RBX1 (108 aa) is a zinc-coordinating E3 ubiquitin ligase component central to Cullin-RING ligase (CRL) complexes. It coordinates 3 Zn ions through 12 residues and mediates interactions with E2 enzymes, Cullin scaffolds, GLMN, and NEDD8. The N-terminal IDR (res 1-34) is disordered; the C-terminal RING-H2 domain is the structured druggable core.

Hypothesis: Designing binders against multiple functional surfaces, including the unconventional Cullin scaffold face, increases hit rates and enables novel CRL disruption modes.

Epitope Analysis: 10 PDB complexes (1LDJ, 1LDK, 1U6G, 2HYE, 2LGV, 3DPL, 3DQV, 3RTR, 4F52, 4P5O) were analyzed to map all PPI surfaces. Four hotspot strategies were designed:

• Hinge + E2/GLMN face (41 res): block E2 conjugation
• Cullin scaffold face (27 res): disrupt CRL assembly [novel]
• Extended E2 + Free surface (43 res): maximize contact area
• Hinge-only (8 res): focused NEDD8 blocker

Backbone Generation: RFdiffusion3 in de novo binder design mode. Only the RBX1 chain was provided in the contig (B20-104). Binder generated entirely de novo with hotspots directing contacts to each epitope. gamma=0.5, step_scale=1. Four backbones produced (121-146 aa).

Sequence Design: ProteinMPNN (576 seq, fixed target chain) + ColabDesign RSO (98 seq). RSO starts from the RFdiff3 backbone (a never-before-seen scaffold) and performs sequence design via gradient descent through AF2, analogous to ProteinMPNN but using differentiable optimization instead of sampling. RSO jointly optimizes iPTM, pLDDT, and RMSD. RSO significantly outperformed ProteinMPNN on identical backbones.

Validation (7 filters):

1. AF2 self-consistency: backbone RMSD between designed pose and AF2 re-prediction. <2A=confirmed, >5A=fails.
2. AF3 Server with 3 Zn ions
3. Rosetta InterfaceAnalyzer (dG, SC, BSA)
4. PLIP atomic contacts
5. CamSol solubility (>1.0)
6. PRODIGY binding affinity (calibrated Kd)
7. Sequence filters (instability, GRAVY, charge, poly-AA)

Key finding: iPTM alone is insufficient. design_25_0 (iPTM 0.82, RMSD 5.1A) was deprioritized because AF2 predicts a different fold. PRODIGY reranked significantly: design_38_3 (iPTM 0.80, Kd 518nM) dropped while design_26_2 (iPTM 0.77, Kd 8.5nM) rose.

Results: 816 designs generated, 25 passed all wet-lab filters. Top: design_3_1 (iPTM 0.70, Kd 15nM, RMSD 1.9A), design_26_2 (tightest Kd 8.5nM). All 120aa, helical, soluble, stable.

Fasta, Structure & Properties https://drive.google.com/drive/folders/1PEdhtll8wmMigRGaEvqN1AHLtfALHL64?usp=drive_link