1
The Role of Fragment-based Discovery in Lead Finding

Roderick E. Hubbard

1.1 Introduction

Fragment-based lead discovery (FBLD) is now firmly established as a mature collection of methods and approaches for the discovery of small molecules that bind to protein or nucleic acid targets. The approach is being successfully applied in the search for new drugs, with many compounds now in clinical trials (see summary in [1]) and with the first fragment-derived compound now treating patients [2]. The approach has also had a number of other impacts such as providing starting points for lead discovery for challenging, unconventional targets such as protein-protein interactions [3-5], increasing the use of biophysics to characterize compound binding and properties, and providing small groups, particularly in academia, with access to the tools to identify chemical probes of biological systems [6,7].

The other chapters in this book will discuss the details and new advances in the methods and provide examples of how fragments have been used in specific projects. In this chapter, I will draw on my own experiences and view of the literature to discuss three main areas. First, I will review current practice in FBLD, highlighting how and when fragments have an impact on the drug discovery process. Second, I will then review how the ideas have developed, with particular emphasis on the past 10 years. I will discuss how fragment methods and thinking have been extended and refined and how these developments have affected the lead discovery process in drug discovery. Finally, I will discuss some of the areas where we can see that improvements in fragment methods could have further impact on discovery.

The discussion will focus on fragment-based discovery against protein targets. Although there are a few examples of fragments being used against RNA [8-10] and DNA [11] targets, the majority of reported campaigns are against proteins. Two types of protein target will be considered. The first shall be called conventional targets. These are proteins such as kinases where although it is never straightforward to achieve the required selectivity and balance of physicochemical properties in the compound, the proteins usually behave in most of the experiments and assays. Crystal structures are usually readily obtained, large amounts of pure, homogeneous, and functional proteins can be generated for biophysical studies, and the activity assays are robust and well understood. The second class of target shall be called unconventional targets. There are two types here - the first are protein-protein interaction targets such as the proapoptotic Bcl-2 family or Ras, where experience over the years has eventually established reasonably robust assays and although crystal structures take some time to determine and the protein does not always behave in biophysical assays, it is possible to establish structure-based discovery. The main challenge here is the nature of the binding sites, with often large, hydrophobic, and sometimes flexible sites. The second type of unconventional targets are the results of recent advances in our understanding of mammalian disease biology and consist of new classes of enzymes (such as the ubiquitin processing machinery [12]), disrupting multiprotein complexes, and proteins that are intrinsically disordered in some way (such as the one described in [13]). Here, the primary challenges are often in producing sufficient, homogeneous, functional protein for study, knowing what the post-translational modification state or even which complex is the true target and establishing robust assays to report on activity or binding. This last issue is often not appreciated - it can take a long time to establish the assays on new classes of target, not only because there is intrinsic variability in the behavior of the system but also because there is often not a tool compound available with which to validate the assay.

1.2 What is FBLD?

There are two distinctive features of fragment-based discovery compared to other approaches to lead finding. The first is that the discovery process begins with screening a small (usually 1-2000 member) library of low molecular weight (typically less than 20 heavy atom) compounds for binding to a particular site on the target. Key is the molecular weight of the fragments - they are big enough to probe interactions in the protein but small enough to minimize chances of unfavorable interactions. The second distinctive feature lies in the approach to optimizing these hits to lead compounds, either through careful, usually structure-guided, growth of the fragment or through merging information from fragments and elsewhere to generate optimized hits.

In many ways, fragments can be viewed as a state of mind - an approach to use the fragments as chemical tools to dissect what the requirements are for the chemical matter that affects a particular target in the desired way (affinity, selectivity) and using a combination of rational, usually structure-guided, and often biophysics-based methods for generation of the optimized compounds. We can define a fragment approach as one of intent - and that intent affects the strategy, methods, and thinking that is applied during the early parts of a discovery project. Detection and characterization of such weakly binding compounds can be problematic for some classes of target, with concerns over false positive and false negative hits, changes in binding mode, and so on. So, fragment methods engender a questioning, problem-solving approach to research. This is carried through into the usually structure-guided evolution of the initial fragment hits, which allows careful assembly of compounds that bind with high efficiency combined with suitable compound properties.

1.3 FBLD: Current Practice

Figure 1.1 and its legend summarize the contemporary approach to fragment-based discovery followed by most practitioners. There are five main components to a fragment platform: a fragment library, a method for finding which fragments bind, characterizing how the fragments bind by determining structure and biophysical measurements, exploring fragment SAR to identify the best fragment(s) to progress, and using the fragment(s) to generate lead compounds. Figure 1.1 also emphasizes how information about binding motifs is combined with information from HTS hits, literature compounds, or virtual screening hits. Other chapters in this book will provide detail on each of these different areas. In this chapter, I am focusing on the impact fragments have had on the lead discovery process. This is best done with some examples.

Figure 1.1 the FBLD process. There are five main components to a fragment platform. (a) Fragment library: there is an extensive literature on the design of fragment libraries [26,31,32,41]. The choice of compounds is constrained both by the demands of the screening methods (solubility, detection) and by the need to evolve the compounds (elaboration vectors, synthetic tractability) as well as avoiding reactive or toxic substructures. Key is the number of heavy atoms in the compounds. Analyses by Reymond [38,39] suggest that the number of possible lead-like compounds (chemical space) increases by around eightfold for each heavy atom. There are many approximations but this means that a fragment library of 1000 compounds of average MW 190 is equivalent to 108 compounds of MW 280 and 1018 compounds of MW 450. (b) Fragment screening: Table 1.1 summarizes the experiences at Vernalis over the years that are variously described elsewhere [15]. For all techniques, the main limitations are whether the protein target can be prepared in a suitable format for screening and whether the fragments are sufficiently soluble. The most robust method of screening is ligand-observed NMR, which has the dynamic range (typically from 5 mM to 100 nM) seen for fragment binding and particularly important for unconventional targets, as the integrity of the ligand and protein is checked at each experiment. (c) Characterizing fragment binding: for conventional targets, it is often possible to rapidly determine crystal structures of the fragment binding to the protein and, if the biochemical or binding assay is not suitable, use a biophysical method to validate and if possible quantify potency. For unconventional targets, this step is particularly important as the targets can have challenging binding sites, where conformational flexibility or large hydrophobic surfaces can challenge reliable detection of fragment binding. NMR methods can be used for unconventional targets, ranging from binding site localization (HSQC) to NMR-guided models (measuring NOE distances from ligand atoms to protein residues) and full structure determination. These are constrained by the size of the protein and requirement for isotope labeling. (d) Fragment SAR and optimization: there are two well-established methods - (1) SAR by catalog where features of the fragment are used to identify commercially available compounds for purchase and assay and (2) detailed design of bespoke compounds to optimize the fragment itself and explore potential vectors for elaboration. More recently, there have been new methods such as off-rate screening [16] that allow rapid profiling of compounds where substituents have been added to particular positions on the fragment, prospecting for suitable vectors for fragment evolution. This can be particularly important when limited structural information is available. (e) Fragment to candidate: medicinal chemistry optimization, supported where possible by rapid crystal...