Chemical proteomics has brought rigor to covalent drug discovery and drugs to the clinic. Can it deliver a new generation of drug targets?

Nature Biotechnology.  doi: 10.1038/s41587-025-02737-2. Online ahead of print.

By Ken Garber

In April, Vividion Therapeutics dosedthe first patients in a phase 1 solid tumor trial of its small molecule blocker of the interaction between RAS and phosphatidylinositol-3-OH kinase-α (PI3Kα), designed to target the RAS oncogene without interfering with normal PI3Kα signaling. Few drug companies try to block protein–protein interactions with small molecules. Most such interactions are considered undruggable. Vividion discovered its drug by applying chemical proteomics — the use of small molecules to investigate proteins in complex biological systems.

“This methodology allows you to make discoveries that you couldn’t make any other way,” says Doug Johnson, senior director of chemical biology and proteomics at Biogen. Drug companies typically screen compound libraries against purified single protein preparations in biochemical assays to discover new drugs. But chemical proteomics looks at proteins in cells and can find ligands for protein pockets that are invisible or nonexistent in purified protein preparations. At least four other compounds discovered with chemical proteomics have entered clinical trials (Table 1). Chemical proteomic screens have also become essential for the development of covalent drugs because the analysis of protein–small molecule interactions across the entire proteome reveals any off-target binding. Before adoption of these methods, relatively ‘dirty’ covalent drugs often moved forward, causing serious side effects.

The use of chemical proteomics to expand the druggable space is just beginning. Of the more than 5,000 proteins so far implicated in disease, United States Food and Drug Administration–approved drugs target fewer than 700 of them. “We just haven’t been able to hit them using the traditional screen approaches we as an industry have utilized for decades,” says Jeff Jonker, president and CEO of Belharra Therapeutics. “That opens up a whole other world, right, of opportunity for patients and for the industry.”

Founding a field

Activity-based protein profiling (ABPP) is the most widespread chemical proteomic method. It originated in the late 1990s in the lab of chemical biologist Ben Cravatt at the Scripps Research Institute. Cravatt was studying the role of serine hydrolase enzymes in endocannabinoid metabolism. He and grad student

Matt Patricelli tagged a covalent inhibitor of these enzymes — a fluorophosphonate — with biotin, a vitamin, which can pull down proteins in complex mixtures by sticking them to the protein avidin. In experiments, the covalently reactive probe was surprisingly selective for serine hydrolases. “It occurred to us that … those sorts of probes could be useful for profiling the enzyme class in bulk,” says Cravatt. “It was inspired in part by the idea that genome-scale chemical biology could be done at that moment in time.”

In their original paper (1), Cravatt, Patricelli and Yongsheng Liu used these probes to explore the expression and activity of the entire serine hydrolase enzyme family in different cells and tissues. Because fluorophosphonate reactivity with serine hydrolases requires the enzymes to be in a catalytically active state, Cravatt called the technique ‘activity-based protein profiling’. At around the same time, Matt Bogyo at the University of California, San Francisco, was using similar methods (2) to identify and characterize cathepsins, a family of cysteine proteases. Mass spectrometry (MS) has since become the main detection and identification method, and click chemistry — modular reactions that allow for later addition and removal of the probe biotin reporter tag — is often used to label proteins of interest. But the basic technique remains the same: take a reactive probe molecule, expose a complex biological system to it, then detect, enrich and identify the probe-bound proteins.

Patricelli later applied the concept to kinases and ATPases at the startup company ActivX Biosciences. Meanwhile, Cravatt started looking beyond enzyme active sites, at other cysteines. (Of the amino acids in proteins, cysteines are the most nucleophilic— the most willing to give up electrons to form bonds, making proteins with cysteine-containing pockets easy to label and enrich.) Cysteine-reactive covalent drugs such as Pharmacyclics’ Bruton’s tyrosine kinase (BTK) inhibitor — which later became the megablockbuster ibrutinib — began to appear, and MS was becoming more sensitive and accurate. “We thought, why not look as broadly as possible at cysteines across the proteome for their interactions with small molecules?” says Cravatt. “The covalent chemistry targeting cysteine doesn’t have to be a catalytic cysteine. It can be any cysteine that happens to be in proximity to a pocket.”

Cravatt’s lab started using covalent probes to bind and identify such sites. They built the first robust method to quantify the covalent reactivity of thousands of cysteine residues across the proteome, using small molecule fragments (3). Fragment screening can yield hits using much smaller libraries than drug-like molecules, and can access pockets that drug-like molecules can’t. Patricelli had in the meantime joined Wellspring Biosciences, which had licensed KRAS G12C mutant inhibitors developed by chemical biologist Kevan Shokat at the University of California, San Francisco. Wellspring developed the chemotype later incorporated into virtually all clinical-stage KRAS inhibitors (4). Instead of beginning with a reversible inhibitor and then sticking on an electrophile — the standard industry approach — Wellspring found potent compounds covalently, following Shokat’s example. By starting covalently, “you can find these pockets that are very difficult to discover de novo by noncovalent chemistry,” says Cravatt. “Then you can throw the kitchen sink at them.”

Adoption of these methods took time. Twenty years ago, “Bogyo and I were getting laughed out of rooms because we suggested that covalent chemistry could be useful for drug development,” Cravatt says. Now, “Chemical proteomics has converted covalent drug discovery into a rigorous science.”

Hidden pockets

Vividion Therapeutics, which Cravatt co-founded in 2014 with two other Scripps scientists, aimed to drug these hidden pockets. Patricelli joined in 2016 and is now CSO. (Bayer acquired Vividion in 2021, but Vividion still operates separately.) The company’s core screens, which run continuously, test its electrophile fragment library in cell lysates, which are more convenient for screening than live cells, looking for binders to unknown pockets in proteins of interest. Four clinical compounds have emerged from the screen so far, all for cancer. These include a covalent allosteric inhibitor of the WRN helicase (5), for mismatch repair deficient cancers; an allosteric site activator of KEAP-1, an E3 ubiquitin ligase that degrades Nrf2, a transcription factor regulator of the cellular environmental stress response; an inhibitor of the STAT3 transcription factor; and the RAS–PI3Kα interaction inhibitor (6). Vividion has also reported covalent allosteric inhibitors of the JAK1 kinase (7). At least ten JAK inhibitors have been approved for various indications, but they all target the kinase’s ATP-binding pocket and hit other JAK isoforms to some degree, with side effects. Cravatt’s group, using ABPP to screen for reactive cysteines in human T cells (8), revealed a previously undiscovered allosteric pocket in a JAK1 pseudokinase domain, and Vividion used ABPP against that pocket to find a “the first authentic isotype-specific JAK1 inhibitor,” in Cravatt’s words. Pharma companies missed it because they “don’t study that part of the protein,” he says. “They take the kinase domain in isolation and screen it against a high-throughput screening library. So they can’t even assess that pocket for druggability because it’s not part of the protein preparation.” And purification can’t always reproduce the protein’s native structure, which often depends on cellular binding partners. “A lot of our mechanisms only show up in cells,” says Patricelli. “You can’t run a biochemical screen and get the effect.”

Speed is the main limiting factor when using ABPP for drug discovery. Company electrophile libraries are small — thousands of molecules, versus millions in a big pharma reversible small molecule compound library, or billions in a DNA-encoded screening library. “You need to build your libraries very carefully,” says Dan Erlanson, chief innovation officer at chemical proteomics company Frontier Medicines. “If the electrophile is too hot, you just get nonspecific reactivity; if it’s too cold, you don’t get hits. That’s why irreversible covalent chemistry was so slow to develop.”

MS remains the main bottleneck. “The challenge really is the throughput of the mass spec,” says Johnson. Vividion has industrialized the process, to some extent. “We’re running five to six thousand samples a week, week in, week out,” says Patricelli. Vividion also has used tandem mass tags (9), a multiplexing technique, but the process is still not high throughput.

The throughput limitations of MS, Cravatt says, must be balanced against the fact that they’re screening against a complex biological system, not a single target protein. Vividion, for example, screens its thousands of compounds against sites on thousands of proteins. Throughput “depends on what you’re focused on,” says Cravatt. “If it’s a biological system, I think the throughput is actually pretty good. If it’s a single protein of interest, there are obviously much higher throughput platforms you can use.”

Cravatt offers the example of trying to drug a regulatory T cell. For an unbiased screen, “you’re going to hit thousands and thousands of proteins in a regulatory T cell for druggability, in every setting that you want for that regulatory T cell. How would you do that by another platform?”

Discovery by competition

Chemical proteomics discovery platforms are not limited to covalent drugs. To find pockets for reversible drugs, some companies have employed photoaffinity labeling (PAL), first adapted for this purpose (10) by Scripps chemical biologist Chris Parker, then a postdoc in Cravatt’s lab. PAL uses a photoreactive group, typically a diazarine — a small molecule that, on exposure to ultraviolet light, covalently crosslinks with an adjacent protein. For screening purposes, a library of diazarine-linked lead-like compounds (or fragments), with alkyne handles for click chemistry, is added to cells. Those small molecules that bind proteins — noncovalently — can be identified after ultraviolet light-induced crosslinking of their diazarine groups with target protein side chains. The proteins are then enriched using biotin or fluorescent tags, digested into peptides, and fed (with tags removed) through MS for analysis. The photoreactive group may not bind a pocket, but it’s critical. Without it, explains Belharra Therapeutics CSO Sean Buchanan, “that noncovalent bond wouldn’t be sufficiently strong to withstand the mass spec interrogation.” PAL thus enables reversible small molecule screening in cells or cell lysates.

Belharra, founded in 2021 by Parker, Cravatt and two others, has ongoing collaborations with Genentech and Sanofi. “It is not so much a small molecule discovery platform as it is a pocket-finding platform,” says Jeff Jonker. “What we’re doing with Sanofi and Genentech is trying to find pockets on proteins that they know are playing a role in human disease and they just haven’t been able to get traction on.” These targets have not been disclosed, and no drug from Belharra has yet made it to the clinic.

Jnana Therapeutics in Boston used a related approach (without MS) to arrive at its phase 3 phenylketonuria drug targeting the SLC6A19 amino acid transporter (11). Phenylketonuria is caused by a genetic deficiency in phenylalanine hydroxylase, the rate-limiting enzyme in the metabolism of phenylalanine, causing it to build up to toxic levels in the blood. Inhibiting the SLC6A19 transporter would divert phenylalanine into the urine instead of the bloodstream. But “it was not clear whether SLC6A19 contained pockets that were amenable to inhibition with lead-like chemical starting points,” writes Justin Rettenmaier, head of Jnana’s early discovery group, in an e-mail. Jnana turned to chemical proteomics to discover such a pocket.

To find pockets, Jnana scientists screened its library of several thousand photoaffinity-labeled covalent fragments (called reactive affinity probes, or RAPs) in cells overexpressing SLC6A19. They detected protein binding using an antibody-based ELISA system that is 100 to 1,000 times faster than MS approaches, according to Rettenmaier. They then showed that RAP binding interfered with transport function. Knowing now that SLC6A19 had a druggable pocket, the company performed a high-throughput ‘displacement’ screen of its much larger lead-like compound library, looking for reversible compounds that could outcompete the RAP for binding, and obtained hits. Those hits bound a previously unknown cryptic allosteric site on the protein. Following lead compound optimization and preclinical studies, Jnana entered the clinic with a potent and selective noncovalent SLC6A19 inhibitor in August 2023. The company announced positive phase 1/2 results for its drug last September, shortly before Otsuka Pharmaceutical paid $800 million to acquire the company.

Jnana has expanded its original target scope beyond SLC transporter proteins and is now targeting other hard-to-drug proteins, including transcription factor IRF3, in autoimmune diseases. “It was critical for us to have a plug-and-play technology,” writes Rettenmaier. Jnana’s platform, he claims, can match the throughput of biochemical screening, without that method’s need to redesign the protocol each time. And it can find hits that might be missed by standard functional screening, he adds — ones with little or no functional activity. Chemists can then optimize those hits.

So chemical proteomics does not always produce covalent drugs. Covalent probes “don’t have to be the source of your actual drug structure,” Cravatt says. “Once you know that protein X has a pocket that can be drugged covalently, you can take it off the mass spec, put the extra work in, and use that same probe as the displacement assay to screen proteins against thousands and millions of compounds.” Cravatt published an early example back in 2009 (ref. 12). Such screening “requires you to be able to recombinantly express and purify a protein and have it recapitulate its behavior in the cell,” Cravatt points out. “Which is an assumption, but at least you know the pocket exists because you’ve seen it on the native protein.”

A limitation of this screening system is that the reversible compound must keep the covalent probe off the target. Screening might miss some promising compounds that fail. “If you have a low potent compound, it’s hard to compete covalency,” says Biogen’s Johnson. Jnana’s success with SLA6A19, however, is not unique. Vividion screened for a reversible inhibitor of the transcription factor STAT3, another classically ‘undruggable’ protein. Company scientists found the binding site using chemical proteomics, attaching fragments covalently to a cysteine on STAT3. It then screened a reversible compound library against those probes, looking for displacement and loss of signal. Once Vividion obtained hits, it saw no reason to go back to covalency. Its reversible oral STAT3 inhibitor is now in phase 1 trials for solid and hematologic tumors.

Lessons from a disaster

These drug discovery methods are not yet widely used. But companies have embraced chemical proteomics for specificity testing of covalent inhibitors. “I think every covalent chemistry drug that gets developed these days goes through some kind of iteration of a chemical proteomic experiment,” says Cravatt. The industry’s experience with covalent inhibitors of fatty acid amide hydrolase (FAAH) drove home the need for such testing.

Starting about two decades ago, companies viewed FAAH, the main metabolic enzyme for an endogenous endocannabinoid receptor agonist, as a promising target for chronic pain and other neurological indications. Chemists at several companies discovered covalent FAAH inhibitors. But they faced strong upper management resistance due to fears that covalent drugs would bind nonspecifically to cysteine residues on off-target proteins, with unpredictable side effects.

Doug Johnson was part of Pfizer’s covalent FAAH inhibitor team. “To convince our management to move forward, we really had to show it was selective,” he says. “They were definitely worried.” So the team performed ABPP screening (13) to identify protein off targets, to prove to management that the drugs would not react indiscriminately. This involves treating cells, or cell lysates, with the lead compound followed by a reactive probe and looking for loss of probe signal in the MS readout, indicating off-target binding. Where off targets are found, chemists work to dial out those effects.

Covalent FAAH inhibitors have proven very safe, with one glaring exception — Bial Pharmaceuticals’ BIA 10-2474. In January 2016, during the multiple ascending dose portion of the drug’s phase 1 trial, one healthy volunteer was hospitalized for neurological symptoms and died, and four others suffered serious CNS adverse events (14). (The drug had shown excellent safety up to that point.)

Reasons for the disaster remain in dispute. In 2017, an international team performed ABPP for BIA 10-2474 against 60 members of the serine hydrolase family in cells and found several off targets (15). The following year, a joint Pfizer– Scripps team expanded the screen beyond serine hydrolases and found still more off targets for Bial’s drug (16). Bial does not concede causality. “The implication of these off-targets in the clinical trial accident remains unclear,” writes Susana Vasconcelos, Bial’s director of communications, in an e-mail. “At present there is no evidence that can explain the adverse events observed in the clinical trial, either by Bial, by the authorities or by independent researchers.”

“You can never prove why something is toxic,” concedes Johnson, who worked on the second ABPP study. But, compared to Pfizer’s FAAH inhibitor, “clearly the compound was much less selective, and we found additional off targets.”

Although Bial did not use ABPP, it “addressed the selectivity of compound using the health authorities’ accepted approach, of testing the compounds against panels of biological targets,” writes Vasconcelos, who points out that two other companies did not use ABPP for their clinical stage FAAH inhibitors, at least in published reports.

Still, the contrast between the safe Pfizer FAAH inhibitor and the Bial compound drove home the need to use ABPP for covalent drug development. No FAAH inhibitor has been approved, but, except for BIA 10-2474, the main problem has been efficacy, not safety. (Pfizer’s FAAH inhibitor was effective for cannabis withdrawal syndrome and reduced cannabis use in a phase 2 trial [17]).

Safe space

Most covalent drugs so far have been developed for cancer because side effects can be more acceptable and oncologists closely monitor patients for problems like liver toxicity. “If we’re trying to take the next step and design covalent drugs for chronic indications, I do think this chemoproteomic profiling is going to be even more important,” says Johnson.

One recent chronic disease example is Pfizer’s PF-07853578, a degrader of mutant PNPLA3 for treating metabolic dysfunction associated liver disease (MASLD). The PNPLA3 I148M mutation is the most common and influential genetic MASLD risk factor and causes the PNPLA3 protein to accumulate on lipid droplets. Pfizer performed a phenotypic screen for compounds that removed the mutant protein from lipid droplets in liver cells. But company scientists did not know which proteins the hit compounds were binding, nor what they were doing.

To find out, Pfizer chemical biologists added paired enantiomeric versions of those hits, with identical clickable handles, to cells and used MS to identify the target protein, binding pocket and mechanism. That turned out to be covalent binding to PNPLA3 itself, leading to degradation of the protein. Cravatt had earlier pioneered the use of such mirror-image ‘stereoprobes’ (19) to distinguish true from false binding pockets. If both enantiomers bind, it is probably just due to reactivity with a surface cysteine, whereas if only one binds, it has probably lodged in a real pocket that is shaped to accommodate the molecule in that precise orientation.

After working out the mechanism of PNPLA3 removal, the Pfizer team used MS chemical proteomics again to show specificity of the compound for its target against other serine hydrolases. Selectivity would have been hard to show without chemical proteomics, says Jaimeen Majmudar, an associate research fellow in the chemical biology group at Pfizer. That’s because separate biochemical assays would have been needed for off-target enzymes, and not all were available. And Pfizer wanted to test in cells — hepatocytes — rather than against purified proteins. “These enzymes likely live in different complexes and may have different behaviors in hepatocytes and other cell types,” says Majmudar. “Sometimes biochemical assays can give you a false sense of security.” The PNPLA3 degrader that came out of this work, PF-07853578, recently completed phase 1 trials in healthy volunteers.

Meanwhile, companies such as Vividion and Jnana are employing ABPP, in its different guises, for the discovery of covalent and reversible drugs that are advancing towards and through clinical trials. But those targets were prioritized and selected in advance. “One of the hopes of this area that hasn’t really panned out is new target discovery,” says Biogen’s Johnson. “It’s been a little bit hard from an industrial point of view.” That’s because, he explains, a truly selective fragment hit on a random protein is rare. “You really need to invest an awful lot of medicinal chemistry to optimize that hit to really see where it’s going to go,” says Johnson. “You get an amazing data set, but then there’s a lot more work that needs to be done.” Artificial intelligence should help. Both Frontier Medicines and Matchpoint Therapeutics, which was founded in 2021 by scientists at Stanford University and the Dana-Farber Cancer Institute, have developed machine learning algorithms to guide medicinal chemistry, enhance fragment library design, and prioritize targets for future development.

Chemical proteomics companies are continuously running their core platform screens against the proteome, encountering new pockets. Jonker says Belharra’s system “has yielded more than 4,000 pockets on the undrugged proteome” — the roughly 5,000 proteins implicated in disease. Frontier, which was co-founded by University of California, Berkeley chemical biologist Dan Nomura in 2018, has hits on more than 8,000 proteins. But small companies can only take these early hits so far before they need pharma partners. That’s where the field needs to go, says Cravatt, because so many of the active-site pockets have already been drugged. “The future of small molecule drug development is mostly going to be focused on cryptic allosteric pocket discovery and development,” he says.

No one knows when that future will arrive. But to veteran industry medicinal chemist Mark Murcko, a founding scientist at Vertex Pharmaceuticals, traditional screening methods have become less and less useful. “Drug discovery today is about exploring ever more complex cellular mechanisms of disease,” he writes in an e-mail. “Running biochemical high-throughput screens to find chemical starting points is arguably the exactly wrong choice, since you can never recapitulate the intracellular environment in which your targets operate.” Murcko anticipates phenotypic screens that employ ABPP to uncover new targets and drug mechanisms. “Chemical proteomics is one of the few technologies that bridge the gap between compounds and consequences in real biological contexts,” he writes. “And that’s where proper drug discovery begins.”

Published online: 01 July 2025

References

1. Liu, Y., Patricelli, M. P. & Cravatt, B. F. Proc. Natl Acad. Sci.USA https://doi.org/10.1073/pnas.96.26.14694(1999).

2. Greenbaum, D. et al. Mol. Cell. Proteomics https://doi.org/10.1074/mcp.t100003-mcp200 (2002).

3. Weerapana, E. et al. Nature https://doi.org/10.1038/nature09472 (2010).

4. Janes, M. R. et al. Cell https://doi.org/10.1016/j.cell.2018.01.006 (2018).

5. Baltgalvis, K. A. et al. Nature https://doi.org/10.1038/s41586-024-07318-y (2024).

6. Klebba, J.E. et al. Preprint at bioRxiv https://doi.org/10.1101/2024.12.17.629001 (2024).

7. Kavanagh, M. E. et al. Nat. Chem. Biol. https://doi.org/10.1038/s41589-022-01098-0 (2022).

8. Vinogradova, E. V. et al. Cell https://doi.org/10.1016/j.cell.2020.07.001 (2020).

9. Kuljanin, M. et al. Nat. Biotechnol. https://doi.org/10.1038/s41587-020-00778-3 (2021).

10. Parker, C. G. et al. Cell https://doi.org/10.1016/j.cell.2016.12.029 (2017).

11. Wobst, H. J. et al. JCI Insight https://doi.org/10.1172/jci.insight.182876 (2024).

12. Bachovchin, D. A., Brown, S. J., Rosen, H. & Cravatt, B. F.

Nat. Biotechnol. https://doi.org/10.1038/nbt.1531(2009).

13. Johnson, D. S. et al. ACS Med. Chem. Lett. https://doi.org/10.1021/ml100190t (2011).

14. Rocha, J.-F. et al. Clin. Pharmacol. Ther. https://doi.org/10.1002/cpt.2290 (2022).

15. Van Esbroeck, A. C. M. et al. Science https://doi.org/10.1126/science.aaf7497 (2017).

16. Huang, Z. et al. ACS Chem. Biol. https://doi.org/10.1021/acschembio.8b01097 (2019).

17. D’Souza, D. C. et al. Lancet Psychiatry https://doi.org/10.1016/S2215-0366(18)30427-9 (2018).

18. Wang, Y. et al. Nat. Chem. https://doi.org/10.1038/s41557-019-0351-5 (2019).