Abstract
Hepatocellular carcinoma (HCC) is characterized by high molecular heterogeneity, posing challenges to precise diagnosis and treatment. Emerging evidence highlights the critical role of lactate metabolism in tumor progression, yet a robust subtype stratification framework based on lactate metabolism (LM)-related interaction perturbation networks remains lacking. Here, we constructed an interaction network using LM-related genes and performed consensus clustering to identify molecular subtypes of HCC. Four distinct subtypes were identified, with Cluster2 exhibiting the worst prognosis and adverse response to immune checkpoint inhibitors (ICIs), validated by independent HCC cohort. Further multi-dimensional characterization revealed that four subtypes differed significantly in clinical features, immune microenvironment, mutational landscape, and intratumoral microbiome composition. Notably, Cluster2 showed the lowest beta-diversity of microbial community and significantly increased relative abundance of genus Streptomyces and Pseudoxanthomonas, which was substantially associated with immune context. To improve prognosis prediction, we identified a 5-gene signature from Cluster2 via least absolute shrinkage and selection operator (LASSO) regression and validated its robustness in predicting survival across multiple HCC cohorts. Our study establishes a robust LM network-driven subtype system for HCC, which not only enhances multidimensional understanding of tumor heterogeneity but also provides a clinically actionable tool for personalized therapy.