Genome-wide DNA methylation analysis reveals molecular subtypes of pancreatic cancer

Abstract

Pancreatic cancer (PC) is the fourth leading cause of cancer deaths in the United States with a five-year patient survival rate of only 6%. Early detection and treatment of this disease is hampered due to lack of reliable diagnostic and prognostic markers. Recent studies have shown that dynamic changes in the global DNA methylation and gene expression patterns play key roles in the PC development; hence, provide valuable insights for better understanding the initiation and progression of PC. In the current study, we used DNA methylation, gene expression, copy number, mutational and clinical data from pancreatic patients. We independently investigated the DNA methylation and differential gene expression profiles between normal and tumor samples and correlated methylation levels with gene expression patterns. We observed a total of ~23-thousand differentially methylated CpG sites (Δβ≥0.1) between normal and tumor samples, where majority of the CpG sites are hypermethylated in PC, and this phenomenon is more prominent in the 5′UTRs and promoter regions compared to the gene bodies. Differential methylation is observed in genes associated with the homeobox domain, cell division and differentiation, cytoskeleton, epigenetic regulation and development, pancreatic development and pancreatic signaling and pancreatic cancer core signaling pathways. Correlation analysis suggests that methylation in the promoter region and 5′UTR has mostly negative correlations with gene expression while gene body and 3′UTR associated methylation has positive correlations. Regulatory element analysis suggests that HOX cluster and histone core proteins are upstream regulators of hypomethylation, while SMAD4, STAT4, STAT5B and zinc finger proteins (ZNF) are upstream regulators of hypermethylation. Non-negative matrix factorization (NMF) clustering of differentially methylated sites generated three clusters in PCs suggesting the existence of distinct molecular subtypes. Cluster 1 and cluster 2 showed samples enriched with clinical phenotypes like neoplasm histological grade and pathologic T-stage T3, respectively, while cluster 3 showed the enrichment of samples with neoplasm histological grade G1. To the best of our knowledge, this is the first genome-scale methylome analysis of PC data from TCGA. Our clustering analysis provides a strong basis for future work on the molecular subtyping of epigenetic regulation in pancreatic cancer.

Machine Learning and Network Analyses Reveal Disease Subtypes of Pancreatic Cancer and their Molecular Characteristics

“…Recently, the paradigm of oncogenesis has been expanded beyond the classical view that oncogenesis is entirely driven by the accumulation of genetic mutations 67,68 . This paradigm now includes the disruption of epigenetic regulatory mechanisms and variations in miRNA expression [68][69][70][71][72][73] . Unlike with mutations, we currently lack adequate conceptual knowledge and the analytical framework needed to identifying putative driver and passenger changes in epigenetic and miRNA based regulatory processes [73][74][75][76] .…”

The Emerging Roles of ATP-Dependent Chromatin Remodeling Complexes in Pancreatic Cancer

“…A limited number of studies have focused on characterizing the role of CHD7 in PDAC. CHD7 is differentially methylated in PDAC [207]. CHD7 was dysregulated in over 90% of the analyzed PDAC samples, and low CHD7 expression was associated with increased survival in patients receiving adjuvant gemcitabine therapy [208].…”

“…Tumor suppressor role in chronic lymphocytic leukemia [209]. Hypomethylated in PDAC [207]. Required to maintain the differentiation potential of mouse ESCs [210].…”

NMFNA: a non-negative matrix factorization network analysis method for identifying communities and characteristic genes from multi-type pancreatic cancer data

“…In addition to above mentioned methods, several Non-negative Matrix Factorization (NMF) based methods also have been proposed to detect pathogenic factors associated with PC. For instance, Mishra et al [9] used the NMF to distinct molecular subtypes of PC methylome data. Wang et al [10] proposed the maximum correntropy criterion based NMF (NMF-MCC) to cluster the PC gene expression data.…”