Inverse Drug Discovery Identifies Electrophiles Affording Protein Conjugates

Abstract

Traditional biochemical target-based and phenotypic cell-based screening approaches to drug discovery have produced the current covalent and non-covalent pharmacopoeia. Strategies to expand the druggable proteome include Inverse Drug Discovery, which involves incubating one weak organic electrophile at a time with the proteins of a living cell to identify the conjugates formed. An alkyne substructure in each organic electrophile enables affinity chromatography–mass spectrometry, which produces a list of proteins that each distinct compound reacts with. Herein, we review Inverse Drug Discovery in the context of organic compounds of intermediate complexity harboring Sulfur(VI)-fluoride exchange (SuFEx) electrophiles used to expand the cellular proteins that can be targeted covalently.

Introduction

Drug discovery programs seek to identify molecules capable of alleviating the impact of diseases. Through iterative refinement, this effort produces a lead molecule, which engages a specific target to produce disease-relevant functional changes [1]. The most common approaches to drug discovery are in vitro screens against a biochemical target or cell-based phenotypic screens–both employing large collections of structurally diverse small molecules (Fig. 1A, B). Traditional drug discovery approaches are not without their pitfalls. A target-based biochemical screening approach requires extensive knowledge about the target and its relevance to pathophysiology to properly transform screening hits into potent and selective lead compounds. Moreover, targeting a particular protein in vitro might not directly translate to clinically relevant outcomes [2]. Phenotypic screening relies on examining observable changes in cellular phenotype mediated by small molecule treatment. Target identification and mechanistic validation still prove to be non-trivial, although recent chemoproteomic and genetic advancements help address these challenges associated with phenotypic screening [3,4].

Figure 1. Screening Methodologies for Drug Discovery.

A. A target-based approach first involves identifying a small-molecule-amenable causative protein target involved in the pathophysiology of a given disease. A library of compounds is screened against the given target to identify high affinity-binding molecules, which can be developed into a lead molecule primed for validation and optimization in vivo. B. A phenotypic-based approach employs a large library of compounds screened for eliciting a desired phenotypic outcome in a model organism or cell line. Target deconvolution and validation is necessary to determine the mechanism of action of initial hits before medicinal chemistry. C. Inverse Drug Discovery leverages a small library of alkyne-containing latent electrophiles that are screened in various cell lines, followed by identification of the disease-relevant target(s) and optimization of a lead molecule.

Despite the history of using covalent protein ligands in pharmacology to demonstrate proof-of-principle, trepidation remains over their use as drugs because of possible off-target reactivity and toxicity concerns [5,6]. Most covalent drugs gained approval from regulatory agencies before their covalent nature was known. However, the recently FDA-approved drugs ibrutinib, a BTK inhibitor [7], and telaprevir, a serine protease inhibitor [8], were established to be covalent prior to their approval, reflecting a change in attitude in the pharmaceutical industry and within regulatory agencies. Covalent drugs form a chemical bond with a proteinogenic nucleophile as a means to modulate protein activity [9]. A common strategy to minimize reactivity towards closely related protein family members is through using covalent inhibitors that target a distal nucleophile rather than the catalytic nucleophile [10,11].

Activity-based protein profiling (ABPP) exploits active-site directed chemical probes to survey enzymatic family function, which can be leveraged for the discovery of unique ligands against a particular protein [12–14]. Reactivity-based probes extend this approach to survey the global proteome reactivity of a given amino-acid side chain (e.g., the thiol of cysteine), and can similarly provide a successful pathway towards discovery of novel covalent ligands that target a particular protein [15–18]. Quantitative mass spectrometric methods are critical for assessing conjugate formation between proteins and broadly reactive probes harboring a reporter tag such as an alkyne. New covalent ligands can be discovered by monitoring decreases in probe labeling intensity for a given protein when co-treated with an inhibitor molecule exhibiting increased selectivity. Activity-Based Protein Profiling provides valuable information about protein families using highly reactive electrophiles. In contrast, utilizing less reactive electrophiles, or electrophiles that need to be activated by protein binding prior to becoming reactive (latent electrophiles), may allow greater selectivity towards protein nucleophiles in a target-agnostic manner. In this regard, latent electrophiles, i.e., electrophilic species exhibiting environment-dependent protein reactivity towards a single protein or a small number of proteins, but inert towards the majority of the cellular proteome, provide another option to minimize possible off-target activity and subsequent toxicity [19]. Herein, we review an alternate screening strategy, hereafter referred to as Inverse Drug Discovery, that has its conceptual origins in ABPP, and employs latent electrophiles for covalent modification of disease-relevant protein targets (Fig. 1C).

Inverse drug discovery rationale and workflow

Inverse Drug Discovery (IDD) matches structurally unique compounds of intermediate complexity harboring a weak electrophile with proteins capable of undergoing reaction in the context of a living cell or cell lysate. As reversible binding precedes conjugate formation, utilizing structures of intermediate complexity with distinct three-dimensional structures is necessary to ensure reasonable binding affinity (K_D’s in the μM range) with only a subset of the human proteome having an appropriate arrangement of side chain functionality to allow reaction with the latent electrophile. In IDD, the scaffolds of intermediate structural complexity are compounds with a higher molecular weight than a typical fragment (>230 Da) often adhering to Lipinski’s rule of 5 also incorporating a conjugation handle (e.g., an alkyne), which facilitates affinity chromatography–mass spectrometric analysis [20,21]. This methodology generates a list of the proteins reacting with each chemically unique structure harboring a latent electrophile and an alkyne. The designation “inverse” derives from the fact that we are screening the proteome of a living cell against one small molecule. This contrasts with traditional drug discovery approaches, which screen thousands to millions of compounds individually, seeking activity against a distinct target or alteration of a specific cellular phenotype.

An IDD campaign begins with the synthesis of a small collection of alkyne-containing small molecules of intermediate complexity that harbor weak, but activatable electrophiles. Novel synthetic routes to SuFEx-containing electrophiles have dramatically expanded the available scaffolds for IDD [22]. Preequilibrium binding allows for a proximity-driven reaction between the latent electrophile and proteinogenic side chain nucleophiles (i.e., cysteine, lysine, tyrosine) [23]. Reaction of the alkyne tag with biotin azide via a CuAAC “click reaction” or Azide-Alkyne cycloaddition reaction allows affinity enrichment of liganded proteins and subsequent identification of protein targets. After prioritizing the small molecule-protein conjugates for pathobiological relevance, x-ray crystallography allows validation of bond-forming reactions between a given small molecule and a recombinantly-produced protein target, often identifying the amino acid side chain involved in the conjugation reaction [19,24]. As crystallization is not possible for some proteins, peptide mass fingerprinting or cryogenic electron microscopy often proves useful for identifying the conjugation bond formed [25–27].

Mass spectrometry (MS) is critical to the IDD workflow, which harnesses quantitative proteomics to identify the proteins covalently targeted. Following treatment of cells with a small molecule latent electrophile, biotin attachment, affinity enrichment, and tryptic digestion, the peptides afforded are subjected to two or three MS measurements done in tandem, where the mass-to-charge (m/z) ratio of digested and fragmented peptides is detected [28,29]. The m/z ratio and fragmentation pattern allow for identification of the peptide sequences, thus leading to final identification of the target proteins [28]. Knowledge of the modification site is critical for future optimization of the lead molecule; however, the increased hydrophobicity and altered charge state of probe-modified peptides can hinder identification. Novel methods leveraging predictable probe-modified peptide gas-phase fragmentation pathways or electron-transfer dissociation fragmentation have proven useful for identification of the site of modification for cysteine, tyrosine, and lysine side chains, respectively [30–32].

Previous IDD Probes based on Arylfluorosulfates and Sulfuramidimidoyl Fluorides

Sulfur(VI)-fluoride exchange (SuFEx)-functional groups are a prototypical class of latent electrophiles [33]. In the context of an appropriate preequilibrium binding step to a protein having the right constellation of side chains, SuFEx chemistry ensues, which involves nucleophilic attack at the Sulfur(VI) atom followed by concomitant loss of the fluoride ion. Nearby arginine and lysine side chains in the protein undergoing reaction appear to catalyze fluoride displacement [19,34]. SuFEx functional groups are relatively stable towards buffer, and, more importantly, inert towards denatured proteins, and most other macromolecules, including most of the folded proteome lacking the ability to bind and undergo reaction with the latent electrophile [23]. The most-recognized class of SuFEx-derived electrophiles are sulfonyl fluorides, which have historically been used to target serine, lysine, or tyrosine residues [33–39]. However, sulfonyl fluorides are highly reactive and are not ideally suited for IDD [38]. Thus far, IDD has utilized less electrophilic arylfluorosulfates and sulfuramidimidoyl fluorides as weakly-reactive latent electrophiles [19,24] (Fig. 2).

Figure 2. SuFEx chemistry applied to Inverse Drug Discovery.

A. Structure of SuFEx hubs used for proteome reactivity studies with the SuFEx hub highlighted in red and sites of structural diversity denoted as colored shapes. B. Key to SuFEx-derived electrophile●protein conjugate formation is hydrogen-bond donor mediated activation-stabilization of the fluoride ion, and sufficient molecular complementarity between protein and scaffold to allow a proximal nucleophilic residue to attack. C. X-Ray structure of GST01 conjugated to Probe 1 at Tyr229, with Lys57 and Lys59 likely mediating conjugate formation through pK_a perturbation, visualized using UCSF Chimera [85].

The Kelly and Sharpless labs identified arylfluorosulfates as having reactivity towards only a small subset of the human cellular proteome, including intracellular lipid-binding proteins (Table 1) using organic substructures that can bind in such a way as to facilitate the SuFEx conjugation reaction [23,34]. Moreover, the excellent cell permeability and relative buffer stability of arylfluorosulfates—exemplified by stability in basic phosphate buffer for two weeks and neutral buffers for months—highlight their utility as pharmacological tools [23,34,36,40]. Accordingly, arylfluorosulfate probes attached to different organic structures of intermediate structural complexity were found to react with different protein targets (Table 1) [19]. Notably, arylfluorosulfate Probe 1, a quinolinic acid-based molecule, was found to react with GSTO1, a stress responsive protein which has been shown to be upregulated in drug resistance cancer (Fig. 2C), as well as HSDL2, a potential therapeutic target for glioma [19,41,42]. The orientations of side chain residues favorable for SuFEx reactions appear to be a tyrosine nucleophile flanked by adjacent arginine/lysine side chains, as exemplified by GSTO1 wherein Tyr229 is adjacent to Lys 57/Lys59 [19].

Table 1.

Structures of SuFEx-derived electrophiles and corresponding disease-relevant protein targets they selectively react with in whole proteome experiments. Probe numbers in “Name” are those used in the cited paper.

Structure	Name	Proteins Targeted	Disease Indication	Paper
	Biphenyl Arylfluorosulfate Probe 4	CRABP2	Cancer/Osteoarthritis	Chen et al, 2016
		FBP4	Diabetes/Atherosclerosis
	Quinolinic Acid-Based Arylfluorosulfate Probe 1	GSTO1	Drug Resistant Cancer	Mortenson et al, 2018
		HSDL2	Glioma
		TIGAR	Glioma
	Imidazole-Based Arylfluorosulfate Probe 2	GSTP1	Drug Resistant Cancer
		NME1	Cancer
	Imidazole-Based Arylfluorosulfate Probe 3	TTR	TTR Amyloidosis
		CRABP2	Cancer/Osteoarthritis
	Thymidine-Based SAF Probe 2	PARP1	BRCA1/BRCA2-Deficient Cancer	Brighty et al., 2020
	Nitropyridine-Based SAF Probe 13	MIF	Atherosclerosis

Brighty and coworkers recently employed terminal acetylene-containing sulfuramidimidoyl fluorides (SAFs) to extend the capacity of SuFEx-derived electrophiles used in IDD [24]. SAFs are less electrophilic than their arylfluorosulfate counterparts and possess chirality from replacement of the arylfluorosulfates’ S=O bond with a S=NR bond (Fig. 2A). While a small subset of conjugates identified overlapped with previously described arylfluorosulfate SuFEx-reactive proteins, many of the SAF-reactive proteins were unique (Table 1) [19,24,34,43–46]. Of note, a thymidine-based SAF small molecule reacted with the catalytic domain of PARP1, a key DNA repair enzyme commonly targeted by BRCA1/BRCA2-deficient cancer therapies [47,48]. This conjugation led to reduced PARylation in HeLa cells, i.e., inhibition of the enzyme [24]. Coupling multiple organic substructures of intermediate complexity through a SAF connection is an attractive option for developing promising ligands for other therapeutically relevant targets, while minimizing off-target reactivity owing to the inherently lower SAF reactivity.

New Classes of Latent Electrophiles

Electrophile class often dictates the sector of the proteome amenable to covalent labeling, and this has been reviewed elsewhere [11,39]. However, novel chemoproteomic methods have shown that some electrophiles that were thought to be amino-acid-selective are quite promiscuous [49]. Further confounding matters, electrophiles targeting a given amino acid often have differential proteome reactivity, indicating that local stereoelectronic factors intrinsic to a particular protein influence selectivity. As such, recently discovered latent electrophiles may be utilized for IDD applications, as discussed here (Fig. 3).

Figure 3. Novel Classes of Latent Electrophiles.

A. Representative novel classes of recently discovered latent electrophiles. Site of reactivity highlighted in red (cysteine reactive), yellow (tyrosine reactive), and orange (aspartate/glutamate reactive). B. Electrophiles that react through nucleophilic aromatic substitution elicit dramatically different proteome reactivity. C. Metabolically activatable electrophiles, such as 147 (top) formation of quinone methide, and ML210 (bottom) formation of nitrile oxide, react with small subset of the proteome.

Novel latent electrophiles have been recently developed to circumvent off-target reactivity associated with traditional cysteine-targeted electrophiles, which have been extensively reviewed elsewhere [50,51]. Examples include α-chlorofluoroacetamides (reversible covalency), bicyclobutane carboxylic acid (strain release), and thiol-alkyne/methacrylamide reaction (proximity-driven) [52–56] (Fig. 3A). Many of these electrophiles may be amenable to IDD-screening efforts on the appropriate organic substructure of intermediate complexity. Proximity-driven covalent chemistry has been extended to lysine-selective electrophiles, where employing the N-acyl-N-alkyl-sulfonamide (NASA) warhead allows conjugate formation only with sufficient ligand-protein interaction [57] (Fig. 3A). Ligand-directed NASA warheads have found utility for HSP90 inhibition and blocking the HDM2/p53 protein-protein interaction, with other applications surely to come [57,58].

Tyrosine is among the least abundant of all amino acids, and its local environment within a protein has substantial effects on its reactivity [59–61]. In analogous fashion to SuFEx chemistry, the generally more reactive Sulfur(VI)-triazole exchange (SuTEx) chemistry utilizes 1,2,4-triazoles as a leaving group from the Sulfur(VI) hub for reaction with specific tyrosine residues [62] (Fig. 3A). This technology allowed for quantification of both phosphotyrosine sites upon pervanadate treatment and annotation of hyperreactive tyrosine residues, with enhanced proteome coverage compared with traditional sulfonyl-fluoride-based approaches [63]. Importantly, SuTEx ligands could be amenable to the IDD platform, as modulating electronics of the adduct group and leaving group imbues unique proteome reactivity [64]. Exemplifying this is the discovery of a recent SuTEx ligand for modulating prostaglandin reductase 2 [65]. Another underexplored mode of reactivity is through nucleophilic aromatic substitution (S_NAr) [66]. Some electrophiles, such as 2-sulfonylpyridines, are predominantly cysteine selective, whereas others, such as triazine-based scaffolds, can target tyrosines or lysines [67–69] (Fig. 3B). The varied amino-acid selectivity of S_NAr-based electrophiles indicates that protein microenvironment plays a larger role in conjugate formation than nucleophilicity of a given residue. As a result of this promiscuity, S_NAr-based scaffolds may be amenable for IDD.

Covalent ligands directed at carboxylate-containing side chains in proteins have hitherto been relatively unexplored, due to the lack of inherent nucleophilicity. Nevertheless, functionally important glutamate and aspartate may play a critical role in human health and disease, and thus warrant interest for covalent ligand development. The 3-phenyl-2H-azirine moiety covalently labels reactive carboxylic acid residues in vitro and in vivo, such as annotated active site residues [70]. Similarly, alkyl diazirines have distinct reactivity towards carboxylate side chains acids residing on negative electrostatic surfaces [71,72]. Finally, unbiased chemoproteomic methods uncovered novel carboxylic acid-directed ligands, such as hydrazonyl chlorides [73]. Elaborating these electrophiles into more complex scaffolds, and tuning their reactivity may allow target-agnostic discovery of novel ligandable proteins, as has been previously accomplished with fragment-based approaches [74].

A final class of latent electrophiles warranting discussion are those which are metabolically activated to generate reactive electrophilic species capable of protein-conjugate formation proximal to the site of metabolite generation. Some examples include the masked nitrile oxide ferroptosis inducer, ML210, and the quinone methide producing ATF6 activator, 147, among others [75–78] (Fig. 3C). While it is difficult to predict a priori whether electrophilic metabolite generation will occur, understanding the selectivity of these reactive moieties, and others yet-to-be discovered, will surely facilitate expansion of the chemically tractable proteome.

Conclusion and Future Directions

The IDD strategy is a method to connect weak electrophiles attached to organic substructures of intermediate complexity to the human proteins in a cell they react with. Using latent electrophiles to discover covalent ligands for various proteins should reduce off-target reactivity inherent to more reactive functional groups. The target-agnostic nature of the IDD workflow delays decisions about which disease etiology to focus on until after conjugate identification. In addition, choosing organic substructures with a high degree of modularity eases requisite diversification of the initial hit into more potent lead molecules. IDD is conceptually similar to activity-based protein profiling, which typically employs more reactive electrophiles and is related to fragment-based drug discovery efforts. Work in this arena has dramatically expanded the chemical space surrounding the ligandable proteome [15,74,79–82]. While the organic substructure drives proximity-induced conjugate formation through equilibrium binding, the chosen electrophile can also bias proteome reactivity and chemoselectivity, thus both components can be altered to enhance conjugate selectivity [53]. Swapping the sulfonyl fluoride for an arylfluorosulfate on the dcpS targeting-diaminoquinazoline scaffold led to reaction with Ser272 versus Tyr 113/Tyr143, and such a change may abolish reactivity completely in other cases [36]. The gold standard for an IDD-derived pharmaceutical would be an electrophile-harboring scaffold that reacts with only one cellular protein. While some novel electrophiles possess reactivity towards a single target, this selectivity originates from chemical reactivity of the nucleophilic residue, rather than proximity-driven reactivity linked to the 3-dimensional side chain orientation within a protein [83,84]. Such a scenario hinders further medicinal chemistry optimization. Future medicinal chemistry efforts aim to design structures harboring weakly activatable electrophiles capable of reaction with only a single member of the entire human proteome. Advances in understanding the reactivity of unique latent electrophiles will hopefully aid in this endeavor.