Unsupervised class discovery in pancreatic ductal adenocarcinoma reveals cell-intrinsic mesenchymal features and high concordance between existing classification systems

Abstract

Pancreatic ductal adenocarcinoma (PDAC) has the worst prognosis of all common cancers. However, divergent outcomes exist between patients, suggesting distinct underlying tumor biology. Here, we delineated this heterogeneity, compared interconnectivity between classification systems, and experimentally addressed the tumor biology that drives poor outcome. RNA-sequencing of 90 resected specimens and unsupervised classification revealed four subgroups associated with distinct outcomes. The worst-prognosis subtype was characterized by mesenchymal gene signatures. Comparative (network) analysis showed high interconnectivity with previously identified classification schemes and high robustness of the mesenchymal subtype. From species-specific transcript analysis of matching patient-derived xenografts we constructed dedicated classifiers for experimental models. Detailed assessments of tumor growth in subtyped experimental models revealed that a highly invasive growth pattern of mesenchymal subtype tumor cells is responsible for its poor outcome. Concluding, by developing a classification system tailored to experimental models, we have uncovered subtype-specific biology that should be further explored to improve treatment of a group of PDAC patients that currently has little therapeutic benefit from surgical treatment.