CD38 is the dominant NAD+-consuming enzyme in aged mammalian tissues. Circulating NAD+ declines ~50% between ages 40–60, and CD38 deletion or pharmacological inhibition restores NAD+ and extends murine healthspan. Existing CD38 inhibitors (apigenin, 78c, isatuximab) are either non-selective flavonoids or antibodies targeting the extracellular epitope rather than the active site. This plan executes a structure-guided de novo binder design campaign: (1) use PDB 2I65 catalytic-cleft geometry (8 hotspot residues now profiled) as the conditioning target for RFdiffusion backbone generation; (2) run ProteinMPNN to sequence-design the scaffold; (3) validate with AF2-Multimer as the independent oracle; (4) filter by ipTM > 0.7, pLDDT > 80, predicted interface dG < -10 kcal/mol; (5) assess novelty with Foldseek against PDB; (6) score ESM-2 pseudo-log-likelihood to eliminate low-fitness sequences. Success criterion: ≥1 candidate with AF2-Multimer ipTM > 0.75 that places a predicted contact on ≥4 of the 8 hotspot residues and outperforms the isatuximab epitope overlap benchmark on active-site coverage. Planned outputs: RFdiffusion backbone PDB, ProteinMPNN FASTA ensemble, AF2-Multimer complex PDB with pTM/ipTM/PAE, Foldseek novelty report, ESM-2 perplexity ranking table.
Details
- disease
- aging / nad+ decline
- target_ref
- UniProt:P28907
- primary_endpoint
- ≥1 designed sequence with ESM-2 pseudo-perplexity ≤ 15 AND ESMFold pLDDT ≥ 75 over the interface region (residues contacting hotspot within 8 Å)
- identification_strategy
- in_silico_KO
Raw fields (3)
- assay_spec
ProteinMPNN sequence design on RFdiffusion-generated backbones conditioned on NAMPT catalytic-cleft hotspots (Q168/K228/V274/A306/I309/G383) from PDB:2GVJ chain A; 8 backbone trajectories × 5 MPNN sequences each; ESM-2 perplexity filter (pseudo-perplexity ≤ 15); top-5 sequences forward-folded with ESMFold for self-consistency pLDDT ≥ 75; 8 RFdiffusion-generated backbone PDB files conditioned on 2GVJ hotspot residues (planned); 40 ProteinMPNN-designed sequences in FASTA format with per-sequence log-likelihood scores (planned); ESM-2 pseudo-perplexity scores for all 40 sequences; filtered shortlist ≤15 ppl (planned); ESMFold predicted structures for top-5 sequences with pLDDT per-residue (planned); Lead candidate protein_design substrate: FASTA + ESMFold PDB + pLDDT + ESM-2 ppl + hotspot contact summary (planned)
- hypothesis
A de novo protein binder designed against the CD38 catalytic cleft (hotspot residues C119, K121, D155, D156, D179, P184, A200, N229 from PDB 2I65) will competitively inhibit CD38 NAD+ hydrolase activity, thereby elevating intracellular NAD+ in aged tissues and extending healthspan. A tight-binding designed protein (target KD < 100 nM) blocking the adenosine-binding pocket should suppress NAD+ consumption without the selectivity liabilities of small-molecule CD38 inhibitors.
- kill_criteria
Abort or revise if: RFdiffusion hotspot conditioning on a homodimer interface may generate binders that clash with the symmetric chain B — must mask chain B or use monomer of 2GVJ; ESMFold self-consistency is a weak oracle; AF2-Multimer validation is the required next gate after this tick; FK866-competitive binders may not inhibit eNAMPT (extracellular form) — functional differentiation needed in wet-lab triage