Development and Validation of the Predictive and Prognostic ChemoResist Signature in Resected Pancreatic Ductal Adenocarcinoma

OBJECTIVE: To develop and validate a signature to precisely predict prognosis in pancreatic ductal adenocarcinoma (PDAC) undergoing resection and adjuvant chemotherapy. BACKGROUND: PDAC is largely heterogeneous and responds discrepantly to treatment. METHODS: A total of 551 consecutive patients with PDAC from 3 different cohorts of tertiary centers were initially enrolled. Genetic events of the 4 most commonly mutated genes in PDAC and expressions of 12 PI3K/AKT/mammalian target of rapamycin (mTOR) pathway markers were examined. A 9-feature signature for the prediction of chemotherapy benefits was constructed in the training cohort using the least absolute shrinkage and selection operator Cox regression model and validated in 2 independent cohorts. RESULTS: Utilizing the least absolute shrinkage and selection operator model, a predictive and prognostic signature, named ChemoResist, was established based on KRAS single nucleotide variant (SNV), phosphatase and tensin homologue (PTEN), and mTOR expressions, and 6 clinicopathologic features. Significant differences in survival were observed between high and low-ChemoResist patients receiving chemotherapy in both the training [median overall survival (OS), 17 vs 42 months, P < 0.001; median disease-free survival (DFS), 10 vs 23 months, P < 0.001] and validation cohorts (median OS, 18 vs 35 months, P = 0.034; median DFS, 11 vs 20 months, P = 0.028). The ChemoResist classifier also significantly differentiated patient survival in whole patients regardless of chemotherapy. Multivariable-adjusted analysis substantiated the ChemoResist signature as an independent predictive and prognostic factor. For predicting 2-year OS, the ChemoResist classifier had significantly higher areas under the curve than TNM stage (0.788 vs 0.636, P < 0.001), other clinicopathologic characteristics (0.505-0.668), and single molecular markers (0.507-0.591) in the training cohort. Furthermore, patients with low ChemoResist scores exhibited a more favorable response to adjuvant chemotherapy compared with those with high ChemoResist scores (hazard ratio for OS: training, 0.22 vs 0.57; validation, 0.26 vs 0.50; hazard ratio for DFS: training, 0.35 vs 0.54; validation, 0.18 vs 0.59). The ChemoResist signature was further validated in the total cohort undergoing R0 resection. CONCLUSIONS: The ChemoResist signature could precisely predict survival in PDAC undergoing resection and chemotherapy, and its predictive value surpassed the TNM stage and other clinicopathologic factors. Moreover, the ChemoResist classifier could assist with identifying patients who would more likely benefit from adjuvant chemotherapy.