Identification of a novel choline metabolism-related gene-pair signature to predict prognosis, immune landscape, and chemotherapy sensitivity in hepatocellular carcinoma

Hepatocellular carcinoma (HCC), a lethal malignancy of the digestive tract, exhibits poor prognosis with limited therapeutic advancements. Although choline metabolic dysregulation has been mechanistically linked to oncogenesis and metastatic dissemination, its clinical utility in prognosticating immune microenvironment characteristics and therapeutic responsiveness remains elusive in HCC. Transcriptomic profiling of 49 choline metabolism-related genes (CMRGs) was performed using TCGA-HCC data, revealing significant differential expression between tumor and adjacent non-tumorous tissues. Through permutation-based combinatorial algorithms, we generated choline metabolism-related gene pairs (CMRGPs) and employed machine learning-optimized feature selection via LASSO-Cox regression to establish a prognostic model. Model validation was conducted through a triple-cohort framework comprising internal (TCGA) and external validation sets (ICGC and GSE14520). We constructed a 10-CMRGP prognostic signature through systematic pairing of 49 differentially expressed CMRGs and machine learning-driven selection. The prognostic signature demonstrated significant efficacy in distinguishing HCC patients with divergent clinical outcomes (high-vs. low-risk subgroups, with AUC of TCGA training set, TCGA testing dataset, TCGA entire dataset, and ICGA cohort 0.749, 0.736, 0.738, and 0.733 respectively). The high-risk cohort exhibited upregulated expression of multiple immune checkpoints, heightened tumor mutational burden, and differential chemosensitivity profiles, which showing enhanced sorafenib sensitivity but refractory response to cisplatin. Moreover, the nomogram integrating risk score and tumor stage demonstrated superior predictive accuracy. Our findings demonstrate the potential use of this CMRGP signature in prognostic stratification, immune microenvironment characterization, and treatment response prediction, which could lay the foundation for future personalized precision therapy.