Stratification of Pancreatic Ductal Adenocarcinomas Based on Tumor and Microenvironment Features

BACKGROUND & AIMS: Genomic studies have revealed subtypes of pancreatic ductal adenocarcinoma (PDA) based on their molecular features, but different studies have reported different classification systems. It is a challenge to obtain high-quality, freshly frozen tissue for clinical analysis and determination of PDA subtypes. We aimed to redefine subtypes of PDA using a large number of formalin-fixed and paraffin-embedded PDA samples, which are more amenable to routine clinical evaluation. METHODS: We collected PDA samples from 309 consecutive patients who underwent surgery from September 1996 through December 2010 at 4 academic hospitals in Europe; nontumor tissue samples were not included. Samples were formalin fixed and paraffin embedded. DNA and RNA were isolated; gene expression, targeted DNA sequencing, and immunohistochemical analyses were performed. We used independent component analysis to deconvolute normal, tumor, and microenvironment transcriptome patterns in samples. We devised classification systems from an unsupervised analysis using a consensus clustering approach of our data set after removing normal contamination components. We associated subtypes with overall survival and disease-free survival of patients using Cox proportional hazards regression with estimation of hazard ratios and 95% confidence interval. We used The Cancer Genome Consortium and International Cancer Genome Consortium PDA data sets as validation cohorts. RESULTS: We validated the previously reported basal-like and classical tumor-specific subtypes of PDAs. We identified features of the PDA, including microenvironment gene expression patterns, that allowed tumors to be categorized into 5 subtypes, called pure basal like, stroma activated, desmoplastic, pure classical, and immune classical. These PDA subtypes have features of cancer cells and immune cells that could be targeted by pharmacologic agents. Tumor subtypes were associated with patient outcomes, based on analysis of our data set and the International Cancer Genome Consortium and The Cancer Genome Consortium PDA data sets. We also observed an exocrine signal associated with acinar cell contamination (from pancreatic tissue). CONCLUSIONS: We identified a classification system based on gene expression analysis of formalin-fixed PDA samples. We identified 5 PDA subtypes, based on features of cancer cells and the tumor microenvironment. This system might be used to select therapies and predict patient outcomes. We found evidence that the previously reported exocrine-like (called ADEX) tumor subtype resulted from contamination with pancreatic acinar cells. ArrayExpress accession number: E-MTAB-6134.