Targeted protein degradation is evolving into a key modality in drug discovery. Harnessing the endogenous protein degradation system in cells, protein degraders bring a target protein into the proximity of an E3 ubiquitin ligase, which mediates the subsequent ubiquitination and proteasomal degradation of the target. Proteolysis-targeting chimeras (PROTACs) and molecular glue (MG) degraders are two major classes of protein degraders (Figure 1a). PROTACs are bivalent molecules consisting of a ligand to a target and a ligand to an E3 ligase. The molecular weight of PROTACs generally falls in the 700–1100 Da range, which could be a hurdle for drug delivery and bioavailability. In contrast, MG degraders are monovalent small molecules with a molecular weight usually less than 500 Da and broadly have more drug-like properties. The mechanism of action of MG degraders is first exemplified by the plant hormone auxin [1]. Later, it is recapitulated by immunomodulatory drugs (IMiDs) [2,3] and investigational compounds such as aryl-sulfonamides [4, 5, 6] and cyclin K degraders [7, 8, 9]. Proteins targeted by MG degraders include transcription factors IKZF1/3 and RNA splicing factor RBM39, which lack well-defined ligand-binding pockets and are intractable to conventional small molecule inhibitors and PROTACs. The potential of MG degraders to clear non-ligandable proteins and their success in clinical contexts have generated tremendous enthusiasm for further exploration and prospective development.

Despite the growing interests in MGs, we still lack a comprehensive understanding of the thermodynamic characteristics that drive the ternary complex formation. Confusions between MGs and PROTACs in the literature have blurred the unique properties of MGs, hindering their rational discovery. Cao et al. systematically studied the known MG systems such as auxin, IMiDs, and a dual-nanobody cannabidiol sensor [10], and found that the MG-enabled ternary complex primarily depends on three interfaces, the interface of MG with the MG receptor (E3 or target), the interface between the MG receptor and its dimerization partner (target or E3), and the interface of MG with the dimerization partner (Figure 1a). The measurable affinities associated with the first two interfaces relax the requirement of strong interactions at the third interface for the stable ternary complex to form. Based on these properties, MG degraders are defined as a unique class of protein degraders, which strengthen the intrinsic basal affinity between a target and an E3 ligase without showing a detectable affinity towards at least one of the protein binding partners. The definition differentiates MG degraders from bifunctional PROTAC molecules and provides a basis for developing MGs to target non-ligandable proteins. Here, we review recent progress in rational MG degrader discovery, compare canonical and non-canonical MGs, and discuss opportunities and challenges in future therapeutic applications.

While MGs have been mainly discovered by serendipity in the past, prospectively discovering MGs is becoming a major theme in targeted protein degradation. We classify the reported strategies for rational MG discovery into two major categories: ligand modification and de novo discovery (Figure 1b). The ligand modification approach introduces chemical substituents on an existing ligand of either an E3 ligase or a target protein to create new interfaces for molecular glue activity. The de novo discovery strategy, on the other hand, does not rely on any known chemical structures. Instead, compound libraries are screened using in vitro biophysical assays or in vivo cell-based assays to identify new chemical scaffolds with molecular glue activities.

Thalidomide and its analogs, lenalidomide and pomalidomide, known as IMiDs, are the most well-characterized MG degraders. IMiDs reprogram cereblon (CRBN), a substrate receptor of the CULLIN4-RING ligase (CRL4) E3 complex, to degrade the Ikaros family zinc finger proteins, IKZF1 and IKZF3, and casein kinase 1A1 (CK1α) [2,3,11,12]. CK1α degron adopts a β-hairpin loop and forms three hydrogen bonds with CRBN next to its lenalidomide-binding pocket (Figure 2a) [13]. Molecular dynamics studies show that the binding of lenalidomide provides hydrophobic shielding of the pre-existing hydrogen bonds between CK1α and CRBN, thereby greatly increasing the energy required to break the hydrogen bonds formed between the E3 and the target protein [14]. These studies are consistent with our observation that there is an intrinsic affinity of ∼2 μM between CK1α and CRBN (Table 1) [10], highlighting the significance of the pre-existing target-E3 interactions for the glue action.

IMiDs anchor to CRBN via its glutarimide ring, while the phthalimide group engages the substrate. Structure–activity relationship (SAR) studies are performed on the phthalimide group to repurpose CRBN to degrade other proteins (Figure 3a). For instance, CC-885 is one of the analogs with potent antitumor activity and induces the degradation of a novel substrate GSPT1 [15]. A more selective GSPT1 degrader CC-90009 is currently under Phase I clinical trial for the treatment of acute myeloid leukemia [16]. IKZF2 has also been successfully targeted by this approach [17∗, 18, 19, 20]. The degron sequence of IKZF1/3 and IKZF2 differs only by a single amino acid, Q146 versus H141. Guided by structural information, IKZF2 degraders like ALV2 and NVP-DKY709 are optimized to specifically engage histidine in IKZF2 via a π-system but would clash with the glutamine residue in IKZF1/3. In addition to IKZFs, CK1α, and GSPT1, many other proteins, such as GSPT2, PDE6D, and ZBTB16 [21, 22, 23, 24], are degraded by IMiD derivatives. Although these proteins do not share sequence homology, most of them possess a glycine in the degron sequence and adopt the same β-hairpin loop as observed in CK1α. This structural motif provides a weak basal interaction between CRBN and the neo-substrate, and is now referred to as the glycine loop [21]. Although most neo-substrates of CRBN that have been documented so far feature a glycine loop, it is conceivable that alternative weak protein–protein interaction interface centering around the IMiD-binding pocket could bypass such a requirement for IMiD-like compounds to function as MGs.

In addition to tuning substrate specificity, SAR studies also improve the degradation activity of IMiDs. More potent analogs such as iberdomide and mezigdomide have been developed [25,26]. A recent cryo-electron microscopy study proposes that the degradation activity of IMiDs correlates with their ability to induce the allosteric transition of CRBN from an open to closed conformation [27]. Although there is still debate regarding factors driving the conformational change [17], future development of CRBN small molecule modulators should consider both the CRBN-binding kinetics and the allosteric state of CRBN.

While most efforts in the field focus on derivatizing CRBN ligands, degraders building upon other E3 ligands are emerging. For instance, 13–7 is a highly cooperative molecule generated from a ligand of E3 ligase VHL (Figure 3b) [28]. 13–7 has a weak affinity to BRD9 but shows a nanomolar affinity in the presence of VHL and BRD9. This is reminiscent of a linker-optimized PROTAC molecule that was used to establish the concept of cooperativity by the Ciulli group [29]. Considering that there are more than 600 E3 ligases in the proteome, the number of ligases that have been exploited for MGs is still very limited. Ligands for other E3 ligases such as DCAF1 [30, 31, 32], SOCS2 [33], KEAP1 [34], and TRIM58 [35] have been reported, which would greatly expand our toolbox for MG discovery. It is of particular interest to explore tissue specific E3 ligases, which might reduce the side effects associated with systematic degradation of a target protein [36].

Target-oriented drug discovery identifies small molecule ligands that enhance or inhibit the function of a specific target protein. However, there have been instances where a ligand unexpectedly induces the degradation of its target. For example, two FDA approved drugs fulvestrant and selinexor are found to degrade their targets, estrogen receptor and nuclear export protein exportin 1, respectively [37,38]. These findings suggest that evaluating known inhibitors or chemically modifying existing protein ligands might represent feasible strategies to discover protein degraders.

Several chemically diverse compounds have recently been reported to degrade cyclin K by recruiting CDK12-cyclin K to the DDB1-CUL4-RBX1 E3 ligase [7, 8, 9,39,40]. Among them are multi-CDK inhibitor CR8 and CDK12-selective inhibitor SR-4835 and 919278 [39]. These compounds occupy the ATP-binding pocket of CDK12 similarly to other CDK inhibitors, such as roscovotine, but possess an extra ‘gluing moiety’, which protrudes out of the kinase and directly engages DDB1 (Figure 2b). Additionally, CDK12 and DDB1 form substantial protein–protein interactions with the C-terminal tail of CDK12 extending to a cleft in DDB1 [7]. The basal affinity between CDK12 and DDB1 is ∼50 μM (Table 1) [7,39]. Albeit being non-productive under physiological conditions, the weak CDK12-DDB1 basal interaction primes the formation of a stable ternary complex upon binding to the cyclin K degraders.

JQ1 is a well-characterized inhibitor of bromodomain proteins including BRD2 and BRD4 (Figure 3b) [41]. Chemical modifications on the chlorophenyl position of JQ1 leads to GNE-0011, an analog which induces proteasome-dependent degradation of BRD2/4 [42,43]. Subsequent CRISPR screen identifies DCAF16 as the E3 ligase responsible for GNE-0011 induced BRD2/4 degradation [44]. An independent group reports GNE-0011 acts as a covalent MG degrader [45]. The authors found that BRD4 weakly associates with DCAF16, which serves as a structural template to facilitate covalent modification of DCAF16. Replacing the amine tail of GNE-0011 with more reactive covalent warheads further improves the activity of the degrader, as seen in MMH1 (Figure 3b). Interestingly, substituting the tert-butyl group of JQ1 with amide-linked cyanoacrylamide leads to compound RCS-03-104-4, which recruits another E3 ligase DCAF11 for the degradation of BRD4 (Figure 3b) [46].

Besides covalent MGs, a series of non-covalent BRD4 degraders are discovered unexpectedly when designing BRD4-targeting PROTACs (Figure 3a) [47,48]. E7820, a ligand of the E3 ligase DCAF15, is used as the warhead in the PROTAC design. However, the resulting compound, IBG1, recruits a different E3 ligase, DCAF16, to promote BRD4 degradation [49]. The E7820 and JQ1 moieties engage the tandem bromodomains (BD1 and BD2) of BRD4 in cis, while the linker region is sandwiched between BD1, BD2, and DCAF16, making critical contacts to the two proteins to stabilize the ternary complex (Figure 2c). BRD4 and DCAF16 form a network of hydrogen bonds and hydrophobic interactions and show an intrinsic affinity of 1–4 μM (Table 1). Surprisingly, IBG4, an analog of IBG1, induces BRD4 degradation via E3 ligase DCAF11. BRD4 and DCAF11 exhibit a basal affinity of 3 μM, and this interaction is significantly enhanced in the presence of IBG4 [49]. Even though IBGs are initially designed as PROTACs, their mechanism of action resembles canonical MGs and enhances the intrinsic interaction between DCAF16/DCAF11 and BRD4.

Converting protein ligands to degraders has proved to be a feasible approach, yet we still lack rational chemical design principles to achieve the conversion. To address this challenge, but-2-ene-1,4-dione (‘fumarate’) moiety is identified as a minimal covalent handle to engage E3 ligase RNF126 [50]. When appended to other protein-targeting ligands, the fumarate handle induces the degradation of their corresponding targets. Similarly, a recent study found that attaching an NSD2 ligand or XIAP ligand with a primary alkylamine group yields reversible covalent degraders that lead to NSD2 or XIAP degradation through the E3 ligase FBXO22 [51,52]. While additional efforts are required to evaluate the potency and specificity of covalent MG degraders, these studies provide proof-of-concept evidence to rationally design monovalent degraders from a ligand.

The ligand modification approach has expedited rational MG discovery, yet we are confined to available E3 or protein ligands. On the contrary, de novo MG discovery offers opportunities to explore a much broader chemical space. Simonetta and colleagues reported an example of de novo discovery of MGs through high-throughput screening (Figure 1b) [53]. In normal cells, β-catenin is doubly phosphorated and subsequently degraded by the E3 ligase β-TrCP. However, in colorectal cancers, the phosphorylated sites are often mutated, thus impairing the ability of β-catenin to bind to β-TrCP. To identify compounds that can restore their binding, the authors performed high-throughput screen (HTS) using a quantitative β-catenin:β-TrCP binding assay. This assay provides a large dynamic window to screen small molecule enhancers, since β-TrCP and hypo-phosphorylated β-catenin still retain an interface that has an affinity of ∼600 nM (Figure 2d). Through this campaign, compounds such as NRX-103094 and NRX-252114 were developed with an activity in enhancing the β-TrCP:β-catenin binding affinity by 2–3 orders of magnitude (Table 1). Remarkably, the enhancer compounds do not exhibit detectable affinity to either β-catenin or β-TrCP and only bind the two proteins in the context of the ternary complex. The relatively strong basal interaction between β-TrCP and monophosphorylated β-catenin provides a significant amount of binding energy; therefore demanding less contribution from the MG:β-TrCP and MG:β-catenin interactions. This study further underscores the importance of the basal interaction between E3 ligase and target in degrading non-ligandable proteins. Of note, a major improvement of the hit compounds involves addition of a chemical moiety that remodels the conformation of β-catenin peptide. This highlights the plasticity of the E3-target interface, enabling the optimization of a glue.

MG degraders rely on the cellular ubiquitination and degradation machinery to degrade their target proteins. This feature is exploited in cell-based phenotypic screens to discover MG without prior knowledge of a ligase or a target protein. Specifically, neddylation is an essential step to activate CRLs. To identify CRL-dependent MGs, Mayor-Ruiz et al. mutated UBE2M, an E2 for NEDD8 conjugation, and created cells with low levels of neddylation [9]. About 2000 cytotoxic compounds were screened in both wild-type and UBE2M-deficient cells. Compounds that were less effective in UBE2M-mutated cells than wild-type cells were found and further elucidated to engage DDB1 and CDK12 to degrade cyclin K. This method was later used to screen a covalent ligand library and identified an NFKB1 degrader [54]. Other cell-based methods, such as dynamic tracing of E3 ligase abundance and cell morphology profiling, have been developed and successfully implemented to identify RBM39 and GSPT2 degraders [22,55].

Our definition of canonical MG mainly refers to chemically simple compounds that are in compliance with Lipinski’s rule of five [56]. Their contribution to the overall binding energy that stabilizes the target:MG:E3 ligase ternary complex is limited in comparison to PROTACs. Therefore, the intrinsic affinity between target and E3 ligase weighs more in the ternary complex (Table 1 and Figure 3g). As the field continues to grow, more chemically complex MGs have been identified, such as the covalent MG degraders against DCAF16, RNF126, and FBXO22, and the linker-less PROTAC 13–7 bridging BRD9 and VHL. We consider both covalent MGs and linker-less PROTAC as non-canonical MGs, as they can bind to each individual protein partner and do not fit the definition of canonical glue. The covalent bond or the larger size of (linker-less) PROTAC boosts the energetic contribution from the compounds, which in return alleviates the requirement of the target:E3 ligase intrinsic interactions for ternary complex formation. In these non-canonical MG scenarios, the intrinsic affinity of the target and E3 might become optional. For instance, no detectable affinity has been reported for the RNF126:substrate, FBXO22:substrate, and BRD9:VHL complexes, while a complementary interface between BRD4 and DCAF16 is still needed for the BRD4 covalent glues to function [45]. Interestingly, such a complementary interface provides a structural scaffold to facilitate the action of the covalent MGs, which holds the promise to reduce the non-specific reactivity with other proteins. For other modalities that are even larger in size, such as cyclic peptides, helicons, or viral proteins that induce protein proximities (Figure 3d–g), no intrinsic protein–protein interface is absolutely required as these proteinaceous molecules might be large and complex enough to simultaneously engage both proteins with high affinities. Overall, canonical MGs, non-canonical MGs, bifunctional PROTACs, cyclic peptides, and proteinaceous molecules constitute a continuous spectrum of proximity inducers, which are characterized by the tradeoff between their molecular size and the requirement of the basal affinity of the two proteins they act on (Figure 3g).