P-212 The genetic and epigenetic abnormalities of plasma cfDNA as liquid biopsy biomarkers for diagnosis and prognosis stratification of pancreatic ductal adenocarcinoma

Yongyan Wang,R. Jiang,S. Xiao,Yuankai Shi,Junnian Zheng,Zhenhui Li,Jun Peng,Jian Sun,G. Yang,J. Yuan,Wenhui Lou

DOI: https://doi.org/10.1016/j.annonc.2021.05.266

IF: 51.769

2021-01-01

Annals of Oncology

Abstract:Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive cancer. The 5-year survival rate is low, only at 39% for localized disease and 3% for distant disease. Serum CA19-9 level has been used as a diagnostic marker for PDAC but the lower sensitivity in early-stage patients and false positive in non-cancerous conditions limit its clinical utility for PDAC detection. Therefore, an effective detection approach is still needed to improve disease outcome. Recent technological advances in the detection of tumor-originating circulating free DNA (cfDNA) provide new opportunities for diagnosis and surveillance of PDAC. In this study, we performed parallel genetic and epigenetic profiling of cfDNA from Chinese PDAC patients, as well as healthy individuals, by targeted ultra-deep sequencing and targeted bisulfite sequencing. For methylome profiling, we first identified PDAC-specific differentially methylated regions (DMRs) in tissues and then employed machine learning approach to select the most informative features. Diagnostic models were built to distinguish PDAC patients from healthy individuals. Models were also built to predict prognosis in PDAC patients. A total of 1173 DMRs were identified from comparison of 32 pairs of PDAC cancer vs. para-cancerous tissues. For the diagnostic model, the training cohort consisted of 119 PDAC patients (median age of 63) and 113 age-matched healthy individuals (median age of 60). A random forest classifier achieved an AUC of 0.95±0.04 with 10-fold cross-validation using a panel of 14 DMR markers. Meanwhile, cfDNA mutation profiling achieved a sensitivity of 48.8% and a specificity of 98.2% in the training cohort, providing an inferior diagnostic performance compared to the methylation-based assay. Further analyses were performed in an independent validation cohort, including PDAC patients (n=51) as well as healthy control (n=96). The cfDNA methylation model achieved a sensitivity of 86.3%, and a specificity of 97.6%; stage I sensitivity was 81.0%. On the other hand, the diagnostic model achieved a sensitivity of 44.4% and a specificity of 98.0% based on mutation profile, and serum CA19-9 level alone achieved a sensitivity of 80.4% in this cohort. Next, we tried to further improve the performance of PDAC diagnostic model by combining methylation profile, mutation profile and serum CA19-9 level. A sensitivity of 95.7% was achieved in model combining genetic and epigenetic markers and further increased to 97.8% by including serum CA19-9 level. For the prognostic model, we also found that dual-omics-based prognostic score incorporating both mutation burden and methylation-based prognostic score (MPS) was an independent prognostic predictor. In conclusion, our results suggest that cancer-derived epigenetic abnormalities provide promising biomarkers for the detection of PDAC. We could further improve performance of the PDAC classifier by incorporating mutation profile and serum CA19-9 level. Moreover, integration of genetic and epigenetic alterations may improve prognosis stratification and guide decision making in pretreatment setting.