How can AlphaFold 3 ternary complex geometry predictions be used to rationally design molecular glue degraders that recruit non-canonical E3 ligases to previously undruggable neo-substrates, and what

AlphaFold 3 (AF3) ternary complex predictions enable the rational design of molecular glue degraders (MGDs) by providing high-resolution geometric models of ligand-induced protein-protein interactions (PPIs), identifying novel binding pockets at interfaces, and facilitating stereochemical optimization (Direct, High; PMID: 38718835, PMID: 41831109). Experimental validation connects these predicted spatial constraints to degradation through frameworks that quantify cooperative binding via surface plasmon resonance (SPR) or evaluation of "productive" ubiquitination based on precise distance mapping between target lysines and E3-resident ubiquitin (Direct, High; PMID: 37443112, PMID: 35101445).

Rational Design Using AlphaFold 3 Geometry Predictions

AF3 uniquely models the simultaneous arrangement of proteins, ligands, and ions, allowing for the direct prediction of complete ternary complexes without prior knowledge of binary binding modes (Direct, High; PMID: 38718835, PMID: 41831109).

• Stereochemical Selection: AF3 can discriminate between active and inactive stereoisomers. For the G3BP2 degrader MRT-5702, AF3 confidence metrics (ipTM and pLDDT) and volume overlap analysis were used to correctly assign the (S,S)-configuration to the active degrader (Direct, High; PMID: 41559416).
• Predicting Selectivity: Computational docking using AF3 and Rosetta helps rationalize how small changes in linker length or substitution can exploit isoform-specific residues distal to the binding site, such as the preference of ZXH-3-26 for BRD4 over other BET family members (Derived, Medium; PMID: 29892083, PMID: 40001507).
• Confidence Metrics: The reliability of these predictions is assessed using "MG Chain ipTM" (interface predicted TM-score) and "Avg. pLDDT of MG," with thresholds of >0.68 and >70, respectively, serving as indicators of high-quality model recovery (Direct, High; PMID: 41160881).

Recruitment of Non-Canonical E3 Ligases to Undruggable Neo-Substrates

Structure-based modeling facilitates the transition of non-canonical E3 ligases into targeted protein degradation (TPD) platforms.

• DCAF10 and Ac-Gly/MO Degrons: AF3 modeled the direct engagement of acetylated N-terminal glycine (Ac-Gly) residues of Src-family kinases (SFKs) into a deep β-propeller tunnel in DCAF10, establishing a novel N-degron pathway (Direct, High; PMID: 41484149).
• KBTBD4 and Molecular Mimicry: The E3 ligase KBTBD4 is co-opted by the small molecule UM171, which mimics the structural features of neomorphic cancer mutations (e.g., R313PRR) to recruit HDAC1/2 for degradation (Direct, High; PMID: 40175372).
• SKP2 (CRL1): Non-covalent PROTACs have been designed using SKP2 recruiters, expanding TPD beyond CRL2/CRL4 complexes to target proteins like BRD4 and AR (Direct, High; PMID: 41637562).
• DDB1-Direct Recruitment: Hydrophobic gluing moieties can be added to parental inhibitors to recruit the DDB1 adaptor directly, as seen in the conversion of CDK12 binders into potent Cyclin K degraders (Direct, High; PMID: 40626960).

Experimental Validation Frameworks

Validation connects the three-body spatial constraints predicted in silico to measurable biochemical and cellular outcomes.

• Productive Complex Modeling: A structure-based workflow classifies ternary complexes as "productive" if at least one surface-exposed lysine on the target is within 16 Å of the C-terminus of E3-loaded ubiquitin. This percentage of productive models correlates with measured ubiquitination potency (Direct, High; PMID: 35101445).
• Quantifying Cooperativity ($\alpha$): Intrinsic cooperativity is defined as the ratio of binary to ternary binding affinities ($K_{LP}/K_{LPT}$). SPR and spectral shift assays measure this by monitoring binding through the weaker partner (often the target) to ensure an adequate signal-to-noise ratio (Direct, High; PMID: 40243574, PMID: 40944659).
• Buried Surface Area (BSA): Molecular dynamics (MD) simulations relax AF3 models to calculate the BSA of ternary interfaces. For SMARCA2 degraders, BSA shows a strong correlation with experimental binding affinity ($K_{LPT}$) (Direct, High; PMID: 37443112).
• Single-Molecule Monitoring: YaxAB biological nanopores provide label-free monitoring of molecular glue efficacy by discriminating between ternary complexes, binary complexes, and individual proteins based on distinct ionic current blockade signals (Direct, High; PMID: 39482865).

Collectively, AF3-guided design paired with quantitative kinetic and spatial mapping provides a roadmap for the prospective discovery of MGDs against historically intractable targets (Derived, High; PMID: 38718835, PMID: 35101445).

How do specific lysine-to-ubiquitin distance constraints in the modeled CRL4A complex influence the accuracy of predicting target ubiquitination rates?

What are the structural determinants of the CRBN LON-domain interface that allow for the recruitment of G-loop-independent neosubstrates like G3BP2?

Which biophysical methods are most effective for differentiating between intrinsic and apparent cooperativity in ternary systems with weak binary affinities?

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID:41772859 — ** Interface-Adjacent Pockets (IAPs): The GlueFinder framework utilizes AF3 to identify preexisting or weak interfa...*
  Failed: entities — The paper uses AF2Complex, not AlphaFold 3, for identify interface-adjacent pockets (IAPs).
  Possible alternatives (unverified): PMID:41152377 (81% topic match); PMID:41484149 (76% topic match)
• PMID:40326883 — SPR and spectral shift assays measure this by monitoring binding through the weaker partner (often the target) to ensure...
  Failed: conclusion — The paper focuses on Free Energy Perturbation (FEP) calculations and does not discuss monitoring binding through the weaker partner in SPR or spectral shift assays to optimize signal-to-noise.

Hypothesis 1

Stabilizing the CRBN N-terminal belt (residues 48–63) into a rigidified, closed-state conformation via small-molecule 'stapling' will synergistically enhance the degradation efficiency of LON-domain-dependent neosubstrates like G3BP2 by eliminating the entropic penalty of the open-to-closed conformational transition and expanding the productive ubiquitination zone.

Mechanistic rationale

• Apo-CRBN exclusively adopts an open conformation in solution, where the N-terminal LON domain and C-terminal TBD are separated. (Direct, High; PMID: 36378961)
• The closed conformation of CRBN is a prerequisite for neosubstrate recruitment and is normally stabilized by an 'N-terminal belt' (residues 48–63) that becomes ordered only upon ligand binding. (Direct, High; PMID: 36378961)
• Molecular glues like mezigdomide enhance efficacy by contacting both the TBD and the distal LON domain (e.g., residues Phe102 and Phe150), effectively 'stapling' the domains together in the closed state. (Direct, High; PMID: 36378961)
• G3BP2 is a G-loop-independent substrate that utilizes the LON-domain loop (E146–I154) as its primary interface, bypassing the canonical CULT-domain interactions used by zinc-finger substrates. (Direct, High; PMID: 41559416)
• The transition from open to closed CRBN represents a critical mechanistic bottleneck; increasing the population of closed conformers via high-affinity ligands translates directly to improved degradation kinetics. (Derived, Medium; PMID: 36378961)
• AlphaFold 3 structurally models these complexes but tends to favor static closed-state snapshots, potentially underestimating the entropic costs of domain closure for novel ligands. (Derived, Medium; PMID: 38718835, PMID: 41831109)

Predictions

• Ensemble modeling (SILCS-xTAC) will reveal that 'stapled' ligands have lower internal ligand grid free energy (LGFE) variances across predicted PPI conformers compared to monovalent TBD binders. (Indirect, Low; PMID: 41416887)

Study design

In silico design of a focused library of CRBN glues with variable-length 'stapling' arms extending toward the LON domain, followed by AF3-based ternary complex prediction with G3BP2. Efficacy will be validated by measuring binding cooperativity via SPR and cellular G3BP2-HiBiT degradation rates. (Derived, Medium; PMID: 38718835, PMID: 37443112, PMID: 41559416)

Confounders & controls

• Use CRBN N-terminal belt deletion mutants (Δ48–63) as negative controls; these should be resistant to the synergistic effects of 'stapled' glues. (Derived, Low; PMID: 36378961)
• Measure ternary complex formation with 'inactive' enantiomers (e.g., MRT-5702A/C) to control for non-specific interface stabilization. (Direct, High; PMID: 41559416)
• Monitor HIF-1α levels to ensure that high-dose closed-state stabilization does not inadvertently inhibit native VHL-like protein-protein interactions. (Indirect, Low; PMID: 35983982)

Risks/limitations

• AF3 exclusively predicts the closed state of CRBN, which may hide the true entropic barrier for ligands that are poor at inducing closure. (Derived, Medium; PMID: 38718835)
• High-resolution local maps for the CRBN-G3BP2 interface often show conformational flexibility (3.5–4.5 Å), suggesting that 'rigid stapling' may be difficult to achieve in practice. (Derived, Low; PMID: 41559416)

Falsification criteria

• The hypothesis is invalidated if G3BP2 degradation is shown to be insensitive to the ordering of the N-terminal belt. (Derived, Low; PMID: 36378961)

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID: 40326883 — Small molecules engineered with bifunctional binding vectors targeting the CULT Trp-cage and the N-terminal belt residue...
  Failed: entities — The paper studies IKZF2 and IKZF1 but does not contain any mention of G3BP2.
  Possible alternatives (unverified): PMID:35983982 (61% topic match); PMID:40707481 (61% topic match)
• PMID: 36378961 — Small molecules engineered with bifunctional binding vectors targeting the CULT Trp-cage and the N-terminal belt residue...
  Failed: entities — The paper does not mention or study G3BP2; it focuses on Ikaros (IKZF1).
  Possible alternatives (unverified): PMID:35983982 (61% topic match); PMID:40707481 (61% topic match)
• PMID: 35101445 — Stabilizing the closed state will increase the percentage of 'productive' ternary complex models where the target G3BP2 ...
  Failed: entities — The paper studies the CDK family (CDK2, 5, 7, 8, 9, 12, 13) and does not mention G3BP2.
• PMID: 36378961 — AF3 exclusively predicts the closed state of CRBN, which may hide the true entropic barrier for ligands that are poor at...
  Failed: entities — The paper does not mention or evaluate AlphaFold 3 (AF3); it focuses on cryo-EM structures and HDX-MS.
• PMID: 36378961 — If ligands that demonstrate superior closed-state stapling by HDX-MS fail to show improved DC50 values for G3BP2 compare...
  Failed: entities — The paper contains HDX-MS data for CRBN and Ikaros but does not mention G3BP2.
  Possible alternatives (unverified): PMID:29892083 (56% topic match); PMID:35320687 (56% topic match)
• PMID: 41559416 — If ligands that demonstrate superior closed-state stapling by HDX-MS fail to show improved DC50 values for G3BP2 compare...
  Failed: mechanism — The paper does not mention or utilize HDX-MS (Hydrogen-Deuterium Exchange Mass Spectrometry).
  Possible alternatives (unverified): PMID:29892083 (56% topic match); PMID:35320687 (56% topic match)

Hypothesis 2

Small-molecule occupancy of the unoccupied proximal subpocket in the DCAF10 β-propeller tunnel, designed to present an electron-rich π-system mimicking the N-terminal acetyl group, will selectively redirect the CUL4A-DDB1-DCAF10 machinery to degrade non-acetylated, myristoylation-deficient Src by bypassing the requirement for NatA-mediated modification.

Mechanistic rationale

• DCAF10 is a specialized N-recognin that recruits acetylated N-terminal glycine (Ac-Gly) residues into a deep β-propeller tunnel for ubiquitination by the CUL4A complex. (Direct, High; PMID: 41484149)
• Native Src-family kinases (SFKs) like Src are poor substrates for NatA in vitro, showing significantly higher Km values compared to canonical substrates, which may limit their natural degradation via the DCAF10 pathway unless myristoylation is inhibited. (Direct, High; PMID: 41484149)
• AlphaFold 3 structural modeling indicates that while acetylated Lyn and Fyn engage the DCAF10 tunnel deeply, Src residues (N4) lead to a shallower and less stable interface, suggesting an unoccupied space within the tunnel available for chemical expansion. (Derived, Medium; PMID: 41484149)
• Hydrophobic aromatic moieties, particularly those containing double bonds or electron-rich systems, can recruit DDB1-associated ligase components by forming stabilizing interactions within specific subpockets. (Indirect, Low; PMID: 40626960)
• The transition from open to closed E3 ligase conformations is a rate-limiting step in neosubstrate recruitment; a molecular glue that 'staples' the interface can reduce the entropic penalty and enhance degradation kinetics. (Indirect, Low; PMID: 36378961)

Predictions

• AlphaFold 3 ternary modeling of the DCAF10-Glue-Src(G2) complex will predict a high MG Chain ipTM (> 0.68) and low PAE at the interface, indicating a stable drug-induced proximity. (Derived, Medium; PMID: 41160881, PMID: 38718835)
• Treatment with the designed molecular glue will induce a dose-dependent BRET signal in cells expressing NanoLuc-Src(G2) and Halo-DCAF10, confirming ternary complex formation without the need for pre-acetylation. (Derived, Medium; PMID: 41559416)
• In vitro ubiquitination assays using reconstituted CUL4A-DDB1-DCAF10 will demonstrate efficient ubiquitin transfer to Src(G2) lysines positioned within 16 Å of the E2-Ub complex only in the presence of the molecular glue. (Derived, Medium; PMID: 35101445, PMID: 41484149)

Study design

Iterative in silico library generation of GBB-derived imidazopyridine derivatives with hydrophobic 'stapling' extensions followed by AF3-based screening against the DCAF10 tunnel. Lead compounds will be tested in HeLa cells for degradation of endogenous Src and a HiBiT-Src(G2) reporter under NMT inhibition, with SPR-based quantification of intrinsic cooperativity (α) through the weaker partner (Src). (Derived, Medium; PMID: 39499896, PMID: 41831109, PMID: 37443112, PMID: 40243574)

Confounders & controls

• Use DCAF10 pocket mutants (F172G, K257A, I475G) as negative controls; these should be resistant to glue-induced Src recruitment. (Derived, Low; PMID: 41484149)
• Include an N-methylated 'inactive' variant of the glue to control for non-specific protein-protein interface effects. (Indirect, Low; PMID: 36007246)
• Cotreat with ZYG11B/ZER1 siRNAs to prevent redundancy through the Gly/N-degron pathway and isolate the DCAF10-dependent signal. (Direct, High; PMID: 41484149)

Risks/limitations

• AF3 exclusively models the closed state of E3 ligases, which may overestimate the initial binding stability of glues that are poor at inducing the open-to-closed transition. (Derived, Medium; PMID: 38718835, PMID: 36378961)
• The endogenous level of DCAF10 is relatively low in some tissues, which may require overexpression of the ligase components for biochemical detection. (Direct, High; PMID: 41484149)

Falsification criteria

• The hypothesis is falsified if the molecular glue induces Src binding but results in 'unproductive' ternary conformers where target lysines are > 16 Å from the ubiquitin donor. (Derived, Medium; PMID: 35101445)
• If cellular degradation of Src by the glue is entirely dependent on the presence of NatA-mediated acetylation (rescued by NatA knockdown), then the glue fails to bypass the modification requirement. (Direct, High; PMID: 41484149)

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID: 38718835 — AlphaFold 3 structural modeling indicates that while acetylated Lyn and Fyn engage the DCAF10 tunnel deeply, Src residue...
  Failed: conclusion — The paper describes the AlphaFold 3 method generally but does not contain any data or discussion regarding DCAF10, Lyn, Fyn, or Src.
• PMID: 39482865 — Treatment with the designed molecular glue will induce a dose-dependent BRET signal in cells expressing NanoLuc-Src(G2) ...
  Failed: conclusion,entities — This paper describes label-free nanopore sensing of the mTOR-FRB/FKBP interaction and does not mention BRET signals, NanoLuc, or DCAF10.