* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Structural biology and drug discovery for protein–protein
Survey
Document related concepts
NK1 receptor antagonist wikipedia , lookup
Discovery and development of direct Xa inhibitors wikipedia , lookup
Discovery and development of integrase inhibitors wikipedia , lookup
Magnesium transporter wikipedia , lookup
Neuropsychopharmacology wikipedia , lookup
Discovery and development of proton pump inhibitors wikipedia , lookup
Drug interaction wikipedia , lookup
DNA-encoded chemical library wikipedia , lookup
Discovery and development of antiandrogens wikipedia , lookup
Neuropharmacology wikipedia , lookup
Transcript
TIPS-957; No. of Pages 8 Opinion Structural biology and drug discovery for protein–protein interactions Harry Jubb, Alicia P. Higueruelo, Anja Winter and Tom L. Blundell Department of Biochemistry, University of Cambridge, Cambridge CB1 2GA, UK Although targeting protein–protein interfaces of regulatory multiprotein complexes has become a significant focus in drug discovery, it continues to pose major challenges. Most interfaces would be classed as ‘undruggable’ by conventional analyses, as they tend to be large, flat and featureless. Over the past decade, encouragement has come from the discovery of hotspots that contribute much of the free energy of interaction, and this has led to the development of tethering methods that target small molecules to these sites, often inducing adaptive changes. Equally important has been the recognition that many protein–protein interactions involve a continuous epitope of one partner and a well-defined groove or series of specific small pockets. These observations have stimulated the development of stapled ahelical peptides and other proteomimetic approaches. They have also led to the realisation that fragments might gain low-affinity ‘footholds’ on some protein– protein interfaces, and that these fragments might be elaborated to useful modulators of the interactions. Outwith the druggable genome The concept of the ‘druggable genome’ [1] has concentrated thinking in target-based drug discovery in two very different ways. On the one hand, it has raised awareness that current methods and libraries work well only with restricted groups of targets. These are often enzymes, ion channels or receptors, with well-defined and preformed concave binding sites. This has encouraged a focus on targeting large protein superfamilies involved in cell regulation that possess these ‘druggable’ characteristics, for example Gprotein-coupled receptors (GPCRs), protein kinases and proteases. Furthermore, undue focus on ‘druggability’ has grave consequences for the pharmaceutical industry, because the temptation to pursue targets amenable to modulation may lead to unsuccessful attempts to design drugs to targets which are of poor quality with respect to disease rationale [2]. On the other hand, the realisation that high affinity may have been bought at the cost of low selectivity has stimulated debate on whether alternative strategies of lead discovery might allow a re-evaluation of the druggability concept. Targeting protein–protein interfaces (PPIs) of multiprotein complexes that mediate cell regulation – long regarded by many as undruggable – has become a subject of intense activity in both academia and industry [3]. Indeed, PPI Corresponding author: Blundell, T.L. ([email protected]). targeting may offer new opportunities for drug discovery in an industry where large companies output on average less than one approved novel molecular entity per year [2]. Although structural biology has shown that PPIs are generally large (1500–3000 Å2), flat and relatively featureless [4], making the design of small molecule antagonists a difficult task, there is hope that these difficulties can be overcome. Not only has it been recognised that energetic hotspots contribute much of interface interaction free energy in many PPIs [5] but also there is increased appreciation that many interactions involve the concerted folding and binding of a continuous epitope into a defined groove or series of small specific pockets [6]. Such observations have stimulated the development of stapled a-helical peptides and other proteomimetic approaches to drugging interfaces [7]. They have also led to the realisation that fragments might gain low-affinity ‘footholds’ on PPIs, and that these might be elaborated to useful modulators of multiprotein assemblies where knowledge of the structures of the complexes is available. Here, we discuss the nature of PPIs with respect to druggability. Using examples from our laboratory, we review fragment-based PPI drug discovery and assess the opportunities this approach presents. We argue that structural biology is and will continue to be central to successful for targeting protein–protein interactions. We are of the opinion that these recent successes, coupled with further understanding of the nature of PPIs and of small molecule binding that disrupts such interactions, will provide a path to the development of novel therapeutics which drug the ‘undruggable’. Understanding the nature and classes of PPIs for drug discovery Protein–protein interactions may be characterised as obligate, in which the subunits are not observed on their own in vivo, or as non-obligate, in which they are. This is a function of both the binding energy of the interaction (including cooperative, competitive and allosteric effects) and coexpression/localisation of the interacting partners [8]. Coexpression is a requisite for obligate pairing, and thus antibody–antigen, enzyme–inhibitor and receptor– protein ligand interactions are never obligate. Nooren and Thornton distinguished obligation of interactions from complex lifetime [8]. Permanent complexes maintain stable interactions, remaining in complex for a long time once bound; thus, all obligate interactions are permanent, as are a subset of nonobligate interactions such as antibody:antigen binding. Transient interactions may be in dynamic 0165-6147/$ – see front matter ß 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.tips.2012.03.006 Trends in Pharmacological Sciences xx (2012) 1–8 1 TIPS-957; No. of Pages 8 Opinion Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x equilibrium, or in stronger associations dependent on modulation by other molecules such as cofactors. We are of the opinion that understanding where a PPI lies on the binding energy, modulation and localisation continuum, and how the structure and nature of interactions vary between these classes will be important for identifying multiprotein complexes more amenable to drug targeting. Another descriptor of PPIs relevant to drug discovery is whether a continuous region of polypeptide is involved in interface pairing. Continuous epitopes often comprise a single secondary structure element, as opposed to discontinuous epitopes in which interacting regions of polypeptide originate from different regions of the primary structure [6]. Many continuous epitopes involve flexible or disordered (regions of) polypeptide chains that assemble with a classical globular protein to give a globular complex; so-called concerted folding and binding, developed beautifully by Wright and Dyson [9]. Globular partners of concerted folding-and-binding interactions tend to have better defined, preformed binding sites as grooves to accommodate helices (e.g. Bcl [10]), or small pockets that bind side chains but can be exploited to bind fragments (e.g. RAD51 interactions with BRCA2 [11,12], involved in homologous recombination). These successes, discussed below, give promise for drug discovery against PPI targets that exhibit disorder–order transitions on binding in vivo. The concept of PPI ‘hotspots’ first suggested by Clackson and Wells [11] and supported by alanine-scanning mutagenesis studies [5] has proved helpful. Hotspots often comprise side chains of tyrosine, tryptophan and arginine, which all allow adaptive conformational change to accommodate small molecules as well as strong hydrogen-bonding potential through nitrogen and oxygen atoms and other interaction potential through aromatic or guanidinium moieties [3]. The PPI hotspot concept is useful in drug discovery for identifying regions on a PPI protomer to which a ligand may bind with high efficiency. Combinations of energetic, structural and evolutionary information, for example using Bayesian Networks based on training sets from experimental data [12] and support vector machine (SVM) classifiers [13] may assist hotspot identification. There is some evidence that protein–ligand binding hotspots in PPIs correlate with protein–protein RAD51–BRC4 interaction (a) hotspots, and this has been an active area of investigation. Most algorithms measuring the ability of a site to bind a ligand – better called ‘ligandability’ [14] than druggability – depend on knowledge of the structures of the interfaces, and databases that define interatomic interactions based on structures are helpful in this endeavour [15]. Small grooves and pockets in the interface region that might bind a small molecule can be identified on the basis of their shape and sequence conservation [16]. Although such methods are successful for classical targets, they are retrospective and are less useful for PPIs. This is partly because pockets at PPI interfaces tend to be unlike the classical large cavities of druggable proteins [17] and partly because PPI interface cavities which accommodate molecules are induced or adaptive, depending on flexibility of side chains and to a very much lesser extent small loops [3]. Alternative approaches geared toward PPIs include ANCHOR [18], which identifies preformed recognition motifs, and Dr PIAS (Druggable Protein–protein Interaction Assessment System) [19], which adds information on pockets at the interface. In a recent computational study, Kozakov et al. investigated the relationship between ‘hotspots’ and druggability in six well-studied PPI targets [20]. They found that by mapping a target protein using different types of probe molecules, consensus sites could be found where the majority of test molecules bound. The fact that these sites were indeed known to bind PPI inhibitors lead to the conclusion that these ‘hotspots’ relate to the druggability of this site. Moreover, this algorithm would be able to use structural data alone to find probable druggable sites in a new target. These computational methods are best complemented by experimental methods that measure the hit rate from an X-ray crystallographic fragment screen or, as originally proposed, from an NMR-based screen [21]. Recent experimental approaches include development of a traffic light ligandability score based on hit rate, affinity and hit diversity [14]. In the RAD51/BRC4 interface, defined in our laboratory [22] (Figure 1), computational methods identify the pocket that binds phenylalanine of FxxA, but only ANCHOR predicts the small alanine pocket as druggable as well. Experimental fragment screening gives (b) BRC repeats TRENDS in Pharmacological Sciences Figure 1. X-Ray diffraction studies targeting PPI involving human RAD51 interactions with BRCA2. (a) The structure of RAD51 in complex with the BRC4 region of BRCA2, demonstrating the existence of two well-defined pockets on RAD51 that are occupied by side chains of the conserved FxxA motif. RAD51 is shown as grey van der Waals surface. BRC4 is shown in purple cartoon form. Residues within a 4 Å radius of the FxxA motif are highlighted in red on the RAD51 surface. The interface on the RAD51 is fairly flat, but there are distinct pockets for the phenylalanine and alanine side chains that bind the FxxA motif. (b) Sequences of homologous repeats in human BRCA2 demonstrating the conservation of the FxxA motif. (Figure is reproduced with permission from D.E. Scott, M. Marsh and M. Hyvonen.) 2 TIPS-957; No. of Pages 8 Opinion Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x (a) (b) (c) (d) (e) (f) TRENDS in Pharmacological Sciences Figure 2. Mode of interaction of fragments binding to the phenylalanine-binding pocket of RAD51 in complex with the BRC4 peptide of BRCA2 as observed using X-ray crystallography. (a) Indazole 5, (b) 4-methylester indole 4, (c) naphth-1-ol 9, (d) L-methylester tryptophan 8, (e) 2-aminobenzothiazole 6, (f) naphth-2-ol 10. The fragments exploit both the lipophilic nature of the pocket as well as opportunities to form well-defined H-bonds, particularly with the main chain carbonyl of Leu 214 and the nearby side chain of Gln 217. The affinities are in the range 400 to 1000 mM. This shows that small side chain pockets in PPIs may act as hotspots that undergo little conformational rearrangement on ligand binding. (Figure is reproduced with permission from D.E. Scott, M. Marsh and M. Hyvonen.) a predominance of hits in the phenylalanine pocket of RAD51 (Figure 2). Although debate continues as to whether PPI hotspots are always indicative of ligand-efficient ‘footholds’ for small molecule binding [3], it is clear that competitive small molecule inhibitors can often be developed at these sites. Screening and design of PPI inhibitors Several studies have focused on small molecules that disturb PPIs, comparing them with classical drugs and popular screening libraries [3,23,24]. These studies show that PPI inhibitors tend to be larger and more lipophilic than molecules that bind classically druggable sites [25]. Furthermore, they are poorly represented in commonly used chemical-screening libraries. It seems clear that drug discovery is in an uncharted area with these challenging targets, and that new methods will be required. When the structure of the target or a close analogue is known, classical computational approaches have been used with success to find small molecules binding to PPIs. These approaches include pharmacophore searches [26] and docking of standard catalogues such as NCI [27] and ACD [28], in addition to less conventional molecules such as traditional herbal medicines [29]. Proteomimetics A successful experimental approach for PPI inhibition is to mimic elements of the binding site surface structure of proteins, antibodies or peptides which bind protein targets; particularly, when the binding site is a secondary structure element such as a helix or b-turn. Yin and Hamilton have used a range of chemistries, including porphyrins and a-helical mimetics to achieve this objective [30]. An alternative approach is to use stapled peptides that target a-helical peptide binding sites. These are stabilised by crosslinking adjacent residues in the helix to make them resistant to proteolysis. Successful examples have included the p53:MDM2/X [31] interaction, and the MAML-1:Notch [32] interaction, both of which involve helices at their interaction interfaces. Small molecule fragment screening To explore the chemical space defined by PPIs, a more efficient approach is required, one that is not biased by earlier drug discovery campaigns. One approach is to decrease the size of the screening molecule. This in principle offers the possibility of exploring chemical space more efficiently, using chemical libraries comprising thousands rather than hundreds of thousands or millions of compounds [26,27]. Because initial hits have low affinity, they will not normally disrupt PPIs, unless they are tethered. Tethering, pioneered by Wells and coworkers [33], exploits a thiol-containing fragment linked through a reversible disulfide bond with a cysteine residue in a protein and identified using mass spectrometry [34]. This often requires engineering in cysteine residues local to the target site and has been developed by using ‘extenders’ that irreversibly alkylate the cysteine but provide a thiol group to capture the fragments [35]. There are several examples of successes where tethering has been used to antagonise PPIs [36–38]. 3 TIPS-957; No. of Pages 8 Opinion Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x An alternative approach to tethering is to stabilise the uncomplexed protein components of the multiprotein system in solution and to employ conventional fragmentbased screening [27,31–33]. The fragment-based approach uses a library of small molecules (fragments) with molecular weights of 100–300 Da to explore chemical space of the target protein-binding site. In our hands the Maybridge library has proved useful with PPIs, giving multiple hits on a range of different multiprotein complexes. As the affinity is low, PPI fragment screening requires sensitive biophysical methods – nuclear magnetic resonance (NMR), X-ray crystallography, surface plasmon resonance (SPR), differential scanning fluorimetry (DSF) or isothermal calorimetry (ITC) – to screen and validate fragment binding. Fragment hits are elaborated by crosslinking or ‘growing’, while maintaining strong interactions for each group added. This approach has the advantage of being able to explore new binding sites, such as those found in PPIs [39]. Problems with solubility, stability, polydispersity and binding site occlusion by crystal contacts for biophysical screening have been avoided by engineering the protein target to produce a stable monomer. An excellent example of this has been the engineering of human RAD51/RadA by Hyvönen and coworkers [39]. Although human RAD51 cannot be expressed in a stable, monomeric and unliganded form, the archaeal homologue of RAD51 (RadA) Target protein can be used as a surrogate, and engineered to have humanised interaction surfaces giving stable monomeric, unliganded species. New crystallographic forms with open binding sites for screening have been generated by targeting packing interaction sites for mutagenesis [39]. Biophysical and structural methods for screening of fragments Fragment-based screening is usually carried out in two steps (Figure 3). In our experience, it is best to have a battery of techniques available for initial high throughput screening (HTS), because the choice of technique is hugely dependent on the physical, chemical and structural properties of the target protein. Given that monomerised subunits of multiprotein assemblies are often difficult to prepare, the first assays should be the simplest and have the lowest demands on quantities of the protein. Fluorescence-based thermal shift assay [40–42] and SPR [36,37] are both suited to PPIs, and Figure 4 exemplifies them in the study of the Met receptor interaction with a protein ligand hepatocyte growth factor/scatter factor (HGF/SF). However, protein-observed NMR spectroscopy [43], the first experimental method used for screening fragments, has also been used successfully for targeting protein–protein interactions of Bcl-XL with an a-helical peptides derived from pro-apoptotic molecules BAK and BAD Hit validation SPR Fragment screening Thermal shift ITC Cheminformatics Substructure search Similarity search Bioassays in vivo and in vitro X-ray Compound concentration NMR SPR X-ray Chemical optimisation Fragment growing Fragment library Molecular weight < 300 Da cLogP ≤3 Hydrogen bond donors ≤3 Hydrogen bond acceptors ≤3 Number of rotatable bonds ≤3 NMR Fragment linking TRENDS in Pharmacological Sciences Figure 3. Strategy for a fragment-based screening campaign. Methods for initial high throughput fragment screening include thermal shift (DSF), SPR, ligand-based NMR and X-ray crystallography. Validation of potential fragment hits can be achieved by nanospray mass spectrometry, the kinetic parameters such as association and dissociation rates defined using SPR, and the major thermodynamic parameters derived using ITC. X-Ray crystallography and NMR are used to define binding modes of the fragment and to guide fragment optimisation or purchase of related compounds (analogue search). Assessment of biological activity assists in the chemical optimisation of the fragments on their way to a drug lead. 4 TIPS-957; No. of Pages 8 Opinion Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x (b) (a) 1e+5 Fluorescence signal 2.5e+4 8e+4 2.0e+4 1.5e+4 6e+4 1.0e+4 4e+4 5.0e+3 2e+4 0.0 0 25 30 35 40 45 50 55 30 Temperature (degrees Celcius) 40 50 60 80 90 (b) (a) Fragment I 600 Relative SPR response units [%RU] 70 Temperature (degrees Celcius) 500 NH 400 NH 300 Fragment V 200 O 100 S S OH 0 -100 tI m ag Fr II I tI en tI en a Fr gm en en gm a Fr tV V tI en gm gm a Fr a Fr (b) 10 (a) Key: SPR response units [RU] 8 Key: Fragment 1 Fragment 2 Fragment 3 Fragment 4 6 4 2 0 K d < 100 μM K d < 500 μM K d > 500 μM no binding -2 0 1e–4 2e–4 3e–4 4e–4 5e–4 Concentration [M] TRENDS in Pharmacological Sciences Figure 4. Targeting the Met receptor interaction with NK1, a splice form of HGF/SF that is a partial agonist/antagonist. (Top panel) Thermal shift curves of Met and NK1. (a) The fragments interacting with Met give good unfolding curves suitable for screening, whereas (b) demonstrates that melting curves of NK1 obtained using different buffers at varying pH [30 mM phosphate, pH 7, 150 mM NaCl (——), 50 mM Bis Tris, pH 6.7 (), CAPS pH 10.5 ( )] have complex melting processes, possibly related to the presence of two domains and disulfide bridges, as well as a tendency to dimerise through domain swapping. One of the curves ( ) shows a fragment that interacts with the fluorescent dye. (Middle panel) Surface plasmon resonance (SPR) for fragment screening of binding the Met receptor, NK1 (N-terminal and Kringle1) and SPH (serine protease homology) domains of HGF/SF. (a) Relative binding levels of five fragments showing fragment hits (fragment I as a hit for Met), nonbinders (fragment II), overstoichiometric binders and promiscuous binders to Met (black), NK1 (light grey) and SPH domain (medium grey). Overstoichiometric binders or aggregating fragments were identified by extremely high responses (fragment III), and promiscuous binders show binding to all three proteins (for example fragments IV and V). (b) Structures of a fragment hit (fragment I) and a promiscuous binder (fragment V). (Bottom panel) Fragment hits for Met using SPR. (a) Distribution of steady-state binding constants (Kd) among fragment hits for Met. (b) Concentration-dependent binding of selected fragments to immobilised Met. Steady-state binding levels were fitted using a global fit resulting in Kd values from low micromolar to low millimolar range. (Figure is reproduced with permission from A. Winter, A. Sigurdardottir, E. Gherardi and T.L. Blundell.) 5 TIPS-957; No. of Pages 8 Opinion Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x [10]. NMR screening has been generally successful in fragment-based drug discovery [44]. For validation of the hits derived from the initial HTS, false negatives are minimised by repeating the assay or better by using a second independent approach. False positives are identified through careful analysis of the kinetics, affinity (Kd) and structures of protein–fragment complexes. In principle, rate constants for association and dissociation of protein–fragment complexes can be directly measured using SPR and used to estimate affinity [45–47]. However, most fragments show a squared response curve due to fast on- and off-rates with little or no curvature, and thus steady-state experiments are carried out using different concentrations of the compound and their respective equilibrium binding levels determined [48]. Nevertheless, we have used these methods to validate the binding of fragments targeting the PPI of the Met complex with HGF/ SF (Figure 4). ITC, which requires large quantities of protein, can be used to titrate the change of enthalpy upon binding, from which affinity, free energy and change of entropy can be calculated. Kd values obtained by ITC can differ from those of SPR but are probably more accurate, because the protein is immobilised in SPR experiments on a surface, possibly introducing restrictions in flexibility or induced conformational changes upon ligand binding. Both ITC and SPR can be used to perform competition experiments with a known ligand to confirm that compounds compete for binding at a specific site. Three-dimensional structures, which are best obtained by X-ray crystallography but can also be generated from NMR spectroscopy, give essential insights into compound-binding modes. Although protein-based NMR is not high throughput and usually does not give a detailed picture of ligand binding, it is very sensitive and valuable for screening weakly binding ligands that otherwise might be missed. X-Ray crystallography has the advantage of providing information on not only where but also how the ligand binds [49,50]. This information can then be used to monitor every iterative round of the optimisation process. The ability to use structural information dramatically increases the success of the fragment based drug design process [27,42]. An example of the successful use of fragment-based approaches in targeting protein– protein interactions is illustrated for the interface between the human recombinase RAD51 and the hub protein BRCA2 (Figure 2). In this case, the analysis has progressed by two routes: the exploration of short peptides based on the natural ligand and fragment screening. (a) (b) (c) (d) TRENDS in Pharmacological Sciences Figure 5. Progression of two different fragments bound simultaneously to Bcl-XL into a helix mimetic with nanomolar affinity, ABT737. The protein Bcl-XL is represented as grey surface. (a) Two different fragments in cyan cocrystallised with Bcl-XL (1YSG). (b) Molecule in green sticks (1YSI) is generated from merging and optimising previous fragments (from 1YSG), both are superposed using the protein chain as a guide and are represented by fine lines. (c) Optimised compound ABT737 in purple sticks is bound to Bcl-XL (2YXJ), the molecule from 1YSI is superposed using the protein and is represented by fine lines. (d) Compound ABT737 in purple sticks is bound to Bcl-XL (2YXJ), BAD peptide in orange is superposed using the protein and is represented by the orange helix. Dotted lines represent polar interactions between the protein and BAD in orange and between protein and small molecules in purple. Figures generated with the program PyMol (www.pymol.org). 6 TIPS-957; No. of Pages 8 Opinion Around 20 structures of peptide–protein complexes were studied by X-ray crystallography and these demonstrated subtle changes of conformation that allow peptide Hbonds to be optimised as the sequence changes. These structures along with structures of around 80 protein– fragment complexes (for examples, see Figure 2) have been used to design fragment–peptide chimaeric molecules, which served as templates for the design of other molecules. Growing and linking fragments With structural information in hand, fragment growing or fragment linking can be employed to identify larger compounds from one or more fragment starting points. Monitoring ligand efficiency [51] or group efficiency [52] allows estimation of an individual group contribution toward the free energy of binding. Recent efforts seem to have been characterised by the optimisation of affinity by increasing lipophilicity [53]. The ligand lipophilicity efficiency (LLE) index [54] and other ligand efficiency indices, including polarity of molecules [55], can be used to monitor and hopefully avoid this trend. In general, we have found fragment growing more straightforward, because only one starting point is required [48,49]. However, when more than one fragment is bound in the target site, structural information of one fragment can be used to guide the growing of another. Fragment linking is often limited by linkers that perturb the optimal binding geometries of the fragments, introduce excessive conformational flexibility or lose ligand efficiency by not making efficient interactions [56]. X-Ray crystal structures solved at each iteration of the growth process continue to be central to successful exploitation of this approach [32,33]. Biochemical and cell-based assays are high throughput and low cost. In some fragment-based screening campaigns, biological assays have proved remarkably successful even in the initial screening, but fragments are often too low affinity to evoke a response. However, assays become central to the campaign as fragments are developed. Abbott Laboratories have used these approaches to design molecules with affinities of up to 5 nM that mimic the key a-helix that binds Bcl-XL. Figure 5 shows the published progression of the design process [10], in which two different fragments were bound simultaneously to BclXL (Figure 5a) and were evolved/grown into a single molecule [57]. The smaller contact regions with the protein meant that the ligand efficiency was almost twofold higher than that of a small helix. Although this is a particularly challenging example with impressive success, the optimised molecules are highly lipophilic and do not mimic available specific polar contacts that BAD engages (Figure 5d). Growth of fragments targeting the PPI between the human recombinase RAD51 and the hub protein BRCA2 has advanced using the structure of the RAD51 in complex with the BRC4 region of BRCA2 as a guide (Figure 1) [22]. Concluding remarks and future prospects Although the number of preclinical or clinical stage candidate compounds derived from targeting PPIs is still small, Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x there continues to be extensive activity in the design of inhibitors in both academia and industry. In the past 4 years, improvements in proteomimetics and fragmentscreening technologies have been pushing the boundaries of what is druggable and what is not. We are beginning to gain useful knowledge by computational analysis of the increasing numbers of reports of compounds that modify protein–protein complexes. Much could be achieved by enriching fragment and other chemical libraries with chemical structures that have been successful in targeting PPIs as well as by providing a greater variety of three-dimensional shapes [58]. Improving specific contacts in the earlier phase of the discovery, instead of optimising purely for binding affinity, should decrease the tendency to increase lipophilicity and nonspecific hydrophobic interactions [59]. Success in targeting PPIs clearly depends on target type. We have commented on the characteristically different interfaces involving proteins that undergo concerted folding and binding at the ‘receptor’ protein [6]. These tend to have better defined binding sites, either as grooves to accommodate helices, for example Bcl, or small pockets that bind side chains but can be exploited to bind fragments, for example RAD51. Globular partners that bind flexible/disordered peptides tend to have relatively little induced conformational change on fragment or even larger ligand binding. They constitute many of the most promising targets for fragment-based approaches. By contrast, interactions between globular proteins tend to involve large, flat and featureless interfaces. The greatest success here has been with tethering methods that allow the fragments to induce pockets at hotspots [37]; noncovalent interactions stabilise the pocket conformation, if the fragment is covalently tethered to give an entropic advantage. Nevertheless, the evidence from the Notch receptor ankyrin domain indicates that untethered fragments can be observed to bind on such surfaces (N. Abdel-Rahman, PhD thesis, University of Cambridge, 2010). However, at this time little progress has been made to link or grow them. Acknowledgements We thank Marko Hyvonen, Chris Abell, Ashok Venktaraman, Grahame McKenzie, Anna Sigurdardottir, May Marsh, Duncan Scott, Antony Coyne, Ermanno Gherardi, Alessio Ciulli and Will Pitt for helpful discussions. T.L.B. thanks the Wellcome Trust for support through a Programme Grant and for a Seeding Drug Discovery Initiative Award. A.P.H. and H.J. thank UCB and the Biotechnology and Biological Sciences Research Council (BBSRC) for CASE Studentships, T.L.B. and A.W. thank the EU FP7 for support. References 1 Hopkins, A.L. and Groom, C.R. (2002) The druggable genome. Nat. Rev. Drug Discov. 1, 727–730 2 Bunnage, M.E. (2011) Getting pharmaceutical R&D back on target. Nat. Chem. Biol. 7, 335–339 3 Wells, J. and McClendon, C. (2007) Reaching for high-hanging fruit in drug discovery at protein–protein interfaces. Nature 450, 1001–1009 4 Blundell, T.L. et al. (2000) Protein–protein interactions in receptor activation and intracellular signalling. Biol. Chem. 381, 955–959 5 Bogan, A.A. and Thorn, K.S. (1998) Anatomy of hot spots in protein interfaces. J. Mol. Biol. 280, 1–9 6 Blundell, T.L. et al. (2006) Structural biology and bioinformatics in drug design: opportunities and challenges for target identification and lead discovery. Philos. Trans. R. Soc. Lond. B: Biol. Sci. 361, 413–423 7 TIPS-957; No. of Pages 8 Opinion 7 Fletcher, S. and Hamilton, A.D. (2005) Protein surface recognition and proteomimetics: mimics of protein surface structure and function. Curr. Opin. Chem. Biol. 9, 632–638 8 Nooren, I.M. and Thornton, J.M. (2003) Diversity of protein–protein interactions. EMBO J. 22, 3486–3492 9 Wright, P.E. and Dyson, H.J. (2009) Linking folding and binding. Curr. Opin. Structural Biol. 19, 31–38 10 Oltersdorf, T. et al. (2005) An inhibitor of Bcl-2 family proteins induces regression of solid tumours. Nature 435, 677–681 11 Clackson, T. and Wells, J.A. (1995) A hot spot of binding energy in a hormone-receptor interface. Science 267, 383–386 12 Segura, J. and Fernandez-Fuentes, N. (2011) PCRPi-DB: a database of computationally annotated hot spots in protein interfaces. Nucleic Acids Res. 39, D755–D760 13 Lise, S. et al. (2011) Predictions of hot spot residues at protein–protein interfaces using support vector machines. PLoS ONE 6, e16774 14 Edfeldt, F.N.B. et al. (2011) Fragment screening to predict druggability (ligandability) and lead discovery success. Drug Discov. Today 16, 284–287 15 Bickerton, G.R. et al. (2011) Comprehensive, atomic-level characterization of structurally characterized protein–protein interactions: the PICCOLO database. BMC Bioinform. 12, 313 16 Capra, J.A. et al. (2009) Predicting protein ligand binding sites by combining evolutionary sequence conservation and 3D structure. PLoS Comput. Biol. 5, e1000585 17 Fuller, J.C. et al. (2009) Predicting druggable binding sites at the protein–protein interface. Drug Discov. Today 14, 155–161 18 Meireles, L.M. et al. (2010) ANCHOR: a web server and database for analysis of protein–protein interaction binding pockets for drug discovery. Nucleic Acids Res. 38, W407–W411 19 Sugaya, N. and Furuya, T. (2011) Dr. PIAS: an integrative system for assessing the druggability of protein–protein interactions. BMC Bioinform. 12, 50 20 Kozakov, D. et al. (2011) Structural conservation of druggable hot spots in protein–protein interfaces. Proc. Natl. Acad. Sci. U.S.A. 108, 13528–13533 21 Hajduk, P.J. et al. (2005) Druggability indices for protein targets derived from NMR-based screening data. J. Med. Chem. 48, 2518–2525 22 Pellegrini, L. et al. (2002) Insights into DNA recombination from the structure of a RAD51–BRCA2 complex. Nature 420, 287–293 23 Pagliaro, L. et al. (2004) Emerging classes of protein–protein interaction inhibitors and new tools for their development. Curr. Opin. Chem. Biol. 8, 442–449 24 Fry, D.C. (2006) Protein–protein interactions as targets for small molecule drug discovery. Biopolymers 84, 535–552 25 Higueruelo, A.P. et al. (2009) Atomic interactions and profile of small molecules disrupting protein–protein interfaces: the TIMBAL database. Chem. Biol. Drug Des. 74, 457–467 26 Galatin, P.S. and Abraham, D.J. (2004) A nonpeptidic sulfonamide inhibits the p53-mdm2 interaction and activates p53-dependent transcription in mdm2-overexpressing cells. J. Med. Chem. 47, 4163–4165 27 Enyedy, I.J. et al. (2001) Discovery of small-molecule inhibitors of Bcl-2 through structure-based computer screening. J. Med. Chem. 44, 4313– 4324 28 Wang, J-L. et al. (2000) Structure-based discovery of an organic compound that binds Bcl-2 protein and induces apoptosis of tumor cells. Proc. Natl. Acad. Sci. U.S.A. 97, 7124–7129 29 Nikolovska-Coleska, Z. et al. (2004) Discovery of embelin as a cellpermeable, small-molecular weight inhibitor of XIAP through structure-based computational screening of a traditional herbal medicine three-dimensional structure database. J. Med. Chem. 47, 2430–2440 30 Yin, H. and Hamilton, A.D. (2005) Strategies for targeting protein– protein interactions with synthetic agents. Angew Chem. Int. Ed. Engl. 44, 4130–4163 31 Bernal, F. et al. (2010) A stapled p53 helix overcomes HDMX-mediated suppression of p53. Cancer Cell 18, 411–422 32 Moellering, R.E. et al. (2009) Direct inhibition of the NOTCH transcription factor complex. Nature 462, 182–188 8 Trends in Pharmacological Sciences xxx xxxx, Vol. xxx, No. x 33 Erlanson, D.A. et al. (2000) Site-directed ligand discovery. Proc. Natl. Acad. Sci. U.S.A. 97, 9367–9372 34 Erlanson, D. et al. (2004) TETHERING: fragment-based drug discovery. Annu. Rev. Biophys. Biomol. Struct. 33, 199–223 35 Erlanson, D. et al. (2003) In situ assembly of enzyme inhibitors using extended tethering. Nat. Biotechnol. 21, 308–314 36 Hyde, J. et al. (2003) Discovery and characterization of cooperative ligand binding in the adaptive region of interleukin-2. Biochemistry 42, 6475–6483 37 Raimundo, B.C. et al. (2004) Integrating fragment assembly and biophysical methods in the chemical advancement of small-molecule antagonists of IL-2: an approach for inhibiting protein–protein interactions. J. Med. Chem. 47, 3111–3130 38 Sadowsky, J.D. et al. (2011) Turning a protein kinase on or off from a single allosteric site via disulfide trapping. Proc. Natl. Acad. Sci. U.S.A. 108, 6056–6061 39 Valkov, E. et al. (2012) Targeting protein–protein interactions and fragment-based drug discovery. Top. Curr. Chem. 317, 145–179 40 Brandts, J.F. and Lin, L.N. (1990) Study of strong to ultratight protein interactions using differential scanning calorimetry. Biochemistry 29, 6927–6940 41 Holdgate, G.A. and Ward, W.H. (2005) Measurements of binding thermodynamics in drug discovery. Drug Discov. Today 10, 1543– 1550 42 Leung, S.M. et al. (1996) Thermal activation of the bovine Hsc70 molecular chaperone at physiological temperatures: physical evidence of a molecular thermometer. Cell Stress Chaperones 1, 78–89 43 Shuker, S. et al. (1996) Discovering high-affinity ligands for proteins: SAR by NMR. Science 274, 1531–1534 44 Campos-Olivas, R. (2011) NMR screening and hit validation in fragment based drug discovery. Curr. Top. Med. Chem. 11, 43–67 45 Cole, D.C. et al. (2010) Identification and characterization of acidic mammalian chitinase inhibitors. J. Med. Chem. 53, 6122– 6128 46 Geitmann, M. et al. (2011) Identification of a novel scaffold for allosteric inhibition of wild type and drug resistant HIV-1 reverse transcriptase by fragment library screening. J. Med. Chem. 54, 699–708 47 Nordin, H. et al. (2005) Kinetic studies of small molecule interactions with protein kinases using biosensor technology. Anal. Biochem. 340, 359–368 48 Giannetti, A.M. (2011) From experimental design to validated hits a comprehensive walk-through of fragment lead identification using surface plasmon resonance. Methods Enzymol. 493, 169–218 49 Blundell, T.L. et al. (2002) High-throughput crystallography for lead discovery in drug design. Nat. Rev. Drug Discov. 1, 45–54 50 Murray, C.W. and Blundell, T.L. (2010) Structural biology in fragmentbased drug design. Curr. Opin. Struct. Biol. 20, 497–507 51 Hopkins, A. et al. (2004) Ligand efficiency: a useful metric for lead selection. Drug Discov. Today 9, 430–431 52 Saxty, G. et al. (2007) Identification of inhibitors of protein kinase B using fragment-based lead discovery. J. Med. Chem. 50, 2293–2296 53 Hann, M.M. (2011) Molecular obesity, potency and other addictions in drug discovery. Med. Chem. Comm. 2, 349–355 54 Leeson, P. and Springthorpe, B. (2007) The influence of drug-like concepts on decision-making in medicinal chemistry. Nat. Rev. Drug Discov. 6, 881–890 55 Abad-Zapatero, C. et al. (2010) Ligand efficiency indices for an effective mapping of chemico-biological space: the concept of an atlas-like representation. Drug Discov. Today 15, 804–811 56 Chung, S. et al. (2009) Impact of linker strain and flexibility in the design of a fragment-based inhibitor. Nat. Chem. Biol. 5, 407–413 57 Petros, A.M. et al. (2006) Discovery of a potent inhibitor of the antiapoptotic protein Bcl-xL from NMR and parallel synthesis. J. Med. Chem. 49, 656–663 58 Hung, A.W. et al. (2011) Route to three-dimensional fragments using diversity-oriented synthesis. Proc. Natl. Acad. Sci. U.S.A. 108, 6799–6804 59 Keserü, G.M. and Makara, G.M. (2009) The influence of lead discovery strategies on the properties of drug candidates. Nat. Rev. Drug Discov. 8, 203–212