Identification and Implications for Tumor Heterogeneity of a DNA Methylation-Based Signature Classifying Pancreatic Ductal Adenocarcinoma Based on their Cellular Origin

Abstract

Background Pancreatic ductal adenocarcinoma (PDAC) arises from distinct cellular origins, yet the extent to which DNA methylation patterns from normal pancreatic cells are preserved in tumor cells remains unclear. Identifying cell-of-origin signatures may enhance PDAC classification and therapeutic stratification. Objective To determine whether DNA methylation signatures in normal acinar and ductal pancreatic cells are retained in PDAC cell lines and to develop a robust classifier for distinguishing tumor origins. Design We performed DNA methylation profiling using the Illumina Infinium Mouse MethylationEPIC array on normal acinar and ductal cells and their PDAC derivatives in genetically engineered mouse models (GEMMs). Differential methylation analysis, and hierarchical clustering were used to identify and validate a conserved cell-of-origin DNA methylation signature. A logistic regression model was developed for classification. Results We identified 178 CpG sites that remain preserved during tumorigenesis and effectively distinguished acinar- and ductal-derived PDAC cell lines. This signature was validated across independent sample sets, primary tumors, and orthotopic allografts. It successfully classified cell lines of unknown origin, including PDAC samples from KPC mice, and revealed the impact of oncogenic mutations on tumor fate. A logistic regression model supported these findings, confirming the robustness of the classification approach. Furthermore, the cell of origin influenced key PDAC characteristics, including treatment response, highlighting its potential role in molecular subtyping and patient stratification. Conclusion A preserved DNA methylation signature during pancreatic carcinogenesis distinguishes PDAC origins and influences tumor behavior. These results highlight the potential of DNA methylation profiling for tumor classification and personalized treatment strategies. They also raise important questions about the relevance of KPC mice as a preclinical model and the mechanisms driving PDAC heterogeneity.