686PEarly detection of pancreatic ductal adenocarcinoma using methylation signatures in circulating tumour DNA

Abstract Background A noninvasive cfDNA test to detect early-stage pancreatic ductal adenocarcinoma (PDAC) is highly desirable to reduce PDAC’s high mortality rate. DNA methylation markers are promising to differentiate PDAC tissues from benign pancreas tissues, which may lead to liquid biopsy tests for PDAC early screening. Methods We have collected freshly frozen clinical PDAC tissues (N = 46), para-tumour pancreas tissues (N = 30), PDAC plasma samples (N = 120), chronical pancreatitis plasma samples (N = 90), and normal plasma samples (N = 100). Genomic DNA and circulating tumour DNA (ctDNA) samples were isolated and purified. DNA methylation profiles were generated by reduced representation bisulfite sequencing (RRBS). We quantified DNA methylation status of Methylation Haplotype Blocks (MHB) using Methylation Haplotype Load (MHL). Tumour-specific DNA methylation signatures were identified for PDAC by comparing MHL scores between PDAC and para-tumour pancreas tissues, PDAC tissues with normal cell-free circulating DNA (cfDNA), and PDAC ctDNA with normal cfDNA samples. Results We identified 350 of MHBs by combining unique MHBs identified in each comparison. Using MHB selected from tissue-vs-tissue comparison to build classifiers, we reiterated training and cross-validating PDAC tissue classification models using SVM method, and achieved the average sensitivity of 86% specificity of 88%, and AUC of 0.91. Other machine learning methods generated similar results. We further compared our markers with previously published DNA methylation markers for PDAC, and found that genes associated with published markers are highly enriched in the genes associated with our markers (p  Conclusions Using multiple metrics on MHBs, we identified PDAC-specific DNA methylation markers, some of which are functionally overlapped with previously reported markers. These markers are candidate biomarkers for non-invasive PDAC screening. Legal entity responsible for the study Peking Union Medical College Hospital and Singlera Genomics Inc. Funding China Science and Technology Exchange Center. Disclosure All authors have declared no conflicts of interest.