Gene Family Targeted Molecular Design

GENE FAMILY TARGETED MOLECULAR DESIGN

GENE FAMILY TARGETED MOLECULAR DESIGN Edited by

KAREN E. LACKEY

Copyright # 2009 by John Wiley & Sons, Inc. All rights reserved Published by John Wiley & Sons, Inc., Hoboken, New Jersey Published simultaneously in Canada 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, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permission. Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002. Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com. Library of Congress Cataloging-in-Publication Data:

ISBN: 9780470412893

Printed in the United States of America 10 9 8 7 6 5 4 3 2 1

CONTENTS

Preface

xi

Contributors

xv

1

Drug Discoveries by Gene Family

1

Karen E. Lackey

1.1 General Drug Discovery Components, 1 1.2 Further Reading for Expert Knowledge, 10 References, 13 2

G-Protein-Coupled Receptors

15

Stephen L. Garland and Tom D. Heightman

2.1 Introduction, 15 2.2 GPCR Structure and Function, 18 2.2.1 Subfamilies, 18 2.2.2 Structural Information and Homology Models, 22 2.2.3 Mechanisms of Receptor Modulation, 27 2.2.4 G-Protein Coupling and Assay Formats, 33 2.3 Challenges Facing the Area of GPCR Drug Design, 35 2.3.1 Hit Generation Strategies: Chemogenomics and Privileged Structures, 36 2.3.2 Case History: Calcitonin Gene-Related Peptide Receptor Antagonist (MK-0974, telcagepant), 40 2.3.3 Case History: Mixed Dopamine/Serotonin Receptor Antagonist As An Atypical Anti-Psychotic, 43 v

vi

CONTENTS

2.3.4 Case History: Chemokine Receptor CCR5 Antagonist Maraviroc (CelsentriTM), 45 2.3.5 Case History: The Discovery of Cinacalcet (Sensipar1/ Mimpara1), a CaSR-Positive Allosteric Modulator, 47 2.4 Conclusions and Outlook, 49 References, 50 3

Ion Channels Gene Family: Strategies for Discovering Ion Channel Drugs

53

Maria L Garcia and Gregory J. Kaczorowski

3.1 Introduction, 53 3.2 Ion Channel Subfamily Descriptions, 54 3.2.1 Voltage-Gated Ion Channels, 54 3.2.2 Inward Rectifier Potassium Channels, 55 3.2.3 Voltage-Gated Potassium Channels, 58 3.2.4 Calcium-Activated Potassium Channels, 61 3.3 Structure of Potassium Channels, 64 3.4 Criteria for Selection of Targets and Establishing Screens, 68 3.5 A Case Study in Ion Channel Drug Discovery, 70 3.6 Perspective on Ion Channels as Drug Targets, 77 References, 78 4

Integrins

85

David D. Miller

4.1 Introduction, 85 4.2 Integrin Inhibitor Discovery, 89 4.2.1 Cyclic Peptides, 91 4.2.2 Peptidomimetic Chemistry, 92 4.2.3 Preferred Peptidomimetic Fragments, 102 4.2.4 I-Domain Integrin Inhibitors, 103 4.2.5 Protein Structure-Based Design, 110 4.3 Challenges—Past and Future, 112 References, 113 5

Strategies for Discovering Kinase Drugs Jerry L. Adams, Paul Bamborough, David H. Drewry, and Lisa Shewchuk

5.1 Introduction, 119 5.2 Protein Kinase Structural Features, 120 5.3 Generating and Optimizing Kinase Inhibitors, 124 5.3.1 ATP Binding Pocket, 124 5.3.2 Non-ATP Binding Pockets, 133

119

CONTENTS

vii

5.4 Establishing Screens for Understanding Kinase Activity and Selectivity, 135 5.5 Case Studies of Successful Kinase Drug Discovery, 140 References, 150 6

Protease-Directed Drug Discovery

159

Richard Sedrani, Ulrich Hommel, and Jo¨rg Eder

6.1 Introduction, 159 6.2 Aspartic Proteases, 162 6.2.1 HIV Protease Inhibitors, 164 6.2.2 Renin, 166 6.3 Metalloproteases, 168 6.3.1 Angiotensin-Converting Enzyme, 168 6.3.2 Matrix Metalloproteases, 170 6.4 Serine Proteases, 175 6.4.1 Dipeptidyl Peptidase 4 (DPP4), 176 6.4.2 Trypsin-Like S1 Serine Proteases of the Coagulation Cascade, 179 6.5 Cysteine Proteases, 183 6.5.1 Cathepsin K, 184 6.5.2 Caspases, 188 6.6 Perspective on Proteases as Drug Targets, 190 References, 191 7

Small-Molecule Inhibitors of Protein–Protein Interactions: Challenges and Prospects Adrian Whitty

7.1 Introduction, 199 7.2 Structure and Properties of PPI, 201 7.2.1 Constitutive versus Transient PPI, 201 7.2.2 Physicochemical Properties and Residue Propensities of PPI, 202 7.2.3 Binding Energetics and ‘‘Hotspots’’, 204 7.3 Structural and Physicochemical Challenges to Inhibiting PPI with Small Molecules, 207 7.3.1 Key Role of Adaptivity at the Interface, 210 7.3.2 Constraints of ‘‘Drug-Like’’ Chemical Space, 212 7.4 Identifying Hits and Leads Against PPI Targets, 213 7.4.1 Fragment-Based Screening, 214 7.4.2 Validating and Optimizing Hits and Leads, 218 7.5 Assessing the Druggability of New PPI Targets, 225 References, 227

199

viii

8

CONTENTS

Transporters

235

Anne Hersey, Frank E. Blaney, and Sandeep Modi

8.1 Introduction, 235 8.2 Methodologies in Transporter Drug Design, 239 8.2.1 Structure-Based Methods, 239 8.2.2 Ligand-Based Methods, 245 8.3 Therapeutic Transporter Targets in Drug Discovery, 248 8.3.1 Vacuolar ATPases, 248 8.3.2 Gastric (P-) ATPases, 251 8.3.3 Neurotransmitter Transporters as Drug Targets, 252 8.4 Transporters as Liability Targets, 256 8.4.1 P-glycoprotein, 259 8.4.2 OATP1B1, 261 8.5 Application of Methods for Designing Interactions with Liability Targets, 262 8.6 Perspective, 267 References, 267

9

Nuclear Receptor Drug Discovery

275

Hiroyuki Kagechika and Aya Tanatani

9.1 Introduction, 275 9.2 Nuclear Receptor Superfamily and Their Functions, 276 9.3 Agonism and Antagonism in Nuclear Receptor Functions, 279 9.3.1 AR Antagonists Effective Toward Mutated Receptors, 283 9.3.2 VDR Agonists and Antagonists, 286 9.3.3 Carboranes as Novel Hydrophobic Pharmacophores, 290 9.4 Medicinal Chemistry of Retinoid Nuclear Receptors, 293 9.4.1 Retinoid and Their Nuclear Receptors, 293 9.4.2 Retinobenzoic Acids, 300 9.4.3 RXR-Selective Ligands, 306 9.5 Clinical Application of Retinoids, 309 9.6 Perspective, 311 References, 312 10

Summary and Comparison of Molecules Designed to Modulate Druggable Targets in the Major Gene Families Karen E. Lackey

10.1 Target Class Concept, 317 10.2 Summary of the Unique Features of Each Target Class, 318 10.2.1 GPCR/7TM, 318 10.2.2 Ion Channels, 320 10.2.3 Integrins, 322

317

CONTENTS

ix

10.2.4 Kinases, 323 10.2.5 Proteases, 324 10.2.6 Protein–Protein Interactions, 325 10.2.7 Transporters, 327 10.2.8 Nuclear Receptors, 328 10.3 Perspective, 330 References, 331 Appendix

333

Index

341

PREFACE

Approximately half of the anticipated small-molecule drug targets fall into just six gene families: G-protein-coupled receptors (GPCRs), protein kinases, zinc metallopeptidases, serine proteases, nuclear hormone receptors, and phosphodiesterases. A system-based research approach groups proteins into classes based on sequence and common motifs that form 3D space receptive for small-molecule interactions. The goal of small-molecule drug discovery is to modulate the activity of a biological target via interactions with an externally administered molecule at optimal drug intervention points in disease pathology to afford the maximum therapeutic index. Key to achieving this mechanistic approach to drug discovery is the design, synthesis, and evaluation of biologically effective compounds. Working within a system of related targets allows scientists to apply learning from one member to accelerate identification of ligands for other disease-associated targets. The purpose of this book is to provide a description of compound design methods for generating small molecules that interact with important biological targets in the following major gene families: G-protein-coupled receptors/7-transmembrane receptors, ion channels, integrins, kinases, proteases, protein–protein interactions, transporters, and nuclear receptors. Each chapter will cover affinity for the intended target, the mechanism of the interaction, small molecule examples, and ways to change the molecule to attenuate the activity. We hope to provide a solid foundation of information that allows readers to then approach more expert technical literature with a greater understanding. At the end of the introductory chapter, I have summarized some books that provide an in-depth coverage of the functional areas of drug discovery that contribute to this stage of research.

xi

xii

PREFACE

The reader will readily come up to speed in a new area of research or be able to compare their work across other gene families. Nature has developed classes of related proteins and has borrowed similar motifs to keep all of the different functions of a cell and complex system working. The selectivity achieved by nature is impressive, and these are the lessons we can exploit to repair, enhance, diminish, or eliminate an activity using a small molecule. By understanding the major gene families, the scientist can design molecules that target the intended protein and minimize interactions with the other protein classes. Beyond the scope of this book, but nonetheless important in selection of targets and molecule design, are the subsequent steps of drug development. The stages of drug generation beyond discovery include extensive toxicity measurements, development, clinical evaluation, registration, and commercialization and marketing of effective medicines for specific disease treatments or prevention. Synthetic, structural, computational, and medicinal chemists in academia, biomedical companies, and the pharmaceutical industry will benefit from a gene family-focused description of molecular design. There is a tremendous amount of literature on the topic and yet very little work has been done to condense the information into a manageable format as practical guidance for a chemist to get started in the area of designing compounds that intervene in important points of disease pathology. Also, overwhelming amounts of information are available in each research area making it a daunting task to do cross-comparisons of the different gene families. The pros and cons of different discovery methods (i.e., use of high-throughput screening, protein–ligand crystal structures, transient transfection assays, etc.) are included to help the reader understand the value and context of the biological evaluation of compounds currently available within each gene family. Another benefit added to this book is the biographies of the contributing authors compiled in Appendix. By reviewing the educational backgrounds and careers of each of the experts in the field, the reader can peruse the many paths of study one can take in a scientific journey for drug discovery. The reader can also learn how the scientific fields are integrated to design molecules for drug discovery. I sincerely hope you enjoy reading the fascinating way by which the scientists have managed to create small molecules that effectively modulate their intended biological targets in the major target classes covered in the chapters of this book. April 2008

KAREN E. LACKEY

ACKNOWLEDGMENT

Carol A. Petrick assisted in essential administrative aspects in the preparation of this work.

xiii

CONTRIBUTORS

Jerry L. Adams, GlaxoSmithKline, 1250 South Collegeville Road, Collegeville, PA 19426, USA. Paul Bamborough, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, UK. Frank E. Blaney, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, UK. David H. Drewry, GlaxoSmithKline, PO Box 13398, Five Moore Drive, Research Triangle Park, NC 27709, USA. Jo¨rg Eder, Novartis Institute for BioMedical Research, Novartis Pharma AG, WKL-136.6.93, CH-4002 Basel, Switzerland. Maria L. Garcia, Merck Research Laboratories, R80N-C31, PO Box 2000, Rahway, NJ 07065, USA. Stephen L. Garland, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, UK. Tom D. Heightman, Structural Genomics Consortium, University of Oxford, Old Road Campus Research Building, Old Road Campus, Roosevelt Drive, Headington, Oxford OX3 7DQ, UK. Anne Hersey, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, UK. xv

xvi

CONTRIBUTORS

Ulrich Hommel, Novartis Institute for BioMedical Research, Novartis Pharma AG, WKL-136.6.93, CH-4002 Basel, Switzerland. Gregory J. Kaczorowski, Merck Research Laboratories, R80N-C31, PO Box 2000, Rahway, NJ 07065, USA. Hiroyuki Kagechika, School of Biomedical Sciences, Tokyo Medical and Dental University, 1-5-45 Yushima, Bunkyo-ku, Tokyo 113-8549, Japan. Karen E. Lackey, GlaxoSmithKline, PO Box 13398, Five Moore Drive, Research Triangle Park, NC 27709, USA. David D. Miller, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, UK. Sandeep Modi, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire SG1 2NY, UK. Richard Sedrani, Novartis Institute for BioMedical Research, Novartis Pharma AG, WKL-136.6.93, CH-4002 Basel, Switzerland. Lisa Shewchuk, GlaxoSmithKline, PO Box 13398, Five Moore Drive, Research Triangle Park, NC 27709, USA. Aya Tanatani, Department of Chemistry, Ochanomizu University, 2-1-1 Ohtsuka, Bunkyo-ku, Tokyo 112-8610, Japan. Adrian Whitty, Department of Chemistry, Boston University, Metcalf Center for Science and Engineering, 590 Commonwealth Avenue, Room 299, Boston MA 02215, USA.

1 DRUG DISCOVERIES BY GENE FAMILY KAREN E. LACKEY

I am fascinated by the notion that a small molecule with a fraction of the molecular weight and size of a comparatively enormous biological protein can change the activity of that protein in directed ways. For years, the scientific community dedicated to drug discovery has developed a knowledge base of overwhelming proportions to hone these interactions of small molecules to seek out the intended target and seemingly avoid all other proteins in its path. We expect these molecules to enter the body in a convenient manner, interrupt the disease pathology, and quietly exit the body upon job done. While there are certainly successes out there, some of which are described in this book, there are so many more unmet medical needs that require our sense of urgency. When the human genome was solved, the expectations were that the targets to affect, avoid, or modulate human diseases would be tackled with drugs emerging at an unprecedented rate. Sometimes, I believe, the human genome project merely uncovered just how much we do not know and how far we need to go to understand the role of genes and protein products in normal and diseased tissues. Approximately 22,000 genes can be organized into the major protein classes they code and that are related by common structural and protein sequence features, called gene families, some of which are depicted in Fig. 1.1 (Hopkins and Groom, 2002). 1.1

GENERAL DRUG DISCOVERY COMPONENTS

Some areas of drug discovery methods transcend all gene families and will be briefly described here for general background to the subsequent chapters that are Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

1

2

DRUG DISCOVERIES BY GENE FAMILY

FIGURE 1.1 The gene families of proteins are classified by function and common structure motifs as can be seen by the representative structures for kinases, integrins, GPCRs, ion channels, proteases, and nuclear receptors. (See the color version of this figure in the Color Plates section.)

dedicated to the larger gene families. Broadly speaking, genomics is the study of all genes in an organism including the sequence, structure, function, and regulation. When referring to genetics, it is typically the inherited variation of genes that is being considered and is one component of genomics. Figure 1.2 summarizes the gene expression process with the general concepts of molecular biology where the rapid technology advances have made a major impact over the last few decades providing significant value to drug screening (Watson et al., 2003). In the nucleus, the DNA is transcribed into messenger RNA, which is transported to the cytoplasm. Translation occurs when transfer RNA (tRNA) reads the genetic code where three base pairs code for each amino acid of a polypeptide chain that forms the protein

Product DNA Transcription

Primary RNA RNA splicing

Messenger RNA RNA transport translation C

N

Protein Processing P

CHO

Functional protein

C

N P

FIGURE 1.2

The normal process and products of gene expression.

3

GENERAL DRUG DISCOVERY COMPONENTS

product. Protein products are the biological machinery where the normal functions are carried out, such as activation, inactivation, scaffolding, and signaling in a healthy cell. Drug intervention can occur at any part of this process, as well as in the modulation of the activity of the proteins produced. The discovery of drugs for diseases is typically selected based on medical need and research feasibility. A biological target is a component within a normal functioning cell or tissue where aberrant activity has given rise to a disease. For example, a genetic instability leads to mutations in cell signaling that can cause cancer. The target would become the biological component of the cascade to stop the unwanted or unchecked signal. A thorough understanding of disease pathology and/or the underlying genetics underpins the successful selection of targets. In some cases, genetic information can be used in patient selection in a field called pharmacogenomics. Whether the target is selected via genetic factors or by determining an ideal intervention point in the cellular protein activities, the ability to modulate the intended target using a small molecule is called tractability or druggability. The assessment of the tractability is shown in an oversimplified scheme in Fig. 1.3. Only a portion of the protein products of the human genome and infectious agents can be modulated by a small molecule or biological product (vaccine, antibody, etc.), and thus named the druggable genome (Imming et al., 2006). One way to increase the capability and understanding of biological target modulation involves organizing the protein products of the gene into families based on similar structure and function. This strategy allows teams of scientists to characterize the biology of the drug targets and build a knowledge base relevant to diseases that can readily be transferred from one member of the gene family to another.

Gene mRNA Protein Disease link (target) Confirm or modify approach

Test small molecules on target or related protein Lead or tool created I Drug

FIGURE 1.3 The process of target selection and validation is linked with assessing tractability or druggability.

4

DRUG DISCOVERIES BY GENE FAMILY

Compounds

Target assay

In vitro selectivity

Cellular and functional assays

In vivo studies: PK/PD

Toxicity assessment

FIGURE 1.4

A generic compound evaluation pathway for a drug discovery project.

It seems that target validation (defined as the ability to effectively treat the disease via modulation of the intended biological target) is an ongoing process from the time of target selection all the way until the desired activity is observed in a patient. The drug discovery industry continuously challenges the data through a series of increasingly more complex biological evaluations. A compound evaluation pathway is typically used to assess the ability of the small molecules and to assist with decision making, as summarized in Fig. 1.4. The labels in the diagram also help with nomenclature used in the field to describe the stages of identifying small molecules that interact with the biological target of interest. Treatment strategies from the molecular knowledge of a disease are used to prioritize targets and drug intervention points. Sometimes, it is necessary to create test compounds for multiple points in a particular disease pathology to determine the one that is the most tenable for a treatment of a patient. It could be that targeting one component of a pathway provides a more tractable target (i.e., a knowledge base exists for the gene family) or that using a test molecule in the assays that mimic the disease shows that it is insufficient to be an effective medicine. It still remains elusive to always predict which parts of a disease pathology are the most likely to be the best for producing the desired effects. Sometimes it is the wrong target, and sometimes the problem lies with compound’s activity profile. In early stages of research when little is known about the disease at the molecular level, it is very difficult to discern between the two types of failures. To begin the drug discovery process, the protein target is isolated and a screen is devised to test specific compounds, focused sets, and/or enormous sets of compounds of over 500,000 compounds often referred to as high-throughput screening or diversity screening. A compound must be potent toward the intended biological target, and the affinity is determined by an in vitro assay. In vitro assays are designed to measure antagonism (inhibition) of activity to stop an unwanted activity or agonism to provide more of an activity that has been lost or subdued by a

GENERAL DRUG DISCOVERY COMPONENTS

5

disease. There are many mechanisms that will be considered in the following chapters to achieve complete or partial antagonism or agonism that are specific to a gene family. Fundamental to all of them is a need to be able to assess the activity at the intended target in a way that allows a chemist to compare the inherent activity to either increase or decrease it depending on the drug intervention strategy. The screens often reflect the biological activity of the target, for example, by measuring the inhibition of an enzymatic reaction. Screens are also built to measure binding affinity—how tight the compound sticks to the target, with the assumption that by binding to certain parts of a protein, the target can be modulated. Some targets cannot be screened as an isolated assay and must be screened in a cellular setting. Throughout each of the following chapters, details regarding the type of assays that are used in each gene family of targets and reasons for the choices will be discussed. The compound evaluation pathway is designed to enable decisions on what compounds to synthesize. Designing compounds based on the data that are generated in each of the assays used in a screening scheme involves computational chemistry, structural biology, in silico predictions, and mathematical analyses. Different gene families rely on specialized design tools in combination with physicochemical assays. The overall process is an iterative learning cycle in which medicinal chemists generate ideas for structural changes to simultaneously optimize activities across all of the assays employed. Chemists working on a gene family system use their knowledge of the target class assisted by computational techniques to design compounds to find biological activity. Sets of compounds, often called arrays, are created to test hypotheses regarding structure–activity relationships (SAR). Where there is more protein structure information or a long history of designing active compounds, the sets of compounds are generally small as the chemist hones in on the best molecule. When less is known or available to guide the design, much larger sets of compounds are generally synthesized (>500). Taking advantage of special pockets or distinct binding interactions can help to narrow the scope of what the drug will modulate. By focusing on gene families, methods to design selectivity for one family member compared with another are discussed. It is noteworthy that building in this selectivity still requires extensive toxicity testing to verify minimal off-target effects. After a screening event, the data are analyzed to determine if any compounds provide the readout intended. The sets may be tested in a general format to cast as wide a net to find as many active starting points as possible. Testing many compounds at a single concentration is often used. Anything that provides a hint of activity is typically called an assay ‘‘hit.’’ The hits are analyzed to check that they are truly active and interesting to work on. They are often narrowed to only include chemical series that will be further worked on via synthetic modifications. Depending on the gene family, the hit validation can be as simple as generating a value that is called the effective dose to give a 50% response. The reason this value is used is because it makes comparing compounds much more reliable based on using a reproducible point on the curve. It is expressed as a concentration or a pXC50 (the negative log of the IC50 for better numeric comparisons of activity).

6

DRUG DISCOVERIES BY GENE FAMILY

It is also important to understand the error range of the data. Often replicates are averaged to provide values for comparison among series of chemically related compounds and that gives a fairly good representation of what activity can be used for analyses. However, to assess how significant an improvement or decrease in activity is, it is important to get an understanding of the variability of the data. Often, pED50 values are useful for preventing overinterpretation of the data. For example, if you compare compound A with an IC50 value of 5 nM with compound B that is 20 nM, you might be tempted to say the compound A is fourfold more active than B. However, if the average range of the assay data on repeat tests is 0.2 log units, then the compounds would be considered equipotent. The data interpretation aspects are different among the various gene families, given the state of the art of the assays systems, and some discussion of the reliability of the data is covered in each chapter. Also, the term high-content lead series is often used to describe a chemical series that meets many of the criteria for modulating the target of interest and often requires the parallel use of several hit generation and follow-up techniques (Bleicher et al., 2003). The activity against an isolated target in an in vitro assay is often verified or further evaluated in a cellular environment. In cases where it is not possible to create an assay for a desired target, whole cells or more complex systems are used to find active small molecules that incorporate the desired cellular outcome, or phenotype, as the assay system. These assays are typically lower throughput (less compounds are screened because the assay is either more difficult to run or too expensive) but more information rich. Also, it might be necessary to run control assays to ensure that the activity that is observed is related to the desired biological target modulation. For many drug discovery projects, it is also important to test for selectivity against other similar targets or important conflicting biological pathways. Each of the gene families has specific selectivity challenges that will be discussed for greater understanding of how to design a compound that is effective for the disease area without unwanted or intolerable side effects. For example, often screening is done on the human protein target, but the activity must also be checked on the animal orthologue (the protein found in the animal that is analogous to the human version) to support the in vivo animal studies of efficacy and toxicity. Data generated from evaluating compounds in cell-based assays are important to integrate with any isolated enzymatic or binding assay. A compound’s activity is affected by cellular context that includes multiprotein complexes, cell-type specific pathways, branching pathways, intracellular localization, and multiple target isoforms or states (phosphorylated, open, closed, etc.). Of course, compensatory bypass or feedback loop mechanisms cannot be detected in an isolated target assay and thus allow the cell assay to be used as a way to judge the tractability of the drug intervention point. In this way, cellular assays can better measure efficacy, not just potency. As a generalized screening tool, cell-based assays can also be set up to select for only permeable, nonmetabolized, and nontoxic compounds. This method of identifying hits and optimizing compounds can be useful in limiting the number of chemical series. On the contrary, it automatically eliminates potentially potent compound interactions that may need properties such as solubility or permeability

GENERAL DRUG DISCOVERY COMPONENTS

7

built in. Each target class strategy of molecular design balances these aspects of cellular screens based on the typical properties or needs of the drug discovery research efforts. The SAR is the study of the correlation between the biological activity and the chemical structure. There can be SAR at every level of the screening cascade, and it guides the entire process of designing an effective medicine. The knowledge derived from SAR forms an understanding of interaction features such that predictions can be made on what kinds of molecules will retain activity and which ones will not within a chemical series. There are nearly always compromises that can be made in potency to add solubility and to create an improved therapeutic index (ratio of the dose required for biological efficacy to the dose at which unacceptable toxicity occurs). All along the project path, it is important to be aware of the SAR to know where on the molecule changes can be made to adjust for newly uncovered risks or to be aware of what cannot be changed without losing all target affinity. This book should help the reader know what types of compounds to synthesize in each gene family, and examples are provided to understand the scope of the generic properties of the small molecules. Each family of targets appears to depend on a few major design techniques based on the properties of the biological targets. However, paramount to the success of creating a drug candidate with optimized gene family protein interactions is the ability to incorporate drug developability characteristics. Requirements of developability properties are different for the variety of intended routes of administration (e.g., oral, inhaled, and intravenous) and typically cover absorption, distribution, metabolism, and excretion, which are abbreviated as ADME properties. There is a whole branch of science supporting the necessary ADME features and components that affect them related to disease pathology and normal body functions. Although the details are beyond the scope of the target class design focus, these developability properties must be optimized while generating potent interactions with the intended biological targets. Developability, metabolism, and pharmacokinetics (DMPK) encompass many types of assays that are used to determine the in vivo properties of potential drug molecule. While the emphasis in this book focuses on designing molecules for specific biological targets within gene families, the ultimate design of the compounds must take into consideration the properties that allow the compound to get to the tissue or site of action. In vitro tests to measure solubility at various pH levels can often be done to predict absorption levels or the handling properties of the drug product. The activity of compounds is measured in a variety of in vitro assays with CYP isoforms (e.g., CYP3A4 or CYP2D6) to predict which compounds might have better or worse in vivo properties based on predictions on metabolism of the parent molecule. The principal evaluation of in vivo properties comes by testing representative compounds in rodent models looking for data to determine half-life, bioavailability, volume of distribution, and clearance of the compound. A thorough discussion of this topic is beyond the scope of the book, but it is important to note that each target class discussed may require certain types of compound classes for proper interactions to modulate the target. These features may offer particular challenges in

8

DRUG DISCOVERIES BY GENE FAMILY

designing an effective medicine and will be pointed out in the chapters that follow. A high-level example, though, is to look at the properties of many drugs that modulate nuclear receptors. They tend to be lipophilic in nature and, in contrast, the kinase compounds that interact at the ATP binding domain are frequently ‘‘flat’’ heterocycles with inherently poor solubility. These features are addressed in drug discovery research projects and it helps to have a knowledge base in the area to facilitate the appropriate design focus. More recently, in silico techniques have made huge advances in the area of predictive properties. Many software packages are available on the computer desktop for either straightforward analyses or more complex analyses for a trained computational chemist. They are used to identify potential issues with active series identified from a screening event to help prioritize the hits with the highest likelihood of generating high-quality compounds. The in silico methods are also used to prioritize a synthetic plan by taking chemistry that is designed to functionalize one or more sites on a central scaffold and create the structures by building the final products as a virtual set (existing only as an electronic collection of structures). The virtual compounds are then put through predictive models in order to select the subset of compounds that would be predicted to retain the desired activity. For example, if the intended target is in the central nervous system (CNS), then an in silico method can be used to virtually screen a large set of possible compounds for CNS penetration property predictions. The purpose of such an exercise is to reduce the number of compounds required for synthesis, thus reducing time, chemical reagent costs and waste, and biological screening costs. To predict oral absorption using in silico predictions, the focus is on oral bioavailability (how much drug reaches the blood compared to the original dose) and predicting interactions with metabolizing enzymes (e.g., enzymes in the liver designed to rid the body of toxins might degrade a compound rapidly or too slowly). Virtual screening techniques are often used for the gene family target interactions. Chemical descriptors form the basis of the structure input and often account for the nature of the compounds, such as molecular size, polarity, hydrogen bonding capacity, lipophilicity, and acidity/basicity. As a medicinal chemist builds knowledge in each of the target classes, the data from virtual and measured screening are used together to form the project strategy in creating a drug. Lipophilicity is measured by log P, which is a partition coefficient that describes the ratio of the concentrations of a compound between two phases: octanol and water. Why this matters in drug design is due to the fact that the cell membrane permeability (ability of compounds to pass through the cell membrane to reach intracellular targets) requires more lipophilic nature, which has a tendency to be less soluble in blood/aqueous media. Conversely, hydrophilic compounds tend to be water soluble and amenable to blood transport, but are poorly cell permeable. This measure of lipophilicity requires a balance of design between the various effects to create a drug that gets to the site of action and retains the key functional groups necessary for target interaction for the desired biological effect. A calculated log P value, designated clog P, has become fairly easy to calculate and is often used to prioritize which of the compounds should be synthesized given the ideal range

GENERAL DRUG DISCOVERY COMPONENTS

9

of 2–4 (for oral drugs), where the larger the number the greater the lipophilicity (Wenlock et al., 2003). Solubility is an important physical property for a drug for both biological effectiveness and drug product formulation and processing. Solubility testing can be done in a variety of conditions to mimic neutral aqueous, gastric fluid simulation or intestinal fluid simulation, or any pH necessary for drug delivery. For oral drugs need to be well absorbed by the body, unless they are designed to specifically combat a GI tract disease for which systemic exposure is unnecessary (e.g., certain bacteria infections of the gut). Understanding the solubility characteristics of series can also assist in interpreting the biological data from screening. Care should be taken to ensure that the compounds are fully solubilized in the test solution to study the SAR trends. In silico methods such as structure-based design and pharmacophore modeling are best covered under the specific target class chapter. When the three-dimensional structure of an enzyme or receptor is known, the information can be used to determine the interactions between it and a bound compound. Computer models can be generated to mimic other similar molecules bound to the active site so that designing can be based on maximizing interactions (while taking into account all of the other properties). It is often referred to as finding where the molecule can tolerate a substitution that would allow the synthetic chemist to attach functional groups to improve the properties without destroying the positive interactions with the ligand/protein. Crystallography can be used to design ideal drug molecules and will be highlighted in the gene families where it currently has the greatest impact. Often, the three-dimensional structure of a protein target is not known, but using binding or catalytic activity of compounds from screening events, a pharmacophore model can be built. A high-level explanation follows: A set of compounds that are known to have activity for the intended target can be used to gain insight into the steric and electronic structural features necessary for a compound to bind. Even if they are poorly active, a model can be built to identify the specific features that garner any activity. Iterative searching and testing in the in vitro assay allows the scientists to refine the pharmacophore model and hone in on the key features for better and better affinity. Also, models can be built based on the similarity of proteins or protein folds from one member of a protein family to another. Often, the models help scientists understand how changes in the small molecule affect the activity observed in the assay and designs incorporate better interactions through appending functional groups. In typical drug discovery programs, assays and in silico predictions such as solubility, in vitro metabolic stability, and cell permeability are used as filters for determining ideal pharmacodynamic and pharmacokinetic properties (Leeson et al., 2004). There seems to be an ongoing discussion in the medicinal chemistry field over the value of rules in drug design, but most scientists agree that trends can be derived for general predictions (Leeson and Springthorpe, 2007). However, within a given chemical series, there can be anomalies and those should be considered when using assays for guidance in progressing compounds through more complex and resource-intensive biological evaluation systems. Specific examples of screens

10

DRUG DISCOVERIES BY GENE FAMILY

for in vitro developability assays include solubility in multiple solvents (neutral aqueous, simulated gastric fluid, and simulated intestinal fluid), Caco-2 permeability, p450 enzyme assays, and microsomal metabolism studies using methods similar to published reports (Guo and Shen, 2004; Li 2004). It is invaluable to have training sets of compounds to investigate correlations between in vitro assays designed to predict drug-like characteristics and actual in vivo data within chemical series. The entire discussion on protein family affinity and modulation coupled with incorporating the desired parameters for delivering the small molecule to the site of action is only a fraction of the drug discovery process. Most researchers will agree that the more the quality is put into the compounds in the early stages of identifying a potential drug candidate, the more likely it will be successful in reaching a patient in need. However, as can be seen by the remarkable financial investments (Goozner, 2004) made in the field, I still believe that the combination of knowing which intervention point to modulate (target selection and validation), achieving selective target interactions with no off-target activities, meeting all of the druglike criteria for the intended route of administration, and being able to synthesize sufficient, pure quantities of material in the correct form is a formidable task to expect of one small molecule. However, there are success stories out there. We owe it to ourselves to understand how to modulate every potential biological target to enable successful drug discovery for the patients still in need.

1.2

FURTHER READING FOR EXPERT KNOWLEDGE

1. Textbook of Drug Design and Discovery, 3rd ed., edited by Povl KrogsgaardLarsen, Tommy Liljefors, and Ulf Madsen, Taylor & Francis, Inc., New York, 2002. This book is a thorough guide to drug discovery with a compilation of chapters going into details on mechanics of small molecule–protein interactions, kinetics, and development of pharmacophore models and descriptors. A terrific explanation of quantitative SAR with physicochemical parameters and how to interpret the data is also included. A general explanation of receptorbased targets in multiple gene families is provided, with subsequent chapters discussing in more detail on ion channels and therapeutic outcomes from modulation of targets. Creating and interpreting data from imaging studies with labeled compounds provides insights into understanding drug distribution and why so much effort goes into understanding physicochemical properties of molecules. The chapter on enzymes with mechanism and kinetics of inhibition offers many drug discovery approaches to these important biological reactions. More advanced topics in drug design include prodrugs (a precursor to the active parent drug typically done to mask an unwanted property of the compound and dependent upon the human metabolism), peptides and their mimics, and the use of metals in modulating biological targets that are dependent on them for their normal processes. The textbook ends with specific examples of drug discovery in the research area of antiviral and anticancer agents where many of the concepts discussed throughout the

FURTHER READING FOR EXPERT KNOWLEDGE

11

book are shown in practice of selecting targets in the disease pathology and determining how to modulate them for a beneficial effect. 2. Biopharmaceutical Drug Design and Development, edited by Susanna WuPong and Yongyut Rojanasakul, Humana Press, Inc., Totowa, NJ, 1999. It provides excellent descriptions of using biotechnology in generating very specific target modulation and the unique properties of these biologicals in creating useful drugs to combat diseases. It is an easy to read scientific book that offers a marked contrast to small molecule design in the gene family focused molecular design. Vaccines and recombinant human products, for example, have excellent specificity but offer difficult challenges to manufacture, deliver to the site of action in the body, and management of stability issues throughout the entire process of ‘‘bench to bedside.’’ Many technology advances have been made and continue to be made to provide effective medicines in this class of drugs, and this book provides a foundation of knowledge to appreciate the more expert literature. 3. Guidebook on Molecular Modeling in Drug Design, edited by N. Claude Cohen, Academic Press, San Diego, CA, 1996. There are many computational approaches needed to operate in the potential chemical space that can be reached with synthetic chemistry to maximize interactions with the intended biological target(s). This guidebook delves into the details of the tools used with a variety of molecular capabilities and applies them to examples in ligand generation. The complexity of drug–biological target interactions that take into account pharmacophores, desolvation, and binding properties is discussed from the perspective of computer-assisted drug discovery and modeling. While the book is over 10 years old, it has a good foundation for anyone interested in understanding the breadth of contributions from computations chemistry to the design of drugs. 4. Protein–Ligand Interactions. From Molecular Recognition to Drug Design, edited by H.-J. Bo¨hm and G. Schneider, in: Methods and Principles in Medicinal Chemistry, Vol. 19, edited by R. Mannhold, H. Kubinyi, and G. Folkers, Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, 2003, ISBN 3-527-30521-1. This book is a must-have expert reference for any serious medicinal chemist who wants to understand the intimate interactions of a small molecule and a large biological protein. From the list of abbreviations provided in the front of the book to the well thought-out order of chapters describing details of small molecule modulation properties, the readers will, at a minimum, understand the nomenclature of the business partners they work with and, at best, will utilize the appropriate principles that affect the SAR of the drug they are hoping to discover. Most of the formulas are broken down into the important features of the design tools, so even if you were not going to use the calculated approaches described, the lessons that are included add insights for the medicinal chemist. The combination of computational chemistry and compound screening explanations is useful for understanding all of the factors that go into compound design at the molecular recognition detailed level.

12

DRUG DISCOVERIES BY GENE FAMILY

5. Receptor-Based Drug Design, edited by Paul Leff, Marcel Dekker, New York, 1998, ISBN 0-8247-0162-3. This book is part of a long series of textbooks dedicated to drugs and pharmaceutical sciences. It is a useful reference book for a more in-depth look at a subset of the gene families where the biological target is a receptor. Details and insights are provided on the functions of a receptor such that thorough interpretation of assay systems can be achieved. Mechanisms of interactions range from determining binding affinity to understanding many forms of modulation. Understanding the outcome of the effect of small molecule on the receptor in an information-rich assay system should help guide the reader to more expert advice on judging the tractability of a receptor for drug discovery. While advances in assay systems for receptors have advanced over recent years along with a large increase in the number of ligand-bound protein crystal structures, this book will still provide expert details for more in-depth understanding of the reader’s research project and current scientific literature. 6. Analogue-Based Drug Design, edited by Janos Fischer and C. Robin Ganellin, Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, 2006, ISBN 3-527-31257-9. This book is one in a series of six titles dealing with different aspects of drug discovery. It covers the general approach that a synthetic chemist can use to explore chemical space around an established active compound with the potential for differentiating the final product. Nineteen specific case studies are provided in detail to demonstrate how the approach works, the parameters used to differentiate analogues, and what the properties were of the final drug candidate. 7. Structure-Based Drug Discovery, edited by Harren Jhoti and Andrew Leach, Springer, Fordrecht, The Netherlands, 2007, ISBN 1-4020-4406-2. Understanding the cellular proteins at the molecular level from threedimensional protein structures allows researchers to design specific interactions to create potent, small molecule drug candidates. This book details the process improvements in protein crystallization and structure determination, and provides an advanced view of scaling the technologies for parallel production for higher throughput. A general guide to using structure in discovering small molecule fragments and strategies for converting the fragments into lead series is also included. Multiple methods of screening for fragments of drug-like molecules are covered including NMR and X-ray crystallography. An alternative to linking fragments together to increase affinity is discussed, whereby the screening event begins with finding the small core scaffold followed by functionalization to improve affinity. All these methods of finding ligand-efficient hits depend upon understanding the mechanisms of the interactions between the small molecules and the protein, so the chapter on biophysical methods is useful. The final chapters cover computational ways to extend the value of the structure-based design methods.

REFERENCES

13

REFERENCES Bleicher KH, Boehm H-J, Mueller K, Alanine AI, 2003. A guide to drug discovery: hit and lead generation: beyond high-throughput screening. Nat. Rev. Drug Discov. 2(5):369–378. Goozner M, 2004. The $800 Million Pill. University of California Press, Berkeley, CA. Guo Y, Shen H, 2004. pKa , solubility, and lipophilicity: assessing physicochemical properties of lead compounds. Optimization in Drug Discovery. Humana Press, Totowa, NJ, pp. 1–17. Hopkins AL, Groom CR, 2002. The druggable genome. Nat. Rev. Drug Discov. 1:727. Imming P, Sinning C, Meyer A, 2006. Drugs, their targets and the nature and number of drug targets. Nat. Rev. Drug Discov. 5(10):821–834. Leeson PD, Springthorpe B, 2007. The influence of drug-like concepts on decision-making in medicinal chemistry. Nat. Rev. Drug Discov. 6(11):881–890. Leeson PD, Davis AM, Steele J, 2004. Drug-like properties: guiding principles for design—or chemical prejudice? Drug Discov. Today: Technol. 1(3):189–195. Li AP 2004. In vitro approaches to evaluate ADMET drug properties. Curr. Top. Med. Chem. 4(7):701–706. Watson JD, Baker TA, Bell SP, Gann A, Levine M, Losick R, 2003. Molecular Biology of the Gene, 5th ed. Addison Wesley, Boston, MA. Wenlock MC, Austin RP, Barton P, Davis AM, Leeson PD, 2003. A comparison of physiochemical property profiles of development and marketed oral drugs. J. Med. Chem. 46(7):1250–1256.

2 G-PROTEIN-COUPLED RECEPTORS STEPHEN L. GARLAND

2.1

AND

TOM D. HEIGHTMAN

INTRODUCTION

G-protein-coupled receptors (GPCRs) comprise a large family of cell surface receptors with highly diverse cognate ligands and biological functions. GPCRs are also referred to as seven transmembrane (7TM) receptors, because of their characteristic configuration of an anticlockwise bundle of seven transmembrane a-helices. Functionally, they may be considered as conduits of specific extracellular information through the cell membrane into the cytoplasm. Binding of an extracellular ligand to the receptor induces conformational changes in the seven transmembrane helices, leading to activation of a G-protein complex, triggering an intracellular signaling cascade with concomitant second messenger release, and ultimately leading to a physiological response that may be immediate (e.g., ion channel modulation) or delayed (e.g., transcriptional modulation). Activating ligands range from small particles (a photon, a calcium ion), through small amino acid-derived neurotransmitters (serotonin, dopamine, noradrenaline), lipids (endocannabinoids, sphingosine phosphate, lysophosphatidic acid), oligopeptides (neurokinins, opioids, ghrelin) to large peptides (parathyroid hormone, chemokines). These endogenous ligands contribute to processes as wide ranging as regulation of blood pressure, cognition and motor function in the CNS, signaling of reflexes such as pain and nausea, transfer of immune signals, and regulation of energy intake and expenditure (see Table 2.1). Such a high diversity of ligands has led to the evolution of an equal diversity in ligand binding pockets, and the apparent ease with which small molecule agonists and antagonists have been discovered for monoamine receptors, for example,

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

15

16 TABLE 2.1

G-PROTEIN-COUPLED RECEPTORS

GPCRs and Their Link to Disease

Receptor (Mode)

Subfamily

Endogenous Ligand

Disease

5HT1a (ag) 5HT1b/d (ag) 5HT2x/Dx (antag) b1-adrenoceptor (antag) b2-adrenoceptor (ag) D1 (antag) H1 (antag) H2 (antag) M3 (antag) AT2 (antag) GnRH (ag)

A/aminergic A/aminergic A/aminergic A/aminergic

Serotonin Serotonin Serotonin, dopamine Adrenaline/noradrenaline

Depression Migraine Schizophrenia Heart failure

A/aminergic

Adrenaline/noradrenaline

Asthma

A/aminergic A/aminergic A/aminergic A/aminergic A/peptide A/peptide

Parkinson’s disease Allergy Peptic ulcer COPD Hypertension Prostate cancer

NK1 (antag) m-Opioid (ag) CCR5 (antag) BLT1 (antag) CB1 (antag) EP1 (ag) GLP-1 (ag) PTH1 (ag) Calcitonin (ag) CaSR (pos mod) GABAb (ag)

A/peptide A/peptide A/chemokine A/lipid A/lipid A/lipid B B B C C

Dopamine Histamine Histamine Acetylcholine Angiotensin-II Gonadotropin releasing hormone Neurokinin Enkephalin MIP-1b Leukotriene B4 Cannabinoid Prostaglandin E1 GLP-1 PTH Calcitonin Calcium Ion g-Aminobutyric acid

CINV Pain HIV infection Asthma Obesity Peptic ulcers Type 2 diabetes Osteoporosis Osteoporosis Secondary HPT Spasticity

COPD ¼ chronic obstructive pulmonary disease; CINV ¼ chemotherapy-induced nausea and vomiting; HPT ¼ hyperparathyroidism.

does not necessarily apply to receptors where the ligand is a large peptide; however, it is also important to note that the function of many receptors may be modulated by allosteric ligands binding at sites distinct from the endogenous ligand, and hence not constrained by the molecular property requirements of the orthosteric binding site. Many early drugs act on GPCRs, including noradrenergic receptors (‘‘betablockers’’ acting on beta-1 subtype receptors, asthma drugs acting on beta-2), histamine receptors (antihistamines at H1 receptors, antacids at H2 receptors), and opioid receptors (morphine), and this early success has been followed with more recent treatments targeting receptors such as angiotensin (hypertension), CCR5 (HIV infection), and cannabinoid (obesity). It is estimated that 30–50% of all marketed drugs target GPCRs (Drews, 2000; Wise et al., 2002) (Fig. 2.1). The receptor targets for many early drugs were discovered post hoc; by contrast, modern drug discovery has been driven by the association of new genome-derived receptors with their endogenous ligands (a process known as

17

INTRODUCTION

Aminergic Receptors HO

O O

N H

O

N

O H O S N

NH

O

N O

HO N H

OH

N H

N H

HO CoregTM (carvedilol) beta1 antagonist

RequipTM (ropinirole) D1 antagonist

ImitrexTM (sumatriptan) 5HT1D agonist

SereventTM (salmeterol) beta2 agonist O O

N

S

N O

OH

NH

OH

O

H

N

N H

O

HO

F

N

SpirivaTM (tiotropium) M3 antagonist

F

O

N

HN N

O H H

CF3

O

N

N N N NH

O O

Peptidergic Receptors O

S O

H

ZantacTM (ranitidine) H2 antagonist

Cl

+

+

S

AllegraTM (fexofenadine) H1 antagonist

N

Me O

N H

H2N

N

N

N

N Me CelsentriTM(maraviroc)

EmendTM(aprepitant) NK1 antagonist

F

CozaarTM((losartan)) Angiotensin II antagonist

Me

NH

CF3

CCR5 antagonist t i t

N

N N

HN

O

O NH O

H N

N H

NH

HN

HN

HO O

O

N

N H

O

O

O N H

O

ZoladexTM(goserelin) Growth Hormone agonist

O

Lipid Receptors

OH

O

OH

HO

H N

nh

OH

O

Morphine Opioid agonist

O

Family C Receptors

O

OH

O

Cl

H2N

O O N N

Cl

H

H

O

H N

OH

O

H

N H

O

N

O

O HO

N N H

Cl AcompliaTM(rimonabant) CB1 inverse agonist

F F

OH

CytotecTM(misoprostol) EP receptor agonist

F

N H

SensiparTM(cinacalcet) CaSR positive modulator

Cl Baclofen GABAb agonist

FIGURE 2.1 Examples of marketed drugs targeting GPCRs. (See the color version of this figure in the Color Plates section.)

deorphanization) and delineation of their role in pathways relevant to disease (Bakker and Leurs, 2005). Due to their upstream signaling role in numerous physiological regulatory pathways, together with their extracellular location and ligand binding pockets amenable to small-molecule recognition, GPCRs make a highly attractive target class for therapeutic intervention. However, this tractability is mitigated by complicating factors such as GPCR homo and heterodimerization, and emerging regulatory protein networks.

18

2.2 2.2.1

G-PROTEIN-COUPLED RECEPTORS

GPCR STRUCTURE AND FUNCTION Subfamilies

GPCRs can be categorized into six subfamilies named Family A through F (or numbered 1 though 6) (Attwood and Findlay, 1994). From a drug discovery perspective, Family A, Family B, and Family C are the most relevant, of which Family A is the most important due to its size. Family D (fungal pheromone) and Family E (slime mold cAMP receptors) are not found in higher species, so they are not discussed further here. The Frizzled/Smoothened groups of Family F could very well be interesting drug targets in the future but have yet to be studied in much detail. Suffice to say there are 10 Frizzled and 1 Smoothened receptor in the human genome, the Wnt proteins are ligands for Frizzled and Patched for Smoothened, in a few, but by no means all instances they have been shown to signal via G-proteins and the receptor/ ligands appear to have a role in controlling cell fate and proliferation (Table 2.2). A thorough phylogenetic analysis of human sequences has yielded an alternative means of GPCR classification called the GRAFS system (Schioeth and Fredriksson, 2005). This system places the receptors in five main subfamilies: Glutamate, Rhodopsin, Adhesion, Frizzled/taste2, and Secretin. This is essentially equivalent to the A–F classification, except that the nonhuman D and E families are missed out, and the Family B group is split into two to yield the peptide hormone receptors (secretins) and the adhesion class. This partitioning appears justified because the receptors have major differences in the N-terminal domains (NTDs), even though there are similarities in the transmembrane regions. The adhesion receptors can have very large NTDs characterized by a GPCR proteolytic site (GPS) domain that appears cleaved in some and potentially all receptors, but with the NTD remaining associated with the receptor, presumably via disulfide bonds (Foord et al., 2002). They are sometimes referred to as EGF-7TMs to reflect the presence of epidermal growth factor (EGF) domains in the NTD for certain examples or LN-7TM/LNBs to reflect their long N-terminal domains. There are 33 adhesion receptors in the human genome, of which 30 are currently orphans. As with the Frizzled receptors, they are relatively poorly studied and hence omitted from further discussion here. TABLE 2.2

GPCR Families and Their Sizes in Human A (Rhodopsin)a

Liganded Orphan Total

189–217 67–95 284

B (Secretin)

B (Adhesion)

C (Glutamate)

F (Frizzled)

15 0 15

3 30 33

15 7 22

11 0 11

The numbers in this table were obtained from Gloriam et al. (2007) and the International Union of Pharmacology (IUPHAR) list (Foord et al., 2005 and http://www.iuphar-db.org/GPCR/ReceptorListForward). The number of orphan receptors in Family A/Rhodopsin group will vary depending on the criteria used, specifically whether repeated evidence from unique sources is required. a Excluding olfactory receptors and taste type 2 receptors.

GPCR STRUCTURE AND FUNCTION

19

All of the GPCR families are believed to share a common topology, with an extracellular N-terminal segment, seven helices of ca. 25–35 residues running in alternating directions through the cell membrane (in anticlockwise manner when viewed from the extracellular side) linked by six interconnecting loops (three intracellular and three extracellular) and a C-terminal tail to the intracellular side of the membrane. An eighth helix (HLX8) is also seen in the currently available crystal structures. This helix sits along the membrane face to the intracellular side, after helix seven and along to the C-terminus, rather than running through the membrane. Hence, GPCRs have seven transmembrane helices and are often referred to as 7TMs; indeed, this is considered by many to be the more appropriate name for the class as not all of the receptors may be G-protein coupled and/or they can demonstrate alternative means of signaling (Fig. 2.2). GPCR or 7TM nomenclature typically refers to the NTD (N-terminal domain), TM1 through TM7 for the transmembrane helices, ICL1 through ICL3 for the intracellular loops, ECL1 through ECL3 for the extracellular loops, and the C-term (C-terminal domain). The variation between the structural types for Families A, B, and C relates mostly to the size of the N-terminal domain. For Family A receptors, this is relatively short, whereas for Family B peptide hormone receptors it is typically 120–150 residues in length and for Family C it is approximately 900–1200 residues. These structural aspects are discussed in more detail later in the chapter. For most members of Family A, the orthosteric (endogenous ligand) binding site is in the TM bundle. The residues from the ECLs may play a role in forming the top of the site. As a rule, the NTD contributes little, if anything, to the site. There are a small number of exceptions to this such as the LH, FSH, TSH, and LG receptors. For Family B, both the NTD and the TM bundle form a part of the orthosteric site

FIGURE 2.2 Semischematic representation of a G-protein coupled receptor. The picture is generated from the coordinates of the b2-adrenergic receptor crystal structure 2rh1 with transmembrane helices colored as follows: TM1 (orange), TM2 (green), TM3 (light blue), TM4 (dark blue), TM5 (violet), TM6 (red/brown), and TM7 (pink). The extracellular and intracellular loops are colored yellow, although ICL3 is missing due to excision of T4lyzozyme. Also shown are helix 8 (purple) and the helical section of ECL2 (cyan). (See the color version of this figure in the Color Plates section.)

20

G-PROTEIN-COUPLED RECEPTORS

(i)

(ii)

(iii)

FIGURE 2.3 Schematic diagram showing the differences in binding of endogenous ligands by Family A (i), Family B (ii), and Family C (iii) GPCRs. The approximate position of the orthosteric site is shown in green. (See the color version of this figure in the Color Plates section.)

and for Family C, the NTD defines the entire of the orthosteric site. The structural differences are shown schematically in Fig. 2.3. The Family A receptors are also referred to as the rhodopsin family after a key member. Rhodopsin is the light receptor present in the eye and enables vision. The ligand—effectively a photon—is rather unusual, and hence so is the receptor in that it has a 11-cis-retinal molecule covalently bound. Light causes cis/trans isomerization of the double bond, which in turn causes a conformational change and switches the receptor from the inactive to active state to yield a signal within the cell. In addition to the opsins, there are also a large number of olfactory Family A receptors, again exploiting GPCR activation upon ligand binding as a detection mechanism, with sequence variation in the binding site providing a means to tune the detector. The opsin and olfactory receptors are not generally regarded as drug targets. However, Family A is sufficiently large that there remain
280 nonsensory receptors that are of potential interest to the pharmaceutical industry (Bjarnadottir et al., 2006;

GPCR STRUCTURE AND FUNCTION

21

Gloriam et al., 2007), of which only
30 have so far been successfully tackled. Despite the fact that the industry has been really quite successful with this class of targets, the majority remain to be exploited and certain of the subfamilies still present challenges. This topic is discussed further in Section 2.3. The binding site occupied by the retinal molecule in rhodopsin is thought to be present in other GPCRs, albeit with different amino acid sequences lining the cavity and hence different ligand specificities. The main difference is that the drug target receptors do not have moieties covalently attached and the site is available for small-molecule binding. There will most likely need to be a degree of receptor flexibility, however, to open up an access channel to the binding site, for example, through a breathing motion of the extracellular loops. This flexibility has yet to be explored in any great detail but may explain some receptor selectivities and, in particular, binding kinetics. There are a number of key amino acid sequence motifs that are highly conserved in Family A. Most notably are an asparagine to the cytoplasmic side for TM1; an aspartate at the same level for TM2; TM3 has a cysteine to the extracellular side that is involved in a disulfide bond to ECL2 and the DRY motif to the cytoplasmic side; TM4 has a tryptophan in the middle of the helix; and TM5, TM6, and TM7 each contain a proline. Hydrophobicity plots can be used to help identify broadly where each TM segment is within the receptor sequence and then coupled with knowledge of these key residues for sequence alignment and homology model building. A standard numbering convention for residues within the TM regions is generally adopted (Ballesteros and Weinstein, 1995). This ascribes the value 50 to the most highly conserved residue within each helix; residues are then indexed by reference to this in the form N.nn, where N is the helix number and nn is the position. For example, a residue 10 amino acids to the N-terminal side of the most conserved position in helix 3 would be position 3.40. The most conserved residues are Asn (1.50), Asp (2.50), Arg (3.50), Trp (4.50), Pro (5.50), Pro (6.50), and Pro (7.50). The predominance of prolines in the definition is perhaps to be expected given their propensity to distort a helical arrangement and hence the likelihood that their presence in the sequence has some key structural/functional role and thus must be conserved. The Family B peptide hormone receptors are also known as the secretin family, after the first member to be cloned. There are 15 such receptors in the human genome. They are activated by peptides of ca. 30–40 amino acids in length and have been implicated in a wide variety of physiological responses (Hoare, 2007). More than half have already been pursued as drug targets. In certain cases, the peptide agent itself has been developed and is used therapeutically. For example, Calcitonin and PTH1 (parathyroid hormone-1) peptides are used for osteoporosis, and a reptilian analogue of GLP-1 (exenatide/ByettaTM) has recently been developed for Type 2 diabetes. However, being peptide agents, these suffer from issues around route of administration, typically being injectable or nasal formulations. Identification of small-molecule agonists or antagonists as potential oral therapies remains a challenge for this family.

22

G-PROTEIN-COUPLED RECEPTORS

Family C is also referred to as the glutamate class after the eight metabotropic glutamate receptors (mGluR1–8). Other key drug targets within this family include two subtypes of gamma-aminobutyric acid (GABA) receptor (GABAbR1 and R2) and the calcium-sensing receptor (CaSR). The remaining members of the family are three taste receptors (T1R), a promiscuous L-a-amino acid receptor and seven orphans. Receptor dimerization appears to play quite a considerable role in the pharmacology of Family C receptors. This is most notable for the GABAb receptor, which is an obligate dimer of the two receptor subtypes. That is, a functional receptor is obtained only through heterodimerization of GABAbR1 with GABAbR2. The particular importance of forming the heterodimer in this case appears to be related to receptor trafficking to the cell surface. Receptor dimerization is discussed in more detail later in the chapter. The phylogenetic classification of receptors based on whole sequence similarity is useful to categorize them into broad groups, but, having done so, it is arguably more beneficial to group the receptors based on the homology within the binding site. This grouping arises because small molecules bind to only a small subset of amino acids within the receptor sequence, namely, those that line the binding site. The ligand does not ‘‘see’’ the other residues and is largely unaffected by sequence variation elsewhere unless it confers a conformational change to the binding site. Changes on the exterior of the TM bundles, for example, would be predicted to face the lipid membrane and have little effect on ligand binding, except potentially through modifying allosteric interaction in dimerization or complexation with accessory proteins. Hence, if the 7TM scaffold is common for all of the receptors (i.e., there is no significant change in receptor structure), it is possible to define a small subset of residue positions that form the binding site in each receptor and regenerate the phylogenetic trees on the basis of the amino acids present in those positions. This receptor grouping can assist with the identification of the endogenous ligand for orphan receptors, which become ‘‘misgrouped’’ when the whole sequence is considered, the selection of counterscreens for selectivity profiling, and directly for drug design using a chemogenomic approach. This area is still somewhat in its infancy, although there are already interesting success stories in the literature. This topic is discussed further in Section 2.3. Redefining phylogeny in this way based on a reduced set of amino acids is obviously binding site specific. A number of allosteric binding sites have been characterized for 7TMs. A different classification scheme is called for in each case and is potentially quite important, though, since the selection of counterscreens based on a conventional phylogeny, for example, may not be appropriate for allosteric compounds. 2.2.2

Structural Information and Homology Models

Until comparatively recently, direct experimental determination of GPCR structure via X-ray crystallography has not been available. There are a number of reasons for this deficiency related to the difficulties in obtaining sufficient protein expression

GPCR STRUCTURE AND FUNCTION

23

and purity, retaining correctly folded/functional protein, the inherent flexibility of 7TMs, and finding a system in which they will crystallize in a membranous or pseudomembrane environment. After a considerable amount of effort in the area, progress is beginning to be made, but we still have only a very small number of structures. The first X-ray structure of a 7TM was that of bovine rhodopsin published in 2000 (Palczewski et al., 2000). Initial success here was assisted by the natural abundance of rhodopsin in the eye and its lack of dynamics relative to other 7TMs. Due to its rather specialized role as a photoreceptor, rhodopsin is effectively locked in the inactive state with virtually no constitutive activity toward its G-protein. Some of the difficulties experienced with other receptors may well be due to their inherent flexibility, which serves to destabilize the crystal. The crystal structure of rhodopsin has been hugely useful and made significant impact, notably in the area of receptor homology modeling. If there was a drawback, however, it was that rhodopsin is rather an unusual 7TM and there may have been some concerns about how appropriate it was as a template for the rest of Family A. This concern has been largely addressed through the very recent determination of X-ray structures for the beta-2 adrenergic receptor, which is in really quite good agreement, although there are some differences. The beta-2 adrenergic receptor is a classical aminergic 7TM and has been the focus of drug discovery efforts for many years, notably through the discovery of beta-2 agonists such as salbutamol (VentolinTM) or salmeterol (SereventTM) used as broncodilators for the treatment of asthma. The beta receptor family (b1, b2, b3) is also the target of beta-blockers (antagonists) such as propranolol (InderalTM) for the treatment of hypertension and angina. The receptor is known to express well in recombinant systems, which has added to its focus as a target for crystallization. Two crystal structures of beta-2 have been solved very recently (Cherezov et al., 2007; Rasmussen et al., 2007). There are a number of elements in common between the two approaches but a major difference in how one aspect is addressed. In both cases, cocrystallization with an inverse agonist, carazolol, was used to assist stabilization. This was further enhanced by targeting the highly flexible ICL3. Rasmussen and coworkers used an antibody fragment to this region for cocrystallization; Cherezov and coworkers replaced it with a protein amenable to crystallization, namely, T4-lysozyme. Both approaches serve to increase receptor stability and provide greater polar surface area with which to form good crystal lattice contacts. Crystals were formed in lipidic environments but were too small and radiation sensitive for data to be collected on conventional synchrotron beams. Hence, another key development for the field has been the successful use of microdiffraction techniques (Kobilka and Schertler, 2008). While immensely important, it is as well to be aware of the potential limitations of the structures. As with all crystallographic data, crystal packing forces can serve to alter the protein conformation relative to the native environment. It is particularly important to consider these around the areas of the protein that have been modified to enable crystallization in these cases. The receptors are also effectively locked into a single conformation in the crystal, but there is a wealth of data to suggest

24

G-PROTEIN-COUPLED RECEPTORS

they are mobile proteins. Carazolol is an inverse agonist that suppresses around half but not all the basal activity of the receptor, so the conformation of the beta-2 structures is potentially not completely inactive. This mechanism may explain some of the differences seen between beta-2 and rhodopsin in the packing of the helices. The biggest changes from a drug discovery perspective, however, appear to be in ECL2. For rhodopsin, this area of the protein is a b-sheet structure that effectively forms a lid over the binding site and would limit ligand access, but is presumably not an issue since the ligand in this case is covalently attached. In one of the beta-2 structures, this region forms a helix constrained by two disulfide bonds in such a way that there is probably a channel providing access to the binding pocket (the region is not resolved in the second structure). As a result, the choice of structure used as template for homology modeling can have a profound effect on the residues predicted to form part of the binding site in this region. However, the structures already give us a very important platform from which to study GPCR function, for example, to understand the role of highly conserved features such as the ‘‘ionic lock’’ (DRY motif) at the bottom of TM3, to investigate the role of a water network that appears to link a number of these residues, or to understand how certain mutations can give rise to constitutive activity. From a technical angle, we still need generic protocols that enable structures to be solved for the majority of receptors, particularly those from Family B and Family C since the limited homology to Family A complicates sequence alignment and may infer different helical packing. Ideally, crystallographic systems need to develop in a way that is amenable to soaking-type experiments, for example, to enable rapid determination of structures with a range of ligands bound. We have, after all, only seen a structure with a single drug-type molecule bound thus far. Developments in this area would be particularly insightful if compounds with a range of efficacies could be used so that we can give a structural angle to our understanding of receptor activation mechanism(s) or if we could use ligands with high selectivities over related subtypes, which may be achieved by induced fit. The lack of direct structural information for the TM bundle does not mean to say that there has not been a structural basis for ligand optimization in the GPCR area until recently. On the contrary, there is a wealth of literature related to homology model building and ligand docking that dates back to 1991, if not earlier. During this time, models have effectively evolved through five generations (Barton et al., 2007). First-generation models employed de novo techniques to optimize the packing of the helices. Second-generation models used bacteriorhodopsin as a template for homology model building. This protein is also heptahelical and spans the membrane but did not share any real sequence similarity to GPCRs and indeed was subsequently found to be quite different in the packing arrangement. Third-generation ˚ electron diffraction map of rhodopsin (Schertler et al., 1993) models used the 9 A ˚ elecand this model was further refined to fourth-generation models based on 6 A ˚ tron diffraction maps of frog rhodopsin collected as 4 A slices through the membrane (Unger et al., 1997). Finally, we arrived at fifth-generation models upon publication of the first crystal structure of a GPCR (Palczewski et al., 2000). When reading the literature, it is worth being aware of the differences between

GPCR STRUCTURE AND FUNCTION

25

these models and their likely quality. Despite the fact that use of the rhodopsin or beta-2 crystal structures is now an expected standard, models based on earlier data persist in the literature until rather later than 2000 and they should be treated with a little caution; better models could probably be obtained by rebuilding them based on more up-to-date data. The lack of direct structural determination for 7TMs has meant greater reliance on other techniques to pinpoint interactions between small molecules and the receptor. This dependence was very important when the models were theoretical, based on a non-GPCR or low-resolution data but even for fifth-generation models this remains an issue. Rhodopsin or beta-2 structures should be a good template upon which to base homology models, at least for Family A, although lack of structural data for the target of interest means binding modes can be generated through ligand docking but we need an alternative means of validation. Site-directed mutagenesis (SDM) has been particularly useful for this in the 7TM area. Many experiments are described in the literature and there have been attempts at collating them: see GRAP (Kristiansen et al., 1996) and TinyGRAP (Edvardsen et al., 2002) databases. SDM is typically applied in a cyclical process of docking, mutant design, and screening, followed by data interpretation and refinement of the proposed mode. Such data can considerably improve the models, improve our confidence in them, and aid their application. It is worth noting that it can be important to distinguish changes in ligand binding and receptor function through use of both binding and functional assays and that highly conserved residues, which probably have structural/functional significance, should generally be avoided since effects seen will almost certainly relate to change in the receptor and not ligand binding. The choice of residue change varies depending on the experiment being performed, but typically alanine is used to probe the importance of a side chain in interacting with a ligand. Use of glycine would give more complete removal of the side chain but the residue has rather different conformational properties that might complicate interpretation. This type of experiment is equivalent to alanine scanning often seen with peptide ligands but here performed in the 7TM binding site. Other types of SDM experiments might probe a residue change seen between receptor subtypes or orthologues to identify causes of selectivity. Chimeric constructs can be generated to probe this effect in larger sections of the receptor (Gearing et al., 2003). Receptor chimeras can be very useful to localize the region of importance to be followed up by SDM to identify the specific changes of interest. The fact that whole receptor structures and, in particular, those containing the membrane-spanning portion are difficult to obtain has not prevented some other useful crystallographic information from being determined. As was mentioned in Section 2.2, the orthosteric site extends into the NTD for Family B and is entirely within the NTD for Family C. These sections of the receptor are extracellular, form a discrete fold, and have proven amenable to crystallization and structure determination. At the present time, we have a crystal structure of the N-terminal domain of the GIP receptor (Parthier et al., 2007) and a very recent structure of the PTH1 NTD

26

G-PROTEIN-COUPLED RECEPTORS

(Pioszak and Xu, 2008) as representative of Family B. These have the endogenous ligand cocrystallized and give an idea of the contact interface between the two. There are earlier NMR structures of the NTD of CRF2b that show a similar fold. The structure is stabilized by three disulfide bonds, a hydrophobic core, and a salt bridge. The folded structure comprises b-sheets in a short consensus repeat called a Sushi domain. The key residues involved in maintaining this structure are highly conserved throughout the peptide hormone Family B receptors, suggesting the structure is common to all. Hence, the GIP or PTH crystal structures should be a good template for homology model building for other members of the family. NTDs of Family C were originally predicted by homology and subsequently shown by crystallography to be related to the bacterial periplasmic binding proteins. There are currently crystal structures for the NTD of mGluR1, mGluR3, and mGluR7 (Kunishima et al., 2000; Muto et al., 2007) with a range of orthosteric ligands bound. This includes both agonists and antagonists, so in this case we have a good picture of the conformational changes that occur upon receptor activation. For Family C 7TMs, the bilobal NTD closes around the endogenous agonist in what is called the ‘‘venus flytrap’’ mechanism. What remains unclear is how this conformational change in the NTD confers a conformational change in the TM bundle and effects signaling on the intracellular face of the receptor. Building full receptor models for Family B and Family C requires a model of the TM bundle and the NTD. The latter is fairly straightforward given the generally quite high homology in the NTD regions and defined structural features such as disulfide bonds that serve to fix the sequence alignment. Rhodopsin or beta-2 could form the basis of the models for the helical regions, but this process is hampered by the very low sequence identity between the families. With Family B, for example, it is often said that there is no real homology apart from the conserved disulfide bond formed between the top of TM3 and the ECL2. However, it appears that there may be a little more in common: the EGxY motif at the bottom of TM3 could be a surrogate for the DRY motif, with the basic side chain reaching across from TM2. This is in close proximity in receptor models and has been shown to be essential for activation. TM4, TM6, and TM7 have conserved W, P, and N, respectively, in what could be equivalent positions to Family A. These observations provide a starting point for homology model building for Family B based on Family A crystal structures, and a few examples have appeared in the literature. In a similar way, there are a few published models for the TM bundle of Family C but the sequence alignment to Family A is again very difficult, most likely subject to error, and the use of other data such as SDM results to validate the models is very important. Another challenge for this area, however, is joining the NTD homology model onto the TM bundle model to build a full receptor model. This total picture is difficult since we have a fairly ill-defined region of structure between the two. It is particularly important for Family B if we wish to study the binding mode of the endogenous ligand since this extends across both the NTD and the TM bundle (traditionally referred to as the juxtamembrane or J-domain). There has been an extensive investigation of the peptide interaction with the receptor, which suggests that the carboxy terminus binds to the NTD as an a-helix and the amino terminus binds

GPCR STRUCTURE AND FUNCTION

27

to and activates the TM bundle. The NTD interaction is of moderate to high affinity but does not appear to be involved in receptor activation. C-terminal fragments of the endogenous ligands can act as reasonably high-affinity antagonists of the receptor. The interaction between the amino terminus of the peptide and the TM bundle has much lower affinity. This yields the two-domain model of peptide binding at Family B 7TMs: binding to the NTD provides an affinity trap, increasing the local concentration of the amino terminus of the peptide in the vicinity of the TM bundle, overcoming the low affinity of this interaction to enable sufficient binding to occur. In addition to the crystal structures of rhodopsin and beta-2 receptors and NTDs for Family B and Family C receptors, there is a range of other structural data that may prove useful for drug design. There are a number of crystal structures for the G-proteins themselves, for instance for the Galpha-beta-gamma trimer (Wall et al., 1995; Lambright et al., 1996). Drug design does not generally target the intracellular side of the receptor, so these have not been used, although recent observations of an intracellular allosteric site in chemokines may change this. Drug design may also be driven from knowledge of endogenous ligand structure and is well precedented for drug discovery around small-molecule ligands (Black, 2004). Some of the 7TM ligands are fairly large peptides and potentially have a defined fold. Particular examples are the chemokine ligands for which a number of NMR structures have been determined. However, difficulties competing with large endogenous ligands may favor an allosteric approach in this case. NMR has also been used to generate structures of ECL or ICL fragments for modeling purposes but this requires the rather bold assumption that the conformation in solution is the same as when part of the receptor. In a similar way, NMR structures of smaller peptide ligands have been determined in solution. These often seem to show a helical arrangement, particularly for the C-terminal section. This situation may be real—formation of an amphipathic helix that sits on the membrane surface would offer advantage in terms of increasing local concentration near the entrance to the receptor binding site—but such results should be treated with a little caution as the conditions employed for the structure determination can favor helix formation. Solid-state NMR has been applied to determine ligand-bound conformations and environments. It can provide accurate information but suffers from the significant drawback of requiring extensive isotopic labeling of the protein and/or its ligand. A number of other techniques such as the substituted cysteine accessibility method (SCAM), site-directed spin labeling, fluorescent labeling, and photoaffinity labeling have been used to generate structural information for 7TMs but are not discussed here. 2.2.3

Mechanisms of Receptor Modulation

There is an increasing body of evidence to suggest that GPCRs exist and might function as dimers or possibly even higher order oligomers. For some years now, experiments performed in recombinant systems have apparently shown the formation of homodimers or heterodimers. The question was whether the effects seen in

28

G-PROTEIN-COUPLED RECEPTORS

such artificial systems, usually with highly overexpressed receptors, were relevant in physiological systems. The possibility is quite attractive. If it is true, molecules that can regulate a GPCR within a heterodimer through allosteric effects between one or more of the partners and the G-protein offer the potential to function in a highly selective and tissue-specific manner. An issue for drug discovery aiming to exploit this effect, however, is that such molecules cannot be discovered through conventional screens using one receptor at a time. Rimonabant (AcompliaTM, SR141716A) is a CB1 orthosteric antagonist or inverse agonist used to treat obesity. This activity could explain the effect on feeding directly but an overlapping distribution of the Orexin-1 receptor has been observed in the brain. This receptor is also thought to have an effect on feeding. Coexpression of both CB1 and Ox-1 in CHO cells yields a 100-fold greater potency of the peptide agonist Orexin-A to signal via MAP kinases ERK1 and ERK2. Furthermore, treatment with rimonabant causes reversal to basal levels. The effect was pathway dependent since stimulation of inositol phosphate (IP) production was unaffected. These results suggest activity via both receptor heterodimerization and agonist trafficking. There is also an example from the chemokine family. While inactive when screened alone, coexpression of CCR2b and CCR5 receptors yields a system in which CCR5-selective antagonists can inhibit the binding of MCP1 (which is selective for CCR2b) and CCR2-selective ligands can inhibit the binding of MIP1b (CCR5 selective) presumably via a CCR2/CCR5 heterodimer. It remains a challenge to validate the presence/importance of heterodimerization in native cells/tissues. The most compelling evidence to date probably comes from a very recent publication around 5HT2a (Family A) and mGluR2 (Family C) receptors (Gonzalez-Maeso et al., 2008). The authors have demonstrated the existence of heterodimers or potentially higher order oligomers via several methods in both native tissue and recombinant systems. They have further shown that heterodimer formation has functional significance in native tissue through investigation of 5HT2a agonist affinity in the presence of an mGluR2/3 agonist and vice versa. They find an allosteric interaction that is eliminated upon antagonist binding to either receptor. The authors have gone on to localize the site of the interface between the receptors through construction of mGluR2/3 chimeras. While mGluR3 has high sequence homology/identity with mGluR2, it does not appear to interact with the 5HT2a receptor. Placing the TM4/5 region of mGluR2 into mGluR3 produces a receptor that functions as a heterodimer, suggesting that this section contains the contact interface. Aside from interaction with other GPCRs, the pharmacology of GPCRs can be further modulated by accessory proteins to yield even greater diversity in the receptor repertoire. The best characterized of these are probably the receptor activity modifying proteins (RAMPs). RAMP interaction with receptors can lead to a variety of actions, including chaperoning the receptor to the cell surface and the generation of novel phenotypes. RAMP heterodimerization with CLR and the related CT receptor is required for the formation of CT gene-related peptide (CGRP), adrenomedullin (AM), or amylin receptors. The RAMP may modulate other functions

GPCR STRUCTURE AND FUNCTION

29

such as receptor internalization, recycling, and the strength of activation of downstream signaling pathways. The RAMPs are a family of three proteins, approximately 150 amino acids in length. Their structure constitutes an N-terminal signal peptide, an extracellular N-terminus of about 90–100 residues before a single membrane-spanning region of around 22 residues, and ending with a small C-term of around 10 residues. RAMP2 has four cysteines and RAMPs 1 and 3 have six cysteines that are thought to form disulfide bonds. However, they lack homology with other proteins for which we currently have structural information, so modeling them is difficult. A screen of eight other Family B receptors has suggested that VPAC1 is capable of interacting with all three RAMPs, PTH1 and Glucagon interact with RAMP2, and PTH2 interacts with RAMP3, whereas VPAC2, GHRH, GLP1, and GLP2 do not appear to partner any of them. It is unclear, however, whether the interaction leads to a change in pharmacology, with the possible exception of VPAC1, where agonist binding does not appear to be altered but RAMP2 overexpression augments phosphatidylinositol (PI) hydrolysis relative to cAMP production. The potential role for RAMPs may extend beyond Family B. There is additional evidence, for example, that RAMP1 or RAMP3 (but not RAMP2) is required for cell surface expression of the Family C calcium-sensing receptor. There are a number of other examples of accessory proteins interacting with GPCRs. For CLR-based receptors, efficient signaling (via Gs) also requires expression of receptor component protein (RCP); this interacts with ICL2 and may influence the stability of CRL þ RAMP complexes. Odorant receptors cannot be expressed unless RTP1, RTP2, or REEP1 is coexpressed. The melanocortin-2 (MC2) receptor requires MC2 receptor accessory protein (MRAP) to move out of the endoplasmic reticulum and undergo posttranslational processing; this interaction is specific to MC2 over the other members of the melanocortin family, the b2-adrenergic receptor and TRH. The Frizzled response to Wnts can depend on low-density lipoprotein receptor-related protein 5 (LPR5). As with receptor dimerization, it is something of a work in progress to understand the prevalence and full relevance of accessory proteins in modulating 7TM pharmacology. The already extensive range of synthetic agonists, antagonists, and inverse agonists discovered over the years has expanded further through the identification of ligands that bind to a number of allosteric sites on 7TMs. Utilizing such sites can confer a number of advantages. For example, it may be possible to obtain improved selectivity or the site might be more tractable. It is possible to identify compounds that sensitize a receptor to its endogenous ligand but may have no activity in their own right. These compounds are described as ‘‘positive allosteric modulators’’ (PAMs); the term ‘‘ago-allosteric modulators’’ seems to be favored for compounds that both sensitize the receptor and have a lesser degree of agonism. PAMs can potentiate agonist potency, efficacy, or both. PAMs can offer exquisite temporal and spatial control because the receptor is only activated at the same time and in the same tissues as it is by the endogenous ligands. Thus, the compound effectively borrows the natural and potentially complex control afforded to the

30

G-PROTEIN-COUPLED RECEPTORS

endogenous ligand; a synthetic agonist would lead to more prolonged and widespread receptor activation that may not be desirable. There are some potential disadvantages with an allosteric mode of action. There is no evolutionary pressure for the site to be conserved, so there is scope for more species differences to be observed that may hamper downstream target validation and compound development. The allosteric effects seen are defined for a particular combination of receptor, any regulatory proteins, and G-protein partner involved in the signal readout, something called ‘‘probe dependence.’’ In principle, the activity can vary if any of these elements change, for example, through use of a receptor orthologue, different cell line with a different complement of potential protein partners, or different G-protein (or other mechanism) for signaling. It remains to be seen how widespread this phenomenon is and how greatly it will affect drug discovery. The exquisite control afforded by such factors could very well be considered an advantage, provided we can manage them. All three major families of 7TMs (A/rhodopsin, B/secretin, C/glutamate) have proven amenable to this approach as antagonists, agonists, and positive allosteric modulators, although the area is best developed for Family A and Family C, with only a relatively small number currently described for Family B. Table 2.3 shows a representative for each category. While the nature of allosteric compounds is quite well described in terms of their pharmacology, the detail of the binding sites somewhat lags behind, undoubtedly due to the relative lack of structural data in the area. There have been a number of attempts to characterize the sites using receptor chimeras and point mutation. For Family C, the compounds are thought to bind in the TM bundle in a manner that is reminiscent of Family A (e.g., see case study for cinacalcet); the compounds are allosteric by virtue of the fact that the orthosteric site is entirely within the NTD in this case. For Family A, the situation is a little more complex since, if we assume no role for the NTD in ligand binding, both endogenous ligand and allosteric modulator need to occupy discrete regions of the receptor TM bundle. Modeling studies are starting to show that this appears possible, either through greater use of the extracellular loops for binding one or other of the ligands, or by virtue of the fact that the ligand binding pocket within the TM regions is actually large enough to accommodate two ligands of drug-like size. Very recently, workers at AstraZeneca have described an intracellular binding site for a number of chemokine receptors, including CXCR2 and CCR4. This site appears to be formed by the section of receptor running from the end of TM7 along to HLX8. It is unclear at this stage whether the site is general to other 7TMs but it certainly has the scope to add another layer of complexity for 7TM drug design in that compounds would need to be cell penetrant to access this site and the region is thought to undergo a fairly substantial conformational change upon receptor activation. As has been alluded to above, 7TMs are quite flexible proteins and appear to undergo a fair degree of conformational change upon agonist binding at the extracellular side such that a signal is conveyed at the intracellular side. Even in the absence of agonist binding, however, most receptors have a level of basal activity that can be quite high in some cases (e.g., ghrelin >50%). This ‘‘constitutive

31

GPCR STRUCTURE AND FUNCTION

TABLE 2.3 Family

Representative Examples of Allosteric Modulators of 7TMs

Mode

Receptor

Structure F

A

Status

F

Antagonist CCR5 Me

Me

NH

O

N

N

N N

Me

Marketed as SelzentryTM/ CelsentriTM for treatment of HIV infection

maraviroc nBu O N

A

Agonist

M1

Preclinical Me AC-42

S NH 2

A

Pos mod

Adenosine A1

O

Cl

Tested on limited numbers of healthy volunteers for dose finding

T62

N(nPr) 2 N N

B

Antagonist

CRF1

Me

Entered Phase IIa studies; discontinued but demonstrated efficacy in anxiety

Me

N N

Me

NMe 2 R121919 O

O

B

Agoallosteric GLP1 modulator

Cl

N

S

Cl

N

N

Preclinical

2-Sulfonyl-quinoxalines

(continued)

32

G-PROTEIN-COUPLED RECEPTORS

TABLE 2.3 Family

C

(Continued )

Mode

Receptor

Structure

Me

N

Antagonist mGluR5

Status

Preclinical

MPEP Ph Ph

C

Agonist

mGluR7

N H

H N

Ph Ph

Preclinical

AMN082

C

Pos Mod

CaSR

F3 C

H N Me

cinacalcet

Marketed as Sensipar1/ Mimpara1 for secondary hyperparathyroidism and hypercalcemia in patients with parathyroid carcinoma.

activity’’ implies that the energy barrier is small and that the receptor is capable of moving between inactive and active states in isolation. Hence, when ligands bind, they effectively alter the equilibrium such that the receptor favors either an activated state (agonists) or an inactive state (antagonists). Antagonists can restore natural levels of constitutive activity (neutral or silent antagonists) or can go on to eliminate some or all of the constitutive activity (inverse agonism). Synthetic agonists can have the same level of efficacy as the endogenous ligand (‘‘full agonist’’) or yield a lesser degree of signaling (‘‘partial agonist’’). Despite extensive work in the area, the exact mechanism of receptor activation is not well understood due to the limited amount of structural data currently available. It is generally assumed that receptor activation occurs via a common mechanism despite the diverse range of endogenous ligands and binding sites. This position may require further consideration, though, in light of an increasing body of evidence to suggest that receptors may be activated in a G-protein or other pathway-specific manner (‘‘trafficking’’). There is also potential for the area to be further complicated by the role of heterodimerization in receptor activation, particularly if activation of one receptor in the pair can yield signaling in the second (‘‘transactivation’’).

GPCR STRUCTURE AND FUNCTION

33

The current state of the art in the area is probably the ‘‘toggle switch model,’’ which has been developed based on accumulated biophysical data such as sitedirected spin labeling, various fluorescent technologies, and use of modified ligands anchored at specific sites in the main ligand binding site (Schwartz et al., 2006). In this model, TM6 and TM7 move upon receptor activation, with the extracellular side coming closer together and the intracellular side moving further apart. The helices are thought to pivot about their highly conserved prolines in a vertical seesaw-type motion, possibly associated with rotation and partial straightening of the bends caused by the prolines in the helices. Data suggest that TM6 moves away ˚ , with TM7 moving by a lesser amount. TM5 does not from TM3 by around 8 A appear to move on the intracellular side, despite the presence of a highly conserved Pro. It is thought that the movements serve to expose key motifs such as the DRY segment at the bottom of TM3 and elements of TM7-HLX8, allowing recognition by signaling molecules. The switch is provided by a highly conserved Cys-Trp sequence prior to the Pro on TM6 (the CWLP motif). In the rhodopin crystal structure, the Trp is seen to sit vertically between TMs 3 and 6 in a way that would prevent the proposed inward movement of the helices. A highly conserved Phe on TM5 is thought to serve as a ‘‘lock’’ for the Trp. When this is released, the Trp can move to a different rotameric state and alleviate the steric constraint so the helices may move closer together. The structural models described above are likely to be a simplification of potentially very complex pharmacology: for more detailed discussions see Kenakin (2004). Activated GPCRs need to be deactivated to prevent perpetual signaling, or to adapt to a constant stimulus. Receptor deactivation involves specific phosphorylation by a GPCR kinase (GRK), which allows arrestin binding to the receptor, blocking further G-protein coupling. This process is referred to as desensitization. Arrestins can also initiate clathrin-mediated endocytosis, leading to internalization of the receptor. This internalization may be followed by translocation to degradation compartments (lysosomes) or recycling back to the plasma membrane. 2.2.4

G-Protein Coupling and Assay Formats

The G-proteins activated by GPCRs are trimeric, consisting of a-, b-, and g-subunits. At last count, there are 21 Ga subunits encoded by 16 genes, 6 Gb subunits encoded by 5 genes, and 12 Gg subunits, creating a large number of theoretical combinations, although only a small number of these form active complexes. The heterotrimeric G-proteins are typically classified into four families based on the primary sequence similarity of the Ga subunit: Gas, Gai, Gaq/11, and Ga12/13. Gs proteins activate adenylyl cyclases, increasing intracellular cAMP levels and thereby stimulating activity of the cAMP-dependent protein kinase (PK) A. Gi proteins inhibit the activity of adenylyl cyclases and activate extracellular signal-regulated kinases 1 and 2 (ERK1/2). Gq proteins activate phospholipase (PL) Cb, triggering the inositol triphosphate cascade and leading to activation of PKC and intracellular

34

G-PROTEIN-COUPLED RECEPTORS

Ca2þ mobilization. The G12 family consists of the G12 and G13 proteins, which activate the monomeric GTPase RhoA. Some GPCRs can couple to only one type of G-protein (e.g., Gs or Gi), but many GPCRs couple to a broader range of G-protein families, such as Gi/G12, Gq/G12, or Gi/Gq/G12. The G-protein coupling profile of GPCRs is not always fully characterized, and coupling to G12/13 proteins has not been elucidated for many GPCRs (Deupi and Kobilka, 2007; Siehler, 2007). Assays for GPCRs have evolved from measuring binding affinity with radioligands to determining the functional efficacy of a ligand by measurement of downstream signaling responses. The G-protein signal transduction and associated assay formats are summarized in Fig. 2.4. The [35S]GTPgS assay works by measuring the reduction in binding of this radiolabeled stable analogue of GTP to the G-protein a-subunit: if the receptor is activated, GTP binding will be high, whereas if the receptor is inactivated (e.g., by binding to an antagonist), GTP binding will be reduced. In principle, since the assay captures the receptor–G-protein interaction independently of the downstream signaling pathway, it is adaptable to measure coupling to all G-protein subtypes; in practice, useable signal-to-noise ratios are only achieved with Gi-coupled receptors.

FIGURE 2.4 G-protein signal transduction pathway steps (orange ellipses) and typical associated assay formats (magenta text). (See the color version of this figure in the Color Plates section.)

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN

35

Other assay technologies have arisen from interception with signalling cascades specific to the G-protein a-subunits. For Gs and Gi/Go, receptor activation or inactivation can be determined by direct measurement of intracellular cAMP concentrations, typically using LANCETM or AlphascreenTM technology, or by reporter gene assays, which provide a readout of downstream transcriptional up- or downregulation. For Gq, receptor function has typically been determined by measuring intracellular Ca2þ concentrations, in the last decade using FLIPR technology, which is now being replaced by luminescence-based assays such as aquorin. Recent advances in understanding the role of arrestins in the modulation of receptor function mean that luminescence assays measuring the proximity of GPCRs and b-arrestin are now practicable. Since b-arrestin binds to all receptors following activation, this provides an attractive universal assay format, whose application in industrial screening is receiving significant interest. Each of these assay formats has advantages and disadvantages. This includes differing tendencies to present compounds that interfere with the assay pathway as ‘‘false positives.’’ Radioligand binding assays tend to produce a low false positive rate but are unlikely to distinguish between agonists and antagonists and detect allosteric modulators. By contrast, functional assays with readouts at the bottom of a complex biochemical cascade not only tend to deliver significantly higher false positive rates, but are also capable of identifying allosteric modulators. In cellular functional assays, this can be exacerbated by the presence of endogenous receptors that may signal through the same pathway. A further feature of functional assays in highly expressed recombinant cell lines is a tendency to exaggerate agonist efficacy. Given this observation and the ability of some receptors to couple to multiple G-proteins, recombinant functional assays should be chosen to reflect the native coupling system, and early validation of hit series in disease-relevant native tissue systems is prudent to ensure that the efficacy observed is predictive of in vivo effects.

2.3

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN

The challenges facing drug design are numerous, and many apply generally across target classes. As will have become apparent from the preceding sections, the field of GPCRs throws up many additional challenges, including (1) variable tractability; (2) complex pharmacology related to dimerization, accessory proteins, and signaling pathways; (3) lack of correlation between screening assays, native tissue efficacy, and in vivo responses; (4) complex phylogenetic relationships and role of allostery; and (5) lack of structural information, particularly relevant in relation to mechanism for activation, directing drug design, for example, high selectivity via induced fit, structural characterization of Family B/C TM bundles. A detailed illustration of these aspects is beyond the scope of this chapter; instead, in the following sections we have chosen to illustrate challenges where medicinal chemists can and do make a direct impact to successfully identify leads, candidates, and marketed drugs.

36

2.3.1

G-PROTEIN-COUPLED RECEPTORS

Hit Generation Strategies: Chemogenomics and Privileged Structures

Many early drugs targeting GPCRs were derived from modification of endogenous monoamine neurotransmitters with simple, low molecular weight structures. This approach has been successfully extended to receptors targeted by other low molecular weight ligands, such as prostanoids, sphingosine phosphate, and nucleosides. However, for peptidergic receptors, capturing pharmacophore elements of large peptides into small drug-like molecules remains a major challenge. Because of the high diversity of ligands and their binding sites across the GPCR superfamily, application of multiple approaches to hit finding is often necessary. Diversity screening adds most value where there is little knowledge surrounding the target or the ligand, particularly for orphan receptors. It is successful in identifying progressable hits for
50% of GPCR targets, including the CCR5 and CGRP receptor antagonist case histories described below. A key drawback is the time and the cost for the execution of high-throughput screens due to the requirements of large-scale reagent generation and assay development. In addition, the diversity in many corporate screening collections is often highly biased by compounds from legacy lead optimization programs and may be limited or contain unsuitable structures for the target of interest. Receptor subfamily-focused chemistry and screening may generate hits more rapidly where some target knowledge is available. In this context, it is helpful to group GPCRs according to ligand and hence binding site similarity: screening compounds with structural similarity to ligands at a closely related receptor generally increases the likelihood of generating hits, although of course there are many exceptions. A more proactive approach involves the design of chemical libraries that adapt and expand the pharmacophore of known ligands for a particular target subfamily: in this way, existing knowledge is exploited while adding the benefit of chemical novelty. Subfamilies that have been exploited in this way include the aminergic family, prostanoids, and some peptidergic receptors. Molecular substructures that occur in ligands active at more than one receptor are often referred to as ‘‘privileged structures’’ and have provided a rich, if conservative, source of templates around which to base the synthesis of prospective hit generation libraries. It should be noted that of course fragments that appear ‘‘privileged’’ for GPCRs often show activity at molecular targets in other target classes, particularly ion channels. The endogenous ligands serotonin, dopamine, histamine, and noradrenaline all have in common an arylethylamino group. To obtain selectivity across the aminergic family, over the years numerous conformationally constrained analogues of the arylethylamino substructure have been synthesized and tested. The benzazepines form one example of this strategy, and examples of this template are outlined in Fig. 2.5. The dopamine D1 receptor agonist (þ)-SKF 38393, discovered some 30 years ago, derives its selectivity from the geometry of the amino group, together with the position of the phenyl group that blocks binding to other receptor subtypes. 6-Chloro substitution together with N-methylation reduced dopamine receptor affinity and introduced selective adrenergic a2 receptor agonism to give SKF86466. Further modifications to the phenyl substitution gave the selective 5HT2A receptor

37

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN NH2

HO

N

NH HO

O O S

HO

NH

HO Cl (R)-SKF38393 D1 agonist 1978

Dopamine

HO SKF103829 5HT2A partial agonist 1996

SK&F 86466 a2 agonist 1986

O

O O S N

N N H

NC

5HT6 antagonist 2002

D3 antagonist 2000

Me O S

Cl

FIGURE 2.5

NH

N

Me N O N H

Mixed D2, D3, 5HT2A, 5HT6 antagonist

O NMe N H

N

N O

GSK189254 H3 antagonist Phase II Narcolepsy

The benzazepine template as a privileged structure for aminergic GPCRs.

partial agonist SKF103829. A study of more elaborate substitution of the amino group led to selective dopamine D3 receptor antagonists. In complementary work, larger substituents on the phenyl ring gave a series of selective 5HT6 receptor antagonists. The combined results of these studies, together with advances in synthetic methodology, identified the benzazepine as a versatile ‘‘privileged structure’’ for GPCR receptors with ideal properties for the design of aminergic hit-finding libraries. Clinical candidates that contain the benzazepine fragment targeting aminergic receptors include the mixed dopamine/serotonin antagonist, described in more detail below, and the selective histamine H3 receptor antagonist GSK189254, which is in Phase II trials for narcolepsy. The concept of privileged structures for peptide receptors is well demonstrated by the spiropiperidine-indoline core, which, depending on its substitution, is active across a number of peptide receptors including ghrelin, oxytocin, somatostatin, tachykinin, and melanocortin (Fig. 2.6). The structure contains a constrained phenylpiperidine unit, and a broader analysis across peptide receptor ligands shows the potential of this motif to present common pharmacophore elements in similar but systematically scanned three-dimensional space. Thus, the piperidinyl benzimidazolone present in NPY-Y5 antagonist contains an extra bond between the piperidine and phenyl rings, and the motif in CGRP antagonist contains a further two atoms. In addition to dictating the distance between the piperidine and aryl rings, the linking group provides opportunity for variation in conformation and hydrogen bonding

38

G-PROTEIN-COUPLED RECEPTORS

H2N

HN

O

N

HN

O

NH

O

NH O

O

O

N O

NH

O

O

N

N

N

N

Cl N S O O

MK-677 Ghrelin agonist

FIGURE 2.6

S O O

MC4 agonist

N

N

N

N

O N H

NPY-Y5 antagonist

O

N H

CGRP antagonist

Aryl piperidine privileged structures for peptidergic GPCRs.

donor and acceptor motifs. Three-dimensional molecular overlays indicate the similarities and differences in the pharmacophores scanned by these relatively rigid templates, which can aid an understanding of selectivity for a particular receptor, as well as suggesting opportunities for novel variations as shown in Fig. 2.7 (Bondensgaard et al., 2004). The preceding discussions emphasize the potential value in exploiting similarities in ligand structures for closely related receptors. However, some molecular motifs are found in ligands for unrelated receptors that appear to have serendipitously developed similar binding sites. This is exemplified by the 2,6-dimethylpiperazinyl indoline moiety, originally prepared as a substructure in a series of 5HT1B

FIGURE 2.7 Three-dimensional overlay of aryl piperidine privileged structures, showing incremental positioning of aryl and hydrogen bonding groups. (See the color version of this figure in the Color Plates section.)

39

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN EtO Cl

O

N S

N

N

N

O S O

NH N

N

O

O 5HT1B pIC50 6.1 Ghrelin pEC50 9.8

5HT1B pIC50 7.9 Ghrelin pEC50 6.8

FIGURE 2.8 Piperazinylindoline core as a cross-GPCR privileged structure.

receptor antagonists (Fig. 2.8). A high-throughput screen revealed additional moderate affinity and efficacy at the human ghrelin receptor: structural modifications including a change in the linking group geometry (amide to sulfonamide switch) allowed a dramatic switch in selectivity in favor of the ghrelin receptor. Modeling of receptor structures complements ligand-based chemogenomics, allowing virtual screening and providing a filter to influence ligand/library design. Information about the binding of a ligand at one receptor can be used to select or design ligands targeting a second receptor that has some sequence similarity in the binding site. The receptors do not need to be related in terms of their endogenous ligand or their whole sequence similarity. An elegant example of this has been described recently by researchers at Roche (Martin et al., 2007). Analysis of the homology of amino acids defining the putative consensus drug binding site of SST5R identified opioid, histamine, dopamine, and serotonin receptors as the closest neighbors. By screening a set of known ligands for these receptors, the histamine H1 receptor antagonist astemizole was found to have low mM antagonist activity at SSTR5. This was used as the seed structure for subsequent optimization, leading to compounds with low nM potency for SSTR5 and high selectivity over H1 receptors (Fig. 2.9). The somatostatin receptor class is traditionally viewed as quite intractable, with no small molecules having previously been described for SSTR5, and hence the achievement is all the more noteworthy.

F N

O H2N

S O O

H N

H N

N N

N N

OEt Astemizole H1 pIC50 ~ 7.4-8.4 hSST5R pKi = 5.4

FIGURE 2.9

EtO OMe

NH2

H1 pKi = 5.6 hSST5R pKi = 7.9

Lead hopping from astemizole to selective SSTR5 antagonists.

40

G-PROTEIN-COUPLED RECEPTORS

Receptor modeling has been applied quite extensively for structure-based drug design, with some notable successes, but is hampered by the relative lack of structural information, is labor intensive, and is best performed in conjunction with additional studies such as SDM to help validate the proposed interactions. While there have been attempts to apply high-throughput docking to 7TMs, it is hard enough getting good enrichment, let alone correlation with biological activity, for systems with a wealth of structural information; it is all the more difficult given the added uncertainties associated with using homology models. Furthermore, the flexibility of 7TMs will prove problematic in this context. The approach appears most successful where models have been optimized around a known ligand, which is then removed. This process potentially improves success rate but limits the scope— clearly, such an approach cannot help discover compounds for orphan receptors and more generally focuses the search around just one conformation of the receptor. We also tend to see greater success in virtual screens for antagonists than agonists, probably due to the conformational state of the rhodopsin crystal structure on which the models are based. Typically, 5–10-fold enrichment is achieved, occasionally being claimed as high as 40-fold, but it is important to distinguish the ability of a method to pull out actives from structurally similar inactives rather than from a pool of random and/or non-drug-like molecules. Perhaps for related reasons, there appears to be only one example in the literature of the successful application of de novo drug design to 7TMs (Ali et al., 2005) and even then the compounds were of fairly modest potency (low mM). However, ligand-based approaches such as pharmacophore modeling have been used extensively. 2.3.2 Case History: Calcitonin Gene-Related Peptide Receptor Antagonist (MK-0974, telcagepant) The difficulties in identifying truly small-molecule drugs for receptors that have evolved to recognize large peptide ligands should not be underestimated. Put simply, to achieve sufficient potency at the target, lead molecules tend to acquire molecular weight, permeability, and solubility properties that preclude good absorption and metabolic stability. However, significant progress has been made in overcoming these hurdles as evidenced by the marketed oral drugs targeting the angiotensin II receptor, the NK1 receptor, the CCR5 receptor (see introduction), and oral drug candidates active at targets including the ghrelin, motilin, orexin, and oxytocin receptors. A holistic analysis of these successes can give useful insights into the molecular property space that allows peptide receptor potency and ADME to be balanced, as well as providing an expanding feedstock for potential privileged structures. The case history below for the discovery of the oral CGRP receptor antagonist MK-0974 shows a successful approach to balancing potency, oral absorption, metabolic stability, and free plasma concentration—although the drug candidate selected shows nonideal physicochemical properties with a molecular weight of 566 and poor solubility, sufficient exposure was obtained in Phase II studies to demonstrate proof of concept and proceed to Phase III.

41

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN

MK-0974, an antagonist of the CGRP receptor, was discovered by scientists at Merck in the United States and is in Phase III trials for migraine (Paone et al., 2007). The CGRP receptor is a family B GPCR formed from two units: a classical 7TM domain known as calcitonin receptor-like receptor (CLR), combined with RAMP-1. CLR can also combine with other RAMPs, forming the adrenomedullin receptors, which are the closest homologues to CGRP. CGRP is a 37-amino acid neuropeptide that is implicated in the pathogenesis of migraine: its levels are elevated in migraineurs but are normalized on treatment with triptans, while intravenous infusion of CGRP can trigger migraine. CGRP antagonists are believed to block the neuroinflammatory effects of CGRP on the trigeminal nervous system, which forms the primary nociception pathway for the face and the head. The first selective CGRP antagonist to enter the clinic was BIBN4096 (Fig. 2.10), discovered by scientists at Boehringer-Ingelheim in Germany, which established clinical proof of concept in Phase II studies after intravenous administration. This compound was derived from optimization of the capped dipeptide HTS hit 1, with weak micromolar affinity for the receptor, by modification of the capping groups, including the piperidinyl-quinazolinone peptidergic privileged fragment, to achieve higher potency. HO

Br

HO

Br

Br

Br

O

O

H N

N H

O

N H

O

N N

1

O

H N

N H

O

N N

2

N

HTS hit IC50 17,000 nM

O

BIBN IC50 0.03 nM

NH2

N H

NH2

F3C Me N

O N

O

N

O

O N H

O

H N

N H

3

N

O

N H

N

N H

N

4 Ki 55 nM cAMP IC50 65 nM

HTS hit Ki 4800 nM cAMP IC50 5000 nM

O

N H

F3C O

N

N H F

F

N

O

O

N

N H

6 Ki 0.77 nM cAMP IC50 2 nM Protein shift 5-fold

N O

N N H

O N

5 Ki 2 nM cAMP IC50 4 nM Protein shift 28-fold

N O

N H

FIGURE 2.10 Derivation of CGRP antagonists BIBN4096 and MK-0974.

42

G-PROTEIN-COUPLED RECEPTORS

FIGURE 2.11 Three-dimensional overlay of energy minimized piperidinyl-spirohydantoin and -quinazolinone substructures, showing similar positioning of key hydrogen bonding groups. (See the color version of this figure in the Color Plates section.)

Following the successful proof of concept study, scientists at Merck set upon the challenging goal of identifying an orally bioavailable drug candidate. A highthroughput screen again identified the starting point 3, which contains a phenylbenzodiazepine b-turn mimetic: another peptidergic privileged fragment. The initial optimization focused on improving the modest micromolar potency. It was observed that the spirocyclic right-hand side grouping can be overlaid with a variety of groups including the piperidinyl-quinazolinone found in BIBN 4096 such that the hydrogen  O and NH groups are similarly positioned (see Fig. 2.11). Incorporation bonding C  of this group and optimization of the benzodiazepine substituent from methyl to trifluoroethyl gave the lead 4 with improved potency in terms of both binding and functional antagonism in a cAMP assay. At this stage, a marked difference in the affinity for CGRP orthologues was observed: compound 4 shows almost no measurable affinity for rat receptors, indicating that alternative species would have to be used for in vivo pharmacodynamic assays. Lead 4 showed moderate bioavailability of
10% in rats, and attention turned to physicochemical properties. Removal of both phenyl rings of the benzodiazepine left-hand side to give caprolactams led to a drop in potency, but micromolar activity was retained (Fig. 2.12). Exploration of the optimum location and configuration of a single phenyl ring, investigation of the optimum stereochemistry, together with replacement of the right-hand side with a phenylimidazolone grouping led to the identification of lead 5, combining nanomolar potency with oral bioavailability of 27%. The

O

N

N

O

N H

N

N

Ki 55 nM cAMP IC50 65 nM

O

N

N H

N

7

O

N H

8 Ki 3.7 µM cAMP IC50 3.1 µM

O N H

9 Ki 7.8 µM cAMP IC50 2.4 µM

FIGURE 2.12 Effect on CGRP antagonism of successive removal of phenyl rings from benzodiazepine unit.

43

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN

optimization endgame was driven by the observation that leads such as 5 show marked reductions in receptor affinity in the presence of human serum (a so-called protein shift), indicating that high systemic concentrations and hence high doses would be required for in vivo efficacy. Protein shift data on a range of variants of the right-hand side highlighted only a modest shift for the azabenzimidazolone series; the loss of potency resulting from this change was regained by the addition of the 2- and 3-fluoro substituents on the left-hand phenyl ring. This gave the clinical candidate MK-0974 (6), which effectively balances potency and lipophilicity to give a favorable protein shift profile combined with modest oral bioavailability in rats (20%) and dogs (35%). MK-0974 showed similar efficacy to triptans in Phase II studies, and the outcome of Phase III studies is eagerly awaited. 2.3.3 Case History: Mixed Dopamine/Serotonin Receptor Antagonist As An Atypical Anti-Psychotic Combining activity at more than one biological target into a single molecule is attractive, as it allows multiple therapeutic mechanisms to be exploited with a reduced side effect and toxicological burden compared to a cocktail of drugs each active at a single target, in addition to making life easier for patients and hence improving compliance. However, this also presents a significant challenge, as it may require the superposition of diverse pharmacophores into a single molecule, which tends to lead to structures with high molecular weights. Where there is close homology between the receptors in question, the probability of success in achieving multitarget activity with favorable ADMET properties is highest, since the individual pharmacophores may be closely related and hence show significant overlap. Drugs described as ‘‘promiscuous’’ provide encouragement to the approach: as an example, the atypical antipsychotic clozapine (Fig. 2.13) shows affinity at a

OMe

N

OMe N

N N

N

Cl

Ph N

N

OMe

N O

N H

OEt

Clozapine

N Cl

O

F

O N O Cl

N N

H2N

NH2 O

N

N O OH

O

H1/CCR3 antagonist Astrazeneca

FIGURE 2.13

N

H1/NK1 antagonist Hoechst Marion Roussel

H1 antagonist/5-LO inhibitor UCB

Examples of GPCR ligands active at more than one target.

44

G-PROTEIN-COUPLED RECEPTORS

minimum of 15 aminergic receptors. In addition, the pharmacophores of some receptors appear to be quite simple and are readily superimposed into more complex pharmacophores. As an example, histamine H1 receptor antagonism has been combined within other GPCR activity, including CCR3 or NK1 receptor antagonism, and beyond GPCRs with 5-lipoxygenase inhibition to tackle allergic responses via dual mechanisms (Fig. 2.13). The discovery of the mixed dopamine/serotonin receptor antagonist described below shows the successful application of a panel of assays and cross-target structure–activity relationships to derive compounds with robust activity at several receptors combined with good general selectivity and desirable ADMET properties. An atypical antipsychotic (13) with activity at several dopamine and serotonin receptors was discovered by scientists at GSK (Garzya et al., 2007). An analysis of the receptor interaction profile of marketed antipsychotics led to the identification of five key receptors for which selective antagonism should provide effective treatment of schizophrenia (Table 2.4). Selectivity over other monoamine receptors such as dopamine D1, histamine H1, muscarinic M1, alpha and beta adrenergic receptors was sought to minimize side effects, including sedation, hypotension, weight gain, movement disorders, and seizures associated with marketed antipsychotics. Focused screening of compounds with aminergic receptor activity identified the tetrahydroisoquinoline sulfonamide 10 (Fig. 2.14). This compound had previously been synthesized as a dopamine D3 ligand, and further profiling revealed promising activity at the other receptors of interest (D2 6.0, D3 8.0, 5HT2A 7.9, 5HT2C 7.5, 5HT6 7.6). Modification of the amino group by exploring ring sizes and N-substitution led to the N-methyl benzazepine 11, showing the desired receptor profile. Issues identified with 11 included poor metabolic stability and potent interactions with the 2C19 and 2D6 CYP450 isoforms. Exploratory chemistry focused on replacement of the butyl side chain with more metabolically stable isosteres. A chlorophenyl ring could be tolerated, with a moderate loss of general affinity (compound 12). However, combining this change with introduction of TABLE 2.4 Desired Aminergic Receptor Profile for New Generation Atypical Antipsychotic Receptor

Target Affinity pKi

D2

7–8

D3 5HT2A

>8 >8

5HT2C

>8

5HT6

>8

Rationale Antipsychotic activity (treatment of positive and negative symptoms) Contributes to antipsychotic activity and atypicality Counteracts D2-mediated extrapyramidal side effect (EPS) Anxiolytic effect Antidepressant effect Addresses cognitive deficits

45

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN O

O S

O

O

NH

N H

S

10

Me O S

N H

NMe

11

Me N O

O

O

NMe

S

N H

13

NMe

N H

12

Cl

Cl

Receptor pKi

10

11

12

13

D2

6.1

7.5

6.8

7.3

D3

7.9

8.7

7.8

8.5

5HT2A

7.9

8.3

6.7

8.8

5HT2C

7.5

8.0

7.7

8.3

5HT6

7.6

8.9

8.3

8.1

FIGURE 2.14 Key optimization steps from the focused screening hit 10 to a newgeneration atypical antipsychotic compound 13.

substituents to the 8-position of the benzazepine ring system reintroduced the potency; when combined with the chlorophenyl left-hand side, this led to molecules that also showed good DMPK properties (13). In rats, compound 13 showed moderate blood clearance of 39 mL/min/kg, good oral bioavailability and halflife (F ¼ 69%; t1=2 ¼ 1:8 h), and good CNS penetration (brain/blood ¼ 3.4:1). A compound from a related series achieved clinical proof of concept for schizophrenia. 2.3.4 Case History: Chemokine Receptor CCR5 Antagonist Maraviroc (CelsentriTM) Ligands for GPCRs cover a wide range of chemical diversity, and it should be no surprise that many also show affinity for proteins in other gene families (Schnur et al., 2006). Empirically, the most overlap occurs with ion channels, which is consistent with their similar transmembrane helical structure. During the course of GPCR discovery programs, it is not uncommon to find significant hERG and sodium channel blocking activity, particularly if the ligand contains a basic center. The range of selectivity assays available to GPCR programs is continually increasing, and it is prudent to use an up-to-date assay panel to check for unwanted activity at regular intervals during optimization. The case history below summarizes

46

G-PROTEIN-COUPLED RECEPTORS

how scientists at Pfizer met the challenge of designing an orally bioavailable antagonist for the CCR5 receptor, successfully negotiating the issue of potent hERG blockade in the lead series. Maraviroc, an antagonist of the CCR5 chemokine receptor, was discovered by scientists at Pfizer UK (Wood and Armour, 2005). CCR5 is a coreceptor formed by binding of HIV gp120 protein to host CD4 cell surface receptors. Chemokine binding to CCR5 triggers conformational changes in the viral gp41 fusion protein, unmasking the fusion peptide and facilitating its insertion into the host cell lipid bilayer and subsequent viral entry. Targeting the viral fusion process has become a focus of research for a new generation of HIV antiretroviral therapies. The starting molecules for medicinal chemistry were identified by HTS, measuring binding displacement of the radiolabeled chemokine MIP-1b. Two hits 14 and 15 shown in Fig. 2.15 were selected based on a balance of potency, ligand efficiency, and potential for physicochemical optimization (IC50/LE 0.4 mM, 0.29 and 1.1 mM, 0.20 kcal/mol/non-H atom, respectively). Key issues included relatively high molecular weight and lipophilicity, a lack of antiviral activity, and, in the case of imidazopyridine 14, potent inhibition of the CYP 2D6 isoform, consistent with coordination of the pyridyl nitrogen to the heme and interaction of the tertiary amino group with an active site aspartyl side chain. Removal of the pyridyl nitrogen reduced the inhibition of 2D6 while enhancing potency, and introduction of a core amide reduced lipophilicity, giving the lead 16 with improved properties and showing promising antiviral activity. The potency was enhanced further with the aliphatic cyclobutyl group replacing phenyl, and exploration of a number of conformationally constrained analogues of the core piperidine led Me

Me N

N

N

N

H N N

O

Me

14

16

N N

N

N

N

N

O

15 Cl F

Cl

F

Me N Me O

Me

NH N

N

18

Me

N

H N

N

N

O

17

N

FIGURE 2.15 Key optimization steps from the CCR5 HTS hits 14 and 15 to maraviroc (18).

CHALLENGES FACING THE AREA OF GPCR DRUG DESIGN

47

to the identification of the tropane 17. These modifications effectively resolved the 2D6 issue, while maintaining good receptor affinity and antiviral activity. However, at this point potent hERG inhibition was uncovered, together with high clearance in hepatocytes. Within the tropane series, exploration of more polar replacements for the benzimidazole led to the discovery of the triazole group, which improved metabolic stability and hERG affinity. Final optimization of the amide substituent, taking care to balance lipophilicity with permeability and metabolic stability, led to the discovery of maraviroc (18). It is of interest to note that this balance was achieved with some compromise: predicted bioavailability was 10% driven by 20% permeability and 50% first-pass loss. However, in clinical trials sufficient levels of drug were achieved to meet antiviral criteria, and the molecule successfully entered the HIV/AIDS prescription market in 2007. 2.3.5 Case History: The Discovery of Cinacalcet (Sensipar1/Mimpara1), a CaSR-Positive Allosteric Modulator As was discussed above, allosteric interaction at 7TMs is well established and it can provide advantage in a number of instances, notably through increased tractability or improved selectivity. The discovery of cinacalcet (below) is an interesting example in this context. The compound is a positive allosteric modulator of the calcium-sensing receptor. Targeting the orthosteric site in this case is problematic because the endogenous ligand is an ion and hence difficult to mimic with something drug-like. It is noteworthy that the researchers chose a functional assay in native tissue as their initial screen. A binding assay (assuming one could be configured for Ca2þ) might not detect allosteric molecules, whereas use of a recombinant assay may have failed to account correctly for additional protein partners having an effect via receptor dimerization, through accessory proteins, or through use of an artificial G-protein coupling. It may be impractical to suggest such a route for the primary screen in many cases but confirmation of recombinant activity in native systems is best performed early. Allosteric activators will require greater work to fully delineate their mode of action (agonist versus positive modulator versus ago-allosteric modulator) but this is arguably the price we pay for greater control. If the mode of action is positive or ago-allosteric modulation, a choice will have to be made about whether to seek optimization of shift in potency, efficacy, or both and there will be a need to understand the basal tone of the endogenous ligand in the target tissue or at least what level of effect to target in vivo. This certainly adds to the challenge of the drug discovery process. The calcium-sensing receptor is a Family C 7TM that senses extracellular Ca2þ through binding of the ion to the NTD. This binding causes the NTD to close via the ‘‘venus flytrap’’ mechanism, giving rise to receptor activation, which in turn serves to regulate calcium homeostasis. Increase in extracellular calcium causes a CaSRmediated decrease in parathyroid hormone levels and increase in calcitonin levels. PTH plays a role in increasing extracellular calcium, whereas calcitonin decreases it by stimulating bone formation and decreasing renal calcium resorption. Hence,

48

G-PROTEIN-COUPLED RECEPTORS

H N

H N Me

fendiline

FIGURE 2.16

R

F3C

H N

Me

R-467 (R = H) R-568 R 568 (R = Cl)

Me

cinacalcet

Progression of key compounds in the discovery of cinacalcet.

compounds that mimic calcium action at CaSR have been pursued as potential treatments for diseases associated with raised calcium levels, which include primary or secondary hyperparathyroidism (HPT) and parathyroid carcinoma. Ligands that mimic calcium activation of CaSR are described as calcimimetics. Type I calcimimetics bind to the orthosteric site, whereas type II are positive allosteric modulators. Cinacalcet is a type II calcimimetic that was approved in 2004 for the treatment of secondary HPT in patients with chronic kidney disease on dialysis and hypercalcemia in patients with parathyroid carcinoma. As such, it is the first positive allosteric modulator of a 7TM to be successfully launched. The path to the discovery of cinacalcet is summarized in Fig. 2.16. An initial screen for compounds with agonist-like activity, capable of increasing intracellular Ca2þ levels in bovine parathyroid cells, led to the discovery of fendiline (EC50 ¼ 12 mM). R-467 and R-568 were subsequently identified by preparing and screening fendiline analogues. In each case, both stereoisomers were found to be active but the R enantiomer was more potent than the S, and R-568 was more potent than R-467. It was found that the compounds were only active in the presence of at least 0.5 mM extracellular calcium and that the effect was saturated if calcium levels exceeded 1.5 mM. The compounds caused a left shift in the dose–response curve but had no effect on the maximum response. CaSR dependence of the activity was confirmed by screening in wild-type HEK293 cells versus HEKs transfected with CaSR. The compounds were also shown to be selective over other Family C 7TMs (mGluR1a, mGluR2, and mGluR8). R-568 was progressed initially. When orally administered to Sprague–Dawley rats, it caused a dose-dependent drop in serum Ca2þ and PTH. The compound was subsequently taken into clinical trials, where it was shown to be efficacious at lowering serum Ca2þ and PTH levels in patients with primary or secondary HPT. These results provided a proof of concept for the mechanism of drug intervention in the clinic but the compound was discontinued due to a poor metabolic profile (CYP2D6 activity) and a poor oral bioavailability (ca. 1%). Subsequent optimization led to the discovery of cinacalcet, which has 74% oral bioavailability in humans. However, the compound is metabolized by CYP3A4, CYP2D6, and CYP1A2 isozymes. The effect of coadministration of CYP3A4 and CYP2D6 inhibitors was investigated in Phase I studies. We can model the NTD of Family C receptors with a reasonable degree of accuracy due to the availability of multiple crystal structures. However, modeling

49

CONCLUSIONS AND OUTLOOK

OH

CN Cl NH

N H

Cl

O

H N

O NPS 2143

Calhex 231

FIGURE 2.17

CaSR negative allosteric modulators (calcilytics).

the TM bundle is rather more difficult due to the limited homology with rhodopsin or beta-2. In this case, the compounds are thought to bind in the TM region. Despite the difficulties, receptor models have been built, and SDM data point to a key charge–charge interaction with the Glu-837 residue on TM7 and provide further detail of lipophilic/pi-stacking-type interactions, although reports vary on the role of certain amino acids such as Phe-668 and Phe-684 from TM3. Such data provide a good starting point for probing ligand interaction at the receptor but more work is required to understand the detail, in terms of both predictive ligand design/optimization and the molecular mechanism for PAM activity at a 7TM. In the context of the latter, it is interesting to note that the binding site, while not identical, appears to overlap to a fair extent with that occupied by negative allosteric modulators (antagonists, calcilytics) such as NPS2143 and Calhex 231 (Fig. 2.17) but it is currently beyond the scope of the models to intentionally design compounds that switch mode of action.

2.4

CONCLUSIONS AND OUTLOOK

Two key challenges must be addressed by drug hunters of the future, namely, the confirmation of biological targets as valid players in disease as early and economically as possible and the delivery of optimum molecules that reduce the high failure rates in early development. The field of GPCR research continues to provide fertile ground for discovery, as evidenced by the case histories described above and many others. These successes have been achieved despite a shift of interest away from the highly tractable aminergic subfamily, indicating a growing ability of medicinal chemists to balance ADMET properties with activity at challenging pharmacophores. Recent advances in understanding of GPCR allosterism continue to enhance the tractability of once recalcitrant targets. Finally, the tremendous progress made in determining receptor structures in the last year is certain to make a significant impact on GPCR drug discovery. As the structural repertoire extends beyond beta adrenergic receptors, one can envision a growing understanding of common features required for agonism versus antagonism, of similarities between binding sites of different receptors that may assist the rapid identification of ligands, and of features that will allow bespoke design of selectivity profiles.

50

G-PROTEIN-COUPLED RECEPTORS

REFERENCES Ali MA, Nirmala B, Findlay JBC, Fishwick CWG, 2005. The first de novo-designed antagonists of the human NK2 receptor. J. Med. Chem., 48:5655–5658. Attwood TK, Findlay JBC, 1994. Fingerprinting G-protein-coupled receptors. Protein Eng. 7(2):195–203. Bakker RA, Leurs R, 2005. From the human genome to new drugs: the potential of orphan G-protein-coupled receptors, in: Dingermann T, Steinhilber D, Folkers G, Mannhold R, Kubinyi H, editors. Molecular Biology in Medicinal Chemistry. Wiley, Weinheim, pp. 95– 122. Ballesteros JA, Weinstein H, 1995. Integrated methods for the construction of three-dimensional models and computational probing of structure–function relations in G protein-coupled receptors. Methods Neurosci. 25:366–428. Barton N, Blaney FE, Garland S, Tehan B, Wall I, 2007. Seven transmembrane G-proteincoupled receptors: insights for drug design from structure and modelling, in: Taylor JB, Triggle DJ, editors. Comprehensive Medicinal Chemistry II, Vol. 4. Elsevier, pp. 669–701. Bjarnadottir TK, Gloriam DE, Hellstrand SH, Kristiansson H, Fredriksson R, Schioeth HB, 2006. Comprehensive repertoire and phylogenetic analysis of the G-protein-coupled receptors in human and mouse. Genomics 88(3):263–273. Black J, 2004. Drugs from emasculated hormones: the principle of syntopic antagonism. Biosci. Rep. 24(4–5):302–22. Bondensgaard K, Ankersen M, Thøgersen H, Hansen BS, Wulff BS, Bywater RP, 2004. Recognition of privileged structures by G-protein coupled receptors. J. Med. Chem. 47:888–899. Cherezov V, Rosenbaum DM, Hanson MA, Rasmussen SGF, Thian FS, Kobilka TS, Choi H-J, Kuhn P, Weis WI, Kobilka BK, Stevens RC, 2007. High resolution crystal structure of an engineered human b2-adrenergic G protein-coupled receptor. Science 318(5854):1258– 1265. Deupi X, Kobilka B, 2007. Activation of G protein-coupled receptors. Adv. Protein Chem. 74 (Mechanisms and Pathways of Heterotrimeric G Protein Signaling):137–166. Drews J, 2000. Drug discovery: a historical perspective. Science 287:1960–1964. Edvardsen O, Reiersen AL, Beukers MW, Kristiansen K, 2002. tGRAP, the G-protein coupled receptors mutant database. Nucleic Acids Res. 30:361–363. Foord SM, Jupe S, Holbrook J, 2002. Bioinformatics and type II G-protein-coupled receptors. Biochem. Soc. Trans. 30(4):473–479. Foord, SM, Bonner TI, Neubig RR, Rosser EM, Pin J-P, Davenport AP, Spedding M, Harmar AJ, 2005. International Union of Pharmacology. XLVI. G protein-coupled receptor list. Pharmacol. Rev. 57(2):279–288. Garzya V, Forbes IT, Gribble AD, Hadley MS, Lightfoot AP, Payne AH, Smith AB, Douglas SE, Cooper DG, Stansfield IG, Meeson M, Dodds EE, Jones DNC, Wood M, Reavill C, Scorer CA, Worby A, Riley G, Eddershaw P, Ioannou C, Donati D, Hagan JJ, Ratti EA, 2007. Studies towards the identification of a new generation of atypical antipsychotic agents. Bioorg. Med. Chem. Lett. 17(2):400–405. Gearing KL, Barnes A, Barnett J, Brown A, Cousens D, Dowell S, Green A, Patel K, Thomas P, Volpe F, Marshall F, 2003. Complex chimeras to map ligand binding sites of GPCRs. Protein Eng. 16(5):365–372.

REFERENCES

51

Gloriam DE, Fredriksson R, Schioeth HB, 2007. The G protein-coupled receptor subset of the rat genome. BMC Genomics 8:338–355. Gonzalez-Maeso J, Ang RL, Yuen T, Chan P, Weisstaub NV, Lopez-Gimenez JF, Zhou M, Okawa Y, Callado LF, Milligan G, Gingrich JA, Filizola M, Meana JJ, Sealfon SC, 2008. Identification of a serotonin/glutamate receptor complex implicated in psychosis. Nature 452(7183):93–97. Hoare SRJ, 2007. Allosteric modulators of class B G-protein coupled receptors, Curr. Neuropharmacol. 5:168–179. Kenakin TP, 2004. Principles: receptor theory in pharmacology. TiPS 25(4):186–192. Kobilka B, Schertler GFX, 2008. New G-protein-coupled receptor crystal structures: insights and limitations. Trends Pharmacol Sci 29(2):79–83. Kristiansen K, Dahl SG, Edvardsen O, 1996. Proteins 26:81–94. Kunishima N, Shimadat Y, Tsuji Y, Sato T, Yamamoto M, Kumasaka T, Nakanishi S, Jingami H, Morikawa K, 2000. Structural basis of glutamate recognition by a dimeric metabotropic glutamate receptor. Nature 407(6807):971–977. ˚ crystal Lambright DG, Sondek J, Bohm A, Skiba NP, Hamm HE, Sigler PB, 1996. The 2.0 A structure of a heterotrimeric G protein. Nature 379(6563):311–319. Martin RE, Green LG, Guba W, Kratochwil N, Christ A, 2007. Discovery of the first nonpeptidic, small-molecule, highly selective somatostatin receptor subtype 5 antagonists: a chemogenomics approach. J. Med. Chem. 50(25):6291–6294. Muto T, Tsuchiya D, Morikawa K, Jingami H, 2007. Structures of the extracellular regions of the group II/III metabotropic glutamate receptors. Proc. Natl. Acad. Sci. USA 104(10): 3759–3764. Palczewski K, Kumasaka T, Hori T, Behnke CA, Motoshima H, Fox BA, Le Trong I, Teller DC, Okada T, Stenkamp RE, Yamamoto M, Miyano M, 2000. Crystal structure of rhodopsin: a G protein-coupled receptor. Science 289:739–745. Paone DV, Shaw AW, Nguyen DN, Burgey CS, Deng JZ, Kane SA, Koblan KS, Salvatore CA, Mosser SD, Johnston VK, Wong BK, Miller-Stein CM, Hershey JC, Graham SL, Vacca JP, Williams TM, 2007. Potent, orally bioavailable calcitonin gene-related peptide receptor antagonists for the treatment of migraine: discovery of N-[(3R,6S)-6-(2,3-difluorophenyl)2-oxo-1- (2,2,2-trifluoroethyl)azepan-3-yl]-4- (2-oxo-2,3-dihydro-1H-imidazo[4,5-b]pyridin- 1-yl)piperidine-1-carboxamide (MK-0974). J. Med. Chem. 50(23):5564. Parthier C, Kleinschmidt M, Neumann P, Rudolph R, Manhart S, Schlenzig D, Fanghanel J, Rahfield H-U, Stubbs MT, 2007. Crystal structure of the incretin-bound extracellular domain of a G protein-coupled receptor. Proc. Natl. Acad. Sci. USA 104(35):13942– 13947. Pioszak, AA, Xu HE, 2008. Molecular recognition of parathyroid hormone by its G-protein coupled receptor. Proc. Natl. Acad. Sci. USA 105(13):5034–5039. Rasmussen SGF, Choi H-J, Rosenbaum DM, Kobilka TS, Thian FS, Edwards PC, Burghammer M, Ratnala VRP, Sanishvili R, Fischetti RF, Schertler GFX, Weis WI, Kobilka BK, 2007. Crystal structure of the human b2 adrenergic G-protein-coupled receptor. Nature 450(7168):383–387. Schertler GF, Villa C, Henderson R, 1993. Projection structure of rhodopsin. Nature 362:770–772. Schioeth HB, Fredriksson R, 2005. The GRAFS classification system of G-protein coupled receptors in comparative perspective. Gen. Comp. Endocrinol. 142:94–101.

52

G-PROTEIN-COUPLED RECEPTORS

Schnur DM, Hermsmeier MA, Tebben AJ, 2006. Are target-family-privileged substructures truly privileged? J. Med. Chem. 49:2000. Schwartz TW, Frimurer TM, Holst B, Rosenkilde MM, Elling CE, 2006. Molecular mechanisms of 7TM receptor activation—a global toggle switch model. Annu. Rev. Pharmacol. Toxicol. 46:481–519. Siehler S, 2007. G12/13-dependent signaling of G-protein-coupled receptors: disease context and impact on drug discovery. Expert Opin. Drug Discov. 2(12):1591–1604. Unger VM, Hargrave PA, Baldwin JM, Schertler GF, 1997. Arrangement of rhodopsin transmembrane a-helixes. Nature 389:203–206. Wall MA, Coleman DE, Lee E, Iniguez-Lluhi JA, Posner BA, Gilman AG, Sprang SR, 1995. The structure of the G protein heterotrimer Gia 1b 1g 2. Cell 83(6):1047–1058. Wise A, Gearing K, Rees S, 2002. Target validation of G-protein coupled receptors. Drug Discov. Today 7:235–246. Wood A, Armour D, 2005. The discovery of the CCR5 receptor antagonist, UK-427, 857, a new agent for the treatment of HIV infection and AIDS. Prog. Med. Chem. 43:239.

3 ION CHANNELS GENE FAMILY: STRATEGIES FOR DISCOVERING ION CHANNEL DRUGS MARIA L GARCIA

3.1

AND

GREGORY J. KACZOROWSKI

INTRODUCTION

Ion channels represent a large superfamily of proteins with diverse structural features and functions. Ion channels are integral membrane proteins that, when gated open, allow the movement of ions at rates of >106 molecules per second, which approach diffusion control. To sustain this high rate of passage, ion channels must possess an aqueous pore within the low dielectric environment of the membrane. Since the first living organisms emerged by creating a diffusion barrier that insulated the inside of the cell from the external environment, ion channels have been critical for survival during the evolutionary process. In mammals, complex arrays of information must be processed accurately, and often very rapidly, to sustain life. In excitable cells, electrical signals regulate among other functions nerve conduction, muscle contraction, and neurotransmitter and hormone secretion. In nonexcitable cells, changes in intracellular calcium concentrations are responsible for regulating patterns of gene expression and cell proliferation, whereas in epithelial cells, large amounts of electrolyte movements are required to maintain proper ionic homeostasis (Hille, 2001). All these processes require the perfect coordination of multiple ion channels functioning together to regulate overall cellular physiology. When the activity of one of these ion channels becomes modified, the system’s equilibrium is altered and the general physiological impact can be severe and lead to a disease state (Ashcroft, 2000).

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

53

54

ION CHANNELS GENE FAMILY

Although ion channels appear to be highly conserved when comparing similar proteins from unicellular organisms to those of mammals, the more elaborate functions required by the later species need greater functional diversity in these proteins. In a simplified manner, ion channels transition between two major states: (1) closed, a conformation in which ion permeation does not occur and (2) open, which allows permeation. The gating mechanism that allows channels to enter the open state can be triggered by changes in membrane voltage, the binding of ligands, or via mechanical forces induced by membrane stretch. Only the family of voltage-gated ion channels will be discussed in more detail below. Ligand-gated ion channels open in response to binding of a specific second messenger and are usually selective for one or more ions such as Naþ , Kþ , Ca2þ , or Cl
. These channels are present at synapses where they convert the chemical signal carried by specific presynaptically released neurotransmitters into an electrical signal at the postsynaptic membrane. Examples of such types of ligand-gated ion channels are the nicotinic acetylcholine, g-aminobutyric acid, glutamate, and glycine receptor proteins. The most extensively studied of this channel class is the nicotinic acetylcholine receptor present at the vertebrate neuromuscular junction. Binding of two acetylcholine molecules at the outer vestibule of the channel on the muscle fiber membrane surface causes a conformational change that opens the pore, allowing Naþ influx into the cell and leading to membrane depolarization. This depolarization opens voltage-gated Naþ channels, whose activity then initiates and propagates an action potential that ultimately induces a mechanical twitch in the muscle. Nicotinic acetylcholine receptors are a pentameric complex of homologous subunits that form a central pore, mostly contributed to by residues from the second transmembrane domain of each subunit. Other postsynaptic ligand-gated channels, such as glycine or g-aminobutyric acid receptors, share similar architecture with the acetylcholine receptor. These two ligand-gated channels, however, are permeable to anions, but not to cations. Cation-selective glutamate-gated channels, on the contrary, have a different polypeptide folding scheme and are tetramers of homologous subunits. Glutamate is the predominant excitatory meurotransmitter in the vertebrate central nervous system. At least three classes of postsynaptic glutamate receptors have been identified pharmacologically: (1) N-methyl-D-aspartate, (2) kainate, and (3) AMPA. These receptors are more permeable to Ca2þ than monovalent ions, and the channels play an important role in controlling intracellular Ca2þ concentration.

3.2 3.2.1

ION CHANNEL SUBFAMILY DESCRIPTIONS Voltage-Gated Ion Channels

This large family, consisting of >140 members, follows in size the families of G-protein-coupled receptors and protein kinases, within the group of molecules specializing in signal transduction processes (Yu and Catterall, 2004). A common feature of all members of the voltage-gated ion channel family is found in the general architecture of the pore region, whereas other modules attached to this region

ION CHANNEL SUBFAMILY DESCRIPTIONS

55

FIGURE 3.1 Voltage-dependent ion channels possess similar architectural features. Ion selectivity resides within the pore domain. When a gating domain is covalently attached, different functional properties result. Functional diversity also occurs if auxiliary subunits associate with the ion channel. Other mechanisms, such as phosphorylation (green diamonds), contribute to the regulation of ion channel function. From Garcia and Kaczorowski, Potassium channels as targets for therapeutic intervention, Sci. STKE 2005, pe46 (2005). (See the color version of this figure in the Color Plates section.)

of the protein allow for great functional diversity as shown in Fig. 3.1 (Garcia and Kaczorowski, 2005). The pore region is constructed by association of 4 twotransmembrane regions that are linked by the pore loop. Depending on the channel under consideration, the central pore can be formed by either association of four independent (identical or closely related) subunits, two independent subunits from members of the two-pore channel family, or by folding of a single polypeptide in which four structurally related tandems are covalently linked together. Auxiliary subunits that modulate channel activity may also be present in the complex. Eight groups of proteins, that is, inwardly rectifier potassium, two-pore potassium, voltage-gated sodium, voltage-gated calcium, voltage-gated potassium, calcium-activated potassium, cyclic nucleotide-modulated, and transient receptor potential channels comprise this large family of ion channels depicted in Fig. 3.2. Given the high complexity and diversity of this superfamily, a detailed discussion of the entire voltage-gated ion channel family is not feasible in this chapter, and we will restrict our commentary to certain prototypical classes. 3.2.2

Inward Rectifier Potassium Channels

Inward rectifier (Kir) potassium channels contain the minimum structural motif that is required to confer potassium selectivity to an ion channel, and as such they are comprised of a tetramer formed by the association of four-pore domain subunits

56

ION CHANNELS GENE FAMILY

FIGURE 3.2 The voltage-gated ion channel superfamily. Four-domain calcium and sodium channels are shown as blue branches, potassium channels as red branches, cyclic nucleotidegated channels as magenta branches, and transient receptor potential and related channels as green branches. From Yu and Catteral, the VGL-chanome: a protein superfamily specialized for electrical signaling and ionic homeostasis. Sci. STKE 2004, re15 (2004). Reprinted with permission from AAAS. (See the color version of this figure in the Color Plates section.)

(Haider et al., 2007). There are 15 Kir members within seven subfamily classes listed in Table 3.1. The name inward rectifier is derived from the fact that these channels conduct more Kþ in the inward than in the outward direction. This rectifying phenomenon results from voltage-dependent block of the channel by intracellular Mg2þ and polyamines (Matsuda et al., 1987; Ficker et al., 1994; Lopatin et al., 1995). Thus, although Kir channels are not gated by voltage, changes in membrane potential can regulate their function. Depending on the degree of rectification, these channels can be classified as strong or weak rectifiers (Nishida and MacKinnon, 2002). Examples of strongly rectifying channels are Kir2.1 and the muscarinic acetylcholine receptor regulated Kir3.1/3.3, both of which are present in heart, whereas the kidney epithelial Kir1.1 and the pancreatic beta cell Kir6.2 represent two types of weak inwardly rectifying channels.

57

ION CHANNEL SUBFAMILY DESCRIPTIONS

TABLE 3.1 IUPHAR Kir1.1 Kir2.1 Kir2.2 Kir2.3 Kir2.4 Kir3.1 Kir3.2 Kir3.3 Kir3.4 Kir4.1 Kir4.2 Kir5.1 Kir6.1 Kir6.2 Kir7.1

a

Inwardly Rectifying Family of Potassium Channels HGNCa

Chromosomal Localization

KCNJ1 KCNJ2 KCNJ12 KCNJ4 KCNJ14 KCNJ3 KCNJ6 KCNJ9 KCNJ5 KCNJ10 KCNJ15 KCNJ16 KCNJ8 KCNJ11 KCNJ13

11q24 17q24.3 17p11.1 22q13.1 19q13 2q24.1 21q22.1 1q23.2 11q24 1q23.2 21q22.2 17q24.3 12p11.1 11p15.1 2q37

a

International Union of Pharmacology (IUPHAR) and HUGO Gene Nomenclature Committee (HGNC) names for the members of this group together with their human chromosomal localization are provided.

The inward rectifiers mediate cellular Kþ efflux at physiological membrane potentials and, among other functions, set and regulate the resting membrane potential in neurons, modulate insulin secretion from pancreatic beta cells in response to metabolic signals, regulate heart rate, and also participate in transepithelial Kþ flux in kidney. Mutations in these proteins are linked to autosomal inherited diseases in man, such as Bartter’s syndrome type II (Kir1.1) (Simon et al., 1996), Andersen– Tawil syndrome (Kir2.1) (Dhamoon and Jalife, 2005), and familial persistent hyperinsulinaemic hypoglycaemia of infancy (Kir6.2) (Ashcroft, 2000). Although most Kir channels are tetrameric complexes of identical or related Kir subunits, functional ATP-dependent Kþ (KATP) channels are formed by the association of four Kir6 pore and four auxiliary sulfonylurea receptor subunits. KATP channels of pancreatic beta cells are the target of sulfonylureas, drugs used as first-line treatment to normalize glucose levels in patients with type-II diabetes. By blocking this channel, sulfonylureas depolarize beta cells and enhance calcium entry through voltage-gated calcium channels leading to enhanced insulin secretion. Both subunits of the KATP channel are needed for functional expression of the channel (Crane and AguilarBryan, 2004). Association of the subunits in the endoplasmic reticulum early on after biosynthesis appears to promote trafficking of the complex to the cell surface by masking intracellular retention signals that are present within each subunit (Zerangue et al., 1999). KATP channels with a different subunit composition than those found in beta cells are also present in smooth muscle and heart. These channels have been a focus of attention in the pharmaceutical industry for several years. It was speculated that agonists of smooth muscle KATP channels would represent novel therapeutic agents for treatment of smooth muscle disorders such as

58

ION CHANNELS GENE FAMILY

asthma, hypertension, and urinary incontinence, and certain approved drugs (e.g., diazoxide, minoxidil sulfate) function by this mechanism (Mannhold, 2004). The most extensively investigated agent within this series of agonists, cromakalin, underwent clinical trials in the recent past as an antihypertensive, but its efficacy and safety profile did not appear to provide a significant advantage over other clinically approved agents. Unlike KATP channels, the pharmacology of the other members of the Kir family is quite limited, and small-molecule, high-affinity blockers have not been disclosed. The only potent and selective blocker of Kir1.1 and Kir3.1/3.3 channels is tertiapin, a 21-amino acid peptide isolated from bee venom (Jin and Lu, 1998). Tertiapin has been used to demonstrate the presence of a constitutively active acetylcholine-modulated IK(Ach) (Kir3.1/3.3) current in atrial myocytes from patients with chronic atrial fibrillation (Dobrev, 2005). 3.2.3

Voltage-Gated Potassium Channels

Unlike Kir channels, voltage-gated (Kv) potassium channels open in response to changes in voltage across the cell membrane. Kv channels are transcribed from 40 genes and comprise 12 subfamilies, Kv1–Kv12 listed in Table 3.2. The distinguishing feature of Kv subunits is denoted by two covalently attached modules consisting of a pore domain, similar to that of Kir channels, and a voltage-sensing domain that contains four transmembrane regions, S1–S4. An extensive body of evidence suggests that the S4 segment in the S1–S4 domain is the major contributor to voltage sensing for channel activation (Long et al., 2005). Kv channels are made up by association of four identical or closely related subunits, which provides large functional diversity for this family of ion channels (Dolly et al., 1994). In addition, auxiliary beta subunits also associate with Kv channels that can alter the biophysical and pharmacological properties, as well as cellular localization of the particular Kv channel with which they associate. It is worth noting that some Kv subfamilies, such as Kv5, 6, 8, and 9, do not appear to form functional channels when expressed by themselves in heterologous systems, but they are able to associate with subunits from the Kv2 family and modify the properties of the resulting heteromeric complex (Yu and Catterall, 2004). Kv channels undergo inactivation, another nonconducting conformational state typical of many ion channels, by one of the two different mechanisms that confer either fast or slow inactivation. Fast inactivation, also called N-type inactivation, results from blocking of the conduction pore due to binding of channel residues to a receptor site at the inner mouth of the channel. This mechanism is depicted by a ball-and-chain scheme where the inactivation ball swings into the inner channel pore and this movement is facilitated by the chain (Hoshi et al., 1990). The inactivation ball either is present on the cytoplasmic N-terminus region of a given Kv subunit, or is provided by a beta subunit (Rettig et al., 1994). Slow inactivation of Kv channels, also called C-type inactivation, appears to involve residues present at the outer mouth of the pore where a conformational change takes place in the channel that leads to occlusion of the pore (Lui et al., 1996). Kv channels are widely distributed. In excitable cells, they maintain the resting membrane potential, repolarize cells after a depolarization event, modulate the

59

ION CHANNEL SUBFAMILY DESCRIPTIONS

TABLE 3.2 IUPHAR Kv1.1 Kv1.2 Kv1.3 Kv1.4 Kv1.5 Kv1.6 Kv1.7 Kv1.8 Kv2.1 Kv2.2 Kv3.1 Kv3.2 Kv3.3 Kv3.4 Kv4.1 Kv4.2 Kv4.3 Kv5.1 Kv6.1 Kv6.2 Kv6.3 Kv6.4 Kv7.1 Kv7.2 Kv7.3 Kv7.4 Kv7.5 Kv8.1 Kv8.2 Kv9.1 Kv9.2 Kv9.3 Kv10.1 Kv10.2 Kv11.1 Kv11.2 Kv11.3 Kv12.1 Kv12.2 Kv12.3 a

a

Voltage-Dependent Family of Potassium Channels HGNCa

Chromosomal Localization

KCNA1 KCNA2 KCNA3 KCNA4 KCNA5 KCNA6 KCNA7 KCNA10 KCNB1 KCNB2 KCNC1 KCNC2 KCNC3 KCNC4 KCND1 KCND2 KCND3 KCNF1 KCNG1 KCNG2 KCNG3 KCNG4 KCNQ1 KCNQ2 KCNQ3 KCNQ4 KCNQ5 KCNV1 KCNV2 KCNS1 KCNS2 KCNS3 KCNH1 KCNH5 KCNH2 KCNH6 KCNH7 KCNH8 KCNH3 KCNH4

12p13.32 1p13 1p13.3 11p14 12p13 12p13 19q13.3 1p13.1 20q13.2 8q13.2 11p15 12q14.1 19q13.33 1p21 Xp11.23 7q31 1p13.2 2p25 20q13 18q23 2p21 16q24.1 11p15.5 20q13.3 8q24 1p34 6q14 8q23.2 9p24.2 20q12 8q22 2p24 1q32.2 14q23.1 7q36.1 17q23.3 2q24.3 3p24.3 12q13 17q21

International Union of Pharmacology (IUPHAR) and HUGO Gene Nomenclature Committee (HGNC) names for the members of this group together with their human chromosomal localization are provided.

60

ION CHANNELS GENE FAMILY

shape of action potentials, and determine the frequency of cell firing. Consequently, the activity of these channels is associated with control of neuronal activity and neurotransmitter release, heart rate, cardiac and smooth muscle contraction, and endocrine secretion. In nonexcitable cells, Kv channels regulate the membrane potential and provide the driving force for calcium entry through non-voltage-dependent calcium channels, a process that is required, for example, for T lymphocyte proliferation, apoptosis, and tumor progression. A number of autosomal inherited diseases are due to changes in Kv channel function (Ashcroft, 2000; Shieh et al., 2000). Episodic ataxia type-1 is a dominant disease due to missense mutations in the neuronal Kv1.1 polypeptide that alter channel function. Heterozygote individuals would be expected to have reduced outward Kþ conductance that will lead to prolongation of action potentials, which in turn cause repetitive firing and enhanced neurotransmitter release, all of which are consistent with the symptoms of the disease. The most studied inherited Kv channel diseases are those that give rise to long QT syndromes. LQT syndromes are disorders characterized by prolonged or delayed ventricular repolarization that is manifested on the electrocardiogram as an increase in the QT interval. A markedly prolonged action potential predisposes the heart to a specific type of life-threatening arrhythmia known as torsade de pointes that can precipitate ventricular fibrillation and cause sudden death. Mutations in Kv7.1 (KCNQ1) or its auxiliary beta subunit, minK, which constitutes the slow activating, delayed rectifier cardiac Kþ current (IKs), are associated with the most common form of long QT syndrome, LQT1, whereas mutations in the rapidly activating delayed rectifier Kv11.1, hERG (IKr), are associated with LQT2. In addition, mutations in MiRP1, a putative auxilliary subunit that is thought to be associated with hERG, are linked to LQT6. Two forms of LQT1 have been identified: an autosomal dominant Romano–Ward syndrome, and the autosomal recessive Jervell and Lange– Nielsen syndrome. In addition to the cardiac disorder, Jervell and Lange-Nielsen patients suffer from profound congenital bilateral deafness, which is thought to result from loss of endolymph secretion by the marginal cells of the stria vascularis of the inner ear where Kv7.1 is expressed. Mutant Romano–Ward syndrome channels, expressed alone or with minK, fail to produce functional currents and cause a downregulation of the current amplitude when coexpressed with wild-type subunits, which explains the dominant nature of Romano–Ward LQT1. LQT2 mutations confer a dominant nature to the disease, and only heterozygous individuals have been identified. Some mutations alter the trafficking of functional channels to the membrane, or form functional channels with altered properties, leading to a reduction in the outward current during the repolarization phase of the cardiac action potential. Drugs that block Kv7.1 or Kv11.1 are responsible for the acquired form of LQT syndrome (Fermini and Fossa, 2003). H1 antagonists, such as terfenadine and astemizole, antipsychotics, such as sertindole, tricyclic antidepressants, and certain antibiotics and antiemetic agents have been shown to block Kv11.1 at similar concentrations to those found in plasma of individuals susceptible to lethal arrhythmias and have resulted in the recall of these drugs. Consequently, lack of inhibition of Kv11.1 is considered critical for future compound development and for determining their approval by regulatory agencies. Benign familial neonatal

ION CHANNEL SUBFAMILY DESCRIPTIONS

61

convulsions exhibit autosomal dominant inheritance, and the two forms of the disease have been mapped to mutations in Kv7.2 and Kv7.3 channels. These channels exist in a heteromultimeric complex that constitutes the muscarinic acetylcholine receptor-sensitive M-current, a critical determinant of electrical excitability in many neurons. Inhibition of M-current causes neuronal hyperexcitability that can lead to epileptic seizures. It appears that a 25% reduction in M-current is sufficient to cause the neuronal excitability characteristic of the disease. Retigabine, a smallmolecule activator of neuronal Kv7.2/Kv7.3 channels, is undergoing Phase III clinical studies for the treatment of epilepsy. Data reported from this study indicate that patients taking the highest dose of the drug had fewer seizures during the study compared to those taking placebo. Regitabine also appears to be effective in reducing seizure frequency (American Academy of Neurology, press release, April 2007). Blockers of the M-current, such as linorpidine, XE991, and DMP543, have shown efficacy in animal models of learning and memory and have been considered for treatment of Alzheimer’s disease. Several mutations in the Kv7.4 channel have been identified in families with autosomal dominant progressive hearing loss. In contrast to Kv7.1, Kv7.4 does not contribute to endolymph secretion, but it is critical for the function of sensory outer hair cells. Although Kv7.4 is able to coassemble with Kv7.3 in in vitro heterologous expression systems, it is not clear whether association of the subunits occurs in vivo. The Kv channel family possesses a rich pharmacology. Both small-molecule and peptide inhibitors of these channels have been identified and characterized, and some of these agents have entered clinical development (Shieh et al., 2000; Coghlan, 2001; Panyi et al., 2006). In general, the peptides isolated from venous of scorpions, spiders, snakes, and sea anemonae constitute the most potent and selective Kv channel inhibitors discovered to date. Although not appropriate for consideration in traditional drug development paradigms, these peptides that block channels by either physically occluding the pore or by altering the gating mechanism constitute important tools for determining the contribution of a given channel(s) to a physiological process and also for target validation. In this way, Kv1.3 has been identified as a relevant target for treating autoimmunity (Beeton et al., 2001, 2005, 2006; Chandy et al., 2004, 2006), whereas inhibitors of Kv2.1, a channel responsible for controlling action potential duration in pancreatic beta cells, are predicted to enhance glucose-dependent insulin secretion from this target tissue (Herrington et al., 2006). Identification of potent and selective Kv channel inhibitors that could enter clinical development represents a major challenge for medicinal chemistry, given the high similarity between these family members and their wide tissue distribution. 3.2.4

Calcium-Activated Potassium Channels

Calcium-activated potassium (KCa) channels represent a different group of proteins in which binding of not only a cytosolic ligand to the channel, in this case typically Ca2þ , but also sometimes Naþ , Cl
, or Hþ triggers conformational changes that lead to channel opening. Five families, KCa1–5, with a total of eight members, comprise

62

ION CHANNELS GENE FAMILY

TABLE 3.3 IUPHAR KCa1.1 KCa2.1 KCa2.2 KCa2.3 KCa3.1 KCa4.1 KCa4.2 KCa5.1

a

Calcium-Activated Family of Potassium Channels HGNCa KCNMA1 KCNN1 KCNN2 KCNN3 KCNN4 KCNT1 KCNT2 KCNU1

Chromosomal Localization 10q22 19p13.1 5q22.3 1q21.3 19q13.2 9q34.3 1q31.3 8p11.2

a

International Union of Pharmacology(IUPHAR)and HUGO Gene Nomenclature Committee (HGNC) names for the members of this group together with their human chromosomal localization are provided.

the KCa group of channels and are listed in Table 3.3. Structurally, KCa channels can be subdivided into two families. One group relates to the high-conductance KCa1 and KCa4–5 channels, and is characterized by the presence of a large terminal region containing two regulatory conductance (RCK) domains that is responsible for modulation of channel activity by cytoplasmic factors (Xia et al., 2004). KCa2 and KCa3 channels belong to the second group of channels with small and intermediate conductance, respectively. These channels are activated by binding of Ca2þ to calmodulin, which is associated with a specific regulatory site in the Cterminal region of the channel and constitutes the calcium sensor (Schumacher et al., 2001). Another feature that distinguishes members of the KCa family is the dependence of channel gating on voltage. KCa1 and KCa5, but not KCa4 channels, contain the typical arrangement of positively charged residues in S4 that are present in voltage-dependent channels, and consequently gating of KCa1 and KCa5 channels depends on changes in membrane potential. These channels also possess an additional membrane-spanning domain near the N-terminal region that places the N-terminus outside the cell. Despite the presence of a typical S4 transmembrane segment, KCa2 and KCa3 channels do not appear to be gated by voltage and consist of six transmembrane regions. The structural and functional diversity of KCa channels is also reflected in the large number of physiological processes that they are thought to regulate. Thus, smooth muscle contractility, synaptic neurotransmitter release, spatial integration and action potential generation, after-hyperpolarization in neurons, volume regulation in erythrocytes, and lymphocyte proliferation are some of the functions assigned to KCa channels. To date, no human disease has been found to be associated with any KCa channel. It is interesting to note that KCa2.3 protein and mRNA levels are increased in skeletal muscle following denervation and in patients with myotonic muscular dystrophy. However, it is unknown whether the increase in KCa2.3 expression is the cause or consequence of the muscle dystrophy. KCa1 channels have been extensively characterized because their large singlechannel conductance provides them with a signature easy to identify. These channels

ION CHANNEL SUBFAMILY DESCRIPTIONS

63

are present in smooth and skeletal, but not cardiac muscle, in neurons, in neuroendocrine tissue, and in epithelial cells. KCa1 channels are activated by both intracellular Ca2þ and membrane depolarization. It appears that strong depolarizations are able to open these channels in the nominal absence of Ca2þ and that the role of Ca2þ is to promote a conformational change that facilitates channel opening by voltage (Ledoux et al., 2006; Salkoff, 2006; Salkoff et al., 2006). Thus, in the presence of Ca2þ , the voltage dependence of channel activation is shifted to more hyperpolarized potentials. KCa1 channels are formed by the association of four pore-forming moieties with four auxiliary b subunits. The pore-forming subunit is expressed from a single gene, but multiple splicing contributes to the large diversity in functional phenotypes of these channels. In some cases, splicing can lead to retention of channels in intracellular compartments, a process that appears to be controlled by different factors, such as hormone levels and stress, and is thought to represent a mechanism by which to regulate channel activity. Four b subunit genes, 1–4, have been identified (Orio et al., 2002). These b subunits alter the biophysical and pharmacological properties of the pore-forming subunits with which they associate and thereby contribute to enlarging the functional diversity of KCa1 channels. b1 is exclusively associated with smooth muscle KCa1 channels and functions in the regulation of smooth muscle tone by enhancing the Ca2þ sensitivity of the channel (Ledoux et al., 2006). b4 appears to be present solely in neuronal KCa1 channels and confers slow gating properties to the channel, minimizing its contribution to membrane repolarization and leading to broadening of the action potential (Brenner et al., 2005). In chromaffin cells, KCa1 channels inactivate and this effect is due to the presence of b2, which possesses an inactivation ball at its cytoplasmic N-terminus conferring a N-type inactivation mechanism similar to that found with Kv channels (Xia et al., 1999). Fast inactivation is also present when the b3b splice variant is coexpressed, and, in addition, this subunit causes strong outward current rectification (Zeng et al., 2003). A number of potent and specific inhibitors of KCa1 have been identified and characterized. These agents include pore blocking peptides, such as charybdotoxin and the highly selective, iberiotoxin, and small-molecule inhibitors, such as penitrem A, paxilline, and verruculogen (Garcia and Kaczorowski, 2001). By and large, the small-molecule blockers do not discriminate in their blocking ability, with regard to the subunit composition of the channel. However, the binding site for peptide blockers, or their access pathway at the channel’s outer vestibule, can be physically occluded depending on the beta subunit that is associated with the channel. Thus, whereas b1 can enhance the affinity of the channel for charybdotoxin, channels in which b4 are present are insensitive to the peptide. In addition to KCa1 blockers, a great deal of interest exists in identifying potent and selective KCa1 activators that could be developed for treating conditions associated with smooth muscle hyperexcitability, such as asthma, hypertension, and urinary incontinence (Garcia et al., 2007). Although many groups in industry and academic institutions have reported the discovery of such molecules, their potency and off-target activities have prevented their use in proof of concept (POC) studies for target validation. With the advent of new technologies that allow testing large

64

ION CHANNELS GENE FAMILY

numbers of compounds in functional ion channel assays, the expectation remains that such ideal channel modulators could be identified in the future (Herrington et al., 2005). Although similar in overall structure and gating mechanisms, the lower conductance KCa2 and KCa3 channels can be distinguished by their pharmacological properties (Wulff et al., 2007) and tissue distribution. KCa2 channels are predominantly of neuronal origin and are highly sensitive to the bee venom peptide, apamin, and the small-molecule blocker, UCL1684. KCa3 channels, on the contrary, are found in nonneuronal tissues and can be blocked by the peptide charybdotoxin or the small molecules, clotrimazole and its analogue TRAM-34. Because they modulate volume regulation in erythrocytes (Begenisich et al., 2004), inhibitors of KCa3 channels are under development for treatment of sickle cell anemia, and their application has also been proposed for the prevention of restenosis and plaque formation (Tharp et al., 2006). However, the presence of KCa3 channels in vascular endothelium, and their possible role in regulating blood pressure (Si et al., 2006), may prevent the development of KCa3 blockers for the other therapeutic indications if unwanted side effects, such as hypertension, are associated with their use.

3.3

STRUCTURE OF POTASSIUM CHANNELS

Most structural information of voltage-gated ion channels comes from study of the potassium channel family, due to the pioneering work of MacKinnon’s laboratory. Since the initial high-resolution structure of the simple bacterial KcsA potassium channel (Doyle et al., 1998), other members of this family with different structural and functional properties have been crystallized and their structures determined. These studies have provided a clear understanding of the molecular features that control ion selectivity and permeation in potassium channels. In addition, insights into gating mechanisms are also emerging for voltage-dependent and calciumactivated potassium channels. All potassium channels contain a minimal structural feature, such as that found in Kir channels, that confers ion channel selectivity. This pore region consists of an outer transmembrane helix, an extracellular turret, the pore helix and selectivity filter, and the inner helix shown in Fig. 3.3. The inner helices of four subunits adopt an inverted teepee-like structure and form a right-handed bundle that lines the pore on the intracellular side of the selectivity filter. A large body of evidence suggests that bundle crossing is related to the activation gate. Above the bundle crossing, a wide ˚ in diameter, exists that traverses more than half of the memwater-filled pore, 10 A brane bilayer. This aqueous region of the pore, and the C-terminal negative charge of the pore helix that points toward the center of the cavity, allow the rapid diffusion of ions and the stabilization of a hydrated potassium ion at the center of the membrane, before it can reach the selectivity filter. Ion selectivity occurs at the selectivity filter, ˚ in length, is located between the central cavity and the extracellular which is 12 A solution, and contains four potassium binding sites in a row. At each of these binding sites, a dehydrated potassium ion can interact with eight oxygen atoms provided by

STRUCTURE OF POTASSIUM CHANNELS

65

FIGURE 3.3 Ion conduction in potassium channels. (a) Two of the four pore domain subunits of a potassium channels are shown. Each subunit contains an outer helix close to the membrane, an inner helix close to the pore, a pore helix (red), and a selectivity filter (yellow). The blue mesh indicates the electron density for potassium ions and water along the pore. (b) A close-up view of the selectivity filter. Dehydrated potassium ions can be seen at positions 1 through 4 in the filter, whereas a hydrated ion is present in the central cavity below the filter. (c) The electron density in the filter corresponds to the two configurations of potassium ions (1,3 and 2,4) alternating with water molecules. (d) Potassium conduction. Reprinted from FEBS Letters, 555, R. MacKinnon, Potassium channels, page 63, copyright (2003), with permission from Elsevier. (See the color version of this figure in the Color Plates section.)

the protein (Zhou et al., 2001a). The arrangement of protein oxygen atoms at each binding site is very similar to the arrangement of water molecules around a hydrated potassium ion, providing a structural feature that minimizes the energetic cost of dehydration of potassium as it diffuses into the selectivity filter. At a given time,

66

ION CHANNELS GENE FAMILY

two potassium ions separated by one water molecule are present in the selectivity filter and they reside in either of the two main locations, 1,3 or 2,4, until a third ion enters on one side and causes the displacement of an ion on the opposite side, again referring to Fig. 3.3. Electrostatic repulsion between closely spaced ions thus leads to high conduction rates in the channel (Morais-Cabral et al., 2001; Zhou and MacKinnon, 2003). A number of Kv channel inhibitors bind to the pore region. The outer vestibule of the channel is the site of interaction for scorpion, snake, and sea anemone peptides that block ion conduction by physical occlusion of the pore (Gross and MacKinnon, 1996; Ranganathan et al., 1996). Because there is structural diversity in this region of Kv channels, peptide inhibitors can display selectivity between channel subtypes. In addition, the high number of contact points between peptidyl blockers and the channel provides very high affinity for the interaction. Tetraethylammonium ion also binds in the outer vestibule to a residue located at the entry of the pore. The nature of this residue determines the level of sensitivity of a given Kv channel to this blocker, when applied from the extracellular side (Heginbotham and MacKinnon, 1992). Several structural classes of Kv inhibitors, including intracellularly applied quaternary ammonium ions, appear to bind within the water-filled cavity below the selectivity filter (Hanner et al., 2001; Zhou et al., 2001b; Yohannan et al., 2007). In general, binding of these compounds requires opening of the activation gate, and, therefore, these inhibitors are called state-dependent blockers. Most therapeutically used ion channel modulators, such as antihypertensive, anticonvulsant, and antiarrhythmic agents, belong to this category of inhibitors. The pore region of a potassium channel involved in ion selectivity and conduction is physically linked to separate modules that control the channel gating mechanism. From X-ray crystallography and functional studies, the molecular basis of two gating mechanisms, those mediated by voltage or calcium, is beginning to be understood. In Kv channels, the voltage sensor, constituted of the S1–S4 transmembrane regions, forms an independent domain inside the membrane and is mechanically linked to the pore through the S4–S5 linker helices shown in Fig. 3.4 (Jiang et al., 2003; Long et al., 2005a; Long et al., 2005b). The voltage sensor contains an excess of positively charged amino acids that respond to changes in the transmembrane electric field. It appears that the negatively charged lipid phosphodiester groups of the phospholipid bilayer can provide interactions that energetically stabilize the voltage sensor and facilitate its operation, suggesting that the existence of arginine residues in the voltage-sensing domain results from an adaptation of the protein to the phospholipid composition of the membrane (Schmidt et al., 2006). Peptides isolated from venoms of different organisms and a number of small, lipid soluble molecules can modify the gating properties of Kv channels and other voltage-gated ion channels. Gating-modifier peptides act at an extracellular site contributed by the S3–S4 linker and through a voltage-sensor trapping mechanism can enhance or inhibit the voltage dependence of channel activation or inactivation (Swartz and MacKinnon, 1997a; Swartz and MacKinnon, 1997b; Lee and MacKinnon, 2004; Ruta and MacKinnon, 2004). As discussed earlier, KCa1 and KCa4–5 channels contain two RCK domains at the cytoplasmic C-terminal that define their unique ligand dependence. These cytosolic

STRUCTURE OF POTASSIUM CHANNELS

67

FIGURE 3.4 Crystal structure of a mammalian voltage-gated potassium channel. Stereo views of the Kv1.2–b2 subunit complex. The four subunits are colored differently. In (a), TM indicates the integral membrane component of the complex. (b) A single subunit of the channel and b subunit are viewed from the side. (c) A view from the extracellular side of the pore. From Long et al. Crystal structure of a mammalian voltage-dependent Shaker family Kþ channel. Science 309: 897–903 (5 August 2005). Reprinted with permission from AAAS. (See the color version of this figure in the Color Plates section.)

regulatory modules are interchangeable within this class of ion channels (Xia et al., 2004). RCK domains contain a conserved pattern of a and b helices. The isolated RCK domain from Escherichia coli forms a homodimer with a bilobed architecture. The lobe of each RCK domain adopts a Rossmann fold, and dimerization produces a

68

ION CHANNELS GENE FAMILY

deep cleft between two lobes where ligands bind (Jiang et al., 2001). In the case of KCa1, the two RCK domains, connected to the pore region by a short linker, have been proposed to form a gating ring where calcium binding causes a conformational change that promotes channel opening by pulling on the linker connecting the gating ring to the pore region (Niu et al., 2004). Insights into the molecular mechanisms of ligand activation in KCa1 channels have been inferred from studies with the bacterial calcium-gated potassium channel, MthK (Jiang et al., 2002; Dong et al., 2005; Ye et al., 2006). The X-ray structure of MthK reveals an octameric gating ring, formed by eight RCK domains, where eight calcium ions can bind to induce a conformational change that opens the channel. In its ligand-free closed state, the diameter of the gating ring is smaller than when calcium is bound, and this region exerts minimal mechanical stress on the pore. The expansion of the gating ring upon calcium binding exerts force on the pore-lining inner helices that will open the intracellular gate at the bundle crossing by bending the inner helices at the conserved glycine residue. Although further studies are needed to understand the intricate molecular mechanisms that control the function of ion channels, progress concerning certain structural features of these mechanisms may help in the future to determine how small-molecule channel modulators interact with the channel and may lead to improvement in the potency and selectivity of such agents when medicinal chemistry efforts can adopt a structural-based approach to guide their efforts.

3.4 CRITERIA FOR SELECTION OF TARGETS AND ESTABLISHING SCREENS Although many drugs targeting ion channels have been developed in the past, historically, it has been difficult to find a new generation of drugs for this class of proteins using molecular approaches. There are several reasons why this situation exists. Target validation is crucial; systems physiology is complex and the precise role of a given candidate protein may be unclear unless studied under certain defined conditions. In addition, channel complements can differ across species leading to difficulties in translating animal pharmacology models to human clinical pathophysiology. Lead identification is challenging; in the past, ion channels have been difficult targets to screen functionally. Typically, very few potent and selective hits are identified in library screening, and it is very important to have a robust high-throughput screening strategy that is able to detect a ‘‘needle in a haystack.’’ Finally, identification of the best lead is key because without adequate initial potency and selectivity across the ion channel superfamilies, experience has shown that it is very difficult to optimize a poor lead structure into a viable drug candidate. In this respect, the mechanism of drug block is of paramount importance, and the ability of a compound to display useful mechanisms of action that allow optimization of the principle of functional selectivity (e.g., by modifying parameters of a compound’s action such as state dependence and/or use dependence in its interaction with the target channel) has been shown to be important in order to increase

CRITERIA FOR SELECTION OF TARGETS AND ESTABLISHING SCREENS

69

the therapeutic index (Scholz, 2002; Kyle and Ilyin 2007). Therefore, it is critical to implement the best screens and counterscreens at the beginning of the discovery process. Strategies to obtain proof of concept could involve genetic validation (e.g., construction of animal models with ablation or overexpression of candidate genes or analyzing consequences of human mutations), modulation of channel expression through molecular biological approaches (e.g., regulating promotor activity, using siRNA techniques, or employing dominant negative interference), or pharmacological validation. Indeed, the most meaningful target validation comes from obtaining pharmacological proof of concept. In this way, one could take the approach of improving on existing ion channel drugs with proven efficacy by enhancing potency/selectivity to eliminate non-mechanism-based side effects, or by optimizing drug metabolism/pharmacokinetic properties of a new agent. Or, as another approach to achieving pharmacological POC, peptide channel blockers could be employed to validate an ion channel target. For example, investigation of Ziconotide, a synthetic analogue of a naturally occurring peptide constituent of a venom isolated from a cone snail, and a potent blocker of the N-type voltagegated calcium channel, Cav2.2, is an excellent example of using a peptide to validate an ion channel target (Lewis and Garcia, 2003). In this case, Ziconotide was in fact developed clinically as a drug for treating intractable pain and is approved for use with an intrathecal route of peptide administration. Results of the Ziconotide effort strongly support the hypothesis that a small-molecule blocker of Cav2.2 administered systemically should also be therapeutically useful in pain management. Other examples of using K channel blocking peptides for obtaining proof of concept (e.g., with Kv1.3 for immunosuppression, with KCa3 for sickle cell anemia, or with Kv2.1 for diabetes) have been alluded to above. In all such studies, it is critical to insure that target validation translates across species to man. Once an ion channel is selected as a target for therapeutic intervention, it is important to establish robust and mechanistically oriented functional high-throughput screening assays to identify leads for medicinal chemistry optimization. Although ligand binding screens have been used extensively in the past, and are still used to support some ion channel selectivity assays (e.g., monitoring binding of Ca entry blockers to detect L-type calcium channel modulators, or following binding of batrachotoxin to identify compounds with sodium channel activity), the most informative assays for lead generation are functional screening paradigms. These can be established using one of the three different approaches: direct biochemical channel assays, indirect biochemical channel assays, or direct electrophysiology-based channel assays. The first approach is based on isotopic, or, more recently, on nonisotopic flux measurements, or on cellular imaging to measure movement of the candidate ion (e.g., Kþ , Naþ , Hþ, Ca2þ , or Cl
) directly through the channel of interest. Radioactive tracers or atomic absorption spectroscopy can be employed to monitor ion fluxes, while calcium-or pH-sensitive dyes record changes in intracellular calcium or hydrogen ion levels, respectively, to detect movement of these ions. Such assays are typically of very high capacity and can be established in a high-density format, depending on the probe that is utilized. In this approach, the trigger used to open the

70

ION CHANNELS GENE FAMILY

channel is important; high potassium is typically employed to depolarize the plasma membrane, thereby triggering channel opening, and when the system has been configured by cotransfection with an inward rectifier potassium channel, the application of different potassium levels can be used to clamp the membrane potential as desired. Indirect biochemical ion channel assays typically monitor plasma membrane potential, and the assays are configured in such a way that activity of the target channel directly affects this parameter by causing either cell depolarization or hyperpolarization (Garcia and Kaczorowski, 2006). Such assays can be of very high capacity by using high-density format plates with fluorescent probes; there are multiple fluorescence plate reader platforms to choose from by which to support this type of screening, depending on the capabilities desired. Several types of membrane potential dyes are available (dye pairs that depend on fluorescence resonance energy transfer readouts, no wash dyes, etc.), and it is important to select the best dye for the assay’s purpose. Again, the trigger used to initiate the assay is important (Felix et al., 2004; Liu et al., 2006), and novel electrical field stimulation techniques have recently been developed to broaden the implementation of this assay technology (Bugianesi et al., 2006; Huang et al., 2006). More important, these types of cellular membrane potential-based assays can be configured to detect either channel blockers or agonists. Finally, medium-throughput, direct electrophysiological ion channel assays using patch voltage clamp measurements have recently become feasible by employing new automated patch clamp platforms (Herrington et al., 2005). Such techniques are the most sensitive monitor of ion channel activity and are very versatile because they allow a wide range of voltage clamp protocols to be employed in a screening format mode. This approach is also ideal for rapid determination of mechanism of action of a compound. At present, the current platforms, which use formats of 16, 48, or 356 wells, are not suitable for high-throughput screening campaigns, but they are excellent for characterizing primary screening actives, ion channel counterscreening, directed library screening, or supporting medicinal chemistry. Together, these approaches provide a powerful new paradigm for identifying potent and selective ion channel modulator leads and for supporting medicinal chemistry efforts to optimize those leads into drug development candidates.

3.5

A CASE STUDY IN ION CHANNEL DRUG DISCOVERY

Modulation of calcium signaling in T lymphocytes is commonly accepted as a mechanism by which to achieve immune regulation (Chandy et al., 2006). Blocking increases in intracellular calcium that occur after T cell receptor stimulation suppresses T cell activation in vitro. It has been shown that function of the voltagegated K channel, Kv1.3, is required to support human T cell activation; Kv1.3 sets the resting potential in human T cells, providing a driving force for calcium entry, and peptidyl blockers of this channel (margatoxin, MgTX; Stichodactyla helianthus peptide, ShK) promote cell depolarization, thereby preventing increases

A CASE STUDY IN ION CHANNEL DRUG DISCOVERY

71

FIGURE 3.5 Human T cell activation. Sustained Ca2þ influx through Ca2þ -releaseactivated calcium (CRAC) channels is required for lymphokine release and T cell proliferation. Kv1.3 channels hyperpolarize the membrane and facilitate Ca2þ entry through CRAC channels. Blockade of Kv1.3 with peptides or small molecules prevents T cell activation in vitro and in vivo. (See the color version of this figure in the Color Plates section.)

in intracellular calcium upon T cell stimulation. This leads to a block of cytokine production that prevents T cell proliferation as summarized in Fig. 3.5. Early on, it was recognized that there were species differences among T cells in the role played by Kv1.3 and that rodent models were unsuitable for validating Kv1.3 as a target for immunomodulation. However, further validation of this target was achieved from studies with miniswine, a model that has been used before as a surrogate for man. In this case, the role of Kv1.3 in peripheral T cells from the two species was demonstrated to be identical. Using infusion of MgTX into mini-swine, it was shown that blockade of Kv1.3 inhibited both a delayed-type hypersensitivity (DTH) response to tuberculin and an antibody response to an allogenic challenge, as well as reduced thymic cellularity and thymic development of T cell subsets in this species (Koo et al., 1997). These results demonstrate that in vivo blockade of Kv1.3 in an appropriate model can inhibit immune responses. A CHO cell line was stably transfected with hKv1.3 to establish a 86Rbþ flux screen, a direct biochemical assay, to search both natural product and synthetic small-molecule libraries for Kv1.3 channel blockers. The screening protocol is summarized in Fig. 3.6. Out of approximately 100,000 natural products tested in one campaign, a novel nortriterpene, termed correolide and shown in Fig. 3.7, was identified and purified from extracts of the tree Spachea correae (Felix et al., 1999). Correolide inhibits Rbþ flux through Kv1.3 with an IC50 value of 85 nM and a Hill coefficient of 1, and correolide analogues demonstrate a defined structure–activity relationship (SAR) in their blocking ability. Although initially this compound

72

ION CHANNELS GENE FAMILY

FIGURE 3.6 Scheme of a functional Rbþ efflux assay used for identifying potassium channel modulators. (See the color version of this figure in the Color Plates section.)

FIGURE 3.7

Two structural classes of small-molecule Kv1.3 blockers.

A CASE STUDY IN ION CHANNEL DRUG DISCOVERY

73

may structurally appear to be membrane active, it does not affect membrane permeability, blocks the channel in electrophysiological experiments with a defined mechanism of action requiring the channel to open before block can occur, and displays marked selectivity against the many receptors and other voltage- and ligandgated ion channels that it was tested against. However, although correolide is most potent as a Kv1.3 inhibitor, it blocks all other members of the Kv1 family, with 5–15-fold lower potency, and it also blocks the Kv2 family of channels. Correolide was radiolabeled and used as a ligand for further study of Kv1.3 and other members of the Kv1 family present in brain. [3H]Dihydrocorreolide (DiTC) binds to Kv1.3 in a specific, saturable, and reversible fashion (Kd of 11 nM) with a stoichiometry of 1:1 per channel tetramer and because of these properties, it has been used to map the drug binding site on the channel. Further validation of correolide’s specificity was shown by the ability of this compound to depolarize human T cells to the same extent as that caused by Kv1.3 peptide blockers, suggesting that it was a candidate for development as an immunosuppressant. Correolide was the first potent small-molecule inhibitor of the Kv1 family of channels to be discovered from a natural product source. A significant medicinal chemistry effort was undertaken with correolide because of the complexity and limited supply of the molecule, with the goal of preparing a completely synthetic analogue with a simplified structure, and 11 patents have appeared describing the results of this work. An initial approach was to simplify the pentacyclic natural product via removal of the E-ring to generate an enone; subsequent modification led to a new series of tetracyclic Kv1.3 inhibitors. Using the Rbþ flux functional assay and human T cell proliferation assay to determine SAR, a correolide analogue, correolide A (Fig. 3.7), was identified that was 15-fold more potent as an immunosuppressant than the parent compound (Bao et al., 2005). Other synthetic approaches were designed to remove the acetyl moieties that line one side of the molecule and to simplify the A-ring lactone. The A-ring was converted to an oxa-pene without loss of activity. Although most of the acetyl substitutions could be removed without causing substantial loss in Kv1.3 blocking potency, removal of the C-4 acetyl group led to a marked decrease in channel inhibitory activity. However, substitution of the exocyclic acetyl group at C-4 with an ortho-bromobenzyl group, or removal of the carbonyl group in this linkage to give the analogous ether in the oxa-pene series, resulted in potent Kv1.3 blockers that had in vivo activity (Fig. 3.7; see discussion below) (Koo et al., 1999). In the end, it was possible to simplify the structure of correolide as described in the patent literature to a tricyclic template that could be produced by total synthesis and identify analogues with good in vivo efficacy, but potency enhancement was limited, and this chemical series turned out to be most valuable as reagents for in vitro drug mapping studies and for obtaining further proof of concept in vivo. The correolide binding domain was mapped on Kv1.3 through a site-directed mutagenesis approach. Initially, using DiTC as a probe, it was found that high-affinity ligand binding could be conferred to the correolide-insensitive Kv3.2 channel after substitution of three nonconserved amino acids in the S5 and S6 helices of this channel with the corresponding residues found in Kv1.3 (Hanner et al., 1999). Because of this finding, site-directed mutagenesis was undertaken along S5 and

74

ION CHANNELS GENE FAMILY

S6 of Kv1.3 to determine whether in this region drug binding occurs (Hanner et al., 2001). Control ligand binding studies were performed with an iodinated peptide that bound to the pore of Kv1.3 to ensure high levels of channel expression and proper folding of the multiple mutated proteins that were constructed and transiently expressed, while DiTC binding was the probe for the drug–channel interaction. Two residues in S5 and five residues in S6 were found to have the most profound effect on DiTC binding; several of these critical S6 residues are known to be associated with conformational changes that occur during channel gating. Based on these data, DiTC was docked in Kv1.3 using molecular modeling of the S5–S6 region with extrapolation from the coordinates of the KcsA channel structure (Hanner et al., 2001). Correolide appears to bind in the water-filled cavity below the selectivity filter to a hydrophobic pocket formed by specific amino acid side chains from S6 residues. High-affinity binding is likely to be due to the complementary nature between the bowl-like shape of the channel cavity and the shape of correolide, and be driven primarily by hydrophobic interactions. The conformational change that occurs during gating in this Kv1.3 region is consistent with the known state-dependent blocking properties of correolide. In addition, the stoichiometry of drug binding, the lack of H-bond interactions with the channel as denoted by the correolide SAR, and the decrease in affinity observed after removal of the C-4 acetyl group that eliminates a favorable van der Waals interaction with the channel are all consistent with the proposed docking scheme. Moreover, the orientation of correolide in the cavity as shown in Fig. 3.8, based on analysis of C-4 substituted derivatives, places the 3keto group of the E-ring ester pointing toward the selectivity filter, with the saturated hydrocarbon side of the molecule interacting with the wall of the channel, and the other side of the molecule containing the acetyl groups lying in the water filled cavity; the two S5 residues that affect DiTC binding may do so by disrupting the helical packing between S5 and S6 and altering the shape of the cavity. Correolide and some of its analogues, correolide B and correolide C, have been used to further examine the role of Kv1.3 in human and mini-swine T cells in both in vitro and in vivo studies (Koo et al., 1999). After demonstrating by electrophysiology techniques that correolide and two C-4 substituted analogues are potent blockers of Kv1.3 channels in T cells from each species, it was also shown that this series had no effect on KCa3, a distinct lymphocyte calcium-activated K channel that has also been linked to controlling T cell activation. This important specificity control allows more precise definition of the physiological role of Kv1.3 in T cells through use of such pharmacological probes. In a series of experiments with human T cells, these Kv1.3 inhibitors blocked stimulated increases in intracellular Ca levels, PMA/ionomycin-induced IL-2 production (but not cytokine production stimulated through the CD28, non-calcium-dependent pathway), and anti-CD3-mediated T-cell proliferation. The latter process was reversed by addition of a cytokine cocktail, ruling out general cytotoxicity, which is another measure of the specificity of correolide’s action. This in vitro profile is identical to that produced by the peptide, MgTX, and has become the hallmark signature of a selective Kv1.3 blocker. Subsequent in vivo experiments tested the effect of the two C-4 substituted correolide analogues on a DTH response to tuberculin in mini-swine. When dosed by oral i.v. or i.m.

A CASE STUDY IN ION CHANNEL DRUG DISCOVERY

75

FIGURE 3.8 Docking of correolide C in Kv1.3. A model of Kv1.3 with a bent S6 helix was generated using the crystal structure of the KcsA channel as a template and modified using experimental data constraints. In the displayed orientation, the 3-keto group of the E-ring ester points to the selectivity filter, the saturated hydrocarbon face of the molecule interacts with the hydrophobic wall of the channel, and the other face, with four acetyl groups, lies in the waterfilled cavity. The bromobenzyl group provides binging energy through van de Waals interaction with Pro425 in the channel. (See the color version of this figure in the Color Plates section.)

routes of administration, both compounds blocked the DTH response, just as had been found previously with MgTX, and did so to the same degree as elicited by FK506, a well-known immunosuppressant. These results confirm that Kv1.3 blockers act as novel immunosuppressant in vivo. Little toxicity was noted during the in vivo mini-swine studies with correolide analogues; cardiovascular parameters, body temperature, blood chemistry and cellular profiles of treated animals remained normal during the study. However, some treated animals experienced intestinal distress during dosing, and these symptoms reversed when treatment was discontinued. Following up on these observations, in vitro studies demonstrate that correolide elicits twitches in guinea pig ileum by stimulating the enteric nervous system and enhancing neurotransmitter release (Vianna-Jorge et al., 2000). The effect was not seen in other guinea pig smooth muscle tissues, or in rat or mouse ileum. Correolide-induced twitching was blocked by atropine, which was also effective in treating the symptoms of diarrhea in the miniswine, as well as by antagonists of NK1 and NK2 tachykinin receptors. It was hypothesized that blockade of Kv1 channels by correolide increases the excitability of intramural nerve plexuses, promoting release of acetylcholine and tachykinins from excitatory motor neurons, which in turn leads to Ca-dependent action potentials and twitching of muscle fibers. A subsequent study confirmed that Kv1 family

76

ION CHANNELS GENE FAMILY

channel blockers, both peptides and small molecules, increase the peristaltic activity of guinea pig ileum by stimulating acetylcholine and tachykinin release from the enteric nervous system and that block of Kv1.1, in particular, is associated with this pathophysiology (Vianna-Jorge et al., 2003). These data, when considered together with the potential neurotoxicity that might result due to the high prevalence of Kv1.1 and Kv1.2 channels in the brain, suggest the importance of developing blockers that are selective for the Kv1.3 channel. A small campaign that assayed approximately 100,000 synthetic compounds was also preformed using the 86Rbþ flux screen to search for novel small-molecule Kv1.3 blockers. One such class of high-affinity inhibitors that was discovered is denoted by the initial hit, 4-phenyl-4-[3-(2-methoxyphenyl)-3-oxo-2-azaprop-1-yl] cyclohexanone (PAC; Fig. 3.7), which is representative of a disubstituted-cyclohexyl (DSC) template (Schmalhofer et al., 2002). PAC blocks Kv1.3 in the functional Rbþ flux assay and in voltage clamp experiments with human T cells, with an IC50 determination of 300 nM and a Hill coefficient of 2, and it inhibits DiTC binding to Kv1 family channels and suppresses the calcium-dependent pathway of T cell activation in in vitro assays. A Hill coefficient greater than 1 is atypical for an ion channel blocker but is characteristic of this series of compounds (see below). PAC displays very good selectivity as it was found to block only members of the Kv1 family but not affect many other ion channels, receptors, or enzyme systems that were tested. A medicinal chemistry effort produced a series of benzamide analogues that helped further define the features of this series, and compounds were identified that were modestly more active in the Rbþ flux and T cell proliferation assays than the parent compound (Miao et al., 2003). These efforts resulted in two issued patents. Block of Kv1.3 was found to display a defined SAR, indicating specificity for interaction between this chemical series and the channel. Substitution at the C-1 ketone of PAC generates cis and trans isomer pairs. While the activity of the cis analogues is not sensitive to the nature of the C-1 substituent, the corresponding trans isomers display a wide range of activities. Whereas many of the DSC analogues do not display selectivity in interacting with members of the Kv1 channel family, the trans analogues appear to distinguish between Kv1.x channels based on their rates of C-type inactivation (Schmalhofer et al., 2002). Together, these results demonstrate that DSC derivatives represent a new class of small-molecule Kv1.3 blockers with immunosuppressant activity and that it may be possible to exploit the selective interaction of trans isomer derivatives with channel conformations related to C-type inactivation to develop Kv1.3 selective blockers that would be useful for the safe treatment of autoimmune diseases. To further explore the unusual >1 Hill coefficient observed in functional assays with the DSC series and the interaction of trans analogues with Kv1.3 (i.e., the potential Kv1 selectivity enhancement linked to conformational changes associated with C-type channel inactivation), trans-1-(N-n-propylcarbamoyloxy-)-4-phenyl4-(3-(2-methoxyphenyl)-3-oxo-2-azapropyl-1-yl)cyclohexane (trans-NPCO-DSC; Fig. 3.7) was radiolabeled with 3H and a binding assay was established with Kv1.3 channels (Schmalhofer et al., 2003). Trans-NPCO-DSC binds specifically to Kv1.3 channels in a saturable, time-dependent, and fully reversible manner.

PERSPECTIVE ON ION CHANNELS AS DRUG TARGETS

77

Saturation binding isotherms and competition binding experiments support the idea of two binding sites for the DSC structural series on Kv1.3 that display positive allosteric cooperativity. This is the first demonstration of the existence of multiple binding sites for a single inhibitor on a voltage-gated ion channel. There is also a correlation between the high-affinity interaction of trans-NPCO-DSC and the rate of C-type inactivation, and this predicts a mechanism by which trans isomers would display specificity for Kv1.3 over other Kv1.x channels that do not C-type inactivate. This hypothesis was validated in binding experiments with trans-NPCO-DSC to Kv1.2 and Kv1.5 channels. Together, these results suggest that further work on the DSC analogue series might have resulted in Kv1.3 subtype selective inhibitors that could be developed for therapeutic utility. Unfortunately, the DSC series displayed poor drug metabolism/pharmacokinetic properties, and given the lack of a small animal model to guide development efforts for Kv1.3 blockers, and the fact that a suitable compound was not identified to test the program’s hypothesis in an in vivo graft rejection model, this drug development program was not continued. However, invaluable lessons were learned from this effort in terms of what is required for successful drug discovery and development on an ion channel target, and efforts in this direction have been strengthened and continue on other targets.

3.6

PERSPECTIVE ON ION CHANNELS AS DRUG TARGETS

A new era in ion channel drug discovery has begun. Almost all of the ion channel families have been identified, cloned, and expressed functionally. Human genetics has revealed important new targets among these families, as well as confirmed the importance of previously identified targets. The pathways that regulate ion channel activity are being defined, and the roles that these proteins play in complex physiological processes are becoming clearer. There are an abundance of selective ion channel modulators, including peptides, natural products, and small molecules, that may be used as probes to obtain proof of concept for the ion channel target in question, and more are being revealed through research efforts in the community. High-throughput screening strategies using functional ion channel screens have become more sophisticated and allow the testing of very large synthetic libraries in a short period of time. Follow-up secondary screenings to determine specificity and mechanism of action have become routine using automated techniques, including electrophysiology. With the advent of high-resolution ion channel structures, it is possible to understand ion permeation and channel gating; soon co-crystallization of ion channels with their modulators will provide a means to test whether docking models with small molecules and peptides are correct and will inform medicinal chemistry as to whether more rational drug design approaches are possible with these targets. Yet, there are still important features of ion channel drug development that must be improved upon. The discovery of high-quality leads for medicinal chemistry during HTS is still akin to finding ‘‘a needle in a haystack.’’ The design of small-molecule libraries with potential channel modulatory activity for directed screening efforts is a

78

ION CHANNELS GENE FAMILY

current challenge. The discovery of agents that selectively modify channel gating, rather than act by simply plugging the channel pore to interfere with conduction, is another important area for future research. Finally, as a critical aid in successful clinical development, the ability to ascertain target engagement, either by directly monitoring the interaction of small molecules with the target channel, or by having a direct readout of channel activity, so as to establish a pharmacodynamic relationship for modulator activity, is the final step required for bringing drug development efforts on this target class to fruition. With such new tools, the medical benefits demonstrated by drugs from the early era of ion channel drug discovery and development can be realized with new generations of therapeutic agents.

REFERENCES Ashcroft F, 2000. Ion Channels and Disease. Academic Press. Bao J, Miao S, Kayser F, Kotliar AJ, Baker RK, Doss GA, Felix JP, Bugianesi RM, Slaughter RS, Kaczorowski GJ, Garcia ML, Ha SN, Castonguay L, Koo GC, Shah K, Springer MS, Staruch MJ, Parsons WH, Rupprecht KM, 2005. Potent Kv1.3 inhibitors from correolidemodification of the C18 position. Bioorg. Med. Chem. Lett. 15:447–451. Beeton C, Wulff H, Barbaria J, Clot-Faybesse O, Pennington M, Bernard D, Cahalan MD, Chandy KG, Beraud E, 2001. Selective blockade of T lymphocyte K(þ) channels ameliorates experimental autoimmune encephalomyelitis, a model for multiple sclerosis, Proc. Natl. Acad. Sci. USA 98:13942–13947. Beeton C, Pennington MW, Wulff H, Singh S, Nugent D, Crossley G, Khaytin I, Calabresi PA, Chen CY, Gutman GA, Chandy KG, 2005. Targeting effector memory T cells with a selective peptide inhibitor of Kv1.3 channels for therapy of autoimmune diseases. Mol. Pharmacol. 67:1369–1381. Beeton C, Wulff H, Standifer NE, Azam P, Mullen KM, Pennington MW, Kolski-Andreaco A, Wei E, Grino A, Counts DR, Wang PH, LeeHealey CJ, S Andrews B, Sankaranarayanan A, Homerick D, Roeck WW, Tehranzadeh J, Stanhope KL, Zimin P, Havel PJ, Griffey S, Knaus HG, Nepom GT, Gutman GA, Calabresi PA, Chandy KG, 2006. Kv1.3 channels are a therapeutic target for T cell-mediated autoimmune diseases, Proc. Natl. Acad. Sci. USA 103:17414–17419. Begenisich T, Nakamoto T, Ovitt CE, Nehrke K, Brugnara C, Alper SL, Melvin JE, 2004. Physiological roles of the intermediate conductance, Ca2þ -activated potassium channel Kcnn4. J. Biol. Chem. 279:47681–47687. Brenner R, Chen QH, Vilaythong A, Toney GM, Noebels JL, Aldrich RW, 2005. BK channel beta4 subunit reduces dentate gyrus excitability and protects against temporal lobe seizures. Nat. Neurosci. 8:1752–1759. Bugianesi RM, Augustine PR, Azer K, Dufresne C, Herrington J, Kath GS, McManus OB, Napolitano CS, Rush A, Sachs J, Simpson N, Wismer MK, Kaczorowski GJ, Slaughter RS, 2006. A cell-sparing electric field stimulation technique for high-throughput screening of voltage-gated ion channels. Assay Drug Dev. Technol. 4:21–35. Chandy KG, Wulff H, Beeton C, Pennington M, Gutman GA, Cahalan MD, 2004. Kþ channels as targets for specific immunomodulation. Trends Pharmacol. Sci. 25:280–289.

REFERENCES

79

Chandy KG, Wulff H, Beeton C, Calabresi PA, Gutman GA, Pennington M, 2006. In: Triggle D, editor. Voltage-Gated Ion Channels as Drug Targets. Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. Coghlan MJ, Carroll WA, Gopalakrishnan M, 2001. Recent developments in the biology and medicinal chemistry of potassium channel modulators: update from a decade of progress. J. Med. Chem. 44:1627–1653. Crane A, Aguilar-Bryan L, 2004. Assembly, maturation, and turnover of K(ATP) channel subunits. J. Biol. Chem. 279:9080–9090. Dhamoon AS, Jalife J, 2005. The inward rectifier current (IK1) controls cardiac excitability and is involved in arrhythmogenesis. Heart Rhythm 2:316–324. Dobrev D, Friedrich A, Voigt N, Jost N, Wettwer E, Christ T, Knaut M, Ravens U, 2005. The G protein-gated potassium current I(K,ACh) is constitutively active in patients with chronic atrial fibrillation. Circulation 112:3697–3706. Dolly JO, Rettig J, Scott VE, Parcej DN, Wittkat R, Sewing S, Pongs O, 1994. Oligomeric and subunit structures of neuronal voltage-sensitive Kþ channels. Biochem. Soc. Trans. 22:473–478. Dong J, Shi N, Berke I, Chen L, Jiang Y, 2005. Structures of the MthK RCK domain and the effect of Ca2þ on gating ring stability. J. Biol. Chem. 280:41716–41724. Doyle DA, Morais Cabral J, Pfuetzner RA, Kuo A, Gulbis JM, Cohen SL, Chait BT, MacKinnon R, 1998. The structure of the potassium channel: molecular basis of Kþ conduction and selectivity. Science 280:69–77. Felix JP, Bugianesi RM, Schmalhofer WA, Borris R, Goetz MA, Hensens OD, Bao JM, Kayser F, Parsons WH, Rupprecht K, Garcia ML, Kaczorowski GJ, Slaughter RS, 1999. Identification and biochemical characterization of a novel nortriterpene inhibitor of the human lymphocyte voltage-gated potassium channel, Kv1.3. Biochemistry 38:4922–4930. Felix JP, Williams BS, Priest BT, Brochu RM, Dick IE, Warren VA, Yan L, Slaughter RS, Kaczorowski GJ, Smith MM, Garcia ML, 2004. Functional assay of voltage-gated sodium channels using membrane potential-sensitive dyes. Assay Drug Dev. Technol. 2:260–268. Fermini B, Fossa AA, 2003. The impact of drug-induced QT interval prolongation on drug discovery and development. Nat. Rev. Drug. Discov. 2:439–447. Ficker E, Taglialatela, M, Wible BA, Henley CM, Brown AM, 1994. Spermine and spermidine as gating molecules for inward rectifier Kþ channels. Science 266:1068–1072. Garcia ML, Kaczorowski GJ, 2001. In: Rusch AA, editor. Potassium Channels in Cardiovascular Biology., Academic/Plenum Publishers, New York. Garcia ML, Kaczorowski, GJ, 2005. Potassium channels as targets for therapeutic intervention, Sci. STKE 2005(302):pe46. Garcia ML and Kaczorowski G, 2006. Critical success. Eur. Pharm. Rev. 11:42–45. Garcia ML, Shen D-M, Kaczorowski GJ, 2007. High-conductance calcium-activated potassium channels: validated targets for smooth muscle relaxants? Expert Opin. Ther. Pat. 17:1–12. Gross A, MacKinnon R, 1996. Agitoxin footprinting the shaker potassium channel pore. Neuron 16:399–406. Haider S, Khalid S, Tucker SJ, Ashcroft FM, Sansom MS, 2007. Molecular dynamics simulations of inwardly rectifying (Kir) potassium channels: a comparative study. Biochemistry 46:3643–3652.

80

ION CHANNELS GENE FAMILY

Hanner M, Schmalhofer WA, Green B, Bordallo C, Liu J, Slaughter RS, Kaczorowski GJ, Garcia ML, 1999. Binding of correolide to K(v)1 family potassium channels. Mapping the domains of high affinity interaction. J. Biol. Chem. 274:25237–25244. Hanner M, Green B, Gao YD, Schmalhofer WA, Matyskiela M, Durand DJ, Felix JP, Linde AR, Bordallo C, Kaczorowski GJ, Kohler M, Garcia ML, 2001. Binding of correolide to the K(v)1.3 potassium channel: characterization of the binding domain by site-directed mutagenesis. Biochemistry 40:11687–11697. Heginbotham L, MacKinnon R, 1992. The aromatic binding site for tetraethylammonium ion on potassium channels. Neuron 8:483–491. Herrington J, McManus O, Kiss L, 2005. The road ahead and how to get there. Eur. Pharm. Rev. 10:79–84. Herrington J, Zhou YP, Bugianesi RM, Dulski PM, Feng Y, Warren VA, Smith MM, Kohler MG, Garsky VM, Sanchez M, Wagner M, Raphaelli K, Banerjee P, Ahaghotu C, Wunderler D, Priest BT, Mehl JT, Garcia ML, McManus OB, Kaczorowski GJ, Slaughter RS, 2006. Blockers of the delayed-rectifier potassium current in pancreatic beta-cells enhance glucose-dependent insulin secretion. Diabetes 55:1034–1042. Hille B, 2001. Ion Channels of Excitable Membranes, 3rd ed. Sinauer Associates, Inc., Sunderland, MA. Hoshi T, Zagotta WN, Aldrich RW, 1990. Biophysical and molecular mechanisms of Shaker potassium channel inactivation. Science 250:533–538. Huang CJ, Harootunian A, Maher MP, Quan C, Raj CD, McCormack K, Numann R, Negulescu PA, Gonzalez JE, 2006. Characterization of voltage-gated sodium-channel blockers by electrical stimulation and fluorescence detection of membrane potential. Nat Biotechnol 24:439–446. Jiang Y, Pico A, Cadene M, Chait BT, MacKinnon R, 2001. Structure of the RCK domain from the E. coli Kþ channel and demonstration of its presence in the human BK channel. Neuron 29:593–601. Jiang Y, Lee A, Chen J, Cadene M, Chait BT, MacKinnon R, 2002. Crystal structure and mechanism of a calcium-gated potassium channel. Nature 417:515–522. Jiang Y, Lee A, Chen J, Ruta V, Cadene M, Chait BT, MacKinnon R, 2003. X-ray structure of a voltage-dependent Kþ channel. Nature 423:33–41. Scholz A, 2002. Mechanisms of (local) anaesthetics on voltage-gated sodium and other ion channels, Br. J. Anaesth. 89:52–61. Jin W, Lu Z, 1998. A novel high-affinity inhibitor for inward-rectifier Kþ channels. Biochemistry 37:13291–13299. Koo GC, Blake JT, Talento A, Nguyen M, Lin S, Sirotina A, Shah K, Mulvany K, Hora Jr D, Cunningham P, Wunderler DL, McManus OB, Slaughter R, Bugianesi R, Felix J, Garcia M, Williamson J, Kaczorowski G, Sigal NH, Springer MS, Feeney W, 1997. Blockade of the voltage-gated potassium channel Kv1.3 inhibits immune responses in vivo. J. Immunol. 158:5120–5128. Koo GC, Blake JT, Shah K, Staruch MJ, Dumont F, Wunderler D, Sanchez M, McManus OB, Sirotina-Meisher A, Fischer P, Boltz RC, Goetz MA, Baker R, Bao J, Kayser F, Rupprecht KM, Parsons WH, Tong XC, Ita IE, Pivnichny J, Vincent S, Cunningham P, Hora Jr D, Feeney W, Kaczorowski G, Springer MS, 1999. Correolide and derivatives are novel immunosuppressants blocking the lymphocyte Kv1.3 potassium channels. Cell Immunol. 197:99–107.

REFERENCES

81

Kyle DJ, Ilyin VI, 2007. Sodium channel blockers. J. Med. Chem. 50:2583–2588. Ledoux J, Werner ME, Brayden JE, Nelson MT, 2006. Calcium-activated potassium channels and the regulation of vascular tone. Physiology (Bethesda) 21:69–78. Lee SY, MacKinnon R, 2004. A membrane-access mechanism of ion channel inhibition by voltage sensor toxins from spider venom. Nature 430:232–235. Lewis RJ, Garcia ML, 2003. Therapeutic potential of venom peptides. Nat. Rev. Drug Discov. 2:790–802. Liu Y, Jurman ME, Yellen G, 1996. Dynamic rearrangement of the outer mouth of a Kþ channel during gating. Neuron 16:859–867. Liu CJ, Priest BT, Bugianesi RM, Dulski PM, Felix JP, Dick IE, Brochu RM, Knaus HG, Middleton RE, Kaczorowski GJ, Slaughter RS, Garcia ML, Kohler MG, 2006. A highcapacity membrane potential FRET-based assay for Nav1.8 channels. Assay Drug Dev. Technol. 4:37–48. Long SB, Campbell EB, MacKinnon R, 2005a. Crystal structure of a mammalian voltagedependent Shaker family Kþ channel. Science 309:897–903. Long SB, Campbell EB, MacKinnon R, 2005b. Voltage sensor of Kv1.2: structural basis of electromechanical coupling. Science 309:903–908. Lopatin AN, Makhina EN, Nichols CG, 1995. The mechanism of inward rectification of potassium channels: ‘‘long-pore plugging’’ by cytoplasmic polyamines. J. Gen. Physiol. 106:923–955. Mannhold R, 2004. KATP channel openers: structure–activity relationships and therapeutic potential. Med. Res. Rev. 24:213–266. Matsuda H, Saigusa, A, Irisawa, H, 1987. Ohmic conductance through the inwardly rectifying K channel and blocking by internal Mg2þ. Nature 325:156–159. Miao S, Bao J, Garcia ML, Goulet JL, Hong XJ, Kaczorowski GJ, Kayser F, Koo GC, Kotliar A, Schmalhofer WA, Shah K, Sinclair PJ, Slaughter RS, Springer MS, Staruch MJ, Tsou NN, Wong F, Parsons WH, Rupprecht KM, 2003. Benzamide derivatives as blockers of Kv1.3 ion channel. Bioorg. Med. Chem. Lett. 13:1161–1164. Morais-Cabral JH, Zhou Y, MacKinnon R, 2001. Energetic optimization of ion conduction rate by the Kþ selectivity filter. Nature 414:37–42. Nishida M, MacKinnon R, 2002. Structural basis of inward rectification: cytoplasmic pore of the G protein-gated inward rectifier GIRK1 at 1.8 A resolution. Cell 111:957–965. Niu X, Qian X, Magleby KL, 2004. Linker-gating ring complex as passive spring and Ca(2þ)dependent machine for a voltage- and Ca(2þ)-activated potassium channel. Neuron 42:745–756. Orio P, Rojas P, Ferreira G, Latorre R, 2002. New disguises for an old channel: MaxiK channel beta-subunits. News Physiol. Sci. 17:156–161. Panyi G, Possani LD, Rodriguez de la Vega RC, Gaspar R., Varga Z, 2006. Kþ channel blockers: novel tools to inhibit T cell activation leading to specific immunosuppression. Curr. Pharm. Des. 12:2199–2220. Ranganathan R, Lewis JH, MacKinnon R, 1996. Spatial localization of the Kþ channel selectivity filter by mutant cycle-based structure analysis. Neuron 16:131–139. Rettig J, Heinemann SH, Wunder F, Lorra C, Parcej DN, Dolly JO, Pongs O, 1994. Inactivation properties of voltage-gated Kþ channels altered by presence of beta-subunit. Nature 369:289–294.

82

ION CHANNELS GENE FAMILY

Ruta, V, MacKinnon R, 2004. Localization of the voltage-sensor toxin receptor on KvAP. Biochemistry 43:10071–10079. Salkoff L, 2006. A tail of multiple calcium-sensing domains. J. Gen. Physiol. 128:387–388. Salkoff L, Butler A, Ferreira G, Santi C, Wei A, 2006. High-conductance potassium channels of the SLO family. Nat. Rev. Neurosci. 7:921–931. Schmalhofer WA, Bao J, McManus OB, Green B, Matyskiela M, Wunderler D, Bugianesi RM, Felix JP, Hanner M, Linde Arias AR, Ponte CG, Velasco L, Koo G, Staruch MJ, Miao S, Parsons WH, Rupprecht K, Slaughter RS, Kaczorowski GJ, Garcia ML, 2002. Identification of a new class of inhibitors of the voltage-gated potassium channel, Kv1.3, with immunosuppressant properties. Biochemistry 41:7781–7794. Schmalhofer WA, Slaughter RS, Matyskiela M, Felix JP, Tang YS, Rupprecht K, Kaczorowski GJ, Garcia ML, 2003. Di-substituted cyclohexyl derivatives bind to two identical sites with positive cooperativity on the voltage-gated potassium channel, K(v)1.3. Biochemistry 42:4733–4743. Schmidt D, Jiang QX, MacKinnon R, 2006. Phospholipids and the origin of cationic gating charges in voltage sensors. Nature 444:775–779. Schumacher MA, Rivard AF, Bachinger HP, Adelman JP, 2001. Structure of the gating domain of a Ca2þ-activated Kþ channel complexed with Ca2þ/calmodulin. Nature 410: 1120–1124. Shieh CC, Coghlan M, Sullivan JP, Gopalakrishnan M, 2000. Potassium channels: molecular defects, diseases, and therapeutic opportunities. Pharmacol. Rev. 52:557–594. Si H, Heyken WT, Wolfle SE, Tysiac M, Schubert R, Grgic I, Vilianovich L, Giebing G, Maier T, Gross V, Bader M, de Wit C, Hoyer J, Kohler R, 2006. Impaired endothelium-derived hyperpolarizing factor-mediated dilations and increased blood pressure in mice deficient of the intermediate-conductance Ca2þ-activated Kþ channel. Circ. Res. 99:537–544. Simon DB, Karet FE, Rodriguez-Soriano J, Hamdan JH, DiPietro A, Trachtman H, Sanjad SA, Lifton RP, 1996. Genetic heterogeneity of Bartter’s syndrome revealed by mutations in the Kþ channel, ROMK. Nat. Genet. 14:152–156. Swartz KJ, MacKinnon R, 1997. Hanatoxin modifies the gating of a voltage-dependent Kþ channel through multiple binding sites. Neuron 18:665–673. Swartz KJ, MacKinnon R, 1997b. Mapping the receptor site for hanatoxin, a gating modifier of voltage-dependent Kþ channels. Neuron 18:675–682. Tharp DL, Wamhoff BR, Turk JR, Bowles DK, 2006. Upregulation of intermediate-conductance Ca2þ-activated Kþ channel (IKCa1) mediates phenotypic modulation of coronary smooth muscle. Am. J. Physiol. Heart Circ. Physiol. 291:H2493–H2503. Vianna-Jorge R, Oliveira CF, Garcia ML, Kaczorowski GJ, Suarez-Kurtz G, 2000. Correolide, a nor-triterpenoid blocker of Shaker-type Kv1 channels elicits twitches in guinea-pig ileum by stimulating the enteric nervous system and enhancing neurotransmitter release. Br. J. Pharmacol. 131:772–778. Vianna-Jorge R, Oliveira CF, Garcia ML, Kaczorowski GJ, Suarez-Kurtz G, 2003. Shakertype Kv1 channel blockers increase the peristaltic activity of guinea-pig ileum by stimulating acetylcholine and tachykinins release by the enteric nervous system. Br. J. Pharmacol. 138:57–62. Wulff H, Kolski-Andreaco A, Sankaranarayanan A, Sabatier JM, Shakkottai V, 2007. Modulators of small- and intermediate-conductance calcium-activated potassium channels and their therapeutic indications. Curr. Med. Chem. 14:1437–1457.

REFERENCES

83

Xia XM, Ding JP, Lingle CJ, 1999. Molecular basis for the inactivation of Ca2þ- and voltagedependent BK channels in adrenal chromaffin cells and rat insulinoma tumor cells. J. Neurosci. 19:5255–5264. Xia XM, Zhang X, Lingle CJ, 2004. Ligand-dependent activation of Slo family channels is defined by interchangeable cytosolic domains. J. Neurosci. 24:5585–5591. Ye S, Li Y, Chen L, Jiang Y, 2006. Crystal structures of a ligand-free MthK gating ring: insights into the ligand gating mechanism of Kþ channels. Cell 126:1161–1173. Yohannan S, Hu Y, Zhou Y, 2007. Crystallographic study of the tetrabutylammonium block to the KcsA Kþ channel. J. Mol. Biol. 366:806–814. Yu FH, Catterall WA, 2004. The VGL-chanome: a protein superfamily specialized for electrical signaling and ionic homeostasis. Sci. STKE 2004:re15. Zeng XH, Xia XM, Lingle CJ, 2003. Redox-sensitive extracellular gates formed by auxiliary beta subunits of calcium-activated potassium channels. Nat. Struct. Biol. 10:448–454. Zerangue N, Schwappach B, Jan YN, Jan LY, 1999. A new ER trafficking signal regulates the subunit stoichiometry of plasma membrane K(ATP) channels. Neuron 22:537–548. Zhou Y, Morais-Cabral JH, Kaufman A, MacKinnon R, 2001a. Chemistry of ion coordination and hydration revealed by a Kþ channel–Fab complex at 2.0 A resolution. Nature 414: 43–48. Zhou M, Morais-Cabral JH, Mann S, MacKinnon R, 2001b. Potassium channel receptor site for the inactivation gate and quaternary amine inhibitors. Nature 411:657–661. Zhou Y, MacKinnon R, 2003. The occupancy of ions in the Kþ selectivity filter: charge balance and coupling of ion binding to a protein conformational change underlie high conduction rates. J. Mol. Biol. 333:965–975.

4 INTEGRINS DAVID D. MILLER

4.1

INTRODUCTION

Regulated cell–cell and cell–matrix adhesion and communication are important biological processes whose correct function is vital to the development and survival of multicellular organisms. Nature has evolved an array of cell-adhesion receptor types such as selectins, cadherins, and integrins, each of which has a distinct functional role. The integrins are a small family of complex proteins found in the cell membrane of almost all nucleated cells (Hynes, 2002; Takada et al., 2007). Their major purpose is to provide a signaling connection across the cell membrane, and the family name ‘‘integrin’’ derives from the proteins’ structure: an integral membrane protein complex linking the extracellular matrix to the cytoskeleton (Hynes, 2004). Generically, an integrin comprises an a-subunit and a b-subunit, which heterodimerize noncovalently to form the receptor. To date, 18 a-subunits and 8 b-subunits have been discovered in humans and of the 144 possible heterodimeric combinations of these, 24 are known to form into recognized receptors. Figure 4.1 shows a summary of the integrin family. In the now widely accepted nomenclature for integrins, their individual name is made up from their constituent subunits (e.g., a3b1, aDb2); however, their synonyms, summarized in Table 4.1, frequently derived from their biological role and origin of discovery, for example, VLA-4, are still often used. Considering the successful sequencing of the human genome, it is not expected that more will be discovered. Figure 4.2 shows a typical integrin with a large extracellular region made up of several domains, a single membrane-spanning helical domain, and a short intracellular tail. Integrins are capable of bidirectional signaling, often referred to as

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

85

86

INTEGRINS α11

α2

α1

α10

αE

αIIb

α5

α4 β7

β5

β1 α7 α9

α8 α6

β3

αv

α3

β8 β6

β2

αM αL

β4

αD

αX

FIGURE 4.1 The integrin family. Eighteen alpha- and 8 beta-subunits combine noncovalently to form 24 unique heterodimers. Integrins that recognize the Arg-Gly-Asp motif are shown in circles. The nine alpha-sub–units that contain an I-domain are shown in boxes.

‘‘outside-in’’ and ‘‘inside-out’’ signaling (Arnout et al., 2005). Most of the time integrins exist in a resting, inactive state, but upon activation they recognize and bind extracellular ligands in a highly specific manner. The ligand binding results in conformational changes that propagate from the outside of the cell to the inside where the intracellular tail interacts with further signaling molecules, such as talin, that link to the actin cytoskeleton and signaling pathways that control a spectrum of cellular processes. The inside-out signaling, initiated by the intracellular ligand binding to the cytoplasmic tail, activates the integrin by bringing about conformational changes in the extracellular domain that change the integrin from a low affinity to high affinity state and thereby facilitate ligand binding. The elucidation of detailed integrin structure, the molecular mechanisms by which they transmit signals, and an understanding of their signaling ligands and their functions have been the focus of intensive investigation in the 30 or so years since research into cell adhesion led to the discovery of the integrin family (Luo et al., 2007). Early cell surface electron microscopy studies indicated that the gross structure of the extracellular region comprises a protruding head domain attached to two multidomain legs, or stalk domain that connects it to the membrane. The head is made up of a seven-bladed beta-propeller domain contributed by the a-subunit and an ‘‘I-like domain’’ from the b-subunit. It is the large interface between these two parts that provides the binding energy to hold the dimerized a–b-receptors together. Nine of the 18 a-subunits (see Fig. 4.1) contain an extra domain of about 200 residues, which is inserted between the second and third repeats of the beta-propeller and which is known as the I-domain (for inserted domain). The I-domain and its very similar I-like domain (also known as b-I-domain) in the b-subunit are structurally related to von Willebrand factor A domains and hence are sometimes referred to as integrin A domains.

87

INTRODUCTION

TABLE 4.1

Integrins, Their Synonyms and Ligands

Integrin

Synonym

Ligand

a1b1 a2b1 a3b1 a4b1

VLA-1 VLA-2, GPIa/IIa, ECMR II VLA-3 VLA-4

a5b1 a6b1 a7b1 a8b1

VLA-5, fibronectin receptor VLA-6 VLA-7 VLA-8

a9b1 a10b1 a11b1 avb1 aDb2 aLb2 aMb2 aXb2 avb3

VLA-9

aIIbb3

GPIIb-IIIa, fibrinogen receptor

Collagen, laminin Collagen, laminin, thrombospondin Laminin, thrombospondin VCAM-1, fibronectin, MAdCAM-1, thrombospondin, osteopontin, JAM2 Fibronectin, osteopontin Laminin Laminin Fibronectin, vitronectin, osteopontin, tenascin, LAP-TGF-b VCAM-1, osteopontin, tenascin Collagen, laminin Collagen Fibronectin, osteopontin, LAP-TGF-b ICAM, VCAM-1 ICAMs ICAM, iC3b, factor X, fibrinogen, ICAM, collagen, iC3b, fibrinogen Vitronectin, fibronectin, thrombospondin, tenascin, osteopontin, fibrinogen, von Willebrand’s factor, laminin, bone sialoprotein, Del-1, fibrillin, LAP-TGF-b, PECAM-1 Fibrinogen, fibronectin, von Willebrand’s factor, vitronectin, thrombospondin Laminin Vitronectin, osteopontin, bone sialoprotein, Del-1 Fibronectin, osteopontin, LAP-TGF-b MAdCAM-1, VCAM-1, fibronectin, osteopontin E-cadherin LAP-TGF-b

a6b4 avb5 avb6 a4b7 aEb7 avb8

LFA-1 Mac-1, CR3 Vitronectin receptor

VLA ¼ very late antigen; ECMR ¼ extracellular matrix receptor; VCAM ¼ vascular cellular adhesion molecule; MAdCAM ¼ mucosal addressin cellular adhesion molecule; JAM ¼ junction adhesion molecule 2; LAP-TGF-b ¼ latency-associated peptide-transforming growth factor b; ICAM ¼ intercellular adhesion molecule; LFA-1 ¼ lymphocyte function-associated antigen-1; Mac-1 ¼ macrophage-1 antigen; CR3 ¼ complement receptor 3; PECAM-1 ¼ platelet/endothelial cell adhesion molecule-1.

The I-domains and I-like domains contain a metal ion-dependent adhesion site (MIDAS) in which amino acid side chains from three loops in the protein coordinate a divalent cation, such as a Ca2þ or Mg2þ, upon which ligand binding is dependent. The MIDAS and the surrounding protein surface are important binding

88

INTEGRINS

FIGURE 4.2 A generic integrin structure. (a) Bent, ‘‘inactive’’ conformation. (b) Extended ‘‘active’’ conformation. (c) Integrin containing I-domain in the a-subunit. bTD, b-terminal domain; EGF, epidermal growth factor domain; PSI, plexins, semiphorins, and integrins domain; Mg, metal ion-dependent adhesion site (MIDAS). (See the color version of this figure in the Color Plates section.)

sites for ligands. In non-I-domain integrins, both the MIDAS region in the b-subunit I-like domain and the beta-propeller in the a-subunit contribute to the ligand binding site, whereas in I-domain integrins the ligand binds solely to the I-domain. In both cases, the carboxylate of a ubiquitous acidic Asp or Glu residue in the ligand, which is a requirement for binding, provides the final partner in the octohedral coordination of the cation in the MIDAS. Complementary biochemical and X-ray crystallography-derived structure information has provided an insight into the detailed and large-scale structural and conformational changes that come about during integrin activation and ligand binding. In the head-piece region, crystal structures have been reported for several I-domains, either uncomplexed or in complex with ligands or ligand fragments. Each adopts a common fold with the MIDAS on the upper surface and a connection to the beta-propeller on the lower surface. Two major conformational states have been observed in the structures, a ‘‘closed,’’ low affinity, inactive state and an ‘‘open,’’ high affinity, active state that can bind ligand. In the active state, there is a downward shift in a C-terminal helix (a-7), which conformationally alters the MIDAS and increases its affinity for ligand by up to 10,000-fold. In turn, ligand binding stabilizes the active conformation. The displaced helix contains an exposed Glu that acts as a ligand for the I-like domain and it is through this connection that the I-domain passes its signal to the I-like domain. In the reverse process of activation, it is also through this connection that the I-like domain controls the conformation of the I-domain.

INTEGRIN INHIBITOR DISCOVERY

89

The landmark publication of the X-ray crystal structure of the ectodomain of avb3 (Xiong et al., 2001) contributed significant insight into understanding the extensively researched structural changes that occur in integrins during activation (Liddington, 2002). Briefly, in its inactive state, the integrin is understood to adopt a bent conformation, as shown in Fig. 4.2a, where the head domain is folded back toward the membrane and is largely inaccessible to ligand. Similarly, the cytoplasmic tails are in close association and do not bind signaling partners. In the active state, the integrin becomes extended, as depicted in Fig. 4.2b, the ligand binding sites in the head domain become receptive to ligand, the legs move apart, and the cytoplasmic tails can dissociate and bind their signaling partners. Although it is convenient to think of a simple binary model of activation like this, the reality is much more complex and the structural state of activation of integrins should be considered a continuum that ranges from inactive to fully activated. The tissue expression of integrins is not universal. Depending on the biological role they perform, some integrins such as a5b1 are widely expressed, whereas others are found in particular cell families; for example, the b2 integrins are expressed only on leukocytes or on very specific cell types, for example, aIIbb3 is only found in blood platelets. Complementary to integrins’ tissue expression patterns is the wide range of protein ligands that they bind, which is summarized in Table 4.1 (Humphries et al., 2006). These can broadly be divided into three categories: extracellular matrix proteins such as collagens; soluble proteins such as fibrinogen; and proteins expressed on cell surfaces such as vascular cell adhesion molecules (VCAMs). Although not physiological ligands, viral and bacterial pathogens also use integrins as adhesion sites in cellular invasion. Some integrins are highly selective in the ligands they recognize and others display considerable promiscuity in both the number and categories of ligands they will bind. A common structural characteristic of all the ligands is that they have a small, usually contiguous recognition sequence that includes an absolutely required aspartic or glutamic acid residue to bind to the MIDAS. Some sequences appear on different ligands and many are recognized by multiple integrins. The most frequently observed motif is arginine–glycine–aspartic acid (RGD), which features on a wide range of ligands and is recognized by at least eight integrins. The combined diversity of integrin type, cellular expression, and range of ligands results in the integrin family being involved in a rich spread of biological activities, which include cellular differentiation, proliferation, migration, apoptosis, immune response, angiogenesis, and wound healing among others. It is therefore not surprising that inhibition of integrin function in diseases that are dependent on these biological activities has been the focus of considerable drug discovery research (Simmons, 2005; Staunton et al., 2006; Vanderslice and Woodside, 2006).

4.2

INTEGRIN INHIBITOR DISCOVERY

Monoclonal antibodies targeting integrins deserve mention as they have been central to the early discovery and therapeutic exploitation of integrin inhibitors. They have

90 TABLE 4.2 Integrin aIIbb3 a4b1 aLb2 avb3 a5b1

INTEGRINS

Examples of Clinically Evaluated Antibody Therapies Antibody Abciximab Natalizumab Efalizumab Vitaxin Volociximab

Product TM

ReoPro TysabriTM RaptivaTM

Disease Indication Acute coronary syndrome, MI Multiple sclerosis Psoriasis Melanoma Solid tumors

been used experimentally in the elucidation of integrin function and their pharmacological use has extended from preclinical target validation to therapeutic applications. The elucidation of the RGD sequence as a ubiquitous recognition motif for several integrins and, in particular, the role it plays in platelet aggregation and adhesion through fibrinogen-mediated cross-linking of activated aIIbb3 integrins rapidly made RGD mimetics and aIIbb3 the prototypical integrin target with a focus on antithrombotic discovery. Much of the extensive exploratory work around RGD and aIIbb3 laid the foundations for the subsequent design and discovery of inhibitors of related RGD-recognizing integrins, especially avb3. The search for inhibitors of other integrins has broadly developed in parallel with emergent information about and understanding of their biological function, signaling processes, and therapeutic potential. The leukocyte integrins a4b1 in the b1 family and aLb2 in the I-domain

FIGURE 4.3 Modes of action of known integrin antagonists. A depiction of ligands and antagonists binding to integrin head-piece of (1) non I-domain and (2) I-domain integrins (Adapted from Shimaoka and Springer, 2003). (See the color version of this figure in the Color Plates section.)

INTEGRIN INHIBITOR DISCOVERY

91

containing b2 family have received wide attention. Similarly, the methods used to discover inhibitors have advanced in parallel with technological developments. The biochemical characterization of many integrin inhibitors during their discovery combined with the publication of structural information on ligand binding and, in particular, of X-ray crystallographic information on the extracellular domains of avb3 in complex with a cyclic peptide (Xiong et al., 2002) and the extracellular domains of aIIbb3 in complex with small-molecule inhibitors (Xiao et al., 2004) and small-molecule inhibitors in I-domain allosteric sites (IDAS) (e.g., Kallen et al., 1999) has allowed a categorization of binding modes of action for integrin inhibitors, as depicted in Fig. 4.3 (Shimaoka and Springer, 2003). In non-I-domain integrins, inhibitors compete with the endogenous ligand and bind in a site that spans the a/b interface and includes the MIDAS in the b-subunit. In I-domain integrins, there are two binding sites. In the first, the inhibitor acts allosterically by binding to the a/b interface and the b-subunit MIDAS to prevent the I-domain helix Glu performing its ligand role on the I-like domain. In the second mode also, the inhibitor acts allosterically, but this time by binding at a site remote to the I-domain MIDAS and locking the I-domain in a closed, inactive conformation by blocking the downward movement of the a-7 helix. This site is known as the I-domain allosteric site. 4.2.1

Cyclic Peptides

Translating a biologically active peptide sequence into a nonpeptidic, orally bioavailable drug is widely acknowledged as a significant medicinal chemistry challenge. Nevertheless, peptides continue to provide a rich source of molecular interaction information, and systematic approaches to exploit this information are continuously being developed (Hruby, 2002; Hummel et al., 2006). Having established a target peptide epitope, usually derived from the sequence of the natural biological ligand or chemically by peptide synthesis or phage display, the first steps are to determine the smallest fragment that retains biological activity and to explore its SAR by making changes to the linear sequence. These typically include glycine or alanine replacement to establish the significance of individual residues and the introduction of local constraint via modified amino acids, for example, D-enantiomers or N-methyl, and amide isosteres. Although much can be gleaned from these approaches, the resulting linear peptides are often still very flexible, are highly unlikely to be bioavailable, and are seldom metabolically stable due to rapid in vivo proteolysis. Applying a global constraint via cyclization can serve two purposes: first, if the peptide is held in a favorable bioactive conformation, significant improvements in potency and selectivity can be gained; and second, metabolic stability can be improved. Frequently used cyclization methods include head-to-tail, side chain-to-side chain, especially via cysteines, and insertion of a nonpeptide spacer between the N- and C-termini. The early discovery of small, contiguous recognition motifs for many integrins prompted a rapid growth in the search for related peptide and peptidomimetic inhibitors as therapeutic agents with cyclic peptides featuring prominently

92

INTEGRINS O

NH2 HN

N H

O

H N

HN

N H

O

O

NH2

OH O

HN

O

N H

HN

HN

O

S

OH O

O HN

O

N

N

O

S

H N

N H O

N H

O

N H

H2N O

1 Eptifibatide

αIIbβ3

2 Cilengitide

αvβ3

NH2

O N H

NH2

HN

NH

HN O N

N H

H N

O N H O

O

O

N H

NH O

OH

3 ZD-7349

FIGURE 4.4

α4β1

Examples of cyclic peptide integrin inhibitors.

(Weide et al., 2007). Figure 4.4 shows examples. Eptifibatide 1 is a potent inhibitor of aIIbb3 marketed as an antithrombotic for use in acute coronary syndrome (Scarborough, 1998). Its discovery was based on the Lys–Gly–Asp (KGD) motif found in the aIIbb3-selective snake venom-derived disintegrin, barbourin (Scarborough et al., 1993). Guanidination of the lysine e-amine significantly increased aIIbb3 inhibitory activity. Cilengitide 2, an antiangiogenic avb3 inhibitor, has demonstrated clinical antitumor efficacy (Smith, 2003). The conformational constraint imparted by the D-Phe and N-methylvaline is crucial to the cyclic pentapeptide’s avb3 potency and selectivity relative to its Arg–Gly–Asp–Ser (RGDS) progenitor. The highly selective, Leu-Asp-Val (LDV)-based a4b1 inhibitor, ZD-7349 3, is effective in preclinical inflammation models. Its prolonged duration of action is attributed in part to improved stability conferred by the N-methyl- and D-amino acids (Dutta et al., 2000). Despite demonstrable efficacy among the cyclic peptide integrin inhibitors, their physical properties limit their use to parenteral dosing. Enormous research effort has been expended in the search for nonpeptide, orally active antagonists that are expected to have much wider potential therapeutic applicability. 4.2.2

Peptidomimetic Chemistry

aIIbb3 is expressed in large numbers on the surface of blood platelets. Normally it exists in an inactive conformation, but upon activation, integrin binding to repeated

93

INTEGRIN INHIBITOR DISCOVERY NH

H N

H2N

N H

O

NH Guanidine binding site

O

O

H N

Central spacer

CO2H

Carboxylate binding site

(a) O

O

O N

N H

H2N NH

CO2H

H2N

N

HN

O

4 NSL-95301

NHCOOCH2Ph

O

CO2H NHSO2

5

CO2 H tBuHN

N H

CO2H

HN O

NHSO2nBu

OH 6

7 Tirofiban

FIGURE 4.5 The Arg-Gly-Asp (RGD) motif recognized by aIIbb3 (a) and examples of peptidiomimetic aIIbb3 inhibitors.

RGD motifs in fibrinogen leads to cross-linking of the platelets and eventually to platelet aggregation. Under normal circumstances, the resultant clotting is important in preventing blood loss and in blood vessel repair. In pathophysiological situations, inappropriate thrombotic events are linked to a range of cardiovascular diseases involving vascular occlusion such as myocardial infarction, angina, and stroke. Lying at the end of the clotting cascade, the aIIbb3/fibrinogen interaction has been viewed as an attractive drug target and the development of antithrombotic therapies based on the inhibition of platelet aggregation by competitively blocking the binding of fibrinogen to aIIbb3 has received extensive research investment, with much of its focus on the discovery of small-molecule RGD mimetics (Hanson et al., 2003; Andronati et al., 2004). The main features of the RGD sequence required for aIIbb3 binding, as shown in Fig. 4.5, are the basic guanidine of the arginine side chain, the carboxylate of the aspartic acid side chain, and a central spacer that contributes to maintaining the correct distance between the Arg and Asp components. The majority of aIIbb3 antagonist discovery has focused on mimicking these features in a molecule with improved potency, selectivity, and drug-like properties. A great many Arg mimetics have been reported and among the most commonly used are aryl amidines and secondary amines, consistent with the hypothesis that aIIbb3 requires a strongly basic center. Similarly, a broad range of Asp mimetics has been exploited, among which obligate is a free carboxylate that coordinates to the metal cation in the MIDAS of the integrin b-chain I-like domain. Other known carboxylic acid isosteres, such as tetrazoles, are not tolerated. In fact, ablation of activity in molecules by modification of their acids to

94

INTEGRINS

amides is frequently employed as part of hit validation. Replacement of the central glycine with conformationally constraining cores can impart improved potency and selectivity as well as better metabolic stability, due to their reduced peptide nature, and pharmacokinetic properties, due to increased lipophilicity. Figure 4.5 exemplifies aIIbb3 antagonists discovered by different approaches that have been widely applied. NSL-95301 4 was identified through a three-component combinatorial chemistry array designed to mimic the linear RGD sequence. Its (þ)-enantiomer inhibits platelet aggregation with an IC50 value of 0.092 mM (Harada et al., 1997). Conformational modeling of an RGD-containing cyclic peptide led to the identification of benzoxazole or benzimidazole as a potential central core replacement and subsequent elaboration of this template led to 5 (platelet aggregation IC50 ¼ 0.010 mM) (Xue et al., 1996). A simple pharmacophore-guided search of the Merck sample collection for compounds ˚ gave the lead molecule containing a carboxylic acid and amine separated by 10–20 A 6 (platelet aggregation IC50 ¼ 27 mM), which was comparable in potency to ArgGly-Asp-Ser (IC50 ¼ 26 mM) and in which the tyrosine replaces Gly–Asp. Sequential optimization of the Arg side chain, pendent from the tyrosine oxygen, and the tyrosine N-substituent eventually led to the discovery of tirofiban 7, which is both potent (IC50 0.011 mM) and highly selective over other RGD-recognizing integrins, such as avb3 and a5b1, in cell adhesion assays (Hartman et al., 1992). It is interesting to note that the N-sulfonamide group, which contributes significantly to tirofiban’s potency, accesses an exosite in the protein that had not been previously utilized by linear or cyclic RGD peptides and that went on to be widely exploited in other aIIbb3 inhibitor design. The original research group synthesis of tirofiban is described in Fig. 4.6 (Egbertson et al., 1994). It begins with the reduction of 4-pyridylmethylcarboxylate

1. H 2, PtO2

N

BocN

BocN

1. Swern oxidation

CO2H 2. Boc 2O

OH

CO2Me

2. Ph3P=CHCO2Me

3. BH 3

1. NaOH 2. BH3 3. CBr4, PPh3

CO2H NHCBZ

HO

+

BocN Br NaH, DMF

CO2H

BocN O n BuSO

2Cl,

NH2

H2, Pd/C

CO2H

BocN O

NHCBZ

NaOH

.HCl

CO2H

BocN

HCl, EtOAc

O

CO2H

HN

NHSO2nBu O 7 Tirofiban

FIGURE 4.6

Research synthesis route to tirofiban.

NHSO2nBu

95

INTEGRIN INHIBITOR DISCOVERY

followed by a Wittig reaction to homologate the alkyl chain. The ester is then hydrolyzed to allow a reduction of the carboxylic acid to the primary alcohol, which is displaced with a bromide. This material is used to alkylate the CBzprotected tyrosine. The product is then selectively deprotected using hydrogenolysis, and then sulfonylated with n-butylsulfonyl chloride. The final step is another deprotection step, this time under acidic conditions, to afford tirofiban 7. To date, it is the only small-molecule integrin inhibitor that has reached approval and it finds use as an injectable in acute coronary syndrome. The optimization of aIIbb3 lead molecules into orally active drugs suitable for use in chronic disease settings has consistently challenged medicinal chemistry. The shortcoming is largely attributed to their zwitterionic characteristic and to the residual peptidic nature of the compounds. Successful approaches have included the conversion of potent, but poorly bioavailable, drugs into bioavailable prodrugs whose active constituent is metabolically released in vivo. This method of installing better properties has been achieved in single prodrugs by conversion of the carboxylic acid, usually to esters, or in double prodrugs by esterification and modification of the basic center. Examples that have been clinically evaluated are Orbofiban 8 (single prodrug) and Sibrafiban 9 (double prodrug), as shown in Fig. 4.7. Lotrafiban 10, which has a much more rigid core structure, is unusual in being an orally active nonprodrug. It reached Phase III, but its development was discontinued by SmithKline Beecham due to insufficient efficacy and safety concerns. The avb3 is among the most promiscuous integrins. It is expressed on many cell types, such as osteoclasts, vascular smooth muscle, and endothelial and many tumor cells, and it recognizes a broad range of RGD-containing ligands. Consequently, it plays a central role in many biological processes and its aberrant function is associated with several disease states including osteoporosis, rheumatoid arthritis, restenosis, macular degeneration, and cancer and tumor metastasis. Not surprisingly, it

O NH

O

O H2N

O

H N

N

H N O

OEt

N H

OH N

N O

O NH2

8 Orbofiban

9 Sibrafiban

O N

N

HN 10 Lotrafiban

FIGURE 4.7

O N H

O OH

Orally bioavailable aIIbb3 inhibitors.

OEt

96

INTEGRINS

(a) R

NH

O HN

Ph

(b) R′ HN H N

HNOC O

O

O N H

O H N CO2H

H

NH HN NH2

N H N H O

N O CO2H

NH NH2

FIGURE 4.8 RGD conformations. (a) Turn-extended-turn preferred by aIIbb3. (b) Gammaturn preferred by avb3.

has been the focus of intensive research and there is a significant body of medicinal chemistry literature around the target (Cacciari and Spalluto, 2005). Many of the RGD-based peptidomimetic discoveries in the aIIbb3 field influenced early avb3 inhibitor discovery and potent inhibitors were soon being reported. Achieving selectivity over aIIbb3 was a major goal and analyses of cyclic peptides that were selective for avb3 established that in contrast to aIIbb3’s preference for RGD ligands with an extended glycine, this integrin preferentially binds the sequence in which there is a gamma-turn about the central glycine (Haubner et al., 1997). As a result of this gamma-turn, there is a significantly shorter distance between the arginine and aspartic acid termini, which could be exploited in the molecular design of avb3 selective compounds, as demonstrated in Fig. 4.8. An exploration around the nature of replacements for the arginine guanidine, particularly pKa and geometry, established further differences, which facilitated the achievement of even greater selectivity. A strongly basic center is not a requirement for avb3 potency and a wide range of near-neutral or nonbasic mimetics such as benzimidazole, 2-aminopyridine, and other aminoazaheterocycles that retain a guanidine- or amidine-like disposition of nitrogen has been reported. Although aIIbb3 prefers an ‘‘end-on’’ presentation in its guanidine recognition (see examples in Fig. 4.5), selectivity toward avb3 can be achieved by a ‘‘side-on’’ geometry, as exemplified by the tetrahydronaphthyridine featured in Fig. 4.9. With regard to aspartic acid replacement, b-alanine with bulky substituents in either the a- or b- position, which can access the aforementioned ‘‘exosite’’ on the b3 integrin chain, have been widely favored and contribute significantly to inhibitor potency. The compatibility of more compact, nonzwitterionic compounds with high levels of avb3 potency and selectivity has beneficially resulted in the discovery of a higher proportion of orally active, non-prodrug inhibitors. The elegant development of a series of nonpeptide avb3 antagonists starting from the RGD sequence is described in Fig. 4.9 (Coleman et al., 2004). Introduction of central constraint

97

INTEGRIN INHIBITOR DISCOVERY

H2N

NH

H N

CO2H

N H

O

NH

O

O

H N

RGD tripeptide

O

H N

O N

N

CO2H

N H

O

Constrained peptidomimetics

11 O

H N

O N

CO2H

N H

Chain-shortened series

12

N H N

N

CO2H

N

Aliphatic chain-shortened series

13

FIGURE 4.9

Translation of the RGD sequence into an orally active avb3 inhibitor.

and modification of the Arg guanidine and Asp to tetrahydronaphthyridine and dihydrobenzofuranyl-b-alanine respectively had yielded a potent inhibitor 11, albeit with poor pharmacokinetics (F ¼ 6%, Cl ¼ 33 mL/min/kg). Oral bioavailability is described with an F to represent the percentage of drug substance found in the plasma levels after oral dosing as compared with an i.v. dose. The chainshortened analogue 12, without central constraint, had much better pharmacokinetics (F ¼ 99%, Cl ¼ 1.2 mL/min/kg), but at the expense of an unacceptable drop in potency. Reintroduction of central constraint, via an imidazolidinone, in the chain-shortened series (not shown) yielded a potent and orally available inhibitor, which was projected to be suitable for twice-daily dosing. Further optimization led to the aliphatic chain-shortened, unconstrained compound 13, which was potent (avb3 IC50 ¼ 0.00008 mM), possessed a favorable pharmacokinetic profile for once-daily dosing (dog PK: F ¼ 83%, t1/2 ¼ 9.5 h, Cl ¼ 2.5 mL/min/kg), and was active in animal models of bone loss. Compound 13 and a close analogue were selected as development candidates for the treatment of osteoporosis.

98

INTEGRINS

FIGURE 4.10 Pharmacophore-directed discovery of novel avb3 inhibitors. Structures are color-coded according to common pharmacophoric features. (See the color version of this figure in the Color Plates section.)

Once multiple inhibitors, preferably represented by different chemotypes that bind at a common site, are known, clusters of these can be used to build a pharmacophore model, which represents features common to the set and which can be used to select database compounds for directed screening to find new hit molecules to optimize. An example of this was recently reported for avb3 (Dayam et al., 2006). A 3D pharmacophore based on a training set of known avb3 inhibitors, typified by compound 14, shown in Fig. 4.10, was validated against a larger set of known inhibitors and then used to search 3D databases of available compounds. Among the small number of retrieved hits that were tested, a novel iminothiazolidinone inhibitor 15 was found to inhibit avb3 in a binding assay (IC50 ¼ 0.240 mM). The authors use docking studies into the published X-ray structure of avb3 from its complex with cyclo(RGDf-N{Me}V) to postulate a binding mode for the inhibitor, which predicts that its carboxylate coordinates to the Mn2þ in the MIDAS. Many autoimmune diseases are characterized by inappropriate and excessive inflammatory responses that cause damage to affected tissues. Recruitment of leukocytes and their subsequent infiltration into and accumulation within the affected tissue are considered to be key processes in such chronic inflammatory diseases. These processes are largely initiated and regulated by ligands on the vascular surface, recognizing and binding to circulating leukocytes, which results in them attaching increasingly tightly to the vascular wall and then being available to invade the surrounding tissue. Among the most important of these adhesion events is the recognition between a4 integrins and their cell adhesion molecules (CAMs). a4b1 (VLA-4, very late antigen-4) is widely expressed across the leukocyte family and its main protein ligands are the immunoglobulin VCAM-1 and an alternatively spliced variant of fibronectin, CS-1, found in the extracellular matrix. Inhibitors of a4b1 have been extensively explored as potential treatments for chronic inflammatory diseases such as asthma, multiple sclerosis, and inflammatory arthritis. The related a4b7 integrin is predominantly expressed on mucosal lymphocytes and its primary ligand is MAdCAM-1 (mucosal addressin cell adhesion molecule-1); however, it also recognizes VCAM-1 (albeit more weakly than a4b1) and CS-1. The relatively restricted cellular expression of a4b7 and MAdCAM has made their interaction an attractive target in inflammatory bowel disease.

99

INTEGRIN INHIBITOR DISCOVERY

The main peptide motifs recognized by a4 integrins are Gln–Ile–Asp–Ser (QIDS) in VCAM-1 and Ile–Leu–Asp–Val (ILDV) in CS-1 for a4b1 and the closely related Leu–Asp–Thr (LDT) in MAdCAM for a4b7. The conserved Asp is understood to coordinate to the divalent metal in the integrin MIDAS, and IDS and LDV can be broadly considered isosteric. Similarities between LDV and RGD prompted early exploration into the utility of RGD-based peptides and small molecules as templates for a4b1 and as the therapeutic potential for a4 inhibitors grew in importance the IDS, LDV, and LDT peptide sequences increasingly became the foundation for wide-ranging medicinal chemistry research into small-molecule a4 integrin inhibitor discovery, largely following similar peptidomimetic approaches as applied to the RGD integrins (Jackson, 2002; Yang and Hagmann, 2003). A discovery program was initiated based on the ILDV sequence, which weakly inhibited a4b1/VCAM-1 binding with an IC50 value of 66 mM. Sequential replacement of each amino acid identified tyrosine as an alternative to the N-terminal isoleucine. Significant improvements in potency were achieved by deleting the Tyr a-amino group and by reducing the methylene side chain to give the hydroxyphenylacetamide 16 (Fig. 4.11, IC50 ¼ 0.43 mM), in which the phenolic OH was found to be important, probably due to H-bonding with the receptor. Further elaboration of the N-terminal capping group led to the highly potent phenylureidophenylacetic acid (PUPA) derivative 17 (IC50 ¼ 0.0006 mM), which was up to 10,000-fold selective across a panel of non-a4 integrins, though it did retain activity at a4b7 (Lin et al., 1999). Its LeuAsp-Val-Pro-OH analogue (known as Bio-1211), which has particularly tight binding characteristics (Kd ¼ 70 pM) due to a very slow off-rate, was selected as a development drug candidate that reached Phase II trials for asthma but development was stopped due to lack of efficacy, most likely attributable to a poor pharmacokinetic profile. Researchers continued their pursuit of the series and adopted an in silico screening approach to identify nonpeptidic replacements for the LDV peptide fragment (Singh et al., 2002). A 3D pharmacophore model was constructed from 17. Its key features, based on SAR from the previous peptide work, were the PUPA, the

O OH O H

H N

N

NH-L-D-V-OH

O H N

N H

OH

O

O

HO

H N

O

17

H N

O

N H

O

N H

N H

16

ILDV

N H

NH-L-D-V-OH

O

O

H N

OMe O

O

N

CO2H

CO2H 18

FIGURE 4.11

19

The a4b1 recognition motif ILDV and derived peptidomimetic inhibitors.

100

INTEGRINS

phenyl acetamide carbonyl, and the Asp carboxylate. The conformation of the peptide segment was derived from the well-defined shape adopted by the analogous IDS recognition sequence in an X-ray structure of VCAM-1. The pharmacophore was used to search a multiconformational, virtual library of 8624 compounds that was constructed by coupling PUPA to a set of commercially available amino acids, some from nitro precursors. Of the 170 nonpeptide search hits, 12 were selected for synthesis and all were active with inhibitory potency ranging from nanomolar to 20 mM, the most potent being compound 18 (IC50 ¼ 0.0013 mM versus a4b1/ VCAM binding) that was active in an in vivo sheep model of asthma when administered by inhaled aerosol. The emergence of potent and subsequently nonpeptidic, PUPA-based inhibitors prompted several groups to exploit this template in the search for orally bioavailable a4b1 antagonists. GlaxoSmithKline researchers recently reported a potent series of pyridones, exemplified by 19 (pKi ¼ 9.1 in an a4b1 cell adhesion SPA) in which an additional ortho-methoxyl group on the phenylurea contributed to an encouraging, though variable, pharmacokinetic profile in rats (F ¼ 121%, Cl ¼ 9 mL/min/kg, t1/2 ¼ 1.4 h) (Witherington et al., 2006). In the search for a4b7 inhibitors, the Kessler group applied a combinatorial approach in the development of LDT-based peptide information to nonpeptidic compounds (Gottschling et al., 2002). Cyclic peptide arrays demonstrated that the threonine of LDT could be replaced by phenylalanine or phenylglycine (20, Fig. 4.12) and LDT sequences capped at the N-terminus with aromatic substituents, such as 21, known to be effective inhibitors of a4b7/MAdCAM binding. This

cyclo-(F-L-D-Phg-p)

Aromatic

N H

20

O

H N

Linker

Hydrophobe

N H

O Acid

22

O N H

N

H N O

OH

O

O N H CO2H

CO2H N

N

N H

FIGURE 4.12 leads.

CO2H

O

H N CO2H

25

N H

O 23

21

O

O

H N

N H

N

N H

H N O

CO2H

24

Combinatorial array design in the discovery of nonpeptide a4b7 inhibitor

101

INTEGRIN INHIBITOR DISCOVERY

information was combined with a library design (22) in which each of the components was sequentially varied in a semicombinatorial manner. Little scope was found for variation around the isoquinoline–carbonyl (aromatic), Leu (linker), or Asp (acid) components, but consistent with the cyclic peptide results, the hydrophobe could be served by an aromatic group such as in 23. b-Aryl-substituted b-alanine, known as an Asp-mimetic from avb3 research, was found to be an effective replacement for the C-terminal acid–hydrophobe pair and this shorter, simplified molecule (24) retained the potency of its progenitor, 21. Finally, reduction of the Leu-b-alanine amide provided an inhibitor of improved potency (25, 89% inhibition of a4b7/MAdCAM at 1 mg/mL for the most active diastereoisomer) although with apparently reduced selectivity over a4b1/VCAM-1 inhibition. The lead molecule 25 had an Mw ¼ 433.5, log P ¼ 4.2, hydrogen bond donors (HBD) ¼ 3, and hydrogen bond acceptors (HBA) ¼ 3. These features were considered a favorable starting point for further optimization as, unlike the molecular starting points, it met Lipinski’s ‘‘rule of five’’ criteria for physical characteristics, which predict favorable solubility and permeability in drug molecules (Lipinski et al., 1997); however, no pharmacokinetic data were reported to support this. Although the relative merits of a4b1- or a4b7-specific inhibitors versus dual active compounds in various therapeutic contexts still remain to be answered, there have been investigations into understanding and exploiting differences between these integrins in designing selective or dual antagonists. In the absence of protein structural information on a4b1 or a4b7, much of the guiding knowledge has been derived from SAR around active templates. For example, researchers reported on a series of N-substituted thioprolyl biarylalanines in which relative activity at a4b1 and a4b7 was influenced by the choice of N-substituent and the stereochemistry of the thioproline (Lin et al., 2002). A differential binding site hypothesis was proposed to explain the observed results. Compound 26, shown in Fig. 4.13, is a potent and 1000-fold selective a4b1 antagonist. The (R)-isomer of the thioproline is postulated to point the N-sulfonamide toward a binding pocket, which is significantly

S H N

N

Cl

SO2

O

O H N

N S

CO2H O

CO2H

O

O

Cl 26

27

α4β1 IC50 = 0.3 nM

α4β1 IC50 = 0.2 nM

α4β7 IC50 = 306 nM

α4β7 IC50 = 2.1 nM

FIGURE 4.13 Selectivity between a4b1 and a4b7 is determined by the thioproline configuration and the size of its N-substituent.

102

INTEGRINS

larger in a4b1 than in a4b7 and therefore better able to accommodate a large substituent. Analogues bearing smaller groups are markedly less selective. In contrast, the thioproline (S)-isomer, with the opposite configuration, is believed to direct the N-substituent into a different pocket that is comparable in size between the two proteins and is more compact than the first, resulting in a potent dual antagonist in which the optimum N-derivative is an acetyl group (27). 4.2.3

Preferred Peptidomimetic Fragments

The ubiquity of an aspartic acid in the ligand recognition motifs described above and the common theme of it being part of a small contiguous sequence give integrin inhibitor discovery the opportunity to exploit templates and fragments discovered for one target in lead generation for others. This approach has been alluded to above, for example, in the use of substituted b-alanines as Asp-mimetics. Benzodiazepines with a pendent carboxylic acid, which were developed by several groups as Gly–Asp mimetics and which originally featured in a variety of aIIbb3 antagonists, provide among the best examples of this approach as successfully applied within and across the integrin subfamilies. SmithKline Beecham scientists took the gamma-turned Gly–Asp mimetic 1,4-benzodiazepinone core found in their highly potent aIIbb3 inhibitor Lotrafiban and reversed its 4000-fold aIIbb3/avb3 selectivity to generate a correspondingly potent avb3 inhibitor SB-223245 (28) with greater than 10,000-fold selectivity over aIIbb3 (Keenan et al., 1997). This strategy was achieved through a series of sequential structural modifications to reach an Arg guanidine mimetic, as shown in Fig. 4.14, that was consistent with what became the accepted avb3 selectivity dogma, namely, a shorter acid–base (a)

O

O N

N

N

O N H

HN

N

N

O

NH N H

CO2H

10 Lotrafiban αIIbβ3 K i = 2.5 nM αvβ3 K i = 10,340 nM

CO2H

28 SB-223245 αIIbβ3 K i = 30,000 nM αvβ3 K i = 2 nM

(b)

HN

N H

CO2H

O

NH2

H N

O

N

O

N H

N

N H

N

O N O

O 29 αIIbβ3

CO2H

O H N

30 α4β1 IC50 = 150 nM

FIGURE 4.14 Benzodiazepine-preferred fragments. (a) Within the b3 family. (b) Across the b3 and b1 families.

INTEGRIN INHIBITOR DISCOVERY

103

distance, lowered pKa, and a ‘‘side-on’’ nitrogen presentation. A structural rationale for this extreme change in selectivity can be based on analogy to the binding of the RGD cyclic peptide in the avb3 cocrystal structure (Xiong et al., 2002). The constant benzodiazepinone Asp mimetic binds to the metal cation and the MIDAS region in the b3-chain that is common to both integrins, whereas the differentiated Arg mimetics bind to aspartic acid residues in the different aIIb and av a-chains. Although potent and selective, SB-223245 was poorly bioavailable. It did however provide the starting point for successful development of a family of related, but increasingly less peptidic, inhibitors, some of which had excellent pharmacokinetic profiles (Miller et al., 2000). In a similar vein, workers at Biogen were able to convert a b3 integrin family inhibitor into a compound that was active in a completely different integrin subfamily, b1 (Singh et al., 2004). By replacing the aIIbb3-specific guanidine substituent with the a4b1-directing phenylureidophenylacetic acid (PUPA), the known benzodiazepinedione template 29 was reconfigured to give a very potent a4b1 inhibitor 30. 4.2.4

I-Domain Integrin Inhibitors

The members of the b2 integrin family, which share a common b-subunit and are expressed exclusively on leukocytes, play critically important roles in immune responses. aLb2 (LFA-1, lymphocyte function-associated antigen-1), which is expressed on many inflammatory response cell types, recognizes ICAM-1 (intercellular adhesion molecule-1) as a major ligand and the interaction between the integrin I-domain and ICAM-1, whose expression on the endothelium is greatly upregulated at sites of inflammation, is central to the activation of leukocytes and their recruitment to, and invasion into, sites of tissue inflammation. Consequently, inhibition of aLb2 function has been the focus of much research into the discovery of immunomodulatory and anti-inflammatory therapeutics with relevance to a range of diseases such as arthritis, psoriasis, and transplant rejection (Giblin and Lemieux, 2006). In contrast to the integrin inhibitors described in the preceding sections, which competitively antagonize ligand binding, small-molecule aLb2 integrin inhibitors that have been reported in the literature and for which the site of action is known fall into the two distinct allosteric categories described earlier and shown in Fig. 4.3. These are based on where they bind to the integrin and how they antagonize ligand binding, namely, I-domain allosteric or a/b I-like allosteric site. Indeed, the modes of binding and biological modulatory effects of several of these molecules have contributed to the growing understanding of the conformational changes implicit in I-domain integrin activation and inhibition (e.g., Welzenbach et al., 2002). The discovery that the lipid-lowering, hydroxy-methyl-glutaryl coenzyme A (HMG Co A) reductase inhibitor lovastatin (31), shown in Fig. 4.15, inhibited aLb2 through binding to a unique site in the I-domain, remote to the MIDAS, which would become known as the IDAS, or L (for lovastatin)-site, was another

104

INTEGRINS OH HO

HO

O O

O

HO O

O

O

O

O

31 Lovastatin

O

O H

H

O N

H

32

33 LFA703

O OH

O HN

O

N

O O

O

O

N

O

H2N

O

N

O

N

N

N

N H

O

N

O H

34 LFA878

35

FIGURE 4.15

36

Lovastatin and related aLb2 inhibitors.

landmark in integrin research. It offered the prospect of potent and selective integrin inhibitors that are nonpeptidic and hence avoid many of the development challenges faced by the peptidomimetics described in earlier sections. Identified by high-throughput screening, lovastatin blocks aLb2/ICAM-1 binding with an IC50 value of 2.1 mM and is also active in aLb2-mediated cell adhesion assays (Kallen et al., 1999). The confirmation that lovastatin’s aLb2 activity was independent of its HMG Co A reductase activity was obtained by demonstrating comparable integrin inhibition by the des–oxo analogue 32, which cannot hydrolyze to the ring-opened hydroxy-acid form of the lovastatin lactone: a requirement for HMG Co A reductase inhibition (Weitz-Schmidt et al., 2001). Optimization led to compounds such as LFA703 (33) with improved potency (aLb2/ICAM-1 binding IC50 ¼ 0.2 mM), which were inactive against HMG Co A and selective over another I-domain b2 integrin, aMb2 and a non-I-domain integrin a4b1. Encouragingly, LFA703 and analogues exhibited anti-inflammatory effects in in vivo models of inflammation. Oral administration of LFA703 is reported to almost completely block thioglycollate-induced peritonitis in mice with a remarkably low ED50 of 0.4 mg/kg, whereas the more in vitro potent analogue LFA878 (34) (aLb2/ICAM1 binding IC50 ¼ 0.05 mM) showed a dose-dependent anti-inflammatory effect in a rat carrageenan paw oedema model with an oral ED50 of approximately 30 mg/kg (Weitz-Schmidt et al., 2004). The integrin selectivity and in vivo activity of these early molecules augured well for the therapeutic potential of inhibitors that bind to the IDAS. The research group went on to pursue approaches to synthesize novel structures around scaffolds designed to fit the aLb2 IDAS. Docking studies into

105

INTEGRIN INHIBITOR DISCOVERY

an I-domain structure derived from the X-ray of the aLb2 I-domain in complex with lovastatin indicated a favorable fit for a 1,4-diazepane bearing a large lipophilic group, such as naphthyl, that would occupy the same position as the decalin fragment of lovastatin. Combinatorial array synthesis and preliminary optimization led to the discovery of 35 as a novel, potent inhibitor (aLb2/ICAM-1 binding IC50 0.07 mM) (Wattanasin et al., 2003). Further modifications around the template yielded a series of closely related 1,4-diazepane-2,5-diones, typified by 36 (aLb2/ICAM-1 binding IC50 ¼ 0.069 mM), which is highly selective over a4b1 and aMb2 and was confirmed by X-ray crystallography to bind to the lovastatin site in the I-domain (Wattanasin et al., 2005). It is of interest to note that although these compounds are novel, designed inhibitors of aLb2, 36 is no more potent than, for example, LFA878 and suffers from high protein binding (>97%) and limiting oral bioavailability (7% in mice and 8.6% in rats), which would require further optimization. The early discovery of an active dichlorophenylhydantoin 37 (aLb2/ICAM-1 binding Kd ¼ 3.5 mM) through high-throughput screening sparked significant investigation into hydantoin-based aLb2 inhibitors, as shown in Fig. 4.16, by several organizations. The SAR studies and optimization of 37 led to BIRT377 (38) (aLb2/ICAM-1 Kd ¼ 0.025 mM), which is selective over a4b1 and aMb2 and which blocked aLb2-dependent, functional, cellular assays in a specific manner. Moreover, BIRT377 also demonstrated efficacy in an in vivo inflammation model by dosedependent inhibition of aLb2-dependent, staphylococcal enterotoxin B-induced IL-2 production at 25 and 50 mg/kg administered orally (Kelly et al., 1999). The Boehringer group hypothesized that BIRT377 was acting allosterically on the Cl

N

O

Cl

Cl

O

O

N

Cl

Cl

O

O

N

N H

Cl

N

Cl

N N

Cl

O

O

N

N

H

N

NO2

Cl

O

O

N

O

N

Cl

Cl

O

O

42

FIGURE 4.16

Cl

N

O

N X

NC

OCOCH3 41

40

N Br

Br

Br

39

Cl

O S O

O

N

N

38 BIRT377

Cl

N

O

Br 37

Cl

Cl

N NC

43a X = CH2 43b X = NR

Hydantoin-based aLb2 inhibitors.

S CO2H 44 BMS-587101

106

INTEGRINS

I-domain and pursued several approaches, including photoaffinity labeling and docking modeling to test the hypothesis. This research culminated in the solution of a crystal structure of the I-domain in complex with a close analogue 39, which confirmed that the compounds do bind allosterically at what would become known as the IDAS (Last-Barney et al., 2001). Poor solubility and rapid metabolism, predominantly via N-demethylation, hindered the progression of BIRT377 and second-generation discovery focused on bicyclic systems, which would obviate demethylation and provide additional sites to incorporate solubilizing groups. Among the systems investigated, the imidazo[1,2-a]imidazol-2-one (40) (aLb2/ICAM-1 Kd ¼ 0.38 mM) stood out as a stable structure that was reasonably potent (Wu et al., 2004). Retaining the invariant dichlorophenyl group on N1, SAR exploration showed that electron-withdrawing groups, particularly sulfones, in the 5-position could provide increased potency, solubility, and metabolic stability, though progress was hampered by high plasma protein binding. Compound 41, an X-ray structure of which confirmed its binding to the IDAS, was among the most potent inhibitors reported in the series (aLb2/ICAM-1 Kd ¼ 0.014 mM). Pharmaceutical researchers described the discovery of a series of bicyclic[5.5] hydantoins, which were designed by analogy to BIRT377, the most potent of which was 42 (aLb2/ICAM-1 cell adhesion IC50 ¼ 0.085 mM) (Potin et al., 2005). Modification of the bicyclic hydantoin to a spirocyclic cyclopentyl-based template 43a yielded novel aLb2 inhibitors, which were not pursued due to poor developability properties. However, X-ray structural studies suggested that further substitution could be accommodated on the 5-membered ring (Potin et al., 2006) and exploration of the analogous pyrrolidine system (43b) provided a series of potent compounds in which the nature of the pyrrolidine N-substituent could be used to modulate potency and tailor the molecules’ drug-like properties. The optimal set of properties was found in 44, BMS-587101 (aLb2/ICAM-1 cell adhesion IC50 0.020 mM), which had a favorable pharmacokinetic profile (e.g., dog PK: F ¼ 100%, Cl ¼ 3.6 mL/min/kg, t1/2 ¼ 3.5 h) and an excellent safety profile (low metabolic turnover, cytochrome P450 inhibition, hERG inhibition, and mutagenicity). Evaluation of BMS-587101 in two in vivo models of aLb2 function, a mouse ovalbumin-induced lung inflammation assay and a mouse heart-to-ear nonvascularized allograft transplant assay showed positive results that were comparable to those achieved by an aLb2 antibody. BMS-587101 was selected as a clinical candidate and reached Phase II trials for psoriasis before its development was discontinued for unspecified reasons. Over a series of publications, researchers at Abbott describe the elegant development of a screening hit into potent aLb2 inhibitors with oral in vivo efficacy in animal models of inflammation. A key step in the progression used NMR-based fragment screening to identify new groups for incorporation into the molecules. The discovery process began with the high-throughput screening identification of the moderately active arylthioaniline 45 in Fig. 4.17 (aLb2/ICAM-1 binding IC50 1.7 mM) (Liu et al., 2000). Early optimization work exploring the ortho substituents on the B ring established the existence of an additional binding pocket that could accommodate structures such as 46, which was at least 10-fold more potent than the

107

INTEGRIN INHIBITOR DISCOVERY Cl

Cl A

N

Cl

Cl

Me

S

S

S B Cl

O2N

HN

NH2

O2N O

N H

O

O

45 αLβ2 IC50 1.7 µM

N

N

N

N

N

O O

COCH3 47 αLβ2 IC50 44 nM Solubility 0.9 µg/mL Rat F 0%

46 αLβ2 IC50 140 nM

O

O

N

COCH3

S

48

N 49 a-d

MeN O

O

S

S

S

Cl

F3C

Cl

S

O

O

H

Cl O

Cl

O

Cl H

N

N

51 αLβ2 IC50 25 nM Solubility >3000 µg/mL Rat F 29%

52 αLβ2 IC50 6 nM Solubility >3000 µg/mL Rat F 60%

N N

HO2C CO2H

COCH3 50 αLβ2 IC50 40 nM Solubility 4.1 µg/mL Rat F ~13%

FIGURE 4.17

N

CO2H 53 αLβ2 IC50 6 nM

The development of arylthio cinnamide aLb2 inhibitors.

original hit. Further modifications, which included moving the pendent substituent to the para position on the B ring and replacing the undesired aniline, yielded the much more potent cinnamide 47 (IC50 ¼ 0.044 mM). Despite the improvement in potency, 47 was poorly soluble and showed no oral bioavailability. Twodimensional NMR spectroscopy of 47 in complex with 15N-labelled aL I-domain demonstrated that this series of compounds also binds to the I-domain allosteric site. This result set up the program to use fragment-based NMR screening to identify suitable hydrophilic replacements for the A ring that might impart better physical properties and potency (Liu et al., 2001). Compound 48 was prepared as a fragment lacking the A ring of 47. A set of 2500 low molecular weight molecules (<150 Da) was then tested by the NMR for binding to the I-domain in the presence of saturating concentrations of 48. Among the several classes of heterocycles that were found to bind to the A ring site with Kd values in the sub-millimolar–millimolar range, the authors highlight four in particular (49 a–d). The success of the approach was demonstrated by grafting the indole or benzodioxane onto the fragment to yield compounds such as 50, which retained aLpotency, had increased solubility, and showed the beginnings of oral exposure in the series. Further improvements were achieved by solubilizing modifications at the opposite end of the molecule to yield highly potent compounds such as 51,

108

INTEGRINS

in which the carboxylic acid has a significant effect on solubility and oral bioavailabilty (Pei et al., 2001). Work on the cinnamides culminated in the discovery of 52, which is among the most potent aL I-domain inhibitors and which demonstrated an excellent solubility and pharmacokinetic profile (Winn et al., 2001). Positive results were obtained when compound 52 was evaluated in two different in vivo cellmigration models of inflammation that are known to be aLb2/ICAM-1 dependent. Eosinophilia in a murine model of allergen-induced pulmonary inflammation was significantly inhibited by doses down to 1 mg/kg and neutrophil recruitment was significantly blocked by 100 mg/kg in a staphylococcus enterotoxin A-induced neutrophil trafficking model in rats. Finally, researchers describe the discovery of cyclopropyl analogues of the cinnamides, typified by 53 (IC50 ¼ 0.009 mM), which were prepared as a way to overcome potential issues associated with isomerization and degradation of the cinnamides (Link et al., 2001). Although this issue had been observed on incubation of one compound with rat or human liver microsomes, it is reported that modification of the cinnamide ultimately proved unnecessary. To date, reports of IDAS inhibitors are restricted to aLb2. It appears from the available I-domain X-ray crystal structures that this site may be unique to aLb2 and it remains to be seen if this mode of inhibition will be discovered for other I-domain integrins. Whereas the origins of the I-domain antagonists described above lie in highthroughput screening, the first and most prominent example of an a/b I-like allosteric inhibitor was discovered by rational design based on the X-ray crystal structure of aLb2’s recognition epitope in ICAM-1 (Gadek et al., 2002). Mutagenesis data on the binding of ICAM-1 to aLb2 established the side-chain functionalities of residues Glu 34, Lys 39, Met 64, Tyr 66, Asn 68, and Gln 73 to be the key determinants of the recognition epitope. Unusually for integrin ligands, these residues are not close together in the primary sequence, but on inspection of an X-ray crystal structure of the first domain of ICAM-1 (e.g., Shimaoka et al., 2003), they are found to be present sufficiently in 3D proximity in the tertiary structure to be viable targets for a small molecule mimetic (Fig. 4.18a). Glu 34 and Lys 39 were known to be important in the epitope and researchers at Genentech chose to focus initial lead discovery efforts in this region, reasoning that an RXD peptide sequence might meet the required structure and shape criteria to span the two residues. In fact, kistrin, an RGD-containing disintegrin, was found to block aLb2/ICAM-1 binding (IC50 ¼ 0.7 mM). Building cyclic peptides based on the RGDMP binding motif in kistrin and exploring their SAR led to the disulfide linked inhibitor 54 (IC50 ¼ 1.6 mM) in which the Arg has been removed and a metatyrosine introduced. The NMR and modeling studies on this peptide’s conformation provided a structural link to the ICAM-1 epitope. The intended translation of this peptidic information into a small-molecule template was serendipitously overtaken by the discovery, through using aLb2/ICAM-1 as a counter-screen for a different project, that the tryptophan derivative 55 was equipotent to 54 (IC50 ¼ 1.4 mM). Structural similarities based on overlays of 54 and 55 prompted the synthesis of 56, in which the phenolic substituent at the 4-position of the bromobenzoyl group was designed to mimic

INTEGRIN INHIBITOR DISCOVERY

109

FIGURE 4.18 (a) The aLb2 binding epitope on ICAM-1. (b) a/b I-like allosteric aLb2 inhibitors. (See the color version of this figure in the Color Plates section.)

the metatyrosine in the cyclic peptide. Its significant increase in potency (IC50 ¼ 0.047mM) vindicated the approach and further optimization by sequential variation of each of the molecule’s component fragments yielded the highly potent 57 (IC50 ¼ 0.0014 mM). Subsequent publications provided further SAR information on the original template (55), among which was confirmation of the absolute requirement for a carboxylic acid to mimic Glu 34 (Burdick et al., 2003, 2004). In addition to inhibiting potently the isolated protein assay, 57 was active in a cell-based, mixed lymphocyte reaction assay (IC50 ¼ 0.003 mM) and was as effective as an anti-aL antibody when administered continuously over three days in an in vivo mouse contact-hypersensitivity model. Based on its design origins, compound 57 and its analogues were assumed to bind to the I-domain at the ICAM1 recognition MIDAS region and the absence of competition between the IDAS inhibitors and 57 in aLb2 binding supported this. However, a detailed biochemical and biophysical analyses of the binding mode of action built up a body of evidence to show that compounds such as 57 bind across the a/b interface at the I-like domain MIDAS and act allosterically, resulting in the widely accepted model depicted in Fig. 4.3 (Welzenbach et al., 2002; Shimaoka et al., 2003). For example, the compounds were able to bind aLb2 in which the I-domain was deleted and they did not block the binding of ICAM-1 to isolated I-domain. Interestingly, authors of the original paper describing the discovery of 57 have more recently published contradictory evidence that these compounds, exemplified by 57 and a close analogue, 58, do indeed compete directly with ICAM-1 for binding to a high affinity site

110

INTEGRINS

believed to be on the I-domain of aLb2 (Keating et al., 2006). Crystallographic data on a complex of 57, or an analogue, bound to aLb2 may eventually resolve these apparently conflicting data. 4.2.5

Protein Structure-Based Design

Protein structural information has been an influential component of inhibitor design since the earliest days of integrin drug discovery and several examples have already been quoted. Initially, research was generally ligand focused and in the absence of X-ray or NMR structures, the active conformations of the ligand required for integrin binding were often inferred from conformational analysis of small, frequently cyclic, peptides derived from the natural ligand sequences, as has been described in the peptidomimetic section. An early example of an X-ray structure approach to inhibitor discovery described the construction of an artificial immunoglobulin protein containing the RIPRGDMP motif from the snake venom disintegrin, kistrin, in an exposed loop region. The structure of the RGD motif on the globular scaffolding protein, confirmed to be in its bioactive conformation by its ability to inhibit potently fibrinogen binding to aIIbb3, was determined by X-ray crystallography. Conformational information from the scaffold peptide was used to define a search of a database of small molecules, from which several known peptides and peptidomimetic aIIbb3 antagonists were identified, thus providing validation of this discovery approach (Zhao et al., 1995). In addition to the RGD motif in fibrinogen’s a-chain, aIIbb3 also binds a Lys–Gln–Ala–Gly–Asp (KQAGD) sequence found in its g-chain and mimetics of this sequence have been targeted in the hope of achieving antithrombotic agents with selectivity over other RGD-recognizing integrins. Information from the NMR studies on the C-terminal peptide of the g-chain containing the KQAGD sequence, which indicated a turn in this region, was used in the design of small molecules, which would mimic the Lys and Asp, resulting in the discovery of the nipecotamide 59 (Fig. 4.19) that inhibited fibrinogen/aIIbb3 H

NH

O CO2H

N H

H2N

O

N H

N H

H N

H N

N

CO2H

O

H N

H N

N H

O

CO2H

O

N O

61 αvβ3 IC50= 8 nM αvβ5 IC50= 5170 nM

60 αvβ3 IC50= 40 nM αvβ5 IC50= 26 nM

NH 59

O N O N

N

O

O

CBZ N H

CBZ

CO2H HN

N

62 α5β1 IC50= 0.18 nM αvβ3 IC50= 49 nM

FIGURE 4.19

N H

SO2 N

NH

O

CO2H HN

63 α5β1 IC50= 3.7 nM αvβ3 IC50= 16 nM

O

N H

SO2 N

NH

O N

OMe

CO2H HN

NH

64 α5β1 IC50= 0.55 nM αvβ3 IC50= 3300 nM

Protein structure-based design-related integrin inhibitors.

O

INTEGRIN INHIBITOR DISCOVERY

111

binding (IC50 ¼ 0.009 mM) while having no effect on vitronectin/avb3 binding up to 100 mM (Hoekstra et al., 1995). Despite the apparent structural similarity ˚ ) distance between 59 and many of the known RGD mimetics, the short (<10 A between the negative and positive charges in 59 compared to the corresponding dis˚ ) in RGD mimetics is taken as evidence that 59 is a g-chain mimetic. tance (>12 A Complementary to these structure-based design examples for ligands containing a linear, contiguous recognition sequence is the previously described discovery of the glutamate-based aLb2 inhibitors using the crystal structure of the discontinuous epitope in ICAM-1 (Section 4.2.4). The X-ray- and NMR-derived structure information on the I-domain of aLb2 in complex with small-molecule antagonists has played an important part in defining these inhibitors’ binding sites and mode of action as well as in guiding their subsequent optimization. However, with the exception of the discovery of the lovastatin-mimicking diazepanes (e.g., 35 and 36, Fig. 4.15), there have been no significant literature reports on the use of the I-domain structure in de novo design. In addition to contributing to the understanding of integrin structure, function, and small-molecule antagonist binding modes, the publication of the avb3 crystal structure, alone and in complex with a cyclic RGD peptide (Xiong et al., 2001, 2002), laid the foundation for a range of structure-based activities. The unliganded avb3 structure was used in docking studies to develop a binding model for nonpeptide inhibitors (Feuston et al., 2002) and in the development of a homology model of aIIbb3 (Feuston et al., 2003). The latter was used to build a docking model in which, for example, the Arg–Asp distance in inhibitors was confirmed to be greater for aIIbb3 and which was consistent with biological activity data for a set of aIIbb3/avb3 dual and selective inhibitors. In a similar vein, Kessler used the avb3/cyclic peptide crystal structure to build models of a range of avb3 inhibitors bound into the active site. An analysis of the key interactions between the inhibitors and the integrin allowed the refinement of an earlier peptide-derived pharmacophore with the intention of using it in enhanced inhibitor design (Marinelli et al., 2003). The information gained from structural studies has proved valuable in understanding antagonist selectivity, especially among closely related integrins. Docking studies with avb3/avb5 dual active and avb3 selective compounds into an avb3derived homology model of avb5 were used to rationalize the biological results. The ligand binding site in avb5 was found to be partly obscured by a specificity determining loop compared to the same region in avb3. Consequently, whereas compounds such as the equipotent 60 can bind to both integrins, close analogues such as 61 with bulky substituents near the carboxylate are hindered from binding to avb5 (Marinelli et al., 2004). Rational inhibitor design based on protein structure is an established and proven discovery method that has been applied across a wide range of protein types. Unfortunately, as is the case with integrins, crystal structures of all potential targets are not necessarily available. However, in the absence of the target protein structure, homology and pharmacophore models that have been validated as described above can be used with confidence in the rational design of novel, selective inhibitors. The fibronectin receptor a5b1 plays an important part in angiogenesis

112

INTEGRINS

and its involvement in the vascularization of growing tumors has stimulated interest in it as an oncology target. In a similar manner to that described above, a homology and docking model for a5b1 in complex with a known inhibitor 62, which has some selectivity toward a5b1 over avb3, was constructed and a detailed analysis of exploitable differences between the binding sites of a5b1 and avb3 was carried out (Marinelli et al., 2005). A virtual screening of a combinatorial library of compounds containing the 2-aminopyridine, CBZ and carboxylate fragments found in 62 and linked by trifunctional cores, followed by synthesis of high scoring structures, identified the potent a5b1 inhibitor 63 (a5b1 IC50 ¼ 0.0037 mM, avb3 selectivity approximately fourfold) (Stragies et al., 2007). Modifications to the structure designed to increase potency and selectivity eventually led to 64 in which both these aims were achieved (a5b1 IC50 ¼ 0.00055 mM, avb3 selectivity 6000-fold). The dramatic increase in selectivity on changing from sulfonamide to amide was rationalized on the basis that a5b1 can accommodate the conformations adopted by both these functionalities, whereas avb3 prefers a conformation that can only be adopted by the more flexible sulphonamide. The most active compounds from this novel series were selected for further development. Reports, such as these, on the use of X-ray-derived integrin structure in the design and discovery of antagonists are emerging in the literature and the results to date augur well for the impact that this will have on the development of new, therapeutically valuable integrin inhibitors.

4.3

CHALLENGES—PAST AND FUTURE

One of the primary challenges facing small-molecule integrin inhibitor discovery is the question of what mechanism to target. Although there are some reports on downregulating ligand or integrin expression from the earliest work onward, the vast majority of effort has gone into finding compounds that bind to the integrin and block its ligand interaction. Choosing to target the integrin rather than the ligand is perhaps not surprising, given the usually large excess of ligand available to the integrin and the combination of many integrins’ ability to bind multiple ligands with some ligands’ ability to bind multiple integrins. To a degree, the greater success observed with anti-integrin antibodies versus antiligand antibodies vindicates this strategy. Despite the wide distribution of protein–protein interactions in many biological processes and the consequent opportunities they present as therapeutic targets, the discovery of small-molecule inhibitors of such targets has generally proved very difficult and has not kept pace with the development of therapeutic antibodies. The integrin–ligand interactions are, however, somewhat atypical in that their short, continuous recognition sequences and the involvement of a Glu– or Asp–cation interaction ‘‘hot-spot’’ allow mimicry by small, hydrophilic drug-sized molecules. Many such peptidomimetic inhibitors have been reported and although achieving an acceptable pharmacokinetic profile in such compounds can present a different set of challenges, especially for zwitterionic compounds, there is a large body of medicinal chemistry literature demonstrating success in this regard.

REFERENCES

113

Although therapeutic monoclonal antibodies to integrins appear set to continue being important in the development of integrin-targeted medicines, despite several small-molecule integrin inhibitors having been discovered, optimized, and progressed into clinical evaluation, tirofiban is still the only approved, marketed example. The challenge for future small-molecule discovery will continue to be the integration and exploitation of integrin knowledge to devise antagonists with the correct drug characteristics and biological profile to become successful medicines. It may be that this will be achieved by alternative approaches such as tackling activation mechanisms or downstream signaling, perhaps by modulating the phosphorylation of the integrins’ cytoplasmic tails. In any case, the continuing growth in knowledge and understanding of integrin structure, function, regulation, and biology should provide a sound foundation to support drug discovery.

REFERENCES Andronati SA, Karaseva TL, Krysko AA, 2004. Peptidomimetics – antagonists of the fibrinogen receptors: molecular design, structures, properties and therapeutic applications. Curr. Med. Chem. 11:1183–1211. Arnout MA, Mahalingam B, Xiong J-P, 2005. Integrin structure, allostery and bidirectional signalling. Annu. Rev. Cell Dev. Biol. 21:381–410. Burdick DJ, Paris K, Weese K, Stanley M, Beresini M, Clark K, McDowell RS, Marsters Jr JC, Gadek TR, 2003. N-benzoyl amino acids as LFA-1/ICAM inhibitors 1: amino acid structure–activity relationship. Bioorg. Med. Chem. Lett. 13:1015–1018. Burdick DJ, Marsters Jr JC, Aliagas-Martin I, Stanley M, Beresini M, Clark K, McDowell RS, Gadek TR, 2004. N-benzoyl amino acids as ICAM /LFA-1 inhibitors. Part 2: structure– activity relationship of the benzoyl moiety. Bioorg. Med. Chem. Lett. 14:2055–2059. Cacciari B, Spalluto G, 2005. Non-peptidic avb3 antagonists: recent developments. Curr. Med. Chem. 12:51–70. Coleman PJ, Brashear KM, Askew BC, Hutchinson JH, McVean CA, Duong LT, Feuston BP, Fernandez-Metzler C, Gentile MA, Hartman GD, Kimmel DB, Leu C-T, Lipfert L, Merkle K, Pennypacker B, Prueksaritanont T, Rodan GA, Wesolowski GA, Rodan SB, Duggan ME, 2004. Nonpeptide avb3 antagonists. Part 11: discovery and preclinical evaluation of potent avb3 antagonists for the prevention and treatment of osteoporosis. J. Med. Chem. 47:4829–4837. Dayam R, Aiello F, Deng J, Wu Y, Garofalo A, Chen X, Neamati N, 2006. Discovery of small molecule integrin avb3 antagonists as novel anticancer agents. J. Med. Chem. 49:4526–4534. Dutta AS, Gormley JJ, Coath M, Hassall L, Hayward CF, Gellert PR, Kittley RS, Alcock PJ, Ferguson R, Halterman T, Jamieson A, Moors JA, Moores JM, Rees A, Wood LJ, Reilly CF, Haworth D, 2000. Potent cyclic peptide inhibitors of VLA-4 (a4b1 integrin)-mediated cell adhesion. Discovery of compounds like cyclo(MePhe-Leu-Asp-Val-D-Arg-D-Arg) (ZD7349) compatible with depot formulation. J. Peptide Sci. 6:398–412. Egbertson MS, Chang CT-C, Duggan ME, Gould RJ, Halczenko W, Hartman GD, Laswell WL, Lynch Jr JJ, Lynch RJ, Manno PD, Naylor AM, Prugh JD, Ramjit DR, Sitko GR, Smith RS, Turchi LM, Zhang G, 1994. Non-peptide fibrinogen receptor antagonists 2.

114

INTEGRINS

Optimisation of a tyrosine template as a mimic for Arg-Gly-Asp. J. Med. Chem. 37:2537–2551. Feuston BP, Culberson JC, Duggan ME, Hartman GD, Leu C-T, Rodan SB, 2002. Binding model for nonpeptide antagonists of avb3 integrin. J. Med. Chem. 45:5640–5648. Feuston BP, Culberson JC, Hartman GD, 2003. Molecular model of the aIIbb3 integrin. J. Med. Chem. 46:5316–5325. Gadek TR, Burdick DJ, McDowell RS, Stanley MS, Marsters Jr, JC, Paris KJ, Oare DA, Reynolds ME, Ladner C, Zioncheck KA, Lee WP, Gribling P, Dennis MS, Skelton NJ, Tumas DB, Clark KR, Keating SM, Beresini MH, Tilley, JW, Presta LG, Bodary SC, 2002. Generation of an LFA-1 antagonist by the transfer of the ICAM-1 immunoregulatory epitope to a small molecule. Science 295:1086–1089. Giblin PA, Lemieux RM, 2006. LFA-1 as a key regulator of immune function: approaches toward the development of LFA-1-based therapeutics. Curr. Pharm. Des. 12:2771–2795. Gottschling D, Boer J, Schuster A, Holzmann B, Kessler H, 2002. Combinatorial and rational strategies to develop nonpeptidic a4b7-integrin antagonists from cyclic peptides. Angew. Chem. Int. Ed. 41:3007–3011. Hanson J, de Laval X, Kolh P, Supuran C, Pirotte B, Dogne J-M, 2003. Update on GPIIb/IIIa antagonists. Expert Opin. Ther. Patents 13:1173–1188. Harada T, Katada J, Tachiki A, Asari T, Iijima K, Uno I, Ojima I, Hayashi Y, 1997. Development of the new potent non-peptide GpIIb/IIIa antagonist NSL-95301 by utilizing combinatorial technique. Bioorg. Med. Chem. Lett. 7:209–212. Hartman GD, Egbertson MS, Halczenko W, Laswell WL, Duggan ME, Smith RL, Naylor AM, Manno PD, Lynch RJ, Zhang G, Chang CT-C, Gould RJ, 1992. Non-peptide fibrinogen receptor antagonists. 1. Discovery and design of exosite inhibitors. J. Med. Chem. 35:4640–4642. Haubner R, Finsinger D, Kessler H, 1997. Stereoisomeric peptide libraries and peptidomimetics for designing selective inhibitors of the avb3 integrin for a new cancer therapy. Angew. Chem. Int. Ed. Engl. 36:1374–1389. Hoekstra WJ, Beavers MP, Andrade-Gordon P, Evangelisto MF, Keane PM, Press JB, Tomko KA, Fan F, Kloczewiak M, Mayo KH, Durkin KA, Liotta DC, 1995. Design and evaluation of nonpeptide fibrinogen g-chain-based GPIIb/IIIa antagonists. J. Med. Chem. 38:1582–1592. Hruby VJ, 2002. Designing peptide receptor agonists and antagonists. Nat. Rev. Drug Discov. 1:847–858. Hummel G, Reinecke U, Reimer U, 2006. Translating peptides into small molecules. Mol. Biosyst. 2:499–508. Humphries JD, Byron A, Humphries MJ, 2006. Integrin ligands at a glance. J. Cell Sci. 119:3901–3903. Hynes RO, 2002. Integrins: bidirectional, allosteric signaling machines. Cell 110:673– 687 Hynes RO, 2004. The emergence of integrins: a personal and historical perspective. Matrix Biol. 23:333–340. Jackson DY, 2002. Alpha 4 integrin antagonists. Curr. Pharm. Des. 8:1229–1253. Kallen J, Welzenbach K, Ramage P, Geyl D, Kriwacki R, Legge G, Cottens S, Weitz-Schmidt G, Hommel U, 1999. Structural basis for LFA-1 inhibition upon lovastatin binding to the CD11a I-domain. J. Mol. Biol. 292:1–9.

REFERENCES

115

Keating SM, Clark KR, Stefanich LD, Arellano F, Edwards CP, Bodary SC, Spencer SA, Gadek TR, Marsters Jr JC, Beresini MH, 2006. Competition between intercellular adhesion molecule-1 and a small-molecule antagonist for a common binding site on the al subunit of lymphocyte function-associated antigen-1. Protein Sci. 15:290–303. Keenan RM, Miller WH, Kwon C, Ali FE, Callahan JF, Calvo RR, Hwang S-M, Kopple KD, Peishoff CE, Samanen JM, Wong AS, Yuan C-K, Huffman WF, 1997. Discovery of potent nonpeptide vitronectin receptor (avb3) antagonists. J. Med. Chem. 40:2289–2292. Kelly TA, Jeanfavre DD, McNeil DW, Woska Jr JR, Reilly PL, Mainolfi EA, Kishimoto KM, Nabozny GH, Zinter R, Bormann B-J, Rothlein R, 1999. Cutting edge: a small molecule antagonist of LFA-1-mediated cell adhesion. J. Immunol. 163:5173–5177. Last-Barney K, Davidson W, Cardozo M, Frye LL, Grygon CA, Hopkins JL, Jeanfavre DD, Pav S, Qian C, Stevenson JM, Tong L, Zindell R, Kelly TA, 2001. Binding site elucidation of hydantoin-based antagonists of LFA-1 using multidisciplinary technologies: evidence for the allosteric inhibition of a protein–protein interaction. J. Am. Chem. Soc. 123:5643–5650. Liddington RC, 2002. Will the real integrin please stand up? Structure 10:605–607. Lin K-C, Ateeq HS, Hsiung SH, Chong LT, Zimmerman CN, Castro A, Lee W-C, Hammond CE, Kalkunte S, Chen L-L, Pepinsky RB, Leone DR, Sprague AG, Abraham WM, Gill A, Lobb RR, Adams SP, 1999. Selective, tight binding inhibitors of integrin a4b1 that inhibit allergic airway responses. J. Med. Chem. 42:920–934. Lin LS, Lanza Jr T, McCauley E, Van Riper G, Kidambi U, Cao J, Egger LA, Mumford RA, Schmidt JA, MacCoss M, Hagmann WK, 2002. Specific and dual antagonists of a4b1 and a4b7 integrins. Bioorg. Med. Chem. Lett. 12:133–136. Link JT, Sorenson B, Liu G, Pei Z, Reilly EB, Leitza S, Okasinski G, 2001. Discovery and SAR of diarylsulfide cyclopropylamide LFA-1/ICAM-1 interaction antagonists. Bioorg. Med. Chem. Lett. 11:973–976. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ, 1997. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Delivery Rev. 23:3–25. Liu G, Link JT, Pei Z, Reilly EB, Leitza S, Nguyen B, Marsh KC, Okasinski GF, von Gelrern TW, Ormes M, Fowler K, Gallatin M, 2000. Discovery of novel p-arylthio cinnamides as antagonists of Leukocyte Function-Associated Antigen-1/Intracellular Adhesion Molecule-1 interaction. 1. Identification of an additional binding pocket based on an anilino diaryl sulfide lead. J. Med. Chem. 43:4025–4040. Liu G, Huth JR, Olejniczak ET, Mendoza R, DeVries P, Leitza S, Reilly EB, Okasinski GF, Fesik SW, von Geldern TW, 2001. Novel p-arylthio cinnamides as antagonists of leukocyte function-associated antigen-1/intracellular adhesion molecule-1 interaction. 2. Mechanism of inhibition and structure-based improvement of pharmaceutical properties. J. Med Chem. 44:1202–1210. Luo B-H, Carman CV, Springer TA, 2007. Structural basis of integrin regulation and signaling. Annu. Rev. Immunol. 25:619–647. Marinelli L, Lavecchia A, Gottschalk K-E, Novellino E, Kessler H, 2003. Docking studies on avb3 integrin ligands: pharmacophore refinement and implications for drug design. J. Med. Chem. 46:4393–4404. Marinelli L, Gottschalk K-E, Meyer A, Novellino E, Kessler, 2004. Human integrin avb5: homology modeling and ligand binding. J. Med. Chem. 47:4166–4177.

116

INTEGRINS

Marinelli L, Meyer A, Heckmann D, Lavecchia A, Novellino E, Kessler H, 2005. Ligand binding analysis for human a5b1 integrin: strategies for designing new a5b1 integrin antagonists. J. Med. Chem. 48:4204–4207. Miller WH, Keenan RM, Willette RN, Lark MW, 2000. Identification and in vivo efficacy of small-molecule antagonists of avb3 (the vitronectin receptor). Drug Discov. Today 5:397–408. Pei Z, Zhili X, Liu G, Li Y, Reilly EB, Lubbers NL, Huth JR, Link JT, von Geldern TW, Cox BF, Leitza S, Gao Y, Marsh KC, DeVries P, Okasinski GF, 2001. Discovery of potent antagonists of leukocyte function-associated antigen-1/intercellular adhesion molecule-1 interaction. 3. Amide (C-ring) structure–activity relationship and improvement of overall properties of arylthio cinnamides. J. Med Chem. 44:2913–2920. Potin D, Launay M, Nicolai E, Fabreguette M, Malabre P, Caussade F, Besse D, Skala S, Stetsko DK, Todderud G, Beno BR, Cheney DL, Chang CJ, Sheriff S, Hollenbaugh DL, Barrish JC, Iwanowicz EJ, Suchard SJ, Dhar TGM, 2005. De novo design, synthesis, and in vitro activity of LFA-1 antagonists based on a bicyclic[5.5]hydantoin scaffold. Bioorg. Med. Chem. Lett. 15:1161–1164. Potin D, Launay M, Monatlik F, Malabre P, Fabreguettes M, Fouquet A, Maillet M, Nicolai E, Dorgeret L, Chevallier F, Besse D, Dufort M, Caussade F, Ahmad SZ, Stetsko DK, Skala S, Davis PM, Balimane P, Patel K,Yang Z, Marathe P, Postelneck J, Townsend RM, Goldfarb V, Sheriff S, Einspahr H, Kish K, Malley MF, DiMarco JD, Gougoutas JZ, Kadiyala P, Cheney DL, Tejwani RW, Murphy DK, Mcintyre KW, Yang X, Chao S, Leith L, Xiao Z, Mathur A, Chen B-C, Wu D-R, Traeger SC, McKinnon M, Barrish JC, Robl JA, Iwanowicz EJ, Suchard SJ, Dhar TGM, 2006. Discovery and development of 5-[(5S,9R)-9(4-cyanophenyl)-3-(3,5-dichlorophenyl)-1-methyl-2,4-dioxo-1,3,7-triazaspiro[4.4]non-7yl-methyl]-3-thiophenecarboxylic acid (BMS-587101) – a small molecule antagonist of leucocyte function-associated antigen-1. J. Med. Chem. 49:6946–6949. Scarborough RM, 1998. Eptifibatide. Drugs of the Future 23:585–590. Scarborough RM, Naughton MA, Teng W, Rose JW, Phillips DR, Nannizzi L, Arfsten A, Campbell AM, Charo IF, 1993. Design of potent and specific integrin antagonists. Peptide antagonists with high specificity for glycoprotein IIb-IIIa. J. Biol. Chem. 268:1066–1073. Shimaoka M, Springer TA, 2003. Therapeutic antagonists and conformational regulation of integrin function. Nat. Rev. Drug Discov. 2:703–716. Shimaoka M, Salas A, Yang W, Weitz-Schmidt G, Springer TA, 2003. Small molecule integrin antagonists that bind to the b2 subunit I-like domain and activate signals in one direction and block them in another. Immunity 19:391–402. Shimaoka M, Xiao T, Liu J-H, Yang Y, Dong Y, Jun C-D, McCormack A, Zhang R, Joachimiak A, Takagi J, Wang J-H, Springer TA, 2003. Structures of the alpha L I domain and its complex with ICAM-1 reveal a shape-shifting pathway for integrin regulation. Cell 112:99–111. Simmons DL, 2005. Anti-adhesion therapies. Curr. Opin. Pharmacol. 5:398–404. Singh J, van Vlijmen H, Liao Y, Lee WC, Cornebise M, Harris M, Shu I, Gill A, Cuervo JH, Abraham WM, Adams SP, 2002. Identification of potent and novel a4b1 antagonists using in silico screening. J. Med. Chem. 45:2988–2993. Singh J, Adams S, Carter MB, Cuervo H, Lee W-C, Lobb RR, Pepinsky RB, Petter R, Scott D, 2004. Rational design of potent and selective VLA-4 inhibitors and their utility in the treatment of asthma. Curr. Top. in Med. Chem. 4:1497–1507.

REFERENCES

117

Smith JW, 2003. Cilengitide. Curr. Opin. in Investig. Drugs 4:741–745. Staunton DE, Lupher ML, Liddington R, Gallatin WM, 2006. Targeting integrin structure and function in disease. Adv. Immunol. 91:111–157. Stragies R, Osterkamp F, Zischinsky G, Vossmeyer D, Kalkhof H, Reimer U, Zahn G, 2007. Design and synthesis of a new class of selective integrin a5b1 antagonists. J. Med. Chem. 50:3786–3794. Takada Y, Ye X, Simon S, 2007. The integrins. Genome Biol. 8:215.1–215.9. Vanderslice P, Woodside DG, 2006. Integrin antagonists as therapeutics for inflammatory diseases. Expert Opin. Investig. Drugs 15:1235–1255. Wattanasin S, Albert R, Ehrhardt C, Roche D, Sabio M, Hommel U, Welzenbach K, WeitzSchmidt G, 2003. 1,4-Diazepane-2-ones as novel inhibitors of LFA-1. Bioorg. Med. Chem. Lett. 13:499–502. Wattanasin S, Kallen J, Myers S, Guo, Q, Sabio M, Ehrhardt C, Albert R, Hommel U, Weckbecker G, Welzenbach K, Weitz-Schmidt G, 2005. 1,4-diazepane-2,5-diones as novel inhibitors of LFA-1. Bioorg. Med. Chem. Lett. 15:1217–1220. Weide T, Modlinger A, Kessler H, 2007. Spatial screening for the identification of the bioactive conformation of integrin ligands. Top. Curr. Chem. 272:1–50. Weitz-Schmidt G, Welzenbach K, Brinkmann V, Kamata T, Kallen J, Bruns C, Cottens S, Takada Y, Hommel U, 2001. Statins selectively inhibit leukocyte function antigen-1 by binding to a novel regulatory integrin site. Nat. Med. 7:687–692. Weitz-Schmidt G, Welzenbach K, Dawson J, Kallen J, 2004. Improved lymphocyte functionassociated antigen-1 (LFA-1) inhibition by statin derivatives. J. Biol. Chem. 279:46764– 46771. Welzenbach K, Hommel U, Weitz-Schmidt G, 2002. Small molecule inhibitors induce conformational changes in the I-domain and the I-like domain of lymphocyte functionassociated antigen-1. J. Biol. Chem. 277:10590–10598. Winn M, Reilly EB, Liu G, Huth JR, Jae H-S, Freeman J, Pei Z, Zhili X, Lynch J, Kester J, von Geldern TW, Leitza S, DeVries P, Dickinson R, Mussatto D, Okasinski GF, 2001. Discovery of novel p-arylthio cinnamides as antagonists of leukocyte function-associated antigen-1/intercellular adhesion molecule-1 interaction. 4. Structure–activity relationship of substituents on the benzene ring of the cinnamide. J. Med. Chem. 44:4393–4403. Witherington J, Blaney EL, Bordas V, Elliott RL, Gaiba A, Garton N, Green PM, Naylor A, Smith DG, Spalding DJ, Takle AK, Ward RW, 2006. Pyridone derivatives as potent, orally bioavailable VLA-4 integrin antagonists. Bioorg. Med. Chem. Lett. 16:5538– 5541. Wu J-P, Emeigh J, Gao DA, Goldberg DR, Kuzmich D, Miao C, Potocki I, Qian KC, Sorcek RJ, Jeanfavre DD, Kishimoto K, Mainolfi EA, Nabozny Jr G, Peng C, Reilly P, Rothlein R, Sellati RH, Woska Jr JR, Chen S, Gunn JA, O’Brien D, Norris SH, Kelly TA, 2004. Secondgeneration lymphocyte function-associated antigen-1 inhibitors: 1H-imidazo[1,2-a]imidazol-2-one derivatives. J. Med. Chem. 47:5356–5366. Xiao T, Takagi J, Coller BS, Wang J-H, Springer TA, 2004. Structural basis for allostery in integrins and binding to fibrinogen-mimetic therapeutics. Nature 432:59–67. Xiong J-P, Stehle T, Diefenbach B, Zhang, R, Dunker R, Scott DL, Joachimiak A, Goodman SL, Arnout, MA, 2001. Crystal structure of the extra-cellular segment of integrin avb3. Science 294:339–345.

118

INTEGRINS

Xiong J-P, Stehle T, Zhang R, Joachimiak A, Frech M, Goodman SL, Arnout MA, 2002. Crystal structure of the extra-cellular segment of integrin avb3 in complex with an Arg-Gly-Asp ligand. Science 296:151–155. Xue C-B, Rafalaski M, Roderick J, Eyermann CJ, Mousa S, Olsen RE, DeGrado WF, 1996. Design, synthesis and in vitro activities of a series of benzimidazole/benzoxazole glycoprotein IIb/IIIa inhibitors. Bioorg. Med. Chem. Lett. 6:339–344. Yang GX, Hagmann WK, 2003. VLA-4 antagonists: potent inhibitors of lymphocyte migration. Med. Res. Rev. 23:369–392. Zhao B, Helms LR, DesJarlais RL, Abdel-Meguid SS, Wetzel R, 1995. A paradigm for drug discovery using a conformation from the crystal structure of a presentation scaffold. Nat. Struct. Biol. 2:1131–1137.

5 STRATEGIES FOR DISCOVERING KINASE DRUGS JERRY L. ADAMS, PAUL BAMBOROUGH, DAVID H. DREWRY, AND LISA SHEWCHUK

5.1

INTRODUCTION

Protein kinases transfer the terminal phosphate of ATP to the hydroxyl of a serine, threonine, or tyrosine residue of the acceptor protein substrate. These phosphorylated proteins are key components in signaling cascades that regulate all aspects of cellular function. Protein kinases can be broadly classified as receptor (e.g., EGFR, c-ErbB2, PDGFR, VEGFR2) or nonreceptor (e.g., c-src, b-raf, ZAP70) kinases and by the amino acid that they phosphorylate as tyrosine or serine/threonine. In addition to several conserved regions, such as the ATP pocket and peptide substrate binding region, protein kinases may also contain additional regulatory domains (such as the cyclin binding site of the cyclin-dependent kinases (CDKs) and protein docking sites (SH2 and SH3 domains; Fedi and Aaronson, 2000). All of these offer potential intervention points for the design of drugs to interrupt kinase signaling, and recent developments have highlighted how the complexity of kinase structure and regulation can be exploited in drug design. Novel strategies include targeting single or multiple binding sites, the inactive conformation of the kinase domain, allosteric sites, regulatory domains, and signaling accomplices, such as heat shock proteins. Using the currently available crystal structures and protein homology modeling paired with chemical informatic approaches, it is now reasonable to speculate that the entire human kinome of
500 kinases could someday have ligands identified for every relevant kinase drug target or a combination of targets.

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

119

120

STRATEGIES FOR DISCOVERING KINASE DRUGS

Working in the kinome offers a plethora of viable, druggable targets due to the fact that they are involved in virtually every regulatable cellular process (Manning et al., 2002). The corollary to this opportunity, of course, is the complexity of the selectivity needed to achieve efficacy with minimal anticipated side effects. The targeted design of selective kinase inhibitors, which began in the 1980s, is a rapidly maturing area of drug discovery. Protein kinases now figure prominently among the molecular targets currently being pursued by the pharmaceutical industry to treat a wide variety of diseases, including cancer, diabetes, rheumatoid arthritis, hypertension, and stroke. The inhibition of the aberrant activity has been demonstrated in a variety of drug discovery efforts: cancer, asthma, psoriasis, and inflammation to name a few (Cohen, 2002). The ATP site remains the most important and tractable small-molecule inhibitor binding site. While the ATP site is conserved, it is not optimized for ATP binding, thus allowing small molecules to exploit the unique features of the ATP site to achieve significant potency and selectivity. Finding small molecules that bind to this site has been relatively easy; however, these compounds have potential drawbacks. The human genome contains >500 kinases plus many other ATP binding proteins such as ATPases, motor proteins, and ATP-gated ion channels. Because ATP is such a widely used cofactor, obtaining selectivity for one or a few kinases over the vast number of enzymes that have evolved to bind ATP is not a trivial challenge. Initially, these obstacles seemed to make the discovery of selective, potent inhibitors of the kinase ATP site an intractable problem. A detailed description of the ATP site, the identification of chemistry starting points, and some success stories provide insight into strategies used for tackling these hurdles.

5.2

PROTEIN KINASE STRUCTURAL FEATURES

The first protein kinase crystal structure, that of mouse cyclic AMP-dependent kinase (also known as protein kinase A or PKA), was solved in 1991 (Knighton et al., 1991). Since then a vast amount of structural data has been reported. The protein structure database contains >600 kinase entries that are unevenly distributed over the kinome, with the majority concentrating on a few well-established kinases binding to different ligands. Excluding orthologues, over 80 unique kinase structures are available, with more than half of these being solved in the past 2 years, due in part to the major structural genomics efforts. The generality of the observations from the first PKA crystal structure has since been seen in other systems. The common features of the kinase domain will now be described, with emphasis on the features that are important for ligand binding and structure-based design. The kinase domain consists of a continuous stretch of
300 amino acids that can be divided into two globular subdomains or lobes as shown in Fig. 5.1. The N-terminal lobe is smaller, containing
80 amino acids, and is separated from the
200amino acid C-terminal lobe by a short flexible linker, also referred to as the hinge. Direct contacts between the two lobes are relatively small, allowing space for a deep pocket, in which ATP binds. In larger proteins, additional sequences either

PROTEIN KINASE STRUCTURAL FEATURES

121

FIGURE 5.1 (a) Kinase sequence alignment highlighting key structural features. Residues close to ATP are underlined. (b) Ribbon representation of PKA using the same color scheme as (a). (c) Interactions of ATP in the binding site of PKA. (See the color version of this figure in the Color Plates section.)

N-terminal or C-terminal to the canonical kinase domain often fold back and pack against the outside of either lobe. The minimal N-terminal lobe consists of a five-stranded antiparallel beta sheet and one alpha helix (the C-helix). It contains a signature motif (the glycine-rich loop) that is used by gene annotation programs to classify a sequence as a protein

122

STRATEGIES FOR DISCOVERING KINASE DRUGS

kinase. This motif, located
10 residues from the N-terminus, is defined by the sequence GXGXXG, where X is any amino acid. The glycine-rich loop forms a turn between two beta strands and lies over ATP during catalysis. The requirement for glycines implies a need for flexibility. This loop is frequently disordered in unliganded crystal structures and shows a wide variation in conformation in different inhibitor cocrystal structures. The N-terminal lobe also contributes two perfectly conserved charged residues to the catalytic cleft, which can be used to align kinase sequences reliably (Lys72 and Glu91 in PKA). The conserved lysine provides an unambiguous marker of the middle strand of the beta sheet, while the glutamic acid is located at the center of the C-helix. These two residues form a salt bridge, coordinating the a- and b-phosphates of ATP and orienting them for catalytic transfer of the g-phosphate. The C-terminal lobe is mainly helical, with at least six major alpha helices and only a minimal amount of sheet. Sequence alignment of the C-terminal lobe is often problematic: however, a number of common features exist and can be used when building homology models. As in the N-terminal lobe, a number of charged residues in the catalytic cleft are highly conserved. In PKA, for example, the side chains of Asn171 and Asp184 bind the magnesium counter ions required for ATP catalysis. Asp184 is also part of another conserved kinase motif, the ‘‘DFG motif,’’ which marks the beginning of the activation loop and will be discussed below. Two additional conserved residues form a salt bridge buried within the C- terminal lobe (Glu208 and Arg280 in PKA). While this salt bridge has no known function, it may be required for stability of the C-terminal fold. Glu208 is part of the ‘‘APE’’ motif that typically marks the end of the activation loop. The conserved arginine is useful in locating the end of the kinase domain since it is positioned just before the final alpha helix of the C-terminal lobe. The activation loop is a stretch of residues in the C-terminal lobe located between the DFG and APE motifs. It is variable in length and often plays a role in activation of catalytic activity. In cases where phosphorylation of this loop leads to activation, this is mediated by conformational changes, frequently involving the interaction of the phosphorylated residue with positively charged side chains on the activation loop, on the C-helix or on the C-terminal lobe. This loop is also very flexible and is often partially disordered in crystal structures. Kinases adopt a variety of conformations, depending on their activation state and the presence of ligands. The conformation observed in the PKA/ATP/peptide complex, and in other protein kinases bound to ATP or analogues, is often referred to as an active conformation. This is a conformation where the active site residues are correctly positioned to catalyze the transfer of the g-phosphate of ATP. Kinases in inactive states can adopt a range of different conformations. In these cases, the activation loop often adopts a conformation that is incompatible with ATP and/or substrate binding, with rearrangement of one or more of the catalytic residues. In the inactive state of cyclin-dependent kinase 2, the C-helix is rotated away from the ATP site, breaking the key salt bridge between the C-helix glutamic acid and the conserved lysine. Binding of its activator, cyclin A, causes the C-helix to move back into the ATP site, restoring this critical Glu/Lys interaction.

123

PROTEIN KINASE STRUCTURAL FEATURES

In early crystal structures of PKA, ternary complexes with ATP and substrate were solved, which still provide the best model for explaining their interactions. Because the ATP binding site is largely flat, occurring at the interface between the N- and C-terminal lobes, it is often represented as a two-dimensional cartoon, usually viewed from the N-terminal lobe down toward the C-terminal lobe as shown in Fig. 5.2. Three-dimensional versions are shown in Fig. 5.1c. The ATP binding site is centered on the position of the adenine ring. Aliphatic, hydrophobic side chains lie above and below the plane of the adenine, forming a lipophilic sandwich. In PKA, two highly conserved alkyl side chains (Ala57 and Val70) drop down from the N-terminal lobe to contact the adenine. From the C-terminal lobe, interactions are made with Leu173 and Thr183. This flat, lipophilic environment creates a strong preference for binding aromatic rings. The hinge of the kinase lies along one edge of the adenine pocket. Its extended conformation leads to a situation where alternating H-bonding donors and acceptors point toward the edge of the adenine. Reading along the protein sequence, the backbone carbonyl of Glu121 in PKA accepts a hydrogen bond from the adenine NH2 group. This is the likely reason that most protein kinases bind ATP but not GTP. The backbone NH of the following residue, Val123, donates a hydrogen bond to the N1 atom of the adenine. The backbone carbonyl of the same residue, Val123, also points into the ATP site, though this potential hydrogen bonding group is unsatisfied by ATP. In an apo kinase structure, the site will be filled with water molecules, which will fulfill the hydrogen bonding potential of these three hinge residues. However, because these waters are located in an otherwise highly lipophilic environment, it is beneficial to displace them with a hydrophobic group that can still fulfill the Val123 Hinge

R

H N

N H

O

Glu121

O

H N R

NH O

O

NH2 Met120

Outer lipophilic pocket Surface site

Solvent

N

1

N

HO

N

Gatekeeper NH2 Adenine N pocket

Backpocket

Lys72 HO Sugar pocket

DFG pocket

O O O O O O O P P P O O O Phosphate site O

O

F of DFG motif

FIGURE 5.2 Schematic view of the ATP binding site subpockets. Residues shown are from PKA. In sections 1-5, compounds shown are drawn in this orientation, and atoms donating the H-bond interaction to the hinge are marked with asterisks (*).

124

STRATEGIES FOR DISCOVERING KINASE DRUGS

hydrogen bonding interactions. Due to the narrow and flat nature of the pocket, these groups are usually aromatic rings. Another wall of the adenine pocket is formed by the residue that immediately precedes the three hinge residues. In PKA this residue is Met120, while larger aromatic groups (Phe80 in CDK2) or smaller side chains (Thr106 in p38) are found in other kinases. When a large side chain is present, this face of the adenine pocket is blocked. However, in p38, a smaller threonine side chain leaves open volume that inhibitors can access, enabling selectivity to be obtained. For this reason, this residue is referred to as the ‘‘gatekeeper.’’ The pocket behind the gatekeeper is known by a variety of names including the gatekeeper pocket, the selectivity pocket, the back pocket, or the inner lipophilic pocket. This space is usually lined with aliphatic side chains of variable shapes and sizes. Even deeper, below this back pocket, is a region usually occupied by the phenylalanine side chain of the DFG motif. In some kinases, this region is made accessible (the DFG pocket) by conformational changes to the DFG motif that occur upon ligand binding. A third side of the adenine pocket points toward the outer edge of the ATP binding cleft. However, before the solvent interface is reached, inhibitors pass through two additional regions. They first pass below a lipophilic residue on the N-terminal lobe, Leu49 in PKA. This residue is conserved or conservatively replaced by valine in most kinases. Below this residue, the space is often quite narrow due to the continuation of the hinge into a small alpha helix. Therefore, this region prefers to bind flat aromatic or lipophilic groups and is known as the outer lipophilic pocket, or the lipophilic plug. Passing further toward solvent, the helix that immediately follows the hinge presents side chains that can interact with larger inhibitors. Because the side chains of this helix are variable, even in very closely related kinases, they offer many opportunities for selectivity and compose the surface site. Moving out from the adenine pocket, the next subsite encountered is the sugar pocket. As the name implies, this pocket contains a number of groups able to form hydrogen bonds with the ribose ring. In particular, the backbone carbonyl of Glu170 in PKA forms a hydrogen bond to one of the sugar hydroxyls. This carbonyl is usually accessible to interact with inhibitors. Finally, the phosphate region of the ATP site presents a complex picture. Although triphosphate is negatively charged, many of the side chains lining this region are acidic. Asp184 in PKA, for example, is highly conserved since it binds the magnesium counter ions of ATP. Another conserved side chain that interacts with the phosphates and many inhibitors is the conserved lysine, Lys72 in PKA. In view of all the hydrogen bonding potential, it is surprising that few inhibitors interact with this region.

5.3 5.3.1

GENERATING AND OPTIMIZING KINASE INHIBITORS ATP Binding Pocket

The first stage in designing a kinase-directed compound array is to select a core capable of binding in the adenine subpocket, often with a nitrogen containing

GENERATING AND OPTIMIZING KINASE INHIBITORS

125

heterocycle that fulfills the requirement for a hinge-bond acceptor. In addition to the synthetic tractability to functionalize the core, the likely binding modes of the core group must be taken into account to tailor the R-group selection to probe different parts of the binding site. Knowledge of the residues with which each monomer is expected to interact can be incorporated into the selection. For example, on a pyrimidine core, the 2-position is frequently used to access the outer lipophilic pocket and the solvent front, and so should include polar groups. Positions expected to interact with the back of the ATP site could include substituents rich in aromatic and lipophilic groups. Drug design through the use of chemogenomic toolkits has been put into practice and is widely used for kinase compounds (Birault et al., 2006). Kinase inhibitors tend to have certain similar one-dimensional properties and differ from nonkinase compounds (Sprous et al., 2006). Kinase inhibitors have more aromatic carbon and nitrogen atoms than nonkinase compounds, and more complex heterocycles. This stems from the requirement to bind in a flat lipophilic pocket and to make the essential hydrogen bonds. There is no statistical difference in molecular weight, but kinase compounds contain fewer rotatable bonds and saturated atoms (Paolini et al., 2006). 3D pharmacophore methods have been used to build focused screening sets. These methods reduce the interactions between ligand and protein to spatial arrangements of key features such as hydrogen bonds and hydrophobic centers. Pharmacophores can be built from ligand information, either by overlaying multiple compounds or by using fingerprinting methods (Schnur et al., 2004). Protein structural information and SAR, when available, can be used to further refine the model. For protein kinases, the conserved hydrogen bonds to the hinge usually form part of the pharmacophore model, but other features can be included to improve potency or selectivity. Published examples of successful lead hopping using pharmacophore models include examples for ALK5 (Singh et al., 2003), CHK1 (Lyne et al., 2004), EGFR (Traxler et al., 1997), IKK2 (Morwick et al., 2006), and p38 (Angell et al., 2008). Other virtual screening approaches have been used, for example, docking, as discussed in a review article (McInnes, 2006). Examples of the successful use of docking to identify kinase hits have been published and include CDK2 (Wu et al., 2003), CK2 (Vangrevelinghe et al., 2003), CHK1 (Foloppe et al., 2006), and ALK (Li et al., 2006). De novo design, designing a kinase inhibitor for a particular kinase target from scratch using structural information and lessons from other kinase inhibitors, has also been used to advance the field (Honma et al., 2001a). The opposite approach of nonselective kinase inhibitors has also been investigated. A pharmacophore model has been built using crystal structure of four structurally distinct ‘‘frequent hitter’’ kinase inhibitors and was used to predict unselective classes of protein kinase compounds (Aronov and Murcko, 2004). The authors found that removing one of the common features in the pharmacophore enabled them to improve the selectivity of their lead molecules. Fragment-based screens (small molecular weight compounds) have been used to identify hits, most notably for the MAP kinases. NMR has been used to screen a SHAPES library of drug-like fragments as well as weakly binding p38 inhibitor fragments such as 1 and for JNK3, 2, shown in Fig. 5.3 (Fejzo et al., 1999, 2003). Initial hits were confirmed by competition with a covalent ATP site binder. Commercial analogues,

126

STRATEGIES FOR DISCOVERING KINASE DRUGS SB-203580 Literature p38 inhibitor

SHAPES JNK3 hit (Fejzo, 2003)

SHAPES p38 hit (Kd 2 mM) (Fejzo, 1999)

Optimized JNK3 Isoxazole K i = 0.79 µM

(a) *

*

N

*

N

*

H N

Me

N

O

N N

F HN

HN N

Me

HN N

N

1

O N

2

3

O S

*

(b)

*

N H2N

N

N Cl

NH2

*

N

5

CDK2 fragment

p38 fragments and optimized compounds (Gill, 2004)

*

N

N

Cl

H N

H N

O

O

4

*

NH2

O

NH O

7

NH

F

N O

F

N

6

O

8

FIGURE 5.3 (a) Fragments of JNK3 and p38 inhibitors reported to be active from the SHAPES library by NMR screening. (b) CDK2 and p38 fragments found by crystallographic screening and their optimized molecules. Atoms donating the H-bond interaction to the hinge are marked with asterisks (*).

containing combinations of the primary hit fragments as well as molecules chosen by docking, were purchased and screened. Thiazoles, uracils and isoxazole hits were among those found in the follow-up set, two of which were submicromolar in potency demonstrated by 3 with a Ki value for JNK3 of 0.79 mM. The isoxazoles were later optimized to sub-10 nM potency with the help of crystallography. p38 and CDK2 hits have also been found by using crystallography as a screening tool (Hartshorn et al., 2005). Two libraries were screened; a drug-like fragment set and a targeted kinase set. Although the hit rates of each were not disclosed, hits included a 350 mM pyrazine fragment 4 for CDK2, a 1300 mM aminopyridine 5, and a 35 mM indole 7 for p38 (Fig. 5.3). Further optimization of these fragments using crystallography-guided compound hybridization made use of the DFG-out conformation to generate much larger compounds such as 6 and 8 (Gill, 2004). One millimolar CDK2 fragment has since been optimized to <10 nM. The successful use of fragment screening for kinases shows that this is a powerful alternative way to discover hits. Most published examples of the fragment screening approach have, however, belonged to kinase targets (JNK3, p38, CDK2, and Aurora A) with high hit rates in conventional screening. It remains to be seen whether this approach will also provide solutions for less tractable kinase targets for which hit identification is the main challenge. While design attempts for one kinase may prove unsuccessful, they occasionally produce surprisingly good results against an unexpected target. In a comparison of

127

GENERATING AND OPTIMIZING KINASE INHIBITORS

the crossover activities of a wide variety of small molecules against many different classes of targets, activity between different protein kinases is one of the most frequently occurring (Paolini et al., 2006). For this reason, we have found it useful to screen kinase-directed compound arrays against multiple kinase targets simultaneously. Organizing kinase lead discovery chemistry and cross-screening in this way ensures that these unpredictable events can be found. In addition, screening a focused set containing compounds that have previously shown to have activity against kinases has proven to be a highly successful strategy to discover hits against new kinases. For example, a potent and selective series of IKK-e inhibitors exemplified by 10 was found by cross-screening a series of PLK1 inhibitors like 9 against a panel of kinases (Bamborough et al., 2006; Lansing et al., 2007). Replacing a carboxamide with a nitrile gave an unexpected reversal of selectivity away from PLK1 in favor of IKK-e as shown in Fig. 5.4. A crystal structure in CDK2 was used as a surrogate for *

N

*

O

N

O

N

(a) N Me

S

S O

O

H N

*

O

12 Cl

O

MeO2S

PLK1 selective inhibitor

IKK-ε selective inhibitor

9

10

*N H N

O

*

HN

N

O

(b)

CN

NH2

N H

*

N

Cl

NH

N

13

*

HN

N N

Cl HN

N

14

11 Initial hit (Takami et al., 2004) showing proposed Cl subpockets

HN Cl

O NH Cl

15

FIGURE 5.4 (a) An IKK-e selective inhibitor discovered by cross-screening compounds active against PLK1 (b) Combinatorial pocket replacements for a ROCK inhibitor. Atoms donating the H-bond interaction to the hinge are marked with asterisks (*).

128

STRATEGIES FOR DISCOVERING KINASE DRUGS

IKK-e to determine a plausible binding mode, and homology models of IKK-e and PLK1 provided a possible explanation for the selectivity difference. One approach for optimization of an initial hit is given in a case study for ROCK summarized in Fig. 5.4 (Takami et al., 2004). The compound 11 found by HTS was relatively weak and did not have reproducible cell activity. At the time of this discovery, a crystal structure of ROCK was not available, and so a homology model was built using PKA. Based on docking, the pyridine nitrogen was proposed to H-bond to the hinge, while the other aromatic group was thought to bind in the sugar pocket. A shortlist of possible bioisosteric replacements was gathered for each of the two subpockets. Using different linkers to connect the two lists, a virtual library was constructed and prioritized by docking. The potency of the pyridine template was improved by substituting the linker with a more planar urea to give 15. Scaffold hopping was successfully carried out by replacing the hinge binding pyridine and retaining potency with other groups, including isoquinolines 14, phthalimides 12, and indazoles 13. Another approach for optimizing potency is to take an intact structure of an inhibitor and grow out to fill unoccupied volume within the ATP site. The optimization of a quinazoline scaffold for EGFR was reported using this approach (Ballard et al., 2006). Upon superimposition of two crystal structures containing either a quinazoline or ATP, it was noted that the quinazoline template did not fill the sugar binding subpocket. Potent analogues were found when cyclic amine substituents were introduced at the 5-position in an attempt to fill this region. Some kinases have ATP sites that, due to the surrounding side chains, are different in size and shape compared to other kinases. For example, p38 has a comparatively large back-pocket due to its small gatekeeper residue Thr106. Many inhibitors of p38 take advantage of this extra available space, notably the pyridinyl imidazole series exemplified by SB-203580. Inhibitors occupying this pocket have improved selectivity against kinases in which this pocket is smaller. This feature has been exploited by a wide range of p38 inhibitors, all of which build from the hinge deeper into the ATP site. The ingenuity of medicinal chemists in producing chemically different molecules that contain a unique combination of hinge and back-pocket binding groups is illustrated by the selection shown in Fig. 5.5, with the back-pocket and hinge binding groups marked. These examples are limited to compounds whose structures have been solved crystallographically and deposited in the protein databank. It was originally thought that a protein kinase with a gatekeeper residue larger than threonine would not accommodate inhibitors with a large back-pocket binding group. However, it has now become apparent that protein flexibility can lead to the breakdown of this simple rule. JNK3 has a methionine gatekeeper but is still inhibited by compounds with large back-pocket groups. A crystal structure of unliganded JNK3 revealed a back-pocket that was blocked by the methionine gatekeeper. However, a cocrystal structure with a diaryl imidazole inhibitor indicated that the methionine side chain can move, allowing access to the back-pocket (Scapin et al., 2003). Other movements have been observed in inactive kinase structures including reorientation of the two kinase lobes with respect to one another and significant conformational changes in the activation loop (Xu et al., 1999; Liu and Gray, 2006).

129

GENERATING AND OPTIMIZING KINASE INHIBITORS

SB-203580 (Wang et al., 1998) * N

Diaryl oxazole (McClure et al., 2005) * N N

F

Pyrazine (Tamayo et al., 2005) *

N

N

HN

F

N O

N

N

N

N

HN

O S VX745 Dihydroquinazolinone (Fitzgerald et al., 2003) (Stelmach et al., 2003) Cl O* Cl O* H N N F N N N F Cl Cl F

S

Pyrimidine/urea (Maier et al., 2006) H HO N N* H N

F Cl

N

N N O

O

N H RO3201195 (Goldstein et al., 2006) O*

PG951717 (Sabat et al., 2006) F O N*

NH2 O HO HO

N

F

N

N N

F

N F

Diaryl-imidazole JNK3/p38 inhibitor (Scapin et al., 2003) H N N* Cl N Cl N

N H

N N

FIGURE 5.5 Examples of p38 inhibitors showing interactions with the hinge (*) and backpocket (curved lines).

Many cyclin-dependent kinase inhibitors have been identified and reported, and some examples are shown in Fig. 5.6. To illustrate the point of achieving potency with selectivity within a structurally similar family of kinases, molecular design has successfully targeted single residue differences within the ATP site. In 2001, a series of naphthalenediones (16) were reported to bind to CDK1 and CDK2 (Furet et al., 2001). A crystal structure of one bearing a phenyl-4-sulphonamide bound to CDK2 revealed the sulphonamide hydrogen bonded to the backbone and side chain of Asp86. Similar results were reported for a series of oxindole CDK2 inhibitors, 17 (Bramson et al., 2001). The oxindole forms typical hydrogen bonding interactions with the hinge, while a phenyl 4-sulphonamide group increased the potency from >10 to 0.060 mM. A crystal structure revealed that the sulphonamide of 17 made identical interactions with Asp86 as was observed in the naphthalenedione 16 structure (Fig. 5.8a). Similar interactions have now been reported in many CDK2 series that have incorporated the phenyl-4-sulphonamide group. Utilizing the

130

STRATEGIES FOR DISCOVERING KINASE DRUGS

*

(a) O

*

H N

O S NH

N

H2NO2S

N

H N

N

N O

O

Br

H2NO2S

Naphthalenedione 16 (Furet et al., 2001)

HO

*

H N

H N

O

O

H N

*

H N

N

Purines 18 NU6102 (Davies et al., 2000)

*

H N

NH2

N

H2NO2S

Oxindole 17 (Bramson et al., 2001)

N

*

H N

NH

N N

NO Br

H2NO2S

O

N

Me2NO2S

NH Pyrimidines 19 (Sayle et al., 2003)

Anilinopyrazoles 20 (Tang et al., 2003)

H2N Pyrazolo-[1,5-a]pyrimidines 21 (Williamson et al., 2005)

* H N N

*

H N

*

N

H N

Cl

N H2NO2S N

O

NH2

Imidazo[1,2-a]pyridine 23 (Hamdouchi et al., 2005)

*

H N

(b) N

Triazolo[1,5-a]pyrimidines 24 (Richardson et al., 2006)

*

O

O

H N

N H

N H

N O

N

25

+

NH2

N H

26

*

H N O S H2N O

O

Cl

Benzodipyrazoles 22 (D'Alessio et al., 2005)

N N

Me2NO2S

O N

H2NO2S

N N

N N NH

27

H + N

CDK4-selective compound (Honma et al., 2001b) N

O

H N

N

*

N

O OH

NH

CDK4-selective compound (Beattie et al., 2003)

28 F

FIGURE 5.6 (a) Examples of CDK2 compounds optimized to include the 4-sulphonamide phenyl group. (b) Examples of compounds gaining CDK4 selectivity over CDK2 by introduction of a basic center close to CDK2 Lys89.

phenyl-4-sulphonamide group on a scaffold that permits it to interact with the solvent-face residues seems to be reliably successful in improving CDK2 activity, as can be seen by compounds 18–24. Many of the compound classes described above also inhibit another family member, CDK4. Compounds containing sulphonamide groups that interact with Asp86 are selective for CDK2 over CDK4. However, CDK4 selectivity has also been achieved. By adding a basic group in the appropriate position, the pan-CDK

131

GENERATING AND OPTIMIZING KINASE INHIBITORS

compound 25 has been converted into a CDK4-selective series represented by 26 (Honma et al., 2001b, 2001c). In a series of bis-anilinopyrimidines such as 27, the same effect of converting a CDK2 potent compound into a CDK4 potent compound was achieved when a basic chain was introduced in place of the sulphonamide in CDK2-selective compounds to afford 28 (Beattie et al., 2003). Crystal structures with CDK2 show that the overall binding mode of these compounds is similar although the potency in enzyme assays is reduced for CDK2 and increased for CDK4. This change was rationalized by a single amino acid difference, Lys89 in CDK2 and Thr102 in CDK4, lying just beyond the para-position of the outer lipophilic pocket aromatic substituents. The lysine in CDK2 is predicted to disfavor compounds with basic chains, a hypothesis supported by recent mutagenesis results (Pratt et al., 2006). In addition to side chain movements, backbone flexibility has also been observed, particularly near glycine residues. In p38, one of the hinge residues is a glycine (Gly110). As a result, p38 is unusually flexible in this region and can accommodate bulkier groups than most kinases. This flexibility can be targeted to gain selectivity. For example, binding of the dihydroquinazolinone 29 in Fig. 5.7 leads to a flip in the peptide bond between Met109 and Gly110 relative to its apo conformation (Stelmach et al., 2003). These types of compounds bind potently to wild-type p38 but lose activity when the glycine is mutated to alanine (Fitzgerald et al., 2003). A consequence of the Gly110 flip is that it points the NH of Gly110 into the ATP site, instead of the carbonyl of Met 109. Compounds with electronegative atoms in this region, close to the outer H-bonding carbonyl, are tolerated in p38 but not in most other kinases. This is illustrated in the crystal structures of a class of p38 purine compounds (Sabat et al., 2006). The purine 30 binds to the hinge via its N1 atom, similar to ATP. The 8-position aniline fills the back-pocket close to the Thr106 gatekeeper. To accommodate the bulk of the 2-phenoxy group, a peptide flip such as that described above is needed. This positions the NH of Gly110 Hydrogen bonding vectors in apo (left) and bound (right) conformations of p38 H R O N G110 H N N O H R O

p38 bound to dihydroquinazolinone (Stelmach) shows flipped Gly110 to accommodate bulky dichlorophenyl R

O

O

R

R

O H N

N H

N H

p38 bound to purine (Sabat) shows flipped Gly110 and hydrogen bond from Gly110 NH to ether linker O H N

N H

N H

R

O

O

F Cl O N

O

H N F F

Cl N N H

F

29

N N N F N N H

30

FIGURE 5.7 Schematic depiction of the orientation of the p38 hinge around Gly110 in the apo structure (left) and in two complexes with a dihydroquinazolinone and a purine.

132

STRATEGIES FOR DISCOVERING KINASE DRUGS

alongside the inhibitor C2-position. In this series, C2-phenoxy analogues are more potent than C2-anilines. This SAR can be rationalized by H-bonding between the phenoxy oxygen and the flipped Gly110 NH atom. With all of the advances in molecular design using ligand-bound crystal structures, creating inhibitors that occupy the ATP binding site has become quite sophisticated in establishing potency and selectivity. Using the above examples, Fig. 5.8b shows the

FIGURE 5.8 (a) CDK2 compounds containing the phenyl 4-sulphonamide group. (b) Overlaid p38 inhibitors showing their back-pocket binding groups. (c) Complexes of Abl/ Gleevec (green) and p38 / BIRB-796 (blue) showing the DFG-out binding mode. (See the color version of this figure in the Color Plates section.)

133

GENERATING AND OPTIMIZING KINASE INHIBITORS

remarkable use of the pocket to generate these kinase inhibitors. Panel A shows how the sulfonamide moiety takes advantage of a unique feature of CDK2 ATP pocket to add an interaction that confers potency to a number of different scaffolds. The variety of p38 compounds shown in Fig. 5.5 can be overlaid in a three-dimensional model such as the one shown in Fig. 5.8b to demonstrate the similarity in their binding interactions, and ways to add features to improve properties can be sought that do not interfere with required interactions. Of course, proteins are dynamic and thus their movements can also create unique binding modes that could introduce more molecular design approaches to differentiate kinase targets. 5.3.2

Non-ATP Binding Pockets

Given the problems expected to exist for ATP site inhibitors including the lack of selectivity and competition with high intracellular concentrations of ATP, it is desirable to find compounds that bind elsewhere or that are noncompetitive with ATP. It is assumed that an ATP-competitive compound will bind in the ATP site, though allosteric compounds may also be ATP competitive. However, compounds that bind in the ATP site may not be ATP competitive if they bind to noninterconvertible states of the kinase. Significant effort in this area has been reported, and the examples shown in Fig. 5.9 will be described below.

(a)

Gleevec (Abl)

*

BIRB-796 (p38)

O

Inner hydrophobic pocket

Inner hydrophobic pocket

O

N H

*

*

N

N H

NH

O

O

DFG pocket

Me

N N

DFG pocket

O

H N

H N

*

*

N

DFG pocket

(c)

N

Cl N

HO

DFG pocket

F3C

Cl

N N

NH

NH

F3C

N

(b)

O

NH

NH

N

Inner hydrophobic pocket

Inner hydrophobic pocket

O

NH

AAL-993 (VEGFR2)

N

O

N

Me N N

NH

*

31

N

Sorafenbib (B-raf)

Cl N

HO

O HN

O

O Cl

32

33

O

O H N

NH

F

Br Expand into DFG pocket, displacing Phe side chain

CF3

DFG-out pocket

F

I

F

PD184352

FIGURE 5.9 (a) Schematic of the binding modes of DFG-out compounds to Abl, p38, B-Raf, and VEGFR. Arrows indicate the hydrogen bonds to the DFG Asp backbone and to the salt-bridge glutamic acid side chain. (b) An example of a DFG-in binding compound elaborated to bind in a DFG-out manner. (c) Noncompetitive inhibitor of MEK1 kinase.

134

STRATEGIES FOR DISCOVERING KINASE DRUGS

The first noncompetitive inhibitor to make it to the clinic, PD184352, was found by screening a compound library in an enzyme cascade assay that measured phosphorylation of myelin basic protein in the presence of MEK1 and MAPK (SeboltLeopold et al., 1999; Allen et al., 2003). PD184352 is a potent, selective inhibitor of MEK1 with an IC50 value of 0.019 mM. Crystal structures of close analogues of PD184352 complexed with MEK1 and MEK2 have been solved that revealed binding in a non-ATP site (Ohren et al., 2004). The ligand bound crystal structures show that this inhibitor series bind simultaneously with Mg-ATP. The binding site is alongside the ATP site, on the opposite side of the conserved lysine, in a pocket formed by the outward movement of the C-helix. Progress on MEK1 clinical candidates can be found in a recent review (Wallace et al., 2005). An alternative approach to identify non-ATP competitive kinase inhibitors is to target an ‘‘inactive conformation,’’ while still binding in the ATP site. Gleevec 31 (Fig. 5.9a) was the first published example of this type of inhibitor (Schindler et al., 2000; Nagar et al., 2002). A crystal structure of Gleevec 31, complexed with its target Abl, revealed a binding mode in which the DFG motif undergoes a large rearrangement. Gleevec 31 binds to the hinge using its pyridine ring and projects a benzamide group deep into the back-pocket. When superimposed on the cocrystal structure of a traditional ATP site inhibitor, PD173995, the benzamide occupies the ˚ to a same space as Phe382 of the DFG motif. As a result, this residue moves
10 A new location on the outside of the ATP site. This movement of the DFG motif and activation loop exposes a deep channel running through the kinase, into which the piperazine of Gleevec projects. A similar binding mode has been seen for BIRB-796 (Fig. 5.9a) in complex with its target p38 (Pargellis et al., 2002). The crystal structure shows a comparable rearrangement of the DFG motif. While the morpholine ring of BIRB-796 does occupy the adenine subpocket, simpler, weaker analogues do not. Because of this, they are sometimes described as ‘‘allosteric’’ inhibitors, although this description may be questioned since all of them occupy the inner hydrophobic subpocket of the ATP site. Comparing the binding modes of Gleevec 31 and BIRB-796 by overlaying the protein structures of Abl and p38 reveals close similarity in the interactions that each makes with its kinase target (Fig. 5.8c). Both compounds fill the inner hydrophobic pocket and the DFG-out pocket with aromatic rings. Joining the two rings are an amide (Gleevec) and a urea (BIRB-796). These linkers are structurally equivalent and make analogous hydrogen bonds to the proteins. Other examples include the serine/threonine b-Raf kinase inhibitor Sorafenib (Khire et al., 2004) and the receptor tyrosine VEGFR2 kinase inhibitor AAL-993 (Manley et al., 2004). The biaryl urea moiety of Sorafenib has a very similar binding mode to the biaryl urea of BIRB-796 (Wan et al., 2004). The number of compounds known to bind to DFG-out conformations is growing, as is the number of examples seen in X-ray structures. The fact that the first marketed kinase inhibitor (Gleevec) binds in this way suggests that targeting the DFG-out kinase deserves full consideration as an alternative to targeting the DFG-in ATP site. All examples for this modulation of kinase activity whose structures have been disclosed contain a urea or an amide linked to two aromatic rings. As a chemically

ESTABLISHING SCREENS FOR UNDERSTANDING KINASE ACTIVITY

135

tractable synthetic group, this diaryl amide or diaryl urea offers a way to target this conformation. One aryl ring of this moiety occupies the inner hydrophobic pocket of the ATP site, which is itself a feature of many ATP-site inhibitors. Therefore it is no great leap to combine the two and attach aryl amides and ureas to existing ATP site templates. Because of the commonalities in binding mode between different targets, it offers the same potential advantages as the ATP site-directed chemistry in targeting multiple kinases simultaneously. Okram et al. (2006) found that adding extra groups to ATP site inhibitors such as 32 to target the DFG-out binding mode as indicated for 33 gave improved potency against several targets, including Abl, but led to inhibition of more kinases than their smaller progenitors. Additional activities introduced included cRaf, Lck, and PDGFRa. This finding agrees with our own results, which indicate that this extra functionality leads to a high probability of inhibiting a distinct subset of kinases. Not all DFG-out compounds inhibit the same set of kinases. For example, Gleevec does not inhibit p38, and BIRB-796 does not inhibit Abl. However, they seem to fish for their targets in a relatively small pool, suggesting that not all kinases can be tackled with this approach. In addition, the high molecular weight and the effect of attaching an aryl urea or amide on solubility must be considered, especially when it is being combined with an insoluble ATP site template.

5.4 ESTABLISHING SCREENS FOR UNDERSTANDING KINASE ACTIVITY AND SELECTIVITY The discovery of small-molecule kinase inhibitors and their optimization to preclinical development candidates often follows a defined compound progression path. The major elements of this path discussed in detail below are as follows: (1) inhibition of catalytic activity for the isolated purified kinase, (2) inhibition of cellular kinase activity or a functional surrogate, (3) determination of pharmacokinetic properties that would support in vivo testing, and (4) testing in an animal model for efficacy. Selectivity and physiochemical properties of the inhibitors are two additional areas that need to be addressed. The importance of physiochemical properties in the design of clinically successful kinase inhibitors is addressed in the case studies included in Section 5.5. Protein kinases catalyze the transfer of the g-phosphate of ATP to a hydroxylic amino acid of the protein substrate to generate ADP and the phospho-acceptor protein. The most widely used in vitro assays employ a recombinant protein kinase, ATP, and an acceptor peptide. While not well suited for HTS, radioactivity assays that measure the transfer of a g-32P-phosphate from ATP to a peptide substrate are widely used because of their sensitivity and simplicity (for a discussion of additional assay formats see von Ahsen and Bomer, 2005). Ideally, a well-designed in vitro kinase assay should provide a rapid and accurate assessment of the inhibitor–kinase interaction that will be predictive of the behavior of the inhibitor in the more complex cellular environment. Three major assay variables that require careful consideration are enzyme form, phospho-acceptor substrate, and enzyme kinetics.

136

STRATEGIES FOR DISCOVERING KINASE DRUGS

Since the majority of kinase inhibitors are ATP competitive, the inhibitor IC50 value is highly dependent upon the concentration of ATP. Therefore, kinase assays are typically run at or near the Km for ATP and peptide substrate. Typical ATP Km values are 10–100 mM, which is two to three orders of magnitude lower than the cellular concentration of ATP (Knight and Shokat, 2005). For the general procedures required to develop a well-characterized kinetic assay, the reader should consult Copeland’s ‘‘Evaluation of Enzyme Inhibitors in Drug Discovery. A Guide for Medicinal Chemists and Pharmacologists’’ (Copeland, 2005). To best mimic the cellular reaction, one would prefer a recombinant full-length human kinase. For cytosolic kinases with minimal nonkinase domains, such as p38, full-length protein assays are readily configured. Such a strategy is not feasible for receptor tyrosine kinases and proteins in which the kinase is a relatively small fragment of the entire protein. In such cases, attempts are made to engineer the smallest possible catalytically active protein in an approach similar to that taken for crystallography. Since the large majority of kinase inhibitors bind in the ATP site and their potency is not greatly influenced by nonkinase domains, the same construct can often be used for both enzyme assays and crystallography. The decision of what form of the enzyme to use must also take into consideration the need for coactivator proteins (such as cyclin for CDK2 and TPX2 for Aurora A) and the activation state (which is typically related to the number of phosphorylated sites on the target kinase protein). While the most activated form of the kinase may afford an assay, which is better characterized kinetically, it may not provide an accurate assessment of potency for compounds that bind to the inactive form of the kinase or interfere with the interaction of an activating protein. Nonetheless, in the absence of specific information to the contrary, the initial goal is to obtain a soluble active protein that can be used to configure a robust activity assay. In keeping with the desire to accurately recapitulate the activity of the kinase in the native cellular environment, one might anticipate a preference for the use of the natural substrate protein (Copeland, 2003). For kinases, such assays are difficult to configure and usually unnecessary. Instead, generic phosphoacceptor peptides can be used or if these prove inadequate, strategies have been developed to discover an acceptable peptide acceptor. For example, sequence information for the residues around the phosphoacceptor site of known substrates can often be used to derive a peptide substrate. An exception to this approach is MEK, for which no peptide substrate has been found, thus requiring the use of its sole protein substrate ERK as phosphoacceptor (Favata et al., 1998). Time-dependent or slow binding behavior, defined as the affinity of the inhibitor increases with increasing time of incubation with the kinase, has been reported for a number of inhibitors. The failure to account for this effect can result in a significant underestimation of potency. Examples include the p38 inhibitor BIRB-796 (Pargellis et al., 2002) and the EGFR/ErbB2 inhibitor laptinib (Wood et al., 2004). In both cases, crystal structures have revealed that the inhibitors bind to a conformation of the kinase distinct from that of apo and/or ATP-bound forms. It is believed that the slow binding behavior is a consequence of an inhibitor-induced conformational shift.

ESTABLISHING SCREENS FOR UNDERSTANDING KINASE ACTIVITY

137

The DFG-out conformation originally observed and correlated with time-dependent inhibition of p38a is not unique to this kinase but instead appears to be an accessible conformation for many protein kinases. Binding to the DGF-out conformation does not necessarily give rise to slow binding behavior and not all time-dependent behavior can be attributed to this particular conformation. Apart from the impact that slow binding inhibition can have on determination of the true IC50 value or the more rigorously determined Ki, this behavior may play an important role in the duration of pharmacological effect as it is often accompanied by an even slower dissociation rate. If the rate of dissociation is sufficiently slow (hours), the effect can be to prolong kinase inhibition after the drug concentration has dropped to noninhibitory levels in plasma and target tissues. In extreme cases, the inhibition appears irreversible if the dissociation rate approaches or exceeds that of the kinase T1/2 (Copeland et al., 2006). The configuration of cellular assays indicative of inhibition of the target kinase is crucial to establishing a robust correlation between enzyme and cellular effects, which can be used to advance lead optimization. The most direct approach is to monitor inhibition of the immediate downstream product—that is, to quantitate the phosphoacceptor protein. In the case of receptor tyrosine kinases (RTKs), dimerization of receptor by an external ligand results in trans phosphorylation, generating phosphotyrosine docking sites that propagate the signal. Autophosphorylation has been used to measure RTK inhibition of VEGFR2, PDGFR, and FGFR2. In contrast to the single substrate of the RTKs, many kinases are known to have multiple substrates. Casein kinase II is reported to be the most promiscuous of all protein kinases with over 200 cellular substrates (Pinna, 2003). In such cases, one would like to monitor a substrate in a pathway related to the desired pharmacology. Examples of pharmacologically relevant direct substrates that have been used to monitor intracellular kinase inhibition are Erk for MEK (Favata et al., 1998) and histone H3 for Aurora B (Girdler et al., 2006). However, the phosphoacceptor substrate most important to the desired pharmacology is not always known at the beginning of an effort, forcing the researcher to make a best guess on which substrate to choose to configure the assay. The low abundance of many kinases and hence the extremely small amounts of product formed in the cell necessitates the development of very sensitive detection methods. Furthermore, the ubiquitous nature of protein kinase signaling that gives rise to literally thousands of phosphoproteins requires these assays to be highly selective. To meet this need, antibody reagents have been developed that can capture the desired signal in a sea of protein phosphorylation. If possible, an ELISA-type assay using an antibody that specifically recognizes the phospho-epitope of the product of interest is used to gain high specificity. Alternatively, a capture antibody is used to recognize the protein, and a separate nonsequence specific phospho-antibody can be used to quantify the signal. Many of these antibodies and recombinant kinases are commercially available. In the examples cited above, inhibition of kinase activity was measured in cells that normally express the kinase of interest. For example, VEGFR2 cellular assays were performed in readily available human umbilical-derived endothelial cells, the

138

STRATEGIES FOR DISCOVERING KINASE DRUGS

only cell type known to robustly express the receptor for VEGF. However, not all cellular assays are this straightforward. In addition to cell-type-dependent kinase expression, there are numerous examples of temporal expression of protein kinases. For example, mitotic kinases, such as Aurora B and PLK1, are absent in quiescent cells and only become detectable during specific phases of the cell-cycle. In cases of low abundance or cell- and temporal-specific expression, it is appealing to consider genetically engineered systems. While such systems have been used, the lack of physiological relevance poses serious questions for their use in directing a lead optimization effort. Ideally, a linear correlation should exist between a well-behaved cellular kinase assay and the IC50 determination for the isolated enzyme in vitro. However, there is no a priori reason to anticipate identical IC50 values for the isolated enzyme versus the intracellular kinase, and in most cases higher IC50 values (less potent inhibition) are seen in the intact cell. In many cases, the magnitude of the shift can be directly related to the ATP-competitive nature of the inhibitors. The Cheng–Prusoff equation calculates the impact of ATP concentration on the IC50 value. IC50 ¼ Ki ð1 þ ½ATP=K M; ATP Þ Estimates of intracellular ATP concentrations are 1–5 mM and typical KM values for protein kinases are 10–100 mM. Applying these ranges suggests a shift in potency of
10–100 fold, which is in agreement with the shifts that are typically observed. Shokat has extensively developed this reasoning and presented numerous rationales for deviation from the above prediction (Knight and Shokat, 2005). These include the physiochemical properties of the inhibitor, as well as nonlinear kinetics due to depletion of substrate and the counterbalancing action of phosphatases, which reverse the reaction you are attempting to measure. While a biochemical cellular phosphorylation assay can be used as the primary cellular assay, a functional assay is often used instead. This situation is especially true when the functional assay is a measure of the desired cellular pharmacology and is simple to run. Examples of functional assays that have been used in kinase lead optimization efforts are inhibition of cellular proliferation (VEGFR2 and EGFR), selective inhibition of protein synthesis (p38 and ROCK), and cell cycle inhibition (CDK2 and Aurora B). Cellular proliferation assays are typically inexpensive to run, well suited to automation, and can often be configured to reflect the desired cellular pharmacology. The danger of these assays is that they may not reflect inhibition of the desired target. For example, if the target kinase is critical for cell division, inhibition of cellular proliferation would be the desired end point. However, there are multiple targets with which a compound might interact that could result in cell inhibition, many of which are kinases. For this reason, appropriate controls are required to validate the utility of the functional assay. In the case of a mitotic kinase such as Aurora B, one would want to establish that the inhibition of proliferation correlated with the biochemical cellular phosphorylation assay (pSer10 of histone H3). A further control might be to examine the effect of compounds on cell cycle to determine if they arrest cell growth at the expected

ESTABLISHING SCREENS FOR UNDERSTANDING KINASE ACTIVITY

139

stage of the cell cycle. In this manner, an Aurora B inhibitor can be differentiated from a CDK2 or PLK inhibitor or from a compound that blocks proliferation through a direct interaction with tubulin (Harrington et al., 2004). For an ATP-competitive inhibitor, very high kinase selectivity is often unobtainable, unnecessary, and undeterminable. That absolute selectivity is not required is demonstrated by Gleevec, the first kinase inhibitor approved for use in humans, which in addition to inhibiting the Bcr-Abl target also blocks cKit and PDGFR. That such selectivity is undeterminable speaks to the technical challenge of developing the methodology to assay
519 putative human kinases. Setting aside for now the discussion of what level of selectivity is required for a specific target, it is generally agreed that if inhibitor selectivity is insufficient, the undesired side effects are intolerable. Hence, determining selectivity and identifying strategies to achieve the required level of selectivity for the given target are top priorities for any kinase drug discovery effort. In the early to mid-1990s, kinase inhibitor selectivity was typically determined by testing a few phylogenetically related kinases plus examples of the major kinase classes that were commercially available. This approach led to many erroneous claims in the early kinase literature about selective inhibitors. These myths were first exposed by Cohen’s lab at the University of Dundee. Using a panel of 28 kinase activity assays, this group demonstrated that these early efforts were wholly inadequate. Many inhibitors reported to be selective were shown to be potent against a broad spectrum of kinases (Davies et al., 2000). A number of complementary approaches to activity-based assays are being pursued. These include immobilized ligand affinity chromatography to capture interacting kinases (Godl et al., 2003). An advantage of this method is the determination of biologically relevant interactions present in intact cells or tissues. While this method may be ill suited for lead optimization because of the amount of kinase inhibition data that would need to be accounted for in SAR studies, it may useful for finding novel effects or alternate kinases for study. A versatile approach to screen for human kinome selectivity has been described by Ambit Biosciences (Fabian et al., 2005). The Ambit technology relies upon phage display of kinase domains and a small set of ATP-competitive immobilized probe molecules. The test inhibitor is added to the kinase displaying phage bound to the immobilized ligands and allowed to compete for displacement. The phages displaced are quantitated either using a plaque assay or by quantitative PCR. The initial assay set contained 113 kinase domains but has expanded to over 300. The advantages of the higher throughput and lower cost of this methodology should be weighed against the potential disadvantages that are inherent in a binding assay that uses a truncated phage-attached kinase domain. In the initial publications using this methodology, Ambit researchers profiled a broad panel of tool inhibitors, including clinical candidates. The binding interactions detected in the phage screen were then validated with soluble kinases using activity assays. Several previously unreported interactions were uncovered including new and potent off-target activities for the p38 inhibitors VX-745 and BIRB-796, both of which have been studied in clinical trials. To date p38 inhibitors have yet to emerge from clinical development, with most being terminated due to safety concerns. The Ambit studies, as

140

STRATEGIES FOR DISCOVERING KINASE DRUGS

well as other recent reports, suggest that off-target activities need to be considered as an explanation for clinically observed side effects. A thorough understanding of off-target kinase activities is equally important for understanding the full therapeutic potential of a clinical candidate. In fact, it has been the rule rather than the exception that the clinical utility of kinase inhibitors extends beyond the kinase originally targeted. Gleevec was discovered as a targeted agent to treat Bcr-Abl-dependent CML. However, the additional off-target inhibition of cKit and PDGF has been utilized to extend Gleevec’s utility into the treatment of GIST (Demetri, 2002). Many creative methods are described in this section for evaluating the target affinity and selectivity of kinase inhibitors that depend on the many factors such as the form and length of the protein kinase, the activity being measured, the phosphorylation state, and whether or not a suitable substrate can be used in a catalytic assay format. Compromises taken to simplify the assay system are done to discover as many novel compounds as possible. The kinase inhibitors discovered in the primary assays are then validated in a cellular context, which introduces more complexity in understanding the effects of signal inhibition and the inhibition profiles. The field of biological screening for kinase inhibitors will continue to improve with more successes in clinical benefit being translated back into our drug discovery programs.

5.5

CASE STUDIES OF SUCCESSFUL KINASE DRUG DISCOVERY

The ultimate measure of success for a drug discovery program is the regulatory approval of a small molecule and the subsequent improvement in the quality of life of patients. Initially, there was concern that kinases would not become therapeutically valuable targets due to the anticipated difficulty in obtaining sufficient selectivity to give an appropriate safety margin. The efforts over the last decade have now shown that the initial reticence was perhaps unwarranted, and sufficient selectivity can be achieved. However, the vast majority of clinical successes have come in the oncology field, where selectivity may be less important and side effects more acceptable due to the life-threatening nature of the disease. Currently, there are eight kinase inhibitors approved for clinical use and they are listed in Table 5.1. Elements of several discoveries will be outlined here. The discovery of Gleevec 31 stands out in kinase drug discovery as it was the first tyrosine kinase inhibitor approved for the treatment of cancer. In chronic myelogenous leukemia, an interchromosomal exchange creates the Bcr-Abl gene, which encodes a protein with elevated tyrosine kinase activity. Compounds that block the activity of this enzyme were hypothesized to be potential treatments. The starting point for medicinal chemistry was pyrimidine 34 that inhibited PKC and PDGFR shown in Fig. 5.10 (Zimmerman et al., 1996, 1997). This molecule was an attractive starting point due to its activity, the potential for rapid and diverse analogue synthesis, and its relatively low molecular weight. During optimization, the addition of a 3-pyridyl substituent to create 35 improved cellular

141

CASE STUDIES OF SUCCESSFUL KINASE DRUG DISCOVERY

TABLE 5.1

Summary of FDA Approved Kinase Inhibitors

Name(s) Fasudil HA1077

Structure

Kinase Targets

H

Rho kinase

Cerebral vasospasm

1995 in Japan

BCR-ABL, PDGFR, c-KIT

Chronic myelogenous leukemia

May 2001

EGFR

Nonsmall cell lung cancer

May 2003

EGFR

Nonsmall cell lung cancer

Nov. 2004

Multikinase inhibitor—Raf, VEGFR2 and 3,PDGFRb, cKIT, FLT3

Renal cell carcinoma

Dec. 2005

Multikinase inhibitor— PDGFRb, VEGFR1, 2, and 3, cKIT, FLT3, CSF1R, RET

Gastrointestinal stromal tumors and advanced renal cell carcinoma

Jan. 2006

N N

Approval Date

Indication

O S O

N

Gleevec Glivec ST1571

N

H

H

N

N

N N O

N

N

Iressa Gefitinib ZD1839

F N

Cl

H

O

O

N

Tarceva Erlotinib OSI-774

H

N

O

N

O O

O

N

Nexavar Sorafenib BAY439006

N

O

N

O

CF3 Cl

O

O N

N

H

H

Sutent Sunitinib SU11248

N

O N H N H O

F N H

N N

H

(continued)

142

STRATEGIES FOR DISCOVERING KINASE DRUGS

TABLE 5.1

(Continued )

Name(s)

Structure

Sprycel Dasatinib BMS354825

Tykerb Lapatinib GW2016

Kinase Targets

Multikinase Chronic Jun. inhibitor –BCRmyelogenous 2006 ABL, SRC leukemia and family (SRC, acute lymphobLCK, YES, lastomic FYN), cKIT, leukemia EPHA2, PDGFRb

N N Cl

N

N

H

N OH

N

N

S

H

O

Cl

ErbB2, EGFR

O

F

N

Cl

H N

H

S O

Approval Date

Indication

Breast cancer

Mar. 2007

O

O

N N

N

H N

N

N

N

H

H

N

N O

N

N

R

N

35

34

H N

N N

N

N

H

N

N O

31 Gleevec

FIGURE 5.10

N

H

H

N

N O

N

N 36

Gleevec—path from hit to approved drug.

R

143

CASE STUDIES OF SUCCESSFUL KINASE DRUG DISCOVERY

activity. An amide group on the phenyl was well tolerated and with the proper substitution further enhanced potency. A key discovery was the finding that addition of a methyl group in 36 dramatically changed the enzyme inhibition profile, eliminating PKC activity while maintaining PDGFR and Bcr-Abl inhibition. However, these molecules showed poor bioavailability and low solubility. Addition of an appropriate R-group basic side chain maintained activity and enhanced solubility and bioavailability (Capdeville et al., 2002). Gleevec was chosen as the compound with the best overall features. It inhibits Bcr-Abl, cKit, and PDGFR; is active in a range of cellular assays; and shows selectivity in vitro and in vivo for Bcr-Abl expressing cells. Dasatinib 37 was approved in 2006 for chronic myelogenous leukemia and acute lymphoblastomic leukemia. Dasatinib 37, like Gleevec 31, not only inhibits BcrAbl but is also a potent Src family tyrosine kinase inhibitor (McIntyre et al., 2005; Bocchia et al., 2006). This profile is considered important since Src family kinases modulate signal transduction in a number of pathways whose aberrant regulation is implicated in a variety of cancers. Screening of the corporate compound collection led to the identification of the aminothiazole series, exemplified by 38 in Fig. 5.11, as micromolar inhibitors of Lck, a member of the Src family (Das et al., 2006). Iterative SAR exploration showed that ortho-substituents on the benzamide portion were critical for potent activity. Due to the comparable activity of t-butyl carbamate 39 to the initial hit (unsubstituted NH2), effort was then directed toward exploration of this region of the molecule. A further improvement

N H2N

O

H

S

N

O

N

H

N O

N

S

H

O

39

38

OH N

N

N

N

N

H

N

N

H N

N

H

N H

Cl

40

N

S O Dasatinib 37

FIGURE 5.11

Discovery of dasatinib.

N

S O

Cl

144

STRATEGIES FOR DISCOVERING KINASE DRUGS

was made by the removal of the 4-prime methyl group from the thiazole. Modeling suggested that removal of this methyl would allow the amide to adopt an optimal conformation for binding to the enzyme. In a parallel synthesis approach, amides, carbamates, ureas, and heteroaryl amines (amide mimetics) were pursued. Pyrimidine 40 emerged with potent Lck enzyme activity, and cellular activity. Cross-screening of this compound against a panel of kinases revealed potent inhibition of several Src family kinases and Bcr-Abl. The compound also showed antiproliferative activity in tumor cell lines. The strategy of the lead optimization program was to emphasize activity in the tumor cell proliferation assays and oral exposure in mice. A number of compounds were discovered that showed potent cellular activity but poor oral exposure, not an uncommon situation in medicinal chemistry. To address this problem, a common and useful strategy was employed. In an area that was likely to be solvent exposed and tolerant of polar functionality, a variety of substituents with varying polarity and basicity were attached. Out of this exercise, dasatinib emerged with both good antiproliferative activity and oral exposure (Lombardo et al., 2004). Subsequent experiments confirmed oral activity in a human tumor xenograft model in mice and lack of toxicity as judged by animal deaths and lack of weight gain. In addition to combining Src family inhibition with the clinically validated Bcr-Abl inhibition, dasatinib also binds to both active and inactive forms of Abl, and inhibits Gleevec-resistant Bcr-Abl mutants (Shah et al., 2004; Manley et al., 2005; O’Hare et al., 2005; Talpaz et al., 2006; Tokarski et al., 2006). Numerous studies have indicated that the EGFR-mediated signaling network is important in cancer cell proliferation and that inhibitors of EGFR may be useful for the treatment of some cancers. In the mid 1990s, quinazolines were identified as EGFR inhibitors via database searching based on a pharmacophore hypothesis (Ward et al., 1994). The query was derived from knowledge of the catalytic mechanism and was designed to look for compounds that mimicked the phenyl ring of the substrate tyrosine, the tyrosine hydroxyl, and the ATP gamma phosphate. Although this structure-based searching methodology led to the identification of this important class of molecules, quinazolines are not competitive with substrate peptide, and thus are not substrate mimetics. Rather, they bind in the hydrophobic adenine site, anchored by hydrogen bond to the hinge region, and are competitive with ATP (Shewchuk et al., 2000; Stamos et al., 2002). IressaTM 41 and TarcevaTM are members of the quinazoline class of kinase inhibitors and were approved by the FDA for treating cancer in 2003 and 2004, respectively. Iressa was identified following lead optimization studies that commenced with the aniline–quinazoline 42 in Fig. 5.12, a potent EGFR kinase inhibitor, and a potent inhibitor of EGFR-stimulated tumor cell growth (Barker et al., 2001). While this compound showed evidence of oral activity in a tumor xenograft model, it also appeared to be rapidly metabolized on the aniline moiety. Lead optimization focused on improving the pharmacokinetic properties in this series. The 3-Cl group on the aniline of 43 removed the issue of hydroxylation of the 3-Me in the original lead. Addition of a fluorine in the 4-position blocked oxidation para to the NH of the aniline. With the metabolically labile sites appropriately altered, this compound had a lower clearance value and improved oral efficacy.

145

CASE STUDIES OF SUCCESSFUL KINASE DRUG DISCOVERY

Cl F N

H N

H

O

N

O

N O

N

O

N 43

42

Cl

Cl

F

F N

H

O O

N

N

N

H OH

N O

N

O

N

Iressa 41

44

FIGURE 5.12

Discovery of Iressa.

Further attempts to improve oral efficacy focused on alteration of the quinazoline methoxy groups. The ability to append a range of groups with different chain lengths, lipophilicity, basicity, and solubility to a key late-stage intermediate 44 proved valuable for optimization. The medicinal chemistry strategy for examining the effects of these groups on SAR is summarized in Fig. 5.13. Keeping the ether linkage, a two-methylene linker to an amino group was compared with similarly substituted three-methylene linkage. There were also compounds made with a hydroxylated 3-carbon linker. The typical amines used to improve solubility in a lead optimization are also shown and include morpholino, piperazino, branched and linear alkyl amines, and others. While the changes are often subtle, the properties vary a great deal and require an extensive review of functional groups to ensure target affinity is retained. To identify compounds with potential for efficacy with once daily dosing, blood concentrations of the new test compounds in mice were measured at 2, 6, and 24 h after oral dosing (200 mg/kg). A morpholino-propyl side chain gave potent enzyme and cell activity, and high blood levels 24 h after dosing. Due to the high and sustained blood level, 41 was advanced to further preclinical studies. Oral efficacy and dose-dependent inhibition of the growth of a number of human solid tumor xenografts led to its selection for clinical investigation.

146

STRATEGIES FOR DISCOVERING KINASE DRUGS Cl F H R1

N

O

N R2

N

O

N

Compounds with 2-carbon linker Cl

Cl F

H

F

N

HO

N

O

H

R1 R2

N

N

O

N

N

O

44

N

Compounds with 3-carbon linker Cl F R1 R2

H

OH

N

N

O

N

O

N

Compounds with hydroxylated 3-carbon linker Types of monomers (R1R2NH) utilized to explore sterics, polarity, and basicity NH

NH O

NH

NH

N

NH2 HO

N

NH

NH

N

O NH2

NH2

HO

NH2 O

NH2

NH

NH O

FIGURE 5.13 Utilization of a key intermediate to optimize oral activity.

EGFR belongs to a family of receptor tyrosine kinases that includes ErbB2 (Her2 or Neu), ErbB3 (Her3), and ErbB4 (Her4) (Johnston et al., 2006). Similar to EGFR, other family members are often overexpressed in different tumor types, and signaling through this family appears to be critical for cancer cell proliferation. TykerbTM was born out of an effort to find dual EGFR/ErbB2 inhibitors. The program goal was to identify compounds that exhibited potent inhibition of tumor cell proliferation in cell lines that overexpressed EGFR and ErbB2 (Lackey, 2006). The compounds also needed to show selectivity, not inhibit normal fibroblast growth, and work in human xenograft mouse models. Finally, in order to show that the compound effect (phenotype) was due to specific kinase inhibition, successful compounds needed to show a correlation between antitumor activity and phosphorylated receptor levels.

147

CASE STUDIES OF SUCCESSFUL KINASE DRUG DISCOVERY

N N

H O

N N

O Tarceva 45

ErbB2: IC50 = 1000 nM EGFR: IC50 = 17 nM

FIGURE 5.14

N

H

O O

N

N N

N

GW2974 46 ErbB2: IC50 = 16 nM EGFR: IC50 = 7 nM

Introduction of large aniline substituents gave dual EGFR/ErbB2 inhibition.

The unique finding from the SAR studies indicated that a large substituent on the 4-anilino moiety of the quinzoline scaffold could give potent dual EGFR/ErbB2 inhibition (Lackey, 2006). Tarceva 45, for example, with a 4-meta-alkyne-anilino quinazoline structure (Fig. 5.14) has an IC50 value of 1 mM on ErbB2 and an IC50 value of 0.017 mM on EGFR. GW2974, 46, which has a much larger group at the 4-position of the quinazoline, N1-benzylindazole, inhibits ErbB2 and EGFR with IC50 values of 0.016 and 0.007 mM, respectively. Compounds such as 46 also showed good selectivity for the ErbB family of kinases, with greater than 50-fold in vitro selectivity over other proliferative kinases (Cockerill et al., 2001; Brignola et al., 2002). Generally speaking, this type of substitution change affects kinase selectivity and inhibition effectiveness significantly, which requires careful attention to SAR studies. Compounds with potent dual activity were progressed into a panel of six parallel cellular assays. These experiments allowed the team to assess inhibition of proliferation of ErbB family-driven tumor cell lines as well as selectivity against normal cells (Rusnak et al., 2001a). The desired profile was activity against the tumor lines with an IC50 value <0.200 mM and 50-fold selectivity over nonErbB-driven cell lines and the normal cells (HFF). 46 met these criteria, showed significant tumor growth inhibition in xenograft models, and blocked phosphorylation of ErbB2 and EGFR in vivo (Cockerill et al., 2001; Rusnak et al., 2001a). Unfortunately, the compound failed preclinical toxicity studies. However, the research leading up to this point provided an outstanding tool compound, useful for elucidating biological mechanisms, and a host of refined assays that would serve to find a new candidate with an improved therapeutic index. A similar medicinal chemistry strategy utilizing key intermediates allowed for the synthesis of a range of substituents at the 6-position of the quinazoline and a pyrido-pyrimidine scaffold (7-aza-quinazoline). An example of the type of pharmacophore SAR map that is done for a lead optimization effort is shown in Fig. 5.15 where the trends are summarized based on accrued screening data for many compounds in a series. As was described in Section 5.4, it is important to correlate the kinase enzyme inhibition data with the cellular activity. Where major differences arise, there is

148

STRATEGIES FOR DISCOVERING KINASE DRUGS Large substitution important for ErbB-2 tyrosine kinase inhibition

Large R groups tolerated, however, cellular in vitro potency is decreased

Optimal chain length

R O S O

Furan and thiazole were the best 5-membered ring heterocycles

HN Z

O, N, NR retain enzyme potency

X N

O W

N

Substitutions decrease enzyme potency

O Y

Small substitutions preferred, F is best

Cl, Br are best, substitutions affect cellular activity

N and C are equally good in enzyme and cell

FIGURE 5.15 Example of an SAR map derived from assessing activity in the lead optimization assays that eventually led to the discovery of lapatinib.

usually a mechanistic (e.g., slow off rate) or physical (e.g., poor solubility) reason for the trend. To prioritize cell active compounds for in vivo efficacy studies, the program team developed a pharmacokinetic assessment protocol that allowed rapid ranking of the oral bioavailability of compounds. With potent, cell active, bioavailable compounds in hand, a decision was made to scale six compounds and progress them in parallel into early toxicity studies. The six compounds were structurally similar and were chosen based on 22 distinguishing criteria (Lackey, 2006). The criteria included measures of efficacy (cell assays, in vivo experiments), toxicity (cellular toxicity, Ames test, 7-day rat studies, blood chemistry), metabolism and pharmacokinetic parameters (P450 inhibition, bioavailability, time of drug exposure above the IC50 concentration), along with chemical issues (cost of goods, scale-up of the compound). Although the six compounds look similar chemically, medicinal chemists know that apparently small changes in structure can have profound affects on key biological and practical parameters. Having an enumerated set of criteria with which to compare compounds allows a program team to progress compounds with the best overall profile. On the basis of these criteria, GW2016 (Tykerb, lapatinib) shown in Table 5.1 presented the most favorable balance of properties and was chosen for clinical development (Rusnak et al., 2001b; Petrov et al., 2006). The discovery of Sorafenib, a Raf kinase inhibitor 47, involved high-throughput screening (HTS), combinatorial chemistry, and medicinal chemistry (Lyons et al., 2001; Wilhelm et al., 2006). Raf kinases have been highlighted as targets for oncology due to their involvement in signal transduction and as regulators of cellular proliferation and survival (Nottage and Siu, 2002). HTS identified a diaryl urea molecule with an IC50 determination equal to 17 mM on Raf1 (Smith et al., 2001). While this molecule was not very potent, it was an attractive starting point due to the ability to make analogues to explore SAR. A traditional, sequential analogue approach (changing one thing at a time) allowed improvement to achieve a potency of 1.7 mM.

149

CASE STUDIES OF SUCCESSFUL KINASE DRUG DISCOVERY

N

H

H

H

H

N

N

N

N

N

O

O

O

O

O

48

O

49 N

F3C Cl

H

H

N

N O

F3C O Cl

Sorafenib 47

H

H

N

N O

O

50 O

N HN

N CH3

FIGURE 5.16 Discovery of Sorafenib.

A broader combinatorial approach (larger changes as well as changing both sides of the molecule at once) was carried out in parallel to the above studies and identified a differentially substituted urea 48 shown in Fig. 5.16 with a potency of 1.1 mM (Lowinger et al., 2002). This finding was surprising since compounds that were close derivatives had been made and were not very active. When these single-point changes, individually not considered worthy of follow-up due to poor activity, were combined, reasonable activity emerged. This example highlights the importance of using combinatorial exploration in lead discovery and optimization to look for synergy between different parts of the molecules that can be used for breaking out of SAR minima. In the next round of optimization, replacement of the phenyl with a pyridyl ring improved potency to an IC50 value of 0.23 mM and also, importantly, lowered the clog P and improved solubility. This compound 49 also exhibited cellular and in vivo activities and warranted further SAR exploration. These efforts led to the replacement of the isoxazole ring with substituted phenyls and the discovery that lipophilic meta substituents enhanced potency to afford 50 with an IC50 value of 0.046 mM. Finally, substitution on the pyridyl ring with a methyl carboxamide 47 gave a further boost in potency. In addition to potent in vitro activity against Raf1, Sorafenib 47 inhibits wild-type bRaf, oncogenic bRaf V600E, VEGFR1, 2, and 3,

150

STRATEGIES FOR DISCOVERING KINASE DRUGS

PDGFRb, Flt3, p38, and cKit with potencies under 0.10 mM (Wilhelm and Chien, 2002; Wilhelm et al., 2004). It is likely that this profile, inhibition of multiple kinases involved in proliferation and angiogenesis, is important and contributes to its efficacy. Sorafenib’s activity in a range of cellular assays and tumor growth inhibitory effects when dosed orally in a variety of xenograft models warranted its progression to clinical trials (Hotte and Hirte 2002). With these and other emerging successes in kinase drug discovery, the field is exciting and full of enormous potential.

REFERENCES Allen LF, Sebolt-Leopold J, Meyer MB, 2003. CI-1040 (PD184352), a targeted signal transduction inhibitor of MEK (MAPKK). Semin. Oncol. 30:105–116. Angell R, Bamborough P, Cleasby A, Cockerill S, Jones KL, Mooney C, Somers DO, Walker AL, 2008. Biphenyl amide p38 kinase inhibitors 1: discovery and binding mode. Bioorg. Med. Chem. Lett. 18:318–323. Aronov AM, Murcko MA, 2004. Toward a pharmacophore for kinase frequent hitters. J. Med. Chem. 47:5616–5619. Ballard P, Bradbury RH, Harris CS, Hennequin LF, Hickinson M, Johnson PD, Kettle JG, Klinowska T, Leach AG, Morgentin R, Pass M, Ogilvie DJ, Olivier A, Warin N, Williams EJ, 2006. Inhibitors of epidermal growth factor receptor tyrosine kinase: novel C-5 substituted anilinoquinazolines designed to target the ribose pocket. Bioorg. Med. Chem. Lett. 16:1633–1637. Bamborough P, Christopher JA, Cutler GJ, Dickson MC, Mellor GW, Morey JV, Patel CB, Shewchuk LM, 2006. 5-(1H-Benzimidazol-1-yl)-3-alkoxy-2-thiophenecarbonitriles as potent, selective, inhibitors of IKK-epsilon kinase. Bioorg. Med. Chem. Lett. 16:6236–6240. Barker AJ, Gibson KH, Grundy W, Godfrey AA, Barlow JJ, Healy MP, Woodburn JR, Ashton SE, Curry BJ, Scarlett L, Henthorn L, Richards L, 2001. Studies leading to the identification of ZD1839 (IRESSA): an orally active, selective epidermal growth factor receptor tyrosine kinase inhibitor targeted to the treatment of cancer. Bioorg. Med. Chem. Lett. 11:1911–1914. Beattie JF, Breault GA, Ellston RP, Green S, Jewsbury PJ, Midgley CJ, Naven RT, Minshull CA, Pauptit RA, Tucker JA, Pease JE 2003. Cyclin-dependent kinase 4 inhibitors as a treatment for cancer. Part 1: identification and optimisation of substituted 4,6-bis anilino pyrimidines. Bioorg. Med. Chem. Lett. 13:2955–2960. Birault V, Harris CJ, Le J, Lipkin M, Nerella R, Stevens A, 2006. Bringing kinases into focus: efficient drug design through the use of chemogenomic toolkits. Curr. Med. Chem. 13:1735–1748. Bocchia M, Forconi F, Lauria F, 2006. Emerging drugs in chronic myelogenous leukaemia. Expert Opin. Emerg. Drugs. 11:651–664. Bramson HN, Corona J, Davis ST, Dickerson SH, Edelstein M, Frye SV, Gampe Jr RT, Harris PA, Hassell A, Holmes WD, Hunter RN, Lackey KE, Lovejoy B, Luzzio MJ, Montana V, Rocque WJ, Rusnak D, Shewchuk L, Veal JM, Walker DH, Kuyper LF, 2001. Oxindolebased inhibitors of cyclin-dependent kinase 2 (CDK2): design, synthesis, enzymatic activities, and X-ray crystallographic analysis. J. Med. Chem. 44:4339–4358.

REFERENCES

151

Brignola PS, Lackey K, Kadwell SH, Hoffman C, Horne E, Carter HL, Stuart JD, Blackburn K, Moyer MB, Alligood KJ, Knight WB, Wood ER, 2002. Comparison of the biochemical and kinetic properties of the type 1 receptor tyrosine kinase intracellular domains. Demonstration of differential sensitivity to kinase inhibitors. J. Biol. Chem. 277:1576–1585. Capdeville R, Buchdunger E, Zimmermann J, Matter A, 2002. Glivec (STI571, imatinib), a rationally developed, targeted anticancer drug. Nat. Rev. Drug. Discov. 1:493–502. Cockerill S, Stubberfield C, Stables J, Carter M, Guntrip S, Smith K, McKeown S, Shaw R, Topley P, Thomsen L, Affleck K, Jowett A, Hayes D, Willson M, Woollard P, Spalding D, 2001. Indazolylamino quinazolines and pyridopyrimidines as inhibitors of the EGFR and c-ErbB-2. Bioorg. Med. Chem. Lett. 11:1401–1405. Cohen P, 2002. Timeline: protein kinases—the major drug targets of the twenty-first century? Nat. Rev. Drug Discov. 1:309–315. Copeland RA, 2003. Mechanistic considerations in high-throughput screening. Anal. Biochem. 320:1–12. Copeland RA, 2005. Evaluation of enzyme inhibitors in drug discovery. A guide for medicinal chemists and pharmacologists. Methods Biochem. Anal. 46:1–265. Copeland RA, Pompliano DL, Meek TD, 2006. Drug-target residence time and its implications for lead optimization. Nat. Rev. Drug Discov. 5:730–739. D’Alessio R, Bargiotti A, Metz S, Brasca MG, Cameron A, Ermoli A, Marsiglio A, Polucci P, Roletto F, Tibolla M, Vazquez ML, Vulpetti A, Pevarello P, 2005. Benzodipyrazoles: a new class of potent CDK2 inhibitors. Bioorg. Med. Chem. Lett. 15:1315–1319. Das J, Chen P, Norris D, Padmanabha R, Lin J, Moquin RV, Shen Z, Cook LS, Doweyko AM, Pitt S, Pang S, Shen DR, Fang Q, de Fex HF, McIntyre KW, Shuster DJ, Gillooly KM, Behnia K, Schieven GL, Wityak J, Barrish JC, 2006. 2-Aminothiazole as a novel kinase inhibitor template. Structure–activity relationship studies toward the discovery of N-(2-chloro-6-methylphenyl)-2-[[6-[4-(2-hydroxyethyl)-1-piperazinyl)]-2-methyl-4-pyrimidinyl]amino)]-1,3-thiazole-5-carboxamide (dasatinib, BMS-354825) as a potent panSrc kinase inhibitor. J. Med. Chem. 49:6819–6832. Davies SP, Reddy Reddy H, Caivano M, Cohen P, 2000. Specificity and mechanism of action of some commonly used protein kinase inhibitors. Biochem. J. 351:95–105. Demetri GD, 2002. Targeting the molecular pathophysiology of gastrointestinal stromal tumors with imatinib. Mechanisms, successes, and challenges to rational drug development. Hematol. Oncol. Clin. North Am. 16:1115–1124. Fabian MA, Biggs III WH, Treiber DK, Atteridge CE, Azimioara MD, Benedetti MG, Carter TA, Ciceri P, Edeen PT, Floyd M, Ford JM, Galvin M, Gerlach JL, Grotzfeld RM, Herrgard S, Insko DE, Insko MA, Lai AG, Lelias JM, Mehta SA, Milanov ZV, Velasco AM, Wodicka LM, Patel HK, Zarrinkar PP, Lockhart DJ, 2005. A small molecule–kinase interaction map for clinical kinase inhibitors. Nat. Biotechnol. 23:329–336. Favata MF, Horiuchi KY, Manos EJ, Daulerio AJ, Stradley DA, Feeser WS, Van Dyk DE, Pitts WJ, Earl RA, Hobbs F, Copeland RA, Magolda RL, Scherle PA, Trzaskos JM, 1998. Identification of a novel inhibitor of mitogen-activated protein kinase kinase. J. Biol. Chem. 273:18623–18632. Fedi P, Aaronson SA, 2000. Signal transduction through tyrosine kinase growth factor receptors. In: Gutkind, JS, editor. Signaling Networks and Cell Cycle Control: The Molecular Basis of Cancer and Other Diseases. Humana Press, Inc., Totowa, NJ. Chapter 2, pp. 27–38.

152

STRATEGIES FOR DISCOVERING KINASE DRUGS

Fejzo J, Lepre CA, Peng JW, Bemis GW, Ajay, Murcko MA, Moore JM, 1999. The SHAPES strategy: an NMR-based approach for lead generation in drug discovery. Chem. Biol. 6:755–769. Fejzo J, Lepre C, Xie X, 2003. Application of NMR screening in drug discovery. Curr. Top. Med. Chem. 3:81–97. Fitzgerald CE, Patel SB, Becker JW, Cameron PM, Zaller D, Pikounis VB, O’Keefe SJ, Scapin G, 2003. Structural basis for p38alpha MAP kinase quinazolinone and pyridol-pyrimidine inhibitor specificity. Nat. Struct. Biol. 10:764–769. Foloppe N, Fisher LM, Howes R, Potter A, Robertson AGS, Surgenor AE, 2006. Identification of chemically diverse Chk1 inhibitors by receptor-based virtual screening. Bioorg. Med. Chem. Lett. 14:4792–4802. Furet P, Meyer T, Mittl P, Fretz H, 2001. Identification of cylin-dependent kinase 1 inhibitors of a new chemical type by structure-based design and database searching. J. Comput. Aided Mol. Des. 15:489–495. Gill A, 2004. New lead generation strategies for protein kinase inhibitors—fragment based screening approaches. Mini Rev. Med. Chem. 4:301–311. Girdler F, Gascoigne KE, Eyers PA, Hartmuth S, Crafter C, Foote KM, Keen NJ, Taylor SS, 2006. Validating Aurora B as an anti-cancer drug target. J. Cell. Sci. 119:3664–3675. Godl K, Wissing J, Kurtenbach A, Habenberger P, Blencke S, Gutbrod H, Salassidis K, SteinGerlach M, Missio A, Cotten M, Daub H, 2003. An efficient proteomics method to identify the cellular targets of protein kinase inhibitors. Proc. Natl. Acad. Sci. USA 100:15434–15439. Goldstein DM, Alfredson T, Bertrand J, Browner MF, Clifford K, Dalrymple SA, Dunn J, Freire-Moar J, Harris S, Labadie SS, La Fargue J, Lapierre JM, Larrabee S, Li F, Papp E, McWeeney D, Ramesha C, Roberts R, Rotstein D, San Pablo B, Sjogren EB, So OY, Talamas FX, Tao W, Trejo A, Villasenor A, Welch M, Welch T, Weller P, Whiteley PE, Young K, Zipfel S, 2006. Discovery of S-5-amino-1-(4-fluorophenyl)-1H-pyrazol-4-yl-3(2,3-dihydroxypropoxy)phenylmethanone (RO3201195) an orally bioavailable and highly selective inhibitor of p38 MAP kinase. J. Med. Chem. 49:1562–1575. Hamdouchi C, Zhong B, Mendoza J, Collins E, Jaramillo C, De Diego JE, Robertson D, Spencer CD, Anderson BD, Watkins SA, Zhang F, Brooks HB, 2005. Structure-based design of a new class of highly selective aminoimidazo[1,2-a]pyridine-based inhibitors of cyclin dependent kinases. Bioorg. Med. Chem. Lett. 15:1943–1947. Harrington EA, Bebbington D, Moore J, Rasmussen RK, Ajose-Adeogun AO, Nakayama T, Graham JA, Demur C, Hercend T, Diu-Hercend A, Su M, Golec JM, Miller KM, 2004. VX680, a potent and selective small-molecule inhibitor of the Aurora kinases, suppresses tumor growth in vivo. Nat. Med. 10:262–267. Hartshorn MJ, Murray CW, Cleasby A, Frederickson M, Tickle IJ, Jhoti H, 2005. Fragmentbased lead discovery using X-ray crystallography. J. Med. Chem. 48:403–413. Honma T, Hayashi K, Aoyama T, Hashimoto N, Machida T, Fukasawa K, Iwama T, Ikeura C, Ikuta M, Suzuki-Takahashi I, Iwasawa Y, Hayama T, Nishimura S, Morishima H, 2001a. Structure-based generation of a new class of potent Cdk4 inhibitors: new de novo design strategy and library design. J. Med. Chem. 44:4615–4627. Honma T, Yoshizumi T, Hashimoto N, Hayashi K, Kawanishi N, Fukasawa K, Takaki T, Ikeura C, Ikuta M, Suzuki-Takahashi I, Hayama T, Nishimura S, Morishima H, 2001b. A novel approach for the development of selective Cdk4 inhibitors: library design based on locations of Cdk4 specific amino acid residues. J. Med. Chem. 44:4628–4640.

REFERENCES

153

Honma T, Hayashi K, Aoyama T, Hashimoto N, Machida T, Fukasawa K, Iwama T, Ikeura C, Ikuta M, Suzuki-Takahashi I, Iwasawa Y, Hayama T, Nishimura S, Morishima H 2001c. Structure-based generation of a new class of potent Cdk4 inhibitors: new de novo design strategy and library design. J. Med. Chem. 44:4615–4627. Hotte SJ, Hirte HW, 2002. BAY 43–9006: early clinical data in patients with advanced solid malignancies. Curr. Pharm. Des. 8:2249–2253. Johnston JB, Navaratnam S, Pitz MW, Maniate JM, Wiechec E, Baust H, Gingerich J, Skliris GP, Murphy LC, Los M, 2006. Targeting the EGFR pathway for cancer therapy. Curr. Med. Chem. 13:3483–3492. Khire UR, Bankston D, Barbosa J, Brittelli DR, Caringal Y, Carlson R, Dumas J, Gane T, Heald SL, Hibner B, Johnson JS, Katz ME, Kennure N, Kingery-Wood J, Lee W, Liu XG, Lowinger TB, McAlexander I, Monahan MK, Natero R, Renick J, Riedl B, Rong H, Sibley RN, Smith RA, Wolanin D, 2004. Omega-carboxypyridyl substituted ureas as Raf kinase inhibitors: SAR of the amide substituent. Bioorg. Med. Chem. Lett. 14:783–786. Knight ZA, Shokat KM, 2005. Features of selective kinase inhibitors. Chem. Biol. 12:621– 637. Knighton DR, Zheng JH, Ten Eyck LF, Ashford VA, Xuong NH, Taylor SS, Sowadski JM, 1991. Crystal structure of the catalytic subunit of cyclic adenosine monophosphatedependent protein kinase. Science 253:407–414. Lackey KE, 2006. Lessons from the drug discovery of lapatinib, a dual ErbB1/2 tyrosine kinase inhibitor. Curr. Top. Med. Chem. 6:435–460. Lansing TJ, McConnell RT, Duckett DR, Spehar GM, Knick VB, Hassler DF, Noro N, Furuta M, Emmitte KA, Gilmer TM, Mook RA, Cheung M, 2007. In vitro biological activity of a novel small-molecule inhibitor of polo-like kinase 1. Mol. Cancer Ther. 6:450–459. Li R, Xue L, Zhu T, Jiang Q, Cui X, Yan Z, McGee D, Wang J, Gantla VR, Pickens JC, McGrath D, Chucholowski A, Morris SW, Webb TR, 2006. Design and synthesis of 5-arylpyridone-carboxamides as inhibitors of anaplastic lymphoma kinase. J. Med. Chem. 49:1006–1015. Liu Y, Gray NS, 2006. Rational design of inhibitors that bind to inactive kinase conformations. Nat. Chem. Biol. 2:358–364. Lombardo LJ, Lee FY, Chen P, Norris D, Barrish JC, Behnia K, Castaneda S, Cornelius LA, Das J, Doweyko AM, Fairchild C, Hunt JT, Inigo I, Johnston K, Kamath A, Kan D, Klei H, Marathe P, Pang S, Peterson R, Pitt S, Schieven GL, Schmidt RJ, Tokarski J, Wen ML, Wityak J, Borzilleri RM 2004. Discovery of N-(2-chloro-6-methyl- phenyl)-2-(6-(4-(2- hydroxyethyl)-piperazin1-yl)-2-methylpyrimidin-4-ylamino)thiazole-5-carboxamide(BMS-354825), a dual Src/Abl kinase inhibitor with potent antitumor activity in preclinical assays. J. Med. Chem. 47:6658–6661. Lowinger TB, Riedl B, Dumas J, Smith RA, 2002. Design and discovery of small molecules targeting raf-1 kinase. Curr. Pharm. Des. 8:2269–2278. Lyne PD, Kenny PW, Cosgrove DA, Deng C, Zabludoff S, Wendoloski JJ, Ashwell S, 2004. Identification of compounds with nanomolar binding affinity for checkpoint kinase-1 using knowledge-based virtual screening. J. Med. Chem. 47:1962–1968. Lyons JF, Wilhelm S, Hibner B, Bollag G, 2001. Discovery of a novel Raf kinase inhibitor. Endocr. Relat. Cancer 8:219–225. Maier JA, Brugel TA, Clark MP, Sabat M, Golebiowski A, Bookland RG, Laufersweiler MJ, Laughlin SK, Vanrens JC, De B, Hsieh LC, Brown KK, Juergens K, Walter RL, Janusz MJ,

154

STRATEGIES FOR DISCOVERING KINASE DRUGS

2006. Development of N-2 4-pyrimidine-N-phenyl-N’-alkyl ureas as orally active inhibitors of tumor necrosis factor alpha (TNF-alpha) synthesis Part 2. Bioorg. Med. Chem. Lett. 16:3514–3518. Manley PW, Bold G, Bruggen J, Fendrich G, Furet P, Mestan J, Schnell C, Stolz B, Meyer T, Meyhack B, Stark W, Strauss A, Wood J, 2004. Advances in the structural biology, design and clinical development of VEGF-R kinase inhibitors for the treatment of angiogenesis. Biochim. Biophys. Acta 1697:17–27. Manley PW, Cowan-Jacob SW, Mestan J, 2005. Advances in the structural biology, design and clinical development of Bcr-Abl kinase inhibitors for the treatment of chronic myeloid leukaemia. Biochim. Biophys. Acta 1754:3–13. Manning G, Plowman GD, Hunter T, Sudarsanam S, 2002. Evolution of protein kinase signaling from yeast to man. Trends Biochem. Sci. 27(10):514–520. McClure KF, Abramov YA, Laird ER, Barberia JT, Cai W, Carty TJ, Cortina SR, Danley DE, Dipesa AJ, Donahue KM, Dombroski MA, Elliott NC, Gabel CA, Han S, Hynes TR, Lemotte PK, Mansour MN, Marr ES, Letavic MA, Pandit J, Ripin DB, Sweeney FJ, Tan D, Tao Y, 2005. Theoretical and experimental design of atypical kinase inhibitors: application to p38 MAP kinase. J. Med. Chem. 48:5728–5737. McInnes C, 2006. Improved lead-finding for kinase targets using high-throughput docking. Curr. Opin. Drug. Discov. Devel. 9:339–347. McIntyre A, Summersgill B, Spendlove HE, Huddart R, Houlston R, Shipley J, 2005. Activating mutations and/or expression levels of tyrosine kinase receptors GRB7, RAS, and BRAF in testicular germ cell tumors. Neoplasia 7:1047–1052. Morwick T, Berry A, Brickwood J, Cardozo M, Catron K, DeTuri M, Emeigh J, Homon C, Hrapchak M, Jacober S, Jakes S, Kaplita P, Kelly TA, Ksiazek J, Liuzzi M, Magolda R, Mao C, Marshall D, McNeil D, Prokopowicz A, III, Sarko C, Scouten E, Sledziona C, Sun S, Watrous J, Wu JP, Cywin CL, 2006. Evolution of the thienopyridine class of inhibitors of IkappaB kinase-beta: part I: hit-to-lead strategies. J. Med. Chem. 49:2898–2908. Nagar B, Bornmann WG, Pellicena P, Schindler T, Veach DR, Miller WT, Clarkson B, Kuriyan J, 2002. Crystal structures of the kinase domain of c-Abl in complex with the small molecule inhibitors PD173955 and imatinib (STI-571). Cancer Res. 62:4236–4243. Nottage M, Siu LL, 2002. Rationale for Ras and raf-kinase as a target for cancer therapeutics. Curr. Pharm. Des. 8:2231–2242. O’Hare T, Walters DK, Stoffregen EP, Jia T, Manley PW, Mestan J, Cowan-Jacob SW, Lee FY, Heinrich MC, Deininger MW, Druker BJ, 2005. In vitro activity of Bcr-Abl inhibitors AMN107 and BMS-354825 against clinically relevant imatinib-resistant Abl kinase domain mutants. Cancer Res. 65:4500–4505. Ohren JF, Chen H, Pavlovsky A, Whitehead C, Zhang E, Kuffa P, Yan C, McConnell P, Spessard C, Banotai C, Mueller WT, Delaney A, Omer C, Sebolt-Leopold J, Dudley DT, Leung IK, Flamme C, Warmus J, Kaufman M, Barrett S, Tecle H, Hasemann CA, 2004. Structures of human MAP kinase kinase 1 (MEK1) and MEK2 describe novel noncompetitive kinase inhibition. Nat. Struct. Mol. Biol. 11:1192–1197. Okram B, Nagle A, Adrian FJ, Lee C, Ren P, Wang X, Sim T, Xie Y, Wang X, Xia G, Spraggon G, Warmuth M, Liu Y, Gray NS, 2006. A general strategy for creating ‘‘inactiveconformation’’ abl inhibitors. Chem. Biol. 13:779–786. Paolini GV, Shapland RH, van Hoorn WP, Mason JS, Hopkins AL, 2006. Global mapping of pharmacological space. Nat. Biotechnol. 24:805–815.

REFERENCES

155

Pargellis C, Tong L, Churchill L, Cirillo PF, Gilmore T, Graham AG, Grob PM, Hickey ER, Moss N, Pav S, Regan J, 2002. Inhibition of p38 MAP kinase by utilizing a novel allosteric binding site. Nat. Struct. Biol. 9:268–272. Petrov KG, Zhang YM, Carter M, Cockerill GS, Dickerson S, Gauthier CA, Guo Y, Mook Jr RA, Rusnak DW, Walker AL, Wood ER, Lackey KE, 2006. Optimization and SAR for dual ErbB-1/ErbB-2 tyrosine kinase inhibition in the 6-furanylquinazoline series. Bioorg. Med. Chem. Lett. 16:4686–4691. Pinna LA, 2003. The raison d’etre of constitutively active protein kinases: the lesson of CK2. Acc. Chem. Res. 36:378–384. Pratt DJ, Bentley J, Jewsbury P, Boyle FT, Endicott JA, Noble ME, 2006. Dissecting the determinants of cyclin-dependent kinase 2 and cyclin-dependent kinase 4 inhibitor selectivity. J. Med. Chem. 49:5470–5477. Richardson CM, Williamson DS, Parratt MJ, Borgognoni J, Cansfield AD, Dokurno P, Francis GL, Howes R, Moore JD, Murray JB, Robertson A, Surgenor AE, Torrance CJ 2006. Triazolo[1,5-a]pyrimidines as novel CDK2 inhibitors: protein structure-guided design and SAR. Bioorg. Med. Chem. Lett. 16:1353–1357. Rusnak DW, Affleck K, Cockerill SG, Stubberfield C, Harris R, Page M, Smith KJ, Guntrip SB, Carter MC, Shaw RJ, Jowett A, Stables J, Topley P, Wood ER, Brignola PS, Kadwell SH, Reep BR, Mullin RJ, Alligood KJ, Keith BR, Crosby RM, Murray DM, Knight WB, Gilmer TM, Lackey K, 2001a. The characterization of novel, dual ErbB-2/EGFR, tyrosine kinase inhibitors: potential therapy for cancer. Cancer Res. 61:7196–7203. Rusnak DW, Lackey K, Affleck K, Wood ER, Alligood KJ, Rhodes N, Keith BR, Murray DM, Knight WB, Mullin RJ, Gilmer TM, 2001b. The effects of the novel, reversible epidermal growth factor receptor/ErbB-2 tyrosine kinase inhibitor, GW2016, on the growth of human normal and tumor-derived cell lines in vitro and in vivo. Mol. Cancer Ther. 1:85–94. Sabat M, VanRens JC, Clark MP, Brugel TA, Maier J, Bookland RG, Laufersweiler MJ, Laughlin SK, Golebiowski A, De B, Hsieh LC, Walter RL, Mekel MJ, Janusz MJ, 2006. The development of novel C-2, C-8, and N-9 trisubstituted purines as inhibitors of TNFalpha production. Bioorg. Med. Chem. Lett. 16:4360–4365. Sayle KL, Bentley J, Boyle FT, Calvert AH, Cheng Y, Curtin NJ, Endicott JA, Golding BT, Hardcastle IR, Jewsbury P, Mesguiche V, Newell DR, Noble ME, Parsons RJ, Pratt DJ, Wang LZ, Griffin RJ, 2003. Structure-based design of 2-arylamino-4-cyclohexylmethyl-5nitroso-6-aminopyrimidine inhibitors of cyclin-dependent kinases 1 and 2. Bioorg. Med. Chem. Lett. 13:3079–3082. Scapin G, Patel SB, Lisnock J, Becker JW, LoGrasso PV, 2003. The structure of JNK3 in complex with small molecule inhibitors: structural basis for potency and selectivity. Chem. Biol. 10:705–712. Schindler T, Bornmann W, Pellicena P, Miller WT, Clarkson B, Kuriyan J, 2000. Structural mechanism for STI-571 inhibition of abelson tyrosine kinase. Science 289:1938–1942. Schnur D, Beno BR, Good A, Tebben A, 2004. Approaches to target class combinatorial library design. Methods Mol. Biol. 275:355–378. Sebolt-Leopold JS, Dudley DT, Herrera R, Van Becelaere K, Wiland A, Gowan RC, Tecle H, Barrett SD, Bridges A, Przybranowski S, Leopold WR, Saltiel AR, 1999. Blockade of the MAP kinase pathway suppresses growth of colon tumors in vivo. Nat. Med. 5:810–816. Shah NP, Tran C, Lee FY, Chen P, Norris D, Sawyers CL, 2004. Overriding imatinib resistance with a novel ABL kinase inhibitor. Science 305:399–401.

156

STRATEGIES FOR DISCOVERING KINASE DRUGS

Shewchuk L, Hassell A, Wisely B, Rocque W, Holmes W, Veal J, Kuyper LF, 2000. Binding mode of the 4-anilinoquinazoline class of protein kinase inhibitor: X-ray crystallographic studies of 4-anilinoquinazolines bound to cyclin-dependent kinase 2 and p38 kinase. J. Med. Chem. 43:133–138. Singh J, Chuaqui CE, Boriack-Sjodin PA, Lee WC, Pontz T, Corbley MJ, Cheung HK, Arduini RM, Mead JN, Newman MN, Papadatos JL, Bowes S, Josiah S, Ling LE, 2003. Successful shape-based virtual screening: the discovery of a potent inhibitor of the type I TGFbeta receptor kinase (TbetaRI). Bioorg. Med. Chem. Lett. 13:4355–4359. Smith RA, Barbosa J, Blum CL, Bobko MA, Caringal YV, Dally R, Johnson JS, Katz ME, Kennure N, Kingery-Wood J, Lee W, Lowinger TB, Lyons J, Marsh V, Rogers DH, Swartz S, Walling T, Wild H, 2001. Discovery of heterocyclic ureas as a new class of raf kinase inhibitors: identification of a second generation lead by a combinatorial chemistry approach. Bioorg. Med. Chem. Lett. 11:2775–2778. Sprous DG, Zhang J, Zhang L, Wang Z, Tepper MA, 2006. Kinase inhibitor recognition by use of a multivariable QSAR model. J. Mol. Graph Model. 24:278–295. Stamos J, Sliwkowski MX, Eigenbrot C, 2002. Structure of the epidermal growth factor receptor kinase domain alone and in complex with a 4-anilinoquinazoline inhibitor. J. Biol. Chem. 277:46265–46272. Stelmach JE, Liu L, Patel SB, Pivnichny JV, Scapin G, Singh S, Hop CE, Wang Z, Strauss JR, Cameron PM, Nichols EA, O’Keefe SJ, O’Neill EA, Schmatz DM, Schwartz CD, Thompson CM, Zaller DM, Doherty JB, 2003. Design and synthesis of potent, orally bioavailable dihydroquinazolinone inhibitors of p38 MAP kinase. Bioorg. Med. Chem. Lett. 13:277–280. Takami A, Iwakubo M, Okada Y, Kawata T, Odai H, Takahashi N, Shindo K, Kimura K, Tagami Y, Miyake M, Fukushima K, Inagaki M, Amano M, Kaibuchi K, Iijima H, 2004. Design and synthesis of Rho kinase inhibitors (I). Bioorg. Med. Chem. 12:2115–2137. Talpaz M, Shah NP, Kantarjian H, Donato N, Nicoll J, Paquette R, Cortes J, O’Brien S, Nicaise C, Bleickardt E, Blackwood-Chirchir MA, Iyer V, Chen TT, Huang F, Decillis AP, Sawyers CL, 2006. Dasatinib in imatinib-resistant Philadelphia chromosome-positive leukemias. N. Engl. J. Med. 354:2531–2541. Tamayo N, Liao L, Goldberg M, Powers D, Tudor YY, Yu V, Wong LM, Henkle B, Middleton S, Syed R, Harvey T, Jang G, Hungate R, Dominguez C, 2005. Design and synthesis of potent pyridazine inhibitors of p38 MAP kinase. Bioorg. Med. Chem. Lett. 15:2409–2413. Tang J, Shewchuk LM, Sato H, Hasegawa M, Washio Y, Nishigaki N, 2003. Anilinopyrazole as selective CDK2 inhibitors: design, synthesis, biological evaluation and X-ray crystallographic analysis. Bioorg. Med. Chem. Lett. 13:2985–2988. Tokarski JS, Newitt JA, Chang CY, Cheng JD, Wittekind M, Kiefer SE, Kish K, Lee FY, Borzillerri R, Lombardo LJ, Xie D, Zhang Y, Klei HE, 2006. The structure of Dasatinib (BMS-354825) bound to activated ABL kinase domain elucidates its inhibitory activity against imatinib-resistant ABL mutants. Cancer Res. 66:5790–5797. Traxler P, Furet P, Mett H, Buchdunger E, Meyer T, Lydon N, 1997. Design and synthesis of novel tyrosine kinase inhibitors using a pharmacophore model of the ATP-binding site of the EGF-R. J. Pharm. Belg. 52:88–96. von Ahsen O, Bomer U, 2005. High-throughput screening for kinase inhibitors. Chembiochem 6:481–490.

REFERENCES

157

Wallace EM, Lyssikatos JP, Yeh T, Winkler JD, Koch K, 2005. Progress towards therapeutic small molecule MEK inhibitors for use in cancer therapy. Curr. Top. Med. Chem. 5:215–229. Wan PT, Garnett MJ, Roe SM, Lee S, Niculescu-Duvaz D, Good VM, Jones CM, Marshall CJ, Springer CJ, Barford D, Marais R, 2004. Mechanism of activation of the RAF-ERK signaling pathway by oncogenic mutations of B-RAF. Cell 116:855–867. Wang Z, Canagarajah BJ, Boehm JC, Kassisa S, Cobb MH, Young PR, Abdel-Meguid S, Adams JL, Goldsmith EJ, 1998. Structural basis of inhibitor selectivity in MAP kinases. Structure 6:1117–1128. Ward WH, Cook PN, Slater AM, Davies DH, Holdgate GA, Green LR, 1994. Epidermal growth factor receptor tyrosine kinase. Investigation of catalytic mechanism, structurebased searching and discovery of a potent inhibitor. Biochem. Pharmacol. 48:659–666. Wilhelm S, Chien DS, 2002. BAY 43–9006: preclinical data. Curr. Pharm. Des. 8:2255–2257. Wilhelm SM, Carter C, Tang L, Wilkie D, McNabola A, Rong H, Chen C, Zhang X, Vincent P, McHugh M, Cao Y, Shujath J, Gawlak S, Eveleigh D, Rowley B, Liu L, Adnane L, Lynch M, Auclair D, Taylor I, Gedrich R, Voznesensky A, Riedl B, Post LE, Bollag G, Trail PA, 2004. BAY 43–9006 exhibits broad spectrum oral antitumor activity and targets the RAF/ MEK/ERK pathway and receptor tyrosine kinases involved in tumor progression and angiogenesis. Cancer Res. 64:7099–7109. Wilhelm S, Carter C, Lynch M, Lowinger T, Dumas J, Smith RA, Schwartz B, Simantov R, Kelley S, 2006. Discovery and development of sorafenib: a multikinase inhibitor for treating cancer. Nat. Rev. Drug. Discov. 5:835–844. Williamson DS, Parratt MJ, Bower JF, Moore JD, Richardson CM, Dokurno P, Cansfield AD, Francis GL, Hebdon RJ, Howes R, Jackson PS, Lockie AM, Murray JB, Nunns CL, Powles J, Robertson A Surgenor AE, Torrance CJ, 2005. Structure-guided design of pyrazolo[1,5a]pyrimidines as inhibitors of human cyclin-dependent kinase 2. Bioorg. Med. Chem. Lett. 15:863–867. Wood ER, Truesdale AT, McDonald OB, Yuan D, Hassell A, Dickerson SH, Ellis B, Pennisi C, Horne E, Lackey K, Alligood KJ, Rusnak DW, Gilmer TM, Shewchuk L, 2004. A unique structure for epidermal growth factor receptor bound to GW572016 (Lapatinib): relationships among protein conformation, inhibitor off-rate, and receptor activity in tumor cells. Cancer Res. 64:6652–6659. Wu SY, McNae I, Kontopidis G, McClue SJ, McInnes C, Stewart KJ, Wang S, Zheleva DI, Marriage H, Lane DP, Taylor P, Fischer PM, Walkinshaw MD, 2003. Discovery of a novel family of CDK inhibitors with the program LIDAEUS: structural basis for ligand-induced disordering of the activation loop. Structure 11:399–410. Xu W, Doshi A, Lei M, Eck MJ, Harrison SC, 1999. Crystal structures of c-Src reveal features of its autoinhibitory mechanism. Mol. Cell. 3:629–638. Zimmerman J, Buchdunger E, Mett H, Meyer T, Lydon NB, Traxler P, 1996. Phenylaminopyrimidine (PAP)-derivatives: a new class of potent and highly selective PDGF-receptor autophosphorylation inhibitors. Bioorg. Med. Chem. Lett. 6:1221–1226. Zimmerman J, Buchdunger E, Mett H, Meyer T, Lydon NB, 1997. Potent and selective inhibitors of the Abl-kinase: phenylaminopyrimidine (PAP) derivatives. Bioorg. Med. Chem. Lett. 7:187–192.

6 PROTEASE-DIRECTED DRUG DISCOVERY RICHARD SEDRANI, ULRICH HOMMEL,

6.1

AND

JO¨RG EDER

INTRODUCTION

Proteases are essential enzymes, which catalyze the cleavage of peptide bonds. They perform vital functions in all living organisms. Originally, proteases were recognized as being involved in the nonspecific degradation of dietary polypeptides. However, recent advances have provided a new view of the proteolytic world with proteolysis being important for almost any biological process (Barrett et al., 1998). In addition to mediating nonspecific hydrolysis of polypeptides, proteases catalyze the selective, limited, and efficient cleavage of specific protein substrates, thus being important regulators of posttranslational modification. Processes as diverse as the activation or inactivation of cytokines, hormones or growth factors, the cellular localization of proteins, or the shedding of receptors and protein ligands from the cell surface are all regulated by proteases. Consequently, alterations in the activity of proteases underlie many pathological processes, including cancer, arthritis, cardiovascular diseases, osteoporosis, and neurodegeneration. Moreover, many infectious organisms are highly dependent on their ability to generate essential proteins through effective proteolytic processing. The importance of proteolysis in biological systems is also reflected by the finding that nature has separately invented the underlying catalytic mechanism multiple times with different solutions. This provides the basis for the categorization of proteases into four main classes: serine/threonine, cysteine, metallo- and aspartic proteases, depending on their catalytic mechanism and the residue that affects enzymatic

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

159

160

PROTEASE-DIRECTED DRUG DISCOVERY

Scissile bond N-term

C-term

P4

P3

P2

P1

P1’ P2’

P3’

P4’

Substrate

S4

S3

S2

S1

S1′

S3′

S4′

Protease

FIGURE 6.1

S2′

Schematic representation of substrate binding.

hydrolysis. In general, proteases use two fundamentally different mechanisms to stabilize the tetrahedral transition state during catalysis. Metallo- and aspartic proteases use an activated water molecule as a nucleophile to attack the peptide bond of the substrate, whereas in serine/threonine and cysteine proteases, the nucleophile is an amino acid residue of the protease polypeptide chain. The latter mechanism usually requires a histidine as base and a covalent acyl–enzyme intermediate. All proteases bind their substrates in a groove or cleft, where substrate hydrolysis occurs. The substrate specificity is determined by enzyme subsites in the groove, numbered S1, S2, . . ., Sn from the scissile bond toward the N-terminus of the substrate (nonprimed site) and S10 , S20 , . . ., Sn0 from the scissile bond toward the C-terminus of the substrate (primed site), as shown in Fig. 6.1 (Schechter and Berger, 1967). The substrate residues they accommodate are designated as P1, P2, . . ., Pn on the nonprimed and P10 , P20 , . . ., Pn0 on the primed site. Depending on the size and architecture of the subsites, the specificity of proteases can vary greatly with some proteases showing high fidelity in their ability to recognize a unique protein, whereas others are rather indiscriminate against many different substrates. Another level of substrate specificity may be provided by exosites that are substrate binding sites distant to the active site (Page et al., 2005). These exosites are usually very specific for a given substrate or substrate class and may compensate for an indiscriminate active site binding groove. Proteases are one of the largest enzyme families encoded by the human genome with 512 active members (Puente et al., 2003; MEROPS database, http://merops. sanger.ac.uk). They are not only divided into different classes, but also into different families and clans based on their amino acid sequence homology and three-dimensional fold, respectively. This classification is summarized in Table 6.1. The biggest class of human proteases is serine/threonine proteases with 193 active members organized in 18 different families. Human metalloproteases (MMPs) can be divided into 22 families with a total of 162 active members, and there are 141 cysteine proteases spread across 17 families. The smallest class of human proteases is aspartic proteases with only 16 members arranged in 2 families. In addition to the active proteases, there are many inactive proteases encoded in the human genome, which account for about 15% of all proteases. These inactive protease homologues have amino acid substitutions in one or several specific residues that are located in critical active site regions and might have important functions as regulatory or inhibiting molecules.

161

INTRODUCTION

TABLE 6.1

Protease Classification

Protease Classes Aspartic Cysteine

Serine/Threonine

Metalloprotease

Families (Number of members) A1 (9), A22 (7) C1 (14), C2 (14), C12 (4), C13 (2), C14 (13), C15 (2), C19 (55), C26 (1), C44 (1), C46 (3), C48 (8), C50 (1), C54 (4), C56 (4), C64 (5), C65 (9), C67 (1) S1 (111), S8 (10), S9 (18), S10 (3), S12 (1), S14 (1), S16 (2), S26 (4), S28 (3), S33 (3), S53 (1), S54 (5), S59 (1), S60 (1), S63 (8), T1 (8), T2 (4), T3 (9) M1 (12), M2 (3), M3 (3), M8 (3), M10 (24), M12 (39), M13 (7), M14 (23), M16 (4), M17 (3), M18 (1), M19 (3), M20 (3), M22 (2), M24 (8), M28 (6), M41 (5), M43 (2), M48 (2), M49 (1), M50 (1), M67 (7)

Total Number 16 141

193

162

Total ¼ 512

Several proteases have already been followed as drug targets by the pharmaceutical industry (Abbenante and Fairlie, 2005). Inhibitors of the angiotensin converting enzyme or of the HIV protease are well-established drugs in the treatment of hypertension or acquired immunodeficiency syndrome, respectively. In addition, a number of protease inhibitors have recently entered the market or are in late clinical trials for multiple myeloma (proteasome inhibitor), diabetes (DPP4 inhibitors), and hypertension (renin inhibitors). Despite these successes, however, drug discovery in the field of proteases has not resulted in as many drugs as could have been expected, given the significant efforts and investments made in this area. In the past, key hurdles in the field of protease-directed drug discovery have been the inability to appropriately assess inhibitor selectivity and the difficulty to obtain orally active drugs (Turk, 2006). An illuminating case study with respect to inhibitor specificity is matrix metalloprotease inhibitors, which were in clinical trials as anticancer drugs (Overall and Kleifeld, 2006). When the first drug discovery efforts for MMPs were described almost 20 years ago, only three members of the MMP family were known and MMPs were viewed solely as extracellular matrix-degrading enzymes. Today, however, we appreciate the complexity of their biology with 24 MMP family members and their important roles in homeostatic regulation of the extracellular environment and for controlling innate immunity, and not simply to degrade extracellular matrix as their name suggests. Moreover, whereas some MMPs promote tumorigenesis and cancer cell growth by participating in many deregulated-signaling pathways, at least three MMP family members (MMP-3, -8 and, -9) have tumor-protective functions and for cancer therapy, they

162

PROTEASE-DIRECTED DRUG DISCOVERY

are rather antitargets. These targets must be therapeutically avoided to prevent worsening of the disease. Most of the currently known MMP inhibitors lack enzyme specificity due to the lack of exhaustive testing (in part due to the limited biological knowledge at the time they were synthesized) and inappropriate inhibitor design and, consequently, many of them have failed in clinical cancer trials as a result of inefficacy or unacceptable side effects. With the knowledge on all proteases encoded by the human genome, it is now possible to start to address these issues in a more systematic way. In terms of inhibitor design, the most common approach in the past has centered on the peptide sequence of the substrate around the cleavage site, or parts thereof, as a starting point. In addition, the inhibitor backbones were combined either with highly reactive electrophilic functionalities designed to target the nucleophilic catalytic residues of serine, threonine, or cysteine proteases, or with metal chelating groups to target metalloproteases. As peptides usually are metabolically unstable and possess low oral bioavailability, parts of the sequence were replaced by peptidomimetics. In general, this peptidomimetic approach rapidly leads to inhibitors that exhibit excellent affinity for the targeted protease. However, they usually retain some degree of peptidic character. The use of electrophilic isosteres such as aldehydes, ketones, and boronic acids is also often at the expense of selectivity due to their high reactivity with many nucleophilic residues including those of other proteases. Therefore, although the peptidomimetic approach (either with or without the use of electrophilic isosteres) has led to many potent protease inhibitors, it has met with limited success in the clinic, the main reasons being poor bioavailability and unacceptable side effects due to lack of selectivity. To be effective drugs, protease inhibitors need to have minimal peptidic character, high metabolic stability, good membrane permeability, long half-lives in the bloodstream and in cells, low susceptibility to elimination, and good bioavailability (preferably by oral delivery). In addition, they need to be potent for their target and specific toward most, if not all, other proteases. Here we describe past and current approaches to discover, design, and optimize different chemotypes for the inhibition of all the four major protease classes, respectively. These chemotypes are either based on structure-based design or on a variety of screening methods, including high-throughput screening of large compound collections and virtual screening or fragment-based screening, and qualify (or may qualify) as effective oral protease drugs.

6.2

ASPARTIC PROTEASES

Efforts to achieve pharmacological inhibition of aspartic proteases have been made in the field of HIV protease and renin. HIV protease is an essential enzyme in the life cycle of the human immunodeficiency virus and cleaves the viral polyprotein. Interference with the proteolytic activity of this protease renders the viral particles immature and noninfectious, and several HIV protease inhibitors are now on the market for the treatment of HIV infection/acquired immunodeficiency syndrome.

ASPARTIC PROTEASES

163

Renin cleaves angiotensinogen to release the N-terminal decapeptide angiotensin I. It catalyzes the first and rate-limiting step of the renin angiotensin system, an enzymatic cascade leading to the production of the potent vasoconstrictor hormone angiotensin II. In addition to these two aspartic protease drug targets, many pharmaceutical companies are now also working on BACE (b-amyloid precursor proteincleaving enzyme), a membrane-bound aspartic protease expressed in the brain. BACE activity is responsible for the generation of amyloidogenic peptides that constitute the major components of amyloid plaques accumulating in the brain of Alzheimer’s disease patients (Hardy and Selkoe, 2002). The crystal structure of many human aspartic proteases has been solved (Andreeva and Rumsh, 2001). They follow a general fold with three topologically distinct regions, an N-terminal domain, a C-terminal domain, and a six-stranded antiparallel b-sheet interdomain connecting the two other domains. Both, the N-and C-terminal domains, each contribute one catalytic aspartic acid residue to the active site, and most aspartic proteases have a flap that closes down on top of the substrate or inhibitor, shielding the active site from solvent and forming the binding pockets on both sides of the catalytic residues. Peptide bond cleavage occurs by a general acid–base catalytic mechanism, as shown in Fig. 6.2, with one of the two catalytic aspartic residues being protonated in the enzyme/substrate complex (Dunn, 2002). The other aspartic residue acts as a general base by activating a water molecule, which then attacks the carbonyl carbon of the substrate amide bond yielding the formation of a tetrahedral geminal diol intermediate. The subsequent deprotonation of the hydroxyl group by one of the catalytic aspartic acid residues and simultaneous activation of the leaving amine by the other, protonated, aspartic residue ultimately leads to peptide bond cleavage. Initially, many aspartic protease inhibitors were based on substrate-derived polypeptides with the scissile amide bond being replaced by a noncleavable transition state isostere (Dash et al., 2003; Abbenante and Fairlie, 2005). The naturally occurring inhibitor pepstatin A, containing a statin, has been a model compound in this respect and several other inhibitory principles have been established, as shown in Fig. 6.3. While these statin- or hydroxyethylene-based inhibitors in general lead to potent enzymatic inhibition, their use as therapeutic agents is often hampered by

FIGURE 6.2 Catalytic mechanisms of aspartic proteases. The two catalytic aspartic residues are shown in grey and the substrate peptide in black.

164

PROTEASE-DIRECTED DRUG DISCOVERY OH

H N

O N H

R

OH

H N R

(a)

OH

H N

O

(d)

R2

R1

H N

R2 R1

O

OH

OH

H N

OH

(e)

(b)

H N

H N

H N

R

OH

H N

P R

(c)

O

O N H

(f)

FIGURE 6.3 Inhibitory principles for aspartic proteases. Transition state isosteres: (a) statine, (b) homostatine, (c) hydroxyethylamine, (d) hydroxyamide, (e) diol, and (f) phosphinate.

their unfavorable biopharmaceutical properties due to their peptidic character. Therefore, substantial chemical modification is usually required to successfully design orally active drugs. 6.2.1

HIV Protease Inhibitors

Over the past 20 years, many structural and mechanistic insights have been gained by studying the HIV protease and many inhibitory principles have been successfully established and refined. HIV protease inhibitors are the first aspartic protease inhibitors, which have reached the market and as such also serve as prototype inhibitors for the whole enzyme family. HIV protease is a homodimeric enzyme, comprising two noncovalently associated subunits. Both subunits contribute equally to the formation of the active site cleft, with each subunit providing one catalytic aspartic acid residue. Many HIV protease inhibitors are based on the protease’s natural cleavage sites in the HIV gag and gag-pol gene products, using a PhePro core that reflects the scissile bond. These peptides have often been used as a starting point to design transition state isosteres and oral bioavailability was achieved through subsequent modifications. So far, nine HIV protease inhibitor drugs are clinically available (Barbaro et al., 2005). Saquinavir (Invirase1) 1 was approved in 1996 as the first HIV protease inhibitor and is one of the most prominent examples of a structure-based designed drug (Roberts et al., 1990). As shown in Fig. 6.4, the first step was the replacement of the Phe-Pro scissile bond with a hydroxyethylamine transition state analogue in the pentapeptide derived from the natural cleavage site, giving rise to inhibitor 2. Searching for the minimal sequence

165

ASPARTIC PROTEASES

(a) Leu

Asn

Phe

Pro

Ile

Z

Leu

2

Asn

OH

IC50 0.75 µM

(c) Z

Asn

N H

N OH

O

O

Ile

N H

O OtBu

O

N

IC50 0.14 µM

1 Saquinavir

H

OH O

H2N

3

NHtBu

H

O

H N

N

(b)

N

N H

N H

K I 0.00012 µM

Z = benzyloxycarbonyl tBu = tert-butyl

FIGURE 6.4 Rational design of the HIV protease inhibitor saquinavir 1. Individual steps in the design and discovery of the final molecule starting from a substrate-derived peptide sequence. (a) Replacement of scissile bond with hydroxyethylamine transition state analogue. (b) Search for minimal sequence required for potent inhibition. (c) Optimization of individual side chains and terminal substituents.

required for potent inhibition led to the tripeptide Z-Asn-Phec[CH(OH)CH2N]Pro-Ot-Bu 3 with an IC50 value of 0.14 mM for inhibition of the protease. Further optimization of individual side chains and terminal substituents with respect to steric and electronic properties resulted in the replacement of the P10 prolyl residue by (S,S,S)-decahydro-isoquinoline-3-carbonyl and the addition of a quinoline-2carbonyl moiety to effectively fill the S3 pocket. One synthetic route toward saquinavir 1, which has a KI value of 0.12 nM, is shown in Fig. 6.5. Beginning with a chloromethyl derivative of phenylalanine, the ketone is reduced to the alcohol and then cyclized under basic conditions. The highly functionalized cyclic amine is used to ring open the oxirane and then to remove the protection group via hydrogenation. A standard amide bond coupling reaction is used, followed by another deprotection to accomplish the final amide bond forming reaction. Many of the other peptidomimetic HIV protease inhibitors are based on similar design principles. Nelfinavir (Viracept1), for example, has the same P10 moiety, but an extended P1 substituent to increase lipophilicity and improve oral bioavailability (Kaldor et al., 1997). Further systematic modification of the peptide core led to inhibitors with even higher potency and/or better oral bioavailability such as Ritonavir (Norvir1), Lopinavir (Kaletra1), and Indinavir (Crixivan1) (Eder et al., 2007). Moreover, the availability of numerous crystal structures in complex with inhibitors has also facilitated the discovery of potent and selective nonpeptidomimetic inhibitors. Of these, Tipranavir (Aptivus1), a sulfonamide-based inhibitor, is already on the market (Kandula et al., 2005).

166

PROTEASE-DIRECTED DRUG DISCOVERY

H NaBH4

KOH

Cl

CbzNH

Cl

CbzNH (Diastereomeric ratio 3:1)

O

+

CbzNH

HN H

O

OH

O N H

H

2-propanol, 80°C

H

OH O N H

N H

1. H2, Pd-C

H

O CbzNH

N H O

H2 N

N H

OH O

CbzAsnOH, DCC, HOBt

N

H2 N

H

OH O

H

H2 , Pd-C

N

CbzNH

2. Quinoline-2-carboxylic acid DCC, HOBt

N H

N

H

O

H N

N H

O

O

N H

OH O

H2 N

N H

1 Saquinavir

FIGURE 6.5

6.2.2

Synthesis scheme for saquinavir 1.

Renin

Similar to the approach taken for the HIV protease inhibitors, many renin inhibitors were based on the peptide sequence derived from the enzyme’s natural cleavage site in angiotensinogen. In this case, however, the scissile bond was replaced with either a statine or homostatine transition state analogue. The structure of renin complexed with such inhibitors has revealed how the active site of renin can accommodate substrates spanning the specificity pockets S4–S40 . An interesting feature of the renin active site is that the S1 and S3 pockets form a contiguous and large hydrophobic cavity with a subpocket termed S3sp (Rahuel et al., 2000). As in other aspartic proteases, a flexible flap covers the active site cleft and plays an important role in the enzyme–substrate interactions. Moreover, protein flexibility and secondary structure rearrangements within the active site, for example, in S10 and S20 , are important in understanding enzyme specificity (Cooper, 2002). Firstgeneration renin inhibitors started with tetrapeptides (Phe-His-Leu-#-Val, the arrow marks the scissile bond) and several highly potent homostatine transition state mimetics were synthesized and tested in human volunteers, including CGP038560A (IC50 value of 1 nM) (Wood et al., 1989), Remikiren (IC50 value of 0.7 nM) (Fischli et al., 1991), and Zankiren (IC50 value of 1.1 nM) (Kleinert et al., 1992). However, insufficient oral bioavailability, due to their rather peptidic character, prevented their further development. The combined use of structural information from renin inhibitor cocomplexes and computer-assisted molecular modeling allowed the reduction of the molecular weight and peptidic character of renin inhibitors. One of these is Aliskiren 4 (SPP100), the first orally bioavailable, selective, and highly potent (IC50 value of 0.6 nM) renin

ASPARTIC PROTEASES

167

FIGURE 6.6 Synthetic route toward Aliskiren 4 (Ru¨eger et al., 2000; Go¨schke et al., 2003) and the crystal structure of the renin/4 complex (2v0z, Rahuel et al., 2000). Residues from the flap covering the active site are omitted for clarity. Pictures of three-dimensional structures were generated using PyMOL (Delano, 2002).

inhibitor that is now on the market for the treatment of hypertension (Wood et al., 2003). The synthetic route is given in Fig. 6.6. Based on a dipeptide-like hydroxyethylene transition state mimetic derived by molecular modeling methods, compounds were synthesized in the first step that lacked the P1–P4 spanning backbone of previous peptidic inhibitors. In particular, the optimal use of the large S1/S3 pocket and the addition of the methoxypropoxy side chain, which is optimally filling the S3sp subpocket, resulted in highly potent compounds with selectivity over all other human aspartic proteases. Further optimization, especially in the P20 position, led to the final compound, with the terminal carboxamide group being involved in an additional hydrogen bond interaction with the enzyme and the geminal methyl residues providing hydrophobic van der Waals interactions with the S20 pocket, as shown in Fig. 6.6. In addition to transition state mimetics, novel compounds such as 3,4-disubstituted piperidines have been identified by screening large compound libraries. The binding of these compounds to the active site aspartic residues (Asp 32 and Asp 215) occurs via the protonated piperidine nitrogen (Oefner et al., 1999). Their further optimization led to 3,4,5-trisubstituted piperidines (Ma¨rki et al., 2001), which were found to induce an open flap conformation, thereby opening a large hydrophobic pocket. In addition, novel compound classes derived from further optimization, such as disubstituted amino-aryl piperidines and ketopiperazines, have been reported recently (Cody et al., 2005; Holsworth et al., 2005).

168

6.3

PROTEASE-DIRECTED DRUG DISCOVERY

METALLOPROTEASES

Metalloproteases represent the second largest class of proteases in the human genome, with 162 potentially active members. They are distinct from the other proteolytic enzymes in that they have an absolute requirement for a divalent metal ion in the active site. This is a zinc ion in most of the enzymes, but in a few cases also copper, nickel, and manganese have been reported. For the majority of metalloproteases, the consensus sequence for the zinc binding motif is HEXXH. Two histidine side chains from this motif and an additional histidine or glutamic acid contribute three ligands to the tetrahedral coordination of the metal ion. The fourth ligation comes from the catalytic water molecule, which is close to the catalytically important glutamic acid side chain of the zinc binding motif. Metalloproteases have attracted much interest in the pharmaceutical industry for more than 30 years, in part with spectacular success. Inhibitors of angiotensinconverting enzyme (ACE) are widely prescribed today to treat high blood pressure and are given to millions of patients worldwide. Captopril, one of these inhibitors, was the first marketed protease inhibitor. In addition, matrix metalloproteases have been targets for cancer therapy, rheumatoid arthritis, and osteoarthritis, and a number of MMP inhibitors have been studied in clinical trials. The subsequent discussion concentrates on ACE and MMPs, since drug discovery in the field of metalloproteases is best exemplified by these cases. The catalytic mechanism of metalloproteases has been studied extensively and, in particular, the crystal structures of a wealth of thermolysin/peptide complexes have shed light onto different stages on the reaction path (Matthews, 1988). The catalysis is shown in Fig. 6.7. The incoming substrate binds with the carbonyl group of its scissile bond to the zinc ion and thereby pushes the zinc-bound water closer toward the side chain of the catalytic glutamic acid. As a consequence, the water molecule becomes activated and adopts a position suited for the nucleophilic attack of the scissile bond. After reshuffling of a proton to the N-terminus of the leaving peptide, the scissile bond is cleaved and the new fragments released. As will be seen subsequently, it is the understanding of this mechanism that has, for a long time, governed the design of inhibitors targeting metalloproteases such as ACE and MMPs. 6.3.1

Angiotensin-Converting Enzyme

Angiotensin-converting enzyme, also known as peptidyl dipeptidase A, catalyzes the generation of the vasopressive peptide hormone angiotensin II by cleaving

FIGURE 6.7 Catalytic mechanisms of metalloproteases (thermolysin).

METALLOPROTEASES

169

the two C-terminal residues from the inactive precursor angiotensin I. ACE is also known to inactivate the vasodilator bradykinin and has therefore gained interest in the early 1970s as target for the treatment of hypertension and congestive heart failure. Drug discovery efforts started with the discovery of ‘‘bradykinin-potentiating peptides’’ (BPPs, 5) in the venom of the Brazilian viper Bothrops jararaca. As these peptides also inhibit ACE in vitro and have blood pressure-lowering effects in vivo, extensive work went into understanding the structural features important for ACE inhibition. This work revealed the importance of the terminal carboxylate function, the preference for proline in the ultimate, alanine in the penultimate, and an aromatic side chain in the antepenultimate position. Despite the failure to develop an orally active derivative of these peptides, the knowledge gained from these studies was important for the rational design of nonpeptidic follow-ups. Structure-based drug design was still in its infancy in the early 1970s and only few crystal structures of proteins had been solved by then. One of them, which later turned out to be central for guiding medicinal chemistry efforts on ACE, was the zinc-dependent metalloprotease carboxypeptidase A (CPA) (Reeke et al., 1967). This structure indicated for the first time that the catalytic zinc ion was coordinated by the carbonyl group of the scissile bond of the substrate. Furthermore, replacing the scissile bond with a carboxymethylene group in a CPA substrate yielded the relatively potent inhibitor L-benzylsuccinic acid 6 (KI value for CPA of 0.5 mM)) (Byers and Wolfenden, 1972). Based on these findings, it was realized that ACE could be inhibited using the same design principles that are summarized in Fig. 6.8 (Ondetti et al., 1977). As a consequence, a series of short dipeptide analogues of the snake venom peptides were synthesized. This effort resulted in the succinyl amino acid derivative 7, which was only moderately active in vitro (IC50 value of 22 mM) but was potent enough to show oral activity in hypertensive rats. Further replacement of the zinc-coordinating carboxylate by a sulfhydryl group, as in compound 8, not only dramatically increased its in vitro potency (IC50 value of 0.023 mM), but also further improved the in vivo profile of the inhibitor. This compound was launched in 1981 under the name captopril. Further development in the field of ACE inhibition was directed toward replacing the thiol group of captopril 8, which was considered to be the reason for unwanted side effects observed in the clinic at high doses. Different research groups, therefore, returned to the weaker carboxylate as the zinc binding motif (Patchett and Cordes, 1985). A switch from the succinyl-proline to the N-alkylated alanine-proline scaffold afforded, at the same time, an extension into the nonprimed site (S1 subsite). The additional interaction energy gained is compensating for the weaker interaction of the carboxylate with the catalytic zinc ion. These efforts resulted in the successful design of enalaprilat 9 and its close analogue lisinopril, both inhibiting ACE with low nanomolar potency (IC50 value of 0.0012 mM). It is interesting to note that the ACE inhibitor design concept, which was based on considerations derived from the three-dimensional structure of CPA, led to potent ACE inhibitors with little activity toward CPA. Only recently, the threedimensional structure of the testis ACE/enalaprilat complex (Fig. 6.9) has been determined showing that its active site is in fact only distantly related to that of CPA (Natesh et al., 2004). However, the structure nicely corroborates the general

170

PROTEASE-DIRECTED DRUG DISCOVERY

(a)

(b)

S1′

Zn2+

S2′

Zn2+ S ′ H 1

+

+

HN

R

O

O

N H

O

N H

R

Glu-Lys

O

O

O

N H

H3N

+

O O

O

O

O N H

N H

5

O

7

O

8

O

9

O

O CH2

O

N H

O

CH2

O

O

O

O O

O

S

6

CH2

N H

O CH2

O

CH2

N H

O

O N H

FIGURE 6.8 By-product designs for carboxypeptidase A (CPA). (a) Schematic view of the interaction of CPA with its substrate, products, and the by-product analogue benyzlsuccinate 6. (b) Schematic view of the interaction of ACE with different inhibitors: pentapeptide BPP5a 5, a succinylamino acid 7, captopril 8, and enalaprilat 9. Figure adopted from Ondetti et al., 1977.

principles followed in the design of ACE inhibitors and shows in particular the importance of the metal chelating group. The design of different metal chelating groups has hence become the cornerstone of many drug design efforts for metalloenzymes, including matrix metalloprotease. 6.3.2

Matrix Metalloproteases

MMPs are not only degrading extracellular matrix components, but are also involved in a process called proteolytic processing, whereby receptors, cofactors, proteins, and protein inhibitors become activated upon proteolytic cleavage. MMPs, therefore, have emerged as potential drug targets for a variety of indications including cancer, rheumatoid arthritis, and osteoarthritis. However, none of the MMP inhibitors studied in clinical trials so far has been launched to the market.

METALLOPROTEASES

171

FIGURE 6.9 Enalaprilat 9 and crystal structure of the testis ACE/9 complex. Parts of the structure are omitted for clarity (1uze, Natesh et al., 2004).

The 24 different MMPs are members of the M10 family and belong to the metzincin superfamily. Together with members of the closely related M12 family, which includes the ADAM (a disintegrin and metalloprotease) and the ADAMTS (a disintegrin and metalloprotease with thromobospondin Type 1 motifs), the MMPs share the extended zinc binding motif HEXXHXXGXXH. The zinc ion of the metzincins is coordinated by three histidine residues. The catalytically important glutamic acid is flanked by the imidazole side chains of the first two histidines and it interacts with the catalytically important water molecule, similar to the situation as described above for thermolysin/ACE. A peculiarity of the metzincin superfamily is the presence of a beta-turn underneath the active site, which contains a conserved methionine residue and gives this family its unique name (Sto¨cker et al., 1995). A wealth of information is available from a number of crystal structures of MMPs complexed with both natural and synthetic inhibitors (Maskos, 2005). These structures provide detailed insight into the determinants of MMP/inhibitor interactions and have been used extensively in the structure-based drug design. Most of the MMPs contain a hydrophobic S1’, a shallow S2’, and a solvent exposed S3’ subsite. By contrast, the nonprimed site does not contain well-formed pockets and, therefore, drug discovery efforts have focused on utilizing interactions within the S1’ pocket in addition to coordination of the active site zinc ion. Hydroxamic acids have been recognized early on as good chelators for the catalytic zinc ion. This observation can be rationalized by the multiple interactions made between this functional group and the MMP active site, as given in Fig. 6.10. The concerns about a nonspecific metal chelating property of such inhibitors in general and their potential metabolic instability in particular prompted many research groups to also look for an alternative metal binding groups. As for

172

O

2+

O

Zn

O

Glu223

HO

OH

N

H

O

O

N H

O

R

Ala161

O

Ala186

N H

2+

O

O

Zn

H N Ala160

Glu202

(e)

Glu402

(b)

O

O

OH

2+

10

H

Zn

OH

N

S

N

H

S

2+

HN

O

Zn

O

O

Ala167

N H

N

OH

N H

R

(f)

(c)

H

N

Ala163

R

Glu223 OH

N

Glu198

O

S

H N

O

N

11

O

O

O

2+

Zn

R

2+

Zn

FIGURE 6.10 Hydroxamate- and nonhydroxamate-based inhibitors. Key interactions of MMP-inhibitors with their target enzyme are shown. (a) MMP-13/hydroxamate (Zhang et al., 2000). (b) MMP-9/retrohydroxamte (1gkc, Rowsell et al., 2002). (c) MMP-8/carboxylate (1bzs, Matter et al., 1999). (d) MMP-8/pyrimidine-2,4,6-trione (1jj9, Brandstetter et al., 2001). (e) MMP-3/thiadiazole 10 (1usn, Finzel et al., 1998). (f) MMP8/thiadiazine 11 (1jh1, Schro¨der et al., 2001).

Glu198

(d)

(a)

METALLOPROTEASES

173

ACE, carboxylates have been used in the design of MMP inhibitors as well. However, in general, these inhibitors are less potent when compared to their hydroxamic acid analogues (Miller et al., 1997). Additional interaction energy have thus to be gained at other sites, for example, in the S1’ pocket, to achieve the same level of potency. In addition, sulfur- containing groups such as thiols, thiadiazoles 10, and thiadiazines 11 have been investigated and low nanomolar inhibitors identified (Fig. 6.10) (Finzel et al., 1998; Schro¨der et al., 2001). Interestingly, these types of inhibitors coordinate the zinc ion in different ways, thus offering new opportunities for targeting the specificity pockets otherwise not reachable by hydroxamic acidderived compounds. Recently, new metal chelating groups based on hydroxamic acid isosteres have been reported (Puerta and Cohen, 2003) and structural models with their zinc ion interactions in surrogate systems are available. Batimastat 12 is a prototypic hydroxamic acid-based MMP inhibitor with broad specificity for many family members. Its binding to the active site of MMP-3 is shown in Fig. 6.11. The fragment-based approaches using in silico or NMR screening have also been employed as a means to identify novel zinc binding groups. By screening a series of prototypic fragments, these studies confirmed previous findings that the hydroxamic acid moiety is superior to many other metal chelators (Hajduk et al., 2002). In this context it is interesting to note that acethydroxamic acid, the minimal component of a classical hydroxamic acid-based inhibitor, has only weak affinity for the protease (KD value for MMP-13 of 17 mM) (Hajduk et al., 1997). By contrast, biphenyl and biphenylether derivates bind to the S1’ site of MMP-13 with KD values in the range of 100–400 mM. This underscores the relative importance of the two main interaction sites of today’s MMP inhibitors and emphasizes the importance of the S1’ subsite for inhibitor design. The size of the S1’ pocket varies between the different MMPs and attempts have been made to exploit this pocket in the design of selective inhibitors. Figure 6.12 gives

FIGURE 6.11 Batimastat 12 and crystal structure of the MMP-3/12 complex. Carbon atoms of the protein and the inhibitor are colored cyan and green, respectively (1mmb, Grams et al., 1995). (See the color version of this figure in the Color Plates section.)

174

S

12

O

H N

N

15

N

O

O

N H

N H

MMP-1, -2, -3, -7, -8, -9, -10, -12, -14, -16: n.a. MMP-13: 8 nM

N H

O

MMP-1: 5 nM MMP-2: 4 nM MMP-3: 20 nM MMP-7: 6 nM

S

N H

O

F

HO

O H N

HO

O

S

O

H N O

N

O

16

O

O

O

TACE: 0.56 nM MMP-2: 2050 nM MMP-13: 1417 nM

MMP-2: 1700 nM MMP-13: 2.3 nM

13

O

H N

N

Br

N

14

N

O N H N

HO N H

O

S

N

N

O

O

17

O

S

O

TACE: 2.2 nM MMP-2: 664 nM MMP-13: 2277 nM

O

O

MMP-1, -2, -3, -7, -8, -9, -10, -12, -14, -16: >100 uM MMP-13: 6.6µM

N

N H

O

FIGURE 6.12 MMP-inhibitor selectivity and selective TACE inhibitors. The in vitro potency of selected MMP inhibitors is expressed as an IC50 value. The in vitro potency of selected TACE inhibitors is expressed as dissociation constant or IC50 value for 16 and 17, respectively.

F

HO

SERINE PROTEASES

175

the summary of the data. While earlier compounds such as batimastat 12 have a broad inhibition profile owing to their relatively small P1’substituent, compounds such as 13, which contain larger P1’ moieties, show higher selectivity for specific MMPs such as MMP-13 (Hu et al., 2005). A new development in the field of MMP inhibition is the discovery of compounds that do not bind to the catalytic zinc ion, yet are still potent MMP inhibitors, for example, compound 14 and 15 (Engel et al., 2005). These inhibitors bind to the S1’ pocket of MMP-13 from the ‘‘bottom,’’ making additional interactions outside the canonical substrate recognition region. Due to the greater structural diversity of MMPs in this region, compounds of this type can be highly selective. Apart from the desire to generate inhibitors with high selectivity within the MMP family, it should be noted that the close structural similarity with members of the M12 family poses an additional challenge in the design of selective compounds. Most of our understanding with regard to how to design MMP versus ADAM/ADAMTS selective compounds stems from studies on TNF-converting enzyme (TACE/ADAM-17), which has been the target of intense research on its own. Several crystal structures of TACE complexed with hydroxamic acid-based inhibitors have been solved, which indicate a large and L-shaped S1’ pocket. This property has been exploited in the design of the selective TACE inhibitors as exemplified by compound 16 (Duan et al., 2002). More recently, it was shown that selectivity for TACE over MMPs can also be obtained by compounds such as 17, which target the S1 site of TACE (Condon et al., 2007).

6.4

SERINE PROTEASES

Numerous serine proteases have been followed as drug targets for a range of indications. One prominent example is dipeptidyl peptidase 4 (DPP4) for the treatment of Type 2 diabetes. Also, Factor Xa and thrombin, which are S1 proteases of the blood coagulation cascade, have been pursued by many pharmaceutical companies for more than 20 years to treat thrombosis and related vascular diseases. The cleavage of peptide bonds by serine proteases involves nucleophilic attack of the deprotonated active site serine hydroxyl on the amide carbonyl carbon with the formation of a tetrahedral intermediate, as shown in Fig. 6.13. The negatively charged oxygen of this intermediate is stabilized by the formation of hydrogen bonds with protein backbone hydrogens, which form the so-called oxyanion hole. The release of the amine part of the amide bond liberates the C-terminal peptide moiety of the substrate protein, whereas the N-terminal part remains covalently bound to the protease in the form of an acyl–enzyme complex. The latter undergoes hydrolysis resulting ultimately in the release of the N-terminal peptide part of the substrate protein with regeneration of the active protease. An essential role for catalysis is played by the so-called catalytic triad, which, in addition to the nucleophilic serine, comprises an aspartic acid and a histidine residue. The latter two act as general acids/bases in the enzymatic reaction. For a serine protease to be active, the three active site residues Asp–His–Ser need to have an optimal spatial alignment, regardless of their position in the amino acid sequence.

176

PROTEASE-DIRECTED DRUG DISCOVERY

FIGURE 6.13 Catalytic mechanism of the cleavage of peptide bonds by serine proteases.

The classical approach for designing serine protease inhibitors makes use of the catalytic mechanism of these enzymes, more specifically of the nucleophilic nature of the catalytic serine. Many inhibitors are based on the so-called serine traps, electrophilic functions that make a covalent bond with the serine hydroxyl, preferably in a reversible manner. Commonly used serine traps are aldehydes, nitriles, boronates, trifluoro acetyl, alpha-keto acids, and alpha-keto amides. The challenge of designing a covalent serine protease inhibitor can thus be viewed as combining an electrophilic group with suitable moieties optimally fitting the specificity pockets of the target protease to obtain potency and selectivity against the related proteases. The electrophile, however, needs to be chosen with care. It should be sufficiently reactive to form a covalent bond with the serine hydroxyl when the inhibitor is bound to the target enzyme, but should not be overly reactive to avoid excessive instability in physiological media and unspecific reactions with other proteases and, in general, with other proteins. In the majority of cases, serine protease inhibitors are designed to occupy the specificity pockets in the nonprimed site of the active site. 6.4.1

Dipeptidyl Peptidase 4 (DPP4)

The concept of serine protease inhibitor design based on serine traps has been successfully applied to DPP4, a so-called exopeptidase that cleaves dipeptides from the N-terminus of a range of proteins. Two important physiological substrates are glucagon-like peptide (GLP-1) and glucose-dependent insulinotropic peptide

SERINE PROTEASES

177

(GIP). These two peptide hormones play an important role in glucose control after ingestion of nutrients. They have a very short half-life due to their rapid proteolytic inactivation by DPP4. The inactivation of DPP4 leads to the stabilization of GLP-1 and GIP and thus results in improved glycemic control. DPP4 exhibits a strong preference for proline in P1. Alanine is also accepted, and is actually present in that position in GLP-1. In contrast to the tight specificity for P1, the enzyme tolerates a wide range of substituents in P2. Moreover, as an N-terminal exopeptidase, DPP4 possesses two glutamic acid residues that make hydrogen bonds with the terminal ammonium ion. Based on the knowledge about its substrate preferences, the design of a covalent DPP4 inhibitor can be viewed as combining a proline analogue containing an electrophilic group, able to covalently bind serine 630, with a suitable P2 residue incorporating a basic amine to make hydrogen bonds with glutamic acids 205 and 206. Based on that principle, DPP4 inhibitors were designed and synthesized even in the absence of information from three-dimensional structures. The first highly potent inhibitors described were dipeptide derivatives using the boronic acid functionality as a serine trap (Flentke et al., 1991), as exemplified by compound 18 (KI value of 2 nM). Although potent, these compounds were not suitable for further development due to their unfavorable pharmacokinetic properties and their instability at neutral pH, resulting from the intramolecular complexation of the primary amine by the boronic acid. However, attempts were made to reduce the electrophilicity of the serine trap. To this end, the boronic acid functionality was replaced by a nitrile (Ashworth et al., 1996a). Despite the fact that this is a weaker electrophile, highly potent DPP4 inhibitors such as compound 19 (FE999011, KI value of 3.8 nM) were obtained. However, such compounds still carry liabilities. Depending on the nature of the P2 amino acid, they can exhibit instability due to the formation of 20 by the intramolecular nucleophilic attack of the primary amine on the nitrile. In an attempt to disfavor this cyclization, the cyanopyrrolidine motif was combined with an N-substituted glycine (Villhauer et al., 2003). These were the first DPP4 inhibitors with a secondary instead of a primary amine in P2. The design of these compounds was based on the observation that substrates with a N-methylglycine (sarcosine) in P2 were cleaved (Heins et al., 1988). These efforts culminated in the identification of vildagliptin 21. This compound exhibits high potency against DPP4 (IC50 value of 3.5 nM) and good selectivity over other serine proteases of the same family. Vildagliptin has good pharmacokinetic and glucoselowering properties in animal models, has successfully undergone clinical trials in diabetic patients and has been submitted for registration. The synthetic route for the preparation of vildagliptin is shown in Fig. 6.14. (Villhauer et al., 2003). Recently, a crystal structure of the related p-Iodo-Phe-Pro-CN (KI value of 25 nM) bound to porcine DPP4 was published (Engel et al., 2003). This structure clearly confirmed the validity of the design principle. As expected, the cyanopyrrolidine occupies the S1 pocket and there is a formation of an imidate by the attack of the catalytic serine hydroxyl on the nitrile function of the inhibitor. The amine of the iodophenylalanine makes hydrogen bonds to glutamic acid 205 and 206 and the iodophenyl itself lies in the S2 pocket, which is large and able to

178

(b)

(a)

Cl

N

N

O

O

Cl

+

B O

HN

O

NH2

+

NH2

H2N N

N

Cl

O

K2CO3, THF

KI = 3.8 nM

19 FE999011

O

N

N

Cl N

O

20

NH2

N

NH

IC50 nM

K2CO3, THF

O

O

R

H N

DPP4 3.5

HO

O N

Cl

DPP2 >500,000

21 Vildagliptin

H N

TFAA, THF

O N

N

N

PPCE 210,000

23 Ki = 0.4 nM

22 P32/98

N

Ki = 130 nM

S NH2

N

O

NH2

O

FIGURE 6.14 (a) Potent DPP4 inhibitors. (b) Synthesis of vildagliptin 21.

HO

KI = 2 nM

18

O

S

N

SERINE PROTEASES

179

accommodate a variety of residues. An important characteristic of the DPP4 inhibitors mentioned so far is the utilization of an electrophilic moiety reacting with the active site serine. This approach has led to active and selective inhibitors, but, as mentioned above, it was also noted that the reactivity of this moiety could be a liability due to instability, particularly in the presence of the critical amine required in P2. Therefore, attempts were made to identify DPP4 inhibitors lacking the electrophilic function. The isoleucine thiazolidide P32/98 22, having a KI value of 130 nM, was one of the first inhibitors of this type reported (Pederson et al., 1998). It is interesting to note that the corresponding nitrile 23 is approximately 300 times more potent (KI value of 0.4 nM) (Ashworth et al., 1996b). Further efforts to improve the activity of DPP4 inhibitors lacking an electrophile eventually led to the identification of potent inhibitors with good selectivity. Researchers at Merck identified the b-amino acid 24, a micromolar DPP4 inhibitor, by screening their sample collection (Xu et al., 2004). They replaced the substituted proline on the right-hand side by a thiazolidine to obtain compound 25. The choice of the thiazolidide was presumably driven by the fact that this is a well-known P1 motif in DPP4 inhibitors, occurring, for example, in P32/98 22. Despite the fact that derivative 25 was much smaller than compound 24, it essentially retained the activity compared to the original starting point. During optimization of the left-hand side of the molecule, as given in Fig. 6.15, the 2,4,5-trifluorophenyl motif was introduced, leading to a remarkable increase in activity with compound 26. In the next step, keeping the trifluorophenyl motif, optimization concentrated again on the amide portion, leading to the very potent derivative 27 (Brockunier et al., 2004). Further work led to sitagliptin 28 (Kim et al., 2005), exhibiting the same level of potency. Interestingly, the heterocyclic amide portion of 27 and 28 had now evolved into a motif that was very different from any other known P1 moiety. As it turned out, the crystal structure of sitagliptin bound to DPP4 revealed that it is the trifluorophenyl motif that tightly binds into S1, whereas the amide extends into S2 (Kim et al., 2005). Sitagliptin is very selective over other S9 family members such as DPP2 (IC50 >100 mM), DPP8 (IC50 >48 mM), and DPP9 (IC5 0 >100 mM). The compound received approval as antidiabetic agent in 2006 and is marketed under the trade name Januvia. The history of optimization of DPP4 inhibitors, as outlined here, illustrates some principles of the design of serine protease inhibitors. The classical approach using knowledge about the substrate specificity and exploiting the enzymatic mechanism by introducing electrophilic groups into inhibitors has led to potent and selective compounds such as vildagliptin, acting through a transient and reversible imidate intermediate. The use of electrophilic groups is however not a necessary requirement for the design of potent inhibitors. Indeed, approaches directed toward the development of active principles lacking a reactive functionality have also been followed, as demonstrated by the discovery of sitagliptin. 6.4.2

Trypsin-Like S1 Serine Proteases of the Coagulation Cascade

Proteases of the coagulation cascade, in particular the coagulation factors thrombin (Factor IIa) and Factor Xa, have been very actively pursued for many years as drug

180

N

IC50 = 1900 nM

24

NH2 O

O

H N

F

NH

F

F

NH2 O F

From a b-amino acid pyrrolidide hit to sitagliptin 28.

IC50 = 18 nM

NH

F

IC50 = 19 nM

N

F

N N

N CF3

N

IC50 = 119 nM

28 Sitagliptin

NH2 O

Ph

F

IC50 = 3000 nM

S

NH2 O

26

N

F

25

NH2 O

F

27

FIGURE 6.15

O

O

Cl

N

S

181

SERINE PROTEASES

P2 N H

H N

P1 ′

O N H

O

H N

O

P2 N H

P1 ′

O

H N

N H

O

H N

O

NH +

H2N

NH3

+

NH2

S1 Pocket

S1 Pocket

-

O

O Asp

O

O Asp

FIGURE 6.16 The S1 pockets of trypsin-like serine proteases exhibit specificity for the basic residues arginine and lysine.

targets for the treatment of diseases such as myocardial infarction, unstable angina, stroke, deep vein thrombosis, and pulmonary embolism. These diseases arise from abnormal coagulation and, therefore, interference with the coagulation cascade, particularly inhibition of the serine proteases of this cascade, is believed to result in therapeutic benefit. Thrombin and Factor Xa belong to the S1 family of serine proteases. More specifically, they belong to the so-called trypsin-like serine proteases, which exhibit specificity for substrates having amino acids with basic side chains, arginine or lysine, in the P1 position, as described in Fig. 6.16. These positively charged residues form a salt bridge with a negatively charged aspartic acid (Asp 189 in the case of thrombin and Factor Xa) located at the bottom of the deep S1 pocket of these proteases, making this particular site a focal point for inhibitor design. Electrophilic groups have been widely used as serine traps in the design of early thrombin inhibitors. These consisted of a combination of either an irreversible chloromethylketone (29, PPACK) or a reversible aldehyde (30, Efegatran) serine trap with a basic moiety fitting the S1 pocket. Heteroarylketones such as 31 or boronic acids such as 32 were also used, and the structures are shown in Fig. 6.17. However, all of these compounds are less than optimal as drug candidates. In particular, as mentioned earlier in this chapter, they suffer from poor pharmacokinetic properties due to their peptidic nature. Also, electrophilic groups can represent a significant liability. Based on the limited success with peptide-like compounds bearing serine traps, noncovalent inhibitors were sought. These efforts led to the identification of melagatran with a KI value for thrombin of 3 nM (Gustafsson et al., 2004). Melagatran, as shown in Fig. 6.18, bears a benzamidine as a basic group, mimicking the lysine/arginine side chains in P1 of the substrates. The compound exhibits high oral bioavailability in dogs. Unfortunately, although the pharmacokinetic properties in humans after i.v. injection were favorable, its oral bioavailability in humans was low. Synthesis of the double prodrug ximelagatran 33 solved this issue (Gustafsson et al., 2004). The prodrug itself is essentially inactive against thrombin, but oral administration in humans results in approximately 20% bioavailability of the active principal melagatran. Ximelagatran was the first oral direct thrombin inhibitor, which received regulatory

182

PROTEASE-DIRECTED DRUG DISCOVERY

N

H N

O

H N O

Cl

O

N

H N

O

NH

NH

29 PPACK

H N

H N

N O

H

O

NH H2N

O

H N

NH

H2N

30 Efegatran O N S

O

H Ac N

H N

N O

O

NH H2N

OH B OH

NH

NH

H2N

31

NH

32

K i Thrombin: 0.2 nM

K i Thrombin: 0.027 nM

FIGURE 6.17 Thrombin inhibitors.

approval in 2003 but was withdrawn from the market in 2006, in part due to unexpected liver side effects. Due to the preference of trypsin-like serine proteases for lysine and arginine binding into the S1 pocket, potent inhibitors of these enzymes require basic P1 groups. These functionalities in turn have been a major hurdle in terms of achieving favorable pharmacokinetic properties with inhibitors. Except for the use of prodrugs, this could seem like an irresolvable dilemma. Nevertheless, the recent discovery of the Factor Xa inhibitor rivaroxaban 36, currently in advanced clinical trials, impressively demonstrates that it is indeed possible to obtain potent inhibitors for trypsin-like serine proteases without relying on serine traps or on basic P1 moieties. Indeed, the high-throughput screening hit 35 (IC50 value for FXa of 120 nM) was optimized to 36 (IC50 value for FXa of 0.7 nM) (Roehrig et al., 2005). A crystallographic analysis revealed that the chlorothiophene moiety occupies the S1 pocket. In particular, the chlorine points toward the face of the aromatic ring of tyrosine 228, located at the bottom of S1, making an interaction with it. Similar binding modes have been reported for other chloroaryls not only with Factor Xa (Casimiro-Garcia et al., 2006), but also with thrombin (Schwienhorst, 2006). Thus, the use of chloroaryls as substitutes for the so far more common basic, charged P1 residues, may represent a new general principle for the design of trypsinlike serine protease inhibitors with more favorable pharmacokinetic properties.

183

CYSTEINE PROTEASES O

O HO

H N

EtO

H N

N

O

H2N

NH

33 Melagatran

H N

N

P1

O

O

H N

O

H2N

N

OH

34 Ximelagatran

Ki Thrombin: 4 nM

O

O O

N O O

O NH

H2N

HN

O

N

N

O

S

Cl 35 IC50 FXa: 120 nM

P1 HN

O

S 36 Rivaroxaban

Cl

IC50 FXa: 0.7 nM

FIGURE 6.18 Noncovalent thrombin inhibitors including the discovery of rivaroxaban 36 from a high-throughput screening hit.

6.5

CYSTEINE PROTEASES

The human genome encodes for 141 potentially active cysteine proteases that usually are intracellular enzymes. Almost 100 of these are deubiquitinating proteases, which cleave the isopeptide bond of the e-amino group of a protein’s lysine side chain linked to the C-terminus of the small polypeptide ubiquitin, a conjugation that targets proteins for destruction by the proteasome. However, in terms of drug discovery, mainly the lysosomal cysteine cathepsins and caspases have been followed. The latter are a protease family involved in apoptosis (programmed cell death) and inflammation. The mechanism by which cysteine proteases cleave peptide bonds closely resembles that of serine proteases, except that the nucleophile attacking the amide carbonyl carbon is the deprotonated thiol of a cysteine side chain. Also here, the reaction proceeds with the formation of a tetrahedral intermediate, the charge of which is stabilized by an oxyanion hole, followed by formation of an acyl–enzyme complex that is ultimately hydrolyzed to regenerate active enzyme, as shown in Fig. 6.19. As in the case of serine proteases, a catalytic triad is required for enzymatic activity. In the case of papain, an archetypal cysteine protease, this triad

184

PROTEASE-DIRECTED DRUG DISCOVERY

FIGURE 6.19

Catalytic mechanism of the cleavage of peptide bonds by cysteine proteases.

is constituted of Asn–His–Ser. As in the case of serine proteases, the design of cysteine protease inhibitors makes use of the nucleophilic thiolate by incorporating electrophiles in appropriate positions, which make a covalent bond with the sulfur atom. Among these are electrophilic moieties, which irreversibly trap the enzyme such as chloromethylketones, bromomethylketones, vinylsulfones, and vinyl esters. Aminomethylketones, cyanomethylamines, and 2-pyrimidylnitriles can serve as functional groups that act as reversible covalent inhibitory principles. 6.5.1

Cathepsin K

Cathepsin K is a lysosomal cysteine protease of the papain superfamily. It has been an important therapeutic target for the past 10 years. The discovery of cathepsin K inhibitors serves as an excellent example illustrating the design of cysteine proteases in general. Cathepsin K is primarily located in osteoclasts, cells which play a crucial role in bone resorption. This protease cleaves Type I collagen and other components of the bone matrix. A lack of balance between the bone resorption and bone formation in favor of the former leads to osteoporosis, and, as a consequence, inhibition of cathepsin K holds great promise as an approach to treat this disease. The main features of the active site of cathepsin K are drawn in Fig. 6.20. They were revealed by its crystal structure in complex with the irreversible inhibitor vinylsulfone 37 (McGrath et al., 1997): (1) the S1 pocket is largely solvent exposed and is more of a wall than a real pocket, (2) the S2 subsite is a deep, hydrophobic

185

CYSTEINE PROTEASES Asp61 O

S3

H N

HO

N O

H N

Gly66

O

O

H N

N O N H

S

H N

O

O

S1

N H

O

O O N H

S S

O

Tyr67

37

Cys25

S2

S1′

FIGURE 6.20 Schematic representation of the complex of cathepsin K and vinylsulfone 37.

pocket accepting in particular the isobutyl residue of leucine, and (3) the S3 subsite is predominantly formed by Asp 61 and Tyr 67, offering the possibility for hydrogen bond formation with basic moieties of the inhibitor as well as the possibility to target p–p interactions. In addition, it is important to note that the b-sheet-type hydrogen bonds between the P2 leucine and Gly 66 closely mimic the interaction between the enzyme and its substrate. The formation of at least one of these hydrogen bonds is an important aspect in inhibitor design. The first reported approach toward cathepsin K inhibitors made use of structural information on the prototypical cysteine protease family member papain using peptide aldehyde inhibitors (Yamashita et al., 1997) (Fig. 6.21). Crystal structures of complexes with papain showed leupeptin 38 binding to the nonprime site, whereas, surprisingly, the leupeptin analogue 39 binds to the prime site. Both of these inhibitors are covalent inhibitors, the aldehyde forming a hemithioketal with the active site cysteine. On the basis of this information an overlay of 38 and 39 was generated, resulting in the hypothetical 1,3-bis(acylamino)-2-propanone inhibitor 40. From this starting point, using molecular modeling, compound 41 was designed and found to exhibit high affinity for cathepsin K (KI,app value of 22 nM). Interestingly, it is essentially inactive against papain (KI;app >10 mM) and rather selective over other cysteine cathepsin proteases such as cathepsin L (KI;app value of 340 nM), cathepsin B (KI;app value of 1300 nM), and cathepsin S (KI,app value of 890 nM). However, despite the acyclic 1,3-bis(acylamino)-2-propanone derivatives being very potent and selective cathepsin K inhibitors, these compounds had limited utility due to poor pharmacokinetic properties and low potency in cellular systems. A strategy to overcome these issues is shown in Fig. 6.22. To obtain potent inhibitors with improved biopharmaceutical properties, compounds such as 41 were conformationally constrained by cyclization of the 1,3-bisacylaminoketone template, giving rise to inhibitors 42 and 43 (Marquis et al., 1998). This approach eventually led to the clinical candidate relacatib 45 (Yamashita et al., 2006) via 44 (Marquis et al., 2001). The relacatib 45 exhibits high potency against cathepsin K and high oral bioavailability in rats. The main interactions between the relacatib 45 and cathepsin K, as observed in the crystal structure of the protease/inhibitor

186

PROTEASE-DIRECTED DRUG DISCOVERY

Papain Prime side

Papain Nonprime side H2N

NH

HN O

O

H N

N H

O H H

N H

O

O

N H

O

O

H N

N H

O

38 Leupeptin

O

39 Cbz-Leu-Leu-Leu-H Overlay NH

H2N HN O

H N

N H

O

O N H

O

O

N H

O

H N

N H

O

O

40 Modeling on Cat K

H N

O O

O

O N H

O

N H

H N

O O

41 Cat K: Ki,app 22 nM

FIGURE 6.21

Design of 1,3-bis(acylamino)-2-propanone cathepsin K inhibitors.

complex, include (1) formation of a hemithioketal between the ketone and the catalytic cysteine, (2) positioning of the isobutyl group into the deep, hydrophobic S2 pocket, and (3) positioning of the benzofuran into the S3 pocket where it is involved in p–p stacking interactions with Tyr 67. A second approach for the design of cathepsin K inhibitors, which has been widely followed, makes use of a nitrile as electrophilic cysteine trap and resembles the strategy used for DPP4 inhibitors. The goal was to find an electrophile capable

187

CYSTEINE PROTEASES Cyclize

CbzNH

N H

NHCbz

N H

O

C4:Epimerization

O

O

O

CbzNH

O 41 Cat K: K i,app 22 nM

O

N ( )n

NHCbz

N H

42 n = 1 Cat K: K i,app 2.3 nM 43 n = 2 Cat K: K i,app 2.6 nM S2

O O

N H

H N O

S3

O O N S O

44

O O

N H

H N O

N

O O N S O Me

45 Relacatib

S1′

Ki, app (nM)

Ki, app (nM) Cat K 0.16 Cat B 500 2.2 Cat L Cat S 4.0 %F rat 42.1

S2′ N

Cat K Cat B Cat L Cat S %F rat

0.041 13 0.068 1.6 89.4

FIGURE 6.22 From acyclic 1,3-bis(acylamino)-2-propanone inhibitors to relacatib 45.

of forming a reversible covalent bond with the thiolate nucleophile of the catalytic cysteine, but being otherwise relatively unreactive, to avoid undesirable modification of other proteases and proteins. The nitrile function fits these criteria and it had been previously shown that peptide nitrile, N-acetylamino-N-(cyanomethyl)phenylpropanamide, is a relatively potent inhibitor of papain with a KI value of 0.73 mM (Lewis and Wolfenden, 1977). The NMR studies with the same compound labeled with 13C on the nitrile carbon allowed an unambiguous demonstration that there is indeed formation of the postulated thioimidate and that the formation of this covalent adduct is reversible (Moon et al., 1986). In contrast to the ketones discussed above, which are able to span the nonprimed and primed regions of the active site, the nitriles are largely limited to binding to the nonprimed site of cathepsin K only, where these inhibitors mainly utilize the S2 and S3 pockets. The path from a peptide nitrile starting point to the identification of the clinical candidate 49 is illustrated in Fig. 6.23 (Palmer et al., 2005). Starting from the relatively potent dipeptide nitrile 46, the P2 position was probed by introducing various replacements for the isobutyl group, which is a preferred substituent in that position. It was found that the cyclohexyl derivative 47 exhibits similar potency against cathepsin K as the dipeptide 46 but is more selective over other cathepsins, in particular cathepsins L and S. With this improved P2 motif in hand, efforts were concentrated on replacing the Cbz moiety. The substituted biphenyls were explored and it was found that a basic nitrogen borne by the second aryl ring was important for activity, as in 48, most likely because it makes an ionic interaction with Asp 61 in the S3 pocket. The compounds with

188

PROTEASE-DIRECTED DRUG DISCOVERY O

O CbzNH

CbzNH

N H

N H

N

47

46

K I, app (nM)

K I, app (nM) Cat K Cat B Cat L Cat S

N

Cat K Cat B Cat L Cat S

34.7 123,000 2790 143

H N

84 64,000 95,000 31,000 _ O

+

HN

Asp61

N

N

O

N S H N

O

O

N H

H N N

S3

O

O

N H

N

S2 48

49

<0.25 420 7900 4900

Cat K Cat B Cat L Cat S

L-006235 K I, app (nM)

K I, app (nM) Cat K Cat B Cat L Cat S

0.2 1000 6000 47,000 %F rat 68

FIGURE 6.23

Peptide nitriles as cathepsin K inhibitors.

the biphenyl motif were potent and selective, but the oral bioavailability had to be improved. This was achieved by replacing the second aryl ring by a thiazole, leading to the clinical candidate 49. 6.5.2

Caspases

Caspases have an absolute preference for an aspartic acid residue in the P1 position of their substrates. Eleven human caspase family members are known. They play an important role in inflammation and apoptosis and have been the target of intense efforts in pharmaceutical research. One of the most prominent members in this context is caspase 1, also called interleukin-1b converting enzyme (ICE), which is responsible for the correct processing of the precursors for the cytokines IL-1b and IL-18. Here, we will only briefly describe key elements in the development of some prototypic caspase inhibitors. As for other cysteine proteases, inhibitors with peptidic sequences and an electrophilic group binding to the catalytic cysteine were used early on to develop structure–activity relationships and as tools to gain insight into the architecture of

CYSTEINE PROTEASES

189

the active site. These studies identified key interactions between the inhibitor and the protease, which were subsequently used for drug design. For instance, an N-methylation scan of the amide groups in the tripeptide sequence Z-Val-Ala-Asp suggested that the amide groups of the P1 and P3 residues are involved in hydrogen bonding of the inhibitor to Ser 339 and Trp 341 of the protease (Dolle et al., 1994). Figure 6.24 shows the three-dimensional structure of Tyr-Val-Ala-Asp-H 50 complexed with caspase 1, which corroborates these findings. In all caspase/peptide complexes known today, the inhibitor binds in an extended conformation and makes hydrogen bonds to the protein typical for an antiparallel b-sheet. The design of inhibitors, therefore, focused on the identification of P2–P3-constrained dipeptide mimetics, which allow for the preservation of the P1 and P3 amide groups. A variety of scaffolds has been tried, with the most prominent ones being the pyridinone 51 and the pyridazinodiazepine 52. A similar approach has been previously reported for the successful design of orally active peptidomimetic elastase inhibitors (Brown et al., 1994). When linked to the essential aspartic acid residues in P1, the pyridazinodiazepine scaffold affords an inhibitor with potency similar to that obtained with the tripeptide Z-Val-Ala-Asp-H, however, having some oral bioavailability in dogs (17%). Again, a prodrug approach has been used to further increase bioavailability. In this case, the P1 aspartic acid was masked by the formation of an internal hemiacetal, resulting in pralnacasan 53. This inhibitor has been tested in clinical trials for the treatment of rheumatoid arthritis and osteoarthritis. Unfortunately, these trials had to be stopped in 2003 due to liver toxicity seen in animal studies. These findings and the general concern that inhibitors containing electrophilic groups might have a higher risk of generating adverse side effects have prompted other groups to look into alternative ways to inhibit caspases. Caspases require homodimerization for autocatalytic processing and thus activation (Boatright et al., 2003). Moreover, the correct arrangement of the active site loops appears to be dependent on the proper alignment of the individual subunits of the homodimer. Using a mass spectrometry-based fragment screening technology called Tethering (Erlanson et al., 2000), compounds have recently been identified binding to a conserved cysteine at a site of the dimer interface, which is coupled to the active site of the protease (Hardy et al., 2004). It was shown that binding of these compounds locks the active site into a zymogen-like conformation and thereby inhibits the enzyme in an allosteric way. The approaches toward inhibitors of cathepsin K and caspase-1 outlined herein exemplify the successful design of covalent, reversible cysteine protease inhibitors, which are potent and selective, and which exhibit favorable pharmacokinetic properties. They also illustrate the use of different motifs to obtain optimal interactions between the inhibitor and the enzyme in the specificity pockets. So far, there are only very few reports describing noncovalent cysteine protease inhibitors. This state-of-the-art approach contrasts with the situation described above for serine proteases, for which many noncovalent inhibitors have been identified. In the cysteine proteases research field, the identification of inhibitors lacking reactive functional groups continues to be a major challenge.

190

PROTEASE-DIRECTED DRUG DISCOVERY

FIGURE 6.24 Caspase-1 inhibitors. The substrate-based inhibitor Ac-Tyr-Val-Ala-Asp-H is shown in (a) indicating atoms involved in hydrogen bonding with the protein. The crystal structure of the caspase 1/inhibitor complex (1ice, Wilson et al., 1994). Carbon atoms of the protein and the inhibitor are colored cyan and green, respectively. Hydrogen bonds between the inhibitor and the protein are indicated by dashed lines in magenta. Inhibitors based on the pyridone (b) and pyridazinodiazepine (c) P2–P3 scaffold are shown. (See the color version of this figure in the Color Plates section.)

6.6

PERSPECTIVE ON PROTEASES AS DRUG TARGETS

After more than 30 years of drug discovery in the field of proteases, only few drugs acting on a small number of targets have reached the market so far. The main hurdles in the past have been the difficulty to obtain orally active inhibitors with good pharmacokinetic profiles and to appropriately test them for selectivity against all other related family members. Those protease inhibitors, which are however now successful drugs for the treatment of a variety of diseases, serve as excellent examples for future protease-directed drug discovery. Moreover, with

REFERENCES

191

the knowledge about all proteases encoded by the human genome, it is now possible to establish suitable assay systems to exhaustively test potential drug candidates for selectivity against other proteases within the same and across different families. Depending on their catalytic mechanism, proteases are grouped into four different classes. Although each of the different protease classes requires different types of inhibitors, there are important commonalities. In the majority of cases, protease inhibitors exploit the respective catalytic mechanism and are, in addition, based on the detailed understanding of the protease’s substrate specificity and the architecture of its active site. For the latter, it is crucial to have the three-dimensional structure early on in the drug discovery process. We have highlighted for each protease class the major principles used to target the respective catalytic machinery. In the past, these have been combined with the substrate-derived peptide sequences or peptide analogues that, however, pose considerable problems in achieving appropriate pharmacokinetic properties, and turning a peptide-like structure into a drug-like molecule has often been a formidable task. Today, the substrate-derived structures are usually replaced by or even combined with building blocks identified by screening of large compound collections or more recently by fragment-based screening. Protease inhibitors that exploit the catalytic mechanism are, in general, specific for members of the corresponding class of proteases. However, since a significant portion of their binding energy is derived from a common active principle, they are often less specific toward a number of other proteases of the same family or class. A new development in the field is now the design of inhibitors that are not based on electrophilic or metal binding groups as exemplified by the Factor Xa inhibitor rivaroxaban 36 or the MMP-13 inhibitor 15. In these cases, sufficient binding energy can be generated by appropriately filling the key active site pockets. An interesting aspect of this work is the finding that the active site of several proteases (e.g., MMP-13) can be adaptable to changes in the size of their inhibitors, which offers further opportunities in the design of target specificity. Despite the limited number of protease drugs on the market today, there are an increasing number of protease inhibitors currently undergoing clinical trials for important indications such as thrombosis, hepatitis C, cancer, osteoporosis, and Alzheimer’s disease. Several of these will ultimately reach the market and there are even more potential protease targets being followed at earlier stages of the drug discovery process. Therefore, based on the extensive experience made over the past years, the future appears to look bright for protease-directed drug discovery.

REFERENCES Abbenante G, Fairlie DP, 2005. Protease inhibitors in the clinic. Med. Chem. 1(1):71–104. Andreeva NS, Rumsh LD, 2001. Analysis of crystal structures of aspartic proteinases: on the role of amino acid residues adjacent to the catalytic site of pepsin-like enzymes. Protein Sci. 10:2439–2450.

192

PROTEASE-DIRECTED DRUG DISCOVERY

Ashworth DM, Atrash B, Baker GR, Baxter AJ, Jenkins PD, Jones DM, Szelke M, 1996a. 2-Cyanopyrrolidides as potent, stable inhibitors of dipeptidyl peptidase IV. Biorg. Med. Chem. Lett. 6:1163–1166. Ashworth DM, Atrash B, Baker GR, Baxter AJ, Jenkins PD, Jones DM, Szelke M, 1996b. 2-Cyanothiazolidides as very potent, stable inhibitors of dipeptidyl peptidase IV. Biorg. Med. Chem. Lett. 6:2475–2748. Barbaro G, Scozzafava A, Mastrolorenzo A, Supuran CT, 2005. Highly active antiretroviral therapy: current state of the art, new agents and their pharmacological interactions useful for improving therapeutic outcome. Curr. Pharm. Des. 11:1805–1843. Barrett AJ, Rawlings ND, Woessner JF, 1998. Handbook of Proteolytic Enzymes. Academic Press, San Diego. Boatright KM, Renatus M, Scott FL, Sperandio S, Shin H, Pedersen IM, Ricci J-E, Edris WA, Sutherlin DP, Green DR, Salvesen GS, 2003. A unified model for apical caspase activation. Mol. Cell 11(2):529–541. Brandstetter H, Grams F, Glitz D, Lang A, Huber R, Bode W, Krell H-W, Engh RA, 2001. The ˚ crystal structure of a matrix metalloproteinase 8-barbiturate inhibitor complex 1.8-A reveals a previously unobserved mechanism for collagenase substrate recognition. J. Biol. Chem. 276(20):17405–17412. Brockunier L H J, Colwell Jr LF, Habulihaz B, He H, Leiting B, Lyons KA, Marsilio F, Patel R, Teffera Y, Wu JK, Thornberry NA, Weber AE, Parmee ER, 2004. Substituted piperazines as novel dipeptidyl peptidase IV inhibitors. Bioorg. Med. Chem. Lett. 14:4763–4766. Brown FJ, Andisik DW, Bernstein PR, Bryant CB, Ceccarelli C, Damewood Jr JR, Edwards PD, Earley RA, Feeney S, Green RC, Gomes EP, Kosmider BJ, Krell RD, Shaw A, Steelman GB, Thomas RM, Vacek EP, Veale CA, Tithill PA, Warner P, Williams JC, Wolanin DJ, Woolson SA, 1994. Design of orally active, non-peptidic inhibitors of human leukocyte elastase. J. Med. Chem. 37(9):1259–1261. Byers LD, Wolfenden R, 1972. A potent reversible inhibitor of carboxypeptidase A. J. Biol. Chem. 247(2):606–608. Casimiro-Garcia A, Dudley DA, Heemstra RJ, Filipski KJ, Bigge CF, Edmunds JJ, 2006. Progress in the discovery of factor Xa inhibitors. Expert Opin. Ther. Patents 16:119– 145. Cody WL, Holsworth DD, Powell NA, Jalaie M, Zhang E, Wang W, Samas B, Bryant J, Ostroski R, Ryan MJ, Edmunds JJ, 2005. The discovery and preparation of disubstituted novel amino-aryl-piperidine-based renin inhibitors. Bioorg. Med. Chem. 13:59–68. Condon JS, Joseph-McCarthy D, Levin JI, Lombart H-G, Lovering FE, Sun L, Wang W, Xu W, Zhang Y, 2007. Identification of potent and selective TACE inhibitors via the S1 pocket. Bioorg. Med. Chem. Lett. 17(1):34–39. Cooper JB, 2002. Aspartic proteinases in disease: a structural perspective. Curr. Drug. Targets 3:155–173. Dash C, Kulkarni A, Dunn B, Rao M, 2003. Aspartic peptidase inhibitors: implications in drug development. Crit. Rev. Biochem. Mol. Biol. 38:89–119. DeLano WL, 2002. The PyMOL Molecular Graphics System. http://www.pymol.org. Dolle RE, Singh J, Rinker J, Hoyer D, Prasad CVC, Graybill CL, Salvino JM, Helaszek CT, Miller RE, Ator MA, 1994. Aspartyl a-(1-phenyl-3-(trifluoromethyl)-pyrazol-5-yl)-oxy) methyl ketones as interleukin-1b converting enzyme inhibitors. Significance of the P1 and

REFERENCES

193

P3 amido nitrogens for enzyme-peptide inhibitor binding. J. Med. Chem. 37 (23):3863–3866. Duan JJW, Chen L, Wasserman ZR, Lu Z, Liu R-Q, Covington MB, Qian M, Hardman KD, Magolda RL, Newton RC, Christ DD, Wexler RR, Decicco CP, 2002. Discovery of b-lactam hydroxamic acids as selective inhibitors of tumor necrosis factor a converting enzyme: design, synthesis, and structure–activity relationships. J. Med. Chem. 45 (23):4954–4957. Dunn BM, 2002. Structure and mechanism of the pepsin-like family of aspartic peptidases. Chem. Rev. 102:4431–4458. Eder J, Hommel U, Cumin F, Martoglio B, Gerhartz B, 2007. Aspartic proteases in drug discovery. Curr. Pharm. Des. 13:271–285. Engel M, Hoffmann T, Wagner L, Wermann M, Heiser U, Kiefersauer R, Huber R, Bode W, Demuth HU, Brandstetter H, 2003. The crystal structure of dipeptidyl peptidase IV (CD26) reveals its functional regulation and enzymatic mechanism. Proc. Natl. Acad. Sci. USA 100:5063–5068. Engel CK, Pirard B, Schimanski S, Kirsch R, Habermann J, Klingler O, Schlotte V, Weithmann KU, Wendt KU, 2005. Structural basis for the highly selective inhibition of MMP-13. Chem. Biol. 12(2):181–189. Erlanson DA, Braisted AC, Raphael D, Randal M, Stroud RM, Gordon EM, Wells JA, 2000. Site-directed ligand discovery. Proc. Natl. Acad. Sci. USA 97(17):9367–9372. Finzel BC, Baldwin ET, Bryant Jr GL, Hess GF, Wilks JW, Trepod CM, Mott JE, Marshall VP, Petzold GL, Poorman RA, O’Sullivan TJ, Schostarez HJ, Mitchell MA, 1998. Structural characterizations of nonpeptidic thiadiazole inhibitors of matrix metalloproteinases reveal the basis for stromelysin selectivity. Protein Sci. 7(10):2118–2126. Fischli W, Clozel JP, el Amrani K, Wostl W, Neidhart W, Stadler H, Branca Q, 1991. Ro 42-5892 is a potent orally active renin inhibitor in primates. Hypertension 18: 22–31. Flentke GR, Munoz E, Huber BT, Plaut AG, Kettner CA, Bachovchin WW, 1991. Inhibition of dipeptidyl aminopeptidase IV (DP-IV) by Xaa-boroPro dipeptides and use of these inhibitors to examine the role of DP-IV in T-cell function. Proc. Natl. Acad. Sci. USA 88:1556–1559. Go¨schke R, Stutz S, Heinzelmann W, Maibaum J, 2003. The nonchiral bislactim diethoxy ether as highly stereo-inducing synthon for sterically hindered g-branched a-amino acids: a practical, large-scale route to an intermediate of the novel renin inhibitor aliskiren. Helv. Chim. Acta. 86:2848–2870. Grams F, Crimmin M, Hinnes L, Huxley P, Pieper M, Tschesche H, Bode W, 1995. Structure determination and analysis of human neutrophil collagenase complexed with a hydroxamate inhibitor. Biochemistry 34:14012–14020. Gustafsson D, Bylund R, Antonsson T, Nilsson I, Nystro¨m J-E, Eriksson U, Bredberg U, Teger-Nilsson A-C, 2004. A new oral anticoagulant: the 50 year challenge. Nat. Rev. Drug Discov. 3:649–659. Hajduk PJ, Sheppard G, Nettesheim DG, Olejniczak ET, Shuker SB, Meadows RP, Steinman DH, Carrera Jr GM, Marcotte PA, Severin J, Walter K, Smith H, Gubbins E, Simmer R, Holzman TF, Morgan DW, Davidsen SK, Summers JB, Fesik SW, 1997. Discovery of potent nonpeptidic inhibitors of stromelysin using SAR by NMR. J. Am. Chem. Soc. 119(25):5818–5827.

194

PROTEASE-DIRECTED DRUG DISCOVERY

Hajduk PJ, Shuker SB, Nettesheim DG, Craig R, Augeri DJ, Betebenner D, Albert DH, Guo Y, Meadows RP, Xu L, Michaelides M, Davidsen SK, Fesik SW, 2002. NMR-based modification of matrix metalloproteinase inhibitors with improved bioavailability. J. Med. Chem. 45(26):5628–5639. Hardy J, Selkoe DJ, 2002. The amyloid hypothesis of Alzheimer’s disease: progress and problems on the road to therapeutics. Science 297:353–356. Hardy JA, Lam J, Nguyen JT, O’Brien T, Wells JA, 2004. Discovery of an allosteric site in the caspases. Proc. Natl. Acad. Sci. USA 101(34):12461–12466. Heins J, Welker P, Scho¨nlein C, Born I, Hartrodt B, Neubert K, Tsuru D, Barth A, 1988. Mechanism of proline-specific proteinases: (I) substrate specificity of dipeptidyl peptidase IV from pig kidney and protein-specific endopeptidase from flavobacterium meningosepticum. Biochem. Biophys. Acta 954:161–169. Holsworth DD, Powell NA, Downing DM, Cai C, Cody WL, Ryan JM, Ostroski R, Jalaie M, Bryant JW, Edmunds JJ, 2005. Discovery of novel non-peptidic ketopiperazine-based renin inhibitors. Bioorg. Med. Chem. 13:2657–2664. Hu Y, Xiang JS, DiGrandi MJ, Du X, Ipek M, Laakso LM, Li J, Li W, Rush TS, Schmid J, Skotnicki JS, Tam S, Thomason JR, Wang Q, Levin JI, 2005. Potent, selective, and orally bioavailable matrix metalloproteinase-13 inhibitors for the treatment of osteoarthritis. Bioorg. Med. Chem. 13(24):6629–6644. Kaldor SW, Kalish VJ, Davies JF, Shetty BV, Fritz JE, Appelt K, Burgess JA, Campanale KM, Chirgadze NY, Clawson DK, Dressman BA, Hatch SD, Khalil DA, Kosa MB, Lubbehusen PP, Muesing MA, Patick AK, Reich SH, Su KS, Tatlock JH, 1997. Viracept (nelfinavir mesylate, AG1343): a potent, orally bioavailable inhibitor of HIV-1 protease. J. Med. Chem. 40:3979–3985. Kandula VR, Khanlou H, Farthing C, 2005. Tipranavir: a novel second-generation nonpeptidic protease inhibitor. Expert Rev. Anti Infect. Ther. 3:9–21. Kim D, Wang L, Beconi M, Eiermann GJ, Fisher MH, He H, Hickey GJ, Kowalchick JE, Leiting B, Lyons K, Marsilio F, McCann ME, Patel RA, Petrov A, Scapin G, Patel SB, Sinha Roy R, Wu JK, Wyvratt MJ, Zhang BB, Zhu L, Thornberry NA, Weber AE, 2005. (2R)-4-oxo-4-[3-(trifluoromethyl)-5,6-dihydro[1,2,4]triazolo[4,3-a]pyrazin-7(8H)-yl]-1(2,4,5-trifluorophenyl)butan-2-amine: a potent, orally active dipeptidyl peptidase IV inhibitor for the treatment of type 2 diabetes. J. Med. Chem. 48:141–151. Kleinert HD, Rosenberg SH, Baker WR, Stein HH, Klinghofer V, Barlow J, Spina K, Polakowski J, Kovar P, Cohen J, 1992. Discovery of a peptide-based renin inhibitor with oral bioavailability and efficacy. Science 257:1940–1943. Lewis CA, Wolfenden R, 1977. Thiohemiacetal formation by inhibitory aldehydes at the active site of papain. Biochemistry 16:4890–4895. Ma¨rki HP, Binggeli A, Bittner B, Bohner-Lang V, Breu V, Bur D, Coassolo PH, Clozel JP, D’Arcy A, Doebeli H, Fischli W, Funk CH, Foricher J, Giller T, Gruninger F, Guenzi A, Guller R, Hartung T, Hirth G, Jenny CH, Kansy M, Klinkhammer U, Lave T, Lohri B, Luft FC, Mervaala EM, Muller DN, Muller M, Montavon F, Oefner CH, Qiu C, Reichel A, SanwaldDucray P, Scalone M, Schleimer M, Schmid R, Stadler H, Treiber A, Valdenaire O, Vieira E, Waldmeier P, Wiegand-Chou R, Wilhelm M, Wostl W, Zell M, Zell R, 2001. Piperidine renin inhibitors: from leads to drug candidates. Il Farmaco 56: 21–27. Marquis RW, Yamashita DS, Ru Y, LoCastro SM, Oh H-J, Erhard KF, DesJarlais RL, Head MS, Smith WW, Zhao B, Janson CA, Abdel-Meguid SS, Tomaszek TA, Levy

REFERENCES

195

MA, Veber DF, 1998. Conformationally constrained 1,3-diamino ketones: a series of potent inhibitors of the cysteine protease cathepsin K. J. Med. Chem. 41:3563–3567. Marquis RW, Ru Y, LoCastro SM, Zeng J, Yamashita DS, Oh H-J, Erhard KF, Davis LD, Tomaszek TA, Tew D, Salyers K, Proksch J, Ward K, Smith B, Levy M, Cummings MD, Haltiwanger RC, Trescher G, Wang B, Hemling ME, Quinn CJ, Cheng H-Y, Lin F, Smith WW, Janson CA, Zhao B, McQueney MS, D’Alessio K, Lee C-P, Marzulli A, Dodds RA, Blake S, Hwang S-M, James IE, Gress CJ, Bradley BR, Lark MW, Gowen M, Veber DF, 2001. Azepanone-based inhibitors of human and rat cathepsin K. J. Med. Chem. 44:1380–1395. Maskos K, 2005. Crystal structures of MMPs in complex with physiological and pharmacological inhibitors. Biochimie 87(3–4):249–263. Matter H, Schwab W, Barbier D, Billen G, Haase B, Neises B, Schudok M, Thorwart W, Schreuder H, Brachvogel V, Loenze P, Weithmann KU, 1999. Quantitative structure– activity relationship of human neutrophil collagenase (MMP-8) inhibitors using comparative molecular field analysis and x-ray structure analysis. J. Med. Chem. 42(11): 1908–1920. Matthews BW, 1988. Structural basis of the action of thermolysin and related zinc peptidases. Acc. Chem. Res. 21:333–340. McGrath ME, Klaus JL, Barnes MG, Bromme D, 1997. Crystal structure of human cathepsin K complexed with a potent inhibitor. Nat. Struct. Biol. 4:105–109. Miller A, Askew M, Beckett RP, Bellamy CL, Bone EA, Coates RE, Davidson AH, Drummond AH, Huxley P, Martin FM, Saroglou L, Thompson AJ, van Dijk SE, Whittaker M, 1997. Inhibition of matrix metalloproteinases: an examination of the S1’ pocket. Bioorg. Med. Chem. Lett. 7(2):193–198. Moon JB, Coleman RS, Hanzlik RP, 1986. Reversible covalent inhibition of papain by a peptide nitrile. 13C NMR evidence for a thioimidate ester adduct. J. Am. Chem. Soc. 108:1350–1351. Natesh R, Schwager SLU, Evans HR, Sturrock ED, Acharya KR, 2004. Structural details on the binding of antihypertensive drugs captopril and enalaprilat to human testicular angiotensin I-converting enzyme. Biochemistry 43(27):8718–8724. Oefner C, Binggeli A, Breu V, Bur D, Clozel JP, D’Arcy A, Dorn A, Fischli W, Gru¨ninger F, Gu¨ller R, Hirth G, Ma¨rki H, Mathews S, Mu¨ller M, Ridley RG, Stadler H, Vieira E, Wilhelm M, Winkler F, Wostl W, 1999. Renin inhibition by substituted piperidines: a novel paradigm for the inhibition of monomeric aspartic proteinases? Chem. Biol. 6:127–131. Ondetti MA, Rubin B, Cushman DW, 1977. Design of specific inhibitors of angiotensinconverting enzyme: new class of orally active antihypertensive agents. Science 196 (4288):441–444. Overall CM, Kleifeld O, 2006. Towards third generation matrix metalloproteinase inhibitors for cancer therapy. Br. J. Cancer 94(7):941–946. Page JM, Macgillivray RT, Di Cera E, 2005. Determinants of specificity in coagulation proteases. J. Thromb. Haemost. 3(11):2401–2408. Palmer JT, Bryant C, Wang D-X, Davis DE, Setti EL, Rydzewski RM, Venkatraman S, Tian Z-Q, Burrill LC, Mendonca RV, Springman E, McCarter J, Chung T, Cheung H, Janc JW, McGrath M, Somoza JR, Enriquez P, Yu ZW, Strickley RM, Liu L, Venuti MC, Percival MD, Falgueyret JP, Prasit P, Oballa R, Riendeau D, Young RN, Wesolowski G,

196

PROTEASE-DIRECTED DRUG DISCOVERY

Rodan SB, Johnson C, Kimmel DB, Rodan G, 2005. Design and synthesis of tri-ring P3 benzamide-containing aminonitriles as potent, selective, orally effective inhibitors of cathepsin K. J. Med. Chem. 48:7520–7534. Patchett AA, Cordes EH, 1985. The design and properties of N-carboxyalkyldipeptide inhibitors of angiotensin-converting enzyme. Adv. Enzymol. Relat. Areas Mol. Biol. 57:1–84. Pederson RA, White HA, Schlenzig D, Pauly RP, McIntosh CHS, Demuth HU, 1998. Improved glucose tolerance in Zucker fatty rats by oral administration of the dipeptidyl peptidase inhibitor isoleucine thiazolidide. Diabetes 47:1253–1258. Puente XS, Sanchez LM, Overall CM, Lopez-Otin C, 2003. Human and mouse proteases: a comparative genomic approach. Nat Rev Genet 4(7):544–558. Puerta DT, Cohen SM, 2003. Examination of novel zinc-binding groups for use in matrix metalloproteinase inhibitors. Inorg. Chem. 42(11):3423–3430. Rahuel J, Rasetti V, Maibaum J, Rueger H, Goschke R, Cohen NC, Stutz S, Cumin F, Fuhrer W, Wood JM, Grutter MG, 2000. Structure-based drug design: the discovery of novel nonpeptide orally active inhibitors of human renin. Chem. Biol. 7:493–504. Reeke GN, Hartsuck JA, Ludwig L, Steitz TA, Lipscomb WN, 1967. The structure of ˚ resolution and the complex with Glycyl– carboxypeptidase A VI. Some results at 2.0-A ˚ resolution. Proc. Natl. Acad. Sci. 58(6):2220–2226. Tyrosine at 2.8-A Roberts NA, Martin JA, Kinchington D, Broadhurst AV, Craig JC, Duncan IB, Galpin SA, Handa BK, Kay J, Krohn A, 1990. Rational design of peptide-based HIV proteinase inhibitors. Science 248:358–361. Roehrig S, Straub A, Pohlmann J, Lampe T, Pernerstorfer J, Schlemmer K-H, Reinemer P, Perzborn E, 2005. Discovery of the novel antithrombotic agent 5-chloro-N-({(5S)-2-oxo-3[4-(3-oxomorpholin-4-yl)phenyl]-1,3-oxazolidin-5yl}methyl)thiophene-2-carboxamide (BAY 59-7939): an oral, direct factor Xa inhibitor. J. Med. Chem. 48:5900–5908. Rowsell S, Hawtin P, Minshull CA, Jepson H, Brockbank SM, Barratt DG, Slater AM, McPheat WL, Waterson D, Henney AM, Pauptit RA, 2002. Crystal structure of human MMP9 in complex with a reverse hydroxamate inhibitor. J.Mol.Biol. 319:173–181. Ru¨eger H, Stutz S, Go¨schke R, Spindler F, Maibaum J, 2000. A convergent synthesis approach towards CGP60536B, a non-peptide orally potent renin inhibitor, via an enantiomerically pure ketolactone intermediate. Tetrahedron Lett. 41:10085–10089. Schechter I, Berger A, 1967. On the size of the active site in proteases. I. Papain. Biochem. Biophys. Res. Commun. 27(2):157–162. Schro¨der J, Henke A, Wenzel H, Brandstetter H, Stammler HG, Stammler A, Pfeiffer WD, Tschesche H, 2001. Structure-based design and synthesis of potent matrix metalloproteinase inhibitors derived from a 6H-1,3,4-thiadiazine scaffold. J. Med. Chem. 44 (20):3231–3243. Schwienhorst A, 2006. Direct thrombin inhibitors – a survey of recent developments. Cell. Mol. Life Sci. 63:2773–2791. Sto¨cker W, Grams F, Baumann U, Reinemer P, Gomis-Ru¨th FX, McKay DB, Bode W, 1995. The metzincins—topological and sequential relations between the astacins, adamalysins, serralysins, and matrixins (collagenases) define a superfamily of zinc-peptidases. Protein Sci. 4(5):823–840. Turk B, 2006. Targeting proteases: successes, failures and future prospects. Nat. Rev. Drug Discov. 5(9):785–799.

REFERENCES

197

Villhauer EB, Brinkman JA, Naderi GB, Burkey BF, Dunning BE, Prasad K, Mangold BL, Russell ME, Hughes TE, 2003. 1-[[(3-Hydroxy-1-adamantyl)amino]acetyl]-2-cyano-(S)pyrrolidine: a potent, selective, and orally bioavailable dipeptidyl peptidase IV inhibitor with antihyperglycemic properties. J. Med. Chem. 46:2774–2789. Wilson KP, Black JA, Thomson JA, Kim EE, Griffith JP, Navia MA, Murcko MA, Chambers SP, Aldape RA, Raybuck SA, Livingston DJ, 1994. Structure and mechanism of interleukin-1 beta converting enzyme. Nature 370(6487):270–275. Wood JM, Criscione L, de Gasparo M, Buhlmayer P, Rueger H, Stanton JL, Jupp RA, Kay J, 1989. CGP 38 560: orally active, low-molecular-weight renin inhibitor with high potency and specificity. J. Cardiovasc. Pharmacol. 14:221–226. Wood JM, Maibaum J, Rahuel J, Grutter MG, Cohen NC, Rasett V, Ruger H, Goschke R, Stutz S, Fuhrer W, Schilling W, Rigollier P, Yamaguchi Y, Cumin F, Baum HP, Schnell CR, Herold P, Mah R, Jensen C, O’Brien E, Stanton A, Bedigian MP, 2003. Structure-based design of aliskiren, a novel orally effective renin inhibitor. Biochem. Biophys. Res. Commun. 308:698–705. Xu J, Ok HO, Gonzalez EJ, Colwell Jr LF, Habulihaz B, He H, Leiting B, Lyona KA, Marsilio F, Patel RA, Wu JK, Thornberry NA, Weber AE, Parmee ER, 2004. Discovery of potent and selective b-homophenylalanine-based dipeptidyl peptidase IV inhibitors. Bioorg. Med. Chem. Lett. 14:4759–4762. Yamashita DS, Smith WW, Zhao B, Janson CA, Tomaszek TA, Bossard MJ, Levy MA, Oh HJ, Carr TJ, Thompson SC, Ijames CF, Carr SA, McQueney M, D’Alessio KJ, Amegadzie BY, Hanning CR, Abdel-Meguid S, DesJarlais RL, Gleason JG, Veber DF, 1997. Structure and design of potent and selective cathepsin K inhibitors. J. Am. Chem. Soc. 119:11351– 11352. Yamashita DS, Marquis RW, Xie R, Nidamarthy SD, Oh H-J, Jeong JU, Erhard KF, Ward KW, Roethke TJ, Smith BR, Cheng H-Y, Geng X, Lin F, Offen PH, Wang B, Nevins N, Head MS, Janson CA, Gao E, Tomaszek T, McQueney M, James IE, Gress CJ, Zembryki DL, Lark MW, Veber DF, 2006. Structure–activity relationships of 5-, 6-, and 7-methylsubstituted azepan-3-one cathepsin K inhibitors. J. Med. Chem. 49:1597–1612. Zhang X, Gonnella NC, Koehn J, Pathak N, Ganu V, Melton R, Parker D, Hu S-I, Nam K-Y, 2000. Solution structure of the catalytic domain of human collagenase-3 (MMP-13) complexed to a potent non-peptidic sulfonamide inhibitor: binding comparison with stromelysin-1 and collagenase-1. J. Mol. Biol. 301(2):513–524.

7 SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS: CHALLENGES AND PROSPECTS ADRIAN WHITTY

7.1

INTRODUCTION

The large majority of existing drugs—and essentially all those that can be taken by mouth as a pill or capsule—are so-called small-molecule drugs; that is, they are synthetic organic compounds of molecular weight less than
1000 Da. Most of the protein targets that have been successfully addressed by oral drugs share the property that, as part of their natural physiological function, they interact with a small organic molecule such as a metabolite or a neurotransmitter. In most of these cases, the drug works by binding to its protein target in the same site that has evolved to bind the natural small ligand, thus blocking the protein’s natural function (or, in the case of drugs that are agonists rather than antagonists, by mimicking the function of the natural ligand to artificially stimulate the protein’s function). A large fraction of the potential drug targets that have not yet been exploited belong to other gene product families. These are proteins that have evolved to bind not to small organic ligands but instead to other proteins. It has been estimated that there are between 40,000 and 200,000 such protein–protein interactions in the human ‘‘interactome’’ (Bork et al., 2004). For reasons that will be discussed in detail in later sections, despite many attempts the pharmaceutical industry has been remarkably unsuccessful in identifying oral smallmolecule drugs that address such protein–protein interaction (PPI) targets. In fact, excluding a very small number of structurally complex natural products

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

199

200

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

(Schreiber and Crabtree, 1992; Jordan, 2002), there are no oral small-molecule drugs against PPI. PPI targets that are extracellular can be addressed using protein drugs such as antibodies (Groner et al., 2004; Hale, 2006), decoy receptors, or other engineered proteins (Marshall et al., 2003; Johnson-Leger et al., 2006). Protein drugs (a.k.a. ‘‘biologics’’ or ‘‘biopharmaceuticals’’) cannot be administered orally, because they do not survive the gut and are poorly absorbed, and so their use in chronic diseases is limited to serious conditions for which patients are willing to tolerate repeated injections. The majority of PPI targets, however, are intracellular proteins involved in the myriad transient and stable protein–protein complexes that mediate cell signaling, cell structure and movement, and other cellular functions (Pawson and Nash, 2000, 2003). The ability to selectively disrupt interactions between signaling proteins would provide a means to regulate cell signaling and function that could have benefit in many disease states. The large number of PPI targets that reside and function inside the cell cannot generally be addressed using biologics, even if parenteral administration is acceptable for the disease in question, because large molecules such as proteins typically cannot cross the cell membrane to access the interior of the cell. Our ability to exploit this wealth of intracellular PPI drug targets, therefore, depends on our ability to develop inhibitors of these interactions that are small molecules with drug-like physicochemical and pharmaceutical properties. Because of the pharmaceutical industry’s historical lack of success at targeting PPI, this class of targets has generally come to be thought of, in industry jargon, as ‘‘undruggable,’’ meaning that it is impossible—or very difficult—to identify druglike small-organic molecules against them. Consequently, many biologically compelling targets have been judged intractable for oral drugs. If approaches could be found to reliably identify oral drugs against PPI targets, the number of gene families that are considered druggable would be greatly increased leading to thousands of new drug targets and, ultimately, to many new drugs. The ability to selectively disrupt specific PPI interfaces and thereby block particular interactions within the cell would provide a means to control many important cellular processes. In addition to rendering a substantial additional fraction of the genome druggable, the ability to target PPI with small molecules would provide a means to replace many parenteral biopharmaceuticals with oral drugs, allowing these proven therapeutic strategies to be extended to additional, somewhat less serious indications for which patients are unwilling to tolerate repeated injections. Finally, an improved ability to target PPI would provide new and different ways to inhibit intracellular signaling enzymes. Most signaling enzymes, such as kinases, phosphatases, caspases, heat shock proteins, and so on, in addition to binding their small organic or peptide substrates, perform their functions in complex with other proteins. The ability to target the PPI sites on such a target or its protein binding partner would, therefore, provide alternative options for modulating the target’s activity, potentially in more subtle or selective ways than by ablating all activity by targeting its active site. It is clear that an ability to develop orally available small-molecule inhibitors of PPI would have a huge impact on drug discovery. The potential value of this

STRUCTURE AND PROPERTIES OF PPI

201

opportunity has not been lost on the pharmaceutical industry and, despite many unsuccessful attempts, industry programs targeting PPI continue occasionally to be reported. A number of academic groups have also worked to address individual elements of the problem. In recent years, the application of special approaches—to be discussed in detail below—has begun to show some success, and at least one small-molecule PPI inhibitor has entered the clinic (Oltersdorf et al., 2005). However, the challenge is still generally regarded as very high risk in most instances. The following sections will discuss the challenges associated with developing small-molecule antagonists against PPI from a structural and physicochemical perspective. I will describe current approaches to addressing this difficult problem, with their strengths and limitations, and highlight areas in which additional research or new technology has the potential to substantially increase prospects for success. I have not attempted to comprehensively describe the various examples of PPI inhibitors that have been reported in the literature, as these have been reviewed extensively (Cochran, 2000; Toogood, 2002; Arkin and Wells, 2004; Fletcher and Hamilton, 2007). Rather, I have focused on the fundamental structural and energetic issues that govern molecular recognition by PPI targets of their natural protein ligands and of small-molecule inhibitors, drawing on selected examples from the literature to illustrate key points, and on discussing how these considerations affect the practical task of drug discovery against these difficult but enticing targets.

7.2

STRUCTURE AND PROPERTIES OF PPI

Interactions between proteins involve arrayed atoms and groups on each binding partner that come together to form an interface at which multiple noncovalent contacts stabilize the resulting complex. A great deal has been learned about the structure and properties of PPI interfaces through biophysical, structural, and bioinformatic approaches. In the following section, I shall summarize the current state of knowledge on the structure and properties of PPI, highlighting the issues that are most relevant to druggability. 7.2.1

Constitutive versus Transient PPI

Interactions between proteins occur through a wide range of mechanisms that present broadly different problems for inhibitor design, from the polymerization of actin in the cytoskeleton to the transient embrace of its substrate by a proteinase enzyme (Ofran and Rost, 2003). Different studies of PPI interface structure and energetics have sometimes reached conflicting conclusions due not only to the use of different methodologies but often also because they based their analyses on different sets of interactions. One important distinction that a number of studies have highlighted is that between constitutive or ‘‘permanent’’ interactions, and those that are transient (Jones and Thornton, 1996; Lo Conte et al., 1999; Ansari and Helms, 2005). Constitutive interactions are those between the polypeptide subunits of proteins that exist only as oligomers. Transient interactions, in contrast,

202

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

involve the reversible association of proteins that exist for part of the time as separate, dissociated molecules (Headd et al., 2007). This distinction is important because the structures and properties of the PPI interfaces in these two cases are somewhat different, having evolved under different evolutionary pressures, and present different problems for inhibitor development. Binding sites involved in transient interactions must confer a tendency to bind to their specific partners while avoiding nonspecific interactions with other proteins. The surface composition of these transient interfaces must therefore not be so different from the composition of noninteracting protein surfaces as to be incompatible with the stable existence of the monomeric protein (Tsai et al., 1997; Lo Conte et al., 1999; Ansari and Helms, 2005). Constitutive interfaces do not have this constraint; the binding sites that mediate constitutive complexes can be structured in ways that are incompatible with a stable existence of the monomer in solution and that would be highly prone to nonspecific associations or aggregation if the oligomer were to dissociate. 7.2.2

Physicochemical Properties and Residue Propensities of PPI

The area of solvent-accessible protein surface that is buried at the interface of a PPI complex—calculated by summing the solvent-exposed surface areas of the two separate monomeric binding partners and subtracting the solvent-accessible surface area of the complex—varies over a wide range. In their seminal analysis of 75 transient protein–protein complexes, Lo Conte and coworkers found that almost half ˚ 2 (i.e.,
800 A ˚2 had buried surface areas that fell within the range 1600  400 A per binding partner), which they defined as ‘‘standard size.’’ A few interfaces ˚ 2 of buried surface area) and the remainder as were classed as ‘‘small’’ (<1200 A 2 ˚ ‘‘large’’ (>2000 A ) (Lo Conte et al., 1999). It has subsequently been confirmed in several studies that transient PPI interfaces typically fall within this ‘‘standard size’’ range, while constitutive PPI tends to be somewhat larger. Analysis of data sets dominated by constitutive complexes initially suggested that PPI interfaces were substantially more hydrophobic than protein surfaces in general (Vakser and Aflalo, 1994; Young et al., 1994). Constitutive interfaces have compositions that are not very different from that of the hydrophobic core of a protein and are greatly depleted in charged residues compared to the protein’s solvent exposed surface (Lo Conte et al., 1999). However, studies that specifically addressed the properties of transient complexes established that in these cases the PPI interface, taken as a whole, is not discernibly more hydrophobic than nonbinding protein surfaces, though the number of charged residues (except for Arg) is somewhat lower (Lo Conte et al., 1999; Ansari and Helms, 2005). In general, PPI interfaces that are larger—either in absolute terms or relative to the total surface area of the protein—tend to be more hydrophobic, and smaller ones more polar (Tsai et al., 1997; Glaser et al., 2001; Ansari and Helms, 2005). More important, even in transient complexes for which the overall interface tends to be fairly balanced with respect to polar versus nonpolar surface, the distribution of polarity is not uniform. Instead, the interface can often be seen to contain two distinct zones, a core that is enriched in hydrophobic contacts surrounded by a rim of more polar

STRUCTURE AND PROPERTIES OF PPI

203

interactions (Bogan and Thorn, 1998; Lo Conte et al., 1999; Chakrabarti and Janin, 2002; Bahadur et al., 2004). This separation of the binding site into central and peripheral zones with different physicochemical properties is crucial to the energetics and the selectivity of protein–protein binding, as described below. A standard size interface for a transient PPI involves, on each binding partner, an average of
90 atoms contributed by
22 amino acid residues (Lo Conte et al., 1999). Of these interface atoms—defined by Lo Conte and coworkers as all atoms that approach an atom on the binding partner to within the sum of their van der ˚ (approximately the diameter of a water molecule)—roughly Waals radii plus 2.8 A half make direct van der Waals contact with the binding partner and approximately one-third become fully buried upon binding (i.e., lose all solvent-accessible surface area) (Lo Conte et al., 1999; Chakrabarti and Janin, 2002; Bahadur et al., 2004). Atoms buried at the interface are closely packed like those in the interior of proteins; indeed, if water molecules are taken into consideration, then the entire interface is tightly packed (Lo Conte et al., 1999). Most often the interface involves interactions between polypeptide loops and turns rather than between a-helices or b-sheets. Where organized secondary structure is involved, helix–helix and sheet–sheet interactions are relatively common but helix–sheet interactions are rare (Jiang et al., 2003), perhaps because it is harder to achieve close packing between these dissimilar secondary structural elements (Ansari and Helms, 2005). A more detailed understanding of the atomic-level interactions that mediate protein–protein binding has been achieved by examining the propensities of different amino acid side chain and main chain groups to participate at PPI interfaces. Lo Conte and coworkers found the transient interfaces they studied were substantially enriched in aromatic residues (His, Tyr, Phe, and Trp, which on average collectively contributed 21% of the interface area) compared to noncontact protein surfaces and were also somewhat enriched in residues with aliphatic side chains (Leu, Ile, Val, and Met). In contrast, the interfaces were relatively poor in the charged residues Asp, Glu, and Lys. Arg is an interesting exception to this relative dearth of charged residues; it is in fact the amino acid that on average made the largest single contribution to interface area at
10% (Bogan and Thorn, 1998; Lo Conte et al., 1999) and is even more common at large interfaces (Glaser et al., 2001). Other studies of transient interactions have confirmed most of the above findings (Headd et al., 2007). There is some variation in residue frequency with interface size, however. In general, hydrophobic residues (L, I, F, V) are relatively ˚ 2) and polar residues (S, D, N, G, enriched at large interfaces (buried ASA > 5000 A 2 ˚ P, K, E) at smaller ones (<1000 A ) (Tsai et al., 1997). In a larger study that examined 621 PPI of all kinds with known high-resolution X-ray crystal structures, the residue pair found to interact most frequently across the interface was Trp–Arg (Glaser et al., 2001). Trp–Pro and Phe–Ile are also common pairings (Glaser et al., 2001; Ansari and Helms, 2005). All these pairs involve side chains containing planar sets of atoms packing against each other, a type of interaction that contributes to tight packing at the interface (Yan et al., 2008). Studies that specifically focused on transient PPI, which as stated above typically are considerably more polar than constitutive interfaces, show that these interfaces

204

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

have a greater prevalence of salt bridges, the four most common interaction pairs in one study being Glu–Arg, Asp–Arg, Asp-Lys, and Glu-Lys (Ofran and Rost, 2003; Headd et al., 2007). The next four most common interactions all involved Tyr, respectively, with Arg, Asn, Lys, and Asp. The Tyr–Arg interaction was about equally likely to involve a hydrogen bond with the Tyr hydroxyl group or alternatively to make a p–cation interaction. Arg–Trp was the next most common interaction, being especially prevalent at the core of the interface (Headd et al., 2007). In general, the aliphatic portion of Arg often engages in hydrophobic interactions with nonpolar residues on the opposite binding partner (Glaser et al., 2001), though in constitutive PPI Arg has been shown to also engage in hydrogen bonding (Janin et al., 1988). In addition to the side chain–side chain interactions discussed above, a substantial portion of the contacts in transient PPI are made by backbone atoms (Lo Conte et al., 1999; Mintz et al., 2005; Headd et al., 2007); Headd and colleagues found that they contributed
23% of the interface area and that about one-thirds of all interactions involved backbone atoms on one or both binding partners. Although the studies cited above show that transient protein–protein interactions make ample use of both charged and hydrophobic pairings, these interactions are not distributed randomly within the interface but tend to occur as pairs or clusters of charged residues on the periphery and hydrophobic patches in the core (Headd et al., 2007). 7.2.3

Binding Energetics and ‘‘Hotspots’’

A key insight into the energetics of protein–protein binding came with the discovery that, in the binding of human growth hormone to its receptor, the bulk of the binding energy resulted from interactions of a small number of amino acid residues clustered near the center of the interface, with the majority of interface residues contributing little or nothing to affinity (Clackson et al., 1998). It has subsequently been established as a general feature of transient protein–protein binding events that the observed affinity is driven by energetic ‘‘hotspots’’ that constitute only a small fraction of the total interface area (Bogan and Thorn, 1998; DeLano, 2002; Keskin et al., 2005). In transient PPI, hotspots tend to be more hydrophobic than the interface as a whole, though individual intermolecular interactions involved can be hydrophobic or polar (Bogan and Thorn, 1998). The amino acid residues most commonly found at PPI hotspots are Trp, Tyr, and Arg; (Bogan and Thorn, 1998; Ansari and Helms, 2005). Ile, His, Asp, and Pro are also overrepresented in comparison to their occurrence in proteins as a whole. Most underrepresented are Val, Leu, Ser, and Thr (Bogan and Thorn, 1998). A common feature of hotspot residues is that they almost invariably become completely buried upon binding. Not all fully buried residues contribute to hotspots, however, so complete loss of solvent accessibility upon binding is a necessary but not a sufficient condition for a hotspot residue. Bogan and Thorn inferred from the buried location of hotspot residues that exclusion of water is critical to the generation of substantial binding energy at the interface. They formulated what is termed

STRUCTURE AND PROPERTIES OF PPI

205

the ‘‘O-ring’’ hypothesis, which postulates that a ring of relatively polar interactions serves to exclude water from the central region of the interface, ensuring a lower dielectric environment in the core region that strengthens interactions deriving part of their interaction energy from electrostatic forces (Bogan and Thorn, 1998). Some PPI interfaces are ‘‘wet’’ throughout, however, and water-mediated interactions can contribute to binding energy (Janin, 1999; Lo Conte et al., 1999; Rodier et al., 2005). It is possible, therefore, that the tendency for hotspot residues to be buried reflects the importance of direct close packing (i.e., close packing without the entropic cost of immobilizing a water molecule as part of the interaction), rather than or in addition to the dielectric effects proposed by Bogan and Thorn. With regard to the role of solvent, water molecules at the interface help ensure tight packing (Lo Conte et al., 1999) and bridge some polar interactions (Janin, 1999). On average, there are
18 water molecules per interface, or one per ˚ 2, though their abundance varies over a wide range. The majority of inter
100 A face waters make hydrogen bonds with both binding partners, and in fact watermediated hydrogen bonds are more common than direct hydrogen bonds (which ˚ 2 of interface area) (Lo Conte et al., 1999). Direct hydrogen occur at one per 170 A bonds most commonly involve a carbonyl oxygen on one side of the interaction, though in almost one in three one or both interacting groups are charged (Lo Conte et al., 1999). Our understanding of structural basis for energetic hotspots has now advanced to the point where computational methods for predicting them have been developed (Kortemme et al., 2004; Moriera et al., 2006, 2007). Nevertheless, there are indications that the interplay between structure and binding energy in PPI hotspots is quite complex. Several studies have suggested that a single PPI interface can contain more than one discrete hotspot, residing within a so-called hot region (Keskin et al., 2005; Reichmann et al., 2005). Moreover, in at least some cases, it has been reported that the effects on the free energy of binding for different hotspot residues are not additive (Bernat et al., 2004). Cooperativity in the roles of individual hotspot residues has also been proposed to allosterically link hotspots in hot regions of the same protein (Reichmann et al., 2005; Moza et al., 2006). The question of how proteins that engage in transient PPI protect themselves against nonspecific interactions mediated by the same binding surfaces was specifically examined by Bahadur et al. (2004). These researchers concluded that the key factors were the high degree of shape complementarity that minimizes energetically costly cavities at the interaction site and also a high density of hydrogen bonds per ˚ 2 polar area), which strongly unit of buried polar surface area (about one per 75 A selects for chemically complementary surfaces. Thus, while constitutive interactions tend to be mostly driven by hydrophobic forces (Tsai et al., 1997), transient PPI involve a complex balance of hydrophobic versus polar interactions structurally arranged to provide strong selectivity based on high degrees of both physicochemical and shape complementarity. It might be expected that transient protein–protein interactions that occur with particularly high affinity might differ in some qualitative way from much weaker ones, with the former perhaps tending to more resemble the structural

206

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

and physicochemical properties of constitutive complexes. This is not so, however. Nooren and Thornton (2003) examined a set of weakly interacting transient PPIs and found that structurally they closely resembled stronger transient PPI. They ˚ 2), found that the interfaces were of standard size (average buried SA
1500 A with a density of hydrogen bonds, fraction of interface that is nonpolar (63  7%), and closeness of packing similar to those found for higher affinity complexes (Nooren and Thornton, 2003; Bahadur et al., 2004). Thus, the differences in affinity between strong and weak transient complexes are apparently achieved by more subtle factors—such as the number and strength of the individual interatomic contacts at the interface, and the free energy changes associated with desolvation of the interacting surfaces and with any conformational changes that occur upon binding—that do not leave a strong structural fingerprint. The current picture of the energetics of transient PPI thus involves a core of hydrophobic docking that contributes significantly to the stability of transient ˚ 2 of nonpolar PPI. One estimate of hydrophobic interaction energy is 86 cal/mol/A contact area (Bahadur et al., 2004). This core region is surrounded by a ring of charged and polar interactions (Chakrabarti and Janin, 2002)—which require more precise juxtaposition for optimal binding—that serves to ensure selectivity of binding (Ansari and Helms, 2005) and to stabilize the unbound protein from nonspecific interactions and aggregation. This division of labor helps account for why transient interfaces are relatively enriched in charged and polar residues and poor in hydrophobics compared to constitutive PPI, because for transient interactions selectivity of recognition is more important and affinity somewhat less so (Ansari and Helms, 2005). This model has several implications for the discovery of small-molecule inhibitors against PPI. The existence of concentrated energetic hotspots raises the possibility that even quite small synthetic inhibitors might be able to generate substantial binding energy—and thus bind with high affinity—by interacting with just a small portion of the PPI binding site. As will be discussed below, however, there are additional topological and dynamic considerations that to some extent complicate this issue. We can also infer that small-molecule inhibitors of constitutive PPI might generally need to be highly hydrophobic in order to complement the hydrophobic nature of those interfaces. In contrast, inhibitors of transient PPI—which includes the majority of intracellular signaling targets of highest interest—will likely require more balanced physicochemical properties to enable strong binding to the mixed polar–apolar composition of transient interfaces. Such interfaces derive binding energy not only from hydrophobic contacts but also from hydrogen bonds (direct or water mediated) and from charge–charge interactions. Thus, transient PPI targets may provide better prospects for inhibition by ‘‘drug-like’’ molecules, which typically engage in a mix of interactions with their targets and require a delicate balance between hydrophobic and polar character to achieve good pharmaceutical properties. In addition, the residue propensities reported for PPI interfaces may provide valuable clues to what type of functionality might interact best at the interface. For example, given the substantial overrepresentation of aromatic amino acids and arginines at PPI interfaces, it might be expected that small molecules containing

STRUCTURAL AND PHYSICOCHEMICAL CHALLENGES TO INHIBITING PPI

207

similar functionality might be particularly favorable. Some support for this notion is provided by the observation that aromatic compounds (particularly biphenyls) were strongly overrepresented among the hits obtained from screening a compound library against a set of difficult targets that included some PPI targets (Hajduk et al., 2000). Finally, it has been hypothesized that nonhotspot residues are not necessarily optimal for binding to an individual partner but play a role in allowing multispecificity of binding—that is, the ability to engage in selective binding through essentially the same site to two or more different binding partners at different times (Humphris and Kortemme, 2007). Some shared interfaces are large, and each binding partner selects a distinct subset of interface residues with which to make energetically significant interactions, while others are smaller and each binding partner utilizes the same hotspot residues. It has recently been suggested that the latter, because they have more compact hotspots, might be better prospects for inhibition with a small molecule (Humphris and Kortemme, 2007).

7.3 STRUCTURAL AND PHYSICOCHEMICAL CHALLENGES TO INHIBITING PPI WITH SMALL MOLECULES For a given PPI target to be susceptible to inhibition by a small-molecule drug, one or the other binding partner must have a binding site that is capable of binding a small molecule with high affinity. The question of PPI druggability thus reduces to understanding what properties in both the putative PPI binding site and the putative small-molecule ligand will allow for strong binding between them. In the following section, we will consider the structural requirements on the protein side and on the small-molecule ligand side in greater detail. The challenge of achieving strong binding between a protein and a small organic molecule is, in large part, topological. The interaction between a protein and its ligand involves the coordinated effects of multiple individual interactions between atoms or groups of atoms on each binding partner. The principal types of intermolecular interactions involved in stabilizing protein–ligand complexes are hydrophobic contacts, salt bridges, and hydrogen bonds. These interactions individually are rather weak (a fraction to a few kcal/mol). Multiple points of contact are therefore needed to generate sufficient overall binding energy between protein and ligand for strong binding, requiring a substantial area of contact between them. When one of the binding partners is a small organic molecule, to achieve a sufficient contact area the protein must to some extent wrap around the small ligand so that a substantial fraction of the surface of the latter is in direct contact with the protein. Only in this way can a sufficient number of binding groups on the protein converge upon the small volume of space occupied by the ligand (Fig. 7.1). In a comprehensive study of the structural factors that characterize drug binding sites in proteins, Nayal and Honig (2006) established that size and shape are indeed the dominant characteristics. Among the 99 crystal structures of proteins in complex with drugs or drug-like small molecules that were included in their analysis, they found that on average the drug binding pocket ˚ 3 and a maximum depth of roughly 7–11 A ˚ . They also found had a volume of 930 A

208

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

FIGURE 7.1 Topological comparison of active sites evolved to bind (a) small molecules versus (b) protein ligands.

that the binding pockets tended to be fairly complex in shape, consistent with a requirement for high shape complementarity between protein and ligand. In their analysis of hits derived from NMR screening against 23 targets Hajduk and colleagues similarly found that factors reflecting the topology of the binding site were the predominant differentiators between sites that did and did not bind smallmolecule ligands. In particular, they concluded that the ligand binding sites dis˚ 2 and a ratio of pocket surface played an optimum nonpolar contact area of
75 A 1 ˚ area to pocket volume of
0.3–0.6 A , suggesting that pockets that are either too spherical or too elongated are less likely to make good binding sites for small molecules (Hajduk et al., 2005a). More important, this study showed that the interplay of different factors that correlate with small-molecule binding is complex, with highly favorable values for some parameters able to offset other characteristics that fall somewhat below the typical threshold for a small-molecule binding site. Davis and Teague (1999) have argued that, not unlike the forces that govern binding in transient PPI, hydrophobic interactions provide the main driving force for protein drug binding, with polar interactions and shape selectivity serving to provide selectivity. It is easy to understand why proteins that have evolved to interact with endogenous substrates or ligands that are themselves small organic molecules have historically proven most amenable to inhibition by small-molecule drugs. The functions of these proteins, in general, require them to bind strongly to their natural ligands or substrates (though in these functional interactions only a fraction of the intrinsic binding energy is typically expressed as affinity, with the remainder being utilized

STRUCTURAL AND PHYSICOCHEMICAL CHALLENGES TO INHIBITING PPI

209

for catalysis or allosteric activation (Jencks, 1975)). Such proteins therefore almost invariably contain a cleft or cavity that has evolved to bind the natural small ligand in a site where multiple binding groups can converge upon it. A synthetic drug can exploit this topology by utilizing some of the same binding groups on the protein, or others that the preexisting cavity positions conveniently, to achieve the highaffinity binding necessary for potent inhibition (Fig. 7.1a). Because the drug is optimized for binding rather than for activity as a substrate or for other biological function, it is typically possible to identify synthetic inhibitors that bind with substantially higher affinity than does the natural substrate or ligand. For example, a typical kinase might interact with its ATP substrate with an affinity that is in the low-to-mid micromolar range, and yet it is frequently possible to identify synthetic inhibitors that bind to the ATP binding site in the kinase active site with low nanomolar or even subnanomolar affinities. For a PPI target, in contrast, evolution has been working to achieve a goal that is quite different from that of the drug hunter. To bind with appropriate affinity and selectivity to its natural protein ligand, the target must complement the shape and composition of that ligand’s cognate binding surface, which is very different in molecular volume and shape from a small organic molecule. In the absence of a cavity or cleft of appropriate dimensions, few binding elements on the relatively flat protein surface can be brought in contact with a small synthetic ligand at any one time, and thus only weak binding can generally be achieved (Fig. 7.1b). When discussing druggability, it is useful to categorize PPI interfaces according to their topology. Some protein–protein interactions involve a short peptide or phosphopeptide segment or loop on one binding partner that binds into a cleft or crevice on the other. The experimental hallmark of such cases is that binding can be recapitulated using a short synthetic peptide or phosphopeptide of the appropriate sequence, either in a direct binding assay or by showing that it can inhibit the interaction between the protein binders. Examples include the large class of heterodimeric cell adhesion receptors known as integrins, many of which mediate cell adhesion by binding to short peptide loops on their protein ligands such as the canonical Arg-Gly-Asp (‘‘RGD’’) motif (Humphries, 1992), SH2 and SH3 domains, which recognize phosphopeptide loops or strands on many intracellular signaling molecules (Pawson et al., 2001), and the interaction of proteinases with their protein substrates (Tyndall and Fairlie, 1999) or with proteinaceous inhibitors such as serpins (Huntington et al., 2000). Other PPI involve a larger cleft on one partner that accommodates an entire alpha-helical segment of its ligand. Examples include interactions among the B cell lymphoma (Bcl) class of signaling proteins that help regulate pro- and antiapoptotic signaling and thus play a key role in cell survival and cell death (Sattler et al., 1997), and the interaction of the oncogene p53 with its ligand HDM2 (Dudkina and Lindsley, 2007). Finally, a large number of biologically important PPI involve two large and relatively flat protein surfaces coming together. Interactions of this kind are typical for cytokines and growth factors binding to their receptors, but also describe many intracellular PPI that are not exclusively driven by phosphopeptide recognition such as the interactions of G-proteins with each other and with G-protein-coupled receptors (Scott et al., 2001). The best

210

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

characterized among this class of interaction is that between human growth hormone and its cell surface receptor (Wells, 1996; Clackson et al., 1998). Interactions involving an isolated peptide loop or strand on one binding partner necessarily require a concave groove in the other, resulting in a highly curved contact interface. Such a site thus contains a preorganized cleft that is of roughly the appropriate dimensions to converge a substantial array of binding functionality onto a small ligand. It is not surprising, therefore, that finding small organic ligands that bind strongly to targets such as proteinases and integrins has typically been relatively straightforward (Shimoaka and Springer, 2003; Fear et al., 2007). Not to say that finding oral drugs against such targets is easy, but multiple successful examples exist (Imming et al., 2006), and the obstacles to success generally do not involve difficulty in finding high-affinity small molecule binders. Binding sites that accommodate an entire alpha helix also have preorganized concavity that converges binding functionality on a small-molecule ligand, but absent significant conformational flexibility is necessarily rather larger than is ideal to snugly envelop a conventional drug-like small molecule. Many attempts have been made to design nonpeptidic small molecules that present substituents with a spacing and orientation that mimics the side chains on an alpha helix (Jain et al., 2004; Davis et al., 2007), though in general the resulting molecules do not appear promising as starting points for the development of oral drugs. The recent report of a bona fide small-molecule antagonist of a subset of Bcl-2 family proteins (Oltersdorf et al., 2005), discussed in more detail below, illustrates that drug-like inhibitors of alpha-helix-accommodating binding sites can be achieved in some cases. It is the third class of PPI binding sites—that is, interactions between relatively flat surfaces on each protein binding partner—that represents the most daunting challenge for small-molecule drug discovery. In these cases, the sites have evolved to complement a large, relatively flat surface rather than to focus binding energy upon an isolated strand or helix, and thus generally do not contain any preexisting cleft or cavity that is suitable to accommodate a small-molecule ligand. Nevertheless, recent years have seen several highly instructive examples of small-molecule inhibitors that act at such sites, including antagonists of the transient PPI targets interleukin-2 (Tilley et al., 1997; Braisted et al., 2003), nerve growth factor (Niederhauser et al., 2000), and CD80 (Uvebrant et al., 2007), and molecules that disrupt constitutive PPI in inducible nitric oxide synthase (Ohtsuka et al., 2002) and in the trimeric cytokine TNFalpha (He et al., 2005). It is not yet clear for any of these cases that the inhibitors can be optimized to achieve the combination of high potency and selectivity plus good pharmaceutical properties necessary for a drug. Nevertheless, the success that was achieved in finding small, and in some cases quite potent, small-molecule inhibitors of these proteins suggests that even this most difficult class of PPI targets might be tractable in some instances. 7.3.1

Key Role of Adaptivity at the Interface

The above description of PPI interfaces and the structural requirements for small molecule drug binding sites ignores the critical fact that proteins are not static

STRUCTURAL AND PHYSICOCHEMICAL CHALLENGES TO INHIBITING PPI

211

structures, but instead are highly dynamic molecules that engage in substantial side chain and main chain motions across a variety of timescales. Proteins can be considered to exist as an ensemble of conformational states that are populated according to their relative free energies and that interconvert at rates governed by the activation barriers between them (Hilser et al., 1998). Thus, the potential for a given protein to bind a drug-like small molecule with high affinity and selectivity cannot be fully determined from the structure and the surface topology seen in a crystal structure of the protein. Instead, it is necessary to include in the assessment all the conformations of the protein that are energetically accessible in solution (i.e., all conformational states that have a free energy that is within a few kcal/mol of the most stable conformation). An example of the importance of structural adaptivity to ligand binding at PPI interfaces is provided by the characterization of small-molecule inhibitors of the interaction of the cytokine interleukin (IL)-2 with its receptor. The first such inhibitor was discovered through attempts to make a small molecule that would mimic two residues on IL-2, Arg38 and Phe42, known to be important for binding to the receptor. Surprisingly, the resulting molecule, Ro26-4550, was determined to inhibit the IL-2/IL-2Ralpha interaction by binding not to the receptor but to IL-2 itself (Tilley et al., 1997; Emerson et al., 2003). A cocrystal structure of Ro264550 with IL-2, obtained by Arkin et al. (2003), showed that the ligand induces substantial conformational changes among the surface residues of IL-2 to form a binding site that is complementary to the small molecule but is not seen in the crystal structure of IL-2 alone (Fig. 7.2). The conformational transformation that accompanies binding appears to involve only low-energy structural changes, suggesting that the ligand has captured an alternative conformation from among the ensemble of low-energy conformational states available to the protein. The same group was able to identify additional compounds that inhibit the IL-2/IL-2Ralpha interaction, one with the impressively high affinity of 60 nM, by taking advantage of this structural adaptivity of the PPI hotspot on IL-2 (Braisted et al., 2003; Waal

FIGURE 7.2 Conformational changes observed at the surface of IL-2 upon binding the small-molecule inhibitor RO26-4550. Reproduced from Whitty and Kumaravel (2006). # 2006, Nature Publishing Group. (See the color version of this figure in the Color Plates section.)

212

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

et al., 2005). A molecular dynamics study of the impact of protein motions on surface topology suggests that surface pockets open and close quickly, and may be as large as conventional ligand binding pockets (Eyrisch and Helms, 2007). The above results are particularly significant for rather flat PPI interfaces, of which IL-2/IL-2Ralpha is a characteristic example, which absent significant conformational adaptation do not have a topology that lends itself to binding small molecules with high affinity. In such cases, evidence of structural adaptivity at the interface—for example, from high crystallographic B factors, structural differences observed between different crystal structures, or dynamic information from NMR—gives an indication that the target is more likely to be druggable than might appear from analysis of a single, static crystal structure (Brown and Hajduk, 2006). 7.3.2

Constraints of ‘‘Drug-Like’’ Chemical Space

It is important to recognize that the ability of a given site on a protein target to bind small-molecule ligands depends not only on the nature of the binding site but also on the properties of the specific set of small molecules being considered. In addition to the properties needed to bind to the target protein with high affinity and selectivity, the physicochemical and structural properties of oral drugs are also limited by the need to achieve good ‘‘pharmaceutical properties.’’ The historical difficulty in finding small-molecule inhibitors of PPI is in part a function of current thinking about the physicochemical constraints associated with ‘‘drug-like’’ molecules. In particular, the previous section shows that the topological challenge of converging a substantial array of binding interactions on an organic ligand is greatly exacerbated by the small size of drug-like molecules (Schuffenhauer et al., 2006). Protein targets for which no energetically accessible conformational state contains a pocket of the appropriate dimensions to accommodate a ligand of this size and physicochemical composition almost certainly will not be druggable by conventional drug-like molecules. For proteins in which the only surface cavities that can form are either too small or too large to snugly envelop a conventional drug-sized small molecule, a significantly larger ligand is required to form sufficient contact area—and thus sufficient binding energy—with the protein to achieve high-affinity binding (Jain et al., 2004; Hajduk, 2006; Schuffenhauer et al., 2006). Thus, Oltersdorf et al. (2005) were able to perform the impressive feat of identifying an inhibitor that binds with subnanomolar affinity to the binding site on Bcl-2 and Bcl-XL that accommodates the alpha-helical BH3 ligand, but in doing so arrived at a molecule with the rather high molecular weight of 813 Da. The reported in vivo experiments required that the inhibitor be administered to mice by intraperitoneal injection rather than orally, implying that the compound does indeed have poor pharmaceutical properties. The prospects for inhibiting PPI targets with small molecules clearly would be greatly improved if it were possible to find larger, more structurally complex organic molecules that can bind strongly to protein surfaces lacking conventional drug-sized pockets and yet possessing sufficiently good pharmaceutical properties to be useful as

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

213

drugs (Jain et al., 2004; Hajduk, 2006). A key question, therefore, is whether there exist regions of chemical space that lie outside the conventional drug-like region that might be richer in PPI binders and yet also possess the key properties of solubility, cell permeability, and stability against degradation by Cyp enzymes in the liver. In addressing this question, an important lesson can be learned from natural product chemistry (Reayi and Arya, 2005). Many plants, molds, fungi, and other organisms synthesize organic molecules that are pharmacologically active when ingested by higher species including humans, a substantial number of which have been developed as drugs (Newman and Cragg, 2007). These biologically active natural products typically have considerably greater structural complexity than conventional synthetic drugs, with many more chiral centers and more complex ring structures that together confer an extraordinary diversity of three-dimensional shape and presentation of binding functionality (Veber et al., 2002; Reayi and Arya, 2005). More important, some natural products are orally available. It is currently unclear in many cases whether these molecules are absorbed passively, or whether they instead are taken up through active transport mechanisms (Keller et al., 2006). One possibility is that some of these compounds, by virtue of their complex structure, are amphiphilic due to an ability to adopt distinct conformations in which excessive hydrophobic or polar surface area that might otherwise confer poor pharmaceutical properties is buried or intramolecularly complemented, thereby modulating aqueous solubility or lipophilicity as the environment demands (Ottiger et al., 1997). Veber, in particular, has argued that oral bioavailability does not decrease with increasing molecular weight, provided that the number of rotatable bonds and hydrogen bonding groups remain acceptably low (Veber et al., 2002). It is certainly important to determine whether cell permeation of such compounds occurs through active or passive mechanisms. However, in either case the possibility exists that, by uncovering the structural features that confer permeability, we might identify new regions of chemical space that contain molecules that are larger than conventional drug-like molecules – and therefore have more potential to bind to PPI targets – and yet can function as drugs (Macarron, 2006; Driggers et al., 2008). Significant efforts have been applied toward constructing synthetic compound libraries with natural product-like character (Schreiber, 2000; Milroy et al., 2007) However, creative synthetic strategies must be coupled with a systematic evaluation of the drug-like properties of the various chemotypes that result, if the compounds are to be validated as potential starting points for drugs.

7.4

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

Conventional library screening has historically shown little success in identifying bona fide hits against difficult PPI targets. The low success rate suggests that one or both of the following is true: (i) the chemical compositions of the many different (though not necessarily wholly dissimilar) libraries that have been screened were not appropriate to this type of target, and/or (ii) small molecules that bind strongly to PPI targets are exceedingly rare, with respect to the whole of chemical space, in

214

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

comparison to molecules that bind conventionally druggable targets. In the following section, I will describe the relatively new approach of ‘‘fragment-based screening’’ that has emerged as the method of choice for finding hits against difficult PPI targets and has shown several impressive successes (Carr et al., 2005; Hajduk and Greer, 2007). I will also address some of the issues surrounding the validation of these initial hits and their advancement to leads and ultimately to drugs. 7.4.1

Fragment-Based Screening

A conventional library screen tests a set of compounds of close to drug-like size (
300–500 kDa) for their ability to bind to the target. To be considered a ‘‘hit,’’ a library compound typically must bind to or inhibit the target with an affinity (KD) or 50% inhibitory concentration (IC50) of no greater than 10–50 mM. Such hits typically contain several distinct elements of binding functionality and, to interact strongly enough to be detected at the low mM concentrations tested, they must be able to simultaneously make multiple energetically favorable interactions with the protein target. Thus, conventional library screening relies upon the likelihood that the compound collection contains a number of molecules that are already a fairly good fit to the binding site on the target. In fragment-based screening (FBS), in contrast, the target is screened against a library of very small chemical compounds or ‘‘fragments’’ (molecular weight typically 100-300 Da, sometimes alternatively called ‘‘seeds,’’ ‘‘needles,’’ ‘‘shapes,’’ or ‘‘monophores’’). Compounds from the library that can bind to the target typically do so only very weakly (0.1–10 mM), because they possess little functionality and can make only one or two productive interactions with the protein. Once a fragment hit has been identified, however, it can be structurally embellished to add binding functionality that makes additional interactions with the target, or two or more fragments can be connected together to make a larger molecule that benefits from the combined interactions of its components (Fig. 7.4). Thus, a fragment hit can be a valuable stepping stone toward larger and more highly functionalized molecules that bind to the target with the low micromolar or better affinity required to constitute a good starting point for lead optimization. FBS samples chemical space much more efficiently than is the case for conventional screening (Hann et al., 2001). That is, FBS tests many more binding possibilities in fewer assays than does screening a conventional compound library. The origin of this greater efficiency is illustrated in Fig. 7.3, which shows a hypothetical FBS screen in which a library of N fragments, where in the figure N ¼ 5, is screened for binding to a protein target. The figure illustrates a case in which two fragments are found to bind in two nonoverlapping but nearby sites (Fig. 7.3a). To identify a more elaborate and stronger binding molecule that combines the binding from both of the fragment hits, a second library can be made in which the two fragments are connected by a set of L linkers that position the fragments with a range of separations and relative orientations. The best binders from this secondary screen will be the molecules of composition (fragment 1)–(linker)–(fragment 2) that best complement

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

215

FIGURE 7.3 Fragment-based screening. (a) Schematic representation of the fragmentbased screening process. (b) A library of N fragments samples the same chemical space as a conventional library of N 2  L larger molecules of structure Fragment–Linker–Fragment. Reproduced from Whitty and Kumaravel (2006). # 2006, Nature Publishing Group. (See the color version of this figure in the Color Plates section.)

the extended binding site encompassing the subsites recognized by the two initial fragment hits. To encompass all of the possible combinations of fragment 1, linker, and fragment 2 that were sampled in this FBS exercise in a library of ‘‘ready-made’’ larger molecules would necessitate assaying N  L  N discrete compounds (Fig. 7.3b), whereas to survey this same set of chemical possibilities using the fragment screen required only N þ L compounds to be assayed. A typically sized library of
1000 fragments, if combined with 100 linker possibilities, would thus require only 1100 separate compounds to be assayed, while surveying this same set of binding possibilities using prefabricated molecules of composition (fragment 1)–(linker)– (fragment 2) would require screening a library of 108 discrete molecules. Although the above argument is a somewhat simplistic representation of how FBS is typically done, it serves to illustrate the very great efficiency with which FBS samples chemical space for binding functionality that can interact productively with a given target. The relative inefficiency of conventional screening derives from the fact that finding a hit requires that the library should contain a molecule in which multiple elements of binding functionality are combined with just the right spacing and relative orientation to be reasonably complementary to the binding site on the target. FBS allows each binding moiety to interact with the target with no constraints on relative spacing or orientation with respect to other fragments that bind in other tests, with the problem of how to extend or combine fragments into higher affinity molecules relegated to a subsequent step. Several technical problems must be overcome to enable FBS; chiefly, how to identify molecules that interact with the target only very weakly, how to validate fragment hits to identify true binders and eliminate false positive results arising

216

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

FIGURE 7.4 Strategies for elaborating fragment hits into larger, more potent drug leads.

from assay artifacts, how to select which fragment hits have the greatest potential for advancement into bona fide leads, and finally, how to structurally elaborate fragment hits into more complex and higher affinity binders. Various approaches to overcoming these problems have been developed (Carr et al., 2005; Erlanson, 2006; Hajduk and Greer, 2007). With regard to the size of fragment libraries, it is interesting that, regardless of the screening technology used and the precise composition of their library, most practitioners of FBS have independently arrived at a quantity of 1000–3000 fragments as representing the number of compounds required to ensure a sufficient number of good quality fragment hits. This figure can be compared with the 105–106 compounds contained in a conventional HTS compound library (historically less successful at identifying hits against difficult PPI targets), as empirical evidence for the more efficient way in which FBS samples chemical space (Hann et al., 2001; Schuffenhauer et al., 2005). The challenges involved and the approaches used in advancing the initial fragment hits to more elaborate and higher affinity hits and leads are largely independent of the screening method used. Several years’ experience with FBS has led to a consensus on the best ways to address some of the steps involved, while on others different opinions exist (Rees et al., 2004). Early efforts at FBS emphasized linking fragment hits together to make larger molecules as the preferred means to advance fragments (Fig. 7.4a) (Shuker et al., 1996). Connecting pairs of fragment hits with a range of linkers can, if necessary, be attempted without any detailed information on the relative binding locations and orientations of the fragments on the target, simply by synthesizing and screening a small library of molecules in which the fragment hits are connected by a set of linkers that position the fragments at a range of distances and relative orientations. Linking often proved

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

217

unsuccessful, however, and though it works on occasion, it has become clear that this does not represent a general solution to the problem. One possible reason for the modest success rate of fragment linking arises from the intrinsic structural flexibility of proteins, which raises the possibility that to bind optimally different fragment hits may induce subtly different conformations of the target. In support of this hypothesis is anecdotal evidence to suggest that linking works best when linkers are designed based on a high-resolution crystal structure of the target that shows both fragments bound simultaneously. Another issue with linking fragments is that it constrains the resulting molecules to an extended ‘‘binder–linker–binder’’ topology, and thus prevents the discovery of the kind of compact structures that are common in approved drugs. For several reasons, therefore, some practitioners of fragment-based lead discovery adopted the alternative advancement strategies of fragment ‘‘merging’’ and ‘‘growing.’’ ‘‘Merging’’ involves taking two or more fragment hits that bind in partially overlapping sites and exploiting elements of structural overlap between the fragments to design a larger molecule that encompasses their combined structure and thus captures all of their combined interactions with the target (Fig. 7.4b). ‘‘Growing’’ depends upon identifying a single fragment that from its binding mode and the interactions it makes with the protein is considered particularly promising, and using structure-based methods to build out from that fragment into adjacent sites on the protein where additional binding interactions can be formed (Fig. 7.4c). In an FBS-based lead discovery program, the team will typically select 4–10 fragment hits to advance, planning on significant attrition and aiming to develop 3–5 distinct lead series. In choosing which hits to advance through merging or growing, high-resolution X-ray structures of each fragment in complex with the target are invaluable, so much so that many practitioners will not take a project beyond the screening stage unless crystal structures can be obtained. In choosing which hits to advance by merging or growing, the principal considerations are (i) binding mode, (ii) synthetic tractability (specifically, which fragments can easily be chemically modified or extended in directions that project toward likely looking binding pockets on the target), and sometimes (iii) ‘‘ligand efficiency.’’ Ligand efficiency is a measure of the amount of binding energy generated per unit molecular weight (or per heavy—i.e., nonhydrogen—atom, as a simple surrogate for molecular weight) (Hopkins et al., 2004). A typical drug generates about 0.3 kcal/ mol per heavy atom in its structure (Hajduk, 2006). Although fragment hits are much lower affinity binders than hits derived from conventional library screening, a good fragment hit will in fact have a significantly greater ligand efficiency (Babaoglu and Shoichet, 2006), because it contains only those elements that contribute to binding with little or no excess structure. If the affinities of fragment hits can be measured, then calculating their ligand efficiencies can give an indication of which provide the best energetic base from which to build out into a larger drugsized molecule. For example, Hajduk’s insightful analysis suggests that to represent a strong starting point for drug discovery a fragment hit with a molecular weight of 200 Da. must bind with an affinity of
1 mM or better, while a modestly

218

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

larger hit with a molecular weight of 250 Da would require an affinity of 30 mM to have energetically equivalent prospects for advancement (Hajduk, 2006). It should be emphasized, however, that the best fragment hit to advance is often not the one that binds with the highest affinity or ligand efficiency, but rather the one possessing an acceptable ligand affinity plus a favorable binding geometry. 7.4.2

Validating and Optimizing Hits and Leads

Validating early hits is especially challenging for very weak binders, such as those that often result from lead identification efforts against PPI targets. All biochemical assays have some false positive rate—that is, there is some chance that an inactive compound will appear to give a positive result in the assay. With a highly druggable target, true hits are abundant and only a small fraction of the apparent hits are false positives, making it relatively easy to identify true binders for further advancement. A protein of low druggability, such as a difficult PPI target, renders true hits much rarer but does not change the false positive rate that is intrinsic to a particular combination of assay, compound collection, and screening concentration. Consequently, low druggability targets generally give a much lower ratio of true hits to false positive hits when screened in a given biochemical assay. The high ratio of false positive results observed when screening conventional compound libraries against PPI targets presents a significant practical problem. This problem is greatly exacerbated by the fact that, to compensate for the low hit rate, PPI targets are often screened at high compound concentrations that greatly increase the likelihood that conventional drug-like compounds will interfere with the assay. It is conceivable, therefore, that some early efforts to screen for PPI inhibitors using conventional assays and conventional compound libraries were not so unsuccessful as they appeared, failing not because true hits were absent but because they were rare and could not readily be distinguished from a much larger number of false positives. Our understanding of the origins of assay artifacts has advanced significantly in recent years, and a number of counterscreens and other steps to identify and eliminate false positives have been developed (McGovern et al., 2003; Feng et al., 2007). In addition to the substantially greater rate of true hits against PPI targets that is typically seen with fragment-based screening methods, the success of this approach can probably also be attributed to a lower false positive rate due to the mechanistically diagnostic nature of popular FBS detection techniques, such as X-ray crystallography and NMR, and to the fact that the small and highly soluble compounds that comprise fragment libraries are intrinsically less prone to several common types of assay artifact. As stated by Arkin and Wells (2004): ‘‘In general, a novel molecule can be described as ‘validated’ when it has been shown to bind noncovalently with 1:1 binding stoichiometry to the target of interest.’’ Thus, even after it has been shown that a hit does not result from a biochemical assay artifact, and thus is a ‘‘true’’ assay positive, additional biophysical characterization is often required to establish that the molecule functions by an appropriate mechanism, that is, that it interacts

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

219

with the target by binding reversibly at a defined site and with a defined orientation. A three-dimensional structure of the ligand in complex with the protein target, obtained by NMR or by X-ray crystallography, constitutes strong evidence for a specific binding mode. However, sometimes a high-resolution structure can be difficult to obtain, or questions may exist concerning whether the binding mode seen in the structure reflects the mechanism of action that is responsible for the biochemical activity of the compound. In cases like these, a variety of biophysical methods have been applied to characterize the binding mode and mechanism of action of weak binders in solution. Among the most useful and widely used are various protein-based and ligand-based NMR methods (Huth et al., 2005; Dalvit et al., 2006), surface plasmon resonance (Biacore) technology, analytical ultracentrifugation (Arkin and Lear, 2001), dynamic lightscattering, calorimetry, and fuorescencebased methods (Boehm et al., 2000; Arkin and Wells, 2004; He et al., 2005). The existence of a clear structure–activity relationship (SAR) around the hit is also a crucial validation that the compound functions in a way that will be amenable to optimization. It is, of course, possible to advance hits without the benefit of definitive structural or biophysical evidence of their detailed modes of action, trusting that molecules with undesirable mechanisms will fall by the wayside as others are successfully advanced. Nevertheless, initial hits can be scarce and are often weak binders with PPI targets, and a significant investment of effort is required to advance them. It is therefore important to know what methods exist to objectively validate these early starting points. A given drug target can be considered as posing a particular molecular recognition problem for small-molecule binding. To screen a compound library for smallmolecule ligands that bind to the target above a given threshold of affinity is to ask how many solutions to that molecular recognition problem exist in the library being screened. As the affinity threshold is raised to select for increasingly potent binders, the number of chemical structures that can satisfy this requirement decreases, so there are many fewer molecules that are sufficiently complementary to the target to bind with low nM affinity, and even fewer with affinities at the pM level. The inverse relationship that exists between the affinity threshold and the number of distinct chemical structures that can bind with that affinity can be represented as shown in Fig. 7.5 (Whitty and Kumaravel, 2006). The relationship describes an inverted triangle or funnel shape, with many weak binders at the top and a few strong binders at the bottom (white triangle in Fig. 7.5). Despite the often steep drop-off in number of ligand structures as the affinity threshold is increased, for highly druggable targets there are generally enough different chemical structures that can bind with high affinity to allow identification of one or more that additionally possess the appropriate combination of pharmaceutical properties to have a chance to become a drug. As noted above, screening conventional compound libraries for inhibitors of the most challenging PPI targets has usually failed (for exceptions, see Tilley et al., 1997; Toogood, 2002). This result indicates that these PPI targets present a very stringent molecular recognition problem and that there are few, if any, molecules in the libraries tested that represent ‘‘solutions’’ to this problem even at the fairly

220

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

FIGURE 7.5 Relationship between the affinity of compound binding and the number of distinct chemical structures that can bind with that affinity. (White triangle: conventional screening library, highly druggable target; black triangle: fragment library, poorly druggable target; gray and crosshatched triangles: prospects for optimizing a fragment lead against, respectively, a difficult but tractable target and an intractable target.) Adapted from Whitty and Kumaravel (2006).

modest threshold of
10 mM affinity. Much greater success with typical hit rates from 3–30% (Schuffenhauer et al., 2005) is seen when screening these targets against low molecular weight fragment libraries using assays with a detection limit in the 0.1–10 mM range, indicating that a significant number of chemical structures satisfy the recognition problem posed by these difficult PPI targets if the affinity threshold is relaxed by two or three orders of magnitude. Thus, the ‘‘selection funnel’’ describing these difficult PPI targets differs in shape from that observed for more druggable targets, being much narrower at the mM affinity level but quite wide at the mM level (black triangle in Fig. 7.5). The difference in the shapes of the selection funnels for highly druggable versus PPI targets, shown schematically in Fig. 7.5, raises the following question: if mM hits against PPI targets are very rare, does this imply that the selection funnel diminishes to the vanishing point at a high-affinity threshold, such that low nM binders are impossible to achieve (illustrated by the crosshatched triangle in Fig. 7.5)? There is no general answer to this question for all PPI targets. For some targets, a dearth of screening hits might imply that no strong binders are possible, while in another case even a rare weak hit might be optimizable to achieve a molecule with the potency required to be a drug (gray triangle in Fig. 7.5). Examining case histories of different attempts to identify and optimize hits and leads against PPI targets can shed light on some factors that affect the progression of weak PPI binders. Two contrasting examples are provided by published attempts to develop antagonists against the difficult PPI targets Bcl-2 (Oltersdorf et al., 2005) and TNFalpha (He et al., 2005). Applying the FBS technique known as ‘‘SAR by NMR’’ to the signaling protein Bcl-XL gave, among the best fragment hits, 40 -fluorobiphenyl-4-carboxylic acid and 5,6,7,8-tetrahydronaphthalene-1-ol (Oltersdorf et al., 2005). NMR structural

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

221

FIGURE 7.6 Application of the FBS technique SAR by NMR plus traditional medicinal chemistry optimization strategies led to Bcl-2 inhibitor ABT737. Reproduced with permission from Oltersdorf et al. (2005). # 2005, Nature Publishing Group. (See the color version of this figure in the Color Plates section.)

analysis showed these molecules bound to distinct but nearby sites in the hydrophobic groove that binds the alpha-helical BH3 ligand (Fig. 7.6a). These two fragments were weak binders, having affinities of KD
0:3 and 4.3 mM, respectively. Replacing the carboxyl group of the biphenyl fragment with an acylsulphonamide linker, and using it as an attachment point to enable testing of multiple extensions of the molecule through parallel synthesis, led to the discovery that a 3-nitro-4(2-phenylthioethyl)aminophenyl substituent could reach into the site bound by the tetrahydronaphthalenol hit. The resulting compound, which spanned both sites occupied by the original fragment hits, bound with a greatly increased affinity of 36  1:6 nM (Fig. 7.6c). Further modification of the molecule, to increase affinity for Bcl-XL and Bcl-2 while reducing binding to human serum albumin and

222

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

increasing solubility, led to the compound ABT737 (Fig. 7.6d) which binds to the target with KD < 1 nM (Oltersdorf et al., 2005). Prior screening campaigns against Bcl-XL using conventional compound libraries had established that 10 mM binders against this target were rare and difficult to discover. However, the scarcity of conventional hits clearly did not indicate that the BH3 binding site on Bcl-XL was intrinsically unsuitable to bind a synthetic ligand with high affinity. Rather, the results imply that the topology and composition of the BH3 binding site is such that only a few very specific chemical structures can complement it sufficiently well to bind strongly, or at least that compounds that can do this were nonexistent or exceedingly rare in the libraries that were originally screened. Because the BH3 binding groove on Bcl-XL accommodates a structured alpha helix, this binding site is substantially larger than the typical size of a smallmolecule binding pocket. The site is also highly hydrophobic, which further narrows the range of molecular structures in a given compound library that will effectively complement it. ABT737 solves these problems in three ways. When bound to the target, the phenylthioethyl substituent folds back against the nitroaniline ring to assume an intramolecularly p-stacked conformation that increases the effective bulk of the ligand and allows it to pack tightly into the large binding site (Fig. 7.6b). In addition, with a molecular weight of 813 Da, the molecule substantially exceeds the guideline of MW < 500 Da from Lipinski’s ‘Rule of 5’ (Lipinski, 2000). This extra bulk means that ABT737 has achieved strong binding to the target at the cost of a relatively low ligand efficiency (Hajduk, 2006). Finally, to compensate for the high hydrophobicity required in the ligand to complement the hydrophobic binding site, and thereby achieve acceptable overall solubility and reduce binding to serum albumin, solubilizing groups were added at strategic points on the molecule. These accommodations illustrate two general points about inhibitor development against PPI: (i) the noncanonical (with respect to conventionally druggable binding sites) topology presented by some PPI binding sites can require unusual (for a drug) molecular geometries in the ligand to achieve a snug fit, and (ii) the problems presented by a PPI target site can sometimes be overcome by increasing the molecular weight of the ligand. The molecule that resulted from this effort does not solve the problem of binding to Bcl-XL by venturing into radically new chemical space, but rather makes trade-offs in different properties at the edge of the acceptable ranges for drug-like molecules to achieve remarkably good overall pharmaceutical properties considering the difficulty of the target. Most important, this example illustrates that the fact that initial hits are rare and difficult to achieve for a given PPI target does not necessarily imply that there are no high-affinity solutions to the binding problem. The challenge faced in inhibiting TNFalpha is quite different to that presented by Bcl-XL. TNFalpha is a constitutively homotrimeric cytokine that signals through two distinct cell surface receptors (Palladino et al., 2003; Chatzantoni and Moutzaki, 2006). A number of highly successful monoclonal antibody and other biologic drugs have validated TNFalpha as a high-value drug target in multiple inflammatory and immune disorders (Mousa et al., 2007; Sandborn, 2007). Developing an oral

IDENTIFYING HITS AND LEADS AGAINST PPI TARGETS

223

small-molecule drug that could recapitulate the activity of these successful biologics by blocking the interaction of TNFalpha with its receptors would represent a significant advance by obviating the need for repeated injections, thereby increasing convenience and compliance for patients already served by the biologic drugs and also potentially enabling anti-TNF therapy to be extended to new patient populations for which an injected drug is not appropriate. Many companies and academic groups have attempted to identify small-molecule antagonists of the TNFalpha/TNF-receptor interaction, either by screening conventional compound libraries or by alternative means, but until recently no bona fide small-molecule inhibitor had been reported (Palladino et al., 2003). This lack of success is not, perhaps, surprising as the interaction between TNFalpha and its receptor involves a large flat interface with no apparent concavities on either binding partner that might support strong binding of a drug-like small molecule (Banner et al., 1993). Recently, however, He et al. (2005) reported the discovery, using a fragmentbased approach, of a molecule that inhibited TNFalpha activity both in biochemical binding assays and in cell-based functional assays. This inhibitor, called SP307 (Fig. 7.7a), is relatively small (MW ¼ 548 Da), and inhibited TNFalpha binding and signaling with a potency of IC50
10 mM. Most surprisingly, when the cocrystal structure of this molecule with TNFalpha was solved, it was discovered that rather than binding to an induced surface cleft on the TNFalpha trimer, SP307 actually displaced one of the protein subunits to stabilize an unprecedented complex containing two TNFalpha subunits plus one inhibitor molecule (Fig. 7.7b). In this complex, the inhibitor is bound to the exposed hydrophobic core of the protein dimer and partly occupies the space that the third subunit of the TNFalpha trimer would fill if it were present. Extensive biochemical and biophysical characterization supports the notion that the subunit displacement mechanism is not a crystallographic artifact, but is the mechanism responsible for the inhibitory activity of SP307 in solution. Thus, this molecule has achieved what might be considered even a more difficult feat than blocking the transient interaction between TNFalpha

FIGURE 7.7 SP307 disrupts the constitutively trimeric cytokine TNFa by displacing one subunit. Reproduced with permission from He et al. (2005). # 2005, AAAS. (See the color version of this figure in the Color Plates section.)

224

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

and its receptor, by disrupting the very high-affinity constitutive PPI that holds together the TNFalpha trimer. Careful structural analysis showed that recognition of SP307 by the protein is entirely hydrophobic and shape based (He et al., 2005). No hydrogen bonds or other polar interactions contribute to binding. Comparison of the structure of the complex to that of native TNFalpha trimer indicated that significant conformational adaptation in the residues lining the core of the protein is required for ligand binding. Specifically, the ligand makes contact with a total of six tyrosine residues, three from each TNFalpha subunit, which have undergone substantial side chain motions to create a binding pocket with a shape that is complementary to the inhibitor (Fig. 7.7c). The key role of tyrosine residues is reminiscent of the prominence of this amino acid in PPI hotspots and in the complementarity determining regions of antibodies selected from phage libraries (Fellouse et al., 2004, 2006). Interestingly, as was seen in the Bcl-XL inhibitor discussed above, SP307 adopts a folded conformation when bound to TNFalpha in which two aromatic portions of the inhibitor engage in intramolecular p-stacking. The compact conformation that the inhibitor assumes again indicates that to occupy the noncanonical binding site presented by a PPI target may require a shape that is atypical for conventional drug-like molecules. Attempts to optimize the potency and other properties of SP307 identified a number of related compounds that shared its activity in biochemical assays and that could be shown in X-ray cocrystal structures to also displace a subunit of the TNFalpha trimer. However, no molecules were found that exceeded the
10 mM potency displayed by SP307. This failure appears to be due to the fact that SP307 is already highly complementary in shape to the induced binding site in the core of the TNFalpha dimer, and there are no proximal pockets or nearby polar binding groups that extended analogues of the molecule could reach to form additional binding interactions. Thus, while it is impossible to prove that the next analogue prepared would not have had improved affinity, SP307 appears to represent an example of a weak binder that is terminally optimized for its PPI binding site and cannot be further modified to achieve substantially increased affinity (as represented by the crosshatched triangle in Fig. 7.5). Given enough examples of weak hits such as SP307 or the early leads against Bcl-XL, and a determination of which of them can and cannot be optimized to achieve the level of affinity typically required for a drug, it should be possible to begin to understand in greater detail what features in a PPI binding site and in a ligand correlate with the potential for advancement. With the limited data in the literature to date, however, any such conclusions are necessarily highly speculative. It is not clear, for example, whether helical protein such as Bcl-XL are generally more likely to be druggable with conventional drug-like compounds than are beta sheet proteins such as TNFalpha or whether the situation will prove to be much more complicated. Further analysis of this question as additional examples of PPI inhibitors become published is very important to inform guidelines for selecting targets and identifying early hits and leads that provide the best prospects for success, rather than having to pursue each case to its conclusion to determine the outcome empirically.

ASSESSING THE DRUGGABILITY OF NEW PPI TARGETS

7.5

225

ASSESSING THE DRUGGABILITY OF NEW PPI TARGETS

The previous sections have established that potent small-molecule inhibitors of PPI targets can, at least in some instances, be identified. In this section, I will discuss approaches to predicting the druggability of a new PPI target short of attempting an expensive full-fledged drug discovery effort, an issue that has been the subject of intense study and rapid progress over the past several years. Early attempts to predict the druggability of novel protein targets were based on structural similarity to known classes of druggable proteins (Hopkins and Groom, 2002). The reasonable, if rather conservative, assumption underlying this approach is that a protein is likely to be druggable if it shares the same fold as a protein for which drug-like small-molecule inhibitors are already known. This approach is effective for new genes that can be assigned, on the basis of sequence homology, as a member of a gene family for which the level of druggability is already known. However, the resulting estimate for the number of druggable human gene products is certainly too low. This is because, as Hajduk et al. (2005b) have pointed out, this approach works only for genes that are well annotated and provides no information about the roughly half of the human genome of unknown structure and function. Also, the analysis by necessity excludes the nebulous but real possibility that in the future we may develop novel chemical scaffolds that fall outside conventional drug-like chemical space while retaining good pharmaceutical properties, thereby expanding the range of protein classes against which oral drugs can be developed. Other approaches to predicting druggability are based on analysis of the threedimensional structures of potential new protein targets. These approaches fall into two major categories: those based on analyzing shape and those based on estimating binding energies. Shape-based approaches survey the surface of the protein in question for the presence of clefts or pockets of a size and shape comparable to those of known drug binding sites (Glaser et al., 2006; Weisel et al., 2007). Energy-based approaches analyze the physicochemical composition of binding pockets to estimate the likelihood of binding a small-molecule ligand with high affinity (An et al., 2004). These structure-based methods have the advantage that they can be applied to any protein for which the three-dimensional structure is known, whether or not the particular fold has been previously shown to bind small-molecule drugs. However, the structural and physicochemical factors that correlate with an ability to bind strongly to small-molecule ligands are complex and as yet incompletely understood (Nayal and Honig, 2006). Moreover, the interplay between different parameters is complex. In addition, most of the available methods treat the protein structure, or at least the main chain atoms, as static, and thus do not anticipate the possibility that a ligand might bind to a conformation that differs from that seen in the crystal structure. Incorporation of conformational adaptivity into computational predictions of druggability can dramatically affect the outcome (Brown and Hajduk, 2006; Eyrisch and Helms, 2007). In general, established structure-based methods are quite good at predicting which site on a given protein is most likely to be amenable to binding by a small-molecule ligand but are not

226

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

very good at making absolute predictions of the likelihood that a protein will bind small molecules at all (Hajduk et al., 2005b). Short of screening a new target against a full compound library, currently the best approach to estimating druggability is to perform a limited fragment screen, using a detection technique such as SAR by NMR or crystallographic screening that has a low false positive hit rate, and then assess the druggability of the protein from the number and quality of the hits obtained (Hajduk et al., 2005a, 2005b). Hajduk et al. (2005a) analyzed the results of heteronuclear NMR-based fragment screens against 23 protein targets including a number of challenging PPI targets. The fragment hit rates across these targets varied over two orders of magnitude, from 0.01 to 0.94%. Significantly, a strong correlation was observed between hit rate and the ability to ultimately identify a high affinity (KD < 300 nM) ligand to that site. Importantly, several of the targets with a relatively high druggability score had been shown to give few or no useful hits when screened against conventional compound libraries. This result is consistent with the idea that using fragment screening data in this analysis provides a more accurate indication, compared to conventional library screening, of the small-molecule binding potential of a protein, presumably due to a more comprehensive coverage of chemical space (Hann et al., 2001; Schuffenhauer et al., 2006). These fragment-screening results were correlated with structural characteristics of the binding sites to establish a ‘‘druggability index’’ that was used to estimate the druggability of proteins based on their three-dimensional structure alone. Application of this index to a set of 35 protein targets that were not included in the training set and for which high-affinity smallmolecule ligands were already known correctly classified 94% of those proteins as druggable. The druggability index was subsequently applied to 1096 human proteins for which high-resolution crystal structures are available in the Protein Data Bank (Hajduk et al., 2005b). The analysis suggested that about 35% of the included structures contain at least one site capable of binding a small-molecule ligand with high affinity. In keeping with historical experience, there was significant variation in the predicted druggability of different gene families, with 75–89% of vitamin or steroid binding proteins and lyases predicted to be druggable but only 14–24% of proteins that bind to DNA, RNA, or carbohydrates (Fig. 7.8). Interestingly, 29% of PPI targets were predicted to be druggable, not greatly different from the average for all protein classes. Given that this analysis treated the proteins as rigid structures, and therefore did not capture binding events that involve significant conformational change or induced fit at the binding site, the percentages quoted above likely underestimate the true druggability of at least some of these protein classes (Brown and Hajduk, 2006). Of the methods currently available, for a protein target site of unknown druggability a preliminary screen with a fragment library is probably the best way to assess the potential for high-affinity binding of a smallmolecule ligand and, compared to computational approaches, is more likely to identify a binder that requires or induces a conformational change in the protein. Further research in this area is clearly needed. The ability to predict druggability with confidence would greatly reduce the risk associated with undertaking drug discovery efforts on PPI targets and would reduce the time and resources spent on

227

REFERENCES

Percent druggable

100% 80% 60% 40% 20%

DNA binding

Glycosyl hydrolase

RNA binding

Carbohydrate binding

Antigen binding

Ester hydrolase

Cystein

Protein binding

Enzyme inhibitor

Hydrolase

Ligase

Metallopeptidase

Serine peptidase

Isomerase

Endopeptidase

Protein kinase

Transferase

Lipid binding

Nucleotide binding

Lyase

Oxidoreductase

Steroid binding

Vitamin binding

0%

Drug Discovery Today

FIGURE 7.8 Predicted druggability of different gene families suggests that protein binding targets fall only slightly below the mean. Reproduced with permission from Hajduk et al. (2005b). # 2005, Elsevier, Ltd.

intractable proteins. Such predictive methods must evolve to keep pace with our growing understanding of the role of protein adaptivity in enabling small-molecule binding and with the discovery of new drug-like areas of chemical space. With continuing advances in these areas, it is hoped that the number of PPI targets that prove to be druggable will be considerably higher than the current estimates described above, quite encouraging though they already are, thereby expanding the druggable genome to allow modulation of critical intracellular and extracellular signaling proteins with convenient, orally administered small molecules.

REFERENCES An J, Totrov M, Abagyan R, 2004. Comprehensive identification of ‘‘druggable’’ protein ligand binding sites. Genome Inform. 15:31. Ansari S, Helms V, 2005. Statistical analysis of predominantly transient protein–protein interfaces. Proteins 61:344. Arkin M, Lear JD, 2001. A new data analysis method to determine binding constants of small molecules to proteins using equilibrium analytical ultracentrifugation with absorption optics. Anal. Biochem. 299:98. Arkin MR, Wells JA, 2004. Small-molecule inhibitors of protein–protein interactions: progressing towards the dream. Nat. Rev. Drug Discov. 3:301. Arkin MR, Randal M, DeLano WL, Hyde J, Luong TN, Oslob JD, Raphael DR, Taylor L, Wang J, McDowell RS, Wells JA, Braisted AC, 2003. Binding of small molecules to an adaptive protein–protein interface. Proc. Natl. Acad. Sci. USA 100:1603.

228

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

Babaoglu K, Shoichet BK, 2006. Deconstructing fragment-based inhibitor discovery. Nat. Chem. Biol. 2:720. Bahadur RP, Chakrabarti P, Rodier F, Janin J, 2004. A dissection of specific and non-specific protein–protein interfaces. J. Mol. Biol. 336:943. Banner DW, D’Arcy A, Janes W, Gentz R, Schoenfeld HJ, Broger C, Loetscher H, Lesslauer W, 1993. Crystal structure of the soluble human 55 Kd TNF receptor–human TNF beta complex: implications for TNF receptor activation. Cell 73:431. Bernat B, Sun M, Dwyer M, Feldkamp M, Kossiakoff AA, 2004. Dissecting the binding energy epitope of a high-affinity variant of human growth hormone: cooperative and additive effects from combining mutations from independently selected phage display mutagenesis libraries. Biochemistry 43:6076. Boehm HJ, Boehringer M, Bur D, Gmuender H, Huber W, Klaus W, Kostrewa D, Kuehne H, Luebbers T, Meunier-Keller N, Mueller F, 2000. Novel inhibitors of DNA gyrase: 3D structure based biased needle screening, hit validation by biophysical methods, and 3D guided optimization. A promising alternative to random screening. J. Med. Chem. 43:2664. Bogan A, Thorn KS, 1998. Anatomy of hot spots in protein interfaces. J. Mol. Biol. 280:1. Bork P, Jensen LJ, von Mering C, Ramani AK, Lee I, Marcotte EM, 2004. Protein interaction networks from yeast to human. Curr. Opin. Struct. Biol. 14:292. Braisted AC, Oslob JD, Delano WL, Hyde J, McDowell RS, Waal N, Yu C, Arkin MR, Raimundo BC, 2003. Discovery of a potent small molecule IL-2 inhibitor through fragment. JACS 125:3714. Brown SP, Hajduk PJ, 2006. Effects of conformational dynamics on predicted protein druggability. Chem. Med. Chem. 1:70. Carr RA, Congreve M, Murray CW, Rees DC, 2005. Fragment-based lead discovery: leads by design. Drug Discov. Today 10:987. Chakrabarti P, Janin J, 2002. Dissecting protein–protein recognition sites. Proteins 47:334. Chatzantoni K, Moutzaki A, 2006. Anti-TNF-alpha antibody therapies in autoimmune diseases. Curr. Top. Med. Chem. 6:1707. Clackson T, Ultsch MH, Wells JA, de Vos AM, 1998. Structural and functional analysis of the 1:1 growth hormone:receptor complex reveals the molecular basis for receptor affinity. J. Mol. Biol. 277:1111. Cochran AG, 2000. Antagonists of protein–protein interactions. Chem. Biol. 7:R85. Dalvit C, Caronni D, Mongelli N, Veronesi M, Vulpetti A, 2006. NMR-based quality control approach for the identification of false positives and false negatives in high throughput screening. Curr. Drug Discov. Technol. 3:115. Davis JM, Tsou LK, Hamilton AD, 2007. Synthetic non-peptide mimetics of alpha-helices. Chem. Soc. Rev. 36:326. Davis MD, Teague SJ, 1999. Hydrogen bonding, hydrophobic interactions, and failure of the rigid receptor hypothesis. Angew. Chem. Int. Ed. Engl. 38:736. DeLano WL, 2002. Unraveling hot spots in binding interfaces: progress and challenges. Curr. Opin. Struct. Biol. 12:14. Driggers EM, Hale SP, Lee J, Terrett NK, 2008. The exploration of macrocycles for drug discovery—an underexploited structural class. Nat. Rev. Drug Discov. 7:608. Dudkina AS, Lindsley CW, 2007. Small molecule protein–protein inhibitors for the p53– MDM2 interaction. Curr. Top. Med. Chem. 7:952.

REFERENCES

229

Emerson SD, Palermo R, Liu CM, Tilley JW, Chen L, Danho W, Madison VS, Greeley DN, Ju G, Fry DC, 2003. NMR characterization of interleukin-2 in complexes with the IL-2Ralpha receptor component, and with low molecular weight compounds that inhibit the IL-2/IL-Ralpha interaction. Protein Sci. 12:811. Erlanson DA, 2006. Fragment-based lead discovery: a chemical update. Curr. Opin. Biotechnol. 17:643. Eyrisch S, Helms V, 2007. Transient pockets on protein surfaces involved in protein–protein interaction. J. Med. Chem. 50:3457. Fear G, Komarnytsky S, Raskin I, 2007. Protease inhibitors and their peptidomimetic derivatives as potential drugs. Pharmacol. Ther. 113:354. Fellouse FA, Wiesmann C, Sidhu SS, 2004. Synthetic antibodies from a four-amino-acid code: a dominant role for tyrosine in antigen recognition. Proc. Natl. Acad. Sci. USA 101:12467. Fellouse FA, Barthelemy PA, Kelley RF, Sidhu SS, 2006. Tyrosine plays a dominant functional role in the paratope of a synthetic antibody derived from a four amino acid code. J. Mol. Biol. 357:100. Feng BY, Simeonov A, Jadhav A, Babaoglu K, Inglese J, Shoichet BK, Austin CP, 2007. A high-throughput screen for aggregation-based inhibition in a large compound library. J. Med. Chem. 50:2385. Fletcher S, Hamilton AD, 2007. Protein–protein interaction inhibitors: small molecules from screening techniques. Curr. Top. Med. Chem. 7:922. Glaser F, Steinberg DM, Vakser IA, Ben-Tal N, 2001. Residue frequencies and pairing preferences at protein–protein interfaces. Proteins 43:89. Glaser F, Morris RJ, Najmanovich RJ, Laskowski RA, Thornton JM, 2006. A method for localizing ligand binding pockets in protein structures. Proteins 62:479. Groner B, Hartmann C, Wels W, 2004. Therapeutic antibodies. Curr. Mol. Med. 4:539. Hajduk PJ, 2006. Fragment-based drug design: how big is too big? J. Med. Chem. 49:6972. Hajduk PJ, Greer J, 2007. A decade of fragment-based drug design: strategic advances and lessons learned. Nat. Rev. Drug Discov. 6:211. Hajduk PJ, Bures M, Praestgaard J, Fesik SW, 2000. Privileged molecules for protein binding identified from NMR-based screening. J. Med. Chem. 43:3443. Hajduk PJ, Huth JR, Fesik SW, 2005a. Druggability indices for protein targets derived from NMR-based screening data. J. Med. Chem. 48:2518. Hajduk PJ, Huth JR, Tse C, 2005b. Predicting protein druggability. Drug Discov. Today 10:1675. Hale G, 2006. Therapeutic antibodies–delivering the promise? Adv. Drug Deliv. Rev. 58:633. Hann MM, Leach AR, Harper G, 2001. Molecular complexity and its impact on the probability of finding leads for drug discovery. J. Chem. Inf. Comput. Sci. 41:856. He MM, Smith AS, Oslob JD, Flanagan WM, Braisted AC, Whitty A, Cancilla MT, Wang J, Lugovskoy AA, Yoburn JC, Fung AD, Farrington G, Eldredge JK, Day ES, Cruz LA, Cachero TG, Miller SK, Friedman JE, Choong IC, Cunningham BC, 2005. Small-molecule inhibition of TNF-alpha. Science 310:1022. Headd JJ, Ban YE, Brown P, Edelsbrunner H, Vaidya M, Rudolph J, 2007. Protein–protein interfaces: properties, preferences, and projections. J. Proteome Res. 6:2576. Hilser VJ, Dowdy D, Oas TG, Freire E, 1998. The structural distribution of cooperative interactions in proteins: analysis of the native state ensemble. Proc. Natl. Acad. Sci. USA 95:9903.

230

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

Hopkins AL, Groom CR, 2002. The druggable genome. Nat. Rev. Drug Discov. 1:727. Hopkins AL, Groom CR, Alex A, 2004. Ligand efficiency: a useful metric for lead selection. Drug Discov. Today 9:430. Humphries MJ, 1992. Peptide recognition motifs involved in the binding of integrins to their ligands. Kidney Int. 41:645. Humphris EL, Kortemme T, 2007. Design of multi-specificity in protein interfaces. PLoS Comput. Biol. 3:e164. Huntington JA, Read RJ, Carrell RW, 2000. Structure of a serpin–protease complex shows inhibition by deformation. Nature 407:923. Huth JR, Mendoza R, Olejniczak ET, Johnson RW, Cothron DA, Liu Y, Lerner CG, Chen J, Hajduk PJ, 2005. Alarm NMR: a rapid and robust experimental method to detect reactive false positives in biochemical screens JACS 127:217. Imming P, Sinning C, Meyer A, 2006. Drugs, their targets and the nature and number of drug targets. Nat. Rev. Drug Discov. 5:821. Jain R, Ernst JT, Kutzki O, Park HS, Hamilton AD, 2004. Protein recognition using synthetic surface-targeted agents. Mol. Divers. 8:89. Janin J, 1999. Wet and dry interfaces: the role of solvent in protein–protein and protein–DNA recognition. Structure 7:R277. Janin J, Miller S, Chothia C, 1988. Surface, subunit interfaces and interior of oligomeric proteins. J. Mol. Biol. 204:155. Jencks WP, 1975. Binding energy, specificity, and enzymic catalysis: the circe effect. Adv. Enzymol. Relat. Areas Mol. Biol. 43:219. Jiang S, Tovchigrechko A, Vakser IA, 2003. The role of geometric complementarity in secondary structure packing: a systematic docking study. Protein Sci. 12:1646. Johnson-Leger C, Power CA, Shomade G, Shaw JP, Proudfoot AE, 2006. Protein therapeutics– lessons learned and a view of the future. Expert Opin. Biol. Ther. 6:1. Jones S, Thornton JM, 1996. Principles of protein–protein interactions. Proc. Natl. Acad. Sci. USA 93:13. Jordan MA, 2002. Mechanism of action of antitumor drugs that interact with microtubules and tubulin. Curr. Med. Chem. Anticancer Agents 2:1. Keller TH, Pichota A, Yin Z, 2006. A practical view of ‘druggability.’ Curr. Opin. Chem. Biol. 10:357. Keskin O, Ma B, Nussinov R, 2005. Hot regions in protein–protein interactions: the organization and contribution of structurally conserved hot spot residues. J. Mol. Biol. 345:1281. KortemmeT, Kim DE, Baker D, 2004. Computational alanine scanning of protein–protein interfaces. Sci. STKE pl2. Lipinski CA, 2000. Drug-like properties and the causes of poor solubility and poor permeability. J. Pharmacol. Toxicol. Methods 44:235. Lo Conte L, Chothia C, Janin J, 1999. The atomic structure of protein–protein recognition sites. J. Mol. Biol. 285:2177. Macarron R, 2006. Critical review of the role of HTS in drug. Drug Discov. Today 11:277. Marshall SA, Lazar GA, Chirino AJ, Desjarlais JR, 2003. Rational design and engineering of therapeutic proteins. Drug Discov. Today 8:212.

REFERENCES

231

McGovern SL, Helfand BT, Feng B, Shoichet BK, 2003. A specific mechanism of nonspecific inhibition. J. Med. Chem. 46:4265. Milroy LG, Zinzalla G, Prencipe G, Michel P, Ley SV, Gunaratnam M, Beltran M, Neidle S, 2007. Chemical variation of natural-product-like scaffolds: design, synthesis, and biological activity of fused bicyclic acetal derivatives. Angew. Chem. Int. Ed. Engl. 46:2493. Mintz S, Shulman-Peleg A, Wolfson HJ, Nussinov R, 2005. Generation and analysis of a protein–protein interface data set with similar chemical and spatial patterns of interactions. Proteins 61:6. Moriera, IS, Fernandez PA, Ramos MJ, 2006. Computational alanine scanning mutagenesis— an improved methodological approach. J. Comput. Chem. 28:644. Moriera IS, Fernandez PA, Ramos MJ, 2007. Unraveling hot spots: a comprehensive computational mutagenesis study. Theor. Chem. Acc. 117:99. Mousa SA, Goncharuk O, Miller D, 2007. Recent advances of TNF-alpha antagonists in rheumatoid arthritis and chronic heart failure. Expert Opin. Biol. Ther. 7:617. Moza B, Buonpane RA, Zhu P, Herfst CA, Rahman AK, McCormick JK, Kranz DM, Sundberg EJ, 2006. Long-range cooperative binding effects in a T-cell receptor variable domain. Proc. Natl. Acad. Sci. USA 103:9867. Nayal M, Honig B, 2006. On the nature of cavities on protein surfaces: application to the identification of drug-binding sites. Proteins 63:892. Newman DJ, Cragg GM, 2007. Natural products as sources of new drugs over the last 25 years. J. Nat. Prod. 70:461. Niederhauser O, Mangold M, Schubenel R, Kusznir EA, Schmidt D, Hertel C, 2000. NGF ligand alters NGF signaling via p75(NTR) and TrkA. J. Neurosci. Res. 61:263. Nooren IM, Thornton JM, 2003. Structural characterisation and functional significance of transient protein–protein interactions. J. Mol. Biol. 325:991. Ofran Y, Rost B, 2003. Analysing six types of protein–protein interfaces. J. Mol. Biol. 325: 377. Ohtsuka M, Konno F, Honda H, Oikawa T, Ishikawa M, Iwase N, Isomae K, Ishii F, Hemmi H, Sato S, 2002. Ppa250 [3-(2,4-difluorophenyl)-6-[2-[4-(1H-imidazol-1-ylmethyl) phenoxy] ethoxy]-2-phenylpyridine], a novel orally effective inhibitor of the dimerization of inducible nitric-oxide synthase, exhibits an anti-inflammatory effect in animal models of chronic arthritis. J. Pharmacol. Exp. Ther. 303:52. Oltersdorf T, Elmore SW, Shoemaker AR, Armstrong RC, Augeri DJ, Belli BA, Bruncko M, Deckwerth TL, Dinges J, Hajduk PJ, Joseph MK, Kitada S, Korsmeyer SJ, Kunzer AR, Letai A, Li C, Mitten MJ, Nettesheim DG, Ng S, Nimmer PM, O’Connor JM, Oleksijew A, Petros AM, Reed JC, Shen W, Tahir SK, Thompson CB, Tomaselli KJ, Wang B, Wendt MD, Zhang H, Fesik SW, Rosenberg SH, 2005. An inhibitor of bcl-2 family proteins induces regression of solid tumours. Nature 435:677. Ottiger M, Zerbe O, Guntert P, Wuthrich K, 1997. The NMR solution conformation of unligated human cyclophilin A. J. Mol. Biol. 272:64. Palladino MA, Bahjat FR, Theodorakis EA, Moldawer LL, 2003. Anti-TNF-alpha therapies: the next generation. Nat. Rev. Drug Discov. 2:736. Pawson T, Nash P, 2000. Protein–protein interactions define specificity in signal transduction. Genes Dev. 14:1027. Pawson T, Nash P, 2003. Assembly of cell regulatory systems through protein interaction domains. Science 300:445.

232

SMALL-MOLECULE INHIBITORS OF PROTEIN–PROTEIN INTERACTIONS

Pawson T, Gish GD, Nash P, 2001. SH2 domains, interaction modules and cellular wiring. Trends Cell Biol. 11:504. Reayi A, Arya P, 2005. Natural product-like chemical space: search for chemical dissectors of macromolecular interactions. Curr. Opin. Chem. Biol. 9:240. Rees DC, Congreve M, Murray CW, Carr R, 2004. Fragment-based lead discovery. Nat. Rev. Drug Discov. 3:660. Reichmann D, Rahat O, Albeck S, Meged R, Dym O, Schreiber G, 2005. The modular architecture of protein–protein binding interfaces. Proc. Natl. Acad. Sci. USA 102:57. Rodier F, Bahadur RP, Chakrabarti P, Janin J, 2005. Hydration of protein–protein interfaces. Proteins 60:36. Sandborn WJ, 2007. Clinical perspectives in Crohn’s disease. Moving forward with anti-TNF-alpha therapy: current needs and future treatments. Rev. Gastroenterol. Disord. 7:S2 – S23. Sattler M, Liang H, Nettesheim D, Meadows RP, Harlan JE, Eberstadt M, Yoon HS, Shuker SB, Chang BS, Minn AJ, Thompson CB, Fesik SW, 1997. Structure of Bcl-xL-Bak peptide complex: recognition between regulators of apoptosis. Science 275:983. Schreiber SL, 2000. Target-oriented and diversity-oriented organic synthesis in drug discovery. Science 287:1964. Schreiber SL, Crabtree GR, 1992. The mechanism of action of cyclosporin A and FK506. Immunol. Today 13:136. Schuffenhauer A, Ruedisser S, Marzinzik AL, Jahnke W, Blommers M, Selzer P, Jacoby E, 2005. Library design for fragment based screening. Curr. Top. Med. Chem. 5:751. Schuffenhauer A, Brown N, Selzer P, Ertl P, Jacoby E, 2006. Relationships between molecular complexity, biological activity, and structural diversity. J. Chem. Inf. Model 46:525. Scott JK, Huang SF, Gangadhar BP, Samoriski GM, Clapp P, Gross RA, Taussig R, Smrcka AV, 2001. Evidence that a protein–protein interaction ‘hot spot’ on heterotrimeric G protein betagamma subunits is used for recognition of a subclass of effectors. EMBO J. 20:767. Shimoaka M, Springer TA, 2003. Therapeutic antagonists and conformational regulation of integrin function. Nat. Rev. Drug Discov. 2:703. Shuker SB, Hajduk PJ, Meadows RP, Fesik SW, 1996. Discovering high-affinity ligands for proteins: SAR by NMR. Science 274:1531. Tilley J, Chen L, Fry D, Emerson S, Powers G, Biondi D, Varnell T, Trilles R, Guthrie R, Mennona F, Kaplan G, LeMahieu R, Carson M, Han R-J, Liu C-M, Palermo R, Ju G, 1997. Identification of a small molecule inhibitor of the IL-2/IL-2Ralpha receptor interaction which binds to IL-2. JACS 119: 7589. Toogood PL, 2002. Inhibition of protein–protein association by small molecules: approaches and progress. J. Med. Chem. 45:1543. Tsai CJ, Lin SL, Wolfson HJ, Nussinov R, 1997. Studies of protein–protein interfaces: a statistical analysis of the hydrophobic effect. Protein Sci 6:53. Tyndall JD, Fairlie DP, 1999. Conformational homogeneity in molecular recognition by proteolytic enzymes. J. Mol. Recognit. 12:363. Uvebrant K, da Graca Thrige D, Rosen A, Akesson M, Berg H, Walse B, Bjork P, 2007. Discovery of selective small-molecule CD80 inhibitors. J. Biomol. Screen 12:464. Vakser IA, Aflalo C, 1994. Hydrophobic docking: a proposed enhancement to molecular recognition techniques. Proteins 20:320.

REFERENCES

233

Veber DF, Johnson SR, Cheng HY, Smith BR, Ward KW, Kopple KD, 2002. Molecular properties that influence the oral bioavailability of drug candidates. J. Med. Chem. 45:2615. Waal ND, Yang W, Oslob JD, Arkin MR, Hyde J, Lu W, McDowell RS, Yu CH, Raimundo BC, 2005. Identification of nonpeptidic small-molecule inhibitors of interleukin-2. Bioorg. Med. Chem. Lett. 15:983. Weisel M, Proschak E, Schneider G, 2007. Pocketpicker: analysis of ligand binding-sites with shape descriptors. Chem. Cent. J. 1:7. Wells JA, 1996. Binding in the growth hormone receptor complex. Proc. Natl. Acad. Sci. USA 93:1. Whitty A, Kumaravel G, 2006. Between a rock and a hard place? Nat. Chem. Biol. 2:112. Yan C, Wu F, Jernigan RL, Dobbs D, Honavar V, 2008. Characterization of protein–protein interfaces. Protein J. 27:59. Young L, Jernigan RL, Covell DG, 1994. A role for surface hydrophobicity in protein–protein recognition. Protein Sci. 3:717.

8 TRANSPORTERS ANNE HERSEY, FRANK E. BLANEY,

8.1

AND

SANDEEP MODI

INTRODUCTION

Transporter proteins are a group of implicit membrane-bound proteins found in all living organisms. Despite the fact that 7TM receptors are usually thought of as the largest target class within the ‘‘druggable genome,’’ a recent analysis in fact shows that there are over twice as many transporters as G-protein-coupled receptors (GPCRs) (see Fig. 8.1). It is perhaps surprising therefore that relatively few drugs target these; however, it should be noted that two of the largest marketed drug classes, the selective serotonin reuptake inhibitors (SSRIs) and the gastric Hþ/Kþ-ATPase inhibitors act on transporter targets. As their name implies, transporters control the movement of a bewildering variety of substances across membranes in both directions. Thus, while essential nutrients are passed into the cell, potentially toxic xenobiotic compounds are removed. The downside to this biological process is that many useful drug molecules are also removed. In bacterial cells, this is a major mechanism of drug resistance because many highly active antibacterial compounds are removed by transporters known as efflux pumps before they ever have a chance to act. Metabolite levels are controlled by active transport efflux mechanisms and macromolecules synthesized within the cell such as proteins and polynucleotides are also passed from the cell via special transporters. One class of transporters, the ion channels, regulates cytoplasmic ion concentration. Yet another group of these proteins is responsible for the control of synaptic levels of neurotransmitters through special release and reuptake transporters. Thus, they are responsible for a great many aspects of biochemical control throughout the body.

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

235

236

TRANSPORTERS

FIGURE 8.1 Pi-chart showing the various ‘‘druggable’’ target classes. Transporter proteins when combined with ion channels are by far the largest class.

Transporters vary greatly in size and are frequently multidomain. They often also function as homo- or heterooligomers. Those that carry a single molecule across the membrane are known as uniporters. However, many transporters involve the passage of two or more species simultaneously. Often one of these will be an inorganic ion. If the passage of the substrates or ions is in the same direction, they are called symporters whereas if they move in opposite directions, they are known as antiporters. Molecules interacting with transporters can be classified as either substrates or inhibitors. A substrate is a molecule that is recognized by the protein and is actively transported across the membrane. Because the movement of substrates through a transporter can involve large conformational changes, the protein exists in multiple states. The closed/open states of ion channels are well known, but the changes that occur in other transporters are less certain. Inhibitors, on the other hand, may bind tightly to the transporter but they are not actively transported. Instead, they frequently block the passage of substrates by binding to one or more of these conformational states. It should be remembered that it is often necessary in inhibitor design to stabilize or favor only one of these conformational states. A large number of transporters are involved in disease states and hence are potential therapeutic targets. Some of these will be discussed in further detail subsequently. However, a great number of transporter proteins are also responsible for adverse effects involving the ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of drug molecules. Important examples include the organic anion transporters (OAT) and P-glycoprotein (Pgp). As mentioned earlier, they are also responsible for the failure of several otherwise highly effective anti-infective compounds. There is therefore scope for coadministration of such drugs with a selective efflux pump blocker.

237

INTRODUCTION

Unlike the 7TM receptors that all adopt a broadly similar folding pattern, there is a huge variation in the size, functional mechanism, and topology of transporters. Accordingly, it is appropriate to discuss some aspects of their classification. It has been recognized for some time that transporters can be classified as either active or passive and as such can be further divided into various families. Active transporters use a chemical reaction to create an ion or solute gradient, which facilitates movement of the solute across the membrane. They can be further classified as either primary or secondary. One of the earlier recognized families is the primary active ATP binding cassette (ABC) transporters (Borst and Elferink, 2002). The passive or facilitated transporters on the other hand move their solutes via an implicit electrochemical gradient. The passive transporters form part of a major class of proteins known as the SoLute Carrier (SLC) transporters, which was devised by the Humane Genome Organization (HUGO) Nomenclature Committee. It contains some 47 families and includes not only the passive transporters but also the ioncoupled transporters and exchangers. Some SLC members are important pharmaceutical targets, for example, inhibition of some of the SLC6 neurotransmitter reuptake proteins is an important treatment for depression. Hediger et al. (2004) discussed the classification of SLC transporters in a special dedicated issue of Pflugers Arch. Saier and coworkers have used a combination of phylogenetic analysis and the varying functionality of transporters to derive their system of transporter classification (Busch and Saier, 2002). The full system is available at www.tcdb.org. This site has a wealth of information on all aspects of transporters, including sequences, available structures, associated disease states, and various useful software programs. The system is similar to the enzyme classification system in that the proteins are initially grouped into classes and subclasses. Figure 8.2 shows that there are 7

Transporters 4. Group 1. Channels 2. Electrochemical 3. Primary Active Translocators and Pores Potential-driven Transporters 4.A PhosphoTransporters

5. Transmembrane Electron Carriers

1.A α-Helical protein channels 1.B β-Barrel protein porins 1.C Toxin channels 1.D Nonribosomally synthesized channels 1.E Holins 1.F Vesicle Fusion proteins

5.A Two - electron transfer carriers 5.B One - electron transfer carriers

2.A Protein porters 2.B Nonribosomally symthesized porters 2.C Ion gradientdriven energizers

FIGURE 8.2

transfer - driven systems 3.A P-P bond 4.B Nicotinamide hydrolysis- driven ribonucleoside systems uptake 3.B Decarboxylpermeases action-driven systems 8. Auxillary 3.C Methyltransfer- Transport driven Proteins systems 8.A Auxillary 3.D transport Oxidoreductionproteins driven 8.B Protein/ systems peptide toxins 3.E Light that target absorptionion channels driven systems of class 1.C

9. Others 9.A Transporters of unknown classfication 9.B Putative uncharacterized transporters 9.C Functionally characterized transportes of unknown sequence

The classification of transporters according to Saier.

238

TRANSPORTERS

major classes that are further divided into 23 subclasses. It should be noted here that Class 1 contains all the well-known ion channels, although these are normally thought of as a separate genomic target class and have been discussed in another chapter of this book. Most transporters require a source of energy to function, which is used in the scheme as a means of classification. Thus, Class 2 transporters rely on some form of electrochemical potential, such as an ion gradient, to move their substrates, whereas Class 3 proteins rely on a particular chemical reaction. The most commonly used reaction is the hydrolysis of ATP, although other reaction types such as decarboxylation or even photochemical transformations are known. Not all transporters are proteins; subclass 2.B, the nonribosomally synthesized porters, consist of either depsipeptides or nonpeptide organic substances. They function by complexing cations in the cytoplasm and then moving across the membrane where the complex presents a hydrophobic exterior. An example is the valinomycin carrier family, an antibiotic from Streptomyces fulrissimus. This contains a triple repeat of the sequence, D-valine-L-lactate-L-valine-D-hydroxyisovalerate, which complexes with Kþ ions and controls their concentration in the cell. The 23 subclasses are further subdivided into families and subfamilies. Currently, there are in excess of 550 families in the classification system. Some of these can be thought of as superfamilies. The voltage-gated ion channel (family 1.A.1), for example, has 91 subfamilies at the time of writing, whereas the major facilitator superfamily (MFS) (family 2.A.1) and the ABC superfamily (family 3.A.1) have over 280 subfamilies each. In the Saier classification, the MFS is largely synonymous with the SLC transporters. They usually consist of 12, 14, or 24 transmembrane helices and individual members can act either as uniporters, symporters, or antiporters. The mechanism of transporters is often written in terms of equilibrium equations. Thus, the MFS proteins can either be Uniport: SðoutÞ $ SðinÞ; or Symport: SðoutÞ þ ½Hþ or Naþ
ðoutÞ $ SðinÞ þ ½Hþ or Naþ
ðinÞ; or Antiport: S1 ðoutÞ þ S2 ðinÞ $ S1 ðinÞ þ S2 ðoutÞðS1 may be Hþ or a soluteÞ: A recent paper by Ren et al. (2007) highlights yet another transporter database, TransportDB available at www.membranetransport.org. Although it dates back to 1966, the development of the current version started in 2002. It uses structured query language (SQL) to access the data and is regularly updated, especially with data from newly sequenced genomes. Another useful feature of this system is that it contains data on transporter family substrates. In the latest version, data are available on all the transporter proteins from 296 organisms.

METHODOLOGIES IN TRANSPORTER DRUG DESIGN

8.2

239

METHODOLOGIES IN TRANSPORTER DRUG DESIGN

Rational drug design can be subdivided into two main categories, namely, structure-based design and ligand-based design. The former is dependent on the availability of a three-dimensional (3D) structure of the target protein (the transporter in this case), whereas the latter is based on calculated or experimental information about the drugs or ligands themselves and assumes no knowledge of the target. 8.2.1

Structure-Based Methods

For transporters, X-ray crystallography is especially difficult because implicit membrane-bound proteins are inherently insoluble and hence very difficult to crystallize. Despite this challenge, over the last decade the crystal structures of several transporter proteins or their subdomains have started to appear. The pioneering work of MacKinnon and his coworkers has led to a number of voltage-gated channel structures (Doyle et al., 1998; Jiang et al., 2002, 2003; Long et al., 2005). Several other channel and porin structures have also emerged for Class 1 transporters. Of more immediate importance was the appearance in 2003 of two major facilitator superfamily proteins, the glycerol triphosphate transporter (GlpT) (Huang et al., 2003) and lactose permease (LacY) (Abramson et al., 2003). Despite a total lack of sequence identity, both the proteins showed a similar transmembrane (TM) topology consisting of 12 TM helices. An insight into the important Naþ/Cl dependent SLC6 (2.A.22) family came from the publication of a bacterial leucine transporter homologue (Yamashita et al., 2005), which led to the generation of homology models of several neurotransmitter reuptake proteins involved in central nervous system (CNS) disorders. Major efforts have also been made on the determination of several (mainly bacterial) members of the ATP-dependent families. A listing of transporter structures with references can be found at www.tcdb.org, although it has not been updated. Alternatively, the structural information available from transportDB is regularly updated and has the added advantage of having links to viewable images. Further details of available structures can be found in the Protein DataBank (www.rcsb.org) by searching for the keyword ‘‘transporter’’. Homology modeling is a very well-established technique and has been described in detail by Leach (2000). When applying this technique, the most important step is the initial sequence alignment because errors arising here will be propagated throughout the process. For example, it may happen that a gap occurs in the alignment of the unknown sequence in a helical region of the known structure. If it is desirable to maintain the helix, then this will give rise to an approximate 100 shift in the orientation of the residues following this region. Some examples of transporter homology models will be discussed subsequently. It is well known that within protein structures, sequences of little or no identity can adopt the same fold. The ability to recognize or predict such occurrences would be a powerful tool in model building. This is what protein threading sets out to do, but the technique has been mainly applied to globular proteins (Bowie et al., 1991).

240

TRANSPORTERS

Simpler techniques such as hydropathy plots however are often capable of predicting the topology of membrane-bound proteins. One important set of tools attempts to find ‘‘distant’’ sequence relationships between proteins. The ‘‘hidden Markov model’’ (HMM) algorithm is one such approach that has been used successfully in the prediction of implicit membrane proteins. For example, Ho¨glund et al. (2005) used hidden Markov models to predict in excess of 400 novel SLC6 transporters from various animal, plant, and bacterial sources. Another widely used distant homology prediction program is PsiBlast (Altschul et al., 1997). We have used this program to show a relationship between the organic anion transport proteins (OATPs) and the known structure of glycerol triphosphate transporter. These distant relationships can be used to generate sequence alignments, which can in turn act as inputs for homology model building. Even in the absence of any homology it is possible to generate 3D models using a variety of de novo folding tools. Many of these were developed in the 1990s to aid in the generation of 7TM receptor models, prior to the publication of the first crystal structure of rhodopsin. It is not necessary to discuss the theoretical details behind these tools here as they have been reviewed recently (Barton et al., 2006; Blaney, 2007). Essentially, as most transporter proteins cross a membrane as a set of alphahelices, the problem becomes one of predicting the folding of this helical TM bundle. This challenge reduces to (a) predicting the residues that form each helix, (b) predicting those residues in each helix that face the core of the bundle and those that interact with the lipid bilayer, and (c) predicting the interhelical tilts of the bundle. The structural prediction of interhelical loops or domains is a separate issue. The techniques that have been used include the following:  Hydrophobicity (and other) profiles  Multiple sequence alignments for identification of conserved regions and key residues  Fourier transform methods for predicting properties of a-helical transmembrane regions  Secondary structure prediction algorithms  Molecular mechanics energy-based tools (MM/MC/MD) and distance geometry algorithms for regional refinements  Experimental data from site-directed mutagenesis (SDM), NMR, and biophysical probes The membrane lipid bilayer consists of charged head groups on either side with long hydrophobic aliphatic chains in between. The residues of a protein, which interact with this bilayer, are therefore also hydrophobic. By assigning a value of hydrophobicity (taken from one of the well-established scales such as Kyte–Dolittle) (Kyte and Doolittle, 1982) to each residue and using a sliding window, typically 19 residues wide, one can calculate the average hydropathy along the length of the sequence. The positive peaks in these plots greater than 20 residues can generally be assumed to be transmembrane helices and thus hydropathy plots are a valuable tool in the

METHODOLOGIES IN TRANSPORTER DRUG DESIGN

241

FIGURE 8.3 Kyte–Dolittle-based hydropathy plots of some transporter sequences showing the variation in the number of transmembrane helices. (a) Serotonin reuptake transporter. (b) P-glycoprotein. (c) Kþ/Hþ-antiporter (family 2.A.63). (d) zinc uptake transporter (family 2.A.5).

definition of the protein topology. Typical plots for a few transporters are shown in Fig. 8.3. Often it is found within a family of membrane proteins that there is a greater degree of conservation in the membrane-spanning regions than in the interceding loops. Also, if these regions are a-helical, then the residues facing the bilayer are generally hydrophobic though not necessarily conserved, whereas some residues in the core of the bundle may have a common functional role and will be highly conserved. Thus, multiple sequence alignments are useful in understanding functions of a membrane protein. The inherent periodicity of an a-helix also means that one can apply a Fourier transform to define a moment of the property assigned to the residues in that helix. Thus, if one uses the residue hydropathy values discussed above, a hydrophobic moment can be defined for a helix that will show the face oriented toward the lipid bilayer. Researchers used hydrophobic moments as part of a helix packing strategy in the construction of an early dopamine transporter model (Edvardsen and Dahl, 1994). Conservation values have also been applied to membrane helical sequences to define conservation moments, which would be representative of the inward-facing side of a transmembrane helix (Donnelly et al., 1993; Blaney and Tennant, 1996). In our modeling of the OATP1B3 transporter, conservation moment calculations showed that an arginine conserved in the OATP family should be inward facing and this was used as a guide to the placement of this helix in the final model, as shown in Fig. 8.4. Secondary structure is an important technique in the de novo modeling of proteins. Given that one is often interested in the transmembrane region of transporters which is generally (though not always) alpha-helical, secondary structure prediction is somewhat superfluous. However, it can be useful in loop regions or in nonmembrane-spanning

242

TRANSPORTERS

METHODOLOGIES IN TRANSPORTER DRUG DESIGN

243

domains. The best methods usually involve multiple sequence alignments and genetic algorithms, but even these are at best only 65–70% accurate (Rost and Sander, 1993). Energy-based methods form an essential aspect of all protein molecular modeling and have been around for several decades. Ideally, the energy (and all other physical properties) of a molecular system should be calculated from first principles, by considering the electronic structure of the system using quantum mechanics (QM). The computational costs of such calculations are so prohibitive that only the small molecule ligands themselves can be treated. However, electrostatic properties derived from QM calculations are important in protein studies, and therefore approximations, for example, Poisson–Boltzmann algorithms, have been developed for use in larger molecular systems. As an example, Poisson–Boltzmann electrostatic potential calculations were used to guide manual packing of the TM helices in an early SERT model (Ravna and Edvardsen, 2001). For protein energetics and geometry considerations, approximate methods, known as molecular mechanics, split a molecular system into a summation of bonded and nonbonded terms. The former consists of bond stretching, angle bending, and torsional rotation terms, which are formulated as simple harmonic or anharmonic functions. Coulombic and attractive/repulsive van der Waal interactions make up the nonbonded terms. Each atom of the system is assigned a particular ‘‘atom type’’ and a set of parameters, known collectively as a force field, is used to calculate the total energy of the system. Several well-known force fields such as CHARMM (Brooks et al., 1983) and AMBER (Cornell et al., 1995) have been developed and are widely available as academic or commercial versions. Further details on molecular mechanics theory and applications can be found in any standard computational chemistry textbook such as Leach (Leach, 2000). There are several techniques that make use of molecular mechanics. Proteins are generally considered to be in low energy conformations and a common stage in model building is to perform some energy minimization on the initial model as a means of refinement. Energy minimization is also often used in ligand docking to calculate interaction energies between the ligand and the protein. Techniques such

3 FIGURE 8.4 Stages in the construction of a model of human OATP1B3. (a) An hydropathy plot of the human OATP1B3 sequence showing the predicted 12 transmembrane spanning helices. PsiBlast calculations showed a distant homology between this and the glycerol triphosphate transporter. Fourier analysis was applied to each helix to predict the inwardfacing residues. (b) A multiple sequence alignment for TM helix 11. Conservation and hydrophobic values were assigned to each residue and a Fourier transform was applied to the alignment. (c)The result is a helical wheel with each spoke representing a particular column in the multiple alignment. The spokes are color-coded by residue type. At the bottom left, a hydrophobic moment (cyan) is displayed while at the upper right corner, the conservation moment is shown at the position of the conserved arginine. This arginine is therefore oriented toward the interior of the bundle. (d) The final model showing the arginine in the interior of the bundle. A typical ligand has been docked into this showing potential interactions with the arginine and a number of phenylalanines. (See the color version of this figure in the Color Plates section.)

244

TRANSPORTERS

as molecular dynamics and Monte Carlo simulations can be used to study the motion of proteins. This is especially useful in the sampling and refinement of loop conformations. Again, the reader should refer to some standard textbooks on the subject for further details. By combining the above methods, it is often possible to build a protein model in the absence of any significant homology. Such models however should be treated with a good deal of skepticism and require extensive validation from other experimental data. Experimental data can come in many forms. For example, ligand structure–activity relationship (SAR) is often available and any good model should be able to explain the data found from biological assays. It is often possible to provide further validation of binding hypotheses using SDM experiments. Here single (or multiple) residues, which may be implicated in some binding interaction with a ligand, are mutated to another residue, which would change that interaction. If the expected change in binding or function is observed experimentally with the mutant, this will provide some evidence that the hypothesis, and hence the model, is correct. Details of known SDM results were used to build a model of the serotonin reuptake transporter (Ravna and Edvardsen, 2001). They used the cocaine molecule to initially place 5 TM helices containing residues of known importance from SDM results. The remaining helices were added and the whole model refined by manual adjustment based on electrostatic considerations, followed by energy minimization. A particularly elegant example of the use of experimental data is that of Kaback and coworkers who, for over 20 years, have worked on the lactose permease transporter from Escherichia coli. Although the crystal structure was finally solved in 2003 (Abramson et al., 2003), prior to that they had used a wide variety of biophysical probes to gain information on interresidue placements, which were then used to construct a TM helix packing 3D model. Many of these probes were based on the systematic mutation of each residue to cysteine. The thiol group can then be crosslinked to a variety of sulfhydryl reagents containing spin labels, fluorescent groups, metal binding sites, and charged or hydrophobic probes (Frillingos et al., 1998; Kaback and Wu, 1999). The effects of the changes compared with the original protein are then studied by spectroscopic or other chemical methods. NMR has become increasingly important in the study of transporter–ligand interactions (Watts, 1999). Using rotational resonance NMR, Middleton et al. (1997) were the first to determine the conformation of a bound inhibitor, TMPIP, to its in situ native state target Hþ/Kþ-ATPase, an important transporter in the treatment of gastric ulcers. More recently, the same group constructed an homology model of this transporter, based on the crystal structure of the related rabbit Ca2þ-ATPase (Toyoshima et al., 2000; Toyoshima and Nomura, 2002), and used docking studies to show that the docked conformations of analogues were in keeping with the NMR-determined structures (Kim et al., 2005). While three-dimensional structure-based design is more routinely used in the field of therapeutic transporter targets, studies with liability transporters have fallen more in the area of ligand-based methods. More recently however, 3D models are starting to emerge with some key liability targets. Some noteworthy examples include the rat organic cation transporter model (rOCT1), based on LacY (Popp

METHODOLOGIES IN TRANSPORTER DRUG DESIGN

245

et al., 2005), and the glucose transporter GLUT1, based on GlpT (Salas-Burgos et al., 2004). Another example is the OATP family where, using the PsiBlast program, we were able to show a distant relationship between these and the GlpT. A similar relationship was described between OAT1 and GlpT (Perry et al., 2006), while different searching algorithms were used in developing models of the OATPs based on both GlpT and LacY (Meier-Abt et al., 2006). Interestingly, their alignments were slightly different from our own. Using this information in conjunction with hydropathy and a Fourier-based helical analysis, we were able to construct models of the OATP transporters. So far, docking of compounds has been in broad agreement with experimental observations (see Fig. 8.4d). Another important target is Pgp, which is involved in solute transport in several parts of the body including the blood–brain barrier (BBB). Good sequence identity has been observed between Pgp and several ATPases, the most notable being MsbA. Homology modeling is certainly possible, but unfortunately it was discovered that there was an error in the software used to solve the structure (Chang et al., 2006). At the time of writing the structures had been withdrawn from the protein Databank, but they have subsequently been redeposited (Ward et al., 2007). 8.2.2

Ligand-Based Methods

Knowledge of the molecular structures and properties of transporter ligands are used in pharmacophore or quantitative structure–activity relationship (QSAR) analyses to determine the properties of molecules most likely to interact with a particular transporter. All methods have their own pros and cons. Structure-based design is probably most appropriately used for designing compounds that bind to therapeutic transporter targets, whereas QSAR analyses, particularly on large global data sets, are best used for designing out transporter liabilities. Pharmacophore models on the other hand have been shown to have application in modeling both types of transporters. When the 3D structure of the transporter is not known, it is possible to gain insight into the steric and electronic structural features necessary for affinity to the transporter by pharmacophore modeling. The pharmacophore is a 3D spatial representation of the key features that confer activity such as presence of hydrogen bond donors or acceptor groups, charged groups, and areas of hydrophobicity. Molecules can then be assessed in terms of how well they fit the pharmacophore features. The use of these methods to model transporters has been well summarized (Ekins and Swaan, 2004; Chang et al., 2005; Chang and Swaan, 2006). They can be used to gain insight into the molecular features important for binding and their relative distribution in the molecule, but they are very general in nature. The properties related to hydrophobic groups and hydrogen bond donor and/or acceptor groups are, in principle, better able to distinguish specific requirements for binding, such as the shape, polarity, and specific hydrogen bonding regions of the molecule (Osterberg and Norinder, 2000; Yates et al., 2003; Crivori et al., 2006), than the equivalent 2D descriptors. However, it is likely that many molecules will fit the pharmacophore but will not be a substrate or inhibitor for the

246

TRANSPORTERS

transporter. These methodologies also have the disadvantage that the computational time to calculate the properties of the molecules is much longer than that for simpler descriptors. The principle of QSARs is to find a relationship between the biological data and the molecular descriptors using a statistical method. It is a commonly used methodology for identifying relationships for developability properties of molecules and once the relationship has been established, it can be used on new compounds to identify those molecules that have good or poor developability properties (Clark, 2001). The statistical techniques used to model the QSARs for transporter interactions with molecules have been largely confined to partial least squares (PLS) and partial least squares discriminant analysis (PLS-DA) (Wold et al., 1993). The PLS methodology has advantages over other modeling techniques; when compared with multiple linear regression (MLR), it is not necessary to reduce the descriptor set by removing correlated variables prior to the analysis. For 3D techniques such as comparative molecular field analysis (CoMFA), which generate a lot of descriptors, this methodology is particularly appropriate. Also in common with MLR but unlike techniques such as neural networks (NN), it is possible to understand the relationships with the original descriptors both in terms of its magnitude and direction. This is of particular importance when trying to use an existing relationship to design new molecules that do or do not interact with a particular transporter. The statistical methods commonly deployed in QSAR analyses have been described by Hill and Lewicki (2005). Having identified a QSAR that seems to be describing the relationship between the structural features in molecules and the strength of their interaction with a transporter, it is important to validate the model on new compounds not utilized in the original model building process. One can then assess the strengths and weaknesses of the model and understand the predictive power of the model as it is applied to molecules with unknown activity. This test is important no matter whether it is a global model, simple property rule, or a specific model on a local data set. An in silico QSAR model in a drug discovery environment is usually used to make predictions on virtual or unsynthesized molecules and then, on the basis of the predictions, a chemist can decide which molecules to go ahead and make. In addition to the validation performed by the model developer, it is usually advisable for a chemistry team wanting to use a model to make predictions on a new series of compounds to validate the model with compounds from the specific chemotype of interest to them (Hersey et al., 2004). To illustrate how a model might be used, let us take, as an example, a model that predicts whether a compound is a substrate for Pgp as determined in an in vitro substrate screen (Hersey et al., 2004). For a classification model, it is usual to look at the matrix of predicted versus observed substrates, sometimes called the confusion matrix. This is shown in Table 8.1. The data as in Table 8.1 can be assessed in two ways. First, by considering if a prediction is made, how likely is it for that prediction to be correct and, second, by assessing how many of the true (experimental) substrates or nonsubstrates the model is able to identify. In Table 8.1, the percentages correctly predicted are shown in terms

247

METHODOLOGIES IN TRANSPORTER DRUG DESIGN

TABLE 8.1 The Relationship Between the Percentages of Compounds Predicted as Pgp Substrates and Nonsubstrates When Considered as Percentages of the Observations and Predictions Number of Observations Observed Substrate Observed Non-substrate % of Predictions Predicted Substrate Predicted Non-substrate % of Observations Observed Substrate Observed Non-substrate

Predicted Substrate 55 21 Observed Substrate 72.4 16.0 Predicted Substrate 78.6 21.0

Predicted Nonsubstrate 15 79 Observed Nonsubstrate 27.6 84.0 Predicted Nonsubstrate 21.4 79

of predictions and observations. In this example, if a prediction is made that a molecule is a substrate, the prediction is likely to be incorrect 28% of the time but only 16% of the time if the prediction is that the molecule is a nonsubstrate. The chemists using the model might decide not to make all those compounds predicted as substrates. In doing so, they could have (on average) failed to synthesize 28 in every 100 compounds that would have been nonsubstrates experimentally. Likewise, by synthesizing all those predicted as nonsubstrates and perhaps subsequently testing them in a Pgp substrate assay, the chemists would have made 16% of their compounds that still had undesirable properties. However, it must be remembered that by not using a model or any SAR knowledge, they would be likely to synthesize a much higher percentage of poor compounds. The matrix of the percentages in terms of the observations shows the percentage of the true substrates and nonsubstrates predicted by the model. In this specific case, the model is quite good at correctly identifying the known substrates and nonsubstrates (80%). However, although a model can be good at getting its predictions correct, it can miss making a prediction on a larger percentage of compounds in a particular class. In this example, this corresponds to 20% of measured substrates and nonsubstrates not predicted by the model. The simplest QSAR models involve the use of whole molecule properties such as partition coefficient (log P), ionization constant (pKa), topological polar surface area (tPSA), size (molar volume or molecular weight), and hydrogen bond donor or acceptor descriptors (e.g., the Abraham A and B descriptors), which are calculated from the two-dimensional (2D) structure of molecules (Platts et al., 1999; Ertl et al., 2000). CoMFA (Cramer et al., 1988) is a commonly used 3D QSAR method that has been applied to the modeling of many transporters. It requires the use of a molecular modeling package to obtain the 3D structures of the lowest energy conformers. The molecules are then aligned in a suitable way and the electrostatic and steric interaction fields at points surrounding the aligned molecules are generated and

248

TRANSPORTERS

used as descriptors in a QSAR model. The method is very sensitive to how the molecules are aligned and hence it is most easily used for structurally related and small compound sets. CoMSIA (comparative molecular shape indices analysis) (Klebe, 1998) is a similar method that has also been used to model transporters. It differs from CoMFA in the way that the fields are calculated. CoMFA uses a Lennard–Jones potential for steric fields and a coulombic potential for the electrostatic field, whereas CoMSIA uses Gaussian functions. VolSurf (Cruciani et al., 2000) is a computational method that calculates interaction energies between a molecule and chemical probes and has been used to model compound interactions with Pgp. The 3D information obtained is then interpreted as a set of descriptors that can be understood and interpreted in molecular terms. Volsurf differs from the other 3D QSAR techniques in that it is not necessary to superimpose the molecules on each other. The descriptors include those of molecular size and shape, regions of hydrophilicity and hydrophobicity, and regions representing hydrogen bond donors and acceptors. Simple Rules, as the name suggests, are a simple way of classifying compounds that are substrates or nonsubstrates for a particular transporter. They are property rules, which the majority of substrates or nonsubstrates obey. This approach was also used by Lipinski in his now well-established and utilized ‘‘rule of 5’’ that predicts which compounds are likely to have good oral absorption (Lipinski et al., 1997). He observed that only 5% of oral drugs failed the rules. In a similar way a ‘‘rule of fours’’ has been proposed (Didziapetris et al., 2003) for Pgp where compounds with (N þ O)  8, MW > 400, and acidic pKa > 4 are likely to be Pgp substrates whereas compounds with (N þ O)  4, MW < 400, and basic pKa < 8 are likely to be nonsubstrates. N þ O is the count of nitrogen and oxygen atoms in the molecule. Rules of this nature can be very useful for chemists to reduce Pgp liability because, for example, they need to synthesize compounds with fewer hydrogen bonds, smaller in size and more acidic (lower acidic pKa). However, they are possibly less useful when also considering the constraints imposed by designing for drug target affinity where basicity, hydrogen bonding, and higher molecular weight are often the properties conferring biological activity.

8.3 THERAPEUTIC TRANSPORTER TARGETS IN DRUG DISCOVERY 8.3.1

Vacuolar ATPases

The V-, F-, and P-ATPases are members of the 3.A.2 cation translocator transporter family. The P-ATPases include the Naþ/Kþ-ATPase, the Ca2þ-ATPase, and the Hþ/ Kþ-ATPase transporters, the latter of course being the gastric pump protein involved in acid secretion in the stomach. F-ATPases are involved in ATP synthesis. V-ATPases are structurally similar but use ATP hydrolysis to pump protons into the associated cell or organelle. Many enzymes require an acidic environment to function properly and the V-ATPase proton pumps are a major mechanism for

249

THERAPEUTIC TRANSPORTER TARGETS IN DRUG DISCOVERY

facilitating this. One particular class of these transporters is the vacuolar ATPases associated with osteoclast cells. An acidic environment is required to dissolve bone material in the microcompartments attached to the bone surface. This process primarily involves acid cysteine protease enzymes such as cathepsin K. Inhibitors of this proton pump should therefore be useful in the treatment of lytic bone disease disorders such as osteoporosis, hypercalcemia, or rheumatoid arthritis. They also have potential use in the treatment of glaucoma and periodontitus. The definition of the osteoclast proton pump as a V-ATPase was first recognized in 1989 (Blair et al., 1989). Unfortunately, V-ATPases are ubiquitous to the body and the real problem was recognized as one of deriving selectivity over other essential V-ATPases such as those found in the kidney or cardiovascular system. Two classes of macrolide antibiotics, the bafilomycins and the concanomycins (as shown in Fig. 8.5) from varying streptomyces species were found to be potent inhibitors of the V-ATPases. However, they were completely nonselective and were quickly found to be toxic as a result of their interactions with other V-ATPases. The mechanism of macrolide inhibition has been studied using homology modeling and docking. The full V-ATPase structure is a complex multidomain protein but it can be divided into two main components, the V0 transmembrane domain comprising a total of eight different subunits and the V1 cytoplasmic domain, which is a bundle of transmembrane helices forming a proteolipid ring. The two domains are connected by a ‘‘stalk’’ region, as shown in Fig. 8.6a. The mechanism of proton pumping has been reported in detail (Wang et al., 2005). ATP hydrolysis occurs in the V1 domain and this causes a rotational movement in the stalk region. Thus, the rotor movement of the stalk gives rise to interhelical rotations in the c subunits of

Me

Me

Me

O

O Me R

O

OH

Me

O

Me

Bafilomycin A1

R=H

Bafilomycin B1

R=

OH

O

CH3O

O Cl

OMe

Me O

O

OH

Me

Me

H

O

O

Me

1

N H

Cl

Me HO

Me

Me R1

O

O HO

OH

O

H Me

O

O

Me

Me

O

OH

OH

Me

Me Me O

Me

CH3O

O

Cl

O

OH

N H

R

N

N H

Cl

Me

2 SB242784

Me Concanamycin A

R = C 2 H 5 R1 = CONH2

Concanamycin B

R = CH3

R1 = CONH2

O N H O

N

N

O

O R

O O H N

N

O

H

Destruxin A

R = Vinyl

Destruxin B

R = Isopropyl

Destruxin C

R = Oxirane

CH3O

H3 C N N H

S O

OMe

N

3 Omeprazole

FIGURE 8.5 Some V-ATPase ligands.

CH3

N

OMe

N N H

S O

4 WY47766

250

TRANSPORTERS

FIGURE 8.6 (a) Schematic structure of the vacuolar V_ATPase (adapted from Sun-Wada et al. 2004). (b) Model of adjacent subunits forming the proteolipid ring of N. crassa showing residues that have been mutated and that affect bafilomycin binding. (c) Bafilomycin docked into its putative binding site. This binding prevents rotation of the helices, thus disabling the movement of the acid residues that are believed to be involved in proton movement (See the color version of this figure in the Color Plates section.)

the V0 domain. Each subunit contains a conserved acidic residue that changes from being buried to exposed during the helical rotation cycle. The protonation– deprotonation of this side chain results in the movement of hydrogen ions down the channel. A series of SDM experiments have been carried out in the c subunits of Neurospora crassa and Saccharomyces cerevisiae and several residues involved in the binding of bafilomycin have been identified (Bowman and Bowman 2002; Bowman et al., 2004). Using the crystal structure of the proteolipid ring of V-ATPase from Enterococcus hirae (Murata et al., 2005), Bowman et al. (2006) have constructed a model of the V0 domain of N. crassa. We have used a similar model to study the docking of bafilomycin and have concluded, as did Bowman, that binding occurs between the residues in TMs 1 and 2 of one c subunit and TM4 of the adjacent one. This binding essentially prevents rotation of the helices and thus blocks proton transfer (Fig. 8.6b and c).

THERAPEUTIC TRANSPORTER TARGETS IN DRUG DISCOVERY

251

Numerous attempts have been made to simplify the macrolide structures and have been reviewed in the literature (Farina and Gagliardi 1999a, 1999b). Generation of the indole enol ether analogue (Fig. 8.5, compound 1) by a group at SmithKline Beecham resulted in a compound with an IC50 value of 1.9 mM. Further modification to replace the ester with a hindered amide gave compound 2 (SB242784) with an IC50 value of 0.003 mM. This compound also showed over 1000-fold selectivity for the osteoclast V-ATPase over other related transporters. Other researchers made use of the similarities between V- and P-ATPases to modify the antiulcer drug, omeprazole. Thus, WY47766 showed weak activity with an IC50 value of 200 mM. Another series of natural products, the destruxins, have been reported to have V-ATPase activity, although they do not appear to have been developed further as osteoclast inhibitors. There is still much work that needs to be done on the development of synthetically accessible selective V-ATPases. It is hoped that with the availability of further SDM data together with the emergence of bacterial homology models of the binding site, a human equivalent will soon be available to help in future design. 8.3.2

Gastric (P-) ATPases

The involvement of gastric acid secretion in peptic ulcers has been established for over 100 years. Reduction of acid secretion was therefore the main treatment until it was discovered that the bacterium Helicobacter pylori could also be involved. Antibiotic treatments are therefore used where appropriate. Early treatment of excessive acid secretion involved the use of histamine H2 antagonists. These were very effective, although tolerance did develop in many patients. What became more apparent was that as the incidents of peptic ulcer disease declined, a new condition known as gastroesophageal reflux disease (GERD) emerged, which was not effectively controlled by H2 antagonists. This led to the emergence of the proton pump inhibitors (PPIs) as alternative treatments, with the first of these, omeprazole, being launched in 1989. The target of the PPIs was discovered to be the gastric Hþ/ Kþ-ATPase, which is a multidomain protein structurally similar to the V-ATPases. These transporters exist in two main conformational states, the E1 and E2 states, which interconvert during phosphorylation–dephosphorylation resulting in ion movement. The sulfoxide-based PPIs act as prodrugs, which under acidic conditions rearrange with loss of water to give a sulfenamide intermediate (Fig. 8.7). The thiol group of cysteine 813 then reacts with this species, resulting in covalent disulfide bond formation stabilizing the E2 conformational state. Because of the covalent bond, this class of PPI is essentially irreversible. Covalently bound inhibitors can cause a variety of problems and significant effort has been expended in the design and synthesis of reversible inhibitors. These are commonly called acid pump antagonists (APAs) and are typified by the imidazopyridine compound SCH28080 and the SmithKline compounds SKF97574 and SKF96067 (Fig. 8.8). These compounds show some advantages over the older PPIs, although as yet no candidate is on the market. As several recent reviews have discussed the past and

252

TRANSPORTERS R3

R3

R3

O R2

R4

R2

H+

N

N

N

S

N

R4

Prot-SH

+

- H2 O

O

R2

R4

N

N

S N

+

NH

S

Cys813 S

NH R1

R1

R1

FIGURE 8.7

Proton pump inhibitors (PPI).

ongoing developments in the field of PPIs and APAs, no further details on the chemistry of these will be discussed here (Sachs et al., 2007; Jain et al., 2007). The gastric Hþ/Kþ-ATPase shows some 29% sequence identity to the Ca2þATPase, the structure of which was solved in 2000 (Toyoshima et al., 2000). Several groups have used the calcium ATPase to construct homology models of the gastric Hþ/Kþ-ATPase (Kim et al., 2005; Munson et al., 2005, 2007). Kim et al. compared their homology model with the experimentally observed NMR data on TMPFPIP. Munson’s group has made extensive use of the models to explain the movement of ligands and ions during the binding process. This is certainly one field where active use is being made of homology models in structure-based design. 8.3.3

Neurotransmitter Transporters as Drug Targets

Neurotransmitters are small organic molecules responsible for the passage of signals between neurons and as such are intimately involved in all functions of CH3

CN N

NH

CH3 N

O CH3

O

N

5 SCH28080

R

6 SKF97574 R = OCH2CH2OH

N

F

+

CH3

F

7 SKF96067 R = OCH3

N

F O

8 TMPFPIP

F F

FIGURE 8.8

APA ligands associated with gastric Hþ/Kþ-ATPase.

THERAPEUTIC TRANSPORTER TARGETS IN DRUG DISCOVERY

253

the central and peripheral nervous systems. Imbalances in the levels of these compounds can lead to major disease states, especially in the brain where they are responsible for disorders such as depression, anxiety, schizophrenia, epilepsy, Parkinson’s disease, and other related conditions such as attention-deficit hyperactivity and obsessive-compulsive disorders. They are also the target of many drugs of abuse such as cocaine and ecstasy. The main neurotransmitters involved are dopamine, serotonin, and noradrenaline (norepinephrine) and the amino acids GABA, glutamate, and glycine. Synaptic levels of these compounds are strictly controlled by a number of transporters of the MFS (or SLC) superfamily. They fall into two broad classes: (a) transporters that collect cytoplasmic neurotransmitters into intracellular vesicles prior to their release into the synapse, and (b) reuptake transporters that remove neurotransmitters from the synapse. The vesicular transporters, which are members of the SLC17, SLC18, and SLC32 subfamilies, are not currently the targets of known drugs, although selective substrates or inhibitors would certainly be of interest. The synaptic reuptake transporters that comprise the SLC1 (glutamate) and SLC6 (dopamine, serotonin, norepinephrine, GABA, and glycine, commonly referred to as DAT, SERT, NET, GAT, and GlyT, respectively) subfamilies have been the targets of many CNS drugs for several decades, although no drugs currently target the glutamate transporter. The tricyclic drugs such as imipramine 9 and amitriptylline 10 were among the first antidepressants and the latter especially was quickly recognized to be an inhibitor at both the serotonin and norepinephrine reuptake sites. The undesirable side effects of these drugs together with the increased recognition of the role of serotonin in depression led to the development of the first selective serotonin reuptake inhibitors (SSRIs), zimelidine 11 and fluoxetine 12. Other important members of this drug class quickly followed, including paroxetine 13 and citalopram 14 and, more importantly, its single S-enantiomer, escitalopram, sertraline 15, and fluovoxamine 16. The structures of these compounds are shown in Fig. 8.9. The design and development of other SSRIs is the subject of a number of recent reviews (Spinks and Spinks, 2002; Vaswani et al., 2003) and hence will not be discussed further here. The modern tendency has been to combine the SSRI activity with a variety of other pharmacological effects in the same molecule and this subject was covered in an excellent recent review (Moltzen and Bang-Andersen, 2006). Another exciting event in the field of neurotransmitter transporter ligand design was the publication of two crystal structures of bacterial homologues of the synaptic reuptake proteins. The first was a homologue of the SLC1 glutamate transporter from Pyrococcus horikoshii (Yernool et al., 2004) whereas the second was an SLC6 leucine transporter from Aquifex aeolicus (Yamashita et al., 2005). The importance of these structures in understanding the mechanism of action of the SLC1 and SLC6 transporters at the molecular level was discussed in a recent review (Gether et al., 2006). Homology models of all the human SLC6 transporters have now been built and are being actively used in novel ligand design. While interest in designing new SSRIs per se has decreased, work continues to find ‘‘augmentation therapies,’’ that is, molecules having additional pharmacological effects to the primary SERT role. One approach has been to develop mixed

254

TRANSPORTERS N

Br

F

H N

O F3C

10

N

9

11

N

NC

HN

O

Cl

O

N

H N

Cl O

O

12

N

N

NH2

O

O

16

15

OH

17 N

14

F

O

Cl N H

S

21

O

Cl

H 2N

O

O

N

19

N H

N

HN O

N

20

18 N

O

H N

N

SO2 N

O N

N

N

N

22

N H

F

N

N N

N

O

23

NC

24

O

F N

O

H

O

N H

FIGURE 8.9

O O

25

N H

26

N

27 F

Uptake inhibitors for neurotransmitter transporters.

SERT/NET or SERT/NET/DAT inhibitors. Currently, only a few drugs such as venlafaxine 17, duloxetine 18, and milnacipran 19 fall in the former category. DOV21947 20 has been reported to be a triple reuptake inhibitor. This active field of research continues and the availability of homology models should have an important impact on future developments. The concept of using mixed SERT-5HT1A activities arose from the early work showing that patients dosed with a combination of paroxetine 13 and pindolol showed major improvements regarding the time of onset of antidepressant activity (Artigas et al., 1994). We have used a combination of homology models of the 5HT1 receptor subtypes and a SERT model to improve the profile of compounds related to 21 (SB-649915). Rational modification using the 5HT1 models led to 22 (SB-744185) with a superb profile of 5HT1A/B/D activity (Fig. 8.10a). At the same time, the activity of 22 at SERT was studied by docking it into a 3D model of SERT (Serafinowska et al., 2008). The primary binding site was assumed to be the TM1 aspartate_98. Docking predicted that the interactions in SERT were largely hydrophobic, although the almost perfect shape complementarity led to an ideal fit to the transporter (see Fig. 8.10b). Several pieces of evidence have emerged that suggest that a mixed SERT-5HT2A or SERT-5HT2C antagonist would help lower some of the side effects associated

THERAPEUTIC TRANSPORTER TARGETS IN DRUG DISCOVERY

255

FIGURE 8.10 (a) SB-744185 docked into the model of the 5-HT1A receptor. In addition to the salt bridge between the protonated nitrogen and the TM3 aspartate, numerous other H-bonding and hydrophobic interactions are observed. (b) SB-744185 in the SERT model. The primary binding is to aspartate 98. (See the color version of this figure in the Color Plates section.)

with SSRIs. Some compounds of this type have appeared, such as 23 (LY-367265, SERT-5HT2A) and 24 (SERT-5HT2C). Other approaches include mixed SERT-NK1 and SERT-GABA antagonists. These approaches are more speculative as no supporting clinical data have appeared.

256

TRANSPORTERS

A final SSRI-related topic worth mentioning is the existence of an allosteric binding site in SERT (Plenge et al., 1991). Escitalopram 14 and paroxetine 13 appear to stabilize their own binding to the normal high affinity site by also binding to the allosteric site. Using the SERT homology model, it may be possible to gain more information on the nature of this site and potentially design novel compounds that may act on it. The use of other selective transporter reuptake inhibitors is less well publicized than for SERT. Nevertheless, interest continues in selective NET inhibitors. Reboxetine 25 from Upjohn is marketed for depression whereas another selective NET inhibitor, atomoxetine 26 (LY-139603), is prescribed for attention deficit/hyperactivity disorder (ADHD). The case for the use of a selective DAT inhibitor is less clear, although there is some potential for such a compound being useful in the treatment of cocaine addiction. One promising example is vanoxerine 27, where 3D_QSAR methods were used to study the effect of flexible substituents on the DAT/SERT selectivity of a set of analogues (Gilbert et al., 2007). Reliable CoMFA models were developed, which could be used in further analogue design. In another study, a set of CoMFA models of the binding of a set of 76 N-substituted benztropine analogues was built to explore the molecular properties giving rise to DAT inhibition (Kulkarni et al., 2004). Although there were a number of compounds that were potent inhibitors of DAT, there was also interest in designing compounds that retained inhibition but were less lipophilic. These would potentially have improved developability properties. The authors used their CoMFA model to identify a substitution position, where the model suggested they should be able to vary the physicochemical properties without adversely affecting the binding affinity at DAT. Synthesis of these compounds subsequently resulted in a number of compounds with high DAT affinity (10–50 nM) and lower lipophilicity as predicted by the model. There is still a lot of effort ongoing in the development of inhibitors for GlyT and GAT. These transporters have been potentially implicated in schizophrenia, psychosis and dementia (GlyT), and epilepsy (GAT).

8.4

TRANSPORTERS AS LIABILITY TARGETS

Drug molecules are distributed throughout the body to different extents, depending on their molecular properties. Some of this distribution occurs through passive diffusion across membranes, but the role of transporter systems in controlling this distribution is now believed to be significant (Ayrton and Morgan, 2001; Shitara et al., 2005). Although understanding of the role of various transporter proteins in affecting distribution is still developing, Fig. 8.11 shows the distribution of some transporter proteins in the body and gives an indication of the range of transporters involved in drug distribution. As many of the transporter proteins involved in determining the distribution of compounds are able to bind to compounds with very diverse structures, there is

TRANSPORTERS AS LIABILITY TARGETS

257

FIGURE 8.11 Location of transporters affecting drug distribution (adapted from Ayrton and Morgan, 2001).

also the possibility for variable pharmacokinetics of coadministered compounds (drug–drug interactions). During the drug development process, studies are undertaken to predict the nature and extent of drug interactions by studying the interactions of drugs with these different transporters (Shitara et al., 2005). Transporterbased interactions are documented in a recent preliminary concept paper by the Food and Drug Administration (FDA) (Zhang et al., 2006). Various reported interactions are attributed to other mechanisms of actions, such as protein displacement or enzyme inhibition, which may be due in part to the inhibition of transport proteins, such as Pgp, OAT, OATP, and organic cation transporter (OCT). OATPs are plasma membrane transport proteins that mediate active influx of compounds, including bile acids and steroids (Niemi, 2007). Of the various transporters, Pgp is the most well understood and may be appropriate to evaluate during drug development. Transporters affect the passage across the gastrointestinal tract (GI) and BBB. The passive permeability of molecules across the GI tract and BBB is mainly governed by the physicochemical properties of drugs. However, the most important transporter present at the BBB is Pgp, which binds to large, lipophilic compounds. Pgp is known to play a significant role in excluding compounds from the CNS (Begley, 2004) and to a lesser extent affects passage across the GI tract by active efflux. The phenomenon of multidrug resistance is a major hurdle when it comes to the delivery of drugs to the brain, not to mention the problem of cancer chemotherapy in general. Therefore, the development of strategies for bypassing the influence of these ABC transporters has become a big challenge for the pharmaceutical industry. It will be discussed subsequently that most of these transporters prefer large, lipophilic molecules, and therefore it is not surprising that for CNS penetration

258

TRANSPORTERS

and GI absorption, we observe that a balance of physicochemical properties is required for drugs to penetrate these barriers. Both too low lipophilicity and too high lipophilicity can lead to low penetration. The primary function of the intestinal transporters is to absorb drugs during the digestion of food. In addition, the intestinal epithelium is an important site of secretion for many drugs and occurs via efflux transporters. Drug absorption across the GI tract can be highly dependent on affinity to these transport proteins that can cause efflux or influx of a compound depending on the affinity or inhibition to the transporters present. Transport proteins (e.g., amino acid transporters and monocarboxylic acid transporters) are involved in the active influx of many compounds, such as amino acids and monosaccharides from the lumen into the blood. Most of these active transporters (e.g., monocarboxylic acid transporters) prefer small molecular weight compounds (Begley and Brightman, 2003). Transport proteins able to actively efflux drugs from gut epithelial cells back into the lumen include those from several families such as MDR, MRP, and OATs (Ayrton and Morgan, 2001; Shitara et al., 2005). In the GI tract, OATP transporters are expressed in the apical membrane of intestinal enterocytes, where they can help in the absorption of a drug (Niemi, 2007). Transporters in the liver and kidney play an important role in the excretion of many drugs. Many drugs, as well as endogenous compounds, are transported from the blood through the sinusoidal membrane of hepatocytes by transporter processes rather than by passive diffusion. Following uptake into the hepatocytes, many compounds are metabolized by Phase I and II metabolism to more polar molecules. The conjugates formed after Phase II metabolism are then excreted via transporters into the bile through the bile canalicular membrane or into urine through the renal tubular epithelial cells. A number of transporter proteins have been shown to play a role in these processes (Shitara et al., 2005). For example, organic anion transporters, Pgp, and organic cation transporters are expressed in the kidney and organic anion transporting polypeptides (OATPs), multidrug resistance-associated proteins (MRPs), and breast cancer-resistant protein (BCRP) are expressed in the liver. In the liver, some of the OATPs are localized to the basolateral membrane of the hepatocytes, which may help the uptake of drugs from blood into hepatocytes (Niemi, 2007). One of the transporters in OATPs family having the most significant effect on drug elimination is OATP1B1, which is expressed predominantly in the liver and is responsible for the hepatic uptake of a number of drug molecules including those in the statin family such as cerivastatin. The magnitude and clinical significance of this type of drug–transporter interaction is exemplified in the interactions of cyclosporin A (the inhibitor) and the statins (substrates), for example, cerivastatin, pravastatin, and rosuvastatin, where six–eightfold changes in AUC are observed (Shitara et al., 2003; Noe et al., 2007). As described above, there are a number of transporters that affect drug distribution. However, Pgp and OATP1B1 are generally considered to be most important and hence a detailed discussion will be restricted to these two.

259

TRANSPORTERS AS LIABILITY TARGETS O O

N

O

N

O

N

O

O

O N

O

S

O

N

N

Bunitrolol

N

O

O

Carvedilol

O

O

O

Amprenavir O O

O

O

O

O

O

N N

N

O

O

N

O

N

Cl

O

O

O

N

N

Loperamide

O

O

N

O

O

Clarithromycin O

Indinavir N Cl Cl N

N N

O

N N

N

O

N

N

O N

O

N

O

O

N

N O

Talinolol

N

Itraconazole

Ondansetron

O

O

O

Teniposide

O

O

O

S

O O

O O O

O

O

FIGURE 8.12 Structures of some drug molecules that are known to be Pgp substrates.

8.4.1

P-glycoprotein

The structures of a few Pgp substrates are shown in Fig. 8.12. They are structurally very diverse and are usually big and lipophilic with a number of hydrogen bond acceptor groups. Pgp is probably the best known and most studied liability transporter. It is expressed on the luminal membrane of the small intestine and blood–brain barrier and in the apical membranes of the liver and kidney. Due to its presence in these membranes, Pgp plays a key role in limiting CNS penetration and intestinal absorption and in the renal and intestinal excretion of many drugs. It exports drugs from cells and utilizes ATP to provide the energy needed for the process. The interaction between the Pgp and its substrates seems to be via the lipid environment of the membrane. Pgp has been shown to affect the disposition of many drug molecules, including HIV protease inhibitors such as amprenavir. Amprenavir was shown by an in vitro assay to be a Pgp substrate. When it was administered to mdr1a/1b knockout mice and to mice pretreated with the Pgp inhibitor GF120918, the whole body

260 TABLE 8.2

TRANSPORTERS

Substrates and Inhibitors of Pgp

Substrates

Adriamycin, aldosterone, amprenavir, atorvastatin, bleomycin, bunitrolol, cyclosporine, carvedilol, chloroquine, cimetidine, clarithromycin, itraconazole, digoxin, eletriptan, erythromycin, fexofenadine, indinavir, itraconazole, loperamide, mithramycin, morphine, nelfinavir, nifedipine, ondansetron, progesterone, quinidine, ranitidine, ritonavir, saquinavir, talinolol, tamoxifen, teniposide, terfenadine, vinblastine, vincristine

Inhibitors

Amiodarone, astemizole, bepridil, chlorpromazine, cortisol, erythromycin, felodipine, fexofenadine, GF120918, gramicidin, itraconazole, lidocaine, methadone, monensin, nifedipine, omeprazole, pentazocine, quinidine, reserpine, ritonavir, saquinavir, tamoxifen, tangeretin, terfenadine,testosterone, trifluorperazine, valinomycin, verapamil, yohimbine

autoradiography (WBA) studies showed that amprenavir is able to penetrate the CNS when Pgp has been knocked out, although not in wild-type mice where the Pgp is able to function normally (Polli et al., 1999). Several examples have demonstrated an in vitro–in vivo correlation between the Pgp inhibition and a measured clinical drug– drug interactions, which shows that Pgp affinity is proportional to the magnitude of the corresponding drug–drug interaction (Yasuda et al., 2002; Aszalos, 2004; Tang et al., 2005; Punit and Rodrigues, 2006; Rautio et al., 2006). When discussing transporters, substrates are usually defined as compounds that are transported across a membrane by a particular transporter, and inhibitors are compounds that inhibit the transporter’s ability to transport compounds across a membrane. Therefore, it is quite important to know if given compound is a substrate or an inhibitor as it can have different effect on drug–drug interactions. Table 8.2 shows a list of compounds identified as Pgp substrates and inhibitors. Pharmaceutical companies and academic groups have developed high-throughput screening methods for assessing whether compounds are substrates or inhibitors for particular transporters. The most commonly used Pgp inhibition assay utilizes a human uterine sarcoma cell line that has been induced and selected to overexpress Pgp using doxorubicin (Harker and Sikic, 1985). This cell line rapidly effluxes the nonfluorescent dye precursor, calcein-AM. The inhibition of Pgp results in an accumulation of calcein-AM in the cell, which is subsequently deesterified to yield calcein, a fluorescent dye that is trapped in the cell. This competition assay can only be used as an indication of binding. It is unable to distinguish real inhibitors from substrates that bind tightly. There is an in vitro Pgp substrate screen commonly used, which measures the flux of a compound across a monolayer of hMDR1-MDCK (Type II) cells, a transfected cell line expressing Pgp (MDR-1 human type) on the apical surface. By using this assay, it is possible to indicate if a given compound is a Pgp substrate or not. The permeability of compounds from apical to basolateral (A–B) and from basolateral to apical (B–A) sides of the monolayer is measured and the flux ratio is calculated in the absence and presence of a Pgp inhibitor using formula given below: Flux Ratio ¼ BA Papp ðnm=sÞ=AB Papp ðnm=sÞ

261

TRANSPORTERS AS LIABILITY TARGETS

The Pgp efflux mechanism actively transports compounds from the basolateral to the apical side of the membrane. Therefore, a Pgp substrate has a greater permeability from basolateral to apical than apical to basolateral and the flux ratio  1. This ratio is reduced in the presence of a Pgp inhibitor (Polli et al., 2001). 8.4.2

OATP1B1

The structures of a few OATP1B1 substrates are shown in Fig. 8.13. They are structurally diverse and are usually big and lipophilic and are mostly either neutral or acidic. OATP1B1 has a role in the hepatic uptake and hence elimination of drugs and some physiological substrates. It is located on the sinusoidal membrane in the liver and it mediates the transport of drugs from blood into the liver. Of all the OATP family of transporters, OATP1B1 is the most studied and most important. The potential for drug–drug interactions due to inhibition of hepatic OATP1B1 is high and may lead to marked increases in systemic exposure either through changes in first pass extraction or systemic clearance. It is of particular significance due to the high number of people being prescribed statins. For any new therapeutic agents likely to be coadministered with a statin, it is necessary to understand the degree of interaction with OATP1B1. For example, the coadministration of cyclosporin A (an OATP1B1 inhibitor) with cerivastatin (OATP1B1 substrate) was shown to result in a fourfold increase in the AUC and fivefold increase in Cmax (Shitara et al., 2003; Noe et al., 2007). The cyclosporin A is inhibiting the OATP1B1 transporter resulting in reduced uptake into the liver of cerivastatin, thus reducing its clearance and

O

F

F

O

O

O

N

N

N

N

N

O

O

O

O

O O

O

N

O

O

N O

O

Atorvastatin

O

Bilirubin Cerivastatin

O O

O

Methotrexate

N O

N

O O

N N

O

O

N

N

O O

Cholic acid

N N

O O O N

N O

O

O N N O

O

O

O N

O

O

O

N O

FIGURE 8.13

S

N

N

N

N

O

O N

O

O

O

O

O

O

N

Phalloidin

N

Pravastatin

Saquinavir

N

O

Structures of some drug molecules that are known to be OATP1B1 substrates.

262 TABLE 8.3

TRANSPORTERS

Substrates and Inhibitors of OATP1B1

Substrates

Atorvastatin, benzylpenicillin, bilirubin, cerivastatin, cholic acid, demethylphalloin, lovastatin, methotrexate, microcystin, phalloidin, pravastatin, rifampin, rosuvastatin, saquinavir, simvastatin

Inhibitors

Bilirubin, estradiol, estrone, glycyrrhizin, human serum albumin, hyperforin, indinavir, nelfinavir, pioglitazone, pravastatin, ritonavir, rosiglitazone, saquinavir, sirolimus, rapamycin, troglitazone

resulting in elevated plasma concentrations. A few of the known substrates and inhibitors for OATP1B1 are mentioned in Table 8.3. The most common assay for inhibition of OATP1B1 was developed in the University of Zurich (Eckhardt, 1999), where OATP1B1 was shown to transport a fluorescein conjugate of methotrexate. The production of this fluorescein can be stopped with bromosulfophthalein (a potent inhibitor) and Draq-5 (a nuclear stain). By use of fluorescent image analysis, the relative amounts of substrate taken up by the cells can be determined. 8.5 APPLICATION OF METHODS FOR DESIGNING INTERACTIONS WITH LIABILITY TARGETS Using a set of 100 Pgp substrates, nonsubstrates and inducers, the number of electron donor (hydrogen bond acceptor) groups did not distinguish compounds that were Pgp substrates from those that were not (Seelig, 1998). However, a relationship with the spatial separation of the electron donor groups could distinguish the substrates and nonsubstrates. These were expressed in two rules or recognition elements (denoted by Type I or Type II). Type I was when two electron donor ˚ and Type II was when either two electron groups were separated by 2.5þ/0.3 A ˚ or when the outer of three electron donor groups were separated by 4.6 þ/0.6 A ˚ donor groups were separated by 4.6 þ/0.6 A. The more the hydrogen bond acceptor groups forming Type I and Type II units, the more strongly the compounds were observed to interact with Pgp. This phenomenon was further rationalized in terms of the hydrogen bond donor groups on the amino acid side chains in the Pgp. The transmembrane sequences that had the highest number of amino acids with hydrogen bond donor side chains generally corresponded to those that had been shown experimentally to be involved in drug binding. However, if one analyzes the Pgp substrate data from more recent research (Crivori et al., 2006) and compares the number of compounds identified as substrates and nonsubstrates from the efflux assay with those predicted by the rule of fours (described above), a count of the hydrogen bond acceptors and the Abraham B descriptor, it is possible to see that they all have some value in quickly and simply distinguishing substrates from nonsubstrates. These data are shown in Table 8.4. The percentages in Table 8.4 are calculated from the number of compounds predicted in a class and measured to be in that class divided by the total number predicted as that class. All rules have their advantages and disadvantages. When the HBA

263

APPLICATION OF METHODS FOR DESIGNING INTERACTIONS

TABLE 8.4 Measured N Y

Prediction of Pgp Substrates from Simple Property Calculations Predicted N (HBA<5)

Y (HBA>5)

30 (77%) 9 (23%)

3 (25%) 9 (75%) Predicted

Measured

N (B<2)

Y (B>¼2)

N Y

22 (92%) 2 (8%)

11 (41%) 16 (59%) Predicted

Measured N Y

N (rule of 4) 7 (100%) 0 (0%)

Y (rule of 4) 1 (13%) 7 (87%)

Can’t tell a 25 (66%) 13 (34%)

Y ¼ Pgp substrate, N ¼ non-Pgp substrate, HBA ¼ hydrogen bond acceptor count, B ¼ abraham beta value, rule of 4 (Didziapetris et al., 2003). a Prediction is unable to distinguish between a substrate and non-substrate.

rule predicts a compound to be a substrate or nonsubstrate, it is correct 75% of the time. When the B rule predicts a nonsubstrate, it is correct 90% of the time but is less good than HBA at predicting substrates correctly. In contrast, the rules of fours are very good at getting a correct prediction for both substrates and nonsubstrates, but they only make a prediction for 15% of compounds. This information is shown here to indicate the extent to which simple properties can be used to predict the interaction of molecules with transporters. Although the predictability of these simple properties can be lower than for more sophisticated models, they have general applicability to most classes of compounds and hence have value in helping chemists understand the key descriptors determining transporter interaction. In principle, it is also possible to derive similar simple rules for other transporters. However, although for Pgp there is a reasonable amount of data available (hundreds – thousands of compounds) in the literature and from in-house screening initiatives, this is not currently the case for other transporters where data on fewer than hundreds of consistently measured data points can be collated from the literature sources. To exemplify general rules for other transporters, (albeit small) data sets have been taken from the literature. The correlation of the binding of compounds to the transporter with some simple physicochemical properties was investigated with the objective of providing some general guidelines as to which types of compounds are likely to bind to a range of transporters. The properties selected were log P, molecular weight (MW), and hydrogen bond donor and acceptor counts (HBA, HBD). These are very small data sets and therefore the trends (Table 8.5) should be used with caution; however, those observed are broadly in line with current knowledge about the transporters themselves.

264

30 31 38 36 30 51

hOCT1 hOCT2 hOATl-4 rOAT1-3 OATP1B1 Pgp

0.46 0.16b none 0.64b none none

b

LogP

0.47 0.23b 0.47b none none 0.57b

b

Siae none none none none none none

A none none 0.38b none none 0.65b

B Bases and quats Bases and quatsa Acids Acids Acids, neutrals, zwitterions Bases and neutrals

b 2

a

a

Charge Class

quats ¼ quaternary nitrogen compounds r value for the correlation hOCT2 correlations - ibuprofen was omitted OAT1-4 data was combined to calculate r2 due to the number of pKi values for each transporter Pgp data was efflux ratio n ¼ number of compounds A ¼ Abraham descriptor (hydrogen bond donor) B ¼ Abraham descriptor (hydrogen bond acceptor) Molecular weight is used as the descriptor for size

n

Transporter

TABLE 8.5

Bednarczyk (2003) Suhre et al. (2005) Shitara et al. (2005) Shitara et a1. (2005) Chang et al. (2005) Crivori et al. (2006)

Data Source

inhibitors inhibitors inhibitors inhibitors substrates substrates

Inhibitors/Substratesi

265

APPLICATION OF METHODS FOR DESIGNING INTERACTIONS

1

1

0

0

pKi hOAT

pK i hOAT

For example, OCT1 binds to basic, lipophilic compounds and unlike Pgp (as discussed above) there does not appear to be a strong dependence on hydrogen bonding that is in agreement with the molecular features identified from the more sophisticated 3D QSAR study (Suhre et al., 2005). The rat and human OATs bind to acidic compounds. There are some differences between the rat and human OATs. For each of the human OATs, (1, 3, and 4) inhibition increases with increasing size (MW), whereas there is no effect with lipophilicity. For the rat OATs (1, 2, and 3), inhibition increases with lipophilicity, but there is no effect with size. These results are shown in Fig. 8.14. Analysis of the OATP1B1 data (Chang and Swaan 2005) did not identify any simple trends except that acids, neutral and zwitterionic compounds, were substrates for the transporter. This finding could be due to the data being measured

-1

-1

-2

-2

-3

-3 200

300

400

500

-3

600

-2

-1

0

1

2

3

4

clogP

MW 1 1 0.5 0

0

pKi rOAT

pKi rOAT

-0.5 -1 -1.5 -2

-1

-2

-2.5 -3

-3 200

300

MW

400

500

600

-3

-2

-1

0

1

2

3

4

clogP

FIGURE 8.14 Correlations of pKi with clogP and MW for human and rat OAT. Square ¼ OAT1, circle ¼ OAT2, triangle ¼ OAT3, and star ¼ OAT4.

266

TRANSPORTERS

in a variety of protocols using different cell lines. Our experience (data not reported) is that in addition to binding these classes of compounds, OATP1B1 binding is favored by lipophilic molecules. This result is consistent with the observation of other researchers that OATP1B1 inhibitors consist of bulky compounds (Shitara et al., 2005). Much of the work published on the 3D QSARs of transporters is on Pgp (e.g., Ekins et al., 2002; Yates et al., 2003; Crivori et al., 2006) whose role in limiting transport across the blood–brain barrier and gastrointestinal tract by active efflux has been well established (Polli et al., 1999; Ayrton and Morgan, 2001). However, over the last few years, 3D QSARs on other liability transporters such as the OCT family (OCT1 and OCT2) (Bednarczyk et al., 2003; Suhre et al., 2005) and the OATP family (Yarim et al., 2005) have been published. In addition, 3D QSARs on transporters of pharmacological targets such as DAT (Kulkarni et al., 2002, 2004) and SERT (Roman, 2003) are available. Most of the compound sets used for these analyses are very small (10–30 compounds) but they can identify regions of the molecules where steric and electrostatic interactions are important for binding. It therefore follows that this information can be used to design molecules where these interactions are not favored. As these models are generally built using small data sets of structurally similar compounds, it is debatable whether the models currently reported in the literature are extendable to structural classes outside those on which the model has been built. As more transporter data become available, it should be possible to extend the chemical space for which these models are applicable. The application of 3D QSARs to the modeling of transporters has been comprehensively reviewed (Ekins and Swaan, 2004; Chang et al., 2005, 2006). We have taken the set of substrate binding data for the human OCT1 (Benarczyk et al., 2003) and use it to exemplify the pros and cons of simple correlations against more sophisticated 3D QSAR analyses. The data in Table 8.6 show that although the pharmacophore model built using catalyst and the descriptor-based QSAR using descriptors generated by Cerius give very good correlations (expressed as r2 values) on the training set, the values on the test set are considerably lower. However, while the log P gives a poorer correlation on the training set, there is no appreciable deterioration in the predictive ability on the test set. This observation suggests that the 3D QSAR models do not have very good predictive power outside the training set domain. As log P increases, the interaction with human OCT1 also increases, a relationship that can potentially be applied to other compound series. The downside is that a correlation with log P does not offer the chemist the molecular design

TABLE 8.6 Comparison of a Simple Property (log P) and 3D QSARs in Predicting the Interaction with OCT1. Table Shows the r2 Values for the Models Transporter

Data set

OCT1

Training Test All

Pharmacophore 0.74 0.29 0.64

Cerius2

ClogP

0.94 0.20 0.90

0.49 0.47 0.46

REFERENCES

267

insight to know where a lipophilic group should be removed or polar group added to reduce the interactions with the undesired transporter.

8.6

PERSPECTIVE

Despite the wide variety of transporter proteins in the genome, relatively few of these have been the targets for drug design. This situation is likely to change in the next few years as more knowledge of transporter 3D structures becomes available. Current treatment of CNS disorders involving uptake proteins has been confined to the SERT and NET transporters, but there is also good validation for the involvement of GABA and GlyT. Likewise, the osteoclast ATPase is a well-validated target, but no drug has as yet reached the market. Numerous sugar transporters have been implicated in diseases such as diabetes. With the availability of the bacterial lactose permease structure, it should be possible to re-explore this field. Many rarer diseases also involve transporters and the reader is referred to the associated human diseases link at Saier’s web site (www.tcdb.org). Another area where work is required is the understanding of efflux mechanisms. In both humans and harmful bacteria, efflux pumps act to restrict the bioavailability of otherwise important drugs. The coadministration of such drugs with an efflux inhibitor would be an obvious way of tackling this problem. The distribution of drug molecules is very dependent on their affinity to transport proteins. There can be efflux or influx of a compound, depending on the affinity to the transporters present in the membrane. Many of the transporter proteins involved in determining the distribution of compounds are able to bind to compounds with very diverse structures, thereby increasing the possibility for variable pharmacokinetics of coadministered compounds (drug–drug interactions). From a drug development viewpoint, it is quite important to predict the nature and extent of these drug interactions at an early stage in the drug development process. As 3D structural information of these targets become more and more available, it will help us to understand the SAR in more detail. By a combination of structure-based and ligand-based methodologies, it is increasingly possible to predict which compounds are likely to bind to a transporter and in some cases the degree of binding to the transporter. These design strategies are able to help identify compounds that are not substrates or inhibitors for transporters and will not therefore be subject to drug–drug interactions or have their in vivo distribution limited by their interaction with transporters.

REFERENCES Abramson J, Smirnova I, Kasho V, Verner G, Kaback HR, Iwata S, 2003. Structure and mechanism of the lactose permease of Escherichia coli. Science 301:610–616. Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ, 1997. Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res. 25:3389–3402.

268

TRANSPORTERS

Artigas F, Perez V, Alvarez E, 1994. Pindolol induces a rapid improvement of depressed patients treated with serotonin reuptake inhibitors. Arch. of Gen. Psychiatry 51:248–251. Aszalos A, 2004. P-glycoprotein-based drug–drug interactions: preclinical methods and relevance to clinical observations. Arch. Pharmacal. Res. 27:127–135. Ayrton A, Morgan P, 2001. Role of transport proteins in drug absorption, distribution and excretion. Xenobiotica 31:469–497. Barton, N, Blaney FE, Garland S, Tehan B, Wall I, 2006. Seven transmembrane G proteincoupled receptors: insights for drug design from structure and modelling. In: Triggle DJ, editor, Comprehensive Medicinal Chemistry II, Vol. 4, Elsevier, pp. 669–701. Begley DJ, 2004. ABC transporters and the blood–brain barrier. Curr. Pharm. Des. 10:1295–1312. Begley DJ, Brightman MW, 2003. Structural and functional aspects of the blood–brain barrier. Prog. Drug Res. 61:39–78. Benarczyk D, Ekins S, Wikel JH, Wright SH, 2003. Influence of molecular structure on substrate binding to the organic cation transporter, hOCT1. Mol. Pharmacol. 63:489–498. Blair HC, Teitelbaum SL, Ghiselli R, Gluck S, 1989. Osteoclastic bone resorption by a polarized vacuolar proton pump. Science 245:855–857. Blaney FE, 2007. Approaches to the molecular modeling of 7-transmembrane helical receptors. Current Protocols in Pharmacology, Section 9.8, John Wiley & Sons, Inc. Blaney FE, Tennant M, 1996. Computational tools and results in the construction of G-proteincoupled receptor models. In: Findlay JBC, editor, Membrane Protein Models, Bios Scientific Publishers Ltd, Oxford, pp. 161–176. Borst P, Elferink RO, 2002. Mammalian ABC transporters in health and disease. Annu. Rev. of Biochem. 71:537–592. Bowie JU, Luthy R, Eisenberg D, 1991. A method to identify protein sequences that fold into a known three-dimensional structure. Science 253:164–170. Bowman BJ, Bowman EJ, 2002. Mutations in subunit C of the vacuolar ATPase confer resistance to bafilomycin and identify a conserved antibiotic binding site. J. Biol. Chem. 277:3965–3972. Bowman EJ, Graham LA, Stevens TH, Bowman BJ, 2004. The bafilomycin/concanamycin binding site in subunit c of the V-ATPases from Neurospora crassa and Saccharomyces cerevisiae. J. Biol. Chem. 279:33131–33138. Bowman BJ, McCall ME, Baertsch R, Bowman EJ, 2006. A model for the proteolipid ring and bafilomycin/concanamycin-binding site in the vacuolar ATPase of Neurospora crassa. J. Biol. Chem. 281:31885–31893. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M, 1983. CHARMM:a program for macromolecular energy, minimization, and dynamics calculations. J. Comput. Chem. 4:187–217. Busch W, Saier MH, 2002. The transporter classification (TC) system. Crit. Rev. Biochem. Mol. Biol. 37:287–337. Chang C, Ray A, Swaan P, 2005. In silico strategies for modelling membrane transport function. Drug Discov. Today 10(9):663–671. Chang C, Swaan PW, 2006. Computational approaches to modelling drug transporters. Eur. J. Pharm. Sci. 27:411–424. Chang G, Roth CB, Reyes CL, Pornillos O, Chen Y, Chen AP, 2006. Structure of the ABC transporter MsbA in complex with ADP-vanadate and lipopolysaccharide. [Retraction of document cited in CA143:111210.] Science 314:1875.

REFERENCES

269

Clark DE, 2001. Prediction of intestinal absorption and blood–brain barrier penetration by computational methods. Comb. Chem. High Throughput Screen. 4(6):477–496. Cornell WD, Cieplak P, Bayly CI, Gould IR, Merz Jr KM, Ferguson DM, Spellmeyer DC, Fox T, Caldwell JW, Kollman PA, 1995. A second generation force field for the simulation of proteins, nucleic acids, and organic molecules. J. Am. Chem. Soc. 117:5179–5197. Cramer RD III, Patterson DE, Bunce JD, 1988. Comparative molecular field analysis (CoMFA) 1. Effect of shape on binding of steroids to carrier proteins. J. Am. Chem. Soc. 110:5959–5967. Crivori P, Reinach B, Pezzetta D, Poggesi I, 2006. Computational models for identifying potential P-glycoprotein substrates and inhibitors. Mol. Pharmaceutics 3(1):33–44. Cruciani G, Crivori P, Carrupt P-A, Testa B, 2000. Molecular fields in quantitative structure– permeation relationships: the VolSurf approach. J. Mol. Struct.: THEOCHEM 503:17–30. Didziapetris R, Japertas P, Avdeef A, Petrauskas A, 2003. Classification of P-glycoprotein substrate specificity. J. Drug Target. 11(7):391–406. Donnelly D, Overington JP, Ruffle SV, Nugent JHA, Blundell TL, 1993. Modeling a helical transmembrane domains: the calculation and use of substitution tables for lipid-facing residues. Protein Sci. 2:55–70. Doyle DA, Cabral JM, Pfuetzner RA, Kuo A, Gulbis JM, Cohen SL, Chait BT, MacKinnon R, 1998. The structure of the potassium channel: molecular basis of Kþ conduction and selectivity. Science 280:69–77. Eckhardt U, 1999. Polyspecific substrate uptake by the hepatic organic anion transporter Oatp1 in stably transfected CHO cells. Am. J. Physiol. 276:G1037–G1042. Edvardsen O, Dahl SG, 1994. A putative model of the dopamine transporter. Mol. Brain Res. 27(2):265–274. Ekins S, Swaan PW, 2004. Computational models for enzymes, transporters, channels and receptors relevant to absorption, distribution, metabolism, excretion and toxicology. Rev. Comp. Chem. 20:333–415. Ekins S, Kim RB, Leake BF, Dantzig AH, Schuetz EG, Lan L, Yasuda K, Shepard RL, Winter MA, Schuetz JD, Wikel JH, Wrighton SA, 2002. Application of three-dimensional quantitative structure–activity relationships of P-glycoprotein inhibitors and substrates. Mol. Pharmacol., 61:974–981. Ertl P, Rohde B, Selzer P, 2000. Fast calculation of molecular polar surface area as a sum of fragment-based contributions and its application to the prediction of drug transport properties. J. Med. Chem. 43:3714–3717. Farina C, Gagliardi S, 1999a. Selective inhibitors of vacuolar Hþ-ATPase of osteoclasts with bone antiresorptive activity. Expert Opin. Ther. Pat 9:157–168. Farina C, Gagliardi S, 1999b. Selective inhibitors of the osteoclast vacuolar proton ATPase as novel bone antiresorptive agents. Drug Discov. Today 4:163–172. Frillingos S, Sahin-Toth M, Wu J, Kaback HR, 1998. Cys-scanning mutagenesis: a novel approach to structure–function relationships in polytopic membrane proteins. FASEB J. 12:1281–1299. Gether U, Andersen PH, Larsson OM, Schousboe A, 2006. Neurotransmitter transporters: molecular function of important drug targets. Trends Pharmacol. Sci. 27:375–383. Gilbert KM, Boos TL, Dersch CM, Greiner E, Jacobson AE, Lewis D, Matecka D, Prisinzano TE, Zhang Y, Rothman RB, Rice KC, Venanzi CA, 2007. DAT/SERT selectivity of flexible GBR 12909 analogs modeled using 3D-QSAR methods. Bioorg. Med. Chem. 15:1146–1159.

270

TRANSPORTERS

Harker WG, Sikic BI, 1985. Multidrug (pleiotropic) resistance in doxorubicin-selected variants of the human sarcoma cell line MES-SA. Cancer Res. 45:4091–4096. Hediger MA, Romero MF, Peng J, Rolfs A, Takanaga H, Bruford EA, 2004. The ABCs of solute carriers: physiological, pathological and therapeutic implications of human membrane transport proteins. Pflugers Archi. 447:465–468. Hersey A, Luscombe C, Modi S, 2004. In-silico transporter modelling and its application in drug discovery. In: Borchardt RT, Kerns EH, Lipinski CA, Thakker D, Wang B, editors, Pharmaceutical Profiling in Drug Discovery for Lead Selection. AAPS Press. Hill T, Lewicki P, 2005. Statistics: methods and applications. StatSoft, Inc., www.statsoft.com/ textbook/stathome.html. Hoglund PJ, Adzic D, Scicluna SJ, Lindblom J, Fredriksson R, 2005. The repertoire of solute carriers of family 6: identification of new human and rodent genes. Biochem. Biophys. Res. Commun. 336:175–189. Huang Y, Lemieux MJ, Song J, Auer M, Wang D, 2003. Structure and mechanism of the glycerol-3-phosphate transporter from Escherichia coli. Science 301:616–620. Jain KS, Shah AK, Bariwal JR, Shelke SM, Kale AP, Jagtap JR, Bhosale AV, 2007. Recent advances in proton pump inhibitors and management of acid–peptic disorders. Bioorg. Med. Chem. 15:1181–1205. Jiang Y, Lee A, Chen J, Cadene M, Chait BT, MacKinnon R, 2002. Crystal structure and mechanism of a calcium-gated potassium channel. Nature 417:515–522. Jiang Y, Lee A, Chen JN, Ruta V, Cadene M, Chait BT, MacKinnon R, 2003. X-ray structure of a voltage-dependent Kþ channel. Nature 423:33–41. Kaback HR, Wu J, 1999. What to do while awaiting crystals of a membrane transport protein and thereafter. Acc. Chem. Res. 32:805–813. Kim CG, Watts JA, Watts A, 2005. Ligand docking in the gastric Hþ/Kþ-ATPase: homology modeling of reversible inhibitor binding sites. J. Med. Chem. 48:7145–7152. Klebe G, 1998. Comparative molecular similarity indices: COMSIA. In: Kubinyi H, Folkes G, Martin YC, editors, 3D QSAR in Drug Design. Kluwer Academic Publishers, UK. Kulkarni SS, Newman AH, Houlihan WJ, 2002. Three dimensional quantatative structure– activity relationships of mazindol analogues at the dopamine transporter. J. Med. Chem. 45:4119–4127. Kulkarni SS, Grundt P, Kopajtic T, Katz JL, Newman AH, 2004. Structure–activity relationships at monoamine transporters for a series of N-substituted 3a-(bis[4-fluorophenyl] methoxy)tropanes: comparative molecular field analysis, synthesis and pharmacological evaluation. J. Med. Chem. 47:3388–3398. Kyte J, Doolittle RF, 1982. A simple method for displaying the hydropathic character of a protein. J. Mol. Biol. 157:105–132. Leach AR, 2000. Molecular Modelling: Principles and Applications. Prentice Hall, Harlow, UK. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ, 1997. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv.Drug Del.Rev. 23:3–25. Long SB, Campbell EB, MacKinnon R, 2005. Crystal structure of a mammalian voltagedependent shaker family Kþ channel. Science 309:897–903. Meier-Abt F, Mokrab Y, Mizuguchi K, 2006. Organic anion transporting polypeptides of the OATP/SLCO superfamily: identification of new members in nonmammalian species, comparative modelling and a potential transport mode. J. Memb. Biol. 208:213–227.

REFERENCES

271

Middleton DA, Robins R, Feng X, Levitt MH, Spiers ID, Schwalbe CH, Reid DG, Watts A, 1997. The conformation of an inhibitor bound to the gastric proton pump. FEBS Lett. 410:269–274. Moltzen EK, Bang-Andersen B, 2006. Serotonin reuptake inhibitors: the corner stone in treatment of depression for half a century - a medicinal chemistry survey. Curr. Top. Med. Chem. 6:1801–1823. Munson K, Garcia R, Sachs G, 2005. Inhibitor and ion binding sites on the gastric H,KATPase. Biochemistry 44:5267–5284. Munson K, Law RJ, Sachs G, 2007. Analysis of the gastric H,K ATPase for ion pathways and inhibitor binding sites. Biochemistry 46:5398–5417. Murata T, Yamato I, Kakinuma Y, Leslie AGW, Walker JE, 2005. Structure of the rotor of the V-type Naþ-ATPase from Enterococcus hirae. Science 308:654–659. Niemi M, 2007. Role of OATP transporters in the disposition of drugs. Pharmacogenomics 8:787–802. Noe J, Portmann R, Brun ME, Funk C, 2007. Substrate-dependent drug–drug interactions between gemfibrozil, fluvastatin and other organic anion-transporting peptide (OATP) substrates on OATP1B1, OATP2B1, and OATP1B3. Drug Metab. Dispos. 35:1308– 1314. Osterberg T, Norinder U, 2000. Theoretical calculation and prediction of P-glycoproteininteracting drugs using Molsurf parameterization and PLS statistics. Eur. J. Pharm. Sci. 10:295–303. Perry JL, Dembla-Rajpal N, Hall LA, Pritchard JB, 2006. A three-dimensional model of human organic anion transporter 1: Aromatic amino acids required for substrate transport. J. Biochem. 281:38071–38079. Platts JA, Butina D, Abraham MH, Hersey A, 1999. Estimation of molecular linear free energy relation descriptors using a group contribution approach. J. Chem. Inf. Comput. Sci. 39:835–845. Plenge P, Mellerup ET, Laursen H, 1991. Affinity modulation of [3H]imipramine, [3H] paroxetine and [3H]citalopram binding to the 5-HT transporter from brain and platelets. Eur. J. Pharmacol. 206(3):243–250. Polli JW, Jarrett JL, Studenberg SD, Humphreys JE, Dennis SW, Brouwer KR, Woolley JL, 1999. Role of P-glycoprotein on the CNS disposition of amprenavir (141W94), an HIV protease inhibitor. Pharm. Res. 16:1206–1212. Polli JW, Wring SA, Humphreys JE, Huang L, Morgan JB, Webster LO, Serabjit-Singh CS, 2001. Rational use of in vitro P-glycoprotein assays in drug discovery. J. Pharmacol. Exp. Ther. 299(2): 620–628. Popp C, Gorboulev V, Mueller TD, Gorbunov D, Shatskaya N, Koepsell H, 2005. Amino acids critical for substrate affinity of rat organic cation transporter 1 line the substrate binding region in a model derived from the tertiary structure of lactose permease. Mol. Pharmacol. 67:1600–1611. Punit MH, Rodrigues DA, 2006. In vivo animal models for investigating potential CYP3A- and Pgp-mediated drug–drug interactions. Curr. Drug Metab. 7:687–704. Rautio J, Humphreys JE, Webster LO, Balakrishnan A, Keogh JP, Kunta JR, Serabjit-Singh CJ, Polli JW, 2006. In-vitro P-glycoprotein inhibition assays for assessment of clinical drug interaction potential of new drug candidates: a recommendation for probe substrates. Drug Metab. Dispos. 34:786–792.

272

TRANSPORTERS

Ravna AW, Edvardsen O, 2001. A putative three-dimensional arrangement of the human serotonin transporter transmembrane helices: a tool to aid experimental studies. J. Mol. Graphics Modell. 20:133–144. Ren Q, Chen K, Paulsen IT, 2007. TransportDB: a comprehensive database resource for cytoplasmic membrane transport systems and outer membrane channels. Nucleic Acids Res. 35:106–114. Roman DL, Walline CC, Rodriguez GJ, Barker EL, 2003. Interactions of antidepressants with the serotonin transporter: a contemporary molecular analysis. Eur. J. Pharmacol. 479:53–63. Rost B, Sander C, 1993. Prediction of protein secondary structure at better than 70% accuracy. J. Mol. Biol. 232:584–599. Sachs G, Shin JM, Vagin O, Lambrecht N, Yakubov I, Munson K, 2007. The gastric H,K ATPase as a drug target: past, present, and future. J. Clin. Gastroenterol. 41(6, Suppl. 2): S226–S242. Salas-Burgos A, Iserovich P, Zuniga F, Vera JC, Fischbarg J, 2004. Predicting the threedimensional structure of the human facilitative glucose transporter Glut1 by a novel evolutionary homology strategy: insights on the molecular mechanism of substrate migration, and binding sites for glucose and inhibitory molecules. Biophys. J. 87:2990– 2999. Seelig A, 1998. A general pattern for substrate recognition by P-glycoprotein. Eur. J. Biochem. 251:252–261. Serafinowska HT, Blaney FE, Lovell PJ, Scott CM, Smith PW, Starr KR, Watson JM, 2008. Novel 5-HT1A/1B/1D receptors antagonists with potent 5-HT reuptake inhibitory activity. Bioorg. Med. Chem. Lett. in press. Shitara Y, Itoh T, Sato H, Li AP, Sugiyama Y, 2003. Inhibition of transporter-mediated hepatic uptake as a mechanism for drug–drug interaction between cerivastatin and cyclosporin A. J. Pharmacol. Exp. Ther. 304:610–616. Shitara Y, Sato H, Sugiyama Y, 2005. Evaluation of drug–drug interaction in the hepatobiliary and renal transport of drugs. Annu. Rev. Pharmacol. Toxicol. 45:689–723. Spinks D, Spinks G, 2002. Serotonin reuptake inhibition: an update on current research strategies. Curr. Med. Chem. 9:799–810. Suhre WM, Ekins S, Chang C, Swaan PW, Wright SH, 2005. Molecular determinants of substrate/inhibitor binding to the human and rabbit renal organic cation transporters hOCT2 and rbOCT2. Mol. Pharmacol. 67:1067–1077. Sun-Wada G, Wada Y, Futai M, 2004. Diverse and essential roles of mammalian vacuolar-type proton pump ATPase: toward the physiological understanding of inside acidic compartments. Biochim. Biophys. Acta-Bioenergetics, 1658:106–114. Tang C, Lin JH, Lu AYH, 2005. Metabolism-based drug–drug interactions: What determines individual variability in cytochrome P450 induction? Drug Metab. Dispos. 33:603–613. Toyoshima C, Nomura H, 2002. Structural changes in the calcium pump accompanying the dissociation of calcium. Nature 418:605–611. Toyoshima C, Nakasako M, Nomura H, Ogawa H, 2000. Crystal structure of the calcium pump ˚ resolution. Nature 405:647–655. of sarcoplasmic reticulum at 2.6 A Vaswani M, Linda FK, Ramesh S, 2003. Role of selective serotonin reuptake inhibitors in psychiatric disorders: a comprehensive review. Prog. Neuro-Psychopharmacol. Biol. Psychiatry 27:85–102.

REFERENCES

273

Wang Y, Inoue T, Forgac M, 2005. Subunit a of the yeast V-ATPase participates in binding of bafilomycin. J. Biol. Chem. 280:40481–40488. Ward A, Reyes CL, Yu J, Roth CB, Chang G, 2007. Flexibility in the ABC transporter MsbA: alternating access with a twist. Proc. Nat. Acad. Sci. 104(48):19005–19010. Watts A, 1999. NMR of drugs and ligands bound to membrane receptors. Curr. Opin. Biotechnol. 10:48–53. Wold S, Johansson E, Cocchi M, 1993. Partial least-squares projections to latent structures. In: Kubinyi H, editor, 3D QSAR in Drug Design, ESCOM, Leiden, pp. 523–550. Yamashita A, Singh SK, Kawate T, Jin Y, Gouaux E, 2005. Crystal structure of a bacterial homologue of Naþ/Cl-dependent neurotransmitter transporters. Nature 437:215–223. Yarim M, Moro S, Huber R, Meier PJ, Kaseda C, Kashima T, Hagenbach B, Folkers G, 2005. Application of QSAR analysis to organic anion transporting polypeptide 1a5 (Oatp1a5) substrates. Bioorg. Med. Chem. 13:463–471. Yasuda K, Lan L, Sanglard D, Furuya K, Schuetz JD, Schu EG, 2002. Interaction of cytochrome P450 3A inhibitors with P-Glycoprotein, J. Pharmacol. Exp. Ther. 303:323–332. Yates CR, Chang C, Kearbey JD, Yasuda K, Schuetz EG, Miller DD, Dalton JT, Swaan PW, 2003. Structural determinants of P-glycoprotein-mediated transport of glucocorticoids. Pharm. Res. 20(11):1794–1803. Yernool D, Boudker O, Jin Y, Gouaux E, 2004. Structure of a glutamate transporter homologue from Pyrococcus horikoshii. Nature 431:811–818. Zhang L, Strong JM, Qiu W, Lesko LJ, Huang SM, 2006. Drug interaction studies study design, data analysis, and implications for dosing and labelling states. Mol. Pharmacol. 3:62–69.

9 NUCLEAR RECEPTOR DRUG DISCOVERY HIROYUKI KAGECHIKA

9.1

AND

AYA TANATANI

INTRODUCTION

Small hydrophobic molecules, such as steroid hormones, thyroid hormones, vitamin A (retinoids), and vitamin D, have significant physiological roles in growth, development, metabolism, and homeostasis. While hydrophilic signaling molecules, including peptide hormones and growth factors, have their own receptors on the cell surface, the hydrophobic signaling molecules can pass through the cell membrane and act as a switch for the target gene expression, mediated by binding to and activating their specific nuclear receptors. A schematic of these processes is shown in Fig. 9.1. Historically, nuclear receptors have been found by mechanistic investigations of steroid hormone actions, and several cDNAs were isolated in the 1980s. Since cDNA encoding retinoic acid receptor (RAR) was identified by the analysis of clones bearing the highly conserved sequence with those of steroid nuclear receptors in 1987, the existence of nuclear receptor superfamily was suggested. Their physiological functions and the implication in various intractable diseases have been elucidated in detail over the last two decades (Laudet and Gronemeyer, 2001). Now, nuclear receptors have become one of the most significant molecular targets for drug discovery in the fields of cancer, autoimmune diseases, metabolic syndrome, and others (Moore et al., 2007). In this chapter, structures and functions of nuclear receptors are summarized, as well as the medicinal chemistry of retinoids are reviewed to provide a typical example of drug discovery research targeting nuclear receptors.

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

275

276

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.1

9.2

Mechanism of hydrophobic and hydrophilic signaling molecules.

NUCLEAR RECEPTOR SUPERFAMILY AND THEIR FUNCTIONS

Nuclear receptors are ligand-inducible transcription factors (Laudet and Gronemeyer, 2001; Moore et al., 2007). Among various transcriptional factors, only nuclear receptors have the binding sites to the specific small molecules. The endogenous ligands for nuclear receptors are hydrophobic hormones, such as steroid hormones, thyroid hormones, activated vitamin A and D, and some metabolic signals. As listed in Table 9.1, there are 48 members of nuclear receptor family reported to be encoded in the human genome (Nuclear Receptor Nomenclature Team, 1999) and are characterized by their similarity in structures and functions. Some of them have subtypes with different amino acid sequences, but with functional relevancy to each other. As shown in Fig. 9.2, the structures of nuclear receptors are classified into functional domains A–F, which include the DNA binding domain (DBD, C region) and the ligand binding domain (LBD, E region). Thus, nuclear receptors have their own specific ligands and DNA binding site, and consequently regulate their specific target genes ligand dependently. DBD is about 80 amino acids, and it is the most highly evolutionally conserved region. DBD is composed of two zinc finger motifs and binds to the specific DNA sequences known as hormone response elements. Nuclear receptors can be classified by their binding features to DNA. One class is steroid hormone receptors that form homodimers and bind to an inverted repeat sequence. There are five

NUCLEAR RECEPTOR SUPERFAMILY AND THEIR FUNCTIONS

TABLE 9.1

277

Nuclear Receptor Superfamily Identified in the Human Genome

Name Steroid hormone receptors ERa, b GR MR

Unified Nomenclature NR3A1, NR3A2 NR3C1 NR3C2

PR AR RXR-related receptors TRa, b RARa, b, g

NR3C3 NR3C4

RXRa, b, g PPARa, d, g LXRa, b FXR VDR

NR2B1, NR2B2, NR2B3 NR1C1, NR1C2, NR1C3 NR1H2, NR1H3 NR1H4 NR1I1

PXR/SXR CAR Other receptors (orphan receptors) Rev-erba, b RORa, b, g HNF4a, g TR2, 4 TLX PNR

NR1I2 NR1I3

COUP-TFI, II, III ERRa, b, g NGFIBa, b, g SF1 LRH1 GCNF1 DAX1 SHP

NR2F1, NR2F2, NR2F6 NR3B1, NR3B2, NR3B3 NR4A1, NR4A2, NR4A3 NR5A1 NR5A2 NR6A1 NR0B1 NR0B2

NR1A1, NR1A2 NR1B1, NR1B2, NR1B3

NR1D1, NR1D2 NR1F1, NR1F2, NR1F3 NR2A1, NR2A2 NR2C1, NR2C2 NR2E1 NR2E3

Endogenous Ligand Estradiol (estrogen) Cortisol (glucocorticoid) Aldosterone, deoxycorticosterone Progesterone (progestin) Testosterone (androgen) Thyroid hormone All-trans-retinoic acid (retinoid) 9-cis-Retinoic acid Fatty acids Oxycholesterols Chenodeoxycholic acid 1a,25-Dihydroxyvitamin D3 Xenobiotics Xenobiotics

(Retinoid-related receptor) Palmitic acid (Testis receptor) (Photoreceptor-specific receptor) (Estrogen-related receptor)

classes of steroid hormones: androgens, estrogens, progestines, glucocorticoids, and mineralocorticoids (Fig. 9.3). These hormones are biosynthesized from cholesterol by chemical modifications of the A-ring and the side chain. Another class is nuclear receptors that form heterodimers with retinoid X receptor (RXR) and bind to the direct repeat sequence of DNA. This class includes nuclear receptors for some hormones such as retinoid (RARs), vitamin D (VDR), and thyroid hormone (TRs), as well as those acting as metabolic sensors such as peroxisome proliferator-activated

278

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.2 (a) Structures and functions of nuclear receptors. (b) Dimer formation and binding to hormone response element (HRE).

receptors (PPARs), liver X receptors (LXRs), and farnesoid X receptor (FXR). In binding to DNA, each heterodimer distinguishes its own response element depending on the spacer between the direct repeat sequences. Interestingly, RXR-selective agonists have different functions in the heterodimers, depending on the heterodimer partner receptors, and this topic is discussed more later in the chapter. Besides homodimeric steroid hormone receptors and nuclear receptors forming RXR heterodimers, there are orphan receptors, whose endogenous ligands are unknown, binding to DBD as monomers or other type of heterodimers.

FIGURE 9.3

Structures of hydrophobic hormonal molecules and their nuclear receptors.

AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS

279

FIGURE 9.4 Mechanism of ligand-dependent transactivation of nuclear receptors. (See the color version of this figure in the Color Plates section.)

The C-terminal LBD (about 250 amino acids) binds to the specific ligand and the coactivator proteins, and also has significant roles in receptor dimerization. Therefore, the LBD is significant for the ligand-dependent transcription activation and named as the activation function-2 (AF-2) domain. On the contrary, the N-terminal A/B region is significant for constitutive transcription activation, known as AF-1 domain. Recent progress in the molecular biology of transcription factors, X-ray crystallographic studies of nuclear receptor LBDs, and the development of various types of synthetic ligands clarified the ligand-dependent transcription activation of nuclear receptors (Bourguet et al., 1995). The typical mechanism is illustrated in Fig. 9.4. The LBD forms a-helix triple sandwich structures with 11–13 a-helices and 2–4 b-strands. Binding of the ligand to LBD causes the drastic change in the shape of receptors, especially in the C-terminal helix 12 (H12), where it changes its position and orientation, as though putting a lid on the ligand binding pocket. Thus, in the apo form, the gene transcription is suppressed by the binding to the corepressor proteins, while the activated holoreceptor with H12 folding removes the corepressors and instead binds to coactivators to elicit the transcription of the target genes.

9.3 AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS Early on, steroid hormones were studied from the viewpoint of endocrinology, and thus natural and synthetic analogues of steroid hormones have been developed prior

280

NUCLEAR RECEPTOR DRUG DISCOVERY

to the detailed knowledge of their nuclear receptors. After the discovery of the nuclear receptors, the researchers could search the synthetic analogues by screening the binding affinity or regulatory ability of these receptors. The function of nuclear receptors, especially of orphan receptors, often could not be clarified until their specific ligands were found. Some compounds were proven to be specific ligands for nuclear receptors after they had been in clinical use. For example, fibrates as hypolipidemic agents and thiazolidinediones as hypoglycemic agents were ligands for PPARa and g subtypes, respectively (Miller and Etgen, 2003; Henke, 2004). PPARa plays a pivotal role in the uptake and oxidation of fatty acids and in lipoprotein metabolism, while PPARg is predominantly expressed in adipose tissue and has key roles in adipocytes. Since these first-generation ligands have some disadvantages in their clinical use, novel PPAR ligands with unique biological profile have been developed, including subtype-selective and subtype-dual agonists, partial agonists, and antagonists. The development of molecules with a variety of mechanisms is important since the regulation of these nuclear receptors related to metabolism (PPARs, LXRs, FXR, etc) will have clinical utility in diseases with metabolic abnormalities. One of the efficient methods for a ligand search is the combination of chemical library and high-throughput screening, and some ligands with unique structures have been found this way. The elucidation of the crystal structures of nuclear receptor LBDs and the ligand-dependent activation mechanism enabled the design of selective ligands of nuclear receptors. The analysis of the interaction of the ligand with the amino acid residues of the receptors in the ligand binding pockets affords useful information for molecular design of ligands and sometimes clarifies the selectivity between nuclear receptors or the subtype selectivity. Further, the method of computer-assisted drug design combined with the proper chemical database is also a powerful tool for ligand searching for novel structures. Such an in silico method (virtual screening) is illustrated in Fig. 9.5. In the case of the receptors whose crystal structures are unknown, the three-dimensional protein structures were constructed by homology analysis. The possible three-dimensional structures of each compound in the database were examined in the ligand binding pocket of the receptor LBD to give some ligand candidates. Example 1 in Fig. 9.5 is the focused library study for the RXR ligand search. In this case, thiazolidinedione moiety is fixed as the key functional group that corresponds to the carboxylic acid of 9-cis-retinoic acid (9cRA, endogenous RXR agonist), and the proper aromatic skeleton was searched. Thus, from the virtual library that consists of approximately 300,000 molecules bearing different combinations of R1–R5 and X in the generic structure 1, 322 molecules were selected by the in silico screening method. From the ligand candidates, several molecules were further selected by considering their chemical structures (e.g., drug-likeness, stability, and potential structural modification) and then synthesized. Among them, TZ335 exhibited potent and selective RXR agonistic activity. Example 2 shows a similar in silico study by using a database of commercially available molecules. In this method, the researchers do not need to synthesize the molecules. They simply purchase the ligand candidates and examine their biological activities. MAY0282 was obtained as RXR agonist by

AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS

281

FIGURE 9.5 Examples of in silico ligand search.

this method, but it also demonstrated weak RAR antagonistic activity. Thus, the in silico study does not always afford successful results utilizing the present computational techniques. An RAR agonist, Bfx80, with benzofloxane moiety was obtained by the structural modification of MAY0282. Since the carboxylic acid moiety of retinoic acids seems to be necessary to bind to the nuclear receptors, the finding of TZ335 and Bfx80 as RXR and RAR agonists, respectively, is significant. Thus, in silico study is useful if the researchers use it appropriately and carefully, and it leads to a unique structure. Nuclear receptor antagonists have been also developed, and several antagonists, especially for steroid hormone receptors, are in clinical use. For example, estrogen and androgen antagonists are in the clinic for endocrine therapy of breast cancer and prostate cancer, respectively. Several possible mechanisms of nuclear receptor antagonism can be considered for drug intervention that include decreasing the receptor dimerization ability, affecting the interaction with DNA response elements, recruiting of corepressors, and inhibiting the coactivator binding. Since the ligand-induced repositioning of H12 is crucial for the nuclear receptor activation, as depicted in Fig. 9.4, the strategy of targeting the change in the interaction between ligands and H12 is useful for the molecular designs of AF-2 antagonists. The X-ray crystallographic studies on the antagonist-bound LBD of nuclear receptors afforded the molecular basis of antagonism. For example, the

282

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.6

Structures of ER ligands with unique biological profile.

crystal structures of ERa–LBD complexes with estradiol (endogenous agonist) and hydroxytamoxifen (synthetic antagonist) showed different position and orientation of H12 (Brzozowski et al., 1997; Shiau et al., 1998). Thus, hydroxytamoxifen shown in Fig. 9.6 has a bulky 4-(dimethylaminoethoxy)phenyl group that precluded H12 acting as the lid of the ligand binding pocket and allowed an alternative stable position of H12. There is another type of antagonist without a bulky substituent that can disturb the proper H12 position. For example, aldosterone and progesterone act as mineralocorticoid receptor (MR) agonist and antagonist, respectively. Progesterone is structurally smaller than aldosterone and can bind to the ligand binding pocket of MR. The docking study indicated that aldosterone interacts strongly with both helices 3 and 5 via several hydrogen bonds, while progesterone can interact only with helix 5. Thus, progesterone cannot tightly position these helices near the ligand binding pockets, and consequently it induces a different stable receptor conformation from that of the active form (Souque et al., 1995). Similar antagonism was observed in the estrogen receptor (ER) antagonist 5R,11R-diethyl-5,6,11,12-tetrahydrochrysene-2,8-diol (THC). THC acts as an agonist for ERa subtype, while it acts as an antagonist for ERb subtype. Crystal structures of THC-bound ERa and b LBDs were studied. In the case of THC-bound ERa LBD, H12 was located at the proper position and the coactivator protein was able to bind to the receptor. However, while THC did not sterically disturb the proper position of H12, it did induce the inactive form of ERb LBD by stabilizing nonproductive conformations of key residues near the ligand binding pocket (Shiau et al., 2002). Such antagonists that lack a bulky substituent are called passive antagonists. The biological activities of nuclear receptor hormones were determined not only by their binding affinities to the receptors, but also by the conformational features of the ligand–receptor complexes and their interactions with corepressors/coactivators. This information indicated that unique ligands with some specificity in receptors, subtypes, or interaction with other proteins (e.g., heterodimer partners or cofactors) can be developed by a combination of several bioassay systems, although the logical molecular design of such compounds is rather difficult. Recently, some ligands for nuclear receptors showed tissue-specific biological properties. The first example was obtained in ER ligands. Estrogen and ERs regulate a wide range of developmental and physiological responses, including reproductive functions, cardiovascular modulation, and bone

AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS

283

density (Nilsson and Gustafsson, 2002). Endogenous and synthetic ER agonists are used in hormone replacement therapy of postmenopausal women to prevent osteoporosis. However, estrogen therapy has been associated with an increased risk of breast and endometrial cancers, and dangerous blood clotting. Estrogen antagonists can inhibit the growth of estrogen-dependent breast cancer in both pre- and postmenopausal women. Tamoxifen is a first-line treatment of ER-positive breast cancer. Tamoxifen is a prodrug, which is metabolized to 4-hydroxytamoxifen and binds to ERs with high affinity. Although tamoxifen acts as ER antagonist in breast cancer, it acts as agonist in uterine tissue. Thus, tamoxifen is the first synthetic nuclear receptor ligand that exhibited tissue-dependent activities, and such molecules related to ER are called selective ER modulators (SERM). The mechanism of tissue selectivity may arise from the conformation of SERM-bound ER (especially the position and orientation of H12 folding) and the tissue-dependent properties of corepressors/coactivators, such as expression ratio and nuclear receptor association (Shang and Brown, 2002). Raloxifene is a second-generation SERM, developed for the treatment of breast cancer, while it acts as ER agonist in bone tissue. Thus, raloxifen has significant osteoporosis protective effects and is now in clinical use for the prevention of osteoporosis (Turner et al., 1994). Since the tissue selectivity of SERMs was clarified, various nuclear receptor ligands with such activities have been reported and are classified as selective nuclear receptor modulators. It is unclear whether the word ‘‘modulator’’ is proper to the ligands with such biological properties, but it is significant to develop the tissue-selective ligands with different therapeutic profile from the natural and conventional ligands for nuclear receptors of both hormones and metabolic sensors. 9.3.1

AR Antagonists Effective Toward Mutated Receptors

Androgen is a steroid hormone required for the development and maintenance of masculine characteristics in vertebrates, including prostate development and normal prostate function (Roy et al., 1999). Endogenous ligands for androgen receptor (AR) are testosterone and its more potent metabolite dihydrotestosterone shown in Fig. 9.3. Approximately 80–90% of prostate cancers are androgen dependent at initial diagnosis, and endocrine therapy of prostate cancer is directed toward the reduction of serum androgens and the inhibition of AR functions (Denis and Griffiths, 2000). The AR antagonists so far known can be classified into two structural types: steroid and nonsteroid types (Gao et al., 2005; Gao and Dalton, 2007). Steroidal analogues, like cyproterone acetate, often exhibited cross-reactivity with other steroid hormone receptors, and it is generally difficult to distinguish effects on androgenic activities from those on anabolic activities. To overcome the disadvantages of steroidal analogues, a number of nonsteroidal AR antagonists have been developed. Among them, flutamide and bicalutamide (Casodex) are clinically used in the treatment of prostate cancer shown in Fig. 9.7. These AR antagonists have a common anilide structure with electron-withdrawing substituents and are

284

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.7 (a) Conventional nonsteroidal/anilide-type AR antagonists. (b) Design of novel AR antagonists, ISOP and PYROP.

useful for the treatment of prostate cancer in the early stage. Prostate cancer often advances to a ‘‘hormone-refractory’’ state in the continued therapy using AR antagonists. Mutation of AR is considered as one possible reason for the hormone-refractory state. Thus, new types of nonsteroidal AR antagonists effective toward mutated AR should be developed. As described above, some antagonists may disturb the proper folding of H12. AR antagonists such as flutamide and hydroxyflutamide are rather small in molecular shape, and the docking studies indicated that these antagonists may induce the folding of H12, but different from that by an AR agonist in the position and the orientation. Thus, such antagonists are classified as misfolding inducers of H12, as summarized in Fig. 9.8. Some mutated ARs are considered to be constitutively active, and these misfolding inducer-type antagonists elicit further activation of these mutated ARs. Therefore, the antagonists that can prohibit the H12 folding itself would be effective toward mutated ARs. As H12 folding inhibitor-type antagonists of AR, isoxazolone derivatives were reported in the literature (Ishioka et al., 2003). Compound 2 was obtained as an AR ligand candidate by in silico screening of a database of commercially available molecules. The binding structure of 2 to AR suggested that the introduction of a bulky substituent on the isoxazoline

AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS

FIGURE 9.8

285

Design of mutated AR antagonists.

ring would affect the H12 folding. Various isoxazolone derivatives were synthesized from 4-aminobenzaldehydes and b-ketoesters. Among them, 3-phenylisoxazolones (ISOP series) showed high binding affinity for wild-type AR and potently suppressed the growth of an androgen-dependent Shionogi carcinoma cell line SC-3. Further, ISOPs acted as antagonists toward the mutant AR (T877A) in the LNCaP cell line, although its potency is not high. The results indicated that the strategy to develop H12 folding inhibitors as antagonists of mutated AR might be effective. Successful of ISOP compounds led to the more potent AR antagonists with different structures. The molecular design leading to a novel structure is shown in Fig. 9.7b. From the structure–activity relationships of known AR antagonists and ISOP derivatives, a general structural feature for AR antagonistic activity consists of two aromatic rings connected by a suitable linking group, with one benzene ring bearing a dialkylamino group. Based on the successful introduction of a heterocyclic ring in the ISOP compounds, one benzene ring was replaced with a heterocyclic pyrrole moiety, and instead, the spacer group between two aromatics was replaced with much simpler linking group, such as amino, amido, and sulfonamido. Then, the terminal pyrrolidine ring of ISOP compounds was connected through a carbonyl group to the pyrrole ring, since the amide structure is expected to afford superior chemical stability and pharmacological behavior. In this manner, several 4-substituted pyrrole-2-carboxamides were synthesized as AR antagonist candidates (Wakabayashi et al., 2005). Among synthesized compounds, the compounds bearing an amino group as the spacer showed potent inhibitory activity of SC-3 cell growth, and PYROP displayed high binding affinity (Ki 5.0 nM) for wild-type AR, which is comparable to hydroxyflutamide (Ki 3.0 nM) and bicartamide (Ki 10.0 nM). Further, PYROP exhibited antagonistic activity against the mutant T877A AR, having potency comparable with that of (R)-bicalutamide. Recently, bicartamide acted as an agonist toward W741L/W741C AR mutants (Bohl et al., 2005). Although it is unknown whether PYROP antagonizes W741L/ W741C AR or other AR mutants, PYROP is considered to be promising as a scaffold for the development of novel agents for anti-AR therapy of prostate cancer.

286

9.3.2

NUCLEAR RECEPTOR DRUG DISCOVERY

VDR Agonists and Antagonists

Vitamin D is an important regulator of calcium homeostasis, bone development and metabolism, cell growth and differentiation, and the immune system (Feldman et al., 2004). Most of these actions are mediated by the active metabolite 1a,25-dihydroxy vitamin D3 (1,25-VD3) and its specific nuclear receptor vitamin D receptor (VDR). On the basis of pleiotropic activities of vitamin D, VDR agonists have clinical potential in the field of cancers, dermatological diseases, and autoimmune diseases. One significant disadvantage of vitamin D3 in clinical use is the hypercalcemic side effect. More than 3000 vitamin D analogues, representatives are shown in Fig. 9.9, have been synthesized so far, with the aim to reduce the side effects. The compounds developed in the early years are closely related derivatives with a modified alkyl side chain, including seven vitamin D drugs. For example, calcipotriol (MC903, Dovonex) has limited calcemic side effects and has become a first-line drug for the topical treatment of psoriasis. 1a,25-Dihydroxy-19-nor-vitamin D3 is a highly stable analogue with the modification of the conjugated triene structure of 1,25-VD3, and it is a potent inducer of cell differentiation with low calcemic activity. During the last decade, various analogues with the A-ring modification have been synthesized (Yamada et al., 2003). The reported crystal structures of VDR with bound agonists also supported the result that the modification at 2-position of A-ring caused the increase of the binding affinity of the ligands (Hourai, et al. 2006). 2-Methylene1a,25-dihydroxy-19-nor-vitamin D3 (2MD) stimulated bone formation in ovariectomized rats without toxic hypercalcemia (Shevde et al., 2002). 1a,25-Dihydroxy-2b(hydroxypropoxy)vitamin D3 (ED-71) also increases bone mass. Both compounds are therapeutic candidates for the treatment of osteoporosis and are now in clinical trials. Most of the vitamin D analogues have a seco-steroidal skeleton, like the natural 1,25-VD3 structure. As observed in successful developments of various nonsteroidal

FIGURE 9.9

Structures of VDR agonists and antagonists.

AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS

287

AR or ER ligands, non-seco-steroidal VDR ligands might display different biological profiles and pharmacological behaviors than those of seco-steroidal compounds. By screening the chemical library using a VDR transactivation assay, one unique compound with a bisphenol structure (LG190090) was obtained as VDR-dependent bioactive molecule. Studies on the structure–activity relationship and the detailed biological activities afforded the potent VDR agonist LG190178 shown in Fig. 9.10. LG190178 has various vitamin D activities, including the inhibition of proliferation of human cancer cells and primary human keratinocytes, and monocytic differentiation of human promyelocytic leukemia cells HL-60. Further, LG190178 induced more potently the expression of kidney 24-hydroxylase, one of the VDR target genes, in rodents, with less increase in serum calcium, than 1,25-VD3 (Boehm et al., 1999). LG190178 is a bisphenol derivative, and several aza analogues, that is, aniline derivatives, were also examined (Hosoda et al., 2006). In this study, the replacement of one phenol moiety with an aniline moiety is the key reaction to develop various unsymmetrical analogues as summarize in the synthetic scheme in Fig. 9.10. Normal functional transformation of the hydroxyl group of 6 to an amino group requires four steps and the total yield is only 35%. When compound 6 was reacted with ortho-toluidine (neat) at 180
C, the reaction afforded directly the monoaniline derivative 10 in a 58% yield. Among the synthesized mono- or bisaniline derivatives of LG190178, compound (R, S)-5 exhibited higher activity than LG190178 in VDR

FIGURE 9.10 Non-seco-steroid-type VDR agonist LGD190178 and its aza analogs. Key intermediate 10 for the synthesis of unsymmetrical aza analogues 4 and 5 was prepared from bisphenol 6 by one step.

288 TABLE 9.2

NUCLEAR RECEPTOR DRUG DISCOVERY

Activity of Aza Analogues of LG190178

Structurea (R, R)-3 (S, R)(R, S)(S, S)(R, R)-4 (S, R)(R, S)(S, S)(R, R)-5 (S, R)(R, S)(S, S)(R, R)-LG190178 (S, R)(R, S)(S, S)Hydroxyflutamide

VDR Binding Affinity Ki (nM) 190 580 120 350 340 1100 220 420 150 180 9.5 20 320 740 12 11 NTb

HL-60 Cell Differentiation EC50 (nM) 48 170 6.6 43 320 770 47 110 30 55 4.1 16 26 120 8.6 23 NTb

AR Binding Affinity Ki (nM)

SC-3 Growth Inhibition IC50 (nM)

1100 1100 2300 540 2500 1100 1900 400 1200 1200 910 730 1000 1100 1000 1100 940

7.1 32 4.1 22 36 290 19 55 5.0 17 1.3 4.7 NTb NTb NTb NTb 180

a See Fig. 9.10. In the configurations of 3–5 and LG190178 shown in parenthesis, the first and second symbols correspond to the chiral carbons of 2-hydroxy-3,3-dimethylbutyl and 2,3-dihydroxylpropyl groups, respectively. b Not tested.

binding affinity and HL-60 cell differentiation assay, which is listed in Table 9.2. The stereochemistry at both chiral carbon atoms is significant for the activity. Interestingly, these compounds have weak AR binding affinity and inhibited androgendependent SC-3 cells, like AR antagonists. The VDR agonists also inhibited SC-3 cells or other androgen-dependent prostate cancer cells. At present, the mechanism of (R, S)-5 with the VDR and AR dual ligand activity is unknown. These compounds could become lead compounds for VDR/AR-specific activity or the combination of dual ligands for clinical utility toward prostate cancer or other VDR-related diseases. In comparison to the various derivatization of VDR agonists, only a few VDR antagonists have been developed. The VDR antagonists have potential clinical use for the treatment of metabolic bone disease, such as Paget’s disease (Ishizuka et al., 2005). There are two reported types of VDR antagonists, that is, ZK168281 and TEI-9647, bearing the terminal ester group and 26,23-lactone structure on the side chain, respectively. The mechanistic studies indicated the difference in the action of these two antagonists, and ZK168281 is a more complete antagonist that exhibited potent antagonism in various assay systems, while TEI-9647 is a rather selective antagonist (Toell et al., 2001). ZK168281 and TEI-9647 seem to act as H12 folding inhibitor-type and passive antagonists, respectively (Yoshimoto et al., 2008), although there are some experiments that

AGONISM AND ANTAGONISM IN NUCLEAR RECEPTOR FUNCTIONS

289

suggest that the a,b-unsaturated ketone moiety acted as Michael reaction acceptor. TEI-9647 can potentially form a covalent bond in the ligand binding pocket of VDR (Takenouchi et al., 2004). Further, TEI-9647 exhibited species specificity and acted as an antagonist and an agonist toward human and rat VDR in the transactivation assay. Recently, the third type of VDR antagonists, DLAMs, were reported (Nakano et al., 2006). DLAMs have a lactam ring on the side chain, and the bulkiness of N-substituent of the lactam ring can interfere with the proper H12 folding. In the first report on DLAM compounds, they were synthesized by the convergent synthetic method of a coupling reaction of a CD ring bearing the lactam ring with an A-ring precursor (Fig. 9.11a). This synthetic strategy is flexible for the modification at the A-ring, while the coupling and the subsequent deprotection reactions did not provide

FIGURE 9.11

Synthetic methods of DLAM compounds.

290

NUCLEAR RECEPTOR DRUG DISCOVERY

consistent results. Therefore, an alternative synthetic method suitable for modification at N-substituent on the lactam ring and synthesis of substantial quantities was developed (Fig. 9.11b). As expected, the activities of DLAM compounds depended on the N-substituent and the stereochemistry of the lactam ring. Thus, (23S, 25S)-N-benzyl (DLAM-1P) and N-phenethyl (DLAM-2P) derivatives exhibited potent VDR antagonistic activity. A docking study of DLAM-1P suggested that the side chain of DLAM-1P located at the position of H12 in the VDR– LBD. Thus, DLAM-P1 would act as the H12 folding inhibitor-type antagonist, and actually antagonized the activation of both human and rat VDRs induced by 1,25-VD3. Similarly, the VDR antagonist bearing bulky adamantine moiety was developed. Interestingly, the adamantine derivative exhibited cell-type-selective activity, and it may be classified as a selective vitamin D modulator (Inaba et al., 2007). Besides vitamin D actions, VDR can bind to lithocholic acid and act as the sensor for metabolism of bile acids (Makishima et al., 2002). The structural development of lithocholic acid as VDR ligand, besides the detailed studies on the physiological roles of VDR in bile acid metabolism, is now in progress. 9.3.3

Carboranes as Novel Hydrophobic Pharmacophores

Endogenous ligands for nuclear receptors are small hydrophobic molecules. Steroid hormones and vitamin D3 have hydrophobic steroidal and seco-steroidal skeletons, respectively. LXR and FXR bind to oxysteroid and bile acid, respectively. Retinoid receptors and some nuclear receptors for metabolic sensors bind to carboxylic acids attached to the hydrocarbon skeletons, such as retinoic acid and fatty acids. As observed in the clinical utilities of nonsteroidal ligands for steroid hormone receptors, the development of a novel hydrophobic pharmacophore would be important for medicinal chemistry of nuclear receptors. From this consideration, unique ligands bearing a boron cluster (carborane) have been developed. Carborane (C2B10H12,) is a boron cluster with icosahedral geometry in which two carbon and 10 boron atoms are hexacoordinated as shown in Fig. 9.12. There are three isomers, ortho-, meta-, and para-carboranes, corresponding to the possible positions of the two carbon atoms. Carboranes have attracted attention as an efficient tool for boron neutron capture therapy (BNCT) of cancer in the field of medicinal chemistry (Soloway et al., 1998). In most of the carborane derivatives examined so far, the carboranes are linked to less important sites of bioactive substances. Thus, carborane is regarded only as boron carrier. Another chemical property of carborane is its bulkiness and exceptional hydrophobicity (Endo et al., 2003a), which is assumed to be the proper hydrophobic pharmacophore of nuclear receptor ligands. First, carborane was introduced to the hydrophobic portion of the aromatic retinoid Am80, described later in this chapter. The replacement of the hydrophobic cyclic alkyl group of Am80 with ortho-carborane, yielding 11, caused a decrease of retinoidal activity. Considering the bulkiness of carborane ring, BR403, having a shorter linking group than 11, was synthesized and showed similar potent RAR agonistic activity as retinoic acid, an endogenous ligand (Endo

291

FIGURE 9.12 (a) Structures of carboranes. (b) Design of carborane derivative BR403 with RAR agonistic activity. (c) Various nuclear receptor ligands with a carborane moiety.

292

NUCLEAR RECEPTOR DRUG DISCOVERY

et al., 2001a). The hydrophobic part of RAR ligand structures interact with H12 of RAR LBDs. It is significant that a bulky carborane can be introduced into this position without loss or change of activity. The computational docking study indicated that BR403 could bind to the ligand binding pocket of RARa, as well as Am80. Since carborane has a rigid icosahedral structure, and plural substituents can be placed in precisely defined spatial positions and orientations, it is expected to be widely useful as a hydrophobic pharmacophore. Indeed, several nuclear receptor agonists and antagonists have been developed recently, including ER and AR antagonists as shown in Fig. 9.12c (Endo et al., 2001b; Fujii et al., 2005). Interestingly, BE120 is 10 times more active than estradiol in ERa binding affinity, and BE360 exhibited SERM activity, acting as an ER agonist in bone tissue without affecting the uterine tissue (Endo et al., 2003b). In these steroid analogues, carborane mimics the CD ring of steroid skeleton. This result means that carborane could be introduced to seco-steroid structure of vitamin D3. Several vitamin D analogues with a para-carborane instead of the CD ring of 19-nor-1a,25-dihydroxyvitamin D3 were synthesized. Among them, compounds 12a and 12b are potent as lead compounds in the HL-60 cell differentiation assay and VDR binding affinity. Further structural modification afforded non-seco-type vitamin D analogues such as 13. Carborane itself is very stable to various reagents, and therefore various analogues can be synthesized easily. It is interesting that both the hydrophobic and electronic properties of carborane derivatives varied, depending on the position of substituents and is summarized in Fig. 9.13. Eight isomers of para-carboranylphenols have been synthesized. Among them, compounds having a carborane on the carbon atom are the most hydrophobic and more hydrophobic than an adamantyl group.

FIGURE 9.13 Hydrophobicity of carborane moieties. The code name of each compound consists of the carborane isomer (o, m, and p) and the position of phenyl substituent on the carborane.

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

293

para-(Carboran-9-yl)-phenol is the least hydrophobic and less hydrophobic than a cyclohexyl group. There are two orders of magnitude between the most and least hydrophobic carboranes. In electronic properties, carboranes on the carbon atom of phenol moiety are considered electron withdrawing, comparable to halogen atoms. The results indicated the usefulness of carborane as hydrophobic pharmacophore, and several potent ligands for nuclear receptors have been developed. Besides their biological functions as agonists/antagonists, these carborane derivatives may have useful applications for BNCT targeting nuclear receptors.

9.4 9.4.1

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS Retinoid and Their Nuclear Receptors

Retinoids are defined as biological isosteres of all-trans-retinoic acid (ATRA), an active metabolite of vitamin A shown in Fig. 9.14. They modulate a wide range of biological events, including cell differentiation, proliferation, apoptosis, and morphogenesis in vertebrates (Sporn et al., 1994). Among various nuclear receptor ligands, retinoids are unique. There are two classes of retinoid nuclear receptors, RARs and RXRs, both having three subtypes (a, b, and g). Endogenous ligands for RARs and RXRs were identified as ATRA and its 9-cis-isomer (9cRA), respectively. Since 9cRA can bind also to RARs with as high affinity as to RXRs, 9cRA is a pan-agonist for all six retinoid nuclear receptors. Most of the retinoidal activities are elicited by the binding of retinoids to the RAR site of RXR–RAR heterodimers. RXRs are the silent partners of RARs, and RXR agonists alone cannot activate the RXR–RAR heterodimers, though RXR agonists allosterically increase the potencies of RAR ligands. Besides the so-called retinoidal activities, RXRs play significant roles in nuclear receptor actions by heterodimerizing with various nuclear receptors, such as VDR, TR, PPAR, LXR, and FXR. Therefore, RXR ligands can regulate these nuclear receptor functions. Originally, retinoids were defined as substances related to vitamin A (retinol). They are chemically diterpenoids derived from a monocyclic parent compound containing five carbon–carbon double bonds and a terminal functional group by IUPAC–IUB. The retinoid developed in the early years could be classified as retinoid by this definition. However, development of novel retinoids without the polyenecarboxylic acid structure made the chemical definition rather meaningless. The biological definition of retinoid was proposed as a substance that can elicit specific biological responses by binding to and activating a specific receptor or a set of receptors. It is significant that the definition was proposed before the discovery of retinoid nuclear receptors, while specific activities of ATRA made the researchers postulate the existence of specific receptors. Although the chemical definition of retinoid is still important, the biological definition is the most practical in the field of medicinal chemistry. ATRA is an endogenous substance of critical importance for growth in animals, and therefore it is important to find novel retinoids with potential clinical applications,

294

FIGURE 9.14

(a) Metabolism of vitamin A. (b) Classical retinoids with a hydrocarbon skeleton.

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

295

since the direct application of the endogenous substances in clinical therapy is often accompanied by hypercrinism or hypervitaminosis, in analogy with steroid hormones. Modern medicinal chemistry of retinoids started in the early 1970s in the field of cancer chemoprevention and clinical treatment of dermatological diseases. Since the potency of ATRA had been confirmed in therapeutic experiments on skin papilloma and carcinoma of animals, a number of polyenecarboxylic acids have been synthesized and their activity in an in vivo tumorgenesis model was examined. The mechanisms of action of retinoids, including the existence of their specific receptors, are unknown at this stage, and therefore simple chemical modification on the structure of ATRA have been examined. Structurally, ATRA consists of three parts, the cyclohexenyl ring, the polyene chain, and the terminal polar group, and various chemical modifications were performed at each part. A unique approach is to use partial structures of natural carotenoids, 3,30 -dihydroxyisorenieratene and an oxidative product of b-carotene. The cyclohexenyl ring of ATRA that was derived from b-carotene was replaced with the terminal groups of these carotenoids. Thus, aromatic retinoids 14 and 15 with the same polyene chains as ATRA were synthesized and evaluated by using the therapeutic index defined in terms of the ratio in dose causing hypervitaminosis compared with a 50% reduction of papillomas. These compounds were particularly active with 10 times more favorable therapeutic indices than ATRA, and the ethyl ester of 15 (etretinate) is now clinically used in the systemic treatment of psoriasis (Ellis and Voorhees 1987). The introduction of a benzene ring into the side chain of ATRA was also a successful approach for obtaining new derivatives. A cyclopropane ring can also be introduced as a trans- or cis-olefin equivalent. These compounds are more stable than polyenes and are regarded as conformationally restricted analogues of the olefinic bonds. For example, 16 has a benzene ring corresponding to the (11E, 12sZ, 13E)-diene system, and it exhibited potent activity in the tracheal organ culture assay. This finding opened up a new structural aspect in the design of novel retinoids. Thus, TTNPB with a stilbene skeleton and its naphthalene analogue TTNN showed very potent activity in the in vivo papilloma assay. These aromatic retinoids (called arotinoids) also had potent anticancer activity in experimental animals, while their activities in HL-60 cell differentiation assay is equipotent to that of ATRA. The strong activity of the arotinoids may reflect the stable and/or lipophilic nature, resulting in a long duration of in vivo plasma levels, or some other mechanisms besides retinoid actions. The introduction of an aromatic ring into ATRA structure was successful to obtain potent retinoids, but the principal issue remains their high toxicities. The intrinsic activity of such a hormonal substance cannot be altered easily, but pharmacodynamic or pharmacokinetic parameters can more readily be modified. A substantial structural modification, not a minor change, is often necessary to be effective. Successful molecular design of a novel structure based on lead compounds should include skeletal changes in order to obtain different physicochemical properties. ATRA has several disadvantages in the chemical structure and properties. First, the unstable polyene structure of ATRA and conventional analogues, described in the previous section, made them difficult to treat or use

296

NUCLEAR RECEPTOR DRUG DISCOVERY

as drug candidates or biological tools. ATRA is unstable and reactive toward heat, light, oxidation, and various chemical reagents. Second, ATRA can be easily isomerized and affords a mixture of isomers in solution, even when the researchers treat it very carefully. Third, the hydrophobic hydrocarbon skeleton of retinoids including etretinate and TTNPB is disadvantageous in retinoid therapy, leading to long-lasting hypervitaminosis A and teratogenicity. Thus, a stable benzanilide4-carboxylic acid skeleton with a polar amide linkage was designed. The retinoidal activity of the designed terephthalic monoanilides was first evaluated in terms of the potency to induce the differentiation of human promyelocytic leukemia cell line HL-60 to mature granulocytes listed in Table 9.3 (Kagechika et al., 1988). Compounds (Am00, Am10, and Am25), having no or a small alkyl group, are inactive in this assay, while the introduction of a bulky isopropyl or tert-butyl group caused HL-60 cell differentiation activity such as the results for compounds Am32 and Am40. More potent activities were observed in the dialkyl-substituted compounds, Am55 and Am66, with two bulky meta-alkyl groups. Although a para-alkyl substituent (R3) seems to be less effective, Am68 having one meta- and one para-isopropyl groups is more active than Am66 bearing two meta-isopropyl groups. The higher potency of Am68 can be interpreted in terms of the steric interaction between the neighboring two alkyl groups. The rotation of the isopropyl groups in Am66 is nearly free, while that in Am68 should be restricted. Among the conformationally fixed analogues of Am68, Am80, having a tetramethylbutano group like TTNPB, was considered to correspond to the preferred conformation of Am68 and is more active TABLE 9.3 Structure–Activity Relationships of Benzanilide-4-Carboxylic Acids in HL-60 Cell Differentiation Assay

R2

COOH

R1 H N O R5

R3 R4

Code #

R1

R2

R3

R4

R5

EC50 (M)

Relative activity

ATRA Am00 Am10 Am20 Am25 Am30 Am32 Am34 Am40 Am50 Am55 Am66 Am68

H H H H H H isoPr H H H H H

H Me Et Et H isoPr H tertBu H tertBu isoPr isoPr

H H Et H isoPr H H H tertBu H H isoPr

H H H H H H H H H tertBu isoPr H

H H H H H H H H H H H H

2:4  109 Inactive Inactive 2:6  107 Inactive 9:1  107 6:8  107 Inactive 7:0  107 >106 3:6  108 3:8  108 2:1  109

100 — — 0.77 — 0.098 0.13 — 0.098 <102 15 4.2 140

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

297

FIGURE 9.15 Design of Am80. The EC50 values in HL-60 cell differentiation assay are shown in parentheses.

than Am68 or other fixed analogues such as Am140 and Am150 with a bicyclic alkyl group. The evolution of the SAR is illustrated in Fig. 9.15. Structural relationships of retinoidal amides show some interesting features and some of these are shown in Fig. 9.16. Am580 is an isomer of Am80, bearing opposite aromatic substitution on the amide bond. While the electronic effects of the amide bond on the aromatic ring or the carboxyl group should be different, Am580 exhibited activity similar to or somewhat more potent than that of Am80. On the other hand, introduction of a methyl group on the amide nitrogen of the potent retinoids Am80 and Am580, yielding Am90 and Am590, respectively, resulted in a 104-fold reduction of activity. Detailed structural investigations using NMR spectroscopy and X-ray crystallography revealed that the dramatic diminution of the activity caused by N-methylation was attributed to conformational changes between secondary and tertiary benzanilides (Kagechika et al., 1989). Potent retinoids, Am80 and Am580, exist in the trans-amide form, while N-methylated amides, Am90 and Am590, exist in the cis-amide form, both in the crystal and predominantly in solution (e.g., >97% cis in CD2Cl2). Thus, the elongated conformation of the secondary amide derivatives is important for retinoid activity, like the trans-stilbene structure of TTNPB. Cis conformational preference is a general property of aromatic N-methylated amides (Fig. 9.16b). Simple benzanilide and acetanilide also change the conformation by N-methylation from trans to cis. Similarly, thioamide, amidine, urea, thiourea, and guanidine have similar steric properties (Tanatani, et al. 1998). Further, such conformational properties of the amide and related compounds often affect the functions of molecules, including proteins, and bioactive molecules, as well as polymers and

298

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.16 (a) Effects of the amide bond structure in retinoidal benzanilides on HL-60 cell differentiation-inducing activity. The EC50 values in HL-60 cell differentiation assay are shown in parentheses. (b) The cis conformational preference of N-methylated amides and related functional groups.

molecular devices. Both trans- and cis-structures of secondary and tertiary amides are useful building blocks for the three-dimensional construction of aromatic molecules. As expected in the molecular design of retinoidal aromatic amides, Am80 and Am580 showed unique pharmacological properties, and they exhibit rapid clearance from the body, compared to conventional highly lipophilic retinoids such as ATRA and TTNPB. The dose of 4 mg of Am80 was administered to two healthy volunteers. Peak plasma concentrations of 16 and 12 mM were obtained after 2–3 h of oral administration, and only a trace amount of Am80 was detected at 24 h. This result is clinically beneficial, since disadvantageous side effects such as hypervitaminosis A or teratogenicity result from the intrinsic activity of retinoids. Further chemical and biological properties will be discussed later in the chapter. In the retinoid structure, an olefinic bond can be replaced with an amide bond. Besides aromatic retinoids, the analogues of ATRA itself with an amide bond were synthesized. Compounds 17 and 18 are dienoyl dienamides bearing the amide bond at the 9,10-position of ATRA (Shimasaki et al., 1995). Dienamides are generally useful precursors for inter- and intramolecular Diels–Alder reactions that can be applied to natural product syntheses, and several synthetic methods have been reported. However, most of them can be applied only to a limited range such as tertiary dienamides or aromatic ring-conjugated systems. The synthetic method developed for dienoyl dienamides 17 and 18 is shown in Fig. 9.17. A Curtius rearrangement of the dienoic

299

FIGURE 9.17

Synthesis of dienoyl dienamide analogues of retinoic acid.

300

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.18 Unsuccessful replacement of olefinic bond of 1a,25-dihydroxyvitamin D3 with an amide bond.

acid azide trapped by methanol and following with an acylation afforded a dienamide with N-methoxycarbonyl group that can be selectively removed by lithium iodide. 1 H-NMR studies showed that secondary dienamide 17 exists in trans-amide form, while tertiary dienamide 18 exists in equilibrium between trans and cis conformers (1:2.7 in CDCl3 at 233 K). Contrary to the potent activities of aromatic amides, Am80 and Am580, these dienamides exhibited only weak retinoidal activity in HL-60 cell differentiation assay. The attempt to replace the olefinic bond of a triene structure of vitamin D3 with the amide bond was performed by the similar strategy as was used with the development of Am80. Thus, the conjugated carboxamides with various N-substituents, instead of A-ring of 1,25-VD3, were designed and are illustrated in Fig. 9.18. However, these compounds did not act as a VDR ligand nor as cell differentiation inducers of HL-60. An amide derivative 19 whose structure is more close to 1,25-VD3 was also synthesized (Suhara et al., 2002). Although 19 has a similar ring system with the A-ring of 1,25-VD3, the binding affinity of 19 is only 3  106 times that of 1,25-VD3. Strict hydrophobic or conformational properties around the triene structure of 1,25-VD3 would be needed for the VDR binding. This outcome may be the reason that only a few non-seco-type VDR ligands have been developed so far, as described above. Although the bioisosteric conversion from the olefinic bond to the amide group in VDR ligand resulted in failure, this methodology, including reverse conversion from some peptides to peptidomimetic alkenes, is generally useful in medicinal chemistry. 9.4.2

Retinobenzoic Acids

The similar activities of Am80 and Am580 suggested that the electronic effects of the linking group between two aromatic rings are not significant, and the elongated conformation is significant. From this viewpoint, various benzoic acid derivatives have been developed as potent retinoids and named retinobenzoic acids, which are depicted in Fig. 9.19. The generic structure of retinobenzoic acid is 20, which consists of three parts: the hydrophobic aromatic ring, the benzoic acid moiety, and the linking group. The linking group X can be varied, for example, to  NHCO , N CH CH  NHCONH ,  OCO ,  N ,  CH ,  COCO , and  COCH . In this region, a rather planar and elongated conformation is important for the activity. As observed in retinoidal amides, N-alkylation of a diphenylurea derivative

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

FIGURE 9.19

301

Retinobenzoic acids: generic structure and representative compounds.

Ur80 also changed the conformation and the activity, and Ur90 with a partially folded (cis, trans) structure is more active than Ur80 with an elongated (trans, trans) structure. N-Methylation at another nitrogen atom of Ur80 yielded Ur100 with (trans, cis) structure and resulted in a decrease of the activity. The conformation of the sulfonamide bond is not planar, unlike that of the amide bond in Am80, and consequently the activity of Sa80 is weaker than that of Am80. Ch55 and Re80 have a longer distance between the two aromatic rings than Am80 but are similar to Ur80. In contrast to the lower activity of Ur80 than Am80, Ch55 and Re80 are more potent retinoids than Am80. The linking group can be a heterocyclic ring fused to the benzene ring, like a flavone derivative Fv80. While the linking group can be varied, the benzoic acid moiety has little scope for modification. The carboxyl group is an essential functional group in retinobenzoic structure 20 and could not be replaced with so-called bioisosteric functional groups, such as sulfo, aminosulfonyl, amidino, or tetrazolyl group, without loss of activity. Functional groups such as esters or hydroxymethyl groups, which can be easily converted to the carboxyl group in the assay conditions, afforded some potency. A unique noncarboxylic acid-type retinoid is a tropolone derivative Tp80 (Ebisawa et al., 2001). Tropolone, 2-hydroxy-2,4,6-cycloheptatrien-1-one, is an isomer of benzoic acid and is a seven-membered, nonbenzenoid aromatic

302

NUCLEAR RECEPTOR DRUG DISCOVERY

molecule possessing three double bonds conjugated with a carbonyl group. Tp80 is an isomer of Am80 and induced HL-60 cell differentiation in a similar manner to Am80. Other noncarboxylic acid-type retinoids (TZ335 and Bfx80, Fig. 9.5) have also been developed by using computer-assisted drug design. Another significant part of the retinobenzoic acid structure is the hydrophobic aromatic ring. A medium-sized bulky alkyl group on the benzene ring (left in the generic structure 20) is essential. Introduction of a polar atom decreases the activity. Interestingly, the replacement of quaternary carbon atoms of tert-butyl groups with silicon or germanium atoms, members of the same group (group IVB) of the periodic table, results in retention or increase of the retinoidal activity as can be seen in Table 9.4 (Yamakawa et al., 1990). Among disubstituted compounds, silyl or germyl derivatives are nearly as active as their carbon analogues. Thus, Ch55 and its silyl (Ch55S) and germyl (Ch55G) derivatives showed potent retinoidal activity. An exceptional increase of the activity caused by introduction of a disilyl moiety was observed with Am555S. The increased activities of silyl and germyl derivatives may be partially due to the different chemical properties of these atoms from that of the carbon atom. First, since the C Si or C Ge bond is longer than C C bond, the hydrophobic groups are bulkier than tert-butyl group and would better fit to the ligand-binding pocket of the receptor. Second, the silicon and germanium atoms are more electropositive than the carbon atom, and the surface electronic potentials of the methyl groups attached to these atoms are more negative than those of tert-butyl group. Retinobenzoic acids were developed by using HL-60 cell differentiation as the bioassay, and these compounds exhibited various biological activities, similar to ATRA. After discovery of retinoid nuclear receptors (three RAR subtypes), retinobenzoic acids have been proven to be RAR-selective agonists. Further, retinoidal

TABLE 9.4 HL-60 Cell Differentiation-inducing Activity of Trimethylsilyl (TMS) or Trimethylgermyl (TMG)-Containing Retinobenzoic Acids COOH

R X R

Code # ATRA Am55 Am55S Am55G Am555 Am555S Ch55 Ch55S Ch55G

X  NHCO   NHCO   NHCO   CONH   CONH     COCH  CH    COCH  CH  COCH  CH

R

EC50 (M)

tertBu TMS TMG tertBu TMS tertBu TMS TMG

2:4  109 3:6  108 4:2  108 3:6  108 4:8  108 9:2  109 2:1  1010 1:4  1010 2:1  1010

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

303

amides, Am80 and Am580, can bind potently to RARa, weakly to RARb, and not to RARg. Silyl analogue, Am555S is more RARa selective, and AGN-193836, developed by using Am80 as a lead compound, showed high RARa selectivity (Teng et al., 1996). Crystal structures of holo-RAR LBDs enabled researchers to design subtypeselective compounds, and BMS-961 was developed as an RAR-g-selective retinoid that distinguishes the small differences of amino acid residues in the LBD among the three RAR subtypes (Klaholz et al., 1998). These retinobenzoic acids cannot bind to RXRs and would not convert chemically to any compounds with binding affinity to RXRs, contrary to the easy conversion between ATRA and 9cRA besides their stability. Therefore, retinobenzoic acids have been used as biological tools for investigations in retinoids and their nuclear receptors. Various steroid hormone antagonists have been synthesized and some of them are in clinical use. Although the clinical utility of retinoid antagonists is still unclear, several types have been reported. As described above, the introduction of the bulky substituent on the agonist structure that affects H12 folding of receptor LBD is a useful strategy to develop the nuclear receptor antagonists. Thus, TD550 is an analogue of Am580 bearing a bulky diamantyl group at the hydrophobic region that would interact with H12 of RAR LBDs and inhibited retinoid-induced HL-60 cell differentiation by binding to RARs (Kaneko et al., 1991). Another strategy is based on the ligand superfamily concept (Fig. 9.20b). The nuclear receptors that are structurally and functionally similar form a superfamily and are assumed to have evolved from a single gene (Laudet et al., 1992). The structures of their ligand binding pockets of the receptors are possibly related to each other. This observation means that the structural relationships of agonist/ antagonist for one nuclear receptor might be applicable to another receptor. From this consideration, retinoid antagonists were designed on the basis of estrogen agonist/antagonist structures (Eyrolles et al., 1994). In estrogen, estradiol, diethylstilbestrol, and tamoxifen are endogenous ligand, synthetic agonist, and synthetic antagonist, respectively. Structurally, diethylstilbestrol corresponds to Am80, both being aromatic analogues of endogenous ligands, estradiol and ATRA, respectively. In the case of estrogen, introduction of bulky substituents into the position corresponding to the 7- or 11-position of estradiol, or on the olefinic bond of diethylstilbestrol, afforded antagonistic compounds, such as tamoxifen. The same structural modification indicated that introduction of bulky N-substituent of the amide bond of Am80 would cause the retinoid antagonists. Since N-alkylation causes the conformation of the amide bond, and conformational restriction is needed, benzimidazoles bearing a bulky substituent on the nitrogen atom of the imidazole ring provide the desired activity. Actually, among the synthesized benzimidazoles, compounds having a small alkyl group on the imidazole nitrogen atom acted as retinoid, while BIPh and BIBn exhibited RAR antagonistic activities. The synthetic scheme of BIPh is shown in Fig. 9.21. In this reaction scheme, the condensation of a diamine 22 with an aldehyde 23 is significant. The reaction of 22 in nitrobenzene at 180
C afforded benzimidazole, while the similar condensation reaction in ethanol at room temperature afforded seven-membered derivative 24. Although the formation of the dibenzodiazepine was not desired, 24 was assumed

304

FIGURE 9.20 (a) Structure of retinoid antagonist TD550. (b) Design of retinoid antagonists based on structure–activity relationships of ER ligands (ligand superfamily concept).

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

305

FIGURE 9.21 Retinoid antagonistic dibenzodiazepines.

to be similar type retinoid antagonist with BIPh, based on the ligand superfamily concept shown in Fig. 9.20. Compound 24 itself is unstable and did not have any activity, but the structural modification using 24 as a lead compound afforded a potent retinoid antagonist LE135. Hydrophobic alkylated aromatic rings and the benzoic acid moiety of LE135 are essential for the binding to the receptors, and the unsubstituted phenyl group is important for the antagonistic activity. Replacement of the unsubstituted phenyl group of LE135 with a naphthyl group, yielding LE540, increased the antagonistic potency (Umemiya et al., 1997). Examples of the various types of retinoid antagonists that have been reported are shown in Fig. 9.22. Ro-41-5253 is the first RARa-subtype-selective antagonist (Apfel et al., 1992), while LE135 exhibited RARb selectivity, although the RAR binding affinities of these two subtype-selective antagonists are not high. Several RAR antagonists such as ER-27191 and AGN-193109 showed high binding affinity for RARs, similar to that of ATRA.

FIGURE 9.22

Structures of RAR antagonists.

306

9.4.3

NUCLEAR RECEPTOR DRUG DISCOVERY

RXR-Selective Ligands

An endogenous ligand for RXRs was identified as 9cRA, but 9cRA can bind to RARs with as high affinity as that of ATRA. Therefore, selective RXR ligands are needed for the elucidation of RXR functions. SR11237 was identified as one of the first RXRselective agonist (Lehmann et al., 1992), and then various RXR-selective agonists have been developed, such as LGD1069, LG268, and PA024, and the new term ‘‘rexinoid’’ was proposed for RXR-selective agonists. While retinobenzoic acids with RAR binding affinity have a linking group with two or three atoms between two benzene rings, these RXR-selective ligands have a shorter linking group with one atom and the twisted conformation as depicted in Fig. 9.23. Structure–activity relationships of RAR antagonist LE135 resulted in development of unique RXR agonists with a dibenzodiazepine skeleton (Fig. 9.23b). Hydrophobicity of the unsubstituted benzo group of LE135 is important, and LE540 with a bulkier naphthyl group at this position affords a more potent RAR antagonist than LE135. However, LE590, having two tetramethylbutanyl groups on the benzene rings, exhibited partial (noncompetitive) antagonistic activity, and the inhibitory activity of LE590 did not reach the basal level, even at high concentrations. To clarify the effect of tetramethylbutanyl groups of LE590, HX600, an isomer of LE135, having the cyclic alkyl group on the other benzene ring, was synthesized. Interestingly, HX600

FIGURE 9.23 (a) Structures of RXR agonists. (b) Retinoid synergistic activity of HX630 with Am80 in HL-60 cell differentiation assay.

MEDICINAL CHEMISTRY OF RETINOID NUCLEAR RECEPTORS

307

exhibited completely opposite effects on retinoids; that is, HX600 itself did not affect HL-60 cell differentiation but strongly enhanced the potency of coexisting Am80 at low concentration (Umemiya et al., 1997). A similar synergistic activity was observed with various azepine derivatives, and the synergistic potency was in the order HX630, HX640 > HX600 > HX620. The synergistic activity of the diazepine derivatives is now interpreted as an allosteric effect of binding to the RXR site of RXRRAR heterodimers liganded with RAR agonist (retinoid). Thus, a RXR-selective agonist alone cannot activate RXR–RAR heterodimers, while it further activates the RXR– RAR heterodimers bound to RAR-selective agonist. RXR forms heterodimers with various nuclear receptors, and role of RXR ligands differ, depending on the heterodimer partners. RXR agonists enhanced the potency of RAR agonists in RXR–RAR heterodimer, while they did not show significant activity in RXR–VDR or RXR–TR heterodimers. On the other hand, RXR alone can activate the heterodimers of RXR with nuclear receptors related to metabolism, such as PPARs, LXRs, and FXRs (Chawala et al., 2001). This means that an RXR agonist has similar activity to the agonist for the heterodimer partner receptors. For example, PPARg is the target molecule for antidiabetic thiazolidindiones, such as troglitazone and pioglytazone. LG268 activated PPARg  RXR heterodimer and showed antidiabetic activity in db/db or ob/ob mice (Mukherjee et al., 1997). LG268 also inhibited cholesterol absorption by the RXR–LXR-mediated increase of cholesterol efflux and by the FXR–RXR-mediated reduction of the bile acid pool (Repa et al., 2000). No RXR ligand with apparent subtype selectivity has been reported so far. However, some RXR ligands showed some selectivity in heterodimer partners. For example, PA024 activated both PPARg–RXR and RXR–LXRa heterodimers, while HX630 activated only PPARg–RXR heterodimer and did not affect the activation of RXR–LXRa heterodimer. Only PA024 enhanced the expression of ATP binding cassette transporter (ABCA1) and apoA-1-dependent cholesterol release in undifferentiated THP-1 cells or RAW264 cells, whereas both RXR agonists are active in differentiated THP-1 cells, in which the PPARg mRNA level is upregulated. HX630 activated PPARg–RXR heterodimer, elevating the LXRa level and thereby enhancing ABCA1 expression and the resultant HDL generation. One of the undesirable effects of RXR agonists is an increase in serum triglyceride levels. This effect is considered to be mediated by RXR–LXR heterodimer-enhanced lipogenesis via induction of SREBP-1c expression. If this is the case, HX630, which lacks the ability to activate RXR–LXR heterodimer, may be a promising agent for the treatment of metabolic syndrome. In the cell differentiation assay using HL-60 cells, both PA024 and HX630 exhibited similar synergistic activity with Am80. However, the combinations of Am80 with PA024 and with HX630 showed different gene expression profiles during the induction of cell differentiation (Ishida et al., 2003). Close inspection of DNA microarrays indicated that PA024 and HX630 had different effects on the apoptosis of HL-60 cells when combined with Am80. The combination of Am80 with PA024, not with HX630, produced a gene expression profile similar to that seen with 9cRA (an RAR/RXR pan-agonist) and increased the induction

308

NUCLEAR RECEPTOR DRUG DISCOVERY

FIGURE 9.24

Structures of RXR antagonists.

of HL-60 cell apoptosis. Although the nature of the difference in mechanism of action is unclear, this result seems significant for retinoid therapy using combinations of RAR and RXR agonists. The function of RXR ligands depends on the heterodimer partners and various corepressors/coactivators, as well as some unknown mechanisms, and therefore different RXR agonists do not necessarily exhibit the same biological activities. Several RXR antagonists have been developed, and some of them showed heterodimer partner receptor-dependent behavior and their structures are shown in Fig. 9.24. LG100754, the first reported RXR antagonist, inhibited the RXR homodimer functions, while it elicited RXR agonistic activities in RXR–RAR and PPARg– RXR heterodimers. A structure–activity relationship study afforded LG101506 that selectively activated PPARg–RXR heterodimer and did not show retinoid synergism in RXR–RAR actions or activation of the RXR–LXR heterodimer (Michellys et al., 2003). LG101506 showed a potent ability to lower the glucose level in db/db mice and did not elevate the triglyceride level in Sprague–Dawley rats, whereas the level was significantly increased when an RXR agonist LG268 was used. Thus, some RXR ligands seem to change the agonistic/antagonistic activities, depending on the heterodimer partners and/or related cofactors (Shulman et al., 2004). HX531 was the first RXR antagonist that inhibited RXR–RAR heterodimer action, developed from the structure–activity relationships of HX600. HX531 acted as an antagonist toward PPARg–RXR heterodimer, but not toward PPARa–RXR heterodimer in an in vitro transactivation assay. In the inhibition of PPARg–RXR heterodimer activation, HX531 suppressed both PPARg and RXR agonists. Thus, HX531 can not only distinguish the heterodimer partner of RXR, but also inhibit the agonist of partner receptors. Interestingly, HX531 elicited antidiabetic and antiobesity effects in KKAy mice on a high-fat diet by inactivating PPARg–RXR action in adipocytes and muscle (Yamauchi et al., 2001). This result is apparently strange since both PPARg and RXR agonist showed antidiabetic effect in vivo. The detailed mechanistic investigations indicated both the activation and the moderate reduction of PPARg–RXR functions caused the antidiabetic effects, while only the latter case ameliorated the obesity. Another RXR antagonist, PA452, is a more RXR-selective antagonist than HX531 (Takahashi et al., 2002). Their antagonistic activities in RAR–RXR heterodimer action are different from each other. The data in Fig. 9.25 show that PA452 did not affect the differentiation-inducing activity of Am80 alone but inhibited the

CLINICAL APPLICATION OF RETINOIDS

309

FIGURE 9.25 Effects of HX531 (left) and PA452 (right) on HL-60 cell differentiation induced by Am80 in the presence or absence of PA024. Concentration of Am80 is 3  1010 M (open and closed circular) or 1  1010 M (closed triangle) and that of PA024 is zero (open circular), 3  1010 M (closed circular), or 1  109 M (closed triangle).

retinoid synergistic activity of PA024 in combination with a low concentration of Am80. In this case, the inhibition did not reach the basal level, and the percentage of differentiated cells in the presence of a high concentration of PA024 was more than that induced by Am80 alone. HX531 inhibited both Am80 alone and the combination of RAR/RXR agonists to the basal level. Thus, some RXR ligands seem to change the agonistic/antagonistic activities, depending on the heterodimer partners and/or related cofactors (Shulman et al., 2004). 9.5

CLINICAL APPLICATION OF RETINOIDS

Retinoid therapy using synthetic retinoids has already been realized in the fields of dermatology and oncology (Kagechika and Shudo, 2005). Some synthetic retinoids, such as adapalene, tazalotene, and Am80, have been proven to be clinically useful in the treatment of acne and psoriasis. Among various retinoid therapies, the most impressive example is the treatment of APL patients because of its achievement of the high complete remission rate (Huang et al., 1988). APL is the M3 subtype of acute myeloid leukemia (AML) in French–American–British classification comprising about 10–15% of adult AML cases. APL is characterized by a nonrandom chromosomal translocation that leads to the fusion of the RARa gene to one of five different partners, and 98% of APL patients display a translocation of RARa on chromosome 17 to the promyelocytic leukemia (PML) gene on chromosome 15, generating t(15;17) fusion gene. More than 90% of APL patients achieved complete remission (CR) by the differentiation therapy with ATRA. However, ATRA fails to induce a second remission in APL patients who relapsed after complete remission induced by ATRA (less than 20% CR) (Lengfelder et al., 2005). Several mechanisms

310

NUCLEAR RECEPTOR DRUG DISCOVERY

for ATRA resistance were proposed. Since ATRA is an endogenous hormonal substance, its concentration and distribution is regulated strictly in the body. Repeated administration of ATRA caused the accelerated clearance of ATRA from plasma by an induction of metabolizing enzymes or of the cellular retinoic acid binding protein (CRABP) that transports ATRA to the endoplasmatic reticulum where it is metabolized. Plasma ATRA concentrations decrease on average to one-thirds of their day 1 values during 1 week of continuous therapy, which might be the reason of ATRA resistance. Another putative mechanism is the mutations in the ATRA binding domain of the PML/RARa fusion gene. The mutations may lead to a reduced binding of the ligand and altered regulation of gene expression. Synthetic retinoids were supposed to overcome some of the disadvantage in ATRA therapy. A synthetic retinoid, Am80, has better pharmacological and chemical properties and further showed the lack of binding affinity to CRABP (about 1/100 of that of ATRA). In the preliminary clinical study on 24 APL patients who had relapsed from ATRAinduced complete remission (CR), 14 (58%) achieved a second CR by treatment of Am80 (6 mg/m2/day orally in two divided doses) (Tobita et al., 1997). The efficacy with less toxicity of Am80 was confirmed by second and detailed clinical trial for relapsed APL patients, and then Am80 (Tamibarotene) was approved as the drug for refractory and relapsed APL in Japan in 2005. In the fields of cancer therapy, the usefulness of RXR ligands has been also investigated, although their mechanisms are unknown. For example, LGD1069 effectively prevented primary and secondary rat mammary carcinoma induced by N-nitroso-N-methylurea. In a clinical trial for the treatment of refractory advanced-stage cutaneous T-cell lymphoma, LGD1069 showed 2% CR and 43% partial response (PR) (Duvic et al., 2001) and is clinically marketed as Bexarotene (Targretin). Since RXR can regulate various nuclear receptors related to the metabolism of lipid, glucose, and cholesterol, RXR ligands appear promising for the treatment of metabolic syndrome, although it is still unclear what kinds of RXR ligands are most useful in such diseases. Recent progress in the structural development of retinoids and the elucidation of their biological functions enabled their application to various diseases, such as autoimmune diseases and cardiovascular diseases (Kagechika and Shudo, 2005). Am80 is one of the most potential candidates for retinoid therapy due to its clinically beneficial properties discussed above. For example, Am80 is a potent inhibitor of the production of a multifunctional cytokine, IL-6, whose abnormal expression of IL-6 is related to the pathogenesis of several diseases such as psoriasis, multiple myeloma, and rheumatoid arthritis. Am80 was effective in animal models of collagen-induced arthritis and 2,4-dinitrofluorobenzene-induced contact dermatitis. Am80 suppressed Th1 development and enhanced Th2 development of naı¨ve CD4 T cells in vitro, and Th1-dominant autoimmune diseases could be the disease targets of Am80 (Iwata et al., 2003). As an RARa agonist attenuated the loss of the epithelial barrier during colitis in vivo, Am80 might be effective for Crohn’s disease, a Th1-dominant autoimmune disease characterized by the chronic inflammatory bowel. A clinical trial of Am80 for patients with Crohn’s disease is now in progress.

PERSPECTIVE

311

Besides the transcriptional regulation at retinoic acid response elements of DNA, liganded RARs interact with various transcriptional factors directly and regulate their functions. For example, Am80 regulates Kru¨ppel-like zinc-finger transcription factor 5 (KLF5/BTEB2) through RARs both in vitro and in vivo, which caused the repression of the formation of the granulation tissue and the neointima in wild-type mice with cuffed femoral arteries, as observed in KLF5-knockout mice ðklf þ= Þ (Shindo, et al. 2002, Takeda, et al. 2006). The potency of retinoid as KLF5 inhibitor would be a potential for clinical use in various cardiovascular diseases, such as arteriosclerosis, restenosis, and hypercardia.

9.6

PERSPECTIVE

Nuclear receptors are specific regulators of a variety of important physiological functions, such as growth, development, metabolism, and homeostasis. Although they have been proven to be involved in various refractory diseases in the present day, which include cancer, autoimmune diseases, and metabolic syndrome, few nuclear receptor ligands have been marketed as drugs, with the exception of steroid hormones. It is partially due to their complex functions, and even the specific ligands would elicit many biological activities, as discussed in this chapter. The specific ligands act as the switch for the activation of nuclear receptors, while the biological actions of nuclear receptors associate with various cellular factors. In the transactivation activity at the specific hormone response elements of DNA, monomeric or dimeric nuclear receptors interact with corepressors or coactivators. Therefore, besides the binding affinity and receptor- specificity, the three-dimensional structures of nuclear receptors, induced by ligand binding, determine the agonist/antagonist activity and their tissue selectivity. Thus, it is important to develop several different novel compounds classified as selective nuclear receptor modulators for various nuclear receptors. In some cases, nuclear receptors are posttranslationally modified by phosphorylation, methylation, sumoylation, and so on. These modifications change the stability or cofactor recruitment potency of nuclear receptors. For example, the phosphorylation of RXRa by Ras/Erk/MAPK signal pathway decreased the ability of dimer formation and the transactivation activity, and is considered to cause the promotion of cell growth in hepatocellular carcinoma cells (Matsushima-Nishiwaka et al., 2001). Thus, activation and inactivation of nuclear receptors related to intracellular signal transduction pathway is important. Further, nuclear receptors interact with various other transcriptional factors and affect their target gene expressions. The interaction of RARs with KLF5 is an example of such a novel action of nuclear receptors. It is also interesting to develop novel ligands that have weak or no activity in the transactivation at the hormone response elements but regulate the functions of the target transcriptional factors like KLF5. While the complex functions of nuclear receptors have been better elucidated recently, there exist various orphan nuclear receptors whose specific ligands and functions remain unknown. In twentieth century, steroid hormones have been developed,

312

NUCLEAR RECEPTOR DRUG DISCOVERY

without knowledge of the detailed functions of their nuclear receptors. Recent progress in molecular biology and medicinal chemistry of nuclear receptors resulted in the development of novel drugs, as illustrated with retinoid therapy using Am80. Although further detailed studies are needed, nuclear receptors are one class of the most significant molecular targets for drug discovery in twenty-first century.

REFERENCES Apfel C, Bauer F, Crettaz M, Forni L, Kamber M, Kaufmann F, LeMotte P, Pirson W, Klaus M, 1992. A retinoic acid receptor a antagonist selectively counteracts retinoic acid effects. Proc. Natl. Acad. Sci. USA 89:7129–7133. Boehm MF, Fitzgerald P, Zou A, Elgort MG, Bischoff ED, Mere L, Mais DE, Bissonnette RP, Heyman RA, Nadzan AM, Reichman M, Allegretto EA, 1999. Novel nonsecosteroidal vitamin D mimics exert VDR-modulating activities with less calcium mobilization than 1,25-dihydroxyvitamin D3. Chem. Biol. 6:265–275. Bohl CE, Gao W, Miller DD, Bell CE, Dalton JT, 2005. Structural basis for antagonism and resistance of bicalutamide in prostate cancer. Proc. Natl. Acad. Sci. USA. 102:6201–6206. Bourguet W, Ruff M, Chambon P, Gronemeyer H, Moras P, 1995. Crystal structure of the ligand-binding domain of the human nuclear receptor RXRa. Nature 375:377–382. ¨ hman L, Greene Brzozowski AM, Pike ACW, Dauter Z, Hubbard RE, Bonn T, Engstro¨m O, O GL, Gustafsson J-A, Carlquist M, 1997. Molecular basis of agonism and antagonism in the oestrogen receptor. Nature 389:753–758. Chawala A, Repa JJ, Evans RM, Mangelsdorf DJ, 2001. Nuclear receptors and lipid physiology: opening the X-files. Science 294:1866–1870. Denis LJ, Griffiths K, 2000. Endocrine treatment in prostate cancer. Semin. Surg. Oncol. 18:52–74. Duvic M, Hymes K, Heald P, Breneman D, Martin AG, Myskowski P, Crowley C, Yocum RC, 2001. Bexarotene is effective and safe for treatment of refractory advanced-stage cutaneous T-cell lymphoma: multinational phase II-III trial results. J. Clin. Oncol. 19:2456–2471. Ebisawa M, Ohta K, Kawachi E, Fukasawa H, Hashimoto Y, Kagechika H, 2001. Novel retinoidal tropolone derivatives. bioisosteric relationship of tropolone ring with benzoic acid moiety in retinoid structure. Chem. Pharm. Bull. 49:501–503. Ellis CN, Voorhees JJ, 1987. Etretinate therapy. J. Am. Acad. Dermatol. 16:267–291. Endo Y, Iijima T, Yaguchi K, Kawachi E, Inoue N, Kagechika H, Kubo A, Itai A, 2001a. Structure-activity Study of retinoid agonists bearing substituted dicarba-closo-dodecaborane. Relation between retinoidal activity and conformation of two aromatic nuclei. Bioorg. Med. Chem. Lett. 11:1307–1311. Endo Y, Iijima T, Yamakoshi Y, Fukasawa H, Miyaura C, Inada M, Kubo A, Itai A, 2001b. Potent estrogen agonists based on carborane as a hydrophobic skeletal structure. A new medicinal application of boron clusters. Chem. Biol. 8:341–355. Endo Y, Yamamoto K, Kagechika H 2003a. Utility of boron clusters for drug design. Relation between estrogen receptor binding affinity and hydrophobicity of phenols bearing various types of carboranyl groups. Bioorg. Med. Chem. Lett. 13:4089–4092. Endo Y, Yoshimi T, Miyaura T, 2003b. Boron clusters for medicinal drug design: selective estrogen receptor modulators bearing carborane. Pure Appl. Chem. 75:1197–1205.

REFERENCES

313

Eyrolles L, Kagechika H, Kawachi E, Fukasawa H, Iijima T, Matsushima Y, Hashimoto Y, Shudo K, 1994. Retinobenzoic acids. 6. Retinoid Antagonists with a heterocyclic ring. J. Med. Chem. 37:1508–1517. Feldman D, Glorieux FH, Pike JW, editors, 2004. Vitamin D, 2nd ed. Academic Press. Fujii S, Goto T, Ohta K, Hashimoto Y, Suzuki T, Ohta S, Endo Y, 2005. Potent androgen antagonists based on carborane as a hydrophobic core structure. J. Med. Chem. 48:4654–4662. Gao W, Bohl CE, Dalton JT, 2005. Chemistry and structural biology of androgen receptor. Chem. Rev. 105:3352–3370. Gao WQ, Dalton JT, 2007. Expanding the therapeutic use of androgens via selective androgen receptor modulators (SARMs). Drug Discov. Today 12:241–248. Henke BR, 2004. Peroxisome proliferator-activated receptor a/g dual agonists for the treatment of type 2 diabetes. J. Med. Chem. 47:4118–4127. Hosoda S, Tanatani A, Wakabayashi K, Makishima M, Imai K, Miyachi H, Nagasawa K, Hashimoto Y, 2006. Ligands with a 3,3-diphenylpentane skeleton for nuclear vitamin D and androgen receptors: dual activities and metabolic activation. Bioorg. Med. Chem. 14:5489–5502. Hourai S, Fujishima T, Kittaka A, Suhara Y, Takayama H, Rochel N, Moras D, 2006. J. Med. Chem. 49:5199–5205. Huang M, Ye Y, Chen S, Chai J, Lu J, Zhoa L, Gu L, Wang Z, 1988. Use of all-trans retinoic acid in the treatment of acute promyelocytic leukemia. Blood 72:567–572. Inaba Y, Yamamoto K, Yoshimoto N, Matsunawa M, Uno S, Yamada S, Makishima M, 2007. Vitamin D3 derivatives with adamantane or lactone ring side chains are cell type-selective vitamin D receptor modulators. Mol. Pharmacol. 71:1298–1311. Ishida S, Shigemoto-Mogami Y, Kagechika H, Shudo K, Ozawa S, Sawada J, Ohno Y, Inoue K, 2003. Clinically potential subclasses of retinoid synergists revealed by gene expression profiling. Mol. Cancer Ther. 2:49–58. Ishioka T, Tanatani A, Nagasawa K, Hashimoto Y, 2003. Anti-androgens with full antagonistic activity toward human prostate tumor LNCaP cells with mutated androgen receptor. Bioorg. Med. Chem. Lett. 13:2655–2658. Ishizuka S, Kurihara N, Reddy SV, Cornish J, Cundy T, Roodman GD, 2005. (23S)25-Dehydro-1a-hydroxyvitamin D3-26, 23-lactone, a vitamin D receptor antagonist that inhibits osteoclast formation and bone resorption in bone marrow cultures from patients with paget’s disease. Endocrinology 146:2023–2030. Iwata M, Eshima Y, Kagechika H, 2003. Retinoic acids exert direct effects on T cells to suppress Th1 development and enhance Th2 development via retinoic acid receptors. Int. Immunol. 15:1017–1025. Kagechika H, Shudo K, 2005. Synthetic retinoids: recent developments concerning structure and clinical utility. J. Med. Chem. 48:5875–5883. Kagechika H, Kawachi E, Hashimoto Y, Himi T, Shudo K, 1988. Retinobenzoic Acids. 1. Structure-activity relationships of aromatic amides with retinoidal activity. J. Med. Chem. 31:2182–2192. Kagechika H, Himi T, Kawachi E, Hashimoto Y, Shudo K, 1989. Retinobenzoic acids. 4. Conformation of aromatic amides with retinoidal activity. Importance of trans-amide structure for the activity. J. Med. Chem. 32:2292–2296.

314

NUCLEAR RECEPTOR DRUG DISCOVERY

Kaneko S, Kagechika H, Kawachi E, Hashimoto Y, Shudo K, 1991. Retinoid antagonists. Med. Chem. Res. 1:220–225. Klaholz BP, Renaud J-P, Mitschler A, Zusi C, Chambon P, Gronemeyer H, Moras D, 1998. Conformational adaptation of agonists to the human nuclear receptor RARg. Nat. Struct. Biol. 5:199–202. Laudet V, Gronemeyer H, 2001. The Nuclear Receptor Facts Book. Academic Press. Laudet V, Ha¨nni C, Coll J, Catzeflis F, Ste´helin D, 1992. Evolution of the nuclear receptor gene superfamily. EMBO J. 11:1003–1013. Lehmann JM, Jong L, Fanjui A, Cameron JF, Lu XP, Haefner P, Dawason MI, Pfahl M, 1992. Retinoids selective for retinoid X receptor response pathways. Science 258:1944–1946. Lengfelder E, Saussele S, Weisser, Buchner T, Hehlmann R, 2005. Treatment concepts of acute promyelocytic leukemia. Crit. Rev. Oncol. Hematol. 56:261–274. Makishima M, Lu TT, Xie W, Whitfield GK, Domoto H, Evans RM, Haussler MR, Mangelsdorf DJ, 2002. Vitamin D receptor as an intestinal bile acid sensor. Science 296:1313–1316. Matsushima-Nishiwaki R, Okuno M, Adachi S, Sano T, Akita K, Moriwaki H, Friedman SL, Kojima S, 2001. Phosphorylation of retinoid X receptor a at serine 260 impairs its metabolism and function in human hepatocellular carcinoma. Cancer Res. 61:7675–7682. Michellys PY, Ardecky RJ, Chen JH, D0Arrigo J, Grese TA, Karanewsky DS, Leibowitz MD, Liu S, Mais DA, Mapes CM, Montrose-Rafizadeh C, Ogilvie KM, Reifel-Miller A, Rungta D, Thompson AW, Tyhonas JS, Boehm MF, 2003. Design, synthesis, and structure–activity relationship studies of novel 6,7-locked-[7-(2-alkoxy-3,5-dialkylbenzene)-3-methylocta]2,4,6-trienoic acids. J. Med. Chem. 46:4087–4103. Miller AR, Etgen GJ, 2003. Novel peroxisome proliferator-activated receptor ligands for type 2 diabetes and the metabolic syndrome. Expert Opin. Investig. Drugs 12:1489–1500. Moore JT, Collins JL, Pearce KH, 2007. In: Schreber SL, Kapoor T, Wess G, editors. The Nuclear Receptor Superfamily and Drug Discovery in Chemical Biology, Vol. 3. WileyVCH, pp. 891–932. Mukherjee R, Davies PJA, Crombie DL, Bischoff ED, Cesario RM, Jow L, Hamann LG, Boehm MF, Mondon CE, Nadzan AM, Paterniti Jr JR, Heyman RA, 1997. Sensitization of diabetic and obese mice to insulin by retinoid X receptor agonists. Nature 386: 407–410. Nakano Y, Kato Y, Imai K, Ochiai E, Namekawa J, Ishizuka S, Takenouchi K, Tanatani A, Hashimoto Y, Nagasawa K, 2006. Practical synthesis and evaluation of biological activities of 1a, 25-dihydroxyvitamin D3 antagonist, 1a, 25-dihydroxyvitamin D3-26, 23-lactams. Designed based on the helix 12-folding inhibition hypothesis. J. Med. Chem. 49:2398–2406. Nilsson S, Gustafsson J-A, 2002. Biological role of estrogen and estrogen receptors. Crit. Rev. Biochem. Mol. Biol. 37:1–28. Nuclear Receptor Nomenclature Committee, 1999. A unified nomenclature system for the nuclear receptor superfamily. Cell 97:161–163. Repa JJ, Turley SD, Lobaccaro JMA, Medina J, Li L, Lustig K, Shan B, Heyman RA, Dietschy JM, Mangelsdorf DJ, 2000. Regulation of absorption and ABC1-mediated efflux of cholesterol by RXR heterodimers. Science 289:1524–1529. Roy AK, Lavrovsky Y, Song CS, Chen S, Jung MH, Velu NK, Bi BY, Chatterjee B, 1999. Regulation of androgen action. Vitam. Horm. 55:309–352.

REFERENCES

315

Shang Y, Brown M, 2002. Molecular determinants for the tissue specificity of SERMs. Science, 295:2465–2468. Shevde NK, Plum LA, Clagett-Dame M, Yamamoto H, Pike JW, DeLuca HF, 2002. A potent analog of 1a,25-dihydroxyvitamin D3 selectively induces bone formation. Proc. Natl. Acad. Sci. USA 99:13487–13491. Shiau AK, Barstad D, Loria PM, Cheng L, Kushner PJ, Agard DA, Greene GL, 1998. The structural basis of estrogen receptor/coactivator recognition and the antagonism of this interaction by tamoxifen. Cell 95:927–937. Shiau AK, Barstad D, Radek JT, Meyers MJ, Nettles KW, Katzenellenbogen BS, Katzenellenbogen JA, Agard DA, Greene GL, 2002. Structural characterization of a subtypeselective ligand reveals a novel mode of estrogen receptor antagonism. Nat. Struct. Biol. 9:359–364. Shimasaki H, Kagechika H, Fukasawa H, Kawachi E, Shudo K, 1995. Retinoidal dienamides and related aromatic amides. Replacement of the 9-ene structure of retinoic acid with a trans- or cis-amide group. Chem. Pharm. Bull. 43:100–107. Shindo T, Manabe I, Fukushima Y, Tobe K, Aizawa K, Miyamoto S, Kawai-Kowase K, Moriyama N, Imai Y, Kawakami H, Nishimatsu H, Ishikawa T, Suzuki T, Morita H, Maemura K, Sata M, Hirata Y, Komukai M, Kagechika H, Kadowaki T, Kurabayashi M, Nagai R, 2002. Kru¨ppel-like zinc-finger transcription factor KLF5/BTEB2 is a target for angiotensin II signaling and an essential regulator of cardiovascular remodeling. Nat. Med. 8:856–863. Shulman AI, Larson C, Mangelsdorf DJ, Ranganathan R, 2004. Structural determinants of allosteric ligand activation in RXR heterodimers. Cell 116:417–429. Soloway AH, Tjarks W, Barnum BA, Rong F-G, Barth RF, Codogni IM, Wilson JG, 1998. The chemistry of neutron capture therapy. Chem. Rev. 98:1515–1562. Souque A, Fagart J, Couette B, Davioud E, Sobrio F, Marquet A, Rafestin-Oblin ME, 1995. The mineralocorticoid activity of progesterone derivatives depends on the nature of the C18 Substituent. Endocrinology 136:5651–5658. Sporn MB, Roberts AB, Goodman DS, editors, 1994. The Retinoids, 2nd ed. Raven Press, New York. Suhara Y, Kittaka A, Ono K, Kurihara M, Fujishima T, Yoshida A, Takayama H, 2002. Design and efficient synthesis of new stable 1a,25-dihydroxy-19-norvitamin D3 analogues containing amide bond. Bioorg. Med. Chem. Lett. 12:3533–3536. Takahashi B, Ohta K, Kawachi E, Fukasawa H, Hashimoto Y, Kagechika H, 2002. Novel retinoid X receptor antagonists: specific inhibition of retinoid synergism in RXR–RAR Heterodimer Actions. J. Med. Chem. 45:3327–3330. Takeda N, Manabe I, Shindo T, Iwata H, Iimuro S, Kagechika H, Shudo K, Nagai R, 2006. The synthetic retinoid Am80 reduces scavenger receptor expression and atherosclerosis in mice by inhibiting IL-6. Arterioscler. Thromb. Vasc. Biol. 26:1177–1183. Takenouchi K, Sogawa R, Manabe K, Saitoh H, Gao Q, Miura D, Ishizuka S, 2004. Synthesis and structure–activity relationships of TEI-9647 derivatives as Vitamin D3 antagonists. J. Steroid.Biochem. Mol. Biol. 89:31–34. Tanatani A, Yamaguchi K, Azumaya I, Fukutomi R, Shudo K, Kagechika H, 1998. N-Methylated diphenylguanidines: conformations, propeller-type molecular chirality, and construction of water-soluble oligomers with multi-layered aromatic structures. J. Am. Chem. Soc. 120:6433–6442.

316

NUCLEAR RECEPTOR DRUG DISCOVERY

Teng M, Duong TT, Klein ES, Pino ME, Chandraratna RAS, 1996. Identification of a retinoic acid receptor a subtype specific agonist. J. Med. Chem. 39:3035–3038. Tobita T, Takeshita A, Kitamura K, Ohnishi K, Yanagi M, Hiraoka A, Karasuno T, Takeuchi M, Miyawaki S, Ueda R, Naoe T, Ohno R, 1997. Treatment with a new synthetic retinoid, Am80, of acute promyelocytic leukemia relapsed from complete remission induced by all-trans retinoic acid. Blood 90:967–973. Toell A, Gonzalez MM, Ruf D, Steinmeyer A, Ishizuka S, Carlberg C, 2001. Different molecular mechanisms of vitamin D3 receptor antagonists. Mol. Pharmacol. 59:1478–1485. Turner CH, Sato M, Bryant HU, 1994. Raloxifene preserves bone strength and bone mass in ovariectomized rats. Endocrinology, 135:2001–2005. Umemiya H, Fukasawa H, Ebisawa M, Eyrolles L, Kawachi E, Eisenmann G, Gronemeyer H, Hashimoto Y, Shudo K, Kagechika H, 1997. Regulation of retinoidal actions by diazepinylbenzoic acids. Retinoid synergists which activate the RXR–RAR heterodimers. J. Med. Chem. 40:4222–4234. Wakabayashi K, Miyachi H, Hashimoto Y, Tanatani A, 2005. Novel non-steroidal/non-anilide type androgen antagonists: Discovery of 4-substituted pyrrole-2-carboxamides as a new scaffold for androgen receptor ligands. Bioorg. Med. Chem. 13:2837–2846. Yamada S, Shimizu M, Yamamoto K, 2003. Structure–function relationships of vitamin D including ligand recognition by the vitamin D receptor. Med. Res. Rev. 23:89–115. Yamakawa T, Kagechika H, Kawachi E, Hashimoto Y, Shudo K, 1990. Retinobenzoic acids. 5. Retinoidal activities of compounds having a trimethylsilyl or trimethylgermyl group(s) in human promyelocytic leukemia cells HL-60. J. Med. Chem. 33:1430–1437. Yamauchi T, Waki H, Kamon J, Murakami K, Motojima K, Komeda K, Miki H, Kubota N, Terauchi Y, Tsuchida A, Tsuboyama-Kasaoka N, Yamauchi N, Ide T, Hori W, Kato S, Fukayama M, Akanuma Y, Ezaki O, Itai A, Nagai R, Kimura S, Tobe K, Kagechika H, Shudo K, Kadowaki T, 2001. Inhibition of RXR and PPAR ameliorates diet-induced obesity and type 2 diabetes. J. Clin. Invest. 108:1001–1013. Yoshimoto N, Inaba Y, Yamada S, Makishima M, Shimizu M, Yamamoto K 2008. 2-Methylene 19-nor-25-dehydro-1a-hydroxyvitamin D3 26,23-lactones: Synthesis, biological activities and molecular basis of passive antagonism. Bioorg. Med. Chem. 16 (1):457–473.

10 SUMMARY AND COMPARISON OF MOLECULES DESIGNED TO MODULATE DRUGGABLE TARGETS IN THE MAJOR GENE FAMILIES KAREN E. LACKEY

10.1

TARGET CLASS CONCEPT

With the completion of the Human Genome Project, the scientific community can carefully consider the protein gene family products and posttranslational modifications of these proteins for determining possible human drug targets. Many bacterial, fungal, yeast, parasitic, and viral infections require important nonhuman drug targets, equally important in discovering drugs for unmet medical needs. Andrew Hopkins and Colin Groom coined the phrase ‘‘druggable genome’’ and provided a thoughtful analysis of how many potential human molecular targets were amenable to drug intervention (Hopkins and Groom, 2002). They limited their analysis to protein targets for which one could reasonably expect to find orally available compounds that bind. Limiting the potential disease targets in this way could help drug discovery groups focus on areas that are tractable for generating small-molecule modulators; thus, using a gene family approach would facilitate rapid design and progress of leads. Unfortunately, the downside is that just because a synthetic ligand can be generated does not mean it is important for drug intervention. Moreover, advances in drug delivery systems could help to expand the definition of ‘‘druggable’’ to small molecules that potently bind

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

317

318

SUMMARY AND COMPARISON OF MOLECULES

the intended target, where the delivery system takes care of the desired route of administration (e.g., oral, dermal patch, slow release, and so on) and potentially even the selectivity. One example of recent success is in the use of nanotechnology derived from electronic industry being applied to biomedical solution (Napier and DeSimone, 2007). Directed drug delivery systems may someday obviate the need for protracted optimization for physicochemical properties that may be mutually detrimental to binding to a therapeutically important target. Approximately 50% of currently marketed oral drugs modulate a G-proteincoupled receptor (GPCR), while only about 5% exert their effects through ion channels (Bleicher et al., 2003). The drug binding for successful drugs is often the same as the endogenous ligand site of action, thus blocking or stimulating the protein’s natural function. The more we learn from allosteric modulation and protein mechanisms of action through techniques such as crystal structures and NMR, the greater our repertoire of tools will be to create small molecules with the desired intervention capabilities. This understanding would also extend to creating compounds with polypharmacology or multiprotein target modulation. The state-ofthe-art research for each gene family discussed in the book has made tremendous advances in our understanding of how and where to bind small molecules, which are at different levels of tractability.

10.2 SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS This section will include a synopsis of each of the gene families of protein targets for drug discovery discussed in this book. I deliberately focused on the prominent methods and approaches used to design specific protein modulators. The reader should seek the details and full descriptions for better understanding directly from the chapters. As you read through the summaries below, it becomes apparent that some portions of target classes are more tractable for generating smallmolecule modulators (e.g., inhibitors, agonists, partial agonists, partial antagonists) than others. Without a doubt, advances in protein crystal structure and screening technologies have shaped the druggability of each target class. Most importantly, every target class has work in progress to identify the challenges and improve the discovery of small-molecule drugs. 10.2.1

GPCR/7TM

As discussed in Chapter 2 by Stephen Garland and Tom Heightman, an endogenous extracellular ligand binds to the GPCR, causing a conformational change that activates a G-protein complex triggering a cascade of signals with diverse biological outcomes. Because GPCRs are modulated via diverse natural ligands and via allosteric sites, many effective small-molecule drugs have emerged for this target class. The high tractability has not been without its challenges especially due to receptor dimerization and complex signaling networks. There are six subfamilies (A–F), but the authors of Chapter 2 chose to limit their discussion to three families: A, B,

SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS

319

and C. A second classification system (GRAFS) breaks the GPCRs into five main subfamilies based on a phylogenetic analysis: glutamate, rhodopsin, adhesion, Frizzled, and secretin. All of the GPCRs generally have the following structural features: an extracellular N-terminal domain (NTD), seven transmembrane a-helices (TM1–7) with three extracellular connecting loops (ECL1-3) plus three intracellular connecting loops (ICL1-3), and a C-terminal intracellular tail (C-term). The orthosteric binding site for most members of Family A GPCRs is defined by the TM bundle with some potential contributions from ECLs. Leaving aside the opsin and olfactory receptors, there are about 280 potential drug targets in this family. Key amino acid sequences that form highly conserved interactions for ligand binding are used to build homology models. Crystal structures for the GPCRs have only emerged very recently due to the inherent challenges of crystallizing flexible, membrane-bound proteins. In 2007, two research groups separately solved the protein structure for beta-2 adrenergic receptor (Family A) by each using a creative stabilization approach coupled with advances in microdiffraction techniques. While the structures provide profound insights, caution in drug design is necessary because of expected conformational differences between native protein and the modified version needed to establish crystals. The limited homology between Family A, as compared with B and C, challenges the accuracy of sequence alignments and helical packing. Homology models, while rationally developed over 15 years, are more reliable if based on more recent data. In fact, a plethora of sitedirected mutagenesis data are available to aid the design and improvement of homology models and small-molecule compound binding. Both the NTD and TM bundle form a component of the orthosteric binding site for Family B peptide hormone GPCRs. This 48-member family covers both secretins and adhesion receptors. The NTD of the GIP receptor has been crystallized and the ligand-bound structure was solved and corroborated by earlier NMR studies with another Family B receptor. A two-domain model of endogenous peptide binding was proposed such that the NTD forms a high affinity binding opportunity, and due to proximity, the binding to the TM bundle is sufficient. The orthosteric binding site for Family C GPCRs, also called glutamate receptors, is defined by the NTD. There are 22 members in this family of drug targets, with ligand-bound protein crystal structures of the NTD for just a few of them. They clearly demonstrate a ‘‘venus flytrap’’ mechanism of ligand binding, but little is known on how the signal is conveyed. Although the phylogenetic classification helps understanding the target function, Garland and Heightman make the case that it is more beneficial for drug design to group the GPCRs based on homology within the binding sites. There are several functional mechanisms of GPCR activity discussed by the authors of Chapter 2, which include potential receptor dimerization, accessory proteins, allosteric binding, activation mechanisms, and receptor deactivation. There is evidence to support GPCR heterodimerization in native cells, which greatly affect the identification of small-molecule modulators. Accessory proteins play a role in modulating GPCR pharmacology, and the full extent of the interactions is yet to be understood. There are a number of allosteric sites that have been exploited by synthetic compounds aimed at modulating GPCRs. Several advantages of allosteric modulation were discussed. For example, positive allosteric modulators might

320

SUMMARY AND COMPARISON OF MOLECULES

sensitize a receptor to its endogenous ligand without any inherent activity alone. There are examples of allosteric modulation in Families A, B, and C providing agonism, antagonism, and positive allosteric modulation. Even without an endogenous ligand bound, GPCRs have some level of basal activity suggesting that when ligands bind, they change the equilibrium between active and inactive states. In the absence of detailed full-length protein crystal structures, the mechanism of the receptor activation has been examined by other sophisticated techniques but remains undetermined. To keep the natural signals in check, receptor deactivation occurs via kinase phosphorylation followed by arrestin binding. G-proteins are trimeric combinations from 21 Ga-subunits, 6 Gb-subunits, and 12 Gg subunits. The heterotrimeric G-proteins are also classified into families. The assay systems used to screen the active modulators measure the functional activity of the GPCR via receptor–G-protein interactions, downstream signaling consequences, transcriptional readouts, and arrestin. A diversity of approaches employed in parallel are often necessary to generate lead molecules for GPCR targets. In addition to typical high-throughput screening, the authors discuss the successful application of privileged fragments such as creating conformationally constrained analogues derived from a substructure of an endogenous ligand. Further optimization to build in selectivity and drug-like properties has led to numerous clinical candidate drugs. For example, cores include benzazepine for aminergic receptors and aryl piperidines for peptidergic receptors. The discovery of MK-0974 as a calcitonin gene-related peptide (CGRP) receptor antagonist for the treatment of migraine was described to demonstrate a success in achieving favorable ADMET properties for peptide receptor drugs. It began from a high-throughput screen (HTS) hit and knowledge derived from an advanced peptide clinical candidate, BIBN-4096. By overlaying the two energy-minimized structures, the potency was improved by substitutions on the core of the phenylbenzodiazepine HTS hit structure, followed by traditional medicinal chemistry to improve the DMPK properties. The discovery of GSK773812 as a mixed dopamine/serotonin receptor agonist demonstrates a successful example of polypharmacology for a potential antipsychotic therapeutic agent. Focused screening resulted in the starting point with aminergic activity. Key to the success of the project was careful selection of assays to reflect the desired receptor profile derived from an analysis of marketed antipsychotic agents and the disease pathology. Iterative screening and SAR were used to design the drug candidate compound. The discovery of cinacalcet, a calcium-sensing receptor (CaSR) positive allosteric modulator, shows a successful example of targeting allosteric sites. A cellular screen was used to find the starting hit, feldiline. This compound and analogues were found active only in the presence of a certain range of calcium concentrations, thus uncovering this series’ unique mechanism of action. 10.2.2

Ion Channels

As discussed in Chapter 3 by Maria Garcia and Gregory Kaczorowski, ion channels (IC) are membrane proteins, which regulate the flow of ions and consequently have

SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS

321

an impact on overall cellular physiology. Changes in membrane voltage, binding of ligands, or mechanical forces are triggers that lead the IC to be closed (no permeation) or open (selective or nonselective permeation). Nicotinic acetylcholine is a well-studied example of a ligand that can bind to an IC receptor causing a chemical signal to be translated into a well-controlled electrical signal via the influx of Naþ. Most of the chapter focused on the large voltage-gated IC subfamily and the unique functional diversity of the pore region achieved by a variety of subunit associations that form the total protein complex. For example, inward rectifier potassium channels (Kir) are comprised of four associated pore domain subunits and conduct more Kþ into the cell than out of the cell. Voltage-gated potassium channels (Kv) are made up of four subunits that form both a pore domain and a voltagesensing domain. There are also associated beta-subunits that confer additional activities. There are mechanisms for inactivation (fast and slow) and activation of Kv channels, and the wide tissue distribution means that these channels are responsible for controlling important, diverse body functions. Side effects of some drugs and clinical candidates (from blocking Kv7.1 or Kv11.1) cause cardiac symptoms that result in a recall or halt to clinical trials. Calcium-activated potassium channels (KCa) typically bind Ca2þ, which causes a conformational change resulting in an open channel. KCa channels are structurally and functionally diverse and they regulate a large number of physiological processes such as contractility and lymphocyte proliferation. Both blockers and activators of different members of this subfamily have been discovered for treating different disease conditions. Compared with other gene families, protein structural work has only recently begun making an impact due to difficulties on crystallization. What has been learned from the pioneering work has led to insights into the molecular features that control the mechanism of ion permeation and selectivity. The selection of appropriate IC targets for drug discovery is highly dependent on the validation of the role of the specific protein in both the animal pharmacology models and the anticipated human disease setting. The authors outline the complexity in screening the correct physiological state of the IC and in creating assay systems robust enough for the data to produce SAR. The biological evaluation must also include appropriate selectivity for the desired mechanism of action and disease indication – all of which are scientific challenges specific to IC. In the ion channel case study presented by Garcia and Kaczorowski, the hit generation emerged from a natural product. The results show that molecules can be simplified and can retain target affinity with the SAR. Further, it is clear that in ion channel research, the complexity of translating target affinity and in vivo activity into mechanisms of desired target-mediated activity continues to challenge drug discovery. Selectivity parameters within the target class are important to justify by mechanistic importance to the intended use of the drug. There has been significant debate in the field of medicinal chemistry regarding the comparison of natural products and small molecules as starting points in a drug discovery program. The most compelling argument that emerged from the case study of correolide was the need to truly understand the binding interaction. In doing so, researchers unveiled a novel, allosteric cooperative activity from multiple binding of a single inhibitor.

322

SUMMARY AND COMPARISON OF MOLECULES

A structural understanding of why a compound gives important selectivity allows a chemist to take advantage of that design feature on other, possibly more druggable scaffolds. While this medicinal chemistry example highlighted ion channel drug discovery aspects, it is also a general lesson for assessing starting points and tractability in general. It is important to note that most, if not all, of the marketed drugs that modulate ICs as their primary targets were discovered empirically. By that I mean that compounds were found to be effective in preclinical animal models in a disease pathology setting without knowing their biological target. Because the compounds had efficacy in these models with acceptable therapeutic indices, they were safely progressed to clinical trials. The specific target responsible for the therapeutic effect was determined after the drugs were used and found to be ICs. It will be interesting to see if developments emerging from a more genomic approach of starting with a specific IC and finding a potent, selective small molecule will result in a more rapid discovery of IC drugs. 10.2.3

Integrins

Integrins, described in Chapter 4 by David Miller, are a small family of proteins that serve as a bidirectional signaling connection linking the outside of the cell, through the membrane, to the cytoskeleton. There are 24 known members formed from heterodimers of a- and b-subunits. Like other protein classes, the activity is modulated by endogenous ligands that transmit signals in a highly controlled manner that has taken years of molecular biology research to understand the mechanism. The insights provided by biochemistry and X-ray crystallography have led to domains, folds, and subunit areas of conformational changes that can be targeted for drug intervention. Arginine–glycine–aspartic acid, the RGD peptide sequence, served as a recognition motif for several important integrins in antithrombotic drug discovery. Extensive exploration of RGD mimetics linked binding modes with mechanisms of integrin inhibition. For example, allosteric binding, competitive binding, and stabilizing an inactive conformation offer a variety of ways to modulate integrins in a disease setting. In contrast to the ion channel area case study that began with a natural product, the integrin drugs have emerged from a peptide starting point. Important interactions of an active cyclic or linear peptide can be used to design a small, orally available compound. As the characteristics of peptides, including their zwitterionic nature, makes lead optimization a particularly challenging event, approaches to overcome the issues were employed. Successful examples of a prodrug approach were described by Miller whereby the active parent molecule is created in vivo via a metabolically labile solubilizing group. Strategies for obtaining selectivity took advantage of differences in the protein structure, pKa of the hydrogen binding interactions, and the geometry of the inhibitor. Like the kinase area, strong influence of X-ray structure and docking studies in the design of compounds can be seen by the use of known inhibitors to build pharmacophore models to search databases for small-molecule starting points. Several examples were shown in Chapter 4 of

SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS

323

dramatic potency improvements with particular impact on cardiovascular and autoimmune diseases. A predominant theme of aspartic acid in the contiguous peptide sequence allows lead hopping to occur in the integrin family of targets. Learning structural reasons for obtaining selectivity offered SAR for converting an inhibitor to bind to a different family member through specific functionalization. In contrast to the competitive binding described above as a means to inhibit integrin signaling, an area of allosteric inhibitor research involves binding in the I-domain or I-like domain. Binding to the domain was measured in a highthroughput screen, which provided heterocyclic small molecules as starting points. Optimization of molecules that bind to aLb2/ICAM-1 led to advanced compounds with anti-inflammatory effects. An example of using fragment-based methodology expanding on an HTS hit was provided to show the value of combined techniques for optimizing binding affinity and compound properties. A peptidomimetic strategy was also described for generating I-like domain inhibitors. By searching a small-molecule database using the key interactions of a potent peptide, followed in some cases with a ligand-bound crystal structure, the binding site, potential mode of action, and areas of compound modification could be defined to improve affinity and drug-like characteristics. Homology models and docking studies could be developed to target related integrins furthering the understanding of how to achieve selectivity. 10.2.4

Kinases

The strategies for discovering Kinase inhibitor drugs, discussed in Chapter 5 by Jerry Adams, Paul Bamborough, David Drewry, and Lisa Shewchuk, focused primarily on binding in the ATP site of the catalytic portion of the signaling proteins. With over 500 kinases in the human genome, the number of potential targets or combination of targets for effective drug intervention is as exciting as the selectivity issues introduced are complex. The ATP site is tractable for finding potent small molecules, and the use of crystal structures and homology models to exploit the unique features of the binding pocket has dominated the field of kinase drug discovery. The kinase domain is specifically designed to phosphorylate proteins by the transfer of the terminal phosphate group of ATP to a downstream member of a signaling cascade. It can be divided into two subdomains joined by a flexible linker that creates a deep pocket in which ATP binds with the terminal phosphate pointing outward. The substrate protein docks across the cleft to complete the enzymatic transfer of the phosphate. Kinases adopt different conformations depending on their activation state and presence of ligands, making the use of structure to design selective inhibitors feasible. The authors brilliantly described the features of the kinase domain necessary to consider when optimizing the target affinity, selectivity, and compound property improvements. To bind in the ATP binding pocket, a core scaffold with the ability to form a hinge acceptor is the most straightforward starting point. Grown out from this primary interaction is a collection of binding regions: DFG, back, adenine, phosphate,

324

SUMMARY AND COMPARISON OF MOLECULES

sugar, and outer lipophilic and surface site pockets. Depending on the amino acids lining the ATP site, these regions are variously filled to achieve the desired potency and selectivity. Cross-screening of kinase targets and compounds appears necessary not only to glean selectivity information, but also for the opportunistic hit discovery for other kinase targets. The structure-based design tools in the inhibitor design for the ATP binding site of kinases are sophisticated enough to design potency with selectivity where there are only single residue differences within the ATP site. Inducible pockets and conformational changes are also successfully predicted and exploited. Despite extraordinary success, problems do exist with cellular activity due to inherent ATP concentrations, physical properties of the compounds, and selectivity leading to the need for non-ATP competitive inhibitors. These compounds have been discovered through screening and targeting the inactive conformation of the kinase protein. Both ATP-competitive and noncompetitive kinase inhibitors require careful consideration for establishing the appropriate enzyme assays coupled with a meaningful cellular mechanistic assay that measures the signal inhibition in a more realistic complex signaling environment. Case studies of successful kinase drug discoveries that led to GleevecTM(bcr-abl, c-kit, PDGFR), dasatinib (bcr-abl, scr family, others), IressaTM (EGFR), TarcevaTM (EGFR), TykerbTM (EGFR, erbB2), and sorafenib (raf, VEGFR, others) provide clear evidence that the field has enormous potential. 10.2.5

Proteases

Inhibition of important posttranslational modifications can be achieved by targeting proteases, described in Chapter 6 by Richard Sedrani, Ulrich Hommel, and Jorg Eder. The 512 active proteases in the human genome can be subdivided based on the mechanism and amino acid residue affected by the enzymatic hydrolysis: serine/threonine, cysteine, metallo, and aspartic proteases. These are further categorized based on their sequence homology and 3D fold. Targeting several proteases has already resulted in successful drugs such as ACE inhibitors for hypertension and HIV protease for AIDS. Predominant strategies for inhibitor design originally used peptides or peptidomimetic approaches based on the sequence of the protein substrate being cleaved. Although potent target affinity was often achieved, challenges in compound properties and selectivity were difficult to overcome. As the research field evolved, more inhibitor designs were based on using crystal structures, high-throughput screening hits, virtual screening based on protein homology models, and fragment-based screening. Aspartic protease inhibitor design is greatly facilitated by the availability of protein crystal structures. The aspartic proteases have a similar 3D fold, which forms the catalytic site. Key to understanding how to design an inhibitor is the understanding of the water-mediated peptide bond cleavage. The authors provide two successful case studies for this protease class, highlighting HIV protease inhibitors based on a peptidomimetic strategy and Renin inhibitors based on transition state analogues.

SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS

325

Metalloprotease activity requires a divalent metal ion, usually zinc but also includes copper, nickel, and manganese. The mechanism of bond cleavage has been extensively studied and used to design compounds. The metal ion binds to the carbonyl group on the protein substrate, facilitating the activated water molecule’s nucleophilic attack resulting in bond cleavage. Angiotensin-converting enzyme (ACE) inhibitors, probably the most famous success story of protease inhibitors, began as dipeptide analogues of snake venom. The basis for improved designs for greater potency and in vivo properties centered on metal chelating groups. The matrix metalloproteases (MMPs) degrade extracellular components and activate proteins through proteolytic cleavage. Extensive use of structure-based design has led to several advanced drug candidates that afforded disappointing clinical results, principally due to their unacceptable (mechanism related) MMP family selectivity profiles. The well-characterized mechanism of the catalytic triad for the serine proteases lead to inhibitor design based on creating reversible traps for a nucleophilic serine intermediate. A successful example of inactivating DPP4 highlighted the challenges of designing a dipeptide with a suitably reactive electrophile possessing suitable in vivo properties. Although early work was mechanism derived, more recently protein crystal structure could be used that validated the original design principles. Further optimization using HTS to find a starting point, followed by optimization via structure-based design, produced molecules with impressively potent target affinity negating the need for the appended reactive group. The final class of proteases, cysteine, has a mechanism similar to that of serine proteases, except that the nucleophile in the catalytic triad is the deprotonated thiol of the cysteine side chain. The general design principles were outlined by the authors using cathepsin K as a prototypical example. Beginning with peptides functionalized with an electrophile designed to form an irreversible covalent interaction with the catalytic cysteine, structure-based design was used to optimize binding interactions. Constrained analogues masking the peptide nature of the molecules achieved the desired cellular and in vivo properties. Unlike the serine proteases inhibitors, the design of effective cysteine protease inhibitors is yet to be demonstrated with noncovalent inhibitors. 10.2.6

Protein–Protein Interactions

Strategies to blocking an interaction between two members of a protein complex with a small molecule were described in Chapter 7 by Adrian Whitty. Several marketed biopharmaceuticals have the desired specificity and efficacy, but due to the nature of the biological product, they are inconvenient to administer to patients. Also, many biological products fail simply because of stability and target accessibility. The ability to mimic the activity of these biopharmaceuticals with a small drug-like compound would make an impact on many diseases. Since most signaling enzymes (e.g., kinases, phosphatases) function in a complex with other proteins, the ability to disrupt protein–protein interactions (PPI) would offer an alternate way to inhibit these signals.

326

SUMMARY AND COMPARISON OF MOLECULES

PPI interfaces have been studied using biophysical, structural, and bioinformatics approaches to find plausible interactions for binding a small molecule. Distinguishing a transient versus constitutive relationship between proteins is important in inhibitor design due to the differences in structure and properties of the potential binding site (s). Generally, the area of solvent accessible protein surface that is subsequently buried with two binding partners varies considerably, and studies have further refined zones of hydrophobic and polar sites that offer different physicochemical properties critical to the selectivity of protein–protein binding. Molecular level understanding of which amino acid side chains dominate the protein–protein binding demonstrated clusters of interactions that could be disrupted for inhibiting the actions of the protein complex. Hotspot residues tend to be completely buried in a protein–protein complex and water molecules also play a key role in bridging polar interactions. Transient PPI also require shape complementarity to avoid nonspecific interactions. These factors have implications for small-molecule inhibitors. For example, the presence of ‘‘hotspots’’ suggests that a small compound could bind with affinity and effectively disrupt a PPI. Whitty makes the case that the inhibitors of transient PPI would be more druglike due to the mixed polar–apolar interfaces. The small molecules would need to make hydrophobic contacts, salt bridges, and hydrogen bonds. Because these are generally weak binding interactions, multiple points of contacts are needed. The druggability of PPI can be assessed by the topology of the interfaces. For example, where the mechanism of action requires the specific binding of short amino acid sequence, a small peptide could be used as a starting point for a peptidomimetic compound design strategy. Proteins are highly dynamic molecules that exist in several interconvertible conformations. The binding of a small-molecule inhibitor can take advantage of structural adaptivity and stabilize or induce a low energy conformation through multiple binding interactions. It is possible to conclude that significantly larger ligands are necessary to form sufficient contact area in PPIs, thus suggesting that they will not meet the conventional definition of drug-sized molecules. Whitty makes the case that regions of chemical space reached by natural products may be ideal starting points for PPI agonist because they generally have greater structural complexity. They are often orally available and despite their size, could offer potential ways forward for discovering drugs in important PPIs. Another powerful technique for finding hits for PPI targets is fragment-based screening. Once fragments are found to interact with the protein, further optimization to achieve the desired affinity can follow several strategies depending on the binding surface and design tools available. For example, perturbation of the heteronuclear NMR spectrum of the protein target can identify true binders and define the binding location. Fragment-bound protein crystal structures are instrumental in identifying key fragments and determining ways to optimally link the fragments and are used to optimize the binding affinity. Several other creative ways to find fragments and grow them into PPI agonists and antagonists are described in the literature with their scope and limitations as a design method (Wyss and Eaton, 2007). A successful example of inhibiting an alpha-helix-accommodating binding site was described for the Bcl-2 family proteins. SAR by NMR, a fragment-based

SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS

327

screening technique, was used to identify binders to Bcl-XL. Starting with the two best fragment hits, a combinatorial chemistry approach was used to find the optimal linker resulting in a significantly more potent compound. The compound was further optimized using standard medicinal chemistry to improve the properties, albeit with a fairly high final molecular weight. Despite the lack of screening success for blocking the TNFa/TNF receptor interaction, a fragment-based approach led to the discovery of a small molecule that displaced one member of a trimeric protein complex, thus effectively inhibiting the TNFa signal. Further optimization to a more potent binding affinity appeared to be elusive, but the mechanism of inhibition could be replicated with other compounds. 10.2.7

Transporters

Transporters, described in Chapter 8 by Anne Hersey, Frank Blaney, and Sandeep Modi, form a protein class responsible for the controlled movement of substances across membranes. Substrates are the molecules that, upon interacting with the Transporter, are actively transported across the membrane. Inhibitors bind to transporters, but are not actively moved across the membrane, and can block substrates from passage. A number of conformational changes and activation states exist. Transporters can be direct drug discovery targets for disease intervention, and they are also avoided due to their involvement in ADMET properties of compounds. Classification of transporters resulted from a phylogenetic analysis and an understanding of the functionality. The result was seven major classes: channels and pores, electrochemical potential-driven transporters, primary active transporters, group translocators, transmembrane electron carriers, auxiliary transport proteins, and others. These major classes were further divided into subclasses and then into families (>550) and subfamilies. This classification certainly provides clarity to the quantity and variety of transporters and mechanisms available to either target or avoid in designing compounds. When designing compounds to interact with transporters, the principal methods used are structure-based design and ligand-based design. Despite the transmembrane nature of transporters, recent successes in protein crystallography have emerged providing insightful target-specific information (e.g., glycerol triphosphate transporter) as well as details for building reliable homology models. The homology models must reasonably predict fold and 3D structure from sequence relationships, expected domains and functionality. By understanding how the models are constructed, the transporter protein can be appreciated at the molecular level, and consequently areas to design interactions with small molecule are elucidated. Ligand SAR generated from screening either the binding or the functional assays can also be used to refine the homology models or can be used directly to optimize target affinity. NMR has also been used to study the ligand–transporter interactions, and can further inform the homology models. Most of the structure-based design techniques have been applied to therapeutically aligned transporter targets. However, some 3D models are emerging for key liability targets to design compounds with reduced side effects.

328

SUMMARY AND COMPARISON OF MOLECULES

QSAR analyses of ligand–transporter data from liability transporter targets have been valuable for understanding the compound features and properties that lead to problems. It is a sophisticated statistical method used to find relationships between biological data and molecular descriptors. The power of using the results of the analyses usually lies in reducing the number of compounds that should be synthesized to the most drug-like molecules. The authors use the example of retaining properties for target affinity while avoiding the properties that can lead to being a substrate for Pgp. Pharmacophore models of ligand–transporter interactions built from screening data have benefited both designing for optimizing therapeutic transporter ligands and designing for reducing interactions with liability transporters. These design tools are universal to most of the target class proteins discussed in the book. Compounds designed to selectively inhibit vacuolar ATPases were discussed, starting with a nonselective macrolide antibiotic. A crystal structure-empowered homology model created insights to understand the binding interactions as well as the mechanism of blocking proton transfer. The proton pump inhibitors for gastroesophageal reflux disease target the transporter gastric Hþ/Kþ ATPase and were originally designed as irreversible inhibitors via a disulfide bond formation stabilizing one of the major protein conformation states (EZ). Homology models in structure-based design were used to discover potent, reversible compounds that show advantages, but have not resulted in a clinically used drug as yet. Two classes of transporter proteins are involved in collecting endogenous neurotransmitters into vesicles prior to release into the synapse or removing the neurotransmitters from the synapse. Selective serotonin reuptake inhibitors were successfully developed into antidepressants. The crystal structures of bacterial homologues of the synaptic uptake proteins have made an impact on understanding the mechanism of action, developing robust homology models, and designing inhibitors. With these tools in place, further work in polypharmacy that deliberately targeted several of these transporters could make these drugs significantly better for patients. The authors went on to cover the design of elements needed to avoid interacting with transporters associated with drug liabilities. For example, transporters affect a compound’s distribution throughout the body and can cause drug–drug interactions with coadministered compounds. By studying compound interactions among certain important transporter proteins, it is possible to predict potential liability issues for a particular compound and develop SAR of structural features that are often associated with liability transporters. The activities affect CNS penetration, drug resistance, gastrointestinal tract absorption, and excretion of drugs. While there are many transporters that affect drug distribution, two of the most important ones are P-glycoprotein and organic anion-transporting polypeptides 1B1 (OATP1B1). 10.2.8

Nuclear Receptors

Drug discovery in the nuclear receptor (NR) family of physiologically important proteins is described in Chapter 9 by Hiroyuki Kagechika and Aya Tanatani. There are 48 members in the NR family, each with a unique DNA binding domain

SUMMARY OF THE UNIQUE FEATURES OF EACH TARGET CLASS

329

(DBD) and ligand binding domain (LBD) that regulate their specific target genes via endogenous, cell permeable small molecules. NRs can be classified by their DNA binding features (e.g., inverted or direct repeat sequence of DNA), and recent advances in biology, crystallography, and SAR from diverse synthetic ligands contributed to the understanding of ligand activation of the LBD. In fact, the function of NRs was often not understood until specific ligands were found. Agonists of NRs have been discovered using high-throughput screening, followed by ligand-bound crystal structures of the LBD to improve the specificity. Virtual screening can be done once the interactions that result in selective compounds are understood. Antagonists of NRs target different mechanisms such as reduction of receptor dimerization or inhibition of coactivator binding. Some compounds can act as an agonist for one subtype, whereas it acts as an antagonist for another subtype. The authors provide protein structure and mechanism of action information, which shows the compound design elements that affect tissue specificity and desired drug actions. Androgen receptor (AR) antagonists are important in prostate cancer therapy. Nonsteroidal AR antagonists are used clinically, and new ones are needed for mutated ARs that occur in cancer progression. An understanding of the folding of helix-12 (H12) during hormone activation and receptor inactivation provided a rationale for prohibiting the H12 folding via small-molecule binding in mutated ARs. An example of finding active molecules from in silico searching with defined structural features was provided to obtain an initial hit followed by structure-based design to directly affect the H12 folding. Finding agonists for the nuclear receptor VDR was done by synthesizing and screening vitamin D analogues designed to improve affinity but reduce hypercalcemic side effects. Crystal structures with bound agonists, mostly based on the core vitamin D scaffold, led to advanced compounds with desired properties. For nonvitamin D-like structures, SAR from screening a chemical library in a transactivation assay led to unique series and profiles of activity. Only a few antagonists of VDR have been developed, and details of SAR, structure-based design coupled with mechanism of inhibition, influence the design of these compounds. The authors shared an interesting rationale for focusing on carboranes as novel hydrophobic pharmacophores. By taking into account the properties of endogenous ligands, the design of carboranes (C2B10H12) as an icosahedral scaffold would offer the necessary hydrophobicity and thermal and chemical stabilities. Both agonists and antagonists were successfully designed for several NRs with potent target affinity and tissue specificity. The six retenoid nuclear receptors and their heterodimer partners are involved in specific biological responses critical for growth in animals. Original work in the research field was done based on the beneficial in vivo effects of vitamin A analogues without a knowledge of the biological target. A therapeutic index of beneficial effects in cancer or psoriasis models relative to toxicity (e.g., hypervitaminosis) and improved compound properties were used to guide the SAR. More recently, screening data, NMR spectroscopy, and X-ray crystallography are used to understand the SAR and important conformations needed for potent NR target affinity. The in vivo

330

SUMMARY AND COMPARISON OF MOLECULES

plasma levels and distribution properties of retinoid aromatic amides can be made clinically beneficial as opposed to the highly lipophilic retinoids, providing another advantage to designing molecules unlike the endogenous ligands. Selective RXR agonists (i.e., compounds that do not interact with RAR) are called rexinoids and typically have two substituted benzene rings with a short linker that provides an overall twisted conformation. The biological activity of rexinoids uncovered a unique synergistic effect from the allosteric binding to the RXR site of RXR–RAR heterodimers liganded with an RAR agonist. Understanding the ligand SAR can impact molecular design for generating agonists and antagonists.

10.3

PERSPECTIVE

Although many design methods overlap in the various protein families of the druggable genome, it however becomes readily apparent that some methods are more reliable than others depending on the properties of the biological target. Structurebased design is undoubtedly the design tool with the greatest impact. However, in target classes where protein crystallization has proven a greater challenge, creative solutions to build homology models have resulted in rapid advances in understanding mechanisms of action as well as optimizing protein–ligand interactions. Areas where screening data can be readily generated offer critical information to refine pharmacophore and homology models as well as direct SAR for optimizing compounds. Confirmation of activity in more complex biological systems such as cellular and in vivo settings has proven quite difficult for some target classes (e.g., proteases, kinases) due to the compound properties of the potent molecules. For drug intervention where a dynamic conformational change is necessary, NMR studies can be invaluable for understanding the SAR. If technology advances could further reduce costs, someday it will be interesting to see that compounds can be evaluated directly in protein crystallography in a high-throughput screening mode. Given the impact of structure-based design, comparisons of binding modes of a variety of ligands could give multiple methods of modulating a target. Instead of comparing tens to hundreds of ligand-bound structures, it might be possible to compare over 100,000 structures to get the exquisite selectivity needed for long-term, chronic disease settings where little or no offtarget activity can be tolerated. Diseases and disease pathology is complex, and will most likely require the modulation of more than one biological target (termed polypharmacy) either built into one molecule or prescribed as multiple drugs in combination. This type of combination therapy has proven vital for success in HIV and cancer, and will no doubt emerge in other disease areas. The complexity of the combinations includes the mechanism and duration of action of each component, the relative potencies, drug–drug interactions, and any off-target affects that may produce unwanted side effects. It may be possible to control some of these issues with advances in drug delivery systems.

REFERENCES

331

Ideally, we will understand enough about the major protein classes in the foreseeable future to have a ligand identified for every druggable target. With these small-molecule tools in place, biological target validation for drug intervention into the ideal component of disease pathology will be possible. Of course, it would be far more satisfying to turn these into drugs for unmet medical needs by ensuring that they have the properties necessary to reach the site of action without any toxicity.

REFERENCES Bleicher KH, Bohm HJ, Muller K, Alanine AI, 2003. Hit and lead generation: beyond highthroughput screening. Nat. Rev. Drug Discov. 2:369–378. Hopkins AL, Groom CR, 2002. The druggable genome. Nat. Rev. Drug Discov. 1:727–730. Napier ME, DeSimone JM, 2007. Nanoparticle drug delivery platform. Polym. Rev. 47 (3):321–327. Wyss DF, Eaton HL, 2007. Fragment-based approaches to lead discovery. Frontiers in Drug Design and Discovery, Vol. 3, pp. 171–202.

APPENDIX

Jerry Adams works with the Oncology Drug Discovery group at GlaxoSmithKline. His current focus is on the discovery of kinase inhibitors for the treatment of cancer. After graduating from UC Berkeley in 1979 and obtaining a PhD in synthetic organic chemistry, Dr. Adams spent the first 11 years of his career applying mechanismbased rational design principles to prepare enzyme inhibitors encompassing a variety of therapeutic areas. In 1990, his group undertook a project to identify the molecular target for a novel class of anti-inflammatory drugs termed CSAIDs for cytokine suppressive anti-inflammatory drugs. Specifically, the goal was to determine how these compounds were able to selectively suppress the synthesis of the proinflammatory cytokines IL-1 and TNF. The proteins that they identified were CSBP1/2 for CSAID binding protein. CSBP1 is now known as p38a, and the publication of this work as an article in the journal Nature was a scientific highpoint in his career (Identification and characterization of a novel protein kinase involved in the regulation of inflammatory cytokine biosynthesis. Nature, 1994, 372:739–746). The fulfillment that he has enjoyed from his work in industry would not have been possible without the help of many mentors, especially, Dr. Brian Metcalf who suggested him to work with Dr. John Lee at GSK and to elucidate the molecular target of the CSAIDs, p38a. Paul Bamborough joined the Computational and Structural Chemistry group at GlaxoSmithKline in January 1999. Currently leading a small group of scientists responsible for computational support to protein kinase programs, his work encompasses activities such as protein modeling, virtual screening for lead discovery, data Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

333

334

APPENDIX

analysis, and optimization of leads. His involvement with protein kinases began 10 years ago at Rhoˆne-Poulenc Rorer. Since then, he has worked closely with medicinal chemists and protein crystallographers to support numerous targets, and they have designed many kinase-directed compounds and arrays. Most satisfying was the discovery and optimization of the biaryl amide (BAA) class of selective p38 inhibitors, his first successful program at GSK. His greatest influences have been the colleagues who have made the experience so enjoyable. He would like to acknowledge everyone he knew during his D Phil under Prof. Graham Richards in Oxford and during his postdoctoral research in the labs of Prof. Fred Cohen and Prof. Stanley Prusiner at the University of California, San Francisco, as well as during his stay at RPR and GSK. Frank E. Blaney is a manager in the GlaxoSmithKline European Computational and Structural Chemistry group, based in Harlow, UK. He is currently involved with computational modeling support for a number of diverse projects involving implicit membrane protein targets including 7TM receptors, ion channels, and transporter proteins. In addition, he is involved in structure-based studies of cytochromes p450 and in the development of 3D models of other liability targets, such as hERG and various transporters. Dr. Blaney completed his PhD in 1974 at Queen’s University, Belfast, on synthetic aspects of functionalized polycyclic hydrocarbons. Following this, he moved to the University of Illinois as a NATO fellow, where he worked in the field of natural product synthesis. While working in medicinal chemistry, he became increasingly interested in the area of rational drug design, particularly in the application of computational techniques to it. He has published a paper on the peroxisome proliferator-activated receptor, PPARg (Int. J. Quantum Chem., 1999, 73: 97–111). No X-ray crystal data were available for any nuclear receptors at that time, but starting from the protein sequence and a small stereo ribbon diagram, Dr. Blaney was able to construct a model of the receptor and combine this with docking and quantum chemical calculations to propose a mechanism of action for the thiazolidinedione class of drugs. This was successfully used in the design of novel ligands for the receptor. David H. Drewry began his career in chemistry as an undergraduate at Yale University. He obtained PhD in 1990 from the University of California at Berkeley where he worked on the design, synthesis, and evaluation of enzyme inhibitors. His experience at graduate school convinced him that a career in creating small molecules that modulated the activity of proteins and had potential to treat disease would be exciting and fulfilling. His career in medicinal chemistry began in 1990 with Glaxo, and he is currently working with the Metabolic Diseases Drug Discovery group at GlaxoSmithKline. For 7 years his focus has been on identifying and optimizing small-molecule inhibitors of kinases to use as tools to aid in the elucidation of relevant biological pathways that are useful for treating disease in a range of therapeutic areas. Dr. Drewry’s group has done this by splitting their efforts between the design and synthesis of diverse sets of molecules targeting kinases (discovery of new kinase inhibitory templates) and pursuing a parallel medicinal chemistry approach (array synthesis addressing specific medicinal chemistry questions) to

APPENDIX

335

optimize hits found in enzyme assays. He has published a review article on the use of solid-supported reagents in organic synthesis and an another article on approaches to the design of combinatorial libraries. They remind him that anything he does in working toward the discovery of a drug is but a small piece of a very big picture, and there is much to learn from others in that area of specialty and indeed from those in other fields as well. In the complex, multivariate world of drug discovery, success comes from learning from others, embracing the different perspectives that individuals in an organization bring, and collaboration. Jo¨rg Eder, PhD, has been heading the Novartis Protease Platform since January 2007. The Protease Platform is a multidisciplinary research department dedicated to protease-directed drug discovery. Dr. Eder is a biochemist by training and his career as a scientist began at the Novartis Research Institute in Vienna in 1993. He has led various drug discovery projects in the field of arthritis and inflammation and has also worked for 2 years in Corporate Research Management/Research Planning. In 2002, he became head of the Biology Unit of the Protease Platform, a group working on the expression, purification, and assay development of proteases. One of his recent publications is on the crystal structure of the aspartic protease cathepsin E entitled ‘‘Crystal structure of an activation intermediate of cathepsin E’’ (J. Mol. Biol., 2004, 342:889–899). Maria L. Garcia is a distinguished Senior Investigator at Merck Research Laboratories. Her work is mainly directed toward the identification of novel targets for drug intervention, with special focus on membrane protein targets, and, in particular, ion channels and transporters. She has adapted novel technologies to study the function of these proteins and established assays that allow testing large number of samples in a short time. Her previous training in bioenergetics and active transport facilitated studying ion channels. Finding that limited information was available on this class of proteins, she accepted the challenge of developing drugs that target ion channels. She has published a paper in J. Biol. Chem., 1990, 265:11083–11090, which describes the identification, purification, and characterization of iberiotoxin (IbTX), a high-affinity peptide inhibitor of high-conductance, calcium-activated potassium channels. IbTX turned out to be the most selective inhibitor of a potassium channel, and consequently allowed her group and others to define the role that this channel plays in different tissues. Dr. Ron Kaback, her postdoctoral mentor at the Roche Institute of Molecular Biology, currently at UCLA, and her good friend and collaborator Prof. Roderick MacKinnon, Noble Laureate in Chemistry, 2003, have influenced her career by enforcing the values of science. Stephen L. Garland graduated in chemistry from Oxford University in 1993, and postgraduated from the University of Cambridge in 1996 under the guidance of Dr. Philip Dean. The software developed during his PhD contributed to the founding of DeNovo Pharmaceuticals in Cambridge, UK, and the methodology described in his thesis was employed by ChemBridge Research Labs in San Diego, USA, for the design of their 7TM-directed libraries. Dr. Garland spent 18 months as a

336

APPENDIX

Rhone-Poulenc Rorer Research Fellow in the Pharmacology Department at Cambridge prior to joining the Molecular Modeling group at SmithKline Beecham in Harlow. He initially worked on structure-based design for kinase targets and combinatorial library design for a wide range of target classes. Upon the merger to form GlaxoSmithKline in 2000, he was appointed to his current position as leader of the 7TM Systems Modeling Group. During his time at GSK, he has also led a number of early-stage drug discovery programs and matrix teams and currently provides computational expertise to the Cross-Portfolio Chemistry and Exploratory Chemistry teams as well as 7TM projects worldwide. His scientific interests include low tractability 7TM receptors, particularly those from Families B and C, allosterism, chemical diversity, and library design. His career from bench chemist to molecular modeling allowed him to bring together his interest in medicinal chemistry and his ability to program computers. Undoubtedly, Dr. Garland has been most influenced in his career by his father, who was head of Anti-Inflammatory, Anti-Asthmatic and Cardiovascular research at the Wellcome Research Laboratories in the United Kingdom. Although Dr. Garland has worked on a large number of projects that have yielded development candidates over the past 10 years, the FDA approval letter for Altabax (a novel topical antibiotic) in 2007 is the first one to reach the market. He believes that scientific publications are good, but getting a drug to patients is what really interests him. Tom D. Heightman’s most recent role was Head of Neurology Lead Discovery Chemistry at GlaxoSmithKline. The key activities are to identify hits from highthroughput screening or from systems- or structure-based design approaches to perform hit-to-lead chemical optimization often using parallel synthesis techniques, and to perform lead optimization to identify candidates for preclinical development. Dr. Heightman has been a medicinal chemist in industry for 10 years. He has published a review in Angewandte Chemie, 1999, following his PhD in the design and synthesis of novel beta-glycosidase inhibitors. One of his students made a compound that should have been a potent inhibitor based on stereochemical mechanism. However, the compound was inactive, challenging the current paradigm. Dr. Heightman docked the compound into a homology model and discovered that the stereochemistry was not valid for all beta-glycosidases. The docking was later confirmed by X-ray crystallography on several enzyme–inhibitor complexes, and they devised a phylogenetic classification of glycosidase stereochemistry to assist the design of new inhibitors. Dr. Heightman’s friends and family who have suffered from diseases such as multiple sclerosis, cancer, Alzheimer’s disease, Parkinsons’s disease, depression, and other diseases have motivated him to do his job. Anne Hersey is Section Head of the ADMET Modelling Group under the Department of Computational and Structural Chemistry at GlaxoSmithKline, Stevenage, UK. She is responsible for the development of ADMET models and their use in the design of compounds with good developability profiles. She obtained a BSc in Chemistry from the University of Kent at Canterbury in 1978 and a PhD from the University of Kent in 1982. From 1982 until 1996, she worked at the Wellcome

APPENDIX

337

Research Laboratories in Beckenham, primarily involved with physicochemical property measurements and their applications. In 1996, she moved to GlaxoWellcome, Ware, UK, where she became interested in using physicochemical parameters and QSAR methods for the prediction of ADMET properties and has been doing the same type of work since then. Dr. Hersey has published a paper in J. Pharm. Sci., written in collaboration with Prof. Michael Abraham’s group, on the prediction of intestinal absorption. It was a collaborative effort on the part of a number of people and was awarded the Ebert prize in 2002 by the APhA for the best paper published in J. Pharm. Sci. in the previous year. Ulrich Hommel, PhD, a biochemist by training, has been heading the Structural Sciences Unit of the Novartis Protease Platform since 2002. He is responsible for studying the interactions of proteases with inhibitors by NMR spectroscopy and determining the proteases and their complexes with inhibitors by X-ray crystallography. Before joining the Platform, Dr. Hommel was contributing for more than 10 years to various drug discovery projects in Novartis research. Among other things, this led to the discovery of allosteric integrin inhibitors, which is published in ‘‘Statins selectively inhibit leukocyte function antigen-1 by binding to a novel regulatory integrin site’’ (Nat. Med. 2001, 7:687–692). Gregory J. Kaczorowski is Senior Director of the Department of Ion Channels at Merck Research Laboratories in Rahway, NJ, USA. This department investigates voltage-gated potassium, calcium, and sodium channels as well as ligand-gated ion channels and transporters as therapeutic targets. In these efforts, he directs a diverse group of scientists specialized in biochemistry, molecular biology, or biophysics. Dr. Kaczorowski graduated in chemistry from the University of Notre Dame and obtained PhD in biochemistry from the Massachusetts Institute of Technology, where he was trained under Prof. Christopher T. Walsh and had studied enzymology and bioorganic chemistry. He was a Helen Hay Whitney Postdoctoral Fellow at the Roche Institute of Molecular Biology under Prof. Howard R. Kaback, studying mechanisms of bacterial active transport, specifically the mechanism of carrier turnover of the Escherichia. coli lactose permease. When he joined Merck in 1980 as an entry level PhD scientist, he was interested in investigating ion channels as therapeutic targets, although he had no background in physiology, pharmacology, or biophysics. He convinced Dr. John Reuben, then heading Harry Grundfest’s laboratory at Columbia University and the Director of the Grass Fellows program at the Marine Biology Laboratory in Woods Hole, to let him ‘‘unofficially’’ join the Grass Program during the summer of 1981 and to teach him electrophysiology. He learned to make intracellular recordings with sharp electrodes using GH3 pituitary cells. The first paper on patch voltage clamping was published at the end of that summer, and using the same cells as a model system, he made his first successful voltage clamp and single channel recordings in 1982; the rest is history. Dr. Kaczorowski would like to thank Dr. P. Roy Vagelos, Dr. Ralph Hirschman, and Dr. Gene Cordes for hiring and mentoring him at the Merck Research Laboratories and for allowing him to follow his dream of establishing an ion channel drug discovery

338

APPENDIX

and development department at Merck. He dedicates this review to Dr. John Reuben and Dr. George Katz who taught him electrophysiology and, more importantly, to be rigorous in ion channel research. Hiroyuki Kagechika obtained his BSc and PhD from the University of Tokyo (Pharmaceutical Sciences), Japan, in 1989 and was employed as an Associate Professor there until 2004. Currently, he is a Professor in the School of Biomedical Science at the Tokyo Medical and Dental University, with his major research interest in medicinal chemistry of nuclear receptors. His interest also lies in chemistry of aromatic amides with unique structures, which is useful for construction of functional aromatic molecules. He is engaged in scientific research and education in the field of chemical biology and medicinal chemistry; in addition, he has provided advice to enhance Japanese scientific systems and has conducted surveys and research on science promotion policies. When he started the project on medicinal chemistry of retinoid in 1983, specific receptors for retinoid were unknown; the purpose of his study was to develop novel drug for differentiation therapy of cancer. He has published a paper in J. Med. Chem., 1988, 29:6279, which describes design, synthesis, and biological activities of a novel synthetic retinoid, Am80. The compound was approved as a drug (general name: tamibarotene) for intractable and relapsed acute promyelocytic leukemia (APL) in Japan (2005). Among the people who most influenced his career is Dr. Koichi Shudo, a professor emeritus of University of Tokyo, Japan, who taught him the true charm of drug development. Karen E. Lackey currently works with GlaxoSmithKline as a Vice President of Molecular Discovery Research Chemistry with global accountability for pioneering exploratory chemistry in support of all therapeutically focused Centres of Excellence in Drug Discovery and for the strategy to design and implement compound collection enhancement to support all screening events and new technology. In her previous role as International Medicinal Chemistry Director, she was responsible for scientific strategy in the kinases, enzymes, ion channels, and cellular pathways medicinal chemistry research areas. Her team provided support to advance more than 55 early stage programs in multiple therapeutic areas, including cardiovascular diseases, oncology, inflammation, psychiatry, neurology, respiratory diseases, gastrointestinal diseases, metabolic diseases, infectious diseases, and virology. She has been involved in drug discovery for over 20 years. David D. Miller is the Director of Chemistry in the Discovery Chemistry Department of the Respiratory and Inflammation Centre of Excellence of Drug Discovery at GlaxoSmithKline. He is currently responsible for system-based research into novel ion channel modulators and kinase inhibitors, with a particular emphasis on respiratory and inflammatory diseases. Trained as a pharmacist, he began his career in industrial medicinal chemistry with Wellcome dealing in antiparasite and then moved to oncology research with increasing focus on biological target class modulator discovery, encompassing proteases and integrins. The publications on the development and exploitation of chiral glycine enolates that came from his postdoctoral

APPENDIX

339

work in Dieter Seebach’s laboratory in the ETH were particularly special for him. ‘‘The chiral glycine enolate derivative from 1-benzoyl-2-(tert-butyl)-3-methyl-1,3imidazolidin-4-one is alkylated in the 5-position with relative topicity lk’’ (Helv. Chim. Acta, 1985, 68: 949–952) and the ‘‘Addition of chiral glycine, methionine, and vinylglycine enolate derivatives to aldehydes and ketones in the preparation of enantiomerically pure a-amino-b-hydroxy acids’’ (Helv. Chim. Acta, 1987, 70: 237–261) capture this work. Early in his career with Wellcome, he worked with a highly experienced and successful medicinal chemist who was outstanding in probing into a problem to reach the key question which, if answered, would unravel the whole problem. Dr. Miller learned a great deal from him about choosing which questions to answer during research projects. Sandeep Modi is an investigator in the ADMET Modeling Group under the Computational and Structural Chemistry Department at GlaxoSmithKline, Stevenage, UK. In support of several drug discovery projects, which have various ADME issues, he is also responsible for the development of local/global ADMET models. In 1986, he obtained a master’s degree in organic chemistry from Delhi University, and in 1990 a PhD from Tata Institute of Fundamental Research, India. After accomplishing postdoctoral research at the University of Cambridge in 1993, he moved to Professor Gordon Robert’s Lab in the Centre of Mechanisms of Human Toxicity Unit in Leicester. In 1997, he joined GlaxoWellcome, Ware, UK, to work on expression, purification, and SAR for several cytochrome human P450s, and then worked on in silico models on CYPs. Since then, he has learned several QSAR methods for the prediction of ADMET properties. Dr. Modi has published a paper in Nat. Struct. Biol. in 1996 on the catalytic mechanism of cytochrome P450 BM3, showing that reduction during the cycle leads to a big movement of the bound substrate. Richard Sedrani, PhD, an organic chemist by training, has been leading the Medicinal Chemistry Unit of the Novartis Protease Platform since 2002. His group design and synthesize potent, selective, and bioavailable inhibitors of proteases, which have a therapeutic relevance, with the ultimate objective of identifying drug candidates. Prior to joining the Protease Platform, Dr. Sedrani was involved in several projects, including the project that he led in the area of immunology and transplantation. His work led, among others, to the discovery of Certican1, a drug used for the prevention of graft rejection after solid organ transplantation, which is published in ‘‘Chemical modification of rapamycin: the discovery of SDZ RAD’’ (Transplantation Proc., 1998, 30:2192). Lisa Shewchuk joined Glaxo as a crystallographer in 1993, after obtaining her PhD in biochemistry under Prof. Christopher Walsh at Harvard Medical School and a postdoctorate degree in protein crystallography under Prof. Brian Matthews at the University of Oregon. Dr. Shewchuk is currently working as Section Head of the Structural Sciences Department under Molecular Discovery Research at GlaxoSmithKline in Research Triangle Park, NC. Her department is responsible for determining the three-dimensional structures of proteins, complexed with

340

APPENDIX

drug-like molecules at atomic resolution. These structures highlight potential routes for chemists to improve the potency, selectivity, and physiochemical properties of their molecules. Her department has had a strong focus on kinases and nuclear receptors for the past 10 years, solving more than 20 different members for each of these families. Aya Tanatani obtained PhD from the University of Tokyo in 1998. After her research at the University of Illinois at Urbana-Champaign, USA, and Kanagawa University, Japan, as a postdoctoral fellow, she worked at Institute of Molecular and Cellular Biosciences, the University of Tokyo, as an assistant professor. In 2008, she moved to Ochanomizu University where she began her research in her own laboratory as an associate professor. Her major research interest is medicinal chemistry of nuclear receptors and chemistry of aromatic foldamers. Adrian Whitty is an Associate Professor in the Chemistry Department at Boston University. He worked for 14 years at Biogen Idec as Director of Physical Biochemistry, where he led a group responsible for the structural, biophysical, and mechanistic study of drug targets and protein and small-molecule drug candidates. Adrian obtained a BSc in chemistry from the University of London, a PhD in organic chemistry from the University of Illinois at Chicago, and a postdoctoral degree under late Prof. William P. Jencks from Brandeis University. He joined Biogen (now Biogen Idec) in 1993. His research included elucidation of enzyme mechanisms and enzyme-inhibitor interactions as well as mechanistic investigations of integrins, immune cell costimulatory molecules, and a number of cytokine and growth factor receptors. The unifying theme of his work has been to understand how binding energy is generated through protein–protein or protein–small-molecule interactions and how it is used to achieve biological function and specificity. In recent years, a major topic of study has been how the interactions between receptor components in the two-dimensional environment of the cell membrane govern the ability of cells to sense and respond to the extracellular environment (e.g., Schlee et al., Nat. Chem. Biol., 2006, 2:636–644). Another area of focus has been the development of smallmolecule inhibitors that block protein–protein interactions, especially as approached using fragment-based methods for lead identification. Between 2002 and 2005, Adrian coled a collaboration between Biogen Idec and Sunesis Pharmaceuticals, aimed at discovering small-molecule inhibitors of a set of difficult protein–protein interaction targets including TNFa (a part of which is described in He et. al., 2005, Science, 310:1022–1025).

INDEX

AAL-993 134 ABC superfamily 238 Abraham A descriptors 247 Abraham B descriptors 247, 262 ABT737 222 Acid pump antagonists (APAs) 251 AcompliaTM 28 Activation Function (AF) domain 279, 281 Activation loop 122 ADAM/ADAMTS 171, 175 Adapalene 309 Adhesion 18 Adrenomedullin receptor 28 ADMET properties 7, 236, 327 Affinity threshold 219 AGN-193109 305 AGN-193836 303 Aldosterone 282 Aliskiren 166 All trans retinoic acid (ATRA) 293, 298, 310 Allosteric binding 15, 22, 209, 319 Allosteric cooperativity 77, 321 AlphascreenTM technology 35

AM68 296 AM80 290, 297, 309, 310 AM580 297 AMBER 243 Ambit technology 139 Amitriptylline 253 Amprenavir 260 Amylin receptor 28 Analogue-Based Drug Design 12 Analytical ultracentrifugation 219 Angiotensis converting enzyme (ACE) 168 Angiotensinogen 166 Antiporters 236, 238 Apamin 64 AptivusTM 165 Arrestin 33, 320 Assay artifacts 216, 218 Assay hit(s) 5, 217 Astemizole 60 ATP binding cassette 237 Atropine 75 Augmentation therapies 253 Autosomal inherited diseases 57, 60

Gene Family Targeted Molecular Design, Edited by Karen E. Lackey Copyright # 2009 John Wiley & Sons, Inc.

341

342 BACE (b-amyloid precursor protein cleaving enzyme) 163 Bacteriorhodopsin 24 Bafilomycins 249 Barbourin 91 Batimastat 175 BCL-2 212, 220 BCL-XL 212, 220 Beta-2 adrenergic receptor 23, 29, 319 Bexarotene 310 Bfx80 281, 302 BIBN4096 41, 320 Bicalutamide 283, 285 Bicartamide 285 Biopharmaceutical Drug Design and Development 11 BIRB-796 134, 136, 139 BIRT377 105 BMS-961 303 BMS-587101 106 Boron neutron capture therapy (BNCT) 290 BR403 290 Bradykinin-potentiating peptides (BPP) 169 ByettaTM 21 Calcipotriol 286 Calcitonin 21 Calcitonin receptor-like receptor (CLR) 41 Calcium-sensing receptor (CaSR) 22, 47 Calorimetry 219 Captopril 168, 169 Carazolol 23, 24 Carboxypeptidase A (CPA) 169 Casein Kinase II 137 CasodexTM 283 Catalytic cleft 122 Catalytic triad 175, 183, 325 CB1 28 CCR5 receptor 28, 36, 46 CCR2b receptor 28 CD80 210 Cell Adhesion molecules (CAMs) 98 Cell-Adhesion receptors 85, 209 Cell-based assays 6, 137 Cellular retinoic acid binding protein (CRABP) 310 CGP038560A 166

INDEX

CHARMM 243 Charybdotoxin 63, 64 Chemical space 214, 215 Cheng-Prusoff equation 138 Cilengitide 91 Citalopram 253 Clotrimazole 64 Collagens 89 Comparative molecular field analysis (CoMFA) 246, 256 Comparative molecular shape indices analysis (CoMSIA) 248 Compound Evaluation Pathway 4, 5 Computational predictions 225 Concanomycins 249 Constitutive interactions 201, 326 Correolide A 73 CRF2b 26 CrixivanTM 165 Cromakalin 58 Crystallographic B factors 212 CT gene-related peptide (CGRP) 28 C-Type channel Inactivation 58, 76, 77 Cyclic AMP-dependent Kinase 120 Cyclin-dependent kinase (CDK) 122, 126, 129 Cyclosporine A 261 Cyproterone acetate 283 Dasatinib 143 Delayed-type hypersensitivity 71 De novo design 125 Deorphanization 17 Developability 10 Deubiquitinating proteases 183 Dihydrotestosterone 283 Dipeptidyl peptidase See DPP4 Diversity Screening 4, 36 DMP543 61 DMPK 7, 45 Dopamine 36 DOV-21947 254 DovonexTM 286 DPP4 175, 177 Drug-drug interactions 257, 260 Druggability 3, 207, 218, 225 Druggability Index 226 Druggable Genome 3, 212, 224, 235, 317

343

INDEX

Drug-like molecules Duloxetine 254

212

Elastase inhibitors 189 Electrophysiological ion channels assays 70 Enalaprilat 169 Energy-based approaches 225, 243 Eptifibatide 91 ER-27191 305 ERK1/2 28, 33 Escitalopram 253, 256 Estradiol 282 Etretinate 296 Exenatide 21 Factor Xa 179 False positive rate 218 Fendiline 48 Fibronectin 98 Fibrinogen 89, 92, 110 FK506 75 Fluovoxamine 253 Fluoxetine 253 Flutamide 283 Flux measurements 69, 71, 76 Flux ratio 260 Fragment based approaches 173, 326 Fragment based screening 107, 125, 189, 214 Frizzled 18, 29 G alpha-beta-gamma trimer 27 Gamma-aminobutyric acid (GABA) receptor 22 Gastric ATPase inhibitors 235 Gastroesophageal reflux disease (GERD) 251 Gatekeeper 124 Gene expression 2 Genetics 2 Genetic validation 69 Genomics 2, 120 GF120918 259 GIP receptor 25 GleevecTM 134, 140 Glutamate 18, 22, 30 Glycerol triphosphate transporter (GlpT) 239

GRAFS system 18, 319 Ghrelin 30 GPCR kinase (GRK) 33 Growing fragments 217 GSK773812 320 Guidebook on Molecular Modeling in Drug Design 11 GW2016 148 GW2974 147 5HT2a 28 hERG 45, 47 Hidden Marko model (HMM) 240 High-content lead series 6 Histamine 36 Histamine H2 antagonists 251 Hits, screening 207, 213 HIV protease 162, 164 Hormone response elements 276 Hot spots 204, 206, 326 HX531 308 Hydropathy plots 240, 241 Hydrophobicity plots 21 Hydroxyflutamide 284 Hydroxytamoxifen 282 Hypoglycemic agents 280 Hypolipidemic agents 280 Iberiotoxin 63 I-Domain allosteric site (IDAS) 91, 103, 104, 108 IKK 127 Imipramine 253 In silico methods 8, 9, 99, 246 In silico predictions 9 In vitro assays 4, 6, 135 In vivo properties 7 InderalTM 23 Indinavir 165 Interaction energy 206 Interactome 199 Intercellular adhesion molecule-1 (ICAM-1) 103, 108 Interface atoms 203 Interleukin-1B converting enzyme (ICE) 188 Interleukin-2 210, 211 Intestinal transporters 258 IressaTM 144

344

INDEX

ISOP series 285 Isosteres 44, 93, 128, 162, 173, 300 JanuviaTM 179 JNK3 128 Juxtamembrane (J-domain)

26

KaletraTM 165 Kir Channels 56, 321 Kristin 108, 110 Kruppel-like zinc finger transcription factor 5 (KLF5) 311 Kyte-Dolittle scale 240 L-006235 188 Lactose permease 239, 244 LANCETM technology 35 Lapatinib 136, 148 Lead series, high-content 6 Leupeptin 185 LFA703 104 LFA878 104 LG100754 308 LG101506 308 LG190090 287 LG190178 287 LGD1069 310 Ligand-Dependent transactivation 279 Ligand efficiency 46, 217 Ligand-inducible transcription factors 276 Ligand superfamily concept 303 Linorpidine 61 Lipophilicity 8 Lisinopril 169 Long QT syndromes 60 Lopinavir 165 Lotrafiban 95 LovastatinTM 103, 105 L-site 103 L-Type calcium channel modulators 69 LY-139603 256 LY-367265 255 Lymphocyte function-associated antigen-1 (LFA-1) 103 Major Facilitator Superfamily (MFS) 238, 253 Maraviroc 46 Margatoxin See MgTX

MAY0282 280 MC903 286 Mechanism of modulation 32 MEK1 134 Melagatran 181 Merging fragments 217 Metzincin 171 mGluR1 26 mGluR2 28 mGluR3 26 mGluR7 26 MgTX 70, 74 MIDAS 87, 89, 91, 93, 98 Milnacipran 254 MK-0974 320 Molecular mechanics 243 Monte Carlo Simulations 244 MthK 68 Mucosal addressin cell adhesion molecule-1 (MAdCAM-1) 98, 101 Multidrug resistance-associated proteins (MRPs) 258 Multiple linear regression (MLR) 246 Muscarinic acetylcholine receptor-sensitive M-current 61 Natural products 213, 321, 326 Nelfinavir 165 Nerve growth factors 210 Neural networks 246 Nitric oxide synthase 210 NMR screening 208, 226 NMR studies 187 Noradrenaline 36 Nortriterpene 71 NorvirTM 165 N-Type inactivation 58 N-Type voltage-gated calcium channel

69

OATP1B1 261 OATP1B3 241 Orbofiban 95 Orexin-1 receptor 28 Organic anion transporters (OAT) 236, 258 Organic cation transporters (OCT) 236, 258, 266 Omeprazole 251 O-ring hypothesis 205 Orphan receptors 40, 278

INDEX

Orthosteric binding 15, 19, 25, 30 Oxyanion hole 175, 183 P32/98 179 P38 126, 134 P53 209 Papain 183, 185 Parathyroid hormone-1 (PTH1) 21, 25, 29 Paroxetine 253, 256 Partial least squares (PLS) 246 Partial least squares discriminant analysis (PLS-DA) 246 Partition coefficient 247 Paxilline 63 PD173995 134 PD184352 134 Penitrem A 63 Pepstatin A 163 Peptidomimetic discoveries 91, 96, 99, 162, 323, 324 P-glycoprotein 236, 245, 259 Pharmacogenomics 3 Pharmacokinetic profile 97 Pharmacophore model 9, 98, 99, 125, 245 Pioglytazone 307 PKA 120, 123 PLK 127 Poisson-Boltzmann algorithms 243 Polar surface area 205, 247 Polypharmacy 330 Positive allosteric modulators (PAMs) 29, 49 Potassium efflux 57 Privileged Structures 36, 37 Probe dependence 30 Prodrugs 95, 182, 189, 251, 283, 322 Progesterone 282 Propranolol 23 Protein-ligand Interactions: From Molecular Recognition to Drug Design 11 Protein shift 43 Proteolysis 159 Quantitative Structure-Activity relationship (QSAR) 245, 265, 328 Quantum mechanics 243 R-467 48 R-568 48

345 Raloxifene 283 RAMPs 28, 29, 41 Reboxetine 256 Receptor activity modifying proteins See RAMPs Receptor-Based Drug Design 12 Rectification 56 Relacatib 185 Remikiren 166 Renin 163, 166 Retigabine 61 Retinobenzoic acid 300 Retinoic acid receptor (RAR) 275 Retinoid synergism 308 Rexinoid 306, 330 Rhodopsin 18, 20, 21, 23, 24, 26, 29, 49 Rimonabant 28 Ritonavir 165 Rivaroxaban 182 RO26-4550 211 RO41-5253 305 ROCK 128 Rossmann fold 67 Rotational resonance NMR 244 Rule of Five 101, 222, 248 Rule of fours 262 RXR-related receptors 277 Salbutamol 23 SalmeterolTM 23 Saquinavir 165 SAR by NMR 220, 226 SB-203580 128 SB-649915 254 SB-744185 254 SCH28080 251 Scissile bond 160, 163 Secretin 18, 21, 29 Selection funnel 220 Selective ER modulators (SERMs) 283 Selective serotonin reuptake inhibitors (SSRIs) 235, 253 Selectivity 6, 22, 29, 45, 64, 68, 74, 112, 139, 325 SereventTM 23 Serine trap 176, 177, 181 Serpins 209 Serotonin 36 Setindole 60

346 Sertraline 253 SH2 domains 119, 209 SH3 domains 119, 209 Shape-based approaches 225 Shape complementarity 205 ShK 70 Sibrafiban 95 Simple Rules 248 Sitagliptin 179 Site-directed mutagenesis (SDM) 25, 40, 49, 73, 240, 244, 250 SKF38393 36 SKF86466 36 SKF96067 251 SKF97574 251 SKF102839 37 SKF189254 37 Solubility 9 SoLute Carrier (SLC) Transporters 237, 253 Solvent-accessible protein surfaces 202, 204 Somatostatin receptor 39 Sorafenib 148 SR11237 306 SR141716A 28 State-dependent blockers 66 Steroid hormones 277 Stichodactyla helianthus peptide See ShK Structural Adaptivity 211 Structure Activity Relationships (SAR) 5, 7, 99, 147, 219, 244, 327 Structure Based Drug Discovery 12 Surface Plasmon resonance technology 219 Sushi domain 26 Symporters 236, 238 Synaptic reuptake transporters 253 Talin 86 Tamoxifen 283 Tamibarotene 310 TarcevaTM 144, 147 TargretinTM 310 Tazalotene 309 TEI-9647 288 Tertiapin 58 Testosterone 283 Tethering 189

INDEX

Textbook of Drug Design and Discovery 10 Therapeutic index 69 THC 282 Threading, protein 239 Thrombin 179 Tipranavir 165 Tirofiban 94 TMPIP 244, 252 TNF-alpha 210, 222 TNF-converting enzyme 175 Toggle switch model 33 Topology 207, 209, 217, 241, 326 Torsade de pointes 60 Tractability 3, 4, 17, 35, 47, 125, 217, 317 Trafficking 32, 57, 60 TRAM-34 64 Transactivation 32 Transient interactions 201, 203, 326 Transition state mimics 164, 167 Transport proteins 258 Troglitazone 307 Trypsin-like serine proteases 181 TTNPB 295 TykerbTM 146, 148 TZ335 280, 302 UCL1684 64 Undruggable 200 Uniporters 236, 238 Validated molecule 218 Valinomycin 238 Van der Waals interactions 167, 203, 243 Vascular cell adhesion molecules (VCAM) 89, 100 VCAM-1 98, 101 Venlafaxine 254 VentolinTM 23 Venus fly trap mechanism 26, 47, 319 Verruculogen 63 Vesicular transporters 253 Vildagliptin 177 Vitronectin 111 ViraceptTM 165 Virtual Compounds 8, 100, 128 Virtual Screening 8, 39, 40, 112, 280, 329 VolSurf 248 Voltage sensor 66

347

INDEX

Von Willebrand Factor A domains VX-745 139 WY47766

251

XE991 61 Ximelagatran

181

86

Zankiren 166 ZD-7349 92 Ziconotide 69 Zimelidine 253 Zinc binding motif 168, 169 Zinc finger motif 276 ZK168281 288

FIGURE 1.1 The gene families of proteins are classified by function and common structure motifs as can be seen by the representative structures for kinases, integrins, GPCRs, ion channels, proteases, and nuclear receptors.

FIGURE 2.1

Examples of marketed drugs targeting GPCRs.

FIGURE 2.2 Semischematic representation of a G-protein coupled receptor. The picture is generated from the coordinates of the b2-adrenergic receptor crystal structure 2rh1 with transmembrane helices colored as follows: TM1 (orange), TM2 (green), TM3 (light blue), TM4 (dark blue), TM5 (violet), TM6 (red/brown), and TM7 (pink). The extracellular and intracellular loops are colored yellow, although ICL3 is missing due to excision of T4-lyzozyme. Also shown are helix 8 (purple) and the helical section of ECL2 (cyan).

(i)

(ii)

(iii)

FIGURE 2.3 Schematic diagram showing the differences in binding of endogenous ligands by Family A (i), Family B (ii), and Family C (iii) GPCRs. The approximate position of the orthosteric site is shown in green.

FIGURE 2.4 G-protein signal transduction pathway steps (orange ellipses) and typical associated assay formats (magenta text).

FIGURE 2.7 Three-dimensional overlay of aryl piperidine privileged structures, showing incremental positioning of aryl and hydrogen bonding groups.

FIGURE 2.11 Three-dimensional overlay of energy minimized piperidinyl-spirohydantoin and -quinazolinone substructures, showing similar positioning of key hydrogen bonding groups.

FIGURE 3.1 Voltage-dependent ion channels possess similar architectural features. Ion selectivity resides within the pore domain. When a gating domain is covalently attached, different functional properties result. Functional diversity also occurs if auxiliary subunits associate with the ion channel. Other mechanisms, such as phosphorylation (green diamonds), contribute to the regulation of ion channel function. From Garcia and Kaczorowski, Potassium channels as targets for therapeutic intervention, Sci. STKE 2005, pe46 (2005).

FIGURE 3.2 The voltage-gated ion channel superfamily. Four-domain calcium and sodium channels are shown as blue branches, potassium channels as red branches, cyclic nucleotidegated channels as magenta branches, and transient receptor potential and related channels as green branches. From Yu and Catteral, the VGL-chanome: a protein superfamily specialized for electrical signaling and ionic homeostasis. Sci. STKE 2004, re15 (2004). Reprinted with permission from AAAS.

FIGURE 3.3 Ion conduction in potassium channels. (a) Two of the four pore domain subunits of a potassium channels are shown. (See text for full caption.)

FIGURE 3.4 Crystal structure of a mammalian voltage-gated potassium channel. Stereo views of the Kv1.2–b2 subunit complex. The four subunits are colored differently. In (a), TM indicates the integral membrane component of the complex. (b) A single subunit of the channel and b subunit are viewed from the side. (c) A view from the extracellular side of the pore. From Long et al. Crystal structure of a mammalian voltage-dependent Shaker family Kþ channel. Science 309: 897–903 (5 August 2005). Reprinted with permission from AAAS.

FIGURE 3.5 Human T cell activation. Sustained Ca2þ influx through Ca2þ -releaseactivated calcium (CRAC) channels is required for lymphokine release and T cell proliferation. Kv1.3 channels hyperpolarize the membrane and facilitate Ca2þ entry through CRAC channels. Blockade of Kv1.3 with peptides or small molecules prevents T cell activation in vitro and in vivo.

FIGURE 3.6 Scheme of a functional Rbþ efflux assay used for identifying potassium channel modulators.

FIGURE 3.8 Docking of correolide C in Kv1.3. A model of Kv1.3 with a bent S6 helix was generated using the crystal structure of the KcsA channel as a template and modified using experimental data constraints. In the displayed orientation, the 3-keto group of the E-ring ester points to the selectivity filter, the saturated hydrocarbon face of the molecule interacts with the hydrophobic wall of the channel, and the other face, with four acetyl groups, lies in the water-filled cavity. The bromobenzyl group provides binging energy through van de Waals interaction with Pro425 in the channel.

FIGURE 4.2 A generic integrin structure. (a) Bent, ‘‘inactive’’ conformation. (b) Extended ‘‘active’’ conformation. (c) Integrin containing I-domain in the a-subunit. bTD, b-terminal domain; EGF, epidermal growth factor domain; PSI, plexins, semiphorins, and integrins domain; Mg, metal ion-dependent adhesion site (MIDAS).

FIGURE 4.3 Modes of action of known integrin antagonists. A depiction of ligands and antagonists binding to integrin head-piece of (1) non I-domain and (2) I-domain integrins (Adapted from Shimaoka and Springer, 2003).

FIGURE 4.10 Pharmacophore-directed discovery of novel avb3 inhibitors. Structures are color-coded according to common pharmacophoric features.

FIGURE 4.18

(a) The aLb2 binding epitope on ICAM-1.

FIGURE 5.1 (a) Kinase sequence alignment highlighting key structural features. Residues close to ATP are underlined. (b) Ribbon representation of PKA using the same color scheme as (a). (c) Interactions of ATP in the binding site of PKA.

FIGURE 5.8 (a) CDK2 compounds containing the phenyl 4-sulphonamide group. (b) Overlaid p38 inhibitors showing their back-pocket binding groups. (c) Complexes of Abl/ Gleevec (green) and p38 / BIRB-796 (blue) showing the DFG-out binding mode.

FIGURE 6.11 Batimastat 12 and crystal structure of the MMP-3/12 complex. Carbon atoms of the protein and the inhibitor are colored cyan and green, respectively (1mmb, Grams et al., 1995).

FIGURE 6.24 Caspase-1 inhibitors. The substrate-based inhibitor Ac-Tyr-Val-Ala-Asp-H is shown in (a) indicating atoms involved in hydrogen bonding with the protein. The crystal structure of the caspase 1/inhibitor complex (1ice, Wilson et al., 1994). Carbon atoms of the protein and the inhibitor are colored cyan and green, respectively. Hydrogen bonds between the inhibitor and the protein are indicated by dashed lines in magenta. Inhibitors based on the pyridone.

FIGURE 7.2 Conformational changes observed at the surface of IL-2 upon binding the small-molecule inhibitor RO26-4550. Reproduced from Whitty and Kumaravel (2006). # 2006, Nature Publishing Group.

FIGURE 7.3 Fragment-based screening. (a) Schematic representation of the fragment-based screening process. (b) A library of N fragments samples the same chemical space as a conventional library of N 2
L larger molecules of structure Fragment–Linker–Fragment. Reproduced from Whitty and Kumaravel (2006). # 2006, Nature Publishing Group.

FIGURE 7.6 Application of the FBS technique SAR by NMR plus traditional medicinal chemistry optimization strategies led to Bcl-2 inhibitor ABT737. Reproduced with permission from Oltersdorf et al. (2005). # 2005, Nature Publishing Group.

FIGURE 7.7 SP307 disrupts the constitutively trimeric cytokine TNFa by displacing one subunit. Reproduced with permission from He et al. (2005). # 2005, AAAS.

FIGURE 8.4 Stages in the construction of a model of human OATP1B3. (See text for full caption).

FIGURE 8.6 (a) Schematic structure of the vacuolar V_ATPase (adapted from Sun-Wada et al. 2004). (b) Model of adjacent subunits forming the proteolipid ring of N. crassa showing residues that have been mutated and that affect bafilomycin binding. (c) Bafilomycin docked into its putative binding site. This binding prevents rotation of the helices, thus disabling the movement of the acid residues that are believed to be involved in proton movement.

FIGURE 8.10 (a) SB-744185 docked into the model of the 5-HT1A receptor. In addition to the salt bridge between the protonated nitrogen and the TM3 aspartate, numerous other H-bonding and hydrophobic interactions are observed. (See text for full caption.)

FIGURE 9.4

Mechanism of ligand-dependent transactivation of nuclear receptors.