Background
Pancreatic adenocarcinoma (PAAD) is one of the most lethal carcinomas, and the current histopathological classifications are of limited use in clinical decision-making. There is an unmet need to identify new biomarkers for prognosis-informative molecular subtyping and ultimately for precision medicine.
Methods
We profiled genomic alterations for 608 PAAD patients in a Chinese cohort, including somatic mutations, pathogenic germline variants and copy number variations (CNV). Using the CNV information, we performed unsupervised consensus clustering of these patients, differential CNV analysis and functional/pathway enrichment analysis. Cox regression was conducted for progression-free survival analysis, the elastic net algorithm used for prognostic model construction, and rank-based gene set enrichment analysis for exploring tumor microenvironments. Findings: Our data did not support prognostic value of point mutations in either highly mutated genes (such as KRAS, TP53, CDKN2A and SMAD4) or homologous recombination repair genes. Instead, associated with worse prognosis were amplified genes involved in DNA repair and receptor tyrosine kinase (RTK) related signalings. Motivated by this observation, we categorized patients into four molecular subtypes (namely repair-deficient, proliferation-active, repair-proficient and repair-enhanced) that differed in prognosis, and also constructed a prognostic model that can stratify patients with low or high risk of relapse. Finally, we analyzed publicly available datasets, not only reinforcing the prognostic value of our identified genes in DNA repair and RTK related signalings, but also identifying tumor microenvironment correlates with prognostic risks. Interpretation: Together with the evidence from genomic footprint analysis, we suggest that repair-deficient and proliferation-active subtypes are better suited for DNA damage therapies, while immunotherapy is highly recommended for repair-proficient and repair-enhanced subtypes. Our