Beyond Histology: A Dual-Cohort Genomic Analysis of 2901 Endometrial Carcinomas Reveals Class-Level Mismatch Repair Effects and Refines Molecular Classification.

Background: Endometrial carcinoma (EC) is now classified primarily by molecular subtype-POLE-ultramutated, mismatch repair-deficient (dMMR), TP53-mutant/copy-number-high (CNH), and "no specific molecular profile" (NSMP)-a framework that has reshaped prognostic counseling and adjuvant therapy decisions. Yet several practically important questions remain insufficiently addressed in real-world cohorts: whether all four mismatch repair genes confer an equivalent favorable prognosis, whether all POLE alterations carry the same survival benefit or only specific pathogenic variants, and whether molecular subtypes retain prognostic value after adjustment for histology and tumor burden. Methods: We addressed these questions in 2901 patients pooled from the MSK-IMPACT 50K Clinical Sequencing Cohort (n = 2372; discovery) and the TCGA UCEC PanCancer Atlas (n = 529; validation)-the largest dual-cohort genomic analysis of EC reported to date. We performed individual MMR gene and combined dMMR survival stratification, multivariable Cox regression adjusted for age, histology, and sample type, and a pathogenicity-aware sensitivity analysis for POLE variants, with tumor mutational burden (TMB) compared across subgroups. Results: Across both cohorts, all four MMR gene-mutant subgroups (MLH1, MSH2, MSH6, PMS2) conferred equivalently favorable overall survival (OS) (six-group log-rank p = 7.66 × 10-12 in discovery; p = 6.78 × 10-3 in validation), confirming dMMR as a class-level prognostic designation independent of which MMR gene is altered. Multivariable Cox regression demonstrated that POLE-ultramutated status retained an independent favorable effect (HR = 0.62, p = 0.038 in MSK; HR = 0.35, p = 0.028 in TCGA) after adjustment for age, histology, and sample type, while the favorable dMMR effect was largely accounted for by histologic context. Critically, a pathogenicity-aware sensitivity analysis revealed that the exceptional survival of the POLE subgroup is confined to canonical exonuclease-domain hotspot mutations (event rate 0.9% in MSK), whereas POLE variants of uncertain significance behave indistinguishably from NSMP-like tumors. Consistent with this finding, TMB was markedly elevated in canonical pathogenic POLE cases (median 138.7 mut/Mb in MSK; 247.4 in TCGA) but not in POLE-VUS-only cases (median 29.0 and 15.0, respectively; p < 0.001 between groups in both cohorts), confirming that the ultramutator phenotype is confined to canonical pathogenic POLE variants. We additionally characterize Uterine Clear Cell Carcinoma as a distinct histologic entity (n = 73; 3.0%) and report the POLE + TP53 co-mutant group (n = 90; 3.8%). Conclusions: These findings refine the molecular classification of EC in clinically meaningful ways: they support class-level immunotherapy eligibility based on dMMR status regardless of the specific MMR gene altered, demonstrate that POLE-ultramutated classification requires variant-level pathogenicity assessment, and identify TP53-mutant/CNH patients as the population with the most urgent unmet therapeutic need.