Why Some Proteins Are Undruggable

A curated map of every drug the FDA has approved found that they act through just 667 proteins from the human genome.^[1] That genome encodes roughly twenty thousand. Almost the entire proteome has never been touched by a medicine, and not because the untouched part does not matter.

"Undruggable" is the word the field uses for much of what is left, and it is worth being precise about what that word describes. It is not a verdict on the biology. It is a statement about the chemistry we happen to have. A protein earns the label when no one has yet found a way to make a small molecule do something useful to it, which is a fact about our toolkit that the field has quietly been treating as a fact about the target.^[2]

What "Druggable" Really Asks For

Conventional drug discovery runs on occupancy. A small molecule finds a site on the target, sits in it, and for as long as it stays there the protein cannot do its job. That model carries a requirement most people never think about, because it is usually satisfied: the molecule has to be able to hold on.

A small molecule is, by construction, small. It touches the protein over a tiny area, which means its grip has to be concentrated rather than spread out. A groove does that. Inside a pocket the molecule is surrounded on several sides at once, and the sum of many weak contacts becomes a hold strong enough to compete with whatever the protein normally binds. On an open, gently curved surface the same molecule makes contact at essentially one point, and it does not stay.

A pocket

The groove wraps around the molecule. Many weak contacts on several sides add up to a grip that holds.

A flat surface

Contact happens at a single point. There is nothing to wrap around, so the molecule drifts off again.

This is why the drugged proteome looks the way it does. Enzymes, receptors and ion channels dominate the list of approved drug targets because their active sites are grooves that evolved to grip small molecules in the first place. We did not choose those targets because they were the most important proteins in human disease. We chose them because they were the ones our chemistry could hold.

Four Ways A Target Becomes Undruggable

The label covers several distinct problems that are worth separating, because they have different consequences and different escape routes.

1. There is no pocket

The commonest case. Proteins that act by binding other proteins meet their partners across wide, flat, feature-poor interfaces, often hundreds of square angstroms with no deep point of attachment anywhere. Protein-protein interactions govern essentially every biological process, which makes them enormously attractive as targets and, for decades, made them a byword for impossible.^[3][4] The hub protein 14-3-3 is the textbook illustration: several hundred interaction partners, a central role across many diseases, and a long spell being described in the literature as undruggable.^[5]

2. There is no fixed shape

Intrinsically disordered proteins do not fold into a defined structure under physiological conditions. They are abundant, they are central to signalling and regulation, and they break the assumption that one sequence gives one structure giving one function. You cannot design a molecule to fit a shape that the target does not reliably hold, and even where a binder is found the protein often stays disordered in the complex.^[6][7]

3. There is nothing to switch off

Inhibition assumes the protein has an activity that can be blocked. Many of the most consequential disease proteins do not. A transcription factor works by binding DNA and recruiting partners; a scaffolding protein works by holding other proteins in proximity. There is no catalytic site to occupy, so even a molecule that bound successfully might change nothing. Transcription factors sit at the convergence point of many signalling pathways, which makes them unusually attractive targets, and they have been considered undruggable largely for this reason together with a shortage of well-defined ligand-binding sites.^[8][9]

4. The pocket exists, but it is not selective

A quieter failure, and a common one. The target has a perfectly good binding site, but so do forty close relatives, and the site is conserved enough that anything binding one binds many. The same applies to a disease-driving mutant that differs from its healthy counterpart by a single residue on an otherwise identical surface. Here the chemistry works and the pharmacology does not, which from a program's point of view is the same outcome.

The Label Expires: What KRAS Proved

KRAS is the most frequently mutated oncogene in human cancer. It was also, for about four decades, the standing example of an undruggable target: a small protein with a smooth surface, no obvious groove, and a grip on its natural ligand far too tight to compete with. Generations of medicinal chemists tried and failed.

Then the surface turned out to have more structure than it appeared to. Structure-based design identified a pocket between the switch I and switch II regions that had been perceived as undruggable, and produced a nanomolar binder for it.^[10] Covalent inhibitors built against the switch II pocket of one particular mutant, KRAS G12C, went on to reach patients with non-small-cell lung cancer.^[11] A target that absorbed forty years of failure became a marketed medicine.

The lesson is usually told as optimism, and it is worth being careful with it. KRAS proves the label is provisional rather than permanent, which is genuinely important. But the breakthrough leaned on a reactive cysteine that exists in one mutant and not in others, and it arrived after four decades. "Undruggable" does not mean never. It means not yet, on a timeline nobody can commit to, which is not something a portfolio can plan around.

Targeted Degradation Asks A Different Question

Every failure mode above descends from one assumption: that the way to deal with a protein is to occupy the site that controls it. Targeted protein degradation drops that assumption. Instead of blocking the protein, it marks the protein for destruction and lets the cell's own ubiquitin-proteasome system remove it.^[12] Three things follow, and each one dissolves a different version of the undruggability problem.

You need a handle, not a switch

An inhibitor has to bind the one site that governs the protein's activity. A degrader only has to hold the protein somewhere, for long enough, in a position that lets a ubiquitin tag be attached. The binding site no longer has to be functionally special. That single change retires failure mode three outright: a transcription factor with no catalytic site to block can still be removed, because removal never depended on blocking anything.

Removing the protein removes all of it

Most disease proteins do more than one thing. A kinase has catalytic activity and also holds complexes together; inhibit it and the scaffolding role survives untouched, which is a well-known route to a disappointing clinical result. Degradation takes away the enzyme and the scaffold at once, because there is no longer a protein there. For targets whose pathology runs through their interactions rather than their chemistry, this is the difference between a partial effect and the intended one.

The effect outlives the drug

Occupancy pharmacology needs the molecule present and bound more or less continuously. Degradation is event-driven: once the target is tagged and destroyed, the degrader is released to do it again, and the protein stays gone until the cell resynthesizes it. One molecule can clear many copies of its target, and the biological effect persists past the exposure rather than tracking it.

Where The Pocket Problem Comes Back

Degradation is often presented as if it had abolished undruggability. It has not, and the gap is worth naming plainly, because it is exactly where most programs on hard targets run aground.

A small-molecule degrader still has to grab the target. It carries one warhead for the target and one for the ligase, and the target-side warhead is subject to every constraint described earlier in this article. Give it a protein with no pocket and it has nothing to hold. Failure mode one survives the shift from inhibition to degradation intact; it has simply moved from the active site to the handle.

What closes the gap is changing what does the gripping. Proteins recognize other proteins across exactly the kind of broad, shallow interface that defeats a small molecule, because that is what protein surfaces are built to do. A genetically encoded binder domain, a DARPin, a nanobody or a designed miniprotein, fused to an E3 ligase, turns that native capability into a degrader. The foundational work on these bioPROTACs showed considerable latitude in both the binder scaffold and the ligase, so the approach generalizes rather than depending on one lucky pairing.^[13]

HuR makes the point concretely. It is an RNA-binding protein whose overexpression tracks with poor prognosis, and it had resisted small-molecule antagonism: an attractive target with nothing to bind. A single-domain antibody against it was built into a TRIM21-based bioPROTAC, and degrading HuR reversed the tumour-promoting behaviour of cancer cells in vivo.^[14] The protein did not become druggable. The thing doing the binding changed.

Which Problem Does Your Target Have

"Undruggable" is too coarse to act on. The useful question is which of the four failure modes applies, because the answer points at a different route.

If there is a decent pocket and the problem is only that blocking it does not do enough, a small-molecule degrader is the natural move: you already have the handle. If the pocket exists but is shared across a protein family, degradation can help, since selectivity can come from the degradation step rather than from binding alone. If the protein has no catalytic site worth blocking but does present a bindable surface, degradation is the whole answer. And if there is no pocket at all, or the only distinguishing feature is a flat interface, the question is not which small molecule to use but whether the binder should be a protein.

Worth saying that the last case is not rare or exotic. It is where most of the unexplored proteome sits, which is the real content of that figure of 667.

Where ProxAI Works

ProxAI designs genetically encoded degraders for the fourth case: targets with no pocket worth the name, where the only thing to bind is a flat interface. Two programs, both from a single design round.

Against the CDC20 D-box degron interface, a 9-amino-acid de novo peptide drove a 9.3× cell-death response versus negative control in HeLa cells, at 2.7× the potency of Apcin, a known small-molecule CDC20 inhibitor. Read the CDC20 case study.

Against a kinase-effector interface in an inflammation pathway, a 90-amino-acid de novo minibinder delivered 82% suppression of the target effector while leaving the target's remaining functions intact. Read the inflammation case study.

If you want the mechanism in more detail, start with what bioPROTACs are, or compare the modalities in bioPROTAC vs. PROTAC vs. molecular glue.

FAQ

References

1. [1]

   Santos R, Ursu O, Gaulton A, et al. A comprehensive map of molecular drug targets. Nat Rev Drug Discov. 2017;16(1):19-34. doi:10.1038/nrd.2016.230

2. [2]

   Xie X, Yu T, Li X, et al. Recent advances in targeting the "undruggable" proteins: from drug discovery to clinical trials. Signal Transduct Target Ther. 2023;8(1):335. doi:10.1038/s41392-023-01589-z

3. [3]

   Skwarczynska M, Ottmann C. Protein-protein interactions as drug targets. Future Med Chem. 2015;7(16):2195-2219. doi:10.4155/fmc.15.138

4. [4]

   Hughes GR, Dudey AP, Hemmings AM, Chantry A. Frontiers in PROTACs. Drug Discov Today. 2021;26(10):2377-2383. doi:10.1016/j.drudis.2021.04.010

5. [5]

   Stevers LM, Sijbesma E, Botta M, et al. Modulators of 14-3-3 protein-protein interactions. J Med Chem. 2018;61(9):3755-3778. doi:10.1021/acs.jmedchem.7b00574

6. [6]

   Santofimia-Castaño P, Rizzuti B, Xia Y, et al. Targeting intrinsically disordered proteins involved in cancer. Cell Mol Life Sci. 2020;77(9):1695-1707. doi:10.1007/s00018-019-03347-3

7. [7]

   Choudhary S, Lopus M, Hosur RV. Targeting disorders in unstructured and structured proteins in various diseases. Biophys Chem. 2022;281:106742. doi:10.1016/j.bpc.2021.106742

8. [8]

   Bushweller JH. Targeting transcription factors in cancer, from undruggable to reality. Nat Rev Cancer. 2019;19(11):611-624. doi:10.1038/s41568-019-0196-7

9. [9]

   Zhuang JJ, Liu Q, Wu DL, Tie L. Current strategies and progress for targeting the "undruggable" transcription factors. Acta Pharmacol Sin. 2022;43(10):2474-2481. doi:10.1038/s41401-021-00852-9

10. [10]

    Kessler D, Gmachl M, Mantoulidis A, et al. Drugging an undruggable pocket on KRAS. Proc Natl Acad Sci U S A. 2019;116(32):15823-15829. doi:10.1073/pnas.1904529116

11. [11]

    Tang D, Kang R. Glimmers of hope for targeting oncogenic KRAS-G12D. Cancer Gene Ther. 2023;30(3):391-393. doi:10.1038/s41417-022-00561-3

12. [12]

    Zhao L, Zhao J, Zhong K, Tong A, Jia D. Targeted protein degradation: mechanisms, strategies and application. Signal Transduct Target Ther. 2022;7(1):113. doi:10.1038/s41392-022-00966-4

13. [13]

    Lim S, Khoo R, Peh KM, et al. bioPROTACs as versatile modulators of intracellular therapeutic targets including proliferating cell nuclear antigen (PCNA). Proc Natl Acad Sci U S A. 2020;117(11):5791-5800. doi:10.1073/pnas.1920251117

14. [14]

    Fletcher A, Clift D, de Vries E, et al. A TRIM21-based bioPROTAC highlights the therapeutic benefit of HuR degradation. Nat Commun. 2023;14(1):7093. doi:10.1038/s41467-023-42546-2