Prognostic Relevance of Pancreatic Adenocarcinoma Whole-tumor Transcriptomic Subtypes and Components.

PURPOSE Our team previously defined six quantitative transcriptomic components, and a classification in five subtypes by association of these components. In this study, we compared the robustness of quantitative components and qualitative classifications from different transcriptomic profiling techniques, investigated their clinical relevance and proposed a new prognostic model. EXPERIMENTAL DESIGN 210 patients from a multicentric cohort and 149 patients from a monocentric cohort were included in this study. RNA micro-arrays profiles were obtained from 165 patients of the multicentric cohort. RNA sequencing (RNA-seq) profiles were obtained from all the patients. RESULTS For the patients with both RNA micro-array and RNA-seq profiles, the concordance in subtype assignment was partial with an 82.4% coherence rate. The correlation between the two techniques projections of the six components ranged from 0.85 to 0.95, demonstrating an advantage of robustness. Based on the Akaike Information criterion, the RNA components showed more prognostic value in univariate or multivariate models than the subtypes. Using the monocentric cohort for training, we developed a multivariate Cox regression model using all six components and clinicopathological characteristics (node invasion and resection margins) on DFS. This prognostic model was highly associated with DFS (p<0.001). The evaluation of the model in the multicentric cohort showed significant association with DFS and OS (p<0.001). CONCLUSIONS We described the advantage of the prognostic value and robustness of the whole-tumor transcriptomic components than subtypes. We created and validated a new DFS based multivariate Cox regression prognostic model, including six PAC transcriptomic component levels and pathological characteristics.