Neoantigens are critical targets for cancer immunotherapy, yet the relationship between experimentally validated neoantigen burden and antigen processing machinery (APM) expression in determining clinical outcomes remains unclear. We mapped CEDAR-annotated neoantigens (CENs) onto mutation data from 43,980 patients across 14 cancer types using cBioPortal. APM gene expression was correlated with survival outcomes across 13 cohorts. Machine learning approaches (elastic net stability selection, random survival forest, univariable Cox regression) identified prognostically important APM genes across 11 cohorts. Findings were validated in the IMvigor210 immunotherapy trial (n=348 metastatic urothelial carcinoma patients) and single-cell RNA-sequencing data (GSE161529; n=29 breast cancers). Overall, 40.4% of patients harbored at least one CEN, with high prevalence in pancreatic (>75%) and skin cancers (>70%). CENs predominantly arose from driver oncogenes including PIK3CA, KRAS, BRAF, TP53, and EGFR. High APM expression was associated with improved survival, particularly in CEN-positive tumors. Machine learning identified immunoproteasome components (PSME1, PSMB8, PSMB9, PSMB10) as the dominant prognostic contributors within the 12-gene APM signature. A simplified 4-gene immunoproteasome score performed equivalently to the full APM score in leave-one-cohort-out cross-validation (median C-index 0.545 vs 0.545; p=0.31). In IMvigor210, immunoproteasome-high patients achieved a 3.2-fold higher response rate to atezolizumab (19.8% vs 6.2%; p=0.010). Single-cell analysis confirmed that tumor-intrinsic immunoproteasome expression correlated with increased CD8+ T cell infiltration (p=0.0014) and total immune fraction (p=0.0002). The 4-gene immunoproteasome signature demonstrates robust prognostic and predictive value across bulk sequencing, clinical trial, and single-cell platforms, warranting prospective validation as an immunotherapy biomarker.
Lai et al. (Sat,) studied this question.