Functional assessment of polymerase proofreading variants to define sensitivity to immune checkpoint blockade (ICB) pan-cancer.

2645 Background: Cancers with inactivating alterations in DNA proofreading polymerases POLE and POLD1 (POLd) are ultra-hypermutated and benefit from ICB. Most mutations in DNA polymerases are synonymous; discerning pathogenic (P) vs variants of unknown significance (VUS) remains challenging. A strategy to define functional mutations in POL with mutational signature deconvolution and AI- protein conformational modeling could better predict ICB benefit. Methods: Solid tumors from patients at MSK and DFCI were selected with DNA panel sequencing datasets to enrich for somatic POL P/likely-P(LP) variants or VUS. Co-mismatch-repair deficiency (MMRd) was defined with MSIsensor and IHC. Tumors were assigned as POL high (>50%), intermediate ([int] 1-49%), or null (0%) based on the proportion of single-base-substitutions (SBS) aligning with POLd COSMIC signatures via the mutational-patterns algorithm. POLE/D1 VUS were functionally evaluated with Swiss-PdbViewer and AlphaFold-derived AI models to infer proofreading efficiency impact. Patient overall survival (OS) from stage IV diagnosis and progression-free survival from ICB start (PFS-ICB) were evaluated with multivariable (MV) Cox regression including MMRd status. Results: POL variants were found in 546 patients across 18 diverse cancers. 84% were P/LP POL alterations, 16% were POL VUS, and 36% overall had concomitant MMRd. Tumors with P/LP variants had high (69%), int (23%), or null (8%) POLd signature intensities, while 62% of tumors with VUS had int POLd signatures. For the subgroup of patients with POL P/LP variants, OS from diagnosis significantly differed based on POLd signature intensity: not reached [NR], 94.5 months [mo], and 32.9 mo for POLd signature high, int, and null, respectively. In MV analysis of POL P/LP tumors, patients with POL signature-null tumors had significantly worse OS than POL-int (HR 0.35, 95%CI 0.16-0.80; p=0.01) and POL-high (HR 0.18, 95%CI 0.08-0.42; p<0.001) patients. POLd signature intensity predicted ICB benefit. Patients with POL P/LP tumors with high and int signatures exhibited favorable PFS-ICB (NR in both) versus POL P/LP–null-signature tumors (12.7 months). In MV model including MMR status, patients with POL-null SBS exhibited significantly worse PFS-ICB versus pts with combined POL signature high or int (HR 0.31, 95% CI 0.10-0.98; p=0.047). Study findings were consistent after adjustment for cancer type. Tumors with POL VUS predicted to be inactivating based on mutational signature deconvolution combined with Swiss-PdbViewer and AlphaFold AI models demonstrated similar OS/PFS to patients with tumors with P/LP variants and high/int signature scores. Conclusions: A strategy combining mutational signature deconvolution with AI proteomic functional assessment predicts survival and ICB benefit in tumors with functional DNA polymerases POLE and POLD1 mutations.