AI-driven structural analysis clarifies novel TNFRSF17 variants of uncertain significance as drivers of resistance to BCMA-targeted immunotherapies in multiple myeloma

Background The advent of B-cell maturation antigen (BCMA)-targeted immunotherapies has markedly transformed the prognosis for patients with relapsed/refractory multiple myeloma (r/r MM). However, the emergence of therapeutic resistance, frequently driven by tumor-intrinsic alterations in the TNFRSF17 (tumor necrosis factor receptor superfamily member 17) gene encoding BCMA, remains a critical clinical challenge. While the event of complete antigen loss via biallelic gene deletion is well-described, the functional impact of novel, non-truncating variants of uncertain significance (VUS) discovered through comprehensive sequencing is often unclear, complicating therapeutic decisions. Methods Whole-genome sequencing (WGS) was performed on purified CD138 + plasma cells from two patients with r/r MM who experienced disease progression after treatment with the anti-BCMA chimeric antigen receptor (CAR) T-cell therapy idecabtagene vicleucel (ide-cel) or ciltacabtagene autoleucel (cilta-cel). We developed a novel AI-based computational workflow to assess the impact of identified TNFRSF17 mutations on protein structure response to treatment across a panel of clinically relevant BCMA-directed agents. Our method combines AI-based protein-complex structure prediction with binding free energy calculations. Results WGS revealed a monoallelic deletion of 16p13, encompassing the TNFRSF17 gene locus, in both patients, accompanied by distinct subclones on the remaining allele: three VUS (p.Cys37Tyr, p.Cys24_Cys28del, and p.Pro23delins9) in patient #1, and the p.Cys8Trp VUS in patient #2. All detected variants in the protein sequence are localized within the extracellular, epitope-bearing domain of BCMA. Our AI-driven structural analysis predicted that these mutations induce significant conformational changes, primarily by disrupting the stabilizing disulfide bond network. Quantitative energetic calculations demonstrated that these variants would severely impair or completely abrogate binding of specific therapeutic agents, including ide-cel and belantamab cilta-cel, consistent with the patients’ clinical course of relapse. Notably, our analysis showed differential effects across various agents, with the binding affinity of some therapeutics predicted to remain robust to the structural changes. Conclusion The integration of WGS into diagnostics with AI-driven structural and energetic modeling provides a powerful framework for the functional interpretation of VUS in TNFRSF17 . This approach can rapidly elucidate patient-specific, epitope-dependent resistance mechanisms and has the potential to guide the rational sequencing of BCMA-targeted therapies. By transforming VUS from a diagnostic challenge into an actionable clinical biomarker, this methodology represents a critical step toward precision immunotherapy in MM.