Next-generation sequencing (NGS) provides a molecular rationale to inform prognostic stratification and to guide personalized treatment in cancer patients. Here, we determined the prognostic and predictive value of actionable mutated genes in metastatic colorectal cancer (mCRC). Among a total of 294 mCRC tumors examined by targeted NGS, 200 of them derived from patients treated with first-line chemotherapy plus/minus monoclonal antibodies were included in prognostic analyses. Discriminative performance was assessed by time-dependent estimates of the area under the curve (AUC). The most recurrently mutated genes were TP53 (64%), KRAS or NRAS (49%), PIK3CA (15%), SMAD4 (14%), BRAF (13%), and FBXW7 (9.5%). Mutations in FBXW7 correlated with worse OS rates (p = 0.036; HR, 2.24) independently of clinical factors. Concurrent mutations in TP53 and FBXW7 were associated with increased risk of death (p = 0.02; HR, 3.31) as well as double-mutated TP53 and SMAD4 (p = 0.03; HR, 2.91). Analysis of the MSK-IMPACT mCRC cohort (N = 1095 patients) confirmed the same prognostic trend for the previously identified mutated genes. Addition of the mutational status of these genes upon clinical factors resulted in a time-dependent AUC of 87%. Gene set enrichment analysis revealed specific molecular pathways associated with SMAD4 and FBXW7 mutations in TP53-defficient tumors. Conclusively, SMAD4 and FBXW7 mutations in TP53-altered tumors were predictive of a negative prognostic outcome in mCRC patients treated with first-line regimens.
Keywords: left–sided colorectal cancer; machine learning; metastatic colorectal cancer; mutational profiling; next–generation sequencing; prognosis.