A folate metabolism-related prognostic signature for predicting survival in patients with hepatocellular carcinoma

Background Hepatocellular carcinoma (HCC) is highly heterogeneous, and reliable molecular prognostic biomarkers remain limited. Folate metabolism contributes to nucleotide synthesis, methyl-donor generation, and redox homeostasis, but the prognostic relevance of folate metabolism-related genes (FAMGs) in HCC remains unclear. Methods Differential expression and functional enrichment analyses of FAMGs were performed in TCGA-LIHC. A prognostic model was developed using univariate Cox and Elastic Net Cox regression and evaluated by bootstrap resampling and external validation in LIRI-JP and GSE10141. SHapley Additive exPlanations (SHAP) was used for model interpretation. Immune characteristics and predicted drug responses were assessed using CIBERSORT, ssGSEA, IPS, TIDE, and oncoPredict. GSS, a candidate gene outside the model, was further examined by single-cell transcriptomic analysis and in vitro experiments. Results A total of 974 differentially expressed FAMGs were identified, and 231 prognosis-associated candidates were screened to construct a 20-gene model. The model achieved a C-index of 0.767 (95% CI, 0.718–0.815) in TCGA-LIHC and an optimism-corrected C-index of 0.676 after 999 bootstrap resamples. EPO, MYCN, PPAT, PON1, and PRDX1 showed the highest global SHAP importance. External validation showed prognostic discrimination in LIRI-JP (C-index = 0.735) but limited performance in GSE10141 (C-index = 0.495). Risk groups differed in immune-related features and predicted drug sensitivity, whereas TIDE-predicted immune checkpoint inhibitor response rates were not significantly different. Although GSS was not included in the final model, higher GSS expression was associated with poorer prognosis. Single-cell analysis showed increased GSS expression and a higher proportion of GSS-positive cells in malignant hepatocytes. In vitro , GSS knockdown inhibited Hepatocellular carcinoma (HCC) cell proliferation, migration, and invasion, whereas overexpression produced the opposite effects. Conclusion A 20-gene FAMG-related prognostic model was developed for HCC and showed variable performance across external cohorts, indicating that further validation is required. GSS may contribute to malignant phenotypes in HCC and warrants further mechanistic investigation.