A RFdiffusion-AA–designed anti-GPNMB(PKD-domain) × IL-15 × anti-CD16a TriKE bypasses glembatumumab's internalization ...

IF a RFdiffusion-AA–designed de novo protein binder targeting the PKD (polycystic kidney disease) domain of GPNMB — an epitope structurally distinct from and distal to the glembatumumab-binding N-terminal region and predicted to be less sterically occluded by N-glycans — is formatted as a Trispecific Killer cell Engager (TriKE) incorporating an IL-15 crosslinker between the anti-GPNMB(PKD) arm and an anti-CD16a (FcγRIIIa) arm, and is administered intravenously (3 mg/kg, bi-weekly) to 24-month-old male C57BL/6 mice with established vascular senescent cell burden,

THEN a ≥40% reduction in GPNMB⁺/p16INK4a⁺ senescent vascular endothelial cells (flow cytometry, aortic digests; immunofluorescence, en face preparations), a ≥30% decrease in plasma SASP cytokines (IL-6, MMP-3, PAI-1 by Luminex), measurable improvement in aortic vascular compliance (pulse wave velocity by Doppler ultrasound), and partial restoration of intra-aortic NK cell cytotoxic function (CD107a degranulation assay on tissue-resident NK cells) will be observed at 8 weeks post-initiation of treatment,

BECAUSE the following causal chain operates:

1. GPNMB is significantly and selectively upregulated on senescent vascular endothelial cells and lesional leukocytes relative to normal vascular tissue, driven by lysosomal stress and SASP signaling, rendering it a viable cell-surface handle for immune-mediated clearance (GPNMB validated as a senescent vascular seno-antigen)[https://pmc.ncbi.nlm.nih.gov/articles/PMC9937554/].

2. The GPNMB extracellular domain contains a structurally defined PKD domain whose surface topology — particularly on the membrane-proximal face — is predicted by AlphaFold2 structural analysis to be less occluded by the clustered N-glycosylation sites concentrated in the N-terminal heparin-binding and RGD regions targeted by glembatumumab; this creates an accessible, protein-surface epitope for RFdiffusion-AA–designed binders that explicitly model glycan shielding during backbone generation (RFdiffusion-AA enables all-atom modeling of non-protein interactions during binder generation)[https://doi.org/10.1038/s41586-023-06415-8].

3. RFdiffusion backbone generation conditioned on the PKD domain surface, followed by ProteinMPNN sequence optimization, produces diverse binder libraries with nanomolar-to-picomolar affinities, with experimental success rates of 10–20% for challenging targets after AlphaFold2/3 self-consistency filtering (RMSD <2.0 Å, pLDDT >80, interface PAE <5 Å) (RFdiffusion achieves picomolar affinities and functional cell killing in de novo designs)[https://pmc.ncbi.nlm.nih.gov/articles/PMC10983868/] (AI-based antibody-antigen complex prediction success rates ~10–13%)[https://pmc.ncbi.nlm.nih.gov/articles/PMC12360200/].

4. The validated anti-GPNMB(PKD) binder arm is integrated into a TriKE format with an IL-15 crosslinker domain positioned between the GPNMB-targeting arm and an anti-CD16a scFv; the tethered IL-15 drives prol...

SENS category: GlycoSENS

Key references: • doi.org/10.1038/s41586-023-06415-8]. • doi.org/10.1126/science.add2187]. • doi.org/10.1038/s41586-021-03819-2].