A glutamine metabolism-related gene signature for liver fibrosis prediction integrating single-cell and bulk transcriptomic analysis with multi-cohort validation

Liver fibrosis is a critical precursor to cirrhosis and liver cancer, yet effective non-invasive predictive tools remain lacking. The role of glutamine metabolism in liver fibrosis has not been systematically investigated. Single-cell analysis revealed predominant enrichment of glutamine metabolism genes in epithelial cells. Using GSE14323, we applied LASSO and SVM algorithms to screen and construct a prognostic risk model. The model was externally validated in three external datasets. A nomogram was developed, and comprehensive analyses of immune infiltration, immune scores, cell death pathways, and experimental validation were performed. Single-cell transcriptomic analysis demonstrated that glutamine metabolism-related genes were predominantly enriched in epithelial cells of fibrotic livers. Using 71 glutamine metabolism-related genes, we applied machine learning algorithms and identified a five-gene signature (ALDH4A1, GLS, GLUD1, GOT1, OTC). The nomogram constructed from these five genes achieved an AUC of 0.946 in the training cohort and maintained excellent predictive performance with AUC values above 0.8 across three independent external validation cohorts. WGCNA revealed that the key genes were significantly enriched in biological processes including chemotaxis, leukocyte adhesion, and extracellular matrix organization. Furthermore, comprehensive immune infiltration analysis demonstrated significant differences between high- and low-risk groups across 22 immune cell types, accompanied by alterations in multiple immune checkpoint molecules. Experimental validation in two independent mouse models of liver fibrosis (MCD and HFD) and human liver tissues confirmed that GOT1 and GLS were consistently upregulated at both mRNA and protein levels in fibrotic livers. We developed a robust glutamine metabolism-related five-gene signature for liver fibrosis prediction with excellent accuracy and generalizability across multiple independent cohorts.