Abstract
BACKGROUND: Given tumor heterogeneity, current diagnostic and staging systems for liver hepatocellular carcinoma (LIHC) do not fully capture the complexity of clinical management. Aberrant activation of the G2/M checkpoint disrupts normal cell-cycle control and can enable proliferation of genomically damaged cells, thereby promoting tumor progression. Targeted inhibition of this pathway may force tumor cells into mitosis and subsequently trigger apoptosis. METHODS: We defined G2/M checkpoint (G2MC) subtypes using unsupervised clustering of gene expression profiles from the TCGA-LIHC cohort, together with corresponding clinical data. We then systematically compared biological and clinical differences between subtypes using survival analysis, clinical feature analysis, immune infiltration profiling, tumor mutational burden (TMB) analysis, and drug sensitivity assessment. Next, we developed a G2MC subtype classifier using machine-learning approaches, including an artificial neural network (ANN), and validated its clinical utility and predictive performance in an independent clinical cohort. Finally, we examined the biological effects of CENPA in vitro using two cancer cell lines. RESULTS: We developed a G2MC pathway activity–based classifier to stratify LIHC patients by prognosis and predicted treatment response. The two G2MC subtypes showed distinct G2/M checkpoint–related expression patterns and mutational landscapes. Compared with subtype C1, subtype C2 had a higher G2MC score and was associated with worse survival, higher pathological grade, more advanced clinical stage, higher AFP levels, a lower predicted response to immunotherapy, and greater sensitivity to 5-fluorouracil and sorafenib. The classifier’s clinical applicability was further supported in an independent cohort of 30 clinical cases. In vitro experiments also provided additional evidence for the biological relevance of CENPA. CONCLUSION: We developed a classifier to identify G2MC subtypes in LIHC, which may support more precise prognostic assessment and treatment selection. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12935-026-04237-y.