Development and validation of a tissue-based DNA methylation risk-score model to predict the prognosis of surgically resected pancreatic cancer patients

BACKGROUND: Pancreatic cancer (PC) is a highly malignant tumor associated with low survival rates. It is challenging to predict the survival of surgically resected patients with PC. A prognostic staging tool could be beneficial to guide treatments and also aid post-treatment surveillance. This study aimed to identify tissue-based DNA methylation risk-score model to predict the prognosis of surgically resected pancreatic cancer patients. METHODS: We performed a monocentric, retrospective study that included 50 patients with stage I–II PC from The First Affiliated Hospital of Soochow University (SU cohort). Both tumor and adjacent normal tissues were obtained from each patient and subjected to capture-based targeted methylation profiling. RESULTS: In total, 1,162 DNA methylation blocks (DMBs) were differentially methylated in tumor tissues compared with adjacent long-distance tissues (P<0.05). Least Absolute Shrinkage and Selection Operator (LASSO) and stepwise regression analyses revealed a significant correlation between the methylation signature (risk score) and overall survival (OS). Patients in the high-risk group showed significantly poorer OS than those in the low-risk group in the survival analysis [P≤0.001; area under curve (AUC) at 1 year, 0.789; AUC at 2 years, 0.852]. The risk score was also validated using clinical and methylation data of 166 PC patients from The Cancer Genome Atlas pancreatic ductal adenocarcinoma (TCGA-PDAC) dataset. Patients in the high-risk group showed significantly poorer OS than those in the low-risk group (P=0.004; AUC at 1 years, 0.677; AUC at 3 years, 0.611). When clinical parameters were considered, the risk score was the only independent prognostic parameter (P<0.001) in the Cox regression analysis. Furthermore, low-risk patients had higher levels of immune infiltration, anti-tumor immune activation, and increased sensitivity to gemcitabine and paclitaxel. In contrast, high-risk patients had lower KRAS mutation rates and benefited more from cisplatin. CONCLUSIONS: In our study, we constructed and validated a tissue-based DNA methylation risk-score model to predict prognosis and identify PC patients with a high mortality risk at the time of surgery. This model might provide a tissue-based prognostic assessment tool for clinicians to aid their treatment decision-making.