RBX1 (RING Box Protein 1) is an 108 amino acid RING-H2 domain protein and functions as the catalytic subunit of Cullin-RING E3 ubiquitin ligase complexes, recruiting E2 enzymes and mediating ubiquitin transfer. RBX1 interfaces with Cullin and E2 enzymes, representing distinct therapeutic targeting opportunities via disrupting the assembly and catalytic activity of CRL complexes and downstream ubiquitin transfer. We have chosen to target the Cullin-RBX1 interface as disruption of this interaction prevents the assembly of the entire CRL complex upstream of E2 recruitment which offers a more complete inhibition than targeting the E2 binding site alone.

Methods Interface and hotspot selection: Upon inspection of the CUL1–RBX1–SKP1–F-boxSKP2 complex (PDB: 1LDJ) and identification of the RBX1–CUL1 interface, residues PHE81, SER85, and LYS89 were selected as candidate hotspots. These residues were identified based on direct contacts with Cullin (≤4 Å), relative SASA confirming solvent exposure and B-factor analysis indicating structural order and limited flexibility. PHE81 was selected as the primary hydrophobic anchor given its bulky aromatic sidechain and restricted rotameric flexibility, making it well-suited for stable packing interactions. SER85 was included based on direct experimental evidence of Cullin contact, while LYS89 contributes electrostatic character consistent with the mixed hydrophobic–electrostatic nature of the native interface, further supported by APBS electrostatic potential calculations. Notably, all three hotspots are positioned on one face of the α2 helix of the RING domain, with side chains aligned in a common direction, forming a coherent interface with Cullin consistent with a helical mimicry strategy.

Target preparation: Taking the target structure (PDB: 2LGV), the intrinsically disordered N-terminal region of RBX1 (res 1-39) was removed prior to design. The structured RING-H2 domain (res 40-108) retaining all three zinc ions was used as the design target. Native zinc ions were retained to preserve RING domain geometry during Rosetta scoring steps.

Computational design: De novo binders were generated using BindCraft (Pacesa et al., 2024), employing AlphaFold2-Multimer–guided design with iterative ProteinMPNN sequence optimisation (default_4stage_multimer_mpnn protocol). We directly targeted the Cullin interface with hotspots A81,A85,A89 and binder lengths of 60-80 residues with a target of 20 accepted designs. All submitted sequences are de novo with maximum sequence identity of 66% to UniRef50, satisfying the minimum edit distance requirement.

Results Filtering and ranking: Accepted designs passed BindCraft default filters that required i_pTM ≥ 0.60, zero relaxed clashes, surface hydrophobicity ≤ 0.40 and ≤ 6 unsatisfied interface hydrogen bonds. Designs were then ranked by Average_i_pTM. Top designs were independently validated using Boltz2 complex prediction (Tamarind Bio) with the trimmed RBX1 construct (res 40-108) and all three zinc ions explicitly modelled, providing the most physically accurate complex representation.

The top design was validated using Boltz-2 with the trimmed RBX1 RING domain and all three zinc ions explicitly modelled, achieving confidence score of 0.95, ipTM of 0.96, and ligand/zinc ipTM of 0.984, indicating high confidence in both protein-protein interface geometry and zinc coordination sphere integrity. Critically, the complex interface predicted distance error (ipde) of 0.39 confirms well-defined interface geometry at near atomic resolution. Boltz-2 recapitulated the primary binding mode with a sub-angstrom target agreement (all atom RMSD = 0.51Å) to the Bindcraft model. A modest binder displacement (rms_cur = 1.83 Å) was observed likely representing a physically-informed refinement of the interface to accommodate for explicit zinc coordination geometries.