Abstract
BACKGROUND: DNA methylation plays a crucial role in the onset and progression of cancer. However, the complex technology and high costs required for methylation detection limit its clinical application. DNA methylation regulators are essential for maintaining the precision and stability of gene methylation, and their aberrant expression can lead to abnormal methylation levels. Whereas the role of combinatorial methylation regulators in pancreatic cancer (PCA) risk remains unclear, we developed a model using 20 DNA methylation regulators to predict patient prognosis and assess treatment response. METHODS: Gene expression and clinical data from 331 PCA patients [The Cancer Genome Atlas (TCGA)-PCA, n=177; Gene Expression Omnibus (GEO)-PCA, n=154] were analyzed. TCGA data were used as the training set, and GEO data were used as the validation set. Inclusion criteria were complete survival data. Univariate and least absolute shrinkage and selection operator (LASSO)-Cox regression identified prognostic DNA methylation regulators. The model's predictive accuracy was validated using time-dependent receiver operating characteristic (ROC) curves. Differences in immune cell infiltration and drug sensitivity were also assessed. RESULTS: A total of 331 PCA patients were analyzed, with a median overall survival (OS) of 1.2 and 1.4 years, respectively. Univariate Cox regression identified seven DNA methylation regulators (DNMT3A, TET3, MBD3, MBD2, ZBTB38, UHRF1, UNG) associated with prognosis, of which MBD3 and UHRF1 were selected via LASSO-Cox regression to construct the final model. The model demonstrated robust prognostic performance, with low-risk patients in both cohorts showing significantly longer OS compared to high-risk groups (P<0.001). ROC analysis confirmed reliability, yielding area under the curve (AUC) values of 0.662 (1-year), 0.684 (2-year) and 0.673 (3-year) in TCGA, and 0.629 (1-year), 0.663 (2-year) and 0.624 (3-year) in GEO. Drug sensitivity analysis further revealed that the low-risk group exhibited enhanced responses to epirubicin (P<0.001), irinotecan (P<0.001), and Poly(ADP-ribose) polymerase (PARP) inhibitors (niraparib P<0.001, olaparib P<0.001), suggesting potential therapeutic implications. CONCLUSIONS: Our findings suggest that the prognostic model, which is based on MBD3 and UHRF1 expression, may improve prognostic stratification in PCA patients and assess drug efficacy. This model represents a step toward epigenetic-based oncology, though its impact on treatment decisions remains to be validated.