Abstract A026: Bulk deconvolution analysis identifies subtype-specific methylation differences in pancreatic cancer patients

Abstract

While the two transcriptional subtypes (basal-like and classical) of pancreatic ductal adenocarcinoma (PDAC) offer an opportunity for precision medicine, current subtype-specific therapy options are limited to existing frontline therapies with poor outcomes. To determine whether there are epigenetically-driven differences driving the distinct biological phenotypes of the PDAC subtypes, we probed the methylation data of PDAC patients in The Cancer Genome Atlas (TCGA) for subtype-specific patterns. Methylation and gene expression data were obtained from the pancreatic cancer cohort (n=149 samples analyzed) in TCGA. Methylation data were pre-processed by removing known cross-reactive probes and probes on sex chromosomes. Gene expression data were log2 transformed from raw counts. Non-negative matrix factorization (NMF) was used to deconvolute each dataset into 7 distinct features: basal-like, classical, normal stroma, activated stroma, exocrine, endocrine, and immune. Methylation sites uniquely altered in the basal and classical features were then cross-referenced with the top 250 genes that define the basal-like and classical transcriptional subtypes of PDAC, as determined in Moffitt et al., Nature Genetics, 2015. NMF deconvolution identified 1,739 differentially altered methylation sites across the 7 features in the PDAC TCGA dataset. Of these sites, 1,665 were unique to one or both PDAC subtypes, with 22.8% of sites uniquely altered in the basal-like feature, 15.2% uniquely altered to the classical feature, and 62.0% uniquely altered in both subtypes. Surprisingly, only a small subset of the methylation sites uniquely altered in each of these PDAC-specific features overlapped with the top 250 genes that define the basal-like and classical subtypes of PDAC: only 2.9% of basal-like sites mapping to 9 distinct basal-like genes and 2.0% of classical sites mapping to 5 distinct classical genes. The 1,739 differentially altered methylation sites were validated in primary flow-sorted PDAC cells from Espinet et al., Cancer Discov., 2021 and in PDAC patient-derived xenografts from Lomberk et al. Nat. Commun., 2018. These findings suggest that there are substantial epigenetic changes at the methylation level that contribute to the biological phenotypes of the PDAC subtypes not captured at the transcriptional level. Current efforts are focused on identifying biological pathways altered by these differentially methylated sites. Additionally, the methylation coverage of the genome in this dataset is limited due to its array-based format. To overcome this challenge, we are conducting enzymatic whole genome methylation-sequencing (which yields superior quality to bisulfite sequencing) coupled to transcriptomic profiling experiments in a new cohort of freshly resected PDAC tissues. These efforts will provide a large-scale comprehensive examination of how the methylation landscapes affect the biology of the PDAC subtypes, with the goal of identifying new precision medicine opportunities. Erin E. Grundy, Margaret A. Hall, Yun J. Yoo, Eric G. Bogart, Natalie K. Horvat, Gregory B. Lesinski, Richard A. Moffitt. Bulk deconvolution analysis identifies subtype-specific methylation differences in pancreatic cancer patients [abstract]. In: Proceedings of the AACR Conference on Pancreatic Cancer: New Frontiers in Biology and Therapeutic Development; 2026 Sep 25-28; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_2):Abstract nr A026.