Molecular Glues: Mechanisms and Intracellular Pathways

Molecular glues are small molecules that force two proteins to stick together inside a cell, usually by wedging into the contact surface between them and reshaping it so the pairing becomes energetically favorable. Most molecular glues studied today work by recruiting a target protein to the cell’s protein-disposal machinery, tagging it for destruction. This mechanism has turned proteins once considered impossible to drug into viable therapeutic targets, and the field has expanded rapidly from a single accidental discovery into an active area of rational drug design.¹ But not all molecular glues destroy their targets, and the intracellular pathways they engage turn out to be more varied than the early research suggested.

What a Molecular Glue Actually Does

A conventional drug typically blocks a protein’s active site the way a cork plugs a bottle. If the protein lacks a deep, well-shaped pocket, there is nothing for the cork to fit into, and the protein is effectively “undruggable.” Molecular glues sidestep this limitation entirely. Instead of plugging a pocket, a molecular glue binds to one protein surface and modifies it just enough to create a new, artificial contact point with a second protein. The glue itself is part of the interface, acting as a bridge that neither protein would form on its own.²

In the most common therapeutic application, the second protein is part of the cell’s ubiquitin-proteasome system. This is the cell’s internal recycling facility: a cascade of enzymes tags unwanted proteins with a small marker protein called ubiquitin, and then a barrel-shaped complex called the proteasome chews up anything carrying enough ubiquitin tags. Molecular glue degraders hijack this system by tricking an E3 ubiquitin ligase, the enzyme responsible for attaching ubiquitin tags, into recognizing a protein it would normally ignore. Once tagged, the target protein gets fed into the proteasome and destroyed.³

This matters because the glue does not have to permanently block the target protein’s function. It only has to hold the target and the E3 ligase together long enough for ubiquitin tagging to occur. The glue can then release and go recruit another copy of the target. This catalytic behavior means very low concentrations of a molecular glue can degrade substantial amounts of a target protein, which is a real pharmacological advantage.

How Thalidomide Revealed the Mechanism

The molecular glue concept emerged from one of medicine’s darkest chapters. Thalidomide, prescribed as a sedative in the late 1950s, caused severe birth defects and was pulled from the market. Decades later, it resurfaced as a cancer treatment, particularly for multiple myeloma, and scientists eventually worked out why: thalidomide and its derivatives bind to an E3 ubiquitin ligase component called cereblon (CRBN) and reprogram it to target specific proteins the ligase would never otherwise touch.⁴ These redirected targets are called “neosubstrates” because they are new to the ligase, not part of its normal job description.

The thalidomide family, which includes lenalidomide and pomalidomide, creates novel recognition surfaces on cereblon that grab onto proteins containing a specific structural feature: a small loop (a beta-hairpin) with a glycine residue in a particular position. Many zinc-finger transcription factors carry this motif, which is why the drugs can degrade proteins like Ikaros (IKZF1) and Aiolos (IKZF3), transcription factors essential for certain blood cancers to survive.⁵ The discovery was largely accidental. No one designed thalidomide to be a molecular glue; it was only recognized as one after years of mechanistic detective work.

The E3 Ligase Partnership

The human genome encodes more than 600 E3 ubiquitin ligases, but nearly all molecular glues studied so far exploit just a handful. Cereblon is the best-characterized partner, thanks to the thalidomide family. When a drug like pomalidomide binds cereblon’s C-terminal domain, it reshapes the protein surface in a way that allows zinc-finger domains on neosubstrates like SALL4 to dock against cereblon through a combination of contacts. Structural studies show that SALL4 makes contacts with both the C-terminal and N-terminal domains of cereblon when the ligase is in a “closed” conformation, with the drug molecule sitting at the core of the new interface.⁶

A second E3 ligase system involves a different substrate receptor called DCAF15. A class of anticancer sulfonamide compounds, including indisulam and E7820, work by gluing the RNA-binding protein RBM39 to the DCAF15-containing E3 ligase complex. Once recruited, RBM39 gets tagged with ubiquitin and degraded by the proteasome. RBM39 is involved in RNA splicing, and its loss disrupts cancer cell survival. Researchers confirmed this mechanism by showing that mutations in RBM39 that prevent its recruitment to DCAF15 make cancer cells resistant to the drug’s effects.⁷

The fact that only two or three E3 ligases dominate the field reflects a practical bottleneck, not a biological limitation. Recent work has identified glue-mediated interactions involving additional ligase components such as DCAF16, FBXO22, and UBE2D, as well as non-ligase adaptor proteins like 14-3-3σ.⁸ Expanding the E3 ligase toolkit is one of the field’s highest priorities because each ligase has its own tissue distribution and substrate preferences, meaning access to more ligases could open therapeutic opportunities in tissues where cereblon or DCAF15 are poorly expressed.

Cooperativity and Why Some Glues Work Better Than Others

A molecular glue needs to do more than just bring two proteins into the same neighborhood. For efficient degradation, the ternary complex (glue plus the two proteins) has to be more stable than either binary pairing alone. This extra stability is called cooperativity. When a glue molecule binds the E3 ligase, it should make the subsequent binding of the target protein easier, and vice versa. The stronger the cooperativity, the more efficiently the complex forms and the more potent the degradation.

Computational studies have tried to predict this property using simulations that estimate the free energy gained when the ternary complex forms. For a series of cereblon-based compounds, simulations correctly ranked the most cooperative glues as the most potent and the least cooperative as the weakest, though different computational methods sometimes disagreed on the fine-grained ranking between closely matched compounds.⁹ Separately, researchers have shown that for a related series of compounds sharing a common binding mode, established computational methods for predicting binding affinity can predict which compounds will be better degraders of specific neosubstrates like GSPT1 and Aiolos.¹⁰

These results are encouraging but come with a caveat: cellular degradation depends on more than just how tightly the three-part complex holds together. The ubiquitin-tagging machinery needs to be oriented correctly, the proteasome needs to be accessible, and competing cellular processes can interfere. A compound that looks great in a binding simulation might underperform in a living cell if any of these downstream steps are suboptimal.

Molecular Glues That Do Not Destroy Their Targets

The degradation pathway gets most of the attention, but molecular glues can also stabilize protein-protein interactions without sending either partner to the proteasome. These non-degrading molecular glues work independently of ubiquitination and can modulate how protein complexes assemble, where they localize inside the cell, and how signaling cascades propagate.¹¹

Several natural products fall into this category. Rather than co-opting the cell’s disposal system, they lock two proteins together in a way that either enhances or blocks a signaling interaction. The immunosuppressant drug rapamycin, for instance, acts as a molecular glue between the protein FKBP12 and the kinase mTOR, inhibiting mTOR signaling without degrading either protein. This stabilization-based mechanism is fundamentally different from degradation: instead of eliminating a protein from the cell, the glue changes what the protein can do by trapping it in a particular partnership.

The distinction matters for drug design. Degradation is irreversible at the single-molecule level. Once a protein is chopped up by the proteasome, it is gone until the cell manufactures a new copy. Stabilization, by contrast, is reversible when the glue molecule dissociates. Each approach suits different therapeutic goals. If a disease is driven by too much of a harmful protein, degradation makes sense. If the problem is that two beneficial proteins are not interacting enough, stabilization is the better fit.

How Researchers Find New Molecular Glues

For years, molecular glue discovery was almost entirely serendipitous. Thalidomide, lenalidomide, and the sulfonamide drugs were all identified as degraders long after they entered clinical use for other reasons. The field has since developed more systematic approaches, though rational design remains harder for molecular glues than for conventional drugs because the glue must optimize a three-way interaction rather than a simple two-body lock-and-key fit.

One productive strategy combines phenotypic screening with chemical proteomics. In one example, researchers screened a library of 750 covalent compounds for their ability to kill leukemia cells, identified hits, and then used mass spectrometry to figure out which proteins were being affected. One compound, EN450, turned out to selectively degrade the NFKB1 p105 subunit, a component of the NF-κB signaling pathway that drives inflammation and cancer cell survival. Out of more than 4,500 cysteine sites profiled across the proteome, the compound engaged 81 targets, but only one protein was substantially reduced in abundance.¹² That kind of selectivity in a covalent compound was unexpected and demonstrated that molecular glues can emerge from unbiased screens when paired with the right follow-up tools.

Computational approaches are also gaining ground. A platform called GlueFinder mines structural databases for binding pockets that sit right at protein-protein interfaces, spots where a small molecule could plausibly nucleate a glue interaction. When applied to three cancer-relevant targets (EGFR, HER2, and KRAS), the tool predicted candidate glues that could recruit dozens to over a hundred distinct E3 ligases to each target.¹³ These are predictions, not validated drugs, but they illustrate the shift from accidental discovery toward data-driven design.

The Selectivity Problem and Off-Target Effects

Selectivity is the central challenge for molecular glue degraders. Because the glue reshapes the surface of an E3 ligase, it can potentially recruit multiple neosubstrates, not just the one you want to degrade. Common off-targets for cereblon-based glues include the Ikaros family transcription factors (IKZF1, IKZF3), the translation termination factor GSPT1, the kinase CK1α, and the developmental transcription factor SALL4. The thalidomide birth defect tragedy is now understood to involve SALL4 degradation, among other mechanisms, which underscores why neosubstrate selectivity is not an academic concern but a safety-critical one.

The compound CC-885 illustrates the difficulty. Identified through large-scale screening as an extremely potent GSPT1 degrader, achieving greater than 90% degradation at sub-nanomolar concentrations, it was ultimately abandoned because it also degraded IKZF1, IKZF3, and CK1α, and proved highly toxic to normal blood cells and liver cells.¹⁴

Researchers have been trying to decode the chemical features that determine which neosubstrates a given compound will hit. An analysis of a cereblon-modulating compound library found that traditional medicinal chemistry intuitions, like adding or removing hydrogen bonds to tune selectivity, are less predictive for molecular glues than they are for conventional drugs. The three-way nature of the complex means that small chemical changes can have unpredictable effects on which proteins get recruited. The researchers did, however, identify a simplified scoring system using just three chemical parameters that could predict undesired neosubstrate activity.¹⁵

Drug Resistance in the Molecular Glue Context

Like any cancer therapy, molecular glues face the risk of resistance. Because the drug works by creating a new protein-protein interface, mutations at that interface can break the interaction and restore the target protein’s stability. A systematic profiling study across multiple neosubstrates found that resistance mutations are not limited to the direct contact surface. Some mutations in distal parts of the target protein led to only modest decreases in degradation, yet those small changes were enough to let cancer cells survive. The implication is that even a partial rescue of a degraded protein can be clinically meaningful if the cancer depends on that protein for growth.¹⁶

The same study found that many of the protein regions co-opted by molecular glues for degradation are not essential to the target protein’s normal function. This means the protein can tolerate mutations in those regions without losing fitness, making resistance mutations more likely to arise and persist. It is a fundamentally different resistance landscape from conventional enzyme inhibitors, where resistance mutations often come at a cost to the protein’s activity.

Chemical Probes for Mapping Molecular Glue Targets

Understanding what a molecular glue actually binds to inside a living cell is harder than it sounds. Cells contain thousands of proteins, and a small molecule can interact with many of them at low affinity. To address this, researchers have developed photo-affinity probes: modified versions of molecular glues that carry a light-activated chemical group. When exposed to ultraviolet light, the probe forms a permanent bond with whatever protein it is sitting on at that moment, allowing the complex to be captured and identified by mass spectrometry.

One instructive example is photolenalidomide, a modified version of lenalidomide that preserves the parent drug’s ability to degrade the same targets and inhibit cancer cell growth. When used as a probe, it captured the known targets IKZF1 and cereblon from myeloma cells, confirming the tool works as expected. But it also identified a previously unknown target, a translation initiation factor called eIF3i, in a different cell type.¹⁷ Findings like this reveal that molecular glues can have broader cellular interactions than initially appreciated, and that the full picture of a glue’s biology requires profiling across multiple cell types.

A separate study used a related approach, photo-affinity labeling in live cells, to profile the interactome of sanglifehrin A, a natural product molecular glue. That work identified cyclophilin B as the compound’s primary cellular target.¹⁸ These proteomic mapping strategies are becoming standard tools in the field, critical for catching off-target interactions early in drug development.

Covalent Molecular Glues

Most molecular glues interact with their targets through reversible, non-covalent forces: hydrogen bonds, hydrophobic contacts, and electrostatic attraction. A newer class of covalent molecular glues adds an irreversible step. These compounds first recognize their targets through conventional non-covalent interactions, then form a permanent chemical bond with a reactive amino acid (typically a cysteine) on one of the partner proteins. The covalent attachment locks the interface in place, which can stabilize interactions that would otherwise be too fleeting to trigger degradation or functional modulation.⁸

The covalent approach is particularly appealing for targets that present flat, featureless surfaces with no deep pockets. If a weakly complementary surface can be engaged long enough by a covalent anchor, the compound has time to recruit the E3 ligase and initiate ubiquitin tagging. Among the targets being explored through this strategy is KRAS G12C, a mutant form of the KRAS oncogene that drives a significant fraction of lung and pancreatic cancers. KRAS has been a poster child for “undruggable” proteins for decades, and the fact that covalent molecular glues can engage it is a meaningful proof of concept for the approach.

Applying the Concept Beyond Cancer and Beyond Animals

Most molecular glue research focuses on oncology, but the underlying principle of induced protein proximity applies to any biological context where protein removal or stabilization would be useful. Neurodegenerative diseases, for instance, are often characterized by the accumulation of misfolded or aggregated proteins. If a molecular glue could selectively tag those toxic aggregates for proteasomal degradation, it could address a root cause of diseases like Alzheimer’s or Parkinson’s rather than merely managing symptoms.¹⁹

Plant biologists have also taken notice. The ubiquitin-proteasome pathway is deeply conserved across eukaryotes, and plants use their own E3 ligases to regulate growth, stress responses, and development. Researchers have begun exploring whether molecular glues and related degraders could be adapted for plant biology, potentially serving as tools for crop improvement or as precision herbicides that degrade specific weed proteins without harming crop plants.²⁰ The field is early-stage, but it illustrates how a mechanism first uncovered through a drug safety catastrophe has branched into areas no one originally envisioned.