DNA methylation driven genes shape prognosis and tumor immune landscape in rectal cancer

DNA methylation alterations play a crucial role in rectal cancer development. This study identifies and characterizes DNA methylation-driven genes in the TCGA-READ cohort, constructs a prognostic model, and validates its robustness and biological relevance in independent external cohorts. Using TCGA-READ (n = 160) as the discovery cohort, we identified 490 methylation-driven genes via MethylMix analysis. A LASSO-derived prognostic model was constructed integrating expression of CCNI2, LINC00899, and ST6GALNAC1. Model performance was assessed by Kaplan–Meier analysis, time-dependent ROC, calibration, and decision curve analysis. Mechanistic insights were explored using differential expression, PPI network analysis, GSEA, and immune infiltration estimation. External validation of the prognostic model was performed in the independent dataset GSE103479 (n = 155). To address gene expression and functional relevance, expression differences for CCNI2, LINC00899, and ST6GALNAC1 were examined in external datasets (GSE71187 and GSE87211), and correlation-based, cross-dataset-averaged GSEA was conducted against MSigDB GO/KEGG/Hallmark collections. The three-gene model stratified TCGA-READ patients into high- and low-risk groups with progressive AUCs of 0.636, 0.714, and 0.898 at 1, 3, and 5 years. High-risk tumors showed reduced expression of multiple protective genes (hub: SULF1), depletion of negative regulators of oncogenic pathways, and reduced cytotoxic immune infiltration. In GSE103479, external validation demonstrated consistent risk stratification trends and supported broader applicability. Gene-specific analyses revealed CCNI2 correlated with proliferation and antigen processing pathways; LINC00899 associated with metabolic reprogramming (oxidative phosphorylation, glycolysis); ST6GALNAC1 linked to epithelial maintenance and fatty acid metabolism. A methylation-driven 3-gene signature provides prognostic value in READ and is supported by independent-cohort validation and external bioinformatics evidence for gene expression and functional relevance.