Fragment-based drug discovery is a rapidly evolving area of research, which has recently seen new applications in areas such as epigenetics, GPCRs and the identification of novel allosteric binding pockets. The first fragment-derived drug was recently approved for the treatment of melanoma. It is hoped that this approval is just the beginning of the many drugs yet to be discovered using this fascinating technique.
This book is written from a Chemist's perspective and comprehensively assesses the impact of fragment-based drug discovery on a wide variety of areas of medicinal chemistry. It will prove to be an invaluable resource for medicinal chemists working in academia and industry, as well as anyone interested in novel drug discovery techniques.
Read an Excerpt
Fragment-Based Drug Discovery
By Steven Howard, Chris Abell
The Royal Society of ChemistryCopyright © 2015 The Royal Society of Chemistry
All rights reserved.
Different Flavours of Fragments
CHRIS ABELL AND CLAUDIO DAGOSTIN
1.1Fragment-Based Drug Discovery
Fragment-based drug discovery is becoming a powerful technology in the arsenal of the pharmaceutical companies to aid discovery of new small-molecule therapeutics, establish the druggability of biological targets, and discover alternative inhibition sites on already established ones. With the market approval of Zelboraf® (Vemurafenib – the first drug discovered via fragment-based drug discovery) the approach has a measure of validation and interest in the field continues to grow.
Fragments are small molecules that may become parts of a larger molecule, but in some cases were fragments of known drugs, that have been used as starting points to find new inhibitors for different biological targets. A fragment is a small, typically aromatic, organic molecule of molecular weight <250 Da, which is very soluble and chemically stable. The use of such small molecules at the beginning of a medicinal chemistry program allows an effective exploration of chemical space and a more rational design of the final molecule through iterations guided by structural knowledge of the site of binding. In the best cases, the approach can offer a less expensive and faster route to potent and better quality lead compounds compared with more traditional approaches such as high-throughput screening (HTS) of higher molecular weight compounds.
1.2 Different Types of Fragments
Views on the composition of fragment libraries have evolved over time. Some libraries were assembled by taking known drugs and dissecting them into fragments or by computationally finding diverse fragments. It was apparent from the start that solubility would be very important because of the concentrations used for screening (often tens of mM). Another key criterion was lack of reactive groups although current interest in covalent inhibitors is beginning to impact on thinking here. Fragment libraries, although generally diverse, were mainly composed of flat aromatic or heteroaromatics. Astex introduced the Rule of 3 to map across to the Lipinski Rule of 5, and soon after, commercial Rule of 3 compliant libraries became available.
There remain different views on the best size for a fragment library. Some organisations have libraries of over 20 000 fragments whereas others use small libraries (for a recent study of an NMR library see ref. 9). We typically screen 800–1300 compounds (generally by thermal shift assay), from which we identify 20–100 fragments for further screening by ligand-based NMR. Structure–activity relationship (SAR) analysis is then conducted around the NMR hits by buying or making similar compounds. At Astex, similar sized libraries are screened by X-ray crystallography and NMR spectroscopy. In large pharmaceutical companies there has been a tendency to screen larger fragment libraries, e.g., by surface plasmon resonance (SPR); an advantage being that SARs will emerge from the initial screen. A consideration in deciding how many fragments to screen is the expected hit rate. This can be very high (20%+) in the case of a kinase screen or vanishing low when looking for compounds binding to a surface site to disrupt a protein–protein interaction.
Fragment libraries have been designed based on natural products e.g., "fragments of life" libraries including highly soluble metabolites, natural product derivatives and biaryl molecules designed to mimic the architectural motifs of proteins (α, β and γ turns). This library was tested against leukotriene A4 hydrolase by screening about 200 fragments using X-ray crystallography. Thirteen hits were obtained of which 11 bound to the active site, while 2 biaryls were found binding in a shallow pocket on the surface of the protein.
In another report, the Waldmann group reported a chemoinformatic analysis of more than 180 000 natural product structures that was used to obtain about 750 000 fragments of different MW and properties rich in sp3 centres. This virtual library was then filtered according to an HTS filter and Rule of 3 derived parameters and re-ordered in 2000 clusters. Part of the library (193 fragments) was tested on p38α MAP kinase using biochemical assays, leading to 12 hits being identified with potencies between 6 and 0.4 mM and ligand efficiencies (LEs) ranging from 0.19 to 0.28. Most of the hits predictably bound to the hinge region of the kinase ATP site, but interestingly there were some that bound to an allosteric site, as a novel class of type-III inhibitors. The library was also tested on a less tractable class of targets, the tyrosine phosphatases, including Mycobacterium tuberculosis protein tyrosine phosphatase A and B. The best hits had an IC50 of <100 µM and LEs of almost 0.50. However, despite the growing interest in 3D fragments there is no compelling evidence yet that this will result in more or better hits.
1.3 How We Identify Fragments
Given their size, fragments generally bind very weakly (100 µM to 10 mM) to biological targets. Consequently, there is a particular emphasis in the biophysical techniques to detect binding. These include NMR (ligand- and protein-based), X-ray crystallography, SPR, isothermal titration calorimetry (ITC), thermal shift, and native mass spectrometry. These different approaches mean that the start of a fragment-based drug discovery (FBDD) project can be quite different in different labs, but the end game is generally similar. The screening approach used is partly dictated by available facilities, and partly by organisational constraints, although in all the cases, a single technique is rarely used in isolation. Typically a combination of techniques is employed in order to minimise false positives and false negatives and thus provide reliable chemical starting points for subsequent design and synthesis.
Virtual screening using in silico docking methods has been used, but usually in support of subsequent biochemical and biophysical assays. The different binding modes of simple fragments may only exhibit small differences in free energy of binding compared to larger compounds, so the inaccuracy inherent in scoring functions does not allow a reliable ranking of fragment-binding modes.
In our lab, the first screen generally employs the Sypro Orange fluorescence-based thermal shift screen (Figure 1.1); where the selected hits are those giving the largest increase in thermal stability of the protein. This offers a quick way to identify 50–100 compounds to proceed on to NMR screening. However, the shifts are generally small and the screen should be repeated to improve the reliability of the hits. This is a very simple application of this screen, and it is clear that in more complex systems, e.g., when studying protein–protein interactions, good hits can give even significant negative shifts. A recent study on Checkpoint kinase 2 used the AlphaScreen kinase biochemical assay and a thermal shift screen. A good correlation was found for compounds which were hits in both these orthogonal screens. Such high-concentration biochemical screens can give a high frequency of false positives so it is important to do a counter-screen to check for interfering compounds.
In many companies, SPR has become the primary screening technique to identifying fragment hits. It has advantages of ease of setting up the assay, relatively high throughput, and requires low amounts of protein compared to other techniques (e.g., NMR and X-ray) used. The methodology is likely to become more popular as the cost of the instrumentation decreases. A later chapter by Giannetti is dedicated to this approach.
NMR techniques are among the most powerful and widely used to detect binding of fragments to target proteins. Ligand-based and protein-based methods each have advantages and limitations, and can complement each other. However it is the ligand-based methods that have found most general application. In our lab we use a combination of three such techniques (described in detail in Krimm's chapter): WaterLOGSY, STD and CPMG. WaterLOGSY detects binding through transfer of magnetisation from the water molecules to the ligand bound to the protein. STD uses a pulse to saturate signals from methyl groups in the protein. The disturbance in magnetisation is then transferred through an NOE to the bound ligand, with the resulting difference spectrum only showing a peak of the ligand when this binds to the protein target. Finally CPMG, is a relaxation-time-edited NMR experiment that exploits differences in transverse relaxation time (T2). Proteins (and bound ligands) have a small T2 while free ligands have a large T2. Thus monitoring T2, binding can be detected when the signal of the ligand decreases.
In a given screen, some fragments show binding by all three techniques (e.g.,Figure 1.2), some by only two, some by one. We progress preferentially those giving a positive outcome from all three experiments. An example of the power of fragment screening by NMR was in a study on Pin1. After a million-compound HTS failed to give any hits, an NMR screen of 1200 fragments using the three 1D ligand-based techniques identified 5 hits from 4 different series and yielded a lead compound of 260 nM IC50. Astex used a multistage NMR screen against HSP90 to select 125 compounds from a 1600 fragment library for X-ray crystallography. Fragments that bound to the protein were initially identified in a WaterLOGSY experiment. The presence of 200 µM ADP made it possible to identify those compounds that bound to the nucleotide site from their effect on the ADP signal. When Mg2+ was then added, this increased the affinity of HSP90 for ADP 20-fold, and led to displacement of those fragments.
There are more sophisticated NMR techniques that can give additional information. For example, we have used the inter-ligand Overhauser effect (ILOE). This can detect two fragments binding simultaneously to the protein, and in close proximity to each other, through the analysis of their negatives NOEs. This is particularly useful if the intention is to increase potency through linking or merging fragments binding in the same cavity.
Fragment screening using protein-based NMR is a powerful approach that has been used to great effect in some laboratories. To be most effective, it does require a pre-assignment of the protein NMR spectrum, involving multiply-labelled protein. It is generally applied to proteins of molecular weight below 30 kDa. An elegant example using this technique has been described recently where two scaffolds were identified to bind in different regions of the Mcl-1 protein through 2D 1H–15N-HMQC. Using a merging approach, Mcl-1 inhibitors were developed with an equilibrium dissociation constant (KD) < 100 nM.
Having confirmed fragment binding by NMR, we usually attempt to quantify the interaction using ITC. Specifically we are interested in the KD and stoichiometry, paying relatively little attention to the enthalpy and entropy of binding – which are generally difficult to explain (although that does not stop some authors attempting to do so). The approach does require a significant amount of protein, and someone skilled in the art. Even then, some proteins or chemical series are not amenable to ITC, often due to low heats of binding. From the KD we derive the LE. The LE continues to be a useful metric although increasingly surpassed by parameters that account for lipophilicity. If the fragment library has been assembled thoughtfully, there should also be analogues of the fragment that are commercially available that can be used to rapidly scope out SARs.
The use of X-ray crystallography for fragment screening, in spite of its early use e.g., by Astex has not been adopted generally. This was initially a surprise, as putting an X-ray fragment screen early in a screening cascade is arguably the most powerful way to drive the process. There may be many reasons for the reluctance to adopt this approach, some of which are probably as much to do with organisational issues, as they are to do with science. Other issues will be around the cost of X-ray equipment and associated robotics, and the specialised skills required to crystallise new proteins in a form suitable for fragment screening. In our academic group we try to incorporate X-ray crystallography as soon as we can, but in practice this is usually after we have hits validated by NMR spectroscopy.
For us, it is vital that we do get X-ray structures of our fragments bound to the target protein. This is the major go/no go gateway for project progression. We consider that structural information on the precise mode of fragment binding is needed in order to inform the subsequent chemical elaboration of the fragment. Several projects have been dropped at this point, because either we were not able to get suitable, soakable crystals of the target protein, or because even with these in hand, we could not identify fragments bound to these structures. This consideration highlights one of the key issues for success in this area within academic teams; the need for very strong collaborations between different disciplines. This includes biophysical scientists, structural biologists and chemists, with each project team having a major investment in the project. Getting the first structure of a fragment–protein complex is often the most difficult aspect of the project but is always a significant event for the team.
Ideally the chemist would like to have crystal structures of several fragments bound to the protein of interest, and generally also fragments bound at several sites (even if this can complicate the understanding of the KD values). Many early studies focused on protein kinases, which led to some level of predictability about the site where fragments would bind. At the other end of the spectrum, our studies on fragment binding to cytochrome P450s have identified many different binding sites ranging from the heme iron across to the protein surface (Figure 1.3). Where and how fragments bind has a major influence on the approach the chemists can take to the subsequent chemical elaboration. The early papers on FBDD against kinases mainly described fragment growing from a common fragment-binding site (usually the ATP binding site). The availability of structures of fragments bound at several sites in a protein gives options for fragment merging, linking or growing.
Fragment growing has been the major method of fragment elaboration described in the literature, perhaps because of the initial focus on protein kinases. A good fragment should offer immediate sites for elaboration. Initial elaboration should be incremental and not too ambitious, and should be guided by appropriate structural information and docking studies. Most fragments located by X-ray crystallography will have LEs of 0.3 or greater, and an aspirational aim should be to maintain this level of LE throughout the iterative synthetic cycles. This is challenging, and sometimes seen as too demanding, e.g., adding a phenyl ring requires an approximate 20-fold increase in potency in order to maintain an LE = 0.3. Such incremental elaboration is perhaps not helped if the synthetic chemistry is done through outsourcing when multiple analogues may be ordered at once.
In elaborating a fragment, the aim is to pick up new interactions e.g., π-stacking interactions, hydrogen bonds etc., as effectively as possible using any information that can be inferred from SARs around the fragment, while also avoiding obvious clashes with the protein surface. Fragment elaboration is an iterative process. Successive molecules are designed and synthesised, and their structure bound to the target protein is solved. Looking at a new co-crystal structure may do nothing more than confirm the predicted binding, but in some cases the binding may lead to changes in the protein conformation. In this case, new opportunities for structure-based design that were not obvious in the previous structures, may present themselves.
The iterative nature of fragment elaboration is potentially a strength of the process. It gives the chemist multiple opportunities to tailor the properties of the molecules being assembled. Fragments are typically small and highly water-soluble, and as such represent a good starting point in terms of physico-chemical properties. The temptation at this stage is to seek a quick increase in binding by adding mainly hydrophobic groups. This should be avoided as it leads to molecules with low solubility, less selectivity, and, ultimately, poor drug-like properties. The elaboration process does depend upon close co-operation between the chemist and structural biologist. Good cross-discipline relationships are essential to minimise cycle times and ensure momentum within the project. In our best cases, structural biology can obtain a ligand-bound protein structure within 24 hours of a ligand being synthesised. However in other cases, cycle time will be governed by the frequency of trips to the synchrotron. Additional delay can be imposed on this cycle time in industry when the synthetic elaboration is outsourced.
Excerpted from Fragment-Based Drug Discovery by Steven Howard, Chris Abell. Copyright © 2015 The Royal Society of Chemistry. Excerpted by permission of The Royal Society of Chemistry.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.