Machine learning-based tumor malignancy prediction based on PBMC methylome landscape in canine mammary tumor

Abstract

Abstract Background Genome-wide dysregulation of CpG methylation accompanies tumor progression and characteristic states of cancer cells, prompting a rationale for biomarker development. Understanding how the archetypic epigenetic modification determines systemic contributions of immune cell types is the key to further clinical benefits. Results In this study, we characterized the differential DNA methylome landscapes of peripheral blood mononuclear cells (PBMCs) from 76 canines using methylated CpG-binding domain sequencing (MBD-seq). Through gene set enrichment analysis, we discovered that genes involved in the growth and differentiation of T- and B-cells are highly methylated in tumor PBMCs. We also revealed the increased methylation at single CpG resolution and reversed expression in representative marker genes regulating immune cell proliferation (BACH2, SH2D1A, TXK, UHRF1). Furthermore, we utilized the PBMC methylome to effectively differentiate between benign and malignant tumors and the presence of mammary gland tumors through a machine-learning approach. Conclusions This research contributes to a better knowledge of the comprehensive epigenetic regulation of circulating immune cells responding to tumors and suggests a new framework for identifying benign and malignant cancers using genome-wide methylome.