Integrated single-cell and bulk RNA-seq analysis reveals an NK-cell-related immune landscape and a prognostic signature for hepatocellular carcinoma

BACKGROUND: Hepatocellular carcinoma (HCC) is characterized by high heterogeneity and poor prognosis. Despite the importance of natural killer (NK) cells in innate immunity, their role in the HCC microenvironment remains unclear. This study aims to develop an NK-cell-related signature to optimize prognostic prediction and personalized therapy. METHODS: Single-cell RNA sequencing (scRNA-seq) identified candidate genes. Differential expression and univariate Cox analyses were performed on an integrated cohort comprising The Cancer Genome Atlas Liver Hepatocellular Carcinoma (TCGA-LIHC) and the Gene Expression Omnibus (GEO) dataset GSE76427. To ensure data consistency, technical batch effects were eliminated using the ComBat algorithm. The integrated cohort was randomly assigned to training and testing sets (1:1). A prognostic model was constructed via Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression (seed = 2) and validated in an independent ICGC-LIRI-JP cohort. Survival differences and model accuracy were evaluated using Kaplan-Meier and Receiver Operating Characteristic (ROC) curves. Finally, the Cell-type Identification by Estimating Relative Subsets of RNA Transcripts (CIBERSORT) algorithm and the Tumor Immune Dysfunction and Exclusion (TIDE) database were applied to characterize immune cell infiltration and predict immunotherapy response, respectively. RESULTS: From 790 initial candidates, 18 prognosis-related DEGs associated with NK-cell-related cellular programs were identified, which classified HCC into two distinct subtypes. A three-gene prognostic model (AP1S3, RPL23, and HM13) was established, showing high predictive accuracy in both the internal testing set and an independent external ICGC-LIRI-JP cohort. High-risk patients exhibited significantly poorer survival and specific clinical profiles. Notably, the risk score served as an independent prognostic factor, and a combined nomogram offered superior predictive power. Further characterization of the tumor microenvironment revealed that a high-risk score correlated with a remodeled stroma and a distinct potential for immunotherapy response. Additionally, high-risk patients showed increased sensitivity to targeted agents, such as sorafenib. Finally, the expression patterns of these signature genes were validated via RT-qPCR and immunohistochemistry (IHC), further supporting their role as key components of the identified prognostic landscape. CONCLUSION: The established risk score model, reflecting an NK-cell-related immune landscape, effectively predicts HCC prognosis and provides an integrated molecular landscape for optimizing personalized interventions.