniques.
14.5.7 PVC (Tygon) Flexible PVC hose is used extensively in food and biopharmaceutical applications for the interconnection of components, especially in coupling columns to chromatographs. PVC hose is available with preformed sanitary (Tri-clover) fittings for extra cleanliness. The flexibility of the hose allows some adjustment of the column without recourse to replumbing the supply pipework. However, PVC is not resistant to attack by many organic solvents commonly used in reversed phase chromatography, such as acetonitrile, and may also be weakened by prolonged exposure to high concentrations (> 20 %) of ethanol.
14.6 Elastomeric Materials (Seals)
427
14.5.8 Fluoropolymers A variety of fluoropolymers are available and are commonly used in pipe work and valve body construction. Although expensive fluoropolymers have outstanding chemical resistance. Careful design and smooth surfaces also help to ensure that fluoropolymer components show very low non-specific binding and are easily cleaned making them suitable for sanitary systems.
14.6 Elastomeric Materials (Seals) Elastomeric materials are required to seal the columns and pipework. Clearly, this is important both from the point of view of preventing leakage of materials or solutions into the workspace, and to prevent ingress of contaminants into the flow path. Earliest columns and systems commonly used natural rubber seals or their man-made equivalents such as styrenebutadiene (SBR, Buna). Advances in elastomer engineering coupled with more stringent regulatory guidelines have greatly reduced the use of such compounds. The compounds outlined below represent the materials most commonly used for sealing columns. Each of the polymers listed below has its own benefits and disadvantages. In general, however, it is important to ensure that all sealing materials are biocompatible and nonleaching. Biocompatibility can be determined by implantation (US Pharmacopeia IV) tests [ I l l and cytotoxicity testing. The USFDA produces a set of guidelines [12] for the composition of elastomeric materials, but it should be borne in mind that these guidelines are in the form of a list of permissible chemical components and the maximum permissible concentrations of those components. Within these guidelines it is possible to have a broad range of ostensibly identical elastomers with widely differing properties.
14.6.1 EPDM (Ethylene Polypropylene) This is perhaps the most common elastomeric material. It is manufactured from polymerized ethylene propylene with the addition of a controlled amount of plasticizers, stabilizers, and bulking agent (usually carbon black). It is widely available and can be molded (for small components and ‘0’ rings) or extruded and joined (for large ‘0’ rings and tubular seals). It is important to be aware that the label EPDM represents a general formulation. Within that specification there is a wide variety of EPDM elastomers available with greatly differing chemical resistance and biocompatibility. It is best to test a sample of the specific elastomer to be used with a variety of process fluids to check for changes in size (usually swelling), hardness, and leachates.
428
14 Process Hygiene in Production Chromatography and Bioseparation
14.6.2 Santoprene (Norprene, Marprene) Santoprene is similar to EPDM in that it is also based on ethylene polypropylene. However, the bulking agent used in the manufacture of Santoprene is polypropylene rather than carbon black. Santoprene has good chemical resistance and is widely used for peristaltic tubing. Once again, there is a variety of grades and formulations available, and care must be taken to ensure that the grade being used will be acceptable to the regulatory authorities.
14.6.3 Silicone Silicone is commonly used in flexible hoses, although it is also available as fixed seals. Two varieties of silicone are available; peroxide-cured and the more expensive platinum-cured. Platinum-cured silicone has fewer potential leachables, a smoother surface, and lower levels of protein binding compared with peroxide-cured silicone. The chemical resistance of silicone can be a problem. If it is to have long-term exposure to concentrated acids or in systems stored for long terms in NaOH solutions chemical resistance tests should first, be carried out on samples of the polymer.
14.6.4 Elastomeric PTFE Elastomeric forms of PTFE are commercially available which have extremely high resistance to solvents and are totally bio-inert. Although used for small seals and components, the very high cost of these compounds makes them unsuitable for use in large columns unless there are no suitable alternatives. This is often the case in preparative HPLC columns where high solvent resistance is necessary.
14.7 Mechanical Construction 14.7.1 Connections and Seals Laboratory-scale chromatographs are interconnected with flanged or ferruled screwthread fittings. A similar principle is also applied to process-scale HPLC systems which are usually interconnected with Swagelok connections (or similar). Such systems are not constrained by the requirements for process hygiene and can therefore make use of screw-threaded fittings, which offer benefits in terms of maximum operating pressures, reliability, cost, and ease of use. Unfortunately, a screw thread is also
14.7 Mechanical Construction
429
impossible to clean in place. Metal-to-metal contact will allow a leakproof seal, but will not be smooth enough to eliminate small crevices which can harbor contamination. The most common form of sanitary fitting used is the Tri-clover (also knows as a triclamp or Ladish connector), although Kamlock connectors are also commonly used on large systems. These connectors rely upon an elastomeric seal between the two pieces of pipe or tubing to be connected. The seal is pressed on both faces to form a continuous surface across the joint. It is important to use the correct size seal, and also to be careful not to overtighten the clamp, which may make the seal deform into the flow path. Care should also be taken in selecting the seal material since some elastomers will swell in the presence of commonly used CIP solutions and this swelling will result in extrusion of the seal into the flow path. As a final note on the question of sanitary connections it has been the all too-frequent experience of the author to witness sanitary triclamp connections attached to flexible hose by means of hose-tail adapters. Whig a hose-tail adapter is undoubtedly the most adaptable and simple method of interconnection available, it is also perhaps the least sanitary, and the use of hose-tail or hose-barb adapters will immediately invalidate any effort on the part of the system designer to produce a sanitary chromatograph! It is also not unknown for some equipment manufacturers to simply mold, weld, or screw a triclamp connector onto an otherwise unchanged component and claim it to be sanitary.
14.7.2 Pipework SpoolsNalves In a small-scale laboratory system it is usual to interconnect components using flexible tubing. However, above the laboratory scale this is impractical and it is unavoidable that fixed pipework spools will be required to interconnect valves, instrumentation, buffer inlet feeds, and fraction collection outlets. For pilot-scale operation, plastic pipework can be used. In order to manufacture a plastic pipework system it is necessary to have separately molded sanitary ends fitted to the pre-shaped spools. Care must be taken when fixing these sanitary end pieces to the pipe, as a poor joint will act as a very effective deadspot and will be liable to accumulate dirt or bio-burden. For larger-scale production systems is it usually more cost-effective to manufacture the pipework spools from electropolished stainless steel. This greatly facilitates the construction of complex spools shapes with welded sanitary fittings (Fig. 14-2). In addition, the overall mechanical strength of the system is better able to cope with a production environment. If a system is designed to only use two-way valves this will add greatly to the complexity of the pipe Work spools, since manifolds have to be fabricated for diversion of flow (for example, to by-pass a column). Such manifolds will contain T junctions which, by definition, form unswept areas. In general, the construction guidelines applied in these cases is that the length of the ‘T’ should be no more than six pipe diameters, although it is anticipated that this will be reduced to three pipe diameters. Clearly, a length of three pipe diameters is significant for larger diameter
430
14 Process Hygiene in Production Chromatography and Bioseparation
Fig. 14 -2. Hygienic design of production-scale chromatographs should take into account the layout and construction of the pipework and valves. Careful consideration at the design stage is essential to the final validation of the system. (Photograph courtesy of Millipore (UK) Ltd.)
pipework and thus this design criterion can be accommodated for high flowrate equipment with wider-bore pipework. However, there is an increasing move towards smaller sanitary systems as very small volume (and very high value) biotechnology products are moving into production. Given that a low flowrate system of say 400 ml min-' would typically have a pipework diameter of 3 mm (114 inch), the three pipediameter criterion would mean that insufficient pipe may be available to be able to orbitally weld a diaphragm valve in place. Until recently only two-way valves, such as diaphragm valves or pinch valves, were recognized as being hygienic. Older designs for three-way valves would inevitably incorporate dead spaces and hold-up volumes; neither were such valves easily available with sanitary connections. In recent years several valve manufacturers, often working closely with suppliers of bioseparation equipment, have introduced a variety of sanitary design three-way valves. These will usually have been extensively tested by a variety of clearance studies.
14.7 Mechanical Construction
43 1
There are several major benefits in using three-way valves, including the elimination of ‘T’ joints within pipework spools and a decrease in the number of valves required (thus decreasing cost and increasing system reliability). The initial reluctance to use such valves within a sanitary process is now being overcome as the benefits of three-way valves become apparent. The latest concepts in sanitary valve design incorporate modular multiport valves within a single body. This gives even further potential for a reduction in dead legs and pipework volume. However, careful design is critical to ensure cleanliness in operation.
14.7.3 Column Seal Designs It is vital within a chromatography column to ensure that all areas in contact with the process fluid are sanitary in nature. The selection of suitable materials is a concern which has been discussed earlier. Although the actual materials utilized are ubiquitous, in the area of seal design there are almost as many variants as there are column manufacturers. The one underlying principle is that the seal should not produce any crevices, or dead spaces. The most basic column seal design is the ‘0’ring. This has the advantage of low cost and easy replacement. If standard sizes are used, replacements can be sought from a variety of sources. However, this now places the burden of material validation onto the purchaser and may invalidate the warranty of the column. The performance of an ‘0’ring as a seal is high. Such seals are reliable and not prone to overtightening. However, the nature of an ‘0’ring means that there has to be a locating groove into which the seal will fit; this then necessitates a certain degree of dead-space around the seal. This dead-space forms unswept areas which are only cleanable by passive diffusion. ‘0’ring seals also require some form of insert for a locating groove. In this region there will inevitably be poor flow distribution for both sample and CIP solution. The efficacy of CIP in the area around the seal can be improved by operating the column in both forward and reverse directions during the cleaning cycle. However, the ‘0’ring seal fails one of the first objectives of sanitary design, that no dead spaces be deliberately created.
14.7.3.1 Space-filling Seals In order to overcome the problem of dead spaces in the sealing area, manufacturers have now produced a variety of space-filling seal designs. These range from the most simple interference fit to complex profile-actuated seals.
Interference fit Although the most simple to use (and possible the most reliable), these seals are of limited use on all but the smallest columns. The fit of the seal must be tight enough to allow the seal to function at the highest specified operating pressure of the column. Because the seal has to make such a tight fit against the column tube it is
432
14 Process Hygiene in Production Chromatography and Bioseparation
often difficult to adjust the level of the distribution plates on column fitted with interference seals. In addition the seal, by definition, will be made as soon as the adjuster is inserted into the column tube. This can cause difficulties with entrapment of air underneath the top distributor plate and seal. Mechanically Actuated An alternative to simple interference fit seals are mechanically actuated space filling seals (Fig. 14-3). These rely on pressure from two actuating plates or rings to force the seal against the column tube. This mode is analogous to the mode of operation of conventional ‘0’ ring seals. However, unlike ‘0’ ring seals, the shape of the spacefilling seal has been carefully designed to occupy any dead spaces on actuation. These seals have the benefits of ‘0’ rings (easy to operate, allow rapid adjustment) with added cleanliness. There are some disadvantages to mechanically actuated space-filling seals. When an ‘0’ ring is actuated, the actuation force is transferred onto a very small contact area between the column tube and the seal; hence, a strong reliable seal can be made. In space-filling seals, some of the actuation energy is dissipated in the deformation of the seal to occupy the dead spaces. Hence, these seals require very careful design to avoid leakage due to under-actuation or poor distribution of the actuating force. SEAL ADJUST TUBE SEAL ACTUATION PLATE
ADJUSTABLE S E
COLUMN TUBE
Fig. 14-3. Actuated space-filling seals are an effective and reliable method of minimizing deadspaces in laboratory and pilot-scale columns (up to 250 mm diameter). However, in larger columns the mechanical force required to maintain the seal requires tightening mechanisms at multiple points around the seal and is less reliable. (Reproduced with permission from Millipore (UK) Ltd.)
14.7 Mechanical Construction
433
Dynamically Actuated On larger column diameters (> 450 mm), simple mechanical actuation of space-filling seals may not give enough compression to allow the seal to operate efficiently. Columns larger than this is usually have dynamically actuated seal mechanisms. These can be pneumatically (Fig. 14-4) or hydraulically operated. In both cases, the principle of action is the same, a hollow elastomeric tube is filled with air or liquid to inflate the seal against the column tube wall. In order to work correctly, the wall of the seal has to be carefully designed to allow selective deformation. The seals are manufactured from extruded lengths which can then be cut and annealed to form seals of the correct size. Pneumatic seals have the advantages of cleanliness of operation and are easy to use and control. However, the elastomers used will have some degree of air permeability and careful formulation is required to ensure that this is kept to a minimum, otherwise the seal can deflate over a period
SNAP RIM
/
MINIMUM DEAD SPACE INFLATABLE SEAL
Fig. 14-4. Inflatable seals allow reliable sealing of production-scale columns (up to 2000 mm diameter) while maintaining a minimal dead space between the distributor and the column tube. Seals can be inflated by gas (pneumatic) or liquid (hydraulic). (Reproduced with permission from Millipore (UK) Ltd.)
434
14 Process Hygiene in Production Chromatography and Bioseparation
of time. It is common practice to reduce the pressure in the seals before long-term storage (since the gas permeability is proportional to pressure). One less obvious advantage of pneumatic seals is that a breach in the seal integrity is usually immediately apparent and easily traceable by the gas bubbles ! Hydraulically actuated seals do not require a clean air supply for operation and are more stable over a long period. However, the liquid within the seal must be regarded as a potential contaminant of the column. Since the liquid within the seal is not subjected to routine CIP procedures it should be sterile and contain a bactericidal agent which should be compatible with the column in case there is a loss in seal integrity.
14.7.4 Distribution Plates The flow distribution system of a chromatography column end cell is designed to produce plug flow within the column. Thus, the distributor plays a crucial role in ensuring that the feed sample is applied evenly and equally across the face of the gel bed. This is also a requirement for the CIP solution. The design of the distributor system is critical in ensuring that the entire column is exposed to sufficient CIP solution for effective sanitization to occur.
14.7.4.1 Single Port Single port distributor plates or probably the most common design and have a general design (Fig. 14-5) which is applicable over the entire spectrum of column sizes. Single port distributor plates have an advantage in that they are less susceptible to the presence of air. The conical shape in a typical top end cell can accommodate a relatively large amount of air with only a small loss in chromatographic performance. However, the presence of entrapped air in either top or bottom distributor plates will have a very deleterious effect on a CIP step, since the air bubble will shield part of the distributor surface from the CIP solution. 14.7.4.2 Multi-Port Multi-port inlets can display a further problem since it is possible for a gradual accumulation of air to form an air-lock in one of the distributor ports. Although this will usually show as a gross reduction in chromatographic performance, it is also a potentially serious problem for CIP steps since a significant proportion of the gel will be underexposed to the CIP regime.
14.7 Mechanical Construction MINIMUM DEAD SPACE INFLATABLE SEAL
\
SUPPORT RIBS
COLUMN TUBE
I
I
\
435
INLET
BED SUPPORT
SINTER FIXING
SCREW
,ANTI-JETTING
DEVICE
/ FLOW DISTRIBUTION CHANNELS
Fig. 14-5. The design of the distributor plate is of critical importance in maintaining scalability from pilot to production scale. The distributor plate design shown has been used successfully on columns ranging from 10 mm up to 2000 mm in diameter with excellent scalability. (Reproduced with permission from Millipore (UK) Ltd.)
14.7.5 Expanded Bed Columns A recent development in process chromatography has been the use of expanded bed resins. The distributor design for expanded bed columns is critical to the performance of the column and the nature of the technique dictates that the distributor consists of a finite and relatively low number of ‘holes’ within a plate. The use of this method is also aimed specifically at capture of product from crude solutions often
436
14 Process Hygiene in Production Chromatography and Bioseparation
containing whole (viable) cells. These two factors clearly combine to form an additional hurdle in CIP procedures. Thus, the CIP of the distributor system needs to be very specifically addressed (and validated) whenever expanded bed chromatography columns are used.
14.7.6 Closed System Columns Recently, column manufacturers have developed process-scale columns which have been designed to allow packing and unpacking to be carried out without removing the upper distribution plate (Fig. 14-6). These closed system columns offer potential advantages in maintaining good hygiene, since the resin is less exposed to potential
Fig. 14-6. (a) Closed system columns allow the resin to be packed and unpacked in situ, reducing the probability of external contamination. Such columns work best with slightly compressible gels and optimization of packing procedures is usually required.
14.7 Mechanical Construction
437
Fig. 14-6. (b) Closed system columns require a separate pumping skid for the gel slurry which must also be designed and built to hygienic standards. (Photographs courtesy of Millipore (UK) Ltd.)
infection. However, it should be noted that such systems require additional pipework and holding tanks which will themselves add to the validation workload for the overall process.
14.7.7 Bubble Traps Bubble traps are usually fitted either to chromatography systems or to specific columns. The functions of bubble traps are twofold: first (and most important) to protect the resin in the column from air in the process stream; and second, to reduce flow pulsation within the chromatography system and thus avoid pressure shocks to the column.
43 8
14 Process Hygiene in Production Chromatography and Bioseparation
It is particularly important to protect both the end cell (distributor) and the resin from air. Air in the distribution system may cause problems with CIP as noted above, air in the gel bed will usually require a re-pack. In larger systems, degassing buffers becomes problematic and expensive. Most air will be introduced into the system as a continuous stream of small bubbles within the process line rather than as a single, gross bolus of air (due perhaps to a buffer or feed tank emptying). Thus, a bubble trap is essential and will in turn (if it is performing properly) accumulate air. The presence of air within the bubble trap makes it both unclean, and also uncleanable, unless specific steps are taken to purge the bubble trap fully with CIP solution to remove all air from inside and ensure complete contact on all surfaces. Once the bubble trap has been through a CIP cycle it must be re-charged with sterile air. In a clean room environment it is usually sufficient to re-charge the bubble trap by using air from the ambient environment; however, for total cleanliness the air used for recharging should be sterilized (usually by filtration).
14.8 CIP Validation The most time-consuming part of developing a CIP protocol is inevitably in the validation of the protocol. The well-known and often-quoted definition of validation given by the FDA is: ‘establishing documented evidence which provides a high decree of assurance that a specific process will consistently produce a product meeting its predetermined specifications and quality attributes’ [ 111. In the case of a CIP protocol the ‘product’ is a chromatography column or system which has a bio-burden reduced to (or below) the ‘pre-determined specifications’. Clearly, this puts the onus on the operator to define the acceptable levels of cleanliness and also to prove that these levels are consistently and reliably achieved. Much has been written about the validation of separation processes (for recent reviews see [13,14]. However, a vital part of the overall process qualification is the resin regeneration and CIP process. In this area much of the original data are generated by in-house research and are regarded by the companies involved as proprietary, and therefore less likely to be found in the public domain. However, some useful ‘discussion documents’ have been published in this field [15-171 and some research work published [18]. The CIP part of the process validation may seem to be unduly arduous and protracted; however, it should always be borne in mind that, once in production, product is more likely to be lost or rejected through the presence of unacceptable levels of contaminants than through the failure of the separation chemistry. It is therefore essential that the CIP routine be considered at earliest possible stage in process development and the same rigorous approach used on purification development and scale-up be applied to the CIP cycle [ 191.
14.8 CIP Validation
439
14.8.1 Challenge Testing One method available to test if a CIP step is effective is to imitate a worst possible case and then carry out the CIP procedure and prove (or otherwise) that it is effective [20]. This is the principle behind challenge testing. Challenge testing can be carried out using either a pure culture of a specified test organism or compound (Table 14-4), or using a sample of normal feedstock which has been ‘spiked’ to produce an abnormally high load of a specific contaminant. The column and system are exposed (infected) with the contaminant and a standard CIP protocol carried out. On-stream samples can be taken during the process to evaluate the progress of each CIP step. At the end of the CIP process the efficacy of the sanitization can be assessed by disassembling the system and taking swabs to determine the presence of contaminant, by flushing the system and column with sterile saline solution and collecting a volume for filtration and analysis, or by filling the system with sterile growth media and incubating for a given period before flushing and sampling. The latter is the most rigorous method and will detect the presence of viable organisms to lower levels than the previous two methods. Although it is most indicative of CIP performance if the challenge tests are carried out using the actual organism or compound most likely to be problematic within the process, this is sometimes not desirable on health and safety grounds. In these cases, standard organisms or compounds are used. The organisms chosen should bear as close a similarity to the most likely contaminant as possible. Although vegetative bacteria can be used for challenge testing, large column viral challenges are most often carried out on scaled-down columns. The aim of these studies is to validate the chromatography as a virus removal step rather than to validate the cleanliness of the specific production equipment. Some examples of ‘standard’ organisms used in challenge tests are given in Table 14-4.
ss RNA ds DNA
Parvoviridae
Togaviridae
Papovaviridiae
Minute Virus of Mice (MVM)
Sindbis virus
Simian virus 40
ads, double-stranded; ss, single-stranded.
ds DNA
Herpesviridae
Pseudorabies virus (PRV)
ds DNA
ss RNA
ss DNA
ss DNA
Herpesviridae
Parvoviridae
Herpessimplex virus
Porcine parvovirus (PPV)
ss RNA
ss RNA
ds RNA
ss RNA
Rhabdoviridae
Picornaviridae
Poliovirus
ds DNA
Vesicular stomatitis virus
Mold
Aspergillus niger
ds DNA
ds DNA
Retroviridae
Yeast
Candida albicans
ds DNA
Human immunovirus (HIV)
Gram +ve bacteria
Reoviridae
Gram -ve bacteria
Pseudomonas aeruginosa
Bacillus subtilis
ds DNA
Retroviridae
Gram -ve bacteria
Escherichia coli
ds DNA
Murine leukemia virus (MuLV)
Gram -ve bacteria
Staphylococcus aureus
Genome
Reovirus
Family
Organism
1000
-
No
Yes
No
Yes
No
Yes
Yes
Yes
Yes
No
No
-
-
-
45-55
45-75
18-26
150-200
18-26
180-200
80-90
80-100
80-110
60-80
25-30
-4000 -
-1000
- 1000 - 800
-
Size (nm)
Enveloped
Table 14-4. Typical organisms used in challenge test studies for the validation of CIP and virus removal.
Icosahedral
Spherical
Icosaheral
Spherical
Icosahedral
Spherical
Bullet
Spherical
Spherical
Spherical
Icosahedral
Mycelial
Rod
Rod
Rod
Rod
Spherical
Shape
High
Low
High
Medium
High
Low
Low
Low
Low
High
Medium
High (spores)
Medium
High (spores)
Medium
Low
Medium
Resistance
14.8 CIP Validation
441
14.8.2 CIP Conditions While a variety of CIP solutions are commonly used (see below) some factors need to be taken into account for all CIP processes. The efficacy of a CIP step will depend upon the temperature and exposure time of the system to the CIP cocktail. It is imperative that these factors be validated and written into the CIP protocol. It is an unfortunate fact that decreasing the exposure time or decreasing the concentration of CIP chemical will both lead to a reduction in operating costs. However, if these are carried out without due caution the resulting increase in column contamination will more than wipe out any potential savings. It is the experience of this author that one process ran into problems when the exposure time and the concentration of CIP solution (NaOH) were reduced independently by two separate groups within a plant after process scale-up. While either change alone may have still been sufficient to CIP the columns, the combination of a reduction in both parameters was catastrophic to the process.
14.8.3 CIP Cocktails While the precise composition of CIP cocktails will vary from application to application, in general there are only a limited set of solutions in common use. Once again, it should be emphasized that a CIP cocktail should be evaluated for the specific process as some of the following solutions will inevitably be incompatible with some processes.
14.8.3.1 Concentrated Salts Although not bactericidal, concentrated salt is often used as a first step in the regeneration of media. High concentrations (-1.5 M) of salts such as NaCI or KCl can be very effective at removing proteins still bound to ion-exchange gels or proteins bound nonspecifically to gel permeation media. High concentrations (-4 M) of chaotropic agents such as urea or guanidinium hydrochloride are also effective in removing strongly bound proteins and precipitated proteins. Such chaotropic solutions may also be the only available option for some protein-based affinity gels where stronger CIP protocols would damage the ligand. 14.8.3.2 Sodium Hydroxide (NaOH) Sodium hydroxide is the most common component of process-scale CIP solution, and is used either individually or in combination. The concentration of NaOH used will vary (from 0.1 to 1.0 M) with the degree and type of contamination and the chemical strength of the gel being used. The exposure times will likewise vary
442
14 Process Hygiene in Production Chromatography and Bioseparation
(from less than 1 h to several days). A 0.5 M NaOH solution with a minimum exposure time of at least 60 minutes is usually sufficient for most processes. In general, the lowest concentration of NaOH possible should be used to prolong gel life. Affinity gels in particular are prone to damage by excessive exposure to strong NaOH solutions. For affinity gels it is common to calculate CIP solutions in terms of column volumes rather than as exposure times in order to avoid prolonged exposure to alkaline conditions. It should also be noted that some affinity linkages such as divinylsulfone (DVS)-activated gels are not resistant to alkaline conditions. For gels such as these, alternative regeneration/CIP procedures such as acid washes (see below) must be used. 14.8.3.3 Organic Solvents Organic solvents are used in CIP procedures to assist removal of lipids binding to gels. This type of contamination is especially problematic with columns used in the early stages of a purification. The most common solvent used is ethanol. The addition of acetic acid (-5 %) to the ethanol solution will assist in solubilizing lipid. Longer-chain alcohols can be used in place of ethanol to increase the solubilization of lipid. Longer-chain alcohols also have the added advantage of being more toxic towards micro-organisms (probably due to their greater efficacy in disrupting lipid membranes). Longer-chain alcohols do have a higher viscosity than ethanol, which may cause settling of a packed bed; they can also be more difficult to remove and can, in extreme cases, cause flocculation of the gel. The inclusion of ethanol in CIP cocktails is commonly intended to act as a sterilization step. However, for efficient sterilization an ethanol concentration of at least 70 % is required. This is much higher than commonly used and, in most cases will raise more problems with chemical resistance and explosion proofing than are justified. Various combinations of ethanol (up to 60 %) and acetic acid (0.5 M) have also been reported as efficient sanitization solutions [2 1.221. At concentrations below 70%, ethanol has little or no sporicidal effect [23] and reduced virucidal effect (a further consideration to note is that ethanol alone will not destroy pyrogens). However, ethanol is a good bacteriostat; therefore ethanol solutions of 10-20 % can be effectively used for the preservation of columns during long-term storage. Ethylene glycol can be used as a chaotropic agent to clean strongly bound species from reversed phase, hydrophobic interaction, or thiophilic gels. 14.8.3.4 Thiomersal (Thimerosal, Merthiolate, Ethyl-mercurithiosalicylate) Thiomersal is a mercuric compound widely used in the preservation of contact lenses. It has also been used widely in the past as a column sanitization and storage reagent, and as a preservative for vaccines. The concentrations used are small (0.005-0.01 %); however, the mercuric nature of the compound gives rise to the possibility of accumulation in the environment and in exposed organisms. It is vital to ensure that all thiomersal is removed from the column before the feed stock is
14.8 CIP Validation
443
applied to avoid cross-contamination using a suitably sensitive assay [24]. Further disadvantages of thiomersal are that it can bind nonspecifically to some resins, its use incurs high disposal costs, and it has been reported as a causal agent of cellmediated immunity [25]. For these reasons the use of thiomersal is becoming less common.
14.8.3.5 Chlorhexidine (Chlorhexidine Digluconate, Hibitane, (1,6 -Di(4-chlorophenyl-diguanid0)hexane) Chlorhexidine is widely used for column sanitization. Chlorhexidine itself is almost insoluble in water and is usually used as a 0.5 % solution in 20 % ethanol [26]. Chlorhexidine digluconate cannot be isolated as a solid but is available as a 20% aqueous solution [27]. Chlorhexidine has good bactericidal properties against vegetative cells, but has no effect on bacterial spores. It is effective against some lipophilic viruses such as herpesvirus and HIV, but is not effective against noncoated viruses such as human Reovirus [28]. The extensive use and safety history of chlorhexidine over the past 30 years makes it the compound of choice for many bioprocesses.
14.8.3.6 Detergents/Solvents Detergent and solvents used in combination have been found to be very effective in virus inactivation while maintaining mild conditions. A typical virucide cocktail would consist of 0.3 % TNBP (tri-(n-buty1)phosphate) and 1 % Polysorbate (Tween 80) [29]. 14.8.3.7 Acidic Conditions Dilute HCI, low pH glycine buffer, citric acid, and acetic acid are often used in the regeneration of immunoaffinity columns due to the labile nature of these columns in alkaline conditions. Low pH cleaning steps also have the added bonus of virucidal activity [30]. Glycine, citrate, and acetate-based buffers are not recommended for prolonged storage of gels, however, as they form a suitable growth substrate for many micro-organisms. 14.8.3.8 Oxidizing Agents The sterilization of surfaces and some equipment by the use of strong oxidizing agents such as formalin, sodium hypochlorite, and peracetic acid is widespread within laboratories. Although these compounds are very effective in sterilizing hardware, very few chromatography gels are resistant to them. Some exceptions may be found among newer ‘synthetic polymer’ gels such as those manufactured in the form of methacrylate polymers (Toyopearl, Macroprep), polystyrene divinylbenzene poly-
444
14 Process Hygiene in Production Chromatography and Bioseparation
mers (Amberchrom), and polystyrenes (HyperD). However, it is strongly advised to seek confirmation from the manufacturers before using oxidizing agents on gels as the resin chemistry may not be compatible, even though the base matrix is. If the resin is compatible with oxidizing agents such as peracetic acid, this can form the basis of a highly effective CIP procedure [31,32].
14.8.4 Storage The long-term storage of chromatography systems and columns pose further problems. Frequently, the concentrations of chemicals used in CIP cocktails (see above) are too high for long-term storage and would result in damage to either the chromatography hardware (for example over a prolonged period of exposure, 20 % ethanol will attack the acrylic commonly used to manufacture column tubes), or alternatively to the resin itself (many affinity resins will undergo ligand uncoupling after prolonged exposure to NaOH). Thus, columns should be stored in solutions of lower concentration than those used for CIP. If the column has undergone an efficient CIP step before storage this will not be a problem, since the storage solution merely has to act as a bacteriostat rather than a bactericide. It is also good practice to flush columns in long-term storage at regular intervals with clean buffer and to renew the storage solution. If the column is stored in a solution different from that used for CIP, separate validation of the storage solution must be carried out. In this case, the emphasis should be on bacteriostasis and resin lifetime rather than bactericidal efficiency. Chromatography pumping systems are usually more resilient to chemical attack. However, storage solutions should also be of lower concentration than CIP cocktails. It is important to remember that even very dilute salt solutions will concentrate during evaporation and can give rise to corrosion, even on stainless steel, while deionized water is very effective at dissolving salts from stainless steels altering the surface chemistry and thus can also be corrosive. When storing systems (and empty columns) for a protracted period the systems should be rinsed in deionized water, drained down, and thoroughly dried. In such cases, pH probes should be removed from the process line and stored in KCI solutions.
14.9 Conclusion Cleaning is a burden, but unfortunately it is also unavoidable. Effort placed in CIP development and validation may seem unnecessary at the development scale; however, care and consideration at the earliest point in process development will reap great rewards at the production level by increasing process reliability and lifetime whilst decreasing lost production time, lost money and, of course, lost sleep !
References
445
References [1] Jungbauer, A., Boschetti. E., Manufacture of recombinant proteins with safe and validated chromatographic sorbents. J Chromatogr B, 1994, 662, 143-179. [2] Sofer, G., Preparative chromatographic separations in parmaceutical, diagnostic, and biotechnology industries: current and future trends. J Chromatogr A, 1995, 707, 23-28. [3] Burnouf, T., Chromatography in plasma fractionation: benefits and future trends. J Chromatogr B, 1995, 664, 3-15. [4] Darling. A. J., Spalto J. J., Process validation for virus removal. Biopharm, 1996, 9, 42-50. [5] Anspach, F.B.. Hilbeck. O., Removal of endotoxins by affinity sorbents. J Chromatogr A, 1995, 711, 1-92. [6] Kemp. G. D., Applications group internal report LR 02029. Amicon Ltd, Stonehouse, Y.K., 1995. [7] Flavell, P., Selecting the correct Medical-grade polymer. Medical Device Technology, 1996, November, 16-22. [8] Pfister, M., Kohlzr, W. G., Fittings and components for aseptic processes in the chemical and pharmaceutical industry. Chemical Plants and Processing, 1993, March, 24-26. [9] White, P. E., Stainless steel for food and beverage processing-advantages in hygiene and cleanability. Stainless Steel Industry, 1988, September, 2-4. [ 101 Roessell, T., Swain. J., Effective chemical cleaning of stainless steel fabrications. Stainless Steel Industry, 1990, Vol. 18, no. 102, 1-3. [ I l l Center for Drugs and Biologics and Center for Devices and Radiological Health, guidelines on general principles of process validation; Food and Drug Administration, May, 1987. [12] US Department of Health and Human services. Code of Federal Regulations, Title 21, part 177; Food and Drug Administration, . [13] Eckman, B., Validation, in: Handbook of Downstream processing; Goldberg, E. (Ed.), Blackie Academic Professional, 1996. [ 141 Sofer, G., Hagel, L., Process chromatography, London: Academic Press, 1997. [15] Agalloco, J., 'Points to consider' in the validation of equipment cleaning procedures. J Parenteral Sci Technol, 1992, 46, 81-86. [ 161 Parenteral Drug Asociation Industry perspective on the validation of column-based separation processes for the purification of proteins. J Parenteral Sci Technol, 1992, 46, 87-97. [17] Zeller, A. O., Cleaning validation and residue limits: a contribution to current discussions. Pharmaceutical Technology Europe, 1993, November, 18-27. [18] Boschetti, E., Pouradier Duteil, X., Nguyen, C., Moroux, Y., Concerns and solutions for a proper decontamination of chromatographic packings. Chimicaoggi, 1993, MarcWApril, 29-35. [19] Edwards, J., Large-scale -column chromatography - a GMP manufacturing perspective, in: Handbook of Downstream processing; Goldberg, E. (Ed.), Blackie Academic Professional, 1996. [20] Berube, R., Oxborrow, G. S., methods of testing sanitizers and bacteriostatic substances, in: Disinfection sterilization and preservation, 4th edition; Block, S. S. (Ed.), Philadelphia: Lea Febiger, 1991, pp. 1058-1068. [21] Boschetti, E., Girot, P., Guenier, L., Silica-dextran sorbent composites and their cleaning in place. J Chromatogz 1990, 523, 35-42. [22] Girot, P., Moroux, Y., Pouradier Duteil, X., Nguyen, C., Boschetti, E., Composite affinity sorbents and their cleaning in place. J Chromatog6 1990, 510, 213-223. [23] Larson, E. L., Morton, H. E., Alcohols, in: Disinfection sterilization and preservation, 4th edition; Block, S. S. (Ed.), Philadelphia: Lea Febiger, 1991, pp. 191-203. [24] Shrivastaw K. P., Singly S., A new method for spectrophotometric determination of thiomersal in biologicals. Biologicals, 1995, 23, 65-69.
446
14 Process Hygiene in Production Chromatography and Bioseparation
[25] Seal, D., Ficker. L., Wright, P., Andrews, V., The case against thiomersal. The lancet, 1991, 338, 315-316. [26] Adner, N., Sofer, G., Biotechnology product validation, part 111: chromatography cleaning validation. Pharm Tech Eul; 1994, April, 21-28. [27] Denton, G. W., Chlorhexidine, in: Disinfection sterilization and preservation, 4" edition; Block, S. S. (Ed.), Philadelphia, Lea Febiger, 1991, pp. 274-289. [28] Springthorpe, V. S., Grenier, J. L., Lloyd-Evans, N., Sattar, S. A,, Chemical disinfection of human rotaviruses: Efficacy of commercially available products in suspension tests. J Hygeine, 1986, 97, 139-161. [29] Horowitz, M. S., Bolmer, S. D., Horowitz, B. Elimination of disease-transmitting enveloped viruses from human blood plasma and mammalian cell culture products. Bioseparation, 1991, 1, 409-417. [30] Burstyn, D. G., Hageman, T. C. Startegies for viral removal and inactivation, in: Viral safety and evaluation of viral clearance from biopharmaceutical products; Brown, F., Lubiniecki, A. S. (Eds.), Developmental Biological Standards, Karger, 1996, vol. 88, pp. 73-79. [3 11 Jungbauer, A,, Lettner H., Chemical disinfection of chromatography resins, part I: preliminary studies and microbial kinetics. Biopharm, 1994, 7, 46-56. [32] Jungbauer, A., Lettner, H., Guerrier, L., Boschetti, E., Chemical disinfection of chromatography resins, part 11: in situ treatment of packed columns and long term stability of resins. Biopharm, 1994, 7, 37-42.
Bioseparation and Bioprocessing Edited b y . Ganapathy Subramanian Copyright BWILEY-VCH Verlag GmbH, 1998
15 Strategies and Considerations for Advanced Economy in Downstream Processing of Biopharmaceutical Proteins Joachim K. Walter
15.1 Introduction Downstream processing (DSP), that is the recovery and purification of biotechnically derived proteins, as well as proteins derived from fluids or tissues of biological origin, represents a sequential arrangement of individual unit operations which form a rather complex synthesis of chromatographic and filtration methodologies. Each single unit operation of DSP is designated to contribute selectively and significantly to the removal of impurities, including contaminating peptides, proteins, and lipids. Devoted care needs to be taken of drug safety regarding potential harmful infectious contaminants, hence the removal of DNA and (potential) virus or virus-like-particles (VLP) is crucial and requires respective efforts in validation.
15.2 Economic Potential in Downstream Processing The yield of product at each purification step strongly depends on the potency and resolution of the separation media chosen, as well as on the design of the operational procedures around that purification step. At an economic basis, typical step yields are in the range of 92-98 %; significant lower yields down to 80 % are accepted in the light of an unique purification effect of such an operation, typically being located at the initial phase of a downstream process. The number of purification steps required, and their individual step yields determine the overall yield. On this account the design of the downstream process contributes considerably to the economy of the manufacturing process. Currently the overall yield of typical purification processes composed of about 8-10 unit operations is in the range of 50-80%, a yield that - at its lower end can diminish minor process achievements in fermentation processes. This is one of the main driving forces to accomplish yield improvement during downstream processing. Since typical yields are already in the range of about 95 % or higher for an individual purification step, potential improvements on single-step yields are to be considered as marginal. Nevertheless, the use of appropriate technologies might contribute to reduce hands-on time and investment in hardware, e.g., the application
448
15 Strategies and Considerations for Advanced Economy ..............................................................................................................
100 90
....................................................................
80
.........................................
70
$
a
60
......................
50
.............................
40
.............................................
30
.....................................................
20
................................................................................
10
..............................................................................................
O ! 1
................. ........................
I
2
3
4
5
6
7
8
9
10
Number of Unit Operations /+Step
Yield 95 % +Step
YieM 90 % *Step
Yield 85 % *Step
Yield 80 %
1
Fig. 15-1. Influence of the number of unit operations and their step yields on the overall yield of bulk product.
of membrane-based chromatography, in the mode of negative chromatography, even being used as disposable in-line filters. Due to the large volumes of several thousand liters used in fermentation processes and to significant improvements in cell productivity resulting in product titers in mammalian cell culture of currently up to 1000 mg l-l, the development of downstream processes might be faced with severe technical or economic limitations. Despite the fact that a further increase of the dynamic loading capacity of the chromatography matrices for the product would solve certain difficulties, the steric hindrance in case of large protein molecules and the aggregation of proteins at high concentrations during elution due to limited solubility rarely allows the usage of protein loads significantly beyond 100 mg ml-* of matrix. The application of high linear flowrates in a chromatographic process is clearly another approach to shorten process time. However, adsorption kinetics of the protein product to the matrix and appropriate access of the product molecule to the inner surfaces of typical chromatographic beads remain as limiting factors. Highly expensive separation media are of impact on production economy at large scale. As a practical approach, the protein purification can result in a split of an individual fermentation batch. By preference, such a split is limited to those steps with limited capacity and implies re-combination of the resulting pools, a strategy that helps to tighten the time frame of manufacturing and avoid the generation of several batches of bulk product. A significant contribution to process amelioration should be expected by the optimization of those steps that show yields lower than 95 %. At times, an even more dramatic loss of product might proceed with the cell harvest process using microfiltration or related techniques. Much effort has been undertaken to reduce product loss
15.3 Strategic Development of Unit Operations
449
during the harvest procedure, i.e., the separation of the crude product solution from cells. Beside a continuous optimization of conventional cell harvest equipment, e.g., microfiltration and efforts to master clogging and membrane fouling respectively, or centrifugation and respective efforts to conquer the generation of cell debris but still guarantee equivalent cell removal, the merger of cell removal and product-specific adsorption to a chromatographic matrix - known as fluidized or expanded bed chromatography - is an attractive approach both to accomplish loss of product and reduce the number of steps [l]. The other option for process amelioration - a reduction in the number of steps - is even more conducive to a distinct increase in total yield as well as process economy. However, such a reduction is limited with respect to product safety by the need for a validation of removal factors for potential DNA and viral contaminants [2-51. The effectiveness of methods and technologies for the inactivation and removal of potentially harmful contaminants such as viral particles and cellular DNA is crucial with regard to the number of labor-intensive procedures which are necessary in order to obtain the required reduction factors and to satisfy validation needs [6].
15.3 Strategic Development of Unit Operations The development of a complex downstream process for biopharmaceuticals is oriented according to the application profile of the protein product: cGMP manufacturing which yields a highly pure product featuring drug safety and state-of-the-art process validation is devoted to therapylin vivo diagnosis. Thus, the extent of unit operations and measures to obtain product quality suitable for an in vitro diagnostic is significantly reduced [7]. Hence, prior to the start-up of any process development, premises must be formulated which are considered as guidelines for the duration of the development: technical basis, raw materials, process design, and process validation.
15.3.1 Technical Basis The fundamental technical pre-requisite for the development in order to achieve a maximum benefit and economy from the process is the identification and knowledge of the prospective manufacturing scale, plant, and equipment. Certainly only those unit operations which can be scaled to the intended scale are allowed for development, and with due regard to all technical contingencies and conditions the transfer to production scale and thence to the production plant is improved considerably. The design and construction of equipment and piping must allow for a simple, complete and easily validated sanitization as cleaning-in-place (CIP). In general, all process components must be planned from the very beginning with respect to the feasibility of validation measures.
450
15 Strategies and Considerations for Advanced Economy
15.3.2 Raw Materials and Equipment A focal point of the development is on the selection of raw materials and process media, and their impact on suitability and functionality at the intended technical scale. Each of the raw materials must be available in appropriate quality and quantity: salts for buffer preparation, solutions and solvents must be specified according to applicable pharmacopoiea, e.g., DAB (Deutsches Arzneimittel Buch) or USP (United States Pharmacopoiea). Process media such as chromatographic matrices and membranes for the different types and modes of filtration are selected according to their potency and commercial availability. (Table 15-1). Table 15-1. Selection criteria for media applied to DSP. 0
Suitability and functionality at technical scale separation performance, regeneration, life time
-
0
Reproducibility of all relevant process parameters - selectivity - resolution - yield - dynamic binding capacity - chromatographic profiles, filtration profiles - flowrates - system pressures - conductivity, pH-value
0
Commercial availability - batch size, batch consistency, deliverability
0
Authorities’ acceptance - Drug Master File (DMF) - GLP/GMP documentation
The influence of process media on process economy is fundamental: economically, the efficacy of the purification steps of the initial phase in DSP (‘recovery’) determines the total number of unit operations which are necessary to obtain the required product quality. The individual step yields contribute essentially to the total product yield, and contribute substantially to the overall economy of the downstream process. Accordingly, the selection for raw materials focuses on their commercial availability in a reproducible and reliable quality. Batch sizes might become crucial at larger scales, as the substantial analysis for release of the raw material including functional testing can be expensive. For the display of their resolution capability, excellent separation media require an equipment of optimal design and functionality. For large-scale operations, advanced chromatographic columns are constructed for an automated, fully controlled in situ filling. Although extremely expensive, such columns allow an excellent return on investment due to both their outstanding handling and reproducibility. Even columns with volumes of up to several hundred liters of matrix can easily be handled by a single person, as no heavy parts need to be moved. Table 15-2 shows data on the preparation of columns for gel permeation
15.3 Strategic Development of Unit Operations
45 1
Table 15 -2. Preparative gel permeation chromatography. Column size
Pressure Flowrate Slurry Theoretical Asymmetry HETP plate number factor N (m-') (AF) (cm) (mF'a) (cm h-I) (%)
Packing time (fin)
Diameter 180 mm Height 900 mm
0.44
34*
48
13680
1.43
0.0073
115
Diameter 180 mm Height 900 nun
0.45
44
45
13515
1.12
0.0074
125
Diameter 180 mm Height 900 mm
0.42
38
68
13930
1.14
0.0072
72
Diameter 600 mm Height 900 mm
0.40
32.5
62
16010
1.14
0.0063
69
Diameter 600 mm Height 900 mm
0.46
40
65
16807
1.14
0.0060
66
Scale-up of preparative gel permeation chromatography was performed using ChromaflowTM Columns (Euroflow, U.K.) and Superdex 200 pg (Pharmacia, Sweden). Total volume of the packed gel bed was 30 L for the 180-mm diameter and 255 L for the 600-mm diameter column. The packing direction was top-down except for the marked 180/900 (*) column, which was prepared bottom-up and resulted in a less favorable asymmetry factor of 1.43. This difference was reproducible. Otherwise, the columns allow the reliable in-situ preparation of columns at large technical scale even under varying conditions as shown in Table 15-2. The obtained resolutions are comparable with those which can be achieved with laboratory-scale columns for the identical matrix. The 600-mm diameter column could be optimized to N > 16000.
chromatography, which is a nearly inextricable task for conventional columns at the featured scale. Furthermore, the performance of resolution is an invaluable fundamental for a reliable scale-up. However, as with other large-scale equipment, the time consumption for design, specification, construction, and qualification is not to be underestimated, and will easily be in the range of 6-12 months. Due to such lead times, already results from early stages in development must be highly reliable, as they might urgently be requested for decision finding in an investment of great consequence.
15.3.3 Process Design DSP is an integral part of biopharmaceutical manufacturing. Although most operational procedures within the DSP might be dealt with as individual unit operations from a technical point of view, the impact and influence of the adjacent scientific and technological disciplines need to be considered (Fig. 15-2). The process design comprises the sequence of various unit operations, employment of personnel capacities and process validation.
452
15 Strategies and Considerations for Advanced Economy
m Cell Bank
V
---.
.-
,
lnocuium
Raw Materlal Tutin
Scale up Fermenter
Fernentation
Downstream Pmcesslng
&
I
Midstream Purification
Final Purification
Labelling Packaging
Waste Water Treatment
Final Produat
Fig. 15-2. Biotechnical manufacturing of pharmaceutical proteins.
The suitable combination of efficient separation technologies is one key function for the realization of utmost process economy. The sequential application of individual unit operations promotes the strategy of a modular process design. Along a general arrangement of the downstream process (Table 15-3), individual operations might be shifted up- and downstream within the process in order to achieve a maximum benefit regarding product quality and process economy. The skilful arrangement of different types of chromatography is an illustrious example: ionexchange chromatography typically applies a low-conductive buffer for loading in order to maximize the dynamic binding capacity, while elution is promoted by increasing the conductivity. With hydrophobic interaction chromatography it is exactly the reverse: binding is enhanced at a high conductivity while elution occurs at low conductivities - hence a promising succession of two powerful separation tools is feasible, excluding an intermediate ultrddiafiltration step as an additional operation. A graphic description for different options which can be found in a chromatographic separation is gel permeation chromatography (GPC) (Fig. 15-3). The product load determines the separation efficacy of a GPC column and, depending on the separation qualifications, the total load, i.e., the combination of product concentration and volumetric load, needs to be fixed in a narrow range. This range is crucial
15.3 Strategic Development of Unit Operations
453
Table 15-3. Downstream processing of biopharmaceutical proteins. DSP section
Unit Operations
Process volume
("/.I Cell Harvest
V Capture
V Midstream
V
Polishing
V Filling
Product concentration (mg a-')
TFF Microfiltration
100
0.2
Ultrddiafiltration Chromatography
100-10
0.2-2
Chromatography Ultrddiafiltration Virus removal Virus inactivation
10-2
2-10
Chromatography Ultrddiafiltration Sterile filtration
1-0.2
20-100
20-0.2
1-100
and, due to the unprofitable relation of product load to column size and volume, GPC can become a delicate downstream operation with respect to economy. The economic benefit of the elimination of an entire process step is evident, but the ingenious selection of buffer components and the total number of different buffers is also of broad economic and logistic impact. The availability of buffer make-up tanks and storage tanks might be limited, but it is even more the turnover procedure including CIP and SIP. Buffer make-up and regeneration procedures under GMP are time-consuming and labor-intensive, as they comprise beside the actual handling appropriate logistics for raw material testing, release and storage, LAL testing of the prepared buffer, and buffer storage validation. Therefore the limitation to simple buffer compositions of inexpensive ingredients, e.g., physiological buffers such as phosphate, glycine/ NaOH, acetate or citrate buffers, and highly effective regeneration solutions, e.g., 0.5-1.0 M NaOH, is a valuable task for basic process development. Thus, the resistance of separation media in chromatography and filtration is stringent. The application of highly selective but unfortunately vastly expensive matrices, e.g., affinity chromatography of antibodies on Protein A or G, is of impact on long-term production economy at large scale. This leads to considerations regarding batch versus cycle mode of processing. There is no general answer to this question, as the advantages of a batch mode (short process time, limited sampling and intermediate analysis, no need for definition and validation of pooling criteria, clear lot designation) need to be balanced against the preferences of a cycle mode (reduced initial investment in equipment and separation matrix, space-saving reduced size of equipment). The lifetime of the chromatographic matrix is another criterion which must be thoroughly determined, and which will definitely be one decisive factor for the mode selection. With respect to regulatory concerns a defined single batch strategy might be recommended; however, in the case of manufacturing campaigns limited to only a few product batches, cycling will be - and will remain - a viable option.
454
15 Strateaies and Considerations for Advanced Economv
1.5% load
T
-c
_ _
0
co
A
B C
I00
120
Chromatograms 1
= 4 mglml c = 3 rnglml c = 2 mglml
hl
J 0
20
40
60
80
140
160
I80
retention time (min)
T
c = 2 mghl
A
B C
Chromatograms2
retention time (min) Fig. 15 -3. Exploitation of gel permeation chromatography. The chromatograms show GPC separation of dimeric (A) from monomeric (B) monoclonal antibody (Mab, IgG2,). The chromatograms illustrate two different strategies in maximizing the column load. Chromatograms 1 feature an identical volumetric load (1.5 % of the packed gel bed volume) at different protein concentrations and look quite comparable regarding resolution. For chromatograms 2 the volumetric load varied from 1.0 to 3.5 % at a constant product concentration. Resolution clearly decreases with an increasing volumetric load: there is no resolution left towards the lower-molecular weight species (C), but - depending on the requested performance - still an acceptable separation of the dimer fraction from the monomers. linear flow rate: 25.0 cm h-' Technical data: column diameter 5.0 cm column height 86.0 cm matrix: Superdax 200 pg (Pharmacia) buffer: 50 mM citrate pH 6.25 column volume 1687 ml
15.3.4 Process Validation The modular design of DSP supports all aspects of process validation. In particular, drug safety is improved early on in development due to the application of proven measures and technologies for the removal and inactivation of potential contaminating DNA and virus. Although a number of preventive measures can be taken to minimize the extent of contamination, the DSP for any clinical-grade protein derived from biological sources must be validated accordingly (Table 15-4).
15.3 Strategic Development of Unit Operations
455
Table 15-4. Measures for the prevention and clearance of potential contaminants in biopharmaceutical manufacturing. 0
Virus preventiodclearance virus-free producer cell in animal cell culture - virus-free sera and raw materials - validation of respective unit operations in DSP for virus inactivation and virus removal
-
0
0
DNA preventiodclearance - gentle cell harvest to minimize cell rupture - anion-exchange chromatography, preferably no product binding - product-specific adsorption chromatography with no DNA binding Pyrogen prevention/clearance sanitary environment - pyrogen-free media and buffers in fermentation and DSP - ultrafiltration of media and buffer at a membrane cut-off < 10 kDa - anion-exchange chromatography, preferably no product binding - product-specific adsorption chromatography with no pyrogen binding -
The use of filtration technologies, e.g., ultrafiltration and nanofiltration, for the mechanical removal of viral particles can easily be implemented to a large extent independent of the process stage. The simple but highly efficient removal of contaminating residual cellular DNA and DNA fragments by means of anion-exchange chromatography provides for a valuable and time-saving approach. Chromatographic removal of DNA might be based upon two competing methodologies: the use of chromatographic beads or membranes (Fig. 15-4). Assuming a position in the midstream section of DSP, the total load of DNA is typically fairly low at 50-500 pg per mg protein (Fig. 15-5). Consequently, the overall process time which is correlated with the effective volumetric flowrate gains importance. Membranes feature a superior pressure/flow relation; therefore, the total membrane area can be reduced to a dimension which ensures a reliable dynamic capacity for DNA clearance. In contrast, the size of a conventional chromatographic column containing beads is most likely balanced on the applicable volumetric flowrate. Hence, the application of strong anion-exchange membranes can diminish a loss of product based on nonspecific adsorption in consequence of a considerable reduction of total surface compared with conventional chromatographic beads. In addition, the application of adsorptive membranes for the purpose of DNA removal is of practical impact on the performance and validation work: otherwise than labor-intensive and hardware-expensive column chromatography, the membranes are easily handled, installed as in-line filters, and can be disposed after use, not requiring comprehensive regeneration procedures.
456
15 Strategies and Considerations for Advanced Economy
f w
0
8
100
0
200
300
400
500
600
700
800
900
1000
1100
1200
1300
Volume (mi)
-
Fig. 15 4. Anion-exchange chromatography for DNA removal. A monoclonal antibody, dissolved in a 10 mM phosphate/250 mM NaCI buffer pH 7.25, 25 mS cm-I at a concentation of 0.63 mg r n - I , was spiked with 25 pg ml-' DNA isolated from CHO-cells. A total of 900 ml was filtered through a Sartobind QlOO anion-exchange membrane (Sartorius, Gottingen, Germany) at a flowrate of 30 ml min-I. The effluent was monitored photometrically parallel at 254 nm (DNA) and 280 nm (protein). Beside photometrical determination, DNA was analytically measured at 50-ml fractions. A slight continuous increase of the 254 nm line starting at 620 ml effluent indicates the DNA breakthrough. The dynamic loading capacity for the Sartobind Q membrane was determined by analytical DNA measurement with 156 pg DNA cm-*. In a further experiment the dynamic binding capacity was determined for a bead-type anion-exchange matrix: Q Sepharose FF (Pharmacia, Uppsala, Sweden). A chromatographic column ( h = 11 cm, d = 1.0 cm, v = 8.6 ml) was loaded comparable with the Sartobind Q membrane. However, a linear flowrate of 4," = 100 cm h-l equivalent to 78 ml h-l, was applied to the column, which is a magnitude below the flowrate applied to the membrane (1800 ml h-l). Due to the vastly increased residence time, the respective dynamic binding capacity for Q Sepharose FF amounted to 581 pg DNA d - l . Under these coditions, a membrane area of 3.75 cmz is equivalent to 1 ml bead-type matrix. The process time is reduced 23-fold, excluding the benefit in handling for set-up.
15.4 Economy in Process Completion
l!CeU Hawest
ProteinA 0 Sartoblnd Q
451
#batch
Fig. 15-5. DNA clearance in downstream processing. The DNA burden was determined at different stages in the DSP of a monoclonal antibody (MAb). The variation of DNA load within the different cell harvests reflects the variability of a biological process: it is due to cell number and cell viability in the fermentation fluid as well as cell rupture during the harvest procedure Already, Mab-specific affinity chromatography on Protein A reduces the DNA to a very consistent level. Filtration through an anion-exchange membrane (Sartobind Q ) clears the DNA below the detection limit.
15.4 Economy in Process Completion Process automation and online data acquisition allow the transfer of noncritical but time-consuming process sections to periods with reduced personnel employment, e.g., night shift or weekend. Automation might make works accessible which can not or not fully be performed during such times. The interconnection of process sections might help to reduce the overall process time. The warranty of a suitable process control, including continuous documentation of relevant process parameters, already during early stages in development helps to limitate the number of scaleup experiments, assuming that a proper linear scale-up is followed. Thus, an initial risk analysis can be performed at a very early stage in development, which helps to identify weak points and to improve the process robustness. The course of the process development of a protein drug may be reflected by the stages in product development (Table 15-5). Early studies in the product development (Toxicology, Clinical Phase I and IIa) are content with a limited amount of the desired protein product: the purpose of these stages in product development is the evidence of safety, tolerance, and efficacy. The protein drug is produced at smaller scale, but needs to feature appropriate quality and safety. The magic term is ‘dedi-
458
15 Strategies and Considerations for Advanced Economy
Table 15-5.Course of development for a biotechnical protein drug. 0
Genetechnical development DNA gene sequence - transfection - cell screening, amplification - cell banking
-
0
0
Assay development analytics - physical assays - functional bioassays - immunoassays Feasibility study fermentation - DSP -
0
Development of the basic process - cell biology: MCB, MWCB media optimisation, process operation - fermentation: - DSP: process design - Toxicology studies
0
Process consolidatiodvalidation validation of DNNvirus clearance - DSP:
0
IND filing - Phase I and IIa clinical studies
0
Process optimizationkale-up process transfer to designated manufacturing scale and site
-
0
Validation of the manufacturing process Phase IIb and 111 clinical studies
-
0
BLA filinghicensing
cated equipment’, which should be a guarantee to exclude the cross-contamination with another product. This principle is fundamental for large-scale manufacturing, but sounds rather hindering for a production campaign limited to a single or few batches only in a developmental phase, or in the manufacturing of a drug with a limited market need. Chromatographic matrices as well as filters and membranes applied to ultrddiafiltration, nanofiltration, and microfiltration represent a vast surface which comes into close contact with the processed fluids. The regeneration procedures for such materials are key operations in the cGMP manufacturing, guaranteeing a most reliable product quality. Nevertheless, the use of filters, membranes, and chromatographic matrices is undoubtly dedicated to a distinctive product: the required efforts in validation in order to eliminate any potential cross-contamination would easily exceed a potential financial benefit given by matrix re-use for different protein products. However, respective cleaning and regeneration operations are to be applied to all other surfaces exposed to process fluids: piping, probes, pumps as well as chromatography columns, filter housings, and vessels. Cleaning fluids based on caustic
15.4 Economy in Process Completion
459
and acidic solutions and detergents are feasible for stainless steel equipment (the recommended grades are SS3 16L and DINl.4435, both electropolished with an interior finish of Ra < 0.8 pm). Their application as a CIP procedure and its performance must be re-validated accordingly. The time requirements and financial impact are significant due to labor and analytical expenditures. The analytical tools need to be developed product-specifically and in place at the right time. The feasibility of all such efforts is given for a manufacturing process which goes into operation after a considerable development time: the product has successfully passed the various clinical phases and has been shown to provide both efficiencyhafety and product quality in an extensive range of analytical investigations. At the time when a designated protein product is at the beginning of its clinical trials, i.e., after development at laboratory scale and first scale-up in order to obtain reasonable amounts of product for a clinical start-up, not all of the described validation and analytical tools might be available. Nevertheless a high through-put of different protein products in a pilot plant designated for clinical use requires an appropriate validation of the product changeover. While generic regeneration methods applied to product-dedicated chromatography and filtration media can be used as ‘process modules’, the cleaning of equipment (tanks, piping, etc.) which will be used for different products does not allow for compromises. An exciting solution both to technology and economy is the use of disposable hardware - piping as well as containers - which can remove a significant burden from pilot-scale manufacturing of clinical-grade biopharmaceuticals. Appropriate disposables feature a technical solution even for the cubic meter scale. Attention is drawn to materials, compatibility, leachables, temperature resistance, and flexibility. Design and sizes of process containers are different and dependent on the purpose of their application - either as a simple storage device or as designated containment for mixing/adjustment of process fluids. Polypropylene or polyethylene tanks are used widely as inexpensive hardware for storage of process fluids. Although they are dedicated to distinctive process buffers or product solutions, they do not really represent ,disposables’ as their repeated use within a production campaign again calls for significant efforts in their cleaning and sanitization. In addition, such containers occupy valuable storage space when not in use, and their supply and storage prior to a scheduled campaign may be problematic. Flexible bags such as FlexboyTM,BioPharmTM, or BioPharm XLTM (Stedim, Aubagne, France) are manufactured from inert polymers and eradicate the obvious shortcomings of plastic tanks. One excellent advantage is that the bags are delivered sterile and pyrogen-free. They are available in sizes from 50 mL to 800 L and are equipped with customer-specified piping (tubing), including sterile filter(s), sample ports, and connectors. Equipped in this way ‘bag technology’ effectively supports any need for operations in closed systems under full cGMP. Beside their application as storage devices for samples, buffers, product intermediates, and bulk product, designed bags provide an excellent mixing behavior, and thus can be applied conveniently to ultddiafiltration operations (Fig. 15- 6). With regard to process logistics, such ‘bag technology’ facilitates the availability of required containers at any time. Moreover, such bags utilize a minimum storage space and handling; thus the cost-benefit relation-ship in comparison with plastic tanks is undoubted in favor of the bags.
460
15 Strategies and Considerations for Advanced Economy
10
9 0
7
3 2 1 1
2
4
3
5
0 6
7
factor of buffer exchange -100
I bag mndudiity
--fl.100 I bag pH-value
- c b 100 Itank mndudiiity
- - o IOOltank -pH-value
-
Fig. 15 6. Bag technology in ultrddiafiltration. Ultrddiafiltration was performed using a stirred 100-L PP tank in comparison with a non-stirred 100-L BioPharmTMbag. In the experiment on conductivity, a 1 M NaCI solution was diafiltered against deionized water. In the experiment on pH, a 0.1 M phosphate buffer of different pH was diafiltered (pH 4.46 versus pH 9.1 ). Performance in the efficiency of mixing was good in both types of hardware.
References [ l ] Born, C., Thommes, J., Biselli, M., Wandrey, C., Kula, M.-R., Bioprocess Eng, 1996, 15, 2129. [2] Walter, J. K., Werz, W., Berthold, W., Biotech Forum Europe, 1992, 9, 560-564. [3] Walter, J. K., Werz, W., Berthold, W., in: Viral Safety and Evaluation of Viral Clearance from Biopharmaceutical Products; Brown, F., Lubiniecki, A. S . (Eds.), Dev Biol Stand, Karger: Basel, 1996, vol. 88, pp. 99-108. [4] Walter, J. K., Nothelfer, F., Werz, W., in: ACS Symposium Series, Validation of Biopharmaceutical Manufacturing Processes, 1998. [5] Walter, J. K., Nothelfer, F., Werz, W., in: Bioseparation and Bioprocessing,.Vol. I; Subramanian, G. (Ed.), Weinheim, Wiley-VCH: 1998. [6] Walter, J. K., Allgaier, H., in: Manipulation of Mammalian Cells; Wagner, R., Hauser, H. J. (Eds.), GBF Braunschweig, Berlin-New York: Walter de Gruyter, 1997, pp. 453-482. [7] Berthold, W., Walter, J. K., Biologicals, 1994, 22, 135-150.
Bioseparation and Bioprocessing Edited by . Ganapathy Subramanian Copyright QWILEY-VCH Verlag GmbH, 1998
Index
a priori assumptions 426 absorbance 127 absorption 267 acetic acid 219, 272 acetonitrile 70, 109, 154 4-acetoxy-cyclopentanone 277 acids 260 -, organic 280 acetate formation 402 activated tentacle chemistry 3 13 activation level 366 activation procedures 3 11 activity -, biological 41, 82 -, specific 59 activity recovery 59 ADALINE 369 adaptive linear estimator 395 adaptive linear networks (ADALINE) 369 adaptive neural networks 389 additives 66, 278, 499 adenine residue 113 adenoviruses 154 Adogen 464 275, 287 adsorbent 66, 113, 210, 345 - beads 204 - cake 212f. - concentration 330 - dispersion 204 - particles 214f. -, polydispersed 204 - sensor 209 - Slurry 215 -, solid 5 - velocity 347 adsorption 90, 251, 305 f. - capacity 204 - efficiency 208 -, hydrophobic 81
isotherms 97, 348 kinetics 332 non-specific 306 performance 208 - process 114, 200 - site 17, 160f., 174 - step 199 adsorption-desorption 146, 155 advancing front model (AFM) 283 affinity 45, 53, 86, 146, 156 - adsorbent 116 - adsorption 308 - beads 317 -, biospecific 45 - chromatography 63, 94, 113, 122, 223, 305, 479 - elution 122 - interactions 305 - ligands 113, 312 - matrix 114 - media 313 - membranes 320 - purification 223 - sites 317 - systems 160 - technique 121 agar 158, 290 agarose 114, 117, 150, 165, 201 f., 307 - adsorbents 202 -, beaded 119 f. -, cross-linked 169 -, macroporous 169, 210 -, non-cross-linked 157 - spheres 118 agglomerate 492 aggregates 84 aggregation 94, 218, 501 air flow 431 air sensors 127 -, -
514
Index
Alamine 336 289 albumin 107 alcalase 242, 244 alcohol dehydrogenase, native 224 alginate 100, 107, 201 Aliquat 336 290 alkaline phosphatase 77, 315 2-alkoxy-1-methylpyridine 3 12 alkylaminohalogenalkanes 307 alumina, porous 172 Amarine 103 Amberlite 27 Amberlite LA-2 275, 292 f. aminimides 123 amino acid 99, 280, 299 D-amino acid hexapeptides 123 L-amino acid hydrochloride 288 aminoacylase 243 6-aminopenicillic acid 293 y-aminopropyl trimethoxy silane 247 ammonia 405 ammonium citrate 105 ammonium sulfate 65, 68 f., 73 AMP 116 AMP-agarose 120 amphiphiles, cationic 295 a-amylase 114, 296 f. analog signal 405 analog-to-digital converter 405 analysis -, final 47 -, mathematical 282 -, numerical 449 -, screening 47 -, sensitive 502 -, specific 502 -, statistical 448 analytical methods 47, 49, 501, 510 -, automated 49 analytical tools 502 anion charges 258 anion exchange chromatography 494 anion exchanger 108, 221, 503 ANN see artificial neural network annexin V, human placental 222 antibiotics 90f., 99, 105f., 200 -, fi-lactam 299 antibody 120, 150, 189f., 481 - capacity 153 - exchange (Abx) 107 -, IgG-type 488 -, monoclonal 107, 116, 156, 189, 314, 465, 491 f., 501, 503
-, non-specific 47 -, polyclonal 156 antigens 116 Antithrombin I11 (rh-AT 111) 107, 121 applicability 299 aprotinin 107, 121, 223 arginine 486 artefacts 477 artificial intelligence 407 artificial neural network (ANN) 363, 419 -, autoassociative (aANN) 404, 407, 426 f., 460 -, recurrent 372 artificial variable-space 460 artificial variables, correlations 427 Asahi Planova 491 ascites 107 aspartate 113 Aspergillus niger 245, 379 Aspergillus oryzae 106 assay variability 476 asynchronous shift 12 automated systems 5 10 -, chromatography 507 automation 131, 506 autoregressive approach 372 axial annular flow reactor 241, 245 axial flow chromatography 133, 136 axial length 331 azeotrope 11
Bacillus circulans 245 Bacillus licheniformis 244 Bacillus thuringiensis fermentation back-extraction 290, 296, 298 back-flushing 150, 216, 239 back-mixing 202 back-washing 260 backpressure 52, 145, 152, 157 backpropagation 422 - network 371 bacteria 199 - cells 269 - cultures 217 - fermentation 220 -, Gram-negative 498 - homogenates 217 bandspreading 42, 58, 63 batch - age 404 - chromatography 356 - operations 200 - sizes 137
394
index
bead - diameter 193 -, macroporous 305
-, mineral
173 volume 158 bed - diameter 136, 192 - expansion 192, 206 -, fluidized 192f., 200, 351 - height 80, 132, 136, 146, 150, 215 - _ , settled 218f. - integrity 145 -, monocomponent 205 -, packed 167, 185, 187, 199, 204, 208, 33 1 - porosity 9 - reactors 203 - stability 206, 208 -, variable 133 -, vertical 135 - voidage 201, 204 beer fermentation 416 beer formation 249 Bellman’s dynamic programming 438 Benzonase 221 f. N-benzoyl-~-arginyl-~-methioninmide 1 04 benz y ldimethy ldodecy I-ammonium bromide 99, 106 benzyltributylammonium chloride 105, 109 bias neuron 367 binary separator 10 binding 81, 92 - affinity 73 - capacity 74, 122, 158, 172, 175f., 214 -, chromatographic 46 - constants 298 - densities 320 -, irreversible 151 -, non-specific 119 - sites 4, 113, 349 - specifity 115 -, static 187 bioburden 505 biochemical theories 422 bioengineering 424 - applications 396 biomass 429 - dry weight 215 biopharmaceuticals 200, 468 -, clinical 499 biopolymer chromatography 92 bioprocess 41 1 - monitoring 416 -
515
biopurification 157 bioreactor 364, 373, 381, 422 biosensors 502 biospecific interaction 9 I , 486 biosurfactant 298 biotechnology - industry 494 - derived products 502 - processes 414 bisoxiranes 309, 3 16 black-box 434 - estimators 419 blank runs 508 block virus removal 494 P-blockers 3 1 blood - factors 156 - fractionation 125 -, hepatic 286 - plasma 95 - stream 267 bond energies 275 boronate 120 bottleneck layer 428, 433, 461 bottleneck nodes 433 boundary layer 187 bovine - albumin 174 - cell line 467 - lens aldose reductase 122 - serum 107 - serum albumin (BSA) 95, 155, 166, 180, 182, 209, 298, 310 - viruses 469 BrainMaker 448 breakthrough capacity 83 breakthrough curve 180, 187 brine 259 Brinkman’s model 165 bubble point test 489 buffer 138, 499, 502, 504 - additives 46 - exchange 49 - specification 138 - strength 145 bulk - concentration 330 - formulation 494 - liquid 330, 345 - phase 333, 335 - suspension 230 I ,4-butanediol diglycidol ether 3 12 f. 2-(2-butoxyethoxy)ethanol (BBE) 99, 105
516
Index
n-butyl acetate 292 f. 4 -t-butylcyclohexanone
104
calcium citrate 290 calcium phosphate 259 calf thymus DNA 503 calibration 471, 506 -, false 314 Candida cylindracea 246 capacity 45, 59, 499 -, dynamic 59, 167, 220, 500 -, high 157 -, ionic 510 -, low 81 -, saturation 59 capillary channels 146 capture 49, 72f., 223 carbodiimides 3 14 carbohydrate 387 carbohydrate polymers 306 f. carbon balance 417 carbon dioxide 253 - evolution rate (CER) 374, 395, 404 carbonyldiimidazole 3 10, 3 12 carboxamide 113 carboxymethyl starch 100, 107 carboxymethyldextranes (CM-D) 100 carboxypeptidase G 3 15 carboxypeptidase Y 104 f. carrier - concentration 279 -, inert 93 - molecules 270f., 275 cascade reactor 240 casein 244, 257 - fraction 108 - glycomacropeptide 257 catalytic membrane reactors 243 catalytic membrane technology 244 cation exchange 189, 200, 222 cell - bank system 465 - biology 412 -, Chinese hamster o ray (CHO) 466 - culture 314, 465, 468, 507 - - broth 199 - -fluid 150 - - supernatant 52 - debris 117, 133, 152, 192, 199, 204, 220 - density 200, 401 - derived vesicles 489 -, diploid 461 - disruption 133, 219, 507
- division - docking
392 470 -, eukaryotic 477 - extraction 261 - growth 376 - harvest 200, 494 - homogenate 200 -, inoculated 392 - layer 475 - lysis 90, 217, 267, 373 -, mammalian 89, 95, 199, 217, 477, 500 - membranes 486 - monolayer 475 -, murine embryo 467 - organells 220 -, plasmid-free 393 - residues 204 - supernatants 189f. - suspensions 200 -, whole 220 cellular metabolism 364 cellulose 150, 164, 169, 202, 241, 308f. - membrane 312 -, regenerated 306 -, un-cross-linked 153 centrifugal extractor 292 centrifugation 90, 199, 220, 224, 253, 267, 411 cephalosporin C 106 ceramics 241 cetramide 99, 103f., 106 cetyltrimethylammonium bromide see cetramide change control policy 509 change control procedures 506 changing processes 41 8 chaotropes 66 charcoal 241 charge 45 -, ionic 182, 189 CHASM 45 chelating group 100 chemical conditions 45 f. chemostat 392 chemotaxis 438 chitin 242 chitosan 164 cholesterol 274 chondroitin sulfate 100, 106 f. chromatography - column 510 -, high-throughput 62 - media 508, 510
Index
-, non-linear 16 - -, preparation 502 - systems 126, 131 chromium 269 a-chymotrypsin 287 f., 296 f. chymotrypsinogen 69 ff. Cibacron Blue Sepharose CL 107 circular system 6 citric acid 272, 290, 299 clarification 223, 250, 261 - techniques 199 cleaning 84, 131, 202, 218, 252, 471, 499ff.. 510 - agent 501 -, analytical methods 503 - evaluation 503 - methods 77 - protocol 216, 503, 508 cleaning-in-place (CIP) 129, 238, 500 - procedure 137, 214f. clearance 506 - factor 474 clinical trials 500f., 506, 510 clogging 74, 152 cloning 443, 467 closed-loop control 414, 439, 443 co-cultivation 467 co-extraction 275 co-solvents, organic I I6 co-transport 275 coalescence 27 1, 282 coenzymes 120 f., 3 15 -, NAD* 113 coenzymes A 116 cofactors 12 1 cold filtering 251 colony stimulating factors 156 column 52 -, axial 145 - bed 150 - bed-height 504 - contactors 285 - cycle 137 - design 132 - -, radial flow 146 - diameter 25, 132, 145 - dimensions 502 - efficiency 34, 43 - fouling 117 - hardware 52 - length 23, 52, 58 - life 63 - lifetime 504
517
loading 132 - operating strategy 136 - packing 50f., 133, 506, 508 -, pulsed perforated 292 - qualification 139 - regeneration 108 -, slurry packed 145 - velocity 187 - volume 52 comparability 5 10 competition, enforced 92 complexation 275, 282 compliance 470 component analysis 426 compressibility 157 compression, axial 30 computer modeling 50, 329 concavalin A 298 concentration 49, 223, 254, 261 - factor 94 - gradients 275 - polarization 238 - profile 11, 206, 345, 347 conductivity 127, 217, 501 f. conformational changes 97 consistency 498 constant separation factor (CSF) 349 contact time 504, 508 contaminants 59, 73, 133, 138, 218, 500 -, critical 60 contamination 424 continuous annular chromatography 95 continuous stirred tank membrane reactor (CSTMR) 240 continuous stirred tank reactor (STR) 240, 391 control profiles 443 control techniques 4 13 controlled release 269 controllers, adaptive 440 f. convection 334 -, intraparticle 160 convergence - criterion 393 -, rapid 366 conversion -, biochemical 429 -, enzymatic 287 - process 434 coordinate - covalent bonds 280 -, spartial 357 - transformation 428, 460 -
518
Index
copolymerization 101, 241 correlations 350, 352, 413 -, engineering 434 corrosion effects 139 costs 60, 412 -, effective 504 -, functions 376 cosurfactant 287, 296 counter-current 4 f., 8 f., 345 - flow 246, 285 - operation 285 -, simulated 9 -,true 9 counter-ion 163, 269, 275, 287, 292 counter-transport 275 counterbalance 12 Crabtree effects 399 cross-contamination 501 cross-linking 165, 308 cross-validation 422, 461 - technique 433 crossflow - filtration 199 f., 23 1, 490 - geometries 306 - modules 320 - velocity 490 crown ethers 280, 288 crystallization 90, 267 CSTR see continuous stirred tank reactor cultivation 475 current Good Manufacturing Practice (cGMP) 468f. cut-off limits 489 cyanogen bromide 114, 119, 307, 312 cyanuric chloride 119, 309 cycle rate, low 145 cycle time 62 cycling 62f. cyclodextrin 104, 280 cyclohexane 277, 296 f. cyclohexanone 278 cylindrical porous frites 146 cytochrome c 68, 296, 298 cytopathic effect 475 cytoplasmic accumulation 383 DARACLEAN 8471 2 19 f. Darcy equation 18 Darcy permeability 182 data - acquisition 453 - analysis 447 - -, software tools 445
-
-, statistical 445
back-up 131 bases 446 evaluation 471 - exploration 435, 446 - management 448 - processing, multivariable 427 - records 454 Data Engine (MIT) 448 de-emulsifying unit 293 deactivation 282 dead volumes 75, 84 dead zone 133, 150, 204 dead-end mode 470, 489 DEAE cellulose 150 DEAE-dextrane 107 debugger 447 I-decanol 272, 291, 293 decyltrimethylammonium bromide 99, 105 defuzzification 406, 459 degradation 422 dehydrogenases 3 15 demineralization 232, 254, 257 denaturants 241 denaturation 73, 76, 94, 151, 295, 470, 475, 507 dense particles 189, 191 density 210, 319 - differences 199 - profile 398 depectinization 259 f. depleted peak 48 deprenyl metabolites 103 depth filtration 23 1, 470, 488 depyrogenation 117, 120 desalting 25, 27, 49, 61, 155 design process 44 design qualification (DQ) 140 desorption 8, 75 destabilization 424 detection limit 21 1 detector cells 475 f. detergents 66, 82 detoxification 274 development process 47 f. deviation 430 dextran 117, 158, 163, 165, 202, 307 -, beaded 119 -, blue 181 -, cross-linked 157 - sulfate 100 dextrin 247 f. diacetyl 421
Index
diafiltration 90, 152, 189, 231, 250, 254, 493 dialysis 61, 250 dicyclohexylcarbodiimide 309 5,lO-dideazatetrahydrofolicacid (DDATHF) 104 diepoxides 307 diethylphthalate 103 differential equation 424, 45 1 differential-algebraic equations (DAE) 337 differentiation 446 diffusion 42, 253, 273, 337, 350f. -, axial 159 - chromatography 157 - coefficient 18, 161 -, connective 159 -, eddy 43 -, enhanced 162 - flux 334, 338 -, free 118 -, intraparticular 42, 153, 159f., 183 -, inward 278 -, liquid 330 -, longitudinal 43 - mechanisms 159f. -, molecular 337 -, particle 160 -, pore 43, 160 - processes 68, 278 - rate 271 - -, finite 283 -, solid 160, 162, 164, 181 -, solute 43 -, surface 160 ff. - time 193 - velocity 307 diffusivities 351 diglycerides 245 diisotridecylamine 215, 292 diltiazem 279 dilution factors 509 dioctylamine 293 Dirac function 92 disaster plans 506 discretization 353, 357 disintegration 470 disodium hydrogen phosphate dihydrate 299 dispersion 206, 321, 334, 350 -, axial 96, 98, 159, 187, 206ff., 216, 285, 329 -, axial coefficient 207, 351 -, coarse 270 - coefficient 96. 334 ->
519
effects 20, 330, 344 factors 506 fine 268 micellar 295 radial 153 solid 204 displacement - chromatography 11, 89, 96 - effects 83, 487 - mode 92 - train 93 displacer 92, 99 -, cetramide 103 - concentration 96, 98 - front 93 -, hydrophobic substances 99 - isotherm 94 -, polyionic 100 -, specifications 99 disposal issues 500 dissociation constant 116, 316 dissociation rate 275, 295 dissymetry 12 distillation 250, 267 distortions 443 distributed control systems (DCS) 131 distribution 210 -, axial 332 -, even 146 -, radial 332, 335, 340 - systems 204, 208 distributor 204 disulfide bridge 47 1 divalent cations 258 DNA 90, 502 - assays 505 - removal 494 - technology 141, 498 -, therapeutic plasmid products 502 documentation 471, 500, 506 dodecyloctyldimethylammonium chloride 99 dodecyltrimethylammonium bromide 296 Donan exclusion 18 Dowex 99 monosphere 26 downstream chromatography 147 downstream detectors 139 downstream processing 89, 110, 141, 145, 202, 267, 471 f., 477 -, design 494 downstream purification 65 drug release 283 drug safety 465 -, -, -, -,
520
Index
dye
- adsorbents 122 -, biomimetic 315 -, immobilized 122
-
matrix 122 dynamic behavior 422 dynamic binding capacity (DBC) 83, 169, 176, 187 dynamic changes 440 ECANSE (Siemens) 448 ecdysteroids 103 economics 77, 141 effector 120 effects, diffusive 331, 334 efficacy, chromatographic 487 efficiency 177, 274 effluent 508 egg trypsin inhibitor 68 electric field 163 electrodialysis 229, 232, 257 electrolyte 232, 337 electron balances 417 electron microscopy 174, 477 electroneutrality 350 electrophoresis 267 -, two-dimensional 107 electrostatic potential 337 eluate 494 eluent consumption 8, 21 eluent flow 146 elution 214, 216, 504 - azeotropes 97 - chromatography 4, 92 - concentrations 53 - gradient 51 -, optimal 55 - position 510 - profile 58, 81 -, selective 116 - scheme 500 -, step-wise 81 - time 502 - volume 214f. emergency backup 507 Empigen BB 222 emulsification 271 emulsifier 293 emulsion 268 ff. - composition 290 - globules 280, 286 -, isotropic, stable 268 -, oil-water-oil 268
-, water-oil-water 269 f. endotoxins 72, 90, 498, 501, 505, 507 - measurement 508 - testing 508 ENJ3029 285 enolase 68 environmental considerations 5 10 enzymatic reaction 274, 278 enzyme - activity 287 -, adsorption immobilization 242 - cofactor 223 -, covalent attachment 241 -, diagnostic 156 -, encapsulating 269 -, entrapment 241 f. -, glycolytic 223 - immobilization 241 -, oligomeric 241 -, proteolytic 251 - recycling 261 -, restriction 156 - retention 240 - stability 243 -, thennoinactivation 241 epichlorohydrin 119, 309 epoch 370 epoxide, chiral 31 equilibration 147, 150, 215 equilibrium - constant 350 - isotherm 337, 348 - model 342 - stage 22 equipment 507, 509 - components 508 - faults 424 - reliability 137 - selection 506 error - compensator 424 - derivatives 371 - distribution 393 - notification 506 Escherichia coli 152, 199, 217, 219f., 222, 369, 384, 424, 431, 498, 500 -, recombinant 401 estimation error 416 ethane 272 ethanol 67, 262, 376 - production 399 ethanolamine 3 10
Index ethylene 272 ethylene glycol 67, 82 - bis((3-aminoethylether)-N,N,N’,”-tetraacetic acid (EGTA) 101 -, high viscosity 82 ethylenediaminetetraacetic acid (EDTA) 238 Eudragits 100, 107 evaluation, statistical 476 evaporation 250, 253 75 exotoxin, Z? aeroginosa expanded bed 174, 189, 199, 202, 208 expanded bed adsorption 201, 220 -, applications 220 -, experimental strategy 212 expanded bed processes 209 -, development 209 -, operation 209, 215 expansion 215 experimental design 42, 454 experimental program 50 expert systems 415, 426 extended cell banks 468 extra-cellular product 383 extract 271 extraction 90, 269 - decanter 292 extrapolation capacity 433 extrapolation measure 424 facility design 500 fatty acids 29, 246 fault analysis 414f., 424, 426 fault detector 385 fed back 236 - errors 372 fed batch 235 feed - concentration 21 - distribution 318 - molecules 92 - solution 272 - tank 236 feedforward 399 feedstock 11, 147, 152, 189 - application 216 - composition 215 -, high biomass 217 -, viscous 217 feedstream 209, 236, 498, 500, 505 Feline 103 fermentation 204, 234, 249, 262, 363 f., 373, 383, 396, 406, 429, 466, 472, 500, 507 - age 395
- Bacillus thuringiensis
396 behavior 395 - broth 107, 267, 270, 279 -, disturbances 407 -, dynamics 407 -, fed-batch 399, 401 - harvest 467 -, mathematical model 402, 406 - model 389 -, mycelial 387 -, non-isothermal 377 - products 267 -, serum-free 466 - time 433 -, time-dependent 407 fermenter 217, 495 - control 453 - hardware 452 fiber lumen, laminar 3 19 Fick’s law 337 filter cartridge 492 filter integrity 490 filtrate 494 - flux 488, 492 filtration 90, 224, 446, 470, 472, 474 - principle 491 - processes 487f. -, sterile 239, 490, 494, 499 - techniques 490 fine tuning 507 finite difference method (FDM) 353 finite volume method (FVM) 353 fixed bed 133, 349 - columns 5, 133 flocculation 220 flow - characteristics 207 - constrictions 203 -, convective 185, 317f., 330f. - direction 128 - distribution 203 - path 48, 146 - properties 204, 502 - rate 179, 252, 484, 500 - velocity 83, 118, 206, 214 flow-rate-to-bed-volume ratio 504 flowthrough 494 fluid - phase 332 -, static 490 - velocity 334 - viscosity 205 -
521
522
Index
fluidization 210 - velocity 192 fluidized mode 133 fluidized state 200 fluorescence probe 429 2-fluoro-1-methylpyridine 3 12 flux -, convective 205, 334 -, low 288 - rate 238, 247 -, relative 163 focus-forming units 475 food-grade material 499 formaldehyde 3 16 formulation 494 FORTRAN 450 fouling 200 fraction, eluted 216 fractional factorial analysis 499 fractionation 23 1, 254 Fractogel 150, 307 friction forces 202 front-end-control 443 front-end-systems 452 frontal analysis 83, 175 frontal chromatography 92 fructose 261 functional group 501 functionality 305 Fusarium graminearum 387 fuzzy - controllers 442 - expert systems 415 - information 396 - logic 397 - model 459 - modeling 422 - neural networks 373 - rules 422, 426 - sets 426, 457 FuzzyCLIPS (NASA) 448 fuzzyTECH (Inform) 448 galactose 245 j3-galactosidase 106, 244, 401 gas - molecules 233 - separation 229, 233 - stream 253 - transfer 253 Gaussian distribution 367 gel - beads 305
compression 145 concentration 182 cross-linked 277 filtration 65, 74, 84, 90, 307 hydrated 119 media 168 network 181 permeation 133 - chromatography 135, 487 -, stable 307 - types 504 Gelman Supor 30 491 genetic algorithms 385, 407 genetic drift 424 genetic engineering I25 genom 477 Gensym G2 451 geometric configuration 164 geometry -, crossflow 318 -, dead-end 318, 323 gigaporous media 182 globules 268 -, isotropic 286 -, organic 283 - size velocities 285 globulin Gc-2 107 glucoamylase 242, 247, 379 gluconic acid 247 f. glucose 245, 247f., 278, 376, 433 glucose 6 -phosphate dehydrogenase (G6PDH) 223 glucose dehydrogenase 247 glutamate 113 glutamine 433 glutaraldehyde 246f., 313f. L-glutathione 29 L-glutathione/glutamic acid 32 gluten 262 glyceraldehyde-3 -phosphate dehydrogenase 369, 392f. glycerol 27, 246 glycidyl methacrylate 310f., 314 glycine 211 glycopeptide 153 glycoproteins 121 glycoproteins gp 120 222 glycosylation pattern 507 Good Manufacturing Practice 125, 498 grades approach 505 gradient 75, 500, 502 - elution 83, 128, 500 -, linear 69 -, -, -
Index - mixing 84, 506 - optimization 54
slope 54f. 371 graft polymerization 308, 3 11 graphical user interface (GUI) 447 f. gravity constant 192 gray-box models 434 grid points 356 growth - factors 156 - medium 396 -, microbial 501 - rate 374 -, specific 429 guanidine hydrochloride 66,76,211,219,224 -
- technique
hafnium oxide 193 Hansenula polymorpha 223 hardlimit function 369 heat - exchanger 424, 480 - inactivation 484 - resistance 483 hemoglobin 499 HEPA-filtration 469 heparin 104, 121 hepatitis 488, 492, 494 heptane 289, 291 heterogeneity 89 hetrazepine 3 1 heuristics 407, 413, 422, 451 hexamethylenediamine 3 13 hexanol 296 hexokinases 3 15 HIC see hydrophobic interaction chromatograph) hidden layer 365 high-performance displacement chromatography 96 histidyl-Sepharose 153 HIV-1 (human immunodeficiency virus) 222 Hoe F2562 275 hold time 481 f., 500f. hold-up 285 hollow-fiber 271 ff., 318, 492 -, hydrophilic 273 - modules 288, 291 -, porous 297 hormones 116, 120 HPDC see high-performance displacement chromatography human - blood-derived products 477
523
cell line 467 gene therapies 154 growth hormones 103, 156 pathogens 465 plasma 498, 504 serum albumin (HSA) 28, 68 -, recombinant 224 tissue 465 humidity, relative 507 Hybnet 451 hybrid models 415, 418f., 422 -, complicated 435 hybrid tools 450 hybridization 505 hybridoma cells 465 -, murine 466 hydration 277 hydrocarbons 267 hydrodynamic 22. 202, 387 hydrodynamic response 206 hydrogel 167, 169, 316 -, cationic 174 hydrogen - balance 417 - bonds 280 - peroxide 239 hydrophilic-lipophilic balance (HLB) 268 hydrophilicity 308 hydrophilization 7 1 hydrophobic heads 287 hydrophobic interaction 27f., 53, 65, 81, 91, 121f., 146, 156, 160 - chromatography 60, 65, 100, 113, 311, 487 f., 494 - -, expanded bed 224 -, entropy-driven 68 - media 501 -, non-polar moieties 65 hydrophobic patches 66, 119 hydrophobic pocket 113 hydrophobicity 45, 68, 105, 113, 202, 280, 288, 297, 308 N-hydroxy succinimide 3 12 hydroxyapatite 100 f., 107 hydroxyethylceliulose (HEC) 312, 314, 316 hydroxypropylacrylate 309 hygiene 126, 133, 469 - levels 137 - factors 138 HyperD 158f. - microstructure 174 -, polycationic 189 Hyperdiffusion 159, 164 -
524
Index ~
hypersurfaces 447 hyphenated techniques hypochlorite 239
103
ICH guideline 503 ideal adsorbed solution (IAS) 97 ideal system 19 IMAC see immobilized metal affinity chromatography imidazolylcarbonate 3 12 iminodiacetic acid (IDA) 101, 316 imitator 443 immiscibility 268 immobilization 173, 243, 272 immobilized metal affinity chromatography 98, 316 immunoadsorbents 499 immunoaffinity - chromatography 153 - resin 152 immunoassays 49 immunoglobulin 109, 121, 188f., 314, 486 -, humanized IgG4 223 -, monoclonal IgGl 70 immunomycin 200 immunosorbent 3 15 IMPAQ 200A 155 impurities 103, 501 f., 504 -, hydrophobic 84 impurity profile 500, 507 inaccuracies 418 inactivation 471 incubation 478 induced coupled plasma atomic emission spectroscopy 21 1 induction 402 -, chemical 384 -, thermal 384 industrial chromatography 158 infection 465 influx 387 inhibition constants 389 inhibitors 113, 120 inlet plate 148 inlet pressure 490 input - layer 365 - variables 428, 433, 457 - vector 430 instability, structural 385 installation qualification (IQ) 140, 47 1, 506 insulin 95, 99, 107, 299, 469 integration 446 I
inter-fiber space 272 inter-phase transport 285 interactive language 447 intercolumn 11 interface -, inner 269 - tensions 268, 276 interface-immobilizing pressure 277 interferons 156, 299 interleukins 219, 156 intermediate 501 - phase 274 - purification 71, 74 International Conference on Harmonization 497 interpolation 419 interstices 186 interstitial liquid 330 intracellular components 217, 392 invertase 244 ion binding 113 ion exchangers 157, 216, 323, 501, 504, 510 -, hollow fibers 323 ion-exchange 17, 65, 84, 113, 146, 156, 292, 305 - adsorbent 223 -, binary 350 - capacity 145 - chromatography 122, 487 - -media 499 - displacement chromatography 98 - hydrogel 174 - isotherms 348, 350 - packings 52 - resin 24, 26 - sites 119 ion-exclusion 17, 27 ion-pair extraction 292 ion-pair interaction 114 ionic strength 46, 114, 116, 190, 296, 299 iris explorer 445 isobufens 104 isoelectric point 177, 297 isohexadecanol 291 isomeric enrichment 280 isooctane 296 f. isopropanol 67, 70, 218 isotachicity 94, 96 isotherm, irreversible 167 isotridecanol 296 isozymes 245 iteration 370 iterative convergence 37 1
Index Kalman filter 389, 418 -, extended (EKF) 418 kerosene 268 a-ketoisocaproate 276 f. kidneys 281 kinetic 396 - changes 387 - control 281 - effects 96 - expressions 424 - models 417 - rate 389, 424 - relationship 419 - resistance 96 Kuhni column 292 a-lactalbumin 108, 166, 256f. lactase 242 ff. lactate dehydrogenase 107, 116 lactic acid 278, 282, 289 Lactobacillus casei 290 Lactobacillus delbreuckii 289 P-lactoglobulin 104, 108, 177, 256 lactose 244, 256, 290 laminar flow 203 Langmuir isotherm 17 -, linear approximation 349 N-lauryl-N-trialkylmethyl amine 275 LCST see lower critical solution temperature leak checks 506 leakage 211 learning, adaptive 367 learning rates 394 learning vector quantization 373 leucine 182, 276 f. leucine dehydrogenase 276 Licosep 8-200 30 ligand 114f., 311 - bleeding 486 -, chemical 158 -, combinatorial 123 -, competitive 116 - coupling 117 -, dedicated 486 - densitiy 308 - - packings 52 -, dye 114 - exchange chromatography 31 -, general 116, 121 -, group-specific 114, 314 -, hydrophobic 65f., 76, 501 -, immobilized 114, 3 12 -, mimetic 114, 120
- protein interaction - purity 122
114
screening methods
123
-
525
- selection 120 -, specific 116
-, thiophilic
315 -, triazine 121 ligate 314 limit, critical 435 limiting factor 56 limiting flux theory 381 linear flow 215 - rate 132, 504 - velocity 216 linguistic expressions 422 lipase 242, 245 f. -, lingual 245 lipids 117, 256, 494 lipoproteins 256 liquid film 267, 280 liquid stream 253 liquid-liquid extraction 290, 295, 299 liver 281 load volumes 132 loadability 57, 59 loading 52, 56, 305 - step 152 log reduction 468 long bed depth 133 lower critical solution temperature 101 lyophilization 90 lysine 113, 313, 486 lysozyme 28, 68 ff., 166, 169, 188, 296 ff. macromolecules 23 1 f. -, biological 41 macroscopic balance equations 417 magnesium chloride 153 magnetic fields 201 magnetically stabilized fluidized bed 201 magnetite 210 malaria vaccine 106 malfunctions 374 L-malic acid 274 manufacturing operations 502 mapping 427, 457, 460 -, inverse 460 -, linear 427 -, static 372 - techniques 49 mass balance 91, 137, 332, 335, 340, 434 -, bulk 342 - equations 333, 419
526
Index
mass concentration 206 mass conservation 353 mass flux 333 mass recovery 59 mass spectrometry -, electrospray ionization (ESI) 103 -, fast atom bombardement (FAB) 103 mass transfer 18, 43. 96. 104. 153. ~, 159, 203, 258, 330, 339, 350 - coefficient 291 - control 281 -, external 280, 284 -, film 332 -, intraparticle 183 - limitations 297 - model 332f. - rate 291 - resistance 98, 285 mass transport 185, 330, 332ff. - mechanisms 204 material - density 174 - handling 138 - storage 138 matrix 116 -, chromatographic 470, 474 - computation 449 - costs 142 -, gel-filled 168 -, hydrophobic 70, 73, 83 - inventory 137 - management 137 -, shielded 313 MB see moving bed chromatography measurement errors 427 media - costs 49 -, porous 163 -, hydrophobic 77 - recycling schemes 95 melanocyte-stimulating hormone a/P-separation 106 membrane 474, 489f. -, affinity 305 -, asymmetric 489 -, n-butyl acetate 292 -, catalytic 229, 239 -, catalytic reactor (CMR) 234 -, ceramic 237, 242 -, chiral 288 - composition 270, 276, 279 - configuration 238, 241 -, dynamic range 322 I
-, flat sheet 315 - flux 243 - fouling 238 - globules 285 -, hollow-fiber 238, 241, 244, 253, 271 f,
-, hydrophobic 253 -, intact
277
-, ion-selective 232 -, liquid 267 -, liquid emulsion 270, 267 -, liquid-impregnated 269 f. - materials 229, 237 - matrix 233 -, microporous 253, 490 - operation 239 -, organic 268, 283 - phase 268f., 272 - poisoning 244 -, polyamide 242 -, polymeric 276 -, polypropylene 276, 288 -, porous 165 - process 229 -, PVC silica 242 -, sheet-type 306, 318 -, spiral 238, 241, 315 - stability 276 -, sulfonated polyethylene/poly(styreneco-divinylbenzene) 272 - surface 490 - swelling 277 -, symmetric 489 - technology 230, 256 - -, hydrophobic 233 -, tubular 238, 241 metabolism 387, 394 -, cellular 383 -, interactive 381 metal - alloy 210 - chelate 154, 315 - - chromatography 113 - ions 120f., 267 - oxides 241 - screen 204 methanol 67, 70, 223, 272 method, combinatorial 123 method of lines (MOL) 350 methotrexate 107 methyl isobutyl ketone (MIBK) 291 (2S)-(-)-methyl-l-butanol 288 microbeads 192 microbiological contaminations 466
Index Micrococcus denitrificans 286 microdroplets 268 f. microfiltration 199, 229f., 246, 250, 305 - membranes 200 microflora 259 microorganisms 501, 508 microwave energy 480 microwave heating 481 midstream 486 milk 95 Mimetic blue 1 6XL 118 mineral oxides 158 miscibility 277 mobile phase 51 model -, complicated 438 -, numerical 185 -, small-scale 502, 504 - viruses 472f., 486 modeling 96, 164, 411 -, classical techniques 417 -, computer resources 415 -, data situation 415 - error 412, 435 -, mathematical 448 - methods 454 -, practical 412 - procedure 414 - project 414 -, software tools 445 - tasks 454 molasses 29 molecular characteristics 45 molecular weight 45 f. Monod model 419 monoglycerides 245 monosodium - aspartate 81 - citrate 291 - glutamate 81 monovalent ions 257 moving bed 14 - chromatography 345 -, modified 6 MSFB see magnetically stabilized fluidized bed Mugele-Evans distribution 285 multicomponent mixture 17 multidimensional state spaces 447 multilayer 3 11 multipoint attachment 100 multivariate analysis 71 MW 404 275
521
mycelial pellets 217 mycoplasma 466, 483 mycoprotein 387 myeloma cell culture 223 myoglobin 28, 69, 182 Myxococcus xanthus 153 Nacolyte 7105 100 NAD+ 116 nanofiltration 229, 232, 257, 488, 491, 494 - membrane 247 P-naphthylamine 103 negative function 187 negative step 122 Nernst-Planck equation 338 network -, porous 194 -, static 392 neural computers 363, 405 neural network 434 -, adaptive 389 - controllers 441 -, dynamic 380 neural simulations 364 Neuralware 448 Neuralworks 448 neurodes 365 neurofuzzy network 422, 426 neuron-fuzzy control 399 neurons 363 -, artificial 366 -, non-linear 363 NeuroShell 448 NeuroSolutions 448 neutralization 282 Newtonian fluid 165 nicotinamide 113 nitrate 276, 290 nitric acid 238 nitrogen balance 417 nitrophenol 269, 277 p-nitrophenyl heptyl ether 288 o-nitrophenyl octyl ether 288 noise filter 429 non-biological reactors 373 non-idealities 331 non-linear relationships 434 non-Newtonian 396 non-specific binding 3 11 nonylphenol 29 1 nopol 288 nuclease 217, 221 f. nucleic acids 72, 217, 498, 503, 507
528
Index
nucleocapsid 470, 477 nucleophiles 114 nucleosides 109 nucleotides 109, 154 numerical computations 429 Nusslet number 167 nylon 306 occlusion 277 n-octane 290, 296 OCtyl-P-D-glyCOpyran 298 oleic acid 109 oligocarbamates 123 oligomyxins A, B, C 106 oligonucleotides 109, 123, 154 oligosaccharides 247 on-line control 414 opacity 209 open loop control 414 operating conditions 327 operating line 94 operation regeneration 47 1 operational qualification (OQ) 140, 471, 506 optical density 376, 396 optimization 214, 438 -, genetic 367 -, linear 450 -, non-linear 450 -, numerical 450 -, off-line 414 - procedures 411 -, quadratic 450 -, stationary 345 orthogonal collocation 354 orthogonal polynominals 354 osmotic shock 221 output layers 371 ovalbumin 68, 166 overflow velocity 381 overloading 500 oxidation 250 oxidative cleavage 3 16 oxirane 310, 312 oxygen, dissolved (DO) 380, 396, 431 oxygen transfer rate (OTR) 383 oxygen uptake rate (OUR) 379, 383, 387, 395, 429, 440 packed mode 133 packing 131, 148 - densities 319f. - efficiency 139 - manifold 148
- material 58, 157 - methods 136 - port plugs 150
Pall Ultipor 491 palmitic acid 99 papain 251 paraffin 268 Paranox-100 269, 286 partial differential and algebraic equations (PDAE) 352 particle -, chromatographic 332 - concentration 319 - diameter 96 - diffusion 185 - dispersion 206 - distribution 205 - highly porous 42 -, monodispersed 205 -, - spherical 117 - radius 183 - shape 117 - size 52, 58, 117, 199, 319, 486 - - distribution 499 - surface 332, 356 particulates 209, 216 partition 273 - coefficient 122, 283 Parvovirus 488, 494 pasteurization 251, 259 path length 337 pattern recognition 396, 445 -, dynamic 392 peak - separation 59 - shape 63, 510 - tracking 47 - width 58 Peclet number 159, 285, 206, 317, 351 pectinase 242, 246 penicillin 379, 437 - acylase 243 - fermentation 394 penicillin G 275, 282, 285, 292f., 299, 373, 394 - acylase 277 Penicillium crysogenum 373 pentaerythritol 100 pentosan polysulfate 100 peptides 99, 105 -, terminal 316 -, therapeutic 502 peptone 290
Index peracetic acid 239 perceptron 370 perfectly classified bed (PCB) 205 perfluorocarbon, hydrophilized 202 performance qualification 47 1 performance relations 4 11 perfusion chromatography 42, 104, 158 perfusion media 62 perfusive particles 43 periodic steady state 6 periplasm, E. coli 107 periplasmic space 220 f. permeability 166, 294 -, hydraulic 169 permeate 231 permeation 475 permeator shell 273 pervaporation 233 phenol 269, 272, 280, 285f., 311 phenolic complex 281 phenyl agarose 122 Phenyl Sepharose 73, 80, 224 L-phenylalanine 277, 280, 286, 299 phenylalanine 28, 286 - hydrochloride 272, 276, 288 phenylboronic acid 121 phenylethyl alcohol 3 1 phosphate 269, 290 - buffer 151 phospholipids 222 R-phycoerythrin 70 physical stability 158 Pichia pastoris 224 pig liver esterase 277 pilot plant 126, 445 pilot scale 212, 505 plant cells 432 plant extracts 95 plaque formation 475 plasmid 364 - instability 385, 404 - pBR Eco gap 392 - retaining ability 387 - stability 401, 424 -, temperature-sensitive 407 plasminogen activator 486 Plasmodium falciparum 106 Plasmodium vivax 106 plate heigth 43 plate number 207 f. plug-flow 203, 318 Poisson distribution 476, 484 f. polarity 45, 280
529
polarization 490 polio 488, 492, 494 polishing 49, 70f., 74 polyacrolein 3 10 polyacrylamide gels 157, 164 f. polyacrylic acid (PAA) 108 polyacrylonitrile 237, 272 polyamide 308, 312, 323 polycarbonate 237 polydextran beads 150 polydispersity 285 polyethersulfone (PES) 306, 308f., 312, 314 polyethylene 306, 308 f. polyethylene glycol 100, 3 13 polyethylene oxide (PEO) 3 14 polyethyleneimine (PEI) 100, 310 polymeric matrix 66, 69, 305 polymers -, amphiphilic 102 -, dendritic 100 -, macroporous 119 -, synthetic 241 -, water-soluble 100 polymethacrylates 101 polymyxin B sulfate 106 polynomial approximation 438 polynomial coefficients 438 polypepetides 123 polyphenols 260 polypropylene 237 -, hydrophobic 293 polysaccharide 114, 120, 158, 247 polystyrene 170 - divinyl benzene 69, 175, 200f. polysulfone 237, 306 polytetrafluoroethylene (PTFE) 237 polyvinylidene fluoride 237, 299 polyvinylpyrrolidone 182, 260 Pontryagin’s maximum principle 438 porcine pancreas 28 pore - diameters 316 - filling 173 - glass 172 -, interconnected 158 - liquid 337 - radius 337 - size 161, 172, 233, 319, 488, 490 -, small 194 - volume 172 - wall 161 POROS 175, 307 POROS Perfusion Chromatography 43
porosity
18, 192
-, bed 331 -, column 330 -, external 330 porous glass 241 porous ribbons 272 porous sheets 288 porous solid supports 286 porous spherical particles 33 1 porous wall 272 positive step 122 post-production cell bank 471 potassium 277 - dihydrogen phosphate 299 - phosphate 69, 81 practical constraints 414 prazinquatel 3 1 precipitate 84, 219 -, cryogenic 153 precipitation 69, 73, 91, 101, 220, 290 prefiltering 377, 491 pressure 63 - difference 321 -, differencial 262 - drop 23, 52, 60, 185, 208, 321, 490 - -, low 145 -, osmotic 232, 262 principal component regression (PCR) 427 principal component analysis (PCA) 427 process analysis 434 - changes 509f. -, chromatographic 3, 125 - conditions 129 -, continuous 36f. - control 131, 502 - - systems 131 f. - -, strategy 127 - data 427, 454, 471 - data exploration 434 description 471 design 411, 498 - development 125, 131 -, discontinuous 36 disturbances 426 -, downscaled model 490 - dynamics 418, 433f., 439 - economics 411, 499 - engineering 125 -, enzymatic 244 - experts 443 - faults 395, 424, 432 - fluid 476
-
-
- hygiene 138 - improvements 454 - intermediates 498
-
kinetics
419
- knowledge 413, 459 - management 131
-, mathematical representation 41 3 -, mathematical model 441 - modeling 413, 434 - model 411, 414 - -, non-mechanistic 420
-
optimization
433, 439
- parameters 141, 499 - -, reliability 474 - performance 411 - - qualification 509 - qualification (PQ) 140, 471 - quality assurance 424 - quantities 433
-, rate-determining steps 147 requirements 499 signals 427 strategy 128, 136 stream 128, 231 throughput 49 validation 471, 507 variables 435, 457 - verification 214 processing - agent 499 -, aseptic 239 - tools 498 Procion Blue 3 15 Procion Red H-E7B 223 product - activity 379 -, biological 9, 267 - comparability 510 - concentration 500, 502 - cost 142 -, diagnostic 497 f. -, efficaceous 510 -, high-value 417 - inhibition 282 - purity 499, 508 - quality 142, 229, 412, 494, 499 - recovery 250, 283 - retention 240 -, therapeutic 497 f. - throughput 510 production - cells 467 - costs 229 -
Index - rates 99 - system 445 - target 142
productivity 412 -, dimensionless 34 -, increased 131 -, low 157 programming tools 41 3 proinsulin 106 properties, diffusive 177 prophet 448 p-propiolactone 474 prorenin 151 prostaglandins 280 protease 72, 121 - deactivation 224 - inhibitor 223 protein 105, 280, 481, 494 - activity 402 - adsorption 171 - - capacity 208 - aggregates 117, 192 -, anticoagulant 222 - binding capacities 311 - biosynthesis 471 - breakthrough 208 -, calcium-dependent, phospholipidbinding 222 - capacity 213 - capture 133 -, complex 478 - concentration 504 - denaturation 67 - diffusivity 166 - drug 471 -, eukariotic 153 - expression 401 - feedstock 213 -, functional 256 -, globular 164 -, glycosylated 471 -, heterogeneous composition 485 -, heterologous 223 -, host cell 505 - load 175 -, longitudinal 488 -, molecular weight 478 -, pharmaceutical 465, 481 - polyphenol complexes 251 - precipitation 189 - production 433 - purification 25, 28 -, recombinant 199, 316
531
-, refolded 154 - recovery 202, 295 - solutes 67 -, spherical 490 - surface 114
-, thrombolytic
107 216 - unfolding 241 - uptake 167 protein A 121, 314, 486 protein G 486 proteolysis 151 proteolytic digestion 501 pseudoaffinity 3 15 Pseudornonas aureginosa exotoxin A pump 509 - packing 136 - systems 506 purification 49, 26 1, 498 - factor 59 - processes 141 - steps 510 purity 59, 94, 501, 510 PV-WAVE 447 PVC composite 306 pyrimidine nucleosides 152 pyrogens 117 pyrophosphate 113 -, unbounded
221
quality - assurance
424, 446, 506 attributes 141 policy 506 - of separation 98 - reviews 509 quartz, crystalline 210 quaternary ammonium salts 275, 280
-
radial basis function 424 radial basis network 372 radial flow chromatography 132, 135f., 145 radial particle coordinate 356 radial variation 283 raffinate 9 f., 16, 27 1 random amplitude 378 random search methods 438 rate - controlling 280, 299 - expressions 419, 434 - limiting factor 333 - limiting step 289 raw materials 498 re-equilibration 500 f.
532
Index
reaction front, spherical 284 reaction zone 286 reactive extraction 292 reactor, biochemical 234 reactor failure 373 receiving phase 267, 290 receptor cells 477 recombinant - albumin 156 - anti-HIV Fab fragment 220 - bacteria 364 - blood factors 90 - cells 393 - CHO cells 433 - DNA 125 - organism 201 - protein 95, 220, 383f., 477, 492 - strains 392 recovery 45, 59, 63, 81, 94, 499, 510 - levels 151 - techniques 199 - yields 99 recycling 117, 152 - pump 12 redox potential 376 reductance balance 417 reduction factor 476 f regeneration 148, 477, 500 regression 457 regulators 502 reliability 77, 470 remote operation 13 1 renaturation 90, 153 replication 397 reprocessing 500 reproducibility 60, 63, 75, 224 repulsion forces 194 repulsive interactions I83 Residence Distribution Times 320 residence time 167, 207, 321 - distribution (RTD) 206 -, variance 207 resolution 45, 52, 58f. - degree 139 resonance, adaptive 373 response times 416 resuspension 212 retardation 8 1 retentate 23 1, 256, 494 retention 68, 81 - factors 99 -, isocratic 54ff.
retrovirus 465 493 reusability 288 revalidation 509 f. reverse diffusion 274 reverse osmosis 229, 232, 250 reversed micelles 278, 280, 295, 300 -, extractions 297 reversed-phase 53, 146 - chromatography 69 - separation 155 Reynolds number 205, 351 rheology 381 Rhizopus nigricans 246 ribonuclease 68 f., 182, 296 ff. ribonuclease A 182, 315 rigidity 117 risk analysis 471 risk assessment 498 RNA molecules 123 robotics 443 robustness 60, 83, 117, 210, 363, 394 routine operations 131 rules-of-thumb 4 13 rupture 276 - HIV
SlOON 285 Saccharomyces cerevisiae 399, 401 f., 419, 429 safety 126, 145, 469, 495 salt concentration 68, 100, 504 salt gradient 122 salting-in 66 salting-out 65, 75 -, ammonium sulfate 65, 68 -, magnesium chloride 66 sample - application 504 - concentration 132 - flow 146 - load 58, 83, 500 - matrix 41, 50 - preparation 49 - processing 60 sand 241 sanitary conditions 148 sanitary design 204 sanitization 137, 202, 471, 499f., 504 sanitizing agent 508 sarcosyl 219 Sarex process 26 saturation capacity 17, 348 scalability 60, 74, 269, 271
533
Index scale-down 126 - studies 510 scale-up 61 f., 84, 150, 208, 224, 329 - factor 214 -, flow distribution 62 - guidelines 506 -, linear velocity 61 -, mobile phase composition 61 -, optimization 62 -, pressure drop 62 scaled isotachic state 93 Schiff bases 314 Schmidt number 351 Scicos (scilab’s connected object simulator) 450 scouting 2 12 f. screening 50, 79 sedimentation 381 sedimented bed 201 segmented bed 201 segregation 285 selectivity 56, 58, 63, 70, 91, 157, 267, 276, 279, 329, 412, 499 - factor 177 -, improved 271 -, limited 271 -, low 295 self-association 155 self-diffusion 288 self-organizing networks 373 self-sharpening 93 semi-discretization 353 sensitivity 267, 271 - analysis 142 sensor faults 424, 429 - detection 427 sensors 383 separation 26 1 -, adsorptive-type 146 -, chiral 280, 288 -, chromatographic 269 - efficiency 98, 157, 276 - factors 97 -, ionic molecules 25, 29 -, isocratic 135, 146 - module 240 -, multi-step 60 -, physical 273 - power 41, 329 -, preparative 49 -, reproducible 56 -, spatial 4 -, sugars 25f.
Sepharose 71, 147, 3 16 f. Sepharose 4B 114, 150 Sepharose CL-4B 147, 150, 153 Sepharose ConA 15 1 Sepharose Fast Flow 150 Sepharose Q 152, 154 Sepharose QAE 15 1 Sepharose S 154, 504 serum albumin 469 serum-free media 468 SERVA 275 set point profiles 412 shallow bed 185 Sherwood number 351 shrinking core model 284 signal - flow 366 - noise 429 - processing 449 silica 150, 193 -, amorphous 202 -, doped 193 - matrix 170 -, porous 172 silver nitrate 272 simple batch 234 simulated annealing 367, 438 simulated environments 45 1 simulated moving bed 3, 6, 14 - chromatography 345 -, numerical simulation 23 simulation 96, 343, 350, 371, 391 -, experiments 352 -, neural 395 -, numerical 23, 329, 352 - tools 357 simulators 329 -, dynamic 353, 357 size - distribution 206 - exclusion 17 - - chromatography 91, 146 - -process 27 -, molecular 86 -, optimal 270 - ranges 206 sludge 381 slurry 381 - packing 148 SMB see simulated moving bed sodium - acetate 216 - bis(2-ethylphenyl) sulfosuccinate
295
534
Index
- bisulfite 239 - chloride 153, 182, 218, 290, 503 - citrate 211
-
hydroxide
84, 210, 291, 299, 307, 503 2 19 - phosphate 219 - sulfate 81, 280 software - control 53 - tools 445 - validation 453 solubility 45 f., 56 solubilization 219 solute - affinity 348 - competition 329 - concentration 188 - decrease 188 - molecules 275 solution, optimal 47 solvent - concentration 46 -, organic 82 sorbent sites 329 sorbitan monooleate 278 sorbitol 247 f. sorption capacity 158, 161, 175 ff. sorption kinetics 160, 177 SOURCE 69 soy trypsin inhibitor 68 spacer 312 -, 1,6-diaminohexane 315 -, polyethyleneimine 3 15 spacer arm 510 -, hydrophilic 3 13 -, hydroxyl group 510 Span-80 269, 285, 287, 289, 293 specific test methods 502 spent resin 148 Spherodex 158, 170 Spherosil XOB 27 spiking experiments 475 squashing function 372 stability 45 f., 76 stabilization 72 stabilizers 90 stainless steel 241 standard deviation 369 standard operating procedures 469, 471, 501 standard test procedures 471 Staphylococcus aureus 314 Start-up 500 - phase 392 - N-lauroylsarcosinate
state estimation 414 ff. state variables 426 stationary phase 398 steady state 6 - design 381 steaming-in-place (SIP) 129, 239 stereoselectivity 280 steric factor 98 steric hindrance 165 steric mass action ion-exchange formalism (SMA) 350 steric mass action model 98 sterile measurement 379 sterility 138, 474 sterilization 117, 120, 250 steroid hormones 280 stirrer speed 431 stoichiometric displacement model (SDM) 350 stoichiometry 417 Stokes’ law 192, 204 Stokes’ radius 488 storage 500, 504 - conditions 214 stratification 203 STREAMLINE 202, 204f., 210f., 214 - Chelating 210f. - DEAE 210f., 216, 218, 222f. - Heparin 210f. - ion exchangers 215 - rProtein A 210f., 215f., 218f., 223 STREAMLINE SP 80, 210f., 216, 218ff., 224 STREAMLINE 50 219f. streptokinase 299 Streptomyces hygroscopicus 200 streptomycin 200 strip solution 272 stripping 81 - liquid 291 - phase 291 structural testing 506 styrene gels 307 subnetwork 364 substrate 113 - conversion 417 - feed 404 - - rate 395 sugars 232, 260 - separation 25f. sulfate 290 sulfonate groups 122, 188 sulfonic acid halogenalkanes 307
Index sulfuric acid 290 supercritical fluid 12 supernatant 224, 477, 487 support -, porous 271 - matrix 115 supported liquid membranes (SLM) 276, 288, 299 surface - activity 277 - area 172, 283 - chemistry 52, 119 - diffusion 332, 337 - distribution 113 - filtration 231 -, hydrophilic 119 -, hydroxilic 121 -, hydroxyl groups 119 -, non-charged 119 - tension 66, 268 surface-active agent see surfactant surfactant 267 f., 270f., 277 -, anionic 295 -, cationic 287, 296 -, hydrated 278 -, hydrophilic part 278 suspension growth 466 swabbing 508 swelling 276, 284 synthetic media 417 system - configuration 500, 509 - design 502 - factors 214 - flexibility 131 - integration 131 - lag 506 -, microbial 396 - performance 277 systematic development 42, 44 -, critical parameters 42 tangential flow 230 490 tangential function 368 tannins 260 target protein 73, 114, 209, 218, 220 -, enzyme 220 - expression 220 -, recombinant 220 temperature - control 450 -, critical 364
- filtration
535
tentacular gels 158 ternary complex 121 tertiary amines 280 test data 393 tetraethyleneglycol-diacrylate 309 thermodynamics 157 thin-layer displacement chromatography 103 -, forced flow (FF-TLCD) 103 thiol exchange 113 threshold system 505 threshold value 396 throughpores 43 throughput 60, 73, 329, 498 -, limited 157 -, low 145 thyroglobulin 177 thyroxine 114 time-dependence 75 time-domain vectors 367 titania 193, 242 -, porous 172 TLCD see thin-layer displacement chromatography TMB see true moving bed p-toluene sulfonic acid 275, 292 topology 364 tortuosity factor 168, 337 tosyl chloride 312 total residence time 207 Toyopearl 307 tracer 206 tracer stimulus method 206 transcriptase 47 1 transfer functions 380 -, concept 441 transferrin 68, 70,95, 107, 180, 182, 3 15,469 transformation 446 -, affine 367 -, continuous 369 -, linear 417, 427 -, nonlinear 369 - function 368, 407 transmembrane 305, 3 19 - pressure 490 transmissible agents 469 transport 269 -, axial 340 -, carrier-mediated 269 -, diffusive 282 -, facilitated 274 -, physical 269 - resistance 280 - selectivity 287
536
Index
triazine dyes 121 f., 223 tricapryl quaternary ammonium salt 287 n-tridecanol 29 1 trifluoroacetic acid 70 trifluoromethanesulfonyl ester 3 12 triglycerides 245 tri-n-octyl amine 291 trioctylmethyl ammonium chloride 275 f., 292, 297 trisacryl 1.50 Triton X-100 269 troubleshooting 45, 63, 508 true moving bed 4f. trypsin 28, 297 tryptophan 296, 316 L-tryptophyl glycine 296 tuning procedures 422 turbidity 405 Tween-80 269 two-phase extraction 299 ultracentrifugation 154, 267, 269 ultrafiltration 62, 224, 229, 231, 244, 246, 470, 488, 493 ultrasonic waves 209 unfolding 76 unit operations 133, 199 unpacking 148 upstream 74, 230, 494 - processing 141 upwards flow 218, 220 uranium 269 urea 66, 76, 82 uridine 104 uridine diphosphoglucuronic acid (UDPGA) 280 f. uridine diphosphoglucuronyl transferase (UDPGT) 274, 281 uridine phosphorylase 152 urine utility costs 142 UV profiles 502 vaccination 461 vaccines 156 validation 110, 125, 127, 131, 139, 415, 421, 435, 465, 470f., 489 -, cleaning 497, 501, 508 - issues 497 -, process 140, 497, 507 - program 140 -, sanitization 508 -, software 497
-, state 509 - strategy 510 -,virus 500 valine I 1 3, 182 values 509 van der Waals interactions 280 Van-Deemter equation 18 variability 507 variables -, algebraic 358 -, differential 358 -, manipulated 398 vector instability 404 velocity -, high 187 - linear 136, 175, 180, 187 -, radial 96 -, relative 347 -, superficial 285 viability -, economic 142, 147, 299 vinylmonomers 3 10 vinylpolymers 307 viral - activity 465 - clearance 469, 497, 504 - -, regulatory acceptability 470 - -, validation 470 - components 470 - infectivity 470 - safety 465, 503 virion 470 virus 72, 502 -, adventitious 465 f. - capsid 485 - clearance 469, 494 - - effectivity 475 - contamination 465 - detection 468 - distribution 475 - features 472, 486 - inactivation 469 f., 477 f., 494, 498, 503 - like particles 465, 471 -, model 472f. - removal 469 f., 478 f., 504, 510 -, specific 472 - structures 477 - titer 474 -, unknown 472 viscosity 68, 161, 199, 215, 276, 321 -, dynamic 192 -, high 82, 84
Index
void - fraction 330 - structure 171 - volume 139, 181, 318, 330 volume 127 -, differential 342 - fraction 330 -, internal 118 -, solid 330 volumetric accuracy 506 volumetric flow 334 washing
147, 214, 216, 475 214f. wastage 145 waste stream 254 wastewater 267 wetted components 509 Whatman DE52 153 whey proteins 100, 245 working cell bank (WCB) 467 workstations 446 - volumes
X-ray crystallography xylenes 291
47
yeast 199, 374 alcohol dehydrogenase fermentation 220 homogenate 224 methylotrophic 223 yield 59 - factor, constant 402 Yoshida’s model 167
-,
537
116
zeolites 26 zinc 269 zirconia 193 -, porous 172 zirconium hydrous oxide-polyacrylate 247 zone broadening 344 zwitterion 287 Zymomunas mobilis 376
242,