Retention and selectivity in liquid chromatography: prediction, standardisation, and phase comparisons

JOURNAL OF CHROMATOGRAPHYLIBRARY- volume 57

retention and selectivity in liquid chromatography prediction, standardisation and phase comparisons

This Page Intentionally Left Blank

JOURNAL OF CHROMATOGRAPHY LIBRARY- volume 57

retention and selectivity in liquid chromatography prediction, standardisation and phase comparisons edited by

Roger M. Smith Department of Chemistry, Loughborough University of Technology,Loughborough, Leicestershire LEI 7 3TU, UK

ELSEVIER Amsterdam

-Lausanne-New York -Oxford -Shannon

-Tokyo

1995

ELSEVIER SCIENCE B.V. Sara Burgerhartstraat 25 PO. Box 211,1000 AE Amsterdam, The Netherlands

ISBN 0-444-81539-2

0 1995 Elsevier Science B.V. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior written permission of the publisher, Elsevier Science B.V., Copyright & Permissions Department, P.O. Box 521,1000 AM Amsterdam, The Netherlands. Special regulations for readers in the U.S.A. - This publication has been registered with the Copyright Clearance Center Inc. (CCC), Salem, Massachusetts. Information can be obtained from the CCC about conditions under which photocopies of parts of this publication may be made in the U.S.A. All other copyright questions, including photocopying outside of the U.S.A., should be referred to the Publisher. No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein.

This book is printed on acid-free paper. Printed in The Netherlands

V

Contents .

Retention prediction based on molecular structure ....................... R.M. Smith Introduction....................................................................................................... Structure and retention ...................................................................................... 1.2.1 Chromatographic functional group contributions .................................. 1.2.2 Related prediction studies...................................................................... Functional group effects on retention indices .................................................... 1.3.1 Retention prediction based on n-alkanes and n-alkylbenzenes.............. 1.3.2 Retention prediction based on alkan-2-ones.......................................... Prediction of retention indices based on alkyl aryl ketones ............................... 1.4.1 Monofunctional compounds .................................................................. 1.4.1.1 Aromatic functional groups ..................................................... 1.4.1.2 Aliphatic functional groups ..................................................... 1.4.1.3 Relationship between substituent indices and octanol-water partition substituent increments............................................... 1.4.2 Polyfunctional compounds .................................................................... 1.4.2.1 General prediction model ........................................................ 1.4.2.2 Meta and para groups ............................................................. 1.4.2.3 Ortho-substituents ................................................................... 1.4.3 CRIPES and expert systems .................................................................. 1.4.4 Comparisons with published retention values........................................ Other retention index prediction studies............................................................ 1.5.1 Retention prediction based on polynuclear aromatic hydrocarbons ...... Conclusion......................................................................................................... Acknowledgements............................................................................................ Appendix 1.1: Coefficients of regression equations for the effect of eluent on parent, aromatic and aliphatic substituent indices ........................................ References .........................................................................................................

Chapter 1 1.1 1.2

1.3

1.4

1.5 1.6 1.7 1.8

1.9

.

Retention prediction of pharmaceutical compounds...................... K. Valk6 Introduction ....................................................................................................... Definition and determination of retention.......................................................... Dependence of retention on the column and mobile phase composition ........... Correlation of retention parameters to the molecular parameters obtained by molecular modelling .......................................................................................... Retention prediction based on topological matrix and information the0ry ........ . . Retention prediction based on the hydrophobicity of drugs ..............................

1

1 2 3 4 5 6 8 11 13 13 15 19 21 23 29 32 33 38 40 41 43 43 44 45

Chapter 2

47

2.1 2.2 2.3 2.4

47 48 50

2.5 2.6

53 60 62

VI

Contents

Retention prediction based on empirical increment values ................................ Retention prediction based on experimental retention values. thermodynamic considerations with multiparameter approaches ..................................... 2.9 Applications of retention predictions of pharmaceutical compounds ................ 2.10 Acknowledgements............................................................................................ 2.11 References ......................................................................................................... 2.7 2.8

68 76 87 90 90

.

Retention index scales used in high-performance liquid chromatography ................................................................................. R.M. Smith 3.1 Introduction ....................................................................................................... 3.1.1 Relative retention times ......................................................................... 3.1.2 Internal and external standards .............................................................. 3.1.3 Retention indices ................................................................................... 3.2 Retention index scales in chromatography ........................................................ 3.2.1 Gas chromatography.............................................................................. 3.2.2 Supercritical fluid chromatography ....................................................... 3.2.3 Liquid chromatography ......................................................................... 3.2.4 Micellar electrokinetic chromatography ................................................ 3.3 Retention index scales in high-performance liquid chromatography................. 3.3.1 n-Alkanes ............................................................................................... 3.3.2 n-Alkylbenzenes .................................................................................... 3.3.3 Alkan-2-ones ......................................................................................... 3.3.4 Alkyl aryl ketones. ................................................................................. 3.3.5. 1-Nitroalkanes....................................................................................... 3.3.6 Polynuclear aromatic hydrocarbons....................................................... 3.3.7 Miscellaneous retention index scales..................................................... 3.3.7.1 Phenolic esters......................................................................... 3.3.7.2 Aliphatic esters ........................................................................ 3.3.7.3 Other retention index scales .................................................... 3.3.8 Comparisons between retention index scales......................................... 3.4 Applications of retention index scales ............................................................... 3.4.1 Reproducibility and transferability of retention indices......................... 3.4.2 Identification ......................................................................................... 3.4.3 Characterization of separation systems .................................................. 3.4.3.1. Column hold-up volume .......................................................... 3.4.3.2 Stationary phase characterization............................................ 3.4.3.3 Mobile phase characteristics ................................................... 3.4.4 Lipophilicity and biological activity ...................................................... 3.4.5 Structure-retentionrelationships............................................................ 3.5 Conclusions ....................................................................................................... 3.6 References ......................................................................................................... Chapter 3

93 93 94 95 97 98 99 100 103 105 107 107 109 109 111 115 115 116 116 117 118 118 122 122 125 125 126 127 133 135 137 139 140

Contents

VII

Chapter 4. 4.1 4.2 4.3

4.4

4.5

4.6 4.7 4.8 4.9

Application of retention indices for identification in high performance liquid chromatography .............................................. R.M. Smith Introduction ....................................................................................................... Advantages and problems.................................................................................. Pharmaceuticals and toxicological drug samples............................................... 4.3.1 Toxicological drug analysis................................................................... 4.3.1.1 Studies based on the alkan-2-ones........................................... 4.3.1.2 Studies based on the alkyl aryl ketones ................................... 4.3.1.3 Studies based on the 1-nitroalkanes ........................................ 4.3.2 Drug metabolites ................................................................................... Natural products ................................................................................................ 4.4.1 Fungal metabolites................................................................................. 4.4.1.1 Mycotoxins.............................................................................. 4.4.1.2 Other hngal metabolites ......................................................... 4.4.2 Plant products ........................................................................................ 4.4.2.1 Spices and flavour components............................................... 4.4.2.2 Plant toxins .............................................................................. 4.4.2.3 Gliadins................................................................................... 4.4.3 Lichen constituents ................................................................................ 4.4.4 Other natural products ........................................................................... Environmental samples...................................................................................... 4.5.1 Chlorinated compounds......................................................................... 4.5.2 PAH and aromatic hydrocarbons........................................................... Miscellaneous samples ...................................................................................... Conclusions ....................................................................................................... Appendix 1: Reported retention indices in HPLC ............................................. References .........................................................................................................

145 145 145 147 147 148 149 156 156 157 157 158 159 160 161 162 163 163 164 164 164 165 165 165 166 167

Chapter 5. 5.1 5.2

5.3

Application of nitroalkanes and secondary retention index standards for the identification of drugs ......................................... M. Bogusz Introduction....................................................................................................... The use of HPLC as a standardized identification method in toxicology .......... 5.2.1 Standardization of retention using straight phase silica......................... 5.2.2 Standardization of retention using reversed phase silica ....................... 5.2.2.1. The concept of secondary standards for retention index scale 5.2.2.2 Assessment of identification potentials of a HPLC/RI system 5.2.2.3 Influence of a biological matrix on the identification potential of a HPLC/RI system ........................................................ 5.2.3 Standardization of detection for toxicological screening procedures .... References .........................................................................................................

171 171 173 173 175 183 199 203 204 205

VIII

Contents

Chapter 6.

209

6.1 6.2 6.3 6.4 6.5

209 210 213 215 217 217 219 221 221 223 228 230 230

6.6

6.7 6.8

Identification using retention indices in gradient HPLC ............... P . Kuronen Introduction ....................................................................................................... Problems ofreversed-phase columns ................................................................ Selection of a retention index standard.............................................................. Principles of gradient elution............................................................................. Chromatographic behaviour of retention index standards ................................. 6.5.1 Isocratic conditions................................................................................ 6.5.2 Gradient elution ..................................................................................... Retention indices in qualitative identification ................................................... 6.6.1 Calculation of retention indices ............................................................. 6.6.2 Reproducibility of gradient retention indices ........................................ 6.6.3 Confirmation of identification ............................................................... Conclusions ....................................................................................................... References .........................................................................................................

Chapter 7. 7.1 7.2

7.3 7.4 7.5

7.6 7.7 7.8

Characterization of retention and selectivity in reversed-phase LC using interaction indices ............................................................. P . Jandera Introduction ....................................................................................................... Interaction indices as the descriptors of retention ............................................. 7.2.1 Retention in binary mobile phases ......................................................... 7.2.2 Retention in ternary mobile phases ........................................................ Calibration of the scale of interaction indices ................................................... Prediction of the retention under changing mobile phase composition using interaction indices.............................................................................................. Interaction indices and the selectivity of separation .......................................... 7.5.1 Non-homologous compounds ................................................................ 7.5.2 Homologous and oligomeric series........................................................ Conclusions ....................................................................................................... Glossary ofthe terms ......................................................................................... References .........................................................................................................

.

Lipophilic and polar indices ............................................................. P . Jandera 8.1 Introduction ....................................................................................................... 8.2 Retention in homologous series as the basis of lipophilic and polar indices ..... 8.3 Molecular structure and lipophilic and polar indices......................................... 8.3.1 Anc and Aq indices as the descriptors of the lipophilicity and polarity of solutes .......................................................................................... 8.3.2 Structural contributions to lipophilic and polar indices ......................... 8.4 Prediction of retention using lipophilic and polar indices ................................. 8.4.1 Selection of the reference calibration homologous series...................... 8.4.2 Precision of the predicted retention data ............................................... 8.5 Characterization of selectivity using lipophilic and polar indices .....................

Chapter 8

235 235 236 238 243 247 250 253 253 255 263 264 266 269 269 269 274 274 276 279 279 279 283

Contents

8.6 8.7 8.8

8.5.1 Binary mobile phases............................................................................. 8.5.2 Ternary mobile phases ........................................................................... 8.5.3 Gradient elution ..................................................................................... Conclusions ....................................................................................................... Glossary of the terms ......................................................................................... References .........................................................................................................

Chapter 9.

Solvent selectivity .............................................................................. S.D. West ............................................................................................ 9.1 Introduction ....................................................................................................... 9.2 Experimental ..................................................................................................... 9.2.1 Chemicals and reagents ......................................................................... 9.2.2 Instrumentation...................................................................................... 9.2.3 Determination of retention data ............................................................. 9.2.4 Selection of solvents and solutes ........................................................... 9.3 Results and discussion 9.3.1 Prediction of retention and resolution with steroids .............................. 9.3.1.1 Retention indices as a function of volume fraction of strong solvent ..................................................................................... 9.3.1.2 Prediction of retention indices for steroids.............................. 9.3.1.3 Prediction of resolution of steroid mixtures ............................ 9.3.1.4 Optimization of resolution of steroid mixtures........................ 9.3.1.5 Resolution and the solvent selectivity triangle concept ........... 9.3.2 Retention and selectivity studies with benzene derivatives ................... 9.3.2.1 Retention index variation with solvent selectivity................... 9.3.2.2 Resolution and the solvent selectivity triangle ........................ 9.3.2.3 Prediction of resolution ........................................................... 9.3.2.4 HPLC resolution as a function of retention index differences. 9.3.2.5 Preadjustment of retention indices for prediction of resolution .......................................................................................... 9.3.3 Reasons for failure of the solvent selectivity triangle ............................ 9.3.4 Extension of theory to gas chromatography........................................... 9.3.4.1 Prediction of resolution ........................................................... 9.3.4.2 McReynolds constants and resolution ..................................... 9.3.4.3 Application of the selectivity triangle to the characterization of GC stationary phase selectivity ........................................... 9.3.5 Characterization of RP-HPLC selectivity with adjusted retention indices....................................................................................................... 9.3.5.1 Calculation of adjusted retention indices................................. 9.3.5.2 Probes for characterization of RP-HPLC solvent selectivity... 9.3.5.3 Quantitative prediction of resolution with any RP solvent ...... 9.4 Conclusions ....................................................................................................... 9.5 Acknowledgments ............................................................................................. 9.6 References .........................................................................................................

IX

283 287 290 292 292 294 297 297 297 299 299 299 299 300 301 301 302 303 307 309 311 311 311 314 3 14 318 320 323 323 324 326 327 327 329 331 334 335 335

X

Contents

.

Chapter 10

Retention and selectivity for polycyclic aromatic hydrocarbons in reversed-phase liquid chromatography ...................................... L.C. Sander and S.A. Wise 10.1 Introduction ....................................................................................................... 10.2 Stationary phase characteristics affecting selectivity in RPLC .......................... 10.2.1 Phase type .............................................................................................. 10.2.2 Isomer separations................................................................................. 10.2.3 Assessing column shape selectivity ....................................................... 10.2.4 Pore size effects ..................................................................................... 10.2.5 Bonding density..................................................................................... 10.2.6 Bonded phase length.............................................................................. 10.2.7 Mobile phase composition..................................................................... 10.2.8 Temperature........................................................................................... 10.3 Retention indexes .............................................................................................. 10.3.1 Retention index data .............................................................................. 10.3.2 Length-to-breadth ratio .......................................................................... 10.3.3 Planar and non-planar PAHs ................................................................. 10.3.4 Methyl-substituted PAHs....................................................................... 10.4 Summary ............................................................................................................ 10.5 References .........................................................................................................

11.1

11.2

11.3

11.4

.

Comparison of novel stationary phases ........................................... J.J. Pesek and E.J. Williamsen Introduction ....................................................................................................... 11.1.1 Characteristicsfor the ideal reversed-phase HPLC stationary phase ..... Characterizationtechniques............................................................................... 1 1.2.1 Spectroscopic techniques....................................................................... 1 1.2.1.1 IR methods .............................................................................. 11.2.1.2 NMR methods ......................................................................... 1 1.2.1.3 ESCA methods ........................................................................ 11.2.2 Thermal methods ................................................................................... 1 1.2.3 Elemental analytical methods ................................................................ 11.2.4 Chromatographic characterization......................................................... Monomeric octadecyl silica............................................................................... 1 1.3.1 Separation mechanisms for ODS phases ............................................... 11.3.2 ODS limitations and adjustments........................................................... 11.3.2.1 Different bonded phase-silica linkages.................................... 11.3.2.2 Differences in ODS columns ................................................... Novel silica phases ............................................................................................ 11.4.1 Non-C18alkane phases .......................................................................... 11.4.2 Materials containing other functional groups ........................................ 11.4.2.1 Cyanopropyl-bondedphases ................................................... 11.4.2.2 Amine and phenyl phases ........................................................ 11.4.2.3 Non-alkyl phases bonded through a reactive olefin................. 11.4.2.4 Polymer bonded phases ...........................................................

Chapter 11

337 337 338 338 341 343 346 349 350 352 353 357 358 360 364 365 368 368 371 371 371 372 372 372 374 375 376 376 377 378 378 379 381 383 383 383 385 385 385 385 388

Contents

XI

11.4.2.5 Liquid crystal phases ............................................................... 11.4.3 Chiral phases ......................................................................................... 11.4.3.1 Ligand exchange chromatography........................................... 11.4.3.2 Direct enantiomeric separation................................................ 11.4.3.3 Pirkle phases ........................................................................... 11.4.3.4 Protein phases ......................................................................... 11.4.3.5 Organometallic phases ............................................................ 11.5 Non-silica-based stationary phases .................................................................... 11.5.1 Alumina-based stationary phases........................................................... 11.5.1.1 Monomeric alumina phases ..................................................... 1 1.5.1.2 Polymeric alumina phases ....................................................... 11.5.2 Carbon-based zirconia ........................................................................... 11.5.3 Polymer-based stationary phases ........................................................... 11.6 Conclusion ......................................................................................................... 11.7 References .........................................................................................................

.

Multivariate characterization of RP-HPLC stationary phases ..... A . Bolck and A.K. Smilde Introduction ....................................................................................................... Multivariate characterization ............................................................................. 12.2.1 Some basic multivariate statistical concepts .......................................... 12.2.1.1 Data matrices; centering and scaling ....................................... 12.2.1.2 Visualization of data................................................................ 12.2.2 Principal component analysis ................................................................ 12.2.2.1 The concept of principal component analysis ......................... 12.2.2.2 The calculation of the principal components........................... 12.2.2.3 A least squares interpretation of principal component analysis .................................................................................... 12.2.3 A principal component example ............................................................ 12.2.4 Three-way analysis ................................................................................ 12.2.5 A three-way analysis example ............................................................... Marker selection ................................................................................................ 12.3.1 The choice of markers ........................................................................... 12.3.1.1 The determinant criterion ........................................................ 12.3.1.2 The induced variance criterion ................................................ 12.3.1.3 Principal component analysis .................................................. 12.3.1.4 Marker selection and three-way analysis................................. 12.3.2 A marker selection example .................................................................. Predictions......................................................................................................... 12.4.1 Ordinary least squares (OLS) ................................................................ 12.4.2 Partial least squares (PLS) ..................................................................... 12.4.2.1 PLS1 ........................................................................................ 12.4.2.2 PLS2 ........................................................................................ 12.4.2.3 PLS predictions ....................................................................... 12.4.3 PLS predictions and marker selection ...................................................

388 390 390 391 391 392 393 393 393 393 395 395 397 399 399

Chapter 12

403

12.1 12.2

403 404 405 406 407 410 410 412

12.3

12.4

414 415 417 420 422 423 423 423 424 425 425 428 429 432 432 433 434 434

XI1

Contents

12.4.4 A PLS example...................................................................................... 12.5 Practical examples ............................................................................................. 12.5.1 Calibration of octadecyl modified stationary phases of different batches ................................................................................................... 12.5.2 Calibration of stationary phases of different types ................................ 12.6 Appendix A ....................................................................................................... A.l Rank ...................................................................................................... A.2 Spectral decomposition ......................................................................... A.3 The singular value decomposition (SVD).............................................. 12.7 References ......................................................................................................... Subject Index

............................................................................................................

435 438 439 443 445 445 446 446 447 451

XI11

Preface This book grew out of a long standing interest in the ways in which retention and the selectivity of separation in liquid chromatography are dependent on the structure of the analyte and on changes in the mobile and stationary phases. These relationships are at the heart of an understanding of the operation of liquid chromatography and of the ways in which the chromatographer can manipulate the conditions of a separation to achieve the analysis of a complex sample. The factors involved in these processes are complex and even 90 years after the pioneering work of Tswett are still not fully understood. Any progress is linked to the development of an understanding of the physical chemical process of solvation and the physicochemical nature of the stationary and mobile phases. Chromatography is also a valuable practical analytical method and much can be learnt by studying relative interactions and by comparing the behaviour of analytes with different chemical structures under different separation conditions. To achieve this objective, techniques for recording relative retentions are needed so that results can be reproduced in different laboratories or by different operators. However, liquid chromatography has a notoriously poor transferability, the same high versatility which enables separations to be precisely optimized also means that small changes between systems can alter the separations. This book addresses some of the ways in which these problems have been overcome to enable retention predictions, identifications and the characterization of the properties of mobile and stationary phases, to be carried out. The work owes much to studies in gas chromatography, in particular the work of Kovhts in providing a retention index scale and of Rohrschneider and McReynolds on the comparison of stationary phases. A theme which leads through the different chapters is the value of relative measurements. Most obviously in the descriptions of the different retention index scales in liquid chromatography and their application to the identification of a wide range of analytes. The indices also form the basis of one of the studies on retention prediction, the other relating retention to the contribution of analytes to partition coefficients. Related methods have been used to compare analytes and their interaction properties. The final group of chapters investigates methods for the comparison of mobile and stationary phases not just by using a simple solvent strength parameter but by examining the comparative interaction of the phases to different types of analytes either in terms of their shapes or physical properties. Bringing these chapters together enables the different approaches to be compared and illustrates the values of each. Hopefully, this will stimulate further research or different approaches for this is by no means the full description of the mechanism of retention. Much more still needs to be done, in particular to understand how complex molecules behave. In this case, the chromatographic behaviour of the analyte under different conditions may itself provide valuable information about the physical properties of the analyte.

XIV

Preface

I would like to thank many of the contributors for useful and interesting discussion of their work and the stimulation it has provided for our own studies. I would also thank my research and project students at Loughborough University of Technology, who have contributed to our own studies in this field. In the same way that their individual contributions have together built our overall study, so I hope that the chapters of this book will contribute to an overall greater understanding of the retention process in liquid chromatography. Roger M Smith May 1994

xv

List of Contributors M. BOGUSZ

Institut fur Rechtsmedizin, Medizinische Fakultat, RheinischWestflilische Technische Hochschule Aachen, Pauwelsstrasse 30, 0-5100, Germany

A. BOLCK

Faculty of Mathematics and Natural Sciences, University Centre for Pharmacy, University of Groningen, Antonius Deusinglaan 2, 9713 A W Groningen, The Netherlands

P. JANDERA

Department of Analytical Chemistry, University of Pardubice, Faculty of Chemical Technology, Ndm. Legii 565, 532 10 Pardubice, Czech Republic

P. KURONEN

Department of Chemistry, P.O. Box 6 (Yuorikatu 20), University of Helsinki, FIN-00014 Helsinki, Finland

J.J. PESEK

Department of Chemistry, Sun Jose State University, Sun Josk, CA 95192-0101, USA

L.C. SANDER

Chemical Science and Technology Laboratory, Organic Analytical Research Division, National Institute of Standards and Technology, Gaithersburg, MD 20899-0001, USA

A.K. SMILDE

Laboratory for Analytical Chemistry, Nieuwe Achtergracht 166, I018 WY Amsterdam, The Netherlands

R.M. SMITH

Department of Chemistry, Loughborough University of Technology, Loughborough, Leicestershire, LEI I 3TU, UK

K. VALKO

Department of Physical Sciences, Wellcome Research Laboratories, Langley Court, Beckenham, Kent, BR3 3BS, UK

S.D. WEST

North American Environmental Chemistry Laboratory, DowElanco, P.O. Box 68955, 9410 Zionsville Road, Indianapolis, IN 46268-1053, USA

E.J. WILLIAMSEN

Department of Chemistry, Sun Jose State University, Sun Jost!, CA 95192-0101, USA

S.A. WISE

Chemical Science and Technology Laboratory, Organic Analytical Research Division, National Institute of Standards and Technology, Gaithersburg, MD 20899-0001, USA

This Page Intentionally Left Blank

Journal of Chromatography Library, Vol. 57: Retention and Selectivity in Liquid Chromatography R.M. Smith, editor 0 1995 Elsevier Science B. V. All rights reserved

1

CHAPTER 1

Retention prediction based on molecular structure Roger M. Smith Department of Chemistry, Loughborough University of Technology, Loughborough, Leicesfershire,LEI I 3TU UK

1.1 INTRODUCTION The retention of a particular analyte in a reversed-phase liquid chromatographic system is dependent on many factors, the structure of the analyte, the nature and chemistry of the stationary phase, the composition of the mobile phase and the temperature. Some of these factors are reasonably well understood, at least on an empirical level, and chromatographers can manipulate eluent composition and even temperature to alter retentions in a predictable manner. However, the effect of the chemical structure of the analyte on retention is probably the least well described parameter. Most chromatographers recognise the broad influence of polarity and size and their effect on hydrophobicity but not the detailed impact of the addition of a methoxyl, carboxamide or other functional group. Nor in most cases is it possible to predict the composition of the eluent required to result in a predetermined retention (capacity) factor (k). Instead the experimental conditions to achieve a particular retention are usually selected by analogy with related compounds or fiom experience of analysing a wide range of samples. Most analytical methods in liquid chromatography are then refined on a trial and error basis. However, in recent years two methods to aid the chromatographer in refining a separation have become available. The first requires no knowledge of the structure of the analyte. A computer programme, often an expert system, uses the retention factors of the components of a mixture fiom a gradient or isoeluotropic set of separations to propose an eluent mixture, which is predicted to provide optimal resolution or overall run times. These techniques include systems such as Drylab, PESOS, ICOS, and DIAMOND, which are based on prediction and mapping methods, and chemometric techniques, including iteration and Simplex optimization methods. These methods have been well reviewed in recent years [l-31. Future developments are likely to see the expert systems being supReferences pp. 45-46

2

Chapter I

plemented by neural networks, which should enable them to “learn” about the properties of a particular column and instrument, before making their predictions [75]. In most assays the structure of the analyte is known and the second approach has been to predict the retention from the molecular structure. This can be carried out directly by the summation of the retention properties of the structural components or by deriving a physical property, such as the octanol-water distribution coefficient (log P), which can then be related to retention by comparison with analytes of known value [3]. As the structures of any impurities or metabolites in a sample are often known, it should also be possible to predict the optimum conditions for their resolution from the main components. This approach has the potential for true prediction as it can propose initial chromatographic conditions, designed from the start to achieve a particular separation. Two different but closely related aspects of this approach form the subject of this and the following chapters. The recent literature also includes numerous papers on retention prediction which related retentions under one set of conditions with those using a different proportion of modifier or temperature. For example, changes in retention with mobile phase composition have been recently discussed by Valko ef al. [4]. A second closely related area has been the selection of robust methods that although they may not be optimized to give the maximum resolution, nevertheless provide methods which are less susceptible to small changes in eluent composition, temperature and or different columns [5,6]. In real life situations this may be an important consideration if the method is to form part of an official method or is required for long-term studies of the stability or quality of a product. Again computer assistance has been provided for the selection of testing conditions and the evaluation of the results.

1.2 STRUCTURE AND RETENTION

The concept that the retention of an analyte in gas or liquid partition chromatography can be expressed as the summation of factors related to its skeleton and individual functional groups was originally proposed by Martin [7].He suggested that the retention of a analyte can be expressed by the summation of contributions from each of the structural components, alkyl-chains, aromatic rings and fictional groups. These substituent values are related to their effects on other equilibria and are recognised as examples of a linear freeenergy relationship. The early work in this area on gas-liquid chromatography and thinlayer chromatography have been reviewed by Kaliszan [S,9]. These concepts have led to a wide range of studies, which have examined the effect of the different substituents on the retention of an analyte in liquid chromatography. These quantitative structure-retention relationships (QSRR) studies have encompassed physical properties, topological indices, and additive functions and have been reviewed in detail [8-121. Similar concepts have long been used for the prediction and calculation of octanol-water partition coefficients (log P) in quantitative structure-activity relationship (QSAR) studies which are important in relating biological activity to structural features. Hansch and Leo [13] have shown that the log P can be calculated by the summation of a value for a parent compound with contributions for each substituent (rc constants) and a

Retentionprediction based on molecular structure

3

similar approach based on fragmentalv) constants has been used by Rekker [14]. There is often a good correlation between the octanol-water partition coefficients and chromatographic retention and numerous studies have used HPLC techniques to measure effective log P values [10,12,15]. The technique works well if a group of analytes are structurally related but compounds of different structural types may show a poorer correlation. However, comparatively relatively little use has been made of the n or f constants to calculate log P values for retention prediction. In a series of studies, Jim0 and Kawasaki [ 16-1 81 predicted the retention factors of alkylbenzenes and substituted aromatic compounds. More recently, the relationship has been used by Valko and co-workers [19] as the basis of a retention prediction system (see Chapter 2). This work has formed the basis of a computer program, which also incorporates the ability to handle partially ionized analytes. Some of the advantages and limitations of this method have recently been evaluated by Fekete et al. [20]. An alternative approach for the prediction of retentions in liquid chromatography is to use the summation of retention increments, which have been determined by comparison of substituted and unsubstituted analytes. These can be expressed either as functional group contributions (Section 1.2.1) or retention index increments (Sections 1.3-1 S). This approach has also been examined in other branches of chromatography. Peng et al. [21] have examined the prediction of the retention indices of analytes, based on their molecu1: structure on an apolar column in gas chromatography. The number of atoms, the aromatic increment and the group retention functions (GFW)were all important. They used a combination of the number of carbon atoms, carbon atom equivalents, and group retention factors for substituents and functional groups. They took into account the effects of rings, is0 and neo-carbons and found that predicted and experimental values were within *3%. In a second paper, they examined these effects for separations on polar columns [22]. A similar approach has also been reported by Evans and co-workers [23] based on the molecular weights and selectivity indices of the analytes.

1.2.1 Chromatographic functional group contributions A frequently applied approach, to relate retention to changes in structure, has been the functional group contribution (z) to the logarithm of the retention factor. The values o f t are determined by comparison of the retention of substituted analytes with the corresponding unsubstituted analyte (Eq. 1.1).

The measurement and application of group contributions have been comprehensively reviewed by Smith [24]. The magnitude of the contributions for individual functional groups differ with the eluent composition and their magnitude usually decreases significantly with increased organic modifier. The contributions also differ with different organic modifiers in the mobile phase. However, these parameters have not been widely used for retention prediction because few studies have examined the relationship between mobile phase composition and the magnitude of the contribution. The contributions were References pp. 45-46

4

Chapter 1

frequently deliberately extrapolated to 100% water as the eluent to give composition independent values (z,), which were then compared with other physical parameters. Probably the most widely investigated functional group change in retention with structure is the methylene group contribution. Numerous studies of homologous series [24] have shown that there is a systematic change in the logarithm of the retention factor with the carbon number. This change is usually similar for all homologous series, irrespective of the other groups present. For example, Figge and co-workers [25] reported a constant change for a series of homologous analytes, n-alkanes, n-alkenes, n-alkylbenzenes, fatty acid methyl esters, alkan-3-ones, 2-n-alkyl-pyridines, 1 -n-alkanols. This relationship also forms the basis of most retention index series and is discussed further in Chapter 3. A difficulty with many of the retention studies, such as the hnctional group contributions, which are based on retention factors (k), is that the increments are very dependent of the experimental conditions, such as temperature and the eluent composition. Frequently these have not been closely controlled and the resulting retention values are often unique to that individual system of mobile phase and column. Many of these problems of reproducibility and transferability between systems can be overcome by using relative retention measurements, such as retention indices (see Chapter 3). A retention index scale effectively compares the increment for a functional group with the corresponding methylene increment in the same system. Both should be similarly affected by the small changes in the strength of the eluent and by temperature, so that retention index based group contributions should be almost independent of the eluent composition and of the make of stationary phase. Unless there are changes in the relative interactions between the methylene or other functional groups and the stationary phase, the retention increments should be largely independent of the brand of stationary phase and carbon loading of the columns, even though these differences can significantly effect retention factors (k).

1.2.2 Related prediction studies

In an extensive series of studies, Jandera has examined the description of the interaction of an analyte in terms of lipophilicity (rice) and polarity interaction indices (4).The values of these terms for a number of functional groups have been determined by comparison with the retentions of the homologous n-alkylbenzenes. This work has been recently reviewed [26] and is described in more detail in Chapters 7 and 8. Galusko proposed that it should be possible to predict the retention of a compound based on the summation of the effects of the bond dipoles and partial molar volumes of the substituents [27]. This system has now been developed into ChromDream, a computer-based prediction system [28]. Their model is based on a two-layer continuum model of reversed-phase chromatography and the differences in molecular solvation energies in the two phases. The retention of an analyte is described by Eq. (1.2), in which Vi are the increments of the partial molar volume fragments in water and Ge,si,H20are the increments of energy of interaction of bond dipoles with water. a, b and c are the parameters of the reversed-phase system and depend on the column and phase ratio and are characteristic of the stationary layer and mobile phase.

Retention prediction based on molecular structure

hk,

=~(cc)~’~

+~(C,AG~.~.J,H,O)+C

i

5

(1.2)

J

The values of a, b and c for a particular separation system have to be determined by using reference compounds. The parameters a and b are related to differences in the SUTface tension and dielectric permeability of the sorbent surface layer and mobile phase, respectively, and can be related to differences between stationary phases. The programme gave accurate predictions over a range of eluent composition for a wide range of aromatic analytes. However, the model does not take stereochemical and intramolecular interactions into account and discrepancieswere found for ortho-disubstituted analytes. A common approach for retention correlation has been to relate a range of physical parameters related to structure, such as shape and connectivity parameters, to retention factors using multivariant analysis [9,10]. The resulting regression equations can then be used for retention prediction. However, although the correlations are frequently excellent, the addition of new model compounds to the data set will often markedly change the coefficients and even cause the significant terms to change. Thus although the regressions can “predict” the retention of analytes that are included the original model data set, they frequently fail to predict accurate retentions for new compounds. A conceptual problem is that there is often no clear connection between the properties that are used as terms in the regression and structural or physical properties, which are generally accepted as being significant in liquid chromatography. The correlation may be valid but only because the parameters are also indirectly related to a parameter of relevance. This purely chemometric approach has been reviewed [9,11] but the real prediction power appears to be limited. 1.3 FUNCTIONAL GROUP EFFECTS ON RETENTION INDICES

Retention indices, which are determined by logarithmic interpolation between the retention factors of a series of homologous standards, provide a reproducible mode of retention measurement (Chapter 3). They can form the basis of reliable and transferable retention comparisons. In comparison to functional group contributions, they are more independent of the eluent composition and stationary phase. Functional group or substituent increments (Is, x) can be determined as the differences between the retention index of a parent compound (ZR-H)and those of substituted derivatives (&) (Eq. 1.3).

Is, x = IR-x- ZR-H As the functional group index increments are not related to the absolute retention times but to the methylene group contribution = 100) by definition) they should be largely independent of eluent composition. If interactions occur between multiple substituents, the differences between the simple summation of the contributions from the individual groups and the experimental retention index value, are defined as the interaction index ZI. Because of the range of polarities of analytes in HPLC, a number of different retention index scales have become established ([29], Chapter 3). The most frequently employed scales have often also been used for retention prediction. As each scale is based on the methylene increment, the results should be equivalent but small differences might be exReferences pp. 45-46

Chapter 1

6

pected from changes in separation conditions. The n-alkanes and n-alkylbenzenes are both highly retained and have been used for non-polar analytes (Section 1.3.1). The alkyl aryl ketones (Section 1.4) cover a wide range of polar and non-polar analytes and have formed the basis of a major prediction study (Section 1.4). The more rapidly eluted alkan2-ones have been used primarily for the prediction of drugs and their metabolites (Section 1.3.2). Although the 1-nitroalkanes have a similar polarity, no prediction studies have been carried out with these standards. Some prediction work has used a retention index scale based on standards with increasing numbers of aromatic rings (Section 1.5).

1.3.1 Retention prediction based on n-alkanes and n-alkylbenzenes

Morishita et al. [30] determined the retention index increments for four aryl substituents using the n-alkane scale (Table 1.1a). The values of the increments reflected the expected hydrophobicities, although the methyl group increment (dZ= 89.3) was lower then the nominal value for a methylene group (dZ= 100) and may have reflected hyperconjugation with the ring. The relative magnitudes of the increments corresponds to those obtained in later studies on the alkyl aryl ketone scale [3 11. They also studied the interactions between two substituents on the aromatic ring, by measuring the difference between the experimental indices and the calculated values obtained by the summation of substituent increments (Table I. Ib). The increments for interTABLE 1 . 1 SUBSTITUENT AND INTERACTION INCREMENTS OF RETENTION INDICES BASED ON THE nALKANE SCALE (a) Substituent increment SIX = I,+x - IA-H. For comparison, substituent index I~,A-x values are given from the regression equations based on alkyl aryl ketone retention index scale [31] Functional group

Substituent increment (SIX)

I ~ ,[311 ~ - ~

89.3 41.9 -195.7 -213.6

104 -87 -290 -302

(b) Interaction increments between substituents on an aromatic ring. Calculated as the differences between experimental retention index and summed index values for substituents and parent Interaction increment (611,~-y)

Groups X

Y

2-Y

3-Y

4-Y

CH3 CH3 CH3 CH3 NH2

CH3 OH NH2 NO2 NO2

-12.7 -11.1 -24.7 -27.4 140.8

3.1 -24.5 -24.3 -2.5 84.5

5.0 -24.1 -23.5 -19.2 48.8

Based on Morishita and co-workers [30]. Conditions: column, Partisil ODS-3; eluent, methanol-water (70:30).

Retention prediction based on molecular structure

7

actions between methyl groups were small (61 = -13 to 5 ) but between amino and nitro groups much larger changes were observed. The ortho-nitro substituent (dZ= 141) on aniline had a major effect and increased the retention by considerably more than the metuor puru-substituents (61 = 85 and 49, respectively) suggesting that hydrogen-bonding was occurring between the ortho-groups. Using the index increments for the substituents and interactions, they calculated the predicted retention indices for a number of trisubstituted aromatic compounds and found a good correspondence with experimental values. The deviations were between -14 and +10 units; for example, 3,5-dimethylaniline ZCdc= 163 and robs= 166; 2-methyl-4-nitroaniline, Zcdc = 79 and Zobs = 80. A similar approach was subsequently used in a study of sulphur compounds by MSckel [32]. He compared the retention of a series of homologous thiols and alcohols with the corresponding alkanes and determined the retention index increments for the replacement of a methylene group by a hydroxyl (OH) group (-510 to -519) or thiol (SH) groups (-1 80 to -206) (the values increased slightly with chain length) using methanol-water (70:30). Slightly confusingly, the retention indices of the parent compounds in this paper were based on the number of non-hydrogen skeletal atoms (C plus S) so that the replacement increment for the thiol group in heptylthiol (]= 599) was calculated as 599 - 800 [(C, + S) x 100 = -201)l. These replacement values correspond to substituent increments for hydroxyl of ZoH = -410 to -419 units and for thiol of ZsH = -80 to -106 units. Thioethers (R-S-R) had retentions similar to the monothiols. The increments for the hydroxyl group were similar to those found later for the aliphatic hydroxyl (Is,R-oH=-362 in methanol-buffer (50:50), see Table 1.7) using the alkyl aryl ketone scale. The retentions of the alkylpolysulphides (R-S,-R) were also examined [32]. When a rnethylene group in tetradecane (Z= 1400) was replaced by a sulphur atom to give hexylheptyl sulphide (I = 1060) the retention decreased markedly. A second replacement by a sulphur atom gave dihexyl disulphide with a similar retention but as the proportion of sulphur atoms in the chain was increased further, the polarity decreased to eventually give dipropyl octasulphide (1= 1259) and dimethyl dodecasulphide ( I = >1400). These changes were explained as the initial formation of a local polar centre and then an increase in retention as the non-polar polysulphide replaced the methylene groups. In subsequent studies, MSckel and co-workers [33] determined the coefficients, which related the retention indices of homologous analytes to the number of carbon atoms (nc) (Eq. 1.4).

They used these values to demonstrate that for most homologues the change ( B ) in the retention index on the n-alkane scale on the addition of a methylene group was close to 100 units. However, smaller values were found for Ph-(CH2),-Ph (88.99 units) and RS9-R (78.84 units). The A term indicated the effect of the functional group. A comparison of the saturated and unsaturated hydrocarbons suggested that the addition of an olefinic group increased retention by 61.85 units but an acetylene group decreased retention markedly (A = -265.96). These increments were translated into “chromatographic free energy” changes. They also found a linear relationship for each series between retention index and calculated total surface area. In a subsequent study [34] they reported the corReferences pp. 45-46

Chapter I

8 TABLE 1.2 INDEX INCREMENTS FOR SUBSTITUENTS ON HOMOLOGOUS SERIES OF ANALYTES Substituent

Correlation coefficients

B

A ~

H Br SH OH CN -04-

0 -1 12.4 -145.9 -653.1 -625.4 -211.6 -108.2

~~

100 97.5 98.4 101.3 101.3 86.1 90.9

Based on data from Mockel and co-workers [34]. Determined from the intercept of plot between 1 , and carbon number. Eluent methanol. IK = A + BNc. The intercept A is equivalent to the substituent index value.

responding values for a wider range of substituted homologues (Table 1.2). The A values reflected the effects due to the different groups. However, these values were not used for retention prediction. Subsequent studies by Aced and co-workers have examined the 1,nbi(alkylthioa1kanes) R-S-(CH2)n-S-R and reported a constant methylene increment of 88 units as n was increased [35]. An example of the application of use of multivariant analysis to correlate physical properties and retention indices has been reported by Dimov [36]. He identified terms which could be used to precalculate the retention indices of isoalkanes in HPLC. His initial equation Icalc = 41.09 + 0.92756PCI (PCI = physicochemical index) gave only a correlation of 0.9861 and was expanded to give a model equation by the addition of terms for different type of atoms (for example nq = quaternary carbons) (Eq. 1.5). Icalc= 42.17 + 0.33448PCI + 58.336n0- 1 1.629ncH3+ 5.056nL

- 3.729nd - 3.44nq- 7.55ni This equation gave a correlation of 0.99959 between experimental and calculated values. The model provided good predictions for m b e r isoalkanes but was not tested for other groups. 1.3.2 Retention prediction based on alkan-2-ones In studies of the application of the alkan-2-one retention index scale, Baker and coworkers explored two main aspects. Firstly, they developed a relationship between octanol-water partition constants and retention indices, which could be used as a predictor of retention and, secondly, they examined the systematic changes in retention indices with the addition of a functional group to a parent compound. In other studies, retention indices based on the alkan-Zones were used as a measure of lipophilicity in QSAR comparisons (Chapter 3).

Retention prediction based on molecular structure

9

The prediction studies examined groups of related pharmaceuticals either as part of QSAR studies or in the examination of drug metabolism. Baker proposed [37] that as the addition of a methylene group (whose x value is 0.50) changes the retention index value by 100 units, other functional groups should show a corresponding effect. The increment for the addition of a function group should therefore be 200 times its x constant (Eq. 1.6)

AI = 200n

(1.6)

Thus, if the retention index of a parent compound (ZR-H) was known, the predicted retention indices for substituted derivatives (ZR-x) could be calculated from the nx constants by using Eq. (1.7).

This assumption was tested using groups of propanolols, barbiturates, anthranilic acids [37], narcotic analgesics and nortropanes [38]. For the barbiturates, barbital (diethylbarbitone) was chosen as the parent compound and the calculated and measured values of the retention indices were determined. The correlation was good for the alkylated and aryl substituted barbiturates but poorer for thiopental and thioamylal. Overall the highest deviation was 43 units with a mean error of 29 units. For the anthranilates [37], which contained a wider range of bctional groups, the mean error between the measured and predicted indices was only 22 units, even though the carboxylate group in each compound would have been nearly completely ionized (Table 1.3). For the propanolol derivatives [37] the calculation was more complicated, as nxvalues were not available for all the functional groups and some had to be determined from model compounds. There was a reasonable correlation between predicted and experimental retention indices, which was attributed to uncertainties in the partition coefficients. When the same approach was applied to the narcotic analgesics and related compounds [38] using morphine as the parent compound, the comparisons showed a greater variation (Table 1.4), which was attributed to stereochemical factors such as the shielding TABLE 1.3 MEASURED AND CALCULATED RETENTION INDICES OF ANTHRANILATES DERIVED FROM CONTRIBUTIONS BASED ON HANSCH CONSTANTS Functional group (X)"

Measured

Calculated

530 565 586 606 656 678

530 (Ired 560 578 634 670 734

Based on values from Baker [37]. aCompounds, C6H4(C02H)-NH-S02-C6H~-x. bIcajc=Ir,+ 20Ozx based on the alkan-2-one scale.

References pp. 45-46

zxconstant

Retention indexb

0.15 0.24 0.52 0.70 1.02

Chapter 1

10 TABLE 1.4 COMPARISON OF MEASURED AND CALCULATED RETENTION INDICES OF NARCOTIC ANALGESICS AND RELATED COMPOUNDS Compound

Morphine Oxymorphine Oxocodeine N aloxone Codeine Dihydrocodeine Hydromorphone Ethylmorphine Hydrocodone Heroin Pentazocine Levallorphan Phenazocine Levorphanol Dextromethorphan

Retention index (0 Observed

Calculated

621 529 615 668 705 71 1 712 783 79s 805 965 1002 1005 1126 1284

621 (Ired 329 459 469 75 1 811 615 85 1 74s 925 1275 1279 1427 1145 1269

Based on Baker and co-workers [38]. Index values calculated as Table 1.3 using the alkan-2-one scale. Reference compound, morphine.

of polar substituents by the 14-hydroxy group in the oxymorphone type of drugs. These effects were used to characterise the stereochemistry of isomeric N-substituted 3-propananilidonortropane analogues. The a-isomer consistently had a higher retention index than the B-isomer, in which the polar 3-propanilido group is less shielded. During a study of metabolites, Baker [39] found that the retention index values of glucuronides were much lower (dI = -244 f 3 1 units) than the parent compounds, such as morphine. In this case the corresponding n value for the glucuronide group was not available. However, by using the shift in the retention index values, Eq. (1.6) could be used to calculate the effective constant (ng~dglucwoni~e = -1.22 f 0.16). In all these examples, a hydroxy group on the parent structure had been converted into the glucuronide but if hydroxylation also occurred as part of the metabolism then the expected change in the retention indices would be -378 units. These changes were subsequently used to study the retentions of metabolites from primaquine [40]. Most of the hydroxylated metabolites were identified by comparison of their retention indices with standards. Two polar metabolites were identified as possible glucuronides because their retention indices (A, I = 538 and B, I = 681) were smaller, by about 244 units, than primaquine (I= 733). However, they were not hydrolysed to the starting material by B-glucuronidase, even through the first metabolite was cleaved by the treatment. By using Eq. (1.7), Hufford et al. [41] were able to predict the retention indices of metabolites formed by microbial action on imipramine. The values for the hydroxylated and desmethyl derivatives closely matched the empirical values (2-hydroxy, Icalc = 789, Iexpt = 769; 10-hydroxy, Zcalc = 725, Iexpt = 720 and N-desmethyl, Icdc = 846, Iexpt = 807) and helped to confirm their identification. They were also able to obtain a tentative as-

Retention prediction based on molecular structure

11

signment of 10-hydroxydesmethylimipramine as a previously unidentified metabolite (Icalc = 529, Zexpt = 484) but no authentic sample was available for confirmation. The influence of stereochemical effects on the ability to use retention index increments to predict retention has been examined using a number of azabicycloakanes and bicycloalkanes [42]. From a regression analysis, Baker et al. were able to determine the empirical effect of selected functional groups, however, these differed from the expected values based on x constants. For example, replacing OH by OCH3should increase retention indices by 234 units but the observed increase was only 160 units and for the introduction of phenolic OH, the expected change was -134 units but the observed value was -84 units. It appeared that the increment for a substituent was dependent on the total number of polar substituents in the molecule. When this was taken into account the empirical Values for these groups became 205 and -144, respectively, much closer to the predicted values. The regression analysis also demonstrated a difference between stereochemical isomers of 57 units. A similar difference of 67 units was observed for a number of related azabicyclooctanes. These results suggest that in complex systems, the ability to obtain good predictions may be affected by interactions between groups, which depend on their relative positions in space. The prediction of retention indices has also been used to assist the identification of natural products from plant and microbial sources. During a study to identify the urushiol congeners from poison ivy and poison oak, Ma and co-workers [43] calculated the retention indices for unsaturated and acetylated derivatives of 3-pentadecylcatechol (PDC), and 3-heptadecylcatechol(HDC) by using Eq. (1.7) (PDS-triene Zexpt = 1416; ZCdc = 1393 and HDC-diacetate Zexpt = 2026 and Zcalc = 2023) (see also Table 4.9). However, they found that the value of x = -0.63 (corresponding to Zx = -125) for the double bond gave a better fit with the experimental value than the standard value (nx = -0.30). This prediction calculation provided a simple method for the characterization of the congeners, which avoided the need for derivatizationor synthesis. Magg and Ballschmiter found systematic changes in the retention indices (IRCOMe) in ergopeptines with changes in the structure [44]. The magnitudes of the changes were almost identical on three different column systems, even though the absolute values of the indices differed. However, the values were not correlated with Hansch x values, although the trends agreed.

1.4 PREDICTION OF RETENTION INDICES BASED ON ALKYL ARYL KETONES

The alkyl aryl ketones [45] have been identified as suitable retention index standards for a wide range of analytes (see Chapter 3), because of their ready detectability, easy availability and similar polarities, and hence retentions, to many aromatic analytes. It was also shown that retention indices based on this scale are highly reproducible and can be transferred between columns more readily than retention factors [46,47]. The retention indices were relatively insensitive to the exact eluent composition and would not be affected by small variations such as might occur in the preparation of an eluent mixture by different operators. References pp. 45-46

Chapter I

12

Because of this independence from the chromatographic conditions, Smith proposed [31] that the retention indices could form the basis of a retention prediction system, whose conclusions would be more generally applicable than values calculated for one specific retention system. The predicted retention index value for an analyte could then be converted into the corresponding retention factor for a column-eluent system by using Eq. (1A), whose constants A and B can be determined fiom the retention factors (k)of a series of alkyl aryl ketones, which have defined retention indices (I = carbon number nc x 100). log k = A + BI

(1.8)

The intention was to develop a system which could predict the retention of a compound fiom its structure and the mobile phase modifier. It should then be possible to use this approach to predict the conditions for a separation or the optimum conditions to achieve a particular resolution. Even if the retention index of a complex analyte could not be predicted, it should be possible to calculate relative retentions, compared to a parent compound in the same way that Baker (in the previous section) was able to estimate the retentions of metabolites and congeners. From the start, it was recognised that because the interactions between substituents are not hlly understood, particularly on heterocyclic and aromatic rings systems, it would probably be not be possible to make completely accurate predictions for complex molecules. However, any deviations between experimental and predicted index values could be used to examine the interactions between the functional groups. The basis of the prediction system was that the retention index of an analyte, in a selected eluent, could be calculated by the summation of the retention index of a parent compound, substituent index values for each substituent plus interaction index terms required to describe interactions between substituents, such as H-bonding, steric and electronic interactions) (Eq. 1.9).

where Ipis the retention index value of a parent compound, IS,R is the substituent index contribution from saturated aliphatic carbons, are the substituent index contributions for substituents on an aromatic ring, Is,R-x are the substituent index contributions for substituents on aliphatic carbons (these substituents will include unsaturated and carbonyl groups), and I1,y-Z are the interaction index contributions between substituents or groups (Y-Z) to account for H-bonding, steric and electronic effects. Although some reports had suggested that there is a nearly linear relationship between percentage composition of the eluent and retention parameters, such as log k, Schoenmakers el al. [48] found a closer correlation can usually be obtained with a quadratic relationship, particularly if a wide range of eluent compositions were being compared. Consequently, each of the terms in the prediction system (Eq. 1.9) was defined for each organic modifier as an experimentally determined quadratic equation (Eq. 1.10): I =ax2 + bx + c (x

= % of

organic modifier in the eluent).

(1.10)

Retention prediction based on molecular structure

13

The a, b, and c coefficients for each modifier derived from the different components of the prediction equation (Eq. 1.9) could then be summed to give an overall quadratic equation (Eq. 1.1 1) for the retention index of the analyte. (1.11) Benzene was selected as the parent compound, because all its substituted derivatives could be detected spectroscopically. A wide range of derivatives was also readily available, substituted both directly on the aromatic ring and on aliphatic side chains, which means that both types of substituent indices could be determined. Suitable polysubstituted standards were also readily available for the determination of the interaction terms. In order to ensure a consistent data set [49], a single batch of stationary phase was used throughout the study. The separations conditions were also standardised. The eluent was buffered to pH 7.0 using a constant ionic strength buffer prepared by weight and the column temperature was controlled at 30°C. Regular test samples were examined and the reproducibility of different columns was checked. Although there were some changes in the retention factors (k),the retention indices were consistent and a variance of less than 10 units was considered to have been reached across the study, which lasted over 2 years (see Chapter 3). All the terms generated in the project were brought together into a database, which could be interrogated using a expert system programme CFUPES (Chromatographic Retention Index Prediction Expert System) to generate predicted retention indices (see later). 1.4.1 Monofunctional compounds Although the retention indices of analytes are largely independent of the proportion of modifier in the mobile phase, there are small changes and the first stage of the project was to determine the baseline values for benzene with different modifiers [31,50]. For methanol, acetonitrile and THF the relationships were curved (Fig. 1.1) and could be matched closely in each case by a quadratic equation (Appendix 1.1). These correlations provided smoothed values, which were used as the defined retention indices of the unsubstituted parent compound (Ip). 1.4.1.I Aromaticfunctional groups

Initially, the retention factors of 16 monosubstituted model compounds and the alkyl aryl ketones (from acetophenone to heptanophenone) were measured in a range of different eluents from methanol-pH 7 buffer (40:60) to (80:20) and acetonitrile-pH 7 buffer (30:70) to (80:20) [31,51]. Although the retention indices were also determined in each case for 90% modifiers, the corresponding retention factors were so small that these values were considered unreliable and were not used in the correlations. Subsequently the retentions of these compounds were also examined using THF-pH 7 buffer (20230) to (60:40) eluents [50]. The monosubstituted model compounds covered a wide range of functional groups, however, it was not possible to examine the aryl carboxylic and sulphonic acid groups, as they were ionized at the pH of the buffer.

References pp. 45-46

14

Chapter 1

'loo

-

n.

=

1

1000

Y

c

.-0 Y

c

2! 900 Q)

a

MeOH

800

1

I

I

I

20

40

60

80

Organic modifier

(%I

Fig. 1.1, Comparison of experimental retention indices (symbols) of benzene with calculated values of parent index values (Ip, lines) derived from quadratic relationships (Table 1.7). Eluents: 0 , methanol-buffer; W, acetonitrile-buffer, THF-buffer. Values from [3 I] and [50].

+,

In each case, the retention indices were calculated by comparison with the retention factors of the alkyl aryl ketones injected in the same series of separations (e.g. for methanolic eluents see Table 1.5). The substituent increments for each functional group were then determined using Eq. (1.3) (Table 1.6). The increments for the relatively non-polar groups were almost constant with proportion of modifier but the polar groups showed greater changes. Particularly for the acetonitrile eluents the relationships were often nonlinear. However, the changes were small compared to the changes in retention factors for these compounds. These empirical substituent increments were fitted to quadratic equations to give the coefficients a, b and c, which could be used to predict the substituent index at any eluent composition. Throughout the study, if the change with mobile phase composition was small ( 4 0 units), a single c term was determined. The calculated substituent index values from the quadratic expressions gave a close fit with the experimental increment values (for example, Fig. 1.2). More recently, naphthalene as an alternative parent compound and some additional monosubstituted compounds have been examined [52] and the fill set of coefficients is listed in Appendix 1.1. The advantage of this approach is that the predicted substituents indices reflect the differences in the interaction selectivity of the modifier. For example, the calculated substituent indices, in three isoeluotropic eluents, methanol-buffer (70:30), acetonitrilebuffer (50:50) and THF-buffer (40:60) are often markedly different (Table 1.7). The changes in the nitro group ( Zs,k-x=-54 in methanol and -56 in THF) and carbomethoxyl group (Zs,k-x = -10 in methanol and -80 in THF), match the published re-

Retention prediction based on molecular structure TABLE 1.5 RETENTION INDICES OF MODEL COMPOUNDS IN METHANOL-pH 7.0BUFFER ELUENTS Compound

Acetophenonea Aniline Anisole Benzaldehyde Benzamide Benzene Benzeneb Benzonitrile Benzyl alcohol Benzyl bromide Benzyl chloride Benzyl cyanide Biphenyl Bromobenzene Chlorobenzene Methyl benzoate Nitrobenzene Phenol Toluene

Retention index (0 Methanol (%)

40

50

60

70

80

805 650 884 774 605 883 885 776 689 991 962 773 1205 1027 998 899 851 685 987

806 658 904 777 589 915 913 788 691 1004 976 766 1222 1051 1021 904 857 683 1019

803 657 917 777 578 938 938 775 698 1019 992 763 1231 1065 1036 904 864 680 1039

803 659 934 775 570 958 961 774 684 1030 992 738 1247 1088 1051 910 874 67I 1065

804 639 954 784 551 983 982 760 675 1059 994 722 1270 1110 1072 914 874 650 1095

Data from Smith and Burr [31]. aDefined value I = 800. bParent index (Ip) values for benzene derived Erom quadratic regression equation (Appendix I. 1).

versa1 in the order of elution of methyl benzoate and nitrobenzene between these two modifiers [53], which is often used as an example of eluent selectivity. The phenolic hy= -229 in methanol, -243 in acetonitrile, and -142 in THF) and acetyl droxyl group (II= -1 13 in methanol, -127 in acetonitrile and -165 in THF) also reflect the marked changes that are frequently observed between separations in isoeluotropic eluents. The way in which the substituent index values change with the proportion of modifier also differs between the modifiers (Fig. 1.3). The chloro group index is unaffected by the proportion of acetonitrile, changes slightly with methanol and markedly with THF. These changes contrast with prediction methods based on log P values or on the extrapolation from a single log ko or zo value in pure water. In those cases, the relative order of elution is predetermined by the original values and the predicted relative order of elution will be the same in each eluent. Although the magnitudes of any changes can depend on the strength of the eluent, each will behave in a corresponding manner. 1.4.1.2 Aliphatic Jirnctional groups

Although by definition, the addition of a methylene group increases the retention index of a compound by 100 units = loo), the index increments for the successive addition of a methylene group to benzene to increase the length of the alkyl side chain were not sysReferences pp. 4 5 4 6

16

Chapter 1

tematic (dIcH2= 87 - 112 units) [50,54] suggesting that an interaction was occurring with the benzene ring. In particular, there was a reduced increment (dZ= 87-92 units) for substitution directly onto the benzylic carbon, i.e. toluene to ethylbenzene. An interaction index correction term (ZI,PHCH~R= -12 for methanol and acetonitrile and -14 in TKF) was therefore defined for alkyl substitution onto a benzylic carbon ) [50,54]. A similar effect had also been observed for the effect of the Hansch n coefficients on alkyl substituents on benzene. The step between the n values for methyl and ethyl substituents of 0.46 is smaller than for the addition of a methylene group to longer chains, which range from 0.56 to 0.58 [13]. The retention indices of a number of 1, 2, and 3-substituted toluenes, ethylbenzenes and propylbenzenes were measured and the substituent index increments determined relative to the calculated retention indices for the corresponding alkylbenzenes [54]. Because they might be partially or completely ionized in the mobile phase, carboxylic and sulphonic acids and amines were not examined. The increments for the substituents differed according to their proximity of the benzene ring. It was assumed that by the 3-position of propylbenzene the effect of the ring should be negligible and these retention index increments (Table 1.8), when available, were used to determine the substituents indices. However, in some cases only substituted toluenes or ethylbenzenes were initially available TABLE 1.6 RETENTION INDEX INCREMENTS FOR SUBSTITUENTS ON AN AROMATIC RING IN METHANOL ELUENTS Substituent

CONH2 NH2 CH20H OH CHO CH2CN CN COCH3b NO2 OCH3 C02CH3 H CH2CI CH3 CI CH2Br Br Ph

Retention index increment (Soa Methanol (?h)

Hansch a constant

40

so

60

70

80

-280 -23 5 -1 96 -20s -111 -1 12 -109 -85 -34 -1 14 0 77 102 113 106 142 320

-324 -255 -222 -230 -136 -147 -133 -1 13 -56 -9 -9 0 63 106 108 91 138 309

-360 -281 -240 -258 -161 -175 -1 63 -138 -74 -20 -34 0 54 101 98 81 127 293

-391 -3 02 -277 -290 -1 86 -223 -187 -161 -87 -27 -5 1 0 31 104 90 69 127 286

4 31 -343 -307 -332 -198 -260 -222 -182 -108 -2 8 -68 0 12 113 90 77 128 288

Values from Smith and Burr [3 13. aIncrement61 = I k - x - Ip (Ip = calculated value for benzene). bBased on defined value of I = 800.

-1.49 -1.23 -1.03 -0.67 -0.65 -0.57 -0.57 -0.55 -0.28 -0.02 -0.01 0.00 0.17 0.56 0.71 0.79 0.86 1.96

Retention prediction based on molecular structure

17

0

0

0

I

I

I

0

-2004

-250-

-300-

-350

I

I

1

(Table 1.8) [54]. The correlation coefficients for changes with the proportion of modifier were calculated and were used to determine the corresponding aliphatic substituent indices (Zs,R-x, Fig. 1.4). The retention indices of the substituted toluenes and ethyl benzenes were used to derive interaction index terms for the effect of the phenyl groups but generally these values were small and except for cyano, hydroxyl and carboxamide groups on toluene could probably be ignored ZI,ph-x < 50 units)[54]. Values from additional model compounds have been derived more recently [51] and the full set of coefficients for the substituent indices are listed in Appendix 1.1. Typical values for the aliphatic substituent indices (Is,R-x) have been calculated for isoeluotropic eluents (Table 1.7) and these were often markedly different from the corresponding aromatic substituents. In particular, the aliphatic hydroxyl group = -362, -459, and -434, in methanol, acetonitrile and THF, respectively) has a greater ef= -229, -243 and -142, respectively). These fect than the phenolic hydroxyl differences corresponded to those observed in the studies described earlier (Tables 1.1 and 1.3). The addition of an aliphatic bromo group causes little change to the retention of an analyte (I=3, -13, and 2, respectively), whereas the aromatic bromo group would increase the retention markedly (Zs, r-Br = 135, 127 and 117, respectively). References pp. 45-46

18

Chapter 1

TABLE 1.7 COMPARISON OF CALCULATED SUBSTITUENT INDICES IN ISOELUOTROPIC ELUENTS Substituent

Substituent index ( I s , ~ - xor IS, -x))a MeOH-bufferb

Benzene Ip Naphthalene

4,k-x

SO2CH3 CONH2 CONHCH3 CON(CH312 NHCOCH3 NH2 OH CHO OCOCH3 CN COCH3 NO2 OCH3 C02CH3 H NHC2H5 F NH(NH312 CH3

c1

Br Ph CHCH-CH3

Hansch n constant

(5050)

MeCN-bufferb (40:60)

THF-bufferb (30:70)

913 1104

927 1101

966 1112

-374 -321 -278 -199 -240 -254 -229 -138

-342 -392 -339 -274 -293 -230 -243 -141

-134 -1 13 -54 -1 1 -10 0 -40 6 44 100 105 135 305 239

-111 -127 -54 -19 -34 0 -1 1 5 46 100 98 127 27 1 215

-299 -381 -366 -381 -278 -216 -142 -165 -127 -133 -165 -5 6 -3 0 -80 0 8 16 17 100 95 117 24 1 190

-0.97 -1.23 -0.67 -0.65 -0.64 -0.57 -0.55 -0.28 -0.02 -0.01 0.00 0.08 0.14 0.18 0.56 0.71 0.86 1.96 Fragmental constant (FJ 1131

IS.R-X

CONH2 OH (primary) OH (sec.-tertiary) CHO CN COCH3 OCH3 C02CH3 Br H CI CH3 CHCH-CH3

-1.82 -1.49 -1.27

-432 -362 -394 -324 -300 -215 -167 -158 3 0 -39 100 217

-540 -459 -494 -326 -280 -226 -190 -186 -13 0 -49 100 199

-532 -434

-2.18 -1.64

-363 -316 -181 -227 -186 2 0 -27 100 -34

-1.27 -1.13 -1.10 -0.72 0.20 0.23 0.06 0.77

a I ~ , and ~ . IS,R-X - ~ calculated from coefficients in Appendix 1.1 [31,50,52,54]. bEluents selected to give similar retention factors for acetophenone:methanol-buffer (50:50), k = 3.23; acetonitribbuffer (40:60), k = 2.91; THF-buffer (30:70), k = 3.48.

Retention prediction based on molecular structure

19

200

100 X P)

-0 C

c C

aJ

0

3

.-c u) n ¶ 4-

cn

-100

-200 0

20

40

60

Organic modifier

80

100

(%I

Fig. 1.3. Comparison of substituent indices of the aryl chloro group (Ar-CI, solid symbols) and carbomethoxyl group (Ar-COzCH3, open symbols) in THF-buffer eluent (0 and 0 )methanol-buffer eluents (0and H) and acetonitrile-buffer eluents (A and A)[50]. Reproduced with permission.

Further studies examined the effect of unsaturation and chain branching in aliphatic side chains [52,55]. The olefinic group (-CH=CH-) had much smaller effect (ZS,CH:CH = 83-122 in 3-phenyl-I-propene and Is,CH:CH = 103-145 in 1-phenyl-1-propene) = 200) in methanolic and acetonitrile modifiers. With than two methylene groups THF as the modifier, the olefinic group in I-phenyl-1-propene had an even smaller contribution (Zs,CH:CH = 47-105) and appeared to make a large negative contribution (-105 to -320) in 3-phenyl-1-propene [52]. There were differences in the retention indices of isomeric alkylbenzenes and isomeric and corresponding isomeric phenylpropanols [%I. These lead to the introduction of terms for alkyl branching (II,brmch = -12 in methanol and -20 in acetonitrile) and for secondary and tertiary hydroxyl groups, which were slightly larger (about 40-50 units more negative) than those for primary hydroxyl groups. The coefficients for these terms and for olefinic groups are included in Appendix 1.1 and typical values are given in Table 1.7.

I.4.1.3 Relationship between substituent indices and octanol-water partition substituent increments Within groups of closely related compounds, log k of analytes in reversed-phase HPLC References pp. 45-46

Chapter I

20

TABLE 1.8 RETENTION INDEX INCREMENTS FOR ALIPHATIC SUBSTITUENTS IN METHANOL-pH 7.0 BUFFER ELUENTS Retention index increment Methanol (%) 40

50

60

70

80

1013

1039

1061

1082

-195

-261

-293

-327

-300

Substituents on ethylbenzene Calculated parent I

1073

1101

1126

1149

1170

COCH; CORa OCH3 ORa

-368 -1 89 -289 -150 -250

-403 -216 -316 -167 -267

-459 -243 -343 -181 -28 1

-486 -272 -372 -194 -294

-530 -303 -403 -202 -302

Substituents on n-propylbenzene Calculated parent I

1173

1201

1226

1249

1270

CN CO2CH3 C02Ra CI Br

-333 -266 -127 -227 -26 8

-366 -302 -158 -258 -42 -5

-391 -334 -186 -286 -5 1 -12

-436 -3 83 -216 -316 -65 -20

-477 -428 -24 1 -34 1 -80 -70

Substituents on toluene Calculated parent I

985

Data from Smith and Burr [54]. alOO subtracted for the methyl group contribution. b200 subtracted for the ethyl group contribution.

have frequently been linearly correlated with the octanol-water partition coefficients (log P)[10,151which can be calculated in an additive manner from the Hansch substituent constants x and the octanol-water log P value of a parent [13]. The values for the substituent indices should be thus related to the increments reported for the prediction of octanol-water partition coefficients, although, as noted earlier, the values will differ because of the selectivity of the modifier. This relationship might provide an mechanism by which estimated Z could be obtained for substituents not determined experimentally (e.g. the non-ionized carboxylic acid group) as was demonstrated by Baker (Eq. 1.6). values and The relationships between the initially predicted substituent index the Hansch x constants (Table 1.7) were therefore determined for a range of different eluent compositions [3 1,501. The correlation coefficients (r) were good for methanolbuffer eluents and improved with increasing buffer proportion (r = 0.9607-0.9849) but poorer for acetonitrile-buffer (0.9627-0.9739) [311. Similar results were also obtained for the full set of predicted indices in the three isoeluotropic eluents (Table 1.7). The hy-

21

Retentionprediction based on molecular structure

*0°1 100

I

.

.

CHS

Br

CI c

c

-200

CHO

OR

CO2R CN COR

OH

Acetonltrlle concentratlon (%) Fig. 1.4. Comparison of experimental values of retention index increments (symbols) and calculated aliphatic substituent indices (ZS,R-X) (lines) in acetonitrils-buffer eluents. Points are experimental values and curves are calculated substituent indices [54]. Reproduced with permission.

droxyl and carboxamide groups.had more negative than expected predicted indices and the sulphonamide group was less negative than predicted (Figs. 1.5-1.7). The fragmental constants F, (Table 1.7) are usually regarded as a closer match with the aliphatic contributions to log P [ 131. However, the correlations were poorer ranging from 0.925 to 0.869 in methanol and 0.924 to 0.860 in acetonitrile with the higher correlations being obtained with low proportions of modifier [52].

1.4.2 Polyfunctional compounds It was known from the studies of Hansch constants as predictors of octanol-water partition coefficients, that for disubstituted aromatic compounds containing polar groups, simple summation of the x terms to calculate log P was not very successful [56]. This was assumed to be due to electronic, steric and hydrogen-bonding interactions between the substituents. Initially, this led to the measurement of individual sets of x y values for each parent substituent. However, this approach is impractical for a wide range of combinations of groups as each group would need a separate set of interaction terms. Clearly a References pp. 45-46

Chapter 1

22

2.0

4-

1.o

E

m

CI

In

c

0 0

l= c

0.0

0 In C

m

=

-1.0

-2.0 -400

-200

200

0

400

Substituent index (I) Fig. 1.5 Comparison of Hansch x constants and calculated substituent indices for methanol-buffer (50:SO) based on Table 1.7.

2.0

-

0

, c

1.o 1.0

Ph

,‘

,,

-

c Q

In

,

C

0 0

l= c

0.0 -

,&‘O

0 In

c m

=

,

NHCOCH,

OH0

,‘ ,,

0 ,’ -1.0 CONHCH, ,’ 0 ,’

’

0

-2.0 I --400 -400

0

SO2NH2 I

I

I

I

-200

0

200

400

Substituent index (I) Fig. 1.6. Comparison of Hansch x constants and calculated substituent indices in acetonitrile-buffer (4050) based on Table 1.7.

Retention prediction based on molecular structure

c

23

1.0 -

U‘

c

,

0’

c Q ln

.‘U

c 0

0

t r

0.0 -

0

.‘0

ln

c Q

=

-1.0 -

‘ -2.0 I -400

s

OH

NWOCH,

0 SOzNH? I

I

I

i

-200

0

200

400

Substituent index (I) Fig. 1.7. Comparison of Hansch II values for substituents and calculated substituent indices in THF-buffer (30:70) based on Table 1.7.

similar problem will occur in the prediction of retention in HPLC and it is possible to examine the systems used in QSAR as models for chromatographic prediction systems 1571. 1.4.2.1 General prediction model

In the studies of log P,it was found that, for the meta- and para-isomers, the difference between the n value of a substituent X with benzene as the parent compound and that with phenol as the parent could be described using the Hammett constant CT of the substituent [56]. This led to a more general equation, which attempted to quantify the effect of a group Y on the Hansch constant of a substituent X in terms of their “susceptibility” constants @) and Hammett constants (0)[56,58-591. nX(phY) -nX(F’bH)

‘PgX

‘PflY

(1.12)

p x andp, are the susceptibilitiesof X and Y to the modifying effects of Y and X, respectively. Values for the susceptibility constants (Table 1.9) were derived experimentally using multiple regression analysis, In a similar study Leo [60,6 11 developed a simplified model to facilitate rapid estimation of the interaction terms. He included additional terms were included to account for intramolecular hydrogen-bonding (FHB),the negative ortho effect (F,) and the presence of alkyl-aryl systems (Fa$). Linear regression analysis gave the following correlation (Eq. 1.9) with experimental partition coefficients. References pp. 45-46

24

Chapter I

TABLE 1.9 VALUES OF u AND p USED IN CALCULATIONS OF INCREMENTS Substituent

P

urneta

upara

Inducers CN NO2 Br CI

0.56 0.71 0.39 0.37

0.66 0.78 0.23 0.23

0.00

Bidirectional CHO COzCH3 COCH3 CONH2 OCH3

0.35 0.37 0.38 0.28 0.12

0.42 0.45 0.50 0.36 -0.27

0.44 0.27 0.27 0.72

0.12 -0.16 -0.07

-0.37 -0.66 -0.17 -0.01

1.06 1.08 0 0

Responders OH NH2 CH3 Phenyl

0.06

0.00 0.00

0.00

0.50

Data from Hansch and Leo [13].

Iog P = C x + Fa - 0.29 F m - 0.1SF,$

(1.13)

in which Fa = p (T +p o y. By analogy, Smith and Burr [57]proposed that a corresponding general equation can be derived for the interaction increments in HPLC in which the different components are expressed in retention index units. 4,x-Y = ( a x P; p*, F& and

-Ia* P t ) -I-G B

+ F0*

(1.14)

Fo* correspond to the terms in Eq. (1.13) but are expressed in retention

index units. It is hoped that in each case, they could be directly related through a common regression equation to the eluent composition (Eq. 1.15). p* =p(ax2+ bx + c)

(1.15)

Leo noted the (T constants are valid for up to 80% organic modifier in aqueous solutions [60] and so they should be applicable in the present eluents. In preliminary calculations, it appeared that the meta- and para-interactions differed so that instead of common (T values as suggested by Leo [60,61], published urneta and apWa values (Table 1.9) [I31 were used. The term for alkyl-aryl substitution (Fa@)was omitted as it was thought that this effect had already be covered by the interaction term for alkyl substitution on a benzylic carbon. In order to study the interactions, the retention factors of 7 3 ortho-, metu-, and parasubstituted toluenes and phenols have been examined [57] However, a number of the

Retention prediction based on molecular structure

25

compounds, such as 4-nitrophenol (pK, = 7.1) had to be excluded from the subsequent calculations because they appeared to be partially ionized in the pH 7.0 eluent. The substituted toluenes showed only minor effects, mainly due to steric or electronic interactions. The phenols were expected to demonstrate stronger electronic effects and many were capable of intramolecular hydrogen-bonding. As with the monosubstituted compounds, the retention factors changed significantly with eluent composition but the retention indices for the substituted phenols were relatively constant across the composition ranges (Table I . 10). For most of the substituents the meta and para-isomers were similar (*50 units) but frequently the ortho-isomer was significantly different (up to 400 units). These differences were also reflected in the corresponding octanol-water partition coefficients (Table 1.10). For example, the retention indices of the methyl 2-, 3-, and 4hydroxybenzoates (Z= 940, 681 and 667) and partition coefficients (log P = 2.55, 1.89 and 1.96, respectively, both reflected a significant reduction in the polarity of the ortho isomer due to hydrogen-bonding. However, when the results for the substituted toluenes and phenols were compared they gave separate but parallel relationships (Fig. I A). The separation corresponded to the differences between the comparative values of substituent indices and the n constants for the phenolic hydroxyl and methyl groups. The ortho- interactions of methyl 2-hydroxybenzoate and other 2-hydroxy-carbonyl compounds were so marked that they appeared to behave as if they were substituted toluenes. The retention indices were used to calculate the interactions increments (61 values) between the substituents, as the differences between the measured I value and the summation of the previously determined parent index values for benzene together with the substituent indices (Zs,k-x) for the individual groups [57]. For most of substituted toluenes, the interaction increments were negligible or small. The most significant changes were found for 2-methylbenzamide (for example, 61= -64, -33 and -32 for the 2-, 3-, 4isomers respectively in acetonitrile-buffer (60:40)). In earlier work, Clark and his coworkers reported that this compound was eluted more rapidly than the 3- and 4-isomers, probably due to steric interaction of the 2-methyl group causing the amide group to be less coplanar with the aromatic ring and hence more polar [62]. Many of the substituted phenols showed much larger interaction increments, which changed significantly with eluent composition (Table I . 11). The smallest effects, usually <50 units, were found for the methyl, methoxy, halogen and phenyl substituted phenols. The largest interactions of up to 400 units were observed with the carbonyl substituents capable of hydrogen-bonding, such as 2-hydroxyacetophenone ( 2 5 9 4 19 units, Fig. I .9), 2-hydroxybenzamide (253-372 units), methyl 2-hydroxybenzoate (253-423 units) and 2hydroxybenzaldehyde (2 13-269 units). The increments for these compounds were much higher than for the corresponding 4- and 3-isomers (for example 3- and 4- hydroxyacetophenone 27-144 units). These differences reflect differences in retention factors reported by Clark and co-workers [62,63] for the same or closely related compounds. Smaller but still significant ortho-interactions were also found for 1,2-dihydroxybenzenes (9 1-370 units) and 2-aminophenol (102-341 units). The large increments noted for the nitrophenols were probably due to partial ionization and were not considered further. Using these values it was possible to calculate the corresponding regression correlation coefficients between eluent composition and interaction increments [57] and these were used in early prediction studies [64]. However, they have limited applicability and indiReferences pp. 45-46

TABLE 1.10 RETENTION INDICES OF SUBSTITUTED PHENOLS Compound

Retention index (I)

Log P

Methanol (YO) 40

Acetonitrile (?h)

[13,601

50

60

70

80

30

40

50

60

70

80

2-Aminophenol 3-Aminophenol 4-Aminophenol

576 483 446

592 477 446

580 492 449*

562 467* 429*

569 474* 515

586 519 463

589 522 47 1

600 496 445

618 474 423

635 468 424

711 519 516

0.52 0.15 0.04

2-Bromophenol 3-Bromophenol 4-Bromophenol

813 862 855

810 860 852

809 849 847

777 830 829

734 793 799

82 1 859 853

805 838 83 1

786 815 807

765 792 785

749 772 765

723 743 747

2.35 2.63 2.65

2-Chlorophenol 3-Chlorophenol 4-Chlorophenol

786 834 827

783 833 822

772 823 815

753 803 798

708 767 768

786 836 825

788 815 804

764 792 783

744 769 758

724 749 739

714 73 1 717

2.19 2.48 2.40

1,2-Dihydroxybenzene 1,3-Dhydroxybenzene 1,4-Dihydroxybemne

599 531 469

589 516 474

583 506 461

570 483 * 451*

568 476* 467*

577 522 467

587 532 492

565 498 46 1

558 467 433

553 453 435

687 495 493

1.01 0.77 0.50

2-Hydroxyacetophenone 3-Hydroxyacetophenone 4-Hydroxyacetophenone

853 679 649

860 661 624

867 64 1 594

875 619 55 1

888 614 428*

849 663 615

867 669 606

853 616 569

854 605 562

855 588 556

857 597 562

1.90 1.39 1.30

2-Hydroxybenzaldehyde 3-Hydroxybenzaldehyde

782 656

783 64 1

783 623

769 600

771 57 1

862 648

875 630

834 610

822 592

845 57 1

867 510*

1.81 1.38

9

% is

5 $ c

$

Q o\

4-Hydroxybenzaldehyde 2-Hydroxybemide 4-Hydroxybenzamide

600 65 1 463

565 636 454

507 605 418*

396* 563 384*

243* 518 350*

600 629 420

577 608 393

550 616 356*

537 577 336*

513 568 324*

462* 540 251*

1.35 1.28 0.33

2-Hydroxybenzonitrile 3-Hydroxybenzonitrile 4-Hydroxybenzonitrile

634 693 649

608 678 632

547 658 599

447* 625 529

510 196*

648 711 666

630 688 653

608 665 628

589 639 597

542 610 575

463 606 520

1.60 1.70 1.60

B3 g. 3

b

2

3

2 3

w

2-Methoxyphenol 3-Methoxyphenol 4-Methoxyphenol

715 701 666

711 692 656

705 678 643

698 663 628

682 635 606

720 704 668

712 688 654

697 668 636

693 653 622

682 625 60 1

678 593 565

1.32 1.58 1.39

Methyl 2-hydroxybenzoate Methyl 3-hydroxybenzoate Methyl 4-hydroxybenzoate

947 755 744

959 738 733

975 720 701

989 700 666

1006 654 565

93 1 738 708

944 707 688

940 68 1 667

945 652 649

94 1 633 622

940 635 608

2.55 1.89 1.96

8

E 3

26 r:

$

1 2-Methylp henol 3-Methylphenol 4-methyl phenol

783 775 777

785 768 772

779 764 766

766 748 752

746 725 733

788 772 774

776 758 758

761 74 1 74 1

745 722 723

732 708 713

714 69 1 700

1.96 1.96 1.94

2-Nitrophenol 3-Nitrophenol 4-Nitrophenol

750 744 63 1

739 739 597

706 72 1 519

622 685 379*

228* 543 -

78 1 754 669

778 739 635

783 716 611

789 688 598

766 659 518

64 1 623 395*

1.79 2.00 1.91

992 1007 1006

98 1 989 99 1

969 973 970

947 943 949

922 918 929

984 985 984

973 950 947

939 914 910

916 888 879

884 854 853

847 822 816

3.09 3.23 3.20

2-Phenylphenol 3-Phenylphenol 4-Phenylphenol

Data from Smith and Burr [57]. *Values regarded as unreliable because of extent of extrapolation.

z2 F m

28

Chapter I

4-

3-

, M0

n

' A 301 * ' 400

4-NY I

I

I

1

600

800

1000

1200

Retention index Fig. 1.8. Relationship of log P and retention indices for derivatives in methanol-buffer (60:40). Analytes; 0 , substituted toluenes; A,phenols; A, suspected ionized phenols [57]. Reproduced with permission.

n

(W" 800 -

u *

n

t-

0

700 -

--a\--&-*-

600

500

400

!

I

I

I

I

1

30

40

50

60

70

80

% MeCN

+

Fig. 1.9. Comparison of calculated (IS& and experimental retention indices for 0, 2-; 0,3-; A , 4hydroxyacetophenone in acetonitrile-buffer eluents [57], Reproduced with permission.

Retention prediction based on molecular structure

29

vidual sets of regression coefficients would be needed for each substituent group. It was therefore necessary to determine if the proposed generalized relationship (Eq. 1.14) described earlier would be valid. 1.4.2.2Meta andpara groups

In order to determine the relationship between p and the chromatographic equivalent p* value in Eq. 1.14, the increments for the substituted phenols were examined. The phenolic group is a responder group @I = 1 . 0 6 , =~ relatively small) (Table 1.9) so that if o could be assumed to be zero, Eq. (1.14) can be redefined for meta- and para-substituted phenols (Eq. 1.16).

Thus in each eluent there should be a close relationship between the empirical interaction increments 61 and ax.In methanol-buffer (5050) a good linear correlation was found for the inducer and bi-directional substituents (para-, Fig. 1.10a and meta-, Fig. 1.10b). However, the amino and hydroxyl substituents, which are responder groups (and the formyl and nitro substituentswhich gave ionized compounds) were clearly outliers. The o values also correctly forecast the sign of the increments. Negative values of apXa for methyl, methoxy, and phenyl groups and of a,, for the methyl group were matched with negative retention increments and the positive value of a, for the methoxyl group was matched by a positive increment. To determine the values of the coefficients a, b and c in Eq. (1.16), the ratios 6 1 x / ( p o H ~were x ) then calculated for each substituent (using poH= 1.06 but excluding the hydroxyl and amino groups) and were correlated with the proportion of modifier. The mean values of the ratios fiom the different substituents were virtually independent of the percentage of methanol and suggested that the relationship for methanol-buffer eluents could be represented by a single value rather than a quadratic expression, hence for methanol-buffer; P*,,~ = loopx, p*meta-x = 170px,and for acetonitrile-buffer; p*,, = 105px, p*meta-x= 190px. The bromo and chloro groups in acetonitrilebuffer eluents were the exception to these correlations and their ratios changed systematically with eluent composition so that P*para-hdogen -Phalogen (244 - 4x1 and P*rneta-halogen =Phalogen(l75 2.6~). Except for the halogens, these correction ratios suggest that for most substituents the interaction increment can be assumed to be a constant irrespective of the proportion of modifier. This corresponds well to the empirical interaction increments (Table 1.11) many of which were almost constant across the eluent ranges forpara- and meta-isomers. Using these ratios, predicted interaction indices could be calculated [57] and these matched reasonable well with the experimental increments for the substituted toluenes and phenols (c30 units difference). However, the interactions between the hydroxyl and amino phenolic substituents had much larger errors around 100 units. Similar problems with these groups were also noted by Fujita [59] and they had to be regarded as outliers. Similar calculations were also carried out on a more limited scale for THF eluents [52] but the correlations were poorer and it was felt that using Hammett values for the prediction of interactions terms would probably be unsatisfactory with this eluent. References pp. 45-46

w

0

TABLE 1.11 INTERACTION INCREMENTS DERIVED FROM SUBSTITUTED PHENOLS Substituent pairs

Interaction increment (dl) Methanol (Yh)

Acetonitrile (YO)

40

50

60

70

80

30

40

50

60

70

80

133 40 3

162 47 16

178 90 47*

199 104* 66*

259 164* 205

102 35 -2 1

135 68 17

173 69 18

213 69 18

250 83 39

341 149 146

OH + 2-Br OH + 3-Br OH + 4-Br

-9

-9 41 33

-1 39 37

-19 34 33

-45 14 20

-4

34 28

-6 27 20

-12 17 9

-22 5 -2

-27 -6 -1 1

-44 -24 -20

OH + 2-CI OH + 3-CI OH + 4-CI

-7 41 34

-6

-6

45 37

-8 42 37

-32 27 28

-10 40 29

6 33 22

-5

44 33

23 14

-14 11 0

-23 2 -8

-24 -7 -2 1

OH + 2-OH OH + 3-OH OH + 4-OH

125 57 -5

134 61 19

160 83 37

190 103* 71*

248 156* 147*

91 36 -19

146 91 51

164 97 60

190 99 65

214 114 96

370 178 176

OH + 2-COCH3 OH + 3-COCH3 OH + 4-COCH3

259 85 55

289 90 54

325 99 52

367 111 43

419 144 -41*

26 1 75 27

310 112 49

323 86 39

346 97 54

363 96 64

380 120 88

OH -+ 2-CHO OH + 3-CHO OH + 4-CHOa

213 87 31

237 95 19

265 105 -1 1

283 114 -90*

319 119 -209*

294 80 32

332 87 34

316 92 32

325 95 40

369 95 37

410 53* 5*

OH + 2-NH2 OH + 3-NH2 OH + 4 - m ~

40

33

25 58 b

2

OH+2-CONH2 OH+4-CONH2

253 65

273 91

284 97*

289 110*

OH+2-CNa OH+3-CN OH+4-CNa

64 123 79

58 128 82

28 139 80

-32* 146 50

OH+2-OCH3 OH + 3-OCH3 OH + 4-OCH3

36 22 -13

38 19 -17

44 17 -18

OH + 2-CO2CH3 OH + 3-CO2CHs OH + 4-CO2CH3

253 61 50

285 64 59

4

296 128*

262 53

315 100

372 112*

354 113*

341 97*

282 -7 *

80 -234*

47 110 65

55 113 78

63 120 83

70 120 78

50 118 83

139 53

54 19 -16

59 12 -17

32 22 -14

47 23 -1 1

54 25 -7

69 29 -2

76 19 -5

87 2 -2 6

327 72 53

371 82 48

423 71 -1 8

253 60 30

294 57 38

315 56 42

340 47 44

355 47 36

367 62 35

-1 -16 -14

-3 -2 1 -17

-5

-26 -1 8

-10 -2 6 -24

-8 -2 6 -2 6

-10 -3 0 -3 0

-15 -3 8 -3 7

-17 4 1 -3 6

-26 -49

-4

CII

I

2

OH + 2-CH3 OH + 3-CH3 OH + 4-CH3

-2

1 -16 -12

OH + 2-N02a OH + 3-No2 OH + 4-N0za

106 100 -13

109 109 -3 3

98 113 -89

43 106 -200*

-317* -2 -

124 97 12

148 109 5

182 115 10

216 115 25

223 116 -25

128 110 -118*

-8 7 6

-8 0 2

-5 -1

-10 -14 -8

-15 -19 -8

-5

18 -5 -8

13 -12 -16

13 -15 -24

-2 -3 2 -33

-2 5 -50 -5 6

OH + 2-Ph OH + 3-Ph OH + 4-Ph

-4

-4

-4

-5

-40

G-

E

b 8

4

P

sP 2

Based on data in Table 1.10 from Smith and Burr [57]. *Values regarded as unreliable because of extrapolation. %ompounds suspected of being partially ionized.

W e

32

Chapter 1

140

-

z c

E

L

0

.-c e

.-cO.

a

120

100 80 60 -

20 -

4. .0

0

0

0

OH

0-cre -20 -40 2 d)

0 NO2 I

140

I

I

1

I

I

1 1

1 .o

b

z c.

e

100-

80-

L0

60 -

.-e c .-0

40 -

0

20

CI

PCI

-

0-

c

-20

-

!

-40 I I -0.75 -0.50 -0.25

I

I

I

I

I

i

0.00

0.25

0.50

0.75

1.0

Qm Fig, 1.10. Relationship between interaction increment (So y d u values in methanol-buffer (50:SO) eluents. (Open symbols were regarded as outliers and were not used in the correlations) [57]. Reproduced with permission. (a) para-substituted phenols; (b) meta-substituted phenols.

1.4.2.3Ortho-substituents When the ortho interaction increments were examined [57], to eliminate electronic effects, it was assumed that the substituents experienced the same effects as the para-

Retention prediction based on molecular structure

33

substituents. For most of the ortho-substituted toluenes, except for 2-toluamide, there were only small residual differences (+lo to -30 units), corresponding to the ortho interactions (F,), which could be largely disregarded. The ortho-interaction increments for the substituents on phenol (Table 1.11) were generally much larger. The major effects appeared to be hydrogen-bonding, which could be divided into three groups, weak interactions from methoxyl and possibly from nitro substituents, medium interactions with hydroxyl and amino groups (150-250 units) which were very dependent on the eluent composition, and strong interactions (200-400 units) with the carbonyl containing substituents, which also could change markedly with eluent composition (by up to 150 units) (Fig. 1.9). These terms will be a combination of hydrogen-bonding (Fm) and ortho (F,) effect but these cannot easily be distinguished. It will be difficult to develop general rules for these pairs of substituents and it appears that prediction of hydrogen-bonding effects will need to described by comparison with the empirical relationships. Because the structural features causing these interactions are well defined, in a prediction system it will be possible to make specific rules for these groups. Similar interaction terms have not been calculated for aliphatic substituents although clearly if hydrogen-bonding were possible, it would be expected to have a significant effect. 1.4.3 CRIPES and expert systems In order to provide a user interface to the coefficient data (in Appendix 1.1) for the different substituents and their interaction data, an expert system program CRIPES (Chromatographic Retention Index Prediction Expert System) was implemented using a commercially available shell, VP Expert [64,65]. This package was designed to ensure that the user asked the correct questions about the structure of the analyte and the relationships of the substituents. A version of the package could also compare the results for two analytes and determine the resolution that would be expected in each case. Finally the user could examine the predictions and select the composition giving the preferred resolution and overall retention times. The regression coefficient data were held as spreadsheets, which enabled them to be updated externally to the main programme. The CRIPES programme was based on Eq. (1.9). Each term in the equation can be described using the three coefficients of a quadratic equation, which can be summed to give an overall relationship (Eq. 1.11) between eluent composition and retention index for each modifier. The package therefore accumulated the coefficients for the individual terms from spreadsheets of data in response to replies from the user to a series of questions (Fig. 1.11). The databases incorporated retention index terms for methanol and acetonitrile-buffer but at the time the programme was developed only empirical interaction indices were available. These were incorporated into the programme and were activated whenever a matching pair of substituents were present. Subsequent work has added additional substituents, terms in THF-buffer eluents, and work is planned to extend the programme to include the calculation of the interaction terms but these developments have not yet been fully assessed. Problems have been encountered with changes that can cause ionization and with the anomalous effects of the hydroxyl and amino groups. The first stage in the expert system programme obtained the coefficients for benzene in the different modifiers from a file. The user then selected substituents from menus of the References pp. 45-46

34 b

Chapter I

USER INPUTS NAME AND SUBSTITUENTS PRESENT

I

IDISPLAY RI AND APPROX K' I Fig. 1 . 1 1, Flow chart of steps in the operation of CRIPES (Chromatographic Retention Index Prediction Expert System) [64,65].

aromatic and aliphatic groups offered by CRIPES. The program asked the user to input the number of each aromatic substituent present in the compound and their positions relative to hydroxyl, amino (assumed initially to interact as an hydroxyl group) or alkyl groups, so that any interaction terms could be added. Finally the program prompted the user for information on the length and branching of any alkyl chain and for the position of aliphatic substituents relative to the phenyl group. These coeEcients for these groups were retrieved from the spreadsheets and summed to provide overall equations for methanol and acetonitrile eluents. The retention index values over the ranges 4040% methanol and 30-80% acetonitrile were then calculated at 10% intervals. The last stage in the program calculated approximate retention factors (k) to give an indication of the time required for separation. This calculation is based on the relationship (log k = AZ+ B) (Eq. 1.8), in which A and B were known from the experimental regression equations of the alkyl aryl ketone standards on that particular column. The ability of the initial programme to prediction retention indices was tested in two ways [64]. Firstly, the retention indices of a number of model compounds were used to confirm that it could obtain the correct indices for individual groups. Then a number of new test compounds were examined and their retention indices were calculated and compared with experimental values (Table 1.12). In most cases there was a reasonable fit

Retention prediction based on molecular structure

35

between experimental and calculated values (Fig. 1.12). Particular difficulties were found with rapidly eluted compounds and the prediction often represented considerable extrapolation from the retention index standards so the experimental values (especially <600) may be unreliable. Problem compounds and groups were noted. Some of these, such as secondary and tertiary amides and secondary amines, which showed large deviations, had been calculated as alkyl substitution on benzamide or aniline. The corresponding groups have subsequently [50,52] been added to the database as separate terms (CONHR, CONR2and NHR) (Appendix 1.1 and Table 1.7). The significant differences between predicted and experimental values with the phenacyl and haloacetophenones suggested that an aliphatic interaction may be present (Table 1.12). Other disubstituted aliphatic compounds (benzyl chloromethyl ether and ethyl phenylcyanoacetate) both showed large deviations suggesting that electronic and steric effect may be also important for aliphatic compounds. Some of the problems with the substituted anilines may have resulted from ionization of the corresponding phenolic compounds, which were provisionally used as models. Aromatic monoesters gave a good match but phthalate esters showed significant deviations suggesting that there was steric interactions between the carboxylate groups. Many simpler compounds, e.g thymol, 2,5dimethylphenol and 3,4-dimethoxyacetophenone,had calculated retention indices close to the experimental values. Although this prediction method based on retention index increments with different modifiers cannot yet compensate accurately for interactions between groups, it should

,

1200 -

,

/

/

X

4j 1000 .-c C

.-0

4.4

C 0,

800

-

3 600

-

E -0 Q) 4.4

-m3 0

0

400

600

800

1000

1200

Experimental retention index Fig. 1.12. Comparison of experimental retention indices and values calculated by CRIPES for methanol-buffer (60:40) [64].

References pp. 45-46

W

m

TABLE 1.12 EXPERIMENTAL RETENTION INDICES (Ie) AND CALCULATED RETENTION INDICES (I,) PREDICTED BY CRIF'ES Compound

Retention index Acetonitrile (YO) 40

50

60

70

80

~

Phenacyl bromide Benzyl2-bromoacetate a-Bromo-p-phenylacetophenone a-p-Dibromoacetophenone 4-Nitrophenacyl bromide a-Chloro-3,4-dihydroxy acetophenone

990

883 645

763

878

75 1

1028 898

888 693 593

1135 1015 874 449

866 841 778 742 701 929

939 839 812 742 653 897

873 788

877 947

883 639 538

832 886 1075 972 784 393

746 775 977 896 618

855 820 748 709 663 940

919 816 815 707 553 895

846 794 720 668 639 925

903 801 797 675 459 893

859 624 919

841 758 980

844 629 899

829 750 963

857 922

745 786

994 880 664

994 830

860 828 762 723 669 928

930 828 820 729 618 896

863

869

852

998

934

992

855

o-Bromoaniline m-Bromoaniline o-Nitroaniline m-Nitroaniline p-Nitroaniline N-Ethylaniline Benzyl acetate Benzyl chloromethyl ether 1-Bromo-2-nitrobenzene

89 1

2-Bromo-2-methylphenol t-Butylhydroquinone p-t-Butylphenol

2-Chloromethyl-4-nitrophenol 4-Chloro-2-nitroaniline 4,6-Dichloro-l,3-dihydroxybenzene 3,4-Dimethoxyacetophenone N,N-dimeth ylbenzamide 2,6-Dimethyl-4-nitrophenol 2,4-Dimethylp hen01 2,5-Dimethylphenol Dimethyl phthalate Ethyl benzoate Ethyl 3-phenylpropionate Ethyl phenylacetate Ethyl phenylcyanoacetate 2-Hydroxybenzyl alcohol 4-Hydroxybenzyl alcohol 2-Hydroxy-5-nitrobenzylbromide N-Methylbenzamide p-Nitrobenzyl alcohol m-Nitrobenzyl alcohol 4-Phenyl-1 -butanol 5-Phenyl-1-pentan01 n-Propyl p-hydroxybenzoate Thymol

653 770

735 791

1036

1052

914

763

589

635

1001

1008

863

897

943 432

974 642

63 1 743 83 I 829 797 977 1030 947

714 759 830 830 850 995 1042 963

353 560 638 668 82 1 903 862 984

663 614 549 561 856 956 850 99 1

846 978 924 567 881 627 696 633 699 717 817 784 987 1021 940

889 768 958 592 925 697 729 714 724 811 811 841 996 1031 952

796 520 303 552 628 643 801 883

350 337 615 614 530 542 830 930

968

972

823

887

905 332* 868 60 1

943 540 906 673

643 528 803 795 764 99 1 1011 92 1 819

264* 549 624 54 1 807 886 779 955

737 686 799 799 834 997 1020 94 1 655

569 637 516 528 829 929 787 95 1

812 974 882 281* 846 574 697 683

892 783 929 484 876 658 702 781

744 783 750 994 999 905 754 80 1 463 233* 558 599 61 1 797 879 775 94 1

768 768 828 996 1008 929 618 297 324 525 682 507 5 19 836 936 773 929

0-

f

-iL 3 0

~~

Values from Smith and Burr [64]. *Values considered to be unreliable because the corresponding retention factors are less than 0.2.

38

Chapter I

provide a better approach than many related techniques, because it takes into account the solvent selectivity differences and can make some compensation for hydrogen-bonding effects. Further developments will concentrate on general relationships for interaction expressions, such as the alp relationship, so that reasonable values can be determined by calculation. 1.4.4 Comparisons with published retention values

In order to demonstrate a broad applicability for retention indices, both as a method for recording retentions and for retention prediction, a number of comparisons have been carried out with index values from different systems. Further examples are described in Chapters 3 and 4 as a way of comparing columns or for specific groups of analytes. During studies of the ability of retention indices to compare retention on different brands of ODs-bonded silica [66],retention indices were measured on eight columns using three isoeluotropic eluents. The results (Table 1.13) showed only small variations between the stationary phases but major difference between mobile phases. As well as previous reports of retention indices, it is frequently possible to calculate or estimate retention indices from published sets of retention factor data. If sufficient alkyl TABLE 1.13 RETENTION INDICES DETERMINED ON DIFFERENT BRANDS OF ODS BONDED SILICA COLUMNS Compound

Retention index (I) Columna H3

H5

T

Methanol-water (70:30) 2-Phenylethanol 779 Nitrobenzene 857 p-Cresol 794 Toluene 1055 Methyl benzoate 910

783 862 810 1062 909

754 875 776 1046 902

743 838 759 988 895

702 843 703 1040 906

Acetonitrile-water (50:SO) 2-Phenylethanol 694 Nitrobenzene 888 p-Cresol 766 Toluene 1030 Methyl benzoate 888

694 891 760 1027 881

692 880 751 1019 890

697 873 745 1009 889

THF-water (4050) 2-Phenylethanol Nitrobenzene p-Cresol Toluene Methyl benzoate

727 913 885 1088 888

738 926 877 1069 893

726 924 873 1051 888

74 1 92 1 885 1076 897

S

Z

P

L

s2

767 849 753 965 902

758 843 777 1055 913

74 1 874 748 1063 910

673 859 719 1027 887

696 870 720 984 881

69 1 874 747 1026 890

675 87 1 74 1 1036 900

713 922 860 1086 888

745 956 889 1042 893

717 927 873 1083 897

906 1083 881

Values from Smith [66] except for column S2, which were from Smith and co-workers [31,50,54,57]. aColumns: H3, 3 p m ODS Hypersil; H5, 5 p m ODS Hypersil; T, Techsil 5 C-18; S, Spherisorb ODs; Z, Zorbax ODs; P, Partisil 10 ODs; L, Lichrosorb RP-18; S2, SpherisorbODS2.

Retention prediction based on molecular structure

39

TABLE 1.14 CALCULATED RETENTION INDICES BASED ON PUBLISHED RETENTION FACTORS OF ALKYL ARYL KETONES Compound

Acetophenone* Propiophenone* Butyrophenone* Methyl benzoate Benzene Naphthalene Toluene Ethylbenzene i-Propylbenzene Phenyl ethylacetate Benzyl ethyl ether Chlorobenzene Bromobenzene Iodobenzene 2-Chlorotoluene 3-Chlorotoluene 2-Bromotoluene 4-Iodotoluene 2-Hy droxy acetophenone 4-Hydroxy propyl benzoate Benzyl alcohol Aniline Benzamide Nitrobenzene

Acetonitrile (“h) 30

40

50

60

70

80

90

800 900 1000 885 889

800 900 1000 887 904 1091 1000 1087

800 900 1000 890 913 1096 1010 1102 1188 930 959 1010 1039 1085 1119 1119 1148 1205 639 833 639 833 525 856

800 900 1000 888 925 1102 1021 1117 1199 925 969 1028 1058 1110 1132 1132 1162 1228 665 814 65 1 836 532 873

800 900 1000 887 907 1097 1007 1107 1197 897 977 1017 1067 1117 1137 1117 1167 1227 667 757 657 807 517 837

800 900 1000 905 92 1 1136 1059 1167 1244 905 1013 1044 1090 1151 1197 1197 1244 1305 690 782 659 797 582 828

800 900 1000 907 907 1127 1007 1107 1227 867 1027 1027 1067 1167 1157 1127 1247 1307 767 767 727 767 647 787

989

957 945 998

896 952 1004 1035 1083 1104 1109

670 90 1 676 75 1 60 1 864

648 870 648 770 561 865

Based on data reported by Hanai et al. [ 671. *Alkyl aryl ketones used as standards for retention index scale.

aryl ketones (or other sets of homologues) are included then these can be used directly as standards. For example, retention indices for a range of compounds on an ODS Chromosorb LC-7 column with acetonitrile-water eluents can be calculated fiom the work of Hanai and co-workers (Table 1.14) [67]. If only acetophenone is included in the data set, methyl benzoate can often be used as a secondary standard to give estimated indices. Its retention index on the alkyl aryl ketone scale appears to be particularly independent of the stationary phase (Table 1.13) [66]. This method was used to derive retention indices (Table 1.15) from the extensive retention factor data set of Schoenmakers and co-workers [68] across a range of eluent compositions. The results compared well with directly measured experimental values (such as o-cresol, Table 1.10) [3 1,571. From some of these data sets it is also possible to calculate substituent and increment indices and in an unpublished study by Smith, Burr and Wang these corresponded closely to the CRIPES data set. These results have important conclusions for retention prediction, as it means that a common set of retention coefficients (as in Appendix 1.1) could be used for many stationReferences pp. 45-46

Chapter 1

40

TABLE 1.15 ESTIMATED RETENTION INDICES CALCULATED FROM PUBLISHED RETENTION FACTORS OF ACETOPHENONE AND METHYL BENZOATE Compound

Acetonitrile(%) 30

Acetophenone* Anisole Benzaldehyde Benzene Benzonitrile Benzophenone Benzyl alcohol Biphenyl Chlorobenzene 4-Chlorotoluene o-Cresol Dimethyl phthalate 1,2-Dinitrobenzene 1,3-Dinitrobenzene 1,4-Dinitrobenzene Diphenyl ether Ethylbenzene 4-Hydroxybenzaldehyde 4-Methylbenzaldehyde Methyl benzoate* Naphthalene Nitrobenzene 2-Nitrophenol 3-Nitrophenol 4-Nitrophenol Phenol 2-Phenylethanol Toluene

800 897 788 895 819 661 1002 789 835 891 870 871

637 866 890 870 833 774 749 698 718 999

40

50

60

70

80

90

800 908 788 918 819 1055 640 1166 1018 1110 788 823 881 872 878 1189 1097 594 876 893 I099 885 837 755 800 689 694 1011

800 923 789 933 818 1060 608 1184 1034 1I34 764 81 1 865 866 873 1200 1115 551 867 895 1109 887 832 728 702 664 668 1030

800 916 794 930 816 1033 629 1166 1022 1111 752 800 836 848 849 1155 1095 539 870 896 I090 815 827 706 688 655 663 1012

800 915 796 938 815 1044 618 1180 1034 1120 735

800 914 787 930 787 1012 606 1151 1024 1116 702 766 744 766 759 1124 1192 412 872 896 I085 849 800 606 565 593 683 1024

800 894 790 915 768 983 632 1101 999 1086 694 733 694 708 694 1069 1056 450 863 894 1052 819 779 554 53 1 575 694 I004

808 818 818 1151 486 867 897 1106 86 1 825 66 1 625 633 647 1032

Calculated from retention factors reported by Schoenmakerset al. [68]. *Based on acetophenone (I=800) and methyl benzoate (Icalculated from Appendix 1.1) as standards.

ary phases, and could be converted to retention factors by direct comparison with a single injection of a mixture of alkyl aryl ketone standards (as was carried out as part of CRIPES). However, if an analyte is being retained by a mixed mode of retention in which reversed-phase retention is combined with interactions with free silanols, then the results may be more dependent on the stationary phase. 1.5 OTHER RETENTION INDEX PREDICTION STUDIES

An alternative but related approach to retention prediction has been described by Hindriks and co-workers [69]. They expressed functional groups in terms of the percentage of methanol required to give a retention factor between 3 and 10. In recent work, this ap-

Retention prediction based on molecular structure

41

TABLE 1.16 RETENTION INDICES OF FRAGMENT @IF) FROM REGRESSION ANALYSIS

Fragment

CH3 CH2 CH C CgH5 Ar-0-R Ar-CI Ar-F Ar-OH Ar-CO-R

ArCONHR Ar-CONR2 R-CO2-R

RIF 87 100 84 -32 486 -88 92 -5 1 -100 0 -358 -339 -277

4,xa 100

305 -111 105 6 -229 -13 -378 -299 -258

Based on data from Wehrens using the alkan-2-one scale [70]. aSubstituent indices (Is. X) in methanol-water (50:50) from Table 1.7 [31].

proach has been refined by Wehrens and co-workers [69]. They have used a regression analysis of the retention indices of 350 CNS-active drugs based on the allcan-2-one retention index scale to determine fragmental retention index values @IF) (Table 1.16), which are used to determine predicted retention indices and hence estimated eluent composition by comparison with standards. The values compare well with the calculated values listed in Table 1.7. In studies of the separation of substituted phenols, Yamauchi [71] determined the mean values for the retention index for compounds across a range of eluent compositions based on the n-alkyl4-hydroxybenzoateretention index scale (Table 1.17). By determining the retention indices of standards, they were then able to predict the retention of the individual compounds with a reasonable accuracy over a range of eluent compositions. However, they did not report the index values for phenol as the parent compound. If a typical value for the increment (61 = 50 in methanol-buffer (50:50)) between phenol and 4-carbomethoxyphenol is taken from Tables 1.5 and 1.10, it gives an estimated index value for phenol of ZP = 60. This can be used to calculate the estimated retention indices (I-) on the alkyl aryl ketone scale and the corresponding substituent increments (Table 1.17). As expected from the earlier studies, 2-acetylamido (-NHCOCH3) causes a much smaller increment (61 = -103) than the 4-substituent (61= -208). Most of the other groups corresponded well to the retention indices recorded earlier for substituted phenols (Table 1.10). 1.5.1 Retention prediction based on polynuclear aromatic hydrocarbons

Although, the polynuclear aromatic scale (see Chapter 3) is not based on a homologous series, the increments of increasing ring number are systematic. Pop1 et al. [72] have used the retention indices measured on this scale to determine increment indices for aryl subReferences pp. 4 5 4 6

Chapter I

42 TABLE 1.17 RETENTION INDICES OF SUBSTITUTEDPHENOLS AND SUBSTITUENT INCREMENTS Substituent

Retention index

61

Estimated

I (Table 1.10)

( I d 60

-

683.

-148 -93 -43 15 110 148 213 244 270 376

-208 -153 -1 03

475 530 580 638 733' 77 1 836 867 893 999

Phenola 4-NHCOMe 3-NHCOMe 2-NHCOMe 4-COMe 4-CO2Me 3-Nq 4-CI 4-Br 4-COPh 4-Ph

-45

so

88 153 184 210 3 16

733 739 822 852 99 1

Values from Yamauchi [71] using the parabens retention index scale. aEstimated value of phenol (60 = 110 - 50) determined from index value for 4-CO2CH3 of I10 by comparison of phenol (Table 1.5) and methyl 4-hydroxybenzoate (Table 1.10) in methanol-buffer (5050) (683 -733 = 50) on the alkyl aryl ketone scale. *Defined terms. TABLE 1.18 RETENTION INDEX INCREMENTS FOR FUNCTIONAL GROUPS ON BENZENE RING BASED ON THE POLYNUCLEAR AROMATIC RING SCALE Group

Group contribution (Alog I): stationary phase Polystyrene gel MeOH-water

I Br

NO2 F CH3 CN COCH3 CHO OH CO2CH3 NH2 SH SO3H C02H

Sepharon gel MeCN-water

0.73 0.49 0.46 0.32 0.28 0.23 0.22 -0.18

0.89 0.65 -1.70 0.42 -0.10 0.12 0.33 -0.54

-0.25 -0.37 -0.40 -0.45 -0.48 -0.70 -0.85

-0.57 -0.84 -0.26 -0.65 -0.71 -0.29 -0.95

MeOH-water

MeCN-water

THF-water

0.36 0.31

0.33 -0.16

0.69 -0.03

0.23

0.22

-0.52

-0.21

-0.60

0.77

-0.58

-0.95

-2.06

-0.50

-0.98

-1.83

Values from crosslinked polystyrene gel column; eluents, methanol-water or acetonitrilewater (70:30) or (60:40) [72]. Values from Sepharon SE macroporous gel column; eluents, methanol-water (80:20), acetonitrile-water (63:37) and THF-water (S0:SO) [73].

Retention prediction based on molecular structure

43

stituents (as Alog I, Table 1.18) in using methanol-water and acetonitrile-water either (70:30) or (60:40) according to their retentions. The values of the indices and increments were virtually independent of the mobile phase composition. The relative magnitudes of the contributions were similar to those determined using other scales but the absolute values naturally differed. The values were then used to predict retention indices for di- and polysubstituted benzenes and naphthalenes. A good correlation was found except for some nitrophenols, polyphenols and amino compounds. However, the pH of the eluent was uncontrolled and as noted in earlier studies, some of these compounds may be partially ionized or the substituentsmay be hydrogen-bonded. Very similar values were later obtained on a commercially available polystyrene gel [73] and were used for the prediction of the retention indices for substituted nitrobenzenes. However, if the substituents could interact, the correlations were often poor with deviations between 0.50 and 0.83 in methanol and acetonitrile eluents and up to 1.83 in THF eluents, with the worst results being obtained with nitrophenols and nitroanilines. Although the nitro groups gave a positive contribution to substituted benzenes in methanol-water, in acetonitrile and THF-water. The situation was confusing and it contributed negatively if the initial substituents were non-polar but positively if they were acidic or basic. Hasan and Jurs [74] employed regression analysis to correlate the retention indices of aromatic hydrocarbons to their shape and size expressed as number of aromatic rings, largest principal axis, smallest principal axis and the frequency of selected clusters. Good correlation could be obtained but different “best” regression were obtained for monomeric and polymeric ODs-bonded stationary phases. The regressions were applied to a further 18 aromatic hydrocarbons and in most cases the prediction was within the error of the calculation (Alog I = *0.26). The principal errors were found for ethyl substituted hydrocarbons but none of these had been included in the training set. 1.6 CONCLUSION

Retention index increments for substituents can be successfully used to accurately prediction the retentions of a wide range of analytes. A common set of relationships can be used for many different brands of stationary phase. Methods have been developed to take into account interactions between substituents including electronic- and hydrogen-bonding effects. Deviations between experimental values and observations can be used to study the magnitude of interactions. The predicted indices reflect the selectivity differences between organic modifiers and match closely values derived in other studies. APPENDIX 1.1 : COEFFICIENTS OF REGRESSION EQUATIONS FOR THE EFFECT OF ELUENT ON PARENT, AROMATIC AND ALIPHATIC SUBSTITUENT INDICES

Coefficients are valid for composition ranges: methanol-buffer (40:60) to (80:20), acetonitrile-buffer (30:70) to (80:20) and THF-buffer (2090) to (60:40). Index value I = a 2 + bx + c. References pp. 45-46

Chapter I

44 Substituent

Coefficients Methanol-buffer a

b

Parent index (I,) Benzene -0.0121 3.887 Naphthalene 0.0179 -0.1330

Acetonitrile-buffer C

748 1065

a

b

THF-buffer a

C

b

C

-0.0154 -0.0068

2.761 1.095

841 1068

0.0193 0.0543

1.727 4.563

904 1208

-0.0964 -0.1228 -0.0643 -0.0564 -0.0800 -0.0721 -0.0800 -0.0307 -0.1 125 -0.0457 -0.0271 -0.0514 -0.0121 -0.0743 -0.0343 -0.0157

0.0 -0.0429 0.0243 0.0150 0.0250 -0.0737

3.024 4.009 -1.037 -1.196 0.300 3.601 2.040 -0.133 6.055 0.437 -0.769 0.454 -1.099 4.423 2.043 -2.283 0.0 0.0 2.409 -3.763 -3.490 -5.890 4.996

-303 -389 -377 494 -315 -259 -131 -133 -307 -105 -218 -23 -86 -258 -213 -97 0 100 -18 186 208 395 6

-0.0357 -0.0250

-6.803 -4.050

-296 -290

-0.0520 -0.0 107 -0.0629 -0.047 1 -0.0186 0.0075 0.0075 0.0 -0.0874

-0.145 -1.793 0.769 -0.129 -0.194 -3.665 -3.525 0.0 1.779

-265 -300 -252 -304 -304 76 101 100 -108

Aromatic substituent indices( Is, A-X) 0.0986 -15.170 138 SO2NH2 -104 0.0093 -4.804 CONH2 -0.0200 -1.120 -272 CONHR -0.1710 -1.703 -27 1 CONR;? 0.0521 -9.267 -7 NHCOR -0.0264 0.541 -215 NH2 -167 -0.0271 0.117 OH 0.0186 -4.469 CHO 39 OCOR -0.0114 -1.429 -34 CN 0.0114 -3.791 -52 COR 0.0050 -2.390 53 NO2 0.0129 -2.263 -3 0 OR 0,0000 -1.7 -155 NHR 0.0229 -2.633 -82 NR2 43 0.0143 -3.774 C02R 0 0.0 0.0 H 0.0 0.0 100 CH3 -0.0243 1.494 F -8 0.0086 -1.669 CI 167 0,0150 -2.190 207 Br 0.0250 -3.870 Ph 436 CH:CHR 0.0 0.0 139

-0.0052 0.1260 0.721 0.1086 0.0271 0.0118 0.0218 0.0025

-2.882 -14.878 -10.287 -13,208 -5.562 -2.405 -4.616 -1.335

-218 2 -143 -199 -216 -153 -93 -92

-0.0025 0.0150 -0.0104 0.0029 -0.0070 0.0102 0.0105 0.0 0.0 -0.0145 0.0 0.0 0.0193 0.2230

-1.251 -2.704 -0.586 -1.097 -0.168 -0.937 -2.096 0.0 0.0 0.400 0.0 0.0 -3.299 2.741

-5 7 -143 -14 -80 -191 -133 -67 0 100 12 98 127 3 72 189

Benzylic groups (Is,k.-d -0.0193 -0.456 CHzOH -0.0171 -1.663 CH2CN -0.0171 0.437 CH2Cl 0.0314 -4.571 CH2Br

-148 -1 8 86 240

0.0513 0.0002 -0.0030 -0.0012

-6.872 -2.543 -1.586 -1.711

-9s -9 140 168

Aliphatic substituent indices (Is,R-x) CONH2 0.0171 4.497 OH (prim) -0.0257 -0.494 OH (secter) -0.0164 -1.979 CN -0.0250 -1.050 CHO 0.1314 -18.531 C02R 0.0071 -3.717 COR -0.0086 -1.851 OR 0.0136 -2.939 CI -0.0021 -1.053 Br -0.0536 4.719 CH3 0.0 0.0 RCHCHR -0.0014 -0.2886

-150 -273 -254 -185 337 -90 -201 -154 19 -99 100 136

0.0491 0.0561 0.0371 0.0002 0.0073 0.0130 0.0161 0.0211 0.0018 0.0064 0.0 0.0100

-9.182 -8.139 -6.41 1 -2.997 -1.768 -3.580 -3.654 -2.644 -1.962 -2.161 0.0 -1.78

-2s 1 -223 -297 -160 -184 -164 -206 -218 21 63 100 154

0.0

Retentionprediction based on molecular structure

45

1.7 ACKNOWLEDGEMENTS

The author thanks Christina M. Burr, Rui Wang and Yuan Wang for gathering and interpreting most of the work on the CRIPES prediction system and the Science and Engineering Research Council for financial support of this study.

1.8 REFERENCES 1

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

J.L. Glajch and L.R. Snyder (Eds.), Computer-Assisted Method Development for High-Performance Liquid Chromatography,Elsevier, Amsterdam, 1990 (reprint of J. Chromatogr.,485 (1989) 1-672). T. Hamoir and D.L. Massart, Adv. Chromatogr., 33 (1993) 97. C.H. Lochmtkller, C. Reese, A.J. Aschman and S.J. Breiner, J. Chromatogr., 656 (1993) 3. K. Valk6, L.R. Snyder and J.L. Glajch, J. Chromatogr., 656 (1993) 501. J.A. van Leeuwen, B.G.M. Vanderginste, G. Kateman, M. Mulholland and A. Cleland, Anal. Chim. Acta, 228 (1990) 145. J.A. van Leeuwen, L.M.C. Buydens, B.G.M. Vanderginste, G. Kateman, A. Cleland, M. Mulholland, C. Jansen, F.A. Maris, P.H. Hoogkamer, J.H. vanBerg, Chemom. Intell. Lab. Syst., 11 (1991) 161. A.J.P. Martin, Biochem. SOC.Symp., 3 (1950) 4. R. Kaliszan, CRC Crit. Rev. Anal. Chem., 16 (1986) 323. R. Kaliszan, Quantitative Structure-ChromatographicRetention Relationships, Wiley, New York, 1987. T. Braumann, J. Chromatogr., Chromatogr. Rev., 373 (1986) 191. R. Kaliszan, J. Chromatogr., 656 (1993) 417. T. Hanai, J. Chromatogr., 550 (1991)313. C. Hansch and A. Leo, Substituent Constants for Correlation Analysis in Chemistry and Biology, Wiley, New York, 1979. R.F. Rekker, The Hydrophobic Fragmental Constant, Elsevier, Amsterdam, 1977. H. Terada, Quant. Struct.-Act. Relat., 5 (1986) 81. K. Jinno and K. Kawasaki, Anal. Chim. Acta, 152 (1983) 25. K. Jinno and K. Kawasaki, Chromatographia, 18 (1984) 90. K. Jinno and K. Kawasaki, Chromatographia, 18 (1984) 499. K. Valk6, G. Szab6, J. Rohricht, K. Jemnitz and F. Darvas, J. Chromatogr., 485 (1989) 349. J. Fekete ,G. Morovjiin, F. Csinadia and F.Darvas, J. Chromatogr., 660 (1994) 33. C.T. Peng, S. F. Ding, R.L. HuaandZ.C. Yang, J. Chrornatogr.,436(1988) 137. C.T. Peng, Z.C. Yang and S.F. Ding, J. Chromatogr., 586 (1991) 85. M.B. Evans, J.K. Haken and T. T6th, J. Chrornatogr., 351 (1986) 155. R.M. Smith, J. Chrornatogr.,656 (1993) 381. H. Figge, A. Deege, J. KOhler and G. Schomburg, J. Chromatogr., 351 (1986) 393. P. Jandera, J. Chromatogr., 656 (1993) 437. S. V. Galushko, J. Chrornatogr., 552 (1991) 91. S.V. Galushko, A.A. Kamenchuk and G.L. Pit, J. Chrornatogr., 660 (1994) 47. R.M. Smith, Adv. Chromatogr.,26 (1987) 277. F. Morishita, H. Kakihanaand T. Kojima, Anal. Lett., 17 (1984) 2385. R.M. Smith and C.M. Burr, J. Chrornatogr., 475 (1989) 57. H.J. Mockel, J. Chromatogr., 317 (1984) 589. H.J. Mockel, G. Welter and H. Melzer, J. Chrornatogr., 388 (1987) 255. H.J. Mbckel, F. Hofler and H. Melzer, J. Chromatogr., 388 (1987) 275. G. Aced, E. Damsmdt and E. Anklam, Fresenius Z. Anal. Chem., 331 (1988) 740. N. Dimov, Anal. Chim. Acta, 201 (1987) 217. J.K. Baker,Anal. Chem., 51 (1979) 1693. J.K. Baker, R.E. Skelton, T.N. Riley and J.R. Bagley, J. Chromatogr. Sci., 18 (1980) 153. J.K. Baker, J. Liq. Chromatogr., 4 (1981) 271

References pp. 45-46

46 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

Chapter I J.K. Baker, J.D. McChesney andL.F. Jorge, Pharm. Res., 3 (1986) 132. C.D. Hufford, G.A. Capiton, A.M. Clark and J.K. Baker, J. Pharm. Sci., 70 (1981) 151. J.K. Baker, G. J. Hite, M. Reamer and P. Salva, Anal. Chem., 56 (1984) 2932. C-Y. Ma, M.A. Elsohlyand J.K. Baker, J. Chromatogr., 200 (1980) 163. H. Magg and K. Ballschmiter, J. Chromatogr., 331 (1985) 245. R.M. Smith, J. Chromatogr., 236 (1982) 313. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, LC-GC Mag., 4 (1986) 314. R.M. Smith, Trends. Anal. Chem., 3 (1984) 186. P.J. Schoenmakers, H.A.H. Billiet and L. de Galan, J. Chromatogr., 185 (1979) 179. R.M. Smith and C.M. Burr, J. Chromatogr., 475 (1989) 75. R.M. Smith andR. Wang, J. Chromatogr., 558 (1991) 7. C.M. Burr and R.M. Smith, Anal. Proc., 25 (1988) 46. R.M. Smith and Y. Wang, in preparation. N. Tanaka, H. Goodell and B.L. Karger, J. Chromatogr., 158 (1978) 223. R.M. Smith and C.M. Burr, J. Chromatogr., 481 (1989) 71. R.M. Smith and C.M. Burr, J. Chromatogr., 481 (1989) 85. T. Fujita, J. Isawa and C. Hansch, J. Am. Chem. SOC., 86 (1964) 5177. R.M. Smith and C.M. Burr, J. Chromatogr., 550 (1991) 335. T. Fujita, J. Pharm. Sci., 72 (1983) 285. T. Fujita, Prog. Phys. Org. Chem., 14 (1983) 75. A. Leo, J. Chem. Soc., Perkin Trans. 2, (1983) 825 A.J. Leo, in W.J. Dunn, J.H. Block and R.S. Pearlman (Eds.), Partition Coefficient. Determination and Estimation, Pergamon Press, New York, 1986, p. 61. M.R. Clark, L.E. Garcia-Roura and C.R. Clark, J. Liq. Chromatogr., 11 (1988) 3213. C.R. Clark and L.E. Garcia-Roura, J. Chromatogr. Sci., 27 (1989) 11. R.M. Smith and C.M. Burr, J. Chromatogr., 485 (1989) 325. C.M. Burr and R.M. Smith, Anal. Proc., 26 (1989) 24. R.M. Smith, Anal. Chem. 56 (1984) 256. T. Hanai, C. Tran and J. Hubert, J. High Resolut. Chromatogr., Chromatog. Commun., 4 (1981) 454. P.J. Schoenmakers, H.A.H. Billiet and L. de Galan, J. Chromatogr., 218 (1981) 261. R. Hindriks, F. Maris, J. Vink, A. Peeters, M. de Smef D.L. Massart and L. Buydens, J. Chromatogr., 485 (1989) 255. R. Wehrens, L. Buydens, H. Hindriks and F. Maris, Chemom. Intell. Lab. Syst., in press. S. Yamauchi, J. Chromatogr., 635 (1993) 61. M. Popl, V. Dolanslj and J. Fhnrich, J. Chromatogr., 148 (1978) 195. F. Smejkal, M. Popl, A. Cihovh and M. Zikvorkovh, J. Chromatogr., 197 (1980) 147. M.N. Hasan and P.C. Jurs, Anal. Chem., 55 (1983) 263.

Added in proof 75 J.R.M. Smits, W.J. Melssen, G.J. Daalmans and G. Kateman, Comput. Chem., 18 (1994) 157.

Journal of Chromatography Library, Vol. 57: Retention and Selectivityin Liquid Chromatography R.M. Smith, editor 0 199s EfsevierScience 8.V. All rights reserved

47

CHAPTER 2

Retention prediction of pharmaceutical compounds

Department of Pharmaceutical Chemistty, School of Pharmaq, University of London, 29-39 BrunswickSquare, London, WCIN IAX, UK

2.1 INTRODUCTION

The principle of chromatographic separation was based on the empirical observation of the separation of plant dye mixtures on a silica gel column by Zwett. Although, it happened almost 100 years ago, the prediction and design of a chromatographic separation is still a trial-and-error method and needs experience and intuition. From the early stages however, separation scientists wanted to understand the separation mechanism and tried to find relationships between the chemical structure of the compounds and their behaviour in a certain chromatographic system. The variation of the stationary phases, mobile phases, flow rates, etc. did not allow precise determination and accuracy of the chromatographic retention, and it was always advisable to express it as a relative value. Computerized “high-tech” HPLC equipment and highly reproducible HPLC columns make possible the precise measurement and repeatability of HPLC retention. It means that from a technical point of view, we are ready to define precisely the retention of a compound under certain HPLC conditions, which also helps in discovering possibilities for the prediction of retention. Why is a retention prediction for pharmaceutical compounds useful and important? In pharmaceutical chemistry, usually the structure of the compound, its possible impurities or metabolites are known when the search starts for the most appropriate analytical method. On the basis of structureretention relationships, the HPLC analytical method can be more easily designed. Retention prediction can also be used for optimization of the separation. The other purpose of the prediction can be the identification of the observed peak in the chromatogram. This task, however, requires a more reliable and accurate re-

*Present address: Department of Physical Sciences, Wellcome Research Laboratories, Langley Court, Beckenham, Kent, BR3 3BS, UK.

References pp. 90-92

48

Chapter 2

tention prediction, and the application of other spectroscopic methods for the peak identification is essential. Although there are three major HPLC separation mechanisms (reversed-phase, absorption and ion-exchange) applied in pharmaceutical analysis, the retention prediction most widely applied is reversed-phase HPLC. The reason for this is that reversed-phase chromatography is used in approximately 80% of pharmaceutical HPLC analyses, the reproducibility of manufacturing reversed-phase columns is much higher than any other type of stationary phase, and the retention mechanism is much clearer in reversed-phase chromatography. Therefore, in this chapter, the retention prediction focuses mainly on reversed-phase chromatography. There are several approaches for retention predictions. One is based on the known chemical structure. With the help of another method, quantitative information (topological indices, connectivity indices, quantum chemical parameters, etc.) can be obtained from the chemical structure. The calculated numbers can be related to chromatographic retention data by mathematical functions, and this can be the basis of retention prediction. The alternative empirical prediction methods are based on experimental retention data for the pharmaceutical compounds and these data are used to predict retention in other similar chromatographicconditions andor structurally similar compounds. In general, much more accurate retention prediction methods have been described for structurally related alkylbenzene homologous or substituted aromatic compounds, than for pharmaceutical compounds. Pharmaceutical compounds can be completely unrelated in structure and contain a much wider variety of polar functional groups, which makes the retention prediction less accurate. In this chapter, the methods presented for retention prediction focus on pharmaceutical compounds that are structurally unrelated. First, the definition and determination of the retention parameters is discussed. The dependence of the retention parameters on the chromatographic conditions is examined in order to reveal the reliable range of the retention prediction. The retention prediction possibilities for pharmaceutical compounds are discussed based on molecular modelling calculations and other molecular descriptors. The relationships between measured or calculated physicochemical parameters and the retention parameters are also demonstrated. A retention prediction method based on the experimentally found retention increment values is presented as well as other empirical methods with a multiparameter approach. As a summary of the results, several applications for retention prediction of pharmaceutically active compounds is demonstrated. The intention has been made to show our results in this field without providing a complete literature review and only related methods are discussed.

2.2 DEFINITION AND DETERMINATION OF RETENTION In order to be able to predict the retention of pharmaceutical compounds in reversedphase high performance liquid chromatography (RP-HPLC) the retention should be defmed precisely. Also the measurements of the retention should be carefully carried out. The most widespread measure of the degree of the retention is the capacity ratio (k) or alternatively it is called the retention factor [l], which can be defined by Eq. (2.1).

Retention prediction of pharmaceutical compounds

k=-

tR

49

(2.1) to where tR is the retention time (the time passed from the injection to the appearance of the solute peak maximum) and to is the dead time. The same expression can be obtained by using retention volume and dead volume values, which can be obtained by multiplying the retention time values with the actual mobile phase flow rate. As is clear from the Eq. (2.1), the retention is expressed relative to the unretained component in the system. Therefore, the exact dead time determination is essential for the accurate determination of the retention factor, k. There are several arguments in the literature about the precise dead time determination of a given HF’LC system. The problem starts with the definition of the dead time, namely the retention time (or if it is multiplied with the flow rate, the retention volume) of an unretained solute. The question arises which, if any, solute is unretained. There are many possibilities described in the literature. The dead volume can be the elution volume of a solvent disturbance peak obtained by injecting an eluent component, or the elution volume of an unionized solute that gives the lowest retention volume, or the elution volume of an isotopically labelled component of the eluent, or isotopically labelled water molecule, or the elution volume of a salt ion, or the volume of the liquid the column contains, or the extrapolated elution volume of a “zero” member of a homolog series. In practice none of the above possibilities are accepted as an absolute definition. Berendsen et al. [2], Wells and Clark [3], and Knox and Kaliszan [4]provided detailed reviews on the various techniques for experimental determination of dead volume in HPLC. Whatever method is used to determine to, the most important thing is to apply the same method throughout the measurements used for the retention predictions. According to Eq. (2.1) for the determination of the retention factor (k), the retention time of the compound should also be accurately measured. Although the retention factor does not depend on the flow rate or column pressure, it is advisable to keep them constant during the measurements. The retention time determination is usually carried out by an integrator or other data handling computer program. It is advisable to connect the injector and the integrator electronically for an accurate start of the time measurements. As the peaks are wider in HPLC than in GC, the slope sensitivity and threshold values should be properly set up. Care has to be taken when the signal to noise ratio is less than 5. As the error of the retention prediction is usually much higher than the error of the measurements, the above-mentioned practical problems for retention measurements are negligible. Unfortunately the retention index system which defines relative retention and is therefore much more independent of slight changes in chromatographic conditions is not as widespread as in gas chromatography. Baker et ul. [5-71 has introduced a retention index scale similar to the gas chromatographic retention index. The scale is based on the relative retention of a homologous series of C34& 2-keto alkanes. The retention index, Z, of a given solute i is calculated from its observed retention factor ki, the retention factor for a 2-keto alkane eluting before the test solute, kN, and the retention factor of a 2-keto alkane eluting after the test compound, kN+ as described by Eq. (2.2).

zi = lOOx

log/$ -logk, 1% k N + 1 - 1% k,

References pp. 9G92

+ lOON

50

Chapter 2

The retention index of a given 2-keto alkane standard is by definition equal to 100 times the carbon number (N) in the formula. A slightly modified retention index scale has been proposed by Smith [8,9]. He suggested replacing the terminal methyl group of the 2-keto alkane series of standards by a terminal phenyl group, which makes the UV detection of the reference homologous series easier. The advantages of using the retention index scale is discussed in detail in Chapter 1. It was also proved that the logarithmic values of the k of a homologous series showed linear relationships with the number of carbon atoms in reversed phase liquid chromatography. Baker and Ma [6] showed that the retention index, for example, of phenacetin was independent of the methanol content of the mobile phase. Both index scales have already been used for the study of quantitative structure-retention relationships. The precision and accuracy of the relative retention determination expressed in the retention index did not exceed significantly the accuracy and precision of the k determination and its determination is more time consuming. The variation in k values as a function of the mobile phase composition can be important in the retention prediction of pharmaceutical compounds, and is often used for method development and optimization. This information disappears when using the index scale. Therefore, in this chapter, the retention prediction based only on the retention factor is discussed.

2.3 DEPENDENCE OF RETENTION ON THE COLUMN AND MOBILE PHASE COMPOSITION In HPLC, the retention of a compound is determined by the strength of the solutestationary phase and solute-mobile phase interactions. Thus, it is important to reveal quantitatively, how the mobile phase composition effects solute retention, as this information is vital for retention prediction. On the basis of numerous experimental data, we can assume that the dependence of the logarithmic retention factor (log k) on the volume fraction of organic component, rp, in the mobile phase can be described by Eq. (2.3). log k = Sj + log k0

(2.3)

where the log k0 value (the intercept of the straight line) refers to the extrapolated log k value for the neat water as mobile phase. The slope value (5') shows the sensitivity of the retention of a given compound caused by the change of the organic phase concentration in the mobile phase (see also Fig. 2.1). Using Eq. (2.3), two important RP-HPLC retention parameters can be determined (the slope, S, and the intercept, log k,,), which can be used for the retention prediction. We should keep in mind that the linearity is valid only in a limited range of the mobile phase composition. The deviation from linearity can be observed at high organic phase concentrations (above 90%, v/v) due to the residual silanol groups. It has been observed that at high organic phase concentrations, the retention of the compounds increased again, mostly for compounds with basic nitrogen or other silanophil groups [lo]. Also it was found that the extrapolated log k values for the zero organic phase concentration are dif-

Retention prediction of pharmaceutical compounds

I

I

I

I

I I

I

I I

I 0

51

50

100

@%

Fig. 2.1. The dependence of log k values on the organic phase concentration (OP% or q5)

ferent when obtained using a different organic phase, and it has no realistic meaning, because deviation from linearity can be observed at this region as well [l l l . In practice the determination of these two parameters (S, log ko) can be carried out by measuring the log k values of compounds by using a minimum of five organic phase concentrations. The organic phase concentration can be decreased by 5 or 10% steps. All of the log k values of the investigated series of compounds should be determined in a minimum of three consecutive measurements, and the average log k values should be plotted against the organic phase concentration. The fit of the data points to straight lines can be checked by calculating the correlation coefficient and the standard deviation of the slope and intercept ( S and log b,respectively) values. Unacceptable fits can be obtained for several reasons. Schoenmakers et al. [12] described a quadratic relationship between the log k and q5 values. It usually fits much better to the experimental data points, as it explains the deviation from the linearity in the extreme organic phase concentration range, and also introduces a second term (the quadratic q5) usually improves the statistical parameters as well. Wells and Clark [ 131 suggested the application of the solvophobic theory proposed by Horvath et al. [ 141 for describing the log k versus q5 relationship. Several results [ 15,161 show that the linearity of the plot is not valid in a wide range of organic phase concentrations, and the log values are not the same when they are derived from data obtained by using acetonitrile or methanol. The deviation from linearity can be observed when the mobile phase pH is varied by the variation of the organic phase concentration in the mobile phase, especially in the case of easily ionizable compounds. The exact definition and determination of the mobile phase pH is therefore very important. In order to avoid the effect of the changing mobile phase pH on the linearity of the log k versus q5 plot, it is suggested that measurements are made References pp. 90-92

52

Chapter 2

CJI 0

0

---

I

t

I

t

I

1

I

I0 20

1 1 1

t

1

I

I

30

40

' 1 1 1 I

I

50 60

I

70

I

I

80 90

O/o

-'f'

Fig. 2.2. Hypothetical possibilities for the plots of the log k versus organic phase concentration ($) relationships. 1, straight line; 2, curved parabolic relationship without silanophil effect; 3, straight lines crossing each other; 4, parabolic relationships for basic molecules showing silanophil effect. (Reprinted with permission from ref. 28)

in a buffer at a pH equivalent to the neutral form of the solute. Figure 2.2 shows the hypothetical possibilities for the plots of log k versus @ for the above-mentioned cases. In spite of these limitations, the relationship described by Eq.(2.3) has great potential to predict the retention of pharmaceutical compounds at various organic phase concentrations (within the linear range) and is extremely valuable in the optimization of separations. The prediction of these two retention parameters (S and log b) is discussed in this chapter in detail. As the retention in chromatography reflects the solute distribution between the stationary and the mobile phases, it also depends on the temperature. The effect of the temperature on retention has been studied less extensively in RP-HPLC than the effect of solvent composition. In fact the use of aqueous eluents and silica-based stationary phases limits the practical column temperature range fiom 5 to 100°C. Over that temperature range one can expect a roughly tenfold decrease in retention factor if the enthalpy of solutestationary phase interaction is -5 kcal/mol, which is a typical value in RP-HPLC. A retention change of similar magnitude occurs when the organic phase component of the mobile phase is changed by approximately 30%. So the effect of the mobile phase composition on the retention is more significant. Usually, chromatographic measurements are carried out at ambient temperature or under controlled temperature at 20 or 25°C. The temperature can also affect the efficiency of the chromatographic system, since the d i f i sion coefficient of a solute increases with the temperature. In most cases the selectivity or the relative retention is not greatly affected by changing the temperature in RP-HPLC. This is not surprising as enthalpies for solute binding by the stationary phase are usually

Retentionprediction of pharmaceutical compounds

53

small and the difference is less than 1 kcal for closely related compounds. The effect of temperature on retention is largely determined by the enthalpy of the solute-stationary phase interaction. The enthalpy can be calculated fiom the slope of plots of log k versus the reciprocal of the absolute temperature; called van’t Hoff plots. Published results [ 17191 indicate that the logarithm of the retention factor is linearly related to the enthalpy in RP-HPLC. The increase in the logarithm of the retention factor with enthalpy, however, is much greater than expected. It can happen that the temperature effect can be important in the retention prediction; then the use of a column thermostat is suggested to control the temperature. Bad column efficiency can be the result of secondary equilibria and it not only decreases the accuracy of the retention determination but it can cause erroneous retention prediction. Therefore, in practice it is always advisable to check the column efficiency and the peak symmetry when the retention measurements are used for prediction. In conclusion, the magnitude of chromatographic retention is determined by the energetics of the solute interactions with both the mobile and stationary phases. Consequently, retention data contain relevant thermodynamic information. The close relationship between the retention factor and the equilibrium constant allows us to extract thermodynamic information from the chromatogram which can be used for understanding and predicting the retention.

2.4 CORRELATION OF RETENTION PARAMETERS TO THE MOLECULAR PARAMETERS OBTAINED BY MOLECULAR MODELLING Many methods have been devised for the determination of molecular structures. In more recent years it has been proven to be possible to determine accurate structures by computational methods, typically ab initio calculations [20] for large molecules. Molecular mechanics calculations can be carried out to determine the three-dimensional structures of molecules having pharmaceutical importance. The calculated molecular parameters can be correlated to retention data. In the early work of Horvhth et al. [21], the retention was attributed to a reversible association of the solutes with the hydrocarbon ligand of the reversed-phase stationary phase. The energetics of the association process was also analyzed and the dependence of the retention factors on ionic strength of the eluent and the hydrophobic surface of the solute were revealed. The correlations of the octanol-water partition coefficients, molecular surface areas, and reversed-phase retention factors were studied by Funasaki et al. [22]. The importance of the molecular cavity surface area and various connectivity indices, which can be calculated from the chemical structure of the compounds was pointed out also by Funasaki et al. [23]. Eng et al. [24] showed the application of holistic conformation and total surface area calculations for the prediction of chromatographicretention parameters for triphenyl derivatives. Mockel et al. [25] investigated the effect of the molecular surface type and area to the retention of various hydrocarbon classes. It cannot be questioned that molecular parameters play an important role in the retention but the extent of possible generalization of the equations and their predictive power for the chromatographic retention have not yet been explored. References pp. 90-92

54

Chapter 2

TABLE 2.1 THE CHROMATOGRAPHICRETENTION DATA AND HYDROPHOBICITY DATA OF STRUCTURALLY UNRELATED PHARMACEUTICALCOMPOUNDS Compound Sulphadimidine Sulphamerazine Barbital Phenobarbital 5 Chloramphenicol 6 Salicylamide 7 Phenacetin 8 Vanillin 9 Benzaldehyde 10 Acetanilide 11 Nicotinamide 12 Benzoic acid 13 Salicylic acid 14 Acetyl salicylic acid 15 Caffeine 16 Hydrochlorothiazide 17 Dexamethasone 18 Deoxycorticosterone 19 Isoniazid 20 Methyl salicylate 21 Hydrocortisone 22 Progesterone 23 Testosterone 1 2 3 4

1% pc

Slope

1.644 0.612 -1.050 -0.430 0.464 0.236 1.128 1.119 1.535 0.529 -0.690 1.769 2.140 1.037 -0.912 -1.717 -0.472 3.795 -2.003 2.528 2.029 1.508 4.874

-0.0280 -0.0283 -0.0402 -0.0319 -0.0414 -0.02 5 5 -0.0226 -0.0244 -0.0303 -0.0270 -0.0382 -0.0284 -0.0301 -0.0272 -0.0299 -0.0456 -0.0139 -0.0147 -0.0382 -0.0244 -0.0129 -0.0192 -0.0143

1%

ko

0.854 0.892 1.063 1.341 1.625 0.871 1.002 0.866 1.575 1.021 0.25 1 1.252 1.425 1.077 0.552 0.887 0.568 1.120 0.060 1.727 0.436 1.831 1.085

$0

30.50 31.52 26.44 42.04 39.25 34.16 44.34 35.49 51.98 37.81 6.57 44.08 47.34 39.60 18.46 19.45 40.86 76.19 1.57 70.78 33.80 95.36 75.87

With permission from ref. 26.

Our investigation [26] was focused on 23 drug molecules with completely different chemical structures (Table 2.1). The retention parameters of the pharmaceutical compounds were measured on LiChrosorb RP-18 10-pm columns, 250 X 4.6mm (Merck, Darmstadt, Germany). The mobile phases were various compositions of acetonitrile and 0.05 M phosphate buffer (pH = 4.6). Lower pH (pH = 2) was used for the retention measurements of the acidic compounds to avoid dissociation. The detailed instrumentation of the measurements and also the retention data obtained have been published earlier [27]. The retention of the compounds was expressed by the logarithmic values of the retention factor, log k and it was plotted as a function of the acetonitrile concentration. On the basis of three to five points, the slope and the intercept values of the straight lines obtained have been calculated and are listed in Table 2.1. It has also been published earlier [28,29] that the r$o values, namely the acetonitrile concentration necessary for obtaining log k = 0 retention showed excellent correlation with the logarithmic values of the octanol-water partition coefficients (log P). These values are also presented in Table 2.1. A Personal Computer (PC) Model approach was used to determine the threedimensional structure of compounds based on energy minimization. After setting up the geometries of the molecules having the smallest mmx-energy, the non-polar (NP), nonpolar unsaturated (NPU) and polar surface areas (PS), their energies were calculated. The water solvation shell was also considered in the calculations of the accessible polar (APS)

55

Retention prediction ofpharmaceutical compounds

and non-polar (ANP, ANPU) surface areas. The calculated total surface energy (TSE) was expressed in kcaVmol. The dipole moment (dm) values and van der Waals radii (vdw) of the molecules listed in Table 2.1 were also calculated. Correlation analysis was carried out to reveal which of these parameters show significant correlation with the reversed-phase retention parameters. The multiple regression equations and their mathematical statistical characteristics also reveal their applicability for retention prediction. It was assumed that some kind of correlation can be revealed between the chromatographic retention parameters ( S and log b)and the calculated molecular parameters, by which retention prediction of the pharmaceutical compounds can be carried out with various mobile phase compositions. The calculated molecular parameters are listed in Table 2.2. The linear regression analysis revealed two significant equations (Eqs. 2.4 and 2.5) by which the S (slope) values could be described by the molecular parameters investigated. S = 1.42 (*0.35) X lo9

X

vdw + 9.29 (* 0.26) X

n = 23, r = 0.845, s = 5.07

X

TSE - 0.04

(2.4)

F = 25.1

X

TABLE 2.2 THE CALCULATED MOLECULAR PARAMETERS FOR THE COMPOUNDS LISTED IN TABLE 2. I ~~~~

No.

NP

NPU

dm

vdw

ANP

TSE

1

261.8 194.7 193.3 206.7 246.1 103.1 286.4 164.1 151.2 196.6 106.3 125.5 108.7 182.6 254.0 124.2 370.2 383.3 108.9 201.2 356.6 410.2 371.9

62.5 68.4 0.0 27.6 27.9 50.8 53.3 40.2 54.1 40.7 36.9 54.3 53.8 42.4 18.3 15.8 12.2 6.9 38.3 31.9 6.8 6.8 6.6

5.44 6.81 0.84 0.68 6.30 3.14 4.83 2.89 2.81 3.03 2.16 1.51 1.93 2.58 1.87 6.42 4.92 3.91 2.47 2.22 4.36 2.28 2.09

10.43 11.60 5.1 1 9.84 11.01 5.60 9.95 7.33 6.38 8.01 6.48 6.88 6.88 7.01 8.03 5.26 17.36 13.52 6.73 7.86 15.21 13.00 13.13

246.7 183.5 182.2 195.3 205.9 97.4 169.7 154.6 142.9 185.4 100.4 118.6 102.7 172.0 239.0 103.7 348.8 361.7 103.0 189.4 336.4 387.0 351.1

-3.7 -8.3 -7.0 -6.3 -6.9 -7.2 4.1 -3.6 0.5 1.5 -5.4 -2.4 -5.5 -3.7 1.9 -14.8 -1.7 0.8 -7.6 0.2 -2.7 4.2 3.4

2 3 4 5

6 7 8 9

10 11 12 13 14 15 16 17 18 19 20 21 22 23

NP, non-polar surface area, A2, NPU, non-polar unsaturated surface area, A2; dm, dipole moment; vdw, van der Waals radii; ANP, accessible non-polar surface area, A2,TSE, total surface energy (kcal/mol). Reprinted with permission from ref 26.

References pp. 90-92

56

Chapter 2

S = 6.98 (& 1.73) X lod3 X ANP - 5.95 (* 1.64) X lo” n = 23, r = 0.857, s = 4.90

X

lo”,

X

NP - 0.04

(2.5)

F = 27.6

where n is the number of compounds, r is the multiple regression coefficient, s is the standard error of the estimate, and F is the Fisher-test value (vdw, van der Waals radii; TSE, total surface energy in kcavmol; ANP, accessible non-polar surface area; NP, nonpolar surface area). Equation (2.4) means that the higher the van der Waals radius and the higher the total surface energy, the more sensitive is the compound retention to changes in the acetonitrile concentration in the mobile phase. This finding is in agreement with the previous suppositions that the S (slope) values are in relation to the size of the molecules. Equation (2.5) expresses the retention sensitivity of the compound to the organic phase concentration as a fimction of the difference between the total non-polar and the accessible non-polar surface areas. This finding can also be understood if we consider that the solvent molecules should replace the compounds from the stationary phase, and the higher the difference between the accessible and non-accessible hydrophobic surface area, the higher the concentration of acetonitrile molecules needed to elute the compound from the hydrophobic stationary phase surface. The three-dimensional plots of hydrocortisone and progesterone (Fig. 2.3) illustrate the situation. The polar regions are darker. The two molecules differ from each other only in the presence or absence of one hydroxyl group. The non-polar surface area of the progesterone is interrupted by the polar hydroxyl group in the case of hydrocortisone. Also the slope values of the compounds differs markedly, -0.0192 and -0.0 129 for the progesterone and hydrocortisone, respectively. Although their size and polarity do not differ very much, the accessible non-polar surface area by the non-polar stationary phase decreased markedly in the case of hydrocortisone.

P r o g e a t e r on

Hydr ocor t i a o n

Fig. 2.3. The three-dimensional plots of the electron clouds for progesterone and hydrocortisone. (Reprinted with permission from ref. 26.)

Retention prediction of pharmaceutical compounds

57

EST I M A T C D

-0 .O(J0

I

-0.045

-

-n.o4u

I

c

-

-0.042

-0.030

-nsn

-0.021

-o.nia

-0.012'

-n.(m' * Slope

Fig. 2.4. The plot of the measured and estimated slope values for the compounds listed in Table 2.1 according to Eq. (2.4). (Reprinted with permission from ref. 26.)

The plot of the measured and calculated S values on the basis of Eqs. (2.4) and (2.5) can be seen in Figs. 2.4 and 2.5, respectively. From the figures and the statistical parameters of Eqs. (2.4) and (2.5) it can be seen that the prediction of the slope (3values is not really accurate. As the slope values commonly range fiom 0.01 to 0.05, the 0.005 standard error means a minimum 10% error. The equations can be suggested for retention prediction in only 10-20% of organic solvent concentration range. The most significant equation for describing the variance of the intercept (log k,,) values was obtained when the calculated clog P values, the molecular parameters describing the slope values (non-polar surface area, NP and the non-polar accessible surface area, NPA) and the reciprocal value of the dipole moment (dm) were taken as independent variables as described by Eq. (2.6). log k,, = 0.286 (k 0.036)~logP + 0.037 (*0.008)NP - 0.040 (* 0.009)ANP

+ 0.652 (* 0.165)[l/(dm)]+ 39.58

(2.6)

n = 23, r = 0.896, s = 0.230, F = 17.4

Although the relationship is significant at higher than 95% probability level according to the Fisher-test value, the predictive power of the equation can be regarded as very poor with 0.230 log k values as the standard error. The chromatographic hydrophobicity index ($,,), discussed in detail in Section 2.6, References pp. 90-92

58

Chapter 2

E S T I M A I ED

Slope

Fig. 2.5. The plot of the measured and estimated slope values of the compounds listed in Table 2.1 according to Eq. (2.5). (Reprinted with permission from ref 26.)

could also be described by the calculated octanol-water partition coefficients (log P ) and the calculated non-polar unsaturated surface area (NPU) as shown by Eq. (2.7). $0 =

11.OO (* 1.06) log P - 0.25 (A 0.92)NPU + 39.58

(2.7)

n = 23, r = 0.924, s = 8.85, F = 58.7

As the hydrophobicity index, $o, means the organic phase concentration at which the retention time is double the dead time (log k = 0), it can be regarded as a retention parameter, which can be calculated from the hydrophobicity and the non-polar unsaturated surface area of the molecules. As can be seen, the standard error is 8.85, which means the error of the mobile phase composition expressed in volume percent. It seems to be relatively high, but in practice it can be acceptable. With 9% change in the mobile phase composition, the chromatographic peak will still be within the measurable range, and can be detected in the chromatogram. The plot of the measured and calculated (by Eq. 2.7) chromatographichydrophobicity index values can be seen in Fig. 2.6. As a conclusion from the relationships described above, the investigated retention parameters could be described by molecular parameters obtained from molecular modelling of structurally unrelated pharmaceutical compounds. The linear regression equations obtained were significant from the mathematical statistical point of view, but the standard deviations seem higher than would be sufficient for retention predictions without any experimental trials. In spite of these drawbacks, the importance of the above correlations is outstanding. The molecular mechanics calculations can be improved easily in the future,

Retention prediction of pharmaceutical compounds

59

ESTIMATED

60

30

1

Fig. 2.6. The plot of the measured and calculated (Eq. 2.7) hydrophobicity indices of compounds listed in Table 2.1. (Reprinted with permission from ref. 26.)

as these calculations are based on ub initio calculations with the assumption that the calculated force fields are transferable to larger molecules. With the help of faster computers and less assumptions, the accuracy of the calculations can be increased, which will probably also increase the predictive power of this type of correlation. It is also remarkable that the relationships can be set up for completely non-congeneric compounds, as most of the published results up to now were achieved by investigating homologous series [30-331. Jinno and Kawasaki [30] used partition coefficients, hydrophobic and electronic constants, van der Waals volume, molecular area, molecular connectivity, length-tobreadth ratio for describing a multiparameter structure-retention correlations for computer-assisted retention prediction of alkylbenzenes, substituted benzenes, phenols, polycyclic hydrocarbons. Kaliszan et ul. [3 1,321 applied the calculated total energy values, maximum charge differences for quantitative structure-retention relationships of substituted benzene derivatives which can be regarded as non-congeneric solutes. Hanai [33] reported the importance of energy effects, pK, values and van der Waals volumes in describing the reversed-phase HPLC retention of aromatic acids. But according to our investigations [34] on the basis of the correlation of the slope (s) and the intercept (log b) values, the substituted aromatic compounds (such as anilines and phenols) can be regarded as structurally related compounds from their partition behaviour in W-HPLC. This means that the examples cited above all refer to structurally related compound series. This emphasizes the importance of the correlations obtained for the 23 structurally unrelated pharmaceutical compounds. References pp. 90-92

60

Chapter 2

2.5 RETENTION PREDICTION BASED ON TOPOLOGICALMATRIX AND INFORMATION THEORY For the description of quantitative structure-retention relationships on which retention predictions can be carried out, the efforts to translate molecular structure into unique characteristic structural descriptors expressed as numerical indices are very important. The most commonly used topological indices as retention descriptors are summarized by Kaliszan [29]. The topological indices can be calculated by means of the chemical graph theory, where a chemical structural formula is expressed as a mathematical graph. The formula shows how bonds connect different atoms in the molecule. The mathematical graph describes abstract vertices joined by edges. Each molecular graph may be represented either by a matrix, a polynomial, a sequence of numbers, or a numerical index (topological index). The molecular connectivity index, at present the most popular topological index, was introduced by Randic [35] for the characterization of molecular branching. Molecular connectivity indices have gained great popularity for describing quantitative structure-retention relationships. Karger et al. [36] found the molecular connectivity useful for the prediction of reversed-phase HPLC retention data of phenols and alcohols containing various normal, branched, and cyclic hydrocarbon moieties. Jinno and Kawasaki [37] reported the lack of correlation (r = 0.176) between log k values obtained in reversed-phase chromatography and the molecular connectivity for a set of benzene compounds containing common substituents such as NH2, NO2, CN, COOCH, and C1. The reason for this may be that the equations describing quantitative relationships between molecular connectivity indices and retention parameters are usually only true for closely generic series of solutes, usually non-polar in character, for which retention data are determined in non-polar chromatographic stationary phase systems. It is because the molecular connectivity indices contain only limited information about the properties of polar fimctional groups, the ability to form hydrogen bonding or the capability for other polar interactions. The predictive power of the derived equations depends on the true relationship which is valid only within the series of the compounds investigated. Matsuda et al. [38] and Hayashi et al. [39] described the application of information theory for retention prediction. The retention prediction for PTH-amino acids using the information theory was first suggested by Jinno [40]. They described the chromatographic retention factor as a hnction of chromatographic conditions and physicochemical properties as shown by Eq. (2.8). log $ = (CXZ + DX+ E)Rj + ( F P + GX+ H) where Rj denotes the retention-solubility parameter of PTH-amino acidj and is related to the aqueous solubility parameter S [40]. The formula shows also a relation to the volume fiaction, X , of a modifier. Coefficients C to H of the polynomials of X can be determined by the linear least squares method for real data. The physicochemical properties of PTHamino acids are reflected by Rj and those of the mobile phase are described by the polynomials of X. They used the retention prediction method for the optimization of the separation of PTH-amino acids. The same approach was also used for the retention prediction of alkylbenzenes [38]. Table 2.3 shows the formulae used for retention prediction of al-

Retention prediction ofpharmaceutical compounds

61

TABLE 2.3 FORMULAE OF RETENTION PREDICTION FOR ALKnBENZENES WITH ACETONITRILEAS A MODIFIER ACCORDING TO MATSUDA et al. [38] log($) = A(x) log(Pi) + B(X)$ + C(x)

A(X) 0 . 1 0 2 ~ 0.746X2 ~ + 0.427, B(x) = 1.022y2- 1.051X+ 0.308, C(X)= -O.980X+ 0.084 5

j (analyte number)

WPi)

5

1 2 3 4

2.60 3.13 3.66 4.19 4.12 5.25

4.0 5.0 6.0 7.0 8.0 9.0

5

6

Toluefle Ethylbenzene n-Propylbenzene n-Butylbenzene n-Pentylbenzene n-Hexylbenzene

With permission from ref. 38.

kylbenzenes with acetonitrile as a modifier. Figure 2.7 shows the predicted retention factor values (k) at various volume fiactions (%) of acetonitrile in water, as mobile phases. In this approach, a great number of experimental data were involved in the retention prediction, and it referred only to the given chromatographic conditions, column, flow rate, etc. The maximum information was extracted from the experimental values, and the retention prediction was carried out only in a limited range of variable conditions for the optimum separation. To gain maximum information fiom a large set of chromatographic retention data, factor analysis was proposed by Righezza and Chretien [4 11. The approach was applied to the retention data k of a large set of compounds in HPLC. The chromatographic information, i.e. affinity and selectivity, was extracted with the help of principal component analysis and correspondence factor analysis (CFA). In conclusion, the application of topological indices or information theory for the prediction of chromatographic conditions are always based on experimental retention data. Quantitative relationships can be set up between the retention factors and the descriptors

Fig. 2.7. Retention prediction of alkylbenzenes for reversed-phase chromatography. X denotes the volume fraction (‘33) of acetonitrile in water. Lines (from bottom to top): toluene; ethylbenzene; n-propylbenzene; nbutylbenzene; n-pentylbenzene; n-hexylbenzene.(Reprinted with permission from ref. 38.)

References pp. 90-92

62

Chapter 2

of the chemical structure of the compounds and other chromatographic conditions. The equations derived can be used for retention predictions only within the series of compounds and HPLC conditions investigated. The advantages of these methods are that they do not require measured physicochemical properties for the compounds.

2.6 RETENTION PREDICTION BASED ON THE HYDROPHOBICITY OF DRUGS

Much better precision and a wider application range in pharmaceutical analysis can be gained by using retention prediction based on measured or calculated physicochemical properties of compounds. Separation in chromatography is the result of differential migration. Differential migration or in other words the movement of individual compounds through the column depends on the equilibrium distribution of each compound between the mobile and the stationary phases. The distribution depends on the composition of the mobile phase and the stationary phase, and on the temperature. There are four major interactions between the solute and solvent molecules involved in the distribution, namely dispersion, dipole, hydrogen bonding and dielectric interactions. The total interaction of a solvent molecule with a sample molecule can be described by the term polarity. Polarity can be defined as the ability of the sample or solvent molecule to interact in all of these forces. Thus, polar solvents attract polar solutes. The polarity can also be measured or related to other physicochemical parameters, such as partition coefficients, dipole moments, etc. These physicochemical parameters can then be related to the chromatographic retention. The separation mechanism on chemically bonded non-polar phases (RP-HPLC) has been described by various theories in the literature. In 1977, Colin and Guiochon discussed three possible retention mechanisms [42]. The RP-HPLC can be regarded as a kind of liquid-liquid partition chromatography, or it is similar to classical liquid-solid chromatography, but interactions between the solute and stationary phases are weaker than in adsorption so that the solute behaviour in the mobile phase is dominant, or partition of the solutes takes place between the mobile and a “mixed” stationary phase formed by adsorption of the organic modifier on the stationary phase [43]. The preponderance of more recent experimental evidence and theoretical developments [44] support a mechanism for RP-HPLC that is largely partitioning, particularly for C-18 stationary phases. Wise et al. [45] have proposed that the selectivity RP-HPLC for polycyclic aromatic hydrocarbons depends on the length-to-breadth ratio of the molecules. Martire et al. [46] have developed a unified molecular theory based on a lattice model to describe the solute distribution process. Dill [47] has developed another statistical-mechanical theory that accounts for bonded-chain re-organization energy. All of these theories indicate that retention energies are dominated by partitioning of the solutes into the bonded-phase. It means that in reversed-phase HPLC, the hydrophobicity of the compounds governs the retention. The hydrophobicity is often expressed by the measured or calculated octanolwater partition coefficients (log P). The log P values are also used in drug design and they are also important from the environmental protection point of view. Therefore, it has a great advantage over other physicochemical properties in that huge databases of meas-

Retention prediction ofpharmaceutical compounds

63

ured log P values are available [48]. Rekker [49] developed a calculation method for log P values based on fragmental constants for the fragments of the molecules. Several programs are on the market [50,51] by which log P values can be calculated from the chemical structure of the pharmaceutical compounds. The relationships between the log P values and RP-HPLC retention data provides the possibility of not only estimating hydrophobicity by chromatography but also predicting reversed-phase chromatographic retention. Numerous publications have been reported on the correlation of RP-HPLC retention data with log P values. There are three main approaches for the correlation of the log P values to the RP-HPLC retention data of pharmaceutical compounds. The first, most simple approach describes linear correlation between the log P values and the log k values obtained in a given RP-HPLC system according to Eq. (2.9). log k = a log P + b

(2.9)

where a and b are constants. The values of the constants depend on the applied reversedphase column, mobile phase composition, and the structure of the series of compounds being investigated. It can be noticed from the published data that Eq. (2.9) was valid only for structurally related series of compounds. When a similar equation successfully described the data of structurally unrelated compound series, the chromatographic system contained octanol and water. As our aim is to study the retention prediction of pharmaceutical compounds under regularly used reversed-phase HPLC conditions, the application of octanol in the chromatography can be disregarded. Although many examples show that high correlation coefficients can be found for Eq. (2.9), it cannot be regarded as being useful for retention prediction as the values of constants vary with the column, mobile phase composition and the compound structure. The second approach overcomes the problem of mobile phase variation by using the extrapolated log k values to 0% organic phase concentration as described by Eq. (2.10). log k,,= a log P + b

(2.10)

where a and b are constants. This approach has the advantage that the retention data of the compounds investigated can be obtained from different mobile phase compositions, so a much wider range of hydrophobicity values can be covered, which means that the equation is more general, and valid for a larger set of compounds than is the case with Eq. (2.9). However, there are two major problems with the general use of Eq. (2.10). First, problems arise with the extrapolation of the log k values to zero organic phase concentration. Snyder et al. [52], Butte et al. [53] and Hammers et al. [54] showed that over a volume fraction range of at most 0.1-0.9, the linear extrapolation from the log k versus volume fraction of the organic phase relationship is acceptable for the estimation of log ko values. However, several results [13,55] showed that the linearity of the plot is not valid for a wide range of organic modifier concentrations, and the l o g b values are not the same when they are derived from data obtained by using acetonitrile or methanol [25]. Therefore, log is suggested instead of log h.Schoenmakers et al. [ 121 suggested a quadratic relationship between log k and the volume fraction of the organic modifier, at which the log values are different. The uncertainty of the derivation of log values References pp. 90-92

64

Chapter 2

makes the application of Eq. (2.10) unreliable. Also the relationship is valid only for structurally related compounds. It is understandable ftom Leo's [56] findings, that the linear relationship between partition coefficients is valid only for structurally related compounds, or for similar partition systems. The third approach can be regarded as the most general for describing the relationships between the log P values and the chromatographic retention parameters [57] by Eq. (2.1 1). log P = U S + b log ko + c

(2.1 1)

where a, b, and c are constants and S is the slope. The equation was found to be valid for structurally unrelated pharmaceutical compounds. The slope and the intercept (log b) values mathematically describe the linear portion of the log k versus organic phase concentration. By applying the least squares method to determine the values of the coefficients (a, b and c), essentially a backward extrapolation is carried out to express the log k A

l o g K'

1 .oo

0.75 0.50 0.25

0 -0.25

-0.50 -0.75

-1.00

I

0

I

I

10

20

I

30

I

40

I I

50

0

60

70

80

1

90

I

100

y

I

(%)

Fig. 2.8. The graphical illustration of the backward extrapolation method for optimizing the RP-HPLC mobile phase composition for describing the best relationship of log k values with the octanol-water partition coeficients (log P). Straight lines 1 4 represent the linear portion of the log k versus organic phase concentration plot. The optimized organic phase concentration for which the extrapolated log k values show the best correlation to log P (vertical line) can be calculated from the regression coefficients of Eq. (2.1 1) (ah).The horizontal line refer to the log k = 0 retention, i.e. the retention time is double the dead time.

Retention prediction ofpharmaceutical compounds

65

0.5400.520-

f

B 111

IU

.N

J

1

I

I

I

1

40

50

60

70

80

90

OP % (acctonitrlle Concentration,

/

v/v )

Fig. 2.9. The plots of the standard error (s) for the log P versus log k relationships as a function of acetonitrile concentration in the eluent obtained on Supelcosil LC-18 (S) and LiChrosorb RP-18 (L) stationary phases. The optimum mobile phase composition was found to be 33.5% (v/v) and 30.1% (v/v) acetonitrile for Supelcosil LC-18 and LiChrosorb RP-18 stationary phases, respectively. (Reprinted with permission from ref. 58.)

values at the organic phase concentration at which the log P (octanol-water partition) can be best modelled. That is why a statistically significant equation was found for structurally unrelated compounds as well. The optimum mobile phase composition can be calculated as a quotient of a and b. Figure 2.8 represents the theoretical meaning of Eq. (2.1 1). As it is based on retention data obtained under more than one set of isocratic mobile phase conditions, a wide range of Rp-HPLC conditions and hydrophobicity values can be covered. The effect of various reversed-phase columns is manifested in the a, b, and c constants. Even the optimum mobile phase composition may differ from column to column [58]. Figure 2.9 shows the difference in the optimum mobile phase compositions for the log k versus log P relationships on two different stationary phases (Supelcosil LC-18 and LiChrosorb Rp- 18). The drawback of Eq. (2.11) is that it cannot be used for retention prediction from the log P values, as we cannot calculate two unknown values (S and log h) from one equation. Figure 2.8 shows the other possibility of using the S and log k0 values for expressing one single chromatographic parameter. From the S and log k0 values, an organic phase concentration can be expressed- at which the compounds have the same retention (log k = 0). It was found [59] that the chromatographic hydrophobicity index values (&) References pp. 90-92

66

Chapter 2

obtained in this way show correlation with the log P values for structurally unrelated pharmaceutical compounds as given by Eq. (2.12) for 140 compounds usine acetonitrile eluents. $0,AcN=9.31logP1-37.94,

n=140, r=0.88, s = 1 2 . 8

(2.12)

From the data of more than 400 pharmaceutical compounds obtained by using methanol as an organic modifier, Eq. (2.13) was obtained. $O,MeOH = 7.08 log P

+ 42,

n = 448, r = 0.787, s = 13.48

(2.13)

For Eqs. (2.12) and (2.13), n stands for the number of compounds, r is the correlation coefficient, s is the standard error of the estimate. It was found [59] that in general the $o = a log P + b relationship exists for a wide range of compounds, and a great variety of reversed-phase columns. It only depends on the type of organic modifier (methanol or acetonitrile). On the basis of the equation, a mobile phase composition can be estimated at which the retention time is expected to be double the dead time. As can be seen fiom the statistical parameters, the error is 12-13% concentration. This can be regarded as a rough prediction of the retention. The plot of log P values and $0,AcN values for 140 compounds (Eq. 2.12) can be seen in Fig. 2.10. The names of the compounds and the data were published by Valk6 and Slegel [59] and cover a wide range of pharmaceutically active sulphonamides, morphines, steroids, salicylates, benzodiazepines, barbiturates, etc. It should be mentioned that the pH of the mobile phase is also very important. In most retention prediction studies, the model compounds are not ionizable, so the mobfie phase pH does not influence the retention prediction significantly. In the case of pharmaceutical compounds, the situation is different. Most are slightly basic or acidic, therefore the mobile phase pH influences their protonation or dissociation, and thus their partition. All of

A AcN 120100-

-4

-2

0

4

2

6

0

10

Log P

Fig. 2.10. The plot of logP values and the chromatographic hydrophobicity index values ($0, A ~ N )for 140 compounds according to Eq. (2.12). (Reprinted with permission from ref. 59.)

Retention prediction ofpharmaceutical compounds

67

the relationships described above are valid only in that pH where the molecules are not ionized, as the measured or calculated log P values refer to the neutral molecules. To overcome the pH problem, and the application of different pHs in the mobile phase for measuring different types of structures, a general mobile phase composition was suggested by Roos and Lau-Cam [60]. They applied 1.5% acetic acid and 0.5% triethylamine in the mobile phase consisting of various concentrations of methanol and water. The data obtained for 140 pharmaceutical compounds were also included in the calculation of Eq. (2.13). In the above mobile phase system, the acidic compounds are in neutral form, due to the low pH, higher acetic acid concentration, while the protonated form of the basic compounds probably form a neutral ion-pair with the triethylamine. It was found [61], that the relationships described by Eqs. (2.10) and (2.1 1) exist also in ion-pair reversedphase chromatography. Similarly, good retention prediction has been reported in reversed-phase ion-pair chromatography using sodium dodecyl sulphate as pairing ion [62]. The retention prediction was made on the basis of the retention data and hydrophobicity parameters of pharmaceutical compounds. A similar approach to the third approach discussed above [58] was presented by Pate1 et al. in 1991 [63]. They also used Eq. (2.9) as the starting relationship. Their second equation was not Eq. (2.3) as in our case, but they described the dependence of the log k values on the quadratic relation to the organic modifier concentration (logk= a + bq5 + ~ $ 2 ) . They also found that the volume percent of the organic phase concentration can be related to the hydrophobicity of the solvent mixture, which can be calculated from the mole fraction of a solvent component and its hydrophobicity values according to Eq. (2.14): (2.14)

where xi is the mole fiaction of the ith solvent component, log Ps,i is the logarithm of the partition coefficient of the ith solvent, and n is the total number of pure solvents present in the solvent mixture. The calculated log P,, values were found to be related to the volume fraction of the organic solvent in the mobile phase according to Eq. (2.15).

9 = A' + B'( UPsm)

(2.15)

Considering the quadratic relationship between the log k values and the 9 values and the linear relation between the log k and the log P values (Eq. 2.9), the following general equation was set up: log k = A0 + @(log P/Ps&+ P(l0g P/Psm2)

(2.16)

Mathematically Eq. (2.16) is very similar to Eq. (2.1 1). By calculating the regression parameters (Ao, Bo and @), the mobile phase composition is optimized for the best modelling of the octanol-water partition of the molecule. Therefore this approach can be considered as analogous to our general approach for describing log P by the measured slope and the intercept values of the compounds. The only difference is that it allows even quadratic relationships between log k and organic phase concentration (while Eq. (2.1 1) References pp. 9&92

68

Chapter 2

supposes only a linear relationship), and the organic phase concentration is expressed by the mobile phase hydrophobicity value. The validity and prediction power of the equation were checked for structurally similar compound series (substituted aromatic compounds), under a variety of reversed-phase HPLC conditions (various types of organic modifiers and reversed-phase stationary phases). Statistically the correlations were always significant, the standard error of the log k prediction ranged from 0.150 to 0.987, which can be considered acceptable. The advantage of Eq. (2.16) is that it allows direct retention prediction at various mobile phase compositions from the known log P values of the compounds and mobile phase additives. To date, Eq. (2.16) can be considered as the most general and reliable retention prediction method for structurally unrelated compounds and applying various reversed-phase chromatographic conditions. Other approaches which use physicochemical descriptors but also other experimentally obtained constants which can be related to physicochemical parameters are discussed later. 2.7 RETENTION PREDICTION BASED ON EMPIRICAL INCREMENT VALUES

The calculation of the hydrophobicity of the molecule based on the chemical structure has more or less been solved. The calculation is based on the constant and additive hydrophobicity contribution of molecular fragments. As the RP-HPLC retention is also governed mainly by hydrophobicity of the compound, the question arises whether the calculation of the RP-HPLC retention can be carried out in a similar way from the retention contributions of the molecular fragments. The contribution of the substituents to retention have often been found constant and additive for a given chromatographic system. Based on this observation, reliable predictions of retention behaviour have frequently been reported in RP-HPLC C65-671. The theoretical basis of a structural retention increment database can be supported with the following equations. It is well known that the logarithmic retention factor (log k) is linearly proportional to the logarithmic distribution coefficients (log K ) of the compounds referring to the chromatographic partition system as described by Eq. (2.17). log k = log K + log( V,/V,)

(2.17)

where VJV, is the so-called phase ratio, the ratio of the stationary and mobile phase volumes. From Eq. (2.17) it is clear that if log K is a linear free-energy related parameter, log k is the same. As log K can be regarded as a logarithmic value of the chromatographic partition coefficients, on the basis of the Collander [64] relationship, linear correlation can be expected for the log P values as mentioned earlier in Eq. (2.9). By analogy with the log P predictions (Eq. 2.1S), the log k values also can be regarded as a sum of the Q log k,values referring to all fragments in the molecule (Eq. 2.19) log P = x n i or log P = C F , f,

(2.18)

Retention prediction of pharmaceutical compounds

69

where xi is the Hansch ;n value [68],J; is the Rekker fragmental constant [49] and Fi represents the number ofJ; fragments in the molecule. log k = C d l o g k;

(2.19)

From Eq. (2.19) it is clear that it is possible to predict the log k values of compounds if the dlog 4 values are available. But two important limitations of the retention prediction should be considered. First, as was also observed in the calculation of log P values, the group or fragment contributions are not always additive, the neighbouring substituents can influence each other’s partition or retention contributions. These effects are often negligible, or they can be involved in the calculations as a dlog k value attributed to the interactions. The second, more important limitation is that the log k value depends on the column and mobile phase composition to a great extent. The dlog 4 values can be collected from the log k values of two compounds differing from each other only in the presence or absence of the i substituent. Let us consider the change in the dlog 4 values for a change in the mobile phase composition. Equations (2.20) and (2.21) describe the relationship of the log k values with the organic phase concentration (@)by linear relationships discussed above.

where c refers to the compound without the i fragment and ci with the i fragment. On the basis of Eqs. (2.20) and (2.21), the dependence of the 61og Ki values on the mobile phase composition (@, volume fraction of the organic modifier), can be described by Eq. (2.22). 61og k, = (S,, - sc)$+ log /$Ic; - log k,

(2.22)

It can be seen that the higher the difference between the S values of the two compounds, the higher will be the dependence of 6 log ki values on the mobile phase compositions. If i refers to a small substituent, which does not influence the contact hydrophobic surface area of the molecules very much, we can expect very similar S values for compound ci and c. Then the difference is zero, the 6 log k; value will be independent from the mobile phase compositions. If we consider an average S value for compound ci, as -0.02 and 25% lower value for compound c, as -0.015, then the difference will be 0.005, which means that a 10% change in the organic phase concentration will cause a 0.05 change in the d log ki values. This value is slightly greater then the general error in log k measurements. It means that the dependence of the fragmental retention contribution on the organic phase concentration is very small, but theoretically cannot be neglected. On the basis of the above considerations, a huge database has been set up for the 6 log k values for structural fragments of pharmaceutical compounds for the most common metabolic changes [69]. More than 400 metabolic transformation routes were collected, and the d log kivalues were calculated from literature data. For example, the most References pp. 90-92

70

Chapter 2

CH3

0

1

4

c?

CONH2

10

9

OH

FH2

15

13

Fig. 2.11. Structure of the compounds investigated. Compound 1-3, pyridopyrimidine derivative and its metabolites; compounds 4-10, diazepam, oxazepam, uxepam and their metabolites; compounds 11-13, 3trifluoromethyla-ethylbenzhydrolderivatives.

common metabolic hydroxylation was characterized by two values of 6 log k, -0.35 and -0.15 for aliphatic and aromatic hydroxylation, respectively. These numbers were obtained from literature data for the differences of log k values of hydroxylated and dehydroxylated compounds obtained in a reversed-phase HPLC system. In general, retention data of a minimum of four pairs of compounds were considered in the Q log ki value. When literature data were not available for a certain compound pair, the 6 log k value for the i metabolic transformation route was estimated fiom the Hansch n value. On the basis of Eqs. (2.12) and (2.13) and our observation that a 10% increase in acetonitrile and methanol causes 0.285 and 0.298 log k change, respectively, it can be calculated that a unit change in log P value will result in 0.204 and 0.270 change in log k values. Thus, for example, the presence of a methyl group which has a rc value of 0.56 can be expected to increase the retention by 0.149 and 0.1 14 log k value with acetonitrile and methanol, respectively, as the organic modifier. Although the change in retention time caused by a given substituent can vary over a wide range, the 6 log k value can be regarded as a constant, neglecting its dependence on the organic phase concentration. Similarly for the log P predictions, when n and f values of a molecular fragment can be changed by an adjacent substituent which influences the prediction, the 6 log k values can also be different, reflecting their neighbouring substituents for the same reason. Moreover, the 6 log k values may vary on different reversed-phase columns from different manufacturers, as the

Retentionprediction of pharmaceutical compounds

71

TABLE 2.4 MEASURED (mtR) AND THE PREDICTED (ptd RETENTION TIME DATA (log k) VALUES FOR THE 15 INVESTIGATED COMPOUNDS Column I (to = 1.40 min)

1 2 3 4

5 6a 7b

7 8 9

LO 11 12 13 14 15

mtR

PtR

11.1 4.7 5.2 13.0 8.7 7.5 5.7 5.7 20.8 5.9 4.9 4.4 12.7 3.7 8.4 6.0

-c

6.0 3.9 8.7 9.6 6.6 5.3 16.0 4.2

9.9 5.0 5.3

'

Column I1 (to = 2.38 min)

Column 111(to = 1.60 min)

bgk

mtR

PtR

bgk

mtR

PtR

logk

0.84 0.37 0.43 0.92 0.71 0.64 0.49 0.49 1.14 0.51 0.40 0.32 0.91 0.22 0.70 0.51

24.4 11.0 12.3 14.3 10.8 9.4 8.0 8.0 19.6 9.4 8.4 6.2 10.4

12.2 8.8 9.9 10.8 8.3 6.8 21.4 6.8

0.97 0.56 0.62 0.70 0.55 0.47 0.37 0.37 0.86 0.47 0.40 0.21 0.53 0.16 0.74 0.54

16.6 6.3 7.3 19.6 12.1 9.2 7.4 7.4 29.8 7.2 5.7 3.9 11.5 3.3 13.5 7.7

-

1.00 0.50 0.58 1.08 0.85 0.71 0.59 0.59 1.28 0.54 0.41 0.15 0.79 0.02 0.87 0.58

5.8

14.3 9.9

-

13.1 4.9 9.1

8.3 5.1 13.0 14.3 9.0 6.4 21.3

-

5.1 8.1 4.7 8.3

Reprinted with permission from ref. 69. All PtR vdues were calculated from the mtR of the parent compounds, except as indicated in the footnotes. Column I, Hypersil ODs; Column 11, LiChrosorb ODs; Column 111, Sepharon RPS. a PtR calculated from mtR of 4. b PtR calculated from mtR of 5. The measured retention times of the parent compounds were the input data for the prediction of the retention times of the metabolites.

log k values can be different. The effect of the adjacent groups and different reversedphase columns on the 6 log k, values, and on the retention prediction has been investigated for a few pharmaceutical compounds [69]. The purpose of the work was the application of the retention prediction method described above for predicting the retention of metabolites of known pharmaceutically active compounds. The chemical structures of the compounds investigated can be seen in Fig. 2.1 1. The retention prediction was carried out for hydroxylation, demethylation, decarboxylation on the various molecules. The retention measurements were carried out on three different RP columns in order to reveal the effect of the stationary phase. Hypersil ODS, LiChrosorb ODS, and Sepharon RPS stationary phases were involved in the study [69]. The measured and predicted retention values for the compounds investigated are listed in Table 2.4. The observed 6 log k, values for the metabolic transformations on the various molecules are listed in Table 2.5. In order to reveal the retention changes on the three columns for the same structural change, the differences between the 6 log k,values obtained for the C-hydroxylation of compounds 1,4,5, and 12 and for the N-demethylation of compounds 4,6,9, and 14 are summarized in Table 2.6. References pp. 90-92

Chapter 2

12

The average 6 log ki values and their standard deviations were also calculated. In comparison with the average and the standard deviation for the measured 6 log k values on the three types of columns and for the same structural change obtained on the same columns, but different mobile phase compositions, it can be seen that the standard deviation is much higher for the 6 log k values measured for the same structural change (i) but in different molecules. This suggests that the weakest point of the prediction is the effect of neighbouring molecular fragments, and not the application of different columns and mobile phase compositions. The average standard deviations were 0.049 (expressed in log k) for demethylation and 0.104 for hydroxylation, which are acceptable, especially when expressed as retention time values (1.8 min). The relatively high standard deviation of 6 log ki values obtained for the same structural change on various molecules can be explained from a physicochemical point of view, namely the N-methyl group is not basic in compounds 4,6, and 9 as in 14, because the first three are carboxamides. For this reason, the change in hydrophobicity and as a consequence in RP-retention cannot be expected to be the same. The effect of C-hydroxylation in 6 log k when there is a possibility of intramolecular hydrogen bond formation is much smaller. That is possibly the case for compound 2. It can be seen from the examples presented that the error in the prediction is higher than the variation of 6 log k values due to a different reversed-phase column or conditions. This retention prediction method was applied only for the retention prediction of metabolites on the basis of the measured retention parameters of the mother compounds. In each case a change of never more than one 6 log 4 value was used for the retention prediction. The method was not considered suitable for the prediction of the retention of a molecule by building up a summation of all its fragmental 6 log k values. With this restriction, the method seems to be reliable and it was compiled in an expert system developed by Compudrug [70] called HPLC-MetabolExpert for predicting the chemical structures and RP-HPLC retention of possible metabolites of drug molecules. A very similar approach for retention prediction of basic drugs were developed by Hindriks et al. [71]. The aim of the work was to develop an expert system for the selecTABLE 2.5 THE OBSERVED Slog k1 VALUES FOR THE METABOLIC TRANSFORMATIONS ON THE VARIOUS MOLECULES ON THE THREE COLUMNS AND THE AVERAGE (A) Slog k VALUES AND THE Slog k VALUE TAKEN FROM THE DATABASE (DB) Metabolic route

6 log k Column I

Column I1

Column III

A

DB

-0.470 -0.279 -0.220 -0.723 -0.204 -0.152 -0.110 -0.183 +0.598

-0.407 -0.230 -0.180 -0.371

-0.498 -0.274 -0.260 -0.773 -0.231 -0.115 -0.136 -0.290 +0.643

-0.458 -0.261 -0.220 -0.622 -0.195 -0.I24 -0.105 -0.223 +0.521

-0.35 -0.15 -0.15 -0.50 -0.20 -0.20 -0.20 -0.25 +0.45

_ _ _ _ _ ~

+OH on 1 +OH on 4 +OH on 5 +OH on 12 -CH3 on 4 -CH3 on 6 -CH3 on 9 -CH3 on 14 -COOH on 1 1

Reprinted with permission from ref. 69.

-0.151

-0.105 -0.068 -0.196 +0.323

Retention prediction of pharmaceutical compounds

73

TABLE 2.6 THE 61og k VALUES CAUSED BY THE METABOLIC HYDROXYLATION AND DEMETHYLATION OF THE COMPOUNDS ON THE THREE COLUMNS Reaction

Compound pair

Column I

Column I1

Column 111

Average

Hydroxylation

1-2 4-6 5-7 Average: Std. dev. 12-13 4-5 6-7 9-10 14-15 Average Std. dev.

-0.470 -0.279 -0.220 -0.323 f0.107 -0.723 -0.204 -0.152 -0.110 -0.183 -0.162 f0.047

-0.407 -0.230 -0.180 -0.272 f0.097 -0.371 -0.151 -0.105 -0.068 -0.196 -0.130 f0.064

-0.498 -0.274 -0.260 -0.344 f0.109 -0.773 -0.231 -0.115 -0.136 -0.290 -0.193 f0.088

-0.458 f 0.038 -0.261 f 0.022 -0.220 f 0.033 -0.313 f 0.030 f0.104 -0.622h0.179 -0.195 f 0.040 -0.124 f 0.024 -0.105 f 0.034 -0.223 f 0.053 -0.162 f 0.026 f0.049

Phenolic OH Demethylation

(Reprinted with permission from ref. 69.

tion of initial HPLC conditions for the analysis of pharmaceuticals. They investigated the retention behaviour of 600 basic compounds belonging to the class of CNS-active and cardiovascular drugs. The relative measure of polarity of the compounds was expressed by retention index values, mentioned earlier [6,8,9], which can be used to characterize the polarity of the molecule and to link the structure elements to some type of polarity descriptor. Under strict conditions, it was shown that the retention index (RI) values are constant. RI values for 300 compounds in combination with chromatographic data on the purity analysis of more than 300 compounds were the basis of the knowledge base. They chose to estimate the polarity of a given molecule on the basis of the presence of polar and non-polar groups. The expert system developed calculates the polarity of a new compound from its structure and expresses the result as a percentage of organic modifier (methanol) in the mobile phase. Before this, the structure is subdivided into fragments or structural elements. These elements are so defined that they can describe a structure in a simple and unambiguous manner. Examples of such elements are phenyl, methyl, hydroxyl and tertiary nitrogen. All initially selected structural elements are shown in Table 2.7. The percentages listed in Table 2.7 are essentially derived from experimental data and from the Rekker’s fragmental constant values [49] for hydrophobicity contributions. The authors [71] explain the effect of the pH of the mobile phase on the fragment value of the protonated groups. They claim that most fragment contributions are independent of the pH. They did not mention the effect of the neighbouring substituents. The differences in reversed-phase columns from various manufacturers are expressed by the so-called zero-level contributions. They determined experimentally the zero level for a NovaPak C18column at 43% of methanol. In other words, methanol (%) = Z(fiagment contributions) + 43%. The same holds for apBondapack C18column, except that the zero level is 2% lower. For most applications, tetramethylammonium phosphate buffer was used to block the silanol sites of the reversed-phase material to avoid the silanol effects on the retention of the basic drug molecules. They stated that the most difficult task References pp. 90-92

Chapter 2

74

TABLE 2.7 SOME STRUCTURAL ELEMENTS AND THEIR EFFECTS ON THE PREDICTED PERCENTAGE OF METHANOL AT pH 7.4 AND pH 4.0 ACCORDING TO HINDRIKS et al. [71] Structural element

9

10

11

12 13 14 15 16 17 18

Phenyl, monosubstituted (c&) Phenyl, disubstituted (CsH4) Phenyl, trisubstituted (c&) CI on aromatic group CI on aliphatic group OH on aromatic group OH on aliphatic group 0 atom in ether. The oxygen is positioned between: a Two aromatic groups b An aromatic and an aliphatic group c Two aliphatic groups 0 atom in ketone. The carbon connected to the oxygen is positioned between: a Two aromatic groups b An aromatic and an aliphatic group c Two aliphatic groups S atom. The sulphur positioned between: a Two aromatic groups b An aromatic and an aliphatic group c Two aliphatic groups Pyridine CH3 CH2 CH C N atom in ring plus double bond N atom in two rings Other N atoms: first one every next one

Methanol (%) pH 7.4

pH4.0

+I 1 +I0

-10

+11 +I0 +9 +7 +I -2 -10

-5 -5 -10

-5 -5 -10

-5 -6 -10

-5 -6 -1 0

+3 +I

+3 +I -3

+9 +7 +1

-2

-3 +3 +5

+3 +2 +I -5 0 -5 -5

-5 +5 +3 +2

+I -5 0 -3 0 -5

Reprinted with permission from ref. 7 1.

proved to be to describe correctly and yet in a simple way the polarity of a sample molecule on the basis of its chemical structure. The approach proved to be successful, but also showed limitations, However, the accuracy of the prediction of the initial methanol concentration was not shown, and also the limitations were not discussed. It was not clear what was the expected retention for the calculated methanol concentration in the mobile phase. Theoretically this approach is very close to the approach of the chromatographic hydrophobicity index, discussed earlier [28], where the mobile phase composition is predicted to provide log k = 0 retention from the log P values of the compounds. The third important approach was published recently by Dimov and Moskovkina [72]. Equations for the dependence of the retention in RP-HPLC on molecular mass and selected structural fragments of 18 benzodiazepine derivatives were proposed. They sug-

Retention prediction ofpharmaceutical compounds

75

gested a biparametric model based on the additivity principle for a general description of chromatographicretention for predictive purposes as shown by Eq. (2.23). n

n+k

/=l

j=n+1

R=bo + x b i B i + C b i T i

(2.23)

where R represents the corresponding retention (k), B, are basic and are tuning contributors to retention. The bo and bj are regression constants. It was accepted [73] that the B term in Eq. (2.23) includes solute properties, allowing the calculation of the retention parameter (R), which does not differ from the experimental value by more than *10-15%. The T term also includes solute properties, which can correlate insignificantly with retention and do not correlate with the properties included in the B term. The retention data of the benzodiazepine derivatives were taken from the literature. The molecular mass, M,, as a general property was tested as a B contributor. Molecular fragments such as C=O, -OH, -F, -NOz, N-R2 and flat rings (phenyl, cyclopropane) were tested as T contributors. An indicator variable was used to represent the presence or absence of the fragments in the molecule. On the basis of the data, the regression parameters were calculated for Eq. (2.23). Very good agreement between the measured and calculated retention parameters were found. The results showed that there are fragments selected from the solute molecule which are responsible for retention and these can be called chromatophores. Their contributions are additive, but in some instances the fragment evaluation can be tuned so that a more accurate equation can be obtained. The evaluation from both the first and second groups of equations allows quantitative considerations of the contributions of different solute fragments, while the difference in a given fragment evaluation could be used to consider intramolecular interactions. In conclusion, the retention prediction of pharmaceutical compounds can be carried out on the basis of fragmental retention contributions. Three methods have been discussed. The first method uses a large database (400 metabolic transformations) of the retention increment values caused by the change in chemical structure, which usually occurs on metabolic transformation. This expert system approach can be applied for the retention prediction of metabolites from the measured retention data of the mother compounds under certain chromatographic conditions. It was shown that the effect of the mobile phase composition, and the type of reversed-phase column are negligible. The effect of the neighbouring structural fragment on the molecule caused higher deviation from the predicted retention. The second approach also uses an expert system, and was developed mostly for basic drug molecules on the basis of the experimental retention data of 600 compounds. From the fragmental hydrophobicity constants, a methanol concentration value was calculated which should be added to the mobile phase to achieve a certain retention, The effect of various columns and mobile phase pH were taken into consideration in the fragmental values and the calculation methods. This system is suggested for building up the retention of the whole molecule from its fragmental contribution values, however, the precision and accuracy of the system has not yet been tested. The effect of intramolecular interactions on the retention prediction is not known. The third approach seemed to be applicable for other chromatographic systems as well, not only for RP-HPLC. The advantage of the method is that it takes into account the intramolecular interactions between the fragments of the molecules, and their role in the Referencespp. 90-92

76

Chapter 2

retention. The disadvantage of the method is that it needs a large amount of experimental data to calculate the regression coefficients of the equation used for the prediction. Different equations should be applied to different chromatographic conditions. It is also not explained how accurate is the retention prediction for non-congeneric compounds or for compounds differing in structure from the investigated series.

2.8 RETENTION PREDICTION BASED ON EXPERIMENTAL RETENTION VALUES, THERMODYNAMICCONSIDERATIONS WITH MULTIPARAMETERAPPROACHES Structure-retention relationships have been studied by numerous chromatographers for predictive purposes. The derived quantitative structure-retention relationships usually allow the prediction of the retention behaviour of a given solute of a given class. The most common way is to convert the retention data to a linear free-energy related (LFER) value usually the log k, and a linear regression equation can be set up using other LFER parameters as independent variables. The relationships are more reliable and show wider range of validity if they are based on multivariate analysis of experimentally determined retention data and physicochemical data. Using more than one independent variable to describe the variation in the retention data increases the reliability and accuracy of the prediction. The drawback of using a great number of variables in the quantitative structure-retention equation is that a large number of compounds with precisely measured retention data are needed to set up the equation and for calculating the regression coefficients. In general, to introduce a new independent variable into the equation, a minimum of five compounds are required with their measured retention data and measured or calculated physicochemical descriptors. This means that when an equation contains three independent variables, the data of a minimum of 15 compounds are necessary to calculate the regression coefficients and the mathematical statistical parameters of the equation. In order to reveal the true value and the predictive power of the equations, the multiple regression coefficients, the standard error of the estimates and the Fischer-test value should be given. The multiple correlation coefficient reflects the proportion of the explained and unexplained variance of the retention parameters for the compounds involved in the calculation. The standard error of the estimates expresses the =k deviations of the predicted retention value from the true value, i.e. it shows the predictive power of the equation. It is always worth checking how it is related to the experimental error of the retention determination. The Fisher-test value shows the probability that the equation reveals a true relationship. It is also very important to check the significance level of the variables. The most common way to show that the application of a given variable significantly increases the explained variance of the dependent variable (usually the retention parameter) is to give the f values of the regression coefficients with 95% probability level. If it is significantly different from zero, the variable can be regarded as significant. In order to estimate the importance of a given variable in a multivariable regression equation to explain the variance of the dependent variable, the so-called b-weight can be calculated. The bweight values show the role of the variable for the prediction and a real rank can be set up among the variables. The independence of the independent variables is also very impor-

Retention prediction of pharmaceutical compounds

77

tant. To reveal the correlation of the variables involved in the equation, the correlation matrix provides valuable information. Briefly, these are the most important mathematical statistical considerations in a multivariate quantitative structure-retention equation. A detailed description of the mathematical basis of step-wise regression analysis can be found in the monograph by Draper and Smith [74]. The application of regression analysis for setting up quantitative structure-retention relationships was given by Woodburn et al. [75]. Retention of several non-polar solutes on two reversed-phase liquid chromatographic supports (C-2 and (2-8) was examined during isocratic, isothermal elution with binary mixtures of methanol-water and acetonitrile-water. The log k values were correlated with the following indices of solute hydrophobicity and molecular topology: octanol-water partition coefficients (log P), hydrophobic surface area (HSA), and first-order molecular connectivity indices k').For each stationary phase-solvent combination, one regression equation was required to describe the data for polycyclic aromatic hydrocarbons and halobenzenes, and another for alkylbenzenes, as can be seen from the following equations:

log k - 2 B ~ 0 . 8 (0h 0.02) log P-2.70 (+ 0.2),

n = 9, r = 0.959

(2.25)

log kc-8 A = 0.84 (*0.05) log P - 1.20 (* 0.2),

n = 11, r = 0.996

(2.26)

log kC-8B = 1.00 (k 0.04) log P - 1.50 (+O.l),

n = 9, r = 0.999

(2.27)

Equations (2.24X2.27) refer to 60:40 methanol-water mixture as mobile phase. A refers to the solute group of benzenes, polyaromatic hydrocarbons (PAHs), and halobenzenes, while B refers to the group of alkylbenzenes. The following equations show the similar correlations obtained in a 50:50 acetonitrile-water mobile phase. log kc-2

= 0.35

(* 0.30) log P - 0.5 1 (* 0.1),

log kc-z,B = 0.51 (* 0.02) log P - 0.91 (h 0.06), log kc.S,A

= 0.55

(* 0.55) log P - 0.19 (5O.l),

= 0.73 (+ 0.03) log P - 0.60 (* 0.07), log k-8,B

n = 11,

Y = 0.993

(2.28)

n = 9, r = 0.998

(2.29)

n = 11, r = 0.995

(2.30)

n = 9, r = 0.999

(2.3 1)

From the above equation it is clear that every mobile phase composition and stationary phase produces different constant values in the correlation analysis. Also the parameters are different with different group of compounds. The same authors [75] found significant correlations between the retention data and hydrophobic surface area, and the molecular connectivity indices, again with different parameters with different mobile and stationary phase compositions. As the physicochemical and topological descriptors investigated showed high correlation with each other, the multiple regression analysis could not be used. The different parameters of the correlations for the data of polycyclic aromatic hyReferences pp. 90-92

78

Chapter 2

drocarbons and halobenzenes, and alkylbenzenes were attributed to the differences in the nature of the interaction of these two groups of solutes with the bonded, n-alkyl chains. Solute molecular size, shape, and conformation as well as hydrophobicity appear to be the dominant factors controlling the solute retention. Several efforts have been made towards describing quantitative structure-retention relationships simultaneously accounting for changes in mobile and stationary phase compositions and including structurally unrelated compounds. A very interesting and promising approach was derived by Jandera [76-781. His starting point was the linear relationship between the log k values with the organic phase composition as described by Eq. (2.3) (log k = S#I+ log b),investigating the retention data of a homologous series of solutes at several different concentrations of the organic solvent used in the binary mobile phase. From the experimental data, he found another two relationships for the parameters Sand log in Eq. (2.3) as described in Eqs. (2.32) and (2.33). log ko = an, + b

(2.32)

+e

(2.33)

S = d log

where a, b, d, and e are constants, n, is the number of carbon atom in the solute of the homologous series. Introducing these parameters to the original Eq. (2.3), the following equation can be set up: log k

= $(d log ko

= $d(an, = (an,

+ e) + (an, + b)

+ b) + $e + (an, + b)

+ b)(l + $4+ $e

(2.34)

This equation means that the dependence of the log k values can be described by the organic phase concentrations considering the a and d constants, which according to Jandera do not depend significantly on the character of the homologous series but on the organic solvent used as the less polar component of the mobile phase. The constants e and b depend also on the type of the homologous series and the stationary phase used. The constants can be estimated by the measured retention data of a homologous series in a given stationary phase with a given organic modifier. These constants are supposed to be generally valid for other non-homologous series of compounds as well. The n, values obtained in this way, which are originally the carbon numbers, are equivalent to the nonspecific contributions, while e refers to the polar contribution to the retention. Jandera has verified experimentally his theoretical approach to retention prediction of substituted aromatic compounds. He also showed [77,78] that the constants obtained showed good correlations with the Hansch type hydrophobicity parameter, n,and the Snyder's P' polarity indices [79]. The experimentally determined constants can be used as molecular descriptors for retention prediction in general. The theoretical approach seems to be very logical and supported by a large number of experimental data, but according to the author of the present chapter one of the early assumptions for the derivation is not always true.

Retention prediction ofpharmaceuticai compounds

79

Namely, the high correlation between the slope (8and the intercept values (log k,,) cannot be expected, except for structurally related compounds [34]. The compound series investigated (substituted aromatic compounds) can be regarded as structurally related according to our investigations. The advantage of the approach is the application of two independent variables to describe the retention, which are not in correlation with each other and sufticiently good prediction can be achieved. The other multiparameter approach which tries to include all three chromatographic variables (solute structure, mobile phase composition, and stationary phase properties) was published by Kaliszan et ul. [80,81]. A set of 12 substituted benzene derivatives with a wide range of substituent properties were selected for the investigations. Changes in composition of the methanol-water mobile phase ranged from 35 to 65% to assure the linear range of the log k versus organic phase concentration plot. A set of the equations (2.3) were investigated as a function of the solute ( i ) and stationary phase 0'). (2.35)

Using multivariate regression analyses, it was found that S!, and log k,,,i can satisfactorily be described by a two-parameter equation involving the quantum chemically calculated total energy of a solute, and its polarity parameters, defined as the maximum excess electronic charge difference in a molecule. An impressive agreement was found between the measured and the calculated retention data of the 12 solutes on three stationary phases and various mobile phase mixtures. As with all the multivariate approaches, this approach also has the limitation that it is valid only within the series of compounds investigated, which can be regarded as similar and simple in structure considering the wide variety of pharmaceutically active drug molecules. Increasing the number of measured data for a wider range of compounds can probably enhance the predictive power of the above-mentioned quantitative structure-activity equations. Another multivariate approach was introduced by Zou et ul. for predicting the S and the log k, values as important chromatographic retention parameters at which the retention prediction can be carried out for various mobile phase compositions within the linear range. They suggested using the solvatochromic parameters as independent variables and solute descriptors. The foilowing molecular descriptors for the solvatochromic effect were suggested: V, as a cavity term, which measures the endoergic process of separating the solvent molecules to provide a suitably sized enclosure for the solute; n* measures the exoergic eXects of the solute-solvent dipoledipole and dipole-induced dipole dielectric interactions, B, and a, measure the exoergic effects of hydrogen bonding involving the solvent as a hydrogen bond donor acid and as a hydrogen bond acceptor base and the solvent as a base and the solute as an acid, respectively. V, can be estimated by simple additivity methods such as those of Bondi [84] or Abraham and McGowan [ 8 5 ] , n*,P, and amare solvatochromic parameters that can be found in a paper by Kamlet et ul. [86] or measured by UV, IR or NMR spectroscopic methods [87]. With the help of these parameters, the following multivariate regression equations were suggested to predict the S and the log k,values: (2.36) References pp. 90-92

Chapter 2

80

where p i and qi (i = 1-5) are regression coefficients, derived using conventional linear regression analysis. With the help of Eqs. (2.36) and (2.37) the relative importance of the solvatochromic parameters for describing the variance of the important S and log k,,, retention parameters can be revealed. The authors [82] described significant correlations for a small number of substituted aromatic compounds in three different mobile phase systems (methanol-water, acetonitrile-water, tetrahydrofurane-water). Table 2.8 shows the parameters obtained for Eqs. (2.36) and (2.37) in three different mobile phases. Table 2.9 shows the compounds investigated with their solvatochromic parameters. Table 2.10 reveals the predictive power of the equations obtained by summarizing the experimental and the predicted log and S values and their differences referring to the acetonitrilewater mobile phases. Unfortunately, only the data of 12 compounds were used to calculate the regression parameters of a multivariate equation with four independent variables, which makes the results questionable, and also the predictive power of the equations was not tested outside the series of compounds investigated. The differences in the regression parameters obtained (pi and qi) obtained with three different mobile phase are explained by the differences in the Hildebrandt solubility parameters [88]. It was also found that the log values extrapolated from the different binary mobile phase systems were not the same. The possible explanation may lie in the sorption of the organic modifiers in the stationary phase, and the extrapolated log values contain contributions from the cavity process and the adsorbed organic modifier. The method is worth considering for pharmaceutical compounds, but further evidence is needed for the underlying mathematical statistical significance of the theory. Galushko [89] recently described a method for calculating the retention and selectivity in RP-HPLC based on the molecular structure of the analyte and the characteristics of the sorbents and mobile phases. The approach is based on solvophobic theory. The stationary TABLE 2.8 THE PARAMETERS OF Eqs. (2.36) AND (2.37) AS WELL AS THE REGRESSION COEFFICIENTS WITH THREE DIFFERENT MOBILE PHASE SYSTEMS OBTAINED BY ZOU et al. [82] Mobile phase

1 2 3

p1

0.161 0.355 1.104 41 ~~

1 2 3

P2

4.829 3.664 2.428 42 ~

-1.029 -1.060 -0.873

-4.089 -3.257 -3.858

P3

P4

P5

R

-0.1192 0.0115 -0.2731

-0.5440 -0.4987 0.1398

-3.417 -2.841 -2.307

0.983 0.991 0.953

43

44

45

~~

~

0.0859 -0.2228 -1.0555

0.1424 0.0026 -0.9422

2.324 2.263 3.325

R ~~

0.971 0.962 0.962

Reprinted with permission from ref. 82. Mobile phases: 1, methanol-water; 2, acetonitrile-water; 3, tetrahydrofuran-water.

Retention prediction of pharmaceutical compounds

81

TABLE 2.9 SOLVATOCHROMIC PARAMETERS USED IN THE CORRELATION STUDY FOR Eqs. (2.36) AND (2.37) BY ZOU et al. [82] Solute

VJlOO

?c*

Bm

am

Aniline Acetophenone Anisole Benzaldehyde Benzene Benzonitrile Diethyl phthalate Ethylbenzene Methyl benzoate Nitrobenzene p-Nitrophenol Phenol n-Propylbenzene

0.562 0.69 0.63 0.606 0.491 0.59 1.153 0.687 0.736 0.631 0.676 0.536 0.785

0.73 0.90 0.73 0.92 0.59 0.90 0.84 0.53 0.76 1.01 1.15 0.72 0.51

0.50 0.49 0.32 0.44 0.10 0.37 0.82 0.12 0.39 0.30 0.32 0.33 0.12

0.16 0.006 0 0 0 0 0 0 0 0 0.93 0.61 0

Reprinted With permission from ref. 82.

phase surface layer is regarded as a layer of a liquid hydrocarbon but as a specific layer containing surface-fixed alkyl radicals and some amount of mobile phase components. The basis of the theory is the general equation which describes the relationship between the retention (log k) and the free-energy change of solvation energies in the distribution system (AG).

-AG lnk=-+log(V, RT

/Vm)

(2.38)

TABLE 2.10 COMPARISON OF EXPERIMENTAL DATA FOR LOG k,AND -5’WITH VALUES CALCULATED FROM Eqs. (2.36) AND (2.37) WITH ACETONITRILE-WATER AS MOBILE PHASE ACCORDING TO ZOU et al. [82] Solute

Acetophenone Anisole Benzaldehyde Benzene Benzonitrile Diethyl phthal. Ethylbenzene Methyl benzoate Nitrobenzene p-Nitrophenol Phenol nPropylbenzene

1% kw

S

Exp.

Calc.

Diff.

Exp.

Calc.

Diff.

1.42 1.86 1.36 1.86 1.54 2.30 2.64 1.82 1.80 1.49 1.06 2.83

1.50 1.76 1.34 1.88 1.48 2.36 2.54 1.95 1.82 1.47 1.09 2.90

0.08 -0.10 -0.02 0.02 -0.06 -0.04 -0.10 0.13 0.02 -0.02 0.03 0.07

-2.28 -2.62 -2.22 -2.57 -2.44 -3.22 -3.37 -2.61 -2.66 -2.81 -2.19 -3.27

-2.40 -2.55 -2.24 -2.56 -2.34 -3.15 -3.14 -2.74 -2.66 -2.79 -2.22 -3.46

-0.12 0.07 -0.02 0.01 0.10 0.07 0.23 -0.13 0 0.02 -0.03 -0.19

Reprinted with permission from ref. 82

References pp. 9&92

Chapter 2

82

where V, lV, is the phase ratio. The energies required to generate a cavity of molecular size in the stationary phase layer and the mobile phase can be described by Eq. (2.39) according to Horviith and Melander [90]. G,

= NAz

+ NAl~(ke1-1)

(2.39)

where N is Avogadro’s number, A is the cavity surface area in the liquid, z is the surface tension, A l is the solvent molecule area and kel is the characteristic constant for every liquid [9 1,921. Considering that the cavity shape is spherical the A molecular area values can be calculated from the V, increments in partial molar volumes of fragments. A large set of experimental values of partial molar volumes for different compounds [93] and simple calculation methods [94] are available. The van der Waals and electrostatic interactions also should be calculated for the estimation of AG. The author suggests that the molecule of a substance can be considered as consisting of dipoles, each of which separately interacts with the surrounding continuum. Thus, by using the dipole moment values, the effective radius of an imaginary sphere in which the dipole is located and epsil dielectric permittivity values, the interaction energy can be described. He says that such an approach does not need quantum chemical methods to calculate the atom charges. The bond dipole moments are determined for almost all bonds for various compounds [95,96]. More precisely, each dipole is not surrounded by a totally closed sphere of solvent molecules. The proposed approach is based on the assumption that in different compounds the parameters of a ball segment in which the same dipole is located vary within a small range, so that to TABLE 2.11 INCREMENTSFOR THE FREE-ENERGY CHANGE ( A G e , s , ~ 2 0FOR ) SOME DIPOLES ACCORDING TO GALUSHKO [89] Dipole csp2-H csp3-H csp2-csp3 csp2-csp csp3-csp C-O

c=o

C-N C=N 0-H (aromatic) 0-H N-H N-O N=O C-cl

c-s

1.49 1.49 1.80 1.80 1.80 1.66 1.74 1.69 2.37 1.35 1.20 1.38 1.55 2.20 2.34 1.80

1.00 1.oo 1.oo 1 .oo 1.00 1.oo 1.05 1.00 1.40 1.00 0.89 1.oo 1.oo 1.42 1.30 1.00

0.70 0.40 0.68 1.48 1.48 0.70 2.40 0.45 3.10 1.51 1.51 1.31 0.30 2.00 1.59 0.90

4.36 1.42 2.33 6.68 11.06 3.15 32.20 1.24 21.25 27.28 38.80 19.20 0.71 11.04 5.80 4.09

Reprinted with permission from ref. 89. aFor calculation of a, = (rl + r2)/2, the van der Waals radii (r) were used: c = 0.18; H = 0.117; 0 = 0.152; N = 0.15; Cl = 0.18 nm.

Retention prediction of pharmaceutical compounds

83

10

4

4

6

8

10

I n kf

Fig. 2.12. Comparison of the retention factors (Ink) in water on Merck RP-18 and ODs-Hypersil stationary phases. Compounds: 1, phenol; 3, nitrobenzene, 4, rn-dinitrobenzene; 6, chlorobenzene; 8, naphthalene; 9, benzophenone; 10, benzene. (Reprinted with permission from ref. 89.)

calculate the electrostatic energy, this parameter can be approximated by the effective radius of the sphere. However, a lot of parameters in the theoretical equation are not known. The author suggests measuring the retention data of a minimum of three model compounds differing in structure (i.e. benzene, phenol, benzophenone) on a given stationary phases to estimate the stationary phase parameters. The paper [89] provides a list of parameters such as the increments of partial molar volumes for some fragments, freeenergy change of increments for some dipoles, as can be seen in Table 2.1 1. It was also found that retention data of ten substituted phenol derivatives obtained on two different reversed-phase stationary phases (Merck RP-18 and ODs-Hypersil) did not show correlation as can be seen in Fig. 2.12. This problem can be overcome by the calculated stationary phase parameters from the retention of standard compounds. The predictive power of the equations and the theory was tested again on 32 substituted aromatic compounds. Some compounds showed large Referencespp. 90-92

Chapter 2

84

TABLE 2.12 THE COMPARISON OF THE CALCULATED AND EXPERIMENTAL RETENTION DATA (Ink) OBTAINED BY GALUSHKO’S RETENTION PREDICTION METHOD [89] (STATIONARY PHASE: MERCK RP-18) Compound

In kcalc

In kexp

Difference

Aniline Dimethyl phthalate Phenol 2,4-Dimethylphenol Benzyl alcohol Quinoline Benzaldehyde Anisole o-Nitroaniline NJ-Dimethylaniline m-Nitrophenol Toluene 2-Phenylethanol Chlorobenzene m-Dinitrobenzene Diethyl 0-phthalate Benzonitrile Benzophenone 1-Phenylethanol Ethylbenzene n-Nitroacetophenone Anethole 0-Cresol Diphenyl ether Acetophenone Biphenyl Nitrobenzene Naphthalene 3-Phenylpropanol Anthracene N-Methylaniline Benzene

2.63 4.84 3.10 5.13 3.26 5.13 3.80 6.20 2.34 7.38 2.74 5.86 4.22 6.08 4.14 6.40 4.00 6.57 3.91 7.12 4.01 8.60 4.30 8.18 4.47 8.34 4.45 7.17 4.85 9.22 5.08 4.95

2.94 5.09 3.13 5.22 3.26 5.58 3.72 5.77 3.81 6.26 3.89 6.27 3.89 6.44 3.99 6.46 4.00 6.96 4.04 7.38 4.18 8.13 4.23 8.58 4.34 8.91 4.42 11.81 4.94 12.84 5.01 4.95

0.31 -0.25 0.03 0.09 -0.45 -0.08 0.43 -1.47 1.12 -1.15 -0.41 0.33 -0.36 0.15 -0.06 -0.39 -0.13 -0.26 -0.17 0.47 0.07 -0.40 0.13

-0.57 -4.64 -0.09 -3.62 0.07 -

Rreprinted with permission from ref. 89.

deviations between the measured and predicted retention values as shown in Table 2.12. This was attributed to the effect of the neighbouring substituents which was not taken into account in the calculations; for example the decrease in the interaction of a molecule with water produced by an intramolecular hydrogen-bond in the case of o-nitroaniline. The suggested method seems to be promising, but as it is difficult to obtain the necessary parameters for pharmaceutical molecules, the estimation of the stationary phase parameters seems to be a bit uncertain. Figure 2.13 shows the determination of the stationary phase parameter on the basis of four model compounds. It is not completely clear from the published paper whether the two arbitrary z values mean two different mobile phase compositions for the estimation of dielectric permittivity, and if three values are considered whether they are on a single straight line or not.

Retention prediction ofpharmaceutical compounds

85

f(€,)

0.5

0.4

0.3

Ic

20

40

60

T3

Fig. 2.13. Determination of the stationary phase parameter of Merck RP-18 stationary phase (Reprinted with permission from ref. 89.)

Also the predictive power of the theory was not tested on pharmaceutical compounds, and the effect of the neighbouring substituents can cause significant deviations in the predicted retention. The advantage of the method would be that it does not require preliminary experiments, only the determination of the stationary phase characteristics. More reliable and general prediction methods are based on experimental retention values. In this case, the retention data of a given compound series are determined on a given stationary phase, with given mobile phase mixtures and additives. The retention prediction is based on the retention data obtained under slightly different chromatographic conditions, so the retention change in the compounds can be monitored, a function can be set up, and the retention prediction can be carried out using these functions. One of the most well-known methods is used by the DryLab program, developed by Snyder [97], for mobile phase optimization. The retention prediction is based on the experimentally measured retention parameters of the compounds by using two different gradient runs on a given column. The difference in retention parameters obtained by a slower and faster gradient run will reveal the slope (3values of the compounds, namely the sensitivity of the retention to the mobile phase composition. The absolute values of the retention give References pp. 90-92

86

Chapter 2

information about its log k,values. On the basis of these two values, the retention can be predicted within reasonable (linear) range of organic phase concentration applied in isocratic mode. Another interesting method which was developed also for optimizing a chromatographic separation is the “Prisma” method described by Nyiredi et al. [98]. In this model the solvent composition is characterized by the solvent strength (S,) and the selectivity points (P,).At constant solvent strength, the correlation between the selectivity points and the retention was described by a quadratic function. For constant selectivity points, the solvent strength and retention data correlate with logarithmic function. These correlations are used to formulate a mathematical model for the dependence of retention times (retention fdctors) on the mobile phase composition. Unknown compounds are estimated in the mathematical model from a sequence of standard chromatograms after having identified individual peaks by an automatic procedure. Only retention times, relative peak areas, and information about the mobile phase compositions are required as input for the peak identification. The peak identification procedure involves a combination of statistical methods which exploit both the basic properties of retention data and the mathematical relation between retention data, selectivity points and solvent strength as derived from the “Prisma” model (Fig. 2.14). The mathematical model completed by the estimated constants predicts the expected retention times for each possible mobile phase composition. Peak start and peak end times are predicted in a similar way to the retention times, once the identification is performed. The most important aspect of a chromatogram can thus be predicted for arbitrary mobile phase compositions. The information required for peak identification and mobile phase optimization is derived from a series of chromatograms generated from the same sample by varying the mobile phase composition according to a standard scheme derived from the “Prisma” model [99]. On the basis of the results obtained from the preliminary experimental runs, a set of equations for the retention surfaces, which relate to the reten-

Fig. 2.14. The regular part of the “Prisma” model for RP-HPLC. ST^ stands for solvent selectivity points and Ps stands for the selectivity points. At a constant ST the correlation between the Ps and the retention data (horizontal function) can be described by a quadratic function. For constant Ps the solvent strengths and retention data correlate (vertical function) with a logarithmic function. (Reprinted with permission from ref. 98.)

Retention prediction of pharmaceutical compounds

87

tion time for each peak with the mobile phase composition, is generated. Apart from the above-mentioned mathematical functions, no other assumption is used. The method requires 7-14 measurements depending on the mobile phase additive. For a ternary system, more preliminary measurements are needed than for the binary system. The suggested method is extremely advantageous for the development of the optimal separation of the components of biological extracts (for example plants) and identification of the peaks. Retention prediction is made only over the experimentally checked range of chromatographic conditions, so it is reliable. In conclusion, the retention prediction methods based on multivariate statistical analysis should be based on strong theoretical backgrounds and sufficient amounts of experimental data. It is always advisable to check whether the mathematical statistical parameters of the equations used for the retention predictions are significant or not and at what level. The theoretical basis of the equation used for the predictions reveals the predictive power of the relationships, whether it is valid for structurally related or unrelated compounds. It also should be checked how the chromatographic conditions (mobile phase and stationary phase) influence the prediction. From the approaches presented above, it is clear that a general quantitative structure-retention relationship for describing chromatographic retention for a wide range of compounds in a wide variety of chromatographic conditions is still not known. Much higher accuracy in prediction can be achieved when only structurally related compounds are considered under certain chromatographic conditions.

2.9 APPLICATIONS OF RETENTION PREDICTIONS OF PHARMACEUTICAL COMPOUNDS

High performance liquid chromatography is gaining wider and wider application in pharmaceutical analysis [loo]. The application of chromatography starts with the design of new drug molecules [ 1011, involving metabolite research, pharmacokinetic investigations, toxicity measurements, drug delivery and formulation research, and quality control in every step of the manufacturing process. Therefore, the prediction of the chromatographic retention of pharmaceutical compounds is of great importance. A few approaches presented in this chapter were used to develop computer expert systems for promoting the application of HPLC in the pharmaceutical industry. The ELUEX expert system developed by CompuDrug Chemistry Ltd. (Budapest, Hungary) suggests an initial mobile phase composition for the very first chromatography of a drug molecule [102,103]. The prediction is based on Eqs. (2.12) and (2.13). On the basis of the calculated log P values from the fragments of the chemical structure of the molecule, the organic phase concentration in the mobile phase can be estimated at which the retention time of the compound will double the dead time. As the calculation of the hydrophobicity (octanol-water partition coefficients, log P) of the molecules is more or less solved, its capability and error are well known. Also Eqs. (2.12) and (2.13) were derived from the retention data of structurally unrelated pharmaceutical compounds obtained on various reversed-phase stationary phases and mobile phase compositions. It seems from the theoretical background [28], that the relationship is valid in general, as the References pp. 90-92

88

Chapter 2

chromatographic partition system is optimized to be the most similar to the octanol-water partition system, when the Collander type of relationship is valid [64]: log K = a log P + b

(2.40)

where P is the octanol-water partition coefficient of a compound, K is the partition coefficient of a compound obtained in a different partition system, a and b are constants. Leo [57] revealed the limitation of the above equation, namely it is valid only for structurally related compounds or similar partition systems. When the chromatographic partition system is set up to be similar to the octanol-water partition system, Eq. (2.40) is also assumed to be valid for structurally unrelated compounds. From the above considerations, it is clear that accuracy of the prediction of the mobile phase composition for the given retention of the pharmaceutical compounds will be higher when more related compounds are considered, or the chromatographic system is more similar to the octanol-water partition system. In the most general case, the error of the prediction was less than 10% organic phase concentration, which is still acceptable for a method which does not require any preliminary experimental data at all. The accuracy of the prediction can be enhanced by application of experimental data. On the basis of three experimental results, the expert system can carry out optimization for the required separation of a mixture of pharmaceutical compounds. The flow chart of the expert system is presented in Fig. 2.15. CHEMICAL STRUCTURE

J

J

log P cakuhtion

Sekchbn of mobile phase pH and additives

J

Basicgroups

+

Acidicgroups

-+

J.

pH=7.8

&,cakuhtion

pH=2

5.

+ TBAH

Bothackiicandksic-+ p H = 2 +ion pair Noneofthem

+

estimation of organic phase volunu

no buffer necessary

J.

First trial

J goodpcak shape Bask

+

4 bad peak shape

pH=7.8+TBAH

+pH=2+ionpOir

pH=2.0

+ p H = 8 + TBAH

Both -, pH = 2.0 + ion pair

+ by wn-crchange + tuUingpH = 7.8

Acidic-,

None

-+

no buffer

leading pH = 2.0

Fig. 2.15. The rule system of the ELUEX expert system for predicting the initial mobile phase composition for the separation of drug mixtures with known chemical structures. (Reprinted with permission from ref. 103.)

Retentionprediction ofpharmaceutical compounds

89

The second most important application of retention prediction is its use for mobile phase optimization. The most reliable optimization methods are based on experimental data. As discussed above, the basic relationship for mobile phase optimization is the dependence of the retention factor on the mobile phase composition. Within a limited range, the linear relationship between the logarithmic retention factor (log k) and the organic phase concentration (OP% or 9 ) is valid. The slope (S)and the intercept (log b)values of the straight lines can be used for calculating the log k values at arbitrary mobile phase compositions. This means that any prediction method for the S and log ko values are of great importance in retention prediction and mobile phase optimization. It was shown that molecular modelling and quantum-chemical parameters can be related to S and log ko. The accuracy of the prediction is, however, very low. Much higher accuracy can be achieved by applying experimental data, for example, as suggested by DryLab software. The retention measurements applying a slower and a faster gradient run (linear increase in the organic phase concentration of the mobile phase). This can reveal more reliable S and log ko values at which the mobile phase optimization can be carried out on the same column and same type of organic modifier. Another purpose of the retention prediction can be the identification of a compound through its chromatographic retention This requires much higher accuracy in retention prediction. An interesting application of retention prediction was developed for promoting the identification of metabolites (HPLC-MetabolExpert, CompuDrug Chemistry, Budapest, Hungary). The retention prediction is based on a database containing more than 400 hundred possible metabolic transformations. Based on experimental data, retention increment values are compiled for each metabolic transformation. From these data, the change in the retention of the metabolite compared to the mother compound is predicted. The accuracy of the retention prediction is increased by the application of measured retention data of a mother compound and the application of only one or two retention increment values. The method is not applicable for predicting the retention of compounds from the increment values only. The accuracy of the prediction is limited by neglecting the effect of the mobile phase composition and stationary phase differences from different manufacturers. But it was shown [69] that neglecting the neighbour effect can cause much bigger errors in the retention prediction. In spite of these limitations, the expert system with its theoretical background can be used for predicting the reversed-phase HPLC retention of metabolites under given chromatographic conditions. It can help to develop analytical methods possible with hyphenated techniques (HPLC-MS) for the identification of metabolites of newly synthesized compounds. The fourth application of retention prediction methods can be the estimation and measurement of chromatographic hydrophobicity indices of pharmaceutical molecules. It is well known that the hydrophobic properties of drug molecules are very important in their absorption, delivery, distributions and drug-receptor binding, etc. The measurements of partition coefficients are very time consuming and require special analytical techniques for the concentration determination of molecules in both partitioning liquids. Often chromatographic methods are applied for the concentration determination. The new chromatographic hydrophobicity index [28] suggested recently, provides a relatively easy measurement of hydrophobicity of drug molecules, independent from the mobile phase composition and from the origin of the reversed-phase stationary phase. The index ranges References pp. 90-92

Chapter 2

90

from 1 to 100, and it means the volume percent of the organic phase, at which the log k values of the compounds are zero (i.e. the retention time is double the dead time). The hydrophobicity index values are dependent only on the type of the organic phase, the pH and the temperature. They also show significant correlation to the octanol-water partition coefficients. They can be used for retention prediction and mobile phase optimization. Finally, the development and application of retention prediction methods can reveal the mechanism of chromatographic retention, which is still not clearly understood. The search for important molecular properties in the retention is still going on. Understanding the chromatographicretention of compounds helps not only retention prediction, but also the development of the chromatographic conditions for pharmaceutical analysis and the design of new stationary phases. A knowledge of the retention mechanism also helps the optimization procedure and hopefully at the end of the day, the chromatographic method developed will not be a trial and error method, but can be computerized, automated completely, even to include the method development.

2.10 ACKNOWLEDGEMENTS

The author gratefully acknowledges support from the Maplethorpe Fellowship at the Department of Pharmaceutical Chemistry, School of Pharmacy, University of London. I thank Dr. Robert Watt for careful reading of the manuscript and valuable comments. The encouragement during the work by Professor William Gibbons is also gratefully appreciated.

2.11 REFERENCES 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

L.S. Ettre, LC-GC Int., 6 (1993) 544. G.E. Berendsen, P.J. Schoenmakers, L. de Galan, G. Vigh, Z. Varga-Puchony and J. Inczedy, J. Liq. Chromatogr., 3 (1980) 1669. M.J.M. Wells and C.R. Clark, Anal. Chem., 53 (1981) 1341. J.H. Knox and R. Kaliszan, J. Chromatogr., 349 (1985) 21 1. J.K. Baker, Anal. Chem., 51 (1979) 1693. J.K. Baker, C.-Y. Ma, J. Chromatogr., 169 (1979) 107. J.K. Baker, G.J. Hite, M. Reamer and P. Salva, Anal. Chem., 56 (1984) 2932. R.M. Smith, J. Chromatogr., 236 (1982) 313. R.M. Smith, Anal. Chem., 56 (1984) 256. N. El Tayar, H. van de Waterbeemd and B. Testa, J. Chromatogr., 320 (1985) 305. M.J.M. Wells, C.R. Clark and R.M. Patterson, J. Chromatogr., 235 (1982) 43. P.J. Schoenmakers, H.A.H. Billiet and L. de Galan, J. Chromatogr., 185 (1979) 179. M.J.M. Wells and C.R. Clark, J. Chromatogr., 235 (1982) 31. Cs. Horvhth, W. Melander and I. MolnL, J. Chromatogr., 125 (1976) 129. M.J.M. Wells and C.R. Clark, J. Chromatogr., 235 (1982) 31. M.J.M. Wells and C.R. Clark, J. Chromatogr., 235 (1982) 43. W.R. Melander, D.E. Campbell and Cs. Horvath, J. Chromatogr., 158 (1978) 215. H. Colin, J.C. Diez-Masa, G. Guiochon, T. Czajkowska and I. Miedziak, J. Chromatogr., 167 (1978) 41. J.H. Knox and G. VasvLi, J. Chromatogr., 83 (1973) 181.

Retention prediction of pharmaceutical compounds 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

91

U.Burkert and N.L. Allinger, Molecular Mechanics, American Chemical Society, Washington, DC, 1982. Cs. HorvAth, W. Melander and I. Molnf, Anal. Chem., 49 (1977) 142. N. Funasaki, S. Hada and S. Neya, J. Phys. Chem., 89 (1985) 3046. N. Funasaki, S. Hada and S. Neya, J. Chromatogr.,361 (1986) 33. G. Eng, R.B. Johannesen, E. J. Tierney, J.M. Bellama and F.E. Brinckman, J. Chromatogr., 403 (1987) 1. H.J. Mockel, G. Welter and H. Melzer, J. Chromatogr. 388 (1987) 255. K. Valkb and P. Slegel, J. Liq. Chromatogr., 14 (1991) 3167. K. Valk6, J. Liq. Chromatogr., 4 (1984) 1405. K. Valk6, in Chromatography, the State of the Art,Akademia, Budapest, 1985, p. 739. K. Valk6 and P. Slegel, J. Chromatogr., 631 (1993) 49. K. Jinno and K. Kawasaki, J. Chromatogr., 316 (1984) 1. R. Kaliszan, K. Osmialowski, S.A. Tomellini, S.-H. Hsu, S.D. Fazio and R.A. Hartwick, Chromatographia, 20 (1985) 705. R. Kaliszan, K. Osmialowski, S.A. Tomellini, S.-H. Hsu, S.D. Fazio and R.A. Hartwick, J. Chromatogr., 352 (1986) 141. T. Hanai, J. Chromatogr., 332 (1985) 189. K. Valk6, J. Liq. Chromatogr., 10 (1987) 1663. M. Randic, J. Am. Chem. SOC.,97 (1975) 6609. B.L. Karger, J.R. Grant, A. Harthopf and P.H. Weiner, J. Chromatogr., 128 (1976) 65. K. Jinno and K. Kawasaki, Chromatographia, 18 (1984) 90. R. Matsuda, Y. Hayashi, T. Suzuki, Y. Saito and K. Jinno, Anal. Lett., 24 (1991) 2083. Y. Hayashi, R. Matsuda and K. Jinno, Chromatographia,31 (1991) 554. K. Jinno, Chromatographia,20 (1985) 743. M. Righezza and J.R. Chretien, J. Chromatogr., 556 (1991) 169. H. Colin and G.J. Guiochon, J. Chromatogr., 141 (1977) 289. J.H. Knox and A. Pryde, J. Chromatogr., 112 (1975) 171. Y.-D. Men and D.B. Marshall, Anal. Chem., 62 (1990) 2606. S.A. Wise, W.J. Bonnett, F.R. Guenther and W.E. May, J. Chromatogr. Sci., 19 (1981) 457. D.E. Martire and R.E. Boem, J. Phys. Chem., 87 (1983) 1045. K.A. Dill, J. Phys. Chem., 91 (1987) 1980. C. Hansch and A. Leo, Substituent Constants for Correlation Analysis in Chemistry and Biology, Wiley, New York, 1979. R.F. Rekker, The Hydrophobic Fragmental Constant. Its Derivation and Application. A Means of Characterizing Membrane Systems, Elsevier, Amsterdam, 1977. C. Hansch, clogP, Pomona College, Claremont, CA, C. Hansch and T. Fujita, J. Am. Chem. Soc., 86 (1964) 1616. ProLogP, An Expert System for the Calculation of LogP, Version 4.1, CompuDrug Chemistry, Ltd., Budapest, Hungary, (1991). L.R. Snyder, J.W. Dolan and J.R. Gant, J. Chromatogr., 165 (1979) 3. W. Butte, C. Fooken, R. Klussmann and D. Schuller, J. Chromatogr.,209 (1981) 7. W.E. Hammers, G.J. Meurs and C.L. de Ligny, J. Chromatogr.,247 (1982) 1. M.J.M. Wells, C.R. Clark and R.M. Patterson, J. Chromatogr., 235 (1982) 43. A. J. Leo, in Biological Correlations- The Hansch Approach, Advances in Chemistry Series, No. 114, R.F. Gould (Ed.), American Chemical Society, Washington, DC, 1972, p. 51. K. Valkb, J. Liq. Chromatogr., 7 (1984) 1405. K. Valk6, in Chromatography’84, H. Kalilsz and L.S. Ettre (Eds.), Akademia, Budapest, 1986, p. 73. K. Valk6 and P. Slegel, J. Chromatogr., 631 (1993) 49. R.W. Roos and C.A. Lau-Cam, J. Chromatogr., 370 (1986) 403. K. Valk6, T. Friedmann, J. Bhti and A. Nagykddi, J. Liq. Chromatogr., 7 (1984) 2073. A.R. Zoest, C.T. Hung, F.C. Lam, R.B. Taylor and S. Wanwimolruk, J. Liq. Chromatogr., 15 (1992) 395. H. B. Patel, D. N. King and T. M. Jefferies, J. Chromatogr., 555 (1991) 21.

92 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103

Chapter 2 R. Collander, ActaChem. Scand., 5 (1951) 774. A.J.P. Martin, Biochem. Soc. Symp., 3 (1950) 4. E. Tomlinson, H. Poppe and J.C. Kraak, Int. J. Pharm., 7 (1981) 225. T.L. Hafkenscheid and E. Tomlinson, Int. J. Pharm., 17 (1983) 1. C. Hansch and T. Fujita, J. Am. Chem. SOC.,86 (1964) 1616. K. Valk6, G. Szab6, J. Rohricht, K. Jemnitz, and F. Darvas, J. Chromatogr.,485 (1989) 349. HPLC-MetabolExpert, CompuDrug Chemistry, Ltd. Budapest, 1989. R. Hindriks, F. Maris, J. Vink, A Peeters, M. de Smet, D.L. Massart and L. Buydens, J. Chromatogr., 485 (1989) 255. N. Dimov and M. Moskovkina, J. Chromatogr., 552 (1991) 59-69. N. Dimov, Anal. Chim. Acta, 201 (1987) 217. N.R. Draper and H. Smith, Applied Regression Analysis, Wiley, New York, 1966. K.B. Woodburn, J.J. Delfino and P.S.C. Rao, Chemosphere, 24 (1992) 1037. P. Jandera, J. Chromatogr., 314 (1984) 13. P. Jandera, J. Chromatogr., 352 (1986) 91. P. Jandera, J. Chromatogr., 352 (1986) 111. L.R. Snyder and J.J. Kirkland, Introduction to Modem Liquid Chromatography, Wiley, New York, 1974, p. 453. R. Kaliszan, K. Osmialowski, S.A. Tomellini, S.-H. Hsu, S.D. Fazio and R.A. Hartwick, Chromatographia, 20 (1985) 705. R. Kaliszan, K. Osmialowski, S.A. Tomellini, S.-H. Hsu, S.D. Fazio and R.A. Hartwick, J. Chromatogr., 352 (1986) 141. H. Zou, Y. Zhang and P. Lu, J. Chromatogr. 522 (1990) 49. J.H. Park, P.W. Cam, M.H. Abraham, R.W. T& R.M. Doherty and M.J. Kamlet, Chromatographia, 25 (1988) 373. S. Bondi, J. Phys. Chem., 68 (1964) 441. M.H. Abraham and J.C. McGowan, Chromatographia,23 (1987) 243. M.J. Kamlet, J.L.M. Abboud andR.W. Tafi, Prog. Phys. Org. Chem., 13 (1981)485. M.H. Abraham, R.W. Doherty, M.J. Kamlet and R.W. TaR, Chem. Ber., 75 (1986) 629. B.L. Karger, L.R. Snyder and C. Eon, J. Chromatogr., 125 (1976) 71. S.V. Galushko, J. Chromatogr., 552 (1991) 91. Cs. Horvhth and W. Melander, J. Chromatogr., Sci., 15 (1977) 393. 0. Sinanoglu, in Molecular Interactions, Vol. 3, Wiley, New York, 1982, p. 284. T. Haligoglu and 0. Sinanoglu,Ann. N. Y. Acad. Sci., 158 (1969) 310. V.P. Belousov, TermodinamikaVodnyh Rastvorov Neelektrolitov, Khimiya, Leningrad, 1983, p. 264. S. Teresawa, H. Itsuki and S. Arakawa, J. Phys. Chem., 79 (1975) 2345. V.I. Minkin, O.D. Osipov and Yu.A. Zhdanov, Dipolnye Momenty v Organicheskoi Khimii, Khimiiya, Leningrad, 1963, p. 248. 0. Exner, Dipole Moments in Organic Chemistry, Georg Thieme, Stuttgart, 1975, p. 33. L.R. Snyder, DryLab, LC Resources lnc., Walnut Creek, CA, USA. Sz. Nyiredy, W. Wosniok, H. Thiele and 0. Sticher, J. Liq. Chromatogr., 14 (1991) 3077. Sz. Nyiredy, K. Dallenbach-Toelke and 0. Sticher, J. Liq. Chromatogr., 12 (1989) 95. G. Szepesi (Ed.), HPLC in PharmaceuticalAnalysis, CRC Press, Boca Raton, 1991. K. Valk6, Trends Anal. Chem., 6 (8) (1987) 214. G. Szepesi and K. Valk6, J. Chromatogr., 550 (1001) 87. P.P. Csokhn, F. Darvas, F. Csizmadia and K. Valk6, LC-GC Int., 6 (1993) 361.

Journal of Chromatography Library, Vol. 57: Retention and Selectivity in Liquid Chromatography

R.M.Smith, editor 0 1995 Elsevier Science 8.V. A# rights reserved

93

CHAPTER 3

Retention index scales used in highperformance liquid chromatography Roger M. Smith Department of Chernisv, Loughborough University of Technology, Loughborough, Leicestershire, LEI1 3TU UK

3.1 INTRODUCTION

In all chromatographic separation methods, the movement of an analyte through the system is defined by its distribution coefficient ( K ) between the mobile and stationary phases. This fundamental property is determined by its structure, physical properties, the nature of the two phases and the temperature. It determines the distance moved by the analyte in planar chromatography or the time taken for an analyte to be eluted from a column.Because the distribution coefficient is independent of the amount of analyte or the presence of other components in a mixture, the rate of elution is characteristic of the analyte and can be used as a means of identification. However, because of experimental variations, it can fiequently be difficult to accurately reproduce this retention value on different days, on different experimental systems, or in different laboratories. A comparison of retentions must therefore be made directly with an authentic sample on the same chromatographic system, which can be time consuming and would require each laboratory to have a complete set of reference compounds. As a consequence, considerable effort has gone into the development of more reproducible chromatographic systems and into methods of recording retention times, volume or distances that are more transferable between systems. Two principal approaches have been used, both based on the concept that many of the experimental variables will exert a proportional effect on both the analytes and standards. Firstly, relative retention measurements can be calculated by comparison to individual internal standards or to a series of standards, such as a homologous retention index scale. The second approach is to measure the differences between the systems, usually by employing a set of test compounds and then correcting or adjusting the final result. This method is less widely used but examples are given in Chapter 5.

References pp. I 4 0 4 44

94

Chapter 3

Reliable relative retention measurements can also be used to compare the properties of analytes and thus to determine the effect of changing analyte structure on physical properties such as partition coefficient, hydrophobicity, and biological activity or on their environmental distribution and degradation [ 1-31. These quantitative structure retention relationships (QSRR) can also be used to predict retentions as in Chapters 1 and 2 or to compare the influence of structure of biological activity (Section 3.4.4).

3.1.1 Relative retention times

Two primary factors can alter the retention times on nominally identical chromatographic separations at different times or between different instruments. There can be differences in the flow rate of the mobile phase through the chromatographic system. In gas chromatography, flow rates are particularly difficult to standardize, particularly on different instruments or columns, and in liquid chromatography the mobile phase delivery from different pumps may often deviate from their nominal values. The second factor is differences in the proportions of the stationary or mobile phases as a consequence of differences in the diameter of the column or in the thickness of the stationary phase layer. These changes will cause changes in the phase ratio between the stationary and mobile phases and in the distribution of the analyte between the two phases. One approach to eliminate these variations, which has been widely adopted in high performance liquid chromatography (HPLC), is to compare the retention time (tR) with that of an unretained standard, whose retention would be also affected by the column proportions and flow rate. This concept defines the retention of analytes as their retention factors (or capacity factors k) (Eq. 3.1), which is defined as the ratio of the adjusted retention time (tR - tM) and the retention time of an unretained marker compound ( t M ) [4].

k = -t~ - t ~

(3.1)

tM

The hold-up time (tM)is sometimes also called the column void volume (or less desirably the dead volume). However, the use of retention factors to obtain reproducible retention values faces two problems. Firstly, because the value of the hold-up time (tM)is small compared to the retention time of the analyte, even very small differences in its measured value will cause significant variations in the calculated retention factors. Secondly, even though many studies have examined and compared techniques for the determination of the hold-up volume [5-91, there is no agreement on a satisfactory and universally acceptable method. Different methods can effectively determine different parameters. For example on a Zorbax Cs column with methanol-water (5050) as the eluent, different marker samples gave a range of values for the “hold-up volume”: “silica smoke”, 1.41 ml; sodium nitrate solution, 1.94 ml; sodium sulphite solution, 1.90 ml; sodium thiosulphate solution, 1.83 ml; potassium dichromate solution, 1.62 ml; perdeuteromethanol, 2.79 ml; and deuterium oxide, 2.70 ml compared to a calculated value of 2.75 ml [8]. The difficulty is that most liquid chromatographic systems combine a complex mixture of different separation modes. Although a very pcilar analyte, such as nitrate ions, should

Retention index scales used in high-performance liquid chromatography

95

not be retained, they suffer ionic repulsion from the charged silanols underlying the bonded-phase surface and cannot enter the full mobile phase volume [6,8]. As a consequence, the apparent hold-up volume can differ if the concentration of the marker solution or buffer strength of the eluent is changed. At the other extreme, larger analytes can suffer size exclusion effects in narrow pores, so would see a reduced mobile phase volume when compared to a small reference material such as uracil [6]. Other proposed test compounds, such as deuterated isotopes of the mobile phase components cannot be easily detected on a routine scale. Because of these effects, it has been suggested that the value of the true hold-up volume for each analyte may be different depending on its interaction with the surface and its ability to access small pores [ 6 ] . As will be seen later (Section 3.4.3.1),it has been suggested that the curve fitting of a homologous series of analytes (such as those used as retention index standards) can be used to calculate the hold-up volume but even this method can give variable results. For most comparison purposes, as long as the method used to determine the hold-up value is defined carefully, the measured value will be consistent and can be used for the calculation of retention factors on a single system. However, it is still notoriously unreliable when different systems are compared. Although differences in the flow rate or phase ratio will alter the absolute retention time of an analyte, every component of a mixture will suffer the same effect and be carried through the column at either a slower or faster rate. Of the remaining variables in a chromatographic separation, temperature is usually easily controlled and the value can be readily transferred between laboratories. The effects of small changes in the mobile phase composition are a more serious problem and need careful control. However, small differences in temperature or in mobile phase composition have a similar influence on all the components of a mixture. It should therefore be possible to compensate for many of these variations by recording retention as relative values compared to the retention of a standard compound in the same system rather than as absolute values. However, these methods cannot compensate for the differences in apparently equivalent stationary phases from different sources, whose interactions properties may differ, as these effects will often alter relative retentions. Instead, the differences in the retentions of test compounds can be used to monitor the properties of stationary phases (see Section 3.4.3.2 and Chapters 10, 11 and 12).

3.1.2 Internal and external standards

The most widely adopted approach to improve retention reproducibility is to use an internal standard. The retention is then reported as the ratio of the retention times of the analyte and of a standard compound, which has been added to the sample. Usually a standard with a similar retention time to the analyte is chosen. Either absolute ItR)or adjusted retention times can be used. The latter are usually preferable as they are directly related to the distribution coefficients ( K ) of the analyte and standard but require the measurement of the column hold-up volume with its the attendant uncertainty. In recent years, the ratio of unadjusted times have frequently been quoted as these can be measured directly by an integrator and calculated as part of the post-run calculations. Ettre has demonstrated that References pp. 140-1 44

96

Chapter 3

the two ratios can be related [101 and, as would be expected, found that the differences between the values were larger at short retention times. Some laboratories use external standards as an alternative. This method has the advantage of removing one step from the sample preparation sequence but will be slightly less reliable as the standard may be separated under marginally different analytical conditions to the sample. However, if the standard solution is injected before and after each set of test samples the relative retentions should be almost as accurate as from internal standardization. This method can also be used for quantitative studies in liquid chromatography as the use of rotary loop-injection valves usually ensures high reproducibility. However, in gas chromatography the reproducibility of the injected sample volume can be poor, particularly with manual injection, and internal standardization is usually preferred. One advantage of external standardization is that with samples of unknown composition, there is no possibility of a sample component being masked by the peak for the standard. The selection of the compound to use as a standard is often determined by the availability of compounds related to the analyte of interest. A compound with a similar structure and properties to the analyte, such as structural isomers or homologues, is preferred so that it will behave in a similar manner in any sample preparation, extraction or derivatization steps. However, the standard must be a compound that is unlikely to be present as a component of a sample. If suitable closely related standards are not available, then the analyst must select a compound based in its chromatographic properties, so that the standard will have a similar retention to the analyte. Ideally it will contain similar functional groups and have a similar pK,, so that it will be similarly influenced by pH and ionic strength effects. Rather than select a compound at random, a number of sets of potential internal standards have been proposed, each member of which covers a range of retentions. In early work, the alkyl aryl ketones were proposed by Kikta and Strang [ 113. They are neutral analytes with a good chromophore and individual ketones have been used in a number of studies. For example, butyrophenone was used as an internal standard in a collaborative study of the determination of propoxur [ 121 and propiophenone was used for the determination of capsaicin, in oleoresins and personal protection aerosols [ 131. The anilides of the fatty acids, which have good stability in dilute solution and high absorbance for spectroscopic detection, were suggested by Verzele et al. [14]. More recently Yamauchi and Mori [15] suggested a set of 30 phenols, which could be used as internal standards, in particular the 4-hydroxybenzoate esters. However, Dolan [ 161 has discussed problems that have been reported with ethyl 4-hydroxybenzoate as an internal standard. In one case, the pH 5.2 buffer in the eluent was inappropriate and in another case sample losses inexplicably occurred on changing the injection vial septum. Skelly and co-workers [ 171 proposed the use of a computer assisted search to suggest an internal standard. The retention times of the analyte and a set of marker compounds were measured and compared with a database of 90 potential internal standards ranging from acetic acid amide (N-4-hydroxylphenyl)to myristylphenone (tridecyl phenyl ketone) and including many of the phenols and the alkyl aryl ketones proposed earlier. The advantage of this system was that in order to carry out a wide range of assays only a limited number of standards would need to be maintained in the laboratory. As they were readily available compounds, they could easily be employed in interlaboratory comparisons.

Retention index scales used in high-performance liquid chromatography

97

3.1.3 Retention indices

Rather than use the relative retentions compared to a single internal standard, which is to some extent dependent on the value determined for the column hold-up volume, an alternative is to compare the analyte to a retention index scale of homologous standards. The retention index of the analyte is then determined by interpolation of its retention between the retentions of the standard compounds. In his early work, Martin [ 181 proposed that the partition of an analyte between two phases was an additive effect and was directly related to its structure. He suggested that the retention could be expressed as the summation of factors (ARM,also defined as group contributions z; see Chapter 1) for individual components ( i ) of the structure. Thus, the retention of analyte (k) can be determined as the retention of a parent compound (kp)plus contributions for the individual substituents (Eq. 3.2). logk=logkp +CAR,(i) Thus, there should be a systematic increase in retention along a homologous series as each addition of a methylene group should systematically increase log k by (the methylene selectivity). A homologous series of standard compounds should thus provide a regularly increasing scale of reference peaks across a chromatographic separation. The first retention index scale was introduced in 1958 by Kovhts for use in gas-liquid chromatography (GLC) [ 191. Since then, alternative retention index scales have been proposed for GLC, HPLC, supercritical fluid chromatography (SFC) and recently for micellar capillary electrophoresis (CE) (see later). In most cases, the retention scale is defined as the carbon number of the standard X 100 (so that hexadecane, I = 1600) and the retention index values of analytes are determined by interpolation between the standards. There are a number of advantages of retention index values compared to other methods of recording retentions. As an interpolated value, they should be more robust than ratioed or direct measurements to small changes in operating conditions and to differences in chromatographic systems between laboratories. In an ideal situation, a laboratory would need only one set of internal standards, which would be applicable to all samples irrespective of the nature of the analyte. If the standard scale was widely accepted, it would enable retention values to be readily transferred between laboratories. However, there are drawbacks in a universal scale. The standards may have different structural features to the analytes and may respond differently to changes in pH or ionic strength or specific interactions, such as silanol or metal ion activity on the column. By design, they contain the same functional group and are designed to separate according to the methylene selectivity. However, they may not separate in the same way as analytes containing different functional groups in mixed mode separations,when there is a mixture of adsorption, ion-exchange or size exclusion separation, together with the partition mechanism. In addition, they usually cannot be used to test the extraction and derivatization stages of sample preparation. In liquid chromatography, because analytes with such a wide range of polarities can be separated, any single homologous series of standards may not cover the full range of retentions of the analytes of interest. The choice of possible

mH2

References pp. 140-1 44

98

Chapter 3

compounds for use as standards may be limited because most HPLC instruments are only fitted with a ultraviolet spectroscopic detector. Although most chromatographers will be familiar with Kovhts retention indices, the term retention indices has often been inappropriately applied or misused. The IUPAC Recommendations for the Nomenclature of Chromatography [4] define retention indices as being based on the n-alkanes as standards as described by Kovhts. If other series of standards are used, they must be clearly differentiated. Frequently, “retention indices” has also been misused simply to denote retention values, as a synonym for RM or log k [20], although this may be partly a problem of translation or transliteration. The term has also been used to denote interaction indices, better described as a measure of the effect of functional groups on retention [21]. Even if a different homologous series to the n-alkanes are used, to avoid confusion an index scale should be based on the total carbon number so that I = nc X 100. Some scales have started at different points, such as toluene, I = 100 rather then 700 or acetone I = 100 rather than I = 300 [22], or have been based on the number of carbons on an alkyl side chain [23,24], which can lead to considerable confusion. As will be seen in Section 3.3.6, one widely used scale is based on systematic changes in the ring number of polycyclic hydrocarbons (n)and the scale is logarithmic ( I = 10”). Although not a homologous series of standards, it is included in this chapter, because the standards represent a systematic skeletal change and are employed in an equivalent manner to a retention index scale. A misuse of the concept of retention indices has been reported in a comparison of cyano-bonded silica for use in solid phase extraction cartridges. The correlation equation between carbon number and logk on one column was used to calculate apparent “retention index” values for alkyl aryl ketones standards on further columns as a test of retentive power [25]. A more appropriate application would have been to compare the intercepts and slopes of the relationships for the standards on each column. Other authors have used the term retention indices when interpolated or corrected relative retentions would be more appropriate. Huovinen, in a study of lichen constituents, described a scale of relative retentions, that compared the retention of an analyte to benzoic acid relative to the retention difference between 2-ethylhexyl phthalate and benzoic acid [26]. They found that this arbitrary two-point scale was able to distinguish compounds in the orcinol series RZ = 0.08-0.58) from dibenzofurans RZ = 0.60-0.94). In recent work, Wallace has used the expression, retention index, for the relative retention times of benzoic acid and benzene on an ion-exchange column [27]. 3.2 RETENTION INDEX SCALES IN CHROMATOGRAPHY

Although originally introduced for gas-liquid chromatography, the application of retention indices has now extended to liquid chromatography, supercritical fluid chromatography and more recently to micellar capillary electrophoretic separations. They have been widely adopted throughout chromatography as a method for recording retentions, for comparing retentions for identification, and as the basis of retention prediction methods. Their role has recently been reviewed by Paciikovii and Felt1 [28] but they concentrated primarily on indices in gas-liquid chromatography.

Retention index scales used in high-performance liquid chromatography

99

A generalized expression linking retention indices with the role of electric interactions based on molecular descriptors and in both gas and liquid chromatography has been described in a series of papers by Lamparczyk and co-workers [29,30]. 3.2.1 Gas chromatography The most important retention index scale in gas chromatography is still the n-alkanes proposed by Kovhts [19,31]. The n-alkanes are readily available in high purity, are nontoxic, chemically stable and inert and their retentions cover a wide range of analyte retentions. They give good peak shapes on non-polar and medium polarity stationary phases. Because they lack functional groups, they have also formed the basis of numerous physical chemical studies, such as the examination of the effects of substitution, etc. Under isothermal conditions, the logarithm of the adjusted retention time increases linearly with carbon number (nc) (log t'R = nc). The retention indices I of the standard is defined as 100 x nc. The retention index for a sample compound is determined by logarithmic interpolation of its retention time (fR,J, between the retention times of the next shorter (tlR,J and next longer retained n-alkanes (tk, + I), containing z and z + 1 carbons atoms, respectively (Eq. 3.3). The adjusted retention times can be replaced by adjusted retention volumes.

If the separation is carried out using temperature programming, the relationship becomes more complicated but a reproducible retention index value can be obtained by using a linear relationship between adjusted retention times and carbon numbers. The corresponding equation is then used for the calculation of the retention index (Eq. 3.4) [32,33]. I = loo[

t'R,i-f'R,z

]+looz

(3.4)

f'R,z+l-flR,z

Considerable study has gone into the comparison and interconversion of I values obtained under isothermal and gradient elution conditions [28]. The early work based on the Kovhts indices has been reviewed by Kovhts [3 13, Ettre [34], and subsequently by Haken [35]. The extent of the acceptance of retention indices can be seen in a comprehensive review by Budahegyi and co-workers [36], which was prepared to commemorate the 25th anniversary of the original retention index paper. The review included 1392 references on retention indices and a firther 443 references were included in an updated review for the 30th anniversary [37]. A more recent review by Takchs [38] noted that out of the 100 000 published papers on gas chromatography, over 2 100 have been devoted to retention index systems, reflecting their widespread interest. The principal application of the Kovhts indices has been for the identification of unknown components in samples. Numerous general tabulations of data derived from a wide range of stationary phases have been published [28,39,40]. Specialist lists have also been compiled, including lists for drugs, solvent of abuse and toxicology [41-44]. Specific listings for 187 N and P containing drugs and their metabolites have been reported [45]. References pp. 140-1 44

100

Chapter 3

A number of these databases are now being provided as computer readable search systems [e.g. 401. The second application of retention indices has been as the basis of the Rohrschneider [46-49] and McReynolds constants [50] for the comparison of the retention properties of liquid and solid stationary phases. In both methods, the retention indices of a set of test compounds measured on the stationary phase being studied are compared with the corresponding retention indices measured on a non-polar squalane column. The increased retention index values of the different test compounds on the column being tested are used as a guide to the different interaction properties of the phase. However, there have been a number of occasions in which the n-alkanes have been felt to be inappropriate as a reference scale. Alternative standard compounds have been described, which are more polar, such as the alkan-2-ones [51,52] or the n-alkanols [53], and thus will give better peak shapes and longer retentions on polar stationary phases. A number of analyte-specific standards have also been adopted, such as fatty acid esters which are used to calculate equivalent chain numbers (ECN) or equivalent chain lengths (ECL) [28]. There has also been a need for standards, which can be detected by specific detectors, that do not respond to the n-alkanes. The electron capture detector is particularly important in environmental studies and has such a high sensitivity that it cannot easily be used in parallel with a flame ionization detector. As a result a number of alternative index scales have been proposed based on halogenated compounds. These include the n-alkyl trichloroacetates [54] and n-alkyl bromides [55,56]. The latter scale has been used to report the retention indices of 221 halogenated compounds [57]. A “universal” series of standards for selective detectors, the n-alkyl bis(trifluoromethy1)phosphine sulphides, has been suggested by Manninen et ul. [58]. These standards will respond to the electron capture detector, the thermionic ionization detector and the flame photometric detector, in both the sulphur and phosphorus modes, as well as the flame ionization detector (FID). These different scales have been reviewed [28,35,59] but compared to the n-alkanes, almost none have gained widespread acceptance except in specialized areas.

3.2.2 Supercritical fluid chromatography As supercritical fluid chromatography (SFC) [60-62] is becoming established as a complementary technique to GLC and HPLC, it is clear that it has similar problems of transferability and reproducibility of retention data to those found in the older techniques. A number of workers have directly transferred the n-alkane scale from gas-liquid to supercritical fluid chromatographic separations [63,64]. These standards are particularly applicable in capillary SFC, when the flame ionization detector can be used. However, in packed column separations, the inclusion of modifiers such as methanol in the eluent increases the background signal of the FID and spectroscopic detectors are usually employed so that standards with a chromophore would be required. Early studies on non-polar octadecylsilyl bonded silica [65] confirmed that the alkyl aryl ketones and alkylbenzenes gave a linear relationship between retention factor and chain length. More detailed examination confirmed these relationship from valerophe-

Retention index scales used in high-performance liquid chromatography

101

'1

.,

Fig. 3.1. Relationship between log k and carbon number X 100 for alkyl aryl ketones separated by supercritical fluid chromatography on an Ultrasphere ODS column. Eluent, supercritical carbon dioxide with different proportions of methanol as modifier. 0%, 0,4.0%, 0 , 8.3%, 0, 12.7%. Conditions: 60T;2470 psi. Reproduced with permission [66].

none to octadecanophenone (Fig. 3.1), fiom decane (Clo) to tetracosane (CZ4)and from toluene to butylbenzene on ODS silica columns and with a range of proportions of methanol or acetonitrile as modifier [66].The retention indices of a set of test compounds were measured using the alkyl aryl ketone scale. There were major changes on the addition of modifier reflecting the masking of active sites on the stationary phase surface (Fig. 3.2) [66], particularly for analytes such as benzoic acid and benzamide which could not 2500

._ C .-c0

1

.,

1500

C

a

1000-

aT

0 0

8

10

12

14

% Methanol (w/w)

Fig. 3.2. Effect of proportion of methanol as modifier on retention indices of analytes (based on alkyl aryl ketones) in supercritical fluid chromatography,Conditions as Fig. 3.1. Compounds: rn, ethylbenzene; 0,benzylamine; 0 ,benzamide; 0,N-propylaniline; A, nitrobenzene. Reproduced with permission [66].

References pp. 140-144

Chapter 3

102

be eluent without modifier and for benzylamine, which was highly retained without modifier. A similar result was obtained for retention indices measured on a PS-DVB polymer column, which should not show strong silanol interactions. A linear relationship was obtained for the higher alkyl aryl ketone homologues but propiophenone was anomalous. Again the use of modifiers caused major changes in the retention indices of the test compounds. Benzamide, p-cresol, benzoic acid and benzyl alcohol changed with methanol but unlike the separation on the ODS column, the basic analytes, benzylamine, N-propyl aniline, were relatively unchanged and behaved in a similar way to the neutral analytes, toluene and methyl benzoate. In later work, it was found that on a polar cyano bonded silica column the linear relationship for the homologues failed and that the smaller alkyl aryl ketones were increasingly retained by interactions between the stationary phase and the analytes (Fig. 3.3) [67]. Unlike separations in liquid chromatography, the retentions were a combination of effects due to chain length, which mirrored volatility, and a normal-phase type interaction with the column material. Subsequently, similar mixed mode retention has been identified for homologous phenyl alkanols, phenyl alkanoic acids and amides [68,69]. In an extreme example, when fatty acid methyl esters were examined on a polar cyano- silica column, it was found that homologous C14-C18esters co-eluted but there were considerable differences between compounds containing increasing unsaturation [70]. To determine their applicability in SFC, retention indices, based on the alkyl aryl ketone scale, were determined for standard test compounds [71,72], barbiturates [73,74] and benzodiazepines [75]. However, the indices were not robust and were very sensitive to changes in the mobile phase and modifier. In a recent study, Young and Games examined the alkyl aryl ketones as retention index standards for the Fusarium mycotoxins on capillary and packed columns [76]. There were difference between values measured on differ-

r

c

'C 1.00 n m 0

0)

0

0.75

-

6

8

10

12

14

16

18

20

22

24

Carbon Number

Fig. 3.3. Relationship between log k and carbon number for alkyl aryl ketones separated by supercritical fluid chromatography on an Ultrasphere CN column at different eluent densities: A , 0.69; 0,0.65; and 0,0.54 g ml-'. Conditions:eluent, carbon dioxide; and temperature 60°C. Reproduced with permission [67].

Retention index scales used in high-performance liquid chromatography

103

ent stationary phases and the values were sensitive to experimental conditions. Not unexpectedly, there were large differences fiom the values reported using reversed-phase HPLC (see Chapter 4). It is apparent that the retention in packed column SFC is a combination of interactions due to molecular sizeholatility and a polar normal-phase type interaction with the stationary phase [67,70]. This latter retention mechanism is significantly influenced by the functional groups and shows little effect for the methylene increment. This mode of separation is not unexpected as supercritical carbon dioxide has a similar polarity to hexane or dichloromethane [61] and thus hydrophobicity changes would play no part in retention. Retention indices may therefore have a limited applicability in this area as they are susceptible to changes in interaction with the stationary phase and will probably not be readily transferable between systems and instruments. 3.2.3 Liquid chromatography

Although, in normal phase chromatography, a linear relationship between the carbon number and log k has been reported for the alkyl aryl ketones [77,78] and for the semicarbazones of aliphatic ketones [79], the slope is negative and represents a decrease in polar interactions with the silica stationary phase rather than a partitioning effect. Despite early suggestions that retention indices could be used in thin-layer chromatography [80], standardization has concentrated on the correction of RF values by comparison with a set of internal standards [81]. Retention indices have therefore only been used for reversed-phase liquid chromatography. By analogy with gas-chromatography, some studies examined the use of n-alkanes as standards but they are difficult to detect using conventional spectroscopic detectors and their low polarity means that they have much longer retention times than most compounds of interest. Most liquid chromatography studies have therefore used alternative homologous series of standards and these are described in detail later in this chapter. The early stages of these studies have been comprehensively reviewed [82]. Smith [83] suggested a number of criteria for a homologous series of compounds to be widely applicable as a retention index scale in liquid chromatography: (a) they should have a strong chromophore at 254 nm so that they can be added to unknown samples in small quantities to act as internal standards; (b) they should not be readily ionized to avoid changes in retention because of pH variations or the presence of ion-pair reagents; (c) a range of members of the series should be readily available at reasonable cost; (d) the most polar member of the series should be eluted with a similar retention to water soluble pharmaceuticals; (e) the standard compounds must be unreactive and stable in common liquid chromatography solvents. Further requirements were suggested by Pacakovh and Felt1 [28]: - the relationship between log k and the number of carbon atoms or characteristic functional groups in the molecules of the homologues must be linear; - they should not specifically interact with silica gel; - the k values should depend little on the mobile phase composition. References pp. 140-144

104

Chapter 3

The first of these extra suggestions is usually satisfied by homologues, the second excludes many amines but the last proposal cannot be satisfied as all analytes in HPLC are strongly affected by the mobile phase composition. Overall these criteria suggest that suitable standards would be aromatic compounds (preferably conjugated to increase the absorbance of the chromophore) but excluding the relatively non-polar hydrocarbons, ethers, halogens or simple nitrocompounds [83]. Polar compounds, such as amines, phenols or acids, would be excluded because of their sensitivity to pH effects. Groups that could be present include alcohols, benzoate and phenyl alkanoate esters, amides and aryl ketones. However, extended homologous series of only a few of these compounds are commercially available, although they would frequently be easy to synthesize. For example, the chromatography of the homologous and isomeric alkyl benzamides has been studied in detail by Wells and Clark [84] but they did not examine their potential as retention index standards. As will be seen later in this chapter, many of the remaining potential series have been examined as standards. The general change in retention with increasing carbon number has been widely examined and has recently been reviewed [85]. In a typical study, Mackel and Masloch measured the retention factors of homologous alkyl bromides, alkylbenzenes, and alkyldisulphides on nine different stationary phases and found good correlations between log k and carbon number [86]. The methylene increments (Alog k between homologues) showed little change with temperature but were strongly dependent on eluent composition. Figge et al. [87] reported that homologous series of n-alkanes, n-alkylbenzenes, fatty acid methyl esters, alkan3-ones, 2-n-alkylpyridines and 1-n-alkanols all gave very similar methylene increments (Fig. 3.4). In a recent study, Varugese and co-workers [88] have examined series of alkyl ester, ether and amide polyoxyethylene non-ionic surfactants and homologous ethyl alkanoates. They found that the methylene increments were virtually the same in each case (0.285, 0.284, 0.293 and 0.282, respectively in methanol-water (55:45). The linear relationship between logk and carbon number holds on different bonded phases. For the n-alkylbenzenes on SAS, ODS and Cz2-bonded phases, Smith [89] reported correlations of 0.9998, 0.9998, and 0.9986, respectively, and on a phenyl bonded phase found correlations of 0.9997 and 0.9998 in methanol-water (30:70) and acetonitrile (20:80) as eluents [90]. Some differences in the methylene increment between different homologous series have been discussed in a recent review of group contributions on HPLC [85]. Guiochon et al. noted that there was a small change in the methylene increment when the length of the alkyl chain was similar to the depth of the C18stationaryphase [91]. Subsequently the effect of temperature on this discontinuity was examined [92]. These effects are small compared to the methylene increment and have been ignored in most retention index studies. From the additive concept expressed in the Martin equation (Eq. 3.2), it is potentially possible to use a systematic increase in the number of other groups besides methylenes to generate a linear retention index scale. For example, there is a linear relationship between the logarithm of the retention factor and number of tin atoms in the polytin derivatives Bu3Sn(Bu2Sn),SnBu3[93] and with the number of sulphur atoms in cyclic polysulphides (S,) [94]. A scale based on the number of rings in polycyclic aromatic hydrocarbons is described in Section 3.3.6. However, changes such as the polychlorination of benzene or

Retention index scales used in high-performance liquid chromatography

105

Fig. 3.4. Retention behaviour of homologous series of various classes of compounds on PMSClg-coated Nucleosil 5-100-C1. Compounds: 1, n-alkanes; 2, n-alkenes; 3, n-alkylbenzenes; 4,fatty acid methyl esters; 5, 3alkanones; 6, 2-n-alkylpyridines; 7, 1-n-alkanols. Mobile phase, methanol-water (90: 10). Reproduced with permission [87].

naphthalene cannot be used as the contributions of the chloro-groups are not additive but depend on their relative positions and on their electronic interactions [95].

3.2.4 Micellar electrokinetic chromatography Although separation in capillary electrophoresis is driven by the migration of charged species in an electrical field, neutral analytes can be separated by the addition of a surfactant to form charged micelles. The analytes are partitioned between the aqueous solution and the non-polar micellar phase in a similar manner to their partitioning between the polar and non-polar phases in reversed-phase chromatography. The migration of an analyte in micellar electrokinetic chromatography (MEKC) is expressed as its retention factor which is determined by the distribution between the two phases (Eq. 3.5). References pp. 140-1 44

106

Chapter 3

rMC and 7AQ are the molar concentrations in the micellar and aqueous phases, respectively, and ts, tMc and tEoF are, respectively, the migration times of the solute, a micelle marker and a neutral insoluble component, designed to determine the electroosmotic flow. The distribution will be determined by the same parameters as in conventional partition chromatography and two recent papers have suggested that retention indices based on a homologous series would also be applicable to standardize these separations [96,97]. Ahuja and Foley [96] examined the possible use of alkyl aryl ketones, 1-nitroalkanes and n-alkylbenzenes as retention index standards with sodium dodecyl sulphate (SDS), SDS/Brij 35 and SDS/SB-12 surfactants. In each case there was a linear relationship between the hydrophobic retention parameter (log k) and carbon number. They concluded that the 1-nitroalkanes were the most suitable for polar analytes. The alkyl aryl ketones were suitable for compounds more hydrophobic than acetophenone but that up to 15% organic modifier would probably be needed in the solvent. The alkylbenzenes were probably too hydrophobic for normal use. Muijselaar et al. [97] also examined n-alkylbenzenes and alkyl aryl ketones as potential standards. Initially, they also studied alkan-2-ones but their chromophores were too weak for easy detection and they were not studied further. They also found a linear relationship between log k and carbon number using sodium dodecyl sulphate (SDS) cetyltrimethylammonium bromide (CTAB), and decyltrimethylammonium bromide (DTAB) as surfactants. They concluded that the alkylbenzenes were preferred for the anionic SDS surfactant but that the alkyl aryl ketones (Fig. 3.5) were more favourable for the cationic CTAB and DTAB surfactants. Using the alkylbenzene standards, they measured the retention indices of a range of aromatic analytes. These differed according to the surfactant, e.g. benzaldehyde I = 607, 553 and 546 with SDS, CTAB and DTAB surfactants, respectively (Table 3.1) but were I

I

I

I 0

2

4

6

8

time (mi11

Fig. 3.5. Electrokinetic chromatography of the separation of (1) aniline, (2) benzaldehyde, (3) acetophenone, (4) nitrobenzene, (5) phenol, (6) propiophenone, (7) butyrophenone, (8) brornobenzene, (9) valerophenone. (10) hexanophenone applying a background electrolyte containing 50 m M DTAB. Detection wavelength 225 nm. Reproduced with permission [97]. 0 1994 American Chemical Society.

Retention index scales used in high-performance liquid chromatography

107

TABLE 3.1 RETENTION INDEX VALUES OF TEST COMPOUNDS IN MICELLAR ELECTROKINETIC CHROMATOGRAPHY Compound

Resorcinol Aniline Phenol Benzene Benzaldehyde Nitrobenzene Acetophenone Toluene Chlorobenzene Bromobenzene Ethylbenzene Naphthalene

Retention index (R-Ar scale) Surfactants

(RCOAr scale)a

SDS

CTAB

DTAB

SDS

HPLC~

478 507 534 598 606 625 652 704 733 770 796 872

670 540 656 604 552 617 574 698 740 774 787 914

658 53 1 652 598 546 616 574 705 748 793 794 925

59 1 628 662 740 751 774 807 871 906 952 983

658 683 915 777 857 800 1019 1021 1051 1100

~

Data from [97]. Conditions: electrolyte, 0.02 M Tris adjusted to pH 8.5 plus 50 mM surfactant; voltage, 20 kV. aRetention indices based on alkyl aryl ketones calculated from slopes of carbon number versus retention index. bHPLC retention indices in methanol-buffer (50:50) from [98].

effectively independent of the concentration of surfactant and temperature. The values also showed a reasonable correlation with the octanol-water distribution coefficients (log P). From the correlations reported in the paper, the retention indices on the alkyl aryl ketone scale for the SDS surfactant solution have been calculated and compared with the reported liquid chromatographic values in methanol-water (5050) [98] (Table 3.1). Although there were differences they were similar to the differences between stationary phases or between the three surfactant solutions.

3.3 RETENTION INDEX SCALES IN HIGH-PERFORMANCE LIQUID CHROMATOGRAPHY A number of homologous series of compounds have been studied as retention index standards for HPLC, each with advantages and disadvantages, principally related to their detectability or the compatibility of their retentions with the analytes of interest. In many cases, because of the similarity of the methylene increments in different homologous series, equivalent results could be obtained and these are discussed later.

3.3.1 n-Alkanes

It wouId be of considerabls interest to use the n-alkanes to create a retention index scale in HPLC, so that results could be directly compared with gas-chromatographic retentions. Refirences pp. 140-1 44

108

Chapter 3

TABLE 3.2 RETENTION INDICES OF UNSATURATED HYDROCARBONS ON REVERSED-PHASE CHROMATOGRAPHYWITH AND WITHOUT 1 0-2M Ag' [99] Compound

I

%.I

1 -0ctene (Z)-2-Octene (9-2-Octene 1,4-Octadiene(Z),Q

682 673 692 575

1,7-Octadiene

563

628 625 678 489 515 463

Conditions: eluent, methanol-water (5:l) with and without lo-* M AgC104. Retention indices on the n-alkane scale.

However, the n-alkanes have two major problems as standards. Firstly, they are very nonpolar, which means that they are highly retained in most reversed-phase systems. As a consequence, their applicability is limited as few compounds of interest would be eluted using the same mobile phase composition. Secondly, as most work in HPLC is carried out using spectroscopic detection, the lack of chromophore means that routine detection of the n-alkanes is inconvenient and an additional refiactive index detector would be needed. Despite these problems, some early studies used the n-alkane scale without comment. Vonach and Schomburg [99] used the changes in the retention indices to study the effect of argentation chromatography on the retentions of alkanes and alkenes (Table 3.2) and of saturated and unsaturated triglycerides. The presence of a double bond caused a decrease in retention of 10-55 units depending on the extent of substitution. Mbckel and Freyholdt [ 1001 reported a linear correlation (r > 0.999) between In k and carbon number for the nalkanes from pentane to dotriacontane on C8- and C18-reversed-phasecolumns. They proposed that these compounds could be used as the basis of a Kovhts type retention index scale. The retention indices of a wide range of branched alkanes from C5to C8were then correlated with their molecular connectivities and partition coefficients over a range of solvent compositions. In later work, Dimov examined a general model for retention for the isoalkanes based on an n-alkane scale, which could be applicable to GLC on squalane and HPLC on a RP-18 column [loll. Morishita and co-workers used the n-alkanes as standards to determine the retention indices of a number of substituted aromatic derivatives [102]. However, because of the practical difficulties of detecting the alkanes, the measurements were made relative to the n-alkylbenzenes and then corrected to the n-alkane scale (Fig. 3.6). They demonstrated a good correlation between the two scales with the same column and eluent. With methanol-water (80:20) on a Lichrosorb RP-8 column, the indices of the n-alkylbenzenes, based on the n-alkane scale, could be calculated as I = 160 + nc X 100, for n > 5 in which nc=number of carbons in the alkyl side chain [102]. Deviations were observed for shorter alkyl side chains so extrapolation from the higher homologues (nc > 6) was used in the subsequent measurements. The discrepancies between the two methods were small (benzene: by extrapolation, I = 269 and by direct interpolation between n-alkane standards, Z = 265). These standards were then used to determine the retention increments for

Retention index scales used in high-performance liquid chromatography

109

0 0

1

2

3

4

5

6

7

8

9

10

11

Carbon number of alkyl chain

Fig. 3.6. Plot of log adjusted retention time against carbon number of alkyl chain. Solid line for 1phenylalkanes, dashed line n-alkanes. Conditions: A, Partisil ODS-3, methanol-water (70:30); B, Finesil C18, methanol-water (80:20) and C, Lichrosorb RP-8, methanol-water (80:20). Reproduced with permission [1021 by courtesy of Marcel Dekker.

functional groups substituted on the benzene ring by comparison of the retention indices of substituted and unsubstituted standards. The increments enabled the retention indices of substituted analytes to be predicted (see Chapter 1). 3.3.2 n-Alkylbenzenes

Compared to the n-alkanes, the n-alkylbenzenes have the advantage that they can be detected spectroscopically.However, they also have a low polarity and hence long retention times compared to most compounds of interest. They have attracted relatively little interest except as secondary standards, although they have been used in a study of polychlorinated biphenyls [22]. In his studies of interactions indices, Jandera compared n-alkanes and n- alkylbenzenes as non-polar standards. He selected the latter scale to determine the effects of substitution but he did not calculate retention index values [103,104] (see Chapter 8). 3.3.3 Alkan-2-ones As part of an investigation of the identification of drugs [ 1051 based on multiwavelength References pp. 1 4 k l 4 4

110

Chapter 3

20% 40%

60%

-X

dI0

iz *

t 0

21! a 0

I

I

I

I

I

I

I

I

I

I

200 400 600 800 1000 1200 1400 1600 I800 RETENTION INDEX

Fig. 3.7. Effect of solvent composition on capacity factor. 0 , Alkan-2-ones, 0 , phenacetin. pBondapak c18 column; eluent, methanol-0.025 M phosphate buffer pH 7.0. Reproduced with permission [ 1061.

detection and relative retentions, Baker and Ma in 1979 [I061 proposed the alkan-2-ones as the first series of retention index standards specifically selected for use in HPLC. They showed that there was a linear correlation between the retention factors of the standards and their carbon number between acetone ( I = 300) and nonadecan-2-one ( I = 1900) over a wide range of eluent compositions from 20 to 90% methanol-pH 7.0 phosphate buffers and 10 to 70% acetonitrile-pH 7.0 phosphate buffer with c18 bonded columns (Fig. 3.7) and over a more limited range on a cyano-bonded column. The alkan-2-ones have moderate chromophores (typically acetone L = 265 nm, log E = 1.2 mol-' crn-' ) and can be detected spectroscopically but the sensitivity is limited. A wide range of homologues from C3 to C23is commercially available, individually or as kits. The retentions of the earlier members of the series were compatible with the retention of polar drugs and the indices, which were calculated using Eq. 3.2, were more robust

Retention index scales used in high-performance liquid chromatography

111

than retention factors to changes in the separation conditions. For example, the retention index of phenacetin as a test compound was fairly constant at about Z = 530 across a wide range of methanol-water compositions (Fig. 3.7) even though the corresponding retention factors changed from about k = 16 to 0.3. The index value was slightly different with acetonitrile-buffer (mean Z = 526) as the modifier and changed markedly on cyanobonded silica; I = 679 (methanol-buffer) and Z = 591 (acetonitrile-buffer). This scale has been widely used for the identification of drugs and other analytes (see Chapter 4), as the basis of quantitative structure-retention measurements (QSRR) (see Chapter 1) and as a reference scale to correlate retention properties with biological activity (see Section 3.4.4). Although they also based their “retention indices” on the alkan-2-ones, Schultz and Applehans [ 1071 calculated the indices in an isocratic separation by linear interpolation between the retention factors of the standards (Eq. 3.6), instead of the generally accepted logarithmic interpolation. I = loo[& -kn)/(k,,+l-k,,)] + loon

(3 4

kd, k,,, and k, + are respectively the retention factors of the analyte and the alkan-2-ones, containing n and n + 1 carbons, eluting immediately before and after the analyte. The dinitrophenylhydrazones of alkan-2-ones have also been examined as a potential retention index scale (see Section 3.3.7.3). 3.3.4 Alkyl aryl ketones

In early studies, Kikta and Grushka [77] demonstrated a systematic increase in the retention factor, with the length of the carbon side chain, for a series of alkyl aryl ketones from acetophenone to myristophenone on elution with methanol-water (50:50) from an alkylbonded column. Kkta and Stang [4] subsequently suggested that because of their regular separation these ketones could be used as a set of universal internal standards. In 1982, Smith [83] proposed that the alkyl aryl ketones might be preferable to the alkan-2-ones as retention index standards because they had stronger chromophores (e.g. acetophenonel 247,279, log E 4.1, 3.1 mol-1 cm-I). About the same time, Kuronen [lo81 also suggested their use as retention index markers for the identification of chemical warfare agents. The alkyl aryl ketones, from acetophenone ( I = 800) to octanophenone ( I = 1400) showed a linear relationship between log k and carbon number from methanol-water (30:70) to (80:20) (Fig. 3.8) [83]. If the retention indices of a series of aromatic analytes, calculated using Eq. (3.3), were compared across the eluent range they showed relatively small changes with eluent composition (Fig. 3.9) even though their retention factors changed by over tenfold. For example, the retention factor of nitrobenzene changed from k = 0.64 in 80% methanol to k = 11.5 in 30% methanol but the retention index only changed from Z = 903 to Z = 8 15 over the same range. A linear relationship was also obtained with other organic modifiers (Fig. 3.10) from acetonitrile-buffer water (10:90) to (1OO:O) [lo91 and tetrahydrofuran (THFkwater (20:80) to (60:40) [109]. The retention References pp. 140-144

Chapter 3

112 log k ' 2.0

1.5

1 .o

80

0.5

0.0

800

1000

1200

1400

Carbon No.x 1 0 0

Fig. 3.8. Retention factors for alkyl aryl ketones compared with retention index (carbon number X 100) using different percentages of methanol as eluent. Column ODs-Hypersil.Reproduced with permission [83].

indices of the test compounds were almost constant with acetonitrile composition but changed systematically with the proportion of THF (Fig. 3.1 1) [ 1091. A comparison of literature retention factors from different columns and the corresponding calculated retention indices indicated that the latter showed much greater discrimination and robustness [82,110] (see Chapter 1). This robustness of the retention indices to small changes in eluent composition prompted a study into the application of a retention index scale to standardize the identification of drug compounds by HPLC [ 110,1111 (Chapter 4). These standards also formed the basis of the CRIPES retention index prediction systems described in Chapter 1. A wide range of alkyl aryl ketones, from acetophenone to octadecanophenone, are commercially available and these standards have found widespread application in a number of laboratories for the identification of drugs, mycotoxins, polychlorinated biphenyls and in other environmental studies (see Chapter 4), retention prediction and QSAR stud-

Retention index scales used in high-performance liquid chromatography

113

I PhMe

1100

PhOEt

1000

MeOBz

PhN02

900

p-Cresol

#

800

P hC H P C H ~ O H PhCHO

700 I

I

I

I

I

1

30

40

50

60

70

80

% MeOH Fig. 3.9. Variation of retention index values of test compounds on the alkyl aryl ketone scale with proportion of methanol in eluent: Me, methyl; Et, ethyl; Ph, phenyl; Bz, benzoyl. Reproduced with permission [83]. 30% THF

1.5

40% MaCN

1.o 40% THF Y

cn

-

60% MeCN

0.5

0.0

-0.5

!

I

I

i

800

900

1000

1100

Carbon number x 100

Fig. 3.10. Relationship between logarithm of retention factors (log k) with carbon number x 100 for alkyl aryl ketones with different proportions of modifier in the eluent. Conditions: acetonitrile on ODS-Hypersil column; THF on ODS-Spherisorb column. Data from [109].

References pp. 140-144

114

Chapter 3 1100

a

1000

gX

-

900

C

C

0

c

C

Q)

c

0

a

800

700

600

20

10

40

80

SO

70

80

80

90

100

% MeCN 1200

b

4

1100

1000

X 0 0

C C

0

900

c

C

0 c 0

a

‘ \ 800

2

\x

700

000

I

10

;0

80

96 THF

i0

so

00

Retention index scales used in high-performance liquid chromatography

115

ies (see Chapter 1) and for the comparison of the characteristics of stationary and mobile phases (see section 3.4.3).

3.3.5. l-Nitroalkanes In 1988, Bogusz and Aderjan proposed the homologous l-nitroalkanes as retention index standards [112]. These compounds have a chromophore with a strong absorbance between 200 and 230nm and a weaker long wavelength band (typically nitromethane: L 278 nm; log E = 1.39 mol-l cm-l ) but a much weaker absorbance at 254 nm so detection at short wavelengths or a diode array detector is recommended. The nitroalkanes are less reactive than the alkan-2-ones and mixtures of the standards in solution can be stored for some time. Their principal advantage is that, like the alkan-2-one standards, the smallest members of the homologues series are relatively polar and have similar retention factors to polar drugs. There was a linear correlation between chain length and carbon number for the standards and the retention indices of non-polar test compounds were largely unaffected by the proportion of organic modifier. The application of these standards has principally been for toxicological drug analysis (see Chapter 5 ) and these studies have recently been reviewed [113]. Aderjan and Bogusz subsequently suggested that the same homologues could also be used for gas chromatography [1141 and this application has also recently been reviewed [ 1151.

3.3.6 Polynuclear aromatic hydrocarbons

In 1974, as part of a study to determine the adsorption energies of aromatic hydrocarbon on an alumina column using n-pentane as the eluent, Pop1 and co-workers [ 1161 suggested the use of a retention index scale based not on a homologous series of analytes but on a series of polycyclic aromatic standards containing increasing numbers of aromatic rings (n). These were assigned a logarithmic scale of retention, logI,= 1, 2, 3, 4 corresponding to benzene, I = 10; naphthalene, I = 100, phenanthrene, I = 1000 and benz[a]anthracene I = 10 000. The assignment of the index value of an analyte was made by interpolation between individual pairs of standards (Eq. 3.7), in which R,, R, and R, + were the corrected elution volumes of the analyte and the standards elutingjust before and after the analytes and containing n and n + 1 aromatic rings.

Fig. 3.1 1. Variation of retention index values of test compounds on the alkyl aryl ketone scale with proportion of modifier in eluent (conditions as Fig. 3.10). Reproduced with permission [109]. (a) Effect of proportion of acetonitrile. Compounds: 1, 2-phenylethanol, 2, p-cresol, 3, methyl benzoate, 4, N-methylaniline, 5 , nitrobenzene and 6, toluene. (b) Effect of proportion of THF. Compounds as in (a).

References pp. 140-1 44

116

Chapter 3

For compounds eluted before benzene the scale was extrapolated from the correlation between log R and log I for benzene and naphthalene. Subsequently, benzo[b]chrysene, I = 100 000, corresponding to log I = 5, was included in the standards thus enabling more highly retained analytes to be measured [117]. Initially, the scale was used to compare the retention of over 100 substituted and polycyclic aromatic hydrocarbons on alumina [ 1161 and silica gel columns [ 1171 on elution with pentane and subsequently was extended to the reversed-phase separation of polynuclear aromatic hydrocarbons [ 1181 and nitroaromatic hydrocarbons [1191 on polystyrene gel columns with methanol-water, acetonitrile-water or tetrahydrofuran-water as the eluent. An almost identical scale was used by Karlesky and co-workers [120,1211: benzene, I = 10; naphthalene, I =100; anthracene = 1000; benz[a]anthracene= 10 000 and dibenz[a,c]anthracene = 100 000. They noted that the differences compared to the previous scale were small. This scale was extended by Thomson and Reynolds [122], who used benzene, naphthalene, phenanthrene,benz[a]anthracene, dibenz[a,clanthracene, and dibenzo[def;p]chrysene (log I = 6). They quoted values as I, which were equivalent to log I, used earlier. For the calculation of indices during a gradient elution, they used a linear interpolation between the adjusted retention times (Eq. 3.8).

I, =I,

+

t',

- t',

t'n+l - t'n The main application of the polynuclear aromatic ring scales has been in the identification of the polynuclear aromatic hydrocarbons and the comparison of the shape selectivity of stationary phases (see Chapter 10). A closely related interest has been in the determination of nitropolynuclear aromatic hydrocarbons and Liu and Robbat [ 1231 suggested a corresponding retention index scale based on nitrobenzene, 1-nitronaphthalene, 9-nitroanthraceneY6-nitrochqsene and 6nitrobenzo[a]pyrene with index values of I = 100-500 (calculated as Eq. 3.8 with I =I, x 100).

3.3.7 Miscellaneous retention index scales A number of other sets of retention index standards have been examined in liquid chromatography but most have been employed by only one laboratory. 3.3.7.1 Phenolic esters

Benzoate esters will normally be too highly retained to be useful as retention index standards and they are also susceptible to hydrolysis, limiting their long-term stability. A preliminary study by Wilken [24] described the use of dialkyl phthalates as retention index standards in which the index values were 10 times the length of a single side chain so that for dimethyl phthalate, I = 10, and dioctyl phthalate, I = 80, but no further work has been located. However, a number of groups have employed the more polar alkyl 4-hydroxybenzoates (alkyl parabens), although as noted earlier questions have been raised about their stability. In a study of the characterization of stationary phases Antle et al. [124]

Retention index scales used in high-performance liquid chromatography

117

compared the relative retentions of selected analytes on different columns on a “retention index basis” based on methyl 4-hydroxybenzoate = 8 and butyl4-hydroxybenzoate = 11 but the method of calculation was not given. Yamauchi [23] suggested that the n-alkyl 4-hydroxybenzoate esters, which had been originally proposed as part of a set of phenolic internal standards [ 151 could form a retention index scale. The scale was based on the number of carbon atoms in the side chain (fiom methyl 4-hydroxybenzoate, Z = 100 to n-nonyl4-hydroxybenzoate,Z = 900) rather than conventionally on the total number of carbon atoms. They reported a linear relationship between carbon number and log k from acetonitrilewater (10:90) to (60:40) to methanol-water (20:80) to (70:30). The methyl ester was usually slightly anomalous and the methylene increment between methyl and ethyl standards was 0.89 compared to 0.98 f 0.2 for the higher homologues. They used this scale to measure the indices of a further 29 substituted phenols and found the values to be largely independent of the proportion of modifier. By using the experimental retentions of the methyl and ethyl esters and uracil, they devised a procedure for the calculation of the expected retention of the substituent phenols in different eluents. In an unpublished study, Rozing and Weinand [125] compared the alkyl aryl ketones, alkyl benzoates and alkyl 4-hydroxybenzoates as retention index standards and selected the last set as they covered the elution of more polar analytes. They based the indices on the number of carbons in the alkyl side chain (methyl 4-hydroxybenzoate; Z = 100). They used these results to compare mobile phases and batches and brands of stationary phases by using phenol, benzaldehyde, benzyl alcohol, nitrobenzene, NN-dimethylaniline and phenetole as test compounds. 3.3.7.2Aliphatic esters

Aliphatic esters as well as being subject to hydrolysis also lack a chromophore and are therefore difficult to detect. They have therefore only been used for the characterization of lipids and related samples of similar polarity and when non-spectroscopic detectors are already in use. In early studies, Plattner and co-workers [126] separated triglycerides on a reversed phase column and found a linear relationship between the carbon number of the acyl groups and log k. The presence of a double bond reduced the retention by the equivalent of two methylene groups. This enabled them to describe the composition of a mixture in terms of the “equivalent carbon number” (ECN), which was equal to the number of carbons in the saturated triglyceride. Porter and co-workers found that the retention volumes of a series of lecithins in non-aqueous reversed-phase HPLC were linearly related to their carbon number [127]. These compounds were used to establish an “effective carbon number scale” based on number of acyl carbon atoms less the number of double bonds. Subsequently, Compton and Purdy determined the retention factors for saturated and unsaturated triglycerides and phospholipids [ 1281. They assigned the saturated triglycerides retention index values of Z, = number of acyl carbons X 100, so that for tripalmitate, Z = 4800. These standards were then used to calculate the corresponding retention indices of unsaturated triglycerides (for example triglyceride of cis-hexadecenoic acid, Z,,= 4158). From the relationship, Z,,= I, - CDb in which Db = the number of double References pp. 140-144

118

Chapter 3

bonds, they were able to determine the retention index increment C per double bond. The values of C for the triglycerides (C = 202-262) were significantly higher than the value for phospholipids (C= 150-174). Takahashi et al. [129] carried out similar studies based on the ECN of a range of saturated and unsaturated triglycerides but after a detailed regression analysis concluded that the double bond coefficient of -2.2 was close to the accepted value of -2.0. In a more recent study, Podlaha and TOrgPlrd [130] proposed an equivalent carbon number scale (ECN) based on the retentions of the saturated triglycerides. In practice the scale was defined by injecting trilaurin (ECN = 36.00), trimyristin (ECN = 42.00) and tristearin (ECN = 54.00). 3.3.7.3 Other retention index scales

Gassiot-Matas et al. [ 1311 studied the potential application of the 2,4-dinitrophenylhydrazone (DNPH) derivatives of homologous alkan-2-ones, from propan-2-one (I = 900) to pentan-2-one ( I = 1loo), as a retention index scale for use in isocratic and gradient elution. These compounds have much stronger chromophores than the original alkan-2ones but are more non-polar. They used them to study a series of DNPH derivatives of alkanals and branched chain ketones. Subsequently, changes in the retention times of 2 4 dinitrophenylhydrazine with mobile phase composition were used by Margarit-Roig et al. [132] to calculate the column hold-up volume. Then, using the linear relationship of the retention factors with carbon number for the DNPHs of formaldehyde (ZR= 1) to nonanal (ZR= 9), they determined retention index (ZR) values for a range of 17 substituted aromatic compounds with polarities from benzoic acid to anthracene. The index values changed little with eluent composition between (40:60) and (60:40) methanol-water or acetonitrile-water on RP-8 and RP-18 columns. A novel series of compounds, the 1-[p-(2,3-dihydroxypropoxy)phenyl]-1-alkanones (D-compounds) have been used by Kuronen and co-workers [133,134] as index standards for the identification of mycotoxins as they were more polar than the alkyl aryl ketones. Their application is described in more detail in Chapter 6.

3.3.8 Comparisons between retention index scales

Relatively few studies have directly compared retention indices measured on different scales. In theory, a direct transfer of values should be possible if the methylene increment is the same for each set of standard compounds. However, as noted earlier, small differences have frequently been observed between homologous series and the value of the increment appears to depend on the functional groups that are present. When Smith [83] first examined the alkyl aryl ketones as potential standards, a series of alkan-2-ones was also measured for comparison. Hexan-2-one had a similar retention to acetophenone but the retention index value increased slightly with the proportion of modifier from I = 787 to 829 in methanol-water (30:70) to (70:30). The methylene increments of the two sets of homologues were very similar. The principal difference was that for easy detection, 1% of the alkan-2-ones but only 0.0025% solution of the alkyl aryl ketones were required because of their stronger chromophore.

Retention index scales used in high-performance liquid chromatography

119

2.00 1 1.50

-

1.00

-

0.50

-

0.00 -

-0.50 0

200

400

600

800

1000

1200

Carbon number x 100

2.0 -

-a

1.5

-

1.0

-

0 O.0 I -0.5

4

I

I

0

200

400

I

600

800

1000

1200

Carbon number x 100

Fig. 3.12. Relationship of log k to carbon number x 100 for the three homologous series, alkyl aryl ketones 0, alkan-2-ones 0,and nitroalkanes A . Reproduced with permission [135]. (a) Eluent, methanol-buffer pH 7.0 (40:60). (b) Eluent, acetonitrile-buffer (20:80).

In a later study [135], a direct comparison was carried out of the homologous 1nitroalkanes, alkan-2-ones, and alkyl aryl ketones and the retention index scales were applied to a set of column test compounds and a number of typical polar drugs. The first members of the two aliphatic series of analytes had shorter retention factors than acetophenone (Fig. 3.12). The three series had very similar methylene selectivities but nitromethane and nitroethane deviated from a linear relationship. A similar effect was also reported by Bogusz and Aderjan [112]. The retention indices for the homologues and analytes were calculated on all three scales (Table 3.3). Almost all the drugs compounds were eluted within the alkan-2-one and nitroalkane scales but an extrapolation below acetophenone was necessary for their retention indices to be determined on the alkyl aryl ketone scale. However, it was demonstrated [I361 that these extrapolated results can be very reproducible, down to I = 600 on the alkyl aryl ketone scale, which is nearly equivalent to nitroethane (I = 577) or butan-2-one ( I = 596). References pp. 140-1 44

120

Chapter 3

TABLE 3.3 COMPARISON OF RETENTlON INDICES CALCULATED USING DIFFERENT RETENTION INDEX SCALES Compound

Retention index scale Alkan-2-one WO)

Nitroalkane I(N02)

Allcan-2-ones Acetone Butan-2-one Pentan-2-one Hexan-2-one Heptan-2-one Nonan-2-one

300a 400 500 600 700 900

118 219 319 422 520 (676)

I -Nitroalkanes Nitromethane Nitroethane Nitropropane Nitrobutane Nitropentane Nitrohexane

(257) 342 426 564 676 798

1ooa 200 300 400 500 600

619 (723) (833)

-

426 516 (607) (700)

589 582 603 626 794

412 404 425 448 593

(170) 308 327 42 1 452 400 504 579

124 147 242 213 22 1 324 40 1

Ally[ aryl ketones Acetophenone Propiophenone Butyrophenone Valerophenone Column test compounris N-Methylanil ine 2-Phenylethanol p-Cresol Nitrobenzene Toluene

PhCOR

@m

Drug compounds Aspirin Paracetamol Theophylline Barbitone Salicylamide Caffeine Phenobarbitone Phenacetin

(
~

Data from [ 1351. Conditions: column, ODS-Hypersil; eluent, methanol-phosphate buffer pH 7.0 (40:60). Indices in parentheses have been calculated by extrapolation. aIndex standards I = carbon number x 100.

Retention index scales used in high-performance liquid chromatography

121

TABLE 3.4 COMPARISONOF CAPACITY FACTORS AND RETENTION INDICES ON DIFFERENT STATIONARY PHASES Compound

Column test compounds N-Methylaniline 2-Phenylethanol pCresol Nitrobenzene Toluene

Drug compounds Paracetamol Theophylline Barbitone Salicylamide Caffeine Phenobarbitone Phenacetin

WO)

k Column ODs-H

ODs-Z

2.60 2.31 2.65 3.47 12.14 0.41 0.40 0.69 0.88 0.64 1.06 2.1 1

ODs-H

ODs-Z

2.29 1.26 1.27 3.53 13.60

589 572 592 630 828

598 504 505 615 878

0.31 0.44 0.40 0.57 1.50 0.44 1.27

(282) (277) 386 428 371 458 559

(189) (267) (245) 326 533 (267) 503

Data from [1351. Retention indices calculated on the alkan-2-one scale. Conditions: eluent, methanol-buffer pH 7.0 (50:SO).

Because they can avoid extrapolation, the results suggested that the aliphatic standards might be more suitable for the identification of polar drug analytes. However, the indices of the drug compounds changed markedly with eluent composition, for example, caffeine, I = 485 changed to I = 371 and barbitone from I = 481 to I = 386 (on the alkan-Zone scale) on changing from 20 to 50% methanol, although there was almost no change in the retention indices of the column test compounds; i.e. 2-phenylethanol, I = 586-572 over the same range. When the separations were compared on two different ODS-silica stationary phases, the retention indices of both the column test compounds and drugs changed markedly (Table 3.4). These effects were more marked for the analytes with the shortest retentions and the order of elution for a number of the drugs changed between the two stationary phases. Thus, although the aliphatic retention index standards may be more appropriate for rapidly eluted analytes, the results appear to be less robust than those of more highly retained analytes. One possible explanation is that many of the polar analytes are being retained by a mixed mode of retention, involving interactions with the surface of the stationary phase as well as hydrophobic interactions. In contrast, each of the homologous standards have the same interactions with the stationary phase and differences in their retention are caused only by the difference in chain length. Thus, the primary factor in the short retention of the drugs is their high polarity, whereas the first members of the homologous series are rapidly eluted because of their small size. In their studies, which are described in detail in Chapter 5 , Bogusz and co-workers [ 1371 proposed that it is possible to correct for differences between retention indices References pp. 140-144

122

Chapter 3

measured on difference stationary phases by scaling the results using standard reference compounds. They concluded from these results that the nitroalkanes were more suitable as a retention index scale for basic drugs than the alkyl aryl ketones. This was disputed in a note by Smith et al. [ 1381, who suggested that once a correction had been applied, there was no difference between the retention index scales as effectively both had been converted to corrected relative retentions compared to the reference compounds. However, the correction technique works well in many examples but it cannot compensate for differences between columns which result in changes in the order of elution (as in Table 3.4 and later examples in Chapter 4).

3.4 APPLICATIONS OF RETENTION INDEX SCALES Many of the applications of retention index scales in HPLC have been very similar to their roles in gas chromatography: (a) identification of analytes; (b) characterization of the separation system (stationary and mobile phases); (c) determination of lipophilicity and relationship to biological activity of analytes; (d) examination of structure-retention relationships. The principal differences in their use have been in emphasis, primarily because in liquid chromatographic the separation system is influenced by both the mobile and stationary phases. The wide elution range of liquid chromatography has also resulted in a range of retention index scales with little direct correlation.

3.4.1 Reproducibility and transferability of retention indices It is often difficult to obtain high reproducibility in HPLC because the conventional retention values are sensitive to small changes in the separation conditions of temperature, mobile phase composition and flow rates, differences in the stationary phases with time, and between columns from the same or different manufacturers and to differences in the measured value of the column hold-up volume. In order to be valuable as a guide to retention properties in a HPLC system, retention indices must therefore be able to demonstrate high reproducibility and transferability between systems and between laboratories. Any differences that remain between measurements can then be attributed to the effects of real differences in the separation conditions. As noted earlier, a comparison of published retention factors with retention indices calculated from the same data showed the latter to have a much lower variation and higher discrimination [ 1 101. On a single system, a number of laboratories have demonstrated high reproducibility and an insensitivity to the experimental conditions. Smith [ 1391 showed that because they were interpolated measurements, the retention indices were effectively independent of the value of the column hold-up time used to determine the retention factors. A 15% change in the nominal value for tMcaused a similar change in the retention factors but the retention indices changed by less than 1% even for rapidly eluted analytes, such as

Retention index scales used in high-performance liquid chromatography

123

p-cresol: nominal value k = 0.81, range 0.57-0.13 but nominal value I = 801, range only 797-803. Small changes in eluent composition, such as those resulting from day to day variations in the preparation of the eluent by different operators have little effect on retention indices (see Fig. 3.8). Temperature changes can markedly change retention factors but usually exert a similar effect on all the analytes in a sample. As a result, relative retentions such as retention indices are more robust and this was confirmed in robustness studies on the separation of drugs by Smith et af.[140]. On changing from 10 to 40°C, the retention factors of nitrobenzene and toluene changed from k = 7.24 to 4.15 and from k = 28.62 to 16.11, respectively, but the retention indices only changed from nitrobenzene Z = 832 to 827 and for toluene from I = 984 to 990. However, partially ionized barbiturates changed more markedly, e.g. quinalbarbitone from I = 948 to 901, because of the secondary effect of temperature on the degree of ionization. For high reproducibility across a range of analytes, the temperature of the chromatographic system should therefore be controlled. As part of the retention prediction studies described in Chapter 1, a detailed examination of the reproducibility of retention indices was carried out over a 2-year period [136]. A single batch of stationary phase was used in a number of different columns and the temperature and composition of the mobile phases were closely controlled. Even with these precautions, there were variations of up to 18% in the retention factors pig. 3.13) but the retention indices were extremely reproducible with a standard deviation of 5 units. This study also re-examined the influence of differences in the measured values of the hold-up time and found that a 15% change caused the retention index of phenylacetamide to change only from 645 to 63 1, even though this analyte is rapidly eluted and was outside the calibration range of the alkyl aryl ketone standards. The intralaboratory reproducibility has also been compared with interlaboratoryreproducibility. Baker and et al. [I411 examined the separation of seven drug compounds using the alkan-2-one scale. Within one laboratory the indices for aspirin ( I = 178, SD = 9.4) and androsterone (Z= 876, SD = 7.2) showed the greatest variations (Table 3.5). Overall, the mean variation was 3.1 units and the mean relative standard deviation (RSD) was 0.99%. In contrast, the adjusted retention times varied by 2.63%. If the same drugs were examined in six laboratories on a range of columns, the differences were much higher and the indices for aspirin showed a wide range, I = 197 f 60, and for androsterone, I = 952 f 133 (Table 3.5). Overall the retention indices of the drugs showed a mean relative standard deviation of 12.6%. The differences in the adjusted retention times were very much higher and had a relative standard deviation of nearly 35%. One of the major causes was considered to be differences between the stationary phases used in the different laboratories. If the results from four laboratories using p-Bondapak CI8 were compared the RSD of the retention indices was reduced to 3.6% and of the adjusted retention times to 12.9% The temperature of the different systems was also not controlled. A more closely controlled collaborative study was carried out by Gill et al. 11421 in which the same stationary phase was used in each laboratory for the separation of a set of barbiturates. The retention factors showed wide variation with a coefficient of variation (CV) of 13%. The retention indices were better with standard deviations from 6 to 12 units and a CV of 0.6-2%. These results were poorer than in interlaboratory studies and References pp. 140-1 44

124

a

Chapter 3

l6

B

C

E

D

14

12

L

0 c

u

10

Q

m $

6

0 4

2

0 20

25

30

Measurement

Measurement

Fig. 3.13. Individual measurements of phenol (a),benzene (0)and toluene (m), throughout a 2-year study using acetonitrilebuffer (40:60) as the mobile phase on four different columns, R E packed with the same batch of Spherisorb ODS2. Reproduced with permission [136]. (a) Retention factors; (b) retention indices on the alkyl aryl ketone scale.

this was principally attributed to a lack of temperature control (see also Chapter 4).Many of the other identification studies have also reported good intralaboratory reproducibility on specific assay systems (Chapter 4). Clearly the retention indices provide a significant improvement in reproducibility compared to retention factors both in short- and long-term studies. However, they are susceptible to effects such as temperature and different stationary phases, which alter the selectivity of the separation but even in these cases the influence is probably less than on the retention factors. Thus, even retention indices cannot compensate for differences in the separation properties of chromatographic systems and in setting up intralaboratory

Retention index scales used in high-performance liquid chromatography

125

TABLE 3.5 COMPARISON OF RETENTION INDICES OF DRUGS IN INTRALABORATORY AND INTERLABORATORY STUDIES Compound

Mean retention index (4 (SD) Intralaboratory

Aspirin Caffeine Phenobarbital Phenacetin Methaqualone Chlordiazepoxide Androsterone

179 457 538 593 760 825 816

Interlaboratory (A-F) (9.4) (1.2) (1.3) (0.7) (0.6) (1.4) (7.2)

197 471 534 612 193 893 952

(60) (45) (24) (50) (54)

(131) (133)

Data from [1411. Retention index values measured using alkan-2-ones as standards. Eluent: methanol-buffer 60:40.

and collaborative studies, it is essential to carry out robustness studies and to identify those factors that need to be closely controlled to achieve good reproducibility. More fiequently, the differences in retention indices on different columns have been used as a measure of the differences in the properties of the column materials in a similar manner to the Rohrschneider constants in gas chromatography (see Section 3.4.3.2). However, by using internal reference standards, Bogusz has described methods for the correction of retention indices on different columns (Chapter 5). These are most valuable for groups of related compounds but are less successful if the change in the stationary phase causes a change in elution order. 3.4.2 Identification A primary application of retention indices in liquid chromatography, as in GLC, has been for the identification of the components of mixtures. This is potentially particularly valuable as, unlike in gas chromatography, it is difficult to routinely couple HPLC separations with mass spectrometry or FT-infrared spectroscopy. Often the only spectroscopic identification in liquid chromatography comes from diode array or multiwavelength ultraviolet detectors, which have a relatively low information content. A wide range of applications to the identification of drugs, pesticides, environmental samples and natural products are given in Chapter 4, with studies on more specialized areas of toxicological drug analysis being described in Chapter 5 and on mycotoxins in Chapter 6. Most of these studies have used either the alkyl aryl ketones or nitroalkanes as standards or have chosen specialized sets of standards to match particular analytes, such as the polynuclear aromatic hydrocarbons (Chapter 10). 3.4.3 Characterization of separation systems A comparison of the retention indices of test compounds determined on different systems References pp. 140-144

126

Chapter 3

can be used to characterize the retention properties of mobile and stationary phases in liquid chromatography in much the same way that Rohrschneider and McReynolds constants have been used for many years in gas chromatography. The values can then be used to identify the changes that might be expected on changing separation conditions such as eluent composition, organic modifiers or type of stationary phase. In addition, because the retention factors of a homologous series of analytes, such as a set of retention index standards, should be related by the Martin equation to the carbon number, it should be possible to determine the effective value of the hold-up volume of a column as that value which results in the best correlation between retention factor and chain length. 3.4.3.1. Column hold-up volume

The first studies to calculate the hold-up volume in liquid chromatography were reported by Berendsen et ul. in 1980 [143]. They suggested that the linearization of the net retention times of a homologous series, which had been used in gas chromatography, could also be adopted in liquid chromatography. They found a precision of 4 % for values calculated from different homologous series but the values differed from those obtained by alternative techniques. They preferred to use the n-alkanols as, despite their lack of chromophore, they could be used in the widest range of eluents. However, when this method was applied to homologous alkylbenzenes, alkyl aryl ketones, phenylalkanols, and Nalkylanilines on a polymer column, Smith and Garside [ 1441 found very different calculated hold-up volumes of 4.74 ml, 4.97 ml, 2.22 ml and 3.21 ml, respectively. In other studies, Wainwright et ul. examined the use of the n-alkylbenzenes but concluded that in gas chromatography, they were unsuitable and that their use in liquid chromatography might also be unwise [145]. Subsequent work by Laub and Madden using n-alkylbenzenes and w-phenylalkanols [ 1461 suggested that the problem was more complicated if different modes of interaction were influencing different homologues. A subsequent paper by Djerki and Laub suggested that calculations based on higher homologues might be more reliable [9]. Van Tulder and co-workers used a computer-based linearization approach and suggested guidelines for the selection of the optimum homologues to be included in the calculation [147]. Furr [148] offered a computer programme for the calculation of column hold-up volumes. The results were demonstrated using a series of retinyl esters with acyl groups from C8-CI8, in a non-aqueous separation. Similar calculations with C1&0 alkyl aryl ketones were then used to determine the I values of retinol, retinyl acetate and retinaldehyde in acetonitrile-water (85:15). The two series gave similar column hold-up volumes of 0.989 and 0.937 ml. More recently, Wiitzig and Ebel offered an alternative nonlinear regression algorithm which was claimed to overcome some of the problems with the earlier programmes [149]. They applied this method to the homologous alkan-2-ones and the calculated retention indices of the standards closely matched their defined values. This method was also found to give consistent values by Rozing and Weinand in their study of retention indices [ 1251. However, there is still disagreement regarding the validity of these methods. M6ckel and co-workers [ 1501, in a study of the n-alkanes, concluded that the higher homologues were being excluded from the pores of the stationary phases distorting the resulting value

Retention index scales used in high-peuformance liquid chromatography

127

for the hold-up volume. They were also concerned [150,151] that the precision of the retention times of the standards, because of variations in the pumping rates, would not provide sufficiently accurate data to allow usefid calculations of the column hold-up volume. 3.4.3.2Stationaryphase characterization

During their initial examination of the aikan-2-ones as retention index standards Baker and Ma [lo61 found differences between the retention indices of drugs measured on columns with different bonded phases and suggested that ketones might form the basis of a system similar to the McReynolds constants to characterize column selectivity in HPLC. This aspect was also explored by Smith [83] in his studies of the alkyl aryl ketones. The retention indices of simple aromatic test compounds differed with different types of stationary phase. It was suggested [ 1521 that if a set of diagnostic column test compounds could be chosen, then it should be possible to establish a “McReynolds” type classification of stationary phases. Because many standards used to measure McReynolds constants [50] contain no significant chromophore, a set of aromatic compounds, benzaldehyde, benzyl alcohol, 2-phenylethanol, p-cresol, methyl benzoate, phenetole, nitrobenzene and toluene, was chosen for evaluation [ 1521. The retention indices were determined on four different column materials, with C22, CI8, C1 (SAS), and phenyl-bonded phases (Table 3.6). Benzyl alcohol and benzaldehyde were then dropped as standards because the former behaved in a similar manner to 2-phenylethanol and the latter had virtually the same retention index on each column. In gas chromatography the properties of a stationary phase are determined by comparison of the retention indices of the column test compounds, compared with their values on a squalane column, defined as the least polar commonly used stationary phase. However, no corresponding reference column material has been identified in HPLC. It was thereTABLE 3.6 RETENTION INDEX VALUES OF COLUMN TEST COMPOUNDS ON DIFFERENT BONDED STATIONARY PHASES Test compound

Benzaldehyde Phenetole Toluene Methyl benzoate Nitrobenzene Acetophenone Benzyl alcohol 2-Phenylethanol p-Cresol

Retention index ( I ) Column ODS

SAS

c22

Phenyl

758 952 998* 90 1 815 800 70 1 772 798

755 914 892 895 816 800 677 745 779

746 895 867 889 793 800 657 733 74 1

747 809 745 848 802 800 543 625 618

Data from [I521 on the alkyl aryl ketone scale. Conditions:eluent, methanol-water (70:30), *(40:60). Columns: ODS, ODS-Hypersil;SAS, SAS-Hypersil, C22, C22-Magnusil; Phenyl, Phenyl-Spherisorb.

References pp. 140-1 44

128

Chapter 3

TABLE 3.7 COLUMN RETENTION CONSTANTS ON DIFFERENT BONDED STATIONARY PHASES [ 1521 Compound

Phenetole Toluene Methyl benzoate Nitrobenzene Acetophenone Benzyl alcohol p-Cresol

Ihexae-water

1039 1017 93 1 840 800 559 61 1

Retention constant (AZ) Column ODS

SAS

c22

Phenyl

-87 -19 -3 0 -2 5 0 142 187

-125 -125 -3 6 -24 0 118 168

-144 -150 -42 -47 0 98 130

-23 0 -272 -83 -3 8 0 -16 7

Retention constant = I - Ihexme-water Retention indices on the alkyl aryl ketone scale. Conditions: eluent, methanol-water (30:70), *(40:60).Columns as Table 3.6.

fore proposed [ 1521 that the hexane-water partition system could be used as the reference partition phase because this would be free of silanol and other specific interactions. From the partition coeacients of the test compound and the alkyl aryl ketones, it was possible on this phase. The retention conto calculate the apparent retention indices (Ihexme-water) stants (Zx - Ihexmewater) could then be determined for the column test compounds (Table 3.7). As expected, the values depended on the different phases, the phenyl-bonded phase showing the greatest effects. However, the values also changed markedly with the proportion of methanol in the mobile phase; phenetole, AZ= -87 in methanol-water (30:70) to +14 in methanol-water (80:20). With the same eluent, the retention factors on the phenyl and ODS columns differed markedly and might reflect a difference in retention ability. A further study of the retention of the column test compounds was carried out using eluents chosen to give similar retention factors [89]. The indices for the aromatic hydrocarbons on the phenyl column were consistently high on the ODS-silica column: toluene, k = 2.3 1 and I = 1047 on ODSsilica with methanol-water (70:30) but k = 2.08 and I = 745 on phenyl-bonded silica with methanol-water (30:70). The values for the column tests compounds in the same eluents also differed: 2-phenylethanol, Z = 795 and 625; p-cresol, I = 8 15 and 6 18; nitrobenzene, Z= 869 and 802, respectively. This study was then extended to a detailed examination of the differences between eight ODS-bonded column, using three isoeluotropic solvents: methanol-water (70:30), acetonitrile-water (5050) and THF-water (40:60) [ 1391. The retention indices of the column test compounds were measured compared to the alkyl aryl ketones. When the results were analysed using principal components analysis, the major independent contributors were toluene, 2-phenylethanol and p-cresol and these were identified as characteristic column test analytes. Nitrobenzene was also retained as in the earlier work [89] it appeared to demonstrate a specific effect on the bonded phenyl column. Although the retention indices of methyl benzoate changed slightly with the mobile phase modifier, they were virtually independent of the stationary phase: methanol-water, Z = 895-913; acetonitrile-water, Z = 884-89 1; and THF-water, I = 884-893. Phenetole duplicated dif-

Retention index scales used in high-performance liquid chromatography CCt

129

2

4-60 Fig. 3.14. Comparison of column-eluent combinations using principal components analysis. Plots of the values of the first and second components from the multivariant analysis of the retention indices of six reference compounds separated on eight columns with methanol-water (70:30),acetonitrilewater (5050) and THF-water 40:60) as eluents. Columns: HI and H2, ODs-Hypersil 5 pm; H3, ODs-Hypersil 3 pm, T, ODs-Techsil, Z, Zorbax ODs, P, ODs-Partisil 10pm; S, Spherisorb ODS and L, Lichrosorb RP-18. Reproduced with permission 11391. 0 1984 American Chemical Society.

ferences identified by other standards and these last two compounds were therefore also dropped fiom the test samples. The unsupervised correlation of the test data showed that the three Hypersil columns were very similar and that the low coverage ODs-Partisil column and ODs-Zorbax columns were markedly different in each eluent [139]. However, the major influence on the differences in the retention indices was the mobile phase and the three modifiers were clearly distinguished in a plot of the first and second principal components (Fig. 3.14). Because it had been claimed that different ODs-bonded phases could cause marked effects on the retention of basic compounds, N-n-alkylanilines were separated under a range of conditions of different percentage of methanol, temperature and eluent pH fkom 7.58 to 9.73, using a methanol-phosphate buffer [153]. Although there were changes in the retention factors, the retention indices were virtually independent of the conditions confirming that if the degree of analyte and silanol ionization was controlled by using a buffer, then reproducible separations could be obtained with good peak shapes. The reproducibility on four different batches of Hypersil ODS was very good: N-n-butylaniline retention factors, k = 43.93, RSD 7.4%; retention indices I = 1085, RSD = 0.19%. If the separations on different brands of ODS-bonded silica were compared, although there were very large differences in the retention factors (N-butylaniline, k = 8.26-82.76), the retention indices were almost identical, I = 1067-1088 (Table 3.8). If instead of using a high pH, hexylamine was added to the eluent to mask the silanol groups at pH 2.5, the retention indices became very sensitive to the conditions suggesting that ionic interactions References pp. 140-1 44

Chapter 3

130

TABLE 3.8 COMPARISON OF RETENTION FACTORS AND RETENTION INDICES OF N-ALKYLANILINES ON DIFFERENT BRANDS OF ODS-SILICA Column

3 ,urn Hypersil ODS 5 ,urn Hypersil ODS Techsil ODS Spherisorb ODS Zorbax ODS Partisil ODS

Retention factor (k)

Retention index (0

PhNHMe

PhNHBu

PhNHMe

PhNHBu

3.54 4.40 3.12 2.77 6.90 1.72

45.67 52.53 3 1.67 22.89 82.76 8.26

790 790 788 767 765 812

1088 1088 1081 1067 1070 1072

Data from [139] based on the alkyl aryl ketone scale. Eluent: methanol-phosphate buffer pH 8.5 (40:60). Compounds: PhHNMe, N-methylaniline; PhHNBu, N-butylaniline.

were present. The value for N-butylamine changed from I = 622 to I = 757 on increasing the proportion of hexylamine from 0 to 1.4% [ 1531. The retention indices of the column test compounds have also been used to compare cyano- and ODs-bonded phases in different proportions of methanol and acetonitrile [ 1541. Although the indices on Ultrasphere cyano and Spherisorb-CN columns were similar, the values on a CPS-Hypersil column were markedly different (Table 3.9) and were similar to those found on a short alkyl-bonded silica (SAS-Hypersil) suggesting that the cyano-group was playing little part in the separations. The data were re-analysed by Park [ 1551 although they considered only the retention factors. In other studies, Figge et ul. [87] have determined the retentions of different homologous series of analytes relative to the n-alkanes on polymer coated columns. The results (Table 3.10) compared the retention indices with those on a standard C18-silicacolumn produced by silanization. The high differences for the alkanols and pyridines on the TABLE 3.9 COMPARISON OF COLUMN TEST COMPOUNDS ON CYAN0 BONDED STATIONARY PHASES Compound

2-Phenylethanol N-Methylaniline p-Cresol Nitrobenzene Methyl benzoate Toluene

Retention index (0 Column CPS

CN-S

Ultra-CN

ODS

SAS

746 782 808 87 1 899 92 1

574 698 574 817 826 793

647 706 642 807 869 804

773 776 797 813 90 1 957

747 779 816 895 892

Data from [I541 based on alkyl aryl ketone scale. Eluent methanol-phosphate buffer pH 7.0 (30:70).Columns: CPS, CPS-Hypersil, CN-S, Spherisorb CN, Ultra-CN, Ultrasphere cyano; ODs, ODS-Hypersil; SAS, SASHypersil.

Retention index scales used in high-performance liquid chromatography

131

TABLE 3.10 SELECTIVITY OF HOMOLOGOUS COMPOUNDS ON POLYMER COATED PRESILANIZED SILICA AND POLYMER COATED ALUMINA Homologues

CH2= CH-R Ph-R R-CO2CH3 R-CO-C~HS R-OH 2-Alkylpyridines

Retention index reductiona Stationary phase C18E

C18-PMS

PBD

XE60

AsY-PBD

110 425 480 590 520 780

130 500 700 840 1060 1110

110 380 550 680 800 900

100

360 520 600 800 950

350 710 650 1000 ~

~~

Data from [87]. aThe retention index differences are the amount by which the retention index of the homologous series (expressed on the n-alkane scale) was reduced compared to its retention on a standard c18 silica, Eluent, methanol-water (80:20). Columns: C18-E, C18-COated silica; PMSC18-E, PMSCl8 coated silica; c18polymethylsiloxane;PBD; polybutadiene; XE60; cyanosilicone; A5Y-PBD; polymer coated alumina.

PMSC18and PBD corresponds to a rapid elution indicating reduced interaction with the polymer coated stationary phase. ODS- and polystyrene-divinylbenzene (PS-DVB) columns have also been compared [144,156]. However, it was noticeable that the relationship between carbon number and retention factor was only linear for the higher alkyl aryl ketone homologues and that extrapolation from propiophenone ( I = 900) to valerophenone ( I = 1 100) gave lower than expected values for acetophenone (found I = 759-776, nominal I = 800). A subsequent comparison of the homologous alkylbenzenes, phenylalkanols, N-alkylanilines and alkyl aryl ketones (greater than acetophenone), found that they all gave good correlations (>0.998) [ 1441. The organic modifier had a marked effect on the peak shapes of the alkyl aryl ketones on the PS-DVB column (Fig. 3.15) and the efficiency of valerophenone increased from N = 651 in methanol-water (9O:lO) to N = 2837 in acetonitrile-water (70:30) to N = 4166 in THF-water (40:60). When the retention indices of the column test compounds were compared with their values on an ODs-column using the same eluent, the values for the polar analytes p-cresol and 2-phenylethanol were significantly lower (Table 3.1 1) [ 1441 indicating marked differences in the properties of the phases. These results were compared by Gawdzik and co-workers [ 1571 with separations on a polymeric di(methacryloyloxymethy1) naphthalene-divinylbenzene polymer (DMN-DVB) column. They also observed deviation from linearity for propiophenone and acetophenone and for the first homologues of the n-alkylbenzenes, alkyl aryl ethers, alkyl benzoates and N-alkylanilines. The methylene selectivities differed for the different homologues from 118 units to 74 units with the greatest deviations when THF was used as the organic modifier. Using the same eluents as in Table 3.11 the DMN-DVB column showed a similar difference between the polar and non-polar analytes. The indices for nitrobenzene, I = 1032, 938 and 985 in methanol-water (90:10), acetonitrilewater (70:30) and THF-water (40:60), were contrasted with p-cresol, I = 595, 653, 728, and 2-phenylethanol, I = 478, 53 1, 399, respectively. The modifier also had similar effects on the effiRefirences pp. 140-I 44

132

Chapter 3

b

a 10

Fig. 3.15. Peak shape of homologous alkyl aryl ketones, acetophenone to heptanophenone, separated on a PLRP-S column with (a) methanol-water (9O:lO) and @) THF-water (40:60). Reproduced with permission [144].

ciency of the column to the PS-DVB study. A second study [1581 used the same approach to compare the effect on the selectivity of the retention of the column test compounds by preparing DMN-DVB polymer containing different ratios of the two monomers. A recent extension of this study has compared the effect of the addition of cg and Clg-alkyl side chains to the DNM-DVB polymer [1591. Although there were significant changes on alkylated and unsubstituted material, the difference between selectivity of C8 and c18 chains was small (Table 3.12). Pesek and co-workers in a series of studies (Chapter 11) have used the alkyl aryl retention index scale and column test compounds to characterize novel stationary bonded and liquid crystal phases based on ally1 bonded silica [ 1061. The major role for the polynuclear aromatic hydrocarbon standards has been in the comparison of differences in the shape selectivity of reversed-phase stationary phases for the separation of the polynuclear aromatic hydrocarbons [161-1631. This work is described by Sander and Wise in Chapter 10. Similar studies have also been carried out by other groups. Chmielowiec and George [ 1641 compared the retention properties of a large number of aromatic hydrocarbons on an amino-bonded column on elution with n-heptane

Retention index scales used in high-performance liquid chromatography

133

TABLE3.11 RETENTION INDEX VALUES OF COLUMN TEST COMPOUNDS ON ODS AND PS-DVB COLUMNS Compound

2-Phenylethanol p-Cresol Nitrobenzene Toluene

Retention index (4 Eluent MeOH-HzO (90:lO) Column

MeCN-H20 (70:30)

THF-HzO (40:60)

PS-DVB

ODS

PS-DVB

ODS

PS-DVB

ODS

393 828 873 933

667 682 853 1128

509 568 875 1061

676 716 853 1050

607 922 1070 1144

711 862 92 1 1064

Values from [I441 based on the alkyl aryl ketone scale. Columns: PS-DVB, Polymer Laboratories PL-RPS, ODs, Spherisorb ODS column.

and a CI8column on elution with 80% acetonitrile-water. Thomson and Reynolds used a similar scale to compare a limited group of hydrocarbons on a number of charge-transfer bonded columns [ 1221. As well as these studies to examine the differences between stationary phases, a number of reports have compared retentions on different stationary phases as part of robustness or reproducibility studies. Some of these are described in Chapter 4 as part of identification studies. In other work, Yamauchi [23] measured the retention indices of a series of substituted phenols, using the 4-hydroxybenzoate esters as standards, on seven alkyland phenyl-bonded silica columns. They noted that analytes with a second phenyl ring had relatively higher retentions on the phenyl than alkyl-bonded columns but for the other analytes there was little difference between CIg, cg and phenyl columns. 3.4.3.3 Mobile phase characteristics The retention indices of analytes have fiequently been reported using different organic TABLE 3.12 COMPARISON OF THE RETENTION INDICES OF COLUMN TEST COMPOUNDS ON ALKYL SUBSTITUTED DI(METHACRYLOYL0XYMETHYL)NAPHTHALENE-DIVINYLBH'JZENE (DNMDVB) COPOLYMERS Compound

N-Methylaniline p-Cresol 2-Phen ylethanol Nitrobenzene Toluene

Retention index (4 Column DMN-DVB

DMN-DVB-Cg

DMN-DVB-Clg

306 492 561 923 1012

281 669 626 900 928

288 688 626 904 935

Data from [159] based on the alkyl aryl ketone scale. Eluent: acetonitrile-buffer pH 2.3 (5050).

References pp. 140-144

Chapter 3

134 RETENTION INDEX 1looT

60:40 MeOH-Buffer

I

I

35:65 THF-Buffer

,5050 MeCN-Buffer

6040 MeOH-Buffer

ELUENT

Fig. 3.16. Variation of retention indices of column tests compounds on the alkyl aryl ketone scale with composition of ternary eluent. (1) Toluene, (2) nitrobenzene, (3) N-methylaniline, (4) p-cresol, (5) 2-phenylethanol, (6) acetophenone ([defined as 800). Reproduced with permission [165].

modifiers in the eluent but few studies have been examined the influence of the modifier or carried out systematic changes in the eluent composition. In the early work on retention index scales, Baker and Ma [lo61 and Smith [83,109] described the variation of retention indices across wide composition ranges. A more detailed comparison was also presented as part of the retention prediction studies in Chapter 1. Some work has been carried out during robustness studies to determine the susceptibility of the retention indices to small changes in composition and these are described in the next chapter. When Smith [139] examine the effect of eluents and stationary phases on the retention indices of the column test compounds, he found that the different solvents were a much more significant cause of differences in the indices than were differences in the stationary phases (Fig. 3.14). This was followed by a systematic study of the changes in retention indices of the column test compounds in isoeluotropic ternary eluents [164]. The changes between the vertices of the selectivity triangle were often not linear (Fig. 3.16). Linear interpolation between binary eluents may therefore give poor guidance on the optimal conditions to resolve two analytes. This problems has been recognized previously and led to the use of the iterative approach or mapping methods as techniques for eluent optimization [166-1681. A subsequent systematic study of the effect of the eluent on retention indices was carried out by West and co-workers [ 169-1 7 11 using retention indices based on the alkan-2one scale to examine the solvent strength of mobile phase solvents (see Chapter 9). By using the retention indices of three marker compounds, nitrobenzene, benzaldehyde and

Retention index scales used in high-performance liquid chromatography

135

toluene, it is possible to compensate for changes in the properties of a wide range of eluents.

3.4.4 Lipophilicity and biological activity

There has been considerable interest in the application of HPLC to measure the physicochemical properties of analytes. Most of this work has concentrated on the determination of lipophilicity, frequently expressed as octanol-water partition coefficients (log P) values because of their importance as indicators of biological function. The relationship between retention factors and log P values has been reviewed in recent years by a number of authors [ 1,172-1 731. A number of the studies have specifically concentrated on the relationship of retention indices and lipophilicity. Part of this interest has also examined the effect of changes in retention indices resulting from changes in the structure of the analytes by the addition or removal of functional groups. These studies are closely related to quantitative retention relationships and retention prediction and have been described in more detail in Chapter 1. Much of the work on lipophilicity and retention indices was initiated by the work of Baker and almost all the studies are based on the alkan-2-one standards. For example, as part of a QSAR study of 4-hydroxyquinoline-3-carboxylicacids as inhibitors of cell respiration [174], the retention indices on the alkan-2-one scale were used as a measure of lipophilicity, which was found to be closely correlated with biological activity (r = 0.92). In their early work, Baker et al. [ 1751 found a good correlation between retention indices and the anti-arrhythmic and inotropic activity of propanolol analogues, the hypnotic activity of barbiturates and the anti-inflammatory activity of anthranilic acid analogues. With the exception of the last example, the correlation of the biological activity with the retention indices was closer than with log P values. The indices were also used by Riley and Bagley to measure log P values in a study of the confirmational aspects of the analgesic activity of 4-anilidopiperidines[ 1761. The biological activity and retention index increments for substituents were found to be related for a number of antihypertensive quinazoline derivatives [ 1771. The correlation was closer with retention index increments than with Hansch partition increments. In a separate study, retention indices of a number of piperidino-quinazolinesulphonamidederivatives correlated well with their calculated partition coefficients [ 1781. In similar work, Schultz and Moulton used retention indices on the alkan-2-one scale to calculate log Koctmol-water values for nitrogen containing aromatic molecules [ 1791. Schultz and Applehans used retention indices calculated by linear interpolation between alkan-Zones [ 1071. These values were then used in a linear regression against known octanol-water partition coefficients (log Kow) to calculate log KO, values as a measurement of hydrophobicity, which were correlated with the population growth inhibition of Tetruhyrnenu by selected complex amino and nitro containing aromatic compounds. Shalaby and co-workers found close relationships between log P values and retention indices on the alkan-2-one scale for pyridipyrimidines and related nitrogen bridged ring compounds [ 1801 and for a series of biologically active quinazolines [ 1811. In a recent References pp. 14&144

136

Chapter 3

study, Wieringa et al. [ 1821 have used the retention indices of monoamine inhibiting 142[(phenoxyphenyl)methoxyl]ethyl]piperazines, based on the alkan-2-ones, as a guide to polarity and found a good correlation between 1/1and biological activity (Fig. 3.17). The retention indices measured on the alkan-2-one scale were used by Nieves and co-workers to compare the hydrophobicities of two series of 1,2,6-thiadiazone-l,1-dioxides and of related pyrazolones [ 1831. They noted some variations in I values with elution conditions but obtained reasonable good correlation (r = 0.97) between the two sets of compounds with methanol-buffer (60:40) as the eluent. Some work has been carried out using other sets of retention indices standards. DiazMarot and co-workers determined retention indices of 30 nitrogen mustard derivatives of steroidal lactams using the alkanal DNPH scale and found that they were largely unaffected by the mobile phase composition [184]. They showed a close relationship to log P values and correlated well with the log P values for monosubstituted aromatic compounds measured in the same system. Despite these successes, Brent et al. [1851 suggested that this use of retention indices (and of HPLC methods in general) to determine lipophilicity was unsatisfactory because the order of elution and hence of retention indices can change with the mobile phase

1

1.60 1.50

I

1

1.40

1

1.30

I-

1.20

I

1 .oo

0.70 0.60 0.50 0.40

;J

GBR12909 b

,

1

5

r-25

28 .

L _I_-

6

7

a

9

-_ 10

1000*DUP/RI Fig. 3.17. Correlation between 1000 X DUP (dopamine re-uptake inhibition)/retention indices of 1-[2[@henoxyphenyl)methoxy]ethyl]-piperazines and Mog LAD (lowest active dose) @noVkg) in ipsiversive circling test as a demonstration that retention indices are a guide to lipophilicity. 1 /log LAD (Crrnolkg)= (0.282 x DUP x lOOO/RI) - 1.030,N = 15, R = 0.96. Reproduced with permission [182].

Retention index scales used in high-performanceliquid chromatography

137

composition and in particular drew attention to examples in the initial work by Baker and Ma in which changes in the proportion of methanol changed the order of elution [ 1061. However, subsequent workers have frequently accepted HPLC as a method for determining log P values for pharmaceutical [ 1721 and environmental samples [ 1861. However, the often close relationship between log P and retention indices is usually only true for related compounds. When Baker [ 1871 compared the retention indices of barbiturates, propanolol analogues and anthranilic acid derivatives with their log P values, each group of compounds gave a linear but parallel relationship (Fig. 3.18). Despite these relationships, Maris et al. [ 1881 reported that there was no correlation between retention indices of a number of drug compounds measured on an ODS column and their separation on a chiral coiumn.

3.4.5 Structure-retentionrelationships The close relationship between reversed-phase chromatography and partition coefficients has led to the comparison of functional groups increments and log P values [85]. This relationship has fkequentlybeen used for retention prediction and these aspects have been discussed in more detail in Chapter 1. The relationship between structure and retention indices for sulphur containing analytes and related compounds, based on the n-alkane scale, has formed the basis of a comprehensive study principally by M6ckel et al. [94,189,190]. They carried out a series of studies of the relationship between retention and solute molecular surface type and area [191-1941 in which the physical parameters were compared with the retention indices of the analytes on the n-alkane scale. As part of this work, they identified differences in the methylene increment slope of the relationship between carbon number and log k and expressed these as free energy increments for the different functional groups [ 19I]. For each set of homologous analytes, there was a close relationship between retention index and total surface areas. Paraffinic areas produced a greater retention than aromatic areas, and the alkene group -CH=CH2 had a larger retention than the ethyl group but the alkyne group C=CH gave a large negative contribution. The authors also provided an explanation for the differences in retention between polymethylated benzenes and the corresponding isomeric n-alkylbenzenes, as the addition of a methyl group directly to the benzene ring replaced a low retention aromatic surface area with a high retention methyl surface area. In a subsequent paper [192] they examined the dialkylchalcones RS,R, RSe,R and RTe,R and found that for the diethyl derivatives the retention indices depended directly on the number of chalcone atoms (Fig. 3.19), such that d l = 139.2 for -%, 161.8 for -Seand 206.8 for -Te-. These results could be correlated with the surface area effects. They also examined the proton flee systems, P4, As4, S9 and Ses. They found differences in the methylene increment depending on the presence of substituents on an alkyl chain used this to calculate quality factorsf, for the different functional groups, -H, -Br, -SH, -OH, -CN,-OR and -SR [193]. These could be converted into molar free energy increments of sorption; respectively, AG = 0, -173, -6, +2570, +2398, +919, and +389 Umol, compared to a change for a methyl group in an References pp. 140-144

138

Chapter 3

alkane of AG = -1207 J/mol. In the final paper in the series they examined the changes in the indices of typical analytes with the water content of the eluent [ 1941. A study of polysulphide ethers [ 1951 found that the retention indices of R-O-S,-O-R increased linearly with the number of sulphur atoms or with the number of carbons in the *

X

w

0

1300

-

1200

-

1100

-

lo00

-

900

-

800

-

z

0

700

t-

z

W

600

0

-

a 500-

400

-

300

-

200

-

100

-

1

-3

I

-2

I -I

I 0

I I

I

I

2

3

LOG P Fig. 3.18. Correlation between observed retention indices on the alkan-2-one scale and octanol-buffer partition barbiturates, (0)propanolol analogues, (A) anthranilic acid analogues. Reproduced with coefficients. (0) permission [187]. 0 1979 American Chemical Society.

Retention index scales used in high-performance liquid chromatography

chalcogen

139

atom no.

Fig. 3.19. Retention indices of Et2S,, EtzSe, and EtzTe, versus chalcogen chain length based on the n-alkane scale. Conditions: eluent, methanol on Radpak A column. For comparison, the retention indices of n-alkanes from pentane to dodecane are marked by crosses. Reproduced with permission [192].

R groups. They also compared the retention indices of a set of related compounds in which methylenes were replaced by oxygen and sulphur; (iPr = isopropy1)iPrCH2(CH2)lo-CH2-iPr, I = 1800; iPr-O-(CH2)lo-O-iPr, I = 780; iPr-S-(CH2)lo-S-iPr, I = 1070; iPr-CH2-Slo- CH2-iPr, I = 1670; iPr-O-Slo-O-iPr, I = 1460; iPr-S-Slo-SiPr, I = 1790. A firther paper examined the corresponding amines R2N-Sn-NR2 [196]. Other studies have examined the relationship between the structure and retention indices of 1,n-bis(a1kylthio)alkanes [ 1971 and 1,n-halo(alky1thio)alkanes)[1981. As noted earlier, Mockel and Freydoldt [ 1001 and Dimov [ I011 have examined the relationships between the structure of the isoalkanes expressed as structural features or connectivities and their retention indices on the n-alkane scale.

3.5 CONCLUSIONS A number of homologous series have been employed as retention index standards. AlReferences pp. 140-1 44

Chapter 3

140

though, the traditional n-alkanes have limited applicability in HPLC, alternative scales can be used which are more polar and contain a readily detected chromophore. The principal series are the alkyl aryl ketones, nitroalkanes and alkan-2-ones and these have been widely applied across a range of elution conditions and with a wide diversity of analytes, for identifications, in QSRR and QSAR studies and for retention prediction.

3.6 REFERENCES 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

29 30 31 32 33 34 35 36 37 38 39 40

T. Braumann, J. Chromatogr, Chromatogr. Rev., 373 (1986) 191. E.T. Tomlinson, J. Chromatogr, Chromatogr. Rev., I13 (1975) 1. R. Kaliszan, Crit. Rev. Anal. Chem., 16 (1986) 323. L.S. Ettre, Pure Appl. Chem., 65 (1993) 819. R.J. Smith, C.S. Nieass and M.S. Wainwright, J. Liq. Chromatogr.,9 (1986) 1387. H. Engelhardt, H. MUIler and B. Dreyer, Chromatographia, 19 (1984) 240. M.C. Hennion and R. Rosset, Chromatographia,25 (1988) 43. A. Alhedai, D.E. Martire and R.P.W. Scott, Analyst, 114 (1989) 869. R.A. Djerki and R.J. Laub, J. Liq. Chromatogr., 10 (1987) 1749. L.S. Ettre, J. Chromatogr., 198 (1980) 229. E.J. Kikta and A.E. Stange, J. Chromatogr., 138 (1977) 41. S.C. Slahck, J. Assoc. Off. Anal. Chem., 67 (1984) 497. M.W. Law, J. Assoc. Off.Anal. Chem.., 66 (1983) 1304. M. Verzele, L. Use and M. Van Kerrebroeck, J. Chromatogr.,289 (1984) 333. S. Yamauchi and H. Mori, J. Chromatogr., 515 (1990) 305. J.W. Dolan, LC-GC Mag. Int.. 6 (1993) 679. N.E. Skelly, S.W. Ban and A.P. Zelinko, J. Chromatogr., 535 (1990) 199. A.J.P. Martin, Biochem. Soc. Symp., 3 (1950) 4. E. Kovdts, Helv. Chim. Acta, 41 (1958) 1915. M. Kuchaf, E. Kraus and M. Jelinkovh, J. Chromatogr., 557 (1991) 399. H. Colin, G. Guiochon and P. Jandera, Chromatographia, 17 (1983) 83. J, Brodsky and K. Ballschmiter, Fresenius Z. Anal. Chem., 335 (1989) 817. S. Yamauchi, J. Chromatogr., 635 (1991) 61. J. Wilken, Fresenius Z. Anal. Chem., 320 (1985) 696. R.S. Shreedhara Murthy, L.J. Crane and C.E. Bronnimann, J. Chromatogr., 542 (1991) 205. K. Huovinen, R. Hiltunen and M. van Schantz, Acta Pharm. Fenn., 94 (1985) 99. G.G. Wallace, Chem. Br., 29 (1993) 967. V. Pacakovd and L. Feltl, Chromatographic Retention Indices. An Aid to Identification of Organic Compounds, Ellis Honvood, Chichester, 1992. H. Lamparczyk and A. Radecki, Chromatographia, 18 (1984) 615. H. Lamparczyk, Chromatographia,20 (1985) 283. E. Kovdts, Adv. Chromatogr., 1 (1965) 229. M. van den Do01 and P.D. Kratz, J. Chromatogr., 11 (1963) 463. G. Guiochon, Anal. Chem., 36 (1964) 661. L.S. Ettre, Anal. Chem., 36 (July) (1964) 31A. J.K. Haken, Adv. Chromatogr., 14 (1976) 367. M.V. Budahegyi, E.R. Lombosi, T.S. Lombosi, S.Y. M~SZ~UOS, Sz.Nyiredy, G. Tar& I. Tim& and J.M. Takdcs, J. Chromatogr. Chromatogr. Rev., 271 (1983) 213. G. Tarjdn, Sz. Nyiredy, M. GyOr, E.R. Lombosi, T.S. Lombosi, M.V.Budahegyi, S.Y. M6sz6ros and J.M. TakCs, J. Chromatogr.,472 (1989) 1. J.M. Takbs, J. Chromatogr. Sci., 29 (1991) 382. The Sadtler Standard Gas Chromatography Retention Index Library, Heyden and Son, London, 1986. ezGC Software, Analytical Innovations, Kettering, Ohio, 1993.

Rettwtion index scales used in high-performance liquid chromatography 41 42 43

44

4s 46 47 48 49

so

51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 13 74 75 76 77 78 79 80 81

82 83 84

141

K. Pfleger, H.H. Maurer and A. Weber (Eds.), Mass Spectral and GC Data of Drugs, Poisons and Their Metabolites, 2nd edition, VCH, Weinheim, 1992. E. Marozzi, V. Gambaro, E. Saligari, R. Marianil and G.F. Lodi, J. Anal. Toxicol., 6 (1982) 185. R.A. de Zeeuw, J.P. Franke, H.H. Maurer and K. Pfelger, Gas Chromatographic Retention Indices of ToxicologicallyRelevant Substances on Packed and Capillary Column with Dimethyl Silicone Stationary Phases, 3rd edition, VCH, Weinheim, 1992. R.A. de Zeeuw, J.P. Franke, G. Machata, M.R. MOller, R.K. MUller, A. Graefe, D. Tiess, K. Pfleger and M. Geldmacher van Mallinckrodt, Gas Chromatographic Retention Indices of Solvents and Other Volatile Substances for Use in Toxicological Analysis, VCH, Weinheim, 1992. C. bra-Tamayo, M.A. Rams and J.M.R. Chacon, J. Chromatogr., 374 (1986) 73. L. Rohrschneider, J. Chromatogr., 22 (1966) 6. L. Rohrschneider, Adv. Chromatogr., 4 (1967) 333. L. Rohrschneider,Anal. Chem., 45 (1973) 1241. L. Rohrschneider, LC-GC Int., 6 (1993) 553. W.O. McReynolds, J. Chromatogr. Sci., 8 (1970) 685. R.G. Ackman, J. Chromatogr. Sci., 10 (1972) 535. S.J. Hawkes, J. Chromatogr. Sci., 10 (1972) 536. V.G. Berezkin and V.N. Retunsky, J. Chromatogr.,292 (1984) 9. T.R. Schwartz, J.D. Petty and E.M. Kaiser, Anal. Chem., 55 (1983) 1839. F. Pacholec and C.F. Poole, Anal. Chem., 54 (1983) 1019. F. Pacholec and C.F. Poole, J. Chromatogr., 302 (1984) 289 A. Yasuhara, M. Morita and K. Fuwa, J. Chromatogr., 328 (1986) 35. A. Manninen, M.L. Kuitunen and L. Julin, J. Chromatogr., 394 (1987) 465. M.B. Evans and J.K. Haken, J. Chromatogr., 472 (1989) 93. R.M. Smith (Ed.), Supercritical Fluid Chromatography, RSC Chromatography Monographs, Royal Society of Chemistry, London, 1988. M.L. Lee and K.E. Markides (Eds.), Analytical Supercritical Fluid Chromatography and Extraction, Chromatography Conferences, Provo, 1990. B. Wenclawiak (Ed.), Analysis with Supercritical Fluids: Extraction and Chromatography, Springer, Berlin, 1992. B.W. Wright, H.T. Kalinoski and R.D. Smith, Anal. Chem., 57 (1985) 2823. B.A. Jones, T.J. Shaw and J. Clark, J. Microcolumn. Separ., 4 (1992) 215. M.M. Sanagi and R.M. Smith, Anal. Proc., 24 (1987) 304. R.M. Smith and M.M. Sanagi, J. Chromatogr., 505 (1990) 147. R.M. Smith, S. Cocks, M.M. Sanagi, D.A. Briggs and V.G. Evans, Analyst, 116 (1991) 1287. R.M. Smith and D.A. Briggs, J. Chromatogr., submitted. R.M. Smith and D.A. Briggs, in preparation. R.M. Smith and S. Cocks, Analyst, 119 (1994) 921. M.M. Sanagi and R.M. Smith, Anal. Proc., 26 (1989) 22. R.M. Smith and M.M. Sanagi, Chromatographia,26 (1988) 77. R.M. Smith and M.M. Sanagi, J. Pharm. Biomed. Anal., 6 (1988) 837. R.M. Smith and M.M. Sanagi, J. Chromatogr.,481 (1989) 63. R.M. Smith and M.M. Sanagi, J. Chromatogr., 483 (1989) 51. J.C. Young and D.E. Games, J. Chromatogr., 627 (1992) 247. E.J. Kikta and E. Grushka, Anal. Chem., 48 (1976) 1098. E. Grushka and E.J. Kikta, Anal. Chem., 46 (1974) 1370. A. Kunugi and K. Tabei, 3. High Resolut. Chromatogr., Chromatogr. Commun., 11 (1988) 600. W.A. Bruggeman, J. van der Steen and 0.Hutzinger, J. Chromatogr.,238 (1982) 335. R.A. de Zeeuw, J.P. Franke, F. Degel, G. Machbert, H. Schutz, and J. Wijsbeek, Thin-Layer Chromatographic Rf Values of Toxicologically Relevant Substances on Standardized Systems, VCH, Weinheim, 1992. R.M. Smith, Adv. Chromatogr., 26 (1987) 277. R.M. Smith, J. Chromatogr., 236 (1982) 313. M.J.M. Wells and C.R. Clark, J. Chromatogr., 235 (1982) 31.

142

Chapter 3

85 86 87 88 89 90 91 92 93

R.M. Smith, J. Chromatogr. A, 656 (1993) 381. H.J. Mbckel and B. Masloch, Fresenius Z. Anal. Chem., 290 (1978) 305. H. Figge, A. Deege, J. Kbhler and G. Schomburg, J. Chromatogr., 351 (1986) 393. P. Varughese, M.E. Gangoda and R.K. Gilpin, J. Chromatogr., 499 (1990) 469. R.M. Smith, J. Chromatogr., 209 (1981) 1. R.M. Smith, J. Chromatogr., 237 (1982) 144. A. Tchapla, H. Colin and G. Guiochon, Anal. Chem., 56 (1984) 621. A. Tchapla, S. Heron, H. Colin and G. Guiochon, Anal. Chem., 60 (1988) 1443. B. Jousseaume, E. Chanson, M. Bevilacqua, A. Saux, M. Pereyre, B. Barbe and M. Petraud, J. Organomet. Chem., 294 (1985) C41. H.J. Mbckel, J. Chromatogr., 317 (1984) 589. A. Opperhuizen, T.L. Sinnige, J.M.D. van der Steen and 0.Hutzinger, J. Chromatogr., 388 (1991) 51. E.S. Ahuja and J.P. Foley, Analyst, 119 (1994) 353. P.G.H.M. Muijselaar, H.A. Claessens and C.A. Cramers, Anal. Chem., 66 (1994) 635. R.M. Smith and C.M. Burr, J. Chromatogr., 475 (1989) 57. B. Vonbach and G. Schomburg, J. Chromatogr., 149 (1978) 417. H.J. Mbckel and T. Freyholdt, Fresenius Z. Anal. Chem., 368 (1981) 401. N. Dimov, Anal. Chim. Acta, 201 (1987) 217. F. Morishita, H. Kakihana and T. Kojima, Anal. Lett., 17A (1984) 2385. P. Jandera, J. Chromatogr., 352 (1986) 91. P. Jandera, Chromatographia, 19 (1984) 101. J.K. Baker, R.E. Skelton and C-Y. Ma, J. Chromatogr., 168 (1979) 417. J.K. Baker and C-Y. Ma, J. Chromatogr., 169 (1979) 107. T.W. Schultz and F.M. Applehans, Ecotoxic. Environ. Safety, 10 (1985) 75. P. Kuronen, in E. Enqvist and A. Manninen (Eds.), Systematic Identification of Chemical Warfare Agents, B. 3. Identification of Non-Phosphorus Warfare Agents, Ministry of Foreign Affairs, Finland, (1982) pp. 43-61. R.M. Smith, G.A. Murillaand C.M. Burr, J. Chromatogr., 388 (1987) 37. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, LC-GC Mag., 4 (1986) 314. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, Anal. Proc., 22 (1985) 331. M. Bogusz and R. Aderja!, J. Chromatogr., 435 (1988) 43. M. Bogusz, J.P. Franke, R.A. de Zeeuw and M. Erkens, Fresenius J. Anal. Chem., 347 (1993) 73. R. Aderjan and M. Bogusz, J. Chromatogr., 454 (1988) 345. J.P. Franke, R.A. de Zeeuw and M. Bogusz, Fresenius J. Anal. Chem., 347 (1993) 67. M. Popl, V. Dolansky and J. Mosteckf, J. Chromatogr., 91 (1974) 649. M. Popl, V. Dolansky and J. Mostecky, J. Chromatogr., 117 (1976) 117. M. Popl, V. Dolansky and J. F&nrich, J. Chromatogr., 148 (1978) 195. F. Smejkal, M. Popl, A. Cihovil and M. ZhworkovA, J. Chromatogr., 197 (1980) 147. D. Karlesky, D.C. Kelly and I M. Warner, J. Liq. Chromatogr., 6 (1983) 471. C.N. Ho, D.L. Karlesky, J.R. Kennedy and I.M. Warner, J. Liq. Chromatogr., 9 (1986) 1. J.S. Thomson and J.W. Reynolds, Anal. Chem., 56 (1984) 2434. T.Y. Liu and A. Robbat, J. Chromatogr., 539 (1991) 1. P.E. Antle, A.P. Goldberg and L.R. Snyder, J. Chromatog., 321 (1985) 1. Personal communication from G. P. Rozing and K. Wienand. R.D. Plattner, G.F. Spencer and R. Kleiman, J. Am. Oil Chem. SOC.,54 (1977) 5 11. N.A. Porter, R.A. Wolfand J.R. Nixon, Lipids, 14 (1979) 20. B.J. Compton and W.C. Purdy, Anal. Chim. Acta, 141 (1982) 405. K. Takahashi, T. Hirano, M. Egi and K. Zama, J. Assoc. Off. Anal. Chem., 62 (1985) 1489. 0. Podlaha and B. Tbregird, J. Chromatogr., 482 (1989) 215. M. Gassiot-Matas, A. Auguet-Sangril and I. Masana-Gonzillez, Afinidad, 39 (1982) 479. L. Margarit-Roig, A. Diaz-Marot and M. Gassiot-Matas, Afinidad, 43 (1986) 525. R. Kostiainen and R. Kuronen, J. Chromatogr., 543 (1991) 39. P. Kuronen, Arch. Environ. Contam. Toxicol, 18 (1989) 336. R.M. Smith andN. Finn, J. Chromatogr., 537 (1991) 51.

94 95 96 97 98 99 100 101 102 103 104 105 106 107 108

109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135

Retention index scales used in high-performance liquid chromatography 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167

143

R.M. Smith and C.M. Burr, J. Chromatogr., 475 (1989) 75. M. Bogusz, G. Neidl-Fischer and R. Aderjan, J. Anal. Toxicol., 12 (1988) 325. R.M. Smith, R. Gill and M.D. Osselton, J. Anal. Toxicol., 13 (1989) 375. R.M. Smith, Anal. Chem., 56 (1984) 256. R.M. Smith, T.G. Hurdley, R.Gill and A.C. Moffat, Chromatographia, 19 (1984) 401. J.K. Baker, L.A. Cates, M.D. Corbett, J.W. Huber and D.L. Lattin, J. Liq. Chromatogr., 5 (1982) 829. R. Gill, A.C. Moffat, R.M. Smith and T.G. Hurdley, J. Chromatogr. Sci., 24 (1986) 153. G.E. Berendsen , P.J. Schoenmakers, L. de Galan, G. Vigh, 2. Varga-Puchony and J. Incddy, J. Liq. Chromatogr., 3 (1980) 1669. R.M. Smith and D.R. Garside, J. Chromatogr.,407 (1987) 19. M.S. Wainwright, C.S. Nieass, J.K. Haken and R.P. Chaplin, J. Chromatogr., 321 (1985) 287. R.J. Laub and S.J. Madden, J. Liq. Chromatogr., 8 (1985) 173. P.J.M. van Tulder, J.P. Franke and R.A. de Zeeuw, J. High Resolut. Chromatogr. Chromatogr. Commun., 10 (1987) 191. H.C. Furr, J. Chromatogr. Sci., 27 (1989) 216. H. Wsltzig and S. Ebel, Chromatographia,31 (1991) 544. H.J. Mbckel, U. Dreyer and H. Melzer, Fresenius J. Anal. Chem., 342 (1992) 673. U. Dreyer, H. Melzer and H.J. Mbckel, J. Chromatogr., 592 (1992) 13. R.M. Smith, J. Chromatogr., 236 (1982) 321. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr.,351 (1986) 259. R.M. Smith and S.L. Miller, J. Chromatogr., 464 (1989) 297. J.H. Park, M.D. Jang and S.M. Kwon, Bull. Korean Chem. Soc., 13 (1992) 75. R.M. Smith, J. Chromatogr., 291 (1984) 372. B. Gawdzik, J. Gawdzik and U. Czenvidska-Bil, Chromatographia,26 (1988) 399. B. Gawdzik, J. Chromatogr., 549 (1991) 77. B. Gawdzik, Chromatographia,35 (1993) 548. J.J. Pesek and S.A. Swedberg, J. Chromatogr., 361 (1986) 83. L.C. Sander and S.A. Wise, Adv. Chromatogr., 25 (1986) 139. L.C. Sander and S.A. Wise, LC-GC Mag., 9 (1990) 378. L.C. Sander and S.A. Wise, J. Chromatogr., 656 (1993) 335. J. Chmielowiec and A.E. George, Anal. Chem., 52 (1980) 1154. R.M. Smith, J. Chromatogr., 324 (1985) 243. P.J. Schoenmakers, Optimization of Chromatographic Selectivity: a Guide to Method Development, Journal of ChromatographyLibrary, Vol. 35, Elsevier, Amsterdam, 1986. J.C. Berridge, Techniques for the Automated Optimization of HPLC Separations, Wiley, Chichester, 1985.

175 176 177 178 179 180 181

J.L. Glajch and L.R. Snyder (Eds.), Computer-Assisted Method Development for High-Performance Liquid Chromatography, Elsevier, Amsterdam, 1990 (Reprint of J. Chromatogr., 485 (1989) 1-672). S.D. West, J. Chromatogr. Sci., 25 (1987) 122. S.D. West, J. Chromatogr. Sci., 27 (1989) 2. S.D. West and D.H. Mowrey, J. Chromatogr. Sci., 29 (1991) 497. H. Terada, Quant. Struct.-Act. Relat., 5 (1986) 81. T. Hanai, J. Chromatogr., 550 (1991) 313. E.A. Coats, K.J. Shah, S.R. Milstein, C.S. Genther, D.M. Nene, J. Roesener, J. Schmidt, M. Pleiss, E. Wagner and J.K. Baker, J. Med. Chem., 25 (1982) 57. J.K. Baker, D.O. Rauls, and R.F. Borne, J. Med. Chem., 22 (1979) 1301. T.N. Riley and J.R. Bagley, J. Med. Chem., 22 (1979) 1167. T. Sekiya, S. Imada, S. Hata and S. Yamada, Chem. Pharm. Bull., 31 (1983) 2779. T. Tanaka, E. Yamada, T. Sone and H. Hidaka, Biochem., 22 (1983) 1030. T.W. Schultz and B.A. Moulton, Environ. Toxicol. Chem., 4 (1985).353. A. Shalaby, Zs. Budvfui-Brby and Gy. Szhsz, J. Liq Chromatogr., 7 (1984) 1133. A. Shalaby, 2s. Budvhri-Bfhy, K. Hankb-Novak and Gy. Szhsz, J. Liq. Chromatogr., 7 (1984)

182

J. Wieringa, A. van den Meerendonk, J. Steenvoorden, G. Heeres, F. van Bakel, T. Roeters, R. van der

168 169 170 171 172 173 174

2493.

144

Chapter 3 Hulst, A. den Blanken, D. Leysen, Th. de Boer, H. Rijk and R. Plate, Rec. Trav. Chim. Pays-Bas, 112 (1993) 143.

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198

R. Nieves, C. Ochoa and M. Stud, J. Liq. Chromatogr., 11 (1988) 2349. A. Diaz-Marot, M. Gassiot Matas, J.J. Bonet-Sugrafles and Y. Dalmases, Afinidad, 47 (1990) 243. D.A. Brent, J.J. Sabatka, D.J. Minick and D.W. Henry, J. Med. Chem., 26 (1983) 1014. W. Klein, W. Kbrdel, M. Weiss, and H.J. Poremski, Chemosphere, 17 (1988) 361. J.K. Baker, Anal. Chem., 5 1 (1979) 1693. F.A. Maris, R.J.M. Vervoort and H. Hindriks, J. Chromatogr., 547 (1991) 45. J. Mdckel, T. Freyholdt, J. Weiss and 1. Molniu; in I. Molnllr (Ed.), Practical Aspects of Modem High Performance Liquid Chromatography, De Gruyter, Berlin, 1983, pp. 161-185. H.J. Mbckel, Adv. Chromatogr.,26 (1987) 1. H.J. Mbckel, G. Welter and H. Melzer, J. Chromatogr., 388 (1987) 255. H.J. Mbckel, F. Hbfler and H. Melzer, J. Chromatogr., 388 (1987) 267. H.J. Mbckel, F. Hdfler and H. Melzer, J. Chromatogr., 388 (1987) 275. H.J. Mbckel, F. Hbfler and H. Melzer, J. Chromatogr., 388 (1987) 285 E. Singer and H.J. Mbckel, Chromatographia,27 (1989) 27. U.Marquardt, H.J. Mdckel and S.F. Nelsen, Chromatographia, 27 (1989) 113. G. Aced, E. Darnstadt and E. Anklam, Fresenius Z. Anal. Chem., 331 (1988) 740. G. Aced, E. A n k h and H.J. Mbckel, Chromatographia,25 (1988) 1067.

Journal of Chromatography Library, Vol. 57: Retention and Selectivity in Liquid Chromatography R.M. Smith, editor 0 1995 Elsevier Science B. V. All rights reserved

145

CHAPTER 4

Application of retention indicesfor identification in high-performance liquid chromatography Roger M. Smith Department of Chemistry, Loughborough Universiv of Technology, Loughborough, Leicestershire, LEI1 3TU,UK

4.1 INTRODUCTION The principal application of retention indices in both gas and liquid chromatography has been for the identification of analytes. By providing a retention value, which is largely independent of the exact operating conditions and which can be reproduced in different laboratories, retention indices enable results to be transferred between laboratories and for identifications to be assigned from library values, even in the absence of authentic samples. Although widely applied in gas chromatography (as discussed in Chapter 3), the application of retention indices in high-performance liquid chromatography (HPLC) has been more limited. However, the indices of a wide range of different analytes have been reported using a range of different scales of standards. As well as studies specifically aimed at the potential of retention indices for identification, mainly in toxicological drug analysis and in the analysis of natural products, other studies have determined index values during studies of biological activity or structure-activity relationships. In addition, the retention indices of a wide range of analytes were reported during the retention prediction studies described in Chapter 1 and during studies of the properties of stationary and mobile phases discussed in Chapter 3 and elsewhere in this book. 4.2 ADVANTAGES AND PROBLEMS Despite many years of development, HPLC still suffers from a lack of long-term reproducibility, although the repeatability of retention times can be excellent. The transfer of methods between laboratories is often unreliable and hence the ability to identify analytes References pp. 167-1 69

146

Chapter 4

other than by direct comparison is limited. These problems stem from a number of sources. (a) The versatility and separation power of HPLC comes partly from the changes in retention and selectivity, which can be produced by manipulating the composition of the mobile phase. However, this sensitivity of the separation to the eluent composition also means that identical conditions will be required if reproducible results are to be achieved. Although this can be achieved within one laboratory, it can be difficult to control in interlaboratory studies. (b) Because of differences in the chemistry of even nominally equivalent bonded phases produced by different manufacturers and variations within the output fiom a single source, the retention properties of individual columns can differ. They also age during use, resulting in changes in their properties. (c) Few HPLC systems are truly standardized, flow rates can differ between pumps for electronic or mechanical reasons, such as leakages in check valves or piston seals, and particular problems occur with gradient systems in which the internal dimensions of the connecting tubing and mixing chambers can alter the gradient profiles reaching the column. Actual column temperatures can differ depending on the stability and heat transfer in an oven. Methods for the determination of the column hold-up volume are not defined and different techniques can give different values. As a consequence, absolute retention times are difficult to reproduce even within one laboratory on a single system. However, most HPLC separations are carried out by the direct comparison of an analyte with the retention of a standard sample on the same column and instrument or by comparison with an internal standard in a system calibrated using an authentic sample. As noted in Chapter 3, this use of retention factors (capacity factors) and relative retention factors to record retentions are both designed to reduce many of the variations between systems. In the previous chapter, retention indices were shown to provide a more robust method of recording retentions, because they are generally insensitive to small changes in eluent composition, temperature or minor column variations. When they introduced the concept of the alkan-2-one standards as the basis of a retention index, scale Baker and Ma [ 11 proposed that their main advantages would be for the characterization of drugs and other compounds. The indices would be largely independent of the percentage composition of the mobile phase (Fig. 3.7) and should be reproducible in different laboratories. However, as discussed in Section 3.4.1, they found significant variations between laboratories largely because of differences in the stationary phases. These differences can be used to characterize the stationary phases but mean that, for interlaboratory comparisons, close control of the stationary phase would be required. This problem can be partially solved by using corrected retention indices (see Chapter 5). The application of retention indices for the identification of analytes can be divided into three main areas plus isolated examples. (a) pharmaceutical and toxicological drug analysis; (b) natural products from plants, fungi and microorganisms; (c) environmental samples, including polynuclear aromatic hydrocarbons and pesticides.

Retention indices for identification in HPLC

147

In most cases, the aim is to identify a compound from within an expected group of analytes. Retention values on their own or even linked with diode array detection would usually be insufficient for complete identification and addition information such as mass spectrometry would be required. In any case, HPLC can potentially separate such a wide range of analytes, that without some additional information it would be difficult to know which set of retention index data to use for comparison.

4.3 PHARMACEUTICALS AND TOXICOLOGICALDRUG SAMPLES The phenomenal rise in popularity of HPLC in the late 1970s and early 1980s can be attributed to the introduction of reversed-phase chromatography and stable microparticular stationary phases, which made possible the routine analysis of relatively polar pharmaceuticals. These compounds frequently could not be analysed easily by gas chromatography because of their low volatility and quantitation in TLC was often inadequate for quality control methods. Most of the assays of drugs are carried out as part of purity checks, quality control, or the monitoring of stability testing. The determination is usually of a single major component and its known impurities and as a result, analyte identifications or method transferability are not important factors. Instead, the method is usually standardized by using a dedicated internal or external reference. In addition, system suitability checks are built into the assays to ensure high reproducibility. Clearly the potential application of retention indices in this area will be limited. However, there are two areas of importance in which drug related analytes need to be identified. In toxicological drug analysis, it is necessary to provide rapid and highly discriminating methods for the identification of drugs in body fluids or forensic samples. Products arising from metabolic, biological, or chemical transformations of a drug also frequently need to be identified. In this case, the starting material is known and any changes in the structure should cause a predictable change in retention. A number of studies of this type were described in Chapter 1, where changes had occurred on hydroxylation or glucuronide formation. Alternatively, by examining the changes in the retention of an analyte, it has also been possible to predict the associated structural changes. Related to these studies is the comparison of retention indices and biological activity, which was discussed in Chapter 3. Related studies on the relationship of Hansch n incremental constants to retentions were discussed in Chapter 1.

4.3.1 Toxicological drug analysis The practical problems of drug identification in analytical toxicology using HPLC were reviewed by de Zeeuw in 1983 [ 2 ] .The difficulties were primarily attributed to variations between stationary phases and differences between laboratories. He noted the success of the Kovhts index scale in gas-liquid chromatography and of standardized RF values in thin-layer chromatography and considered the need for comparable systems in highperformance liquid chromatography. However, at that time three factors still needed to be References pp. 167-169

148

Chapter 4

overcome, before HPLC could be widely used in systematic toxicological analysis; stationary phase variability, the lack of a universal HPLC detector, and the absence of an accepted method for expressing retentions in a standardized form. Although the first work to examine the potential of retention indices for the identification of drugs was carried out using the alkan-2-one standards, the majority of the studies have used the alkyl aryl ketones, in a series of studies by Smith and co-workers, or the 1nitroalkanes, which have been examined by Bogusz. 4.3.1.1 Studies based on the alkun-2-ones The initial interest in the retention indices came about as part of a study into the use of HPLC with a dual wavelength detector to generate a library of data for the identification of drugs [3].This led to the examination of the alkan-2-ones as retention index standards and the measurement of the reproducibility of selected drug samples [I]. Baker et al. [4]examined the potential of a retention index scale based on the alkan-2ones for interlaboratory comparisons but as discussed in Section 3.4.1,the interlaboratory results were much poorer than the intralaboratory results (Table 3.5). It appeared that considerable care would be needed to achieve good reproducibility. In particular, there were considerable differences between indices measured on different stationary phases. Retention index values, based on this scale, were employed to identify drugs administered for the management of arthritis at clinics in Mexico [5].The patients had been told that the tablets were vitamins or natural products but assays had suggested that the ingredients were frequently tranquillizers, anti-inflammatory steroids or non-steroidal antiinflammatory drugs. The retention index system provided a simple screening method for the identification of these drugs. The indices ranged from naproxen I = 304 to diazepam Z = 8 12 (Table 4.1).The precision was * 5 units, which together with the UV absorbance TABLE 4.1 RETENTION INDICES OF DRUGS USED IN THE MANAGEMENT OF ARTHRITIS Compounds

Retention index

Naproxen Phenylbutazone Fenoprofen Triamcinolone Prednisone Ibuprofen Mefenamic acid Indomethacin Prednisolone Dexamethasone Methylprednisolone Chlordiazepoxide Methylprednisolone acetate Diazepam

304 363 439 463 550 566 598 62 1 625 677 707 750 786 812

Data from Baker and Fifer IS]. Conditions: column, p-Bondapak c18;eluent, methanol-buffer (70:30). I values measured using the alkan-2-ones.

Retention indices for identijkation in HPLC

149

ratios at 254 and 280 nm, enabled all 14 drugs of interest to be positively discriminated. Although the retention indices of a number of drugs were reported in later studies from this laboratory, the work was primarily aimed at investigating the relationship between hydrophobicity and biological activity (Chapter 3) or to predict the retention indices of metabolites (see Chapter 1). A recent study by Vervoort and co-workers [6] compared the retention indices of 32 basic drugs on different stationary phases using the alkan-2-one scale. The drugs differed widely in structure and pK, but not all were identified. The influence of the mobile phase on peak shape and column stability was studied. 4.3.1.2 Studies bused on the ulkyl aryl ketones

Smith and co-workers have carried out a number of studies to examine the application of retention indices for the identification of unknown drug samples. The principal interest was to compare the robustness of the identification and to determine if retention indices provided better discrimination for unknown analytes than alternative methods, such as internal standards or relative retentions. The most comprehensive study examined the identification of the barbiturates [7-91. The separation was based on an existing method using an ODs-Hypersil column and a methanol-phosphate buffer (pH 8.5) (40:60). Particular care was taken to ensure a reproducible eluent composition and pH as this was required to achieve optimum resolution. The initial robustness studies examined the effect of changing the experimental conditions on the retention indices of ten barbiturates. A set of column test compounds (see Chapter 3) were also included to examine changes in the overall separation properties. In each case, the retention factors were compared with retention indices. Over a period of days at ambient temperature, the retention indices, which ranged from barbitone I = 579 to methohexitone I = 1001, were highly reproducible with a standard deviation (SD) of 1.22.8 [8]. The results could be correlated with retention indices on the alkan-2-one scale for the barbiturates measured in a quantitative structure-activity relationship study by Baker [lo]. Changing the proportion of methanol in the eluent from 30% to 50% had a marked effect on the retention factors (Fig. 4.la) but a smaller effect on the retention indices (Fig. 4. lb) and retention index values of the column test compounds were virtually unchanged. Increasing the temperature improved the efficiency of the separation. The retention factors of the barbiturates and alkyl aryl ketones showed a linear relationship with reciprocal temperature. The retention indices of the barbiturates decreased with increasing temperature (amylobarbitone from I = 891 at 10°C to I = 847 at 40°C, much more than the column test compounds (nitrobenzene, I = 832 to I = 827) over the same range. It was thought that the increased sensitivity of the barbiturates to temperature was probably caused by a change in the degree of ionization and later robustness studies were carried out at a controlled 30°C. In order to increase the discrimination of the separation, the aqueous phase was buffered to pH 8.5 which is close to the pK, of some of the drugs. Particular care had been taken to specify the composition of the buffer to ensure a consistent pH and ionic strength. If the pH was deliberately altered, even small changes were found to have marked effects on both the retention times and retention indices of the barbiturates (Fig. References pp. 167-1 69

Chapter 4

150

1

100

ao ti

60

c 0

9 r C

.-

I

.C

40

c a,

a

20

0 30

35

40

45

50

Methanol ( X I

1

b

1100

X

0

v

.g 900 .-c C

c

a,

a

aoo

700

!

I

I

I

I

30

35

40

45

50

Methanol 1%) Fig. 4.1, Effect of % of methanol as modifier on the separation of barbiturates and p-cresol as a column test compound, using methanol-buffer pH 8.5 as eluent. Retention values: (a) retention factors; (b) retention indices compared to alkyl aryl ketones. Compounds: 0,pentobarbitone; V, heptabarbitone; A, methohexitone; 0 , p-cresol. Values from [8].

4.2). Although the ionization of the barbiturates was being altered, the retention indices of neutral test compounds were unaffected. The second stage in the robustness study examined the effect of the stationary phase [9]. Columns prepared from the same batch of ODs-Hypersil gave effectively identical results (retention indices, SD 2.0-3.6). Four different batches of ODS-Hypersil gave slightly more variable results (SD 3.3-4.7) for the barbiturates. In both cases, the indices for the neutral column test compounds, except toluene, were more reproducible and for all the samples, the retention indices were more reproducible than the retention factors. The separations were then compared on six different ODs-bonded phases ranging from

Retention indices for identification in HPLC 1100

-- 1000

151

1

I

X 0

.-

-0 c

g 900

.-c c

c

al

a

000

700 7.5

0.0

8.5

9.0

9.5

10.0

PH

Fig. 4.2. Effect of eluent pH on retention indices of barbiturates and test compounds . Eluent methanol-buffer (40:60). Compounds as Fig. 4.1. Data from [S].

ODs-Partisil with a low carbon loading to more highly coated materials. In this study, a much greater variation in the retention indices was found for both the barbiturates and column test compounds (Table 4.2). The SD of the retention indices for the barbiturates increased to 41-59 units. From the robustness study, it was clear that although retention indices could compensate for some of the variations in the method, the brand of stationary phase, the pH of the eluent and the temperature of the column were critical factors. To test these conclusions, the method was tested in a collaborative study involving 10 laboratories in the UK, each of whom was provided with column material from the same batch of ODs-Hypersil and a closely defined method for the preparation of the buffer solution from fixed weights of solid components [ 1 11. A number of different methods were compared for recording and analysing the retention data. The coefficient of variation (CV) for the retention times of “~nknown’~ samples (13.3% for barbitone to 19.5% for methohexitone) and the retention factors (16.5% to 13.9%, respectively) were high. In order to be able to compare the different techniques as they were all measured using different scales, the results were all converted to discrimination numbers (Table 4.3), which represents the maximum number of analytes in a given retention range that can be positively distinguished on the basis of the uncertainty in the reporting methods. As expected, direct measurement and retention factors were poor at discrimination. However, relative retention times, relative adjusted retentions times and corrected retention factors compared to barbiturate standards were all more discriminatory than retention indices relative to the alkyl aryl ketones. It was concluded that differences in the ionization of the barbiturates in different laboratories, due to temperature differences (which were often not controlled) or in the preparation and hence pH of the buffer, were probably the major source of variation. As a result internal standards, such as other barbiturates which would also be partially ionized, References pp. 167-1 69

Chapter 4

152

TABLE 4.2 RETENTION INDICES OF BARBITURATES AND COLUMN TEST COMPOUNDS ON DIFFERENT COLUMN MATERIALS Compound

Retention index (I) H3

H5

T

S

Z

P

Barbiturates Barbitone Phenobarbitone Cyclobarbitone Butobarbitone Talbutal Heptabarbitone Amylobarbitone Pentobarbitone Quinalbarbitone Methohexitone

5 89 662 752 78 1 826 829 865 882 92 1 994

587 660 75 1 780 823 827 864 879 918 990

545 614 705 735 778 780 817 835 874 950

511 567 667 704 747 745 786 807 844 92 1

492 555 653 685 73 1 73 1 769 790 829 917

483 517 631 675 720 708 759 789 82 1 895

Column test compounds 2-Phenylethanol Nitrobenzene p-Cresol Toluene N-Methylaniline

776 828 796 989 790

778 828 798 989 790

753 838 769 985 788

736 823 745 955 767

719 818 726 976 765

75 1 823 752 936 782

Data from Smith and co-workers [9]. Indices based on the alkyl aryl ketones. Conditions: eluent, methanolphosphate buffer pH 8.5 (40:60). Columns: H3, 3 pm ODs-Hypersil, H5,5 pm-ODS Hypersil; T, Techsil 5 C ~ RS,; Spherisorb ODs; Z, Zorbax ODs; P, Partisil 10 ODs.

TABLE 4.3 DISCRIMINATION NUMBERS OF ALTERNATIVE METHODS FOR REPORTING THE RETENTION OF BARBITURATES IN A COLLABORATIVE TRIAL Techniquesa

Range

DiscriminationNumber

Retention times (s) Retention factors Retention indices Relative retention times Relative adjusted retention times Corrected retention factors

67-869 1-25 625-869 0.14-1.84 0.08-19.2 1-25

10 16 34 44 55 64

Data from Gill and co-workers [l 11. The number represented the maximum number of analytes which could be unambiguously identified on the basis of the uncertainty in their retention values over the arbitrary range k = 125. aRetention indices are based on the alkyl aryl ketones. Adjusted retention times were calculated relative to quinalbarbitone. Corrected retention times were calculated relative to phenobarbitone, butobarbitone, amylobarbitone and quinalbarbitone as standards

Retention indicesfor identipcation in HPLC

153

would be more effective for this assay at providing robust relative retention measurements, than comparison with a set of neutral alkyl aryl ketones. A second study examined the separation of a group of local anaesthetics on an ODS silica column using methanol-water-1 .O% (v/v) aqueous orthophosphoric acid-n-hexylm i n e (30:70: 100: 1.4, v/v) as eluent [ 12,131. Again the retention indices showed much smaller changes than the retention factors on varying the temperature and proportion of methanol. The effect of altering the acidity of the eluent from the original value of pH 2.5, had little effect on most of the analytes but the retention of benzocaine dropped from I = 829 at pH 3.0 to I = 812 at pH 2.5 and then to I = 741 at pH 2.0, as expected from its pK, of 2.5. The retention times of all the compounds decreased noticeably as the proportion of hexylamine was increased from 0.0% and 0.35% but with higher levels the retention indices (Table 4.4) stabilized, although the retention factors still decreased. This suggested that above a threshold level, all the uncapped silanols were effectively completely masked by the hexylamine. At the standard eluent pH of 2.5, benzocaine was unaffected by the presence of the hexylamine and behaved as a neutral analyte. The retention indices of the neutral column test compounds were largely unaffected by most of the tests indicating that the changes were primarily in the ionic interactions. The most marked effect on the separation was observed by changing the brand of stationary phase. The retention indices, relative retention factors, and even the order of elution altered markedly (Table 4.4). The change in selectivity between the columns was also reflected by marked changes in the indices of the column test compounds.

TABLE 4.4 EFFECT OF PROPORTION OF HEXYLAMINE AND DIFFERENT STATIONARY PHASES ON THE RETENTION INDICES OF LOCAL ANAESTHETIC DRUGS AND COLUMN TEST COMPOUNDS Compound

Retention index (2) Hexylamine %

Stationary phase

0.0

0.35

0.7a

1.4

Ha

P

Z

Chloroprocaine Lignocaine Cocaine Amylocaine Benzocaine

684 704 815 866 804

508 559 668 744 799

461 530 630 714 812

437 523 605 700 796

461 530 630 714 812

654 609 795 772 808

487 488 643 675 776

Column test compounds 2-Phenylethanol Nitrobenzene p-Cresol Toluene

762 78 1 771 900

765 795 789 921

769 794 796 92 1

772 807 811 944

769 794 796 921

116 771 716 850

678 754 692 88 1

Data from Smith and co-workers [13]. For the effect of hexylamine:column, ODS-Hypersil; eluent, methanolwater-1.0% aqueous othophosphoric acid-hexylamine (30:70:1OO:various). For stationary phase comparison: eluent contains 0.7% hexylamine. Columns: H, Hypersil-ODS; P, Partisil 10 ODs; Z, Zorbax-ODs. astandud conditions.

References pp. 167-1 69

154

Chapter 4

TABLE 4.5 RETENTION INDICES OF THIAZIDE DIURETICS AND RELATED DRUGS ON DIFFERENT ODS SILICA PACKING MATERIALS Compound

Retention index (I) Stationary phase HI

H2

Z

T

TS

L

N

527 555 632 730 754 855 873 885 913 918 929

529 559 636 74 1 766 870 889 900 927 93 1 944

451 486 574 680 706 805 824 836 869 877 877

506 535 623 727 754 852 868 879 915 918 92 1

506 532 616 715 738 836 853 865 897 904 908

510 542 622 724 749 850 878 878 908 912 92 1

51 1 540 626 732 756 856 882 882 918 92 1 923

Non-thiazide diuretic drugs Quinethazone 546 Chlorthalidone 616 Metalazone 785 Chlorexolone 834 Mefruside 862 Clopamide 835

539 625 796 848 873 815

473 543 732 783 82 1 822

529 600 776 82 1 856 702

521 589 764 810 847 826

529 603 776 825 854 774

528 603 779 823 858 679

Column test compounds 2-Phenylethanol 706 p-Cresol 771 Nitrobenzene 870 Toluene 1009

707 771 868 1001

690 739 862 1008

70 1 763 875 1003

698 762 870 1014

708 769 872 1009

699 764 875 1006

Thiazide diuretics Chlorothiazide Hydrochlorothiazide Hydroflumethiazide Trichloromethiazide Methyclothiazide Benzthiazide Cyclothiazide 1 Cyclothiazide 2 Polythiazide Bendrofluazide Cyclopenthiazide

Data from Smith and co-workers [ 141. Eluent, acetonitrile-1% aqueous acetic acid (35:65); temperature 30°C. Columns: HI and H2, different batches of ODS Hypersil; Z, ODS-Zorbax; T, Techsil 5 CIS; TS, Techsphere ODS; L, Lichrosorb RP-18; N, Nucleosil5 CIR.

To test if the retention indices would be equally applicable when acetonitrile was used as the modifier, the robustness of the separation of a group of thiazide and non-thiazide diuretics was examined [ 141. These compounds included a wide range of different structural types and were separated using acetonitrile-1% acetic acid (3550). The temperature and proportion of acetonitrile were varied but the changes had only a small effect on the retention indices, despite large changes in the retention factors. For example, cyclopenthiazide changed from k = 14.87, I =959 in 30% acetonitrile to k = 3.73, I =895 in 40% acetonitrile. Similar decreases in retention indices were observed with the proportion of modifier for all the other diuretics, except clopamide, which increased from I = 809 to 841. Unlike the other analytes, it was probably partially protonated and was also susceptible to changes in the eluent pH (I= 849 at pH 2.5 and I = 706 at pH 3.5) and changes in the proportion of acetic acid.

Retention indices for identification in HPLC

155

As with the previous studies the retention indices (and retention factors) differed on different brands of ODs-silica (Table 4.5) although reasonably consistent results were obtained on different batches of the same material (ODS-Hypersil, columns H1 and H2). The Zorbax-ODS column was noticeably different and this was also reflected in a low retention index value on this column for p-cresol, I = 739 (one of the column test compounds), compared to its index values on the other columns, I = 762-77 1. However, the order of the elution of the diuretics (except clopamide) was the same on each column and polythiazide could be used as an internal standard to correct for any variations. However, under the different eluent conditions investigated earlier, such as the proportion of methanol, it provided only a poor correction for many of the diuretics. Some of these results were subsequently re-analysed by Bogusz [ 151, who showed that it was possible to correct for differences in retention indices between the stationary phases by using selected analytes as marker compounds (Chapter 5). In further studies, he reported retention indices using the alkyl aryl ketone scale for acidic and neutral drugs [ 161. The alkyl aryl ketone and nitroalkane scales were compared for the identification of basic drugs [ 171, although Smith and co-workers [ 181 observed that effectively both corrected indices then became corrected relative retention values compared to the same standards. Retention correction worked well for the barbiturates, which have a quasihomologous relationship and differed only in the size and isomerization of alkyl side chains. In contrast, the local anaesthetics and the diuretics were structurally diverse and each would have different interactions with the column material. Under these circumstances, retention indices can improve reproducibility if the experimental conditions are TABLE 4.6 RETENTION INDICES OF BASIC DRUGS ON POLYMERIC COLUMNS AND THE EFFECT OF THE ELUENT PH Compound

Retention index (I) Column PS-DVB pH 8.9

Ephedrine Clonidine Lidocaine Benzocaine Procaine Propanolol Diazepam Strychnine Cyproheptadine Amitriptyline Promethazine

782 766 725 636 720

-

1013 1087 1297 1356 1296

DMN-DVB pH 12.5

pH 8.9

pH 12.5

435 556 718 619 66 1 84 1 1017 975 1237 1276 1238

997 993 788 784 1345 1062 1613 1426

410 62 1 653 739 784 90 1 1086 1158 1200 1241 1263

1412

Data from Gawdzik [19]. Eluent, methanol-aqueous sodium hydroxide (90:10). Columns: PS-DVB, Hamilton PRP-1; DMN-DVB, di(methacryloyloxymethyl)naphthalene-divinylbenzene copolymer. Retention indices measured on the alkyl aryl ketone scale.

Referencespp. 167-1 69

156

Chapter 4

closely controlled. However, it would not be possible to cany out proportional corrections for differences in the stationary phases because of the order of elution frequently changed under different conditions. In other studies, Gawdzik [ 191 compared the retention indices using the alkyl aryl ketone scale of a range of drugs on a polystyrene divinylbenzene (PS-DVB) and di(methacryloyloxymethy1)naphthalene-divinylbenzene copolymer (Dh4N-DVB). There were significant differences in the retention factors and retention indices on changing the pH of the eluent from 8.9 to 12.5 (Table 4.6), which were attributed to changes in the ionization of the analytes, which, except for benzocaine and diazepam, had pK, values between 8 and 10. The mechanism of the effect is complex and suggested that some ionic interactions were occurring between the drugs and the polymeric materials. Despite these studies on specific groups of drugs, the alkyl aryl ketone standards have not been widely adopted in this field. Hill and Langner [20] discussed the potential application of retention indices using the alkyl aryl ketone scale in combination with diode array spectra to create a database for the identification of drugs using gradient elution but there have been no reports of its implementation. A probable reason is that although the retentions of many drugs fell within the alkyl aryl ketone scale, extrapolation was required for the more polar compounds. Retention indices less than I = 800 might therefore be less reliable. However, in practice extrapolation to I = 600 appears to provide reproducible values as long as the retention factors are not small. For more rapidly eluted drugs (see Table 3.3) alternative aliphatic scales may be more desirable [21] (Chapter 3.3.8). 4.3.1.3Studies based on the l-nitroalkanes

One approach to the problem of polar drugs is to use small aliphatic retention index standards, which are rapidly eluted. In a comprehensive series of studies, Bogusz and coworkers have demonstrated that the 1 -nitroalkanes can be used for this role and their work is described in more detail in Chapter 5. They addressed the problems of differences between stationary phases by adopting corrected retention indices compared to standard drug samples. As part of this work they have reported a library of retention index data covering 225 substances, coupled with diode array spectroscopic values [22]. The separations were carried out using an acetonitrile-triethylammonium phosphate pH 3 .O buffer gradient and the drugs were compared with the retentions of nitromethane to 1nitrooctane standards. The effect of different stationary phases on the retention indices and their interlaboratory applicability have also been examined [23]. The application of this approach to toxicological drug analysis has recently been reviewed [24,25]. The nitroalkanes have also been adopted by Gill and co-workers in an unpublished method [26] for the screening of basic drugs. They used an aqueous sulphuric acidacetonitrile gradient so that detection could be carried out at a short wavelength. Particular care was taken to standardize the conditions of the earlier stages of the gradient to improve reproducibility. 4.3.2 Drug metabolites

Hufford and co-workers [27] used the relationship between retention indices and n constants (Eq. 1.7) to predict and identify the metabolites of imipramine on microbial degra-

Retention indicesfor identification in HPLC

157

dation. The experimental and calculated values matched closely: 1O-hydroxyimipramine, Zexp = 720 and Z cdc = 725; 2-hydroxyimipramine Zexp = 769 Zcarc = 789; and desimipramine, Zexp= 807 and Zcdc = 846. Baker and co-workers [28] have also used retention indices and detector response ratios to aid the identification of the urinary metabolites of primaquine.

4.4 NATURAL PRODUCTS

The Kovhts retention indices have been a useful tool in the gas chromatography of terpenes, because the wide range of analytes from natural sources often means that standard samples are not available. A similar problem also occurs in the HPLC of natural products. The use of relative measurements can provide the long term reproducibility necessary for libraries of retentions values to be established. In addition, the ability to predict the retention of homologues or substituted analogues can be valuable if the compounds in a sample are unknown or if authentic compounds cannot be obtained for direct comparison. Retention indices have been applied to both fungal and plant sources, usually to examine particular groups of analytes, but some studies have employed the indices as a method of screening. Both alkyl aryl ketones and alkan-2-ones have been used as retention index standards and some studies have developed additional specialized standards to match the polarities of the analytes. Relative retentions compared to homologous standards have also been employed in the analysis of lipids and fatty acids, as equivalent carbon numbers, and these were discussed in Chapter 3.

4.4.1 Fungal metabolites

Although the fungi are usually classified on their morphology, there has been considerable interest in chemotaxonomic approaches based on the presence or absence of marker compounds and the profile of the metabolites. This interest and the desire to find an easy method to identify h g a l toxins led Frisvad and Thane [29] to generate a comprehensive set of retention index and diode array spectroscopic data for 182 mycotoxins and other fungal metabolites. These were used to compare samples from Penicillium, Aspergillus and Fusarium species. The fungal extracts were separated using a gradient elution of 10 to 90% 0.05% trifluoroacetic acid in acetonitrile-water. Retention indices were calculated by linear interpolation between the retention times of alkyl aryl ketone retention index standards, which had been added to the sample. The indices were highly reproducible with standard deviations for most analytes of 1-3 units. The chromatograms could be used to compare the profile of metabolites from different species of fungi (Fig. 4.3). Even though the alkyl aryl ketones were eluted together with the sample, they could usually be distinguished by their diode array spectra (Fig. 4.4). This application has been discussed in a review lecture on the use of HPLC in fungal taxonomy [30]. Frisvad commented that large difference of up to 80 units were found between retention indices measured on different columns but only small differences on changing from 0.05 M to 0.005 M TFA in the eluent. References pp. 167-1 69

158

Chapter 4 LC A 2 5 4 . 5 10387 L C R 2 5 4 , s

s50,20

3

04 FUSMRIBA.~

550,20

I

o f

FUSMfl03R.

D

80-

z 40-

t

I

20U

u

r

a 0 bl

-20-

-6 -I0/ 0

-8 0-

8

9 I0

10 2 0 Time

111

12 30

(mtn

14

6 40

)

Fig. 4.3. HPLC separation of Fusarium culmorum CBS 417.86 (upper trace) and F. sporofrichioides AJ 10 (lower trace). The numbers and arrows on the lower axis mark the alkyl aryl ketone reference compounds and their carbon numbers. Groups of compounds: f, fusarins; z, zearalenones; t, trichothecenes. Gradient elution: A, water, B, TFA-acetonitrile. Reproduced with permission [29].

This type of approach has been valuable in the pharmaceutical industry as part of the screening of fungi for novel components, as previously isolated metabolites can be readily identified by their retention indices and diode array spectra. 4.4.1.1 Mycotoxins

There is considerable interest in the analysis of mycotoxins [3 I] and other fungal toxins because of their importance in human and animal food sources and their identification has been the subject of a number of studies using retention indices, in addition to the broader work of Frisvad and Thane [29]. In initial studies of mycotoxins, Kuronen employed the alkyl aryl ketones as retention index standards [32]but more recent work [33] has employed a novel series of retention index markers, the 1-[p-(2,3-dihydroxypropoxy)phenyl]-1-alkanones (D-compounds), in conjunction with diode array detection (Chapter 6 ) . These identifications were assisted by the use of LC-MS with thermospray and fast atom bombardment ionization methods [34]. Hill and co-workers [35] also used the alkyl aryl ketone scale from acetophenone to undecyl phenyl ketone, with a gradient elution fiom 20% to 100% acetonitrile, for the identification of mycotoxins. Because the relationship between carbon number and reten-

Retention indicesfor identijkation in HPLC

159

I00

80 70 U 0 c

a

60 50 40

30 20 10 0

----_ .

.

.

I

220

.

.

.

~

'

~

~

I

.

240 260 280 Wavelength (nm)

~

~

300

I

"

'

320

I

.

'

'

I

~

340

Fig. 4.4. Photodiode array spectra of (A) brefeldin A and (B) propiophenone during elution of metabolites as Fig. 4.3 to show ready recognition of internal standards. Reproduced with permission [29].

tion time was curved, they calculated the indices by a linear interpolation between the retention times of the alkyl aryl ketones, which bracketed the analyte (IB). For mycotoxins which eluted before acetophenone, they used a linear extrapolation from the relationship between acetophenone and propiophenone. Because of concern that the addition of the complete homologous series of standards to real samples might result in co-elution, they calculated normalized retention indices (IN) based on a linear interpolation between propiophenone and hexanophenone. They found that both the bracketed and normalized indices (Table 4.7) were highly reproducible on different columns of the same brand of stationary phase with RSD < 1.0%. The differences between the two scales was usually very small (IB- IN = 0-7). The values were subsequently compared with the indices (I,) (Table 4.7) of mycotoxins recorded by Frisvad and Thrane [29]. These values were consistently higher (by 45-82 units) but a different column, eluent and gradient profile had been used. 4.4.1.2 Otherfungal metabolites

Magg and Ballschmiter determined the retention indices of the 11 ergopeptines, which are peptide derivatives of lysergic acid, on three different stationary phase systems using the alkan-Zone scale [36]. Large differences in the retention indices were found, even for columns containing the same brand of stationary phases (for example, ergotamine, I = 606 and 7 1 8 on two Lichrosorb RP-18 columns of different dimensions and I = 66 1 on a Nucleosil CI8column). However, there was a good correlation between the increments for the addition of different alkyl substituents, when measured on the different columns. For example, the addition of methylene groups to ergotamine to give ergostine caused changes of AZ = 69,79 and 64 units, on the three columns, respectively. References pp. 167-1 69

.

'

I

160

Chapter 4

TABLE 4.7 RETENTION INDICES OF MYCOTOXINS SEPARATED BY GRADIENT HF'LC Compounds

Retention index

IB Aflatoxin M2 Aflatoxin M I Aflatoxin PI Aflatoxin G2 Aflatoxin B2 Aflatoxin GI Parisiticol Aflatoxin B1 Aflatoxicol I Dechlorogriseofulvin Zearalenol 0-Methylsterigmatocystin Roridin A Vermcarin A Rubratoxin B Zearalenone TetrahydrodeoxyaflatoxinB 1 Paxilline

697 727 729 764 786 792 796 826 84 1 873 899 92 1 937 958 1005 1024 1037 1228

IN

764 793 799 795 826 842 873

IF

838 868 865

967 922 93 8 959

1036 1233

1013 1022 1079 1069 1287

~~

Retention indices compared to alkyl aryl ketone standards. IB, bracketed retention indices and IN, normalized retention indices determined on a Zorbax Cs column with water-acetonitrile gradient from Hill and co-workers [35]. IF,on a Nucleosil CIScolumn using a water-TFA-acetonitrile gradient from Frisvad and Thrane [29].

Patterson and Kemmelmeier [37] have used the alkyl aryl ketone retention index scale for the identification of components in insecticidal extracts of Penicillium strains. Rather based on butyrophenone and than use standards in each sample, normalized indices (lN) dodecanophenone (Fig. 4.5) were determined by using normalized retention times in the same way as Hill and co-workers [35]. A comprehensive list of retention indices for over 60 metabolites was reported. The results for selected compounds were compared with those of Frisvad and Thrane [29] and a systematic difference of -1 12 25 units was found. This change was sufficiently constant to be applied as a correction factors, enabling the identification of additional compounds. In some of their samples, the peaks for the alkyl aryl ketone standards were reduced in intensity, suggesting that these standards may not be inert and might have combined with some of the metabolites.

*

4.4.2 Plant products

As with the fungal metabolites, there has been interest in the use of retention indices for the identification of plant constituents as part of taxonomic studies. The indices have also been used to examine commercially important constituents, such as pigments and flavour ingredients, mirroring the use of indices in GLC for the identification of odour components.

Retention indices for identification in HPLC

161

100

90

80

70

60

OO /

50

Y 40

2 6 7

30

20

10

li

Fig. 4.5. HPLC separation of extract of PeniciNiurn citrinin. Gradient elution; water to acetonitrile. X, butyrophenone and Y, dodecanophenone internal standards used to derive normalized retention indices on the alkyl aryl ketone scale as Table 4.7. Reproduced with permission [37].

4.4.2.1Spices andflavour components

Plant materials used in the food industry often contain complex mixtures of natural products whose identification and quantitation is required, as a control on commercial quality, as a guide to the proportion of active odour or flavour ingredients, or as a test of the presence of the material in a finished product. Although the volatile terpene constituents of ginger and related Zingiberaceae species, such as grains of paradise, can be examined by gas chromatography, the pungent principles are thermally unstable and degrade on injection. Smith [38,39] used reversed-phase chromatography coupled with ultraviolet spectroscopic and electrochemical detection to separate the constituents and proposed the use of retention indices based on the alkyl aryl ketone scale for their identification. The gingerols, shogaol and paradols, which are the main constituents, showed different retention indices in methanol-buffer (70:30) and acetonitrilebuffer (60:40)mobile phases (Table 4.8). Turmeric (Curcuma domestica Val. Syn. C . longa Koenig non L.) is a widely used member of the Zingiberaceae family. The mature rhizomes are ground to give an aromatic References pp. 167-1 69

162

Chapter 4

TABLE 4.8 RETENTION INDICES OF PUNGENT PRINCIPLES FROM ZINGIBERACEAE 0

OH

Glngerolo

HO n= 4,6,8

b4.i)

0

Compound

[6]-Gingerol [8]-Gingerol [101-Ginger01 [6]-Shogaol [6] -Parado1 Zingerone Citrals

Retention index Eluent MeOH-water (70:30)

MeCN-Buffer (60:40)

983 1169 1375 1162 1247 680 1032 1057

885 1072 1274 1127 121 1 652 1017 1037

Data from Smith [38]. Conditions:Column ODS-Hypersil.

yellow powder employed as a colouring in curry powder. The separation of the principal pigment constituents, curcumin and its demethoxy and bisdemethoxy derivatives was reported by Smith and Witowska [40]. The order of elution was dependent on the organic modifier in the mobile phase: with methanol-water (70:30), the three components were unresolved I = 1098; in acetonitrile-water (60:40), I =941, 920 and 898, respectively; but the elution was reversed with THF-water ( 4 5 5 9 , I =969,996 and 1022 based on the alkyl aryl ketone scale. A short study by Smith and Beck [41] demonstrated that eugenol (I=932) and methyl eugenol (I= 1031) could be separated using methanol-pH 7.0 buffer (75:25) in extracts fiom pimento berries (Pimenta dioica L), which are the source of the important spice Allspice. 4.4.2.2 Plant toxins

Ma and co-workers [42] used retention indices on the alkan-2-one scale to identify the urushiol congeners of poison ivy and poison oak. By using the relationship to the ~t constants described in Eq. 1.7, it was also possible to predict the retention indices of addi-

Retention indices for identijication in HPLC

163

TABLE 4.9 RETENTION INDICES OF POISON IVY AND POISON OAK CONGENERS Compound

PDC-(15:0) PDC-monoene (15 :1) PDC-diene (15:2) PDC-triene (15:3) PDC-diacetate HDC (17:O) HDC-monoene-( 17:1) HDC-diene (17:2) HDC-triene( 17:3) HDC-diacetate

Retention index (4 Observed

Calculated

1771 1620 1506 1416 1814 1969 1795 1668 1573 2026

Reference 1645 1518 1393 1823 1971 1845 1718 1593 2023

Data from Ma and co-workers [42]. Eluent; methanol-water (85:15) on pBondapak c18, Indices on alkan-2one scale. PDC, pentadecyl-catechol; HDC, 3-heptadecyl catechol. Calculated values determined by incremental changes for unsaturation (-126) and acetylation (+26)

tional unsaturated isomers (Table 4.9). This method had the advantage that unlike GLC, no derivatization was necessary and the retention indices of additional components could be estimated. 4.4.2.3 Gliadins

In early work it was suggested that the alkyl aryl ketone scale could be used for the identification of gliadins (ethanol-soluble proteins) as a means of characterizing varietal variations in wheat samples [43]. Subsequently, direct comparison of the chromatograms to gliadins standards was recommended by Sapirstein [44]. This method was felt to have advantages because the standards covered the range of samples and had similar retention properties.

4.4.3 Lichen constituents Culberson et al. [45,46] examined the retention indices of depsides fiom lichens as a method of identification. They found a systematic relationship between the retention index values, based on the alkan-2-one scale, and the number of side chain carbons for a wide range of alkylated orcinol depsides and phenolic units. These could be used to predict the retention of additional homologues and to identify them in herbarium specimens. They reported a high reproducibility for perlatolic acid between two Ultrasphere ODS columns, I =1525 f 4 and 1523 f 7 with a combined RSD = 0.37%, whereas the corresponding retention factors ranged fiom k=4.37 to 7.79 with a RSD of 31%. A subsequent study by Gowan reported the use of alkan-2-one indices to identify depsides as secondary products of Popridiaceae [47]. References pp. 167-1 69

164

Chapter 4

TABLE 4.10 RETENTION INDICES OF RETINOIDS, TOCOPHEROLS AND CAROTENODS Compound

Retention index (4

Retinol Lutein Zeaxanthin y-Tocopherol a-Tocopherol /?-Cryptoxanthin Lycopene a-Carotene /?-Carotene Retinyl palmitate

1664 1858 2006 2325 2410 2468 2563 2835 2885 2966

Separation on Resolve c18 column, eluent acetonitrile-dichloromethane-methanol-1901 (90: 15: 1O:O. I). Data from Barua and co-workers [48]. Retention indices based on the alkyl aryl ketone scale.

4.4.4 Other natural products

Barua and co-workers have used indices based on the alkyl aryl ketone scale in conjunction with dual wavelength detection at 300 or 325 nm and 450 nm for the identification of carotenoids, retinol, retinyl esters and tocopherols in serum samples. They used nonaqueous reversed-phase liquid chromatography with acetonitrile-dichloromethanemethanol-l-octanol(90:15:10:0.1)as eluent on a CI8bonded column [48]. Typical indices ranged from I = 1664 for retinol to I = 2966 for retinyl palmitate (Table 4.10). The indices were consistent on a single column and changed by a much smaller proportion than retention factors on different columns.

4.5 ENVIRONMENTAL SAMPLES 4.5.1 Chlorinated compounds

As part of an environmental study, Ballschmiter and Brodsky [49] reported the retention indices of perchlorobenzenes and polychlorinated biphenyls (PCBs) as test mixtures for the selectivity of reversed-phase separation systems. They used the alkan-2-ones as the reference scale for and diphenyl bonded columns and the n-alkylbenzenes as standards for a cyano-bonded column. In a subsequent paper [50],they examined the PCBs on a Nucleosil and diphenyl phases, using a version of the alkan-2-one scale in which acetone I = 100, and on a Nucleosil 5CN column using the n-alkylbenzene scale with toluene I = 100. They reported the indices for the full set of PCBs on all three columns using methanol-water and acetonitrilewater eluents. For the Nucleosil.5 CI8column with methanol-water (80:20), the indices for changed from biphenyl, I = 820 to PCB 209 (filly chlorinated), I = 820 to 1765. (If based more conventionally on acetone, I = 300, they range from 1020 to 1965). They examined the effect of substitution and the index increment caused by the addition of a chlorine atoms in different positions.

Retention indices for identification in HPLC

165

4.5.2 PAH and aromatic hydrocarbons

The principal application of the polynuclear aromatic ring retention index scale has been for the identification of PAH hydrocarbons. In preliminary studies [51], Wise and coworkers reported the index values for over 80 hydrocarbons on an amino bonded column in a normal phase mode. They also reported the separation on two ODS bonded columns in a reversed phase mode [52]. A similar study was also reported by Chmielowic and George [53]. As part of an environmental study, Murray and co-workers [54] used the PAH scale to identify aromatic hydrocarbons in refinery effluents, mussel extracts and petroleum oils. Thomson and Reynolds [55] employed the PAH scale on liquid crystal columns to identifj. the components of different processed coal and shale oil products, These methods have been reviewed by Sander and Wise [56]. Because of significant differences in the retention depending on the brand of stationary phases, these studies have led to a consideration of the shape selectivity of ODS columns [57] and this work is described in detail in Chapter 10. As part of studies to examine the application of retention indices based on the alkyl aryl ketone scale, Smith [58] compared the separation of linear and branched alkylbenzenes on ODs- and phenyl-bonded silica columns but the selectivities of the phases were not significantlydifferent.

4.6 MISCELLANEOUS SAMPLES

The identification of explosives and their residues is an important environmental and forensic problem and Harvey and co-workers [59] have used the alkyl aryl ketone scale as a marker for 2,4,6-trinitriphenylmethylnitramine(tetryl), using a programmed elution separation with acetonitrile. The tetryl had an index values of I = 946 and its degradation products in a soil sample appeared at I = 714, 813, 872 and 922. In a subsequent paper, Harvey and co-workers [60] used the retention indices to study the metabolism of tetryl in bush bean plants. The original compounds and seven metabolites could be identified from their indices. Yamamato et ul. [61] have reported that a database of retention index values and ultraviolet spectral maxima has been produced for the identification of over 300 components of cosmetics. The indices were based on the carbon number of the side chain of alkyl aryl ketones eluted using a concave gradient.

4.7 CONCLUSIONS The concept of retention indices is slowly gaining acceptance in HPLC, principally for the identification of unknown analytes in drugs samples or plant products. Many other types and groups of samples have also been reported and a wide range of sources of retention index values are given in Appendix 1. The range of applications are already very wide reflecting the almost unlimited scope and breadth of application of reversed-phase HPLC. More comprehensive libraries of retention index values are slowly being estabReferences pp. 167-1 69

166

Chapter 4

lished but face a major problem of the differences between stationary phases and hence in their reproducibility and the transferability of their values. Some form of correction or internal reference to known analytes may be necessary but selectivity differences which cause changes in the order of elution will be difficult to correct.

APPENDIX 1:REPORTED RETENTION INDICES IN HPLC

Retention indices for groups of compounds determined using the different retention index scales. The extensive list of model aromatic compounds examined as part of the retention studies described in Chapter 1 are not included. n-Alkane scale Alkanes, alkenes, alkynes 62,63,64.95 Alkanols, 65; 66 Alkyl alkanoates 65 Alkylthiols, alkyl halides and alkyl 66 ethers 63,65,61 Alkylbenzenes Alkyl chalcones (RzX,, = S, Se, Te) 68 Alkyl polysulphur, selenium and 66,69 tellurium Alkyl pyridines 65 Alkyl thiols and thioethers 63,66, I0 Aromatic mines, phenols and nitro 61 compounds 1pBis(a1kylthioalkanes) 71,12 Chalcogens (S,,, Sen, Ten) 64,69 Isoalkanes 13,14 Ketones 65 Olefins (argentation chromatography) 62 Nitrobenzenes 61 Phenols 61 Polysulphides, thioethers and 68, 70,15 thioamines Sulphur polysulphides 70 n-Alkylbenzene scale Chlorobenzenes (PCBs) Chlorobiphenyls Alkan-2-ones scale Alkyl aryl ketones Anthranilic acid derivative Arthritis treatments(steroids,nonsteroidal anti-inflammatory agents and tranquillizers Azabicycloalkanes Barbiturates Depsides Drug compounds

49,50 49,50

21 10 5

16 10 45,46 6, 71,18

Ergopeptines Glucuronidesof drugs

4-Hydroxyquinoline-3-carboxylicacids Imipramine and metabolites Narcotic analgesics Nitroalkanes PCBs Pharmaceuticals 1-[2-[(Phenoxyphenyl)rnethoxy]ethyl-

piperazines Propanolol analogues N-Substituted 3-propanilidonortropane analogues Primaquine metabolites Pyrazolones Pyridopyrimidines Quinazolinaminederivatives Quinazolinesulphonamidesderivatives Quinazolinederivatives Substituted benzenes Steroids 1,2,6-Thiadiazinone 1,I,-dioxides Urushiol congeners

Alkyl aryl ketone scale (see also Chapters 1 and 3) Alkan-2-ones N-Alkyl anilines Alkyl aryl ethers Alkylbenzenes Alkyl benzoates Barbiturates Basic drugs Carotenoids Cosmetic ingredients Curcuminoids Diuretics Drugs Eugenol-methyl eugenol

36 I9 80 27 81 21 49,50 1,4 82 10 81

28 83 84 85 86 87 88 5 83 42

21 89,90,91,92 90,91,92 59,90,91, 92,93 90,91,92 7,8,9, 12 14, 17, 19 48 62 40 14 21 41

Retention indicesfor identification in HPLC Explosives Fungal metabolites Gingerols, shogaols, paradols Gliadin proteins Local Anaesthetics Penicillium secondary metabolites Mycotoxins Neutral and acidic drugs Nitroalkanes Retinoids Thiazide diuretic Tocopherols

60,61 29 38,39 43 14,15 37 29,35 16 21 48 14 48

Nitroalkane scale (see also Chapter 5) Alkan-2-ones 21 Alkyl aryl ketones 21 Drugs 4,17,21,22,25 Fatty acid ester scales (ECN scale) Unsaturated triglycerides

167 Polycyclic hydrocarbon and nitroPAH scales (see also Chapter 10) 96 NitroPAH PAH 52,53,54,55, 56,97 Processed coal and oil shale 55 Substituted benzenes 54,98 Substituted nitrobenzenes 99 Parabens scales Substituted phenols

100

DNPH alkan-2-one scale Alkanal and ketone DNPHs Aromatic standards

101 102

D Compounds (1-[4-(2,3-dihydroxypropoxy) phenylj-1-alkanones (see Chapter 6) Mycotoxins 34,33

94,95

4.8 REFERENCES 1

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

J.K. Baker and C-Y. Ma, J. Chromatogr., 169 (1979)107. R.A. de Zeeuw, J. Pharm. Biomed. Anal., 1 (1983)435. J.K. Baker, R.E. Skelton and C-Y. Ma, J. Chromatogr., 168 (1979)417. J.K. Baker, L.A. Cates, M.D. Corbett, J.W. Huber and D.L. Lattin, J. Liq. Chromatogr., 5 (1982)829. J.K. Baker and E.K. Fifer, J. Pharm. Sci., 69 (1980)590. R.J.M. Vervoort, F.A. Maris and H. Hindriks, J., Chromatogr., 623 (1992)207. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, Anal. Proc., 22 (1985)331. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, Chromatographia, 19 (1984)401. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, Chromatographia, 19 (1984)407. J.K. Baker, Anal. Chem., 51 (1979)1693. R. Gill, A.C. Moffat, R.M. Smith and T.G. Hurdley, J. Chromatogr. Sci., 24 (1986)153. T.G. Hurdley, R.M. Smith, R. Gill and A.C. Moffat, Anal. Proc., 23 (1986)161. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr.,355 (1986)75. R.M. Smith, G.A. Murilla, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr., 384 (1987)259. M. Bogusz, J. Chromatogr., 387 (1987)404. M. Bogusz and R. Aderjan, J. Anal. Toxicol., 12 (1988)67. M. Bogusz, G. Neidl-Fischer and R. Aderjan, J. Anal. Toxicol., 12 (1988)325. R.M. Smith, R. Gill and M.D. Osselton, J. Anal. Toxicol., 13 (1989)375. B.Gawdzik, J. Chromatogr., 600 (1992)115. D.W. Hill and K.J. Langner, J. Liq. Chromatogr., 10 (1987)377. R.M. Smith and N. Finn, J. Chromatogr., 537 (1991)51. M. Bogusz and M. Wu, J. Anal. Toxicol., 15 (1991)188. M. Bogusz, M. Erkens, J.P. Franke, J. Wijsbeek and R.A. de Zeeuw, J. Liq. Chromatogr., 16 (1993) 1341. M. BogWz, LC-GC, 9 (1991)290;LC-GC Int.., 4 (1991)22. M. Bogusz, J.P. Franke, R.A. de Zeeuw and M. Erkens, Fresenius J. Anal. Chem., 347 (1993)73. R. Gill, personal communication. C.D. Hufford, G.A. Capiton, A.M. Clark and J.K. Baker, J. Pharm. Sci., 70 (1981)151. J.K. Baker, J.D. McChesney and L.F. Jorge, Pharm. Res., 3 (1986)132.

168

Chapter 4

29 30 31

J.C. Frisvad and U. Thrane, J. Chromatogr.,404 (1987) 195. J.C. Frisvad, Bot. J. Linn. SOC.,99 (1989) 81. V. Betina (Ed.), Chromatographyof Mycotoxins. Techniques and Applications, Journal of Chromatography Library, Vol. 54, Elsevier, Amsterdam, 1993. P. Kuronen, in J. Enqvist and A. Manninen (Eds.), Systematic Identification of Chemical Warfare Agents, B.3. Identification of Non-Phosphorus Chemical Warfare Agents, Ministry of Foreign Affairs, Helsinki, 1982, pp. 43-61. P. Kuronen, Arch. Environ. Contam. Toxicol., 18 (1989) 336. R. Kostiainen and R. Kuronen, J. Chromatogr., 543 (1991) 39. D.W. Hill, T.R. Kelley, K.J. Langner and K.W. Miller, Anal. Chem., 56 (1984) 2576. H. Magg and K. Ballschmiter,J. Chromatogr., 331 (1985) 245. R.R.M. Paterson and C. Kemmelmeier, J. Chromatogr., 483 (1989) 153. R.M. Smith, Chromatographia, 16 (1982) 155. R.M. Smith, in T.H. Ryan (Ed.), Electrochemical Detectors. Fundamental Aspects and Analytical Applications, Plenum, New York, 1984, pp. 83-88. R.M. Smith and B.A. Witowska, Analyst, 109 (1984) 259. R.M. Smith and S. Beck, J. Chromatogr.,291 (1984) 424. C-Y. Ma, M.A. Elsohly and J.K. Baker, J. Chromatogr., 200 (1980) 163. J.A. Bietz and L.A. Cobb, Cereal Chem., 62 (1985) 332. H.D. Sapirstein, M.G. Scanlon and W. Bushuk, J. Chromatogr.,469 (1989) 127. C.F. Culberson, M.E. Hale, T. Twsberg and A. Johnson, Mycologia, 76 (1984) 148. C.F. Culberson, W.L. Culberson, S. Gowan and A. Johnson, Am. J. Bot., 74 (1987) 403. S.P. Gowan, System. Bot., 14 (1989) 77. A.B. Barua, H.C. Furr, D. Janick-Bucknerand J.A. Olson, Food Chem., 46 (1993) 419. K. Ballschmiter and J. Brodsky, Fresenius Z. Anal. Chem., 335 (1989) 309. J. Brodsky and K. Ballschmiter, Fresenius Z. Anal. Chem., 335 (1989) 817. S.A. Wise, S.N. Chester, H.S. Hertz, L.R. Hilpert and W.E. May, Anal. Chem., 49 (1977) 2306. S.A. Wise, W.J. Bonnett and W.E. May in A. Bjlrrseth and A.J. Dennis (Eds.), Polynuclear Aromatic Hydrocarbons: 4th Int. Symp. on Analysis, Chemistry and Biology, Battelle, Columbus, OH, 1980, pp.

32

33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52

791-806. 53 54 55 56 57 58 59 60 61

J. Chmielowiec and A.E. George, Anal. Chem., 52 (1980) 1154. A.P. Murray, C.F. Gibbs and P.E. Kavanagh, Int. J. Environ. Anal. Chem., 16 (1983) 167. J.S. Thomson and J.W. Reynolds, Anal. Chem., 56 (1984) 2434. L.C. Sander and S.A. Wise, Adv. Chromatogr.,25 (1986) 139. L.C. Sander and S.A. Wise, J. High Resolut. Chromatogr. Chromatogr.Commun., 11 (1988) 383. R.M. Smith, J. Chromatogr., 237 (1982) 144. S.D. Harvey, R.J. Fellows, J.A. Campbell and D.A. Cataldo, J. Chromatogr., 605 (1992) 227. S.D. Harvey, R.J. Fellows, D.A. Cataldo and R.M. Bean, J. Chromatogr.,630 (1993) 167. S. Yamamato, M. Hosaka, M. Kanda, M. Yokouchi, J. Yamada and S. Tahara, Bunseki Kakagu, 36 (1987) T108.

62 63 64 65 66 67 68 69 70 71 72 73 74 75

B. Vonach and G. Schomburg, J. Chromatogr., 149 (1978) 417. H.J. Mockel, G. Welter and H. Melzer, J. Chromatogr., 388 (1987) 255. H.J. Mockel, F. HOfler and H. Melzer, J. Chromatogr., 388 (1987) 285. H. Figge, A. Deege, J. Kohler and G. Schomburg, J. Chromatogr., 351 (1986) 393. H.J. Mockel, F. HOfler and H. Melzer, J. Chromatogr.,388 (1987) 275. F. Morishita, H. Kakihana and T. Kojima, Anal. Lett, 17A (1984) 2385. U. Marquardt, H.J. Mockel and S.F. Nelsen, Chromatographia,27 (1989) 113. H.J. Mockel, F. Hofler and H. Melzer, J. Chromatogr., 388 (1987) 267. H.J. Mockel, J. Chromatogr., 317 (1984) 589. G. Aced, E. DamstBdt and E. Anklam, Fresenius Z. Anal. Chem., 331 (1988) 740. G. Aced, E. A n k h and H.J. Mockel, Chromatographia,25 (1988) 1067. H.J. MOckel and T. Freyholdt, Fresenius Z. Anal. Chem., 368 (1981) 401. N. Dimov, Anal. Chim. Acta, 201 (1987) 217. E. Singer and H.J. Mockel, Chromatographia,27 (1989) 27.

Retention indicesfor identification in HPLC 76 77 78 79 80 81 82

169

J.K. Baker, G.J. Hite, M. Reamer and P. Salva, Anal. Chem., 56 (1984) 2932. J.K. Baker, L.A. Cates, M.D. Corbett, J.W. Huber and D.L. Lattin, J. Liq. Chromatogr., 5 (1982) 829. S.D. West, J. Chromatogr. Sci., 25 (1987) 122. J.K. Baker, J. Liq. Chromatogr., 4 (1981) 271. E.A. Coats, K.J. Shah, S.R. Milstein, C.S. Genther, D.M. Nene, J. Roesener, J. Schmidt, M. Pleiss, E. Wagner and J.K. Baker, J. Med. Chem., 25 (1982) 57. J.K. Baker, R.E. Skelton, T.N. Rileyand J.R. Bagley, J. Chromatogr. Sci., 18 (1980) 153. J. Wieringa, A. van den Meerendonk, J, Steenvoorden, G. Heeres, F. van Bakal, T. Roeters, R. van der Hulst, A. van Blanken, D. Leysen, Th. de Boer, H. Rijk and R. Plate, Rec. Trav. Chim. Pays-Bas, 112 (1993) 143.

83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102

R. Nieves, C. Ochoa and M. Stud, J. Liq. Chromatogr., 11 (1988) 2349. A. Shalaby, Zs. Budviiri-BMny and Gy. Szhz, J. Liq. Chromatogr., 7 (1984) 1133. T. Sekiya, S. Imada, S. Hata and S. Yamada, Chem. Pharm. Bull., 31 (1983) 2779. T. Tanaka, E. Yamada, T. Sone and H. Hidaka, Biochem., 22 (1983) 1030. A. Shalaby, Zs. Budviiri-Biirhny, K. Hank6-Novhk and G. Szhz, J. Liq. Chromatogr., 7 (1984) 2493. S.D. West and D.H. Mowrey, J. Chromatogr. Sci., 29 (1991) 497. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr., 351 (1986) 259. B. Gawdzik, J. Gawdzik and U. Czerwiliska-Bil, Chromatographia,26 (1988) 399. B. Gawdzik, J. Chromatogr., 549 (1991) 77. B. Gawdzik, Chromatographia,35 (1993) 548. R.M. Smith and D.R. Garside, J. Chromatogr.,407 (1987) 19. B.J. Compton and W.C. Purdy, Anal. Chim. Acta, 141 (1982) 405. 0. Podlaha and B. TOregiird, J. Chromatogr., 482 (1989) 215. T-Y. Liu and A. Robbat, J. Chromatogr., 539 (1991) 1. M.N. Hasan, and P.C. Jurs, Anal. Chem., 55 (1983) 263. M. Popl, V. Dolanslj and J. F h r i c h , J. Chromatogr., 148 (1978) 195. F. Smejkal, M. Popl, A. Cihovii and M. Zhorkovh, J. Chromatogr., 197 (1980) 147. S. Yamauchi, J. Chromatogr., 635 (1993) 61. L. Margarit-Roig, A. Dim-Marot and M. Gassiot-Matas, Afinidad, 43 (1986) 525. M. Gassiot Mates, A.C. Auguet-Sangrh and I. Masana Gonzales, Afinidad, 39 (1982) 479.

This Page Intentionally Left Blank

R.M. Smith (Ed.), Retention and Selectivity in Liquid Chromatography Journal of Chromatography Library, Vol. 57 0 1995 Elsevier Science B.V. All rights reserved

171

CHAPTER 5

Application of nitroalkanes and secondary retention index standardsfor the identification of drugs Maciej Bogusz Institutf i r Rechtsmedizin, Medizinische Fakultat, Rheinisch-Wesfalische Technische Hochschule, Germany Pauwelsstrasse 30,0-52057,

5.1 INTRODUCTION

Forensic and clinical analytical toxicologists are confionted in everyday casework with one main difficulty: the information concerning the sample about to be analyzed is often sparse and of disputable quality. On the other hand, in almost every case of suspected poisoning with an unknown agent, the spectrum of toxic substances which should be considered is broad and usually several hundred relevant substances are potentially involved. The analytical strategy applied in such a situation must follow two aims at the same time: - The results should be obtained as quickly as possible. The time factor is particularly important in clinical toxicology, when the analytical results may influence the process of decision-making relevant for treatment. - The analytical procedure should enable the detection of all possible toxic substances and their metabolites present in the biosample. It is obvious that both these tasks are mutually conflicting. Quick, simple bed-side or field tests are neither comprehensive nor of value. On the other hand, the systematic search for every poisonous substance that might potentially be involved, may last weeks and cause unacceptable instrumental and personnel costs. This discrepancy dictates the need for compromise, the extent of which must fit the local situation, the primary purpose of the analysis and the particular case. The importance of the above-mentioned difficulties was recognized early in forensic and clinical analytical toxicology. For the most effective identification in so-called “general unknown” analysis, a two-step analytical strategy was developed: - Preliminary tests, mainly based on group-specific immunochemical reactions, are applied to narrow the possible analytical field and to optimize the final identification.

References pp. 205-20 7

172

Chapter 5

Confmation analysis assures the positive individual identification and quantitation of a given substance. The analytical techniques applied for both these steps in toxicological screening are reviewed elsewhere [e.g. 1-31. In the first step of the identification procedure, a collection of specific identification parameters, comprising values from all the possibly relevant compounds, is established. The comparison of the corresponding parameters observed for a particular sample, with the collection known as the database is the principle of every identification procedure. It is obvious that the efficiency of the identification depends upon the specificity of the identification parameters and the completeness of the database. One particular laboratory cannot establish a database for all relevant poisonous substances, even when only one analytical technique is involved. This can only be done with the cooperation of many laboratories, at national and international level. Such a cooperation should first scrutinize the existing methods in order to chose those that are generally applicable. The main criterion for such an applicability is the possibility of standardization. Only standardizable techniques enable the establishment of comprehensive databases, available for general use. The rationale behind standardization is to make the results conform to a set of standards, i.e. to known, generally available substances with predictable and comparable analytical properties. The choice of the appropriate standardization procedure and adequate standards is of utmost importance for successful identification. Among the various methods potentially useful in toxicological screening, chromatographic techniques play a particular role, due to two important features: - chromatographic techniques allow the separation, detection, identification and quantification of several hundred substances in one analytical run; - the methods are standardizable and the establishment of toxicological databases is feasible. From the very beginning of the application of chromatographic methods, such as thin layer chromatography (TLC) or gas chromatography (GC) in toxicology, various authors have tried to express the retention of compounds as secondary parameters, i.e. related to various standards. In textbooks of analytical toxicology, several such compilations of relative retention times in GC, or relative RF in TLC may be found [1,4]. An organized scientific approach to standardization in toxicology was provided by the Committee for Systematic Toxicological Analysis of The International Association of Forensic Toxicologists. This body has been involved for more than 10 years in the optimization of general toxicological screening analyses for unknown substances. Standardized GC and TLC databases, comprising values for ca. 6000 toxicologically relevant substances, were compiled from several sources and published recently [5-71. The GC data of non-volatile compounds were standardized on the n-alkane retention index (RI) scale of Kovhts [SJ, using a set of drugs with known retention index values as secondary standards [9]. The retention index library of volatile compounds was standardized with a homologous series of aliphatic n-alcohols [lo]. In the case of the TLC database, four standard drugs were used for the calculation of corrected RF values in each of 10 developing systems [7]. In general, studies on the standardization of GC and TLC results have shown that the best results are obtained when the standards structurally resembled the substances under investigation [ 11,121. -

Nitroalkanes and secondary retention standards

173

HPLC offers some specific qualities, which assure its place among other chromatographic methods relevant to toxicological screening analysis. The separation potential of HPLC is much higher and the detection limits much lower than those of less sophisticated TLC. On the other hand, the comparison of HPLC with other column chromatographic techniques, such as GC, indicates the universality of HPLC. The method may be applied to non-volatile, thermolabile and highly polar substances. These features make HPLC the method of choice for the analysis of conjugated drug metabolites.

5.2 THE USE OF HPLC AS A STANDARDIZED IDENTIFICATION METHOD IN TOXICOLOGY 5.2.1 Standardization of retention using straight phase silica The major study of the application of straight-phase silica column packing material for toxicological screening was undertaken by British researchers, affiliated to the Home Office Forensic Science Service. Jane [ 131 introduced the use of a polar eluent, a mixture of methanol and ammonium nitrate buffer (9:1, pH 10.1) for the analysis of basic compounds by means of straight phase HPLC. This system was used for years as a standard in UK forensic science laboratories and several applications were described [ 14,151. Law et al. [16] have compared the efficiency of several straight phase silica packing materials: Hypersil, Spherisorb S5W, Nucleosil50-5 and Zorbax BP-SIL, using seven selected basic drugs. Large differences in selectivities were reported, and one packing, Spherisorb S5 W, was chosen for further use and studies. The retention data (Rt and k) of 84 basic drugs were given. The authors have recommended the use of the same packing material among all laboratories wishing to transfer retention data. Gill et al. [17] have performed an interlaboratory study among nine British forensic laboratories. The same batch of Spherisorb S5W was distributed, and preparation of the mobile phase, elution conditions and sample amount were defined. Each laboratory packed the column in-house and used routine instrumentation. Twenty-eight basic drugs were examined, their retentions being expressed as retention times, relative retention times, capacity factors, relative capacity factors and corrected capacity factors. The last method was based on the assignment of reference capacity factors to five selected basic drugs. These reference values were obtained by repeated interlaboratory determinations under standardized conditions. The experimental k values for the same drugs were plotted against the reference values, and the plot served for the calculation of corrected k values for all other compounds. Even under these uniform conditions, the interlaboratory reproducibility of the primary retention parameters, retention times and capacity factors, was poor. Better results were obtained using relative Rt or k values, whereas the most reproducible results and the best discrimination values were obtained using corrected capacity factors. As a general principle emerging from this study, it was stated that the interlaboratory reproducibility of a retention measurement method increases with the number of reference compounds used. In a second interlaboratory study [ 181 comprising 10 laboratories from different countries, each laboratory received two columns packed with Spherisorb S5W material. One References pp. 205-207

174

Chapter 5

column was packed with a common batch, the second one with one of three batches. The laboratories were also asked to use their own column. The elution conditions were standardized, the instrumentation and environmental conditions were different. Sample mixtures containing 28 selected basic drugs (the same as in the previous study) were supplied by the authors of the study. The retentions were expressed as R,, k values and relative k values. The last parameter gave the most reproducible results. As in the previous studies, distinct variations were found between different brands of silica. At the international level, however, the interlaboratory variations on a single batch were even more significant. The differences in ambient temperature and composition of the mobile phase (containing volatile ammonia) were identified as possible factors affecting the reproducibility of results. The interlaboratory variability of results was also substancedependent: Some drugs (prolintane, pipazethate, dipipanone and strychnine) were obviously more sensitive to changes in the separation conditions. The corrected capacity factors were not calculated in this study. The importance of very careful standardization of the mobile phase composition consisting of methanol and ammonium nitrate buffer was shown by the same group of authors, using the same Spherisorb silica packing [19]. Small changes in methanol content were associated not only with changes in the elution time, but also in elution order. Such deviations could not be compensated by the use of secondary standards. In addition, changes in pH, and to a lesser extent in temperature and ionic strength of the mobile phase, may affect the retention behaviour of basic drugs to various degrees. Experiments with 18 different batches of Spherisorb S5W, which had been manufactured over a period of several years, revealed significant differences in the retention behaviour of basic drugs [20]. The same methanol-ammonium nitrate mobile phase was used. For most of the 28 basic drugs examined, the batch-related retention differences could be compensated using relative capacity factors. For some compounds, however (prolintane, pipazethate, dipipanone, strychnine, phenylephrine), the variations were particularly high. These drugs were proposed as a test mixture for silica columns. In the conclusion, the authors stated that not only the chromatographic conditions, but also the column packing (the use of a single batch) should be standardized in order to obtain reproducible results between laboratories. The replacement of ammonia with a more robust constituent of the mobile phase was postulated. An alternative to an ammonium nitrate mobile phase was proposed by Flanagan et al. [21] who used a methanol-ammonium perchlorate mixture for a screening analysis for basic drugs. Minty [22] found this system to be very suitable for quantitative work, rather than for identification. In a further study, Smith et al. [23] replaced ammonium nitrate buffer with CAPSCAPSO-Na buffer, prepared from solid reagents. This mobile phase was more robust, but again the differences between different batches of the same brand of silica were noted. Also, the same compounds (dipipanone, prolintane and pipazethate) were particularly sensitive to small changes in elution conditions. A 1-year long reproducibility study of the retention behaviour of 28 selected basic drugs, examined in previously standardized conditions (CAPS-CAPSO-Na buffer), has identified the ageing processes as the main source of irreproducibility of results within one brand of silica [24]. The same drugs as observed previously were mostly affected.

Nitroalkanes and secondary retention standards

175

The changes in the retention behaviour of drugs were accelerated by water; however, they also occurred when silica was stored as a dry powder. The results of this recent study suggest that much of the batch-to-batch differences observed in the earlier studies might have been due to an ageing process rather than variations in the manufacturing procedures. The results of the studies on straight phase silica are also relevant for reversed-phase HPLC. The silanol interactions, responsible for uncontrolled and unwanted chromatographic effects and observed particularly during the separation of basic drugs, occur also in reversed-phase chromatography. Moreover, the ageing processes observed on the silica surface may also affect the separation potential of reversed-phase packing material. These processes might be enhanced in due time, when more silica on the surface of the reversed-phase packing is exposed. In general, reproducible results between laboratories may be achieved when the following conditions are met: - the elution conditions should be strictly defined and meticulously followed; - the retention should be expressed using secondary parameters, possibly with multiple reference compounds; - the same brand of straight phase column packing material should be used. The last condition, which is clearly impossible to fulfil on an international level, shows the limitations of the standardization efficiency. Also, the observed ageing processes of silica cannot be controlled and compensated.

5.2.2 Standardization of retention using reversed-phase silica

Reversed-phase column packings have been more frequently applied for toxicological screening than straight phase silica. Most often, octadecyl- or octyl-silica was used as the column packing, and a mixture of methanol or acetonitrile with an acidic buffer, containing an amine modifier, as the mobile phase. Several hundreds of toxicologically relevant substances have been examined in various laboratories. Table 1 shows the most important databases established using reversed-phase systems. Each of the published databases was established using different operational conditions, different brands of octyl- or octadecyl-silica columns and a variety of instrumentation. Moreover, the retentions of the analytes were expressed in relative terms, as relative retention times [25,34] or retention indices [3 11, in only three publications. In all other papers, the authors used primary retention parameters, i.e. retention times or capacity factors. In only one publication was an effort made to make the results transferable on an interlaboratory scale through a standardization procedure [3 11. In the other papers, the possibility of interlaboratory use of databases was not addressed at all, or it was recommended that each laboratory should prepare its own collection of data under local conditions [33,37]. The successful application of reversed-phase column packing for identification in toxicology is affected by even more factors than in the case of straight phase silica. The common factor for both kinds of phase, is that the major contributor to the irreproducibility of the results is the use of different brands of column packing. It has been repeatedly Referencespp. 205-207

I76

Chapter 5

TABLE 5.1 REVERSED-PHASE SYSTEMS USED FOR TOXICOLOGICAL SCREENING Column

Operational conditions

Substances

Ref.

RP 18 Merck

ACN-phosphate buffer pH 2.3 isocratic and gradient, UV 220 nm. ACN-phosphate buffer pH 3.2 gradient, UV 210 nm ACN-phosphate buffer pH 3.2 gradient, UV 202 nm ACN-MeOH-phosphate buffer-TEA pH 2.7, isocratic, DAD ACN-perchlorate buffer isocratic, DAD ACN-phosphate buffer-TEA pH 2, gradient, DAD ACN-triethylammonium phosphate buffer pH 3.2, gradient, DAD ACN-phosphate buffer-TEA pH 3.4, isocratic, DAD MeOH-THF-phosphate buffer pH 2.6, isocratic, DAD ACN-sulhric acid, isocratic DAD ACN-phosphoric acid ACN-ammonia, gradient, DAD ACN-phosphate buffer-dimethyloctamine, pH 6.5, isocratic, DAD ACN-phosphate buffer pH 3.2 gradient, DAD

372 acidic and basic drugs

25

42 acidic and basic drugs

26

47 acidic and neutral drugs

27

57 basic drugs

28

48 acidic and basic drugs

29

350 acidic and basic drugs

30

225 acidic and basic drugs

31

121 basic drugs

32

27 neuroleptic drugs

33

161 acidic and basic drugs

34

157 acidic and basic drugs 144 acidic and basic drugs 30 acidic and basic drugs

35 3Cia

100 basic drugs

37

C 18 Bondapak ODS Hypersil LC 18 DB Supelcosil C 18 Jasco RP 18 Hypersil RP 18 Superspher ODS Apex

C 18 Novapak RP8

C 8 Zorbax PRP- 1 C 8 and silica RP 8 Lichrospher

automated HPLC system with a library comprising data for over 300 drugs was developed on the basis of this paper and is commercially available (Bio-Rad Inc., Hercules, CA, USA).

stated that nominally identical but commercially different reversed-phase packings (e.g. octadecyl silicas) may show strikingly different selectivities [3 8-40]. The applicability of various brands of packing for identification, particularly in the case of basic drugs, depends upon the extent of so-called “silanol effects”, i.e. unwanted and sometimes unpredictable interactions of the analytes with residual silanols, exposed on the silica surface. The theoretical and practical aspects of these effects have been reviewed elsewhere [41441.

The extent of silanol effects and of other factors influencing the chromatographic properties of a reversed-phase column depends upon the material and procedures used in the manufacturing of column packing. Type B silica (in contrast to type A polymer) showed a higher concentration and more uniform distribution of silanol groups, which are more effectively alkylated on the silica surface [45]. The reversed-phase packing prepared from type B silica showed less “silanol effects” and was more suitable for the analysis of basic compounds. Different (usually patented) silane modification procedures lead to consequent differences in carbon loading, chain type of silane (monomeric or polymeric) [46] and distribution of free silanols. The manufacturers have invested consid-

Nitroalkanes and secondary retention standards

177

erable effort in recent years in the preparation of so-called base-deactivated columns, which show negligible silanol effects and are specially prepared for the analysis of basic drugs [47-49]. A comparative study has demonstrated that these columns, although not completely free fi-om silanol effects, are very suitable for systematic toxicological screening [50]. In view of the above-mentioned studies, it is understandable that even the use of the same column packing material and nominally identical chromatographic conditions may not prevent an interlaboratory variability of retention parameters, particularly the primary ones. Small deviations in eluent composition, in flow rate, gradient profile or temperature, may cause uncontrolled fluctuations in the retention of analytes, making impossible the effective use of a database of retention times or capacity factors developed in another laboratory. In this situation, various retention index scales in HPLC were introduced in order to compensate for the differences in the selectivities of various brands of column packings and for deviations from the pre-set operational conditions [51]. A homologous series of alkane-Zones was proposed as the first retention index system applicable for toxicological analysis [52]. The RI-values of selected acidic and neutral drugs were much less dependent on the composition of the mobile phase than the capacity factors. Nevertheless, an increase in methanol content of the mobile phase was associated with a decrease in RI values. The possible influence of changes in mobile phase composition and pH on various classes of drugs (acidic, neutral and basic compounds) was discussed. In an interlaboratory study, Baker et al. [53] compared the retention times and RI-values on the alkane-2one scale of seven selected acidic, neutral and basic drugs. Four laboratories used the same brand of ODS column (Bondapak C-18 endcapped), one laboratory used Partisil ODS-2 and one used Mikropak MCH-10 column. The last two columns have a higher number of free silanol groups. The elution conditions were identical. The interlaboratory variability of the RI-values was considerably lower than that of the relative retention times (mean CV 12.6% and 30.2%, respectively). When only Bondapak C-18 columns were taken into consideration,the mean CV value decreased to 3.6%. The retention index scale of alkane-2-ones showed a practical drawback; the ketones have low UV absorbance, therefore the use of relatively high concentrations of standards was required. However, this was difficult to achieve due to the limited solubility of higher, unpolar homologues in the mixture of organic modifier and water. Alkyl aryl ketones, proposed as a retention index scale in reversed-phase HPLC by Smith [54], showed much higher UV-absorptivity than aliphatic ketones. Similar to the alkane-2-one data, the €U-values of several test compounds measured on the alkyl aryl ketone scale were less sensitive to variations in the methanol content of the mobile phase than the corresponding capacity factors. An increase in RI values of the test compounds with increasing methanol content of the mobile phase was observed. Three brands of column packing were used in this study, and the corresponding RI values were distinctly different. In further studies, Smith et a1.[55] demonstrated that RI values of barbituric acid derivatives, although more reproducible than the capacity factors, were susceptible to changes in methanol content, pH and temperature of the mobile phase. The extent of these changes and trends observed for barbiturates were different than for other test compounds, which were of different polarity and chemical character. When different brands Referencespp. 205-207

178

Chapter 5

of reversed- phase packing were used, the scatter of RI-values observed for the same compounds was tremendous. In conclusion, the use of the same batch of packing material was recommended in order to obtain highly reproducible results [56]. The investigation of alkaline local anaesthetic drugs [40] showed that the FU-values of drugs were dependent on the composition of the mobile phase (methanol and hexylamine content, pH and temperature). The comparison of results obtained with different ODS columns (Hypersil, Partisil and Zorbax) revealed distinct differences and different elution order. It should be mentioned that the columns used in this study have different properties: in the “acidity ranking” [41] Zorbax and Hypersil columns are located far on the “acidic” side, whereas Partisil was less acidic. Again, the use of the same batch of packing material for interlaboratory use was recommended. The application of the alkyl aryl ketone retention index scale for identification of acidic thiazide diuretics showed that the indices decreased with increasing concentrations of organic modifier in the mobile phase. This trend was identical to the case of barbiturates. One exception was clopamide, which was the only basic drug among the diuretics examined. Comparison of the results obtained with seven columns showed that the variability in RI-values was much lower than in k-values and similar to the variability of relative k-values. Again, RI-values observed for clopamide on different columns showed different trends from FU-values of other diuretic drugs [57]. Acetonitrile and tetrahydrofuran can be also used as organic modifiers for an alkyl aryl ketone retention index system. The changes in the concentration of these modifiers were associated with the changes in RI-values of the test compounds (58). In an interlaboratory study, Gill et al. [59] distributed the same batch of Hypersil reversed-phase packing to ten laboratories for the examination of barbiturates in exactly defined conditions. Nevertheless, the retention times and capacity factors of the drugs showed considerable interlaboratory variability. The use of secondary retention parameters (relative capacity factors or retention indices, calculated on the alkyl aryl ketone scale) brought an improvement, but the best results were obtained using corrected capacity factors, calculated against four standard barbiturates. This method of standardization was identical to that used in the study of straight phase silica by the same group of authors [ 171. Bogusz et al. [60] have investigated the retention index values of selected acidic and basic drugs in the alkyl aryl ketone scale, examined consecutively on the same columns under identical conditions in two laboratories. The differences in instrumentation contributed more to the variability than the use of different batches of the same brand of column packing. The retention index system based on alkyl aryl ketones showed a practical disadvantage: The first reference compound, acetophenone, elutes later than a number of toxicologically relevant substances, which makes the calculation of their retention indices difficult or even impossible. A homologous series of l-nitroalkanes was therefore proposed as an alternative retention index scale for reversed-phase HPLC [6 I]. Nitroalkanes show high UV absorptivity in the range 200-220 nm which is slightly affected by changes in pH (Fig. 5.1). Also, the retention behaviour of nitroalkanes was predictable in isocratic conditions and was not affected by the changes in the pH of the mobile phase (Figs. 5.2,

5.3). Changes in the concentration of organic modifier (acetonitrile) were associated with the changes in €U-values of selected acidic and basic drugs. The U-values of acidic drugs

Nitroalkanes and secondary retention standards

179

2.0

1.8

1.6

1.4

1.2

1 .o

0.8

0.6

0.4

0.2

A

x

2oo

220

240

260

28 0

Fig. 5.1. UV spectra of 1-nitropropane in acetonitrile-water (1: 1) (-), 3.2) (1:l) (......) and acetonitrilephosphate buffer (pH 8.5) (I:I) (----). of Elsevier Science.

References pp. 205-207

3 0 0 nm

acetonitrile-phosphate buffer (pH From ref. [61] with the permission

180

Chapter 5

I

. l o o C" x 100

300

400

500

bU0

Fig. 5.2. The relationship between the logarithm of k and the carbon number of I-nitroalkanes observed in acetonitrile-buffer pH 3.2 (......) and acetonimobile phases of different pH: acetonitrile-water (4:6)(-), trile-buffer pH 8.5 (----). From ref [61] with the permission of Elsevier Science.

decreased with increasing concentration of acetonitrile, whereas basic drugs showed an opposite trend. Retention indices of neutral substances were minimally affected (Fig. 5.4). The trend observed for acidic drugs was also noted by Baker and Smith for the alkan2-one and alkyl aryl ketone scales [52-541. The first six 1 -nitroalkane homologues (nitromethane to 1-nitrohexane) are available commercially; the higher ones may be easily synthesized [62]. The homologues up to 1nitrooctane cover practically the whole elution range relevant for toxicological analysis using HPLC. The series of 1-nitroalkanes may be also used as a retention index scale for GC. The advantage of this scale over the n-alkanes is that 1-nitroalkanes are easily detectable not only with a flame ionization detector, but also with electron capture and thermionic detectors (Fig. 5.5). Preliminary observations concluded that the RI-values of acidic and basic drugs showed different trends when the elution conditions are changed. In a following study, Bogusz [63] examined the RI-values (nitroalkane scale) of 62 selected acidic, neutral and basic drugs under isocratic conditions with various concentrations of acetonitrile and under gradient conditions (Figs. 5.6,5.7). The mobile phases consisted of acetonitrile, phosphate buffer and nonylamine. The RIvalues of acidic drugs and earlier-eluting basic drugs decreased with increasing concentration of acetonitrile. For the later-eluting basic drugs, the opposite trend was observed. The values obtained in gradient elution and in isocratic conditions were not transferable, and the elution order was in some cases different (Table 5.2). In contrast to the isocratic conditions, in the case of gradient elution there were three variables involved: Acetonitrile concentration, nonylamine concentration, and ionic strength of the mobile phase. The changes in nonylamine concentration during gradient

Nitroalkanes and secondary retention standards

181

elution may influence the retention behaviour of basic drugs, especially those which elute later, when the concentration of amine is greatly reduced. The tentative use of a constant nonylamine concentration throughout the whole gradient elution leads to a rapid change in the pH of the mobile phase, due to the decreasing buffer capacity (Fig. 5.8). The use of a constant concentration of amine modifier would be feasible under gradient elution only if a concentrated buffer of large capacity were used. However, this would be associated with large variations in ionic strength of the mobile phase and would be technically impossible for gradient elution, due to the limited solubility of salts in higher concentration of acetonitrile. This study has indicated that in the case of gradient elution, a simple, one-step gradient profile and possibly the simplest mobile phase should be applied, in order to eliminate all possible sources of interlaboratory variations.

50.C

10.0

1.0

k'

loo C" x 100

200

300

400

500

600

Fig. 5.3. Relationship between the logarithms of k and the carbon numbers of 1-nitroalkanes observed at different concentrations of acetonitrile-phosphate buffer (pH 3.2) on the Lichrocart column. From ref. [61] with the permission of Elsevier Science.

References pp. 205-207

182

Chapter 5

600

/*

*-*/*-*

500

methylaniline niethyl benzoate

400

/* 300

200

brallobarbital

8 phenobarbital

100

I ~

5 10 % MeCN

20

30

40

50

~~

60

70

Fig. 5.4. The retention indices of the test compounds calculated with reference to 1-nitroalkanes at different concentrations of acetonitrile-phosphate buffer (pH 3.2) on the Lichrocart column. From ref. [61] with the permission of Elsevier Science.

Smith et al. [64] have compared all three retention index scales: alkane-2-ones, alkyl aryl ketones and l-nitroaikanes, using isocratic mixtures of methanol, acetonitrile and tetrahydrofuran with buffer as an eluent (Fig. 5.9). Both alkane-2-one and l-nitroalkane scales can also cover the elution range for rapidly eluting drugs. However, in all eluent combinations, the retention indices of rapidly eluting, polar drugs decreased with increased proportions of organic modifier. The elution order of acidic and weakly basic drugs was different on two ODS-columns tested. The need for standardization of elution conditions and of the brand of stationary phase was stressed. The conclusions concerning the applicability of reversed-phase column packing for establishing of a common HPLC database, are virtually the same as in the case of straight phase silica. Interlaboratory use and exchange of retention data requires: - an exact standardization of elution conditions, which should be robust enough for interlaboratorycomparison; - the use of secondary retention parameters, preferably a retention index scale; - the use of the same brand of reversed-phase packing, which is hardly feasible on an international level.

Nitroalkanes and secondary retention standards

c

, rnin

1

183

20

15

10 I

RItKwatr)

1000

1200

1400

1600

2s 1

I

zdo

100

30U

2000

2400

300

2800

3000

3200

Fig. 5 . 5 . Gas chromatogram of a mixture of I-nitroalkanes (c6+6) analyzed on a CP-Sil 5 column with thermionic detection. Temperature programme. The lower scale shows the location of the n-alkanes. From ref. [62] with the permission of Elsevier Science.

5.2.2. I . The concept of secondaty standardsfor retention index scale

The prerequisite for the efficient use of any retention index scale for compound identification is the robustness of the individual RI-values of the compounds stored in a given database. This robustness depends upon one condition: the retention behaviour of reference standards should be qualitatively and quantitatively identical to the behaviour of the compounds being examined, irrespective of operational conditions or changes in the brand of chromatographic column or instrumentation. This may be expected when the reference and compounds being examined are structurally similar. The importance of the use of adequate reference standards was first recognized in GC. n-Alkanes, first applied as a retention index system in GC 35 years ago [8], showed limited applicability for the identification of drugs, especially when temperature programmed conditions were applied. The chromatographic behaviour of non-polar n-alkanes, separated on a non-polar column, differed distinctly from the behaviour of acidic or basic drugs. Even the rigid standardization of instrumental conditions did not ensure reproducible results on an interlaboratory level, due to deviations from the preset operational conditions. As a consequence, the interlaboratory variability of €U-values in GC was found to be at the level of *50 RI units [65], whereas within one laboratory the RIs may be reproduced within 1 RI unit [66]. As a solution, secondary RI standards for identificaReferences pp. 205-207

184

Chapter 5

Fig. 5.6. Retention behaviour of 1-nitroalkanes analyzed in isocratic conditions in mobile phases containing phosphate buffer with 0.05% nonylamine and different concentrations of acetonitrile. From ref. [63] with the permission of Preston Publications.

tion were proposed; a mixture of drugs with known, previously determined RI-values were used for the calibration [9,1 I]. First, a mixture of 1 1 acidic and basic drugs, covering the whole elution range in temperature programmed GC was applied [65]. In a further development, two mixtures of RI-secondary standards were used: 12 acidic and 14 basic drugs, which should be applied to aqueous extracts. The RI-values of the compounds examined were determined through linear interpolation between two secondary standards. The interlaboratory variability was reduced from *50 to *25 RI units [9]. A retention index database, comprising over 6000 substances, was compiled on the basis of this standardization [5]. In HPLC, secondary retention index standards were introduced by Bogusz [67] as “corrected retention indices”. The concept was similar to corrected &values in TLC [7] or to corrected capacity factors [ 17,591. The correction procedure assumes that the secondary RI-standards are eluted in the same order on different brands of RP-silica columns, and that the differences in elution times of analytes observed on particular columns are proportional to differences in elution times of secondary RI-standards.

Nitroalkanes and secondary retention standards

185

The retention index values of six barbiturates were determined under identical, isocratic conditions in the alkyl aryl ketone scale on five different ODS columns. Three other barbiturates were used as secondary standards, and the data from the study of Smith et al. [56] were used as reference values. The corrected retention indices RIc were then calculated using the formula

where RI is the retention index of the analyte, RIlo and R120 are reference RI values of secondary standards, eluted before and after the analyte, RI, and R12 are apparent RI values for secondary standards. The correction procedure compensated very effectively for the differences in RI values caused by different brands of column packing (Table 5.3). I

I

I

I

I

I

I

I

CGRBON NUMBER X 100 Fig. 5.7. Retention behaviour of 1-nitroalkanes analyzed in gradient elution in a mixture of acetonitrilephosphate buffer with 0.05% nonylamine. From ref. [63]with the permission of Preston Publications.

References pp. 205-207

Chapter 5

186 TABLE 5.2 RETENTION INDEX VALUES OF DRUGS EXAMINED IN ISOCRATIC CONDITIONS AND IN GRADIENT ELUTION" Drug name 1. Acebutolol 2. Cocaine 3. Paracetarnol 4. Theophylline 5 . Nomifennsine 6. Bupivacaine 7. Caffeine 8. Barbital 9. Alprenolol 10. Propranolol 11. Salicylamide 12. Diphenhydramine 13. Doxepin 14. Allobarbital 15. Haloperidol 16. Aprobarbital 17. Promethazine 18. Imipramine 19. Phenobarbital 20. Brallobarbital 21. Promazine 22. Phenacetin 23. Cyclobarbital 24. Maprotiline 25. Methadon 26. Amitriptyline 27. Butalbital 28. Carbromal 29. Perphenazine 30. Pentobarbital 3 1. Heptabarbital 32. Amobarbital 33. Phenytoin 34. Chlorprornazine 35. Chlorprothixene 36. Carbarnazepine 37. Propyphenazone 38. Secobarbital 39. Nitrazepam 40. Lorazeparn 41. Oxazepam 42. Fluphenazine 43. Tritiuoropromazine 44. Flunitrazeparn 45. Thiopental 46. Ternazepam 47. Lormetazepam 48. Thioridazine

ISOCR2O

ISOCR40

ISOCR60

Gradient

192 310 48 68 298 318 1I6 184 342 340 246 272 313 282 320 308 347 390 314 332 276 324 364 402 409 42 1 354 420 456 446 454 448 41 1

80 176 ND 20

70 70 ND ND 50 80 ND ND ND ND 108 80 158 80 136 172 284 296 112 128 292 210 164 322 370 474 235 294 310 270 267 246 180 346 406 280 250 276 280 274 230 340 412 366 390 350 400 564

164 165 175 200 208 222 271 275 294 296 307 311 333 334 337 351 353 362 363 363 364 366 378 382 389 389 392 397 411 412 413 417 418 422 424 429 436 437 442 446 449 452 469 475 479 480 486 486

100

1I3 36 66 90 90 158 170 192 174 194 196 250 270 208 226 260 236 248 260 326 308 268 322 250 312 310 320 305 352 376 338 396 360 360 352 330 350 398 430 444 428 446 442

Nitroalkanes and secondary retention standards

187

TABLE 5.2 (continued) Drug name 49. 50. 5 1, 52. 53. 54. 55.

56. 57. 58. 59. 60. 61. 62.

Nordiazepam Clobazam Clorazepate Oxyphenbutazone Nitedipine Wartarin Diazepam Ketazolam Camazepam Indomethacine Giibenclamide Ibuproten Phenylbutazone Prazepam

ISOCR20

ISOCR40

ISOCR60

Gradient

438 452 454 476 482 535 524 522 534 642 656 650 660 672

430 386 352 386 420 480 544 480 516 550 530 600 616 688

49 1 49 1 496 503 513 546 553 553 561 614 619 627 647 674

From ref. [63] with the permission of Preston Publications. “Mean values of triplicate determinations. ND, value not determined because of very short retention time.

The same procedure was applied to the recalculation of data obtained for the thiazide diuretics by Smith et al. on seven different ODS columns [57]. Among 17 compounds examined, three were used as secondary standards, and one column was chosen arbitrarily as a reference. The results of the recalculation showed that the differences in €U-values were compensated to a large extent for all drugs and all columns with the exception of clopamide, the only basic substance among the drugs studied [ 121 (Table 5.4). The application of corrected retention indices for barbiturates or acidic diuretic drugs indicated that this standardization procedure was very effective for chemically related substances with similar polarity and fkctional groups. However, the case of clopamide showed that it is probably not possible to use acidic drugs as standards for basic substances and vice versa. The applicability of corrected retention indices for the identification of acidic and neutral drugs belonging to different chemical and pharmacological classes, was studied by Bogusz and Aderjan [68]. Thirty-seven drugs were divided into two series and the corrected retention indices were determined using an alkyl aryl ketone scale and three drugs as correction standards for each series. Seven ODS columns were used for the study, representing four different brands and three different sizes. Two-step gradient elution in acetonitrile-phosphate buffer (pH 3.2) was applied. The uncorrected RI-values of drugs observed for the same drugs on different columns showed large variability; the mean standard deviation amounted to 25 RI units. When the correction was applied, the mean standard deviation value decreased to 15 RI units. The variability of corrected EU-values observed for various brands of column packings, was similar to those observed for various batches of the same brand (Table 5.5). In this paper, the “discrimination number” (DN) was calculated for the values obtained in the study and for retention values from other studies. DN value may be a guide as to how many compounds can be distinguished in the system, assuming that the distribution of the elution is even [59]. The results are shown in Table 5.6. Application of the correcReferences pp. 205-207

188

Chapter 5 PH O F MOBILE PHASE

2 1.0

0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.8

PERCENT 0F"RCETONITRILE

4 llecN uithout WI

+llecN*M

Fig. 5.8. Changes of pH of a mixture of phosphate buffer containing 0.05% nonylamine after addition of increasing amounts of acetonitrile alone (+) and acetonitrile containing 0.05% nonylamine (+). From ref. [63] with the permission of Preston Publications.

tion procedure was associated with a distinct increase in DN value. For a uniform group of drugs (barbiturates), analyzed under isocratic conditions, this increase was 5-6-fold in comparison with uncorrected RI-values. For structurally different acidic and neutral drugs analyzed in gradient conditions, a twofold increase in DN was noted (Table 5.6). The effectiveness of the correction procedure for various basic drugs was studied by Bogusz et al. [69]. Retention index values of 16 selected basic drugs, covering a broad elution range, were determined on two scales: alkyl aryl ketones and 1-nitroalkanes. Three batches each of two brands of ODS columns (Superspher RP 18, 125 X 4 mm, and Chromspher C 18, 100 x 3 mm) were used. The experiments were carried out in a twostep elution gradient in acetonitrile-phosphate buffer (pH 3.2) containing 0.05% nonylamine. In the correction procedure, three drugs were applied as secondary RI standards. The RI-values expressed in the l-nitroalkane scale showed ca. threefold decrease in variability after correction. The deviations from the reference values observed for two different brands of column packing were of the same order as deviations within one brand (Tables 5.7, 5.8, Fig. 5.10).

Nitroalkanes and secondary retention standards

189

2.00 1 1.50

-

1.00

-

0.50

-

x

ol

0

0.00 -

0

200

400

800

800

1000

1200

Carbon number x 100

Fig. 5.9. Relationships of log k to carbon number X 100 for the three homologous series nitroalkanes, alkm-2ones and alkyl aryl ketones. Eluent, methanol-buffer (pH 7.0) (40:60). From ref. [64] with the permission of Elsevier Science.

The correction of RI-values in the alkyl aryl ketone scale was effective for drugs, eluting in the elution range of standards. However, a number of basic drugs eluted much TABLE 5.3 UNCORRECTED AND CORRECTED I VALUES OF BARBITURATES FOUND ON FIVE DIFFERENT ODs-SILICA COLUMNS Drug

Barbital (579) Cyclobarbital (762) Butobarbital (792) Amobarbital (875) Pentobarbital (890) Secobarbital (930) Phenobarbital (676) Talbutal (835) Methohexital (1001)

Column

Ua C U

C

U C U

C U

C U

C

1

2

3

4

5

572 604 728 759 762 791 846 870 858 88 I 906 928 640

556 602 716 760 750 789 832 868 852 887 892 926 624

520 606 678 762 712 792 796 879 811 891 854 950 580

534 587 708 761 742 79 1 830 878 850 894 890 933 615

582 590 786 792 87 1 875 886 889 927 929 666

810

798

760

790

830

980

970

896

960

1001

From ref [67] with the permission of Elsevier Science. Columns: I, Serva 3 pm; 2, Serva 5 pm; 3, Spherisorb 5 pm; 4, Partisil 10pm; S, Hypersil S p m uncorrected; c, corrected.

References pp. 205-207

190

Chapter 5

TABLE 5.4 RESULTS OF CORRECTION OF RI VALUES OF DIURETIC DRUGS D w

Ref. RI

RI f SD

Chlorothiazide (st) Quinethazone Hydrochlorothiazide Chlorthalidone Hydroflumethiazide Trichloromethazide Methyclothiazide (st) Metolazone Chlorexolone Clopamide Benzthiazide Mefruside Cyclothiazide 1 Cyclothiazide 2 Polythiazide Bendrofluazide Cyclopenthiazide (st)

527 546 555 616 632 730 754 785 834 835 855 862 873 885 913 918 929

506 f 25.9 524 f 23.7 536 f 24 597 f 26.5 618420.7 721 f 19.8 746 f 19.6 773 f 20.3 821 f 20.3 779 f 63.8 846 f 20.8 853 f 16.2 865 f 21.1 875 f 20.1 907 f 19 912 f 17.4 918 f 20.9

RIGf SD

545 f 3.8 556 f 3.3 614 f 3.7 634 f 2.1 730 f 0.7 781 f 2.9 830f4.1 802 f 52.8 855 f 2.1 862 f 4.6 875 f 4.4 885 f 2 916f4.1 923 f 5.3

Original, uncorrected data taken from ref. [57].

earlier than the first homologue in the alkyl aryl ketone scale, and for these compounds the correction could not be applied. The alkyl aryl ketone scale was therefore of limited use for general toxicological screening (Fig. 5.1 1). In an interlaboratory study of corrected retention indices, the 1-nitroalkane scale and the same gradient conditions as in the above-mentioned study were applied [70]. Sixtytwo selected acidic, neutral and basic drugs were examined in two laboratories using different instrumentation. In one laboratory, four ODS columns were used: Chromspher 100 x 3 mm, Superspher 125 X 4 mm, Novapak 100 x 5 mm and Lichrospher TABLE 5.5 ACCURACY OF UNCORRECTED AND CORRECTED RI VALUES FOR ALL EXAMINED DRUGSa Column

dRI

dRIC

27 f 29 31f38 -11f28 5 f 17 Ref. -2f 17 37f 16 31 f 15 17f 18

9 f 18 0 f 23 5f20 1fl7 Ref. -1f8 O f 13 5f15 -2f 13

~~~

Lichrosorb RP 18 Lichrosorb RP 18 Nucleosil C 18 Superspher RP 18 Superspher RP 18 Superspher RP 18 ChromSpher C 18 ChromSpher C 18 ChromSpher C 18

From ref. 68 with the permission of Preston Publications. aAccuracy means nearness to reference data. dRI and dRICrepresent the arithmetic mean difference f SD between the experimental and reference value. Values were calculated for each column separately.

Nitroalkanes and secondary retention standards

191

TABLE 5.6 COMPARISON OF THE DISCRIMINATION NUMBERS (DN) OF VARIOUS METHODS FOR REPORTING THE RETENTION OF DRUGS Source

Drugs investigated

Columns used

Retention factor

DN value

Baker ef al. [53]

7 acidic and neutral drugs

6 ODS columns (4 identical, 2 different)

t

Smith et al. [56]

10 barbiturates

10 different ODS columns

Gill et af. [59]

9 barbiturates

10 identical ODS columns

3.9 4.6 5.3 2.1 4.3 14.8 22.9 10 16 34 44 55 64 2.6 6.6 6.8 24.6 13.6 25.0 14.6 22.9

f RI k RI kC RIC t

k RI Rt

RI Bogusz [67 J

6 barbiturates

5 different ODS columns

Present study [61]

Series I(17 drugs)

9 different ODS columns

Series 11 (20 drugs)

kC k RI kC RIC RI RIC

RI RIC

From ref. [68] with the permission of Preston Publications)

125 X 4 mm. In the second laboratory, a Lichrospher 125 X 4 mm column (the same batch as in the first laboratory) was used. Separate sets of secondary standards were used for acididneutral and for basic drugs. The variabilities of uncorrected and corrected retention index values were expressed in terms of mean standard deviation (MSD). For each substance, the mean value and standard deviation was calculated for the RI-values determined on various columns, and the deviations were averaged. The design of this study has enabled us to inspect several aspects of variability, which are shown in Table 5.9. The results of this study indicate that the procedure of correcting RIs by utilizing a set of drugs with pre-assigned RI values is a valuable tool for handling differences in column characteristics, regardless of interlaboratory conditions. After correction with secondary standards, the same reproducibility was achieved for various brands as for the same batch of column packing. Also, the limits of reproducibility, expressed as a MSD of about 10 RI units were demonstrated. In a fbther study of the application of secondary standards for HPLC screening [3 11, a one-step gradient elution in acetonitrile-triethylammonium phosphate buffer 25 mM (pH 3 .O) was applied. These elution conditions are more robust for interlaboratory use, for two reasons: a simple gradient profile was applied (0-70% of acetonitrile in 30 min, 5 min at 70%), - triethylammonium phosphate buffer is commercially available as a 1 M solution References pp. 205-207

192

Chapter 5

TABLE 5.7 UNCORRECTED RI VALUES FOR BASIC DRUGS, L-NITROALKANE SCALING Mean f SDa

Column

Cocaineb Diphenhydramine Doxepin Promethazine Desipramineb Imipramine Nortriptyline Maprotiline Methadone Amitriptyline Perphenazine Chlorpromazine Fluphenazine Chlorprothixene Triflupromazineb Thioridazine

1

2

3

4

5

6

236 328 363 389 390 408 410 408 430 429 460 47 1 506 485 5 10 55 1

193 300 325 344 358 364 374 374 381 379 403 410 440 422 453 47 1

230 332 360 387 395 408 410 410 421 428 452 468 482 482 506 542

237 328 3 72 40 1 419 425 436 436 439 444 467 472 5 02 486 511 524

240 328 372 404 420 428 437 437 442 448 473 479 510 491 520 534

230 337 382 406 423 433 442 442 442 457 482 489 508 501 517 547

228f 17 3 2 6 i 13 362*20 389f23 4013Z25 4113~25 418i26 418i26 427+23 431 i 2 8 456i28 465f28 491 i 2 7 478 f 28 503 f 25 528 f 30

From ref. [69] with the PERMISSION of Preston Publications. aMean SD = 24.5 (calculated without correction standards); DN value = 6.1 (calculated without correction standards). bCorrection standards.

TABLE 5.8 CORRECTED RI VALUES FOR BASIC DRUGS, 1-NITROALKANE SCALEa Column

Diphenhydramine Doxepin Promethazine Imipramine Nortriptyline Maprotiline Methadone Amitriptyline Perphenazine Chlorpromazine Fluphenazine Chlorprothixene Thioridazine

Mean i SD

1

2

3

4

5

6

328 365 393 41 1 413 41 1 432 43 1 460 470 502 483 544

337 362 381 402 414 414 422 420 448 456 491 470 527

332 360 387 408 410 410 427 428 452 468 482 482 542

312 352 379 402 415 415 419 425 453 459 495 476 521

311 351 380 404 414 415 420 426 454 461 495 474 522

321 360 380 406 417 418 417 434 464 472 494 486 540

From ref. [69] with the permission of Preston Publications. aCorrection standards: cocaine. desipramine, and triflupromazine (as Table 5.7). bMean SD = 5.8; DN value = 18.0.

3 2 4 i 11 358 f 6 383 i6 406 f 4 414 2 414 f 3 423 f 6 427 f 5 455 f 6 464 7 493 f 7 479 f 6 533*11

*

*

Nitroalkanes and secondary retention standards

193

RI

R

600

60C

.........*..--...-...*--..) .......*....

...*a

500

5OC

400

40C

300

300

-

'-................................ ........ .-.---..-....*-..... ...... .......-*.....--...

200 1

I

I

I

1

I

I

I

I

I

1

2

3

4

5

6

1

2

3

4

Column

I

5

I

6

Column

Fig. 5.10. The uncorrected (left) and corrected (right) RI values of drugs analyzed on six ODS-silica columns, 1-nitroalkane scale. The broken lines join the RI values obtained for individual drugs. Solid lines indicate the drugs used as correction standards. From ref. [69]with the permission of Preston Publications.

(cat. 90362, Fluka Chemie AG, CH 9470 Buchs, Switzerland). The use of this reagent may minimize the differences in the composition of the mobile phase, particularly in the concentration of amine modifier, which is an important source of interlaboratory variability. Two sets of drugs were chosen in this study as secondary RI-standards: 8 acidicheutral drugs and 10 basic drugs. The retention indices of these drugs were determined over 3 months under the specified conditions on the 1-nitroalkane scale using a Superspher RP-18 column (endcapped) (Fig. 5.12). The measured RI-values of drugs were then used as a secondary retention index scale for use with other columns under intra- and interlaboratory conditions (Fig. 5.13, 5.14, Table 5.10). The operational conditions applied covered the whole elution range for toxicologically relevant compounds and allowed the separation of acidic, neutral and basic drugs with the same efficiency (Fig. 5.15). Refirences pp. 205-207

194

Chapter 5

RI

RI

1000

100C D

900

900

800

800

700

700

600

600 1

I

I

1

I

I

1

2

3

4

5

6

Column

Column

Fig. 5.11. The uncorrected (lee)and corrected (right) RI values of drugs analyzed on six ODS-silica columns, alkyl aryl using ketone scale. The broken lines join the RI values obtained for individual drugs. Solid lines indicate the drugs used as correction standards. From ref. [69] with the permission of Preston Publications.

TABLE 5.9 OVERVIEW OF MEAN STANDARD DEVIATIONS (MSD) OF RI-VALUES, FOUND WITH DIFFERENT COMBINATIONS OF COLUMNS Columns

MSD (RI units)

All five, two labs Four, one lab Two of same batch, two labs

25.9 26.8 11.8

10.0 9.8 8.9

Nitroalkanes and secondary retention standards

195

3

a

E

10

20

30

Fig. 5.12. Chromatogram of the I-nitroalkane mixture in a gradient of acetonitrile-triethylammonium phosphate buffer (pH 3.0). From ref [31] with the permission of Preston Publications.

A database comprising the retention index values of 225 drugs and other compounds was established under these conditions and used in routine toxicological casework (Figs. 5.16,5.17). The final concept of standardization was proved in an interlaboratory study involving two laboratories [7 I]. In one laboratory, three different reversed-phase silica columns were used: Superspher RP-18 (used as the reference column for the database), Lichrospher Select B (octyl) and Nucleosil ODs-AB. In the second laboratory, four columns were used: Lichrospher RP 18, Enca Pharm RP- 18, Synchropak SCD (octyl) and Inertsil ODs-2. All these columns were defined by the manufacturers as endcapped and basedeactivated. A separate study demonstrated that these columns were suitable for screening toxicological analysis of acidic and basic drugs [50]. The elution conditions were identical to those in the previous paper to establish a HPLC database [31]. The retention index values of 25 acididneutral and 22 basic drugs were determined in both laboratories on all seven columns using 1-nitroalkanes as primary RI standards and mixtures of acididneutral and basic drugs as secondary RI standards. The 1-nitroalkane mixture and drugs mixtures were co-analyzed with each series of determinations, and the actual calibration curves were used for calculation of RI-values (Figs. 5.18, 5.19). The RI-values determined with primary and secondary standards were compared with the listed values from a previously published database [31]. The RI-values of individual drugs determined on the 1-nitroalkane scale on different columns showed large deviations from the listed data, ranging from -188 to +79 units. Also, the variations in RI-values between the columns were large, with a mean SD value of 44.4 units. In contrast, the RIvalues determined using the secondary scale differed by less than *lo units from the listed values in 80% of cases, and less than *20 units in 95% of cases. The mean SD for all the drugs and all columns was 10.3 RI units. The mean deviations from the listed values were also calculated for each column separately (Table 5.1 1). Again, a striking improvement in accuracy (expressed as mean deviation from the listed values) and precision (expressed as SD of the mean deviation) was observed when References pp. 205-207

196

Chapter 5

Fig. 5.13. Chromatogram of the acidicheutral mixture of drug standards.

the secondary retention index standards were used. The largest deviations from the listed values were observed for the Synchropak column, which was prepared with a different type of silica, different stationary phase (octyl) and was of different length to the reference column. For all other columns, the standard deviation was in the range of 8-10 RI

Nitroalkanes and secondary retention standards

197

PI

e ............:........... .......A-

4

i

-4

B %

E C

Q CI CI

0 Y

E 0

I

u

1 -

! I

@4 a'

m ' . . . . l . . . . l . . . . m

;e ;A

10

Time

. . . . . . . . . . . . . . . . .. 8

---

20

30

35 rnin

Fig. 5.14. Chromatogram of the basic mixture of drug standards.

units, and was comparable to the standard deviation value of 8.1 observed for six identical reversed-phase columns (Lichrospher Select B), examined in two laboratories under the same conditions [72]. Hence, the standardization procedure allowed us to compensate for the differences in selectivities of almost all reversed-phase columns. It should be References pp. 205-207

198

Chapter 5

TABLE 5.10 RETENTION INDEX VALUES OF SECONDARY DRUG STANDARDS

Acididneutral standards Paracetamol Barbital Bra1lobarbital Pentobarbital Secobarbital Clobazam Indomethacine Prazepam

234 f 1.9 287 f 1.8 359 f I .6 405 f 1.9 437 f 1.6 484 f 2.2 610 f 3.3 648 f 3.5

Basic standards Morphine Chloroquine Benzoylecgonine Cocaine Diphenhydramine Haloperidol Amitriptyline Thioridazine Meclozine Amiodarone

198 f 1.9 265 f 0.5 295 f 0.8 336 i 1.9 385 f 2.0 409f 1.7 446 f 2.7 504 f 3.9 601 f 3.0 762 f 3.0 ~~

stressed that the columns used in this study were carefully endcapped and claimed to be base-deactivated. On the base of the inter- and intralaboratory variability determined, a search window of h30 RI units was recommended for automatic search procedures. The HPLC database established with secondary retention index standards, was expanded and now comprises RI-values and spectral data for about 400 compounds: thera1

a

W

W W Z N

Z

+ia

3

4 0200-

1

10

20 Time

Cmin.)

30

1 1

Fig. 5.15. Separation of various acidic and basic drugs in gradient of acetonitrile-triethylammonium phosphate buffer. Superspher RP 18 column 5 pm,125 x 4 mm.

Nitroalkanes and secondary retention standards

199

LC A 220,4

550,100 o f WUQ-A0SA. R e s u l t s o f s p e c t r a l 1 i b r a r y sear.ch L f b r a r y : TOXLIB-TRIETHYLAMINE

D

w

z

U

Match

> 990

10 Time

20 (min.)

30

Fig. 5.16. Chromatogram of the street heroin sample, generated after a spectral library search. From ref. [31] with the permission of Preston Publications.

peutic and illicit drugs, their metabolites, herbicides, insecticides, fungicides and endogenous compounds [73]. The €U-values for about 200 compounds were determined earlier, using the same operational conditions, but different instrumentation [3 13. The differences between the €U-values for the same compounds in these two data collections were, as a rule, smaller than * l O units [73].

5.2.2.2Assessment of identfication potentials of a HPLC/RI system The identification potential of a HPLC system, based on €U-values and UV data, was checked using two parameters: discriminating power (DP) [74] and mean list length (MLL) [75,76]. These methods allow us to assess the identification potential of various methods, such as TLC, GC, UV or HPLC and their combinations in order to find out the most powerful identification system [77]. Discriminating power was defined as the probability of separation of randomly selected substances in a given system. The substances are regarded as unseparated (matched), when the difference between them is smaller than the error window, estimated as three times the interlaboratory standard deviation. The higher the identification power of the system, the higher is the DP value. The maximum value is 1. The DP value can be calculated according to the formula DP = 1- 2M/N(N- 1) where M is the total number of matches and N is the total number of drugs. In the calculation of MLL, the assumption is made that the collection of any identification parameter follows a normal distribution. When any unknown substance is comReferences pp. 205-207

200

Chapter 5

'1

L C A 220.4

550,100

of A

184/90R.D

-WEBIMD

- AUIUPSY BLOOD SPIKED UIlM l'and 1 - APROBARBITAL IN SAMPLE 1' - APROBARBITAL 2 mp/l BUOD 2 - SECBUIUBARBITAL IN SAWLE 2' - SECBVIIIBARBITAL 2 4 1 BLOOD B

80

1

70:

2'

60 3

a

50

E 4 0-

' 0

. . . . . . . . . . . . . . . . . . . . . . 10

20 Time

30

40

(min.)

Fig. 5.17. Samples from a case of aprobarbital + secbutobarbital poisoning. The drugs were identified in the library search procedure. From ref. [3 11 with the permission of Preston Publications.

pared with the database, it may qualify for identification at a preset level of probability. All standards which qualifjr for identification form a list. The shorter the list, the more powerfid is the identification method. The procedure of creating a list may be repeated for all substances in the database, and after averaging, the mean value which is the mean list length (MLL) may be calculated as a measure of the identification potential of the system. The calculation of DP and MLL values was performed for 372 substances stored in a HPLC database [78]. The HPLC €U-values and UV data (the wavelength of the absorbtion maxima with the highest wavelength) were taken for calculation. Table 5.12 shows DP and MLL values calculated for single parameters and for their combination. The identification potential of retention indices or UV spectral data, regarded separately, was pretty poor. However, after combining both parameters, as is normally done on-line in a HPLC search routine, the system became very selective, allowing the identification of most compounds in the database. Comparison of the identification power of a HPLC system with other systems (GC, TLC, UV and their combinations) showed that the combination of HPLC-RI and spectral data is as powerful as the combination of GC with UV (off line) or with TLC [77,78]. Fig. 5.18. Calibration curves, prepared with nitroalkanes, acidic and basic drug mixtures for Superspher 18, Lichrospher Select B RP 8, Nucleosil C18 Al3 and for Inertsil ODS-2 columns. From ref. [71, p.13471 by courtesy of Marcel Dekker, Inc.

8

Y

I Y

3

c--

0

c

3

-

cn

Y bl

%

b

Nitroalkanes and secondary retention standards.

*

References pp. 205-207

*

C r " YI

0

e

-

cn

Y

-51

- 3

I- 2

- 2 n

E

20 1

R

SYNCHROPAK

SYNCHROPAK

CALlBRATION CURVES

CALIBRATIONCURVES

-" "I

I

c t

2

I

N 0 N

R e t

e n

e n

t

t

i

1

0

0

n

n i

i n d

n

c

d e

I

I

1

0

5

0

10

NitmPltnaa

e

m

I

20

25

+Addkdny

ENCAPHARM

LICHROSPHER CALIBRATION CURYES

R

.......................................................

e t e

35

40

+Baskdrug

CALIBRATIONCURVES

I

30

Rdatim tim

-

"

IS

...........I

9.:

n t

i 0

n I

n d e x

-1

I

f

200

I

0' 0

5

10

IS

20

25

M

35

Rdation tim

-NitroPlLnna

+Mdkdnrg.

+JhskdNga

40

Nitroalkanes and secondary retention standards

203

TABLE 5.1 1 THE DEVIATIONS FROM THE LIBRARY RI-VALUES, DETERMINED WITH 1-NITROALKANES AND DRUG STANDARDS: MEAN f SD FOR 47 DRUGS EXAMINED ON 7 COLUMNS Columns Super RI(N02) mean dev. SD-NO;! RID(NOd mean dev. SDD(N02)

-7.4 8.8 0.5 8.7

Lichro

Nucleo

-14.2 11.6 -0.7 9.0

-4.3 17.3 -5.3 8.2

Inert 3.7 15.4 -2.6 8.9

ENCA

Select

-8.0 16.3 -1.8 9.5

18.5 12.6 -0.5 9.2

Sync-1

Sync-2

-92.6 23.2 4.3 15.1

-87.6 21.4 1.5 16.5

From ref. [71] with the permission of Marcel Dekker, Inc.

5.2.2.3Influence of a biological matrix on the identijkation potential of a HPLC/RI system All existing HPLC databases were established with pure drugs and do not contain retention and spectral data concerning non-drug matrix substances. On the other hand, any isolation procedure, applied to any kind of biological material, leads unavoidably to the co-extraction of various body constituents, which may influence the separation or chromatographic mobility of toxicologically relevant compounds, affecting the identification potential of the procedure. This is particularly true in the case of TLC, where the chromatographic behaviour of substances may be greatly affected by a co-extracted biological matrix (79-81). In the case of GC, the influence of a biological matrix on RI-values was negligible [82]. A collection of GC-retention index values for 296 non-drug substances likely to be encountered in biological extracts was published [83] and implemented into a comprehensive GC database [ 5 ] . A recent study on the influence of biological matrices [84] followed two objectives: to study the occurrence and identity of non-drug matrix substances, detectable with a routine HPLC-screening procedure; to examine the influence of the biological matrix on the chromatographic behaviour and detectability of selected acididneutral and basic drugs. Samples of autopsy blood, liver and serum were extracted at acidic and basic pH and examined by means of a HPLC screening procedure, used for establishing the database [3 1,733. In the extracts, 2-phenethylamine, tryptamine and indole were identified. Furthermore, four unidentified substances were found regularly in acidic extracts, and five in basic extracts. The RI-values and UV spectra of these substances were implemented in the database. In the next step of this study, known amounts of selected acidic and basic drugs were added to the blank biological extracts and the retention index values were compared with those obtained with pure drugs and with listed values. The differences between the listed RI-values and observed values in the presence of a biological matrix were not greater than between the listed RI-values and values observed for pure drugs (Fig. 5.20).

Fig. 5.19. Calibration curves, prepared with nitroalkanes, acidic and basic drug mixtures for Synchropak SCD, flow 0.8 ml/min and 1 ml/min, for Encapharm RP 18 and for Lichrospher RP 18 columns. From ref. [71, p. 13481 by courtesy of Marcel Dekker, Inc.

References pp. 205-207

204

Chapter 5

TABLE 5.12 DP AND MLL VALUES FOUND FOR HPLC IDENTIFICATION SYSTEM (372 SUBSTANCES) Identification parameter

DP

MLL

Retention index (RI) UV-Maximum RI + UV

0.840 0.804 0.993

29.645 36.301 1.277

However, some of the substances examined were not detected in the presence of extracts, due to co-eluting peaks of matrix substances. The study showed that the retention index database, established with pure drugs, may also be used for identification of drugs extracted from biological material. The purity of the extract was, however, of critical importance.

5.2.3 Standardization of detection for toxicological screening procedures

The studies reviewed in the previous sections were devoted to establishing the interlaboratory use of databases of standardized retention parameters. No effort was undertaken to standardize the detection. In HPLC with diode array detection, the registered UV spectrum of a given peak becomes a second, very powerful parameter of identification, as was demonstrated in the study mentioned above [78]. However, up to now no reports have

. 500

O

-

o

q

n

3

o

200 Rlrefl

-E+

Rl(pure1 Hydromorphone Acepromazine

>:

n

J Rl(serum) Acebutolol

* Chroprothixene

j RUblood)

] Rl(liver1

++ Dibenzepin -4-

Terfenadine

Fig. 5.20. Retention indices of selected basic drugs: Listed values, values observed for pure drugs and in the presence of various biological matrices.

Nitroalkanes and secondary retention standards

205

demonstrated the transfer of a stored spectral library from one chromatographic system to another. This problem was studied by Binder et al. [85], who investigated the possibility of distinguishing closely related and closely eluting drugs (amphetamine and phentermine), using a multi-wavelength detector, equipped with a rotating holographic grating and a single-diode sensor (BarSpec, Revohot, Israel). The reliable transfer of spectral data between detector units was possible when the following conditions were met: - isocratic separation was applied; - the pH of the mobile phase was exactly controlled (k0.02pH units); the acetonitrile concentration was controlled to within &0.5%; - the detector units were calibrated within *1 nm. The data transfer was then possible only between the same make of detector. Ryan [86,87] have investigated the identification of ten barbiturate spectra, taken at pH 3.0 with a diode array detector (Varian, Walnut Creek, USA). These compounds represent a “worst case scenario” because of unspecificity and similarity of U V spectra. Nevertheless, differentiation of spectra was possible on the basis of “Purity ParameterTM”,which was defined as the average wavelength of the spectrum weighted by the square of the absorbance at each wavelength of the spectrum over the spectral range. The limitations of this procedure are similar to those of the previous study: it is applicable to only one commercially available system and is strongly dependent on its optical performance. The influence of co-eluted matrix substances on the efficiency of identification was not studied. As a concluding remark it may be stated that up to now, the exchange of spectral data between different commercially available multi-wavelength detector systems is not possible. Each manufacturer has developed his own software, which is incompatible with others. This makes the establishment of a common database of spectra or their derivatives not feasible at the moment. From a practical point of view, this means that the interlaboratory use and exchange of an automatic, on-line identification routine on the basis of retention parameters and UV spectra is possible only among users of the same commercial identification system. 5.3 REFERENCES 1

6 7

8 9 10

A.C. Moffat (sen. cons. Ed.), Clarke’s Isolation and Identification of Drugs, Pharmaceutical Press, London, 1986. T.A. Gough (Ed.), The Analysis of Drugs of Abuse, Wiley, Chichester, UK, 1991. A.S. Curry (Ed.), Analytical Methods in Human Toxicology, Verlag Chemie, Weinheim, 1986. 1. Sunshine (Ed.), Handbook of Analytical Toxicology, CRC Press, Cleveland, OH, 1969. DFG-TIAFT, Gas Chromatographic Retention Indices of Toxicologically Relevant Substances on Packed or Capillary Columns with Dimethylsilicone Stationary Phases, VCH, New York, 1992. DFG-TIAFT, Gas Chromatographic Retention Indices of Solvents and Other Volatile Substances for Use in Toxicological Analysis, VCH, New York, 1992. DFG-TIAFT, Thin-Layer Chromatographic Rf Values of Toxicologically Relevant Substances on Standardized Systems, VCH, New York, 1992. E. Kovats, Helv. Chim. Acta, 42 (1958) 1915. J.P. Franke, J. Wijsbeek and R.A. de Zeeuw, J. Forens. Sci., 35 (1990) 813. J.P. Franke, J. Wijsbeek, R.A. de Zeeuw, M.R. M6ller and H. Niemeyer, J. Anal. Toxicol., 12 (1988) 20.

References pp. 205-207

206 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64

Chapter 5 J.P. Franke, M. Bogusz and R.A. de Zeeuw, Fresenius J. Anal. Chem., 347 (1993) 67. M. Bogusz, J.P. Franke, R.A. de Zeeuw and M. Erkens, Fresenius J. Anal. Chem., 347 (1993) 73. I. Jane, J. Chromatogr., 109 (1975) 37. B.B. Wheals, J. Chromatogr., 187 (1980) 65. B.A. Bindlingmeyer, J.K. Del Rios and J. Korpi, Anal. Chem., 54 (1982) 442. B. Law, R. Gill and A.C. Moffat, J. Chromatogr., 301 (1984) 165. R. Gill, M.D. Osselton, R.M. Smith and T.G. Hurdley, J. Chromatogr., 386 (1987) 65. R. Gill, M.D. Osselton and R.M. Smith, J. Pharm. Biomed. Anal., 7 (1989) 447. R.M. Smith, T.G. Hurdley, R. Gill and M.D. Osselton, J. Chromatogr., 398 (1987) 73. R.M. Smith, T.G. Hurdley, J.P. Westlake, R. Gill and M.D. Osselton, J. Chromatogr., 455 (1988) 77. R.J. Flanagan, G.C. Storey, R.K. Bhamra and I. Jane, J. Chromatogr., 247 (1 982) 15. P.S.B. Minty, Proc. 29th TIAFT Meeting (B. Kaempe, Ed.), Copenhagen, 1991, p. 449. R.M. Smith, J.P. Westlake, R. Gill and M.D. Osselton, J. Chromatogr., 514 (1990) 97. R.M. Smith, J.P. Westlake, R. Gill and M.D. Osselton, J. Chromatogr., 592 (1992) 85. T. Daldrup, F. Susanto and P. Michalke, Fresenius Z. Anal. Chem., 308 (1981) 413. P.M. Kabra, B.E. Stafford and L.J. Marton, J. Anal. Toxicol., 5 (1981) 177. E.M. Chan and S.C. Chan, J. Anal. Toxicol., 8 (1 984) 173. E.I. Minder, R. Schaubhut, C. Minder and D.J. Vonderschmitt, J. Chromatogr., 419 (1987) 135. K. Jinno, M. Kuwajima, M. Hayashida, T. Watanabe and T. Hondo, J. Chromatogr., 436 (1988) 11. A. Turcant, A. Premel-Cabic, A. Cailleux and P. Allain, Clin. Chem., 37 (1991) 1210. M. Bogusz andM. Wu, J. Anal. Toxicol., 15 (1991) 188. E.M. Koves and J. Wells, J. Forens. Sci., 37 (1992) 42. A. Tracqui, P. Kintz, P. Kreissig and P. Mangin, J. Liq. Chromatogr., 15 (1992) 1381. H. Kslferstein and G. Sticht, Beitr. Gerichtl. Med., 44 (1986) 253. D.W. Hill and K.J. Langner, J. Liq. Chromatogr., 10 (1987) 377. S.R. Binder, M. Regalia, M. Biaggi-McEachern and M. Mazhar, J. Chromatogr., 473 (1989) 325. B.K. Logan, D.T. Stafford, I.R. Tebbett andC.M. Moore, J. Anal. Toxicol., 14 (1990) 154. A.P. Goldberg, Anal. Chem., 54 (1982) 342. T. Daldrup and B. Kardel, Chromatographia, 18 (1984) 81. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr., 225 (1986) 75. D. Chan Leach, M.A. Stadalius, J.S. Berus and L.R. Snyder, LC-GC Int., I (1988) 22. J. Nawrocki, Chromatographia, 31 (1991) 177. J. Nawrocki, Chromatographia, 31 (1991) 193. A. Berthod, J. Chromatogr., 549 (1991) 1. J. Kohler, D.B. Chase, R.D. Farlee, A.J. Vega and J.J. Kirkland, J. Chromatogr., 352 (1986) 275. L.C. Sander and S.A. Wise, LC-GC Int., 3(6) (1990), 24. R.E. Majors, LC-GC Int., 4(3) (1990) 12. R.E. Majors, LC-GC Int., 6 (1993) 196. H.H. Freiser, M.P. Nowlan and D.L. Gooding, J. Liq. Chromatogr., 12 (1989) 827. M. Bogusz, M. Erkens, R.D. Maier and I. Schrdder, J. Liq. Chromatogr., 15 (1992) 127. R.M. Smith, Adv. Chromatogr., 26 (1987) 278. J.C. Baker and C.Y. Ma, J. Chromatogr., 169 (1979) 107. J.C. Baker, L.A. Cates, M.D. Corbett, J.W. Huber and D.L. Lattin, J. Liq. Chromatogr., 5 (1982) 829 R.M. Smith, J. Chromatogr., 236 (1982)313., R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, Chromatographia, 19 (1984) 401. R.M. Smith, T.G. Hurdley, R. Gill and A.C. Moffat, Chromatographia, 19 (1984) 407. R.M. Smith, G.A. Murilla, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr., 384 (1987) 259. R.M. Smith, G.A. Murilla and C.M. Burr, J. Chromatogr., 388 (1987) 37. R. Gill, A.C. Moffat, R.M. Smith and T.G. Hurdley, J. Chromatogr. Sci., 24 (1986) 153. M. Bogusz, R. Aderjan, J.P. Franke and R.A. de Zeeuw, J. Leg. Med., 98 (1987) 263. M. Bogusz and R. Aderjan, J. Chromatogr., 388 (1988) 37. R. Aderjan and M. Bogusz, J. Chromatogr., 454 (1988) 345. M. Bogusz, J. Anal. Toxicol., 15 (1991) 174. R.M. Smith and N. Finn, J. Chromatogr., 537 (1991) 51.

Nitroalkanes and secondary retention standards 65

66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87

207

DFG-TIAFT, Gas Chromatographic Retention Indices of Toxicologically Relevant Substances on SE-30 or OV-1, VCH, Weinheim, 1985. W.H. Anderson and D.T. Stafford, J. High Resolut. Chromatogr. Chromatogr. Commun., 6 (1983) 247. M. Bogusz, J. Chromatogr., 387 (1987) 404. M. Bogusz and R. Aderjan, J. Anal. Toxicol., 12 (1988) 67. M. Bogusz, G. Neidl-Fischer and R. Aderjan, J. Anal. Toxicol., 12 (1988) 325. M. Bogusz, R. Aderjan, J.P. Franke, J. Wijsbeek and R.A. de Zeeuw, in: Forensic Toxicology, Scottish Academic Press, Edinburgh, 1992, p. 90. M. Bogusz, M. Erkens, J.P. Franke, J. Wijsbeek and R.A. de Zeeuw, J. Liq. Chromatogr., 16 (1993) 1341. M. Bogusz, M. Erkens, J.P. Franke and R.A. de Zeeuw, Proc. 30th TIAFT, Fukuoka, 1993, p. 49. M. Bogusz and M. Erkens, J. Chromatogr., 674 (1994) 97. A.C. Moffat, P. Owen and C. Brown, J. Chromatogr., 161 (1978) 179. P.G.A. Schepers, J.P. Franke and R.A. de Zeeuw, J. Anal. Toxicol., 7 (1983) 272. J.P. Franke, R.A. de Zeeuw and P.G.A. Schepers, J. Forens. Sci., 30 (1985) 1074. J.P. Franke and R.A. de Zeeuw, in: The Analysis of Drugs of Abuse (T.A. Gough, Ed.), Wiley, Chichester, UK, 1991, p. 93. M. Bogusz and R.D. Maier, J. Anal. Toxicol., (1994) in press. M. Bogusz, M. Klys, J. Wijsbeek, J.P. Franke and R.A. de Zeeuw, J. Anal. Toxicol., 8 (1984) 149. M. Bogusz, J. Gierz, R.A. de Zeeuw and J.P. Franke, J. Chromatogr., 342 (1985) 241. M. Bogusz, J.P. Franke, J. Wijsbeek and R.A. deZeeuw, J. Anal. Toxicol., 10 (1986) 245. M. Bogusz, J. Wijsbeek, J.P. Franke and R.A. de Zeeuw, J. Anal. Toxicol., 9 (1985) 49. J.D. Ramsay, T.D. Lee, M.D. 0sseltonandA.C. Moffat, J. Chromatogr., 184 (1980) 185. M. Bogusz and M. Erkens, J. Anal. Toxicol., (1994) in press. S.R. Binder, A.K. Adams, M. Regalia, H. Essien and R. Rosenblum, J. Chromatogr., 550 (1991) 449. T.M. Ryan, J. Liq. Chromatogr., 16 (1993) 33. T.M. Ryan, J. Liq. Chromatogr., 16 (1993) 315.

This Page Intentionally Left Blank

R.M. Smith (Ed.), Retention and Selectivity in Liquid Chromatography Journal of Chromatography Library, Vol. 57 0 1995 Elsevier Science B.V. All rights reserved

209

CHAPTER 6

Identification using retention indices in gradient HPLC P. Kuronen Department of Chemisty, University of Helsinki, Vuorikatu 20, FIN-00014,Finland

6.1 INTRODUCTION

High-performance liquid chromatography (HPLC), one of the most commonly used instrumental analysis techniques, has primarily been employed for quantitative analysis, although it can be used for the identification of substances. Identification methods are based on the retention behaviour of substances and/or the use of HPLC coupled on-line with other instrumental methods, usually diode array detector (DAD), mass spectrometry (MS) or Fourier transform infrared spectrometry (FTIR), which are the most popular coupled techniques in HPLC separation. The sensitivity of conventional retention parameters (the retention time, f R and the capacity factor, k) to variations in experimental conditions has generated the need for improvement in the reproducibility of retention data, which is required for the reliable identification of substances by comparing experimental and published data [l]. The introduction of retention indices (RI) in HPLC has created the possibility of better interlaboratory comparability of retention data [l]. The idea of the retention index, used widely in GC [2-4], relates the retention behaviour of substances to a series of selected retention standards. A number of studies on the application of retention indices in HPLC have been published during the last 15 years. Among several RI standards proposed in HPLC [l], the two most widely applied homologous series of standards are the 2-alkanones [S] and 1-phenyl-1-alkanones (alkyl aryl ketones) [6,7], because they are highly stable, readily available and detectable with a UV detector. Subsequently a new series of homologous 1-[4-(2,3-dihydroxypropoxy)phenyl]1-alkanones has been synthesized and evaluated by Kuronen as retention index standards for reversed-phase HPLC (RP-HPLC) [8-lo]. Further, Bogusz and Aderjan have suggested the use of 1-nitroalkanes as alternative retention index standards for RP-HPLC [l 11. All four series of standards are usable across a wide range of eluent composition. The 1 -phenyl- 1 -alkanones and 1 -[4-(2,3-dihydroxypropoxy)phenyl]-1-alkanones have the advantage of showing high UV sensitivity over a relatively wide wavelength range (200References pp. 230-233

210

Chapter 6

300 nm) [lo], whereas l-nitroalkanes have strong UV absorption in only a narrow nonselective wavelength range (200-220 nm) and very weak absorption above that [l 11. 2Alkanones, l-[4-(2,3-dihydroxypropoxy)phenyl]-l-alkanones and l-nitroalkanes allow the indexes to be calculated for earlier eluting, more polar compounds than l-phenyl-lalkanones. Bogusz et al. [12-191 have also studied the use of corrected RI values to compensate for differences in the separations of neutral, acidic and basic drugs on different brands of CI8columns. This correction method appeared useful, however, only in the case of a uniform group of compounds. Besides homologous series, polycyclic aromatic hydrocarbons (PAHs) [20,211 and their nitrated derivatives [22] have found applications as retention index standards for the identification of PAHs and nitro-PAHs in environmental samples. The retention index concept in HPLC has been verified in isocratic systems, and relatively few studies have been carried out using gradient elution. However, isocratic HPLC is suitable when compounds with similar retention behaviour are to be studied, but not for the screening of widely differing compounds. For these, solvent-programmed retention indices, as introduced by Kuronen in 1982 [7], are more useful. If isocratic retention indices were used, runs would have to be made at several solvent compositions, with the production of several sets of indices. For example, the identification of chemical warfare (CW) agents and their degradation products and precursors representing a wide range of polarities and chemical structures and acid-base properties has been made under gradient elution conditions on the basis of retention indices relative to the l-phenyl-l-alkanone and 1-[4-(2, 3-dihydroxypropoxy)phenyl]-l-alkanone homologous series [7-10,23,24]. Further, the use of retention indices in gradient HPLC has been applied to multimycotoxin analysis [25-321, drug compounds [ 13-19] and polycyclic aromatic hydrocarbons [22]. Aromatic lichen substances have been identified under gradient elution conditions by comparison of the internal RI values [33,34]. Each of the aforementioned retention index systems has been used in RP-HPLC which is still the most popular separation mode of the HPLC techniques, although its popularity has slightly diminished. The RP method is now used for about half of all HPLC separations [35]. It allows the analysis of compounds with a wide range of structures and polarities, and is a potential method for multi-component analysis under gradient elution conditions. The present chapter focuses on identification of compounds on the basis of their retention indices in gradient RP-HPLC. First, the problems associated with the difficulty of ensuring reproducibility of the retention properties and selectivities from one commercial RP column to another and even from one batch to another of the same product are discussed. Consideration of alternative retention index standard series in RP-HPLC and the principles of gradient elution are described. Further, a more detailed evaluation of selected standards and the critical factors affecting the reproducibility of the gradient retention indices are reported. 6.2 PROBLEMS OF REVERSED-PHASE COLUMNS RP-HPLC has been the method of choice for the analysis of samples including pharma-

Identification using retention indices in gradient HPLC

21 1

ceuticals, agricultural chemicals and environmental compounds. The stationary phase plays an important role in retention and selectivity in RP-HPLC. The most reproducible and consequently, the most commonly used stationary phases for RP separations have been monomeric dimethyloctadecylsilylsilicas (C18 bonded phases) followed by n-octyl (C,) and shorter n-alkyl, phenyl and cyanopropyl bonded phases. During the last 25 years a variety of different procedures have been reported for the synthesis of chemically bonded silica-based packing materials. Several authors have exhaustively reviewed the preparation and characterization of silica-based bonded phases [e.g. 21,36401. Generally, monomeric phases are prepared by refluxing the activated silica with a reactive monofunctional organosilane in an appropriate solvent, when a siloxane bond is formed between a surface silanol and organosilane (Fig. 6.1). Then the phase is usually re-silanized (endcapped) in a second step by treatment with trimethylchlorosilane. In GC, stationary phases are chemically well established and they have similar properties, irrespective of their commercial origin. Unfortunately, this is not true in HPLC, where one of the most important problems has been the difficulty of ensuring reproducibility of the retention properties and selectivities fiom one commercial product to another,

t

S I-OH

+ n

S I-OH

y

Q

I

d I-OH

H

S I-0-S

t

I

t

I-CH2-R

CH3

Endcapping

Sl-0-S

8t

I

1

I-CHz-R

S I-0-S

1

CH3

4

8

AH3

CH3 \ I-O-Sf-CH3

1

S I-OH

CH3

y 3 S1-0-S I-CH2-R

Sl-0-S I-CHZ-R A

8

y 3 I-CH2-R

1 1

I

A

CH3

S I-OH

[

Fig. 6.1. Reaction scheme for the preparation of monomeric octadecyldimethylsilylsilicas(C1g).

References pp. 230-233

+

CI’

212

Chapter 6

and even from one batch to another of the same product [4145]. This is due to the numerous variables involved in the preparation of the RP column: the physical and chemical properties of the parent silica, the silica pretreatment before the bonding reaction, the functionality of the silane reagent, the reaction solvent, the reaction time, the presence and type of acid acceptor and basic catalyst, the type of endcapping reagent and finally, the column packing procedure [38]. The most effective silylations and endcapping reactions cannot remove all the hydroxyl groups from the silica surface. These different kinds of unreacted silanol groups are not inert and thus some of them are available for silanophilic interactions with polar and basic compounds. Possibly a very small population of highly acidic silanols are able to interact much more strongly than the others [38,47,48] and these strong interactions are responsible for peak shape problems and variations in retentions and selectivities of polar and basic compounds. There are many contradictions in the literature, however, on which of the various silanol types are the most active. Undissociated silanols (SiOH) may interact with solute molecules or ions by hydrogenbonding and ionized silanols (SiO-) interact with solute ions electrostatically. Furthermore, silanols may interact with the components of the mobile phase (water, organic modifier and ions). A number of additional tools have been used to overcome this problem in RP-HPLC, including the pH control of the mobile phase, the use of ion-pairing reagents or additives such as alkyl amines (so-called silanol blockers) [49,50] and finally the use of densely covered silica-based stationary phases [38,51]. It is worth noting that when alkyl amines are used for saturating active silanol sites they can then be difficult to remove from columns, leading to permanently altered columns. RP column materials from different manufacturers are based on silicas with different apparent surface pH values varying widely from 3.0 to 9.9 [52] because of the difficulties of removing traces of acids or alkalies used in the synthesis and washing of the silica. In addition, large variations have been noted in pH values of the silica from batch to batch [53,54]. The acidity of the starting silica greatly determines the suitability of the bondedphase for the analysis of polar and basic compounds. Some commercial RP columns have been classified according to their ability to provide the separation for basic compounds without excessive band tailing or retention [49,55]. This relative column acidity ranking lists the columns in the order of increasing acidity. Generally, less acidic columns give a better separation of basic samples, and more acidic columns a better separation of acidic samples. This is not necessarily so simple and there are compounds that behave very irratically regardless of the acidity of the columns. The conventional silica-based bonded-phase columns are expected to be stable against most common solvents even under slightly elevated temperatures and over a pH range of 2-8. However, the properties of the RP columns have been observed to change even under normal laboratory conditions with time [56]. In this aging process, the loss of the organosilane chain from the surface of the silica as a consequence of the hydrolysis of the silicon-oxygen bond have been observed for the cg and c1g phases [56,57]. Hydrolysis occurs most rapidly when the column is used at low pH and at high temperatures [57], but even the pH variation in the safe pH range 2-8 seems to increase the hydrolysis rate of the phase [58]. In the past 5 years, column manufacturers have developed improved reversed phases.

Identification using retention indices in gradient HPLC

213

For example, special base-deactivated columns have been available from different manufacturers for some years to help with problems of analysing basic compounds. In addition, more and more special columns are available for the analysis of other specific sample types. Further, the pH and thermal stability of the conventional CI8bonded phases have been increased by substituting two bulky sterically protective groups (e.g. isopropyl or tbutyl) for the dimethyl groups on the silicon atom of organosilane (see Fig. 6.1) [59-61]. These new CI8column packing materials based on the newer, high-purity and low-acidity silica provide improved separations of basic solutes [6244]. Japanese scientists have been active in the development of different types of polymeric column packing materials used for different purposes [65]. For example, neutral macroporous polymeric resins have been used for RP-HPLC. These packing materials are more pH stable (from pH 1 to 13) and have no underlying surface silanol groups that affect the separation of polar and basic compounds.

6.3 SELECTION OF A RETENTION INDEX STANDARD

The selection of RI standard compounds in HPLC is much more complicated than in GC, because HPLC separations depend on very complex interactions between solutes, stationary and mobile phases. A number of different homologous series of compounds have been used as standards for a retention index scale in RP-HPLC [1,10,11,28]. In addition, alternative series of RI standards have been used for specialized applications, such as PAHs [20,2 13 and nitro-PAHs [22] for the identification of polycyclic aromatic hydrocarbons. The essential requirements for a good standard series for RP-HPLC with UV detection are: most of the members of the series should be readily available; they must have UV sensitivity over a relatively wide wavelength range and high enough sensitivity that they can be added to samples in very small amounts; the first eluting members should be polar enough that they would elute rapidly from the W column with a polar mobile phase like water or an eluent that contains only a small amount of organic modifier; the members of the series should not contain readily ionizable groups, which would cause changes in retention with pH variations; the compounds should be chemically highly stable in LC solvents; retention times of the compounds should increase regularly with carbon number as the carbon chain is successively lengthened with methylene groups; besides a homologous series, a series of compounds with repeating functional groups could be considered as possible RI standards. It is very difficult to find a series fulfilling the aforementioned requirements. For example, the n-alkanes cannot be easily adopted as a reference scale in HPLC because of their low polarity and in addition, they cannot be detected by the almost universally used UV-Vis detectors. The more readily detectable but also quite non-polar n-alkylbenzenes have been used by Morishita et al. [66] as RI marker compounds for the n-alkane scale. Figure 6.2 shows some retention index standard series used in gradient HPLC. The 2alkanones, first proposed by Baker and Ma [5] as a retention index series for RP-HPLC, are capable of spanning a very wide range of solute retentions, but their use is limited by their having a very weak chromophore (e.g. A,,, = 265 nm, log E = 1.2 for 2-propanone

[Ill. References pp. 230-233

214

Chapter 6

Fig. 6.2. Some retention index standard series for use in gradient HPLC with UV detection. I = 2-alkanones; I1 = 1-phenyl-1-alkanones (K-series); I11 = alkyl 4-hydroxybenzoates (E-series); IV = 1-[4-(2,3-dihydroxypropoxy)phenyl]-I-alkanones (D-series); V = I-nitroalkanes (N02-series); n = 0, 1,2, etc.

Aromatic homologues, especially those with conjugated systems and polar groups, would be good candidates, because they are strong chromophores and are easily detected with the UV detector and they could cover solutes with a wide range of retentions. Of the compounds presented in Fig. 6.2, the I-phenyl-1-alkanones (K-series) [1,6,7,10], alkyl4hydroxybenzoates [7,67,68] and l-[4-(2,3-dihydroxypropoxy)-phenyl]-l-alkanones (Dseries) [8-lo], have about equally strong chromophores, absorbing strongly in a wide wavelength range (200-300 nm). Benzoate esters are slightly more polar than l-phenyl-lalkanones [7, lo]. Methyl 4-hydroxybenzoate is less retained than I-phenyl-1-ethanone (acetophenone, KJ under comparable gradient elution conditions. 1-Phenyl-1-ethanone, on the other hand, is less polar and more highly retained than the first three members of the D-compounds [9,10]. The UV spectra of K- and D-compounds in methanol contain characteristic absorption maxima at 243 nm (log E = 4.10); 280 nm (log E = 3.10) and 219 nm (log E = 4.00); 272 nm (log E = 4.22), respectively [lo]. The latter series is not commercially available. 1-Nitroalkanes (N02-series) proposed by Bogusz and Aderjan [ l l ] as the basis of an RI scale for HPLC have similar retention properties to 2-alkanones [69], thus allowing retention indices for early eluting, more polar compounds to be calculated. 1-Nitroalkanes are of limited usefulness because they have a strong UV absorption in a very narrow and , I , , , = 203 nm for 1-nitropropane [ 11J), non-selective wavelength range, 200-220 nm ( and a very weak absorption above that (e.g. A,,, = 260 nm and log E = 1.59 for nitromethane [70]) when they have to be used in very high concentrations. The disadvantage of the low-wavelength UV detection in the 200-220 nm range is that not only the compounds of interest absorb but also unwanted sample compounds absorb as well. The RI series presented in Fig. 6.2 are all usable across a wide eluent composition and pH variation generally used in RP-HPLC [9,10,11,69]. Many other homologous series have been used as the basis of RI scales, but they will not be covered in this context (see the review of Smith [I] and the references cited therein).

Identification using retention indices in gradient HPLC

215

6.4 PRINCIPLES OF GRADIENT ELUTION Gradient HPLC, in which the composition of the mobile phase is changed continuously during the chromatographic run, is effective for the separation of the samples containing compounds with a wide range of retention times. Maximum resolution is provided per unit time. Usually a binary gradient (A-B) system is used, with the strong solvent B (e.g. methanol, acetonitrile (ACN) or tetrahydrofuran (THF) in RP-HPLC) increasing in concentration during the gradient. Most practical separation problems can be solved using binary gradients although the use of ternary (A-B-C) or even quaternary (A-B-C-D) gradient systems may be useful for better control of the separation selectivity. In addition to solvents the following characteristics are needed to describe the solvent gradient: the initial and final mobile phase compositions (gradient range), gradient shape and gradient steepness or time. The gradient shape can be linear, segmented or follow one of several non-linear patterns (e.g. convex, concave). Gradient steepness is expressed as %/min change in the concentration of the B solvent in binary gradient. Since the mid- 1950s several workers have proposed theories for retention prediction in gradient elution. Subsequently two research groups have proposed more general and widely applicable theories for the prediction of chromatographic behaviour under gradient elution: the approach of Snyder et al. [71-741 and the approach of Jandera and ChurtEek [75,76]. Snyder and co-workers introduced the concept of a linear-solventstrength (LSS) system for gradient elution [71-741. RP-HPLC separations with linear gradients normally exhibit LSS behaviour. Details of the aforementioned theories are not given here and the readers are directed to the above references. Resolution in linear gradient elution is determined by the following relationship:

Here, a is the separation factor, N is the column plate number, E is the average capacity factor (k) during gradient elution; it is the instantaneous k value for a solute at column midpoint. The following equations apply best for the case of linear gradients and RP-HPLC [55]. The parameter E can be related to the conditions used in gradient elution separation:

k=

tGF 1.15V, A @S

(6.2)

Here tG is the gradient time (gradient duration in min), F is the flow rate (ml/min), V, is the column dead volume (= toF, to is the column dead time), A@ is the change in the volume fraction of the strong solvent during gradient elution and S is an isocratic parameter determined by the strong solvent and the sample compound. Plots of log k (isocratic) against percent organic are usually linear, with a slope equal to S/lOO. In RP-HPLC, values of S for small molecules (MW < 500) usually range from 3 to 5 and large molecules such as proteins can have S values from 50 to 100. The quantity A@ is given by A@ = (final %B - starting O/oB)/lOO

References pp. 230-233

(6.3)

216

Chapter 6

According to Eq. (6.1) resolution depends strongly on A and therefore it is important to select chromatographic conditions to maintain within a desirable range, typically 2 < A < 10. Any change in gradient parameters (e.g. gradient time or flow rate) affecting leads to predictable changes in the chromatogram. In RP separations under linear gradient elution conditions for an (average) optimum value of = 5 and S = 4 (for small molecules), Eq. (6.2) becomes

r

or 20VmA@ (6.4a) F An optimum gradient range should be adjusted so that the first band elutes at about 2t0 and the last one leaves the column before the gradient is over. In many cases, the optimization of gradient elution conditions are still being performed empirically (trial-and-error methods), which may be a time-consuming and troublesome task, because more parameters must be taken into account than in isocratic elution. This situation is, however, changing with the increasing availability of computer-assisted method development. Computers have played a role in HPLC method development for more than a decade [77791. The work carried out in this field has been summarized in several books and articles [e.g. 80-851. Only during the last few years, has the use of these computer aids been increasing [83]. The optimization of HPLC separations has been extensively studied by using computer-assisted retention prediction and a HPLC computer simulation system. Computer programs for simulating gradient elution separations have been reported by several research groups [74,76,86-891. Two such programs for optimizing gradient RP-HPLC conditions are DryLab G [74,84,87] and LCSIM [86,87] being especially useful for complex samples. The use of these programs requires experimental data from two or more actual chromatographic runs under gradient (DryLab G) or isocratic (LCSIM) conditions and the results are entered as input to the program. Computer simulations of changing the initial and final mobile phase composition (gradient range), gradient shape and gradient time (or steepness) will give guidelines for finding the optimum gradient conditions, but the prediction of optimum conditions still remains difficult, because no adequate theory about the mechanism of solute retention in RP-HPLC yet exists. For technical reasons, gradient elution is always more prone to run-to-run variability than isocratic elution. However, several experimental problems [50] observed in gradient elution can be minimized and more reproducible separations obtained by paying attention to some details. The solvents used in gradient HPLC must be extremely pure, otherwise trace impurities from eluents may be adsorbed on top of the column and eluted later, giving rise to ghost peaks. When working with low-wavelength UV detection in RP-HPLC separations, the use of ACN-water gradients is better than the use of methanol- or THFwater gradients because methanol and THF absorb too strongly at wavelengths <220 nm causing an upward-drifting baseline. It is important that the column is completely equilibrated with the starting mobile phase before the next gradient run. Usually, RP columns t, z

Identij7cation using retention indices in gradient HPLC

217

equilibrate rapidly with changes in mobile phase compositions, but it is noteworthy that these columns equilibrate poorly at 100% water and therefore it is good practice to limit the gradient to 5 1 0 0 % organic modifier. The use of strongly retained additives in the mobile phase (e.g. triethyl amine) to prevent silanol interactions, can complicate the use of gradient elution, because column regeneration is slow and separations tend to be less reproducible.

6.5 CHROMATOGRAPHICBEHAVIOUR OF RETENTION INDEX STANDARDS 6.5.1 Isocratic conditions

The chromatographic behaviour of the RI standards of Fig. 6.2 has been characterized in isocratic conditions by investigating the variation of capacity factors (k) with mobile phase composition and with chemical structure within the homologous series [5,6,10-11, 68-69,90-911. It has been shown that there is generally a linear (Eq. 6.5) or slightly curvilinear relationship between the logarithm of k (log k) and the total carbon number or the carbon number in the alkyl chain (C,) of the individual homologue. The relationship depends on the series and organic modifier of the mobile phase. logk’=aC, +b

(6.5)

Here a and b are constants for the homologous series under the given experimental conditions. Both the slope a and the intercept b depend on the mobile phase composition. Slope a increases with increasing water content, while the intercept b becomes more negative as more organic modifier is included in the mobile phase. The slopes represent Alog k for a unit increase (CH2 unit) in the basic structural moiety for each member in the homologous series. The slope is a measure of methylene selectivity, which characterizes non-specific interactions, while the intercept reflects the specific interactions of the molecule residue with the mobile and stationary phase [92-961. The correlation coefficients (r) are good for a linear relationship between log k and C, for the 1-phenyl-l-alkanones(0.9995-0.9999 [69], 0.999 1-0.9999 [lo]) using methanol or ACN as the organic modifier. When using THF as the organic modifier, the first homologue (acetophenone) is eluted more rapidly than would be anticipated (e.g. correlation coefficient for 40% THF 0.9976) [89,90]. Further, the correlations for the 2-alkanones and nitroalkanes are generally poorer (0.9943-0.9995) than for the l-phenyl-l-alkanones and the relationships were curved [69]. Nitromethane and nitroethane, on the other hand, are more highly retained than would be expected [11,69]. Similar deviations from linearity have been also noted for 1-[4-(2,3-dihydroxypropoxy)phenyl]-l-alkanones (see Fig. 6.3) [lo] and alkyl 4-hydroxybenzoates [91]. Correlations vary between 0.9936 and 0.9989 for 1-[4-(2,3-dihydroxypropoxy)phenyl]-l-alkanones under different isocratic conditions of methanol-water and ACN-water [lo]. As can be seen from Fig. 6.3, the first two or three members of the series deviate from the linearity of Eq. (6.5). In this case a second-order polynomial fit would better describe the correlation. The values of the

References pp. 230-233

Chapter 6

218

0

0

2

1

2

3

4

I

6

5

Cn

Cn

6

8

10

Fig. 6.3. Plot of log k of I-[4-(2,3-dihydroxypropoxy)phenyl]-l-alkanones against number of carbons in the alkyl side chain (C,) under different isocratic conditions. Column, LiChrosorb Hibar RP-18 column (250 x 4.0 mm id.), particle size 5pm; column temperature, 50°C;flow rate, 1.00 ml/min; UV detection, 280 nm. (a) Mobile phase, methanol-water (“A,v/v). (b) ACN-water (“h,v/v).

slope a for 1-phenyl-1-alkanones(Alog k = 0.260 with 50% methanol-water and 0.240 with 40% ACN-water) and 1-[4-(2,3-dihydroxypropoxy)phenyl]-l -alkanones (Alog k = 0.232with 50% methanol-water and 0.193 with 40% ACN-water) show that the CH2unit has a strong effect in both series but a stronger effect in the 1-phenyl-1-alkanone series than in the 1 -[4-(2,3-dihydroxypropoxy)phenyl]- 1 -alkanone series with both solvent sys-

219

Identijkation using retention indices in gradient HPLC

0

I

10

I

20

30

I

0

10

I

I

20

30

* t (min)

*

t (min)

Fig. 6.4. Separation of K-compounds (KI-KI 1) and D-compounds (DI-DI 1) under conditions of linear gradient elution. Column, LiChrosorb Hibar RP-18 (250 X 4.0 mm id.), particle size 5 pm; column temperature, 50°C; flow rate, 1 ml/min; UV detection, 280 nm. Gradient program, (a) from 40% ACN-water to 100%ACN in 30 min, (b) from 20% ACN-water to 100%ACN in 40 min.

tems [lo]. The presence of two alcoholic hydroxyl groups in the latter compounds to some extent reduces interactions of the CH2 unit with the column. 6.5.2 Gradient elution

Under gradient elution conditions the homologous series of RI standards show more or less non-linear chromatographic behaviour [7-10,15,17,32]. The linear relationship is very difficult to obtain in gradient elution without using complicated, non-linear gradient profiles which provide an extra tool for fine tuning of these separations. However, the use of a simplified gradient profile in HPLC screening is advisable for interlaboratorystudies. Figure 6.4 shows representative chromatograms of the homologous series of K- and Dcompounds analyzed by linear gradient elution chromatography with ACN-water. As shown’ in Fig. 6.4, linear gradient shape is not optimum. The smaller interval between the later eluting bands suggests that in this case the optimum gradient would be convex. Figure 6.5 shows plots of the absolute retention time against the number of carbon atoms in the alkyl group of K- and D-compounds and against carbon atoms of nitroalkanes analyzed under ACN-water linear gradient elution conditions. Figure 6.5 shows that the curves of retention time versus carbon number are non-linear. Further, greater deviations from linearity have been noted when methanol is used as the organic modifier in gradient systems with K- and D-compounds [7-10,321. Differences between the retention times of successive members of the homologues are not equal and they vary with the gradient References pp. 230-233

Chapter 6

220

0

2

4

en

6

8

10

12

Fig. 6.5. Plot of unadjusted retention time ( t ~against ) number of carbons (C,) in the alkyl chain for (a) the Kcompounds and (b) the D-compounds and (c) against number of carbons of nitroalkanes under linear gradient elution conditions. Gradient program, (a) from 40% ACN-water to 100% ACN in 30 min [lo], (b) from 20% ACN-water to 100% ACN in 40min [lo] and (c) from 0 to 70% ACN-buffer in 30 min (from data in ref.

Identification using retention indices in gradient HPLC

22 1

conditions used. The retention of these RI series is not dependent on the pH of the mobile phase in buffer-methanol or ACN gradients run at pH 2.5 and 7.5 [lo].

6.6 RETENTION INDICES IN QUALITATIVE IDENTIFICATION

6.6.1 Calculation of retention indices

The retention index, RI, provides a calculated experimental quantity of solute retention relative to a pair of RI standards. It is calculated from the retention parameters of an appropriate homologous or other series (e.g. PAHs) of compounds and the solute compound of interest. Under isocratic conditions, the retention indices can be calculated (Eq. 6.8) like the Kovats indices [2] in GC which are based on the linear relationship between the logarithm of the adjusted retention time (tR)and the carbon number of n-alkanes. Thus by definition, methane has a Kovats index of 100 and ethane, 200, etc. The Kovats index of the compound x eluting between these compounds is then calculated by interpolation. Thus the first standard compounds of the 2-alkanone and I-phenyl-l-alkanone scales, acetone and acetophenone, have RI values of 300 and 800, respectively. The linear relationship (6.6)between the capacity factors and the number of carbon atoms (C,) in the standards is exploited: log Kstandard) = aC, 100 + b

(6.6)

The isocratic retention indices of samples are then determined by linear interpolation using Eq. (6.7): log k(ana1yte)

= aRI(ana1yte) + b

(6.7)

Further, RI is calculated from

where k, is the capacity factor for a compound x and k,, and k,,+i are the capacity factors for the RI standards of carbon number n and n + i eluting before and after the compound x. Provided that the linearity of the plot of log k versus carbon number in the homologous series has been confirmed, it is possible to use any two members of index series, with carbon numbers n and n + i. As the isocratic indices are based on the k values, determination of the value of to is necessary. There is no standard method for measuring this parameter which depends on the mobile phase composition and the molecular size and concentration of the solute that is used for its determination [reviews 97,981. Thus it is diGcult to determine it accurately and therefore, it influences the accuracy of RI calculation by Eq. (6.8). For gradient systems, the capacity factors have no practical significance because the retention times of specific compounds are dependent on the gradient. When this parame-

References pp. 23&233

222

---

Chapter 6

Cubic SpllnO method Polygon mothod

f

-

I

I

I

I

I

I

tRi(*,

1

I

Retention tima

Fig. 6 . 6 . Comparison of two calculation methods for a gradient retention

index.

ter is not used under gradient elution conditions, a determination of to is unnecessary and the errors connected with this factor are avoided. Although under properly chosen gradient conditions, the retention times of homologous compounds could correlate linearly with their carbon numbers, in practice, however, there is more or less deviation from linearity as shown in Fig. 6.5. Therefore, gradient retention indices calculated by polygonal and spline interpolation [99] of the RI versus tR curve (Fig. 6.6) show differences. In the polygon method, the data points of adjacent index standards are connected in a linear fashion and the solute indices are calculated according to Eq. (6.9), first introduced in temperature-programmed GC by van den Do01 and Kratz [loo]. RI = 1ooc,

+ 1OO( c,,; - c, )

tR(x) -tR(n) tR(n+,)

- tR(n)

where tR0 is the unadjusted retention time of the compound of interest and tRcn)and tR(n+i) are the unadjusted retention times of RI standards with carbon numbers (or carbon numbers in the alkyl side chain) C, and Cn+i.When PAHs or nitro-PAHs are used as standards, n is the number of aromatic rings. The more precise method, as shown in Fig. 6.6, is a cubic spline interpolation in which case gradient indices are calculated using a cubic spline function for interpolation [99] in which a smooth curve is generated by piecing together third order polynomials at the data points. A detailed description of the spline interpolation procedure is given in Ref. [99]. However, indices calculated by the polygon method have been most commonly used in gradient HPLC. Then it is important that the standards are mixed with the sample and that the members of the index standard series bracketing the compound of interest are present. Finally, it is worth noting that the retention indices calculated by extrapolation

Identijkation using retention indices in gradient HPLC

223

for the compounds that fall outside the retention index scale are always more or less inaccurate. Bogusz [12] has proposed the use of corrected retention indices (RIc) to reduce the variability of RI values for selected drug compounds caused by use of different column packing materials [12-141. In this correction procedure, a set of selected drugs distributed uniformly throughout the elution range has been used as correction standards whose RI values in a given scale have been predetermined on a reference column. The concept of corrected RI values is similar to the corrected capacity factors in HPLC [101,102], RI correction in GC [1031 or Rfcorrection in TLC [ 1041. The RIc value of a compound of interest is calculated according to the following equations: RIc =aR1+6

(6.10) (6.1 1)

b = RI; -aRI,

(6.12)

where RI is the retention index (isocratic or gradient) of analyte, RI,* and R120 are the reference RI values for correction standards I and 2, eluted before and after the analyte, and RII and RIP are the observed values for standards 1 and 2.

6.6.2 Reproducibility of gradient retention indices The reproducibility of gradient RP-HPLC retention indices is essential for a reliable identification procedure of compounds. The effects of the following parameters and laboratory set-ups have been evaluated in linear gradient W-HPLC: short-term reproducibility, long-term reproducibility, flow rate of the mobile phase, initial mobile phase composition of the gradient, gradient steepness, pH adjustment of the mobile phase, column temperature, intercolumn reproducibility, number of index standards, use of external standards, make of instrument, sampling mode, sample solvent, different analyst, sample size, different laboratories, background and organic modifier of the mobile phase [9,lo]. The 1-phenyl- 1-alkanone (K-series) and 1-[4-(2,3-dihydroxypropoxy)phenyl]- 1-alkanone series (D-series) were used as index standards and the test compounds, representing a wide range of polarities and chemical structures, were a-chloroacetophenone (CN), 2chlorobenzalmalononitrile (CS), dibenz[bfI-ly4-oxazepin (CR), 3-quinuclidinylbenzilate (Bz), benzilic acid, 10,lO’-oxybis-(5,lO-dihydrophenarsazine), 2,4-dichlorophenoxyacetic acid (2,4-D), 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), butyl 2,4-dichlorophenoxyacetate (2,4-D Bu), butyl 2,4,5-trichlorophenoxyacetate(2,4,5-T Bu) and mustard (see Table 6.1). CI8 columns were used throughout the study and the basic organic modifier of the mobile phase was methanol. The indices (RI, and RID) were calculated using a cubic spline method for the interpolation. References pp. 230-233

224

Chapter 6

TABLE 6.1 NAMES AND STRUCTURES OF THE MODEL COMPOUNDS USED IN STUDIES TESTING THE VARIABILITY OF THE RETENTION INDICES Abbreviation

Name

CN

aChloroacetophenone

cs

2-Chlorobenzalmalononitrile

Structure

CN

HC=C

1 \

CN

CR

Bz

3-Quinuclidinylbenzilate

Precursor of Bz

Benzilic acid

Hydrolysis product of adamsite

1 0,10’-0xybis-(5,10dihydrophenarsazine)

HO 0

H-

2,4-D

2,4-Dichlorophenoxyacetic acid

-H

IdentiJicationusing retention indices in gradient HPLC

225

TABLE 6.1 (continued) Abbreviation

Name

2,4,5-T

2,4,5-Trichlorophenoxyacetic acid

Structure

0

O-CHzC,

4

OH CI 2,4-D BU

Butyl2,4-dichlorophenoxyacetate

CI 2,4,5-T BU

Butyl2,4,5-Trichlorophenoxyacetate

0

O-CH2C

4 \

OBU

CI Mustard gas, H

Bis-(2-chloroethyl)sulfide

CH2CHZCI

/

s\CH2CH2CI

The results of the experiments investigating index variability [9,10] indicate that the mean standard deviation of RIK and RID values, calculated for all the test compounds together, vary from 0.9 to 10.7 RI units (RSD from 0.1 to 3.4%) and fkom 1.6 to 29.3 RI units (RSD from 0.1 to 4.9%), respectively (see Table 6.2). The most critical chromatographic parameters affecting the gradient retention indices are RP columns, especially the source of the columns, column temperature, the organic modifier of the eluent, the pH adjustment of the mobile phase with ionizable compounds (e.g. benzilic acid, 2,4-D, 2,4,5-T, Bz and CR) and the exclusion of those members of the index series strongly determining the shape of the interpolation curve. For example, CI8 columns from different manufacturers show retention index variations as much as 124 RI units for CR. Further, in column temperature tests carried out between 35 and 60°C, index variations as much as 3 0 4 0 RI units have been recorded for some CW compounds. The changes in retention indices by as much as 1 5 4 0 FU units caused by variations in column temperature have also been reported for some mycotoxins [26,28]. The effect of column temperature can be exploited as a means to alter selectivity, because temperature change has a different effect on the retention of compounds [ 10,281. It is worth noting that Yamauchi [68] has reported that the retention indices of different phenols do not change with variation in the temperature of the column oven under isocratic conditions with ACN-water (40:60). The RI values determined using different organic modifiers differ markedly, even by 100-200 RI units, reflecting the differences in References pp. 2 3 k 2 3 3

226

Chapter 6

TABLE 6.2 VARIATION IN RETENTION INDEX VALUES (RIK AND KID) AS A FUNCTION OF SELECTED PARAMETERS AND LABORATORY SET-UPS CALCULATED FOR ALL MODEL COMPOUNDS OF TABLE 6.1 TOGETHER AS THE MEAN STANDARD DEVIATIONS [MEAN SD (INDEX UNITS) AND %RSD1 Parameter

R'K Mean SD

Long-term reproducibility Effect of flow rate of eluent Effect of the starting point of the gradient program Effect of slope of linear gradient Effect of pH adjustment Effect of column temperature Intercolumn reproducibility - Columns from the same manufacturer (diff. batches) - Columns from diff. manufacturers Effect of different LC instruments Effect of using only part of the index standard series Effect of using index standards internally and externally Effect of different sample solvents Interlaboratory comparison - C18 columns of same make - Different CIS columns

RID % RSD

Mean SD

Yo RSD

1.9 1.9

i0.5 0.4 I .4

*1.7 4.9 5.3

k0.3 0.7 1.3

3.8 5.9 7.2

0.9 1.7 2.2

6.9 7.9 8.4

1.1 1.4 1.3

7.2

2.2

9.2

1.7

10.7 2.2 1.5

3.4 0.7 0.4

29.3 6.3 5.1

4.9 1.1 0.6

0.9

0.1

1.6

0.1

1.5

0.4

-

-

2.8 4.3

1.0 1.5

3.1 8.0

0.7 1.4

*1,8

the selectivity and the different effects on the elution of RI standards and test compounds. The index reliability deteriorates slightly when a different gradient program is used. The retention indices for those test compounds not capable of strong silanophilic interactions are clearly more reproducible than the retention indices for solutes capable of strong silanophilic interactions. For example, CR has sometimes shown peak broadening and unpredictable chromatographic behaviour. Generally, the compounds possessing polar and basic functions respond quite differently from those simple non-polar and weakly polar solutes whose retention increases simply with the carbon number. Index variations are greater, regardless of the compound, in cases where the compounds elute very close to or together with an index standard. There are practically no significant differences in the reproducibility of retention indices determined using K- and D-compounds as retention index scales. Figure 6.7 shows the correlation between the retention indices (RI, and RID) of selected CW compounds based on the aforementioned index series in methanol-water and ACN-water solvent systems . Equations (6.13) and (6.14) in Fig. 6.7 have been calculated using second-order curve fitting. The mean difference between the measured and calculated index values (RID)is 1.2 and 2.0 RI units with the methanol-water and ACN-water solvent systems, respectively. Thus retention indices can reliably be transformed from the l-phenyl-l-

Identification using retention indices in gradient HPLC

227

- (a)

1200 1 lo00

-

800

-

RID

r

n

l

'

200

100

l

-

300

l

-

400

l

'

l 600

500

-

~

700

.

~

'

i

800

WK

RID = 222.6093 + 0.9102RZK + 0.W2899RIi

500

(6.13)

Benzilic acid

300 100

1

200

.

1

.

300

1

400

-

1

500

I

'

600

i

700

N K RID

= 242.9455 + 1.38997RZK -0.W2119RZ;

(6.14)

Fig. 6.7.Dependence between RIK and RID for the selected test compounds on LiChrosorb Hibar RP-18 column (250 x 4.0 mm Ld.), particle size 5 pm; column temperature, 50°C; flow rate, 1.00 ml/min; UV detection, 254 nm; gradient programs, (a) from 50% methanol-water (pH 2.5) to 100% methanol in 25 min and (b) from 30% ACN-water (pH 2.5) to 100% ACN in 35 min.

References pp. 230-233

228

Chapter 6

alkanone scale to the 1-[4-(2,3-dihydroxypropoxy)phenyl]- 1-alkanone scale, and vice versa [ 101. A comparison of the 2-alkanones, l-phenyl-1-alkanones and 1-nitroalkanes, carried out with acidic compounds, has shown that the RI values are equally sensitive towards changes in eluent composition, pH and make of CIBcolumn in all three scales [69]. Investigations of Bogusz et al. [12-141 have shown that the use of selected drugs as secondary standards may distinctly reduce the variability of RI values caused by the use of different brands of CI8columns provided that the elution conditions are strictly defined [ 151 and that separate sets of drug standards (acidicheutral and basic) are used for the determination of RI values of acididneutral and basic analytes [ 15,161. Further, they have developed a library of 225 compounds using a gradient elution system and a retention index scale based on l-nitroalkanes [17]. The applicability of this library has been checked in interlaboratory use using different instrumentation and different CI8columns [18,19]. According to these studies, the retention indices, calculated with drugs as RI markers, show distinctly lower deviations from the library values and lower intercolumn variability: the mean standard deviations for all drugs analyzed on seven RP columns in the nitroalkane scale are 44.3 RI units, but only 10.3 units using selected drugs as RI markers [18,19]. However, it must be pointed out that the RI correction procedure will fail in the cases where the elution order of the compounds is different on the reference column and analytical column depending on large differences in selectivities between the columns [69].

6.6.3 Confirmation of identiffcation

The reliability of identification can be enhanced by combining the gradient retention indices with on-line UV-Vis spectra [17,24,26-28,31,32] and/or mass spectra [30]. The DAD allows UV-Vis spectral measurements to be taken instantaneously and continuously during the elution of the solute providing a great amount of multi-wavelength chromatographic and spectroscopic data for the identification and confirmation of the sample components. The detector collects spectral data across the 190-600 or 190-800 nm ranges continuously during the analysis. Several graphical and numerical facilities have been developed for the presentation and analysis ofthe data [105-1071. The most important are the following: on-the-fly UV-Vis spectral scanning, three-dimensional plots, twodimensional plots, UV-Vis spectral overlays, absorbance ratioing (purity parameter) and absorbance ratio plots, derivative spectra and recording of chromatograms simultaneously at several wavelengths. Frisvad and Thrane [27,32] have reported on a database including ACN-water gradient retention indices based on the 1-phenyl-l-alkanonescale and UV-Vis spectra for 400 mycotoxins and fungal secondary metabolites having very characteristic UV-Vis spectra which can be used for the confirmation of identity. An RP-HPLC gradient elution method using an acidified ACN-water solvent system has been applied as a multi-mycotoxin screening method where selected mycotoxins have been characterized using retention indices based on 1-[4-(2,3-dihydroxypropoxy)phenyl]- 1-alkanone series and UV-Vis spectral data produced with the DAD [26,28]. Bogusz et ul. [19,108] have created the

Identification using retention indices in gradient HPLC

229

TABLE 6.3 NAMES AND STRUCTURESFOR THE TRICHOTHECENES STUDIED BY HPLC-TSP-MS AND HPLCDYNAMIC FAB-MS [30]

Trichothecenesa

R1

R2

R3

R4

R5

T-2 toxin (T-2) HT-2 toxin (HT-2) Triacetoxyscirpenol(TAS) Diacetoxyscirpenol(DAS) MonoacetoxyscirpenolWAS) Deoxynivalenol (DON)

OH OH OAc OH OH OH

OAc OH OAc OAC OH H

OAc OAc OAc OAc OAc OH

H H H H H OH

OCOCH~CH(CH~)Z OCOCH2CH(CH3)2 H H H =O

aOAc = acetyl.

library for the toxicological screening of biosamples. This library comprises retention data and UV-Vis spectra of over 300 substances including among other things therapeutic and illicit drugs and their metabolites, pesticides and endogenous compounds. The direct coupling of HPLC with MS has become well-established in recent years offering high sensitivity and selectivity in the analysis of a wide variety of compounds. A number of interfaces are available now: e.g. particle beam, thermospray (TSP), electrospray, heated nebulizer and fast atom bombardment (FAB) [ 1091. Different interfaces have their advantages and disadvantages and no single interface can fulfill all requirements and handle all types of separations. The choice of the interface depends strongly on the analytes and information desired. The use of retention indices in gradient RP-HPLC-MS can greatly improve the reliability of identifications made by HPLC-MS [30]. For example, a homologous series of D-compounds has been used as retention index standards in the identification of trichothecenes DON, MAS, DAS, TAS, HT-2 and T-2 (Table 6.3) by gradient RP-HPLC-TSPand dynamic FAB-MS (30). The retention indices offer an independent identification parameter, in addition to the TSP and FAB mass spectra, from the same run. The Dcompounds are efficiently ionized by both the TSP and dynamic FAB techniques. The variations in the retention index values of trichothecenes (from 1.8 to 16.9 RI units) obtained with the TSP and dynamic FAB methods and calculated by the cubic spline method are mainly caused by different brands of CI8columns, but minor contributions of different instrumentation and the use of the buffer solution (ammonium acetate) in TSP cannot be excluded [30].

References pp. 230-233

230

Chapter 6

6.7 CONCLUSIONS

An extension of the retention index concept to gradient elution is the only feasible approach when compounds with a wide range of polarities and structures need to be screened in the same chromatographic run,because gradient methods offer unique solutions to separation problems. Further, integration of retention index with diode array detection and/or mass detection enables both retention data and UV-Vis andor mass spectral information to be used from the same run for definite compound identification. MS is the only sensitive universal detection technique in HPLC that have hture possibilities. The gradient retention index is a complex h c t i o n of the experimental conditions. Variation in gradient retention indices depends on the structures of index standards and the analytes on the one hand and on the stationary and mobile phases on the other, but also on the very complex interactions between all of these which are not adequately understood by current theory [110-1 121. On the basis of the identification of the most critical parameters for the reproducibility of gradient retention indices, it is important that published retention indices should have accurate information about column packing material, column temperature, composition of the mobile phase and gradient program. The nature of the stationary phase exerts the greatest effect on HPLC retention indices. It is still a problem in RP-HPLC that specifLing the column does not necessarily ensure the reproducibility of the retention properties of different batches of the same product, especially with polar and basic compounds. Even the use of the correction method of Bogusz for the retention indices cannot reduce the intercolumn variability of retention index values with polar and basic compounds if there are large differences in the selectivity between the columns. Therefore, the RP column phases should be standardized in terms of selected bulk and chromatographic properties for better comparison. It must be pointed out, however, that promising developments in column technology have been achieved in recent years facilitating the development of rugged, reproducible analytical methods. An understanding of the parameters that influence retention would be important for the design of new stationary phases. In summary, the RI system under gradient elution conditions markedly reduces the variability due to the chromatographic conditions and can be used for the establishment of databases and for tentative identification on an interlaboratory basis under specified conditions.

6.8 REFERENCES 1

2 3 4

5

R.M. Smith, in: Advances in Chromatography (J.C. Giddings, E. Grushka and P.R. Brown, Eds.), Vol. 26, Marcel Dekker, New York, 1987, p. 277. E. Kovats, Helv. Chim. Acta, 41 (1958) 1915. L.G. Blomberg, in: Advances in Chromatography (J.C. Giddings, E. Grushka and P.R.Brown, Eds.), Vol. 26, Marcel Dekker, New York, 1987, p. 229. V. Pac6kovB and L. Feltl, Chromatographic Retention Indices, an Aid to Identification of Organic Compounds, Ellis Honvood, Chichester, 1992. J.K. Baker and C.-Y. Ma, J. Chromatogr., 169 (1979) 107.

ldentijkation using retention indices in gradient HPLC 6 7

8

9

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

2s 26

27 28 29 30 31 32 33 34 3s 36 37 38 39 40 41 42 43

23 1

R. M. Smith, J. Chromatogr., 236 (1982) 313. P. Kuronen, in: Systematic Identification of Chemical Warfare Agents, B.3. Identification of NonPhosphorus Warfare Agents (J. Enqvist and A. Manninen, Eds.), The Ministry for Foreign Affairs of Finland, Helsinki, 1982, p. 43. P. Kuronen, in: Systematic Identification of Chemical Warfare Agents, B.4. Identification of Precursors of Warfare Agents, Degradation Products of Non-Phosphorus Agents, and Some Potential Agents (J. Enqvist and A. Manninen, Eds.), The Ministry for Foreign Affairs o f Finland, Helsinki, 1983, p. 51. P. Kuronen, in: Air Monitoring as a Means for Verification of Chemical Disarmament, C.2. Development and Evaluation of Basic Techniques, Part 1 (M. Rautio, Ed.), The Ministry for Foreign Affairs of Finland Helsinki, 1985, p. 162. P. Kuronen, Ann. Acad. Sci. Fenn., Ser. A2,224, 1990. M. Bogusz and R. Aderjan, J. Chromatogr., 435 (1988) 43. M. Bogusz, J. Chromatogr., 387 (1987) 404. M. Bogusz and R. Aderjan, J. Anal. Toxicol., 12 (1988) 67. M. Bogusz, G. Neidl-Fischer and R. Aderjan, J. Anal. Toxicol., 12 (1988) 325. M. Bogusz, J. Anal . Toxicol., 15 (1991) 174. M. BO~USZ, LC-GC Int., 4(2) (1991) 22. M. Bogusz and M. Wu, J. Anal. Toxicol., 15 (1991) 188. M. Bogusz, M. Erkens, J.P. Franke, J. Wijsbeek and R.A. de Zeeuw, J. Liq. Chromatogr., 16 (1993) 1341. M. Bogusz, J.P. Franke, R.A. de Zeeuw and M. Erkens, Fresenius J. Anal. Chem., 347 (1993) 73. M. Popl, V. Dolansky and J. Mostecky, J. Chromatogr., 117 (1976) 117. L.C. Sander and S.A. Wise, in: Advances in Chromatography (J.C. Giddings, E. Grushka, J. C u e s and P.R. Brown, Eds.), Vol. 25, Marcel Dekker, New York, 1986, p. 139. T.-Y. Liu and A. Robbat, Jr., J. Chromatogr., 539 (1991) 1. P. Kuronen, Proc. 2nd Int. Symp. on Protection Against Chemical Warfare Agents, Stockholm, 1986, National Defence Research Institute, Umeft, 1986, p. 261. P. Kuronen, in: Air Monitoring as a Means for Verification of Chemical Disarmament, C.4. Further Development and Testing of Methods, Part 111 (M. Rautio, Ed.), The Ministry for Foreign Affairs of Finland, Helsinki, 1987, p. 44. D.W. Hill, T.R. Kelley, K.J. Langner and K.W. Miller, Anal. Chem., 56 (1984) 2576. P. Kuronen, in: Systematic Identification of Mycotoxins, B.5. Identification of Selected Trichothecenes, Aflatoxins and Related Mycotoxins (M. Rautio, Ed.), The Ministry for Foreign Affairs of Finland, Helsinki, 1986, p. 21. J.C. Frisvad and U. Thrane, J. Chromatogr., 404 (1987) 195. P. Kuronen, Arch. Environ. Contam. Toxicol., 18 (1989) 336. R. Russel, M. Paterson and C. Kemmelmeier, J. Chromatogr., 483 (1989) 153. R. Kostiainen and P. Kuronen, J. Chromatogr., 543 (1991) 39. P. Kuronen, in: Chromatography of Mycotoxins, Techniques and Applications, Journal of Chromatography Library, Vol. 54 (V. Betina, Ed.), Elsevier, 1993, p. 36. J.C. Frisvad and U. Thrane, in: Chromatography of Mycotoxins, Techniques and Applications, Journal of Chromatography Library, Vol. 54, (V. Betina, Ed.), Elsevier, 1993, p. 253. K. Huovinen, R. Hiltunen and M. von Schantz, Acta Pharm. Fenn., 94 (1985) 99. G.B. Feige, H.T. Lumbsch, S. Huneck and J.A. Elix, J. Chromatogr., 646 (1993) 417. R.E. Majors, LC-GC Int., 5(2) (1992) 12. L.C. Sander and S.A. Wise, CRC Crit. Rev. Anal. Chem., 18 (1987) 299. P.J. van den Drist, H.J. Ritchie and S. Rose, LC-GC, 6 (1988) 124. J. Nawrocki and B. Buszewski, J. Chromatogr., 449 (1988) 1. H.G. Barth, W.E. Barber, C.H. Lochmikller, R.E. Majors and F.E. Regnier, Anal. Chem., 60 (1988) 387R. J.G. Dorsey and K.A. Dill, Chem. Rev., 89 (1989) 331. J.G. Atwood and J. Goldstein, J. Chromatogr. Sci., 18 (1980) 650. R. Amos, J. Chromatogr., 204 (1981) 469. A.P. Goldberg, Anal. Chem., 54 (1982) 342.

232

Chapter 6

44 4s 46

T. Daldrup and B. Kardel, Chromatographia, 18 (1984) 81. R.M. Smith, T.G. Hurdley, J.P. Westlake, R. Gill and M.D. Osselton, J. Chromatogr., 455 (1988) 77. H.A. Claessens, J.W. de Haan, L.J.M. van de Ven, P.C. de Bmyn and C.A. Cramers, J. Chromatogr.,

47 48 49

Nawocki, J., Chromatographia,31 (1991) 177. Nawrocki, J., Chromatographia,31 (1991) 193. M.A. Stadalius, J.S. Berus and L.R. Snyder, LC-GC, 6 (1988) 494. J.W. DoIan and L.R. Snyder, TroubleshootingLC Systems, Humana Press, Clifton, NJ, 1989. Hetem, J.W. de Haan, H.A. Claessens, C.A. Cramers, A. Deege and G. Schomburg, J. Chromatogr., 540 (1991) 53. H. Engelhardt, H. MUller and B. Dreyer, Chromatographia, 19 (1984) 240. R. Schwarzenbach,J. Chromatogr., 334 (1985) 35. S.H. Hansen, P. Helboe and M. Thomsen, J. Chromatogr., 368 (1986) 39. L.R. Snyder, J.L. Glajch and J.J. Kirkland, Practical HPLC Method Development, Wiley, New York, 1988, p. 62. H.A. Claessens, C.A. Cramers, J.W. de Haan, F.A.H. den Otter, L.J.M. van de Ven, P.J. Andree, G.J. de Jong, N. Lammers, J. Wijma and J. Zeeman, Chromatographia,20 (1985) 582. J.L. Glajch, J.J. Kirkland and J. KOhler, J. Chromatogr., 384 (1987) 81. P.J. Schoenmakers,S. van Molle, C.M.G. Hayes and L.G.M. Uunk, Anal. Chim. Acta, 250 (1991) 1. J.J. Kirkland, J.L. Glajch and R.D. Farlee, Anal. Chem., 61 (1988) 2. J.L. Glajch and J.J. Kirkland, LC-GC Int., 3 (4) (1990) SO. J.J. Kirkland, C.H. Dilks, Jr. and J.E. Henderson, LC-GC Int., 6(7) (1993) 436. J. KOhler, D.B Chase, R.D. Farlee, A.J. Vegaand J.J. Kirkland, J. Chromatogr., 352 (1986) 275. J. KOhler and J.J. Kirkland, J. Chromatogr.,385 (1987) 125. J.J. Kirkland, C.H. Dilks, Jr. and J.J. DeStefano, J. Chromatogr., 635 (1993) 19. R.E. Majors, LC-GC Int., 6(12) (1993) 745. F. Morishita, H. Kakihana and T. Kojima, Anal. Lett., 17 (1984) 2385. P.E. Antle, A.P. Goldberg and L.R. Snyder, J. Chromatogr., 321 (1985) 1. S. Yamauchi, J. Chromatogr., 635 (1993) 61. R.M. Smith and N. Finn, J. Chromatogr., 537 (1991) 51. Handbook of Chemistry and Physics, 52nd edition, Chemical Rubber Company, Cleveland, OH, 1971. L.R. Snyder and J.J. Kirkland,, Introduction to Modem Liquid Chromatography, 2nd edition, Wiley, New York, 1979, Chapter 16. L.R. Snyder, in: High Performance Liquid Chromatography, Advances and Perspectives, Vol. 1 (Cs. Horvath, Ed.), Academic Press, New York, 1980, p. 207. L.R. Snyder and M.A. Stadalius, in: High Performance Liquid Chromatography, Advances and Perspectives. Vol. 4 (Cs. Horvath, Ed.), Academic Press, New York, 1986, p. 195. J.W. Dolan, D.C. Lommen and L.R. Snyder, J. Chromatogr.,485 (1989) 91. P. Jandera and J. Churhkk, Gradient Elution in Column Liquid Chromatography, Theory and Practice, Journal of Chromatography Library, Vol. 3 1, Elsevier, Amsterdam, 1985. P. Jandera, J. Chromatogr., 485 (1989) 113. M.W. Watson andP.W. Carr,Anal. Chem., 51 (1979) 1835. J.L. Glajch, J.J. Kirkland, K.M. Squire and J.M. Minor, J. Chromatogr., 199 (1980) 57. R.C. Kong, B. Sachok and S.N. Deming, J. Chromatogr., 199 (1980) 307. J.C. Berridge, Techniques for the Automated Optimization of HPLC Separations, Wiley, New York,

436 (1988) 345.

so 51 52 53 54

55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 7s 76 77 78 79 80

1985. 81 82 83 84 85

P.J. Schoenmakers, Optimization of Chromatographic Selectivity, A Guide to Method Development, Journal of ChromatographyLibrary, Vol. 35, Elsevier, Amsterdam, 1986. L.R. Snyder, J.L. Glajch and J.J. Kirkland, Practical HPLC Method Development, Wiley, New York, 1988, p. 199. J.L. Glajch and L.R. Snyder (Eds.), Computer-Assisted Method Development for High-Performance Liquid Chromatography, Elsevier, Amsterdam, 1990. A. Drouen, J.W. Dolan, L.R. Snyder, A. Poile and P.J. Schoenmakers, LC-GC Int., 5(2) (1992) 28. P.P. Csokan, F. Darvas, F. Csizmadia and K. Valko, LC-GC Int., 6(6) (1993) 361.

Identijication using retention indices in gradient HPLC 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112

233

J. Schmidt and C. McClain, J. Chromatogr.,419 (1987) 1. J. Schmidt, J. Chromatogr., 485 (1989) 421. Y. Baba, J. Chromatogr., 485 (1989) 143. T. Sasagawa, Y. Sakamoto, T. Hirose, T. Yoshida. Y. Kobayashi, Y. Sat0 and K. Koimmi, J. Chromatogr., 485 (1989) 533. R.M. Smith and C.M. Burr, J. Chromatogr., 475 (1989) 57. R.M. Smith and R. Wang, J. Chromatogr., 558 (1991) 7. H. Colin and G. Guiochon, J. Chromatogr. Sci., 18 (1980) 54. P. Jandera, H. Colin and G. Guiochon, Anal. Chem., 54 (1982) 435. H. Colin, G. Guiochon, 2. Yun, J.C. Diez-Masa and P. Jandera, J. Chromatogr. Sci., 21 (1983) 179. P. Jandera, J. Chromatogr.,314 (1984) 13. P. Jandera, Chromatographia, 19 (1985) 101. R.J. Smith, C.S. Neiass and M.S. Wainwright, J. Liq. Chromatogr.,9 (1986) 1387. R.A. Djerki and R.J. Laub, J. Liq. Chromatogr., 10 (1987) 1794. W.A. Halang, R. Langlais and E. Kugler, Anal. Chem., 50 (1978) 1829. H. van den Do01 and P.D. Kratz, J. Chromatogr., 11 (1963) 463. R. Gill, A.C. Moffat, R.M. Smith and T.G. Hurdley, J. Chromatogr. Sci., 24 (1986) 153. R. Gill, M.D. Osselton, R.M. Smith and T.G. Hurdley, J. Chromatogr., 386 (1987) 65. J.P. Franke, J. Wijsbeek and R.A. de Zeeuw, J. Forensic. Sci., 35 (1990) 813. J.P. Franke and RA. de Zeeuw, in: The Analysis of Drugs of Abuse (T.A. Gough, Ed.), Wiley, New York, 1991, p. 93. D.G. Jones, Anal. Chem., 57 (1985) 1207A. T. Alfredson, T. Sheehan, T. Lenert, S. Amodt and L. Correia, J. Chromatogr., 385 (1987) 213. S. Ebel and W. Mueck, Chromatographia,25 (1988) 1039. M. Bogusz and M. Erkens, Toxichem. Krimtech., 59 (1992) 2. A.L. Yergey, G.E. Edmonds, LAX Lewis and M.L. Vestal (Eds.), Liquid ChrornatographyMass Spectrometry, Techniques and Applications, Plenum Press, New York, 1990. R.P.W. Scott, J. Chromatogr. A, 656 (1993) 51. A. Tchapla, S. Heron, E. Lesellier and H. Colin, J. Chromatogr. A, 656 (1993) 81. K. Valko, L.R. Snyder and J.L. Glajch, J. Chromatogr. A, 656 (1993) 501.

This Page Intentionally Left Blank

R.M. Smith (Ed,), Retention and Selectivity in Liquid Chromatography Journal of Chromatogaphy Library, Vol. 57 0 1995 Elsevier Science B.V. All rights reserved

235

CHAPTER 7

Characterization of retention and selectivity in reversed-phase LC using interaction indices P. Jandera Department of Analytical Chemistry!University of Pardubice, Faculty of Chemical Technology,Ndm.Legii 565, 532 10 Pardubice, Czech Republic

7.1 INTRODUCTION

Our present knowledge of quantitative aspects of the retention mechanism on the molecular level is too limited to allow the calculation of the retention characteristics directly from the physico-chemical parameters of solutes and various methods of correlation of retention, usually expressed as the logarithm of the capacity factor, with some structural descriptors have been proposed for this purpose. Without resorting to such correlations, the retention can be characterized in relation to a set of standard compounds used to calibrate the retention scale by means of retention indices. The coefficients of the correlation equations or the retention indices usually depend significantly on the separation conditions, which requires re-calibration of the chromatographic system when the separation conditions, such as the composition of the mobile phase, are changed. To take this behaviour into account, some authors have introduced a functional dependence of the coefficients on the composition of the mobile phase into the correlation equations. The problem can also be approached using theoretical models of retention in reversedphase LC such as the model using the Hildebrand solubility parameter theory [1-4], or the model supported by the concept of molecular connectivity [ 5 ] , models based on the solvophobic theory [6,7] or on the molecular statistical theory [8]. Unfortunately, sophisticated models introduce a number of physicochemical constants which are often not known or are difficult and time-consuming to determine, so that such models are not very suitable for rapid prediction of retention data. The most characteristic feature of reversed-phase chromatography is higher polarity of the mobile phase in comparison with the stationary phase (usually an alkyl phase chemiReferences pp. 266-267

236

Chapter 7

cally bonded on a suitable support), which should theoretically behave as an almost ideally non-polar material with the properties of long-chain aliphatic hydrocarbons. According to the solvophobic theory [6,7], such a stationary phase can be considered inert with respect to polar interactions and the mobile phase (solvophobic) interactions are understood as the driving force of the formation of associates of the solutes with the non-polar stationary phase. The retention results from a decrease in the contact area of the solute with the mobile phase caused by its transition from the bulk mobile phase to the surface of the stationary phase. Replacement of weaker interactions between a moderately polar solute and polar mobile phase by mutual interactions between strongly polar molecules of the mobile phase in the space element of the mobile phase originally occupied by a solute molecule results in overall energy decrease in the system, which is the driving force of the retention in the absence of strong (polar) interactions of the solute with the stationary phase. In the real world, the stationary phase plays an important role in retention in reversedphase systems. The bonded alkyl chains differ from the free molecules of hydrocarbons in the liquid state by limited mobility and for sterical reasons, it is principally impossible to modify all the available silanol groups at the surface of silica gel by chemical reaction with the silanization reagent. The unreacted silanol groups may affect the retention by specific polar interactions, especially with basic solutes. Finally, organic solvents used as the components of the mobile phases in reversed-phase systems are preferentially sorbed by the stationary phase and can significantly modify its properties [9,10]. In spite of its inherent limitations, we found the solvophobic theory of retention useful as the starting point in the derivation of a simplified semi-empirical description of reversed-phase systems to characterize and predict the retention and the selectivity of separation [l l]. Because the effects of the stationary phase are neglected, this approach can be applied for relative rather than absolute predictions and a suitable set of standard reference compounds is necessary to calibrate the retention ( or the selectivity ) scale. 7.2 INTERACTION INDICES AS THE DESCRIPTORS OF RETENTION

The basic first-approximation assumption of the interaction indices approach adopted from the solvophobic theory is that the stationary phase does not contain polar adsorption sites and that only the non-specific dispersive forces account for the non-polar interactions between the stationary phase and the solutes. Non-polar interactions occur also in the mobile phase together with stronger polar interactions (dipole-dipole, proton-donor and proton-acceptor) between the molecules of the solute and those of the mobile phase on the one hand and mutual interactions between the polar molecules of the mobile phase on the other. The dispersive interactions in the mobile and in the stationary phases are of similar magnitude and are assumed to approximately equilibrate each the other. The polar interactions between the polar molecules of the aqueous-organic mobile phase are in most instances stronger than those between the mobile phase and the solute, which is usually less polar than the components of the mobile phase. This results in a more or less strong repulsion of the molecules of the solute from the mobile phase into the stationary phase, which behaves as a passive acceptor rather than a source of active attraction forces for the sorbed molecules, in contrast to the sorption on polar adsorbents. As the sample solutes

Characterization of retention and selectivity in reversed-phase LC using interaction indices

237

are separated in dilute solutions in analytical HPLC, there is no need to consider mutual interactions between solute molecules in this model. The main driving force of retention can be understood as the difference in the free energy of polar interactions between the molecules of the mobile phase, -AGM-M, and the free energy of polar interactions between the molecules of the solute and the molecules of the mobile phase, -AGM-x. The energy of polar interactions between two molecules is characterized here as the product of the contributions from each molecule. The individual contribution is understood to be directly proportional to the index of interaction, Zi,characterizing the polarity of the molecule i. As every solute is characterized by a unique value of Zi,the proportionality constant, ci > 0, accounts for the specific character of the polar interactions connected with the type of organic solvent used in the mobile phase and affects the constants of the calibration equation of the interaction indices scale based on a set of reference standards [ 113. As one molecule of the solute A can interact with a different number of molecules of compound B, it is necessary to relate the interaction indices to the unit volume element of the space where the interactions occur. According to this definition, the interactions connected with the transfer of one mole of the solute X from the mobile to the stationary phase occur in the volume proportional to Vx, the molar volume of the solute. Then -AGM-x = cMZMCXZXVX and -AGM-M = CMZMCMZMVX, where IX is the interaction index of the solute and,Z that of the mobile phase and cx > 0, cM > 0 are c, of the solute and of the mobile phase, respectively. Neglecting the relatively small entropic contribution to the retention, we obtain the following equation for the total change of the free energy, -AG, of the transfer of one mole of the solute fiom the mobile to the stationary phase [ 111:

where k is the capacity factor of the solute, R is the gas constant, T is the temperature in degrees Kelvin and @ = VdVM is the phase ratio in the column, i.e. the ratio of the volumes of the stationary, Vs, and of the mobile, VM, phases. Equation (7.1) can be rearranged into the form

Equation (7.2) predicts a linear variation of the logarithms of the specific capacity factors, k*, with the interaction indices of the solutes, ZX, in a given chromatographic system. The parameters A and B of Eq. (7.2) depend on the column, temperature and on the mobile phase used. Hence, it should be possible to use the interaction indices for the calibration of the retention scale in terms of log k* in different chromatographic systems with the parameters A and B determined in each system using a set of suitable calibration standards. References pp. 266-267

238

Chapter 7

As the underlying theory takes into account only a simplified and imperfect scheme of complex phenomena controlling the retention in which the role of the stationary phase cannot be neglected, the validity of Eq. (7.2) may be limited in some chromatographic systems, especially those where unreacted silanol groups can affect the retention of basic or polar compounds. The role of the stationary phase appears only in the value of the phase ratio @ = VdVM in Eq. (7.2). The volume of the stationary phase Vs increases and consequently VM decreases with the amount and the length of the bonded alkyls. The retention of a solute has been found experimentally to increase with increasing length of the bonded alkyl only up to a certain "critical chain length" [ 121 and because of the simplifying assumptions accepted in the derivation of the retention equation (Eq. 7. l), Eq. (7.2) does not allow any meaningful conclusion about the effects of the stationary phase on retention. The effect of the structure of the solute on retention is expected at two levels. The retention should decrease with increasing intensity of polar interactions, i.e. with the values of the interaction indices, ,Z On the other hand, it should increase with the volume in which the interactions occur, which is expected to be approximately proportional to the molar volume of the solute, Vx. Only the effects of the size of the structural units of low and approximately equal polarities can be compared in this way, otherwise the behaviour is complicated by the effects of changing polarities.

7.2.1 Retention in binary mobile phases

In aqueous-organicbinary mobile phases, ZM in Eqs. (7.1) and (7.2) can be expressed as a linear function of the concentration of the organic solvent, r# (volume fraction, % much like the Snyder's polarity indices P' [ 131: vol/vol.

IHz0and Zorgare the interaction indices of water and of the organic solvent, respectively. Combining Eqs. (7.1) and (7.3) we obtain the retention equation describing the dependence of the retention (capacity factor) of a solute, k, on the composition of the mobile phase [111:

log k = a-rn$+atp2

(7.4)

This equation is formally identical with the quadratic retention equations derived earlier using the solubility parameter theory [ 1-41 and from the molecular statistical theory [8], but the constants a, m,dare defined in a different way here and depend on the interaction indices of the components of the mobile phase and on the molar volume of the solute: a = log-

Vs VM

+

VXIH~OCM (zHzOcM - z X c X ) 2.31RT

(7.5)

Characterization of retention and selectivity in reversed-phase LC using interaction indices

239

The value of the quadratic term dP2in Eq. (7.4) determines the curvature of the log k versus @ plots. The curvature parameter d is expected to increase with decreasing polarity of the organic solvent, Zorg, according to Eq. (7.7). This is in agreement with the experimental observations, where the parameters d are significantly lower in methanol-water than in acetonitrile-water or tetrahydrofuran-water mobile phases [I 1,141. Equation (7.7) further predicts that the parameter d and the curvature of the log k versus +plots should increase with the size (molar volume, V,) of the solute. For example, the parameter d increases from 1.88 for ethylbenzene to 3.94 for n-hexylbenzene and from 3.67 for nheptane to 4.72 for n-decane on a C8 column in acetonitrile-water mobile phases [14]. Hence, the term often is not very significant and can be neglected to first approximation, at least at lower concentrations of organic solvents in the mobile phase, so that the retention equation is reduced to the well-known and widely used form [ 1,4,15].

e2

The log k versus @ plots are often linear in aqueous solutions of methanol (Zorg = 2 1. I), slightly non-linear in acetonitrile-water mixtures (Zorg = 18.3) and significantly curved in mobile phases containing tetrahydrofuran in water (Zorg = 11.4). According to Eq. (7.6), the parameter m is directly proportional to the difference between the interaction indices of water and of the organic solvent, (ZH20-Zor& Consequently, m is expected to increase with decreasing Zorg, i.e. with decreasing polarity of the organic solvent, which agrees with the experimental data. Both the slope (parameter m, Eq. (7.6)) and the curvature (parameter d, Eq. (7.7)) of the log k versus @ plots are expected to increase with increasing size of the non-polar part in the molecule of a sample compound and the experimental observations are in agreement with this prediction. It should be noted that the intercept and the slope of a regression line fitted to the experimental data points correspond to the constants a and m as defined by Eqs. (7.5) and (7.6) only if the term dP2is lower than approximately 5-10% of a + m@. Equation. (7.2) can be combined with Eq. (7.3) to introduce the dependence of the specific capacity factors k* on the composition of a binary mobile phase. Neglecting the second-order terms as in the derivation of Eq. (7.8), the following expressions are obtained for the parameters A and B: A =A'o-A',@

(7.9)

B = B'o - B'14

(7.10)

with References pp. 266-267

Chapter 7

240

(7.1 1)

(7.12)

Bi =

C M C X IH,O

(7.13)

2.3 1RT

(7.14)

If the plots of log k versus $J are linear over the range of mobile phase compositions of interest, Eqs. (7.11H7.14) can be used to calculate the constants A and B. The coefficient A is in fact a quadratic function of the solvent composition, but the experimental plots of A versus (except for tetrahydrofuran-water mobile phases) can usually be fitted by straight lines with regression coefficients better than 0.999. More sophisticated calculations in the systems with less polar organic solvents used as the components of the mobile phase require quadratic equations to describe the dependence of the parameter A on the composition of the mobile phase. Figures 7.1 and 7.2 show examples of the experimental plots of the logarithms of the specific capacity factors, k*, versus the interaction indices, I,, on a Silasorb SPH C8 col-

1-00 1

-0.50

'

0

I

I

I

I

I

1

2

3

4

5

6

1, Fig. 7.1. Experimental dependences of the specific capacity factors, k*, on the interaction indices for 15 solutes on a Silasorb SPH C 8 column in methanol-water 60:40 (l), 70:30 (2) and 80:20 (3) mobile phases.

Characterizationof retention and selectivity in reversed-phase LC using interaction indices

24 1

;

A bo

0

c (

0.00

OO

O0 0

2

3

-0.50

1

0

4

5

6

1, Fig. 7.2. Experimental dependences of the specific capacity factors, k*, on the interaction indices for 15 solutes on a Silasorb SPH C8 column in acetonitrile-water 60:40 (l), 70:30 (2) and 80:20 (3) mobile phases.

umn in binary mobile phases with various concentrations of methanol and acetonitrile in water. As expected, the data sets are shifted to lower values of logk* with increasing concentration of the organic solvent in the mobile phase. 1.50

I

9

Q

\

1.20 0.90 -

4

0.60 -

0.30

-

0.00

-

’

-0.30 0.40

I

I

I

I

I

0.50

0.60

0.70

0.80

0.90

1.00

W

Fig. 7.3. Plots of the intercepts A of Eq. (7.2) versus the concentration of the organic solvent, q5 (in % vol/vol.lo-’), in methanol-water (1, 4), acetonitrile-water (2, 5) and tetrahydrofuran-water (3) aqueousorganic mobile phases on a Silasorh SPH C18 (1-3) and Silasorb SPH C8 (4,s) columns.

References pp. 266-267

242

Chapter 7

Oa20

I

0.15 -

m 0.10 -

’

0.05 0.40

I

I

I

I

I

0.50

0.60

0.70

0.80

0.90

1.00

+

Fig. 7.4. Plots of the slopes B of Eq. (7.2) versus the concentration of the organic solvent, (in % v~Vvol.lO-~), in methanol-water ( I , 4), acetonitrile-water (2, 5) and tetrahydrofuran-water (3) aqueous-organic mobile phases on a Silasorb SPH C18 (1-3) and Silasorb SPH C8 (4, 5) columns.

Experimental dependences of the parameters A and B of Eq. (7.2) on the concentration of methanol, acetonitrile and tetrahydrofuran in binary aqueous-organic mobile phases are illustrated by several examples in Figs. 7.3 and 7.4 for a Silasorb C18 and a Silasorb C8 column. The plots are close to straight lines as predicted by Eqs. (7.9) and (7.10) and their slopes decrease with decreasing polarities of the organic solvent (increasing Z0& i.e. are higher for methanol-water than for acetonitrile-water and tetrahydrofuran-water mobile phases, in agreement with Eqs. (7.12) and (7.14). Table 7.1 lists some experimental values of the parameters KO,B’o, A’,, B’, in various reversed-phase chromatographic systems. From Eqs. (7.1 1)-(7.14), it follows that the ratio (B’dB’&4’dA’,) should be equal to 2. Most experimental values in Table 7.1 are close to this value which suggests that the model is self-consistent. Equations (7.11)(7.14) also allow the interaction indices of the organic solvents, Zorg, to be determined. First, the interaction index of methanol was determined by extrapolation of the linear plots of Z, versus the number of methylene units for the homologous series of n-alcohols. Then and Zorgof other organic solvents can be obtained by solving the following equations obtained from the combination of Eqs. (7.1 I), (7.12) and Eqs. (7.13), (7.14):

( 2i]zH20

I org = I - -

=( ‘-Z)’H2O

(7.15)

The values of Zorg determined in this way (see Table 7.3) [ 113 are in qualitative agreement with usual behaviour in reversed-phase systems. Methanol as the solvent with the acetonitrile weakest elution strength and the highest polarity has the highest value of Zorg,

Characterization of retention and selectivity in reversed-phase LC using interaction indices

243

TABLE 7.1 EXPERIMENTAL VALUES OF THE PARAMETERS OF EQS. (7.9H7.14) Columns: Lichrosorb RP18 (I), Lichrosorb RP8 (II), Hypersil ODS (111), Silasorb C18 (IV) and Silasorb C8 (V) columns in methanol-water (a), acetonitrilewater (b), dioxane-water (c) and tetrahydrofuran-water (d) mobile phases X = (B’&I’i)/(A’dA’i) Column

Mobile phase

A’o

A‘1

B’O

B’1

X

I

a b

3.378 2.783 3.033 2.733 3.037 2.472 2.459 2.766 2.783 2.783 3.286 3.065 3.316 2.932 2.126

3.580 3.351 3.572 3.860 3.540 3.025 3.030 4.488 3.580 3.351 2.791 3.509 4.253 3.350 2.548

0.315 0.225 0.297 0.229 0.238 0.182 0.221 0.213 0.315 0.225 0.330 0.268 0.188 0.277 0.170

0.188 0.126 0.175 0.184 0.131 0.109 0.107 0.204 0.188 0.126 0.240 0.206 0.133 0.216 0.120

1.78 2.15 2.00 1.76 2.12 2.04 2.06 1.70 1.78 2.15 1.18 1.49 1.81 1.47 1.70

C

I1

d a b C

I11 IV

V

d a b a b d

a b

Adapted from the data in refs. [l I] and [28].

and dioxane are less polar and approximately equivalent and tetrahydrofuran is much stronger eluent with lowest polarity and I,,,. The values of Iorgobtained with different coIumn packing materials are in good agreement, except for tetrahydrofiran, which can be explained by a more significant curvature of the plots of A and B versus the concentration $ in tetrahydrofuran-water mobile phases observed with some columns.

7.2.2 Retention in ternary mobile phases

The approach similar to the description of retention in binary mobile phases can also be applied to multicomponent solvent systems, where Eq. (7.3) has the following form: (7.16) and 4; are the interaction index and the concentration, respectively, of the organic solvent i in the aqueous-organic multisolvent mobile phase. From Eqs. (7.1) and (7.16), the following equation was derived for k [ 161: (7.17)

References pp. 266-267

244

$y

Chapter 7

For a ternary mobile phase containing organic solvents X and Y in concentrations @D in water, Eq. (7.17) becomes [16]:

where the constants a, m, my d,, and dy are defined by Eqs. (7.5)47.7), with either Iorg = lorg,x, or Iorg = zorg,y. The contributions of the terms with d,, dy to log k often can be neglected over a more or less limited range of concentrations $, @y and then the retention can be calculated from the following simple equation [ 16-18]. logk =a-m,$, -my$y

(7.19)

The values of the constants a, m , my and, if necessary, d, and dy can be determined in binary mobile phases, which makes it possible to predict the retention in ternary mobile phases X-Y-water from the retention data in binary mobile phases X-water and Y-water, using Eq. (7.18) or (7.19). When applying this approach, we often find significant differences in the values of the constant a determined by regression of the log k versus @ plots for the individual binary mobile phases containing different organic solvents X and Y, i.e. ax differs from ay, which makes the use of Eqs. (7.18) or (7.19) difficult. A simple but efficient empirical remedy is to use the balanced mean value of the two experimental constants a, and av for different compositions of ternary mobile phases [ 17,181: (7.20)

The errors of the k predicted from the data in binary mobile phases on a given column using Eqs. (7.18) or (7.19) and (7.20) are usually 5% or less [17,18]. In addition to isocratic elution with ternary mobile phases, Eqs. (7.19) and (7.20) were used as the basis for calculations of retention volumes in reversed-phase chromatography using elution with ternary gradients [ 171. Figure 7.5 shows experimental plots of the logarithms of the specific capacity factors, k*, in dependence on the interaction indices Z, in ternary mobile phases with two different ratios of concentrations of methanol and acetonitrile in water, measured on a Silasorb C18 column. The form of the plots is similar to the dependences measured in binary mobile phases (Figs. 7.1 and 7.2) and both the absolute values of log k* and the slopes of the plots decrease with increasing proportion of the concentration of acetonitrile to the concentration of methanol in mobile phases with a constant concentration of water, i.e. with decreasing polarity and interaction index of the ternary mobile phase, in agreement with Eq. (7.2). In mobile phases comprised of water and two or more organic solvents, the parameters A and B of Eq. (7.2) relating the specific capacity factors, k*, to the interaction indices, Zx,depend on the concentrations of the organic solvents X and Y in the aqueous+rganic

Characterization of retention and selectivity in reversed-phase LC using interaction indices

0

2

4

6

245

8

Fig. 7.5. Experimental dependences of the specific capacity factors, k*,on the interaction indices for 29 solutes on a Silasorb C18 column in methanol-acetonitrilewater 55:15:30 (1) and 15:55:30 (2) ternarymobile phases.

mobile phase in a similar way as in binary mobile phases (Eqs. (7.9) and 7.10)). A ternary mobile phase is often prepared from two binary mobile phases containing different organic solvents X and Y in concentrations $x and $,. The possibility of predicting the retention data in the mixed ternary mobile phase from known retention characteristicsof the two original binary mobile phases may be very useful in this case. Introducing Eq. (7.16) for ZM into Eq. (7.2) we obtain the following dependence of the specific capacity factor k* in a ternary mobile phase on the volume proportions ux,wy, in which the two binary mobile phases were mixed (w, + uy= 1) [ 161:

(7.21)

The logarithms of the specific capacity factors and the parameters A and B of the. calibration equation (7.2) in a ternary mobile phase can be predicted from Eq. (7.21) as a linear combination of log k* or of A and B in the binary mobile phases used for the preparation of the ternary mobile phase. The prediction is expected to be subject to an error A, which depends on the mixing ratios ux,wy and on the concentrations of the organic solvents Xand Y in the original binary mobile phases. Equation (7.21) suggests that the error of prediction can be suppressed by matching the compositions of the two mobile phases in such a way that the difference (Iorg,,.$, - Zorg,y$y) is minimized. This is illustrated in Figs. 7.6 and 7.7 by comparison of the experimental plots (points) with the dependences References pp. 266-267

246

Chapter 7 2.50

2.00

4

1.50

1.oo

0.50

A3

1

0.00

0.20

0.40

0.60

0.80

1.oo

csx

Fig. 7.6. Plots of the dependences of the intercept A of Eq. (7.2) on the mixing ratio, ox,of methanol-water (5050) and acetonitrile-water (74.3:25.7 (I), 44.3:55.7 (2), 59.3:40.7 (3)) binary mobile phases used to prepare ternary mobile phases methanol-acetonitrile-water. Points, experimental values; lines, dependences predicted from Eq. (7.21).

0.30

I

0.25 Pp

0.20 n

0.1 5 0.00

0.20

0.40

0.60

0.80

1-00

Q* Fig. 7.7. Plots of the dependences of the slope, E , of Eq. (7.2) on the mixing ratio, w,, of methanol-water (5050) and acetonitrile-water (74.3:25.7(l), 44.355.7 (2), 59.3:40.7 ( 3 ) ) binary mobile phases used to prepare ternary mobile phases methanol-acetonitrile-water. Points, experimental values; lines, dependences predicted from Eq. (7.21).

Characterizationof retention and selectivity in reversed-phase LC using interaction indices

0

3

6

12

247

15

Fig. 7.8. Effect of increasing term (Iorg3+, - Iorgz+y) on the error of prediction of k from Eq. (7.22) neglecting the term A (characterized by relative standard deviation (RSD) of k of 36 compounds) on a Lichrosorb RP18 column in ternary mobile phases prepared by mixing methanol-water binary mobile phases with binary mobile phases containing acetonitrile, dioxane or tetrahydrofuran in water.

of the parameters A and B predicted from Eq. (7.21) on the mixing ratio w, of acetonitrile-water and methanol-water binary mobile phases used to prepare acetonitrilemethanol-water ternary mobile phases. The agreement between the experimental and predicted values of the parameters is better when 50% methanol is mixed with 44.3% acetonitrile, where (ZOrg,, $, - Zorg,y &) = 0.1 (plots 2) than for mixtures of 50% methanol $, - Zorgy$y = 16.8) or 74.3% acetonitrile (Iorg&,with either 59.3% acetonitrile (lorg,x Zorgdy = 7 1.4) (plots 1 and 3). Equation (7.21) can be rearranged to predict the capacity factors k in a ternary mobile phase prepared by mixing two binary mobile phases: logk=w, logk, + m y iogk, - D

(7.22)

Figure 7.8 demonstrates the expected effect of the magnitude of the term (Iorg,&, Zorg,@y) on the error of k predicted from Eq. (7.22) neglecting the term A in ternary mobile phases prepared by mixing methanol-water, acetonitrile-water, dioxane-water and tetrahydrofhran-water binary mobile phases. The experimental results substantiate the importance of minimizing the difference (fors,&,- lorg,&,) in predicting the retention data in ternary mobile phases using simple additivity rules, such as Eq. (7.22).

7.3 CALIBRATION OF THE SCALE OF INTERACTION INDICES

Eq. (7.2) predicts a linear variation of the logarithm of specific capacity factor k* with the References pp. 266-267

248

Chapter 7

interaction index. In spite of the expected validity of this equation only for a series of compounds belonging to the same selectivity class (with the same value of the constant c), the experimental log k* versus I, plots are approximately linear for a broad variety of solutes tested (Figs. 7.1, 7.2 and 7.5) and the effects of the individual parameters (Ix ZOrg V,) on the retention are predicted in reasonable qualitative agreement with the experimentally observed behaviour. Equation (7.2) can be used either to estimate the capacity factors of the solutes whose I, and V, are known or to calculate the values of the interaction indices from the retention data. Interaction indices are not absolute physico-chemical quantities and to permit the use of this equation for quantitative calculations of retention, a calibration scale based on several adequately chosen standard compounds must be defined. The calculation of the specific capacity factors requires the determination of the phase ratio @, which is not straightforward in reversed-phase chromatography, especially with respect to the definition of the volume of the stationary phase. Some workers have simplified the problem by choosing @ = 1 [ 191, other defined the stationary phase as the mass of the packing material in the column either determined directly [20], calculated from the carbon content and the packing density, or as the surface area estimated by the BET method [21]. None of the above methods can accurately describe the widely accepted idea of the stationary phase formed by the chemically bonded material solvated by the components of the mobile phase. It would be very difficult (if possible at all) to determine exactly the amount of the liquid embedded in the stationary phase. The volume of the mobile phase in the column is most simply determined as the retention time of an inert, i.e. unretained, compound (a pure component of the mobile phase or a salt), even though some more sophisticated methods such as linearization of the retention data in a homologous series yield better accuracy of VM, [22]. However, the latter methods are tedious and time consuming. Another complicating problem is the fact that VM and probably also Vs slightly depend on changing composition of the mobile phase [22,23]. Regardless of the real retention mechanism applying in reversed-phase chromatography on chemically bonded phases, the choice of @ results from the definition of the distribution constant in the chromatographic system. From the practical point of view, it should allow a simple use by practicing chromatographers and must not require tedious sets of measurements. Because of the uncertainty of the definition of Vs, the definition of the phase ratio may be understood as a matter of convention rather than of an exact physico-chemical definition for the purpose of the interaction indices approach. For the sake of convenience, Vs is defined here as the fraction of the column that is not occupied by the mobile phase, Vs = VG - V M , where VG is the geometrical inner volume of the empty column. VM is determined as the mean value of the elution volume of an inert compound in mobile phases containing 40-80% of the organic solvent in water [ 111. This formal definition of @ implies that the distribution constant is understood as the ratio of solute concentration in the bulk volume of the packing material (support + bonded phase + occluded liquid) to the concentration in the mobile phase contained in the column. The definition of the scale of I, was originally based on the Snyder’s polarity indices P’ which are a measure of the polarity of solutes derived from their gas chromatographic data [13]. It was obvious that the P’ values must be adjusted to take account of the aque-

Characterization of retention and selectivity in reversed-phase LC using interaction indices

249

ous-organic medium surrounding the molecules of the solutes. For this purpose the experimental values of log k* were plotted as a function of P' for a variety of solutes in many different mobile phases and a set of standard reference solutes was selected for which the Z, values were defined as the arithmetic mean of the values of polar indices calculated by linear regression of the log k* versus P' plots in fifteen mobile phases with different concentrations of methanol, acetonitrile, tetrahydrofuran and dioxane in water [111. The correct selection of the reference standard compounds for calibration of the log k* scale is very important for the predictive power of the interaction indices approach. To first approximation, the specific interactions between the solutes and the components of the mobile phase are neglected. The relative importance of these interactions may depend on the composition of the mobile phase and consequently, more or less significant deviations from the linear log k* -I, plots may be observed with some solutes and chromatographic systems. Such compounds should be avoided as reference standards as they would yield incorrect coefficients A and B of the calibration straight lines fitted by regression analysis to the data in various reversed-phase systems and the predicted capacity factors would be less accurate. Further, the time necessary to perform the calibration experiments makes it impractical to use more than five to six reference standards. These compounds should be selected to cover a rather broad range of mobile phase composition, to be stable, readily available and to absorb in the UV region for convenient detection. Benzene, toluene, nitrobenzene, acetophenone and anisole were initially selected as the calibration standards. Later, various groups of solutes were tested for least deviations from the log k* versus Z, regression lines in different mobile phases. Based on these experiments, a new series of calibration standards was selected: 1,Cdichlorobenzene TABLE 7.2A POLARITY INDICES, P', MOLAR VOLUMES, V, AND AVERAGE VALUES OF THE INTERACTION INDICES, I,, OF REFERENCE STANDARDS FOR CALIBRATIONOF THE RETENTION SCALE (EQ. 7.2) Solute

P'

V, (cm3/mol

4

Set 1 Toluene Benzene Anisol Nitrobenzene Acetophenone

2.3 3.0 3.5 4.5 4.4

1.063 0.889 1.086 1.023 1.169

2.46 2.76 3.85 4.49 5.60

0.959 1.181 1.186 1.023 1.021

1.05 2.20 3.44 4.49 5.32

Set 2 1,4-DichIorobenzene 3-Chlorotoluene 3-Bromonitrobenzene Nitrobenzene Benzonitrile Data from refs. [ll], [24].

References pp. 266-267

-

4.5 4.6

250

Chapter 7

TABLE 7.2B RELATIVE DEVIATIONS (RD) FROM THE REGRESSION LINE OF THE LOG k* VERSUS ,I DATA MEASURED FOR 27 NON-STANDARD COMPOUNDS FROM THE CALIBRATION LINES WITH SET 1 AND SET 2 OF REFERENCE STANDARDS IN METHANOL-WATER 60:40 (a), ACETONITRILEWATER 50150(b), TETRAHYDROFURAN-WATER 30170(c) AND DIOXANE-WATER 50:50 (d) Mobile

RD (“A), Set 1

phase

intercept

slope

a b

6.43 5.87 -12.33 -3.86

10.77 10.18 -23.03 -7.14

C

d

RD (“h),Set 2 intercept

slope

1.10 0.74 -2.48 -1.25

0 3.35 -6.85 -2.69

(I, = 1.05), 3-chlorotoluene (Z, = 2.20) 3-bromonitrobenzene (I, = 3.44), nitrobenzene (I, = 4.49) and benzonitrile (I, = 5.32) [24]. The relative standard deviations for the interaction indices of these standards were between 4.2 and 6.8%. The polarity indices, molar volumes and the interaction indices of the two sets of standards are given in Table 7.2A. The values of I, of the standards are similar to the P‘, but this does not apply generally and, for instance, the values are very different for the organic solvents used as the components of the mobile phases. The data in Table 7.2B show that the second set of reference standards yield significantly lower deviations of the calibration line from the regression line fitted to the data of a larger set of non-standard compounds in different mobile phases, which indicates better suitability of the second set for calibration of the log k* versus Z, dependence [24]. 7.4 PREDICTION OF THE RETENTION UNDER CHANGING MOBILE PHASE COMPOSITION USING INTERACTION INDICES

The inherent limitations of the model used, especially neglecting the role of the stationary phase in the retention mechanism, makes absolute prediction of retention using the present model questionable. However, the non-specific stationary phase interactions are more or less taken into account in the calibration lines. In this case, the values of the coefficients A’,,, Ro, are affected by these interactions and are no longer described exactly by Eqs. (7.1 1H7.14). With correct choice of standard reference compounds used for the construction of the calibration straight line according to Eq. (7.2), a fairly accurate relative prediction of retention on the specific capacity factor scale should still be possible. This means that the capacity factors in a given reversed-phase system can be estimated from the data obtained in another system. With the coefficients A and B of the calibration straight line determined by linear regression of the experimental plots of log k* versus ,Z of the reference standards in a given reversed-phase system, Eq. (7.2) can be used to determine the I, of sample solutes from their capacity factors k. If the constants A and B are known for another system, the interaction indices of the sample solutes can be used for prediction of their k in this system.

Characterization of retention and selectivity in reversed-phaseLC using interaction indices

25 1

TABLE 7.3 INTERACTIONINDICESzx AND MOLAR VOLUMES vx(IN cm3/mol I 0-*) OF NON-REFERENCE COMPOUNDS Solute

Solvents Water Methanol Acetonitrile Tetrahydrofuran Dioxane Aromatic compounds 1,2-Dimethylbenzene 1,3-Dimethylbenzene Ethylbenzene nSutylbenzene n-Pentylbenzene n-Hexylbenzene n-Heptylbenzene n-Octylbenzene n-Nonylbenzene Cyclohexylbenzene Biphenyl Styrene Phenylacetylene Chlorobenzene Bromobenzene 1,2-Dichlorobenzene 1,3-Dichlorobenzene a,a-Dichlorotoluene ap,a-Trichlorotoluene Ethoxybenzene n-Propylphenylether 1,3-Dinitrobenzene 1,4-Dinitrobenzener Benzophenone Methyl benzoate Ethyl benzoate Phenyl acetate Benzatdehyde 3-Nitrobenzaldehyde Benzylalcohol Phenol o-Cresol mCresol p-Cresol Thiophenol Ethylphenylcarbamate Butylphenylcarbamate Benzylamine Aniline

References pp. 266-267

4 47.0 21.1 18.3

VX

18.8

0.18 0.40 0.52 0.81 0.85

2.13 2.15 2.44 1.93 1.68 1.43 1.18 0.92 0.67 1.41 3.43 2.35 1.50 2.46 2.23 2.01 1.73 2.78 1.29 3.36 3.01 4.46 4.71 3.57 4.09 3.13 5.66 5.40 5.78 7.10 8.55 6.53 6.71 6.60 3.68 6.21 4.14 5.11 6.62

1.21 1.22 1.23 1.55 1.72 1.88 2.04 2.21 2.37 1.69 1.56 1.14 1.10 1.01 1.05 1.13 1.14 1.29 1.42 1.28 1.43 1.26 1.03 1.59 .1.25 1.43 1.25 1.02 1.18 1.04 0.88 1.03 1.05 1.05 1.07 1.65 1.78 1.09 0.91

11.4

Solute Quinoline Pyridine Phenylurea Linuron Chlorobromuron Acetanilide Benzamide N-Methylbenzamide N,N-Dimethylbenzamide Phenylsulphonamide Tolylsulphonamide 3,5-Dinitrobenzoatesof Methanol Ethanol 1-Propanol 1Butanol 1-Pentanol 1-Hexanol p-Bromophenacyl esters of Acetic acid Propionic acid n-Butyric acid n-Valeric acid n-Caproic acid n-Caprylic acid

Aliphatic compounds n-Pentane n-Hexane n-Heptane n-Octane n-Nonane n-Decane Di-n-butylether n-Butylbromide Acetone Butan-2-one Pentan-2-one Hexan-2-one 6-Methyluracils 3,6-Dimethyl3-Ethyl3-n-Propyl3-n-ButylMethyl formiate Methyl acetate Methyl propionate

4

VX

4.26 5.75 8.61 3.84 4.30 7.37 8.75 8.37 7.76 8.96 7.87

1.18 0.81 1.09 1.59 1.95 1.18 1.12 1.36 1.51 1.57 1.71

5.12 4.73 4.33 3.94 3.54 3.15

1.62 1.74 1.96 2.07 2.19

5.23 4.88 4.52 4.16 3.81 3.09

1.77 1.91 2.04 2.18 2.31 2.58

-0.27 -0.63 -1.00 -1.38 -1.80 -2.25 2.35 1.29 9.88 7.99 6.69 5.77

1.15 1.31 1.47 1.63 1.79 1.95 1.70 1.08 0.74 0.92 1.07 1.23

10.67 9.67 8.52 7.99 8.94 7.78 6.61

1.40 1.54 1.68 1.82 0.62 0.79 0.96

1.85

252

Chapter 7

TABLE 7.3 (continued) Solute

1,

Methyl butyrate Methyl nonanoate Methyl undecanoate

5.45 2.81 2.01

V.

1.12 1.96 2.29

Solute

1,

VX

Methyl dodecanoate Methyl tridecanoate Methyl pentadecanoate

1.61 1.22 0.42

2.46 2.62 2.96

Adapted from refs. [11,14,24,25,28,30].

If low-molecular organic compounds of similar size are separated and their molar volumes are not known or their determination is inconvenient, approximately constant V, can be assumed to first approximation and can be included into the constants of Eq. (7.2) in the calculations. Interaction indices of the sample solutes studied so far are listed in Table 7.3 as mean values determined in different reversed-phase systems [ 11,14,16,18,24,25]. The capacity factors predicted using the interaction indices usually agree with the experimental values within 10-15% rel., but occasionally the deviations may exceed 20%, mainly for basic compounds, probably because of the effect of the stationary phase interactions, neglected in the present simple model [26]. Another possible reason for the deviations of the calculated values fiom the experimental k may be selective polar interactions in the mobile phase, which could be accounted for by introduction of either specific values of the constants c, (Eq. 7.2) for solutes with specific interactions or different indices I, to describe the retention in various mobile phases. This approach is discussed in detail in Chapter 8, however, it leads to the loss of the simplicity of using a single index for characterization of retention under varying experimental conditions. Several examples of prediction of retention are shown in Table 7.4. for a C8 column and methanol-water and acetonitrile-water mobile phases. An attempt was reported to account for the effects of the entropic changes during the transition of the solutes between the mobile and the stationary phases by introduction of additional terms into Eq. (7.2) [27], but this equation thus becomes too complex for easy practical applications. The interaction indices approach can be used not only in pure aqueous-organic mobile phases, but also in mobile phases containing neutral electrolytes as additives. This is demonstrated by the agreement of the values of the interaction indices in 60% methanol and in the mobile phases with the addition of lithium sulphate (Table 7.5) [25]. The interaction indices of neutral compounds in mobile phases containing lithium sulphate are only slightly higher than in the mobile phases without the addition of the salt. Acids are eluted as strongly deformed peaks close to the column dead volume in pure aqueous-organic mobile phases. The addition of a salt suppresses the repulsive interactions of the residual negatively charged groups on the surface of the packing material and introduces salting out of the acids from the mobile into the stationary phase. The interaction indices increase with the acidity and decrease with the size of the non-polar part in the molecules of the acids. The concentration of the salt does not significantly affect the values of the interaction indices and of the constants A and B in Eq. (7.2). The interaction indices can be significantly affected in the presence of the electrolytes

Characterization of retention and selectivity in reversed-phase LC using interaction indices

253

TABLE 7.4 EXPERIMENTAL (E) AND PREDICTED (p) (EQ. 7.2) CAPACITY FACTORS OF SEVERAL NONSTANDARD COMPOUNDS Values from a Silasorb C8 column with methanol-water 60:40 (I), 70:30 (II), 80:20 (111) and acetonitrile-water 5O:SO (IV), 60:40 0,70:30 (VI) mobile phases Solute

Mobile phase

k 0-Cresol Benzaldehyde Methyl benzoate Ethoxybenzene nButyl phenylcarbamate Benzophenone Ethyl benzoate Styrene Bromobenzene n-Propyl phenyl ether nButylbromide Di-n-butyl ether a,a,a-Trichlorotoluene

E P E P E P E P E P E P E P E P E P E P E P E P E P

I 0.47 0.40 0.52 0.52 1.02 1.04 1.40 1.39 1.49 1.58 1.83 1.78 1.86 1.85 1.82 1.74 1.91 1.70 1.86 1.94 1.92 2.01 3.56 3.70 4.10 4.11

rl

111

IV

V

VI

0.24 0.21 0.30 0.28 0.48 0.48 0.63 0.63 0.64 0.52 0.76 0.66 0.79 0.74 0.82 0.83 0.86 0.86 0.82 0.78 0.88 0.99 1.33 1.24 1.45 1.60

0.13 0.13 0.16 0.17 0.24 0.23 0.33 0.28

0.61 0.58 0.77 0.72 1.10 1.39 1.53 1.77 1.85 2.26 2.04 2.38 1.99 2.35 2.01 2.08 2.03 1.99 2.13 2.45 2.32 2.3 1 3.67 4.53 3.98 4.62

0.38 0.40 0.49 0.48 0.71 0.79 0.99 0.97 0.98 1.01 1.06 1.13 1.15 1.17 1.09 1.16 1.16 1.15 1.21 1.22 1.24 1.30 1.87 1.91 1.85 2.12

0.24 0.24 0.29 0.29 0.41 0.40 0.48 0.48 0.53 0.39 0.57 0.47 0.60 0.52 0.58 0.59 0.62 0.61 0.65 0.54 0.69 0.67 1.01 0.70 0.99 0.87

0.25

0.19 0.34 0.25 0.35 0.30 0.38 0.38 0.39 0.40 0.37 0.31 0.41 0.45 0.58 0.39 0.58 0.55

taking part in secondary equilibria in the chromatographic system, such as in the formation of ion pairs or in pH controlled acid-base equilibria. 7.5 INTERACTION INDICES AND THE SELECTIVITY OF SEPARATION 7.5.1 Non-homologous compounds

Introducing Eq. (7.2) for two solutes i andj and combining with Eqs. (7.9) and (7. lo), the following relationship can be obtained for the relative retention of the two compounds, r,,j= kj/ki, which is conveniently used to characterize the selectivity of separation [ 191: (7.23)

References pp. 266-267

254

Chapter 7

TABLE 7.5 INTERACTION INDICES I, OF SEVERAL NON-STANDARD COMPOUNDS ON A SILASORB C8 COLUMN IN 60% METHANOL (A), 0.04 M Li2SO4 IN 60% METHANOL (B) AND 0.20 M Li2SO4 IN 60% METHANOL (C) Mobile phase: Solute:

a I,

b

C

1,

4

p-Bromophenacyl esters of acids Acetic Propionic Valeric Caproic

5.34 4.89 4.02 3.61

5.37 4.94 4.05 3.62

5.53 5.08 4.18 3.76

3,5-Dinitrobenzoates of alcohols Methanol Ethanol 1-Propanol I-Butanol I-Pentanol 1-Hexanol

5.51 4.88 4.3 1 3.75 3.25 2.79

5.60 4.98 4.39 3.83 3.30 2.82

5.64 4.98 4.40 3.84 3.32 2.84

7.42 6.80 9.47 6.95 8.46 7.84 7.23 8.66 6.65 6.59 6.21 6.15 6.25 6.28 5.61 5.74

7.54 6.92 9.08 6.99 8.78 8.07 7.55 8.42 6.55 6.46 6.16 6.10 6.15 6.22 5.67 5.83

1.09 0.15

1.12 0.15

Organic acids

Anthraquinone-I-sulphonic Anthraquinone-2-sulphonic Benzoic a-Naphthylacetic 3-Indolylacetic 3-Indolylpropionic 3-Indolylbutyric Phenoxyacetic 2,4-Dichloroacetic 2-Methyl-4-chloroacetic 2-Methyl-4,6-dichloroacetic 2,4,5-Trichloroacetic 2-(2-Methyl-4-chlorophenoxy) propionic 2-(2,4-Dichlorophenoxy)propionic

2-(2-Methyl-4-chlorophenoxy)butyric 2-(2,4-Dichlorophenoxy)butyric 1.16 0.16 aA, B are the constants in Eq. (7.2).

The indices i,j, in Eq. (7.23) denote the parameters Vx Zx of the compounds i andj, respectively. According to Eq. (7.23) the separation selectivity of a pair of compounds in reversedphase systems is controlled by the differences in the size and in the polarities of the two compounds. Hence, compounds of similar polarities, Z4 =Ixi, can be separated on the basis of the differences between the size of their molecules. Because the constants B’o, B’l are positive (see Eqs. (7.1 1-7.14)), the separation selectivity theoretically

Characterization of retention and selectivity in reversed-phase LC using interaction indices

255

should increase with increasing difference between the size of the molecules of i andj and decrease with increasing concentration of the organic solvent, $, in the mobile phase, which agrees with the experimental observations. According to Eq. (7.23) the selectivity change per concentration unit is expected to be directly proportional to the difference in the size of the molecules of separated compounds. If the molecules of two compounds i a n d j are of approximately equal size, VxJ= V,, the less polar compound is retained more strongly, i.e.' I <,Z and the separation selectivity decreases with increasing $ proportionally to the difference ZB--Zx. These predictions are in agreement with numerous experimental observations. The retention behaviour is more complex if one compound of the pair to be separated is both bulkier and more polar than the other, i.e. V4 > Vx and ZxJ > I,. In this instance the difference VxJZxJ - V,l, is large and significant effects of the composition of the mobile phase on the selectivity of separation can be expected, sometimes connected even with the reversal of the order of elution, which may occur at a certain concentration $o of the organic solvent in the mobile phase. $0 can be calculated from Eq. (7.23) setting r,,, = 1 and can be either >1 or 4 , if the concentration of the organic solvent is expressed in % vol/vol. 102. Only $o < 1 has physical meaning of the concentration at which the reversal of the order of elution occurs. If calculated $0 > 1, the order of elution remains unchanged over the whole concentration range of $. From Eq. (7.23) it follows that the value of (VxJZxJ - VxZx,) has the opposite effect on r,,, than the difference in the molecular size (VxJ- V,) and consequently a large value of ( Vx& - Vx,Zx,)increases the probability that &, falls below 1, into the range of real concentrations. For example, a change in the order of elution of benzene and anisole was observed experimentally on a Lichrosorb C18 column in 46% methanol, 3 1% acetonitrile and 55% dioxane, while the values of $o predicted from Eq. (7.23) are 43% methanol, 40% acetonitrile and 50% dioxane [19]. It should be noted that the model of interaction indices can predict only selectivity changes caused by the combination of the contributions of the molecular size and polarity. Because of the simplifying assumptions adopted, it is not possible to predict small changes in selectivity caused by specific solvation or by specific polar interactions and as a small variation in selectivity may have a significant effect on the separation, additional fine tuning of the separation conditions is usually necessary.

7.5.2 Homologous and oligomeric series

The interaction indices approach can be used to describe the retention behaviour of homologous series with methylene groups or oligomeric series with other structural repeat units. Both the size (molar volumes, Vx) and the polarity (interaction indices, Zx) of members of such a series increase or decrease regularly with the number of repeat units, n [28301.

Vx = Vo, + A V x n

(7.24)

I , = I,, +AZ,n

(7.25)

References pp. 266267

256

Chapter 7

Vox and Zox are the molar volume and the interaction index, respectively, of the end groups in the series and A Vx Mxare the increments of the molar volume and of the interaction index in the series. The validity of Eq. (7.24) is straightforward and has been verified experimentally [25]. Figure 7.9 illustrates the validity of Eq. (7.25) for four different homologous series [25]. The AVx of the repeat methylene unit in these series is approximately 16 cm3/mol-'. As the methylene unit is non-polar, the polarities of the homologues decrease regularly with n and the values of AZx are negative and close to each other in various homologous series (e.g. Mx= -0.40 for the esters of aliphatic carboxylic acids, Mx= -0.36 for the esters of n-alcohols and Nx= -0.25 for n-alkylbenzenes). Combination of Eqs. (7.1), (7.24) and (7.25) results in the following quadratic expression for the dependence of the logarithms of k on the number of the repeat units, n [28]: vOxcM

(c 1 2

logk = log@+M M 2.31RT

- C X ~ M ~ O X )

(7.26)

The empirical Martin's additivity rule predicts a linear dependence of log k n and linear plots are found experimentally for various homologous and oligomeric series in different chromatographic systems. Sometimes deviations fiom linearity are observed, 12

10 -

IW'

8

6

4 ' 0

I

I

I

I

1

2

3

4

5

n Fig. 7.9. Dependence of the interaction indices, I,, on the number of methylene units, n, in homologous methylesters of n-alkanecarboxylic acids (I), alkane-2-ones (2), 3-alkoxycarbonylpyrrazolines (3) and 3-alkyl-6methyluracils (4) on a Silasorb SPH C18 column in methanol-water and acetonitrilewater binary mobile phases containing 25-90% of the organic solvent.

Characterization of retention and selectivity in reversed-phase LC using interaction indices

I

I

-5 0

2

257

4

6

8

10

12

n Fig. 7.10. Dependence of the product IxV’xon the number of repeat structural units n in homologous n-alkanes (l), n-alkylbenzenes (2), methylesters of n-alkanecarboxylic acids (3), 3,5-dinitrobenzoates of n-alkanols (4) andp-bromophenacyl esters of n-alkanecarboxylic acids (5) on a Silasorb C18 column in methanol-water and acetonitrile-water mobile phases.

mainly if the retention is investigated over a wide range of n. These differences are usually attributed to possible changes in the conformation of the chains of repeat structural units occurring when a certain size of the molecule is exceeded or to the effects of the length of bonded alkyls in the stationary phase, but the quadratic term in Eq. (7.26) offers another explanation: the effect of the interaction of simultaneous regular changes of the polarity and the size of the molecules in the series 1281. This behaviour is illustrated by Figs. 7.10 and 7.1 1, where the values of the product ZxVx are plotted as a function of the number of the repeat structural units in several homologous and oligomeric series. The non-linear shape of these plots is caused by the interaction term hVxAIx and its magnitude depends on the type of the series and both the slopes and the curvatures of these plots are larger with oligomeric than with homologous series, where the polarity AZx of the repeat methylene group is small. This is the reason why the log k versus n plots for oligomeric series are more likely to be curved than the plots for homologous series; see an example of a set of these plots in Fig. 7.12. However, the quadratic term is often small and can be neglected in practice. Neglecting the quadratic term with n2 and introducing,Z from Eq. (7.3), Eq. (7.26) can be rewritten to describe the retention in a homologous or an oligomeric series as a function of two variables, the number of the repeat structural units and the composition of a binary mobile phase [28,30]: l o g k = a o +aln-(rno +rnp)$+(do +d1n)(b2

References pp. 266-267

(7.27)

258

Chapter 7 250

200

150 U

100

50

0

2

0

6

4

8

10

n Fig. 7.1 1. Dependence of the product IxVxon the number of repeat structural units n in oligostyrenes (I), oligoethylene glycols (2) and ethoxylated nonylphenols (3) on a Silasorb C18 column in dioxane-water (l), methanol-water (2) and 1-propanol-water (3) mobile phases.

where

.3 with c1 = ~ ~ 1 21RT. From the comparison of Eqs. (7.4) and (7.27), it follows that the parameters a, m, d of Eq. (7.4) are linear functions of the number of repeat structural units n in a homologous or oligomeric series, which results in linear correlations between the parameters a, m and

d: a0 “1 m=mo-ml-+-a=q+pa a1

a1

(7.34)

Characterization of retention and selectivity in reversed-phase LC using interaction indices

259

1.60

1.20

Y

0.80 0

CI

0.40

0.00

0

4

8

12

16

20

n Fig. 7.12. Dependence of the capacity factors, k, on the number of oligomeric units, n, in oligostyrenes measured on a Separon SIX C18 column in dioxanewater 75:25 (I), 77.5:22.5 (2), 80:20 (3), 82.5:17.5 (4) and 85:15 (5) mobile phases.

a0 dl d = do - d , -+-a a1 a1

= s+ra

(7.35)

The coefficients q, p, s and r depend on the stationary phase, organic solvent and the type of the series. Approximate correlation between the parameters m and a was observed experimentally earlier, even for some non-homologous compounds in methanol-water mobile phases [31,32] but the validity of Eq. (7.34) was much better within individual homologous or oligomeric series in aqueous-organic mobile phases containing methanol, acetonitrile, tetrahydrofixan, dioxane and propanol [25,28,30,33-341. The parameters ml, al are approximately constant with various homologous series and the slope p of the relationship (7.34) acquires a value close to p = 2( 1 = 0.7-0.9 for methanol-water and acetonitrile-water mobile phases [28]. The values of p significantly higher or lower may be found with oligomeric series containing repeat structural units other than the methylene group, depending on the size and polarities of both the repeat unit and the end groups [30]. The concentration of the organic solvent in the mobile phase, $, and the number of repeat units in the molecule, n, simultaneously affect the free energy of the transfer of a member of a homologous or an oligomeric series from the mobile to the stationary phase and the two effects act most often against each other. At a certain combination of $J = $Jc and n = n,, the two effects compensate each the other, which means that theoretically a composition of the mobile phase can be found where all the members of a given series are eluted with the same retention time. Hence a set of log k versus n plots measured at different compositions of a binary mobile phase has a fan-like appearance and a common convergence point with the coordinates n,, log k,. A similar set of fan-like straight lines or References pp. 266267

260

Chapter 7

curves is found if log k values are plotted as a function of r$ for different members of a homologous or oligomeric series and the common convergence point has the coordinates qj,, log k,. If the quadratic term in Eq. (7.27) is neglected and the dependence of retention on the composition of the mobile phase is adequately described by Eq. (7.8), the coordinates can be derived as [25,28,29,30,35]: (7.36)

(7.37)

(7.38)

Fan-like plots with a convergence point have been observed experimentally with numerous series and chromatographic systems [25,28-31,361; two examples are shown in Figs. 7.13 and 7.14. Numerical values of the coordinates of the convergence points calculated from the experimental data for several homologous series, columns and mobile phases are given in Table 7.6. The log k, coordinate of the convergence point for a homologous series is significantly contributed to by the logarithm of the phase ratio in the column, log @, and is

1.60

'

-1.60 -10

I

-5

I

I

0

5

10

n Fig. 7.13. Intersection point of the dependences of logk on the number of methylene units, n, in nalkylbenzeneson a Silasorb CIS column in methanol-water 70:30 (l), 75:25 (2), 80:20 (3) and 9O:lO (4).

Characterization of retention and selectivity in reversed-phase LC using interaction indices

26 1

1.60

'

-1.60 0.60

I

I

I

I

I

0.10

0.80

0.90

1.00

1.10

1.20

(P Fig. 7.14. Intersection point of the dependences of log k on the concentration $ (%vol/vol.lo-*) of methanol in methanol-water mobile phases for ethylbenzene (l), n-butylbenzene (2), n-pentylbenzene (3), n-hexylbenzene (4) and n-heptylbenzene (5) on a Silasorb C18 column.

therefore expected to be relatively independent of the type of the series. The log k, values found experimentally were -1.03 to -1.1 1 for Silasorb C18, -1.29 for Silasorb C8, -1.63 for Lichrosorb RP18 and -1.4 for Hypersil ODS (Table 7.6). These values are somewhat lower than the values of log @ determined using Vs calculated as VG- V,, but can be used as a characteristic of the stationary phase (column) [29]. The coordinate 4, is expected to depend on the type of the organic solvent in the mobile phase rather than on the column or on the homologous series, in approximate agreement with the data in Table 7.6 ( 1.08-1.16 for methanol-water, 0.95 for dioxane-water and 0.86 for tetrahydrofuran-water mobile phases). The values of the coordinates n, in this table depend not only on the type of the series, but also on the organic solvent in the mobile phase. The negative values of n, mean that the convergence points cannot be observed under real chromatographic conditions with various homologous series. The occurrence of the convergence point is not limited to the linear log k versus 4 plots. If the quadratic term in Eq. (7.27) cannot be neglected, the coordinates of the convergence point for a set of log k versus n and of log k versus @ plots can be obtained from Eqs. (7.39H7.41):

d,@:-m1$, +a, = 0

(7.39)

logk, =ao - m 0 @ , +do$:

(7.40)

n, =- do -mo m1-dl

(7.41)

References pp. 266267

262

Chapter 7

TABLE 7.6 COORDINATES OF THE CONVERGENCE POINTS FOR HOMOLOGOUS SERIES Compounds: n-alkylbenzenes (l), p-bromophenacyl esters of aliphatic carboxylic acids (2). 3,s-dinitrobenzoof n-alkylamines (4) on Silasorb ates of n-alkanols (3) and 1,2-naphthoylenebenzimidazole-6-sulphonamides C18 (I), Silasorb C8 (II), Lichrosorb RPl8 (111) and Hypersil ODS (IV) columns in methanol-water (a), dioxane-water (b) and tetrahydrofuran-water mobile phases Homologous series

Mobile phase

Column

1% kc

nC

@C

1

a a a

I I1 IV

2

a

111

3

b b

I I I I

-1.11 -1.29 -1.4 -1.63 -1.10 -1.13 -1.03 -1.08

-5.64 -5.65 -5.8 -6.46 4.68 -5.14 -7.99 -6.95

1.15 1.08 N.D.a 1.16 0.95 0.95 0.86 1.11

C

4

a

Values of log kc, no q5c coordinates were calculated from the experimental data in refs. [28], [29] and [30] using Eqs. (7.36)-(7.38). WD., not determined.

For all homologous and most oligomeric series, the convergence point falls out of the range of real values of $ and n, so that it cannot be observed experimentally (see Figs. 7.13 and7.14). The expression for the selectivity of separation, i.e. for the relative retention of two adjacent members i and j in a given homologous or oligomeric series, r,,, = a,can be derived from Eq. (7.27): log a = al - ml$+ d1$2

(7.42)

Neglecting the quadratic term and introducing Eqs. (7.30) and (7.31) for al and ml, Eq. (7.42) can be rearranged [30]as (7.43) where (7.44) From Eq. (7.43), it follows that log a can be either positive or negative, i.e. the members of a series can be eluted in the order of either increasing or decreasing size. The experimentally observed separation selectivity in a homologous or in an oligomeric series usually increases with increasing AV, and with increasing polarity, Zorg, and decreasing concentration,$, of the organic solvent in the mobile phase, in agreement with Eq. (7.43).

Characterization of retention and selectivity in reversed-phase LC using interaction indices

263

However, some oligomeric series with relatively polar repeat structural units sometimes show exceptions from this rule. For example, ethylene glycols are eluted in the order of increasing size, but the order of elution of ethoxylated nonylphenols is reversed in propanol-water mobile phases [37]. At first glance, this behaviour could seem to be caused by higher concentration and lower polarity of the organic solvent in the mobile phase which was necessary because of a much stronger retention of the ethoxylated nonylphenols with a bulky and relatively non-polar end group. Theoretically, if the polarity of the mobile phase is decreased, it can drop below the polarity of the repeat group, which then shows a higher affinity to the mobile than to the stationary phase and increasing number of the repeat structural units speeds up the elution. However, this cannot explain the behaviour observed, because the polarity of an ether (oxyethylene) group obviously is not higher than that of a mixture of propanol with water and because the retention still decreases when the concentration of the organic solvent in the mobile phase is increased. The model of interaction indices offers the explanation by the effect of the combined structural parameter Q in Eq. (7.43) comprising the size (AVx, Vox)and polarity (Nxlox) parameters of both the repeat structural unit and the end groups [30,311. Q decreases with increasing molar volume of the repeat structural unit, AVx, if log a is positive. The parameter Q is low if hlx is low or negative, i.e. if the repeat structural unit is relatively non-polar such as the methylene group in homologous series, where NX= -0.3 in methanol-water mobile phases. Consequently, approximately constant values of the selectivities, log a, were found for various homologous series in a given mobile phase [28]. On the other hand, the parameter Q in oligomeric series may become large if hlx is positive and AVx is low, i.e. for oligomeric series with relatively small and polar repeat structural units. In this instance Q increases with increasing size, Vex, and polarity, lox,of the end groups. This means that we can expect significant differences in the separation selectivities between two oligomeric series with an equal repeat structural unit, if the end groups in one series are significantly bulkier than in the other. The physical meaning of the effect of the end groups on the separation selectivity possibly can be explained as follows [25]. The end groups in an oligomer form a part of the environment of the repeat structural units, together with the surrounding molecules of the mobile phase and if bulky and polar, they may interact with the repeat units. As the parameter Q is significantly higher with ethoxylated nonylphenols (Q = 13.3) than with oligoethylene glycols (Q = 8. l), the selectivity for an oxyethylene group calculated from Eq. (7.43) using these values of Q is positive with the first and negative with the second series, even though the repeat oligomeric groups are the same, which seems to explain the retention behaviour observed experimentally [30].

7.6 CONCLUSIONS

The semi-empirical model of interaction indices makes use of a single parameter to characterize the polarity and neglects the role of the stationary phase in the separation and does not distinguish between different specific polarity effects such as dipole-dipole, proton-acceptor or proton-donor interactions between the molecules of the solute and of the components of the mobile phase. The retention and the selectivity of separation Referencespp. 2 6 6 2 6 7

264

Chapter 7

are understood to result from the contributions of the polarities and of the size of the molecules to the free energy of transfer of the solute from the mobile to the stationary phase. In spite of the simplifications adopted in the derivation, the model of interaction indices makes it possible to explain some experimentally observed phenomena in binary and ternary mobile phases, if they result from the simultaneous effects of the molecular size and polarity on the selectivity of separation. To use the model of interaction indices for the characterization and prediction of retention successfully, it is necessary to determine the ratio of the volumes of the stationary and mobile phases in the column and to calibrate the scale of the interaction indices using an appropriate set of reference standards. If different chromatographic systems are calibrated in this way, the retention in different mobile phases and (or) columns can be predicted from the interaction indices determined with another column and mobile phase. The relative errors of predicted retention volumes of approximately 10% can be expected, but larger deviations may occasionally be observed, especially with compounds showing strong specific interactions in a given chromatographic system. To decrease the error of prediction, a more sophisticated approach should be adopted taking into account selective effects of polar interactions in mobile phases containing the solvents from different selectivity classes, such as the two-indices approach described in Chapter 8.

7.7 GLOSSARY OF THE TERMS

A

B

IH,O IM

Iorg Iorg ,P Iorg ,xt Iorg ,y

-AG -AGM-M

intercept of the calibration equation (Eq. 7.2) for the scale of interaction indices coefficients of the linear dependence (Eq. 7.9) of the intercept A on the concentration of the organic solvent in the mobile phase, t$ slope of the calibration equation (Eq. 7.2) for the scale of interaction indices coefficients ofthe linear dependence (Eq. 7.10) of the slope B on the concentration of the organic solvent in the mobile phase, q5 interaction index of water interaction index of mobile phase interaction index of an organic solvent interaction indices of organic solvents i, X, Y in multicomponent mobile phases interaction index of a sample solute (of sample solutes i andj, respectively) interaction index of the end group in a homologous or oligomeric series contribution to IX of a repeat structural unit in a homologous or in an oligomeric series total fiee energy of retention free energy of mutual interactions between the molecules of the mobile phase

Characterization of retention and selectivity in reversed-phase LC using interaction indices

265

fiee energy of interactions between the molecules of the solute and the molecules of the mobile phase polarity index according to Snyder's definition [131 term characterizing combined effects of the size and polarity of the repeat structural unit and of the end groups in an oligomeric series (Eq. 7.44) gas constant temperature in degrees Kelvin geometrical inner volume of the empty column volume of the mobile phase in the column volume of the stationary phase in the column molar volumes of a sample solute (sample solutes i and j ) , in 1 mol-' 10 molar volume of the end group in a homologous or in an oligomeric series contribution to Vx of a repeat structural unit in a homologous or oligomeric series intercept of the log k versus 9 relationships (Eqs. (7.4), (7.8) or (7.191) intercepts of Eq. (7.8) fitted by linear regression to the experimental data in binary mobile phases water-organic solvent X and water-organic solvent Y intercept of the linear dependence of a on n in a homologous or in an oligomeric series (Eq. 7.28) slope of the linear dependence of a on n in a homologous or oligomeric series (Eq. 7.30) proportionality constant between ZM and the energy of interaction for a molecule of the mobile phase proportionality constant between Zx and the energy of interaction for a molecule of the solute second power term parameter of the log k versus @ relationships (Eq. 7.4) d i n binary aqueous mobile phases containing organic solvents i, j , X or Y(Eqs. (7.17), (7.18)) intercept of the linear dependence of d on n in a homologous or oligomeric series (Eq. 7.32) slope of the linear dependence of d on n in a homologous or oligomeric series (Eq. 7.33) capacity factor of a sample solute (solute i orj, respectively) specific capacity factor (Eq. 7.2) k* in binary mobile phases water-organic solvent X or waterorganic solvent Y k corresponding to the convergence point of the logk versus n plots or of the log k versus 9 plots for a homologous or oligomeric series Referencespp. 266-267

Chapter 7

slope parameter of the log k versus @ dependences (Eqs. (7.4), (7.8) or (7.17)) m in binary aqueous mobile phases containing organic solvent i, X or Y (Eqs. (7.17), (7.18)) intercept of the linear dependence of m on n in a homologous or oligomeric series (Eq. 7.29) slope of the linear dependence of m on n in a homologous or oligomeric series (Eq. 7.3 1) number of repeat structural units in the molecule of a homologue or an oligomer n of the common intersection point of the log k versus n dependences for a homologous or an oligomeric series slope of the linear correlation between the parameters m and a for a homologous or oligomeric series (Eq. 7.34) intercept of the linear correlation between the parameters m and a for a homologous or oligomeric series (Eq. 7.34) slope of the linear correlation between the parameters d and a for a homologous or oligomeric series (Eq. 7.35) relative retention r,,j = $/ki intercept of the linear correlation between the parameters d and a for a homologous or oIigomeric series (Eq. 7.35) error term in prediction of k* in ternary mobile phases (Eqs. (7.21), (7.22)) phase ratio in the column, Qi = VdY, relative retention rj,,for two adjacent homologues or oligomers (Eq. 7.42) concentration of the organic solvent in an aqueous-organic mobile phase I#J of organic solvents i, X and Y in multicomponent aqueous-organic mobile phases 9 of the convergence point of the log k versus @ dependences for a homologous or oligomeric series volume proportions of binary mobile phases containing organic solvents Xand Ymixed to prepare a ternary mobile phase (Eq. 7.2 1)

7.8 REFERENCES P. Jandera and J. Churiikk, J. Chromatogr., 91 (1974) 207. R. Tijssen, H.A.H. Billiet and P. Schoenmakers, J. Chromatogr., 128 (1976) 65. P.J. Schoenmakers, H.A.H. Billiet, R. Tijssen and L. De Galan, J. Chromatogr., 149 (1978) 519. P. Jandera, J. Chur6Eek and L. Svoboda, J. Chromatogr., 174 (1979) 35. B. L. Karger, J.R. Gant, A. Hartkopf and P.H. Weiner, J. Chromatogr., 128 (1976) 65. Cs. Horviith, W. Melander and I. Molnhr, J. Chromatogr., 125 (1976) 129. Cs. Horvhth and W. Melander, J. Chromatogr. Sci., 15 (1977) 393. D.E. Martire and R.E. Boehm, J. Phys. Chem., 87 (1983) 1062.

Characterization ofretention and selectivity in reversed-phase LC using interaction indices 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

267

R.M. McCormik and B.L. Karger, Anal. Chern., 52 (1980) 2249. H.E. Slaats, W. Markowski, J. Fekete and H. Poppe, J. Chromatogr., 207 (1981) 299. P. Jandera, H. Colin and G. Guiochon, Anal. Chem., 54 (1982) 435. G.E. Berendsen and L. De Galan, .I. Chromatogr., 196 (1980) 21. L.R. Snyder and J.J. Kirkland: Introduction to Modem Liquid Chromatography, 2nd. edition, Wiley Interscience, New York 1979, p. 260. P. Jandera, Chromatographia, 19 (1984) 101. L.R. Snyder, J.W. Dolan and J.R. Gant, J. Chromatogr., 165 (1979) 3. H. Colin, G. Guiochon and P. Jandera, Anal. Chem., 55 (1983) 442. P. Jandera, J. Churiiikk and H. Colin, J. Chromatogr., 214 (1981) 35. P. Jandera, H. Colin and G. Guiochon, Chromatographia, 16 (1982) 132. H. Hemetsberger, H. Klar and E. Ricken, Chromatographia, 13 (1980) 277. V.Ya Davidov, M.E. Gonzalez, A.V. Kiselev and K. Lenda, Chromatographia, 14 (1981) 13. J.H. Knox and G. Vasvari, J. Chromatogr., 83 (1973) 181, G.E. Berendsen, P.J. Schoenmakers, L. De Galan, G. Vigh, 2. Varga-Puchony and J. Inczedy, J. Liquid Chromatogr., 3 (1980) 1669. M. Montes, J.L. Usero, A. Del Arco, C. Izquierdo and J. Casado, J. Chromatogr., 481 (1989) 97. H. Colin, G. Guiochon and P. Jandera, Chromatographia, 17 (1983) 83. P. Jandera, J. Chromatogr., A, 656 (1993) 437. P. Jandera, Chromatographia, 19 (1984) 101. T. Tsuneyoshi, J. Chromatogr., 354 (1985) 37. P. Jandera, J. Chromatogr., 314 (1984) 13. H. Colin, A.M. Krstulovic, M.-F. Gonnord, G. Guiochon, Z. Yun and P. Jandera, Chromatographia, 17 (1983) 9. P. Jandera, J. Chromatogr., 449 (1988) 361. P.J. Schoenrnakers, H.A.H. Billiet and L. De Galan, J. Chromatogr., 185 (1979) 179. T.L. Hafkenschied and M. Tomlinson, J. Chromatogr., 264 (1983) 47. M. Lafosse, P. Marinier, B. Joseph and M. Dreux, J. Chromatogr.. 623 (1992) 277. K. Belsner, M. Pfeifer and B. Wilffert, J. Chromatogr., 629 (1993) 123. M. Czok and H. Engelhardt, Chromatographia, 27 (1989) 5. P. Jandera, H. Pechovii, J. Kriilovskjr, D. Tocksteinovh and J. Churhkk, Chromatographia, 16 (1982) 275. P. Jandera, Chromatographia, 26 (1988) 417.

This Page Intentionally Left Blank

R.M. Smith @ Retention .I.), and Selectivity in Liquid Chromatography Journal of Chromatography Library, Vol. 57 0 1995 Elsevier Science B.V. All rights reserved

269

CHAPTER 8

Lipophilic and polar indices P. Jandera Department of Analytical Chemistry, University of Pardubice, Faculty of Chemical Technology, Nhm. Legii 565,532 10 Pardubice, Czech Republic

8.1 INTRODUCTION

In Chapter 7, the possibilities of using interaction indices for characterization and prediction of retention and selectivity of separation of various solutes in reversed-phase systems are discussed. The model of interaction indices considers the interactions in the mobile phase as the driving force of retention and understands the retention and separation as the result of the differences in the polarities of the molecules characterized by the interaction indices, 1, and in the size of the molecules, i.e. in their molar volumes, V,. The limitations of this approach consist of neglecting the role of the stationary phase in retention adopted in the derivation of the model and in considering only the overall polarity of the interacting molecules. Impossibility of discrimination between the selective polarity effects, i.e. the dipole-dipole, proton-acceptor or proton-donor interactions between the solutes and the components of the mobile phase may result in deviations between the calculated and experimental retention characteristics (capacity factors) for substances with strongly polar groups such as hydroxy-, amino-, etc. To avoid this source of errors in prediction of retention, the model of interaction indices should be modified. A single interaction index is obviously insufficient to take into account different selectivity effects. Different polarity indices should be used for mobile phases containing solvents belonging to different classes of selectivity according to the Snyder’s classification [I]. These considerations were used for the derivation of the approach for characterization and prediction of retention based on so-called lipophilic and polar indices described in this chapter. 8.2 RETENTION IN HOMOLOGOUS SERIES AS THE BASIS OF LIPOPHILIC AND POLAR INDICES To discriminate between various selective polarity effects, the reference standards, selected in order to include solutes from different selectivity classes, are less suitable than a References pp. 294-295

Chapter 8

270

calibration set of the reference compounds that belong to a single selectivity class with regular differences between the overall polarities of the individual reference standards. In this case, the selective polarity effects are accounted for by different polarity indices applying with mobile phases containing various organic solvents. The members of a single homologous series seem best suited for this purpose, especially if the end groups in the homologous series are non-polar. From this point of view, n-alkanes seem to be the standards of choice, but because of lower volatility and better detectability, n-alkylbenzenes are preferred for this purpose. The retention behaviour of homologous series in reversed-phase systems is discussed in detail in Section 7.2.I, where the retention equations were derived, making it possible to predict the capacity factors of sample compounds as a h c t i o n of both the number of methylene groups in the molecule and of the concentration of one or more organic solvent(s) in the aqueous-organic mobile phase. It was shown that the parameters m and a of the linear dependence of log k on the concentration of the organic solvent in the mobile phase, r$ (retention equation (7.8) in Chapter 7) are strongly correlated for the members of a homologous series in various mobile phases (Fig. 8.1):

m=q+pa

(8.1)

If log k of sample soIutes change in a linear manner with $, the combination of Eqs. (7.27) and (7.34) results in the following equation, taking into account the linear correlation of the parameters a and m (Eq. 8.1) [2]:

1

2

3

4

5

6

7

a Fig. 8.1. Correlations between the slopes m and the intercepts a of the linear log k versus q5 plots (Eq. (7.8) in Chapter 7) for homologous 1,2-naphthoylenebenzimidazole-6-sulphonamidesof n-aminoalkanes (1-3) and 3,5dinitrobenmates of n-alkanols (4-7) on a Silasorb C18 column in methanol-water (1,4), dioxane-water (2,5), acetonitrile-water (3,6)and tetrahydrofuran-water (7) mobile phases.

27 1

Lipophilic andpolar indices

logk =(a, +a,n)(l-&)-q@ The introduction of lipophilic and polar indices is based on the validity of Eq. (8.2) in a variety of reversed-phase systems. The parameters al and p of Eq. (8.2) are approximately constant for various homologous series and columns and characterize the nonspecific contribution to retention, whereas the constants q and a, are far more influenced by the type of the homologous series and column (Table 8.1) [2]. This suggests that Eq. (8.2) with the parameters al and p determined for one (calibration) homologous series can be used to describe the retention in another homologous series, if the parameters a, and q are determined for each individual series. In this way, the application of Eq. (8.2) can be extended to an arbitrary sample solute, because any compound can be considered as a member of a homologous series with either n = 0 or n > 0. It would be difficult to determine the values of the constants a, al and p of a single sample solute , but these constants can be substituted by the values of the parameters a, TABLE 8.1 VALUES OF THE PARAMETERS UO, ~ 1p,. 4 OF EQ. (8.2) FOR VARIOUS HOMOLOGOUS SERIES On a Silasorb C18 column in mobile phases containing methanol (a), acetonitrile (b), 1,4-dioxane (c) and tetrahydrofuran (d) in water ~~

Mobile phase

Series

00

a

1 2 3 4 5 6 7 8 9 10 1 3 4 5 6 7 10 1 3 4 5 1

1.92 -0.95 1.71 1.74 2.17 2.29 -3.00 -0.56 -0.17 0.43 1.39 1.54 1.42 0.97 1.64 -0.45 0.29 1.82 1.70 1.40 1.42 1.42

b

C

d

01

0.54 0.48 0.52 0.48 0.47 0.46 0.56 0.60 0.62 0.59 0.28 0.25 0.26 0.25 0.24 0.34 0.42 0.33 0.55 0.54 0.46 0.33

P

4

0.87 0.84 0.86 0.86 0.90 0.88 0.77 0.83 0.84 0.88 0.60 0.56 0.64 0.65 0.43 0.48 0.66 0.92 1.05 1.05 1.02 1.18

0.96 1.08 1.30 1.32 0.96 1.09 2.12 1.37 1.85 1.07 1.36 2.03 1.65 1.14 0.87 2.36 1.70 1.04 1.19 1.16 1.76 0.90

1, n-alkylbenzenes; 2, n-alkanols; 3, 3,5-dinitrobenzoates of n-alkanols; 4, 4-bromophenacyl esters of nalkanoic acids; 5, 1,2-naphthoylenebenzimidazole-6-sulphonamides of n-alkylamines; 6, 4-(N,N-dimethylamino)benzene-4’-azobenzoylamides of n-alkylamines; 7; 3-n-alkyl-6-methyluracils; 8; alkan-2-ones; 9; 3-nalkoxycarbonyl pyrrazolines; 10; methyl esters of n-alkanoic acids. Data from ref. [2].

Referencespp. 294-295

272

Chapter 8

al and p of a suitable calibration homologous series such as n-alkylbenzenes. The individual character of each compound is then concentrated only in the parameters n and q of Eq. (8.2). Of course, the real indices n and q following from the attribution of the compound to its own homologous series cannot be applied in this approach, otherwise the right-hand side of this equation would not equal the left-hand side. Instead, hypothetical equivalents n, and qi to the number of methylene units n and to the parameter q of the calibration homologous series should be introduced. These indices can be calculated from the parameters a, m of Eq. (7.8) in Chapter 7, determined by linear regression of the log k versus 9 plots of the solute measured under the same conditions (same column and components of the mobile phase) as the data for the calibration homologous series [3]: a-a0

nce =-

a1

q i = m - p ( a , +aln,,) = m-pa

(8.4)

The indices nm qi, can be introduced into Eq. (8.2) instead of n, and q to make possible predictive calculations of the capacity factors of the solute in other systems (columns and binary mobile phases), for which the parameters a@al and p of the calibration homologous series were determined experimentally:

The main difference from the interaction indices approach described in Chapter 7 is the use of two indices related to the calibration homologous series instead of Zx and Vx to characterize the retention of the sample in different mobile phases. The lipophilic index n,, represents the number of methylene groups in a hypothetical member of the calibration homologous series with the same lipophilic character as the sample solute, characterizes the lipophilicity of the solute and is a measure of the nonpolar contribution to the retention. It is independent of the composition of the mobile phase. The contribution of the size of the molecule, V,, is combined with the non-specific contribution of the non-polar (hydrocarbonous) part of the molecule to overall polarity in the index rice. Selective polar contributions to the retention by the dipole-dipole, proton-donor and proton-acceptor interactions of the polar functional groups in the molecule of the solute are characterized by the index qi,the polar index that depends on the type of the organic solvent in the mobile phase [3]. Equation (8.5) can be adapted to describe the relative retention (the selectivity of separation), ri,j= kj/ki,of two solutes i andj [4]: logri,j =al(I-&)Anc where

-@Aq

(8.6)

Lipophilic and polar indices

273

and

are relative lipophilic and polar indices, respectively, that characterize the differences in lipophilic and polar contributions to the retention of compounds i andj, i.e. the lipophilic and the polar contributions to the selectivity of separation of these two compounds. (The subscripts i and j of the constants a, m and of the indices nco qi attribute these quantities to the compounds i orj.) The relative indices Ance, Aq can be calculated either fkom the lipophilic and polar indices of compounds i andj or directly from the intercept and slope of the linear log k versus q5 plots (Eq. (7.8) in Chapter 7 ) , a and m, applying for the individual solutes. Unlike in Eq. (8.5), the term aopq5, which is the same for the compounds i andj, does not appear in Eq. (8.6). In the absence of the parameter a. depending on the type of the calibration series, only the constants al and p characterize the calibration homologous series in Eq. (8.6). These constants are fairly independent of the type of the homologous series used for the calibration of the retention scale (Table 8.1). This suggests the possibility of using Eq. (8.6) instead of Eq. (8.5) for the calibration of retention on the basis of relative lipophilic and polar indices. For this purpose, the retention of a sample solutej should be related to that of a standard reference compound, i = st, so that nce,i= rice,,,, qi,i = qi,stand the constants ai = a,,, mi = m,, in Eqs. (8.6)(8.8) belong to the reference standard. With known retention (capacity factor) of the reference standard, k,,, Eq. (8.9) can be used to predict the retention of sample solutes from their relative lipophilic and polar indices Ano Aq at various compositions of the mobile phase: log k = log k,, + a , (1 - p$)Anc - q5Aq

(8.9)

For this purpose, it is necessary to know only the values of the parameters al and p of the calibration homologous series. For the sake of convenience, a member of the calibration homologous series can be selected as the reference standard solute, for example toluene or ethylbenzene of the n-alkylbenzene series. The advantage of using relative indices Anc and Aq consists of improving the precision of prediction of k, as the errors originating from the determination of the constant a. are eliminated and the constants at and p are less prone to be affected by the experimental errors, because their values are almost independent of the type of the homologous series and column. Hence, the relative retention predicted in this way is less affected by the variation in the properties of different batches of the packing material and by the column aging, which is at least partially compensated for by the experimental determination of the capacity factor of the standard, k,,, than the capacity factors predicted using the indices nce and qi* References pp. 294-295

274

Chapter 8

8.3 MOLECULAR STRUCTURE AND LIPOPHILIC AND POLAR INDICES 8.3.1 Anc and Aq indices as the descriptors of the lipophilicity and polarity of solutes To verify the assumption that the indices rice, Anc are suited to characterize the lipophilicity of solutes, the indices of 16 simple aromatic compounds were plotted versus Hansch and Leo hydrophobic substituent constants n [5] and good correlation was found with only one outlying point (Fig. 8.2) [4].On the other hand, the lipophilic parameters Anc are not correlated at all with the polarity parameters P' [ 6 ] . It was shown that the relative polar index Aq is directly proportional to the difference between the interaction indices of a solute with a polar hctional group and of its unsubstituted parent compound and therefore it can be expected to increase with the polarity of the hctional group(s) in the molecule of the solute [7]. In agreement with these considerations, good correlation was found between the values of Aq and P' polarity indices, with two outlying higher values for benzyl alcohol and rn-cresol (Fig. 8.3) probably caused by the presence of hydroxyl groups, which are able to form hydrogen bonds with methanol and water. Selective interactions with the components of the mobile phase were not taken into account in the derivation of the P' indices and lead to increased values of the polar indices qiand Aq. The relative polar indices Aq were found to increase with the polarities of the h c tional groups approximately in the following order: n-alkanes < polycyclic aromatic hydrocarbons < n-alkylbenzenes < benzene, styrene, biphenyl < halogenated benzenes < dialkyl ethers < alkyl aryl ethers, diary1 ethers < aromatic nitriles < aromatic ketones and aldehydes < aromatic amines < aromatic alcohols < phenol, alkylphenols < chloro

I

lm20

0.00

-

-1.20 -

-2.40 -1.50

I

-1.00

I

-0.50

I

I

I

0.00

0.50

1.00

1.50

IT

Fig. 8.2. Correlation between the relative lipophilic indices Anc and Hansch and Leo hydrophobic substituent constants x [ 5 ] for various substituted benzenes. Column, Silasorb C8; mobile phase, methanol-water.

Lipophilic and polar indices

-2

0

275

2

4

6

8

P’ Fig. 8.3. Correlation between the relative polar indices Aq and Snyder’s polarity indices P‘ [6] for various substituted benzenes. Column, Silasorb C8; mobile phase, methanol-water.

0.80

0.60

0.40

0.20

0.00

’

-0.20 -0.20

I

I

I

I

I

0.00

0.20

0.40

0.60

0.80

Fig. 8.4. Correlation between the relative polar indices Aq, in acetonitrile-water, AqACN, and in methanolwater, AqMeOH, mobile phases on a Silasorb C8 column. Classes of compounds and individual solutes: 1, nalkylbenzenes, styrene; 2, halogenobenzenes; 3, alkyl aryl ethers; 4, esters of aromatic carboxylic acids; 5, alkyl aryl ketones, diarylketones; 6, benzaldehyde; 7, benzonitrile; 8, aniline; 9, benzylalcohol; 10, nitrobenzene; 11, phenols; 12, phenylureas (data from ref [4], the straight line with the intercept = 0 and the slope = 1 corresponds to the theoretical correlation in the absence of selective polar interactions).

References pp. 294-295

Chapter 8

276

phenols. For a given class of compounds, the Aq values were found within a relatively narrow range [4]. As the polar indices Aq reflect selective polar interactions of solutes in a specific mobile phase, different values of the polar indices can be expected in mobile phases containing various organic solvents. This is illustrated by Fig. 8.4, where the relative polar indices Aq in acetonitrile-water mobile phases are plotted against the values of Aq in aqueous methanol. The most apparent feature of this plot is a regular increase of Aq in the two sets of indices with increasing polarities of the sample solutes. Compounds with a common functional group are grouped into limited regions of this graph. Some of these regions overlap and the areas in which the data for one class of compounds are spread increase with increasing polarity of the compounds [4]. In the absence of specific polar interactions, Aq in mobile phases containing various organic solvents should be the same and should appear on the straight line with the slope = 1. The experimental data for most compounds in Fig. 8.4 are spaced within the interval *O.O5Aq units on both sides from this line. Because of selective polar interactions exhibited by the compounds that can form hydrogen bonds with -OH groups in water and methanol, such as phenols, ketones, aldehydes, etc., the data points for these compounds are shifted downwards into the methanol region. The point for aniline is shifted in the opposite direction, into the acetonitrile area. The position of the data in the Aq-Aq diagrams can be used to roughly estimate the polarities and the presence of specific fhnctional groups in simple organic compounds from their reversed-phase retention data [4]. 8.3.2 Structural contributions to lipophilic and polar indices

Experimental measurements suggested that structural elements in various classes of sample solutes contribute additively to the nce and qi (and to Anc and Aq) indices [4,8], For these compounds, the indices can be calculated as the sum of the individual contributions of the substituents Anc,i, Aqi :

An, =

Anc,i

(8.10) (8.1 1)

In homologous series, the experimental contributions of a methylene group to the lipophilic indices Anc,i varied from 0.89 to 1.09 and to the polar indices Aqi from -0.08 to +O.O 1 for six different homologous series in methanol-water and in acetonitrile-water mobile phases (Table 8.2), which is in agreement with the contributions 1 and Aqi = 0 implied by using a homologous n-alkylbenzene series for calibration of the lipophilic and polar indices scale [4,7]. The contributions Anc,i and Aqi for oligomeric series can differ significantly from those in homologous series, according to the character of the repeat unit in the series and for example Aqi are positive for oligostyrenes and oligoethylene glycols. The contribution of the repeat unit in the oligostyrene series to the lipophilic index Anc,i is greater

Lipophilic and polar indices

277

TABLE 8.2 STRUCTURAL CONTRIBUTIONS TO THE INDICES An" Aq FOR HOMOLOGOUS AND OLIGOMERIC SERIES AND FOR POLYSUBSTITUTED BENZENES ON C18 AND C8 COLUMNS Anc

= Anc, 0

+ nAn,, ,+ norAnc,or

Aq = Aqo + nAqi + norAqor

Series

Repeat group

Anc,o

An,,

1 2 3 4 5 6 7* 8 9** 10 11 12 13

A A A A A A B C C D E E E

-1.85 -0.39 -0.33 -5.35 -4.32 -2.78 2.00 -5.3 1 3.43 -0.94 -0.97 -2.61 -2.33

0.99 0.96 0.89 0.89 1.02 1.09 2.30 0.66 -0.10 0.94 1.11 1.40 1.60

I

4,or

Aqo

Aqi

0.11 -0.11 0.17 -0.40

1.18 1.28 1.31 1.11 1.58 0.33 0.82 1.56 1.oo 0.03 0.11 0.37 0.56

-0.01 0.00 -0.01 -0.01 -0.08 0.00 0.12 0.28 0.05 -0.05 -0.08 0.00 -0.12

4 0 ,

0.005 4.01

-0.074 0.06

Anc,(]. Aqo, contributions of the parent molecule or of the end groups; Anc,l, Aql ,contributions of the structural repeat unit or of the substituent; Anc,or, Aqor, contributions of the ortho-position of substituents on the benzene ring; n, number of repeat groups or of substituents; nor, number of substituted ortho-positions on the benzene ring. Mobile phases: methanol-water, except for: *dioxane-water and ** 1-propanol-water. Series: 1, n-alkanes; 2, 3,5-dinitrobenzoates of n-alkanols; 3, 4-bromophenacylesters of n-alkanoic acids; 4, n-alkanols; 5, alkan-2-ones; 6, methylesters of n-alkanoic acids; 7, oligostyrenes; 8, oligoethylene glycols; 9, ethoxylated nonylphenols; 10, polymethylbenzenes; 11, polychlorobenzenes; 12, polychloroanilines; 13, polychlorophenols. D, 4 H 3 ; E, -CI. CalcuRepeat groups or substituents: A, -CH2-; B, -C(C6H5)H
than the contribution of the repeat -CH2-CH2-O- unit in oligoethylene glycols, while the opposite applies for the contributions of the repeat units to the polar index Aqi and the contribution of the repeat -CH2-CH2-O- unit in ethoxylated nonylphenols is close to zero. In contrast to the oligoethylene glycol series, the negative contribution An,, I of the oxyethylene unit was found with ethoxylated nonylphenols. This means that the lipophilic character of ethoxylated nonylphenols diminishes with increasing number of oxyethylene groups, which is the reason for their elution in the order of decreasing molecular masses ~91. The contribution of the end groups to Anc in homologous and oligomeric series in Table 8.2 increases with increasing size of the non-polar part of the molecule of a solute, while increasing polarities of the end groups increase their contribution to Aq. It should be noted that because of the selection of n-alkylbenzenes as the calibration homologous series and toluene as the reference standard, both Anc = 0 and Aq = 0 for toluene and theoretically expected values for the contribution of the parent molecule (benzene) in the groups of polymethylbenzenes and polychlorobenzenes are = -1 .O and Aqo = 0, which was confirmed by the experimental data in Table 8.2. References pp. 294-295

278

Chapter 8

The substituents in polysubstituted benzenes may be subject to mutual interactions that may complicate direct application of the additivity rule for the contributions of the individual groups to the lipophilic and polar indices. These interactions become more important with increasing polarities of the substituents and are most significant with adjacent substituents on the benzene ring. This “ortho-effect” can be taken into account by introducing contributions to the An, and Aq indices for each two substituents in the o-position. The corresponding increments depend on the type of the substituents. For example, from the data published by Hammers et al. [lo] for methylbenzenes, chlorobenzenes, chloroanilines and chlorophenols on a Lichrosorb C 18 column in aqueous methanol, the methyl- and chloro-substituents and the number of substituents in adjacent positions were found to contribute additively to the Anc indices (correlation coefficients from 0.994 to 0.999; Table 8.2). Increased contributions of a chloro-substituent to Anc in chloroanilines and in chlorophenols in comparison to chlorobenzenes can be attributed to the effects of the interactions between these substituents and the amino or phenolic groups. Good correlations were found also between the polar indices Aq and the number of substituents in methylbenzenes and in chlorobenzenes, where the contribution of the ortho-effect is practically negligible. For chloroanilines, the ortho-effect is more significant. The correlation was very poor for Aq values of chlorophenols, probably because of the interactions between the chloro-substituents and the phenolic group [4]. Two other examples of the additivity of structural contributions to the Anc and Aq inTABLE 8.3 STRUCTURAL CONTRIBUTIONS OF SUBSTITUENTS An,, ,AND Aq, TO THE INDICES Anc, Aq ON A SILASORB C18 COLUMN IN METHANOL-WATER MOBILE PHASES OF SOME HERBICIDES Series

(A)

Substituent

CH3 iso-C3H7 C4H9 OCH3 OH

Position R1, R2

Position XI, X2, X

A%, I

&I

0.25

0.04

2.38 0.96

0.06 0.04

c1 B

Br CF3 CH3 C2H5

iso-C3H7 ter-CqHg CH30(CH2)3 CI SCH3

0.25 1.25 2.13 3.22 1.69

4

1

A41

1.05 2.68

0 0

-0.05 -1.10 1.33 1.so 2.39

0.24 0.97 -0.08

1.33 2.15

-0.08 -0.14

-0.10 0.06

0.04 0.06 0.06 0.06 0.13

(A) Substituted phenylureas, AncO =-1.4; Aqo = 1.60; R1, R2, substituents on nitrogen; XI, X2, substituents on phenyl in positions 3, 4. (B) Substituted 4,6-diamino-l,3,5-triazines,AncO =-3.53; Aqo = 1.45. R1, R2, substituents of the amino nitrogen; X, substituent in position 2 on the triazine ring. Data from ref. [8].

Lipophilic andpolar indices

279

dices of phenylurea and triazine herbicides are illustrated in Table 8.3 [8]. The contributions Anc,i for the substituents on the benzene ring of substituted phenylureas agree with the contributions in methylbenzenes, chlorobenzenes and chlorophenols, e.g. Anc,,= 1.05 (0.94) for a methyl group, 1.33 (1.1 1) for a chloro- and -1.29 (-1.39) for a phenolic hydroxy substituent. The polar Aq of the phenylurea herbicides are, only slightly or not at all, contributed to by the alkyl-, halogeno- or methoxy substituents and the q, indices of all the phenylurea herbicides studied are close to one another, with the exception of methoxuron with a phenolic hydroxy group. The contributions of the alkyl substituents to the An, indices of substituted triazines are somewhat lower than the values for the substituents on the benzene ring and the Aq indices of the triazine herbicides are close to one another. The differences between the experimental Anc and the values calculated from the additive contributions are 0.13 or less and the differences between the experimental and calculated Aq indices are less than 0.06. However, only a limited number of compounds have been tested so far and it is not possible to generalize the rule of the additivities of the structural contributions to the Anc and Aq indices without experimental verification for additional sets of compounds.

8.4 PREDICTION OF RETENTION USING LIPOPHILIC AND POLAR INDICES 8.4.1 Selection of the reference calibration homologous series The calibration homologous series should be non-polar in order to better distinguish the specific and non-specific contributions to the retention. As has been already mentioned, n-alkanes are theoretically best suited to this requirement, but they cannot be detected using UV detection and lower n-alkanes are gaseous or volatile, which precludes their use for the present purpose. Instead, the n-alkylbenzene calibration series was employed, which is easily available and convenient to use in connection with UV detectors [4]. Because of the strong retention of higher alkylbenzenes in reversed-phase systems, the use of n-alkylbenzene calibration homologous series is limited to mobile phases containing 50% or more of one or two organic solvent@). Homologous alkylmethyluracils, alkoxycarbonylpyrrazolines and alkan-%-oneswere tested as potential calibration series in mobile phases with higher concentrations of water and were found suitable for this purpose in mobile phases containing 25-50% methanol in water or 30-50% acetonitrile in water [ 1I]. Homologous alkan-2-ones are easily available and seem especially suitable to complement the n-alkylbenzene calibration series in mobile phases with lower contents of the organic solvent(s). 8.4.2 Precision of the predicted retention data As the parameter q depends on the column used [2], the parameters qi determined on different columns can be expected to slightly differ each from the other. Moreover, the valReferences pp. 294-295

280

Chapter 8

TABLE 8.4 CAPACITY FACTORS, k, ON DIFFERENT C18 COLUMNS IN 70% MeOH CALCULATED FROM THE Anc AND Aq INDICES DETERMINED ON A SILASORB C18 COLUMN USING EQ. (8.9) ~~

Column

1 2 3 4 5

6 7

A

B

C

kXP

kcal

kXP

kcal

kXP

kcal

1.34 1.38 1.36 1.09 0.75 1.47 1.16

1.46 1.43 1.43 1.16 0.75 1.51 1.18

6.84 6.59 6.59 6.00 2.95 6.60 5.18

6.83 6.69 6.70

14.05 13.42 13.13 10.90 5.80 12.46 10.63

13.03 12.75 12.77 10.37 6.69 13.54 10.52

5.44

3.51 7.10 5.52

Solutes: A, anisole; B, p-bromophenacyl capronate; C, n-hexyl3,5-dinitrobenzoate. Columns: 1, Silasorb C18 (5,um); 2, LiChrosorb SI 100 C18 (10,um); 3, LiChrosorb SI 60 C18 (10pm); 4, Hypersil ODS ( l o p ) ; 5, pBondapak C18 (10pm); 6, Separon SIX C18 (10,um); 7, Silasorb C18 (10pm). Dataffom ref. [12].

ues of the indices ncetq,, can be subject to systematic errors originating from fitting linear log k+ plots to the experimental data that are in fact linear only in a limited range of concentrations of the organic solvent in the mobile phase and to the errors in determination of the constants uo, al and p of the calibration homologous series. These errors may affect the precision of prediction of the retention data. In Table 8.4, the experimental capacity factors of several compounds on various octadecyl silica columns are compared with the values calculated from the Anc and Aq indices determined on a Silasorb C18 column. With two exceptions for the microBondapak column, the agreement between the experimental and the calculated values is better than 510% relative [12]. Similar average error was found for the prediction of the capacity factors in a mobile phase of composition different from that in which the experimental retention data was measured [12]. The two-indices approach can be also applied to gradient-elution chromatography. For reversed-phase systems with linear gradients, the retention volumes can be calculated explicitly. If the log k of a solute under isocratic conditions decreases in a linear manner with increasing concentration of the organic solvent in the mobile phase, $ (the retention can be described by Eq. (7.8) in Chapter 7), the net retention volume VR in gradientelution chromatography can be calculated from the equation [13,14]:

+

V; = L o g [ 2 . 3 1 m B G V M10(a-m&)+ 11 mBG

(8.12)

where a and m are the parameters of Eq. (7.8), VM is the dead volume of the column, AG is the initial concentration $J of the organic solvent at the start of the gradient and BG is the slope of a linear increase in with the volume of the eluate, V, during the gradient elution:

+

Lipophilic and polar indices

28 1

(8.13) Introducing the relationships between the parameters a, m and the lipophilic and polar indices nce qi (8.14)

into Eq. (8.12), we obtain the expression describing the retention volume in gradientelution chromatography as a fbnction of the indices nce and qi and of the parameters A , BG characterizing the profile of the gradient. In Table 8.5, experimental retention volumes of eight phenylurea herbicides are compared with the values predicted on the basis of the nce and q, indices for different gradient profiles and agreement better than 5-7% relative was found for most values, even though the indices determined from the data measured on one column were used to predict the retention volumes on another column [8]. In all the examples discussed so far, the prediction of retention was based on the nalkylbenzene calibration series, however, its use is limited to mobile phases containing more than 50% of the organic solvent(s), because of too strong a retention in other mobile phases. Table 8.6 shows experimental capacity factors for five test solutes in mobile phases containing 25-50% methanol or acetonitrile in water together with the values calculated from the Anc and Aq indices using alkylmethyluracils, alkoxycarbonylpyrrazolines and alkan-2-ones instead of n-alkylbenzenes as calibration homologous series. The results suggest that the effect of the calibration series on predicted k is insignificant and the error in the predicted retention data is comparable to the error of prediction in TABLE 8.5 EXPERIMENTAL (Exp) AND CALCULATED (Cal) ELUTION VOLUMES, V'R, IN cm3, ON A C18 COLUMN IN CHROMATOGRAPHY WITH LINEAR GRADIENTS OF METHANOL IN WATER Solute

v', I1

I

I11

EXP

Cal

EXP

Cal

EXP

Cal

14.5 18.2 22.7 23.3 23.7 25.0 25.5 27.4

13.4 17.0 21.9 22.7 23.1 24.8 25.3 27.2

17.4 22.1 28.1 29.1 29.6 31.4 32.1 35.0

16.3 21.7 28.7 29.7 30.3 32.7 33.4 36.3

13.6 18.9 25.6 26.6 27.3 29.4 30.1 33.4

13.2 18.1 25.6 26.8 27.4 30.1 31.0 34.3

I, AG = 0.2, BG = 0.027; I11 AG = 0.15, BG = 0.020; 111, AG = 0.25, BG = 0.0175. 1, phenuron; 2, methoxuron; 3, chlorotoluron; 4, isoproturon; 5, diuron; 6, linuron; 7, chlorobromuron;8, neburon. Data from ref. [8].

References pp. 294-29s

282

Chapter 8

TABLE 8.6 CAPACITY FACTORS k PREDICTED USING CALIBRATION SERIES OF (A) ALKYLBENZENES, (B) ALKYLMETHYLURACILS, (C) ALKOXYCARBONYL PYRRAZOLINES AND (D) ALKAN-ZONES

1 1 2 2 3 3 4 4 5 5

40%MeOH 40% ACN 40%MeOH 40%ACN 40%MeOH 40%ACN 40%MeOH 40%ACN 40% MeOH 40%ACN

1.88 0.92 3.83 2.81 7.34 3.79 15.08 3.76 21.12 9.12

1.83 0.87 3.62 2.79 7.39 3.80 14.71 3.78 20.12 9.49

1.84 0.87 3.64 2.76 7.36 3.81 14.61 3.76 20.23 9.55

1.84 0.87 3.64 2.76 7.41 3.82 14.48 3.77 20.12 9.58

1.85 0.87 3.65 2.80 7.36 3.83 14.54 3.79 20.23 9.59

Column: Silasorb C18. Solutes: 1, benzyl alcohol; 2, benzonitrile; 3, ethyl phenyl carbaminate; 4, atrazine; 5, phenetole. Data from ref. [l 11.

mobile phases containing more than 50% of the organic solvent(s) with n-alkylbenzenes used as the calibration series [ 1 I]. Rather than for HPLC separations, prediction of the capacity factors of sample solutes in pure water, k,,,, is of interest both for correlations with the parameters characterizing solute lipophilicities [ 15,161 and for the evaluation of various materials as potential sorbents for solid-phase extraction techniques treating dilute aqueous samples [ 171. Direct experimental determination of k, is unfortunately fraught with serious difficulties, as these values are of the order of magnitude of 102-105 or even greater, depending on the size and polarity of the solute. It would seem possible to determine k,,, by extrapolation of the log k versus qj plots to zero concentration of the organic solvent in the mobile phase, but neither linear nor quadratic dependence gives an accurate description of the solute retention in mobile phases with concentrations of the organic solvent close to zero, probably because of preferential sorption of the organic solvent on the surface of the packing material [ 181 and consequently, the values of k,,, determined by extrapolation may be subject to gross errors. To overcome this difficulty, various authors suggested calculating k, from correlations of their logarithms with the logarithms of the solubilities in water [ 191, with the surface tension, dielectric constant and other physico-chemical quantities [20,21]. Correlation of log k,,, with the relative lipophilic indices, ncecan also be used for this purpose [22]. Figure 8.5 shows correlation between the experimental capacity factors in pure water and k, predicted from the relationship nce = -5.3 1 + 1.72 log k,,, found experimentally for nine pesticides on a Silasorb C18 column in methanol-water mobile phases. Slight improvement in the precision of the predicted values of k, was found in comparison with the values calculated from the correlation between the logarithms of k,,, and logarithms of solubilities in water [21]. The average error of the prediction is comparable to the error of the “direct” determination by extrapolation from the data measured in mobile phases containing 3040% methanol (up to 60% relative). Hence, using lipophilic indices for prediction

Lipophilic and polar indices

283

4.00

3.50

?

-

3.00 -

Y M

2

2.50

-

2.00 -

'

1.50 1.50

0 I

I

I

I

2.00

2.50

3.00

3.50

log

4.00

k.".B

Fig. 8.5. Correlation between the experimental, kw,E, and the predicted, k w , p capacity factors in pure water. The predicted values are calculated from the correlation equation between the kw of phenylurea and triazine herbicides with their lipophilic indices nCeon a Silasorb C18 column: n,, =-5.31 + 1.72 log k,, r = 0.996.

of k, offers an alternative to methods of determination of k,, introduced earlier, even though it does not significantly improve the accuracy of the prediction.

8.5 CHARACTERIZATION OF SELECTIVITY USING LIPOPHILIC AND POLAR INDICES 8.5.1 Binary mobile phases

The selectivity of separation of two compounds i a n d j under isocratic conditions can be conveniently characterized by the relative retention, ri,j,The dependence of ri,j on the lipophilic and polar indices is described by Eq. (8.6), where An, and Aq are differences in the iipophilic and polar indices of the two compounds rather than the relative indices of the individual solutes related to a standard reference compound such as toluene. Equation (8.6) can be rewritten to obtain the expression for the relative retention as the product of two contributions: ri,j = a L a p

(8.16)

where a L = a, (1 -&)An,

represents the lipophilic contribution to the selectivity and References pp. 294-295

(8.17)

284

Chapter 8

a p =-$Aq

(8.18)

characterizes the polar contribution to rj,j[4]. The constants al and p relate to the calibration homologous series and depend on the column and on the type of the organic solvent used. For n-alkylbenzenes, al = 0.28-0.54, p = 0.6-0.92 (Table 8.1). Because $20.8 in most practical separations, al(l -&)> 0. The lipophilic contributionaL> 1 if An, > 0 and is controlled by the size of the non-polar (hydrocarbon) part of the molecule; it is directly proportional to the difference between the lipophilic indices An, of the sample solutes. According to Eq. (8.17), aL is expected to decrease with increasing concentration of the organic solvent in the mobile phase, $. Both al and p usually decrease with increasing polarity of the organic solvent in the mobile phase, which means that the effect of the character of the organic solvent on aL cannot be predicted apriori. The polar contribution to the separation selectivity log ap is directly proportional to the difference between the polar indices of the compounds i and j , Aq, and to the concentration of the organic solvent in the mobile phase, $. To illustrate the effects of the lipophilic and polar indices on the selectivity, Fig. 8.6 shows plots of ri,jand of the lipophilic and polar contributions aL,apto the relative retention in dependence on the concentration of the organic solvent in the mobile phase, $, for three different pairs of phenylurea herbicides on a C18 column in methanol-water mobile phases [8]. For most pairs of compounds, the difference between the lipophilic indices is positive

,/'

lc

0 '

0.00

I

I

I

I

0.20

0.40

0.60

0.80

co Fig. 8.6. Changes in lipophilic ( a b curves a) and polar (ap,curves b) contributions to the separation selectivity (relative retention r;,) curves c) of phenylurea herbicides with the concentration of methanol, @, in aqueousmethanolic mobile phases on a Silasorb SPH C18 column. Differences in the lipophilic and in the polar indices between the individual compounds: An, = 1.24, Aq = -0.14 for metoxuron and fluometuron (l), An, = -0.48, Aq = -0.94 for hydroxymetoxuron and desphenuron (2), An, = 0.71, Aq = 0.01 for linuron and chlorobromuron (3).

Lipophilic and polar indices

285

and the difference between the polar indices is negative such as with fluometuron and metoxuron, where Anc = 1.24 and Aq = -0.14, which means that the contributions to the selectivity of separation aL> 1 and ap> 1. Because both differences in the lipophilic and the polar indices increase the selectivity, r,,Jis rather high. The increase in the polar contribution to selectivity with increasing @ (curve lb) is less significant than the decrease in the lipophilic contribution (curve la), which results in a rather steep decrease in rr,Jwith increasing concentration of the organic solvent in the mobile phase (curve lc). On the other hand, if molecules of one solute contain a bulkier non-polar part and a more polar group than the molecules of the other solute, the differences between both the lipophilic and polar indices of the two compounds, Anc and Aq are negative. This occurs with desphenuron and hydroxymetoxuron, where Anc = -0.48 and Aq = -0.94 and consequently the lipophilic contribution to selectivity aLis <1 (curve 2a) and largely compensates the polar contribution ap> 1 to the selectivity (curve 2b). Because both aLand a p increase with increasing concentration of the organic solvent in the mobile phase, the rr,/ also increases with @ (curve 2c). In this example, the value of r,,J< 1 in mobile phases containing less than 22% of methanol, while above this concentration, it is larger than unity and reversal of the order of elution of desphenuron and hydroxymetoxuron results in 22% methanol. The concentration at which the selectivity cross-over occurs can be predicted from the lipophilic and polar indices of the sample solutes. Compounds with equal polar functional groups, but with different size of the molecules possess approximately equal polar indices and differ only in the lipophilic indices, so that the selectivity of separation is controlled only by the lipophilic contribution. Members of various homologous series, but also some pairs of non-homologous compounds behave in this way. For example, the difference in the polar indices of chlorobromuron and linuron Aq is very close to zero and consequently ap= 1 (curve 3b in Fig. 8.6). Practically, it does not contribute to the selectivity of separation (curve 3c), which is equal to the lipophilic contributionaL> 1 (curve 3a) and decreases with increasing concentration of the organic solvent in the mobile phase, @, because of the positive difference in lipophilic indices Anc. Figure 8.7 shows an example of the change in the order of elution observed for the pair of compounds benzophenone and chlorobenzene on a C8 column in methanol-water and in acetonitrile-water mobile phases, where both the Anc and Aq differences in polar and lipophilic indices are positive [12]. Here, the lipophilic contributions to selectivity aL> 1 (curves 1 and 4) and the polar contributions ap< 1 (curves 2 and 5). The resulting selectivity rl,Jdecreases with increasing @ and is >1 in mobile phases containing less than 73% of methanol or 66% of acetonitrile (curves 3 and 6 ) . At these concentrations, the two compounds co-elute in a single peak. When the concentration of the organic solvent in the mobile phase is further increased, rl,Jfalls below unity and a change in the order of elution occurs. The points represent the experimental data. The experimentally observed concentrations of methanol and acetonitrile corresponding to the reversal of the elution order differ less than 5% from that predicted by calculation from Eq. (8.6) setting rl,J= 0. Several numerical values of the indices, the lipophilic and the polar contributions to the retention aLand apand the calculated and experimental relative retention are given in Table 8.7. The data are given here also for two phenylurea herbicides, chlorobromuron and linuron, with the change of the elution order shown in Fig. 8.6. References pp. 294-295

286

Chapter 8

"40 1.25

A '

0.80 0.50

1

I

I

I

0.60

0.70

0.80

0.90

1 .oo

50 Fig. 8.7. Lipophilic (aL,curves 1 and 4) and polar (ap, curves 2 and 5) contributions to the experimental (points) and calculated (curves 3 and 6 ) separation selectivities r , , j , for benzophenone and chlorobenzene on a Silasorb C8 column as a hnction of the concentration of the organic solvent, $, in methanol-water (curves 13) and in acetonitrile-water (curves 4-6) mobile phases.

TABLE 8.7 LIPOPHLIC (ad AND POLAR (ap)CONTRIBUTIONS TO THE SEPARATION SELECTIVITY rl, ON A SILASORB C8 COLUMN Mobile phase CH30HH20 CH3CNH20

CH30HH20 CH3CN-

H20

60:40 70:30 80:20 90:lO 55:45 60:40 70:30 80:20 60:40 70:30 80:20 90:lO 55:45 60:40 70:30 80:20

Pair of solutes

aL

UP

'1,J

rl, J

1: benzophenone, i : chlorobenzene, An, = 0.38, Aq = 0.08 j : benzophenone, i: chlorobenzene, An, = 0.34,

1.22 1.16 1.11 1.08 1.12 1.11 1.09 1.08 1.19 1.15 1.11 1.07 1.13 1.12 1.10 1.08

0.89 0.88 0.86 0.85 0.93 0.92 0.91 0.89 0.96 0.95 0.95 0.94 0.93 0.92 0.91 0.90

1.09 1.02 0.95 0.92 1.04 1.02 0.99 0.96 1.14 1.09 1.05 1.00 1.05 1.03 1.oo 0.97

1.11 0.99 0.94 0.93 I .03 1.01 0.98 0.94 1.09 1.11 1.14 0.92 1.08 1.01 1.02 0.97

A = 0.06 j : chlorobromuron,

i: linuron, An, = 0.35, Aq = 0.03 j : chlorobromuron, i: linuron, An, = 0.35, Aq = 0.06

Exp, experimental values, calc, values predicted by calculation using Eq. (8.6). Data from ref. [12].

exp

Lipophilic and polar indices

287

8.5.2 Ternary mobile phases

The characterization of selectivity using two-indices approach can be applied not only to binary, but also to ternary mobile phases containing water and two organic solvents X and Y. The relationships were derived for the capacity factors and for the relative retention in dependence on the concentrations of each organic solvent, $x and qbY7 in the ternary mobile phase, using the lipophilic and polar indices and the parameters a,, al and p of the calibration homologous series determined in the individual binary solvent systems waterorganic solvent X and water-organic solvent Y [ 121:

(8.19)

The subscripts x and y denote the values of the constants and the indices in the binary mobile phases water-Xand water-Y, respectively. Using Eq. (8.19), the following expression for the relative retention r,,, of the two solutes i andj (the selectivity of separation) in ternary mobile phases is obtained [12]:

1.70

1.60 3

c

1.50

1.40 0.00

0.20

0.40

0.60

0.80

Fig. 8.8. Changes in selectivities r, in homologous series of n-alkylbenzenes (2), 3,5-dinitrobenzoates of nalkanols (3) and 4-bromophenacyl esters of n-alkanoic acids (4) with the concentration of acetonitrile, @ACN, in ternary mobile phases methanol-acetonitrile-water (containing 30% of water) on a Silasorb C18 column. Curve 1, values calculated from Eq.(8.20)with Aq = 0 and Anc = I (a1 = 0 . 5 3 6 , ~= 0.871 in methanol-water; a1 = 0 . 2 7 7 , ~ = 0.599 in acetonitrile-water mobile phases).

References pp. 294-295

288

Chapter 8 /

\

(8.20)

where An,,, ,An,,y and Aq,, Aqy are the differences between the indices nCeand qr of the two solutes i a n d j in binary mobile phases water-organic solvent X and water-organic solvent Y. The first and the third term on the right-hand side of Eq. (8.20) represent the lipophilic contribution to the selectivity of separation, logaL, and the second and the fourth terms the polar contribution, logap, in analogy to Eq. (8.16) for binary mobile phases. The selectivity of separation between the individual members of homologous series is controlled only by the lipophilic contribution. As shown in Fig. 8.8, this contribution in ternary mobile phases changes much in the same way for various homologous series and increases regularly with increasing polarity of the mobile phase, for example with increasing content of more polar methanol in the phases with a constant sum of the concentrations of methanol and acetonitrile (70%). The numerical data are given in Table 8.8, where the experimental selectivities for three homologous series are compared with the values calculated using Eq. (8.20), assuming ideal behaviour of the homologues with Aq = 0 and An, = 1 both in binary and in ternary mobile phases. The selectivity in binary mobile phases is more than sufficient for the separation of the members of homologous series and there is no real need to use ternary mobile phases for this purpose. More interesting is the application of ternary mobile phases for nonhomologous compounds, where the selective polarity effects can be utilized for fine tuning of separation. Figure 8.9 shows several examples of the effect of ternary mobile

TABLE 8.8 CALCULATED (c) AND EXPERIMENTAL (e) SELECTIVITIES, ri,, On a Silasorb C18 column in ternary mobile phases methanol-acetonitrile-water for homologous nalkylbenzenes (I), 3,5-dinitrobenzoates of n-alkanols (2) and 4-bromophenacyl esters of n-alkanoic acids (3) Mobile phase (CH30H:CH3CN:H20) 70:0:30 55:15:30 40:30:30 30:40:30 15:55:30 0:70:30

rl,

J(')

1.60 1.57 1.53 1.51 1.48 1.45

r;,Jte) 1

1.63 1.61 1.58

1.55 1.51 1.45

rl,

J(e)

'I,

Jte)

2

3

1.61

1.58 1.52 1.49 1.43 1.46 1.42

1.58

1.54 1.52 1.48 1.43

Eq. (8.20) with Aqi,j = 0 and An, = 1 in both methanol-water and in acetonitrile-water mobile phases and the constants of the calibration n-alkylbenzene homologous series in methanol-water (a]= 0.536, p = 0.871) and in acetonitrile-water(u1 = 0.277, p = 0.599) binary mobile phases were used for calculations. Data from ref. [I21

Lipophilic and polar indices

289

41

Y

0 '

0.00

I

I

I

0.20

0.40

0.60

0.80

Fig. 8.9. Experimental (points) and calculated (curves, Eq. 8.19) capacity factors k as a function of the concentration of acetonitrile, @ACN, in ternary mobile phases methanol-acetonitrile-water (containing 30% of water) on a Silasorb C18 column for 3-chlorotoluene (I), 1,4-dichlorobenzene(Z), toluene (3), 3-bromonitrobenzene (4), anisole (5) and nitrobenzene (6).

phases on retention and Table 8.9 lists experimental and calculated selectivities of separation for the three pairs of compounds in this figure. The retention of all the compounds decreases with increasing concentration of acetonitrile, @ACN, the more polar of the two organic solvents, in ternary mobile phases at a constant sum of concentrations of acetonitrile and methanol in water. The dependences of k on @ACN have a sigmoidal shape and the differences in the capacity factors of the three pairs of compounds (and correspondingly, the separation selectivities) either decrease or increase with increasing concentration of acetonitrile in the ternary mobile phase. This behaviour can be explained by the effects of the lipophilic and polar indices. The differences between the lipophilic and the polar indices Anc and Aq of 3-chlorotoluene and 1,4-dichlorobenzene(curves 1 and 2 in Fig. 8.9) have negative signs in acetonitrilewater and methanol-water mobile phases and both the lipophilic and the polar contributions to retention decrease with increasing concentration of acetonitrile. The relative retention first decreases with increasing concentration of acetonitrile and decreasing concentration of methanol, until reversal in the order of elution occurs at 60% of acetonitrile in the mobile phase. The relative retention of the pair anisole-nitrobenzene is almost independent of the ratio of the concentrations of acetonitrile and methanol (curves 3 and 4 in Fig. 8.9), as the lipophilic contribution to the separation selectivity increases and the polar contribution decreases with increasing concentration of acetonitrile in the mobile phase. The changes in the lipophilic and in the polar contributions to the retention tend to compensate one another, because the positive differences between the indices An, increase while the References pp. 294-295

290

Chapter 8

TABLE 8.9 EXPERIMENTAL (exp) AND CALCULATED (cal) SEPARATION SELECTIVITIES ri,, IN TERNARY MOBILE PHASES METHANOL-ACETONITRILE-WATER ON A SILASORB C18 COLUMN

Differences in indices in binary mobile phases methanol-water (M) and acetonitrile-water (A)

%MeOH

70

55

15

0

% ACN

0

15

55

70

I

M: An,=-0.17, Aq=-0.05 A: An, = -0.10, Aq = -0.02

r; ,cal Ti.F*P

1.06 1.05

1.04 1.02

1.00 1.00

0.99 0.98

I1

M: An, = 0.14, Aq =-0.12 A:An,=0.31, Aq=-O.O9

r,,,cal ri . P X P

1.31 1.31

1.31 1.33

1.31 1.27

1.30 1.27

III

M: An, = -0.15, Aq = -0.08 A: An, = -0.24, Aq = -0.17

rl . Fa1

1.06 1.06

1.08 1.05

1.16 1.12

1.19 1.17

,

r1,IexP

Solutes: (I) j , 3-chlorotoluene; i, 1,4-dichlorobenzene. (11) j , anisole, i, nitrobenzene. (111) j, toluene, i, 3bromonitrobenzene. Data from ref. [12].

negative differences between the indices Aq decrease with increasing concentration of acetonitrile in ternary mobile phases. In the last example, the difference between the polar indices Aq of toluene and 3bromonitrobenzene has a higher negative value in acetonitrile than in methanolic mobile phases and both the lipophilic and the polar contributions to selectivity increase with increasing concentration of acetonitrile in the ternary mobile phase, in contrast to the other pairs of compounds in the figure. Consequently, the relative retention slightly increases with increasing concentration of acetonitrile (curves 5 and 6 in Fig. 8.9). The numerical values for the data in Fig. 8.9 are given in Table 8.9 and the experimental selectivities agree well with the predicted values. 8.5.3 Gradient elution The selectivity of separation of a pair of compounds i and j in gradient-elution chromatography can be expressed as the difference in the retention volumes A VIR rather than the ratio of the capacity factors (which are not defined under gradient-elution conditions) and can be predicted by calculation after combination of Eqs. (8.12), (8.14) and (8.15) for each of the two compounds i and j . The selectivity in gradient-elution chromatography depends not only on the values of the relative indices An,, Aq, but also on the parameters A G , B G of the gradient and on the indices n,, qi.Reversal in the order of elution can be predicted for some pairs of compounds as in isocratic separations. Figure 8.10 shows several examples of the influence of the linear gradient methanolwater on the separation selectivities of phenylurea herbicides with different values of the lipophilic and polar indices. Gradients with a constant duration (constant gradient volume VG)were considered, so that the slope of the gradient B G and the initial concentration A G are interrelated by the condition

Lipophilic and polar indices

29 1

(8.21) where @G is the concentration at the end of the gradient, corresponding to the volume V,. In the examples in Fig. 8.10, @G = 1 (100% methanol) at the volume of the eluate VG= 42.75 cm3. Linear gradients with a constant gradient volume VGare characterized by a single parameter such as the initial concentration A G , much like the isocratic separation on a given column is characterized by 4J. Curve 1 in Fig. 8.10 shows the difference in the retention volumes of desphenuron and hydroxymetoxuron as a hnction of AG. Here, the differences between the lipophilic and the polar indices of the two compounds, Anc, Aq, are relatively high and of the same signs, so that reversal in the order of elution occurs under isocratic conditions in mobile phases with 23% methanol in water (Fig. 8.6). Similar behaviour is observed in gradientelution chromatography, where the order of elution using gradients starting at 7% methanol or less is reversed with respect to the gradients with higher initial concentrations AG. Furthermore, maxima are found on the AVG versus A G plots for various pairs of solutes indicating that the best separation is achieved with gradients starting at 4 0 4 5 % of methanol in the mobile phase for all the phenylureas studied (curves 1, 3-5), with the exception of the pair desphenuron-phenuron (curve 2). The occurrence of the maxima

1.60

1.20

0.80

0.40

0.00

F-' '

-0.40 0.00

I

I

I

I

I

I

0.10

0.20

0.30

0.40

0.50

0.60

0.70

A, Fig. 8.10, Changes in selectivities (differences in the retention volumes, Au',) in the gradient-elution separation of phenylurea herbicides with the concentration of methanol, A,, at the start of linear gradients with a constant gradient volume ( VG = 42.75 cm3) on a Silasorb SPH CIS column (VM = 3.18 cm3, gradient delay of the instrument = 2.3 1 cm3). Pairs of solutes: 1, hydroxymetoxuron and desphenuron, Anc = -0.48, Aq =-0.94; 2, desphenuron and phenuron, Anc = 0.25, Aq = 0.04; 3, fluometuron and chlorotoluron, An, = -0.01, Aq = 0.12;4, chlorotoluron and isoproturon, Anc = 0.30, Aq = 0.02; 5, isoproturon and diuron, Anc =-0.03, Aq =0.11.

References pp. 294-295

292

Chapter 8

can be explained by the combined effects of lipophilic and polar indices on the rate of decrease of the retention volumes with increasing A G . At lower A G these effects result in a steeper decrease of the retention volume of the earlier eluted compound while at higher AG the retention volume of the more strongly retained of the two compounds decreases more rapidly with increasing AG. The value of A , at which the rate of decrease equilibrates for the two compounds corresponds to the maximum on the plots in Fig. 8.10 and it increases with increasing value of the nCeand decreasing value of the qi indices, so that it is close to AG = 0 for the pair desphenuron-phenuron with low and negative nCeand high qi indices while it is shifted to higher values with increasing nCeand decreasing qi indices of other compounds. It should be noted that the reversal in the order of elution is not observed if the pairs of phenylurea herbicides have similar values of either nce or qi indices, i.e. if either An, or Aq are close to zero. The relationships between the lipophilic and polar indices and the selectivity in gradient elution were utilized for optimization of gradient-elution separation of phenylurea herbicides [8]. 8.6 CONCLUSIONS

The retention of each solute in reversed-phase chromatography can be characterized by two indices, the lipophilic and the polar one. These two indices can be used for prediction of retention and selectivity in chromatography with binary and ternary mobile phases and with gradient elution. In mobile phases with 40-90% of organic solvents, n-alkylbenzenes are most convenient for calibration of the retention scale, but at higher contents of water, various more polar calibration series can be used instead. The accuracy of prediction is generally better than with the method using interaction indices and the errors of prediction of the retention volumes are usually lower than 10% relative. The lipophilic and the polar indices determined on different columns with equal lengths of the bonded alkyl groups are close to one another. However, more significant differences are found between the indices determined on columns with different lengths of the bonded alkyl groups, such as between the data measured on a C18 and on a C8 column. The reversal in the order of elution can be predicted using the two-indices approach and various selectivity effects can be explained by simultaneous influence of the size of the nonpolar part of the solute molecule and of the polarities of the functional groups, characterized by the lipophilic and the polar contributions to selectivity, which is understood as the relative retention (ratio of the capacity factors) under isocratic conditions and as the difference in the retention volumes in gradient-elution chromatography. For some groups of solutes, the lipophilic and the polar indices are comprised of additive increments of structural units, which can be used for prediction of the values of the indices, but more data will be necessary to verify general validity of the additivity rules. 8.7 GLOSSARY OF THE TERMS AG

initial concentration rp of the organic solvent at the start of gradient elution

Lipophilic and polar indices

293

slope of the linear gradient of the organic solvent in water in gradient elution, in %vol./vol.1W m l Snyder's polarity index [6] volume of the eluate from the start of the elution volume of the gradient, i.e. volume of the eluate from the start to the end of the gradient elution volume of the mobile phase in the column net retention volume difference between the retention volumes of the solutes i and j = selectivity in gradient-elution chromatography intercept of the log k versus @ relationship; Eq. (7.8) in Chapter 7 a of solutes i, j a of a reference standard compound intercept of the linear dependence of a on n in a homologous or oligomeric series; Eq. (8.2) a. in binary mobile phases water-organic solvent X or water-organic solvent Y slope of the linear dependence of a on n in a homologous or in an oligomeric series; Eq. (8.2) al in binary mobile phases water-organic solvent X or water-organic solvent Y capacity factor of a sample solute k of a reference standard compound k in pure water as the mobile phase slope of the log k versus @ relationship; Eq. (7.8) in Chapter 7 m of solutes i,j m of a reference standard compound number of repeat structural units in the molecule of a homologue or of an oligomer lipophilic structural index, an equivalent to the number of methylene groups in the molecule of a hypothetical member of the calibration homologous series with the same lipophilic character as the sample solute; Eq. (8.3) n, of solutes i o r j rice of a reference standard compound nce in binary mobile phases water-organic solvent X or waterorganic solvent Y difference between the lipophilic index of the solute and that of the reference standard or between the lipophilic indices of two solutesj and i contribution of a substituent or of a repeat group in a homologous or an oligomeric series to Anc contribution of ortho-position of substituents to Anc Anc in binary mobile phases water-organic solvent X or waterorganic solvent Y References pp. 294-295

Chapter 8

Anc,, of the parent molecule or of the end groups in a homologous or oligomeric series slope of the correlation equation (Eq. 8.1) between the parameters m and a in a homologous or oligomeric series p in binary mobile phases water-organic solvent X or water-organic solvent Y intercept of the correlation equation (Eq. 8.1) between the parameters m and a in a homologous or oligomeric series polar structural index = hypothetical equivalent of a sample solute to the constant q of Eq. (8.2) for the calibration homologous series q, of solutes i o r j q, of a reference standard compound q, in binary mobile phases water-organic solvent X or water-organic solvent Y relative polar index, difference between the polar indices of a solute and of the reference standard, or difference between the polar indices of two solutes i and j contribution of a structural substituent to Aq contribution of ortho-position of substituents to Aq Aq, of the parent molecule or of the end groups in a homologous or oligorneric series Aq in binary mobile phases water-organic solvent X or water organic solvent Y relative retention (selectivity coefficient) r,,]= k/k, lipophilic contribution to the separation selectivity polar contribution to the separation selectivity Hansch and Leo hydrophobic substituent constant [5] volume fraction of the organic solvent in aqueous-organic mobile phase (in %vol./vol. 9 of the organic solvents X or Y in a ternary mobile phase water-XY at the end of the gradient elution, corresponding to the volume of the eluate V = VG

8.8 REFERENCES L.R. Snyder, J. Chromatogr. Sci., 16 (1978) 223. P. Jandera, J. Chromatogr.,314 (1984) 13. P. Jandera, Chromatographia, 19 (1984) 101. P. Jandera, J. Chromatogr., 352 (1986) 91. C. Hansch and A. Leo: Substituent Constants for Correlation Analysis in Chemistry and Biology, Wiley, New York, 1979. L.R. Snyder and J.J. Kirkland: Introduction to Modern Liquid Chromatography, 2nd. edn., WileyInterscience, New York 1979, p. 260. P. Jandera, J. Chromatogr., A, 656 (1993) 437. P. Jandera and M. SpaEek, J. Chromatogr.,366 (1986) 107.

Lipophilic and polar indices 9 10 11 12 13 14 15 16 17 18 19 20 21 22

P. Jandera, J. Chromatogr., 449 (1988) 361. W.E. Hammers, G.J. Meurs and C.L. De Ligny, J. Chromatogr., 247 (1982) 1. P. Jandera and J. RozkoSnA, J. Chromatogr., 556 (1991) 145. P. Jandera, J. Chrornatogr., 352 (1986) 111. P. Jandera and J. Churhzek, J. Chromatogr., 91 (1974) 223. P. Jandera, J. ChurhCek and L. Svoboda, J. Chromatogr., 192 (1980) 37. T. Braumann, G. Weber and L.H. Grimme, J. Chromatogr., 261 (1983) 329. S. Bitteur and R. Rosset, J. Chromatogr., 394 (1987) 279. C.E. Werkhoven-Goewie, U.A.Th. Brinkman and R.W. Frei, Anal. Chem., 53 (1981) 2072. R.M. McCormic and B.L. Karger, Anal. Chem., 54 (1982) 435. E.M. Thurman, R.L. Malcolm and G.R. Aiken, Anal. Chem., 50 (1978) 775. M.J.M. Wells, C.R. Clark and R.M. Patterson, J. Chromatogr., 235 (1982) 43. M.J.M. Wells and C.R. Clark, J. Chromatogr., 284 (1984) 319. P Jandera and J. Kubat, J. Chromatogr., 500 (1990) 281.

295

This Page Intentionally Left Blank

R.M. Smith (Ed.),Retention and Selectivity in Liquid Chromatography Journal of Chromatography Library,Vol. 57 Q 1995 Elsevier Science B.V. All rights reserved

297

CHAPTER 9

Solvent selectivity S.D. West North American Environmental Chemistry Laboratory,DowElanco, Indianapolis,IN 46268, USA

9.1 INTRODUCTION

Within the past 10-15 years, high-performance liquid chromatography (HPLC) has grown in importance as an analytical technique [ 1-31. Reversed-phase (RP) chromatography dominates published applications of HPLC [3], and it has been estimated that 80-90% of modem HPLC separations utilize the RP mode [4]. The RP mode has been used primarily for non-polar and hydrophobic compounds, whereas the normal-phase (NP) mode has been used mainly for specialized applications involving water-sensitive compounds, isomers, polymers, and metallic ions [3]. Despite the enormous popularity of RP-HPLC, the mechanisms of retention and selectivity have not been well understood. A variety of attempts to predict retention have been published, and this topic has been reviewed [2,3]. As in the early days of gas chromatography, one of the major problems in predicting HPLC retention data, or in the utilization of published data, has been the irreproducibility of capacity factors (k)and relative retention times, as well as the lack of universally accepted standard reference compounds for calculating relative retention times. It has been noted that five- to tenfold variations in HPLC k values for the same compound have been observed even when the same mobile phase and column type were used [ 5 ] . The development of the Kovats retention index system [6] largely alleviated this problem for gas chromatography. The Kovats retention index is highly reproducible between laboratories and varies mainly with temperature and stationary phase but very little with other variables that affect GC retention times. The overwhelming acceptance of the Kovats system lead to a breakthrough in many aspects of GC selectivity and structure-retention relationships. A 1983 survey, “TwentyFifth Anniversary of the Retention Index System in Gas-Liquid Chromatography”, listed 1380 references on the subject [7]. At least one book has appeared that was devoted solely to the publication of GC retention indices [8], and a GC retention index library has been established [9]. In RP-HPLC, the retention behavior of a compound depends upon complex interactions between the solute, the stationary phase, and the mobile phase. The ability to deReferences pp. 335-336

298

Chapter 9

scribe these interactions quantitatively would allow retention behavior and resolution to be predicted. With this goal, various retention index systems have been developed for HPLC [lo-141 that are analogous to the Kovats retention index system in gas chromatography. One of the systems proposed for HPLC is the 2-keto alkane retention index system [ 1 11, which utilizes a series of C3 to C23 2-keto alkanes that are commercially available as a kit of highly pure standards capable of spanning a very wide range of solute retentions. This system has been found to substantially reduce the interlaboratory variation of retention data [15], and it appears to offer the same potential for a rapid progression of research in selectivity and structure-retention relationships in HPLC as was achieved with the Kovats system in GC. The 2-keto alkane retention index system has been demonstrated to be useful for HPLC retention prediction with several types of drugs [ 11,15-181, glucuronides [ 191, urushiol congeners [20], steroids [2 11, azabicycloalkanes [22], ergopeptines [23], and substituted aromatics [24]. The study of HPLC selectivity and structure-retention relationships has been further complicated by the large number of liquids available for use as mobile phase solvents. A widely accepted technique for characterizing solvent selectivity has been the solvent selectivity triangle concept [25,26], which establishes the primary selectivity characteristics of various solvents according to their relative ability to engage in proton acceptor, proton donor, and strong dipolar interactions. When the resulting values are plotted on three axes in the form of a triangle, solvents having similar functionalities tend to fall within the same area of the triangular plot. The theory of the solvent selectivity triangle concept is that solvents grouped in the same area of the triangle will have similar selectivity, while solvents from the other groups should exhibit different selectivity for a given separation [27]. This theory has been widely accepted and has often formed the rationale for solvent selection for a given separation, optimization technique, or structure-retention study [e.g. 28-32]. Until recently, definitive studies on the accuracy of this approach had been lacking. However, a number of publications have now questioned the reliability of the selectivity triangle for grouping HPLC solvents according to their resolving abilities. The solvent triangle has not accurately predicted normal phase or reversed-phase selectivity for the separation of polystyrene oligomers [33] or RP selectivity for the separation of steroids [21] or benzene derivatives [24]. The study with the polystyrene oligomers showed that the degree of solute solubility in the pure mobile phase was a better predictor of selectivity than were the groupings of the solvent triangle, while the studies with steroids and benzene derivatives demonstrated that RP-HPLC retention indices based upon 2-keto alkanes can be used to accurately predict and optimize resolution with isocratic or gradient elution. The failure of the solvent selectivity triangle concept with polystyrene oligomers and steroids both occurred with the separation of compounds possessing similar functionalities, whereas the study with benzene derivatives demonstrated the failure of the solvent selectivity triangle to predict selectivity differences for compounds having a wide variety of functionalities. This chapter describes the techniques for characterizing RP-HPLC solvent selectivity for predicting retention and selectivity by utilizing 2-keto alkane retention indices.

Solvent selectivity

299

9.2 EXPERIMENTAL 9.2.1 Chemicals and reagents The steroids, aromatic compounds, and 2-keto alkane standards were obtained fiom Sigma Chemical Co., Aldrich Chemical Co. and Foxboro/Analabs, respectively. The solvents were HPLC-grade and were obtained fiom Burdick and Jackson Laboratories, Inc. The 2-keto alkanes were prepared in methanol at concentrations of approximately 1.O mg/ml, and the steroids and the aromatic compounds were prepared in methanol at concentrations of approximately 0.1 mg/ml. The solutes and standards were prepared in methanol to provide for adequate solubility and to minimize the potential for hydrolysis in aqueous mobile phase mixtures. To determine the hydrolytic stability of benzoyl chloride during exposure to water in the mobile phases, its hydrolysis product (benzoic acid) was also chromatographed. Benzoyl chloride was found to be stable in each of the mobile phase combinations studied, as evidenced by very different retention times for benzoyl chloride and benzoic acid.

9.2.2 Instrumentation The HPLC system was a Hewlett-Packard Model 1084B equipped with a variable wavelength UV detector and a recording integrator. The HPLC column was a Beckman C18 Ultrasphere ODS (25 cm X 4.6 mm id.) with a particle size of 5 pm. The oven temperature was maintained at 30°C. Fifty-microliter aliquots of sample solutions were injected. The mobile phase flow rate was 0.8 mumin, and the chart speed was 0.1 cdmin. The UV detector was operated at 280 nm. Isocratic conditions were maintained throughout the study. The gas chromatograph was a Tracor Model MT-220 equipped with a flame ionization detector. GC columns were 183 cm X 4 mm i.d. glass tubing. Stationary phases were 10% on 80/IOO mesh Gas Chrom Q, and 10 mg of each compound was injected. Nitrogen was the carrier gas at a flow rate of 40 ml/min. The oven was maintained isothermally at 100°C.

9.2.3 Determination of retention data Whenever the HPLC mobile phase composition was changed, the chromatographic system was allowed to equilibrate for at least 1 h prior to sample injection. Retention times were measured to within 0.01 min by a recording integrator. Peak widths at the base line were measured to the nearest 0.1 mm (equivalent to 0.1 min) with the aid of a contact magnifying lens scale (American Scientific Products) using the peak tangent technique [27]. HPLC retention indices (Z) were calculated using a computer program according to Eq. (9. I):

References pp. 335-336

300

Chapter 9

where N and N + 1 are the smaller and larger chain 2-keto alkanes whose adjusted retention times (t’) bracket that of solute x . GC retention indices were calculated as in Eq. (9.1) except that n-alkanes were used as standards [6].

9.2.4 Selection of solvents and solutes

Twelve HPLC solvents were chosen for use in this study based upon the following criteria: suitable miscibility with water for use in RP systems, UV cutoffs not exceeding 280 nm to avoid interference with detection of the solutes and standards, relatively low viscosities to permit acceptable back pressures, and commercial availability as HPLCgrade solvents. The properties of the solvents used in this study are listed in Table 9.1. The 12 solvents include four from Group 11, five from Group 111, and three from Group VI of the selectivity triangle. Based upon solvent polarity P’ values [27,34], these solvents ranged in polarity from P’ = 3.9 to P’ = 7.2. Thirty steroids were initially chromatographed, and twelve were selected that eluted within an acceptable retention time range (normally 3-60 min) over a 15% concentration range of strong solvent in aqueous binary mobile-phase mixtures. The 12 steroids that were selected for extensive study are listed in Table 9.2. Sixteen benzene derivatives were selected for study. The compounds chosen included 13 monosubstituted benzenes with different functionalities and three positional isomers of nitrotoluene (Table 9.3). The abbreviations used for the solvents and aromatic compounds are listed in Tables 9.1 and 9.3, respectively.

TABLE 9.1 SOLVENTS STUDIED Solvent

Abbreviation

Groupa

P’a

$c

Methanol Ethanol 1-Propanol 2-Propanol

MeOH EtOH 1 -PrOH 2-PrOH

I1 I1 I1 I1

5.1 4.3 4.0 3.9

0.589 0.493 0.355 0.438

Dimethylformamide Dimethylsulfoxide 2-Ethoxyethanol 2-Methoxyethanol Tetrahydrofuran

DMF DMSO 2-EE 2-ME THF

111

I11

6.4 7.2 5.0b 5.5 4.0

0.572 0.703 0.524 0.625 0.420

Acetonitrile Dioxane 2-Methoxyethyl acetate

ACN Diox 2-MEA

VI VI VIb

5.8

4.8 4.0b

0.485 0.508 0.490

I11 IIIb 111

aReference27 unless otherwise indicated. P‘ = polarity value. bReference 34 unless otherwise indicated. CVolumefraction.

Solvent selectivity

301

TABLE 9.2 STEROIDS STUDIED Cortisone Hydrocortisone Testosterone Corticosterone Androstenedione Reichstein’s substance S ( I 1-deoxy-17-hydrocorticosterone) 1 7a-Acetoxyprogesterone Ethisterone Prednisone Spironolactone Betamethasone Dexamethasone

9.3 RESULTS AND DISCUSSION 9.3.1 Prediction of retention and resolution with steroids

9.3.1.1Retention indices as a function of volumefvaction of strong solvent

Retention indices (Ix) were determined for the 12 steroids for seven 2.5% increments of six strong solvents in aqueous binary mixtures of the mobile phase. The six strong solvents included methanol and 1-propanol (Group II), tetrahydrofuran and 2-methoxyethanol (Group III), and acetonitrile and 2-methoxyethyl acetate (Group VI). The volume fractions of strong solvents (@)were adjusted with water to result in an acceptable range of retention times for the group of 12 steroids as a whole. For a given solvent, a linear relationship existed for the retention index of a given steroid as a function of 9, as described by the following equation: TABLE 9.3 BENZENE DERIVATIVES STUDIED Compound

Functionality

Abbreviation

Benzene Chlorobenzene Bromobenzene Toluene Styrene Nitrobenzene Acetophenone Methyl benzoate Benzaldehyde Phenol Benzonitrile Benzoyl chloride Anisole o-Nitrotoluene m-Nitrotoluene p-Nitrotoluene

-H -CI -Br “3H3 -CH=CH2 -NO2 -COCH3 -COOCH~ -CHO -OH -CN -COCI -OCH3 -N02, -CH3 -N02,-CH3 -NO2,-CH3

-H

References pp. 335-336

-C1

-Br -CH3

-CH=CH2 -NO2 -COCH3 -COOCH~ -CHO -OH -CN -COCI -0CH3 0-NT m-NT p-NT

3 02

Chapter 9

(9.2) The resulting regression coefficients (a and b) are summarized in Table 9.4. 9.3.1.2Prediction of retention indicesfor steroids

The linearity of Eq. (9.2) permits a prediction of retention indices as a fhction of $ for a given solvent. For example, using Eq. (9.2) and the regression coefficients from Table 9.4, predicted retention indices were compared with experimental retention indices for all 12 steroids using a mobile phase consisting of tetrahydrofuran at $ = 0.325. As shown in Table 9.5, the predicted retention indices compared very favorably with the experimental values. In like manner, the experimental and predicted retention indices for the steroids were compared at seven different values of $ for each of the six solvents, and the resulting percentage errors are summarized in Table 9.6. The average error ranged from 0.2% for methanol to 0.9% for 1-propanol. The overall error for all six solvents averaged 0.5% (n = 504).

TABLE 9.4 REGRESSION COEFFICIENTS FOR THE SLOPE (a)AND INTERCEPT (b) FROM EQ (9.2) Steroid

Cortisone Hydrocortisone Testosterone Corticosterone Androstenedione Reichstein’s substance S 17a-Acetoxy progesterone Ethisterone Prednisone Spironolactone Betamethasone Dexamethasone Average

Regress. coeff. a b a b a b a b

a b a b

a b

a b a b a b a b a b a

Group I1

Group VI

Group I11

MeOH

1-PrOH

-498.9 891.3 -439.2 894.1 -227.0 958.3 -341.3 921.7 -236.3 928.3 -422.3 976.4 -337.7 1079.5 -318.6 1014.0 -538.7 901.2 -383.7 1045.2 499.4 1000.5 497.1 1002.4 -394.5

-967.6 713.0 -903.6 717.8 -822.1 901.3 -904.4 787.3 -691.7 840.0 -980.3 830.1 -913.4 941.0 -918.6 930.8 -926.3 691.9 -1003.6 924.9 -847.4 773.3 -937.6 794.0 -901.4

THF

2-ME

ACN

2-MEA

-649.7 739.2 -671.1 763.9 -657.6 883.8 -687.3 809.2 -428.9 807.7 -682.7 837.9 -507.8 941.5 -534.4 935.4 -652.9 724.1 -625.3 897.9 -743.7 856.6 -878.3 790.0 -636.0

-649.7 939.6 -607.7 954.2 -430.9 1003.3 -494.3 973.1 -340.6 930.8 -552.0 994.9 498.0 1118.9 -541.1 1069.0 -660.3 920.8 -576.1 1080.6 -67 1.4 1032.3 -771.7 1097.5 -566.2

-504.7 669.3 -400.3 608.6 -230.0 783.2 -204.4 654.6 -182.7 820.9 -329.6 719.2 -303.7 996.9 -298.4 867.8 -501.3 653.3 -364.7 934.3 488.0 724.6 -448.9 711.4 -354.7

-756.4 754.2 -716.1 737.2 -568.0 865.6 -665.7 797.2 -408.0 835.4 -698.4 835.0 -514.4 984.8 -585.6 949.9 -794.9 735.5 -572.6 918.9 -823.1 839.9 -859.1 858.3 -663.5

Solvent selectivity

303

TABLE 9.5 COMPARISON OF PREDICTED AND EXPERIMENTAL RETENTION INDICES USING THF AT $= 0.325 Steroid

Retention index Predicted

Experimental ~~

Cortisone Hydrocortisone Testosterone Corticosterone Androstenedione Reichstein’s substance S 17a-Acetoxyprogesterone Ethisterone Prednisone Spironolactone Betamethasone Dexamethasone

528.0 545.8 670.0 585.8 668.3 616.0 776.5 761.7 511.9 694.6 614.9 621.5

526.9 547.0 667.7 584.9 666.5 614.9 776.0 76 1.4 511.5 692.4 613.2 621.0

The true predictive value of any mathematical model, however, lies in its ability to accurately predict results for conditions outside the range of those used to generate the coefficients of the equation. This capability is demonstrated in Table 9.7, which compares predicted and experimental retention indices with those subsequently determined for the steroids using methanol at q~= 0.5 13. 9.3.1.3 Prediction of resolution of steroid mixtures

The ability to predict retention indices can be used to predict whether or not a mixture of compounds can be separated under a given set of conditions. Alternatively, predicted retention indices may be used to select optimum conditions for obtaining complete resolution for a given mixture. An approach for predicting solute resolution from Kovats retention indices in GC has been reported [35] and subsequently verified [36]. This apTABLE 9.6 AVERAGE PERCENTAGE ERROR IN THE PREDICTION OF RETENTION INDICES FOR TWELVE STEROIDS WITH SIX SOLVENTS AT SEVEN VOLUME FRACTIONS PER SOLVENT Solvent

Average % Error

Methanol 1-Propanol Tetrahydrofuran 2-Methoxyethanol Acetonitrile 2-Methoxyethyl acetate

0.2 0.9 0.3 0.8 0.5 0.5

Overall average

0.5

References pp. 335-336

Number of determinations 84 84 84 84 84 84

Total

504

3 04

Chapter 9

TABLE 9.7 COMPARISON OF PREDICTED AND EXPERIMENTAL RETENTION INDICES WITH METHANOL AT

4 = 0.513 Steroid

Cortisone Hydrocortisone Testosterone Corticosterone Androstenedione Reichstein’s substance S 17a-Acetoxyprogesterone Ethisterone Prednisone Spironolactone Betamethasone Dexamethas.one

Retention index Predicted

Experimental

635.4 672.0 841.8 746.6 807.1 759.8 906.3 850.6 624.8 848.4 744.3 747.4

640.9 673.6 847.8 749.6 812.8 763.0 91 1.8 853.8 625.2 855.4 747.4 748.8

proach can be successfully adapted to HPLC to predict the resolution of various pairs of steroids. The approach utilizes retention indices and a simple linear equation to predict whether or not a given pair of compounds can be completely resolved (R, 2 1.5) under specified conditions. The equation is derived from classical chromatographic equations as follows: (9.3)

where R, is the experimental resolution, t‘, and t f 2are the adjusted retention times of the two components, and w1and w 2are the peak widths at baseline. In the derivation of Eq. (9.3), it is assumed that w1 and w2 are equal and that the peaks are Gaussian in shape. Furthermore, values of R, 2 1.5 indicate complete baseline resolution [37]. Thus, for complete resolution, Eq. (9.3) can be expressed as follows: 1.5 = [2(tf2- t’I)]/(Wl+ w2)

(9.4)

Because w1 and w 2 can be considered equal for closely eluting peaks [37] (i.e. w 1+ w, = 2wI),Eq. (9.4)can be solved in terms of tf2: t‘2 (minimum) = 1.5(w1)+ t f l

(9.5)

In this equation, t f 2(minimum) represents the minimum adjusted retention time of the second component that is required to yield baseline resolution from the first component. This minimum value for t’2 can then be substituted into Eq. (9.1) to determine the minimum retention index of the second component. The Z2 (minimum) values calculated in this manner are derived from an equation containing terms that reflect both the efficiency (wl+ w2)and the selectivity ( t f 2- f l )of the chromatographic system.

305

Solvent selectivity

TABLE 9.8 12 (MINIMUM) AS A FUNCTION OF @ FOR SEVERAL 2-KETO ALKANES WITH 2-METHOXYETHYL ACETATE AS THE STRONG SOLVENT Compound

2-Pentanone 2-Hexanone 2-Heptanone 2-Octanone 2-Nonanone Slope (s) Intercept Q

I1

500.0 600.0 700.0 800.0 900.0

12 (minimum) as a function of $J

0.325

0.350

0.375

0.400

0.425

0.450

0.475

538.8 621.4 712.5 809.0 907.5 0.925 70.3

543.0 622.8 712.3 807.6 905.4 0.910 81.5

544.2 629.3 718.6 808.4 907.9 0.907 87.1

553.9 629.4 719.0 812.0 908.1 0.891 100.8

544.6 630.7 719.3 812.9 910.6 0.914 83.7

550.5 632.9 723.8 820.8 910.7 0.908 91.9

542.1 642.2 726.1 819.4 910.2 0.913 88.6

The 2-keto alkanes are ideal standard compounds for determining Z2 (minimum) values because they are defined by Eq. (9.1) as having a retention index of 100 times the number of carbon atoms [ 111, thus providing identical reference points under all conditions. For example, using this approach with 2-octanone (Zl= 800) and methanol at $r = 0.550 resulted in the calculation of an Zz (minimum) value of 813.7. This indicates that a compound must possess a retention index of at least 813.7 to achieve baseline resolution fi-om 2-octanone under the given operating conditions. Representative Zz (minimum) values calculated for a series of 2-keto alkanes are listed in Table 9.8. The difference between Zz (minimum) and Zldecreases with increasing values for Z,.This occurs because the distance between the peak maxima of 2-keto alkanes increases logarithmically with increasing carbon number, whereas the number of retention index units representing the distance between the peaks remains constant at 100. As was noted in GC [35], Z2 (minimum) is a linear function of Zlunder a given set of conditions. The relationship is described by the following equation:

I2 (minimum) = s(Zl)+ y Representative regression coefficients for the slope (s) and y-axis intercept (y) are contained in Table 9.8. The correlation coefficient (r2)for Eq. (9.6) with the 2-keto alkanes in Table 9.8 ranged from 0.9994 to 0.9997. A value for Z2 (minimum) can thus be obtained for any value of Zlby using Eq. (9.6). Equation (9.2) for predicting retention indices may be used in conjunction with Eq. (9.6) to predict resolution. This approach is demonstrated in Table 9.9, where Id is defined as follows: Z,

= 1, (predicted) - Zz (minimum)

(9.7)

A positive value for Id indicates that the predicted retention index of the second component is greater than the minimum retention index calculated by Eq. (9.6), thus predicting baseline resolution. For example, the retention indices for cortisone and hydrocortisone are predicted by Eq. (9.2) to be 635.4 and 672.0 with methanol at @ = 0.513. AcReferences pp. 335-336

306

Chapter 9

cording to Eq. (9.6), the minimum retention index necessary for complete resolution from cortisone (II= 635.4) under these conditions is calculated to be 666.7. Therefore, Eq. (9.7) predicted that cortisone and hydrocortisone could be resolved, since I d = 672.0 666.7 = 5.3. An experimental R, of 1.91 verified this prediction (Table 9.9). In the same manner, this technique predicted positive Id values (i.e. R, > 1.5) for 56 compound pairs and negative I d values (i.e. R, < 1.5) for 10 compound pairs using methaTABLE 9.9 PREDICTION OF RESOLUTION WITH METHANOL AT q5 = 0.513 Compound pair

Predicted retention indexa

12

Id

Experimental resolution

(minimum)’

Cortisone Hydrocortisone

635.4 672.0

666.7

5.3

1.91

Prednisone Hydrocortisone

624.8 672.0

657.3

14.7

2.59

Prednisone Cortisone

624.8 635.4

657.3

-21.9

0.66

Hydrocortisone Betamethasone

672.0 744.3

699.4

44.9

5.95

Betamethasone Corticosterone

744.3 746.6

764.0

-17.4

0.19

Betamethasone Reichstein’s substance S

744.3 759.8

764.0

-4.2

1.31

Betamethasone Androstenedione

744.3 807.1

764.0

43.1

6.98

Corticosterone Reichstein’s substance S

746.6 759.8

766.0

-6.2

1.14

Dexamethasone Reichstein’s substance S

747.4 759.8

766.0

-7.0

1.16

Androstenedione Testosterone

807.1 841.8

820.1

21.7

4.45

Testosterone Spironolactone

841.8 848.4

851.1

-2.7

0.98

Ethisterone 1 7a-Acetoxyprogesteronc

850.6 906.3

858.9

47.4

4.61

aFrom Eq. (9.2). ’From Eq. (9.6),with 12 (minimum) = 0.893(/,) + 90.3.

307

Solvent selectivity

no1 at $ = 0.513. Comparisons of these predictions with experimental resolutions resulted in an accurate prediction in every case. Several examples are given in Table 9.9. 9.3.1.4Optimization of resolution of steroid mixtures

In addition to predicting if a given pair of compounds can be completely resolved under a given set of conditions, it is also possible to predict the optimum conditions for yielding baseline resolution in minimum time. To optimize resolution in this manner, it is necessary to calculate a value of $ where I d = 0 (i.e. where R, = 1.5). In Table 9.10, I d values are calculated using experimental retention indices and 1, (minimum) values for three compound pairs with methanol as the strong solvent at seven values of @. Plotting Id on the x-axis versus $ on the y-axis yielded a linear relationship described by the following equation:

where q50 is the y-axis intercept and therefore represents the volume fraction of strong solvent at which 1, = 0 (and R, = 1.5). The $o regression coefficients for the three compound pairs are listed in Table 9.10. This technique predicted baseline resolution at rp0 = 0.579 for pair number 1, at $o = 0.662 for pair number 2, and at $o = 0.734 for pair number 3. To test the accuracy of this approach, a mixture of the compounds was chromatographed isocratically at each of the three rp0 values calculated to give baseline resolution for a given pair in the mixture. The resulting chromatograms are presented in Fig. 9.1. As expected, baseline resolution was achieved for each compound pair at the predicted value of $@ However, none of the isocratic conditions that resulted in baseline resolution for a given pair resulted in the optimum separation of all of the components in the mixture. At $o = 0.579 (Fig. 9.1, chromatogram A), the retention times of the three latest eluting components were longer than necessary to achieve baseline resolution, resulting in an unnecessarily long analysis time of nearly 40 min. Increasing $, to 0.662 (Fig. 9.1, chromatogram B) or 0.734 (Fig. 9.1, chromatogram C) substantially reduced the analysis time, but at the expense of decreased resolution of compound pair number 1. Figure 9.2 demonstrates that Eq. (9.8) can be used to estimate a gradient system that should optimize the separation of the entire mixture. In this system, $ was held isocratiTABLE 9.10 Id AS A FUNCTION OF $J WITH METHANOL AS THE STRONG SOLVENT

Compound pair no.a 1

2 3

Id as a function of $J

Intercept ($0)

0.550

0.575

0.600

0.625

0.650

0.675

0.700

5.9 17.9 48.3

-3.4 13.8 43.4

-4.5 15.0 37.2

-17.3 12.3 35.4

-15.2 5.5 28.5

-28.3 -0.5 17.3

-45.7 -11.1 6.1

0.579 0.662 0.734

"Pair no. 1, prednisone and hydrocortisone; pair no. 2, androstenedione and testosterone; pair no. 3, testosterone and 17a-acetoxyprogesterone.

References pp. 335-336

308

Chapter 9

--

0

10

20 30 Minutes

40

0

10

20 30 Minutes

40

0

10

20 3 0 4 0 Minutes

Fig. 9.1 Chromatograms demonstrating the use of Eq. (9.8) for predicting isocratic values of q5 that will yield baseline resolution for given compound pairs using methanol as the strong solvent: (A) q5 = 0.579; (B) q5 = 0.662; (C) q5 = 0.734. (Reproduced from Ref. 21 with permission of Preston Publications, A Division of Preston Industries, Inc.)

-

0

10

20 30 Minutes

40

Fig. 9.2. Chromatogram demonstrating the use of Eq. (9.8) for predicting values of q5 for optimizing the gradient resolution of all five compounds in a mixture in minimum time. (Reproduced from Ref. 21 with permission of Preston Publications, A Division of Preston Industries, Inc.)

Solvent selectivity

309

cally at 0.579 for 10 min to provide resolution of pair number 1. At 10 min, cp was increased to 0.662 and held isocratically for 5 min to provide baseline resolution for pair number 2. At 15 min, $ was increased to 0.734 and held isocratically to resolve the last pair. This technique resulted in baseline resolution for all five compounds in minimum time. 9.3.1.5 Resolution and the solvent selectivity triangle concept

Several discrepancies in the solvent selectivity triangle concept were observed for the experimentally determined selectivities of the various solvents in this study. Examination of the slopes (a) describing the change in the steroid retention indices as a function of q5 showed that they varied considerably among solvents from the same selectivity group (Table 9.4). For example, the average slope was 2.3 times greater for 1-propanol than for methanol, even though both solvents are classified in Group I1 and should therefore result in similar selectivity for the steroids relative to the 2-keto alkane standards. Likewise, the slopes for the two Group V1 solvents were substantially different, averaging 1.87 times greater for 2-methoxyethyl acetate than for acetonitrile. In some instances, slopes were actually more similar for solvents in different groups than for those in the same group. For example, the average slope for 2-methoxyethyl acetate in Group V1 (-663.5) more closely resembled that of tetrahydrofuran in Group I11 (-636.0) than that of its Group VI counterpart, acetonitrile (-354.7). Similarly, the average slopes for methanol (-394.5) in Group I1 and acetonitrile (-354.7) in Group VI were more similar than those for their group counterparts, 1-propanol (-901.4) and 2methoxyethyl acetate (-663 3. Consequently, a study was undertaken in which experimental resolutions (R,) were determined by RP-HPLC for three steroid compound pairs using a total of 10 solvents from three different selectivity groups. Resolution was calculated using the following classical equation:

where tz and tl are the observed retention times and w1 and w 2are the peak widths at baseline. In order to ensure that the comparison of resolution was done at constant solvent strength, the capacity factor (k) of the earliest eluting compound in each solute pair was adjusted to 2.00 f 0.03 for each solvent in a binary mixture with water. As shown in Table 9.1 1, the results of this study confirmed that solvents in the same selectivity group frequently did not give similar resolution, even at identical solvent strength. Perhaps the most dramatic examples occurred with compound pair number 1, consisting of spironolactone and ethisterone. Within Group 111, 2-ethoxyethanol, 2-methoxyethanol, and tetrahydrofuran resulted in an experimental resolution of 0.67, 1.15, and 3.26, respectively. Likewise, the resolution of this compound pair by Group VI solvents was greater than baseline for dioxane (R,= 1.70), but less than baseline with 2-methoxyethyl acetate (R, = 0.87) and acetonitrile (R, = 0.59). Similar discrepancies occurred in the attempted separation of compound pair number 2 (prednisone and hydrocortisone) with Group VI solvents. The R, values ranged from 0.25 for acetonitrile to 1.34 for 2-methoxyethyl acetate. References pp. 335-336

310

Chapter 9

TABLE 9.11 COMPARISON OF SOLVENT SELECTIVITYFOR THE EXPERIMENTAL RESOLUTION (Rs) OF THREE STEROID PAIRS Solvent

Solvent group

Polarity

Experimental resolution

(P’) Pair no. la

Pair no. 2a

Pair no. 3a

2-Propanol I-Propanol Ethanol MethanoI

I1 I1 I1 I1

3.9 4.0 4.3 5.1

0.78 1.01 0.71 0.26

1.53 1.64 1.88 1.58

1.35 1.29 0.81 0.93

Tetrahydrofuran 2-Ethoxyethanol 2-Methoxyethanol

I11 I11 111

4.0 5.0 5.5

3.26 0.67 1.15

1.81 2.54 1.93

0.42 1.00 1.03

2-Methoxyethyl acetate Dioxane Acetonitrile

VI VI VI

4.0 4.8 5.8

0.87 1.70 0.59

1.34 0.90 0.25

2.94 2.09 2.83

aPair no, 1, spironolactone and ethisterone; pair no. 2, prednisone and hydrocortisone; pair no. 3, betamethasone and Reichstein’s substance S.

The results in Table 9.1 1 also demonstrated that actual separations were frequently more alike with solvents classified in different selectivity groups than for those within the same group. For example, the resolution of prednisone and hydrocortisone by two of the Group I11 solvents, tetrahydrofuran (R, = 1.81) and 2-methoxyethanol (R, = 1.93), much more closely resembled the resolution obtained with ethanol (R, = 1.88) from Group I1 than that with the other Group I11 solvent, 2-ethoxyethanol (R, = 2.54). Also, the resolution of spironolactone and ethisterone with a Group I11 solvent, 2-ethoxyethanol (R, = 0.67), more closely resembled those obtained with ethanol (R, = 0.71) from Group I1 and acetonitrile (R, = 0.59) from Group VI than those with the other Group I11 solvents, tetrahydrofuran (R, = 3.26) and 2-methoxyethanol (R, = 1.15). These results contradicted the theory of the solvent selectivity triangle concept, which states that solvents in the same group should result in similar selectivity, while those in different groups should yield different selectivities [26,27]. A literature search conducted on publications listed in the Science Citation Index demonstrated widespread usage of the selectivity triangle as a rationale for solvent selection. However, with the exception of a publication by Lewis et ul. [33], definitive studies on the accuracy of the solvent groupings in the selectivity triangle appeared to be lacking. Lewis et ul. studied the separation of polystyrene oligomers using a total of 27 solvents representing all eight of the selectivity groups and concluded that the solvent triangle did not accurately predict selectivity for the separations being studied. These authors reported that the degree of solute solubility in the pure mobile phase solvents was a better predictor of selectivity than were the groupings of the solvent triangle. Thus, studies with steroids and polystyrene oligomers had clearly demonstrated that solvents classified within the same group of the solvent selectivity triangle did not necessarily result in similar selectivity. Both of these studies were conducted with compounds

Solvent selectivity

311

containing similar functionalities. Consequently, it was of interest to determine if the solvent selectivity triangle approach would accurately predict selectivity for compounds containing different hctionalities.

9.3.2 Retention and selectivity studies with benzene derivatives

9.3.2. I Retention index variation with solvent selectivity HPLC retention indices (Z) were determined for the 16 compounds on a CI8 column for each of the 12 strong solvents in aqueous binary mixtures of the mobile phase. The solvents used for the study with the benzene derivatives are listed in Table 9.1. The fraction of the organic solvent in the mobile phase ($) was adjusted to yield a capacity factor (k) of 4.00 f 0.04 for benzene. The values for q5 and I are listed in Tables 9.1 and 9.12, respectively. In most cases, considerable variation in retention indices occurred for solvents classified in the same solvent group, suggesting quantitative selectivity differences for the solutes relative to the 2-keto alkane standards. In many cases, retention indices for a given compound were more similar for solvents in different groups than for those in the same group. For example, the retention index of chlorobenzene with methanol from Group I1 was 812.7, which more closely resembled that of 2-ethoxyethanol in Group I11 (809.2) and acetonitrile in Group VI (813.7) than those of the other Group I1 solvents, ethanol (871.5), 1-propanol (979.8), and 2-propanol (968.5). Likewise, the retention index for nitrobenzene with ethanol from Group I1 (654.0) most closely resembled that with acetonitrile from Group VI (653.6), and the retention index for nitrobenzene with methanol from Group I1 (635.0) most closely resembled those of 2-ethoxyethanol in Group I11 (628.0) and dioxane in Group VI (626.9). The retention index of acetophenone with tetrahydrokan (THF) from Group I11 (583.7) much more closely resembled those of the four solvents from Group I1 (568. I to 579.6) and acetonitrile from Group VI (579.1) than those for any of the other four solvents in Group I11 (513.4 to 531.4). In general, there was very little or no correlation between retention indices and the solvents grouped according to the selectivity triangle concept. 9.3.2.2 Resolution and the solvent selectivity triangle To more thoroughly evaluate the correlation of selectivity with the solvent groupings in the solvent triangle, a study was undertaken in which experimental resolution (R,) was determined by RP-HPLC for the 16 benzene derivatives using all 12 of the mobile phase solvents. In order to ensure that the comparison was done at constant solvent strength, the volume of each strong solvent (q5) in a binary mixture with water was adjusted to yield a capacity factor of 4.00 f 0.04 for benzene. The results of this study confirmed that solvents in the same selectivity group seldom give similar resolution, even at constant solvent strength. The resolution of all 15 compounds from benzene using each of the 12 solvents is shown in Table 9.13. Numerous examples of extreme variation of R, within the solvent groups are evident, with resolution frequently being more alike for solvents classified in different groups than for those within a given group. For example, the resolution of benzene and chlorobenzene with ethanol from Group I1 (6.03) most closely resembled those of dimethylformamide (DMF) in Group 111 (6.14) and acetonitrile in Group VI Referencespp. 335-336

TABLE 9.12 RETENTION INDICES OF SEVERAL AROMATIC COMPOUNDS WITH 12 RF' SOLVENTSa Compound

-H -CI -Br -CH3 -CH=CHz -NO2 -COCH3 -COOCH3 -CHO -OH -cN -COCI -333

0-NT

m-NT p-NT

Group II

Group VI

Group I11

MeOH

EtOH

1-PrOH

2-PrOH

DMF

DMSO

2-EE

2-ME

THF

ACN

Diox

2-MEA

714.7 812.7 844.2 819.1 849.4 635.0 579.6 687.6 553.2 476.8 552.3 687.1 699.5 710.8 736.8 723.8

776.5 871.5 901.3 884.3 907.9 654.0 568.1 678.0 555.9 494.0 557.1 681.9 735.3 732.8 760.4 743.3

870.2 979.8 1010.2 994.2 1024.0 675.1 579.4 691.8 572.0 435.1 572.7 693.0 784.9 767.2 792.6 776.4

870.6 968.5 1005.6 997.7 1022.0 680.0 575.0 694.1 567.1 504.9 562.6 696.4 784.3 761.8 790.8 774.0

634.0 718.4 736.7 725.4 731.8 601.0 531.4 618.6 508.7 514.2 531.1 620.7 622.5 661.8 691.0 681.9

551.2 637.9 649.7 666.9 665.0 546.2 517.9 615.7 494.9 459.9 505.5 616.0 576.7 614.1 626.3 624.0

734.7 809.2 828.0 839.5 843.3 628.0 513.4 630.5 518.1 491.9 520.9 634.1 687.0 687.0 720.3 704.9

697.4 768.1 784.6 789.7 787.2 598.7 515.7 627.0 500.0 432.0 503.8 627.8 654.9 662.5 693.4 682.4

811.9 885.2 900.2 911.3 9 18.4 705.2 583.7 677.0 595.8 598.2 620.3 678.3 755.5 774.8 803.4 786.7

720.4 813.7 840.6 815.0 836.6 653.6 579.1 674.1 567.8 460.8 593.9 673.5 698.3 728.6 750.4 740.0

722.5 801.4 816.9 816.3 823.4 626.9 552.3 664.0 547.8 471.8 560.6 661.4 693.4 698.6 721.4 705.5

757.3 833.7 853.1 848.1 857.3 664.6 569.5 672.5 573.7 499.8 595.6 672.5 719.2 732.9 752.6 739.2

aSee Table 1 for solvent abbreviations.

TABLE 9.13 EXPERIMENTALRESOLUTION OF SEVERAL AROMATIC COMPOUNDS FROM BENZENE WITH 12 RP-HPLC SOLVENTS IN AQUEOUS BINARY MOBILE PHASES Compound

4 1

-Br 4H3 -CH=CH;? -NO2 -COCH3 -COOCH3 -CHO -OH

-cN -COCl 4CH3 0-NT rn-NT P-NT

Group III

Group II

Group VI

MeOH

EtOH

I-PIOH

2-PrOH

DMF

DMSO

2-EE

2-ME

THF

ACN

Diox

2-MEA

8.29 10.87 8.77 10.81 4.78 7.04 1.64 6.81 9.94 7.19 1.71 1.oo 0.28 1.60 0.67

6.03 8.44 6.67 7.78 5.10 6.56 4.06 6.90 8.68 7.39 4.03 1.97 2.03 0.81 1.66

4.61 5.82 4.91 6.69 5.52 7.98 5.40 7.89 10.21 7.88 5.81 3.05 3.33 2.60 3.80

3.67 5.44 4.75 6.07 4.86 6.27 5.04 6.50 8.02 7.18 5.24 2.75 3.29 2.65 2.97

6.14 8.48 7.16 9.62 1.76 5.59 0.82 6.20 4.73 4.20 0.76 0.54 1.74 3.58 2.86

4.84 6.42 8.02 6.78 0.22 1.26 3.01 2.39 2.82 1.55 3.27 0.91 3.42 3.65 3.78

3.97 5.41 5.52 5.50 3.73 6.68 3.98 6.76 8.09 6.56 3.64 1.88 2.03 0.64 1.40

3.32 4.13 4.25 3.98 3.25 4.33 1.97 4.30 6.06 4.70 2.35 1.57 1.27 0.15 0.58

3.34 4.15 4.50 4.95 3.92 7.61 4.80 6.79 8.17 6.27 4.23 2.17 1.44 0.35 1.07

6.71 9.00 6.54 9.19 3.68 6.57 2.37 6.36 10.95 6.29 2.65 1.30 0.46 1.90 1.21

3.78 4.74 4.63 5.11 2.73 4.41 1.90 4.02 5.72 4.16 2.04 1.02 0.87 0.04 0.64

4.65 5.87 5.27 5.83 3.43 6.11 3.64 6.16 7.53 6.27 3.72 1.80 1.12 0.90 0.25

314

Chapter 9

(6.71). Likewise, the resolution of benzene and anisole with methanol from Group I1 (1.OO) most closely resembled those obtained with dimethylsulfoxide (DMSO) in Group I11 (0.91) and dioxane in Group V I (1.02). Numerous other examples are obvious in Table 9.13. The discrepancies in solvent selectivity within solvent groups were not limited to resolution from benzene, but rather were universal for all of the compounds studied. Other examples of extreme variation are shown in Table 9. I4 with all solvent groups for the resolution of nitrobenzene and anisole from benzoyl chloride and for the resolution of three positional isomers of nitrotoluene. Methanol was the only Group I1 solvent and DMSO was the only Group I11 solvent providing resolution greater than baseline (i.e. R, 2 1.5) for nitrobenzene and benzoyl chloride. In Group 11, only methanol failed to completely resolve anisole from benzoyl chloride, while only three of the Group I11 solvents and one of the Group VI solvents provided baseline resolution. For the resolution of 0- and m-nitrotoluene, methanol was the only Group I1 solvent and DMF and 2ethoxyethanol were the only Group 111 solvents providing complete separation. In Group 111, the resolution of nitrobenzene and benzoyl chloride ranged from 0.2 to 3.2, and the resolution of anisole and benzoyl chloride ranged from 0.1 to 2.1. 9.3.2.3 Prediction of resolution

As indicated in Section 9.3.1.3. for steroids, the ability of a solvent to resolve a given pair of compounds can be predicted using Z.(minimum) values and Eq. (9.7). Regression coefficients for the slope (3) and y-axis intercept (y) are contained in Table 9.15 for the values of @ given in Table 9.1. A positive value for indicates that the retention index of the second component is greater than that required for baseline resolution, while a negative Z, indicates that resolution cannot be achieved. This approach for predicting resolution is demonstrated to also provide accurate predictions for benzene derivatives in Table 9.16 by using the previous example of nitrobenzene and benzoyl chloride in which the selectivity triangle concept failed to predict differences in solvent selectivity. From Table 9.12, the retention indices of nitrobenzene and benzoyl chloride with methanol at @ = 0.589 are 635.0 and 687.1, respectively. From Table 9.15, the minimum retention index necessary for complete resolution from nitrobenzene with methanol as the strong solvent is calculated as:

Z.(minimum) = 0.8581(635.0) + 127.9 = 672.8

(9.10)

Z, thus equals 687.1 - 672.8 = 14.3, indicating that the retention index of benzoyl chloride is 14.3 units greater than the minimum required for baseline resolution from nitrobenzene. An experimental resolution (R,)of 2.5 confirmed that complete resolution was easily obtained. In like manner, this technique correctly predicted that DMSO was the only other solvent capable of completely resolving this pair of solutes (Table 9.15). 9.3.2.4HPLC resolution as afunction of retention index diferences A common misconception with retention index systems in both GC and HPLC is that two

liquid phases producing similar retention indices or retention index differences will result in a similar resolution of the mixture of compounds [38 and references therein]. Without

b

8 m Ti

0 3

2

3cu

TABLE 9.14 VARIATION OF EXPERIMENTAL RESOLUTION OF AROMATIC COMPOUNDS WITH SOLVENTS CLASSIFIED ACCORDING TO THE SOLVENT SELECTIVITY TRIANGLE

m

z 0,

$Compounds

cu

Y cu cu


Group II

Group 111

i2.

Group VI

MeOH

EtOH

1-PrOH

2-PrOH

DMF

DMSO

2-EE

2-ME

THF

ACN

Diox

2-MEA

2.5 1.4 1.8 0.9 0.9

1.0 2.2 1.3 0.5 0.9

0.5 3.0 0.8 0.3 0.5

0.4 2.6 0.9 0.4 0.5

0.9 0.1 1.7 1.1 0.5

3.2 1.5 0.6

0.2 1.8 1.5 0.9 0.8

1.o 1.O 1.2 0.8 0.5

0.8 2.1 1.1 0.5 0.7

1.0 1.3 1.2 0.6 0.6

1.3 1.2

1.o 0.3 0.7

0.3 2.0 1.0 0.3 0.7

0.5

0.1

d. Q

TABLE 9.15 27

(MINIMUM) VALUES FOR 2-KETO ALKANES WITH 12 DIFFERENT RP-HPLC SOLVENTS IN AQUEOUS BINARY MOBILE PHASES

3 4 5 6 7 8 9 10 11

I1

MeOH

EtOH

I-PrOH

2-PrOH

DMF

DMSO

2-EE

2-ME

300 400 500 600 700 800 900 1000 1100

560.5 638.6 726.3 817.3 -

570.4 641.8 730.0 820.6 914.4 -

583.3 656.9 749.5 836.2 927.0 1028.7 1122.8

681.0 757.9 843.3 931.3 1027.1 1121.6

493.9 560.8 630.4 722.2 -

-

380.7 556.9 64 1.6 726.0 819.4 915.5 -

485.5 566.1 642.2 724.8 821.1 -

-

474.3 542.6 628.4 -

-

565.2 652.0 745.4 834.2 926.6 1014.9 -

0.8581 127.9

0.8668 129.4

0.9070 117.9

0.8853 141.2

0.8845 129.9

0.8114 141.4

0.8950 105.4

0.8399 142.0

0.9994

0.9987

0.9992

0.9992

0.9982

0.9984

0.9995

0.9987

Slope (s) Intercept (y) Correlation coeff. (2)

aN = number of carbon atoms in the 2-keto alkane standards.

-

-

THF

ACN

Diox

2-MEA

-

474.0 565.7 639.3 728.0 815.7 -

475.2 567.3 649.9 729.6 816.1 -

-

-

471.8 553.8 640.1 730.0 818.7 914.2 -

0.9032 112.3

0.8457 137.8

0.8411 141.9

0.8847 113.0

0.9999

0.9995

0.9997

0.9997

-

-

W

VI

316

Chapter 9

TABLE 9.16 PREDICTION OF RESOLUTION OF BENZOYL CHLORIDE AND NITROBENZENE FROM 12 (MINIMUM) CALCULATIONS Solvent group

Solvent

Retention index

12

-NO2

-cOCI

(minimum)

Id

Resolution

I1

MeOH EtOH 1 -PrOH 2-PrOH

635.0 654.0 675.1 680.0

687.1 681.9 693.0 696.4

672.8 689.1 707.2 711.4

+14.3 -7.2 -14.2 -15.0

2.5 1.0 0.5 0.4

111

DMF DMSO 2-EE 2-ME THF

601.0 546.2 628.0 598.7 705.2

620.7 616.0 634.1 627.8 678.3

643.6 596.6 666.8 641.6 709.9

-22.9 +19.4 -32.7 -13.8 -4.7

0.9 3.2 0.2 1 .o 0.8

VI

ACN Diox 2-MEA

653.6 629.6 664.6

673.5 661.4 672.5

688.8 665.8 698.2

-15.3 -4.4 -25.7

1 .o 1.3 0.3

first relating the retention indices to resolution, however, there is no correlation between the two. This lack of correlation is evident in Table 9.17, which compares differences in retention indices with the experimental resolution (R,) obtained for chlorobenzene and bromobenzene using all 12 of the RP-HPLC solvents. Again, R, varies considerably within each of the solvent selectivity groups even when the difference in retention indices (30 between bromobenzene and chlorobenzene are very similar. For example, d l for these two compounds are 3 1.5, 29.8, and 30.4 for the Group I1 solvents methanol, ethanol, and 1-propanoland 26.9 for the Group VI solvent acetonitrile. Thus, it would appear that these four solvents should yield similar separations of this compound pair; however, the experimental R, values ranged from 1.5 to 2.7 for these four solvents. The reason for the lack of correlation between d l and R, is that the solvents yield varying degrees of resolution for the 2-keto alkane standards that provide the reference points for calculating retention indices. For example, methanol separates the 2-keto alkanes to a greater degree than does l-propanol. Thus, when the retention time difference between the peak maxima of the adjacent 2-keto alkanes is divided into 100 retention index units (by definition), each retention index unit for methanol represents a greater retention time than does each retention index unit for 1-propanol. In the above example with chlorobenzene and bromobenzene, each retention index unit represented 8.74 s for methanol and 4.28 s for 1 -propanol. Consequently, the 31 of 3 1.5 units for methanol represented a difference of 275 s between the peak maxima of chlorobenzene and bromobenzene, whereas the dI of 30.4 for 1-propanol represented a difference of 130 s. The varying degree to which the solvents separate the retention index reference compounds is represented by the slope of the line (SL) for the plot of logarithm of the adjusted retention time versus the carbon number of the 2-keto alkanes in RP-HPLC. The slopes for each solvent are summarized in Table 9.18. By dividing the slope for each sol-

Solvent selectivity

317

TABLE 9. I 7 PREDICTION OF RESOLUTION OF CHLOROBENZENE AND BROMOBENZENE FROM RELATIVE SLOPES (RSL) AND RELATIVE RETENTION INDEX DIFFERENCES (Rar) Solvent

Solvent

Retention index

a1

4

-Br

RSL

Resolution Experimental

Predicted

I1

MeOH EtOH 1 -PrOH 2-PrOH

812.7 871.5 979.8 968.5

844.2 901.3 1010.2 1005.6

31.5 29.8 30.4 37.1

1.000 0.874 0.639 0.593

2.7 2.5 1.5 1.7

2.7 2.2 1.7 1.9

I11

DMF DMSO 2-EE 2-ME THF

718.4 637.9 809.2 768.1 885.2

736.7 649.7 828.0 784.6 900.2

18.3 11.8 18.8 16.5 15.0

1.044 0.975 0.835 0.826 0.703

1.5 0.8 1.3 1.0 0.7

1.6 1 .o 1.3 1.2 0.9

I11

ACN Diox 2-MEA

813.7 801.4 833.7

840.6 816.9 853.1

26.9 15.5 19.4

0.894 0.961 0.875

2.2 1.2 1.5

2.1 1.3 1.5

vent (SL,) by the slope for methanol (SL,,,), a relative slope (RSL) is obtained that corrects for the varying separation of the 2-keto alkanes: (9.1 1)

RSL = SL,/SL,

Relative slopes for the 12 solvents are presented in Table 9.18. These relative slopes can then be multiplied by relative aZ values (RaZ) and the experimental resolution obtained with methanol (R,,,) to accurately calculate the corresponding resolution with any of the remaining RP solvents (Rx): R,=RSL x RaIX R,,

(9.12)

where RaI = aZ with solvent x divided by a Z with methanol. For example, aZ for bromobenzene and chlorobenzene is 3 1.5 with methanol and 26.9 with acetonitrile (Table 9.17), and RSL = 0.894 for acetonitrile (Table 9.18). The resolution of this compound pair was determined to be 2.7 using methanol, and the resolution with acetonitrile was thus predicted as R,

= 0.894

X

(26.9131.5) X 2.7

= 2.1

(9.13)

The experimental resolution (R,) with acetonitrile was determined to be 2.2. Likewise, R, was accurately predicted for each of the solvents, as shown in Table 9.17. Equation (9.13) accurately predicted the varying R, values that were observed within the selectivity triangle groups even in cases where uncorrected a Z values did not suggest that resolution differences should occur. Referencespp. 335-336

318

Chapter 9

TABLE 9.18 SLOPES (SLx)AND RELATIVE SLOPES (RSL) FOR THE 2-KETO ALKANE RETENTION INDEX SCALE WITH 12 DIFFERENT RP-HPLC SOLVENTS IN AQUEOUS BINARY MOBILE PHASES

Group

Solvent

SLX

RSL

I1

MeOH EtOH I -PrOH 2-PrOH

0.2799 0.2447 0.1789 0.1660

1.000 0.874 0.639 0.593

I11

DMF DMSO 2-EE 2-ME THF

0.2923 0.2728 0.2338 0.231 1 0.1968

1.044 0.975 0.835 0.826 0.703

IV

ACN Diox 2-MEA

0.2502 0.2691 0.2449

0.894 0.961 0.875

It is obvious fiom Eq. (9.12) that two solvents resulting in a similar aZ for two compounds will yield similar resolution only if the RSL values are approximately equal. For example, bromobenzene and toluene have retention indices of 844.2 and 819.1 with methanol (dZ= 25.1) and 840.6 and 815.0 with acetonitrile (aZ= 25.6). The ML for acetonitrile and methanol are 0.894 and 1.000, respectively, and an experimental R, of 2.2 was determined for methanol. From Eq. (9.12), an R, of 2.1 is calculated for acetonitrile. An experimental resolution of 2.1 confirmed that these two solvents would give a similar resolution of this compound pair. 9.3.2.5Preadjustment of retention indicesfor prediction of resolution

Following the relationship described in Eq. (9.12), it is possible to preadjust retention indices to compensate for standard slope differences. In this approach, adjusted retention indices (1’)are calculated by multiplying the retention index ( I ) obtained with a given solvent by the relative slope (RSL) for that solvent: Z’=ZX

RSL

(9.14)

Tables of I’ values can thus be used to directly compare selectivity differences that occur with the various solvents. The I’ values calculated from the Z values in Table 9.12 and the RSL values in Table 9.18 are summarized in Table 9.19. The utility of this approach can be demonstrated by considering again the resolution of chlorobenzene and bromobenzene. Differences in adjusted retention indices (31’)can be calculated for chlorobenzene and bromobenzene fiom Table 9.19. The resulting aZ‘ values are summarized for each solvent in Table 9.20. The relative 31’ values ( R W )can then be multiplied by the resolution obtained with methanol (R,) to predict the resolution with each of the other solvents (R,):

TABLE 9.19 ADJUSTED RETENTION INDICES (1')CORRECTED FOR 2-KETO ALKANE SLOPE DIFFERENCES RELATIVE TO METHANOL ~

Compound

Group I1 MeOH

Group 111 EtOH

1-PrOH

2-PrOH

DMF

Group VI DMSO

2-EE

2-ME ~

-H 4 3 1 -Br -CH3 -CH=CH;? -NO2 -COCH3 -COOCH3 -CHO 4 H -CN -cOCI 4CH3 0-NT m-NT P-NT

714.7 812.7 844.2 819.1 849.4 635.0 579.6 687.6 553.2 476.8 552.3 687.1 699.5 710.8 736.8 723.8

768.7 761.7 787.7 772.9 793.5 571.6 496.5 592.6 485.9 431.8 486.9 596.0 642.7 640.5 664.6 649.6

556.1 626.1 645.5 635.3 654.3 43 1.4 370.2 442.1 365.5 278.0 366.0 442.8 501.6 490.2 506.5 496.1

516.3 574.3 596.3 591.6 606.0 403.2 341.0 411.6 336.3 299.4 333.6 413.0 465.1 451.7 468.9 459.0

661.9 750.0 769.1 757.3 764.0 627.4 554.8 645.8 531.1 536.8 554.5 648.0 649.9 690.9 721.4 711.9

537.4 622.0 633.5 650.2 648.4 532.5 505.0 600.3 482.5 448.4 492.9 600.6 562.3 598.7 610.6 608.4

613.5 675.7 691.4 701.0 704.2 524.4 428.7 526.5 432.6 410.7 435.0 529.5 573.6 573.6 601.5 588.6

576.1 634.5 648.1 652.3 650.2 494.5 426.0 5 17.9 413.0 356.8 416.1 518.6 540.9 547.2 572.7 563.7

THF ~~

570.8 622.3 632.8 640.6 645.6 495.8 410.3 475.9 418.8 420.5 436.1 476.8 531.1 544.7 564.8 553.1

ACN

Diox

2-MEA

694.3 770.1 785.0 784.5 791.3 602.5 530.8 638.1 526.4 453.4 538.7 635.6 666.4 671.4 693.3 678.0

662.6 729.5 746.5 742.1 750.1 581.5 498.3 588.4 502.0 437.3 521.2 588.4 629.3 641.3 658.8 646.8

~

644.0 727.4 751.5 728.6 747.9 584.3 517.7 602.6 507.6 412.0 530.9 602.1 624.3 65 1.4 670.9 661.6

W CL

W

320

R,

Chapter 9

= RaZ’ X

R,

(9.15)

where RaZ’ is aZ’ for a given solvent divided by aZ‘ for methanol. R, is accurately predicted for each of the other solvents (Table 9.20) by using the experimentally determined R, of 2.7. A second example that is also contained in Table 9.20 demonstrates the accurate prediction of resolution for chlorobenzene and toluene. The advantage of preadjusting retention indices in this manner is that selectivity differences can be seen at a glance, because selectivity differences correlate directly with I’ differences (a]’). I’ values can thus be used to characterize various solvents according to their actual selectivity for effecting given separations. In addition, the advantage of using retention indices (rather than capacity factors, retention times, or relative retention times) in Eqs. (9.10), (9.12) and (9.15) is that retention indices offer improved interlaboratory reproducibility [ 151. Thus, experimental, published, or predicted retention indices can be used to predict or optimize resolution.

9.3.3 Reasons for failure of the solvent selectivity triangle

The solvent selectivity triangle concept fails to group HPLC solvents according to their resolving properties because it is constructed using data that do not correlate with resolution. The data used to construct the triangle represent retention of each probe molecule as a fraction of the summed retentions of the three probe molecules. However, it is clear from Eq. (9.3) that resolution is related to differences in retention rather than normalized fractions of retention. In fact, chromatographic selectivity is defined as the difference in retention of the peak maxima of two components that are chromatographed together (39) (i.e. t2 - tl). Selectivity combines with chromatographic efficiency (wl+ w2)to define resolution. The use of fractions of summed retentions actually serves to hide differences in selectivity by masking absolute differences in retention units. This masking of selectivity differences can be demonstrated using a previous example involving the resolution of benzoyl chloride, nitrobenzene, and anisole. The retention indices (4 of the three components were summed, and the fraction of the sum represented by each of the components is given in Table 9.2 1 for several of the RPHPLC solvents. From their identical fractions in Table 9.2 1, it would appear that ethanol, 2-ethoxyethanol, 2-methoxyethanol, acetonitrile, dioxane, and 2-methoxyethyl acetate would all result in the same selectivity for resolving this compound mixture; however, the experimental resolution ranged from 0.2 to I .3 for benzoyl chloride versus nitrobenzene and from 1 .O to 2.2 for benzoyl chloride versus anisole (Table 9.21). In contrast, a comparison of differences in adjusted retention indices (X) clearly shows the selectivity differences between these solvents, with aZ’ values correlating well with the observed differences in experimental resolution. Using the aZ’ values with Eq. (9.15) also resulted in the accurate prediction of resolution (Table 9.22). In the triangle approach, the differing magnitudes of the aZ’ values (which are a direct measure of selectivity) are obscured by normalizing retentions to a fraction of unity. It is apparent, then, that usage of the solvent selectivity triangle as a rationale for

!a

8 z 0 3

E

TABLE 9.20 PREDICTION OF THE RESOLUTION OF CHLOROBENZENE VERSUS BROMOBENZENE AND CHLOROBENZENE VERSUS TOLUENE WITH ADJUSTED RETENTION INDICES (I? ~

Group

11

111

IV

Solvent

ai'

I' 4 1

-Br

MeOH EtOH 1-PrOH 2-PrOH

812.7 761.7 626.1 574.3

844.2 787.7 645.5 596.3

DMF DMSO 2-EE 2-ME THF

750.0 622.0 675.7 634.5 622.3

ACN Diox 2-MEA

727.4 770.1 729.5

Resolution

a'

I'

Predicted

Experimental

6.4 11.2 9.2 17.3

0.5

0.5

0.9 0.7 1.3

1.0 0.7 1.2

757.3 650.2 701.0 652.3 640.6

7.3 28.2 25.3 17.8 18.3

0.6 2.2 2.0 1.4 1.4

0.6 2.2 2.0 1.3 1.2

728.6 784.5 742.1

1.2 14.4 12.6

0.1 1.1 1.o

0.1 1.2 1.o

Predicted

Experimental

4 1

4 3 3

31.5 26.0 19.4 22.0

2.7 2.2 1.7 1.9

2.7 2.5 1.5 1.7

812.7 761.7 626.1 574.3

819.1 772.9 635.3 591.6

769.1 633.5 691.4 648.1 632.8

19.0 11.5 15.7 13.6 10.5

1.6 1.o 1.3 1.2 0.9

1.5 0.8 1.3 0.7

750.0 622.0 675.7 634.5 622.3

751.5 785.0 746.5

24.1 14.9 17.0

2.1 1.3 1.5

2.2 1.2 1.5

727.4 770.1 729.5

1.o

~~

Resolution

W

c!

322

Chapter 9

TABLE 9.21 FAILURE OF NORMALIZED RETENTION INDEX FRACTIONS TO CHARACTERIZE SOLVENT SELECTIVITY DIFFERENCES Solvent

EtOH 2-EE 2-ME ACN Diox 2-MEA

Retention fraction

Resolution

-cOCI

-NO2

-OCH3

0.33 0.33 0.33 0.33 0.33 0.33

0.32 0.32 0.32 0.32 0.32 0.32

0.35 0.35 0.35 0.35 0.35 0.35

-COCI VS. -NO2

-COCI VS.4 C H 3

1.o 0.2 1.o 1.o 1.3 0.3

2.2 1.8 1.o 1.3 1.2 2.0

HPLC solvent selection is an unreliable approach. The data obtained with aromatic compounds support the data reported for steroids [21] and polystyrene oligomers [33] in which experimental resolution failed to correlate with the solvent groupings in the selectivity triangle. In the latter study [33], the triangle groupings failed with normal phase as well as Rp solvents. A common usage of the triangle has been to justirjl the selection of certain solvents as being typical of a given group so that all of the solvents in the group need not be studied. For example, methanol, THF, and acetonitrile were frequently selected as being representative of Rp solvents from Groups 11, 111, and VI, respectively [28-321. Numerous examples, however, can be seen in the tables where other solvents in these groups would give much better separations of a given mixture of compounds. Extreme examples even exist, as in the case of the resolution of benzene from p-nitrotoluene, where baseline resolution TABLE 9.22 PREDICTION OF SOLVENT SELECTIVITY DIFFERENCES FROM ADJUSTED RETENTION INDEX DIFFERENCES@l', Compounds

Solvent

ar'

Experimental

Predicteda

-COCI VS. -NO2

EtOH 2EE 2-ME ACN Diox 2-MEA

24.4 5.1 24.1 17.8 33.1 6.9

1.0 0.2 1.o 1.o 1.3 0.3

1.2 0.2 1.2 0.9 1.6 0.3

-COC1

EtOH 2-EE 2-ME ACN Diox 2-MEA

46.7 46.8 22.3 22.2 30.8 40.9

2.2 1.8 1.o 1.3 1.2 2.0

2.3 2.3 1.1 1.1 1.5 2.0

VS. 4 C H 3

aPredicted from Eq. (9.15); R, ride versus anisole.

= 2.5

for benzoyl chloride versus nitrobenzene and R, = 0.6 for benzoyl chlo-

323

Solvent selectivity

could not be attained with any of these three solvents, but baseline resolution could easily be attained with several of the other solvents (Table 9.13). A similar example exists with the separation of chlorobenzene and toluene, where only dimethylsulfoxide and 2ethoxyethanol provide resolution that is greater than baseline (Table 9.20). While studies that used the triangle are by no means invalidated, usage of the triangle as a basis for solvent selection can result in overlooking many common solvents that exhibit the required selectivity for a given separation.

9.3.4 Extension of theory to gas chromatography 9.3.4.I Prediction of resolution

In GC, the Kovats retention index system is based upon the linear relationship between the logarithm of the adjusted retention time and the carbon number of n-alkane standards. In a manner analogous to the relationship described by Eq. (9.11) for HPLC liquid phases, the slope of the line obtained with a given GC liquid phase (SL,) can be related to that obtained with squalane (SL,) as follows: RSL = SLJSL,

(9.16)

where RSL again is the relative slope. The experimental resolution on a given stationary phase (R,) can be predicted using an equation analogous to Eq. (9.12): R,

= RSL X

RaI

X

(9.17)

R,

where R, is the experimental resolution determined on squalane. The utility of this relationship is demonstrated in Table 9.23 in which the experimental resolution of p methylanisole and o-bromotoluene is accurately predicted on OV-10 1, OV-7, OV-25, and OV-225 from an experimental R, of 3.4 on squalane.

TABLE 9.23 PREDICTION OF THE RESOLUTION OFp-METHYLANISOLE (PMA) AND O-BROMOTOLUENE(OBT) WITH SEVERAL GC STATIONARY PHASES GC stationary phase Squalane ov-101 OV-7 OV- 17 OV-25 OV-225

Retention index PMA

OBT

980 1002 1077 1161 1241 1297

1019 1025 1094 1173 1255 1284

aPredicted from Eq. (9.17); R, = 3.4.

References pp. 335-336

ar 39 23 17 12 14 13

RSL

1.000 0.859 0.889 0.882 0.840 0.787

Resolution Experimental

Predicteda

3.4 1.6 1.5 1.1 1.0 0.9

3.4 1.7 1.3 0.9 1.o 1.o

324

Chapter 9

9.3.4.2McReynolds constants and resolution

In GC, differences in retention indices obtained on different liquid stationary phases (4 have also been misinterpreted for many years. Several papers have appeared that utilize comparisons of McReynolds AZ constants [40] to advocate similarities and dissimilarities between various stationary phases [38 and references therein]. However, without first relating the AZ values to relative slope values (RSL), the correlations are based on data that have not been adjusted for varying degrees of separation of the n-alkane retention index standards. Corrected AZ values (AZ’) can be calculated relative to squalane as follows: A Z ’ = A Z X RSL

(9.18)

The slope of the line (b) is given for each liquid phase in the McReynolds publication [19] so that AZ’ values are easily obtained. The effect of this adjustment upon McReynolds constants is demonstrated in Table 9.24. Obviously, the adjustment produces only minor changes for stationary phases that have a slope similar to that of squalane so that RSL is close to 1.000 (e.g. Apiezon M). The effect is substantial, however, for those stationary phases that have an RSL that differs greatly from 1.OOO (e.g. DEGS and Carbowax 20M). In many cases, stationary phases that appear to be similar for retaining a given probe molecule based upon AZ values are shown to yield different retentions when AZ’values are compared (Table 9.25). For example, the McReynolds constant (AZ) for 2-pentanone is listed as 560 on diglycerol and 558 on DEGS Supelco 1045 [40], thereby suggesting that these two stationary phases would retain 2-pentanone to a similar degree; but the adjusted McReynolds constant (AZ’) is 498 for diglycerol and 384 for DEGS Supelco 1045, indicating that diglycerol actually retains 2-penantone more strongly than does the DEGS phase. Several other examples of this nature are given in Table 9.25. Anomalies caused by the relative nature of the McReynolds constants have been alluded to previously [4 1,421. It has been suggested that 2-keto alkanes replace n-alkanes as the standards in GC so that the McReynolds constants can be determined more accurately on polar GC stationary phases [42]. While this suggestion appears to have merit for the reasons discussed by Kersten et al. [42], it also poses some major drawbacks. Since the development of the Kovats retention index system in 1958, tremendous volumes of retention index data using the n-alkane standards have been generated, and they would not correlate with data using the new standards. Also, this approach would require a very substantial undertaking to reevaluate McReynolds constants with different standards. The use of Eq. (9.18) to compensate for the slope differences that occur with the various phases appears to be a workable alternative to switching to 2-keto alkanes for calculating GC retention indices. This approach offers several advantages. First, the n-alkane retention index system is retained. Second, the existing McReynolds constants only require adjustment rather than determination with a new set of standards. Finally, retention indices adjusted in this manner can be used directly for the prediction of resolution by Eq. (9.17).

m

m 4

3cu

F a. s.

cu u

Q

L.bl o\

TABLE 9.24 COMPARISON OF REPRESENTATIVE McREYNOLDS CONSTANTS (AZ)WITH ADJUSTED McREYNOLDS CONSTANTS (AI’)CORRECTED FOR nALKANE SLOPE DIFFERENCES RELATIVE TO SQUALANE ~~~

~

Gc

RSLa

Benzene

Butanol

2-Pentanone

~~

Nitropropane

Pyridine

stationary phase

AI ~

AI‘

AI

AI’

AI

AI’

~~~

Squalane Apiezon M ov-1 OV-17 o v - 2 10 Carbowax 20M Reoplex 400 EGSS-X DEGS

AI

AI’

~~

1.000 0.980 0.854 0.882 0.722 0.773 0.737 0.660 0.666

0

31 16 119 146 322 364 484 492

0 30 14 105 105 249 268 319 328

0 22 55 158 238 536 619 710 733

0 22 47 139 172 414 456 469 488

aRelative slope (calculated relative to squalane using the slopes in ref. [40]).

0 15 44 162 358 368 449 585 581

0 15 38 143 258 284 33 1 386 387

0 30 65 243 468 572 647 83 1 833

AI

Az’

0 40 42 202 310 510 67 1 778 79 1

0 39 36 178 224 394 495 513 527

~

0 29 56 214 338 442 477 548 555

326

Chapter 9

TABLE 9.25 EXAMPLES OF ADJUSTED McREYNOLDS CONSTANTS (Alf)REVEALING DIFFERENT STATIONARY PHASE SELECTIVITIES THAN THOSE SUGGESTED BY NON-ADJUSTED McREYNOLDS CONSTANTS Standard probe

Stationary phase

RSLa

Al

Alf

2-Pentanone

Diglycerol DEGS Supelco 1045

0.889 0.689

560 558

498 384

1,4-Dioxane

LSX-3-0295 Phonic P85

0.720 0.983

29 1 285

210 280

2-Pentanone

ov-11 Dinonyl phthalate

0.887 0.970

145 147

129 142

Nitropropane

OV-17 Hallcornid M-180L

0.882 0.984

243 239

214 235

aRelative slope (calculated relative to squalane using slopes in ref [40]).

9.3.4.3Application of the selectivity triangle to the characterization of GC stationary phase selectivity

The solvent selectivity triangle concept has been extended to the characterization of GC stationary (liquid) phase selectivity [4345]. In this approach, retention indices or AI values for three of the solute probes that were utilized by Rohrschneider [46] or McReynolds [40] were selected as being representative of various molecular interactions that a solute can undergo with the stationary phase. The individual retentions of the three probe molecules were then expressed as fractions of the total of the three retentions, and the resulting fractions were plotted in a triangular diagram to define similarities and dissimilarities in stationary phase selectivities. The selectivity triangle approach for characterizing GC liquid phase selectivity suffers from the same flaw as occurs with this approach in normal-phase HPLC or RP-HPLC in that selectivity differences are masked by expressing selectivity as normalized fractions of summed retentions rather than as absolute retention differences. The masking of selectivity differences is clearly demonstrated by the two examples in Table 9.26. In this table, McReynolds Al constants for butanol (MI), nitropropane (AZ2),and dioxane (AZ3) are summed, and each AI value is expressed as a fraction (XI,X2 and X3) of the sum of the three AZ values. Using this approach, Apiezon J and OV-25 are characterized as having similar selectivity due to similarities in the fractional values (XI, X2, and X3);however, absolute differences in the AI values for the solute probes with the two stationary phases show substantial differences in the actual separating capacity of Apiezon J and OV-25. For example, butanol and nitropropane are obviously separated to a far greater extent on OV-25 (with an absolute AZ difference of 305-204 = 101 units) than on Apiezon J (with an absolute AI difference of only 49-36 = 13 units), yet the triangle approach classifies them as having similar selectivity. In the second example in Table 9.26, the triangle approach suggests that OV-101 and Carbowax 20M have similar selectivity due to similar fractions of retention; however,

Solvent selectivity

327

TABLE 9.26 COMPARISON OF THE CHARACTERIZATION OF GC STATIONARY PHASE SELECTIVITY BY THE TRIANGLE APPROACH VERSUS ABSOLUTE RETENTION DIFFERENCES

XI

Stationary phase McReynolds constanta

Apiezon J OV-25 ov-101 Carbowax20M a&l

AI1

AI2

Al3

Total

36 204 57 536

49 305 67 572

42 251 46 434

127 760 170 1542

X2

X3

&difference 1-2

0.28 0.27 0.34 0.35

0.39 0.40 0.39 0.37

0.33 0.33 0.27 0.28

13 101 10 36

1-3 6 47 11

102

2-3 7 54 21 138

is butanol, N2is nitropropane, A l 3 is dioxane. Values from ref. [40].

absolute differences in AZ clearly show that the probe molecules are separated to a far greater extent on Carbowax 20M. For example, the absolute difference in AZ values for butanol and dioxane is 11 units on OV-101, but it is 102 units on Carbowax 20M. Even though the fractions of AZ values are similar, the absolute retention differences are not. When the actual retention indices of these McReynolds solute probes are adjusted to compensate for the relative slope differences of the n-alkane standard calibration line via Eq. (9.14), a direct measure of selectivity can be obtained for these stationary phases. As shown by the absolute I' differences in Table 9.27, this approach confirms that Apiezon J has a different selectivity than OV-25 and that OV-101 has a different selectivity than Carbowax 20M. It should be possible to characterize GC stationary phase selectivity using adjusted McReynolds retention indices or adjusted McReynolds AZ values in a nearest neighbor technique [47] such that absolute differences in adjusted retentions are compared. This approach would correctly characterize selectivity as absolute retention differences using data that are adjusted for the varying slopes of the n-alkane retention index standards.

9.3.5 Characterization of RP-HPLC selectivity with adjusted retention indices 9.3.5.I Calculation of adjusted retention indices

Due to the nature of the retention index system, retention indices are relative rather than absolute values (i.e. retention indices are expressed relative to 2-keto alkane standards). Because the various RP solvents separate the 2-keto alkane standards to differing degrees, it is necessary to compensate for this difference in order to normalize the retention index scale to remove this anomaly before comparing retention index differences that occur with different RF' solvents [24]. As noted in Section 9.3.2.4, the varying degree to which the solvents separate the retention index standards can be measured by the slope of the line (SL) in which the logarithm of the adjusted retention time was regressed against the carbon number of the 2-keto alkane standards. By dividing the slope obtained with each RP solvent (SLx)by the slope obtained with methanol (SL,), a relative slope (RSL) can be obtained that adjusts for the varying separation of the 2-keto alkanes and normalizes the retention index scale (Eq. 9.1 1). References pp. 335-336

W

N 00

TABLE 9.27 EXAMPLES OF THE CHARACTERIZATION OF GC STATIONARY PHASE SELECTIVITY USING ADJUSTED RETENTION INDICES (1')FOR THREE McREYNOLDS PROBE SOLUTES

GC stationary phase Retention indexa

Apiezon J OV-25 ov-101 Carbowax20M

RSL

11

12

13

626 794 647 1126

70 1 957 719 1224

696 905 700 1088

0.980 0.840 0.859 0.773

Adjusted retention indexb

Absolute I' differences

1'1

f 2

f3

1 vs. 2

1 vs. 3

2 vs. 3

613 667 556 870

687 804 618 946

682 760 60 I 841

74 137 62 76

69 93 45 29

5 44 17 105

Valculated fiom ref. [40]. I1 is the retention index for butanol, 12 is the retention index for nitropropane, and I3 is the retention index for dioxane

9' = Ix RSL.

Solvent selectivify

329

The slopes and the relative slopes for the 12 RP solvents are presented in Table 9.18. Methanol, by definition, had a relative slope of 1.000, and the RSL values for the other solvents ranged from 0.593 for 2-propanol to 1.044 for dimethylformamide. The relative slope values were then used to normalize retention indices relative to those obtained with methanol as the strong solvent. Adjusted retention indices (I’) for a given solute with a given RP solvent were calculated by multiplying the retention index (Z) obtained with the solvent by the relative slope for that solvent (Eq. 9.14). The resulting I’ values are summarized in Table 9.19. Thus, the adjusted retention indices in Table 9.19 have been normalized to remove the anomaly associated with the relative nature of the retention index scale, and the I‘ values for the different solvents directly reflect selectivity differences among the solvents. 9.3.5.2Probes for characterization of RP-HPLC solvent selectivity.

Because the I’ values in Table 9.19 directly reflect solvent selectivity differences, it was possible to characterize solvent selectivity with the adjusted retention indices of representative solute “probes”. A multiple linear regression analysis [48,49] was conducted by regressing adjusted retention indices of the 16 solutes in the 12 solvents against those of a selected set of compounds thought to be representative of the 16 solutes. This approach was similar to Rohrschneider’s method for characterizing GC liquid phases [50]. The regression equation was I’

= ax

+ by+ cz + i

(9.19)

where x , y , and z are the adjusted retention indices (Z’) of three selected probe molecules (nitrobenzene, benzaldehyde, and anisole, respectively), a, b, and c are the respective regression coefficients, and i is the “intercept” of the regression line. The coefficient of determination (similar to the square of the correlation coefficient in simple linear regression) is summarized for each of the 13 aromatic compounds in Table 9.28. A value of R2 which is nearly equal to 1 indicates that the three probes accurately explain (predict) the response of the solutes in the 12 RP-HPLC solvents. The substitution or addition of other solutes into the equation failed to significantly improve the coefficient of determination. However, from a theoretical standpoint, it would seem that phenol would likely serve as a more representative probe if additional solutes that undergo substantial hydrogen bonding were included in the group of solutes studied. The relatively low coefficient of determination for phenol (0.9048) also tends to suggest that this might be the case. Characterization of solvent selectivity with three selected probe values (x, y , and z) is shown in Table 9.29. In GC, the sum of the McReynolds constants is taken as a measure of liquid phase polarity [40]. Similarly, the sum of the I’ values for these three probe molecules can be taken as a measure of polarity of the 12 RP-HPLC solvents. The polarities for the solvents as measured by this technique ranged from 1204.6 for 2-propanol to 1887.7 for methanol under the conditions studied. The regression coefficients (a, b, and c) and the i values (Table 9.30) can be used in combination with the solvent-specific coefficients (x, y , and z ) to predict the adjusted retention index of any of the solutes with any of the solvents. For example, the adjusted References pp. 335-336

Chapter 9

330

TABLE 9.28 COEFFICIENTS OF DETERMINATION (R2) FOR AROMATIC SOLUTES FROM REGRESSING ADJUSTED RETENTION INDICES (Z’) VERSUS ADJUSTED RETENTION INDICES FOR THREE PROBE MOLECULES Solute

Coefficient of determinationa

(R2) Benzene Chlorobenzene Bromobenzene Toluene Styrene Acetophenone Methyl benzoate Phenol Benzonitrile Benzoyl chloride o-Nitrotoluene m-Nitrotoluene p-Nitrotoluene

0.9849 0.9900 0.9811 0.9981 0.9870 0.9835 0.9723 0.9048 0.9937 0.9705 0.9985 0.9973 0.9956

aFromZ’= ax+ by+ cz+ i.

retention index of chlorobenzene with methanol at r$ I’

= -0.0640(635.0)

= 0.589

is calculated to be

- 0.7169(553.2) + 1.7470(699.5) + 26.9 = 811.7

(9.20)

which is nearly equal to the experimental I’ of 812.7 (Table 9.19). Further examples TABLE 9.29 CHARACTERIZATION OF RP-HPLC SOLVENT SELECTIVITYAND POLARITY WITH ADJUSTED RETENTION INDICES OF THREE PROBE MOLECULES (NITROBENZENE, BENZALDEHYDE, AND ANISOLE) RP-HPLC solvent

2-PrOH 1-PrOH THF 2-ME 2-EE DMSO EtOH 2-MEA ACN Diox DMF MeOH

Adjusted retention index (I)

Polarity

XFNO2)

fl-cHO)

4-OCH3)

(x + Y + 2)

403.2 43 1.4 495.8 494.5 524.4 532.5 571.6 581.5 584.3 602.5 627.4 635.0

336.3 365.5 418.8 413.0 432.6 482.5 485.9 502.0 507.6 526.4 531.1 553.2

465.1 501.6 531.1 540.9 573.6 562.3 642.7 629.3 624.3 666.4 649.9 699.5

1204.6 1298.5 1445.7 1448.4 1530.6 1577.3 1700.2 1712.8 1716.2 1795.3 1808.4 1887.7

Solvent selectivity

33 1

TABLE 9.30 MULTIPLE LINEAR REGRESSION COEFFICIENTS FOR THE PREDICTION OF ADJUSTED RETENTION INDICES (1’) Solute

-H 4 1

-Br -CH3 -CH=CH2 4OCH3 -COOCH3

-OH -CN -cOCI 0-NT m-NT p-NT

Regression coefficientsa U

b

C

i

0.3649 -0.0640 -0.2110 -0.0790 -0.4682 -0.1899 -0.4141 2.3058 0.4095 -0.3228 0.5344 0.9374 0.9068

-1.1628 -0.7 169 -0.7670 -0.601 1 -0.5795 1.3150 1.4433 -0.1410 1.0281 1.3556 0.3849 -0.0160 0.1518

1.5794 1.7470 1.9773 1.6059 2.0198 -0.0355 0.2211 -1.3561 -0.3900 0.2045 0.2149 0.2152 0.0723

27.2 26.9 12.3 78.9 45.5 -13.6 -14.1 33.7 5.1 -12.3 7.2 -3.6 8.7

aFrorn Eq. (9.19).

(Table 9.31) demonstrate the accurate prediction of adjusted retention indices of onitrotoluene with all 12 solvents. This approach resulted in an average error of 7.4 adjusted retention index units for all 13 solutes with all 12 solutes (Table 9.32). 9.3.5.3Quantitativeprediction of resolution with any RP solvent Predicting adjusted retention indices for any of the RP-HPLC solvents makes possible the TABLE 9.31 PREDICTED AND EXPERIMENTAL ADJUSTED RETENTION INDICES FOR 0-NITROTOLUENE WITH 12 RP-HPLC SOLVENTS Solvent

MeOH EtOH 1-PrOH 2-PrOH DMF DMSO 2-EE 2-ME THF ACN Diox 2-MEA aFrorn Eq. (9.19).

References pp. 335-336

Adjusted retention index Predicteda

Experimental

709.7 637.8 486.2 452.0 686.6 591.1 577.2 546.1 547.5 648.9 675.0 646.4

710.8 640.5 490.2 451.7 690.9 598.7 573.6 547.2 544.7 651.4 671.4 64 1.3

332

Chapter 9

TABLE 9.32 AVERAGE ERRORS (MEAN DIFFERENCE BETWEEN EXPERIMENTAL ADJUSTED RETENTION INDICES AND THOSE PREDICTED FROM EQ. (9.19) Solute

Average error

Solvent

Average error

-H 4 1 -Br 4H3 -CH=CH2 -COCH3 -COOCH3 -OH

7.1 5.9 9.4 1.9 7.0 8.1 11.3 18.3 4.9 12.0 3.1 3.3 4.4

MeOH EtOH 1-PIOH 2-PrOH DMF DMSO 2-EE 2-ME THF ACN Diox 2-MEA

6.2 3.7 6.8 3.9 7.3 7.9 6.1 10.9 9.7 10.6 6.8 9.4

-CN 4OCl 0-NT rn-NT p-NT

quantitative prediction of resolution with any of the solvents. In this approach, adjusted retention indices (Z’) are predicted (Eq. 9.19), and differences in the predicted I’ values for the two solutes (81’) in a mixture are calculated. Differences in the adjusted solute retention indices (&’) that are predicted for solvent x (X,) are then related to the corresponding experimental difference obtained with methanol (Xm). The result is a relative difference in adjusted retention indices (RaZ’) for the two solvents (9.21) Because the adjusted retention index scale has been normalized relative to methanol, the ratio of adjusted retention index differences (RaZ’) directly reflects solvent selectivity differences relative to those obtained with methanol. Consequently, the resolution of the solute pair with solvent x (R,) can then be predicted by multiplying the ratio obtained relative to methanol by the experimental resolution obtained with methanol (R,,,)

R,

= RaI’ X

R,

(9.22)

For example, chlorobenzene and toluene were chromatographed with methanol at An experimental resolution (R,,,) of 0.5 was obtained, and the adjusted retention indices were 812.7 and 819.1, respectively, so that W, = 819.1 - 812.7 = 6.4. To determine if an improved resolution could be achieved by selecting a different solvent, it was necessary to predict the adjusted retention indices of the solutes with the selected solvent. For example, if THF were selected, the adjusted retention indices for chlorobenzene and toluene were predicted (Eq. 9.18) to be 622.8 and 640.9, respectively. Therefore, &’, = 18.1, so that the resolution with THF was calculated (Eq. 9.22) to be

c$

= 0.589.

R, = (18.U6.4)

X

0.5 = 1.4

(9.23)

333

Solvent selectivity TABLE 9.33 PREDICTED RESOLUTION OF CHLOROBENZENE AND TOLUENE FROM THEIR PREDICTED ADJUSTED RETENTION INDICES (1')AS AN AID TO HPLC SOLVENT SELECTION Solvent

THF EtOH I-PrOH 2-ME 2-MEA DMSO

Predicted I'

I' difference @I:)

4 1

-CH3

622.8 764.8 613.6 644.2 729.2 629.2

640.9 773.7 630.6 660.1 741.8 649.8

18.1 8.9 17.0 15.9 12.6 20.6

aFrom Eq. (9.15), with Ral' = (a1'/6.4) and R,

Resolution (RJ Predicteda

Experimental

1.4 0.7 1.3 1.2 1.0 I .6

1.2 0.5 1.0 1.3 1.0 2.2

= 0.5.

which agreed closely with the experimental value of 1.2 (Table 9.33). It could be seen (Table 9.33) that changing solvents from methanol to ethanol would not substantially improve resolution, while changing to 1-propanol, 2-methoxyethanol, THF, or 2methoxyethyl acetate would result in a resolution that would be improved, but less than baseline. However, changing to DMSO would result in a resolution that would be greater than baseline. The experimental resolution values (Table 9.33) confirmed these predictions. A second example is presented in Table 9.34 for the resolution of benzonitrile and acetophenone. The resolution (R,) and adjusted retention index difference (W,J with methanol were determined to be 0.7 and 27.3, respectively. It was predicted (Eq. 9.22) that an improved resolution for this compound pair could not be achieved by changing to any of the other solvents under the given conditions. The experimental resolution values confirmed these predictions, with resolution ranging from 0.01 to 0.6 (Table 9.34). TABLE 9.34 PREDICTED RESOLUTION OF BENZONITRILE AND ACETOPHENONE FROM THEIR PREDICTED ADJUSTED RETENTION INDICES (1')AS AN AID TO HPLC SOLVENT SELECTION Solvent

Predicted I'

I' difference

Resolution

@fX)

-CN ~~~~

EtOH 1 -PrOH 2-PrOH DMF DMSO 2-EE 2-ME THF ACN Diox

-COCH3

Predicteda

Experimental

0.2 0.1 0.0 0.3 0.0 0.1 0.1 0.2 0. I 0.2

0.2 0.1 0.2 0.01 0.3 0.2 0.2 0.6 0.5 0.2

~

488. I 362.0 334.6 554.5 500.0 440.9 421.2 43 1.6 522.8 533.0

494.1 367.3 335.5 542.6 499.8 435.3 416.4 424.0 520.7 540.5

aFrom Eq. 9.22, with RaI' = (afX/27.3)and R, = 0.7.

References pp. 335-336

6.0 5.3 0.9 11.9 0.2 5.6 4.8 7.6 2.1 7.5

334

Chapter 9

TABLE 9.35 PREDICTED RESOLUTION OF O-NITROTOLUENEAND p-NITROTOLUENE FROM THEIR PREDICTED ADJUSTED RETENTION INDICES (I’) AS AN AID TO HPLC SOLVENT SELECTION Solvent

EtOH I-PrOH 2-PrOH DMF DMSO 2-EE 2-ME THF ACN Diox 2-MEA

I’ difference

Predicted I’

w : )

0-NT

p-NT

637.8 486.2 452.0 686.6 591.1 577.2 546.2 547.5 648.9 675.0 646.4

647.3 491.7 459.0 705.2 605.5 59 1.4 558.9 560.3 660.7 683.1 657.7

9.5 5.5 7.0 18.6 14.4 14.2 12.2 12.8 11.8 8.1 11.3

Resolution Predicted

Experimental

0.7 0.4 0.5 1.3 I .o 1.O 0.8 0.9 0.8 0.6 0.8

0.5 0.3 0.4 1.2 0.5 0.9

0.8 0.5 0.6 0.3 0.3

~

aFrom Eq (9.22), with RaI’ = @I’,J13.0) and R,

= 0.7.

A final example is presented in Table 9.35 for the resolution of the ortho and para isomers of nitrotoluene. Using methanol as the strong solvent, R, and dZ’, were determined to be 0.3 and 3.9, respectively. The best resolution (1.3) for this compound pair was correctly predicted to occur with DMF. The prediction of resolution requires the highly precise prediction of adjusted retention indices, as only small errors in retention can result in relatively high percentage errors in predicted resolution. For example, small errors in retention prediction with dioxane in Table 9.35 resulted in a predicted resolution of 0.6 compared to an experimental resolution of 0.3. However, the predictions are still useful for mobile phase selection, as the absolute resolution error of 0.6 versus 0.3 is not highly significant. 9.4 CONCLUSIONS

A simple linear equation relating resolution to 2-keto alkane retention indices has been derived from classical chromatographic equations. The Z2 (minimum) values are derived from an equation containing terms that reflect both the selectivity ( f 2 - t’J and efficiency (w,+ w2)of the chromatographic system. The ability to predict the minimum retention index difference needed to achieve resolution can be used to optimize gradient or isocratic separations of compound mixtures. Experimental data [21,33,35,51] suggest that each of the HPLC solvents studied offers some unique selectivity that cannot be duplicated by any of the other solvents. Consequently, proposals that utilize only a few preferred solvents for optimization of separations in HPLC can result in overlooking many common solvents that exhibit the required selectivity for a given separation. The solvent selectivity triangle concept fails to group HPLC solvents and GC liquid stationary phases according to selectivity for resolving compound mixtures. The concept

Solvent selectivity

335

fails because the triangle is constructed from data that are not only unrelated to experimental resolution, but which also mask solvent differences by expressing selectivity as normalized fractions of summed retentions rather than as absolute differences in retentions. Selectivity can be correlated with HPLC retention indices in a manner such that experimental resolution can be accurately predicted. By adjusting retention indices obtained with the various solvents for differences in slopes of the 2-keto alkane retention index calibration line, it is possible to characterize RP-HPLC solvents in a manner that directly reflects selectivity differences. The same approach can be applied to GC retention indices obtained on different liquid phases, so that resolution can be predicted from adjusted retention indices after correcting for slope differences of the n-alkane standards. The utility of this approach is enhanced by the ability to predict HPLC and GC retention indices [21,35,51]. A similar adjustment of McReynolds constants to account for standard calibration slope differences for the various GC liquid phases can be made using existing data, thereby eliminating anomalies in GC stationary phase classification. A nearest neighbor technique [47] utilizing differences in adjusted retention data would result in a more accurate comparison of GC stationary phase selectivity than does the triangle approach with retention indices that have not been adjusted for calibration slope differences. RP-HPLC solvent selectivity has been characterized on a CI8 column with retention indices that were adjusted for the relative slopes of the 2-keto alkane standard calibration line, Adjusted retention indices of three probe molecules (nitrobenzene, benzaldehyde, and anisole) were used to characterize solvent selectively for the accurate prediction of retention and resolution for a group of 13 substituted aromatic compounds containing a wide variety of functional groups. Equations derived from chromatographic relationships were used to predict resolution for all 12 of the RP-HPLC solvents in a manner that aided solvent optimization for the separation of compound mixtures.

9.5 ACKNOWLEDGMENTS

Material in this chapter has been reproduced from the Journal of Chromatographic Science by permission of Preston Publications, A Division of Preston Industries, Inc.

9.6 REFERENCES 1 2 3 4

5 6

7

Anonymous, Am. Lab., January (1984) 136. R.E. Majors, H.G. Barth and C.H. Lochmuller, Anal. Chem., 54 (1982) 323R. R.E. Majors, H.G. Barth andC.H. Lochmuller,Anal. Chem., 56 (1984) 300R. R. Melander and C. Horvath, in: High Performance Liquid Chromatography, Advances and Perspectives., Vol. 2 (C. Horvath, Ed.), Academic Press, New York, 1980 and references therein. J.K. Baker, L.A. Cates, M.D. Corbett, J.W. Huber and D.L. Lattin, J. Liq. Chromatogr., 5 (1982) 829. E. Kovats, Helv. Chem. Acta, 41 (1958) 1915. M.V. Budakegyi, E.R. Lombosi, T.S. Lombosi, S.Y. Meszaros, Sz.Nyiredy, G. Tarjan, I. Timar and J.M. Takacs, J. Chromatogr., 271 (1983) 213.

336

Chapter 9

8

W.O. McReynolds, Gas Chromatographic Retention Data, Preston Technical Abstracts Company, Evanston, IL, 1966. J.F. Sprouse and A. Varono, Am. Lab., September (1984) 54. M. Popyl, V. Dolansky and J. Coupek, J. Chromatogr., 130 (1977) 195. J.K. Baker and C. Ma, J. Chromatogr., 169 (1979) 107. R.M. Smith, J. Chromatogr., 236 (1982) 313. M.N. Hasan and P.C. Jurs, Anal. Chem. 55 (1983) 263. H.J. Mockel, G. Welter and H. Melzer, J. Chromatogr., 388 (1987) 255. J.K. Baker, L.A. Cates, M.D. Corbett, J.W. Huber and D.L. Lattin, J. Liq. Chromatogr., 5 (1982) 829. J.K. Baker and E.K. Fifer, J. Pharm. Sci., 69 (1980) 590. J.K. Baker, Anal. Chem., 51 (1979) 1693. J.K. Baker, R.E. Skelton, T.N. Riley and J.R. Bagley, J. Chromatogr. Sci., 18 (1980) 153. J.K. Baker, J. Liq. Chromatogr., 5 (1982) 829. C. Ma, M.A. Elsohly and J.K. Baker, J. Chromatogr.,200 (1980) 163. S.D. West, J. Chromatogr. Sci., 25 (1987) 122. J.K. Baker, Anal. Chem., 56 (1984) 2932. H. Magg and K. Ballschmiter, J. Chromatogr.,331 (1985) 245. S.D. West, J. Chromatogr. Sci., 27 (1989)2. L.R. Snyder, J. Chromatogr., 92 (1974) 223. L.R. Snyder, J. Chromatogr. Sci. 16 (1978) 223. L.R. Snyder and J.J. Kirkland, Introduction to Modern Liquid Chromatography, 2nd edn., Wiley, New York, 1979. J.L. Glajch, J.J. Kirkland, K.M. Squire and J.M. Minor, J. Chromatogr., 199 (1980) 57. P.L. Smith and W.T. Cooper, J. Chromatogr., 410 (1987) 249. W.T. Cooper and Li-Ying Lin, Chromatographia, 21 (1988) 335. R.M. Smith, Anal. Chem., 56 (1984) 256. R.M. Smith, J. Chromatogr., 324 (1985) 243. J.J. Lewis, L.B. Rogers and R.E. Pauls, J. Chromatogr., 264 (1984) 339. J.T. Przybytek, High Purity Solvent Guide, 2nd edn., Burdick and Jackson Laboratories Inc., Muskegon, MI, 1982. S.D.West and R.C. Hall, J. Chromatogr. Sci., 14 (1976) 339. Z. Xiuyou, S. Jian, S. Rong, Z. Lili and B. Ningsheng, Nanjing Daxue Xuebao Ziran Kexue, 3 (1982) 651. J. Tranchant, Practical Manual of Gas Chromatography(J. Tranchant, Ed.), Elsevier, Amsterdam, 1969. J.A. Garcia-Dominguez,J.G. Garcia-Munoz, V. Menendez, M.J. Molera and J.M. Santiuste, J. Chromatogr., 393 (1987) 209. W.R. Supina, The Packed Column in Gas Chromatography, Supelco, Inc., Bellefonte, PA, 1974. W.O. McReynolds, J. Chromatogr. Sci., 8 (1970) 685. W.A. Aue and V. Paramasigamani, J. Chromatogr., 166 (1978) 253. B.R. Kersten, C.F. Poole and K.G. Furton, J. Chromatogr., 411 (1987) 43. T.J. Betts, J. Chromatogr.,354 (1986) 1. P. Shah, H. Na and L.B. Rogers, J. Chromatogr., 329 (1985) 5. M.S. Klee, M.A. Kaiser and K.B. Laughlin, J. Chromatogr.,279 (1983) 681. L. Rohrschneider,Anal. Chem., 45 (1973) 1241. J.J. Leary, J.B. Justice, S. Tsuge, S.R. Lowry and T.L. Isenhour, J. Chromatogr. Sci., 11 (1973) 201. G.W. Snedecor and W.G. Cochran, Statistical Methods, 7th edn., Iowa State University Press, Ames, IA,1980. SAS, Statistical Manual, PROC GLM, SAS Institute, Cary, NC, 1982. L. Rohrschneider,J. Chromatogr.,22 (1966) 6. S.D. West and D.H. Mowrey, J. Chromatogr. Sci., 29 (1991) 497.

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51

R.M. Smith (Ed.), Retention and Selectivity in Liquid Chromatography Journal of Chromatography Library, Vol. 57 0 1995 Elsevier Science B.V. All rights reserved

337

CHAPTER 10

Retention and selectivityfor polycyclic aromatic hydrocarbons in reversed-phase liquid chromatography Lane C. Sander and Stephen A. Wise Chemical Science and Technology Laboratory, Organic Analytical Research Division, National Institute of Standards and Technoloa, Gaithersburg, MD 20899-0001, USA

10.1 INTRODUCTION

Polycyclic aromatic hydrocarbons (PAHs) are widespread environmental contaminants resulting from emissions from a variety of sources including: industrial combustion and discharge of fossil fuels, residential heating (both fossil fbels and wood burning), and motor vehicle exhausts. Because of their mutagenic and carcinogenic properties, PAHs are measured in environmental matrices including air, water, soil (sediment), and tissue samples. PAHs are usually present in environmental samples as extremely complex mixtures which contain numerous alkylated and non-alkylated isomers that vary greatly in relative concentration and carcinogenic andor mutagenic properties. Since its inception in the early 1970s, high performance liquid chromatography (LC) has been used for the separation of PAHs. In 1971, Schmit et ul. [l] first described the separation of PAHs using a chemically bonded octadecylsilane (CIS) stationary phase. Since Schmit’s report, reversed-phase LC on chemically bonded c18 phases has become the most popular LC mode for the separation of PAHs [2-51. The popularity of reversedphase LC for PAH separations is due, in part, to its excellent selectivity for the separation of PAH isomers. Even when using high resolution open tubular column gas chromatography (GC), a number of isomeric PAHs are still difficult to separate on conventional nonpolar stationary phases, e.g. chrysenekriphenylene; benzo[b]fluoranthene/benzo[i]fluoranthene/benzo[k]fluoranthene; and dibenz[u,c]anthraceneldibenz[a,h]anthracene. Ultraviolet (UV) absorption and fluorescence spectroscopy provide extremely sensitive and, more important, selective detection for PAHs in LC. Because of the excellent separation and detection selectivity of reversed-phase LC, this technique has been specified as the method of choice by the US Environmental Protection Agency (EPA) for the determination of PAHs in aqueous effluents [ 6 ] . References pp. 368-369

\

338

Chapter I0

In the late 1970s as part of LC methods development for PAH separations, studies at the National Institute of Standards and Technology (NIST, then the National Bureau of Standards), were focused on the retention of PAHs on various stationary phases in both normal and reversed-phase LC [7,8]. Confronted with the need to compare retention on different columns, retention data were reported using a retention index system described by Pop1 et al. [9] for normal-phase separations of PAHs (see discussion below). However, we observed that the retention indices for PAHs often varied significantly among the different columns. These studies indicated that even though reversed-phase LC on c 1 g stationary phases provide excellent separations of PAHs, not all c 1 8 stationary phases provide the same selectivity (i.e. relative separation) for PAHs. In the early 1980s studies at NIST [7,10] and other laboratories [lo-141 compared different commercial c18 columns from various manufacturers for the separation of PAHs with particular emphasis on the separation of the 16 PAHs identified by the EPA as priority pollutants. These studies found that even though all of the different columns were “generically” c 1 g phases, some provided significantly enhanced selectivity for the separation of the 16 PAHs on EPA’s priority pollutant list. During these early studies, it became evident that such investigations were limited because the exact details concerning the silica substrate and the bonded-phase syntheses were difficult to obtain from the LC column manufacturers. As a result of this limitation, investigations were initiated at NIST to understand more filly the influence of factors such a bonded-phase type, silica substrate characteristics, alkyl chain length, and c1g ligand density on selectivity of PAH separations in reversed-phase LC. In this review chapter, the results of our investigations concerning the stationary phase characteristics affecting selectivity for PAHs separations are summarized. In addition, a retention index system is used to report reversed-phase retention data for over 170 PAHs and alkyl-substituted PAHs.

10.2 STATIONARY PHASE CHARACTERISTICSAFFECTING SELECTIVITY IN RPLC 10.2.1 Phase type The ability of a column to discriminate among PAH isomers (and other classes of compounds) on the basis of the molecular structure is described as “shape selectivity”. Shape recognition in liquid chromatography is the subject of extensive reviews [ 15-17]. Perhaps the single most important parameter affecting column shape selectivity towards PAHs is a property involving phase structure that has been all but ignored over the past 20 years, namely “phase type” (whether a phase was prepared by monomeric or polymeric synthesis chemistry) [18,19]. Bonded phases are prepared through the reaction of silica with chloro- or alkoxysilanes (see Fig. 10.1). The hctionality of the silane reagents and reaction conditions employed in the surface modification greatly affects the properties of the resulting phase. Silicon has a valence of four, and silane reagents may have up to three reactive sites per molecule in addition to the non-reactive primary substituent. Surface modification results from the reaction of mono-, di-, or trihctional silanes with silanols (Si-OH groups) on the silica surface. For example, c1g phases are usually prepared

Retention and selectivity for PAHs in RPLC

339

Monomeric Synthesis (Monofunctional Silane) SiOH

Si(CH3)$

SiOH

Si(CH3).p

SiOH

SiOH

SiOH

CI(CH,),SiR

F -

Si(CH3)p

SiOH

SiOH

SiOH

Si(CH3)p

Monomeric Synthesis (Trifunctional Silane) SiOH

SiOHR

SiOH cinu

Polymeric Synthesis (Trifunctional Silane) SiOH SiOH SiOH SiOH SiOH SiOH

water

CI3SiR

SI-0-$I-0-SI-R

0 SI-041-R SiOH 0-SI-R Sl-O-QI-O-~I-R OH 0-SI-R SiOH OH SI-O-$I-R OH

Fig. 10.1, Synthesis schemes for monomeric and polymeric alkyl stationary phases.

through the reaction of mono-, di-, or trichlorooctadecylsilanewith silica. Even though the primary substituent in each case is the same (octadecyl ligands), phases prepared with these silanes may exhibit different selectivities depending on reaction conditions. Under rigorously anhydrous conditions, silanes react with silica to yield similar phases regardless of silane functionality. Monofunctional silanes necessarily react to form single References pp. 368-369

340

Chapter I0

bond linkages with silica (Fig. 1O.la). This reaction is ultimately limited by steric hindrance effects of the bound ligands as surface modification proceeds. Di- and trifunctional silanes have the potential for forming multiple bonds with silica (Fig, 10.lb). Spectral evidence does suggest that two covalent bonds can be formed per silane molecule, but it is considered unlikely that three bonds per molecule occur due to spacing and orientational constraints of silica silanols [20]. Under otherwise similar reaction conditions, phases prepared with trifunctional silanes have slightly higher phase loadings than phases prepared with monofunctional silanes, although the effect on shape selectivity is insignificant. Very different results occur for syntheses carried out in the presence of water (Fig. 10.1~)[19]. Chloro- and alkoxysilanes are hydrolyzed by water to yield silane silanols. These hydrolyzed silanes are subject to reaction with chloro- or alkoxysilanes; with trifunctional silanes polymerization readily occurs. Phases prepared with trifunctional silanes in the presence of water are termed polymeric phases, and likewise, phases prepared with monofunctional silanes are termed monomeric phases. Because the presence of water is essential in the formation of polymeric phases, syntheses carried out under anhydrous conditions result in monomeric phases, even when trifunctional silane reagents are employed. Polymeric phases should not be confused with polymer substrate columns, which are typically based on porous polystyrene particles rather than modified silica. Considerable controversy exists over the relationship between silanol activity and phase type. Phases prepared using di- and trifunctional silanes have the potential for increased silanol density compared to phases prepared with monofunctional silanes. This is true for both polymeric syntheses and monomeric syntheses carried out under anhydrous conditions, since any unreacted chloro- or alkoxy groups in the stationary phase will hydrolyze upon contact with water in the mobile phase. Although a polymeric phase may have more silanol groups than a monomeric phase, changes in shape selectivity between monomeric and polymeric C18phases are not the result of differences in silanol density. As will be discussed later, the enhanced shape recognition characteristics of polymeric CI8phases are present for non-polar solutes such as PAHs, and retention differences cannot be explained by silanol-solute interactions [151. A separate question concerns how silanol activity is related to silanol type. Two types of silanols can be distinguished: silanols on the substrate surface (silica silanols) and siIanols created on the silane reagent (silane silanols). Tests commonly used to assess silanol activity employing basic solutes such as or phenylthiohydantion-arginine(PTH-Arg) or N,N-diethyl-m-toluamideindicate that monomeric and polymeric phases prepared with trifunctional silanes do not necessarily have high silanol activity [2 I]. The fact that silane silanols are introduced during bonded phase synthesis with polyfunctional silanes is probably less significant to retention behavior than is generally believed. The vast majority of columns used in liquid chromatography are prepared using monomeric surface modification chemistry. In an informal sampling of over 60 commercial CI8 columns, only 7 were identified as having polymeric-like selectivity [17]. The relatively limited number of commercial sources of polymeric C18 phase columns has helped to promote a lack of appreciation for the unique properties of these columns. A prejudice still exists against polymeric phases, based in part on the assumption that the phases exhibit high silanol activity and low efficiency. In fact, of the seven commercially

Retention and selectivify for PANS in RPLC

341

available polymeric CIScolumns, only one manufacturer promotes the product as a polymeric phase. The remaining manufacturers target specific applications (e.g. PAH separations) to express the novel selectivity differences inherent with the phases.

10.2.2 Isomer separations

Differences in column selectivity between monomeric and polymeric CIS phases are readily apparent for separations of PAH isomers [22]. A comparison of separations of molecular weight (MW) 278 isomers is presented in Fig. 10.2. Using a heavily loaded polymeric CISphase, all 11 isomers in the mixture were baseline resolved, whereas under the same mobile phase conditions many of the components co-eluted using a monomeric

Fig. 10.2. Separation of MW 278 PAH isomers on (a) monomeric C18 column, and (b) polymeric C18 column. Values refer to WB values.

References pp. 368-369

Chapter 10

342

A

18

B

7

5

l2 10 9

n' AA

A - J ~ J ~

Fig. 10.3. Separation of MW 302 PAH isomers on (A) polymeric c18 column, and (B) monomeric c18 column. c18 column and the isomer group eluted in a much narrower band [22]. This difference is not a consequence of column strength or absolute retention. To illustrate this point, an attempt was made to improve the separation with the monomeric phase by adjusting the mobile phase composition to spread the solutes uniformly during the gradient. The net result was broad peaks with increased retention but no increase in component resolution. An even more dramatic example of differences in phase selectivity is provided by separa-

Retention and selectivity for PAHs in RPLC

343

tion of MW 302 PAH isomers, on monomeric and polymeric Cls phases (Fig. 10.3). Very little separation of these isomers could be achieved on a monomeric CIScolumn; however, most of the components were resolved with a polymeric CISphase. Comparable trends have been observed for methyl-substituted PAH isomers. Several methylsubstituted isomer sets have been studied, and in general, enhanced separations of methyl PAHs are achieved using polymeric Cls columns compared with monomeric CIScolumns [ 10,18,22]. 10.2.3 Assessing column shape selectivity

Because column manufacturers do not routinely provide details of column preparation, chromatographers must somehow determine the suitability of a column for a specific application. A variety of column testing procedures have been proposed for evaluation of properties such as silanol activity [23-251, trace metal activity [26], column “strength” [23,27], substrate pore size [28-311, and efficiency [23]. A simple empirical test has been developed to assess column shape selectivity toward PAHs [17,21,32]. This test is based on the retention of three PAH solutes, two of which have non-planar conformations. This test material is available as NIST Standard Reference Material 869, “Column Selectivity Test Mixture for Liquid Chromatography (Polycyclic Aromatic Hydrocarbons)” (Standard Reference Materials Program, NIST, Gaithersburg, MD 20899, USA). The retention of benzo[a]pyrene (BaP; planar conformation), relative to 1,2:3,4:5,6:7,8tetrabenzonaphthalene (TBN, non-planar conformation, alternate name dibenzok,p]chrysene) and phenanthr0[3,4-~]phenanthrene(PhPh, non-planar conformation), provides a sensitive measure of the polymeric or monomeric character of the phase. Phases prepared using monomeric surface modification chemistry give the elution order BaP 5 PhPh < TBN; while phases prepared using polymeric surface modification chemistry give the order PhPh < TBN I BaP. The first category has been described as “monomeric-like selectivity”, and the second, “polymeric-like selectivity”. Phases with intermediate properties such as densely loaded monomeric CI8phases, or lightly loaded polymeric CISphases are indicated by the elution order PhPh < BaP < TBN and are said to have “intermediate selectivity”. It should be emphasized that shape selectivity does not correlate with absolute retention. High surface area monomeric CISphases with high absolute retention have little shape recognition ability; the opposite is true for low surface area polymeric phases. Separations of SRM 869 are illustrated in Fig. 10.4 for several commercial C18 columns. The dramatic differences in these separations illustrate the variations in selectivity that exist among columns, and serve to emphasize the importance of column selection in method development. A quantitative measure of phase shape selectivity can be calculated to enable relative comparisons between different CISphases. The shape selectivity factor amNmap(defined as kmN lkBs) has been shown to correlate with retention behavior for PAHs and phase type [18,19,22,32]. This column evaluation test mixture enables columns to be grouped into well characterized classes, facilitating method development. A classification scheme has been proposed based on measurement of aTBNBap values for experimental and commercial C18 columns [32]. Values for a T B N f B e I 1 reflect moderately loaded polymeric C18 phases, and values for Q T B N D ~2 1.7 reflect monGmeric CIS phases. For values References pp. 368-369

344

Chapter I0

aTBwBsp = 2.08 (monomeric)

TBN

BaP + PhPh

TEN

= 1.27 (Intermediate) ~tT8W.p

TBN + BaP

aTBwB,, = 0.48

(polymeric)

Fig. 10.4. Separation of SRM 869 “Column Selectivity Test Mixture” on c18 columns with different shape selectivity characteristics.

22

m

P

s. 0 3

? lu

3

a

SELECTIVITY CLASSIFICATION (aTBNmap) FOR VARIOUS COMMERCIAL C18 COLUMNS Polymeric phases

“Intermediate”phases

i% 3

Monomeric phases

6 0,

I

2

lu

3

Column

aTBN/BaP

Column

aTBNIBap

Column

aTBN/Bap

9

Bakerbond C18 Wide-Pore Hypersil Green PAH Phenomenex Envirosep PP Chromspher PAH

0.56 0.58 0.58 0.59

ES Industries BF-C18 LiChrospher 100 RP-18 Bakerbond C 18 Erbasil C18 M

1.04 1.11 1.27 1.28

Erbasil C18 L Pecospher 5 Cr C18 Partisphere C18 Zorbax ODS

1.76 1.76 1.79 1.80

2 3

BioRad RP 318 Supelcosil LC-PAH Vydac 201TP Spherisorb PAH Erbasil C18 H

0.59 0.63 0.74 0.82 0.91

LiChrospher 60 RP-select B Partisil 5 ODs-2 Partisil 5 ODS Spherisorb ODs-1 Zorbax RX C18 Brownlee ODS 5A Sepralyte C18 Spherisorb ODs-2

1.36 1.40 1.48 1.50 1.50 1.51 1.61 1.68

Serva C 18 Partisil 5 ODs-3 Hypersil ODS (HP) Microsorb C18 J&W Accuphase ODS 2 Novapak C18 Ultrasphere ODs Capcell C18 SG120 A Supelcosil LC-18 IBM ODs Brownlee Spheri 5 RP-18 ODS Hypersil Cosmosil C18-P Ultracarb 5 C18 (20%) J&W Accuphase ODS YMC 120 A “A” Ultracarb 5 C18 (30%) Adsorbosphere C18 HS Supelcosil LC-18-DB

1.84 1.93 1.94 1.95 1.96 1.97 1.98 1.99 2.00 2.00 2.02 2.04 2.04 2.05 2.07 2.08 2.10 2.10 2.18

B 3

%h

n

346

Chapter 10

1 < a m N / B a p < 1.7, the synthesis scheme is less certain, and may indicate light polymerization with di- or tribctional reagents, or densely loaded monomeric phases. Thus, the following classification scheme has been proposed: (a) aTBN/Bap I 1, “polymeric-like”, (b) 1 < (xTBN/Bap < 1.7, “intermediate”, and (c) a m N B a p 2 1.7, “monomeric-like”. A listing of over 40 commercial c18 columns, grouped using this classification scheme, is provided in Table 10.1. In general, better overall separations of isomers and other mixtures of structurally similar compounds can be achieved with polymeric CI8 columns compared with monomeric (218 columns. However, since changes in elution order occur with variations in column selectivity, it is possible that specific critical pairs may be better separated on monomeric c18 phases. Figure 10.5 illustrates the separations of the 16 PAHs listed by the EPA as priority pollutants. This mixture of 16 PAHs is available as NIST SRM 1647. Incomplete separation of several components occurs with monomeric c18 columns; complete baseline resolution is possible with polymeric c18 phases. This separation has been studied in some detail, and it has been found that column selectivity must fall within a narrow window of aTBN/Bap values for separation of all components to be possible (Fig. 10.6). Three pairs of components prove to be most critical in the separation: (1) chrysene and benz[u]anthracene, (2) indeno[1,2,3-cd]pyrene and benzo[ghi]perylene, and (3) benzo[ghi]perylene and dibenz[u,h]anthracene. Columns exhibiting selectivity factors within two ranges (-0.65 < a m N / B a p < -0.9 and amN/Bap < -0.4) provide complete separation of the 16 components, although the elution order for benzo[ghi]perylene and dibenz[u,h]anthraceneis reversed for aTBN/Bap < -0.4. Both of these ranges correspond to polymeric CI8 phases; however, lower aTBN/Bap values indicate increased phase density (loading). Column selectivity can be controlled at several levels. Column manufacturers can of course change reaction conditions and silane reagents to alter phase selectivity. For example, the degree of phase loading with polymeric c18 phases can be varied by controlling the quantity of water introduced in the synthesis. An alternative approach to tailoring selectivity is to blend dissimilar phases. Thus, to achieve a column with aTBN/Bap = 0.65, a = 0.4) could be blended with a monomeric heavily loaded polymeric phase (e.g. amNBaP phase (e.g. aTBN/Bap= 1.8) in fractions appropriate to yield the desired selectivity. Wise et al. [33] have shown that columns prepared with blended silicas have retention behavior that is very similar to columns prepared with unblended bonded silica with comparable selectivity. Unfortunately, neither of these approaches is readily accessible to chromatographers for routine applications. A third, more easily utilized approach is column coupling [33]. When joined together, the selectivity of dissimilar columns is intermediate to the selectivity of individual columns. The resulting retention behavior is most easily predicted for isocratic separations. With gradient elution separations, solute retention is affected by column order, since the analytes will interact with each stationary phase under different mobile phase conditions. 10.2.4 Pore size effects

The possible influence of pore size on retention is often neglected for low molecular weight solutes. Size exclusion effects are not expected to play any significant role for

Retention and selectivity for PAHs in RPLC

I

I

I

0

5

10

347

I

I

I

i

15

20

25

30

I

I

I

15

20

25

Time (min)

I

I

I

0

5

10

30

Time(min)

Fig. 10.5. Separation of the 16 Priority Pollutant PAHs (SRM 1647) using (a) monomeric C18 column, and (b) polymeric Clg column.

PAHs or other small solutes since molecular dimensions are small compared to the pore diameter of most reversed-phase substrates. For example, benzo[a]pyrene has a molecular length of -14 A, and this can be compared with pore diameters of 60-300 8, for common commercial phases. Despite this assumption, changes in column selectivity have been observed for CI8phases prepared on different pore diameter silicas. References pp. 368-369

348

Chapter I0 1.12 1.10

.-2

1.18 1.06 1.04

B- 1.02 CI

L

s 1.00 6 0.98 5

Coelution

0.96 0.92 OSg4

i 0.0

A-A

-rn 0.25

0.50

0.75

1.00

Selectivity

1.25

IPBghiP BghiPpahA 1.50

1.75

2.00

aTeNlsaP

Fig. 10.6. Plot of the relative retention of critical pairs (acp) as a function of column shape selectivity aTBN/Bap for the separation of the 16 priority pollutant PAHs. Points inside shaded regions represent co-elution of critical pairs.

The effect of pore size on column selectivity was studied for four substrates with pore diameters of 60, 100, 150 and 300 A [34]. Monomeric and polymeric CI8 phases were synthesized on each substrate, and columns prepared. To evaluate differences in selectivity, SRM 869 was chromatographed as well as SRM 1647a, “Priority Pollutant Polycyclic Aromatic Hydrocarbons (in Acetonitrile)”. Few differences were evident for any of the separations among the four monomeric CI8phases. The primary difference was variations in absolute retention. This is expected, since the surface area (and thus phase loading) of the small pore substrates was significantly greater than that for the wide pore substrates. Most important were the similarities in column selectivity among the four monomeric phases, suggesting that size exclusion does not play a significant role in the reversedphase separation of PAHs on monomeric phases. The effect of pore size on selectivity was pronounced for the polymeric CISphases. Polymeric phases prepared on narrow pore substrates exhibited retention behavior more like commercial monomeric C18phases than polymeric phases. Thus, separation of the 16 priority pollutant PAHs, which is normally possible only with certain polymeric Clg phases, could be achieved only for the 150 and 300 8, polymeric C18columns. In an examination of polymeric C18 phases prepared on many dissimilar silicas of different pore sizes, narrow pore substrates consistently yielded phases with monomeric-like selectivity. Typically, substrates with pore diameters of 150 8, or larger produced phases with enhanced “polymeric-like” selectivity for PAH isomers. Since monomeric C18phases did not exhibit this variation in selectivity with pore size, size exclusion effects were ruled out as the source of the differences. Instead, phase structure is viewed as being different among the polymeric phases. We envision a size exclusion effect that influences the extent of polymeric surface modification. Polymeric

Retention and selectivityfor PAHs in RPLC

349

reaction schemes utilizing trichlorosilanes and water form silane polymers in solution. These reactive polymers may have an appreciable molecular weight, and consequently surface modification is affected by pore size. An excess of the trichlorosilane is used, and monomer co-exists with silane polymer in solution. Monomer and polymer molecules complete for reaction at the silica surface. Because the smaller silane monomers can diffuse into narrow pores more easily than large silane polymer molecules, more monomer molecules reach the surface of narrow pore substrates, resulting in a high percentage of bound monomers. On the other hand, silane polymers compete with silane monomer molecules more effectively in large pore substrates, and the resulting phases have overall polymeric-like character.

10.2.5 Bonding density The effect of bonded phase loading (density) on selectivity has been studied by a number of research groups, particularly for monomeric C18phases. Phase density can be regulated by altering the synthesis reaction conditions. For monomeric C, phases surface coverage values typically range from about 3.0 to 3.5 pmol/m*. Phases with low alkyl chain density are easily prepared by reducing the concentration of the reactive silane or by reducing the reaction time. High density monomeric phases with surface coverage values of 4 ,umol/m2 or greater have been prepared by Sentell and Dorsey using a reaction procedure driven by ultrasound [35]. High density phases have also been prepared by Kovats using a novel dimethylaminodimethyloctadecylsilane modifying reagent [36]. In general, retention is observed to increase with increases in the percent carbon loading of the phase. These changes are as expected (based on increases in the phase ratio), and selectivity towards PAHs remains relatively constant, at least for monomeric phases with low to normal phase density. Anomalous shifts in retention and selectivity have been reported for high density phases. Sentell and Dorsey have observed that retention is greatest for a bonded phase density of about 3.1 ,umol/m2, and at higher monomeric phase densities retention actually decreases [37]. It should be noted that column shape selectivity also changes for high density monomeric phases. Using the SRM 869 Column Evaluation Test Mixture, Sentell and Dorsey found that the high density phases were more selective towards shape, with a shift in selectivity toward that normally associated with polymeric C18phases [38]. Polymeric C18phases typically have phase densities nearly twice that of monomeric C18phases (5-6 ,umol/m2).Because these phases are prepared through polymerization of a trifunctional silane, phase structure at the molecular level is difficult to conceptualize (and harder to prove!). One possibility is that the additional phase loading results from branched structures extending away from the silica surface. Another possibility is that phase structure is like a monomeric phase, but with substantially increased phase density. We favor the latter model for polymeric phases. Studies utilizing small angle neutron scattering (SANS) have been carried out on both monomeric and polymeric C18phases [39,40]. Bonded phase thicknesses were evaluated (see Table 10.2), and although the value for the polymeric C18phase was greater than for the monomeric phase (i.e. 21 A versus 17 A respectively), the difference was small, and not enough to account for the differences in phase loading. Bonded phase density was also obtained from the measureReferences pp. 368-369

Chapter I0

350

TABLE 10.2 PHYSICAL PROPERTIES OF ALKYL BONDED PHASES AS DETERMINED BY SMALL ANGLE NEUTRON SCATTERING [39]

Phase type

Thickness (A)

Alkyl chain volume fraction

c8 monomeric

10f2 17*3 21 It3

0.65 f 0.15 0.66 f 0.15 0.88 f 0.10 0.63 f 0.15

C18 monomeric c18 polymeric C30 monomeric

25f4

ment, and the polymeric phase was found to be significantly denser than the monomeric phase. At the molecular level, the two phase types do not appear to be fundamentally different, but instead the differences appear to be a matter of degree and the result of alkyl chain packing density. This view of bonded phase structure should be used with caution, since the covalent linkages for polymeric CI8phases are known to differ from monomeric c 1 8 phases, and silanol density is also known to differ between the two phase types. Among phases prepared using polymeric phase syntheses, phase density can be altered by changing the quantity of water added during the reaction [ 191. Phases with loadings as high as 6pmol/m2 have been prepared. Such heavily loaded polymeric phases exhibit even greater shape selectivity, as is evidenced through isomer separations. The effect of phase loading on selectivity is shown in Fig. 10.7 for a separation of MW 278 isomers. The change in selectivity with phase loading is also indicated by separation of SRM 869. The selectivity factor C Z T B N / B ~decreases with increasing phase loading. To date, the smallest aTBNrnaP value observed was -0.2, for an experimental lot of a commercial polymeric Ct8column.

10.2.6 Bonded phase length

The effect of alkyl phase length on retention and selectivity has been studied extensively over many years. For non-rigid, non-polar solutes, retention is observed to increase with increasing bonded phase length, either through a linear or logarithmic relationship [41431. Berendsen observed a critical chain length behavior for solute retention such that k increased with alkyl phase length up to a point, and then leveled off [44]. The chain length at which this leveling off of retention occurs is called the critical chain length. 0thers have observed a similar phenomenon related to changes in selectivity [45]. Despite the existence of this considerable body of research, little effort has been expended in examining possible relationships between phase length and shape selectivity. We studied this effect by preparing monomeric and polymeric phases with alkyl ligands ranging from c8 to C30 [46]. Shape selectivity was probed using SRM 869 and SRM 1647a. For the monomeric phases, selectivity changed little among the c8, CI2 and c18 phases, but amN/Bap was observed to decrease significantly for the C22 and C30phases. The decrease in a T B N / B a is indicative of phase selectivity similar to that expected for CI8 polymeric phases. In addition, the aTBNBaP values were usehl in predicting the extent of

Retention and selectivity for PAHs in RPLC

I

hk

arsNlsaP = 1.80 (monomeric)

351

I

U I-

I arsNlsap = 0.65

I

I

(polymeric)

arsmap =0.38 (polymeric)

Fig. 10.7. Separation of MW 278 PAH isomers on monomeric and polymeric C1g columns with different shape selectivity factors (aTBN/Bapf.

References pp. 368-369

352

Chapter I0

2-o

T

1.6

0 0 0.4 0

4

8

12

16

20

24

28

32

Phase length

Fig. 10.8, Variations in column shape selectivity (aTBN/s&I) as a function of alkyl phase length for monomeric (0) and polymeric phases (0).

separation of the priority pollutant PAHs. Baseline separation of the components of SRM 1647a was possible with the long chain length monomeric phases, but not for phases C18 and shorter. A plot of a T B N B a p versus chain length is shown in Fig. 10.8. For the polymeric phases of various alkyl lengths, selectivity is observed to change significantly for the shorter chain lengths. Values for aTBN/s&I increase for the shorter akyl phases. The monomeric phases take on “polymeric-like” selectivity for long alkyl lengths, and polymeric phases become “monomeric-like” for short chain lengths (1 2 carbons and below).

10.2.7 Mobile phase composition

In reversed-phase LC, the most important consideration for controlling solute retention is mobile phase composition. Numerous studies have been made to examine the relationship between the composition of the mobile phase and retention, and most of the data indicate linear free energy relationships: retention decreases with increasing organic composition of the mobile phase, and plots of In k versus percent organic modifier are linear. Selectivity is also observed to vary as a function of mobile phase composition, although trends are not easily predicted, especially among dissimilar solutes. We have studied changes in selectivity as a function of mobile phase composition for alkyl homologs (alkylbenzenes) and for aromatic solutes [47]. For the reversed-phase solvent systems examined (i.e. aqueous solutions of methanol, ethanol, and acetonitrile), the ratio ln(k,+ l/kfl)for two consecutive alkyl homologs decreases with increasing organic composition. Changes in shape selectivity were also assessed using S R M 869 for various mobile phase systems (Fig. 10.9). Although changes in the shape selectivity factor

Retention and selectivity for PAHs in RPLC 3.0

0.0

353

I

i

I

! 50

60

70

80

90

100

Mobile Phase Composition (% organic modifier)

Fig. 10.9. Variations in column shape selectivity (qBN/Bap) as a function of mobile phase composition.

were small, in each case C Z T B N D decreased ~~ with increasing organic content in the mobile phase. This indicates that shape recognition increases with increasing organic mobile phase composition; however, it is doubtful that this change is large enough to be of practical significance in method development. The change in selectivity with mobile phase composition is of concern in the reporting of retention indexes, and this problem will be discussed in a later section. aTBN/Be

10.2.8 Temperature

Unlike gas chromatography, liquid chromatography rarely uses temperature as a separation parameter. When it is employed, column temperature is controlled to improve retention reproducibility or efficiency. Solute retention decreases with increasing temperature, and occasionally this property is exploited to adjust retention when the mobile phase composition is held constant. The effect of temperature on selectivity has been studied relatively little, particularly for subambient separations. In our laboratory, we have observed dramatic changes in column selectivity towards planar and non-planar solutes with changes in column temperature [48]. These changes occur continuously over the useful temperature range for both monomeric and polymeric CI8columns. SRM 869 was used initially to probe changes in selectivity (see Fig. 10.10). The selectivity factor aTBN/Bap decreased at subambient temperatures, and increased at elevated temperatures. Low values for amNIBap (i.e. amNIBap < 1 at ambient temperature) typically indicate polymeric phase retention behavior, so the observation that aTBNmap decreases with temperature suggests that “polymeric-like” selectivity might be temperature induced. References pp. 368-369

354

Chapter 10

>90 TBN

TBN

TBN

TBN

TBN+B.P

TBN+PhPh

TBN

0

2

4

6

8

I 0 1 2 1 4

Elevatedtemperatures

0

10

20

50

40

60

60

70

Subambient temperatures

Fig. 10.10. Separation of SRh4 869 “Column Selectivity Test Mixture” on a polymeric Clg column at different temperatures.

Likewise, increases in temperature produced increases in aTBNBap and “monomeric-like’’ selectivity. A plot of aTBNBap versus temperature is shown in Fig. 10.11 for a monomeric and a polymeric C18column. Enhanced separations of complex isomer mixtures were achieved at subambient temperatures. The aTBN/Bap values determined at various temperatures were indicative of overall column shape selectivity towards PAH isomers. For example, as previously indicated, an amNBap value of 0.65-0.9 (indicative of a polymeric CI8phase) is usually required for separation of the 16 priority pollutant PAHs. This separation was achieved using a monomeric CIS phase (amNmap = 1.7) operated at -8”C, at which aTBN/Bap= -0.65. Thus, reduced column temperature was used to alter the selectivity of a monomeric phase to mimic that of a polymeric CI8phase. By cooling a polymeric phase, even greater shape discrimination was possible, and this approach was used to resolve 5and 6-methylchrysene, a difficult analytical separation [48,49].

Retention and selectivity for PAHs in RPLC 2.0

355

monomeric

1.5

-"intermediate"

__---_----

0.5

--

monomeric CIS polymeric CIS

0

I

I

0

I

1

-40

-20

0

20

40

60

80

100

Temperature, C O

Fig. 10.11. Column shape selectivity (CZTBN/S~~) plotted as a function of temperature for monomeric C18 column (0)and polymeric Clg column (0).

The increase in the ability of a column to separate solutes on the basis of shape with reductions in temperature is a general trend that is observed for all CI8 columns. At a given subambient temperature, the relative degree of shape discrimination depends on column selectivity observed at ambient temperature. For example, although enhanced shape recognition is displayed at subambient temperatures for both monomeric and polymeric c18 phases (compared to the same columns at ambient temperature), greater shape recognition is possible with the polymeric c18 phase at low temperatures since the column exhibits enhanced shape selectivity at ambient temperature. This is evident from the plots P at any given of ~ T B N I B ~inP Fig. 10.1 for which QTBN/B~P (polymeric) < ~ T B N B ~(monomeric) temperature. We have explained the influence of temperature on selectivity in terms of the changes in phase structure that occur with temperature. The phenomena associated with phase transitions in normal paraffins have been studied in detail by Snyder [50-551. At temperatures above the melting point, normal hydrocarbons exhibit considerable disorder, with random bends and kinks in the alkyl chains. These conformational defects give rise to distinct features which are evident in infrared spectra of the hydrocarbons (see Fig. 10.12). The intensity of the absorptions decreases with decreasing temperature, indicating a decrease in the abundance of the defects. At the melting point, dramatic changes in the IR spectra occur that result from the formation of a crystalline lattice. We have applied the results of Snyder's research to bonded stationary phases. Fourier transform infrared (FTIR) spectra were obtained at various temperatures above and below the melting point for the corresponding bulk alkane (see Fig. 10.13) [56]. Two features are immediately evident from the spectra. Transitions in the IR spectra, corresponding to bends (gauche-gauche, gg) and kinks (gauche-trans-gauche', gtg') decrease with decreas1 9

References pp. 368-369

356

Chapter 10

Conformation

IR TransHion (cm-1)

End Gauch (&)

1345

Kink (t,gtg’k) 1367 Gauche-Gauchc (t,ggh) 1354

I 1400

I

1380

I

1360

I

1340

I 1320

Wavenumbers (cml)

Fig. 10.12.FTIR spectrum of liquid octadecane at 44°C. Assignments for transitions within the interval 14001320 cm-’ are indicated by representative space filling models of octadecane.

ing temperature. Also, the FTIR spectra do not exhibit any evidence of first order phase transition of the type observed for the freezing of normal paraffins. Thus, bonded alkyl ligands become straighter and more extended at low temperatures, but do not associate to

I

1400

I

I

I

f

1380

1360

1340

1320

WAVENUMBERS (cm-1)

Fig. 10.13. FTIR spectra of silica modified with dimethyldocosylchlorosilane (monomeric C22 phase), at various temperatures.

Retention and selectivity for PAHs in RPLC

357

undergo a phase transition. The inability of the bound chains to associate in a crystalline lattice is perhaps not unexpected, since movement of the ligands is restricted by the covalent linkage. Thus, phase ordering results not from enthalpic contributions from lattice formation, but instead from entropic effects of individual chains as conformational defects are eliminated at low temperatures. The structure of this ordered, bonded phase can be conceptualized as a liquid crystalline system of monolayer thickness. Solute interaction with such an ordered phase can be expected to involve molecular shape; solute shape should be less of a factor for interaction with randomly oriented alkyl chains. An argument can be made that bonded phases at ambient temperature possess more order than true liquids, since the covalently bound ligands have restricted degrees of freedom and necessarily extend away from the silica surface.

10.3 RETENTION INDEXES The reporting of liquid chromatographic retention data in a meaningful form remains a difficult problem. Because absolute retention is dependent on a variety of stationary phase, mobile phase, and operational parameters, capacity factors have limited utility in the expression of solute retention. However, if all chromatographic conditions are held constant including mobile phase composition and temperature, then a comparison of k values for different solutes is valid for a specific column. If these criteria are met, a tabulation of k values for a series of solutes should provide an indication of relative elution order and degree of separation. A better approach, however, is the normalization of retention relative to a standard(s) and the expression as a retention index. The retention index approach was popularized for GC with the introduction of the Kovats retention index system in 1958 in which a series of homologous normal hydrocarbons were used as the retention index standards [57]. An LC retention index system for PAHs, which was similar to Kovats indexes for GC, was described by Popl et ul. in 1974 [58]. In this system, retention is expressed relative to PAH standards containing from one to five aromatic rings and assigned the following retention indexes: benzene 10, naphthalene 100, phenanthrene 1000, benz[u]anthracene 10 000, and benzo[b]chrysene 100 000. Popl et ul. [9,58] reported the retention indexes (Z) as log(Z) values using the following expression: IogZ, =

log R: - log R,', log R,',+I - log R,',

(10.1)

where R' is the corrected retention volume, x represents the solute, and n and n + 1 represent the lower and higher eluting PAH standards. Capacity factors can be used in place of corrected retention volumes with no change in the retention index. The five retention standards, benzene to benzo[b]chrysene, are assigned log I values of 1-5. Using this system, a retention index of log Z = 4.5 indicates elution intermediate to benz[u]anthracene and benzo[b]chrysene. In 1979 Lee et ul. [59] developed a similar retention index system based on PAH stanReferences pp. 368-369

358

Chapter I0

dards for reporting GC data. Using the Kovats retention index system, Lee and coworkers found that the retention index for PAH solutes varied by as much as 20 index units when comparing two columns with different stationary phase film thicknesses. To minimize these differences, Lee suggested the use of benzene, naphthalene, phenanthrene, chrysene, and picene as the retention standards. Using this system, differences of only 0.14 index units were observed on the two columns with different stationary phase film thicknesses. The Lee retention index system has found widespread use in GC for the reporting of PAH retention data [60]. As mentioned in the Introduction, we have previously investigated the use of the retention indexes described by Popl et al. [9] for comparing LC data on different columns [7,10]. This work was carried out on a number of different c1g columns, and it indicated a problem inherent to reversed-phase LC data, namely, that retention indexes in this system varied with the type of column utilized. The primary utility of retention indexes in liquid chromatography is retention normalization. Retention indexes for PAHs determined on columns with dissimilar absolute retention, but similar overall selectivity for the compounds of interest should be similar. For example, monomeric c{gphases prepared with low and high surface area substrates will differ in absolute retention behavior, but selectivity for PAHs will remain nearly constant. Thus, retention indexes for PAHs determined with each column will be similar, even though absolute retention may differ considerably. Conversely, perhaps the biggest limitation in the use of retention indexes in LC for PAHs is the variations that occur with changes in column selectivity. Because differences in column selectivity can result in changes in the relative retention of various classes of solutes (including changes in elution order), retention indexes will necessarily depend on selectivity. For PAHs, column selectivity is most strongly affected by phase type (monomeric versus polymeric), column temperature, bonded phase length and density, pore size, and to a lesser extent, mobile phase composition (see discussions above). Thus, retention indexes can only be compared for solutes chromatographed under the same chromatographic conditions. In practice, it is usually sufficient to specify c1g monomeric or polymeric phase chemistry, column temperature, and mobile phase composition to obtain somewhat consistent retention index values. However, we are currently investigating the potential of SRM 869 to specify column selectivity quantitatively, and relate this data to changes in retention indexes. While this work is still in progress, it does appear that columns that exhibit similar aTBNBap selectivity factors (from SRM 869) will also have relatively constant retention indexes for PAHs.

10.3.1 Retention index data

Retention indexes for selected PAHs and alkyl substituted PAHs are presented in Tables 10.3 and 10.4. In the original work of Popl et ul. [9], a six aromatic ring PAH retention standard was not specified; however, we have adopted the use of dibenzo[u,h]pyrene for this purpose. Obviously, dibenzo[u,h]pyrene does not represent the next member of the series of cata-condensed PAHs used as retention standards; however, since the selection of standards is arbitrary, it was selected as the longest retained PAH of the six-ring peri-

Retention and selectivityfor PAHs in RPLC

359

TABLE 10.3 RETENTION INDEXES FOR SELECTED POLYCYCLIC AROMATIC HYDROCARBONS Name

MW

LIB

Benzene Anthracene Naphthalene Phenanthrene Aceanthrylene Acephenanthrylene Fluoranthene F'yrene 11H-Benzo[u]fluorene 1 1H-Benzo [b] fluorene 7H-Benzo[c]fluorene Benzo[ghi]fluoranthene Cyclopenta[cdpyrene Benz[a]anthracene Benzo[c]phenanthrene Chrysene Triphenylene Benz[e]aceanthrylene Benz[ilaceanthrylene Benz[[laceanthrylene Benz[k]acephenanthrylene Benzo[u]fluoranthene Benz[e]acephenanthrylene Benzo[ilfluoranthene Benzo[k]fluoranthene Benzo[u]pyrene Benzo[e]pyrene Perylene Cholanthrene 13H-Dibenzo[u,g]fluorene 13H-Dibenzo[u,h]fluorene 11H-Indeno[2,I-ulphenanthrene Dibenzo[deJ;rnno]chrysene Benzo[ghi]perylene Indeno[1,2,3-cd]fluoranthene Indeno[1,2,3-cd]pyrene Benzo[b]chrysene Benzo[c]chrysene Benzo[g]chrysene Benzo[u]naphthacene Dibenz[u, c]anthracene Di benz[uj] anthracene Dibenz[u, h] anthracene Dibenzo[b,g]phenanthrene Dibenzo[c,g]phenanthrene Pentaphene Picene Benzo[u] perylene

78

1.10

1.00

1.00

178 128

3.16

3.11

2.00 3.00

2.00 3.00

202 202 202 202 216 216 216 226 226

1.57 1.24 1.46 1.35 1.33 1.22 1.27 1.68 1.78 1.34 1.18 1.20

3.39 3.37 3.39 3.55 3.81 3.82 3.49 3.95 3.94

228

1.58

4.00

228 228 228 252 252 252 252 252 252 252 252 252 252 252 254 266 266 266 276 276 276 276 218 278 278 278 278 278 278 278 278 278 278 302

1.22 1.72 1.12 1.43 1.52 1.48 1.53 1.16 1.40 1.39 1.48 1.50 1.12 1.27 1.57 1.63 1.97 1.93 1.35 1.12 1.62 1.40 1.84 1.47 1.32 1.77 1.24 1.47 1.79 1.33 1.12 1.73 1.99 1.18

3.69 4.06 3.75 4.25 4.26 4.26 4.39 4.24 4.29 4.26 4.38 4.51 4.29 4.33 4.43 4.53 4.96 4.91 5.08 4.76 4.93 4.84

3.39 3.38 3.43 3.63 3.75 3.78 3.64 4.07 3.95 4.00 3.91 3.97 3.82 4.34 4.29 4.38 4.43 4.45 4.44 4.37 4.50 4.68 4.51 4.52 4.69 4.62 4.77 4.74 5.61 5.36 5.05 5.23

References pp. 368-369

178

Polymeric column

Monomeric column

5.00

5.00

4.45 4.27 4.99 4.40 4.56 4.73 4.33 4.07 4.67 5.18 4.93

4.85 4.71 4.50 4.73 4.84 4.86 4.80 4.51 4.96 5.02 5.54

360

Chapter I0

TABLE 10.3 (continued) Name

MW

WB

Polymeric column

Monomeric column

Benzo[b]peryIene Dibenz[e,k]acephenanthnthrylene Dibenzo[a,h]pyrene Dibenzo[dejp]chrysene

302 302

1.38 1.61

5.04 5.28

5.56 5.59

302

1.73

6.00

6.00

302 302 302 302 302 302 302 302 302 302 302 302 302 302 302 328 328

1.18 1.14 1.26 1.25 1.15 1.58 1.14 1.28 1.73 1.32 1.24 1.62 1.55 1.74 1.69 1.09 1.09

4.65 4.90 5.07 4.91 4.80 5.27 4.79 4.91 5.74 5.04 4.97 5.00 4.98 5.92 5.86 4.21 4.45

5.57 5.50 5.71 5.51 5.48 5.64 5.35 5.53 5.93 5.52 5.56 5.34 5.40 5.79 5.92 4.77 5.53

Dibenz[n,e]aceanthrylene Indenor1,2,3$g]naphthacene Naphth[2,3-a]aceanthrylene Dibenzo[b,el fluoranthene

Dibenzo[b,k]fluoranthene Dibenzo[i,l]fluoranthene Dibenzo[de,qrlnaphthacene Benzo[rsf]pentaphene Dibenzovq,oplnaphthacene Naphtho[1,2,3,4-deflchrysene Naphtho[ 1,2-k]fluoranthene

Naphtho[2,3-j]fluoranthene Naphtho[2,3-k]fluoranthene Naphtho[2,1&qru]naphthacene Phenanthr0[3,4-~]phenanthrene Dibenzok,p]chrysene

condensed PAHs studied [61]. The data presented in this work are for typical monomeric and polymeric CIScolumns operated at room temperature (-25*C), with selectivity values aTBNBaP = -1.7 and 0.7, respectively. Because of the wide range of compounds studied, retention was measured under different mobile phase conditions. Some of the compounds were chromatographed at two or more compositions. An average value for log,'i was calculated, although this value is less reliable than retention indexes measured at specific mobile phase compositions. 10.3.2 Length-to-breadth ratio

Tables 10.3 and 10.4 also contain length-to-breadth (LA3) values for each of the PAHs studied. While a detailed explanation of molecular descriptors is beyond the scope of this review, LA3 is one of the most useful and commonly specified descriptors. This parameter provides an indication of the overall two dimensional shape of a molecule, and good correlations have been reported between LA3 and both GC and LC retention for PAHs. One of the earliest observations concerning solute shape and chromatographic retention for PAHs was made by Janini and co-workers [62]. Using only a few solutes, they observed that in GC, on a liquid crystalline stationary phase, retention among isomeric PAHs increased with the ratio of the length to the width of the molecule as determined by a rough estimate of these dimensions. Later Radecki et al. [63] expanded on the GC observation of Janini et al. [64] and proposed a way of describing the molecular shape as the ratio of the length to the breadth (LA3) of a box of minimum area drawn to enclose the atoms of

Retention and selectivity for PAHs in RPLC

361

TABLE 10.4 RETENTION INDEXES FOR SELECTED ALKYL-SUBSTITUTEDPOLYCYCLIC AROMATIC HYDROCARBONSa Name

MW

Benzene Methylbenzene Ethylbenzene Propylbenzene Butylbenzene Pentylbenzene Hexylbenzene Heptylbenzene Octylbenzene Nonyl benzene Decylbenzene

78 92 106 120 134 148 162 176 180 196 210

Fluorene 1-Methylfluorene 2-Methylfluorene 4-Methyl fluorene 9-Methylfluorene

166 180 180 180 180

Anthracene 1-Methylanthracene 2-Methylanthracene 9-Methylanthracene 9,lO-Dimethylanthracene

178 192 192 192 206

Phenanthrene 1-Methylphenanthrene 2-Methylphenanthrene 3-Methylphenanthrene 4-Methylphenanthrene 9-Methylphenanthrene

9-Ethy lphenanthrene 9-Isopropylphenanthrene 9-Propylphenanthrene 9-Methyl-l O-ethylphenanthrene 1-Methyl-7-isopropylphenanthrene

178 192 192 192 192 192 206 206 206 220 220 220 234

Fluoranthene 1-Methylfluoranthene 3-Methylfluoranthene 7-Methylfluoranthene 8-Methylfluoranthene

216 216 216 216

Pyrene 1-Methylpyrene 2-Methylpyrene

202 216 216

1,8-Dimethylphenanthrene 3,6-Dimethylphenanthrene

References pp. 368-369

m

Polymeric column

Monomeric column

1.00

1.00

1.64 2.14 2.74 3.27 3.78 4.27 4.75 5.17 5.60 6.08

I .40 1.73 1.35

3.27 3.35 3.26 3.08 3.16 3.43 3.69 3.41 3.68

3.11 3.57 3.69 3.52 3.91

3.00 3.40 3.68 3.34 3.26 3.38 3.81 3.62 3.58 3.67 3.85 3.79 4.11

3.00 3.50 3.71 3.47 3.51 3.96 3.93 3.88 4.10 4.27 4.17 4.65

1.13 1.33 1.22 1.36

3.39 3.73 3.86 3.80 3.85

3.43 3.87 3.91 3.91 3.95

1.27 1.38

3.55 3.98 4.04

3.63 4.15 4.21

1.41 1.71 1.38 1.23

1.45 1.58 1.37 1.25 1.25 1.46 1.26

202

362

Chapter 10

TABLE 10.4 (continued) Name

MW

LJB

Polymeric column

Monomeric column

4-Methylpyrene 1-Ethylpyrene 4,9-Dimethylpyrene

216 230 230

1.10

3.98 4.09 4.40

4.13 4.45 4.77

Benz[a]anthracene 1 -Methylbenz[u]anthracene 2-Methylbenz[u]anthracene 3-Methylbenz[u]anthracene 4-Methylbenz[u]anthracene S-Methylbenz[a]anthracene 6-Methylbenz[a]anthracene 7-Methylbenz[a]anthracene 8-Methylbenz[a]anthracene 9-Methylbenz[u]anthracene 10-Methylbenz[u]anthracene 11-Methylbenz[a]anthracene 12-Methylbenz[u]anthracene

228 242 242 242 242 242 242 242 242 242 242 242 242 256 256 256 256 256 256 256 270 270 270 270

4.00 4.18 4.14 4.39 4.33 4.28 4.15 4.17 4.21 4.37 4.18 4.17 4.14 4.18 4.80 4.37 4.42 4.84 4.26 4.26 4.18 4.41 4.41 4.47

4.00 4.39 4.43 4.51

4.97 5.18

3.69 3.73 3.94 4.09 4.04 4.04 4.17 3.90 4.24

3.91 4.10 4.29 4.41 4.37 4.37 4.37 4.34 4.82

4.06 4.39 4.49 4.28 4.20 4.17 4.17 4.19

3.97 4.46 4.54 4.41 4.36 4.36 4.35 4.61

3.15

3.82 4.19

1,12-Dimethylbenz[u]anthracene 3,9-Dimethylbenz[u]anthracene 5 ,7-Dimethylbenz[u]anthracene

6,8-Dimethylbenz[u]anthracene 7,10-Dimethylbenz[u]anthracene 7,12-Dimethylbenz[u]anthracene 9,1O-Dimethylbenz[u]anthracene 1,7,12-Trimethylbenz[a]anthracene 2,7,12-Trimethylbenz[u]anthracene 6,7,8-Trimethylbenz[a]anthracene 6,8,12-Trimethylbenz[u]anthracene Benzo[cjphenan threne

5,8-Dimethylbenzo[c]phenanthrene

228 242 242 242 242 242 242 256 256

Chrysene 1-Methylchrysene 2-Methylchrysene 3-Methylchrysene 4-Methylchrysene 5-Methylchrysene 6-Methylchrysene 5-Ethylchrysene

228 242 242 242 242 242 242 256

Triphenylene 1-Methyltriphenylene

228 242

I-Methylbenzo[c]phenanthrene 2-Methylbenzo[c]phenanthrene 3-Methylbenzo[c]phenanthrene 4-Methylbenzo[c]phenanthrene 5-Methylbenzo[c]phenanthrene 6-Methylbenzo[c]phenanthrene I112-Dimethylbenzo[c]phenanthrene

1.47 1S O 1.71 1.64 1.43 1.38 1.50 1.57 1.71 1.59 1.45 1.51

1.49 1.84 1.43 1.38 1.48 1.46 1.66

1.21 1.17 1.36 1.36 1.22 1.12 1.26 1.25 1.71 1.85

1.63 1.51 1.48 1.48

1.11

3.88

4.48 4.41 4.36 4.40 4.52 4.42 4.36 4.37 4.39 5.13 4.16 4.88 5.52 4.64 4.65 4.85

Retention and selectivity for PAHs in RPLC

363

TABLE 10.4 (continued) Name

Benzo[a]pyrene 1-Methylbenzo[u]pyrene 2-Methylbenzo[u]pyrene 3-Methylbenzo[u]pyrene 4-Methylbenzo[u]pyrene 5-Methylbenzo[u]pyrene 6-Methylbenzo[u]pyrene 7-Methylbenzo[u]pyrene 8-Methylbenzo[u]pyrene 9-Methylbenzo[u]pyrene 10-Methylbenzo[u]pyrene 11-Methylbenzo[u]pyrene 12-Methylbenzo[u]pyrene

1,2-Dimethylbenzo[u]pyrene 1,3-Dimethylbenzo[u]pyrene 1,4-Dimethylbenzo[u]pyrene 1,6-Dimethylbenzo[u]pyrene 2,3-Dimethylbenzo[u]pyrene 3,6-Dimethylbenzo[u]pyrene 3,11-Dimethylbenzo[u]pyrene 3,12-Dimethylbenzo[a]pyrene 4,5-Dimethylbenzo[a]pyrene 7,1O-Dimethylbenzo[u]pyrene Perylene 1-Methylperylene 2-Methylperylene 3-Methylperylene Picene 13-Methylpicene 2,9-Dimethylpicene

MW

I A

Polymeric column

Monomeric column

4.51

4.68

1.51

4.83 4.93 4.91 4.81 4.60 4.72 4.73 4.89 4.73 4.66 4.60 4.59 5.21 5.26 5.08 5.10 5.37 5.09

5.24 5.33

252

266 266 266 266 266 266 266 266 266 266 266 266 290 290 290 290 290 290 290 290 290 290

1.60 1.50 1.so

1.30 1.48 1.50

1.62 1.51

5.11 5.13 5.33 5.18 5.16 5.14 5.14 5.73

4.94 4.90 4.82 4.33

4.52

1.16 1.23 1.40

4.28 4.50 4.69

4.85 5.03 4.57

5.18

5.02

1.78

4.85 5.91

5.22 6.00

278

292 306

5.26 5.15

1.41 1.33 1.30 1.60 1.49 1.32 1.50 1.60 1.49 1.47 1.36 1.31 1.42

252

266 266 266

5.25

5.01

5.75

5.71 5.65 5.74 5.63 5.68 5.62 5.54 5.51

W B ratios are listed for methyl substituted PAH only.

the PAH molecule. Wise and co-workers extended this observation to reversed-phase LC with polymeric CI8 phases [lo]. Wise et ul. calculated LA3 in a slightly different way from Radecki et al. [63], i.e. for a given solute, the box is drawn about a planar representation of the molecule such that the ratio of the sides of the rectangle is maximized (Fig. 10.14a). Wise et al. [lo] demonstrated that better correlation of LA3 with LC retention was obtained using a maximized L/B rather than a “minimized area” L/B, particularly €or methyl-substituted PAHs. Shape descriptors such as L/B are easily defined for planar molecules since algorithms for their calculation are constrained to two dimensions. One approach to the calculation of LA3 is to rotate the two-dimensional representation of a molecule and calculate various “trial” values for LA3 until the maximum value (or minimum area) L/B is determined. For References pp. 368-369

364

Chapter I0

3

. V

lengm

mlnlrnumd l m d m

along I axla

Y

mutlrnum dlmnslon along x axis

Fig. 10.14. Approaches for specification of the length-to-breadth molecular descriptor. (a) the planar representation is enclosed by a rectangle such that the ratio WB is maximized. (b) Non-planar molecules are oriented with the minimum dimension aligned with the z-axis. LiJ3 is then determined from the planar ny projection.

non-planar molecules, however, different values for LA3 will result depending on the initial orientation. If the molecule is only slightly non-planar, this effect may be insignificant. However, for globularly shaped molecules (for example, phenanthro[3,4-~]phenanthrene) suitable orientation is not as straightforward and ambiguities are possible. In our current implementation for calculation of L/B, an iterative approach is used to evaluate LIB for non-planar molecules. The algorithm starts with an arbitrary molecular orientation. The molecule is sequentially rotated about x,y, and z axes and xy, yz, and xz projections are determined with each rotation. An orientation is set such that when a box is drawn about the molecule to enclose the van der Waals surface, the minimum dimension is aligned with the z-axis and the maximum dimension, with the x-axis (Fig. 10.14b). L/B is then calculated from the x y projection as if the molecule were planar. This algorithm is also useful for planar molecules since the final orientation will be with the plane of the molecule aligned with the xy plane. The z dimension is the “molecular thickness”, and this parameter can be utilized as a descriptor for solute non-planarity.

10.3.3 Planar and non-planar PAHs

Among planar PAH isomers, solute retention in reversed-phase LC is closely related to L B . Long narrow molecules are retained longer than square molecules. However, nonplanar solutes elute earlier than expected based only on L/B. This trend holds true for PAH isomer sets including methyl substituted PAHs. We have attempted to consolidate these observations through an empirical model of retention termed the “slot model”. In this model, the stationary phase is viewed schematically as an environment consisting of slots into which solute molecules can penetrate (see Fig. 10.15). These slots can be thought of as the space between the alkyl chains in CI8phases. For a slot of finite height, width, and depth, solute interaction (penetration)will vary depending on shape characteristics. Planar molecules will fit readily into some slots, whereas non-planar molecules may be excluded from the phase resulting in shortened retention. Among planar isomer sets, molecules with large L/B have the potential for the greatest interaction with the phase since long, narrow solutes will fit into a large fraction of the available slots. On the other hand, square molecules may be excluded from a fraction of the slots on the basis of solute width, resulting in weak interaction and reduced retention. Thus, the observed

Retention and selectivity for PAHs in RPLC

365

"Slot Model"

Fig. 10.15. Schematic representation of solute interaction with the bonded stationary phase. Slots represent spaces between bonded alkyl chains.

trends can be represented in a qualitative sense by the slot model. If this concept is extended further, the differences in shape selectivity that are observed between monomeric and polymeric phases can also be represented by the slot model. We have shown that polymeric phases are thicker, and more dense than monomeric phases [39]. In terms of the slot model representation, polymeric phases have thinner, narrower (and deeper) slots than monomer phases. Because the alkyl chains are more disordered with monomeric phases, the slots are poorly defined and less discrimination on the basis of shape is predicted. A good example of the effect of solute shape on retention is illustrated in Fig. 10.2b for the separation of 11 PAH isomers of molecular weight (MW) 278. Two trends are evident: (1) retention increases with increasing L/B, and (2) when L/B values are similar, non-planar solutes elute before planar solutes. Solute non-planarity results when portions of a molecule are sterically hindered or strained, and such structures are often indicated by molecular modeling. Thus benzo[c]chrysene (LIB = 1.47; modeled with non-planar shape) elutes before dibenz[a,j]anthracene (LA3 = 1.47; planar shape). Long, narrow PAHs such as picene are strongly retained (pentacene could not be eluted, even under 100% acetonitrile mobile phase conditions).

10.3.4 Methyl-substituted PAHs

The LC retention behavior of methyl-substituted PAH isomers has been studied in some detail [ 10,18,49,65,66]. Many of the trends observed for unsubstituted PAH isomers are also observed for methyl-PAHs. For example, among planar methyl-PAHs, log I is strongly correlated with L/B. Figure 10.16 illustrates this relationship for a series of References pp. 368-3 69

366

Chapter 10

benzo[c]phenanthrenes

m

x benz[a]anthracenes a

1

,’

,

1.1

1.2

1.3

3.6

1.4

chrysenes

1.5

1.6

1.1

1.8

1.9

LIB Fig. 10.16. log1 plotted as a function of L/B for methyl-substituted benzo[c]phenanthrenes (m), methylsubstituted benz[a]anthracenes ( X ) and methyl-substituted chrysenes ( 0 ) .

methyl-substituted benzo[c]phenanthrene, benz[a]anthracene, and chrysene isomers. A close examination of the data indicates that most of the points which do not fit the linear relation represent non-planar solutes. The effect of non-planarity on the retention of methyl-substituted PAHs was studied for the methyl-substituted isomers of chrysene, perylene, and picene [49]. In each instance, methyl substitution at certain sites on the parent molecule resulted in a non-planar structure, as indicated by molecular modeling. These non-planar compounds had reduced retention compared to that expected on the basis of LB. The extent of this discrepancy increased with the degree of polymeric phase loading of the stationary phase. Thus, nonplanarity was more of a factor influencing retention for polymeric C18phases than monomeric C18 phases, and heavily loaded polymeric phases showed a greater effect than lightly loaded phases. Interestingly enough, the absolute retention of many non-planar methyl-PAH isomers was less than that of the unsubstituted parent PAH. This behavior is remarkable since based on hydrophobic retention theory and other theories of solute retention, the addition of alkyl groups to a solute molecule is expected to increase retention. The observed retention behavior was explained in terms the “slot” retention model described above. The non-planar structure imposed by methyl substitution is viewed as reducing interaction with the stationary phase (compared with the unsubstituted solute), since the bulkier structure requires more space (larger “slots”) to fit between the alkyl chains. An example of a separation of methyl-PAHs is shown in Fig. 10.17 for methylpicene isomers. The upper figure is for a monomeric CIScolumn, the middle figure is for an “intermediate” column (lightly loaded polymeric CISphase), and the lower figure, is for a heavily loaded polymeric C18 column. All methyl substituted isomers are observed to elute after picene with the monomeric CISphase, but with the polymeric phases certain methylpicene isomers elute earlier than picene. 13-, 6-, and l-methylpicene exhibit nonplanar conformations due to steric hindrance effects of the methyl substitution and these compounds elute earlier than the unsubstituted parent PAH (picene) for the intermediate

Retention and selectivity for PAHs in RPLC

367

8:

11 / 11&4 /

lo\

/ 5 / 5

/

9

7

8

I

I

I

0

5

10

I 15

20

I 25

0

5

10

15

20

25

I 0

I

I

I

I

I

10

20

30

40

50

Time

I

(min)

Fig. 10.17. Separation of picene, and methyl-substituted picene isomers with (a) monomeric c18 column, (b) “intermediate” c18 column (see text), and (c) polymeric c18 column. Values in parentheses refer to L/B values.

References pp. 368-369

368

Chapter I0

and polymeric C18columns [49]. The elution of the planar methylpicene isomers follows that predicted by LA3 values. In general, it can be concluded that the elution of methylPAH isomers follows LA3, but non-planar isomers elute earlier than predicted by LA3 alone. More discrimination among isomers is possible with heavily loaded polymeric C18 columns than with monomeric CIScolumns. Finally, these factors can result in early elution of non-planar methyl substituted PAHs, before the unsubstituted parent PAH.

10.4 SUMMARY

Polycyclic aromatic hydrocarbons constitute a class of compounds with considerable chemical diversity. The widespread occurrence of these compounds in the environment and the variability of biological activity associated with individual PAHs has provided a motivation for the determination of PAH species in complex mixtures. Reporting of retention data for PAHs is complicated by the variability of solute retention with operational parameters. For retention data to have meaning, chromatographic conditions must be specified. Data reported in the form of retention indexes (rather than as k) helps to reduce variability by normalizing retention relative to standards; however, any change in conditions that alters system selectivity will also alter retention indexes. In practice, retention indexes differ significantly between monomeric and polymeric C18columns, and at different temperatures, but differ only slightly at different mobile phase compositions. Work is in progress to develop a model to relate changes in column selectivity with changes in retention indexes. If successful, such a model would enable the estimation of retention data at specified conditions from data collected under dissimilar chromatographic conditions.

10.5 REFERENCES 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

J.A. Schmit, R.A. Henry, R.C. Williams and J.F. Dieckman, J. Chromatogr. Sci., 9 (1971) 645. S.A. Wise, in: Handbook for Polycyclic Aromatic Hydrocarbons, Marcel Dekker, New York, 1983, p. 183. S.A. Wise, in: Handbook for Polycyclic Aromatic Hydrocarbons, Vol. 11, Marcel Dekker, New York, 1985, p. 113. J.C. Fetzer, in: Chemical Analysis of Polycyclic Aromatic Compounds, Wiley, New York, 1993, p. 59. K.D. Bartle, M.L. Lee and S.A. Wise, Chem. SOC.Rev., 10 (1981) 113. EPA Test Method, Polynuclear Aromatic Hydrocarbons - Method 610, US Environmental Protection Agency, Environmental Monitoring and Support Laboratory, 1982. S.A. Wise, W.J. Bonnett, F.R. Guenther and W.E. May, in: Polynuclear Aromatic Hydrocarbons: Chemistry and Biological Effects, Battelle Press, 1980, p. 791. S.A. Wise, S.N. Cheder, H.S. Hertz, L.R. Hilpert and W.E. May, Anal. Chem., 49 (1977) 2306. M. Popl, V. Dolansky and J. Mostecky, J. Chromatogr., 117 (1976) 117. S.A. Wise, W.J. Bonnett, F.R. Guenther and W.E. May, J. Chromatogr. Sci., 19 (1981) 457. K.L. Ogan and E.D. Katz, J. Chromatogr., 188 (1980) 115. E.D. KatzandK.L. Ogan. J. Liq. Chromatogr., 3 (1980) 1151. A. ColmsjO and J.C. McDonald, Chromatographia, 13 (1980) 350. R. Amos, J. Chromatogr., 204 (1981) 469. L.C. Sander and S.A. Wise, J. Chromatogr. 656 (1993) 335.

Retention and selectivity for PAHs in RPLC 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

369

S.A. Wise, L.C. Sander and W.E. May, J. Chromatogr. 642 (1993) 329. L.C. Sander and S.A. Wise, LC GC, 8 (1990) 378. L.C. Sander and S.A. Wise, in: Advances in Chromatography, Vol. 25, Marcel Dekker, New York, 1986, p. 139. L.C. Sander and S.A. Wise, Anal. Chem. 56 (1984) 504. P. Roumeliotis and K.K. Unger, J. Chromatogr., 149 (1978) 21 1. L.C. Sander, J. Chromatogr. Sci., 26 (1988) 380. S.A. Wise and L.C. Sander, J. High Resolut. Chromatogr. Chromatogr. Commun., 8 (1985) 248. M.J. Walters, J. Assoc. Off. Anal. Chem., 70 (1987) 465. E.C. Jennings and R.G. Brownlee, Anal. Chem., 58 (1986) 2895. P.C. Sadek and P.W. Carr, J. Chromatogr. Sci., 21 (1983) 314. M. Verzele and C. Dewaele, Chromatographia, 18 (1984) 84. P.E. Antle, A.P. Goldberg and L.R. Snyder, J. Chromatogr., 321 (1985) 1. R.N. Nikolov, W. Werner and I. Halasz, J. Chromatogr. Sci., 18 (1980) 207. R.N. Nikolov, J. Chromatogr., 364 (1986) 163. S.B. Schram and D.H. Freeman, J. Liq. Chromatogr., 3 (1980) 403. S. Kuga, J. Chromatogr., 206 (1981) 449. L.C. Sander and S.A. Wise, J. High Resolut. Chromatogr. Chromatogr. Commun., 11 (1988) 383. S.A. Wise, L.C. Sander and W.E. May, J. Liq. Chromatogr., 6 (1983) 2709. L.C. Sander and S.A. Wise, J. Chromatogr., 316 (1984) 163. K.B. Sentell, K.W. Barnes and J.G. Dorsey, J. Chromatogr., 455 (1988) 95. K. Szabo, N. Le Ha, P. Schneider, P. Zeltner and E. Kovats, Helv. Chim. Acta, 67 (1984) 2128. K.B. Sentell and J.G. Dorsey, Anal. Chem., 61 (1989) 930. K.B. Sentell and J.G. Dorsey, J. Chromatogr., 461 (1989) 193. L.C. Sander, C.J. Glinka and S.A. Wise, Anal. Chem., 62 (1990) 1099. C.J. Glinka, L.C. Sander, S.A. Wise and N.F. Berk, Mater. Res. SOC.Symp. Proc., 166 (1990) 415. H. Hemetsberger, W. Maasfeld and H. Ricken, Chromatographia, 9 (1976) 303. P. Spacek, M. Kubin, S. Vozka and B. Porsch, J. Liq. Chromatogr, 3 (1980) 1465. N. Tanaka,K. Sakagami and M. Araki, Chem Lett., (1980) 587. G.E. Berendsen and L. de Galan, J. Chromatogr., 196 (1980) 21. C.H. Lochmllller and D.R. Wilder, J. Chromatogr. Sci., 17 (1979) 574. L.C. Sander and S.A. Wise, Anal. Chem., 59 (1987) 2309. L.C. Sander and S.A. Wise, 14th Int. Symp. on Column Liquid Chromatography, Boston, MA, 1990 (Abstract). L.C. Sander and S.A. Wise, Anal. Chem., 61 (1989) 1749. S.A. Wise, L.C. Sander, R. Lapouyade and P. Garrigues, J. Chromatogr., 514 (1990) 111. R.G. Snyder, M. Maroncelli, S.P. Qi and H.L. Strauss, Science, 214 (1981) 188. R.G. Snyder, H.L. Strauss and C.A. Elliger, J. Phys. Chem., 86 (1982) 5145. R.G. Snyder, J. Chem. Phys., 71 (1979) 3229. R.G. Snyder, J. Chem. Phys., 47 (1967) 1316. J.H. Schachtschneider and R.G. Snyder, Spectrochim. Acta, 19 (1963) 117. R.G. Snyder and J.H. Schachtschneider, Spectrochimica Acta, 19 (1963) 85. L.C. Sander, J.B. Callis andL.R. Field, Anal. Chem., 55 (1983) 1068. E. Kovats, Helv. Chim. Acta, 41 (1958) 1915. M. Popl, V. Dolansky and J. Mostecky, J. Chromatogr., 91 (1974) 649. M.L. Lee, D.L. Vassilaros, C.M. White and M. Novotny, Anal. Chem., 51 (1979) 768. D.L. Vassilaros, R.C. Kong, D.W. Later and M.L. Lee. J., Chromatogr., 252 (1982) 1. S.A. Wise, B.A. Benner, H. Liu, G.D. Byrd and A. Colmsjb, Anal. Chem., 60 (1988) 630. G.M. Janini, G.M. Muschik, J.A. Schroer and W.L. Zielinski, Anal. Chem., 48 (1976) 1879. A. Radecki, H. Lamparczyk and R. Kaliszan, Chromatographia, 12 (1979) 595. H.J. Issaq, G.M. Janini, B. Poehland, R. Shipe and G.M. Muschik, Chromatographia, 14 (1981) 655. P. Garrigues, M. Radke, 0. Druez, H. Willsch and J. Bellocq, J. Chromatogr., 473 (1989) 207. P. Garrigues and M. Ewald, Anal. Chem., 55 (1983) 2155.

This Page Intentionally Left Blank

R.M. Smith (Ed.), Retention and Selectivity in Liquid Chromatography Journal of ChromatographyLibrary, Vol. 57 0 1995 Elsevier Science B.V. All rights reserved

371

CHAPTER 11

Comparison of novel stationary phases Joseph J. Pesek and Eric J. Williamsen Department of Chemistty, San Jose State Universi& San Jose, CA 95192-0101, USA

11.1 INTRODUCTION

Reversed-phase HPLC is one of the predominant chemical separation techniques available today as experimental conditions can be varied widely. Because of the ease, flexibility, and large changes in separation efficiency that can occur in changing the mobile phase composition, most chromatographic optimization occurs by changing this experimental variable. Other easily modified variables, such as temperature, have been studied less extensively. However, the number of possible modifications for these variables is limited by the stationary phase. Different stationary phases have been developed to increase both the types of chemical and physical interactions between the stationary phase and the solute@)and the range of permissible modifications of the other variables. Development of new stationary phases will improve existing separations, allow separations that are not currently possible and, increase column durability. The types of stationary phases reviewed in this chapter contain two components: a chemically, and possibly physically, bound organic species on a solid support. Three types of stationary phases are discussed: monomeric octadecyl silica (ODs), other silicabased stationary phases, and stationary phases based on other solid supports. This discussion is not intended to be exhaustive and does not include stationary phases traditionally used in normal-phase HPLC. Comparisons are made to the ideal properties for a reversedphase HPLC stationary phase and to the most common stationary phase, monomeric ODs. These comparisons not only illustrate the advantages and disadvantages of certain stationary phases for the separation of particular classes of solutes but also increase our knowledge of how solutes interact with the mobile and stationary phases.

11.1.1 Characteristics for the ideal reversed-phase HPLC stationary phase

The ideal HPLC column would completely separate any mixture and last indefinitely. Although no ideal stationary phase has been developed, different stationary phases have References pp. 399-401

3 72

Chapter I I

TABLE 11.1 CHARACTERISTICS OF AN IDEAL STATIONARY PHASE (1) (2) (3) (4)

(5) (6)

(7) (8)

Bonding density should be uniform and reproducible from batch to batch on different types of supports and from different sources Bonded phases should be stable over as broad a pH range as possible A broad range of solutes should be differentiallyeluted with adjustment of mobile phase composition Additional refinements in elution order can be made by adjusting the nature of the stationary phase, such as alkyl chain length or functional group($ incorporated into the bonded phase Separations for any mixture should be reproducible on the same column from injection to injection for long periods of operation Separations for any mixture should be reproducible from column to column with the same type of stationary phase under identical mobile phase conditions Interactions between the stationary phase and the solutes should be reversible The stationary phase should remain attached to the support and retain its separation indefinitely

one or more of these admirable qualities. Several characteristics leading to an ideal stationary phase are given in Table 11.1.

11.2 CHARACTERIZATION TECHNIQUES Although the ultimate test of a particular stationary phase is whether the solutes of interest are separated, knowledge of why the separation occurs aids in the development of better stationary phases. A number of spectroscopic, thermal, elemental analysis, and chromatographic methods have been used to characterize stationary phases. Information from these techniques include the chemical and physical state of the solid supports and bonded phases, confirmation of the degree of success in bonding the stationary phase to the solid support, and identification of the chromatographic behavior of the stationary phase.

11.2.1 Spectroscopic techniques Spectroscopic techniques used to provide information in the study of stationary phases and solid supports include IR, NMR, and ESCA. Sampling is often performed in the solid phase because stationary phases are bonded to solid, not easily soluble, supports and crushing the solids would destroy the pore structure, and therefore, positioning of the stationary phase on the support. 11.2.1.1IR methods

The primary type of IR analyses performed is not the more common absorption or transmission measurements, but difhse reflectance infrared Fourier transform (DRIFT). Bonded phase and other species (silanols, silanes, etc.) on the solid support surface can be identified, and information on the success of the bonding and elimination of deleterious groups can be qualitatively obtained. For example, a partial DRIFT spectra of several steps of a bonding reaction with Partisil-40 silica is shown in Fig. 11.1 [11. The broad

Comparison of novel stationary phases

4000

3500

3000

2500

373

2000

1200

c m -1 Fig. 1 1.1. Partial DRIFT spectra of a silica (Partisil-40) after derivatization: (A) native silica; (B) hydride intermediate prepared via a chlorinationheduction sequence; (C) octyl-bonded phase; (D) octadecyl-bonded phase. For better illustration, the ordinates were contracted by a common factor and the curves were moved downward. Ordinate units are therefore arbitrary. (Reproduced from Ref. 1 with permission of the American Chemical Society.)

References pp. 399-401

374

Chapter I1

peak, especially prominent on the unaltered silica (Curve A), between 4000 and 3000 cm-I is due to water on the surface of the silica while the small sharp peak at 3740 cm-1 is due to isolated silanols. In the hydride intermediate (Curve B), the Si-H stretch appears at 2260 cm-' and the silanol peak decreases. In the bonded phases (Curves C and D), the Si-H stretch decreases in intensity and a peak between 3000 and 2800 cm-l, representative of C-H stretching bands, appears. I I . 2.1.2 NMR methods

NMR studies provide more information on the bonded species, other surface species, and the amount of bonding. The cross-polarization, magic angle spinning NMR (CP-MAS NMR) technique is often used to enhance the low signal levels expected fiom NMR analyses of solid substances. NMR analyses on nuclei that are specific to the bonded phases (I3C) or to the support material (29Si,27Al)can be performed. NMR studies provide complementary information to IR; these spectra often have fewer features than IR spectra and provide information on bonds that have low or non-existent intensity in the IR. 13C and 29Sispectra are shown in Figs. 11.2 and 11.3, respectively, of octyl-bonded silica [ 13. Assignments for these peaks are shown above each spectra. NMR studies can also be used to monitor the interaction of the mobile and stationary phases. 2H and I3C NMR studies by Sentell et al. [2] have been used to investigate solvation and temperature effects on the interaction between the stationary phase and mobile phase. Measurements were made both with the mobile phase solutions alone and with a

1

1

8

4

S

a

T

8

Si-CH,-CH,-CH,-CH1-CH,-CH,-CH,-CH, a 3.4

1e.1

a 1.0 11.4

1e.1

8 1.0

11.a

. 50

40

20

30

10

0

PPM

Fig. 11.2. 13C CP-MAS NMR spectrum of an octyl-bonded phase. (Reproduced from Ref. 1 with permission of the American Chemical Society.)

Comparison of novel stationary phases

8

Si'(OSi-),

b

H O S i ' (OSi-),

375

b

I

8

C

I

d

IC

!

A

k

-a

-40

-60

-eo

-100

-120

-140

-160

-ieo

PPM

Fig. 11.3. 29Si CP-MAS NMR spectra of (A) hydride intermediate and (B) octyl-bonded silica. (Reproduced from Ref. 1 with permission of the American Chemical Society.)

paste mixture of mobile phase and stationary phase. The amount of molecular motion due to interaction between the mobile and stationary phases was determined by measurement of the spin-lattice relaxation time, TI. 11.2.1.3 ESCA methods

ESCA is a spectroscopic technique especially useful in the characterization of impurities on the surface. Many bonding reactions require the use of catalysts which contain transition metals. Because any material left on the surface can significantly change the chromatographic environment, a way to monitor for the presence of catalyst after reaction is reReferences pp. 399-401

376

Chapter 11

t

585

580

-

J

575

570

565

binding energy ( e V )

Fig. 11.4. Partial ESCA spectrum of the Cr(2p) region of a weak cation exchange material. (Reproduced from Ref. 41 with permission of Friedr. Vieweg & Sohn Verlagsgesellschaft mbH.)

quired. NMR and IR are not well suited to monitoring transition metal species, but ESCA is. For example ESCA has been used to monitor for the presence of Cr on the support after bonding with a procedure that required the use of potassium dichromate [41]. The ESCA spectrum for the Cr(2p) region of the final product is shown in Fig. 1 1.4. It clearly indicates that no chromium contamination exists on the surface. Besides transition metals, other important elements that can be detected include N, S, 0, and halogens, elements which also have a direct effect on the chromatographic behavior.

11.2.2 Thermal methods

Besides the spectroscopic techniques, thermal techniques, such as differential scanning calorimetry (DSC) and thermal gravimetric analysis (TGA) are useful in qualitatively and quantitatively characterizing solid support surfaces and bonded phases. For example, TGA has been used to determine the concentration of -OH groups on the solid support surface [4]. TGA and DSC studies have also been used to provide information on the success of surface modification. DSC thermograms of Partisil-40 and Vydac TP hydrosilation by chlorinationheduction and reaction with TES are shown in Fig. 11.5 [4]. The large feature between 400 and 500°C occurred due to the heat released by the reaction of hydride groups at the surface with oxygen. The area of the peak represents the amount of hydride available. For this example, the larger amount of hydride at the surface when performing the hydrosilation by the TES reaction instead of by chlorinationheductionwas easily seen.

11.2.3 Elemental analytical methods

Elemental analysis is another useful tool for quantitatively determining surface coverage of an organic stationary phase upon the solid support. The carbon percentage can be related to surface coverage by an equation proposed by Berendsen and de Galan [5]:

Comparison of novel stationaryphases

O . 0O .j 0- 4

~ 0 . 0

3 77

0.0 400.0

500.0

600 .O

Tewerature I'CI

30.0

I

400.0

Teweraturc

I

500.0

0.0 600.0

('c)

Fig. 11.5. DSC curves of hydride-modified silicas in air: (I) Partisil-40; (11) Vydac TP. Curves: (A) chlorinationheduction production; (B) TES silanization product. (Reproduced from Ref. 4 with permission of the Royal Society of Chemistry.)

cwR(pmol/m2)= lo6 Pc (lo2 ~ 4 c n c - pM~~ S B E T

(11.1)

where pc is the carbon percentage by weight of the bonded material (after correction for any carbon present before bonding), n, is the number of carbon atoms in the bonded organic group, SB, is the specific surface area (m2/g) of the hydride substrate, MR is the molecular weight of the stationary phase, and Mc is the atomic weight of carbon. This form of the equation is for an addition bonding reaction. If a hydrogen is lost during the reaction, as in an organosilation reaction, the atomic weight of hydrogen must be subtracted from the weight of the bonded phase.

11.2.4 Chromatographic characterization

Most spectroscopic and thermal characterization techniques do not sample the stationary phase under chromatographic conditions. Although these measurements provide important information on the bonded phase and the surface of the support material, no direct information on how the stationary phase and support will interact with the mobile phase and solutes is provided. One method used to directly measure these interactions is through the use of retention indices. Retention indices are groups of solutes that are selected to emphasize specific physical or chemical interactions of the solute with the mobile phase, bonded phase, and support material. Referencespp. 399-401

378

Chapter I I

TABLE 11.2 REQUIREMENTS FOR HPLC RETENTION INDICES (1)

(2) (3) (4) (5)

(6) (7)

The relationship between log k and the number of carbon atoms or characteristic functional groups in the molecules of the homologues must be linear They should be readily obtainable They should be chemically stable in common mobile phases They should not specifically interact with the solid support They should not undergo ionization (in order to avoid any influence of pH and ion-pairing agents on the k value) The k values should depend little on the mobile phase composition The lowest homologue should be reasonably polar, in order to obtain a wide scale of the retention indices involving as many compounds as possible

Various groups of solutes have been used as retention indices. Pachkovk and Felt1 [ 6 ] have suggested a number of properties that the ideal retention indices for HPLC should have (Table 11.2). Because of the sensitivity of HPLC retention data to changes in the composition of the stationary phase, mobile phase, or experimental conditions, reproducible retention indices have been more difficult to determine for HPLC than for GC [ 6 ] . Examples of groups that have been used as HPLC retention indices are given in Table 11.3.

11.3 MONOMERIC OCTADECYL SILICA

Octadecyl silica (ODS) is the most commonly used stationary phase in reversed-phase HPLC because of its ability to separate a wide variety of different solutes. Although ODS has many admirable qualities, the stationary phase has several limitations.

11.3.1 Separation mechanisms for ODS phases

The major means of interaction between solutes and ODS is hydrophobicity [18]. Besides hydrophobicity, the long, bonded organic phases can also contribute to retention. Under most mobile phase combinations, CI8phases have a “brush-like” structure. This structure TABLE 11.3 EXAMPLES OF RETENTION INDICES USED IN HPLC Retention indices

Reference

Alkylbenzenes Alkan-2-ones Alkyl aryl ketones Methyl esters of saturated fatty acids p-H ydroxybenzoates Dialkylphthalates Polycyclic aromatic hydrocarbons

7,8 9 10,ll 12 13 14 15, 16, 17

Comparison of novel stationary phases

(b)

379

TMS ligand

0 : polar functional group

Fig. 11.6. Position of solute containing polar functional group ( X ) within the stationary phase of (a) ODS ligand; (b) TMS ligand. (Reproduced from Ref. 18 with permission of Friedr. Vieweg & Sohn Verlagsgesellshaft mbH.)

falls apart and becomes a film which shields the silanols on the silica surface when the amount of water in the mobile phase exceeds 50% (v/v) as the CI8molecules interact with themselves more than with the outside environment [ 191. Kaibara et ul. [ 181 suggested that the octadecyl bonded phase, unlike shorter molecules such as trimethyl, form pockets into which molecules can diffuse. This “pocket formation” can be used to explain the difference in polar solute retention for octadecyl and trimethyl stationary phases, as can be seen schematically in Fig. 11.6. Solutes containing polar functional groups are surrounded by the inhospitable non-polar octadecyl groups while smaller stationary phases, such as trimethyl, allow the polar functional group to interact more fully with the more hospitable polar mobile phase. Therefore, retention is less for the ODS phase than that found for a shorter stationary phase with similar hydrophobicity, such as trimethylsilyl (TMS). 11.3.2 ODS limitations and adjustments

The silica support limits the chromatographic conditions used and causes separation problems with certain solutes. As shown in Fig. 11.7 [20], compromises between column Referencespp. 399-401

380

Chapter I I

0

PH

- range eluent

4

14 b

tailing due to 8ilanol activity tailing due to vKa of comgoud

working area:

/ ] a c c e p t a b l e -not

J-1

acceptable acceptability depends on silica material or compound

Fig. 11.7. Peak tailing and column stability as a function of mobile phase pH. (Reproduced from Ref. 20 with permission of Elsevier Science).

stability and peak asymmetry must be made, especially when separating basic solutes. Acceptable column stability occurs only in the pH range of 2-7.5 because of hydrolytic attack of the Si-0-Si bond. For most of this pH region, basic compounds strongly interact with free silanols on the silica surface and peak tailing is seen [20]. Adjustment of the mobile phase pH to either acidic conditions (PH 5 4 ) to suppress the formation of free silanols or basic conditions (pH 2 10) to avoid ionization of the basic solutes can reduce the interaction between silanols and the basic solutes, and therefore peak tailing, but column lifetimes are dramatically shortened. Besides adjustment of mobile phase pH, other methods have been attempted to diminish the interaction between these solutes and silanols. Silanols can be made inaccessible to solutes by bonding other types of stationary phases to the column, endcapping [20], the addition of inorganic or organic salts as blocking agents [20-221, and modifying the underlying silica to make “basic” columns [21,23]. Vervoort et ul. [20] investigated a variety of different experimental conditions and ODS columns for their usefdness in the separation of basic drugs. Thirty-two test compounds were selected on the basis of having pKa values ranging from 3 to 9, retention indices between 400 and 1600, 1-5 nitrogen atoms, and different positions for the nitrogen atom on the solute. Initial experiments found that asymmetric peaks were obtained for compounds with pKa > 6 on a pBondapak CI8column and a mobile phase of methanol10 mM sodium phosphate buffer (PH = 7.4). For those compounds with pKa > 6, partially protonated molecules that were more flexible interacted more with the acidic silanol groups. A column with its residual silanols electrostatically shielded, Suplex pKb-100, was also tested. This technique apparently effectively blocks the silanols as superior peak

Comparison of novel stationaryphases

381

0 : without TEA and IPR, pH 7.4

m :with 0.05% (v/v) TEA,pH 7.4

"

IXI : with 0.01M IPR. pH 3.5

: : 4

T

%-

3-6

6-7

7-8

8-9

PKa Fig. 11.8. Influence of a silanol blocking agent, triethylamine (TEA), and an ion-pair reagent, sodium 1hexane-sulphonate(IPR), on peak asymmetry. (Reproduced from Ref. 20 with permission of Elsevier Science).

shapes with plate numbers ranging from 3000-1 1 000 were obtained without the need for buffers. Mobile phase modifiers can also be used to diminish the effect of silanols on retention. The effect of the addition of a silanol blocking agent, triethylamine (TEA), was compared to the addition of an ion-pair reagent, sodium I-hexanesulphonate (Fig. 11.8). The silanol blocking agent decreased the effect of silanols, as measured by peak asymmetry, more than the ion-pair reagent, especially for solutes with high pKw In fact, peak asymmetry measured in the presence of silanol blocking agents was as small or better than without any mobile phase modification for all solutes. 11.3.2.I Diferent bondedphase-silica linkages

Different linkages between the silica and the C18bonded phase have also been developed to improve column life by reducing the possibility of hydrolysis of the Si-0-Si bonds. Kirkland et al. [24] increased column lifetime by reacting bulky diisobutyl-n-octadecylsilane with highly purified, low-acidity porous silica microspheres. Unlike other commercial columns, the diisobutyl-synthesized column showed no detectable degradation after greater than 27 000 column volumes while being continuously purged with a 5050 (vh) methanol-water mixture with 1.0% trifluoroacetic acid (pH < 1). This column also had increased stability at high pH. While continuously purging with a pH = 9 phosphate buffer, little change was seen in chromatographic parameters, such as plate height and capacity factor, for greater than 13 000 column volumes. Besides protecting the Si-0-Si bonds, the bulky groups also diminish interaction between basic solutes and reactive silanols. Another method to increase column life is to eliminate the relatively weak Si-0-Si and instead form a stronger Si-C bond between the Si and the bonded phase. These syntheses have been performed through either a chlorination reagent followed by an alkylating reagent [3,25,26], a chlorinationheductionscheme to form a surface hydride followed References pp. 399-401

382

Chapter I I

by reaction with a terminal olefin [3,27], or a hydrosilation in the presence of metal catalysts followed by reaction with a terminal olefin [I]. Degradation of the stationary phase when the bonded phase was synthesized via the hydride intermediate was compared to a bonded phase synthesized via the traditional organosilanization method (Fig. 11.9) [ 11. These columns were subjected at room temperature to a wateddioxane solution containing 0.1% TFA (v/v), and the amount of column degradation was monitored by determining the surface coverage by elemental carbon determinations. The ODS phase made by traditional organosilanationtechniques lost about 50% of its initial coverage compared to

Elopood Time. h Fig. 11.9. Relative surface coverage of octyl-bonded Vydac lOlTPB silicas as a function of hydrolysis time: (A) octyl-silica prepared by hydrosilation; (B) octyl-dimethylsilyl silica by silanization by organosilanization. (Reproduced from Ref. 1 with permission of the American Chemical Society.)

Comparison of novel stationary phases

3 83

15% for the ODS phase made by the hydrosilation method. Although this example was for octyl-bonded phases, the results should be similar for octyldecyl-bonded phases. 11.3.2.2Diferences in ODs columns

Although ODS phases from different manufacturers have many features in common, significant variations in the chromatographic behavior and physical characteristics occur due to the different bonding schemes and different means of silica production [191. Silicas differ in their physical properties, such as specific surface areas, average pore diameters, porosities, and packing densities, and in their chemical properties, such as concentrations of isolated and hydrogen-bridged silanols, cations, and other impurities [28]. Bonding schemes differ in the silanes used, in the reaction conditions, and in whether any endcapping is used. Due to different concentrations of octadecyl and other molecules on the silica surface, interaction between adjacent octadecyl molecules, between solutes and octadecyl molecules, and between solutes and silanols vary. To gain an understanding of these differences, Engelhardt and Jungheim [19] compared a number of commercial ODS phases. To test hydrophobic interactions, toluene and ethylbenzene were used. Capacity factors (k) ranged from 2.25 to 4.16 for toluene and fi-om 3.54 to 6.63 for ethylbenzene. A linear relationship between k and % C (w/w) was found with greater amounts of carbon leading to higher k. The hydrophobic selectivity, as measured by the ratio of k’s of ethylbenzene to toluene, was determined to be a function of % C up to approximately 12% C after which there was little change in selectivity. In making this measurement the selectivity within one brand of stationary phase often varied more than the difference between c18 and Cg phases. Engelhardt and Jungheim also investigated the role of silanols in retention [19]. For polar neutral and hydrocarbon solutes, surface silanols play no role in retention, but for basic solutes, the surface silanols play a major, complex, and not completely understood role. To test the hydrolytic stability of columns, a number of columns were stored in standard water-methanol eluent. The selectivity of non-polar or neutral polar solutes was little affected, but the selectivity of basic polar solutes was greatly affected. Engelhardt and Jungheim suggested that this result can be interpreted as meaning little bonded carbon is lost during storage but more active silanols are formed through cleavage of the Si-0-Si bond.

11.4 NOVEL SILICA PHASES

This discussion of silica-based bonded phases will be divided into three convenient categories: non-C18alkanes, materials containing other hctional groups, and chiral phases. The latter is a special category because analyses are generally performed on these phases for a single purpose: to separate optical isomers.

11.4.1 Non-C18alkane phases

The simplest modification from the reference ODS standard is to change the length of the Referencespp. 399-401

3 84

Chapter I I

alkyl chain. Perhaps the second most common reversed-phase material is the octylbonded moiety. One study [29], involving both c8 and c18, compared the effects of mobile phase composition, temperature and flow rate for the separation of triglycerides. Because of the nature of the solutes, the separation was not performed in a reversed-phase mode. The solvent selected was acetonitrile-acetone-tetrahydrofuran (58:38:4) for the octadecyl column while acetonitrile-tetrahydrofuran-water (60:40:1) was used for the octyl column. Under these conditions, it was determined that the ODS phase was more suitable for the smaller triglycerides (partition number <48), while the C8-bonded phase is more effective for the separation of larger triglycerides (partition number >48). This result is consistent with the hydrophobicity difference between the two materials. It was further determined that optimum separation of the triglycerides could be achieved on both columns at a temperature of 30°C and a flow rate of 1.5 mumin. In another study of various octyl phases [30], 29Siand 13C CP-MAS NMR were used to study the surface after bonding by monodentate and bidentate organosilane reagents. The spectroscopic results indicate that there are differences in the extent of bonding as well as the degree of alkyl chain mobility depending on the type of reagent used. In general, the monodentate reagents displayed greater alkyl chain mobility than the bidentate reagents. The chromatographic evaluations of the different octyl phases also showed a few subtle differences. For polycyclic aromatic hydrocarbons (PAHs), a commonly used solute-type for chromatographic stationary-phase evaluation, the trireactive, bidentate bonded species caused longer retention, a result consistent with its greater carbon loading and poorer efficiency. The latter result correlates with the NMR data indicating restricted chain motion, probably because of greater cross-linking, which presumably leads to slower mass transfer, and hence, lower efficiency. For aniline samples, another common solute-type used to detect the presence of unreacted silanols, the results were similar to those obtained for the PAHs. Peak symmetry was very similar for the various octyl phases which indicates that the number of unreacted silanols was similar or those that are present on the surface are not readily accessible. Because of the importance of HPLC in biochemical separations, protein and polypeptide solutes are often used to characterize stationary phases. In one study of insulins and related polypeptides [313, ODS and C8-bonded phases displayed irreversible binding of the proteins on the column. Partial sample recovery was achieved with the polypeptides on these columns. Butyl and propyl bonded phases, however, were able to separate these mixtures to some degree at low temperatures (25°C) and to achieve complete separation at a higher temperature (45°C). In another more extensive survey of polypeptide and protein behavior on c8 columns [32], satisfactory elution was reported for a wide variety of biochemical species. For some proteins, relatively poor efficiency was obtained, while in other cases a high percentage of organic modifier (acetonitrile) was required for a reasonable retention time. Another study of several proteins (ribonuclease, cyctochrome-c, bovine serum albumin, carbonic anhydrase and ovalbumin) on several ODS phases, as well as on CS-, C4-, and C1-bondedmaterials, revealed no simple correlation between the type of alkyl ligand and chromatographic performance [33]. The crucial factor appeared to be proper choice of solvent (acetonitrile-water with TFA versus ethanol-butanolwater with HCl) in conjunction with the column-packing material and the protein to be eluted.

Comparison of novel stationary phases

385

When a variety of small molecules representing different shape characteristics was evaluated on C1, C8 and C18 phases, the rigid planar compounds were preferentially retained by the octadecyl phase [34]. The result of the longer chain appears to restrict the access of the bulkier molecules to the full extent of the hydrophobic surface.

11.4.2 Materials containing other functional groups While alkyl phases represent the most common type of bonded materials used in HPLC, a number of other functional groups are useful in reverse phase separations. Non-alkylbonded phases are used to reduce hydrophobicity or to create other stationary phasesolute interactions. For example, the hydrophobicity of the bonded material very often must be reduced for macromolecular separations involving proteins and peptides. This change can be effected either by shortening the chain length of the alkyl phase or by adding a polar functional group to the bonded moiety. A change in stationary phase is often exercised when mobile phase manipulation is not possible due to solute considerations.

I I . 4.2.I Cyanopropyl-bondedphases Probably the most often encountered of the phases containing other hctional groups is the cyanopropyl (CN)-bonded material. For peptide and protein separations, the CN column shows different and often better selectivity than typical alkyl phases like C8 [32]. In addition, proteins frequently exhibit higher resolution and better peak shape on the cyanopropyl columns than on alkyl columns [33]. In general, protein recoveries are very good on the CN column with the proper choice of solvent. The CN column has also been shown to give very high efficiency and excellent peak shape for the metabolite ethylenethiourea [35]. 11.4.2.2 Amine andphenylphases

Two other column materials, the amine and phenyl phases, provide a different degree of hydrophobicity, as well as selectivity, which can be used to advantage for certain separations. For example, the NH2 column is effective for the separation of certain steroid saponins with the selectivity dependent on the composition of the mobile phase [36]. While good peak shape and symmetry was obtained for ethylenethiourea on the amine column, retention properties (k too small) were not desirable under accessible mobile phase conditions [35]. With respect to the phenyl phase, reasonable resolution and recovery were possible with several proteins under certain mobile phase conditions [33]. The phenyl phase has also been shown to be a more accurate medium for use in the prediction of the n-octanol/water partition coeecient [37]. This parameter is one of the most widely accepted in structure-activity relationship studies used for the prediction of biological or pharmacological activity. 11.4.2.3 Non-alkyl phases bonded through a reactive olefin

One new stationary phase that involves attachment of the ally1 moiety to the silica surface has resulted in a variety of other bonded materials due to the presence of a reactive olefin which can be used as a site for firther modification [3]. During the initial chromaReferences pp. 399-401

386

Chapter I I

Fig. 11.10, Plot of log k for various alkyl aryl ketones versus carbon number x 100 for ally1 bonded stationary phase. (Reproduced from Ref, 3 with permission of Elsevier Science).

tographic characterization of the allyl phase, the alkyl aryl ketone homologous series was chosen for the measurement of its reversed-phase properties [3]. Figure 11.10 shows the plot of log k versus carbon number X 100 obtained in 50:50 methanol-water. The linear plot obtained is typical of reversed-phase behavior observed for other longer hydrocarbon chains like CI8.The first successful conversions of the allyl phase involved bromination and hydrobrominationwhich were tested by gas chromatography and DRIFT [38]. Figure 11.11 is a typical DRIFT result which was obtained when the allyl bonded material was converted by means of an aminomethylation to an N-butylpyrrolidine moiety 1391. The usefulness of the alkyl aryl ketone homologous series for elucidating retention behavior was illustrated when this phase was compared chromatographically to the allyl phase. Figure 11.12 illustrates the difference obtained between the allyl and the new Nbutylpyrrolidine phases. The non-linear behavior of the pyrrolidine phase was explained by the exclusion of the higher homologs of this series due to the large size of this bonded phase in comparison to the three-carbon material. When the allyl moiety was converted to a sulfonic acid material, the changes in the alkyl aryl ketone retention still produced a linear plot [40]. This result is illustrated in Fig. 1 1.13.

Comparison of novel stationary phases

387

Ally1 Bonded Stationary Phase

Product Phase

1800

1750

1700

1650

1600

1550

1500

Cm'l

Any1 Bonded Stationary Phase

3300

3200

3100

3000

2900

2800

2700

2600

Cm-1

Fig. 1 1 . 1 1 . DRIFT spectra of carbon-hydrogen stretching region (bottom) and carbon-carbon bond stretching region (top) for the ally1 phase (unlabeled spectrum) and the product of the aminomethylation reaction. (Reproduced from Ref. 39 with permission of Friedr. Vieweg 19Sohn Verlagsgesellschaft mbH.)

Because both the pyrrolidine and sulfonic acid phases possess an ion-exchange functional group, their presence was tested by separating mixtures of amino acids. Further proof of the cation-exchange properties of the sulfonic acid phase was obtained by the separation of catecholamines shown in Fig. 11.14. Both of the cation-exchange phases required the use of a metal-containing reagent, and the presence of foreign substances on the surface could make the chromatographic results suspect. To determine if any metallic residue was present, the surface of each phase was examined by ESCA. The ESCA spectrum for the Cr(2p) region of the final product, shown earlier in Fig. 11.4, clearly indicates that no chromium contamination exists on the surface. References pp. 399-401

388

Chapter I I .2

0 -.2

.

Y . 0

-.4

0 J

-.6

-.8

-1.0

-1.2

800

1000

1200

1400

Carbon Number x 100

Fig. 11.12. Log k versus carbon number x 100 for alkyl aryl ketones on the N-butylpyrrolidine phase. (Top curve pyrrolidine phase and bottom curve allyl phase.) (Reproduced from Ref. 39 with permission of Friedr. Vieweg & Sohn Veriagsgesellschafl mbH.)

When the allyl phase was converted to a weak cation exchange material [41], the plot of log k versus carbon number for the alkyl aryl ketones was linear in acetonitrile-water but non-linear in methanol-water. However, in each case retention was clearly different than for the starting allyl material. 11.4.2.4 Polymer bondedphases

Figure 11.15 shows a chromatogram for the separation of the homologous series on the polystyrene-divinylbenzene bonded phase [42]. This chromatogram was obtained with gradient elution because a non-linear log k versus carbon number plot was obtained under isocratic conditions for the alkyl aryl ketone series. This result probably indicates that strong x-x interactions occurred between the solutes and the stationary phase. 11.4.2.5Liquid crystal phases

One interesting new type of stationary phase involves the bonding of liquid crystals to silica [43]. These materials show phase transitions with change in temperature [44,45], as well as with mobile phase composition [46,47]. Most important, selectivity appears to be mainly based on molecular shape [48,49]. The dominant factors identified to date are the

Comparison of novel stationary phases

389

Carbon No. x 100

Fig. 11.13. Logk versus carbon number x 100 for alkyl aryl ketones on various columns: I, ally1 phase; 11, endcapped allyl phase; 111, cation exchange material made by conversion of allyl phase via a bulk process; IV, cation exchange material made by conversion of ally1 phase via an on-column process; and V, endcapped cation exchange material made by bulk process. Mobile phase methanol-water (10:90). (Reproduced from Ref. 40 with permission of Elsevier Science).

1

2

3

VR (ml) Fig. 11.14. Separation of catecholamineson cation exchange material made by conversion of ally1 phase via the on-column process. NA, noradrenaline; A, adrenaline; DA, dopamine. (Reproduced from Ref 40 with permission of Elsevier Science). References pp. 399-401

3 90

Chapter 11

7 3 4 5

1

1

6

8

I

2

i 4

6

8

10

12

14

16

18

20

22

24

26

28

Time (minj Fig. 11.15. Separation of alkyl aryl ketones on polystyrene-divinylbenzene silica. 1, acetophenone; 2, propriophenone; 3, butyrophenone; 4, valerophenone; 5, hexanophenone; 6, octanophenone; 7, decanophenone; 8, laurophenone; 9, myristophenone. Mobile phase: 40:60 methanol-water for 5 min, followed by a linear gradient to 100% methanol in 15 min. (Reproduced from Ref. 42 with permission of Marcel Dekker.)

length to breadth ratio and the planarity of the compounds. Polycyclic aromatic hydrocarbons (PAHs) have been the best solutes to probe the retention characteristics of bonded liquid crystal phases in HPLC. PAHs with large length to breadth ratios and a high degree of planarity are preferentially retained on these bonded phases.

11.4.3 Chiral phases

Due to the importance of obtaining optically pure materials in the pharmaceutical and biotechnological industries, significant development of stationary phases for enantiomeric resolution has occurred. The separation of optical isomers can be achieved by two distinct mechanisms: ligand exchange chromatography and direct enantiomeric separation. Numerous applications and stationary phases have been developed for each mechanism, and a variety of solutes have been used for characterizing these bonded materials. 11.4.3.I Ligand exchange chromatography

Ligand exchange chromatography can be used to separate chiral compounds that form metal complexes, i.e. the analyte functions as the chelating agent during the separation process. The ligands (or chiral selectors) can be part of the stationary phase or a compo-

Comparison of novel stationaryphases

391

nent in the mobile phase. When chiral selectors are part of the stationary phase, they are bound to the surface with the metal ion, and the analyte is a part of the mobile phase. Efficiency is generally higher for mobile phase selectors because the process of ligand formation and dissociation controls the plate height [50].However, because not all or even many enantiomeric compounds can be separated by mobile phase additives, true chiral stationary phases in which the ligand is bound to the surface chemically or through adsorption are important. Flow rate, capacity factor, analyte structure, percentage of organic modifier, type of organic modifier and pH all influence the efficiency of optical isomer separation when a chiral stationary phase is used [50]. Ligand exchange chromatography for the separation of enantiomers has been particularly effective for the separation of amino acids as well as the derivatives of amino acids and benzoic acid [50,51]. 11.4.3.2 Direct enantiomeric separation

Direct enantiomeric separations involve immobilizing a chiral compound on silica through adsorption or chemical bonding. One such example has utilized a-chymotrypsin as the chiral selector [52].While the adsorbed chymotrypsin gave satisfactory separation of enantiomers of compounds such as naproxen, the stability of the stationary phase was limited. The chemically bonded material, however, provided enantiomeric separations of mono- and divalent acids, 1-phenylethanol, tryptophan and N-substituted amino acid derivatives with excellent stability. Some examples of a values obtained on the bonded chymotrypsin phase are shown in Table 11.4. In addition to chiral selectivity, experimental factors like pH, ionic strength and the concentration of a cationic modifier were also studied. For smaller chiral selectors that are derived from amino acids or other carboxylic acids, the material can be bound to silica by two methods: either the chiral material can be attached to aminopropyl silica or a chiral reagent can be prepared and then bonded to the silica. The latter technique has been shown to be more effective with respect to both discrimination and efficiency [53]. Efficiency is improved by using a chiral silane reagent and end-capping with hexamethyldisilazane. The test compounds for the phases produced by the two different bonding methods included several 3,5-dinitrobenzoyl amino acid derivatives, an alcohol of a polycyclic aromatic hydrocarbon and an aromatic epoxide. 11.4.3.3 Pirkle phases

Attaching various chiral selectors to aminopropyl silica, referred to as Pirkle phases TABLE 11.4 ENANTIOSELECTIVITY ON BONDED CHYMOTRYPSIN PHASES Solute

a

N-( I-Pheny1ethyl)phthalamic acid Naproxen Warfarin Di-p-toluoyltartaric acid 2-(4-Iodophenoxy)propionic acid 1-Phenylethanol

1.7 1.8

References pp. 399-401

1.3 1.7 1.2 1.2

Chapter I I

392

[54,55], are some of the best and most fiequently used methods for the separation of enantiomers. Two common selectors which can be obtained commercially are (R)-N-(3,5dintrobenzoy1)-phenylglycine and (&‘)-N-3,5-dinitrophenyl)-leucine.One study [56] has evaluated the separation of 26 bay region and non-bay region mono-ol and diol enantiomers of phenanthrene, benz[a]anthracene and chrysene. Bay region mono-ols and diols were much better resolved than the non-bay region compounds. In addition, the enantiomers were generally better resolved on the @)-phase than on the (&‘)-phase. Comparisons of enantiomeric separations on Pirkle-type and other chiral phases have been made. One such study involved the separation of the racemic herbicide 2-(4-chloro2-methylphenoxy)propanoic acid [57]. The Pirkle column was the R-phase cited above and the other chiral material was acid glycoprotein. The glycoprotein column had several advantages over the Pirkle column. Separation of the herbicide occurred directly on the glycoprotein column, while the Pirkle column required conversion to an amide derivative. The separation on the glycoprotein column was performed in aqueous solution, while the Pirkle column required a non-aqueous mobile phase and a tedious derivatization step. In addition, the glycoprotein column lasted considerably longer than the Pirkle column. However, the Pirkle column was more sensitive and less costly. The R-phase is a versatile chiral material and numerous separations have been reported. Another series of test compounds on this phase involved the ester and amide derivatives of fatty acid epoxides and substituted palmitic acids [56]. This study indicated that separations of enantiomers without aromatic substituents at the chiral center was possible on this material. Additional aspects of this study involved an extensive evaluation of the effect of other experimental and structural parameters such as hydrocarbon length of the fatty acid, type of substituent at the chiral center, the type of derivative, the column temperature, sample size, mobile phase composition, column type (ionic versus covalent bonding) and flow rate on the resolution and separation factor. An example of the effects of one of these factors, mobile phase composition, on the separation factor of enantiomers of phenacyl ester of 2-tetradecylglycidic acid is shown in Table 1 1.5. 11.4.3.4 Protein phases

As mentioned above, protein columns are often the choice for certain enantiomeric separations, particularly in drug analysis. An immobilized ovomucoid column (OVM) was compared to an acid glycoprotein column (AGP) for optical isomer separation of 24 drugs [58]. In most cases the OVM column was superior to the AGP column in resolution TABLE 11.5 EFFECTS OF SOLVENT COMPOSITON ON THE RESOLUTION OF ENANTIOMERS OF PHENACYL ESTER OF 2-TETRADECYLGLYCIDIC ACID Solvent mixture

a

Ethanol-hexane (1 :99) Isopropanol-hexane (2:98) t-Butanol-hexane (2:98) THF-hexane (8:92) Dichloromethane-hexane (15%)

1.07 1.09 1.09 1.03 1.03

Comparison of novel stationary phases

393

and efficiency. The OVM material also exhibited excellent stability with respect to pH and mobile phase composition variations. These characteristics make the OVM stationary phase suitable for method development in the analysis of both acidic and basic drugs at the trace level, for stereoselective pharmacokinetic and metabolism studies, and for monitoring process intermediates in bulk syntheses and possible drug racemization. I I . 4.3.5 Organometallicphases

Another area of interest in chiral separations involves organometallic compounds that are used as asymmetric reagents or catalysts. Cyclodextrin-based columns proved usehl for some ferrocenes, but not manganese and chromium cyclopentadienyl carbonyl complexes 159,601. Much better resolution was obtained for the organometallics, particularly with methyl- substituted phenyl moieties, on tris(pheny1carbamate) stationary phases [6 I]. The size and shape of the adsorption site is apparently crucial and the presence of the methyl groups provide the necessary discriminating factor to increase separation of the enantiomers. Changing the concentration of the modifier in the hexane mobile phase (ethanol or 2-propanol) did not change the separation factor. The tris(dimethylpheny1carbamate) phase proved successful for studying the enzymatic transformations of planar chiral organometallic compounds.

11.5 NON-SILICA-BASED STATIONARY PHASES To overcome solute-silanol interactions and pH limitations of silica-based supports, the use of other supports, such as alumina, carbon-covered zirconium, and polymer has been investigated.

11.5.1 Alumina-based stationary phases Alumina has long been used as a support for column chromatography and in normalphase chromatography, although investigations of its use in reversed-phase chromatography are few. The main advantage of alumina over silica is that alumina should remain stable over a greater pH range as suggested by the higher pK, of alumina (1 1.2 compared to 9.77) [62]. In fact, alumina is stable from pH 3-12 [63]. Polymer [64], Cx [65], and CIx [65,66] are a few of the organic groups that have been bound to alumina. 11.5.I . I Monomeric aluminaphases

For example, Cx- and CI8-aluminahave been prepared via hydrosilation [65]. Results from DRIFT, 13C, 29Si,and 27AlCP-MAS NMR, and DSC studies provided results consistent with successful chemical bonding of the organic groups to the alumina surface. In the DSC/air thermooxidative study, a distinct peak appeared at 520-530°C for the hydride-modified alumina. Other studies in this lab found peaks at 365°C for polyhydrosiloxane (polymerization product of triethoxysilane (TES)) in the absence of a support and a peak at 430450°C for hydride-modified silica [67]. The large difference in peak position between the hydride-modified alumina and the polyhydrosilane was attributed to the hyReferencespp. 399-401

394

Chapter I I

2

2 .'oo

4

.'oo

6 .'OO

8 .'oo

1o:oo

Fig. 11.16. Separation of test mixture by C18 alumina in a mobile phase of acetonitrile-water (50:50). Peaks: 1, theophylline; 2, p-nitroaniline; 3, methylbenzoate; 4, phenetol; 5, o-xylene. (Reproduced from Ref. 65 with permission of Elsevier Science).

dride being chemically, not physically, bound to the surface. Because the peak of the alumina-based material occurs at a higher temperature than for the silica-based material, the alumina-based stationary phase has greater thermal stability, and possibly a higher overall stability. As seen in Fig. 1 1.16, preliminary chromatographic studies of the CISalumina bound via hydrosilation successfully separated a test mixture of theophylline, p nitroaniline, methyl benzoate, phenetole, and o-xylene with relatively little peak asymmetry. A number of stationary phases based on a Unisphere alumina, an alumina formed of approximately 200-nm thick platelets bonded together to form spheroidal particles, have been studied by Haky and associates [66,68]. Octadecyl-bonded Unisphere alumina (ODA) successhlly determined lipophilicities of organic bases in a pH range of 9-1 1, much higher than the practical pH range of silica-based materials [68]. In the same study, the retention and estimation of log P (octanol-water partition coefficients) on ODA were compared to values obtained on another monomeric phase, ODs, and two polymeric phases: polybutadiene-coated alumina (PBD) and octadecyl-modified polystyrene-divinylbenzene(ACT-1). Correlation between log P and log k was better on the ODA column. In fact as seen in Table 11.6, some log P values measured on the ODA column compare more favorably to theory than previously measured experimental values. The close agreement of ODA and octanol-water partition coefficients was attributed to hydrogen-bonding by the oxygen atoms on the alumina backbone.

Comparison of novel stationary phases

395

TABLE 11.6 PARTITION COEFFICIENTS OF MODEL COMPOUNDS Compound

Literature Log P [69]

ODA Log P

Calculated Log P

Codeine Procaine Thebaine Cocaine Benzylamine

1.14 1.90 1.91 2.09 1.09

2.04 2.21 2.38 2.92 1.45

2.01 2.34 2.51 2.51 1.56

11.5.1.2Polymeric aluminaphases

The use of a polymeric polybutadiene-coated alumina phase (PBDA) was compared to a polymeric octadecylsilane (ODS) phase for the separation of proteins and peptides [66]. Increased backpressure on successive runs and 50% reductions in peak areas during the separation of proteins on the PBDA phase in comparison to the ODS phase was attributed to irreversible absorption of the proteins within the PBDA phase. The separation of smaller, octapeptides on the PBDA column is similar to that found on the ODS phase. The hydrophobicity of both phases is similar, but PBDA phases have reduced mass transfer with larger molecules due to the unique shape of the alumina particles.

11.5.2 Carbon-based zirconia

Carbon-based chromatographic supports have also been investigated for use in reversedphase HPLC. Many early phases suffered from poor mechanical stability, low surface area, lack of an energetically homogeneous surface, and non-uniform pore structure [70]. Zirconia has some promising solid-support properties. It is insoluble in excess base and its pore structure is maintained under high temperature [71] so that it should not degrade even under extreme conditions. The pH stability of zirconia was demonstrated in a test of polybutadiene-modified zirconia and polybutadiene-modified alumina [72]. Under mobile phase conditions of methanol-0.1 M NaOH (50:50), the alumina-based phase formed a void after 8000 column volumes, but the zirconia-based materials exhibited no voids or any other stability problems even after 30 000 column volumes. In the absence of phosphate and at low pH, zirconia-based materials, however, interact with acidic solutes very strongly with (basic) anion-exchange sites on the zirconia [72]. To block the active sites on the zirconia and to utilize the useful chromatographic properties of carbon, vapor-deposited carbon zirconia stationary phases have been developed [70,73]. Untreated, H-treated, and polymerized polybutadiene-covered carbon vapor-deposited zirconia stationary phases have been tested for their stability and chromatographic potential [73]. All three phase-support combinations had 97% carbon coverage on the porous zirconium dioxide microspherules. For the complete aqueous pH range (014), no deterioration was observed for any column after 350 column volumes at 100°C. The chromatographic behavior differs with the type of modification of carbon-covered support. Retention of solutes is greater on the untreated and H-modified phases than for Refirences pp. 399-401

396

Chapter 11

a

Q

&,, -

/

I

l - - l - - - l ~

0

1

1

12

8

4

1

1

16

,

1

MINUTES b C I

I

CRJ

I

I

I

I

1

0

2

4

6

0

MINUTES

I

24

20

Comparison of novel stationaryphases

397

polymer-modified phase, but loading is better for the polymer-modified phase. These results were attributed to the presence of heterogeneous sites on the untreated and Hmodified phases. Flow rates are more homogeneous for the H-modified than the untreated phase. For all three phase-support combinations, alkylbenzenes, alkyl phenyl ethers, and phenyl alcohols are separated with good efficiency. The separation of isomers was compared on untreated, vapor-deposited carbon zirconia to the same separation on ODS [70]. The mobile phases used were acetonitrile-water and tetrahydrofuran-water. To achieve similar solute capacity factors, slightly higher organic concentrations were used with the carbon-based phase, 50% compared to 40% for acetonitrile and 35% compared to 30% tetrahydrofuran. On the ODS phase, k is smaller for nitrobenzene than for benzene in both mobile phases, while the opposite is true for the carbon-based phase. Increased solute dipolarity decreases retention on the ODS phase but increases it on the carbon-based zirconia phase. For all the groups of isomers studied, the isomers are separated as well or better on the zirconia phase than on the ODS phase and oRen in less time. For example, a separation of a mixture of m-, 0-,and p-xylenes was attempted on both phases in an acetonitrile-water mobile phase (Fig. 11.17). On the ODS phase, the m- and p-xylene are not separated, while on the carbon-based phase, although not quite baseline resolved, distinct peaks for all three xylenes are seen. The elution order for both columns is o-, p-, and m-xylene as compared to an elution order ofpuru, metu, and ortho on a normal-phase silica or alumina-based column. If only electronic interactions between the aromatic ring and the carbon cause the separation, the order of elution should be the same. Therefore, hydrophobic interactions in addition to electronic interactions cause separation with the carbon-based phase.

11.5.3 Polymer-based stationary phases A number of organic, polymer-based stationary phases have appeared. These phases promise to extend the useful pH range of analyses by eliminating charged sites, while also providing a possible additional means of selectivity by providing a size-exclusion effect. Polymeric stationary phases, however, swell and shrink with changes with mobile phase composition [70]. Therefore, pore structure and column backpressures vary in gradient elution. The performance of several hydrophilic, porous polymer gels was compared to C1, C8, and CI8 silica-based stationary phases for the separation of alkanes, cycloalkanes, and planar and bulky polyaromatic hydrocarbons (PAHs) in methanol-water mobile phases [74]. Aliphatic solutes are less retained by 15-75% on the polymer-based phases. However, polymer-based stationary phases preferentially retain rigid aromatic solutes. The suggested mechanism was that separation occurs not only by electronic effects through the participation of Jt-electrons, but also by steric effects. Changes in the percentFig. 11.17. Separation of xylene isomers in an acetonitrile-water mobile phase. (a) Hypersil ODS column; (b) CVD carbon-clad zirconia column. (Reproduced from Ref. 70 with permission of the American Chemical Society.)

References pp. 399-401

398

Chapter I I

0.05

0

co

R1 W

0

1

i

30

60

ELUTION TIME (MIN) Fig. 11.18. Separation of proteins on a TSK Phenyl 5PW RP+ column and acetic acid gradient. (Reproduced from Ref. 76 with permission of Elsevier Science).

age of methanol cause significant changes in structure, and therefore, selectivity of the phases. Polyvinylalcohol Asahipak stationary phases (C18,C8, C4) [75] were compared to silica-based columns (CIS,C8, C4, C,) for the separation of C-peptides, insulins and proinsulins. Separation conditions used were shallow trifluoroacetic acid (TFA)-acetonitrile and triethylammonium phosphate (TEAP)-acetonitrile mobile phase gradients and at temperatures of 25 and 45°C. At 25°C in the TEAP-acetonitrile mobile phase, only the shorter-chain polymer phases are able to separate the proinsulin, while only the C4polymer phase is able to separate the proinsulin at both 25 and 45°C in the TFAacetonitrile mobile phase. No irreversible binding of insulin and proinsulin is found, even when no ion-pairing reagent was added to the mobile phase, unlike the behavior found for c 1 g and C8 silica-based phases. This difference was attributed to the lack of silanols on the polyvinylalcohol surface. Recovery is high for both types of supports. Welinder [76] also studied the use of a variety of different commercial polymer-based stationary phases with acetonitrile mobile phases for the separation of polypeptides. The polymer-based supports appear to use different separation mechanisms. TSK Phenyl5PW RP+ columns separate proteins well with an acetic acid gradient (Fig. 11.18). Because high-molecular-weight samples, such as transferrin and catalase, have good peak shape, not only does the phenyl-bonded phase contribute to separation, but so does the polymer support. Other polymer-supported phases, such as Asahipak C4, C8, and Clg, appear to separate materials only by the hydrophobicity of the attached ligand with no interaction with the polyvinyl alcohol support.

Comparison of novel stationaryphases

399

Octadecyl-polyvinyl copolymer (ODP) and octadecylsilane (ODS) stationary phases were compared for their ability to determine the lipophilicities of benzene derivatives [77]. The lipophilicities measured on the ODP column are comparable to those obtained on the ODS column. However, no masking agents are needed with the ODP column due to the lack of reactive groups, unlike the silanols found on the ODS phase.

11.6 CONCLUSION

Although the ODS bonded phase is still the most widely used column material and reverse phase predominates in a vast majority of applications, improvements or variations of this phase still are being introduced each year. In addition, new bonded organic moieties are constantly being synthesized to allow additional new interactions with the solute and to shield solutes from reactive silanols. Further changes in bonded phase technology are being made by the introduction of materials based on supports other than silica or composed of an organic polymeric backbone that may or may not contain specific functional groups. In spite of all the activity involved in the synthesis of new bonded phases and attempts to improve such characteristics as stability, efficiency or resolving power, there is no set protocol which allows meaningful comparisons between new and existing materials. In most cases, reports of new or improved phases consist of a few separations that are of particular interest to the investigator. Occasionally spectroscopic characterization utilizing such methods as DRIFT and CP-MAS NMR are reported. Therefore, someone wishing to determine which stationary phase is most suitable for a particular separation problem must often choose using a limited amount of non-uniform information. Due to the great diversity of applications reported for HPLC, it is likely that this situation will continue for the foreseeable future.

11.7 REFERENCES 1

2 3 4

5 6 7 8 9 10 11 12 13 14 15 16

J.E. Sandoval and J.J. Pesek, Anal. Chem., 63 (1991) 2634. K.B. Sentell, D.M. Bliesner and S.T. Shearer, in: Chemically Modified Surfaces, Special Publication 139, Royal Society of Chemistry, 1994, p. 190. J.J. Pesek and S.A. Swedberg,J. Chromatogr., 361 (1986) 83. J.J. Pesek, in: Chemically Modified Surfaces, Special Publication 139, Royal Society of Chemistry, 1994, p. 1. G.E. Berendsen and L. deGalan, J. Liq. Chromatogr., 1 (1978) 561. V. Pachkovb and L. Feltl, ChromatographicRetention Indices, Ellis Horwood, 1992. P. Jandera, J. Chromatogr., 91 (352) 11 1. F. Morishita, H. Kahihana and T. Kojima, Anal. Lett., 17 (1984) 2385. J.K. Baker and C.-Y. Ma, J. Chromatogr., 169 (1979) 107. R.M. Smith, J. Chromatogr., 236 (1982) 313. R.M. Smith, J. Chromatogr., 291 (1984) 372. B. Vonach and G. Schomburg, J. Chromatogr., 149 (1978) 417. P.E. Antle, A.P. Goldberg and L.R. Snyder, J. Chromatogr., 321 (1985) 1. J. Wilken, Z. Anal. Chem., 320 (1985) 696. M.L. Lee, D.L. Vassilaros, C.M. White and M. Novotny, Anal. Chem., 51 (1979) 768. D.L. Knarlesky, D.C. Kelly and I.M. Warner, J. Liq. Chromatogr., 6 (1983) 471.

400

Chapter I I

17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 3s 36 37 38 39 40 41 42 43 44 45 46 47 48

C.N. Ho, D.L. Karlesky, J.R. Kennedy and I.M. Warner, J. Liq. Chromatogr., 9 (1986) 1. A. Kaibara, M. Hirose and T. Nakagawa, Chromatographia, 30 (1990) 99. H. Engelhardt and M. Jungheim, Chromatographia, 29 (1990) 59. R.J.M. Vervoort, F.A. Maris and H. Hindriks, J. Chromatogr., 623 (1992) 207. M. Stadalius, J. Berus and L. Snyder, LC GC, 6 (1988) 494. J. Kiel, S. Morgan and R. Abramson, J. Chromatogr., 320 (1985) 313. T. Ascah and B. Feibush, J. Chromatogr., 506 (1990) 357. J.J. Kirkland, C.H. Dilks, Jr. and J.E. Henderson, LC GC, 11 (1993) 290. D.C. Locke, J.T. Schmermund and B. Banner, Anal. Chem., 44 (1972) 90. K. Unger, W. Thomas and P. Adrian, Kolloid-Z. Z. Polym., 251 (1973) 45. J.E. Sandoval and J.J. Pesek, U.S. Patent 5017540,1991. K.K. Unger, Porous Silica, Elsevier, Amsterdam, 1979. L.J.R. Barron, G. Santa-Maria and J.C. Diez Masa, J. Liq. Chromatogr., 10 (1987) 3193. R.K. Gilpin and M.E. Gangoda, J. Chromatogr. Sci., 28 (1990) 277. S. Linde and B.S. Welinder, J. Chromatogr., 548 (1 191) 195. N.E. Zhou, C.T. Mant, J.J. Kirkland and R.S. Hodges, J. Chromatogr., 548 (1991) 179. R. van der Zee, T. Hoekzema, S. Welling-Wester and G.W. Welling, J. Chromatogr., 368 (1986) 283. N. Tanaka, K. Hashizume and M. Araki, J. Chromatogr., 400 (1987) 33. R.T. Krause, J. Liq. Chromatogr., 12 (1989) 1635. C-J. Xu and J-T. Lin, J. Liq. Chromatogr., 8 (1985) 361. J.L.G. Thus and J.C. Kraak, J. Chromatogr., 320 (1985) 271. J.J. Pesek, P. Mahabadi and S.-J. Chan, Chromatographia, 23 (1987) 3. J.J. Pesek, E.L. Tarver and A. Lange, Chromatographia, 24 (1987) 815. J.J. Pesek and G. Guiochon, J. Chromatogr., 395 (1987) 1. J.J. Pesek and W.Y. Wey, Chromatographia, 25 (1988) 969. R. Abuelafiya and J.J. Pesek, J. Liq. Chromatogr., 12 (1989) 1571. J.J. Pesek and E.J. Williamsen, Trends Anal. Chem., 11 (1992) 259. J.J. Pesek and T. Cash, Chromatographia, 27 (1989) 559. J.J. Pesek and A.M. Sioufi, Anal. Chem., 61 (1989) 1928. J.J. Pesek, Y. Lu, A.M. Sioufi and F. Grandperrin, Chromatographia, 31 (1991) 373. J.J. Pesek, M.A. Vidensek and M. Miller, J. Chromatogr., 556 (1991) 373. K. Jinno, Y. Saito, R.M. Chopra, J.J. Pesek, J.C. Fetzer and W.R. Biggs, J. Chromatogr., 557 (1991) 459. K. Jinno, H. Mae, Y. Saito, J.J. Pesek, J.C. Fetzer and W.R. Biggs, J. Microcolumn Sep., 3 (1991) 417. A.M. Rizzi, J. Chromatogr., 542 (1991) 221. H. Bruckner, Chromatographia, 24 (1987) 725. I. Marle, A. Karlson and C. Pettersson, J. Chromatogr., 604 (1992) 185. L. Oliveros, C. Minguillon, B. Desmazieres and P.-L. Desbene, J. Chromatogr., 606 (1992) 9. W.H. Pirkle, D.W. Houseand J.M. Finn, J. Chromatogr., 192 (1980) 143. W.H. Pirkle, J.M. Finn, B.C. Hamper, J. Schreiner and J.R. Pribish, in Asymmetric Reactions and Processes in Chemistry, ACS Symp. Ser. 185, American Chemical Society, Ch. 18. H.B. Weems, M. Mushtaq, P.P. Fu and S.K. Yang, J. Chromatogr., 371 (1986) 211. B. Blessington, N. Crabb and J. O’Sullivan, J. Chromatogr., 396 (1987) 177. K.M. Kirkland, K.L. Neilson and D.A. McCombs, J. Chromatogr., 545 (1991) 43. Y. Yamazaki, M. Uebayasi and K. Hosono, Eur. J. Biochem., 184 (1989) 671. Y. Yamazaki and K. Hosono, Biotechnol. Lett., 11 (1989) 679. Y. Yamazaki, N. Morohashi and K. Hosono, J. Chromatogr., 542 (1991) 129. J.A. Dean (Ed.), Lange’s Handbook of Chemistry, 13th edn., McGraw-Hill, New York, 1985. M.P. Rigney, T.P. WeberandP.W. Car, J. Chromatogr., 484(1989)273. R. Kaliszan, J. Petrusewicz, R.W. Blain and R.A. Hartwick, J. Chromatogr., 458 (1988) 395. J.J. Pesek, J.E. Sandoval and M. Su, J. Chromatogr., 630 (1993) 95. J.E. Haky, A. Raghani and B.M. Dunn, J. Chromatogr., 541 (1991) 303. C.H. Chu, E. Jonsson, M. Auvinen, J.J. Pesek and J.E. Sandoval, Anal. Chem., 65 (1993) 808. J.E. Haky and S. Vemulapalli, J. Liq. Chromatogr., 13 (1990) 31 11.

49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

Comparison of novel stationary phases 69 70 71 72 73 74 75 76 77

C. Hansch and A. Leo, Log P and Parameter Database, Comtex Scientific Cop., New York, 1983. T.P. Weber and P.W. Carr, Anal. Chem., 62 (1990) 2620. F.A. Cotton and G. Wilkinson, Advanced Inorganic Chemistry, 5th edn., Wiley, New York, 1988. M.P. Rigney, T.P. Weber and P.W. Carr, J. Chromatogr., 484 (1989) 273. T.P. Weber, P.W. Carr and E.F. Funkenbusch, J. Chromatogr., 519 (1990) 31. N. Tanaka, K. Hashizume and M. Araki, J. Chromatogr., 400 (1987) 33. S. Linde and B.S. Welinder, J. Chromatogr., 548 (1991) 195. B.S. Welinder, J. Chromatogr., 542 (1991) 83. A. Bechalany, A. Tsantili-Kakoulidou, N. El Tayar and B. Testa, J. Chromatogr., 541 (1991) 221.

40 1

This Page Intentionally Left Blank

R.M.Smith (Ed.), Retention and Selectivity in Liquid Chromatography Journal of ChromatographyLibmy, Vol. 57 0 1995 Elsevier Science B.V. All rights resewed

403

CHAPTER 12

Multivariate characterization of RP-HPLC stationaryphases Annabel Bolcka and Age K. Smildeb aResearch Group Chemometrics, University Centrefor Pharmacy, University of Groningen, Antonius Deusinglaan 2, 9713 A W Groningen, The Netherlands and bLaboratotyfor Analytical Chemistry, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands

12.1 INTRODUCTION

One of the problems in RP-HPLC is the reproducibility of retention values between different batches of stationary phases. Even stationary phase materials of the same brand differ between batches of the same material [ 1-31. One of the major causes for the variations is the presence of free silanol groups at the surface of the stationary phase materials. The amount and place of the free silanol groups differs for every stationary phase. Recently, various manufacturers have launched specially prepared columns, claimed to be free from silanol effects [4,5], but still large differences in retention values on columns of different brands and batches were observed. There is a need for “measurement-independent” retention values to make retention values measured on stationary phases of different brands and batches comparable. In the previous chapters some standardization methods are shown. Besides relative retention, relative capacity and corrected capacity factors, some retention index systems are mentioned. Although the retention indices can be considered as good “measurementindependent” retention values, there are still a few problems. One of the problems is the fact that retention indices are dependent on the dead time. These dead times vary a lot which can cause variations in the retention indices. Another problem is the fact that retention times are based on homologous, not on completely different standards. Some homologs used for retention indices show a discontinuity in their linearity. This means that retention times of, for instance, alkanes with a Iow chain length have a different linear relationship with the retention indices than alkanes with a long chain [6]. Furthermore, the concept of retention indices can only be applied to separations based on partition chromatography and therefore cannot be used in straight-phase chromatograP ~ r731. Y References pp. 447-449

404

Chapter 12

Finally, the indices are still sensitive to the kind of organic modifier and other properties of the mobile phase and the make of stationary phase material [3,9-121. In systematic toxicological analysis, for instance, the retention indices cannot compensate for large differences in selectivities of nominally identical, but commercially different reversed-phase column packings [13-151. Previous chapters gave solutions to some of the problems, an overview on the standardization of chromatographic methods for screening analysis in toxicology is given by Bogusz [ 131. This chapter presents another way of dealing with the problems presented above. With the use of multivariate statistical techniques, retention values of one chromatographic system are transferred to another chromatographic system [ 16,171. This possibility of predicting retention times of solutes at varying mobile phase compositions is stressed. First, a description is given of what multivariate characterization is. Some multivariate techniques are outlined. Secondly, it is described how some specific compounds (markers) can be chosen with different techniques and how these markers can be used to predict retention values of other solutes (non-markers). Finally, some practical examples are given. 12.2 MULTIVARIATE CHARACTERIZATION

To characterize an object (e.g. a nail), measurements can be done on this object. If one characteristic (e.g the length of the nail) is used only one measurement is necessary. This is a univariate characterization. The object can also be described by more characteristics (e.g. length, thickness and weight). If measurements, which together describe the object, are done, the object is characterizedmultivariately. The characterization of a molecule can be done, for instance, only by measuring the molecular weight. But to get a better idea, measurements on the spatial structure, dipole moment, etc. can also be done. A univariate characterization of a stationary phase can be done for example by the measurement of a retention value (retention time or capacity factor) of a solute. It is obvious that one such measurement gives a limited characterization of a stationary phase. It is better to measure the retention value of a solute in duplo or triplo. The mean of the retention values can characterize the stationary phase. The standard deviation is a measure of the random variance or repeatability of the measurement. Measurements done by one analyst can give different results from the measurements done by another analyst. Or measurements done in one laboratory can give different results from measurements in another laboratory. To overcome these problems, the retention value of the solute can be measured by different analysts and/or in different laboratories. The standard deviation of the measurements is a measure for the reproducibility. Another possibility to characterize stationary phases is to measure the retention values (retention times, relative retention times, capacity factors, relative capacity factors or corrected capacity factors) of a number of solutes. This can be done at one mobile phase but also at various mobile phases that differ for instance in modifier and/or pH. Measuring a number of solutes at different mobile phases is a multivariate characterization of the stationary phase, and the data (the retention values) are called multivariate

Multivariate characterization of RP-HPLC stationary phases testsolute 1

405

testsolute 14 at M.Ph 1

Fig. 12.1. A data matrix with the logarithm of the capacity factor of 14 test solutes at 9 mobile phases

data. With chromatographic data, a single measurement is never sufficient to characterize a stationary phase. Many measurements give more information than single ones. Multivariate data are mostly represented by data matrices. Figure 12.1 gives an example of a data matrix. The logarithm of the capacity factors of a number of solutes at various mobile phase compositions are given. A data matrix like that presented in Fig. 12.1 contains a lot of information. It contains information to characterize a stationary phase. This is information fiom the behavior of the solutes on this stationary phase: different solutes have different capacity factors, and this is information fiom the mobile phases: the capacity factor of one solute varies over the mobile phases. All these variations characterize the stationary phase. One way to summarize the information of the data matrix statistically, is by taking the total variance of the data matrix. The total variance is a measure for the variation in the whole data matrix. This total variance consists of some random variance and some systematic variance. Data always contain random variance. Random variations in the data cannot be assigned to a limited number of causes. There are a lot of causes for the random variations, such as measurement errors, changes in experimental conditions, etc., but it is not possible to find a limited number of causes, such as temperature and pressure only; it is always a combination of causes. The random variance in one laboratory can be calculated by taking the variance between the retention measurements and its duplos. The random variance between laboratories can be measured by taking the variance between the retention measurements of the laboratories. The systematic variance can be assigned to a limited number of causes. These causes can be the various mobile phases and solutes, the type of stationary phase, the batch from which the stationary phase comes, etc. The systematic variance can be calculated as the difference between the total variance and the random variance. A data matrix, like Fig. 12.1, contains random variance and systematic variance. The systematic variance of this data matrix is of interest for the characterization of the stationary phase. With multivariate statistical techniques it is possible to find the structure in the data matrices that causes this systematic variance. 12.2.1 Some basic multivariate statistical concepts

Not every chemist is used to the basic statistical concepts necessary to understand the References pp. 447-449

406

Chapter 12

X

X

lrn

11

x ....... ... I1

X

X nl

nm

X Fig. 12.2. A data matrix and a vector.

multivariate techniques presented in this chapter. This section gives some fundamental concepts with examples. 12.2.I . 1 Data matrices; centering and scaling Data matrices have columns and rows. The columns of the data matrix are the variables, the rows are the objects. If a data matrix X has n objects and m variables this is a (n x m) data matrix. An (n X 1) or (1 X m) matrix (a row or a column of X) is called a vector. Figure 12.2 gives a (n X m) data matrix X, xu denotes the element of X corresponding to thejth variable and ith object. Also a (n X 1) vector xj, that is thejth column of the data matrix is represented. The matrix of Fig. 12.1 is a (9 X 14) data matrix of ln(k) values. In the following, a vector is denoted by a bold lower case character (e.g. x), a matrix by a bold capital (e.g. X) and lowercase italic characters are used for running indices and the number of variables, objects and components. Two important statistics to summarize the data are the mean and variance. The mean and variance can be calculated for the whole matrix, the objects and the variables. The mean of a variable F j = Zq,/n. The variance of a variable is X(xg - x,)*/(n- 1). The standard deviation of a variable is the root of the variance of a variable. Data matrices are often centered and sometimes scaled. Because the multivariate techniques presented here are usually concerned with relative differences between ob-

centering

scal i ng

Fig. 12.3. Centering and scaling of a data matrix.

Multivariate characterization of RP-HPLC stationary phases

407

jects, X is often column-centered. This means that the means E j of each variable are first subtracted from each column of X.Column-centered data have variables with zero mean. When the size of the data of the various columns differs much or other measurement units are used, the data are also scaled. Each column of X is divided by its standard deviation. Now each variable gets equal weight. Scaled data have variables with unit variance. Figure 12.3 gives an example of a (3 X 2) matrix that is first column-centered and then scaled. 12.2.1.2 Visualizationof data

Most data matrices have some structure, this cannot always be seen just by looking at the matrix. A possibility is to make a picture of the data. This can be done by taking the rows (the objects) of the data matrices as vectors and plotting these vectors. The values of the

a

b

x2 I

C

x2

Fig. 12.4.A data matrix (a) represented by the objects (b) and variables (c).

References pp. 447-449

01

408

Chapter 12

first column of the data matrix are the values of all vectors on the xl-axis, the values of the second column, the values of all vectors on the x2-axis and so on. An example is give inFig. 12.4. Let X be the (3 x 2) column centered data matrix of Fig. 12.3, then in Fig. 12.4b the xl-the axis and x2-axis are formed by the two variables (xl and xz). The points in the figure are the three objects. It is also possible to represent the two variables in a threedimensional picture with the ol, o2 and 03 axis formed by the three objects. The variables are represented with vector arrows (Fig. 12.4~). The length of a (variable) vector xi can be calculated by (12.1) where ’ stands for transpose and llxll is the norm of x, and xY.. .xn, are the elements of xY For example, the length of the first variable of the column centered data matrix of Fig. 12.3 is (12.2) The variables of the column centered data matrix of Fig. 12.3 have zero mean. If the data are column-centered then x i is a measure for the variance of vector xl, and x i x2 a measure for the covariance between the vectors x1 and x2. The covariance is high if the vectors x1 and x2 have (almost) the same direction (are highly correlated) or highly negative if x1 and x2 have (almost) an opposite direction (are also highly correlated), and zero if x1 and x2 are orthogonal (are uncorrelated). In Fig. 1 2 . 4 ~the two vectors have a covariance of -16, this is not zero, but as can be seen the vectors have not the same or opposite direction, so this is also not the highest possible covariance. The correlation coefficient is the standardized covariance, that is the covariance divided by the root of the variances of both variables x1 and x2. This correlation coefficient will always be between -1 and 1. All variances and covariances of a matrix X can be found in X’X. To be complete, the matrix S = (I/(n- 1))X’X is called the variance-covariance matrix and contains all possible variances and covariances between the columns of X. The diagonal elements are the variances, the other elements the covariances. But often it is convenient to use X‘X instead of ll(n - 1)X’X (and x i xi instead of ll(n - 1) x i xi). The results only differ by a scalar; when comparing variances and covariances this scalar is of no need. Of course, if the exact value of a variance-covariance matrix of a matrix X or the exact value of the variance of a vector xi is required, this scalar cannot be omitted.

Fig. 12.5.An example of the identity matrix I.

Multivariate characterization of RP-HPLC stationary phases

409

If variables are orthogonal and of length 1, the variables are called orthonormal. A matrix is called orthogonal if all variables and objects are orthonormal. This means that X’X = XX’ = I, with I the identity matrix, a diagonal matrix with the diagonal elements 1 and the off diagonal elements 0 (see Fig. 12.5). For good experiments, this also includes multivariate characterizations, and experimental designs are needed. The idea of the experimental designs is to choose a minimum number of experiments that gives maximal information. This means that in the case of multivariate characterization, experiments are chosen efficiently in such a way that not too many measurements are done, but enough to find the systematic information (variation). The most simple and best known design is the factorial design [13]. With factorial designs, the variables are chosen orthogonal. The following example will explain this. Suppose, for a certain multivariate characterization, measurements are done on temperature and concentration. The most simple factorial design measures both variables on two levels (Fig. 12.6). The design can also be presented by a data matrix. In this example, it is possible to make a picture of the objects, but it is difficult to make a picture of the variables. After column centering the data, it can be seen that xi x2 = 0. This means that without making a picture, it can be said that the vectors x1 and x2 are orthogonal. The variables are not correlated. The more two variables are correlated, the more of the same information these variables contain. The purpose of multivariate characterization is to get more information by taking multivariate data, not to get more of approximately the same, that is redundant, information. Because the variables here are orthogonal, the variables give all new information. A factorial design is a design for efficient experiments, because it does not contain redundant information. Another design is a mixture design [14]. This is an efficient design for mixtures, like

, concentration

10

u

-

1

centering

ternperature

I

Fig. 12.6. A factorial design with two variables on two levels.

References pp. 447-449

410

Chapter 12

the components of a mobile phase. In most of the examples of Sections 12.2-12.5 these mixture designs are used. The examples presented in Section 12.2.1 are very simple examples; reality is more complex. In most cases, a data matrix is much more complex with more objects and variables, and variables cannot be chosen orthogonal. Figure 12.1 shows a data matrix to characterize a stationary phase with 14 test solutes as variables and 9 mobile phases as objects. In the case of Fig. 12.1, it is not easy to make pictures, because it is difficult to plot more than three dimensions. On the other hand, some variables are highly correlated and contain a lot of redundant information. When large data matrices are used, as in Fig. 12.1 or much larger matrices of, for example spectral data, it is not easy to see the relevant information in these data matrices. Often they contain a lot of redundant information. Principal component analysis, the technique presented in the next section, is a technique that searches for underlying factors (latent variables) of a data matrix that explain as much of the total variance (total information) of the data matrix as possible. It is a dimension reduction technique. Instead of enormous amounts of variables, it limits the data matrix to a few (two or three) latent variables. 12.2.2 Principal component analysis

Principal component analysis (PCA) is a special case of factor analysis (FA). With factor analysis, and also PCA, it is possible to search for the structure in a data set. Here attention is focussed on PCA, because this is the most used in chemical applications. An extensive description of PCA is given by Jolliffe [20], Mardia et al. [21] and Krzanowski [22]. 12.2.2.1The concept ofprincipal component analysis

Figure 12.7a gives a two-dimensional example. The n objects of X, a (n X 2) columncentered data matrix, are plotted. The xl-axis and x2-axis are formed by the two variables. The variables both have considerable variation, but are also highly correlated, the points in the plot are not randomly spread but lie more or less on a line. If principal component analysis is done on this data set, a maximum of two latent variables or principal components (PCs) can be found. The first PC tl is chosen in the direction of the most variation of the data set. It will explain a lot of variance, more than x1 or x2 separately. The second PC t2 is chosen uncorrelated with t, in the direction of the most variance left. There will be not much variance left for the second PC t2. To explain most of the variance in X it is almost possible to drop tZ.This can be seen in Figs. 12.7b,c. To find the principal components of a (n X m) column-centered data matrix X, the first step is to look for a vector tl = Xpl that is a linear function of X, which has maximum variance, where p1 is a vector of m constantspll,pI2,...,plmand llplll = 1 (see Fig. 12.8). This linear function Xpl is a new variable that is a weighted average of the original variables. The weights are chosen in such a way that the new variable has the largest possible variance and the weight vector has length 1. The length of the weight vector has to be constrained otherwise the weights will be infinite. This constraint does of course not affect the direction of the weight vector. The second step is to look for a linear function Xp2 orthogonal to Xpl which has maximum variance, and so on. The two vectors Xpl and Xp, are orthogonal if

Multivariate characterization of RP-HPLC stationary phases

41 1

i x1

I b

t2

x2

tl x1

C

t2

Fig, 12.7. (a) Representation of a data matrix with two variables (xi and xz). (b) The choice of principal components from a data matrix with two variables (XI and x2). (c) Representation of the data matrix with two principal components (tl and t2).

Referencespp. 447-449

412

Chapter 12 m

1

1

Fig. 12.8. The vector tl = Xpl is a weighted average of the variables of the matrix X with p1 the vector of weights.

(Xpl)’(Xp2) = 0. The kth linear function has maximum variance subject to being orthogonal to Xp,, Xpz,..., XpkThe kth linear function is called the Mh principal component (PC). This principle component is a latent variable, a new underlying variable. An underlying structure, not directly observable becomes manifest in the variables of the data matrix X, the elements of the principal components are in fact estimates of the so-called latent variables. The first few chosen PCs explain more information than the same amount of original variables. The original variables are often highly correlated, they contain a lot of redundant information. The PCs are orthogonal and chosen in the direction of the most variation. There are as many PCs possible as variables, but usually the first few PCs together explain an enormous part of the total variance of X. The total variance of X is the sum Z x; x, of the variances of all variables x, of column centered data. This is the sum of the diagonal elements X’X and is denoted by trace(X’X). If the sum of the variances of the first few PCs is almost the same as the total variance of X, the PCs are chosen as the underlying factors of X that explain almost all the variance in X. The dimension is reduced from m variables to g latent variables. The g latent variables form a so-called rank g approximation of X (see Appendix A. 1). 12.2.2.2 The calculation of the principal components

Let X be a column centered (n X m) matrix and xJ be columnj (variablej) of the matrix X. Then x,, denotes the observation corresponding to thejth variable as measured on the ith object. The first PC Xpl is found by choosing p1 so that Xpl has maximum variance, under the constraint that p1 has to be of unit length. This means that (Xpl)’(Xpl) has to be maximized subject to pi p1= 1:

Multivariate characterization of RP-HPLCstationary phases

max(Xpl)’(Xpl) = max(p; X’Xpl) subject to p; p1 = 1

413

(12.3)

X’X is a symmetric matrix; it can be decomposed according to the spectral decomposition (see Appendix A.2)

with K‘K = KK’ = I and A a diagonal matrix with the so-called eigenvalues in descending order on the diagonal and the off diagonal elements all zero. The vectors kl to k, are the eigenvectors correspondingto the eigenvalues A1 to A2. Now p; X’Xpl can be written as

Taking y = K’p, it follows that

y’hy = x y : A , = 1,

(12.6)

i

because y’y = pi KK’pl = pi Ipl = pi p1 = 1. Thus the maximum of pi X’Xpl is 11, this maximum is attained by taking for the first PC of X(pl) the first eigenvector of X’X, which is k l . This can be seen by

The other PCs are derived in the same way. Take pk = b.Then p i Pk = 1 and p ; p I = 0 with k # 1. And it can be proven that p i x‘xpk has maximum variance subject to the constraints for all k. Furthermore p i X’Xp, = 0. The vectors pk are called the loadings, and the vectors tk = xpk the scores, k = 1,...,n. The (m x m) matrix P is the matrix of all m loading vectors and T = XP is the matrix of all m score vectors. Now it can be seen that X can be decomposed in scores and loadings:

Each component tipi has rank 1 because each latent variable represents one dimension. All latent variables are orthogonal, so m latent variables represent m dimensions (see Appendix A. 1). Each component contains the information explained by one latent variable when the information explained by former variables is excluded. When only the first g components are used, X can be written as

where E is the (n X rn) matrix of residuals. The model is called a rank g approximation of X. This is visualized in Fig. 12.9. Referencespp. 447-449

414

Chapter 12 m

m

PO

+

+

+.......+

lI

n

ta

t2

E

Fig. 12.9. X is decomposed in scores (tl to tg) and loadings (PI to pg). This is a rank g approximation. E is the matrix of residuals.

The singular value decomposition (SVD) presented in Appendix A.3 is an alternative for finding principal components of a matrix X. The solution is not unique. Multiplying Tgwith a full rank matrix S and subsequently P, with (S-l) gives

where

Tg and fig give as good a separation of X as do Tg and Pr

12.2.2.3 A least squares interpretation ofprincipal component analysis

The PCA model

is built in such a way that the scores and loadings explain as much as possible of the total variance of X ( = X x i x,). This means that the total variance of the residual matrix, that is the s u m of the squared residuals Xei e,, is minimized. PCA can be considered as a socalled “least squares” method (see Section 12.4.1). Figure 12.10 makes this clear. In Fig. 12.7 the first PC t, is chosen in such a way that the variance is maximized. Figure 12.10a gives the residuals after this first PC. All other linear hnctions of the variables x1 to x,, give larger residuals (Fig. 12.10b). The variance explained by the first PC is maximized and the variance of the residuals of X after extraction of the first PC is minimized. For an element x,, of X the PCA model can be written as

-_-

...

.. .. . ’ . . ,. . 0

.

._1_-_’_..%

_.L-

’

* . 0

.

-

_. .

i.

. . .. * ’

-__

~

..

* . *

0

9

Multivariate characterization of RP-HPLC stationary phases

415

TABLE 12.1 THE SOLUTES USED IN THE PCA EXAMPLE AN BAB CRE DIA DMP ETB NBA

Aniline n-Butyl p-aminobenzoate p-Cresol Diazepam Dimethylphthalate Ethyl benzene p-Nitrobenzaldehyde

NBZ NIN PE PHB PRE PRS TOL

Nitronaphthalene Nitronaphthalene 2-Phenylethanol n-Propyl4-hydroxybenzoate Prednisone Prednisolone Toluene

g

(12.10)

x rl. . = C t i k p k j+eij

k=l

xu is the observation corresponding to thejth variable measured on the ith object, tikis the

ith element of the kth score vector and p v is theJth element of the kth loading vector. The residual, the real value minus the model value, corresponding to thejth variable measured on the ith object is denoted by eU.

12.2.3 A principal component example As an example with chromatographic data, a data matrix with ln(k) values is used [16]. The capacity factors of 14 test solutes are measured on a octyl silane reversed-phase stationary phase at nine different mobile phases. The purpose is to find the structure in this data matrix. Table 12.1 gives the 14 test solutes. The mobile phases are chosen according to a constraint mixture design [19] as shown in Fig. 12.11.

a

Water

A

b ACN

warn2 0.57

ACN

MeOH

Fig. 12.11. The mixture design of the nine mobile phases.

Referencespp. 447-449

0.00 0.00 0.00 0.23 0.19 0.15 0.46 0.38 0.30

MeOH

0.28 0.24

0.00

Chapter 12

416

Assuming that the relative error of a measured capacity factor is independent of the mobile phase composition, the specific solute and stationary phase, a logarithmic transformation of k gives ln(k) values with a constant variance [23-251. The ln(k) values of the 14 test solutes at the 9 mobile phases are arranged in a (9 x 14) data matrix as presented in Fig. 12.1. The data matrix is then column centered; no scaling was performed because all ln(k) values are measured on the same scale and do not differ greatly in magnitude. It is impossible to make a 14-dimensional picture of the data. With the use of PCA, a few latent variables will be calculated that explain almost all the variance of the data set. It appears that only the first two principal components (PCs) together explain 98% of the total variance of the data matrix. Now a picture can be made based on the two new PC axes (Fig. 12.12). Figure 12.12 (like Fig. 12.7~)is called a score plot. The structure in the score plot resembles the mixture design. It appears that the first PC can be associated with the water fraction in the mobile phase and can be interpreted in terms of solvent strength and solvent strength selectivity. This means that the first PC explains the variations in the data caused by solvent strength and the behavior of the test solutes at the used mobile phases on this stationary phase. It explains 84% of the total variance. The second PC can be associated with the modifier type and can be interpreted in terms of modifier selectivity. This mean that the second PC explains variations caused by the behavior of different modifiers with the used solutes on this stationary phase. It explains 14% of the total variance. Solvent strength, solvent strength selectivity and modifier selectivity are found to be the underlying factors of data matrices of the type shown in Fig. 12.12 [26]. Solvent

S C

1.25

1

wal

Wa2

0

0

0

wa3 0

r S

0.50-

P warn3

2

-0.25-

0

WN13 0

warn1

.

wan2

0

wrn2 0

Wml

Fig. 12.12. A score plot of the first two PCs of the data matrix of Fig. 12.1.For abbreviations, see Fig. 12.11.

Multivariate characterization of RP-HPLC stationary phases

1

417

0.50 AN

I

I 0

a

NBA

0.23

i

I ~

NBZ a

PE

BAB

a

g

a

-0.05

PHB

I

S

a

DIA

I C

*

-0.33

PRE

I

PRS

- 0.60 0.170

a

a

I

I

,

1

,

,

,

r

0.194

0.218

0.241

0.265

0.289

0.313

0.336

0.360

loadings PCI Fig. 12.13. A loading plot of the first two PCs of the data matrix of Fig. 12.1. For abbreviations,see Table 12.1.

strength, solvent selectivity and modifier selectivity characterize the data matrix. These factors are the main cause for the systematic variance. To see what original variables have a high contribution to the new latent variables, the loadings have to be investigated. It is possible to make a loading plot. Figure 12.13 gives such a loading plot, it shows the contribution of the 14 original variables to the first and second PC . Here it can be seen that solutes like diazepam and aniline have high loadings on the first PC. They are the most extreme on the solvent strength and solvent selectivity axis. Prednisone, prednisolone and aniline load very high on the first and second PC. They have the most extreme values in the solute selectivity direction. The solutions presented here are not unique. Multiplying the loadings and scores with full rank matrices does not alter the quality of the approximation. AN, DIA, PRE and PRS will remain, however, the most extreme values in the score plot.

12.2.4 Three-way analysis

Three-way analysis is an extension of PCA and FA. Instead of a decomposition of a twoway data matrix in two directions, there is now a decomposition in three directions of a three-way data matrix. Investigations on three-way analysis and three-way data tables have been done by Law et al, Caroll, Kroonenberg, Harshman and Tucker [27-311. References pp. 447449

418

Chapter 12

n m Fig. 12.14.A three-way data matrix.

Applications in chemistry are given by Geladi, Wold, de Ligny, Smilde and Ohman [32381. In the following, a three-way matrix will be denoted by an italic underlined bold capital (e.g. Let X be a (n X m X Z) three-way data matrix as shown in Fig. 12.14. & can be decomposed in three directions (Fig. 12.15). The difference between object and variable is no longer always clear; that is why the three directions are all called modes. The three modes can for instance correspond to the solutes, the mobile phases and the stationary phases. X is often centered in the direction of the mode for which the differences are of primary interest. In order to analyze a three-way data matrix, a generalization of the PCA model is needed. Such a generalization is given by the trilinear model:

a.

S

xijk

= zaigbjgckg

(12.1 1)

+eijk

g=l

where xi/k is an element of & and e'ikan element of the three-way residual matrix _E. a,,, bjg,and ckg are elements of the loading matrices A (n X s) of the first mode, B ( m X s) of

IFig. 12.15. The decomposition of a three-way data matrix. The PARAFAC model.

I

Multivariate characterization of RP-HPLC stationary phases

419

Fig. 12.16. The unfolding of a three-way data matrix.

the second mode and C (I X s) of the third mode. A, B and C are chosen in such a way that the sum of squared residuals is minimized; this means that again the factors are chosen in such a way that they explain as much of X as possible. The trilinear model is also called the parallel factor analysis (PARAFAC) [27,30] model. Contrary to PCA, the solutions are unique. Another generalization of the PCA model is obtained by applying the principle of unfolding [33]. This is shown in Fig. 12.16. The three-way data matrix is unfolded in a two-way data matrix, then PCA is performed on the unfolded data matrix. The data matrix can of course be unfolded in three directions. The unfolding is done in such a way that the mode which is of primary interest will form the objects in the unfolded matrix. The generalization can be written as (12.12)

where tig is an element of the (n X s) score matrix T and vjkg an element of the (rn X I ) loading matrix Vg The loading matrices V1 to V, can be stacked in the three-way (m x I x s) loading matrix y. Here T and y a r e calculated in such a way that the residual sum of scares is minimized. The solution is not unique. Equation (12.12) is visualized in Fig. 12.17 for the data matrices.

Fig. 12.17. The decomposition of a three-way data matrix. The unfolded PCA model.

Referencespp. 447-449

420

Chapter 12

C18 CN ACP Fig. 12.18. The three-way data matrix. All entries are In@) values. For abbreviations, see text.

It is claimed [38-40] that the PARAFAC decomposition is more restrictive than the unfold decomposition, which seems to be a reasonable claim when considering the bilinear equation of the unfold and trilinear equation of the PARAFAC. This difference in restrictiveness has two different consequences. The unfold decomposition is more flexible than the PARAFAC decomposition, but leaves the principal components the freedom to explain the variation by two or three objects only. This is only reasonable if these objects are really informative, otherwise outlying behavior is modelled. The PARAFAC decomposition is more restrictive, explaining less of the total variation than the unfold solution, but the stationary phases contribute more regularly to the components.

12.2.5 A three-way analysis example

The data described in the example presented in Section 12.2.3 can be extended with data of different stationary phases [ 161. A three-way data cube, as presented in Fig. 12.18 below can be investigated with three-way analysis [37]. The purpose is still to find the structure in the three-way data matrix. The ln(k) values of 14 test solutes at 9 mobile phases and on 6 stationary phases are given. The test solutes are the same as presented in Section 12.2.3. The mobile phases are chosen according to the same mixture design as presented in Section 12.2.3 (Fig. 12.11). The stationary phases are all reversed-phase stationary phases with the same base silica structure (from the same batch). The phases were modified with trimethylsilyl bonded to the silica (Cl); hexylsilyl bonded to the silica, filly capped (C6); octylsilyl bonded to the silica, filly capped (CS); octadecyl bonded to the silica, fully capped (C18); cyanopropylsilyl (CN) bonded to the silica and phenylsilyl (PHE) bonded to the silica, partially capped. One of the possible generalizationsof PCA can be obtained by performing an ordinary PCA on the unfolded data cube as presented in Fig. 12.19. The purpose of this principal component analysis is to show differences between stationary phases; that is why the stationary phases form the objects in the unfolded data matrix. Because differences between stationary phases are of primary interest, the relative differences between stationary phases will be investigated. This means that the data are centered in such a way that for each j and k it holds that ZxUk= 0, where the summation

Multivariate characterization of RP-HPLC stationary phases

42 1

ACPwml ...... ....PHBwml..................................... ....................PHBam2

--

C18 CN PHE Fig. 12.19. The unfolded data matrix of Fig. 12.18.For abbreviations see text.

runs fkom i = 1,...,6, the stationary phases. The first two PCs of the unfolded data cube accounted for 99.7% of the total variation (respectively 98.6% and 1.1%). The score plots of the first two PCs are shown in Fig. 12.20. From Fig. 12.20 can be seen that the C6 and C8 stationary phases are very similar, while the other stationary phases are dissimilar. The most extreme ones are the C18, CN and C1 phases. The first PC may be associated with the hydrophobicity of the stationary phases. The CN and PHE phases are the least hydrophobic, whereas the C6 and C8 have a similar hydrophobicity although somewhat less than C18. C1 takes an intermediate position between C 18 and CN. The second PC is hard to interpret. The first PC explains considerably more than the second PC. Another possibility is PARAFAC. Again, because differences between stationary phases are of primary interest, the relative differences between stationary phases will be investigated, thus the data are centered in such a way that for eachj and k it holds that Z.qk = 0, where the summation runs from i = 1 ,...,6, the stationary phases. One component in this model explains 98.5% of the total variation of the three-way data matrix. Again C 18 and CN are extremes and C 1 hardly scores on the first component. Which of the models is best is not easy to say. It depends on many factors, which are not discussed here. 1

9

c

0.5

1 -

4'

5'

0

r e

0 -

2'. 3

S

cP

6

-0.5-

2 I

-1

-10

-8

I

-6

I

-4

1':

-2

0

I

2

4

6

scores PC1 Fig. 12.20. PCA score plot of the unfolded data matrix presented in Fig. 12.19. Legend: 1, C1; 2, C6; 3, C8; 4, C18; 5, CN and 6, PHE.

References pp. 447-449

422

Chapter 12

12.3 MARKER SELECTION

Retention indices are based on standards, multivariate characterization uses markers. Markers are a subset of a data matrix, with for instance retention values measured on one stationary phase. The markers are variables chosen from the data matrix in such a way that these variables represent the structure of the data best. With the use of multivariate statistical techniques, a model is built to relate the markers to the other variables (the nonmarkers). Such a model is called a calibration model and the markers and non-markers used to build this model are called the training set. The calibration model can be used for the prediction of variable values of other data matrices. Suppose retention values of some solutes measured at different mobile phases on one stationary phase are available (the training set). In order to predict retention values of the same solutes on a new stationary phase, a selection of a special subset of the solutes, that represent all the solutes, is made. The solutes in this subset are the markers. Measurements on these markers are done on both stationary phases and measurements on the non-markers are done on the first stationary phase. A calibration model is built to relate the non-markers to the markers of the training set, the first stationary phase. The retention values of the non-markers on the new stationary phase are then predicted with the calibration model and the information fiom the markers of the new stationary phase. A good selection of markers has to be done. The markers have to be chosen in such a way that they characterize the structure of a chromatographic system, thus in such a way that the chromatographic behavior of the non-markers can be derived from the chromatographic behavior of the markers. If a chromatographic system is represented by a data matrix that contains retention values of some solutes at different mobile phases measured on one stationary phase, then the solutes that cause most of the variation are responsible for the structure in the data matrix. When the markers are chosen, a calibration model is built to relate the markers to the non-markers. The idea is that, although the structure differs between different chromatographic systems, the relation between markers and non-markers stays more or less the same. The chromatographic behavior of the markers changes, from one chromatographic system to another, in such a way that it characterizes the change in structure in the chromatographic systems as a whole. The choice of the training set is as important as the choice of the markers. If only a data matrix with retention values of some solutes at different mobile phases on one stationary phase are used, then only variations caused by solutes and mobile phases are used to characterize the chromatographic system. This has to be considered when comparing stationary phases of different batches, type and age and when retention values on certain stationary phases have to be predicted with models based on another stationary phase. To be sure that some variations caused by stationary phases are also considered in the models, the training set has to be chosen in such a way that it contains measurements on more than one stationary phase. A selection of markers and training sets based on experience is possible, but often situations are too complicated for this. Then multivariate statistical techniques, as presented in this section, can help.

Multivariate characterization of RP-HPLC stationary phases

423

12.3.1 The choice of markers

Let X be a (n X m) column-centered matrix. For instance, retention measurements of 14 solutes at 9 mobile phase compositions on a stationary phase arranged in a data matrix X as depicted in Fig. 12.1 or Fig. 12.12, with the columns mean centered. The solutes are the variables. Strong relationships between those variables are present, hence the choice of markers comes down to choosing those variables which represent the structure of X best. Stated otherwise, select those solutes which have the highest marking power. In the following, a description is given of procedures to choose markers, which are partly based on the work of McCabe [41]. Suppose r variables are selected, then the X matrix can be partitioned as

with X1the (n X r ) matrix of markers and X2 the (n X (m - r)) matrix of non-markers. The corresponding variance-covariance matrix S = (l/(n - 1))X’X can be partitioned as (12.14) with SI1the (r X r ) variance-covariance matrix of the markers. 12.3.I . I The determinant criterion One criterion to select the markers X1is the determinant criterion [42,43]. The idea is to select those variables that have the highest generalized variance. The generalized variance of the selected variables is given by the determinant of Sll, that is det(SI1). The determinant of a matrix is a measure of how much the columns or rows of that matrix are the same. All combinations of r variables, out of m variables, are investigated and the combination with the highest det(SI1) is chosen. The determinant criterion for the selection of markers is suitable for situations where variation of retention is important. It should be stressed that the determinant criterion assumes fixed r because determinants of matrices of different sizes cannot be compared.

12.3.1.2 The induced variance criterion Another criterion is the induced variance criterion [42,43]. The idea is to select those variables that explain much of the total variance of X based on the so-called induced variance. The variance of the jth variable of X is sJ . This is the jth diagonal element of S. The total variance of X is Zjsj withj = 1,...,m. The variation of solutej can be partly explained by the variation in the r markers, because the solute is more or less correlated to the markers. In Section 12.2.1.2, it could be seen that the more variables are correlated the more of the same information these variables contain; in Section 12.4.1 it can be seen that one variable can be regressed on another variable. The more the variation of solutej is explained by the variation of the markers, the more the markers represent the solute. A References pp. 447-449

424

Chapter 12

measure for the correlation between variables is the correlation coefficient. The squared multiple correlation coefficient R2(xjXl) is a measure of the correlation between xj and the r variables of XI. The amount of variation in xj which is explained by the correlation between xj and XI is sji . R2(xj,X1), because the R2 represents the fiaction of explained variation in xi. The term sJ .R2(xj,Xl) is called the induced variance of X, on xi. If xj is a column of XI note then that the induced variance is exactly sji. The induced variance criterion leads to the definition of the percent explained variance P: i s J

P=

j=1

+

m C S J

.R2(Xj,XI)

j=r+l m

( 12.15)

The value of P can be calculated for all combination of r variables, where xi is not an element of X I . The combination of r variables with the highest P is chosen. To illustrate a weakness of the induced-variance and determinant criterion, suppose the following. From a set of four variables, two should be selected. The first two variables have reasonably high correlation and two correlation coefficients (say 0.9) and are therefore exchangeable. The final variable has no high correlation with any of the first three (say 0.4), so this variable can be regarded as “outlying”. If two of the variables are chosen according to the induced variance criterion, the final variable will be part of the optimal subset, because exclusion of that variable will give rise to a low P. The final variable is not chosen on the basis of its predictive power, but on the basis of its bad predictability. Tests on “outlying’ variables are needed. This will become apparent in the following, when particular solutes have outlying behavior. The same holds for the determinant criterion. Both criteria are also sensible for “chance results”. Hence, validation is important, 12.3.1.3Principal component analysis

Another possibility to select variables is to use principal component analysis. Markers have to characterize the structure of the chromatographic system. In Section 12.2 it can be seen that PCA tries to find latent variables that determine the underlying structure of a data matrix (chromatographic system). These latent variables are linear functions of the original variables. Based on these latent variables, there have to be found original variables that explain much of the underlying structure of the data matrix. Original variables that are the most similar to the latent variables are of interest. The original variables most similar to the latent variables can be found graphically, from a principal component loading plot of a data matrix. From the loading plot of the first two or three principle components, the most extreme variables are selected. If more PCs are important, a loading plot of the first and second PC and a loading plot of the third and fourth PC, for example, have to be examined to find the most extreme variables in the plots. An example is given in Fig. 12.21. The problem of this technique is that it is not always easy to depict the most extreme variables. It is difficult to make a selection without procedures to evaluate the selection quantitatively and to recognize “outlying” variables.

Multivariate characterization of RP-HPLC stationary phases

scores PC2

425

X X

*

*

*

6

. W

.*

0

K

Another problem is the non-uniqueness of PCA solutions. As presented in Section 12.2.3, a rotated score plot can give other extreme values. 12.3.1.4 Marker selection and three-way analysis

The variable selection problem can be generalized to the three-way case just by unfolding a three-way data cube to a two-way data matrix. Now a selection can be made from two of the three original modes, e.g. solute/mobile phase combinations can be selected which explain much of the total variation.

12.3.2 A marker selection example The purpose here is to find the solutes of a data matrix of a chromatographic system that represent the chromatographic system best. These solutes are the markers of the chromatographic system. The data matrix of the chromatographic system is the training set for a calibration model that relates the selected solutes to other solutes. It is desired that variations caused by the solutes, mobile phases and stationary phases are considered in the model. The model is used in Section 12.4.4 for prediction purposes, where retention values of solutes on certain stationary phases have to be predicted based on retention values of the same solutes measured on different stationary phases. In a laboratory, stationary phases as described in Section 12.2.5 (Cl, C6, C8, C18, CN and PHE) are often used [ 161. Three of the six stationary phases (C 1, C 18 and CN) are selected for the training set. More than one stationary phase is selected to consider variations caused by stationary phases in the calibration model, but not all stationary phases are selected, because this costs too much and maybe it is possible to predict retention values on the other stationary phases with the calibration model. The mobile phases are chosen according to a constraint mixture design (Fig. 12.22). The solutes chosen are acetophenone (ACP), n-butyl 4-aminobenzoate (BAB), ethyl 4hydroxybenzoate (EHB), paracetamol (PAR), 2-phenylethanol (PE), toluene (TOL), npropyl4-hydroxybenzoate (PHB), ethyl 4-aminobenzoate (EAB) and methyl 4-hydroxybenzoate (MHB). Referencespp. 447-449

426

Chapter 12

Water

22 wal

PI &gA,

r

yml

w a 2 k m T wml i

/

/

/ /

\ \

/

/

i

c1

wm2

*.+

C18 ,'

/

/i

PHB

I..,

[ 7' am2

\,, \

CN

'\

L

ACP -----

......................

23

1

[ 11:

MeOH

ACN

Fig. 12.22. The constraint mixture design of the marker selection example. Fig. 12.23. The data matrix of the marker selection problem.

A data matrix of In@) values as depicted in Fig. 12.23 is used. The data matrix has stationary phase/mobile phase combinations as objects, and solutes as variables. The data matrix is column centered. A first step in the marker choice procedure is the performance of a PCA on the training set. Figure 12.24 gives the score plot of the first two principal components. The first two PCs together explain 99.22%of the total variation (90.88%and 8.34%,respectively). In Fig. 12.24 a grouping of the observations can be seen according to the stationary phases. The data are spread regularly and no clear outliers are detected.

5 C

0

r e

14

4'

82

1 -

6' 74'

-

0.5

83

0 4.5

.

1

-6

-4

1

1 l.

77

46

7

-2

55

0

2

'54

4

I 6

scores PC1 Fig. 12.24. PCA score plot of the data matrix of Fig. 12.23. Legend: 1, Clwml; 4, Clwal; 6, Clwa2; 7, Clwml; 11, Claml; 14, Clam2; 46, C18wm2; 48, C18wml; 50, C18wa2; 53, C18wal; 54, C18am2; 55, C18aml; 67, CNwm2; 74, CNwa2; 77, CNam2; 81, CNwml; 82, CNwal; 83, CNaml.

Multivariate characterization of RP-HPLC stationary phases 1

I

-7'

0.0

-

0.6

-

'

0.4 -

9

0.2

n

-

S

11

0

P

c

421

-0.2

I

;.5

9'

2 -0.4

I

I

I

I

The contributions of the variables are displayed in the loading plot (Fig. 12.25). The most extreme variables are PAR and TOL. These solutes span together with one of the solutes MHB, PHB, EHB or PE the loading space. The loading of paracetamol is very extreme. This is maybe an outlying variable. Paracetamol is more polar than all the other solutes. It needs a weaker mobile phase to get more retention, but that conflicts with the resulting very high retention of the other solutes. The PCA approach for selecting variables is not robust against selecting outlying instead of representative variables. If the induced-variance criterion is used for selecting three markers, the variables selected will be BABYEHB and PAR Again paracetamol is chosen because of the deviating behavior of paracetamol and the dual character of the induced-variance criterion as discussed in Section 12.3.1.2. The solute PAR is not incorporated in each subset because of its predictive quality, but due to its unpredictable behavior. The choice of the number of variables can be made by examining the percentages of explained variation of the markers and principal components. The three variables chosen by the induced variance criterion explain 99.5% of the total variance. A number of three markers seems reasonable. If the determinant criterion is used to select three markers, PAR, TOL and MHB are selected. This subset induces 99.47% of the variance in the whole set. The markers selected with the determinant criterion also have a high induced variance, the markers selected with the induced-variance criterion, however, have a low determinant value. The subsets can only be compared completely when prediction results are known. Prediction methods are discussed in Section 12.4, but using techniques described there, such as PLS, it can be found that the solutes that are not contained in either of the chosen subsets of the determinant criterion and the induced-variance criterion are predicted very well. Only PE is better predicted with the determinant criterion than with the induced variance criterion. In this case the determinant criterion seems more preferable. The solutes MHB, EHB and PHB are members of a homologous series. Such a homologous series is often used to correct differences in retention times due to variation of stationary phases. Within the data matrix the homologs induce 91.2% of the total variation, which is slightly lower than the induced-variance-markers and determinant markers. The determinant of the homologs is considerably lower than that of the determinant References pp. 447-449

428

Chapter 12

markers. The induced variance is slightly lower than that of the induced-variance markers. The homologs have also to be evaluated to their predictive power. It is found that in this case and most other cases, the predictability of markers chosen with the determinant criterion or the induced variance criterion is higher or equal to that of homologs.

12.4 PREDICTIONS

In Section 12.2 some multivariate statistical techniques are given to characterize chromatographic data. This is done by trying to find the underlying structure of the data, and with that some latent variables that represent this structure. In Section 12.3 some techniques are given to find a subset of variables that represent the data best. This selection of markers is necessary for the multivariate techniques presented in this section on prediction. After a subset of markers is selected from a data matrix, a calibration model can be built of the relation between the markers and non-markers of this data matrix (the training set). If the same markers of another data set are measured, then with the calibration model, the non-markers of this data set can be predicted. This method of prediction together with some other prediction techniques are described here. In the previous sections, only one data matrix X was considered. Now the relation between one data matrix X to another data matrix Y is considered. For instance, the retention values measured on a stationary phase are related to the retention values on another stationary phase. The retention values measured on a stationary phase are also related to the mobile phase composition. For simplicity, fust the case with only one y variable is described. Two steps are of interest. First, the structure of the relation between X and y has to be found and modelled. Secondly, based on the model between X and y, new y values have to be predicted from new X measurements. In chromatography, models of the relation between the capacity factor of a solute (y) and the modifier fractions in the mobile phase (X) are well known [44-47]. With QSRR (quantitative structure-retention relationships) [48] related to QSAR (quantitative structure-biological activity relationships) [49] studies, relations between the structure of solutes and their chromatographic retention are modelled. Based on these models, predictions can be made. After the relation between the capacity factor of a solute and the modifier fractions of the mobile phase compositions is modelled, the capacity factor of this solute can be predicted at other fractions of the modifiers in the mobile phase. The models are mostly made with the use of linear regression techniques, called ordinary least squares (OLS). This method is described here. In this chapter, the interest is, in the first place, not focussed on predicting retention values from experimental circumstances like the mobile phase composition or QSRR relations, but the focus is on transferring retention values from one chromatographic system to another chromatographic system. This means that the retention value of one solute (y) or the retention values of more solutes (Y)of one chromatographic system have to be predicted using the retention times of another chromatographic system. This can be done with multivariate statistical techniques related to linear regression or principal component analysis.

Multivariate characterization of RP-HPLC stationary phases

r-----l

I

I

I

I

429

Calibration

I

I prediction

Fig. 12.26. The calibration and prediction model.

Markers have to be chosen that represent the structure of a chromatographic system best (see Section 12.3). Markers and non-markers have to be trained. This means that the retention values of both markers and non-markers are measured once and that the relation between the markers and non-markers is estimated. A statistical model is constructed for a chromatographic system. Once this calibration model is built, only measurements of the markers in a new chromatographic system are necessary to predict the retention times of the other solutes of that chromatographic system. Besides ordinary least squares (OLS), another technique that deals with these kind of problems, partial least squares (PLS), is described (see Fig. 12.26). In spectroscopy often a related technique to transfer models is used [50,51]. This method, multivariate instrument standardization, deals with problems such as transporting models developed on one instrument to other instruments. For instance, spectra on one instrument are shifted compared to spectra on another instrument. In this chapter, not chromatograms (pictures like spectra) but derived data such as retention times and capacity factors are used. This calls for methods as described in this chapter. Two techniques, ordinary least squares (OLS) and partial least squares (PLS) are described here.

12.4.1 Ordinary least squares (OLS)

Ordinary least squares (OLS) is a regression technique. With regression, it is possible to model relations between variables and predict values of other variables. In Sections 12.2.1.2 and 12.3.1.2, these regression techniques were already implicitly used. To illustrate this, linear regression is described. Suppose X is a ( n X rn) matrix and y is a ( n X 1) vector. A linear relation between the variables of X and y is assumed. This means that y can be written as a linear function of X and some error (see Fig. 12.27). Referencespp. 447-449

430

Chapter I 2

Y

.

b

Fig. 12.27.A linear relation between y and X assumed

Thus y can be written as Y=Xp+&

(12.16)

The value of/? can be estimated based on measurements done on X and y. The model is estimated by y=Xb+e

(12.17)

(see Fig. 12.28) with b an (m X 1) vector of estimated constants and e an (n X 1) vector of residuals, the difference between the real y and the estimated Xb. Here b is not chosen in such a way that Xb maximizes the variance in X, as with PCA, but b is chosen in such a way that Xb explains as much as possible of the total variance of y. This means that the correlation between y and Xb is maximized (Fig. 12.29). The correspondence with Sections 12.2.1.2 and 12.3.1.2 now becomes clear. In these sections, it was mentioned that the more correlation there is between two variables, the more of the same information these variables contain or the more variation in one variable is explained by the other variable. A measure for the correlation is the covariance or the correlation coefficient. With linear regression, a linear function Xb of X has to be found that explains as much as possible of the variation in y or has the highest possible correlation with y, Y

Fig. 12.28. The linear regression model for y

0

Multivariate characterization of RP-HPLC stationary phases

43 1

Y

I I

Fig. 12.29.The correlation between y and X is maximized.

Maximizing the correlation is, as with PCA, minimizing the sum of the squared residuals e. But now the residuals between y and Xb. The method is called “ordinary least squares”. b is chosen in such a way that the sum of squared residuals Ze’e is minimized. minxe’e = minx(y-Xb)’(y-Xb)

(12.18)

This leads to the classical Gauss-Markov [52] solution: b = (X’X)-’X’y. After the model is estimated ,values of y in a new situation can be predicted with the X values in the new situation and the model Xb. An example can be given with models [44] where X is a matrix with percentages modifier in a mobile phase, with nine mobile phases as objects and two modifiers as variables. The percentage methanol is the first column x1 of X, the percentage acetonitrile is the second column x2 of X. And suppose y is a vector with the ln(k) values of a solute at the nine mobile phases. A linear relation is assumed between the ln(k) values and the percentage modifier. The model can be written y=Xb+e or slightly differently, y = blxl + b2x2+ e

(12.19)

b = (bl,b2)’is chosen in such a way that the sum of the squared residuals is minimized. A new mobile phase has other percentages of the modifiers. Using the new percentages in the estimated model gives the predicted In@) value (y,,, = Plxlnew+ /?2x2new) of the solute at the new mobile phase. The relation between the ln(k) value and the mobile phase composition can be more complicated. For instance

y = blxl + bllxI2+ b2x2+,812xIx2+ e Referencespp. 447-449

(12.20)

432

Chapter 12

Regression of X on one y variable can be extended to regression of a matrix X on a matrix Y with the model Y = XB.This is not described here.

12.4.2 Partial least squares (PLS)

Partial least squares (PLS) was developed by H. Wold. Descriptions of the original PLS modelling strategies are given by Jtireskog and H. Wold [53]. The PLS algorithms, which are too complicated to describe here, are described by Geladi and Kowalski [54,55], S. Wold et al. [56,57], Htiskuldsson [58] and Manne [59]. 12.4.2.1PLSl

With PLS, as with OLS, it is possible to model relations between variables and make predictions. PLS is, like OLS, a least squares method, it minimizes the sum of squared residuals between y and a model of y based on X,but in contrast to the OLS method presented in Section 12.4.1, PLS is not a direct regression from X on y, but a regression of y on the latent variables of X. PLS also resembles also PCA, but with PCA the idea is to find latent variables that represent the structure within a data matrix X;the idea of PLS is to find latent variables that represent the relation structure between data matrices X and y. Again for simplicity, first a case is described where there is only one y variable. This is called PLSl. Suppose that X is a column centered (n X m) matrix and y is a mean centered (n x 1) vector. For instance, X contains the In(@ values at 9 mobile phases of 14 test solutes, and y contains the In(k) values at the same 9 mobile phases of another solute (Fig. 12.30). A model is estimated of the relation between y and X.This estimation is done in steps. The first step is, as with PCA, to look for a vector tl = Xwlthat is a linear function of X, where w1 is a vector of rn constants wll, wI2,..., wlm and llwlll=1. With PCA, w1 (in Section 12.2.2, this vector was called pl) is chosen in such a way that Xwl has maximum variance, so that Xwlexplains as much as possible of X.With PLS w1is chosen in such a way that the covariance between Xwland y is maximized, so that Xwl explains as much as possible of y. Thus, max cov(tl,y) = max cov(Xwl,y)

(12.21)

Again Xwlis a weighted average of the original variables, tl is the first score vector; 14

Y

I Fig. 12.30.An example of X and y.

Multivariate characterization of RP-HPLC stationary phases

43 3

w1 is not called the first loading vector but the first weight vector. The vector w1 is used to find the first score vector tl. Now X and y are regressed on tl to account for the described variation of tl. The results are the regression vector (or loading vector) p1 and regression scalar q1(for X and y, respectively). Hence the model can be written as

X=tlp; +E

(12.22)

Y = tlql+ f

(12.23)

with

t1= X W l

(12.24)

(see Fig. 12.31). After the first step, the same procedure can be repeated on the residual matrices E and f, and so on [53-591. The resulting model is

X=T,PL + E ,

y=T,q; + f

Both X and y are represented by a rank g approximation of latent variables. This procedure is called PLS1, because only one y variable is used. When more y variables are used, the procedure is called PLS2. This procedure, which is slightly different from the PLSl algorithm, is described below 12.4.2.2 PLS2

Suppose X is a column centered (n X m) matrix and Y is a column centered (n X k) matrix. Both matrices have the same objects but different variables. For instance, X is a matrix of retention times of 14 solutes at nine 9 mobile phases and Y is a matrix of retention times of 6 other solutes at the same mobile phase (Fig. 12.32). The PLS2 procedure is like the PLSl procedure except for the fact that there are now more y variables. This means that latent variables (T) have to be found not only for X,but Y

X

I

tlpl

tl

Fig. 12.31. X and y are regressed on tl.

Referencespp. 44 7 4 4 9

434

Chapter 12 8

Fig. 12.32. An example of X and Y.

also for Y. Latent variables of both matrices, instead of only latent variables of X, that represent the structure best between the data matrices X and Y are chosen. Xwl = tl is a weighted average of the original variables of X and Ycl = u1 is a weighted average of the original variables of Y. Now c1 and w1have to be chosen in such a way that the covariance between Ycl and Xwl is maximized: max cov(tl,ul) = max cov(Xwl,Ycl)

(12.25)

12.4.2.3 PLSpredictions

Suppose the PLS model is estimated for some column-centered (n X rn) X and some column-centered (n X k) Y, and the values of a new (n X rn) data matrix X, are measured. Now the values of the (n X k) Y, can be predicted. With the model, new latent variables can be calculated based on X,,, and based on this, new latent variables Y,, can be calculated. This is visualized in Fig. 12.33. 12.4.3 PLS predictions and marker selection

Let X be a column-centered (n X rn) data matrix. With the use of some variable selection

1i x , , U 1 .

I

I prediction

Fig. 12.33. APLS model.

I

Y,

I I

Multivariate characterization of RP-HPLC stationary phases

43 5

1 I

i

II

model

.,-,*.-,L ____

prediction -

- -

I - -I

Fig. 12.34.A PLS model with markers and non-markers.

criteria a few, say r, variables are selected to represent the structure of X best. The data matrix X is arranged in such a way that the markers are gathered in the (n X r) data matrix X[M] and the non-markers are gathered in X[NM]. This matrix X is called the training set. A PLS model is estimated based on this training set. The values of the markers of a new data matrix X,, are gathered in the (s x r) data matrix X,,,[M]. The values of non-markers gathered in X,,,[NM] can be calculated with X,,,[M] and the PLS model (Fig. 12.34). An example is given in Section 12.4.4.

12.4.4 A PLS example

The same laboratory as in the example presented in Section 12.3.2 is considered. The purpose is to transfer retention values from one chromatographic system to another chromatographic system. This means that for this laboratory, retention values measured on some stationary phases, that are used frequently, can be transferred to other stationary phases, that are also used frequently. The same data are used as in the example presented in Section 12.3.2. Here the training set was a data matrix with ln(k) values of nine solutes measured at six mobile phases on three different stationary phases C1, C18 and CN [16]. The solutes BAB, EHB and PAR were chosen with the induced-variance criterion as markers, the other solutes are the non-markers. The data can now be divided in a X[M]-matrix with markers and an X[NM]-matrix with non-markers (Fig. 12.35), but the data matrix of markers is extended compared to the markers used in Section 12.3.2. The fractions methanol (M) and acetonitrile (A) in the mobile phase are also used in the X[M]-matrix as predictors. Now the model accounts explicitly for the mobile phase variations. All the variables are column centered and scaled. This scaling is done because the ln(k) values of the solutes are measured on a different scale from the fractions modifier. The loading of a variable on the PLS component of a specific dimension measures the contribution of that variable to that PLS component. The contribution of a variable within References pp. 447-449

43 6

Chapter I2

Fig. 12.35. The training set for the PLS example.

the X[M]-matrix, can be interpreted as the amount of variation in that variable that is used to explain the variation in the variables in the X[NM]-matrix. The contribution of a variable within the X[NM]-matrix, can be interpreted as the amount of explained variation of that specific variable (Table 12.2). The value s2,,,,expl is calculated as the variance of the variable after the contribution of the variable to the three PLS dimensions (Table 12.2). In case of the variables of the X[M]-matrix these szuneXpl values can be interpreted as the unused variance in the modelling process. For the variables of the X[NM]-matrix, s ~ , , , , ~can ~ be interpreted as the variance unexplained by the model. Note that at the beginning of the estimation process each variable has variance 1, so that each s2,nexpl equals 1. The square root of s ~ , , , , , is ~~ reported in Table 12.2. After applying three dimensions, 99.42% of the variation in the X[M]-matrix is used to explain 99.21% of the variation in the X[NM]-matrix. The markers BAB and EHB contribute mainly in the first dimension (that is the first PLS component), the mobile phases mainly in the second one. The marker PAR plays a particular role and contributes to a large extent to the third dimension. Especially the non-marker TOL profits from this contribution because a high loading of TOL on this PLS component is observed. The X[M]-matrix and the X[NM]-matrix of Fig. 12.35 form the training set. The PLS model and a new X-matrix, X,,[M], with the same markers is used to predict the values of new non-markers in X,,,[NM]. In this example the same markers (A, M, BAB, EHB, TABLE 12.2 THE CONTRIBUTION OF THE VARIABLES TO THE SPECIFIC PLS DIMENSIONS Variable

Dim 1

Dim 2

Dim 3

Sunexpl

A M BAB EHB PAR

0.146 -0.197 0.603 0.615 0.446

-0.648 0.643 0.169 0.149 -0.339

0.495 -0.43 1 0.221 0.049 -0.72

0.1269 0.1311 0.0276 0.03 16 0.009

ACP PE EAB TOL MHB PHB

0.406 0.406 0.416 0.319 0.418 0.412

0.428 0.497 0.269 0.435 0.335 0.443

0.459 0.206 0.385 0.764 0.03 0.185

0.0872 0.1269 0.1034 0.1315 0.0616 0.0447

Multivariate characterization of RP-HPLC stationary phases

437

PAR) are measured at the same mobile phases on the new stationary phases C6, C8 and PHE. The non-markers of X,,[NM] are also the same as the non-markers of X[NM]. These non-markers of X,,,,[NM] can be predicted with the PLS model. To get an idea of the performance of this prediction, these non-markers can also be measured. Then the observed values can be compared with the predicted values, the difference between these values can be calculated. X,,,[M] and X,,[NM] are then called the test set (see Fig. 12.36). The performance of a prediction (the difference between observed and predicted values) can be calculated in different ways. Here the RMSEP, the root mean square error of predictions, will be used. The RMSEP is the root of the s u m of all the squared differences between the observed measured and predicted values summarized and divided by the total number of non-markers. If yii is the ijth element of the data matrix of observed values of X,,,,[NM] and y;; the ijth element of the data matrix of predicted values of X,,,[NM], then RMSEP = , / z ( y i i -yi;)2 l ( n x ( m - r )

(12.26)

When more prediction methods are used or different markers are compared, this RMSEP can be compared to find the best prediction method for this situation. The RMSEP can be measured for a data matrix as a whole, as in Eq. (12.26), but also for all different solutes. The RMSEP can also be calculated for the training set. The calibration model is then used to predict the non-markers of the training set with the markers of the training set. These predictions can then be compared with the already measured values of the non-markers of the training set. Table 12.3 gives the RMSEP of the different solutes and the mean RMSEP of all solutes and the test and training set.

c1

c18

{ XIMI

PLS

I,

XINMI

m

cb c8 PHE

Fig. 12.36. The training and test set for the PLS example.

References pp. 447-449

43 8

Chapter 12

TABLE 12.3 THE RMSEP VALUES OF THE DIFFERENT SOLUTES Solute

RMSEP train

RMSEP C6

RMSEP C8

RMSEP phe

RMSEP test

ACP PE EAB TOL MHB PHB Mean

0.0653 0.0831 0.0847 0.1767 0.0454 0.058 0.0957

0.0614 0.054 0,0909 0.2053 0.0861 0.0559 0.1062

0.0751 0.0424 0.0801 0.109 0.0471 0.0572 0.0721

0.1014 0.0626 0.0598 0.1385 0.0445 0.0647 0.0848

0.081 0.0536 0.078 0.1562 0.0622 0.0594 0.0888

12.5 PRACTICAL EXAMPLES

There is a need for “measurement-independent”retention values to make retention values measured on stationary phases of different brands and batches comparable. Previous chapters described possibilities to come to more or less measurement-independent retention values. This chapter has presented another way to deal with this. Multivariate statistical techniques were presented to characterize stationary phases and were used to transfer retention values from one chromatographic system to another. With the multivariate statistical techniques it is possible to find the underlying structure of different stationary phases, and retention values of stationary phases of different brands, batches, types and age car: be transferred. Here practical examples are given how to calibrate stationary phases of different brands, different types and different age. During routine analysis, sometimes columns age. This means that chromatograms can change, which can cause some problems. A new column can be used, but it is still possible that the chromatograms are not the same as before, because of column differences or batch differences if the stationary phase is from another batch. In these situations multivariate characterization techniques can help. It is possible with these techniques to relate chromatographic data from different columns to each other. This enables people to interpret new chromatograms in terms of old chromatograms or to predict new chromatograms. Suppose every day a number of solutes are measured on a particular column. To save time, it is possible to measure only a few solutes, instead of all solutes. These markers have to be well chosen. As could be seen in Section 12.3, the markers have to represent other solutes. The markers should contain information about the quality of the column. All variations in the retention values of the solutes caused by the stationary phase have to be represented by the markers, If the stationary phase ages in time or new columns are used, separations change and the mobile phase composition has to be varied to restore the separation. Retention values of other solutes on the aged stationary phases or new columns can be predicted from a calibration model, which describes the relation between the retention of markers and non-markers on the original column. Measurements of the markers on the aged stationary phase or a new column enables one to predict the retention of the non-markers.

Multivariate characterization of RP-HPLC stationary phases

439

The examples in this section deal with the calibration of stationary phases of different batches and types. 12.5.1 Calibration of octadecyl modified stationary phases of different batches

In a laboratory, octadecyl stationary phases are often used, but all measurements done on these phases have to be done again when new fresh octadecyl stationary phases are used, because these phases are from different batches. To prevent that this having to be done in future, a calibration model can be made, which makes it possible to measure only a part of the solutes on new stationary phases and predict the retention values of other solutes. The stationary phases chosen for the calibration model are different batches of octadecyl stationary phases of the same brand, denoted by C18 batches 1-4 [16]. Two batches of silica substrate material (Si batch 1 and 2) were used. Each silica substrate material was subjected to two separate silanizationprocedures. This resulted in four different silica substrate/octadecyl modified materials. Each combination was duplicated, so eight stationary phases were obtained. From these eight stationary phases, six were used as indicated in Table 12.4. The number of six was believed to be a good compromise between investigatingall four combinations in duplo and the amount of experimental effort. Measurements on these six stationary phases were performed by three different analysts, using different equipment, also indicated in Table 12.4, with analystlapparatus combination 13. Sixteen test solutes were chosen for the calibration, some benzene derivates and some other solutes like steroids. The solutes were acetophenone (ACP), acetanilide (ACT), anisole (ANS), p-cresol (CRE), nitrobenzene (NBZ), ethyl aminobenzoate (EAB), toluene (TOL), ethyl hydroxybenzoate (EHB), 2-phenylethanol (PE), methyl hydroxybenzoate (MHB), ethynylestradiol (EE), dimethyl phthalate (DMP), prednisone (PRE), prednisolone (PRS), phenobarbital (PBL) and propyl hydroxybenzoate (PHB). The mobile phases were chosen according to the constraint mixture design presented in Fig. 12.37. The h(k) values of the 16 solutes at the 9 mobile phases were measured on the 6 stationary phases and gathered in a (6 X 9 X 16) data cube a. This data cube can be unfolded in such a way that the direction of the stationary phases is left intact and the staTABLE 12.4 THE SIX STATIONARY PHASES USED FOR THE CALIBRATION OF OCTADECYL STATIONARY PHASES OF DIFFERENT BATCHES Stationary phase

C18 batch

Si batch

Analyst/apparatus

A B C D E F

1 2 3 4 2 1

1 2 1 2 2

1 2 3 3 3 3

References pp. 447-449

1

440

Chapter 12

Water

MeOH

ACN

Fig. 12.37. The mobile phase compositions used in the calibration of the octadecyl stationaryphases.

tionary phases are the objects (Fig. 12.38). The (6 X 144) data matrix X is then column centered. First the structure for the data presented here is investigated. To find the underlying structure of the data set, first principal component analysis was done on the whole data set. Figure 12.39 gives the score plot of the data matrix X. In the score plot, it can be seen that stationary phases A and B are the most extreme. Note that the difference between stationary phases of the same batch, but used by a different analyst/apparatus combination (A, F and B, E), is larger than the differences between batches measured by the same analyst/apparatus combination. Figure 12.40 gives the score plot of the PARAFAC model of the three-way data cube. The data are centered in such a way that the relative differences between the stationary phases will be investigated. The three-way solution, as explained in Section 12.2.4,has a more regular contribution to the components of the stationary phases than the unfolded

wa3

K P w m l..._...,....PHBwml... ,.. ......,....... ......, ..,..._._. , ......,.. . _ _...... _ . ....,._. ._......PHBwmB ,

,

A 6 C D E F ~

Fig. 12.38.The unfolding of the (6 x 9 x 16) data cube to a (6 x 144) data matrix.

Multivariate characterization of RP-HPLC stationary phases

44 1

...5

C 0

0.2

0 .

r 8 5

0.

-

. I

-0.2

-

-0.4-0.6-

-1.2

A.

-1

-0.1)

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

scores PCl

Fig. 12.39. PCA score plot of the unfolded data matrix of Fig. 12.38.

solution. The variation associated with stationary phases A and B is not completely absorbed in the second and first PARAFAC component, as was the case in the unfolded solution. Contrary to the score plot of the unfolded data matrix, the scores are now not uncorrelated. B and C are now the most deviating. The unfolded solution is more flexible but leaves the principal components the freedom to explain retention on stationary phases A and B completely. This is only reasonable if these stationary phases are really extremely informative, otherwise outlying behavior is modelled. Three stationary phases were chosen for the calibration: two for the training set and the other one for the test set. Based on the PCA score plot of the unfolded solution, stationary phases A, B (training) and C (test) were chosen. Four solutes (with abbreviations ANS, DMP, EE and PRE) were chosen as markers, with the induced variance criterion. All mobile phases were included. The PLS model building and prediction is visualized in Fig. 12.41. With two dimensions in the PLS model, 99.5%of the variation in X[M] is used to explain 99.2% of the variation in X[NM]. The final step in the prediction procedure is the calibration of the new stationary phase C with the markers. It is not necessary to predict at the same mobile phase compositions as in the training set, but it is convenient with the 3 8.

S

oA A.

r S

2

-1

C.

-2 -4

-3

-2

-1

0

1

2

scores 1

Fig. 12.40. Score plot of the PARAFAC model of the data cube presented in Fig. 12.38.

References pp. 447-449

442

C

Chapter 12

XIMJ n w

I

v

1 - - - - - - l

XINMJ.,

1

TABLE 12.5 THE RMSEP VALUES OF DIFFERENT SOLUTES AND MOBILE PHASES AND THE OBSERVED AND PREDICTED VALUES OF SOME SOLUTES Solute

Rh4SEP

Mobile phase

RMSEP

ACP ACT ANS CRE DMP EAB EE EHB MHB NBZ PBL PE PHB PRE PRS TOL Mean

0.036 0049

Wml

0.045 0.055 0.034 0.072 0.041 0.047 0.053 0.047 0.054 0.05 1

0.037

wm2 wm3 am1 am2

0.037 004 0.054 0.04 0.07 0.036 0.045

am3 wal W22

Wa3 Mean

0.071 0.071 0.051 TOL

ACP

wml wm2 wm3 am1 am2 am3 wal Wa2 Wa3

Observed

Predicted

Observed

Predicted

1.49 0.88 0.24 1.89 1.13 0.61 1.88 1.28 0.74

1.52 0.89 0.27 1.88 1.17 0.62 1.89 1.24 0.66

26.33 14.03 7.08 35.11 17.81 9.28 42.58 20.72 10.01

24.66 12.72 6.64 35.97 17.24 9.77 37.91 19.36 10.55

Multivariate characterization of RP-HPLC stationary phases

443

data set at hand. The results of the prediction are presented in Table 12.5, where the RMSEP values are given and the predicted versus measured values of some solutes. The RMSEP values are about three times the value of the reproducibility, indicating some systematic prediction error. To get an impression of the performance of the calibration, the observed versus the predicted ln(k) values of ACP and TOL are given for the best and worst calibrations, respectively. The solute TOL is known to be sensitive to stationary phase differences. This solute would be a marker in the second and third best marker subsets. The predictions of the other solutes can be considered good.

12.5.2 Calibration of stationary phases of different types

It would be desirable if retention values of solutes measured on stationary phases could be transferred to stationary phases of a different type. Here, an example is given of reversedphase stationary phases that differ in bctionality of the ligands. Two stationary phases are considered. The first stationary phase is a monofunctional octyl modified phase (C8/1), the second stationary phase is a trihnctional octyl modified stationary phase (C8/3). The chromatograms of the two stationary phases are almost the same, but the retention values of the solutes on the trihnctional octyl modified stationary phase are sometimes slightly higher or lower than the retention values on the monofunctional octyl modified stationary phase. With PCA, any difference in the underlying structure between the two phases will be investigated. With PLS, prediction of the retention values of the trifunctional octyl modified phase based on the monofimctional octyl modified phases will be tried. PLS also tries to diminish the difference between the two phases. The retention values measured on one stationary phase can be used for the retention values of the other stationary phase, but these values are not correct because there is a difference between the retention values of the two stationary phases. The retention values of a stationary phase can also be predicted based on a calibration model of the other stationary phase and some measurements of the stationary phase itself. Maybe these predicted values differ less from the retention values on a stationary phase than the retention values on another stationary phase differ from the retention values on the stationary phase being considered. This difference is measured as the root of the sum of the squared differences between the ln(k) values measured on both stationary phases, divided by the total number of elements included before and after the prediction. After prediction, this is the RMSEP, before prediction this is something like the RMSEP but then the ln(k) values of both stationary phases are compared instead of observed and predicted ln(k) values. For the calibration of the two stationary phases, 14 solutes were measured at 9 mobile phases. The solutes and mobile phases were the same as those presented in Section 12.2.3. The column centered measurements on both stationary phases can be stacked in two (9 X 14) data matrices. Section 12.2.3 gives a PCA example. The PCA score plot presented (Fig. 12.12) is representative of the score plot of both stationary phases considered in this section. Thus, the structure in the data matrices of both stationary phases is the same. The underlying References pp. 447-449

444

Chapter 12

r

7 I

--

L& jb i

C8M’9I

r---XINMl...

prediction l - - - - i

.-

I I I

Fig. 12.42. The PLS model for the monofunctional and trifunctional octyl modified stationaryphases

factors explaining the systematic variance can be interpreted in terms of solvent strength, solvent strength selectivity and modifier selectivity. This is true for both stationary phases. The loading plot presented in Section 12.2.3(Fig. 12.13) is also more or less representative of both stationary phases, apart from slight differences. In both loading plots, aniline was the most extreme solute. It could be that the behavior of aniline is outlying instead of representative. If markers are selected based on one data matrix (that is, for instance, the data matrix corresponding to the monofunctional octyl modified stationary phase) and a PLS model is based on that stationary phase, good predictions with this model are expected for the In(k) values of the non-markers on the other stationary phase, because the stationary phases have the same underlying structure. Based on the determinant criterion, three variables (PRE, ETB and NBA.) are selected as markers for the data matrix of the monofunctional octyl modified phase. Aniline (AN) was excluded from the selection procedure, because of the probable outlying behavior. All the solutes except PRE, ETB and NBA are the non-markers. A PLS model can be made (Fig. 12.42). The capacity factors of the non-markers on the trifunctional octyl modified phase can now be predicted and compared with the observed values. The RMSEP can be calculated and compared with the difference between the observed values of the monofunctional octyl modified stationary phase and the observed values of the trifunctional octyl modified stationary phase. The RMSEP is 0.0084 and the difference between the values of the

i

matrix 1 j

I

Fig. 12.43. A matrix of differences.

differences

Multivariate characterization of RP-HPLC stationary phases

445

Wat a

S

C

Wml a

0

r warn1

wal a

b

2

-0.085

wrn2

warn3

b

8

Wa3 a

Fig. 12.44. A score plot of a matrix of difference between two different fresh columns.

two stationary phases is 0.0186. The RMSEP is almost the same as the reproducibility (= 0.009). The PLS model seems to give good predictions. Although the RMSEP is lower than the difference between the two stationary phases, it is not much. It is interesting to know if there is a systematic difference between the stationary phases. A possibility is to look at the matrix of differences. This is a matrix with the In@) values measured at one stationary phase minus the In@) values measured at the other stationary phase (Fig. 12.43). To find the structure in this data set, one possibility is a PCA on this data set. Figure 12.44 gives a PCA score plot on the matrix of differences between the monofunctional octyl modified stationary phase and the trifunctional octyl modified stationary phase. This score plot does not show much structure. This does not guarantee that there is not some systematic variance between the two stationary phases, but it is an indication. The PLS example of Section 12.4.4 is also a good example of the calibration of stationary phases of different types.

12.6 APPENDIX A A.1 Rank

The number of independent rows or columns in a data matrix is called the rank of the data Referencespp. 447-449

446

Chapter 12

Fig. 12.45. Two variables that are independent and not orthogonal.

matrix. This rank is at the most the minimum of the number of variables (m)and number of objects (n). The rank of the design matrix presented in Section 12.2 1 (Fig. 12.6) is at the most 2. This is the minimum of 2 (variables) and 4 (objects). The rank is equal to the maximum number of linear independent variables or objects. A variable is linear independent of other variables if this variable cannot be written as a linear function of the other variables. A linear function of the variables x1 to x, is a weighted average of these w,x, with w iconstant for every i. Orthogonal variables are variables: wlxl + w2x2 always linear independent. Linear independent variables are not always orthogonal (Fig. 12.45). The rank of the above design matrix is 2, because the 2 variables are orthogonal. If the rank of a data matrix is r then the maximum number of PCs is of course r. Because of random variations in real data, r often equals the minimum of n and m. The variation explained by the last PCs will be very little. It is the variation caused by measurement errors, noise, etc. The so-called effective rank is not equal to the minimum of n and m. +..a+

A.2 Spectral decomposition An important technique associated with PCA is the spectral decomposition. It is described here briefly. More about this can be found in the literature [20-22,601. Let S be a symmetric (n x n) matrix of rank r (r I n). Then S can be decomposed according to the spectral decomposition

with KK' = K'K = I and A diagonal (Fig. 12.46). The diagonal elements ill to 1, are called the eigenvalues and are arranged in descending order on the diagonal of A. The vectors kl to k,,are the corresponding eigenvectors. A.3 The singular value decomposition (SVD)

The singular value decomposition is a generalization of the spectral decomposition. It is

Fi

Multivariate characterization of RP-HPLC stationary phases

A =

447

-..

O‘..x,

Fig. 12.46. A diagonal matrix.

true for all kinds of matrices, not only symmetric matrices. The singular value decomposition is related to PCA. For more about this, see the literature [20-22,60-62]. Let X be an arbitrary (n X m) matrix of rank r (r I min(n,m)). Then X can be decomposed according to the singular value decomposition

with U’U = I, V‘V = W’ = I and D diagonal. The diagonal elements dl to dmare non-negative and called the singular values. The singular values are arranged in descending order in D. The columns of U and V are called the left and right singular vectors. This because XV‘ = UD and U’X = DV’. The relation with PCA becomes clear by taking in the singular value decomposition UD = T and V = P. Every X can be decomposed according to the singular value decomposition

X = UDV’ Then

This is the spectral decomposition. From this it follows that V is the matrix of eigenvectors of X’X corresponding to the eigenvalues stacked in descending order in D2. This matrix V of eigenvalues is thus the matrix of loadings of X. The matrix of score vectors of X is XV = UDV’V = UD. The loadings are chosen orthogonal (P’P=I), the scores are have the property T‘T = P’X’XP = D’U’UD = A (diagonal). Now X can be written as X = TP’.

12.7 REFERENCES 1 2 3 4 5 6 7

J.G. Atwood and J. Goldstein, J. Chromatog. Sci., 18 (1980) 650. L. Hansson and J. Trojer, J. Chromatog., 207 (1981) 1. R.M. Smith, T.G.Hurdley, R. Gill and J.K. Baker, Chromatographia, 19 (1984) 407. H.H. Freiser, M.P. Nowlan and D.L.Gooding, J. Chromatog., 12 (1989) 827. R.E. Majors, LC-GC, 3 (1990) 12. L.C. Sander and S.A. Wise, LC-GC Int., 3 (1990) 29. E. Kovats, Helv. Anal. Chim. Acta, 41 (1985) 1915.

448

Chapter 12

8 9 10 11 12 13 14

R.M. Smith, J. Chromatogr., 26 (1987) 278. J.K. Baker and C.Y. Ma, J. Chromatogr., 169 (1979) 107. R.M. Smith, Anal. Chem., 56 (1984) 256. R.M. Smith, G.A. Murilla, T.G. Hurdley, R. Gill and A.C. Moffat, J. Chromatogr., 384 (1987) 259. R.M. Smith, T.G. Hurdley, R. Gill and M.D. Osselton., J. Chromatogr., 398 (1987) 73. M. Bogusz, J.P. Franke, R. A de Zeeuw and M. Erkens, Anal. Chem., 347 (1993) 73. J.H. Dhont, C. Vinkenoog, H. Compaan, F.J. Ritter, R.P. Labadie, A. Verweij and R.A. De Zeeuw, J. Chromatogr., 47 (1970) 376. R. Gill, M.D. Osselton, R.M. Smith and T.G. Hurdley, J. Chromatogr., 386 (1987) 65. A.K. Smilde, Ph.D. Thesis, University of Groningen, The Netherlands, 1990. K. Smilde, P.M.J. Coenegracht, C.H.D. Bruins and D.A. Doornbos, J. Chromatogr., 485 1989) 69. E.P Box, W.G. Hunter and J.S. Hunter, Statistics for Experimenters, Wiley, New York, 1978. A. Cornell, Experiments with Mixtures: Designs, Models and the Analysis of Mixture Data, Wiley, New York, 1990. T. Jolliffe, Principal Component Analysis, Springer-Verlag,New York, 1986. V. Mardia, J.T Kent, J.M. Bikky, Multivariate Analysis, Academic Press, 1979. J. Krzanowski, Principles of Multivariate Analysis, Oxford Science, New York, 1988. J.W. Weyland, Ph.D. Thesis, University of Groningen, The Netherlands, 1986. S.T. Balke, Quantative Column Liquid Chromatography, Elsevier, Amsterdam, 1984. G.E.P Box and N.R. Draper, Empirical Model Building and Response Surfaces, Wiley, New York, 1987. P.M.J. Coenegracht, A.K. Smilde, H. Benak, C.H.P Bruins, H.J. Metting, H de Vries and D.A. Doornbos, J. Chromatogr., 550 (1991) 397. H.G. Law, C.W. Snyder, J.A. Hattie and R.P McDonald, Research Methods for Multimode Data Analysis, Praeger, New York, 1984 J.D. Caroll, S Pruzansky and J.B. Kruskal, Psychometrika, 45 (1980) 3. P.M. Kroonenberg and J de Leeuw, Psychometrika, 45 (1980) 69. R.A. Harshman, UCLA Working Papers in Phonetics, 16 (1970) 1. L.R. Tucker, in: Problems in Measuring Change, University of Wisconsin Press, Madison, 1963. P. Geladi, Chemometrics Intell. Lab. Syst., 7 (1990) 237. S. Wold, P. Geladi, K. Esbensen and J. Ohman, J. Chemometrics, 1 (1987) 41. C.L. de Ligny, M.C Spanjer, J.C. van Houwelingen and H.M. Weesie, J. Chromatogr., 301 (1984) 31 1. J. Ohman, P. Geladi and S. Wold, J. Chemometrics, 4 (1990) 79. J. Ohman, P. Geladi and S. Wold, J. Chemometrics, 4 (1990) 135. A.K. Smilde, Chemometrics Intell. Lab. Syst., 15 (1992) 143. A.K. Smilde and D. A. Doornbos, J. Chemometrics, 5 (1991) 345. C.H. Lochmuller, S.J. Breiner, Ch. E. Reese and M.N. Koel, Anal. Chem., 61 (1989) 367. A.K. Smilde, P.H. van der Graaf, D.A. Doornbos, A.G.M. Steerneman and A. S h r i n k , Anal Chim. Acta, 235 (1990) 41. G.P Mc Cabe, Technometrics, 26 (1984) 137. A.K. Smilde, C.H.P. Bruins, P.M.J. Coenegracht and D.A. Doornbos, Anal. Chim. Acta, 212 (1988) 95. A.K. Smilde, P.M.J. Coenegracht, C.H.P. Bruins and D.A. Doornbos, J. Chromatogr., 485 (1989) 169. P.J. Schoenmakers, Ph.D. Thesis, Delft, 1981. E.H. Slaats, Ph. D. Thesis, Amsterdam, 1980. R.M. Cormich and B.L. Karger, Anal. Chem., 52 (1980) 2249. H. Colin, A. Krstulovic and G. Guiochon, J. Chromatogr., 225 (1983) 295. R. Kaliszan, QSRR, Wiley, New York, 1987. C. Hansch, Drug Design, Vol. 1, Academic Press, New York, 1971. Y. Wang, David J. Veltkamp and B.R. Kowalski, Anal. Chem., 63 (1990) 2750. Y. Wang, M.J. Lysaght and B.R. Kowalski, Anal. Chem., 64 (1992) 562. C.F. Gauss, Collected Works, GOttingen, 1873. K.G. JOreskog and H. Wold, Systems under Indirect Observations, Parts I and 11, North-Holland, Amsterdam, 1982. P. Geladi and B.R. Kowalski, Anal. Chim. Acta, 185 (1986) I .

15 16 17 18 19 20 21 22 23 24 25 26 21 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54

Multivariate characterization of RP-HPLC sfationaryphases 55 56

57 58 59 60 61 62

449

P. Geladi and B.R. Kowalski, Anal. Chim. Acta, 185 (1986) 19. S. Wold, C. Albano, W.J. Dunn 111, U. Edlund, K. Esbensen, P. Geladi, S. Helberg, E. Johansson, W Lindberg and M. SjOstrOm, Chemometrics, Mathemathics and Statistical Chemistry, Dordrecht, 1984. S. Wold, A. Ruhe, H. Wold and W.J. Dunn 111, J. Sci. Statist. Comput., 5 (1984) 735. A. HUskuldson, J. Chemometrics, 2 (1988) 21 1. R. Manne, ChemometricsIntell. Lab. Syst., 2 (1987) 187. J.M.F. ten Berge, Least Squares Optimization in Multivariate Analysis, DSWO Press, Leiden, 1993. S. Wold, K. Esbensen and P. Geladi, Chemometrics Intell. Lab. Syst., 2 (1997) 283. O.M. Kvalheim, Chemometrics Intell. Lab. Syst., 7 (1989) 39.

This Page Intentionally Left Blank

45 1

Subject Index Aliphatic esters convergence point 262 effect mobile phase 271 interaction indices 257 retention index scale 1 17 retention selectivity prediction 288 n-Alkane retention index scale 130 chalcogens 139 gas chromatography97,99,183,324 in liquid chromatography 107,213 in supercritical fluid chromatography 100 interaction increments 6, 109 retention prediction 6 substituent increments 6-7, 108, 137 n-Alkanes 257 n-Alkanols 126,172,271 Alkan-2-one retention index scale 49, 110,146, 298 biological activity 135 drug identification 148,177 environmental 164 group contributions 9 Hansch II contributions 9 in gas chromatography324 in liquid chromatography213,209 natural products 163 mobile phase characterization298,318 resolution prediction 306 retention prediction 8-9,305 stationary phase characterization 127 Alkan-2-ones 271,281 Alkenes 108 N-Alkylanilines 129-1 30 Alkyl aryl ketone retention index scale aromatic substituent contributions 13 benzene as parent compound 12-14 calculated from literature 39-40 drug identification 149, 177 functional group contributions 12 gradient elution 219 in liquid chromatography 1 1 1, 117,209,214, 217,226 in micellar electrokinetic chromatography 106 interaction contributions 12 natural products 157 monosubstituted aromatic compounds 15

polysubstituted aromatic compounds 21,26 reproducibility 13, 112, 122-123,226 retention prediction 12,282 stationary phase comparisons 127,386 substituted aliphatic compounds 15 supercritical fluid chromatography 100 Alkyl aryl ketones 111, 126,214 external standards 96 internal standards 96 n-Alkylbenzene retention index scale in liquid chromatography 109,213 in micellar electrokineticchromatography 106 retention prediction 6 substituent increments 108-109 n-Alkylbenzenes 61, 104, 109,126,130 convergence point 260-261 effect mobile phase 271 interaction indices 257,260,270,272 retention prediction 279,282,288 Alkyl benzoates 1 17, 13 1 Alkyl halides 100,104 AIkyl hydroxybenzoates retention index scale 41,116-1 17,133 Alkyl polysulphides 7 Allyl-bonded phases 385 Alumina stationary phases 393 Amino-bonded phases 385 Anilides of fatty acids internal standards 96 Anilines 59 Anthranilic acids 9, 135 Argentation chromatography 108 Arthritis drugs 148 Azabicycloalkanes 11 Barbiturates 9,102,123,135,149,178,185,189 Basic drugs see Drugs Benzene 13 Benzodiazepines 74,102 Biological activity retention indices 135 Biological matrices 203 Bonded phases see Stationary phase materials

452

Subject Index

Capacity factors see Retention factors Capillary electrophoresis see Micellar electrokinetic chromatography Carotenoids 164 Catecholamines 389 Chalcogens 137, 139 Chiral stationary phases 137,390 ligand exchange phases 390 organometallic phases 393 Pirkle phases 391 protein phases 391-392 ChromDream 4 Column hold-up volume 49,9495, 126,248 Column test compounds 113-1 14 Column void volume see Column hold-up volume Convergence points 260,262 Corrected retention index values see Retention indices, corrected Cosmetics 105 CRIPES 13,33, 112 predictions 36 Cyano-bonded stationary phases 130,385 D-compounds see 1-[p-(2,3 -Dihydroxypropoxy)phenyl] 1 alkanones Dead volume/time see Column hold-up volume Dinitrophenylhydrazoneretention index scale 111, 118,136 1-[p-(2,3-Dihydroxypropoxy)phenyl]-l-alkanones see also Chapter 6 identification of mycotoxins 229 retention index scale 118,209,214,217 reproducibility 225 Discrimination number 151, 187, 191, 199 Diuretic drugs 154, 178, 187, 190 DRIFT 273,373,387 Drugs 54 see also Chapters 4 and 5 and individual groups gas chromatographyretention indices 172, 184 hydrophobicity 62 identification 146147, 149, 156 metabolites 9, 156 nitroalkane scale 186, 192 retention factors 54 retention indices 120-121, 186, 192 retention prediction 62,65 retention reproducibility 124-125 separation 380 standardisation 173, 175

--

Drylab 1,85,216 Elemental analysis 376 ELUEX 87 Environmental samples 164 Equivalent carbon number 117 EPA methods 337 Ergopeptines 11, 159 ESCA methods 375 Esters as retention index scales aliphatic 117 phenolic 116 Expert systems 1,13,33 Explosives 165 External standards 95 Factor analysis 61 Flavour compounds retention indices 161 Fragmental retention contributions see Functional group contributions Free energy changes 82 Functional group contributions 3, 5 see also Substituent indices and Methylene group contributions metabolites 69-70, 73 molar volumes 82 structural effects 74-75 Fungal metabolites 157,228

Gas chromatography identification 172 resolution prediction 323 retention index scales 99-100, 183,222 selectivitytriangle 326 stationary phase characterization 127,323, 325-326 temperature programmed 99,222 Glucuronides 10 Gradient elution 215,219,290 Group contributions see Functional group contributions Hammett functions 23 Hansch x contributions 2,9,2 1 lipophilic indices 78,274 substituent indices 18 Herbicides 224,281,284 Hold-up volume see Column hold-up volume Homologues 97, 104, 130,255,260,270 see also Methylene group contributions and Retention index scales convergencepoints 260-261

Subject Index lipophilic and polar indices 78,277 retention linearity 126,217,255,269,271 Hydrogen-bonding 23 Hydrophobicity index values 57-58,66,89 relationship to retention 54,62 Hydroxybenzoateesters see Alkyl hydroxybenzoates Hydroxylation 7 1-73 Identification using retention indices see Retention indices for identification Infrared spectra of stationary phases 355-356,372, 387,393 Interaction indices (analytes) 4,236,247,253 binary mobile phases 238 calibration 247,249 calibration standards 249 homologues 256 indices of analytes 251,254 mobile phases 269 retention prediction 250 ternary mobile phases 243 values 250,254 Interaction indices (between functional groups) 6, 23 hydrogen-bonding interactions 24,33 meta andpara interactions 29 on n-alkane scale 6 on alkyl aryl ketone scale 21 ortho interactions 32 phenolic substituents 30 Interlaboratory studies 123, 173, 195 Internal standards alkyl aryl ketones 96 anilides of fatty acids 96 computer aided selection 96 phenols 96 Isoalkanes 8 Isocratic elution 217 2-Keto alkanes see Alkan-2-ones Kovhts indices 97,297,324 LC-MS 158,228 Least squares analysis 429 Length to breadth ratios 62,358,360-361,365 Lichen constituents 163 Linear free energy relationships 2, 76 Lipophilicity 135 Lipophilic indices 4,78, 135,269,271-272 see also Chapter 8 calibration standards 272 hydrophobic substituent constant 274

453 mobile phase effects 285 retention prediction 136,283 selectivity characterization 283 structural effects 274,276 Liquid crystal phases 388 Local anaesthetics 153 Log p see Octanol-water partition coefficient Martin equation 2,97, 104,254 McReynolds constants 127,324 Mean list length 199 MetabolExpert 72,89 Metabolites 156 hydroxylation 7 1-73 retention factor prediction 10,69-70, 89 retention indices 10, 156-157 Methylene group contributions 4, 7-8, 104,242 Micellar electrokinetic chromatography 105 Mobile phases in liquid chromatography binary mobile phases 136,238 characterization 133,269 composition prediction 74 effects on retention factors SO, 7P.80, 146, 216, 270,285,352 effects on retention indices 12-13, 112-1 13, 129,134,149,150, 181,318,330 pH effects 151 prediction of retention effects 69,285,298 prediction of retention indices 302 selectivity characterization 297,309,322,329 solvent selectivity triangle 298,309,3 11, 320 solvent properties 300,3 10-3 11 ternary mobile phases 134,243 test compounds 301,329 Molar volumes 53,82,251 Molecular connectivity 59-60 Molecular parameters effect on retention 55 surface area 53 Multivariant analysis 8 see also Chapter 12 marker compounds 422,425 ODS-bonded phases 439 ordinary least squares 429 partial least squares 432 prediction methods 428,434 principal components analysis 129,410,415, 424 Mycotoxins 102, 158,228 Narcotic analgesics 9-10 Natural products 157 fungal metabolites 157

454 lichen constituents 163 plant products 160 plant toxins 162 Neural networks 2 NIST standards 343,346,349 1-Nitroalkaneretention index scale see also Chapter 5 drug identification 156, 178 in gas chromatography 115, 183 in liquid chromatography 115, 178, 195,209, 214,217 in micellar electrokinetic chromatography 106 1-Nitroalkanes 178, 195,214 Nitropolynuclear aromatic hydrocarbons 116 Nortropanes 9 Nuclear magnetic resonance spectroscopyof stationary phases 374,393 Octanol-water partition coeficient 53, 135 alkan-2-one retention indices relationship 8, 135 alkyl atyl ketone retention indices relationship 28 calculated 58, 87 from liquid chromatography64, 89 hydrophobicity relationship 66 retention factor relationships 2, 54,62-63,88 substituent index relationships 19 Oligomers 255,277,298,310 Parabens see also Alkyl hydroxybenzoates substituent increments 42 Partial least squares 432 PCBs 164 Perchlorinated compounds 164 pH effects on retention indices 151 Pharmaceuticals see Drugs Phenolic esters retention index scale 41, 116 Phenols 26,28,42, 5 9 4 0 , 83 internal standards 95, 117 Phthalates 116 Pirkle phases 391 Plant products 160 gliadins 163 spices 161 toxins 162 Polarity interaction index 4 Polarity indices 4,78,249,268,272 see also Chapter 8 retention prediction 248,279 selectivity characterization283 structural effects 274,276

Subject index Polycyclic aromatic hydrocarbons see Polynuclear aromatic hydrocarbons Polymeric bonded-phases 388,390,394 Polymeric columns 43, 131, 155,397 Polynuclear aromatic hydrocarbon retention index scale41, 115,210, 357 gas chromatography 358 retention indices of PAHs 359,361 retention prediction using 42 stationary phase effects 358 substituent indices 42 Polynuclear aromatic hydrocarbons see also Chapter 10 gas chromatography 337 identification 165 isomer separations 341 IengtWbreadth ratio 62,358,360-361, 365 nitro-derivatives as index scale 116,2 10 mobile phase effects 352 planarity effects 364 retention indices 165,359,361 shape selectivity 132,343,355 standard reference mixtures 343,346,348-349, 353 stationary phase effects 338,340,343 substituted polynuclear aromatic hydrocarbons 365 temperature effects 353 Polystyrene oligomers 298,310 Polytin compounds 104 Pore size effects 346 Prediction see Retention prediction and Retention indices - prediction Principal components analysis 129,410,415 Prisma 86 Propanolols 9, 135 Protein stationary phases 392 PTH amino acids 60 Quinazoline derivatives 135 Regression analysis for prediction 55-59, 61, 66, 77,79 Rekkerfcontributions 3,21,63,69 Relative retention factordtimes 94 Resolution 304,3 11,314 resolution prediction 305-306,314,331 solvent selectivitytriangle effects 309 Retenoids 164 Retention effect of binary mobile phases 238 effect of mobile phase 2,50,78-79 effect of ternary mobile phases 243

Subject Index

455

interaction indices of analytes see Chapter 7 methylene increments 242 solvophobic theory 236 temperature 52 Retention factors 48,68,94 gradient elution 216 prediction 57,60,248,250,252 reproducibility 145-146 Retention index scales see also Chapter 3 and individual series n-alkanes 97,107 alkan-2-ones 49, 109,119,177,300 alkylarylketones111,119,177 n-alkylbenzenes 109 alkyl hydroxybenzoate esters 4 1 comparison of scales 1 18,188-1 89,217,226, 378 D-Compounds see 1-[p-(2,3-dihydroxypropoxy)phenyl]- 1 alkanones dinitrophenylhydrazones 1 18 1-[p-(2,3-dihydroxypropoxy)phenyl]-1alkanones 118,210,214,217 in gas chromatography 99, 183 in micellar electrokinetic chromatography 107 in supercritical fluid chromatography 100 Kovhts indices 97,99 I-nitroalkanes 115, 119,178 nitropolynuclear aromatic hydrocarbons I 16 phenolic esters 116 polynuclear aromatic hydrocarbons 41, 115, 357 requirement of standards for LC 103,213,378 triglycerides 117 Retention indices see also Chapter 1, Substituent indices and Interaction indices (between functional groups) adjusted retention indices 327 benzene as parent compound 12-14 biological activity 135 characterisation of stationary phases 125,127 corrected 155, 184,223,318,327 see also Chapter 5 definition 4,49,98-99,1 15,217,221,299 effect biological matrices 203 functional group effect 5 see also Interaction indices (between functional groups) gradient elution 210,215,219,220 interlaboratory studies 123,195 isoeluotropic mobile phases 18, 123 lipophilicity 135, 138

-

misuse of terms 98 mobile phase characterisation 329 mobile phase composition effects 12, 15, 110113, 123,129,134,181,184,217,311 pHofeluent 151 regression equations for prediction 43 reproducibility 13, 112, 122-123, 133,151, 177,188,190,194,223,226 resolution 3 14 solvent selectivity effects 311,327,329 standards for column comparison 422,425 stationary phase effects 38, 109,121,127, 130, 133,152-154,190,193,201,225,385,439 statistical analysis 433 structurehetention relationships 137 temperature effects 123,225 temperature programmed 99 transferability of indices 122 Retention indices for identification 125 see also Chapter 4 and Appendix 4.1 166 calculated from literature 38,40 drugs 146 isoeluotropic eluents 18 monosubstituted aromatic compounds 15, 107, 113,312,319 phenols 27 polynuclear aromatic hydrocarbons 357,359 polysubstituted aromatic compounds 21,26 steroids 302 substituted aliphatic compounds 15 Retention indices -prediction 33,35,40-41 see also Chapter 1 computer based 2,4 from CRIPES 13,33,36 from Log P 3 from Hansch n values 9 from mobile phase solvent 302,304 in gas chromatography 3 isoalkanes 8 on n-alkane scale 6 on alkyl aryl ketone scales 1 1 et seq. on alkylbenzene scale 6 Retention prediction see also Chapter 2 and Retention indices prediction from free energy values 84 from Hansch n values 69 from physicochemical data 76 from regression expressions 81 from Rekker constants 69 from structural fragments 69 lipophilic and polar indices 279 mobile phase composition 301,304

456 mobile phase effects 33 1 retention factors 34,48, 57,60, 75, 84,253 statistical analysis 428 stationary phases 428 Retention reproducibility 145-146 Reversed-phase stationary phases see Stationary phases Secondary retention index standards 183-184, 198 see also Chapter 5 Selectivity 253,283 Shape selectivity 343,345,355 Silica as stationary phases for drugs 173 Silanol activity 176, 240,380 Small angle neutron scattering 349 Solvation parameters used in prediction 81-82 Solvatochromicparameters 81 Solvent properties Solvent selectivity and solvent selectivity triangle see Mobile phases in liquid chromatography Solvophobic theory 236 Specific capacity factors 237,239 Spices 161 Standard reference materials 343, 346,348-349, 353 Standardisationof retention values of drugs on reversed-phase columns 175 on silica 173 Stationary phase materials alkyl-bonded phases 383 allyl-bonded phases 385 alumina phases 393 amino-bonded phases 385 base deactivated 176,213 bonded chain length 350 chiral phases 390 cyano-bonded phases 130,385 liquid crystal phases 388 ligand exchange phases 390 monomeric phases 340,348,378 organometallic phases phenyl-bonded phases 385 Pirkle phases 391 polymeric bonded phases 388,390,394 polymeric phases 340,348-349,397 protein phases 391-392 silica phases 174, 176 synthesis of bonded phases 176,211,338-339, 349,380-381 zirconia phases 395 Stationary phase properties bonding density 349 characteristicsfor PAHs 338

Subject Index characterization using retention indices 125, 127,377,386 chromatographiccharacterisation 377 comparisons 178,211,350,383,415,439,443 effects on log P 65 free energy parameters 84 elemental analysis 376 ideal reversed-phaseproperties 371 infrared spectra 355-356,372,387,393 peak tailing 380 phase type 338 pore size effects 346 problems with phases 210 retention indices, effects on 38,42, 109, 121, 127, 130, 152-154,190, 193,201 shape selectivity 343,345 spectroscopic properties 372,374,393 statistical comparison see Chapter 12 test compounds 127428,343 thermal properties 376 Statistical analysis of stationary phases see Chapter 12 Steroids 56,301-302,306 Structural hydrophobicity relationship 68 Structure retention relationships from free energy 53 retention indices 137 lipophilic and polar indices 274,276 Styrenes 258 Substituent contributions 3 Substituentretention increment values 68-69, 7275 free energy relationship 82 metabolites 70, 72, 74 Substituent indices 3-4 see also Functional group contributions, Methylene group contributions and Interaction indices (between functional groups) aliphatic substituents 15, 18,20 aryl substituents 13,16, 18 coefficients for calculations 12-13 see also Appendix 1.1 44 comparison Hansch II values 19,21-23 determination 5 in isoeluotropic eluents 18 lipophilic indices scale 278 mobile phase effects 18-19 on n-alkane scale 6,8,108 on alkan-2-one scale 41 on n-alkylbenzenescale 108-109 on polynuclear aromatic hydrocarbon scale 42 polar index scale 278 polyfunctional compounds 21

Subject Index see also Interaction indices (between functional groups) regression equations 43 unsaturation 19, 108 Sulphonamides 262 Sulphur compounds 137 Supercritical fluid chromatography retention index scales 100

Temperature effects 52,225,353,355 Ternary mobile phases 243,287 Test mixtures 343 Thermal analysis of stationary phases 376 Thiazides see Diuretics Thin-layer chromatography 172 Thiols 7 Three way analysis 417

451 Topological matrix molecular connectivity 60 retention prediction 60 Toxicological drug analysis 147, 156, 171, 175, 404 see also Chapter 5 Trichothecanes see Fungal metabolites Triglycerides 108, 117 Urushiols 11, 162 van der Waals radiiholume 56, 59 Void volume see Column hold-up volume Zirconia stationary phases 395

This Page Intentionally Left Blank

459

JOURNAL OF CHROMATOGRAPHY LIBRARY A Series of Books Devoted to Chromatographic and Electrophoretic Techniques and their Applications Although complementary to the Journal of Chromatography, each volume in the Library Series is a n important and independent contribution in the field of chromatography and electrophoresis. The Library contains no material reprinted from thejournal itself.

Othervolumes in this series Volume 1

Chromatography of Antibiotics (see also Volume 26) by G.H. Wagman and M.J. Weinstein

Volume 2

ExtractionChromatography edited by T. Braun and G. Ghersini

Volume 3

Liquid Column Chromatography.A Survey OfModernTechniques and Applications editedby Z. Deyl, K. Macek and J.Jantik

Volume 4

Detectors in Gas Chromatography by J. SevEik

Volume 5

InstrumentalLiquid Chromatography.A Practical Manual on HighPerformance Liquid Chromatographic Methods (see alsoVolume 27) by N.A. Parris

Volume 6

Isotachophoresis.Theory, Instrumentation and Applications by F.M. Everaerts, J.L. Beckers andTh.P.E.M. Verheggen

Volume 7

ChemicalDerivatizationin Liquid Chromatography by J.F. Lawrence and R.W. Frei

Volume 8

Chromatography of Steroids by E. Heftmann

Volume 9

HPTLC -High PerformanceThin-LayerChromatography edited by A. Zlatkis andR.E. Kaiser

Volume 10

Gas Chromatography of Polymers by V.G. Berezkin, V.R. Alishoyev and I.B. Nemirovskaya

Volume 11

Liquid Chromatography Detectors (see also Volume 33) by R.P.W. Scott

Volume 12

Affinity Chromatography by J . Turkova

Volume 13

Instrumentationfor High-PerformanceLiquid Chromatography edited by J.F.K. Huber

Volume 14

Radiochromatography.The Chromatography and Electrophoresis of Radiolabelled Compounds byT.R. Roberts

Volume 15

Antibiotics. Isolation, Separation and Purification editedbyM.J. Weinstein andG.H. Wagman

460 Volume 16

Porous Silica. Its Properties and Use as Support in Column Liquid Chromatography by K.K. Unger

Volume 17

75Yearsof Chromatography -A Historical Dialogue edited by L.S. Ettre and A. Zlatkis

Volume 18A

Electrophoresis.A Survey of Techniques and Applications. Part A Techniques edited by Z. Deyl

Volume 18B

Electrophoresis.A Survey of Techniques and Applications. Part B Applications edited by Z. Deyl

Volume 19

ChemicalDerivatizationin Gas Chromatography by J. Drozd

Volume 20

Electron Capture.Theory and Practice in Chromatography edited by A. Zlatkis and C.F. Poole

Volume 21

Environmental Problem Solving using Gas and Liquid Chromatography by R.L. Grob and M.A. Kaiser

Volume 22A

Chromatography.Fundamentals and Applications of Chromatographic and Electrophoretic Methods. Part A Fundamentals andTechniques edited by E. Heftmann

Volume 22B

Chromatography.Fundamentals and Applications of Chromatographic and Electrophoretic Methods. Part B: Applications edited by E. Heftmann

Volume 23A

Chromatographyof Alkaloids. Part A: Thin-LayerChromatography by A. Baerheim Svendsen and R. Verpoorte

Volume 23B

Chromatography of Alkaloids. Part B Gas-Liquid Chromatography and High-PerformanceLiquid Chromatography by R. Verpoorte and A. Baerheim Svendsen

Volume 24

Chemical Methodsin Gas Chromatography byV.G. Berezkin

Volume 25

Modern Liquid Chromatography of Macromolecules by B.G. Belenkii and L.Z. Vilenchik

Volume 26

Chromatography of Antibiotics. Second,CompletelyRevised Edition by G.H. Wagman and M.J. Weinstein

Volume 27

Instrumental Liquid Chromatography.A Practical Manual on HighPerformance Liquid Chromatographic Methods. Second,Completely Revised Edition by N.A. Parris

Volume 28

MicrocolumnHigh-PerformanceLiquid Chromatography by P. Kucera

Volume 29

Quantitative Column Liquid Chromatography.A Survey of Chemometric Methods by S.T.Balke

461 Volume 30

MicrocolumnSeparations. Columns, Instrumentation and Ancillary Techniques edited by M.V. Novotny and D. Ishii

Volume 31

Gradient Elution in Column Liquid Chromatography.Theory and Practice by P. Jandera and J. Chureieek

Volume 32

The Scienceof Chromatography.Lectures Presented at the A.J.P. Martin Honorary Symposium, Urbino, May 2741,1985 edited by F. Bruner

Volume 33

Liquid Chromatography Detectors. Second,CompletelyRevised Edition by R.P.W. Scott

Volume 34

Polymer Characterization by Liquid Chromatography by G. Glockner

Volume 35

Optimizationof Chromatographic Selectivity.A Guide to Method Development by P. J. Schoenmakers

Volume 36

SelectiveGas Chromatographic Detectors by M. Dressler

Volume 37

Chromatographyof Lipids in Biomedical Research and Clinical Diagnosis edited by A. Kuksis

Volume 38

Preparative Liquid Chromatography edited by B.A. Bidlingmeyer

Volume 39A

Selective SampleHandling and Detection in High-Performance Liquid Chromatography.Part A edited by R.W. Frei and K. Zech

Volume 39B

Selective SampleHandling and Detection in High-Performance Liquid Chromatography.Part B edited by K. Zech and R.W. Frei

Volume 40

Aqueous Size-ExclusionChromatography editedby P.L. Dubin

Volume 41A

High-PerformanceLiquid Chromatography of Biopolymersand Bio-oligomers.Part A Principles, Materials andTechniques by 0.Mikes

Volume 41B

High-PerformanceLiquid Chromatography of Biopolymersand Bio-oligomers.Part B Separation of Individual CompoundClasses by 0.Mike3

Volume 42

Quantitative Gas Chromatography for Laboratory Analyses and On-LineProcess Control by G. Guiochon and C.L. Guillemin

Volume 43

Natural Products Isolation. Separation Methods for Antimicrobials, Antivirals and Enzyme Inhibitors edited by G.H. Wagman and R. Cooper

Volume 44

Analytical Artifacts. GC, MS, HPLC,TLC andPC by B.S. Middleditch

462 Volume 45A

Chromatography and Modification of Nucleosides. Part A: Analytical Methodsfor Major and Modified Nucleosides- HPLC, GC, MS, NMR, W and FT-IR edited by C.W. Gehrke and K.C.T. Kuo

Volume 45B

Chromatography and Modification of Nucleosides. Part B: Biological Roles and Function of Modification edited by C.W. Gehrke and K.C.T. Kuo

Volume 45C

Chromatography and Modification of Nucleosides. Part C: Modified Nucleosidesin Cancer and Normal Metabolism - Methods and Applications edited by C.W. Gehrke and K.C.T. Kuo

Volume 46

Ion Chromatography.Principles and Applications byP.R. HaddadandP.E. Jackson

Volume 47

Trace Metal Analysis and Speciation edited by1.S. Krull

Volume 48

Stationary Phases in Gas Chromatography by H. Rotzsche

Volume 49

Gas Chromatography in Air Pollution Analysis byV.G. Berezkin andYu. S. Drugov

Volume 50

Liquid Chromatography in Biomedical Analysis edited byT. Hanai

Volume 51A

Chromatography,5th edition. Fundamentals and Applications of Chromatography and Related Differential Migration Methods. Part A Fundamentals andTechniques edited by E. Heftmann

Volume 51B

Chromatography,5th edition. Fundamentals and Applications of Chromatography and Related Differential Migration Methods. Part B Applications edited by E. Heftmann

Volume 52

Capillary Electrophoresis. Principles, Practice and Applications by S.F.Y. Li

Volume 53

HyphenatedTechniques in Supercritical Fluid Chromatography and Extraction edited by K. Jinno

Volume 54

Chromatography of Mycotoxins.Techniques and Applications edited by V. Betina

Volume 55

BioaffinityChromatography.Second, completely revised edition by J . Turkova

Volume 56

Chromatography in the Petroleum Industry edited by E. R. Adlard

Volume 57

Retention and Selectivityin Liquid Chromatography.Prediction, Standardisation and Phase Comparisons edited by R. M. Smith

Volume 58

Carbohydrate Analysis. High Performance Liquid Chromatography and Capillary Electrophoresis edited by Z. El Rassi