To construct a prognostic model for patients with hepatocellular carcinoma (HCC) based on multi-omics subtyping of ammonia-induced cell death features, and to explore the mechanism of tumor microenvironment (TME) remodeling and precision treatment strategy.

In this study, The Cancer Genome Atlas (TCGA), conventional transcriptome of International Cancer Genome Consortium (ICGC) cohort and multiple single-cell RNA sequencing (scRNA-seq) datasets were integrated. Firstly, the activity of ammonia-induced cell death pathway of samples was quantified by single-sample gene set enrichment analysis (ssGSEA), and the modules of key ammonia-induced cell death genes were identified by weighted gene co-expression network analysis (WGCNA). The risk scoring prognostic model was constructed by the least absolute shrinkage and selection operator (Lasso) regression analysis. Multiple immune infiltration algorithms, SCENIC, CellChat and drug sensitivity prediction were employed to systematically analyze the "endothelial-myeloid" immune rejection remodeling mechanism of high-risk TME and synthesize lethal intervention targets. Correlation analysis was conducted by Pearson or Spearman correlation coefficients. Survival analysis was performed by Kaplan-Meier method and Log-rank test.

Based on the sample data of 371 cases in TCGA-LIHC cohort, 17 gene co-expression modules were identified by cluster analysis. 1 073 genes in two key modules were defined as candidate genes of ammonia-induced cell death. Functional enrichment analysis showed that the above candidate genes were closely associated with immune regulation in biological function. Based on the genes related to ammonia-induced cell death, a risk scoring prognostic model of ammonia-induced cell death with independent prognostic value was successfully constructed. ROC analysis demonstrated that the model possessed high predictive efficiency. The area under the ROC curve (AUC) for predicting the 1-, 3-and 5-year survival rates of liver cancer was 0.855, 0.824 and 0.787, respectively. External validation of ICGC cohort confirmed that the model showed high cross-cohort consistency in identifying highly-invasive liver cancer subgroups and prognostic risk stratification. High-risk phenotype was mainly characterized with genomic instability driven by TP53 mutation. TME was characterized with "immune rejection" phenotype, with an "endothelial-myeloid" immune rejection barrier. The infiltration of CD4⁺ central memory T cells (CD4⁺ Tcm) was significantly down-regulated. Inhibitory myeloid cells represented by monocyte lineage were significantly up-regulated, and the infiltration scores of fibroblasts and smooth muscle cells were also significantly up-regulated. CD8⁺ T cells in TME were in an inactive resting state, manifested with a significant decrease in the proportion of terminal effector subsets and relative enrichment of transitional and naive cells. The regulatory network suggested that abnormal activation of MYC in high-risk endothelial cells promoted the recruitment of myeloid cells probably through the ANGPTL/JAM signal axis and participated in the construction of physical and immune barriers. In terms of therapeutic strategies, HCC patients of high-risk subtypes showed significant potential sensitivity to WEE1 inhibitors (adavosertib).

In this study, a risk scoring prognostic model of ammonia-induced cell death with independent prognostic value is established, which can effectively identify high-risk HCC subtypes with TP53 mutation. TME in the high-risk group is characterized with typical "endothelial-myeloid" immune rejection, accompanied by activation blockade of CD8⁺ T cells. WEE1 inhibitors are expected to become potential targeted interventional drugs for high-risk patients.