3D Structure and the Drug Discovery Process

RODERICK E. HUBBARD

Vernalis (R&D) Ltd, Granta Park, Abington, Cambridge CB1 6GB, UK and University of York, Structural Biology Lab, York, YO10 5YW, UK

1 Introduction

The past 30 years has seen an accelerating increase in our understanding of the molecular mechanisms that underlie disease processes. This has had a fundamental impact on the process of drug discovery, and most of modern pharmaceutical research is based on target-focused discovery, where the goal is to affect the biological activity of a particular molecular target to provide a cure or treatment for a disease. As the 3D structures of some of these targets have become available, a range of experimental and computational methods have been developed to exploit that structure in drug discovery. These developments and some of their applications are the subject of this book.

In a target-focused approach, the cycle of discovery is very similar with or without a structure for the target. Initial-hit compounds are found that bind to the target and enter a medicinal chemistry cycle of making compound analogues and testing in suitable biological models. From this, the chemist builds hypotheses of what is important for the activity. Using experience (or inspired guesses) the chemist then makes changes that should improve the properties of the compound and the cycle of synthesis, testing and design begins again. These hypotheses develop a model of the conformations the compounds adopt, the chemical surfaces they project and the interactions made with the active site. For example, the optimisation of sildenafil (Viagra), included consideration of the electronic properties of an initial-hit compound and how it could be improved to more closely mimic the known substrate in the active site of phosphodiesterase, many years before the structure of this enzyme was known.

Nowadays an appreciation of the 3D structure of both the compounds and their target are a part of just about every drug-discovery project. This target structure can be experimentally determined, a model constructed on the basis of homology or a virtual model of the receptor created on the basis of the chemical structure of the known active compounds. In addition, computational methods such as virtual screening and experimental methods such as fragment screening can generate many new ideas for compound templates and possible interactions with the active site. The major advantage of experimentally determining the structure of these different compounds bound to the target is to increase the confidence in the hypotheses and increase the scope of subsequent design. This encourages the medicinal chemists to embark on novel and often challenging syntheses in the search for novel, distinctive and drug-like lead compounds. Our ability to predict conformational changes in proteins and the binding energy of protein-ligand complexes remains relatively poor, so there is still plenty of scope for experience, inspiration and guess work in the details of design.

This book will provide an overview of the methods currently used in structure-based drug discovery and give some insights into their application. Essentially, all of the examples and methods focus on proteins as the therapeutic target. There has been considerable progress in the structural biology of RNA and DNA molecules and these classes of molecules are the recognised target for some successful drugs. For DNA, our understanding of the binding of compounds that intercalate or bind to the small groove is reasonably well advanced (for an early example, see Henry; current perspectives are provided in Tse and Boger, and Neidle and Thurston,). There have also been spectacular advances in determining the structure of whole ribosome sub-units and of representative portions of the ribosomal RNA in complex with known natural product antibiotics. These structures have led to some hope that rational structure-based methods may be applied against the ribosome and also other RNA targets where a particular conformation has a role in disease processes (Knowles et al., 2002). Although there has been some progress and it has been possible to discover compounds with reasonable affinity for RNA, there remain considerable difficulties in designing small, drug-like molecules with the required specificity to discriminate between the very similar sites presented on RNA. For these reasons, the discussions in this book focus on proteins as the therapeutic target.

2 The Drug Discovery Process

As discussed in the Preface, drug discovery is an expensive and time-consuming activity that mostly fails. Retrospective analyses of the pharmaceutical industry during the 1990s estimate that each new drug in the market takes an average 14 years to develop, costing in the region of $800 million. In addition one in nine compounds that enters clinical trials makes it to the market. The attrition rate in discovery research is similarly high. Depending on the company, therapeutic area and discovery strategy, at best only one in ten research projects that begin with a starting compound will generate an optimised candidate to enter clinical trials. For these reasons, most companies maintain a pipeline with a large number of projects in the early stages, taking a diminishing number forward at each stage. The discovery process gets more expensive as you proceed, hence careful management of the portfolio is essential. The key is to make the right decision at the right time – knowing when to stop a project is often more important than committing to continuing.

Modern, target-oriented drug discovery is usually organised into a series of stages. The definitions of these stages differ from company to company and the details of the boundaries will vary from project to project. The following discussion provides an illustration of the stages, their purpose and duration and the types of resources involved. Clear criteria need to be established for moving from one stage to another as, in general, the stages become progressively more resource and expense intensive (Figure 1).

2.1 Establishing a Target

Clearly, the starting point for a target-oriented drug-discovery project is to identify a relevant target. In the pre-genomic era, targets were discovered through cellular and protein biochemistry methods, where a detailed understanding of the origins of a disease led to isolation and characterisation of key protein molecules. Examples presented in the applications section of this book include neuraminidase described by Colman for anti-influenza therapies and the factor Xa work described by Liebschutz and colleagues to produce anti-thrombotic agents. The nature and significance of these targets were established before much of the modern machinery of molecular biology and genomics methods were available.

The approach to biological research has undergone dramatic changes in the past decade, with successions of omics technologies becoming available. Genomics has recorded the sequence of nucleic acid bases in many genomes, and continuing bioin-formatics analyses are identifying the coding regions. Comparing the genomes of both pathogen and host organism can identify potential target genes. Transcriptomics methods monitor the identity and levels of RNA transcribed for each gene, and there have been high hopes that comparison of "normal" and diseased cells will identify targets. There is a vast literature in these areas – Egner et al. provide an introduction to the methods, and the recent critique by Dechering points out some of the pitfalls. There has been considerable interest (and investment) in applying these methods to find new targets for different diseases and conditions. As the first genomes began to appear, there was intense interest in identifying what all the genes were. An example of a target discovered in this way is the beta form of the estrogen receptor (see Manas et al. in this book).

Whatever the mechanism of identifying a target, there needs to be some level of validation before nominating it for a drug-discovery project. The phrase "target validation" is much misused – a target cannot be said to be truly validated until a drug that uniquely affects that target is on the market. Even then, there can be issues such as the recent challenges facing COX-2 as a target following adverse effects (see 24 February 2005 news item in Nature,433, 790).

In general, the requirements for a target are to establish a biological rationale for why affecting the target will have the desired therapeutic benefit. This can include assessing the viability of the organisms produced with a particular gene removed, either through knock-out technology or through RNA interference techniques. These are not ideal methods for emulating the actual effect of a drug – with gene knock-outs, there is much redundancy and subtlety in biological pathways and the removal of a gene can often be compensated in other ways as the organism differentiates and grows. An example here is the attempts to discover a function for the beta form of the estrogen receptor. Once the gene had been identified, there were intense efforts to ascribe a function to the gene, with considerable investment in producing and characterising knock-out animals. There were hints, but in the end, it took the development of isoform-specific compounds to provide chemical tools which could probe the biology and identify which diseases or conditions were associated with the receptor (again, see the chapter from Manas et al. in this book).

The best case for a target is to have a compound available that can provide the biological proof of concept. This is a compound that is sufficiently specific for the target of interest that can be studied either in cellular assays or in animal models of disease, to demonstrate that modulating a particular target will have the desired therapeutic benefit, in vivo. Such compounds could come from natural products, as in the case of antibiotics that validate the ribosome as a target and the gel-danamycin derivatives that are demonstrating the potential of Hsp90 as an oncology target.

In addition to biological validation, targets also need to be considered for what is termed, druggability. That is, does the protein have a binding site which can accommodate a drug-like compound with sufficient affinity and specificity? Although some experimental methods may be used to assess these, analyses of experiences with many targets have generated some general principles discussed in the chapter by Hann et al. later in this book. In summary, enzyme active sites tend to be highly druggable consisting of a distinct cleft designed to bind small substrates and with defined shape and directional chemistry. In contrast, most protein–protein interactions are less druggable as they cover quite large areas of protein surface with few shape or chemical features that a small molecule could bind to selectively. Unless particular "hot-spots" of activity can be identified, they are generally regarded as unsuitable drug targets (see Arkin and Wells, 2004 for a discussion).

Finally, for a structure-based project, there is a clear structural gate – that is, the structure of an appropriate form of the target needs to be available. Sometimes (for example, in a small structure-based company) this is set as a strict gate – that is, unless the structure is available hit identification cannot begin. There can be additional constraints. For example, if the project is relying on fragment screening using crystallography followed by soaking with compound mixtures, then the protein has to crystallise in a suitable crystal form with an open binding site.

2.2 Hit Identification

A hit is a compound that binds to the target and has the desired effect. The conventional method for identifying hits is by screening a compound collection which could consist of natural products or substrate mimetics, legacy compounds in a company's collection, compounds synthesised as potential hits against a particular class of target (focused library) or commercially available compounds. The majority of large pharmaceutical companies have invested considerably in automating this initial phase of hit identification, both in the generation of suitable target libraries and in the initial assay. This High Throughput Screening (HTS) approach places considerable constraints on the robustness of the assay and the availability and properties of the available compound collection (see Davis et al. for an up-to-date discussion of the issues).

HTS is also very expensive, consuming large quantities of target and compounds and requiring significant investment in robotic screening devices. Smaller companies that rely on screening usually work with smaller libraries of compounds, and depend on a particular "edge" over the larger companies. That distinctiveness could be either in some detailed knowledge or expertise with the biology of the target class and thus more appropriate configuring of the assay, or through a small library of compounds for that particular class of target. It is in the hit-identification phase that structure-based methods have provided smaller companies an opportunity to establish rapidly effective drug-discovery projects, particularly through the use of virtual screening or fragment-based methods (see later).

In most cases, the hit-identification phase relies on configuring a particular assay to monitor binding or inhibition. Usually, a large number of compounds are being screened, so the first experiment is to measure compounds that exhibit activity (above a certain percentage inhibition) at a set concentration. This is usually followed by confirming the hits, that is where an in vitro assay is run at varying concentrations to determine the IC50* or the Ki or the Kd[dagger] for the compound and the quality of the compound sample checked. Maintaining quality in a compound collection is a major challenge – compounds decompose over time, particularly if held dilute in solution in air. In addition, it is not unusual for 5-10% of compounds purchased from commercial suppliers to either be not what they claim to be, or to contain major contaminants that can give false positive (or false negative) results.

An HTS campaign can require significant resources (compound, target, manpower) and last 6–12 months, depending on how long it takes to configure a robust assay. Where smaller collections of compounds are being used, or structure-based methods applied, the hit-identification phase usually lasts around 6 months and requires a relatively small team of scientists.

The output from a hit-identification campaign is a set of compounds whose chemical structures have been checked and which have reproducibly been shown to have activity.

2.3 Hits to Leads

The hits to leads (H2L) phase is where some of the crucial decisions are made in a project – establishing which chemical series has the potential to be optimised into a drug candidate. This is an important decision as lead optimisation (the next phase) is when significant resources and effort are spent in optimising the properties of compounds. For these reasons, most companies set quite stringent criteria for entering lead optimisation, set for each target and reflecting the projected requirement of the properties of the final drug candidate, often called target-product profile.

The detailed work during the H2L phase varies with the nature of the project and, in particular, the origin of the hit compounds. Wherever the compounds come from, it is usual to re-synthesise the compounds for complete validation of the hit and to either purchase or synthesise close analogues of the compounds. In general, it is during the H2L phase that dramatic changes in chemical template are made and the essential core of the lead series established. The usual aims are to establish preliminary structure-activity relationships (SAR) within one or more series, to explore the indicative physicochemical and ADMET[double dagger] properties of the compounds, to consider the chemical tractability or synthetic accessibility of the compounds and to understand the IP position on the compound series and target. Depending on the project (and the company policy), entry into lead optimisation can be gated by demonstrating some in vivo activity in the series. Setting the right barriers for entry into lead optimisation is one of the most challenging aspects of medicinal chemistry.

This phase usually takes around 6 months, depending on the requirements for biological testing and the degree of synthesis required to establish a lead series with appropriate properties.

2.4 Lead Optimisation

This is the most resource-intensive component in drug discovery, requiring considerable input from synthetic chemistry, modelling, disease biology and assay design. It is not unusual for a lead optimisation (LO) team to consist of over 15 scientists, particularly if more than one lead series of compounds are being progressed. The main challenge is to develop one or more compounds with the desired drug-like properties. As well as having sufficient affinity for the target (nM[section] is the usual goal), the compound needs to have an appropriate selectivity profile, be able to get to the site of action (which for many targets means cell permeability) and have acceptable drug-like properties. In addition, it is important to continue to track that the observed effect on cellular (and later in vivo activity) is from interaction with the identified protein target. Although, in the end, the most important feature is that the compound works in the cell, pharmacodynamic markers are important to check if the compound is affecting the biology through the predicted target,[paragraph] particularly when an understanding of the structure of that target is being used to guide optimisation.