ATP6AP1 drives pyroptosis-mediated immune evasion in hepatocellular carcinoma: a machine learning-guided therapeutic target

Abstract

Background: Hepatocellular carcinoma (HCC) remains a major therapeutic challenge due to its immunosuppressive tumor microenvironment (TME) and resistance to immune checkpoint inhibitors (ICIs). Pyroptosis is a form of cell death with complex dual functions in tumor immunity. However, the precise regulatory mechanisms and interactions between pyroptosis and immune evasion in HCC remain poorly understood. This study aimed to elucidate the role of ATP6AP1 in pyroptosis-mediated TME remodeling and its potential as a therapeutic target.

Methods: We integrated large-scale datasets from TCGA and GEO databases to identify core modules by weighted gene co-expression network analysis (WGCNA), while mutation profiling and survival analysis verified clinical relevance. Multiple machine learning techniques, including GBM (gradient boosting machine), XGBoost (extreme gradient boosting machine), SVM (support vector machine), LASSO (least absolute shrinkage and selection operator) and random forest, as well as functional analysis, were used to systematically investigate the role of ATP6AP1 in HCC. Finally, CIBERSORT was used to analyze the immune infiltration pattern to gain insight into the mechanism.

Results: Through a rigorous multi-algorithm screening process, ATP6AP1 was found to be a highly reliable biomarker with an area under the curve (AUC) of 0.979. We found that it has a recurrent C > T mutation with an incidence of 68%. Notably, its expression level was associated with stage (P < 0.001). We also found that regions with high ATP6AP1 expression were enriched in resting DCS (P < 0.05) and regulatory T cells (P < 0.05), which further promoted immunosuppressed TME.

Conclusions: In our study, the machine learning-trained diagnostic model (AUC = 0.998) and the identified pyroptosis-related core gene ATP6AP1 provided an actionable strategy to overcome immune resistance in HCC. Mechanistically, ATP6AP1 stabilizes V-ATPase, which acidifies lysosomes, impairs antigen presentation, and drives pyroptotic inflammasome activation. This study highlights that ATP6AP1 plays a key role in promoting the lysosomal acidisis-pyroptosis-immunosuppression axis, and targeting ATP6AP1 can reshape the TME and enhance the efficacy of immunotherapy in HCC patients.

Keywords: ATP6AP1; Hepatocellular carcinoma (HCC); Pyroptosis; Regulatory T cells; Resting dendritic cells; Tumor microenvironment (TME).

Fig. 1

TCGA-LIHC difference analysis and functional enrichment analysis. A Differential analysis volcano maps. B Heat maps of the top 50 differential genes. C GO enrichment analysis of the top 10 pathways. D The top 10 pathways were analyzed for KEGG enrichment

Fig. 2

WGCNA analysis. A Optimal selection of soft threshold power and evaluation of scale-free topological fitting index (R2). B Clustering tree maps C Heat maps showing correlations between gene modules and LIHC groups. D–F Scatter plot of correlation between modules and phenotypes

Fig. 3

Mutation Landscape of Pyroptosis-Related Genes in TCGA-LIHC. A Intersection of key module genes from WGCNA and pyroptosis-related genes. B Comprehensive statistics and visualization of LIHC mutation data showing the number of different mutation types at both sample and gene levels. C Mutation profiles of 15 pyroptosis-related key module genes. D Mutation base substitution frequency analysis

Fig. 4

Construction of Prognostic Models and Risk Scores. A Univariate Cox analysis of 15 pyroptosis-related key module genes. B, C Univariate and multivariate independent prognostic analyses incorporating clinical variables and risk scores. D Survival analysis in the training cohort. E Survival analysis in the testing cohort. F Heatmap illustrating associations between LIHC clinical data and risk scores. G Analysis of TCGA-LIHC tumor stages stratified by risk score groups. H Distribution of four immune subtypes across high- and low-risk groups in LIHC

Fig. 5

Combination prediction model of 101 algorithms. A Heat map showing 101 algorithm combinations in training set and validation set AUC values. B–E ROC curves of the optimal model in the training set and the validation set. F–I Optimal model in the training set and validation set confusion matrix

Fig. 6

illustrates a methodologically rigorous computational framework integrating five machine learning algorithms for biomarker discovery. A Gradient Boosting Machine (GBM) analysis for feature selection and ranking. B XGBoost-based importance scoring of candidate genes. C Support Vector Machine (SVM)-guided feature elimination. D, E LASSO regression with covariate selection via regularization parameter (λ) optimization. F, G Random Forest algorithm evaluating gene importance metrics. H Consensus biomarker identification through intersection analysis of five computational approaches. I ROC curves validating diagnostic potential of ATP6AP1 (AUC = 0.979), RSPO3 (AUC = 0.959), and PVALB (AUC = 0.964) in LIHC stratification

Fig. 7

Survival Analysis of Pyroptosis-Related Genes. A Patient survival status and risk score distribution. B Survival analysis for ATP6AP1. C Survival analysis for PVALB. D Survival analysis for RSPO3

Fig. 8

Core Gene Identification. A Overlap between machine learning-selected genes and LIHC differentially expressed genes. B GSEA enrichment analysis for ATP6AP1. C–F Expression levels of ATP6AP1 in TCGA cohort and GEO validation datasets

Fig. 9

Immune Infiltration Landscape in LIHC. A Bar plot showing the proportional distribution of 22 immune cell types across all samples. B Correlation analysis of infiltration levels among 22 immune cell types. C Differential immune cell infiltration between normal and tumor tissues. D Correlation analysis between ATP6AP1 expression and immune cell infiltration levels