|Home | About | Journals | Submit | Contact Us | Français|
Creating first-in-class medications to treat human disease is an extremely challenging endeavor. While genome sequencing and genetics are making direct connections between mutations and human disorders at an unprecedented rate, matching molecular target(s) with a suitable therapeutic indication must ultimately be achieved by pharmacology. Here, we will discuss how the integration of chemical proteomic platforms, such as activity-based protein profiling, into the earliest stages of the drug discovery process has the potential to greatly expand the scope of proteins that can be pharmacologically evaluated in living systems, and, through doing so, promote the identification and prioritization of new therapeutic targets.
Paradoxes abound in the modern world of drug discovery. Genome sequences have provided a complete parts list describing all of the proteins in the human body, and high-throughput screening technologies offer platforms for exposing these proteins to millions of small-molecules. Yet, as has been well documented by others [1, 2], such informational and technical advances have not yielded a corresponding increase in new first-in-class medicines. While the reasons for this are complex and multifold, we will take the stance in this Perspective that at least part of the problem with drug discovery today is that, for the critical step of early-stage target characterization in both academia and industry, pharmacology has been largely displaced by molecular biology and genetics. This has created a methodological disconnect between the early (genetically driven) and late (pharmacologically driven) stages of the drug development process that, for the reasons outlined below, can impede and even prevent the progression of potentially interesting therapeutic targets. We will argue that recent advances in chemical proteomic (‘chemoproteomic’) methods should inspire a re-integration of pharmacology into the earliest stages of target characterization, such that it serves as a driver for, rather than responder to biological discovery. Establishing a renewed commitment to pharmacology that is guided by modern chemoproteomic technologies has the potential to create a much more efficient path to mine the proteome for new drug targets.
Decades ago, pharmacology and the chemical probes that it provides were integral to the process of biological discovery, often providing the first insights into new protein targets and biochemical pathways that affect mammalian physiology and disease [3–6]. Many of these pharmacologic tools represented valuable ‘proof-of-relevance’ probes, which we define for the purposes of this Perspective as meeting the minimalist definition of compounds that can block (or agonize) a protein of interest with good potency and selectivity in both cell and animal models. Some of these pharmacologically driven discoveries led to ground-breaking medicines that are still used to treat human disorders today . But, with the emergence of advanced molecular biology methods in the late 1980s through the 1990s, pharmacology somehow lost its sheen for early-stage target characterization, perhaps appearing as a somewhat cumbersome approach that lacked the technical simplicity and specificity of genetic methods. In the ensuing decades, molecular biology maintained and arguably even extended its dominion over pharmacology, which was relegated to a much later step in the target characterization pipeline that only initiated once substantial biological understanding of the protein target had been achieved (usually through years of molecular biology-driven research). This order of events brings us to the here and now, where we are entering the second decade of research since the first report of complete mouse and human genome sequences, and yet we still lack ‘proof-of-relevance’ small-molecule probes for the vast majority of mammalian proteins. Should this deficiency be attributed to the degree of difficulty in developing such probes or a lack of firm commitment to do so? We believe, perhaps not surprisingly, that the answer is a combination of both, but also that the second issue is more of a contributory factor than generally realized.
No doubt about it – pharmacology is more difficult than genetics. Each protein target, owing to its distinct structure and function, presents a special set of challenges for chemical probe development, and many have argued that only a small fraction of the human proteome is even, in principle, ‘druggable’ [7, 8]. In contrast, genetic methods have few, if any target boundaries. The technical ease with which genes can be selectively knocked down or out of cell and animal models is seductive, and rodent and human genetics can further provide some of the most convincing validation for target relevance to disease. However, genetic disruption of a protein, which often results in the loss of protein expression throughout life and is impractical for numerous protein classes (e.g. developmentally essential genes), may not mirror the effects of pharmacologically blocking its activity in a mature organism. Several elegant inducible and tissue-specific knock-out/overexpression genetic systems have been introduced in recent years to allow for more specific spatial and temporal control of gene expression; however, even these systems do not fully recapitulate the diverse ways that pharmacological probes can modulate protein function. Many drugs, for instance, produce their biological and medical effects, as well as avoid toxicity, by partial blockade (or activation) of a protein or by targeting proteins in a subset of tissues. Drugs can also affect multiple protein targets in vivo to produce therapeutic effects. Modeling such partial target modulation or polypharmacology by genetics is problematic.
Extrapolating from our knowledge of successful drugs and their targets and mechanisms of action, one could argue that pharmacology, no matter how challenging it may be, should be placed front and center in any serious attempt to mine the proteome for new drug targets. Ideally, one would like to generate a proof-of-relevance small-molecule probe for every protein in the mammalian proteome. The big question then becomes – how can we best pursue this ambitious goal, especially in today’s research environment, where the pharmaceutical industry, an historical juggernaut for developing first-in-class pharmacological probes, is rapidly moving away from early-stage target discovery and validation [1, 2]? As will be elaborated on below, we believe that this change presents a tremendous opportunity for the academic research community to create a new and more target-inclusive approach to mammalian pharmacology. Emerging chemoproteomic methods offer ways to develop proof-of-relevance probes for proteins that span the full spectrum of annotation to include those with established activities and proteins that lack functional annotation. Success could usher in a ‘back-to-the-future’ era of scientific research where pharmacology once again serves as a principal driving force for early-stage biological discoveries that, when coupled with insights into mechanism-of-action provided by chemoproteomics, propel our understanding of small-molecule effects on protein function in living systems. This knowledge can then be used to prioritize new targets and, perhaps more accurately, new drug-target pairs for clinical development.
One cannot overstate the importance of complete genome sequences for modern approaches to pharmacology. We now understand the full complement of proteins encoded by the human genome (splice variants and post-translationally modified proteins excepted), and many human proteins can be grouped into structurally and mechanistically related families based on sequence homology. These complete protein families provide a valuable starting point for asking an interesting set of pharmacological questions. Across how many protein families do drugged targets distribute? Within these druggable families, how many members have proof-of-relevance probes? For probes that target multiple members of a given protein family, is this polypharmacology reflected in simple sequence-relatedness among the shared protein targets? Are protein families that lack drugged members more difficult to target with chemical probes or do they simply represent portions of the proteome that have not yet been experimentally investigated? Chemoproteomics is well-suited to address some, if not all of these important questions.
Several reviews and perspectives have discussed the topic of the ‘druggability’ of the human proteome [7, 8], often making the point that the typical protein families considered most amenable to small-molecule pharmacology, such as enzymes, channels, and receptors, represent only a modest fraction of all human proteins. What is rarely mentioned in these perspectives, however, is what fraction of the ‘druggable’ proteome has already been drugged? This is an important issue considering that the sum of all enzymes, channels, and receptors in the human proteome likely totals well over 2000 proteins, a healthy number of candidate drug targets by any account. To better grasp the current state of our pharmacological coverage of the druggable proteome, we will consider two strategies that aim to discover chemical probes for new protein targets – 1) chemoproteomics of large enzyme families, and 2) chemoproteomics combined with phenotypic screening. Together, these approaches have generated the first proof-of-relevance chemical probes for many enzymes, receptors, and channels, while at the same time, underscoring that such probes are still lacking for a substantial fraction of the druggable proteome. We will discuss how chemoproteomic methods can facilitate the completion of pharmacological maps for such portions of the druggable, but as-of-yet undrugged proteome.
Several of the largest protein families in human, including kinases, hydrolases, and oxidoreductases, are enzymes that are related by a common mechanism and/or structure. The chemoproteomic method activity-based protein profiling (ABPP) targets the shared mechanistic and structural features of large enzyme classes using active site-directed chemical probes to create a versatile platform for enzyme and inhibitor discovery . Here, we will review how ABPP is being used to develop proof-of-relevance inhibitors for enzymes using the serine hydrolases as a case study.
Serine hydrolases are one of the largest and most diverse enzyme classes in Nature, with greater than 200 predicted members in humans . Serine hydrolases are bound together, not by sequence or structure, but by a common catalytic mechanism that uses an activated serine nucleophile to hydrolyze ester, thioester, or amide bonds in small-molecule, peptide, and protein substrates. Individual serine hydrolases have been found to perform vital biological functions in both prokaryotic and eukaryotic organisms, including involvement in bacterial cell wall biosynthesis, viral protein processing, blood clotting, lipid metabolism, and termination of neurotransmitter and hormone signaling . Selective inhibitors have played prominent roles in the functional characterization of serine hydrolases and, in several instances, been developed into drugs to treat human disorders such as diabetes , obesity , Alzheimer’s disease , and bacterial  and viral  infections. Based on these translational successes, one might presume that pharmacology tools are available for a large fraction of serine hydrolases; however, this is not the case. Indeed, we estimate that proof-of-relevance inhibitors have been developed for less than 10% of the 200+ human serine hydrolases. This percentage does not appreciably increase if one restricts the analysis to the more ‘druggable’ ~120 metabolic serine hydrolases (i.e., removing from consideration the ~120 trypsin/chymotryspin proteases, which some may consider difficult to drug) (Figure 1A).
Given the biological importance and clinical relevance of serine hydrolases, it is worth asking why do most enzymes from this class still lack proof-of-relevance inhibitors? One potential explanation is that the physiologic functions for the majority of mammalian serine hydrolases remain poorly characterized [10, 11], which could limit interest in targeting these enzymes for inhibitor development. This is a classical ‘chicken-and-egg’ problem, where investment in pharmacology awaits deeper biological knowledge, but acquiring this knowledge could itself depend on the availability of pharmacological tools. Indeed, others have noted that, for protein families such as the nuclear hormone receptors, there is a direct relationship between the extent of biological understanding of a protein (as estimated by publications on the protein) and whether a chemical probe exists for this protein . This relationship certainly appears to hold true for serine hydrolases (Figure 1B), which has motivated us to consider how to place pharmacology ahead of biology for these enzymes, such that proof-of-relevance inhibitors can be developed to study the entire enzyme class, including members that lack functional annotation.
Developing inhibitors for enzymes typically requires knowledge of their substrates for assay configuration. This criterion has historically hindered efforts to create full pharmacological maps for large enzyme classes, such as the serine hydrolases, that possess many uncharacterized members. In recent years, however, chemoproteomic methods, such as ABPP, have emerged to address this challenge. We have shown that reporter tagged-fluorophosphonates (FPs)  serve as activity-based probes for the vast majority of mammalian serine hydrolases . Competitive ABPP can thus form the basis for a near-universal assay format for serine hydrolase inhibitor discovery and optimization, where small-molecules are evaluated for their ability to block FP probe labeling in a range of biological systems [19, 20] (Figure 2A). Such competitive ABPP programs might initiate with a high-throughput screen, where the reaction between fluorophore-tagged FP probes and purified serine hydrolases is measured by fluorescence polarization (fluopol, Figure 2A) [21, 22]. Hits from these screens are then immediately screened for activity and selectivity in proteomes using gel- or mass spectrometry (MS)-based platforms for competitive ABPP (Figure 2A), which collectively assay 50+ serine hydrolases in parallel . An interesting output of these proteomic assays is not only the rank-ordering of hits for the screened enzyme target, but also the discovery of lead inhibitors for other enzymes that were part of the counterscreen . Lead compounds that show promising potency and selectivity can be then be optimized through an iterative process of medicinal chemistry and competitive ABPP, culminating in candidate proof-of-relevance inhibitors that are ready for in vivo testing.
Confirming inhibitor activity and selectivity in vivo can be challenging, especially for enzymes that lack known substrates or product biomarkers. Here, we have found that competitive ABPP can also serve an important purpose, where the technology is used to evaluate serine hydrolase activities in cells and tissues from inhibitor-treated animals [22–25]. Competitive ABPP thus enables the enzyme target to serve as its own biomarker for inhibitor activity in vivo, allowing for a direct and quantitative assessment of the extent and duration of target occupancy at a given dose of inhibitor. To be fair, the ex vivo assessment of inhibitor activity by competitive ABPP is technically more straightforward to perform with irreversible enzyme inhibitors (due to the stability of the covalent enzyme-inhibitor interaction), but should also be applicable to tight-binding reversible inhibitors that display slow off-rates.
Following the general competitive ABPP workflow outlined above, proof-of-relevance inhibitors have been developed for several serine hydrolases (Figure 1) and, at least one of these compounds has even progressed into human clinical trials (the FAAH inhibitor PF-04457845 ). The targeted serine hydrolases belong to diverse branches of the enzyme class and include both characterized and uncharacterized members (Figure 1). Lead inhibitors have also been identified for many additional serine hydrolases by competitive ABPP [19, 23], creating the first semblance of a pharmacological map for this large enzyme class. Although still incomplete, this map possesses multiple features that are informative for family-wide inhibitor development programs. First, sequence relatedness proves to be a rather poor predictor of shared pharmacology among serine hydrolases , indicating that traditional strategies that involve counterscreening against the nearest sequence-neighbor enzymes may fail to uncover important inhibitor cross-reactivity across large enzyme classes. Second, competitive ABPP has identified several mechanism-based chemotypes, such as the 1,2,3-triazole urea  and aza-beta-lactam , that have not yet been extensively explored for serine hydrolase inhibition. These chemotypes offer exciting starting points for targeted libraries to improve pharmacologic coverage of the serine hydrolase class. Third, because competitive ABPP provides a unifying assay for initial library screening, hit optimization, and verification of inhibitor activity in vivo, inhibitor development programs have yielded proof-of-relevance probe at an excellent pace (usually within 1–2 years of program inception). The iterative feedback from competitive ABPP assays in proteomes also expands the scope of probe development across the serine hydrolase class by identifying lead inhibitors for additional family members. In this way, many inhibitor optimization programs can be launched and progressed in parallel.
Competitive ABPP and additional chemoproteomic methods have been used to assess inhibitor activity for many other enzyme classes, including cysteine proteases, kinases, histone deacetylases, and cytochrome P450s.
Bogyo and colleagues have employed competitive ABPP to develop selective inhibitors for the malarial cysteine proteases falcipain-1, PfSU B1, DPAP1 , and DPAP3  and have shown that pharmacologic blockade of these enzymes impairs the malaria parasite’s life cycle.
Kinases, like serine hydrolases, are an extremely large and diverse enzyme class with several hundred members in humans. Several kinases, especially those with genetic ties to cancer, have been the focus of intense drug development programs in the pharmaceutical industry, but in aggregate, these only account for a modest fraction of all human kinases. Academic and biotechnology researchers have begun to fill in the pharamcological gaps in the kinome, often using chemoproteomic methods to optimize the target selectivity of lead inhibitors. Multiple strategies have been introduced for this purpose, virtually all of which exploit the conserved ATP binding pocket to create general assay platforms. Examples of successful kinase inhibitor profiling platforms include: 1) phage-display screening of compounds for competitive blockade of interactions between kinases and immobilized nonselective inhibitors [32, 33], 2) MS screening of compounds for competitive disruption of interactions between kinases and immobilized broad-spectrum inhibitors [34, 35], and 3) ABPP of compounds for competitive disruption of interactions between kinases and acylphosphate-ATP probes [36, 37] (Figure 2B). The latter two approaches, which use MS to identify and quantify affinity-enriched kinases, are distinguished by their applicability to native proteomes, which facilitate matching endogenous kinase selectivity profiles with the cellular activity for inhibitors. Analogous to the aforementioned advances made with serine hydrolases, academic and biotechnology efforts have already begun to deliver valuable proof-of-relevance inhibitors for poorly characterized kinases, including RSK1 [38, 39], RIP1 , TORC1/TORC2 , Mps1 , BKM1 , LRRK2 , ERK5 , Ack1 , HIPK , NEK2 , and isoform-selective PI3-K inhibitors . Interestingly, some of these inhibitors irreversibly inactivate kinases by modifying noncatalytic active-site cysteine residues[39, 47], which has enabled their conversion into activity-based probes.
Multiple chemoproteomic strategies for profiling HDAC inhibitors have also been introduced, including the creation of activity-based probes [49–51] and immobilized broad-spectrum inhibitors . Clickable, photoreactive activity-based probes have been shown to label HDACs in living cells, facilitating the discovery of HDAC activities that were impaired upon cell lysis [49, 50]. These ABPP studies also provided some of the first evidence that the HDAC inhibitor SAHA, historically considered a pan-class I and II HDAC inhibitor, shows selectivity for a subset of HDACs (1, 2, 3, and 6) , a finding that has been confirmed with advanced substrate assays . Chemoproteomic enrichment of inhibitor-interacting proteins has also identified unexpected specificity within HDAC protein complexes and additional potential targets for SAHA that are outside of the HDAC family [51, 52].
Activity-based probes for CYP450s, which have been synthesized in clickable form to enable in vivo profiling , have uncovered instances where clinically approved drugs stimulate probe-labeling of individual CYP450s . This ‘activation’ may reflect heterotropic cooperativity, a special feature of CYP450s, which can possess large active sites capable of simultaneously binding multiple small-molecules.
Cell-based phenotypic screening offers another powerful means for developing proof-of-relevance chemical probes for proteins . One potential advantage of cell-based screening compared to the ‘protein family-centric’ approaches mentioned above is that it can identify small-molecule probes for a protein, or collection of proteins, that lack robust biochemical assays. Connecting bioactive small-molecules emerging from cell-based screens with their protein targets remains, however, a technically challenging endeavor. In recent years, chemoproteomic methods have been introduced that are substantially improving the success rate of target identification in small-molecule phenotypic screens. Affinity chromatography, wherein small-molecules are covalently attached to a solid-support and used to ‘fish out’ interacting proteins from cell lysates, is a well-established method that has succeeded in identifying the protein targets of many bioactive compounds, including natural products [3, 57], enzyme inhibitors [5, 58], and hits from cellular screens [59, 60]. But, historically, these approaches have been limited by sensitivity and a lack of quantitation. Chemoproteomic solutions have been introduced to address these problems. First, modern MS methods provide a tremendous boost in sensitivity such that even low-abundance protein targets of small-molecules can be identified. MS methods can also be used to quantify the extent of target enrichment by affinity chromatography, as elegantly demonstrated by Carr, Schreiber and colleagues in their use of SILAC (stable isotope labeling with amino acids in cell culture) to map the targets of several small-molecules (Figure 2C), including the kinase inhibitor K252a  and the anti-cancer agent piperlongumine . In both cases, multiple protein targets were identified (kinases and a set of oxidative stress enzymes, respectively). ABPP methods have also offered a chemoproteomic means to detect, enrich, and identify the protein targets of bioactive small-molecules [59, 60, 62, 63].
The use of chemoproteomic strategies for target characterization in small-molecule phenotypic screens has yielded some intriguing findings, including the realization that many bioactive compounds appear to produce their pharmacological effects through modulating multiple protein targets [59, 61]. That these protein targets can be unrelated by sequence, structure, or function underscores the importance of chemoproteomic methods that broadly survey the proteins that bind to bioactive small-molecules. It is also interesting to note that, while many of the small-molecules discovered in cell-based screens represent the first proof-of-relevance probes for their respective protein targets, the targets themselves typically belong to druggable classes of proteins (enzymes, channels, receptors). In numerous cases where phenotypic screens were employed to identify compounds capable of perturbing targets or pathways perceived to be undruggable, such as Ras , Wnt , mutant p53 , and Sox2-mediated iPS induction , the identified compounds were shown to target enzymes or channels that would be considered quite tractable by conventional drug discovery efforts. Each of these efforts thus discovered new pharmacologic control points for historically challenging biological pathways, while at the same time, underscoring the considerable gaps that remain in our coverage of the druggable proteome.
The contributions made to date by chemoproteomics to early stage drug discovery can be grouped into a few general categories: 1) providing platforms for assaying small-molecule libraries against poorly characterized members of large protein classes, 2) optimizing the selectivity of lead probes by screening against many proteins in parallel, often directly in native proteomes and sometimes even in vivo, and 3) identifying the protein targets for bioactive small-molecules discovered in phenotypic screens (Figure 3). The value of chemoproteomics can also extend to later stages in the drug development process by providing, for instance, a way to understand unanticipated drug toxicities, as well as in vivo-biomarker assays that report on target engagement in humans  (Figure 3). There are clearly certain portions of the proteome, such as the hydrolases and kinases, where chemoproteomic approaches have become integral components of the drug development process and are helping to create proof-of-relevance probes that span the full membership of these large enzyme classes. This integration reflects, at least in part, the advanced state of chemoproteomic technologies available to profile these enzymes. Chemoproteomics has also become the preferred method for addressing the dreaded ‘target identification problem’ in phenotypic screening programs, and can even be used to more carefully re-assess the proteins that interact with drugs of purportedly known mechanisms of action. Bearing such thoughts in mind, we propose a handful of directions where chemoproteomics could be taken to further advance drug discovery.
An emerging theme from the family-centric and phenotypic screening strategies discussed herein is the impact that chemoproteomics can have on developing first-in-class, proof-of-relevance probes for the large number of proteins that fall into the category of druggable, but not yet drugged. Philosophically, it is worth asking, why were these druggable proteins, until recently, lacking chemical probes? One possibility is that many of these proteins were only discovered upon sequencing the human genome and thus could not have been the focus of medicinal chemistry efforts that predated this landmark event. However, we do not believe that this explanation is satisfactory. Take, for instance, the enzymes involved in glycolysis, which were first characterized many decades ago. Most of the pharmacological tools used to perturb glycolytic enzymes consist of low-affinity, broadly reactive chemotypes (e.g., alpha-keto-halogens, arsenate salts) appended onto simplified substrate-like scaffolds that are unlikely to exhibit suitable selectivity and pharmacokinetic properties for in vivo biological studies (Table 1). That specific glycolytic enzymes such as PKM2  and PGAM1 [62, 68] have now moved to the forefront as potential anti-cancer targets only serves to underscore the frustrating gap in our pharmacological toolbox to assess this classical metabolic pathway. Fortunately, efforts are now underway to develop proof-of-relevance probes for glycolytic enzymes , but we should be self-critical enough to ask retrospectively why has it taken so long for these programs to kick into action? We believe, as stated at the outset of this Perspective, that the answer to the question is at least partly related to the displacement of pharmacology by molecular biology and genetics as modern approaches for perturbing protein function. By pushing pharmacology back to the tail-end of biological discovery, we unnecessarily delay the creation of chemical probes that could serve as valuable tools for assessing protein function and translating this information into new medicines.
We recognize that there have been historical challenges associated with the de novo discovery of useful small-molecule probes in academic settings, but with the availability of public screening centers and advancement of chemoproteomic methods for target discovery and inhibitor optimization, these issues have now largely been addressed. We therefore posit that the infrastructure and technologies are in place to complete our pharmacological map of the druggable proteome. Returning, for instance, to glycolytic enzymes as an example. Many of these enzymes possess aberrantly reactive nucleophlic residues  and use cofactor-binding sites (e.g., ATP, NADH) that could be exploited for designing activity-based probes to assist in assaying the activity and selectivity of inhibitors. Lead inhibitors could also be evaluated for their selectivity using the other chemoproteomic methods described in this Perspective.
As already demonstrated for kinases and hydrolases, the integration of small-molecule screening with chemoproteomics can be extended to develop proof-of-relevance probes for proteins that lack known functions. One might ask – why bother developing small-molecule probes for uncharacterized proteins? The answer is at least two-fold. First, these inhibitors, when applied to cell and organism phenotypic assays, offer powerful tools to discover the functions of proteins, as we  and others [28, 43, 44] have demonstrated. These functional assignments can include elucidating the biochemical pathways regulated by proteins , as well as the identification of specific diseases where probe-target pairs might be progressed for drug development. Indeed, one only need consider some of the most successful recent drug targets, such as dipeptidylpeptidase IV (DPPIV) , for which proof-of-relevance inhibitors combined with a simple glucose-tolerance test in rodents would have designated this enzyme as potential type II diabetes drug target. Subsequent ‘peptidomic’ studies could then be used to discover endogenous substrates of DPPIV , which might not only help to explain the pharmacological effects of disrupting this enzyme, but also serve as pharmacodynamic biomarkers for assessing its inhibition in human clinical studies. DPPIV is also an interesting example because it represents a validated drug target that, to our knowledge, lacks any direct human genetic ties to diabetes. Thus, while human genetics can certainly guide us to new drug targets, it should not, in our opinion, be used blindly as a filter for this purpose, especially at the risk of de-prioritizing druggable proteins from direct pharmacological investigation in relevant disease models.
Even if initial phenotypic screens with a chemical probe fail to reveal the biological activity of a protein, having such probes in hand should enable rapid pharmacologic confirmation of functional assignments determined by other, complementary approaches. Mouse and human genetic studies, for instance, are linking missense, nonsense, and activating mutations in protein-coding genes to disease at an increasing rate, but determining the mechanistic basis and translational relevance for genotype-phenotype connections still depends on the availability of pharmacologic probes. From a drug development perspective, having proof-of-relevance chemical probes for poorly characterized proteins thus sets the stage for “anticipatory pharmacology”, wherein a genetic finding can be rapidly exploited for new medicines. An excellent example of anticipatory pharmacology is crizotinib , an ALK inhibitor that was recently approved for treating non-small cell lung cancer (NSCLC). Crizotinib was originally developed as an inhibitor of another kinase, c-Met , but, upon the discovery that activating mutations in ALK are causative for a substantive fraction of NSCLCs , was quickly repurposed for treating this disease. In this case, we were fortunate that crizotinib happened to inhibit not only its originally intended target c-Met, but also ALK. As a contrast, consider another enzyme, isocitrate dehydrogenase-1 (IDH1), for which activating mutations have also recently been linked to cancer [76, 77]. Despite being a well-studied metabolic enzyme (first molecularly characterized over a decade ago ) and a putatively druggable protein, no inhibitors were available for IDH1 at the time of its discovery as a cancer-relevant protein. Progress toward validating IDH as a drug target must thus now await the development of selective and in vivo-active inhibitors for this enzyme. Considering that IDH1 belongs to a large family of NADPH-dependent dehydrogenases, we anticipate that chemoproteomics methods could play an important role in inhibitor optimization for this enzyme. More generally, ALK and IDH1 serve as interesting case studies to advocate for more systematic efforts to create proof-of-relevance chemical probes for the entire druggable proteome, such that biological discoveries linking these proteins to human disease can be rapidly translated into new medicines.
Crizotinib is one of several examples of multi-target kinase inhibitors that have been approved as therapeutics. In some cases, it appears that the polypharmacologic mechanism of action is important for drug efficacy . The realization that drugs often produce their biological activity through affecting multiple protein targets has inspired consideration of purposeful polypharmacology as a way to develop new medicines [76, 79, 80]. Converting polypharmacology into a predictable science, however, will require methods to fully assess the target profile of drugs in complex biological systems, and, here, chemoproteomics stands out as a particularly powerful approach. There are many recent examples where chemoproteomics has helped to define the set of proteins that are responsible for mediating the cellular effects of bioactive small-molecules. In some cases, the compendium of targets belong to the same family of proteins (e.g., the JQ1 and I-BET151 inhibitors of bromodomains [81, 82]; the piperlongumine inhibitor of glutathione S-transferases and related oxidative stress response enzymes  ), while in others, they show little or no mechanistic or structural homology (e.g., the SC1 agent that promotes embryonic stem cell self-renewal through targeting both RasGAP and ERK1 proteins ). Chemoproteomics is particularly well-suited for uncovering such unanticipated cases of shared pharmacology that span unrelated protein families. Once relevant target sets are defined, ensuing medicinal chemistry can focus on coordinately optimizing compounds to maintain the desired target profile for drug action.
In addition to facilitating the characterization of direct targets of bioactive small-molecules, chemoproteomics can also lend insights into the endogenous protein complexes that contain these protein targets. This information can contribute to drug discovery and target validation in several ways, including providing a more complete understanding of the composition of complexes and pathways in which a protein of interest resides and uncovering unanticipated specificity that chemical probes might display for sub-complexes. Moulick and colleagues, for instance, employed affinity chromatography using small-molecules probes and HSP90-specific antibodies to enrich HSP90-containing protein complexes . This approach identified a known HSP90 “client” oncoprotein in chronic myelogenous leukemia cells, Bcr-Abl, as a preferentially associated protein in probe-enriched, but not immunoenriched HSP90 complexes. Furthermore, the abundance of HSP90-associated Bcr-Abl in a cell line was shown to correlate with sensitivity to HSP90 inhibition. As HSP90 is known to associate with a variety of oncogenic proteins , chemoproteomic enrichment of this chaperone in diverse cancers could be used as a method to identify active oncogenic pathways and suitable therapeutic options. A similar approach has been employed to identify the protein targets and associated complexes of diverse classes of HDAC inhibitors . When coupled with quantitative mass spectrometry, these studies revealed compound specificity for not only the intended HDAC targets but also HDAC-associated protein complexes, thus highlighting the importance of assessing compound activity within endogenous complexes, rather than in reconstituted in vitro systems. Finally, a multi-pronged, quantitative chemoproteomics approach was recently used to identify protein-binding partners for several bromodomain and extra terminal (BET) proteins in MLL-fusion leukemic cells . Affinity-capture of BET proteins with acetylated-histone peptides, BET-specific antibodies and BET-binding small molecules provided a quantitative profile of BET-associated protein complexes, which included a novel MLL-fusion protein containing transcriptional complex. This insight provided a rationale to study the effects of BET-inhibitors in MLL-fusion-positive cancer cells, which proved highly sensitive to BET inhibition in vitro and in vivo. Each of these examples highlights how chemoproteomics can enrich our biological understanding of emerging drug targets by mapping their connectivity to disease-relevant protein complexes and pathways.
Much of this Perspective has focused on the role that chemoproteomics can play in completing our pharmacologic map of the druggable proteome. We also believe, however, that the described technologies can impact future efforts to develop chemical probes for the “undruggable” proteome. New chemistries, such as stapled peptides  and small-molecule mimetics of protein secondary structures [86, 87], are emerging that enable pharmacologic perturbation of historically challenging protein classes, such as transcription factor complexes [86, 88, 89], GTPases  and apoptotic effectors [91, 92]. Confirming target interactions and selectivity for such agents in complex biological systems, however, remains a difficult task. Many of the chemoproteomic methods described herein should be amenable to addressing this problem. Given the modular synthesis of peptides one could consider, for instance, embedding within a stapled peptide a photoreactive, clickable unit to enable crosslinking to protein targets in living cells. Indeed, a similar approach was recently employed to map the interaction surfaces of photoreactive, stapled BH3-domain peptides in vitro . Likewise, the full complement of cellular proteins that interact with a protein-protein interaction disrupter could be mapped using affinity enrichment coupled with SILAC proteomics. The information acquired in such chemoproteomic experiments would serve to guide the optimization of drug activity and selectivity across the proteome to accelerate the translation of emerging chemical probes that target ‘undruggable’ proteins into new medicines. Furthermore, enrichment of protein complexes around these targets could in principle identify alternative, more classically ‘druggable sites’ for pharmacologic intervention, as discussed above.
We began this Perspective with a stated goal of making an argument for the re-integration of pharmacology into early-stage target discovery programs. We believe that chemoproteomic methods have matured to the point where they can efficiently guide the development of proof-of-relevance chemical probes for a substantial fraction of the human proteome. These probes possess many important advantages over genetic approaches for target perturbation, including the potential to disrupt protein function in a temporally controlled manner and without affecting protein expression, to partially activate or inhibit protein targets, and to affect multiple protein targets in parallel. It is also instructive to recognize that protein classes, like hydrolases and kinases, which have been a focal point of chemoproteomic investigations to date, also harbor some of the most rapidly expanding suites of chemical probes. The unbiased nature of chemoproteomic methods has even facilitated the development of probes for uncharacterized members of these enzyme classes [19, 46, 94]. We find this contribution of chemoproteomics to modern pharmacology to be particularly exciting in that it is delivering versatile chemical probes that can be used to assign functions to proteins in virtually any biological model or system. Nonetheless, the majority of human proteins, even those that reside within the druggable proteome, still lack proof-of-relevance probes and addressing this problem will require a more systematic approach to integrate chemoproteomics, as well as other screening platforms that evaluate small-molecule activity and specificity , into the basic fabric of pharmacology. We are skeptical that this integration can occur exclusively within the domain of the pharmaceutical industry, which has been steadily moving away from early-stage drug discovery. Rather, we believe the challenge (or opportunity, as we would prefer to view it) will also need to be met by the academic research community. Undoubtedly, continued collaboration between academic and pharmaceutical institutions to match identified compound-target pairs with relevant therapeutic indications will be necessary to rapidly translate early-stage discovery efforts into new medicines. By embracing the development of proof-of-relevance chemical probes, and the continued advancement and implementation of chemoproteomic methods to ensure probe quality, we should enjoy an era of pharmacology-driven biological discovery that enhances the efficiency of converting basic knowledge on protein and pathway function into new first-in-class medicines to treat human disease.
We thank G. Simon for assistance with figures and helpful discussions. We are grateful for the support of the NIH (CA087660, CA132630, DA025285), the Damon Runyon Cancer Research Foundation (HHMI postdoctoral fellowship to R.E.M.), and the Skaggs Institute for Chemical Biology.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.