G2M-checkpoint related immune barrier structure and signature for prognosis and immunotherapy response in hepatocellular carcinoma: insights from spatial transcriptome and machine learning.

Background: Hepatocellular carcinoma (HCC) treatment remains challenging, particularly for immune checkpoint inhibitors (ICIs) non-response patients. Spatial transcriptome (ST) data and machine learning algorithms offer new insights into understanding HCC heterogeneity and ICIs resistance mechanisms.

Methods: Utilizing ST data from HCC patients on ICIs treatment, we analyzed pathway activity and immune infiltration. We combined 167 machine learning models to develop a G2M-checkpoint related signature (G2MRS) based on differential gene expression. The four cohorts and in-house cohort was used to validate G2MRS, and KPNA2's role was further examined through in vitro experiments in two different liver cancer cell lines.

Results: Our analysis revealed a distinct suppressive immune barrier structure (SIBS) in ICIs non-response patients, associated with upregulated G2M-checkpoint levels. G2MRS, consisting of KPNA2, CENPA, and UCK2, accurately predicted HCC prognosis and ICIs response. Further in vitro experiments demonstrated KPNA2's role in regulating migration, proliferation, and apoptosis in liver cancer.

Conclusions: This study highlights the importance of spatial heterogeneity and machine learning in refining HCC prognosis and ICIs response prediction. G2MRS and KPNA2 emerge as promising biomarkers for personalized HCC management.