Circulating tumor DNA methylation as markers for early detection of pancreatic ductal adenocarcinoma (PDAC).

395 Background: PDAC is a cancer of high mortality and low survival. Its early detection is critical due to symptoms often occur only at advanced stages. However there is no reliable screening tool to identify high-risk patients. ctDNA methylation has recently emerged as a promising new target to differentiate PDAC plasma from normal plasma for its early detection. Methods: Reduced representation bisulfite sequencing libraries were made in 46 PDAC tissues, 30 para-PDAC tissues and 20 PDAC plasmas to screen PDAC-specific markers, which was done by quantifying and comparing methylation levels of genomic regions and individual CpG sites between those groups. Markers were validated in plasma samples from 84 PDAC patients and 64 normal controls to propose a blood classifier. The best-performing markers were developed into a targeted sequencing panel, which was tested on a larger collection of plasma samples from patients of a variety of pancreatic diseases to build and validate a PDAC-predicting model. Results: We profiled genome-wide methylation patterns of tissues samples to identify 171 PDAC-specific markers. We reiterated training and cross-validating PDAC classification models using SVM method, and achieved an average sensitivity of 86% and specificity of 88%. To prove the feasibility of a non-invasive detection in plasma, a targeted methylation assay using those markers was tested on PDAC and normal plasmas, and yielded an average sensitivity of 68.4% and a specificity of 85.8%. We refined the panel by selecting the most discriminatory markers and built the version II of the panel, which was named PANcreatic Cancer Detection Assay, or PANDA, for a more efficient target capture, which was validated in an independent cohort of plasma samples that included 94 PDAC cases, 25 chronic pancreatitis (CP) cases and 80 normal samples from multiple centers. The PANDA achieved an AUC of 0.906 when classifying PDAC from normal, and an AUC of 0.882 when separating PDAC from CPs, both of which are more accurate than CA19-9, the conventional blood marker for PDAC. We further integrated test subjects’ age and their CA19-9 level as features into the PANDA model, which further elevated their AUC to 0.882 and 0.933 when classifying PDAC plasma from either CP plasma or normal plasma, respectively. Conclusions: We have developed PANDA, an NGS based target assay covering PDAC-specific DNA methylation targets by screening and validation on PDAC tissues and plasmas. Combined with age and CA19-9 blood level, PANDA has shown encouraging results to classify PDAC plasma from non-malignant diseases, demonstrating its potential to be optimized into non-invasive diagnostics for blood-based early PDAC screening.