Ligand Discovery by Activity-Based Protein Profiling

Abstract

Genomic technologies have led to massive gains in our understanding of human gene function and disease relevance. Chemical biologists are a primary beneficiary of this information, which can guide the prioritization of proteins for chemical probe and drug development. The vast functional and structural diversity of disease-relevant proteins, however, presents challenges for conventional screening libraries and assay development that in turn raise questions about the broader ‘druggability’ of the human proteome. Here, we posit that activity-based protein profiling (ABPP), by generating global maps of small molecule-protein interactions in native biological systems, is well-positioned to address major obstacles in human biology-guided chemical probe and drug discovery. We will support this viewpoint with case studies highlighting a range of small molecule mechanisms illuminated by ABPP that include the disruption and stabilization of biomolecular (protein-protein/nucleic acid) interactions and underscore allostery as a rich source of chemical tools for historically ‘undruggable’ protein classes.

eTOC

In this Review, Niphakis and Cravatt describe how activity-based protein profiling (ABPP) technologies generate global maps of small molecule-protein interactions in native systems, expanding the druggability of the human proteome. Highlighted are chemical tools discovered by ABPP, including those remodeling protein-protein interactions and acting through cryptic allosteric pockets.

Graphical Abstract

Introduction

Drug discovery is witnessing a wave of new small molecule modalities that deliver their therapeutic effects in diverse and remarkable ways.¹ An increasing number of targets, once deemed undruggable, are finally bending the knee to small molecules capable of accessing cryptic sites, forging covalent bonds with non-catalytic residues or inducing protein-protein interactions among other feats. The success of these efforts often hinges on the discovery of molecules that first bind and then modulate the function of their respective targets. For many proteins, the crux of this process – ligand discovery – can still be a monumental task.

Advances in DNA sequencing² and gene editing³ technologies have propelled our understanding of the genetic basis of human disease and, with it, nominated diverse candidate therapeutic targets; yet, there remains a vast chasm between identifying genes/proteins that drive pathology and developing drugs based on this knowledge to restore health. The disease-relevant proteins illuminated by human genetics originate from a wide array of structural and functional classes that extend far beyond traditional drug targets like enzymes and receptors. This complexity in target biology can impede the earliest steps in the drug discovery process from assay development to hit identification. In contrast to protein classes like kinases and GPCRs, which have benefited from pharmacophore-driven compound libraries⁴ and consolidated functional assays^5,6 the groundswell of human genetics-nominated targets is much harder to address in a uniform manner. While aspects of this challenge have been addressed by large-scale proteomic methods for measuring protein abundances, localizations, and post-translational modifications^7,8, others may relate to the intrinsic ‘druggability’ of some disease-relevant proteins – i.e., the presence or absence of well-defined small molecule-binding pockets. We posit in this Review that an equivalent, if not greater impediment to drug discovery is ‘assayability’ – i.e., the availability or not of methods to screen for ligands that target disease-relevant proteins.

Numerous approaches have been introduced to address the assayability problem, most of which represent generalized ways to measure the binding of small molecules to proteins⁹. Such binding-first assays include affinity selection mass spectrometry (AS-MS)¹⁰, tethering¹¹, DNA-encoded library (DEL) screening¹², differential scanning fluorimetry (DSF)¹³, thermal shift assays^14,15, surface plasmon resonance (SPR)¹⁶, hydrogen-deuterium exchange-mass spectrometry (HDX-MS)^17,18, X-ray crystallography (XRC)- and nuclear magnetic resonance (NMR)-screening^19,20 among others, but notably many of these approaches require purified or engineered proteins. Binding-first assays that are compatible with native proteins in complex matrices, such as cells and tissues, have more recently emerged. For instance, limited proteolysis-mass spectrometry (LiP-MS)²¹ and MS-based cellular thermal shift assays (MS-CETSA)²² can produce interaction maps spanning thousands of proteins in a single sample guided by ligand-induced changes in proteolytic or thermal protein stability, respectively. Activity-based protein profiling (ABPP) represents a complementary chemical proteomic approach that measures small molecule-protein interactions using reporter molecules to survey ligandable pockets in native biological systems. Like other native binding assays, ABPP bypasses the often-arduous task of reconstituting the native state of a protein in vitro and accounts for the various mechanisms by which the cell regulates protein structure and function, including dynamic modification and bimolecular interactions. This Review will discuss the use of ABPP for ligand discovery, from its origins focused on enzyme active sites to its more recent expanded implementation to map cryptic ligandable pockets across the broader proteome. We will provide evidence that ABPP, especially when integrated with advances in covalent chemistry, can offer a streamlined path for chemical probe and drug discovery applicable to a wide range of protein classes, including those historically considered undruggable. We will finally emphasize the atypical modes of action of the novel ligands emanating from ABPP experiments, which include the remodeling of protein-protein and protein-DNA/RNA interactions, as well as the allosteric regulation of multidomain proteins and protein complexes, and ask whether such pharmacological mechanisms may be an enriched byproduct of assaying small molecule-protein interactions in living systems.

Origins and Attributes of ABPP for Ligand Discovery

Since its inception over two decades ago^23,24, ABPP has provided key insights into protein function and ligandability in diverse biological systems. The general principles of ABPP and its broad applicability have been reviewed elsewhere^25–29, and therefore, we will only briefly introduce fundamental concepts as it pertains to ligand discovery. Broadly defined, ABPP is a strategy that uses reactive chemical reagents – also called ‘probes’ – to read out the accessibility of small molecule-binding sites or pockets in proteins^26,29,30 (Figure 1A). When these reagents bind to enzyme active sites, as in the first embodiments of ABPP, this technology and the reagents themselves are deemed ‘activity-based’. This designation stems from their capacity to indirectly report on the activity state of enzymes and the common mechanisms by which enzymes are regulated in biology (e.g., inactive precursor enzymes (zymogens) or endogenous enzyme inhibitors)^23,31.

Figure 1. Overview of activity-based protein profiling (ABPP).

(A) A prototypical ABPP workflow begins with treating native biological systems with an activity-based probe featuring a reactive group, a reporter tag, and a binding group or linker. Together, these features direct the probe to bind and covalently react with protein sites that are complementary to the structure and reactivity of the probe. The reporter tag provides a means to detect, enrich, identify, and quantify the probe-reactive proteins.

(B) Representative reactive groups and binding groups found in original activity-based probes that target the active sites of the indicated enzyme classes. Fluorophores, such as TAMRA, and biotin are common reporter tags allowing visualization and enrichment of probe targets, respectively. In the place of a reporter tag, small biorthogonal handles such as alkynes, can be used, allowing for later-stage introduction of the desired reporter tags by copper-catalyzed azide-alkyne cycloaddition chemistry.

(C) Residue-directed reagents that target specific amino acids based on their unique reactivity have expanded the scope of ABPP beyond enzyme active sites. Such reagents exist for many amino acid side chains including cysteine, lysine, methionine, sulfenic acid and tyrosine.

(D) Competitive ABPP workflows enable ligand discovery in native biological systems through a comparison of control (e.g., DMSO-treated)- versus compound-treated proteomes, where compounds are assigned as ligands for a protein based on blockade of probe labeling.

Reactive sites on each reactive group and residue-directed reagent are highlighted in red.

Original ABPP reagents exploited conserved mechanistic and/or structural features of enzymes to react with many members of a given class (Figure 1B)^{23,24,32–35}. However, commensurate with advances in the sensitivity of mass spectrometry (MS) instrumentation, additional types of reagents that broadly react with nucleophilic amino acid residues like cysteine³⁶ (including S-hydroxycysteine³⁷), lysine³⁸, methionine³⁹ or tyrosine⁴⁰ (Figure 1C) have found utility in chemical proteomic experiments capable of quantifying many thousands of sites in the proteome. These residue-directed profiles contain, but also extend far beyond, the landscape of enzyme active sites originally targeted by ABPP and have been shown to exhibit quantitative relationships with additional types of functionality in the proteome, including redox sensitivity for cysteines^36,41,42 and localization to ligand-binding pockets for lysines³⁸. Thus, while residue-directed reagents, such as reporter-tagged iodoacetamides for cysteine, would not themselves be considered purely activity-based, they can nevertheless inform on diverse features of protein function, including ligand binding, when used in quantitative chemical proteomic experiments that we consider appropriate to term ABPP. When applied to ligand discovery efforts, ABPP generally invokes the simple principle of competitive binding between small molecules and ABPP reagents at a protein pocket (Figure 1D). Ligand binding to a protein is thus determined indirectly through a decrease in reagent binding as measured, for instance, by fluorescence or quantitative MS-based proteomics.

We have previously reviewed the impact of ABPP for the discovery of “orthosteric” inhibitors of enzymes²⁵ (i.e., ligands that bind to active-site pockets) and will therefore only briefly touch on key advantages of this strategy that are also pertinent to its broader use in ligand discovery across the proteome. Arguably the most distinguishing feature of ABPP is that it measures small molecule binding to endogenously expressed proteins in native biological settings. As we will describe below, this property of ABPP has facilitated the identification of ligands that selectively target difficult-to-assay or rare proteoforms of proteins in the cell. Second, ABPP acts a uniform target engagement assay for diverse proteins, including those that remain poorly characterized in terms of their specific biochemical or cellular activities. Ligands can thus be discovered by ABPP for proteins that lack alternative assays, and these small molecules can in turn assist in the ‘chemistry-first’ functional annotation of proteins^43–48, including orphan members of large enzyme classes that had eluded assignment by sequence and structure predictions⁴⁹. Finally, the vast landscape of the proteome surveyed in an individual ABPP experiment, which can range from hundreds (for active site-directed ABPP) to tens of thousands (for residue-directed ABPP) of protein sites, provides a deep understanding of the selectivity of ligands. These attributes of ABPP, taken together, underscore its utility as an end-to-end assay for ligand discovery and development through screening, optimization of potency and selectivity, and ultimately establishing target engagement in preclinical and clinical settings^50–52.

Strategies for Small Molecule Library Design

While ABPP can survey a broad swath of the proteome, the number of compounds that can typically be screened by this method is much more limited. In other words, ABPP experiments, at least when measuring small molecule-protein interactions by untargeted MS-based proteomics, should be considered high-content, but not high-throughput. This limitation has inspired innovations in the design of small, focused libraries that can efficiently survey the proteome for ligandable pockets. Arguably the most prominent success has come from integrating ABPP with covalent chemistry, where screening libraries can be curated to contain molecules featuring reactive groups with preferential reactivity for specific types of proteins or residues. Serine hydrolase inhibitor discovery, for instance, has greatly benefited from screening focused libraries of carbamates^53–56 and ureas⁵⁷, whereas deubiquitinases favor specific cysteine-reactive chemotypes^58,59 (Figure 2A). The screening of compound libraries with tailored reactive groups has also been leveraged in ABPP studies using residue-directed reagents, including those targeting cysteine^60,61, lysine^38,62,63, tyrosine^63,64, methionine⁶⁵ and S-hydroxycysteine³⁷ (Figure 2B).

Figure 2. Strategies for small molecule library design.

(A) Covalent small molecule libraries have been screened by ABPP to identify numerous first-in-class ligands for enzyme classes such as serine hydrolases and cysteine proteases.

(B) Representative reactive groups featured in focused covalent small molecule libraries screened by residue-directed ABPP.

(C) Covalent fragments screened by residue-directed ABPP have generated global ligandability maps in diverse biological systems.

(D) Covalent stereoprobes comprising stereochemically defined sets of compounds constructed from a common scaffold can be screened by ABPP to identify stereoselective ligand-protein interactions across the proteome, furnishing physicochemically matched active and inactive enantiomeric pairs of ligands for diverse protein types.

(E) A general strategy for evaluating the functional effects of covalent stereoprobes in biological systems by comparing active versus inactive enantiomers and wild type versus stereoprobe-resistant point mutants of target proteins.

Reactive sites for each reactive group are highlighted in red.

Of course, matching reactive groups with amino acid residues is nontrivial, and ABPP experiments have revealed, unsurprisingly, that chemotype reactivity and target promiscuity go hand-in-hand. Such promiscuity can be advantageous, as reflected, for instance, in the use of structurally simple electrophilic ‘scout’ fragments^60,66 to gather initial insights on the ligandability of many protein sites^42,60 in diverse biological systems, including human cancer cell lines^42,60,62,67, primary human immune cells⁶⁶, and specific protein classes of interest⁶⁸ (Figure 2C). Nonetheless, the structure-activity relationships (SARs) afforded by ABPP experiments performed with small sets of scout fragments are limited, and the optimization of promiscuous fragment hits into more advanced bioactive tools can be challenging without extensive structure-guided medicinal chemistry effort. Thus, while these studies have enriched our understanding of the scope of sites on proteins that can be targeted by covalent chemistry, most of these sites still lack sufficiently advanced tool compounds for use in cell biology experiments.

The need for such advanced tool compounds is made even more apparent by the recognition that many fragment-liganded sites are non-orthosteric and therefore of unclear functional significance. This conundrum has inspired the design of next-generation libraries that incorporate reactive groups into more structurally diverse and elaborated scaffolds, while, at the same time, not surpassing in size the capacity of ABPP platforms. Covalent chemistry has been integrated, for instance, with diversity-oriented synthesis (DOS)⁶⁹ to produce libraries comprised of focused sets of stereochemically-matched compounds that display recognition and reactivity elements appended to sp³-rich entropically constrained scaffolds (termed ‘stereoprobes’) (Figure 2D).ABPP experiments have showcased key advantages for cysteine-directed stereoprobes over fragments^{63,66,70–73}. First, stereoprobes can assign with greater confidence the tractability of ligandable sites based on their enantioselective reactivity which can be ascribed to non-covalent binding efficiency rather than differences in intrinsic reactivity or physicochemical properties. Second, presumably due to enhanced binding interactions conferred by their more elaborated structures, stereoprobes have been found to show high-occupancy covalent binding to proteins in cells at generally much lower concentrations than electrophilic fragments (5–20 vs 200–500 μM, respectively). Third, the stereoselectivity and site-specificity of stereoprobe-protein interactions provide critical chemical (e.g., active and inactive enantiomeric compounds) and biological (e.g., stereoprobe-sensitive wild type vs stereoprobe-resistant point mutant) controls for functional studies in cells (Figure 2E). These features, taken together, have facilitated a streamlined transition from the global ligandability maps generated by ABPP to functional studies of individual stereoprobe-protein interactions, often with minimal ligand optimization. Stereochemistry has also been incorporated into photoreactive⁷⁴ fragment libraries to enhance the quality of non-covalent ligandability maps generated by chemical proteomics.

Even though only a handful of stereoprobe scaffolds have been analyzed to date by ABPP, they have revealed a remarkably diverse array of stereoselective liganding events in the proteome, many of which occur on proteins, or at sites on proteins, that previously lacked tractable hit matter^{63,66,71,73,75–78}. In subsequent sections, we will highlight case studies of stereoprobe-protein interactions discovered by ABPP that together emphasize the diverse ways that non-orthosteric ligands can impact protein function.

Allosteric Ligand Discovery

The pursuit of orthosteric inhibitors for enzyme classes like kinases has yielded many transformative pharmacological agents, including multiple approved drugs that operate by a covalent mechanism^79,80. Nonetheless, it can be challenging to achieve optimal selectivity for orthosteric kinase inhibitors due to the conserved structures of ATP-binding pockets across this class. Additionally, the high cellular concentrations of ATP can impede the complete engagement of kinases by even potent orthosteric inhibitors. Allosteric kinase inhibitors offer a compelling way to overcome these challenges, as has been showcased by the clinical success of asciminib⁸¹ for treating BCR-ABL-driven chronic myeloid leukemia and the allosteric TYK2 inhibitor deucravacitinib⁸² for treating psoriasis. While several allosteric kinase inhibitors have been identified⁸³, their discovery is often serendipitous because primary biochemical screens rarely discriminate between orthosteric and allosteric binders (the origins of asciminib⁸⁴ and deucravacitinib⁸⁵ can each be traced back to phenotypic screens). Advances in structure-based drug design⁸⁶ and site-directed screening methods such as tethering¹¹ have offered powerful alternative strategies for allosteric ligand discovery. Similarly, the covalent ligandability maps generated by ABPP that provide site-specific binding information are proving to be fertile ground for the discovery of new allosteric inhibitors for several enzyme classes, including kinases, as will be highlighted in several case studies below.

Orthosteric inhibitors of the janus tyrosine kinase (JAK) family, comprising JAK1, JAK2, JAK3 and TYK2, are mainstay drugs for treating various autoimmune disorders⁸⁷. However, achieving high JAK isoform selectivity with orthosteric inhibitors remains challenging, and pan-JAK inhibitors display life-threatening side effects⁸⁸. Mining the cysteine-directed ABPP experiments in human primary T cells identified a scout fragment-sensitive cysteine in the pseudokinase (PK) domains of JAK1 (C817) and TYK2 (C838) enzymes⁶⁶ (Figure 3A). Importantly, neither JAK2 nor JAK3 possesses a cysteine at this position. The potential for compounds to allosterically regulate JAK kinases through binding to the pseudokinase domain was supported by gain- or loss-of-function mutations in this region⁸⁹, as well as the activity of deucravacitinib⁸⁵, which binds the ATP pocket of the TYK2 PK domain. Consistent with these precedents, a more advanced covalent ligand targeting JAK1 (C817) – termed VVD-118313 – was developed by targeted cysteine-directed ABPP and found to selectively inhibit JAK1-dependent cytokine signaling in human immune cells, but not the signaling activity of other JAK kinases⁶⁶. The inhibitory effects of VVD-118313 were not observed with a C817A JAK1 mutant. Curiously, despite retaining engagement of TYK2 (C838), VVD-118313 showed only minimal inhibitory effects on TYK2 signaling, suggesting that this compound acts as a silent TYK2 ligand.

Figure 3. Discovery of covalent allosteric ligands by ABPP.

(A) Screening a library of electrophilic compounds in human cell lysates using cysteine-directed ABPP identified the butynamide hit (–)-1a that targeted an allosteric cysteine (C817) in the pseudokinase (PK) domain (purple) of JAK1. C817 is remote from the ATP binding site (red) in the kinase domain (blue) where current FDA-approved JAK inhibitors bind (PDB 7T6F). Butynamide (–)-1a was further optimized to provide the potent and selective JAK1 (C817) ligand VVD-118313 that suppressed inflammatory cytokine production from human immune cells.

(B) Tryptoline acrylamide stereoprobe MY-9B stereoselectively engages a cysteine in SARM1 (C311) that resides in the autoregulatory ARM domain (blue) adjacent to an NAD⁺ binding site (red) but distant from the catalytic TIR domain (purple) (PDB 7CM6). The stereoprobes inhibit SARM1 biochemical activity and protect axons from degeneration.

(C) A stereoprobe-liganded cysteine in the RNA exonuclease TOE1 (C80) was found by base-editing to reside in a functional allosteric pocket, distant from the catalytic residues (red) (AlphaFold2 Q96GM8), that regulates cancer cell growth. Stereoprobe WX-02–33 site-specifically and stereoselectively inhibited TOE1 nuclease activity and impaired cancer cell growth, phenocopying point mutations at and near C80.

(D) Targeted cysteine-directed ABPP identified an allosteric ligandable cysteine (C727) in WRN helicase that resides >12 Å away from the ATP binding site (red) (PDB 7GQU). Screening a covalent fragment library by MS-ABPP identified hit compound VVD-109063 that inhibited WRN helicase activity in an ATP-competitive manner; however, subsequent analogs bound WRN cooperatively with ATP resulting in enhanced cellular potency. These efforts led to the discovery of VVD-133214, a clinical stage, allosteric WRN inhibitor, for the treatment of MSI-H cancers.

Reactive sites for each covalent ligand are highlighted in red.

Considering the extensive efforts by the pharmaceutical industry to develop JAK inhibitors, it is worth asking – why wasn’t the allosteric inhibitory site on JAK1 discovered earlier? The answer is unlikely to be rooted in the degree of difficulty in ‘drugging’ this allosteric pocket, as it has been found to interact with structurally diverse covalent ligands, including natural electrophilic metabolites, in several ABPP studies performed to date^50,66,77. Instead, we suspect that the ligandable allosteric pocket in JAK1 remained cryptic because it was difficult to assay. Substrate-based biochemical assays for JAK inhibitors often use truncated kinase domains⁹⁰, and it is not even clear that pseudokinase domain ligands would register as hits if such assays used full length JAK proteins. Indeed, VVD-118313 did not inhibit purified JAK1 (aa 438–1154) in a biochemical substrate assay⁶⁶. The cellular mechanism of action for allosteric JAK1 inhibitors remains to be fully elucidated, but appears to involve disruption of auto-transphosphorylation⁶⁶, and recent cryogenic electron microscopy (cryo-EM) studies suggest that covalent ligands engaging JAK1 (C817) are poised to destabilize the trans-activation pose as a means of blocking JAK1-mediated phosphorylation⁹¹. The discovery of isotype-selective allosteric inhibitors of JAK1 by ABPP thus underscores the value of assaying endogenous proteins in their native biological environments, where both the binding and functional effects of small molecules can be more fully understood.

ABPP studies have delivered covalent, allosteric inhibitors for several additional classes of enzymes, including the NAD hydrolase SARM1, the exonuclease TOE1, and the helicase WRN. The liganded cysteines in SARM1 (C311) and TOE1 (C80) showed stereoselective reactivity with tryptoline acrylamide stereoprobes and co-localized with natural (for SARM1) or engineered (for TOE1) mutations that also confer functional effects on these enzymes (Figure 3B and 3C). Interestingly, the co-localizing mutations in SARM1 promote the activity of this enzyme, suggesting that the allosteric pocket, which is found in the autoregulatory armadillo repeat (ARM) domain, might not only support antagonistic, but also agonistic chemistry (a potentially reoccurring theme of allosteric sites that we will return to later). For both SARM1 and TOE1, the inhibitory effects of tryptoline acrylamides were stereoseletive (not observed with inactive enantiomeric compounds) and site-specific (not observed with cysteine-to-alanine/serine mutants). In the case of TOE1, immunoprecipitation-MS experiments revealed that the stereoprobes also promoted binding of TOE1 to several components of the spliceosome, suggesting that the ligands may act, at least in part, by trapping this exonuclease in splicing-related complexes containing its snRNA substrates. Considering the respective roles of SARM1 and TOE1 in axonal degeneration and cancer cell growth, more advanced allosteric inhibitors of these enzymes may have therapeutic potential.

For WRN helicase, Vividion Therapeutics used targeted MS-ABPP to screen eleven solvent-exposed cysteines on this protein, which is considered a compelling oncology target due to its synthetic lethality relationship with microsatellite instability high (MSI-H) cancers^92–94. A covalent hit ligand (VVD-109063) targeting C727 was identified (Figure 3D). Biochemical assays confirmed that VVD-109063 and additional analogs inhibited WRN helicase activity. Considering that C727 is quite distant from the helicase active site and ATP binding pocket, the C727-directed covalent ligands were interpreted to act by an allosteric inhibitory mechanism. Interestingly, a subset of covalent ligands showed much greater engagement of WRN (C727) in cells compared to cell lysates, while other ligands showed impaired cellular activity. These cellular potency shifts for the covalent ligands were demonstrated to reflect ATP-cooperative versus ATP-competitive binding to WRN (C727), respectively. This remarkable finding not only highlights the importance of evaluating small molecule-protein interactions in cells at an early stage in chemical probe and drug development process, but also emphasizes the complexity of allosteric modulation, which can both impact and be impacted by activities at orthosteric sites. Further optimization of the allosteric WRN inhibitors furnished a clinical candidate (VVD-133214) that suppressed the growth of MSI-H cancer cells and tumor models, while showing negligible activity in microsatellite stable cancers or MSI-H cancers expressing a drug-resistant C727A WRN mutant. The discovery of allosteric, covalent WRN inhibitors showcased the value of ABPP across the drug development pipeline, where the approach was used for hit discovery, to provide mechanistic insights into compound action that guided medicinal chemistry optimization, and as a target engagement assay in cellular and in vivo settings.

Beyond Classically Druggable Targets

When one considers the increased pace of discovery of allosteric enzyme inhibitors⁹⁵ (which might themselves be argued to target historically undruggable sites that just happen to reside on druggable proteins), the traditional boundaries for small molecule action across the proteome begin to fade away. This becomes even more evident when one assesses recent advances in identifying functional ligands for notoriously challenging protein classes, like DNA/RNA-binding proteins, E3 ligases, and adaptor/scaffolding proteins, to each of which ABPP has made substantial contributions.

Phenotypic screening has served as a powerful way to discover bioactive molecules that act by non-canonical mechanisms (as noted above for allosteric kinase inhibitors). Focused phenotypic screens of electrophilic compound libraries can be integrated with ABPP to facilitate the streamlined identification of protein targets mediating pharmacological effects of interest. One example is the recent discovery of α-chloroacetamides that stereoselectively suppress the mRNAs encoding androgen receptor (AR) and its major splice variants in prostate cancer cells⁹⁶ (Figure 4A). Cysteine-directed ABPP experiments identified a cysteine (C145) in the RNA-binding protein (RBP) NONO as a shared site of engagement for active compounds, but not structurally related inactive analogs (including inactive enantiomers). Interestingly, mechanistic studies revealed that genetic disruption of NONO blocked rather than replicated the suppressive effects of the active compounds on AR transcripts and cancer cell growth, pointing to a gain-of-toxicity mechanism. Further work demonstrated that the active compounds stabilize NONO-mRNA complexes, leading to a model of delayed processing, maturation, and ultimately loss of transcripts encoding AR and other pro-proliferative gene products. Finally, evidence was provided to indicate that the genetic loss of NONO is better tolerated by cancer cells due to compensatory elevations in paralog proteins PSPC1 and SFPQ. Such paralog increases also occurred in active compound-treated cells but appear unable to override the compound-induced trapping of NONO-mRNA complexes, thus pointing to an intriguing instance of targeted pharmacology that exceeds the impact of genetic disruption. More generally, this work and other studies^71,78 suggest that the combined use of covalent chemistry and ABPP may offer a particularly fruitful path for discovering ligands targeting RBPs, which are often parts of large and dynamic protein-RNA complexes that are challenging to study outside of the cell.

Figure 4. Ligand discovery beyond classically druggable proteins.

(A) Integrated phenotypic screening and cysteine-directed ABPP of a covalent small molecule library led to the discovery of an α-chloroacetamide (R)-SKBG-1 that stereoselectively suppressed androgen receptor (AR) transcript and protein in human prostate cancer cells through engaging C145 of the RNA-binding protein NONO. C145 resides in a hinge between two RNA-binding domains (RRM1 and RRM2) (PDB 3SDE) and ligands targeting this residue were found to stabilize NONO-mRNA complexes, ultimately impairing AR transcription and AR-dependent cancer cell growth. Notably, the effects of NONO (C145) ligands were blocked, rather than recapitulated by genetic disruption of NONO consistent with a gain-of-dysfunction pharmacological mechanism.

(B, C) Integrated size-exclusion chromatography and ABPP identified (B) an azetidine butynamide stereoprobe MY-45B that disrupted the proteasome regulatory complex PA28 (PDB 7DRW) by stereoselectively and site-specifically engaging C22 of PSME1, resulting in remodeling of MHC-peptide complexes in human cancer cells; and (C) a tryptoline acrylamide stereoprobe WX-02–23 that remodels the spliceosome by engaging C1111 of SF3B1. This cysteine is located near the binding site for natural product splicing modulators pladienolide B (PladB) and spliceostatin (PDB 6EN4), and engagement of SF3B1 by WX-02–23 was found to remodel spliceosome complexes, impair splicing, and block cancer cell growth.

(D) Representative covalent ligands discovered for E3 ubiqutin ligases using cysteine-directed ABPP. R-groups represent vectors where ligands for proteins-of-interest were linked to induce protein degradation.

(E) Representative covalent ligands discovered for transcription factors using cysteine-directed ABPP and their respective functional effects on target proteins.

Reactive sites on each covalent ligand are highlighted in red.

Additional types of phenotypic assays have been integrated with ABPP to illuminate covalent ligand-protein interactions that alter protein complexation states⁷¹ and promote E3 ligase-mediated targeted protein degradation (DCAF11⁹⁷, DCAF16^98,99, RNF126¹⁰⁰ and FBXO22¹⁰¹) in human cells, as well as to identify inhibitors of DNA-binding proteins¹⁰². Examples include stereoselective and site-specific covalent ligands that target i) the adaptor protein PSME1, resulting in disruption of its interactions with PSME2 and dissolution of the PA28 proteasome regulatory complex; and ii) the splicing factor SF3B1, resulting in the stabilization of spliceosome binding to helicase DDX42 and global alterations in mRNA splicing in cancer cells. Interestingly, covalent ligand binding to SF3B1 was only detected in living cells, but not cell lysates¹⁰³, possibly indicating that the ligandable pocket is exclusive to a rare and dynamic complexation state (or proteoform) of the spliceosome. Recent structural work has revealed that DDX42 binds to an interface of SF3B1 that also contains several key cancer-relevant mutations¹⁰⁴. The liganded cysteine (C1111) in SF3B1 is located in a nearby pocket at the interface of binding to PHF5A, which is also targeted by complex natural products such as pladienolide B and spliceostatin¹⁰⁵. The discovery of more synthetically tractable covalent SF3B1 ligands may provide an attractive path for medicinal chemists to optimize spliceosome modulators as anti-cancer therapeutics.

Another fruitful avenue has been the ABPP-guided discovery of ligands that regulate protein degradation through covalent binding to proteins of the ubiquitin-proteasome machinery, including E2 and E3 ubiquitin ligases^106,107, deubiquitinases¹⁰⁸ and adaptor proteins^109,110. E3 ligases, in particular, have garnered intense focus for chemical biology and drug discovery applications, as their cellular roles in mediating the disposal of specific protein substrates can be pharmacologically redirected to target disease-implicated proteins^111,112. Despite the massive size of the E3 protein class – with over 600 members in the human proteome¹¹³– only a small fraction has been so far harnessed for targeted protein degradation with small molecules. Screening electrophilic compounds by cysteine-directed ABPP has offered one effective strategy to identify ligands for additional E3 ligases, including DCAF1⁷², RNF4¹¹⁴, RNF114¹¹⁵ and UBR7¹¹⁶ (Figure 4D), that can in turn be retrofitted to induce degradation of proteins of interest¹⁰⁶. The biochemical counterparts of E3 ligases – deubiquitinases – protect proteins from degradation¹¹⁷ and have recently been marshalled for targeted protein stabilization¹⁰⁸. Screening of a covalent library against the deubiquitinase OTUB1 by ABPP identified a ligand for an allosteric cysteine (C23), providing an anchor point for heterobifunctional agents, termed deubiquitinase-targeting chimeras (DUBTACs), that protect target proteins from proteasomal degradation¹⁰⁸. Together, these ligands present exciting opportunities to bidirectionally modulate protein levels for therapeutic applications and beyond.

Transcription factors represent compelling therapeutic targets based on their fundamental roles in regulating gene expression networks that establish cell identity in diseases like cancer and autoimmunity¹¹⁸. However, few protein classes are considered more challenging to address with small molecules than transcription factors¹¹⁸. With the notable exception of nuclear hormone receptors (NHRs), many disease-relevant transcription factors have eluded drug discovery efforts due to their highly disordered structures that dynamically interact with DNA and other chromatin regulatory proteins. Adding to the short list of liganded, non-NHR transcription factors, recent cysteine-directed ABPP studies have identified covalent ligands for FOXA1⁷⁵, SOX10⁴², NFKB1⁴², MYC¹¹⁹ and CTNNB1¹²⁰ (Figure 4E). Interestingly, these transcription factor ligands produce disparate functional effects on their respective targets, such as blocking DNA binding (NFKB1), shifting DNA binding preference (FOXA1), promoting dimerization (SOX10) and inducing degradation (CTNNB1), underscoring the diverse ways that small molecules can affect transcription factor structure and function.

Non-covalent Ligand Discovery

The integration of photoaffinity probes with chemical proteomics represents an expansion of ABPP that provides a method for ligand discovery that is not limited by the presence of reactive residues within or near small molecule-binding sites. Unlike intrinsically electrophilic ABPP reagents, photoaffinity probes show latent reactivity that is typically induced by UV light to convert these probes into intermediates that can form covalent bonds with diverse proteinaceous residues. Several enzymes classes have been addressed by ABPP using photoaffinity probes, including methyltransferases^121,122 and metallo-^123–125 and aspartyl-¹²⁶ proteases. More recently, chemical proteomics with photoaffinity probes has facilitated non-covalent (reversible) ligand discovery across the proteome^127,128 and transcriptome¹²⁹. These experiments have explored several types of photoaffinity groups, including benzophenones, aryl azides, and diazirines^130,131. Diazirines have also been more recently deployed alongside reagents bearing iridium photocatalysts as a complementary way to map spatially demarcated interactions of small molecules in cells^132,133. Advances have also been made in the global mapping of sites of photoaffinity probe engagement on proteins, revealing evidence of ligandability for many non-orthosteric pockets across the proteome¹³⁴. As noted above for covalent ligands, novel sites of photoaffinity probe binding come with the challenge of ascribing functional relevance to both the ligands and the liganded pockets. Coupling photoaffinity probes with phenotypic assays has streamlined the identification of reversible ligands that regulate inflammation¹³⁵, tumor immunity¹³⁶, metabolism¹³⁷, and differentiation¹²⁷.

Challenges and Future Opportunities for Ligand Discovery by ABPP

The virtues of ABPP for ligand discovery are accompanied by several notable limitations. Prominent among these is the limited capacity of ABPP to screen large compound libraries while retaining sufficient sensitivity to detect endogenously expressed proteins en masse. The most comprehensive ABPP protocols rely on untargeted MS-based readouts that can identify and quantify thousands of proteins and ten times as many sites on proteins in a single assay. Adapting MS-ABPP for high throughput applications, however, is problematic due to complex sample preparation workflows and limited data acquisition times of untargeted MS. Targeted versions of MS-ABPP have increased the speed of MS data acquisition in exchange for analyzing a predefined set of proteins⁵¹. Automated sample preparation^138,139 and multiplexing strategies^138,140 coupled with new MS instrumentation capable of rapid and sensitive data acquisition^141–143 have further addressed some of the bottlenecks of untargeted MS-ABPP, although its capacity still falls short of that needed for libraries typically used in high-throughput screening (HTS) campaigns (> 100,000 compounds).

Moving beyond MS-based detection has been key to developing HTS-compatible ABPP assay formats, although these non-MS methods typically require recombinantly expressed protein. Reading out ABPP experiments by fluorescence polarization, for instance, has enabled large-scale small molecule screens against enzymes from multiple families^144–150. A multiplexed ABPP platform termed EnPlex reduces requirements on protein production through immobilization of proteins on spectrally distinguishable microspheres (Luminex beads), allowing for sensitive detection of ~100 enzymes in a single well¹⁵¹. While these methods offer considerably enhanced throughput, recombinant protein expression and purification is generally required. Alternative HTS-compatible ABPP platforms have been introduced that leverage ELISA¹⁵², FRET¹⁵³, BRET^154–156 and DNA-encoded^157,158 readouts enabling screening of recombinant proteins in lysates or cells.

ABPP provides one of the most compelling ways to determine the proteome-wide selectivity of small molecules in biological systems, and most advanced covalent ligands and drug candidates are now evaluated by this method^51,159–161. Nonetheless, alterations in amino-acid residue reactivity in the presence of competitor molecules may not always reflect the direct covalent binding of ligands. Ligands can occasionally indirectly perturb residue reactivity through conformational changes in protein structure or inducing post-translational modifications (e.g., cysteine oxidation)⁴². In this regard, integrating residue- and protein-directed ABPP data can facilitate identifying indirect vs direct liganding events based on their respective and complementary loss- and gain-of-signal readouts⁷⁰. ABPP experiments, especially when performed in residue-directed (e.g., cysteine-directed) formats are also beholden to the detection capabilities of MS, which can vary widely for peptide analytes based on their physicochemical properties (often termed ‘proteotypicity’¹⁶²). Sensitivity limits for detecting low abundance and membrane proteins, which have posed challenges for MS analysis,^163,164 can also produce gaps in ligandability maps. Recent work has shown that a more comprehensive understanding of the protein targets of covalent small molecules can be achieved by combining residue-directed ABPP with a complementary ABPP method that enriches and identifies proteins using alkynylated variants of the small molecules⁷⁰. This “protein-directed” ABPP approach, while sacrificing site mapping information, increases the probability of detecting protein targets of covalent compounds, as multiple tryptic peptides can be quantified per protein. Although synthesizing alkynylated versions of covalent small molecules increases the initial workload for ABPP, we call attention to additional attributes of these alkynylated probes, which can function in convenient off-MS (e.g., gel-based) ABPP assays, as well as potentially allow for the identification of ligands of rare proteoforms of proteins that might not be resolved when using broader reactivity (e.g., residue-directed) reagents.

The extent to which ABPP can enable ligand discovery is also dependent on the quality and diversity of small molecule libraries screened by this method. Covalent ligand discovery efforts, as evident from the case studies highlighted above, have, so far, been dominated by cysteine-targeting agents owing to the exceptional reactivity of cysteine and the maturity of chemistry enabling the development of highly selective ligands for this residue. The relative ease of discovering covalent ligands for cysteines versus other amino acids is counterbalanced by the scarcity of cysteines within ligandable pockets. Advances in integrated covalent chemistry and chemical proteomics have begun to address residues beyond cysteine^37–40,63, and the steep rise of electrophilic chemotypes capable of accessing additional residues^{37,62,64,65,165–168} in proteins has laid a strong foundation for future covalent ligand discovery efforts.

As the number of covalent small molecule-protein interactions mapped by ABPP continues to grow (>300 stereoselective interactions were described in a recent study⁷⁰), the challenge of determining whether and how such covalent liganding events, the vast majority of which are occurring at non-orthoseric sites, affect protein function becomes apparent. The proximity of disease-relevant mutations to ligandable sites, as found for JAK1⁵⁰ and SARM1⁷³, can increase confidence that covalent ligands will confer functional effects on the target proteins. While such instances are admittedly, at this time, anecdotal, we envision a much deeper and more comprehensive integration of ligandability and genetic maps in the near future. Structural biology offers another way to infer the functional relevance of ligandable sites emerging from ABPP studies. With considerable advances in experimental (e.g., cryo-EM) and predicted (e.g., AlphaFold) structural elucidation techniques, the corpus of solved and theorized protein structures is facilitating the localization of ligandable sites to, for instance, protein-protein interfaces^70,71,155 or dynamic regulatory domains⁹¹. These structures also serve as vast training sets for artificial intelligence/machine learning (AI/ML) models to predict functional domains¹⁶⁹ and their potential ligandability^170,171. Such AI/ML predictions will nonetheless have to grapple with the sobering reality that many allosteric sites of small molecule action occur at cryptic pockets that are not visible in apo-protein structures^51,155,172. Accounting for conformational protein dynamics in AI/ML efforts may be particularly important for predicting allosteric sites of druggability¹⁷³.

In closing, we are now ~25 years from the initial description of ABPP, which began as an effort to develop chemical approaches for protein science that were capable of matching the scale of genome-level research. In doing so, ABPP has facilitated the assignment of functions to diverse human disease-relevant enzymes, provided generalized assay formats for ligand discovery and optimization that contribute to virtually every step of chemical probe and drug development, and most recently demonstrated a capacity to reshape our understanding of the global ligandability of the human proteome. In considering the expansion of ABPP from orthosteric (active) sites to diverse ligandable pockets across the broader proteome, we believe small molecule drug development as a whole is on a similar trajectory, where compounds acting by allostery and the remodeling of macromolecular interactions will become a primary source of therapeutics. How one assays systematically for such noncanonical mechanisms of drug action remains a key question, and as much as we are confident ABPP will continue to make major contributions, we are excited to imagine the emergence of additional methodological innovations that can address questions of protein ligandability in native biological settings. Toward this end, we have emphasized the integration of ABPP with covalent chemistry in this Perspective, but similar chemical proteomic principles are also applicable to non-covalent chemistry^{127,128,130,174} which we expect will further augment the landscape of the proteome that can be addressed by small molecules.