整合多组学与网络毒理学揭示环境激素驱动肝细胞癌的多靶点机制

Hepatocellular carcinoma (HCC) is a major malignancy with rising global incidence and mortality. Clinical treatment is limited by molecular heterogeneity and drug resistance. In recent years, endocrine-disrupting chemicals (EDCs) have attracted attention as emerging risk factors, but systematic pathogenic evidence for their roles in HCC initiation and progression remains insufficient.

First, we predicted potential targets of EDCs using SwissTargetPrediction, STITCH, and ChEMBL, and intersected them with differentially expressed genes and key module genes from WGCNA in the GEO database to screen candidate key genes. Second, based on these candidates, we constructed diagnostic models using 14 machine-learning algorithms and evaluated feature importance via the SHAP framework to identify key biomarkers and their functional contributions. Molecular docking and molecular dynamics simulations were used to validate interaction mechanisms between EDCs and key target proteins. We then built a multivariable Cox proportional hazards model in the TCGA-LIHC cohort and performed stratified survival analysis, somatic mutation profiling, and immune evasion characterization. Subsequently, we evaluated the tumor immune microenvironment using CIBERSORT and ssGSEA, and integrated single-cell transcriptomic data to resolve cell-subtype heterogeneity, target expression distributions, and cell-cell communication. Meanwhile, we integrated the GDSC drug-sensitivity database to evaluate associations between risk scores and drug response, and conducted pan-cancer analyses to examine cross-cancer applicability.

We identified 18 genes jointly associated with EDCs and HCC, significantly enriched in AMPK, p53, and FoxO signaling pathways and cell cycle-related pathways. Among models built with 14 machine-learning algorithms, CatBoost showed the best discriminative performance and identified CCNB2 and AKR1C3 as core driver genes. Docking and dynamics simulations indicated strong binding affinities and stable binding conformations between EDCs and target proteins including CCNB1 (-8.9 kcal/mol), AKR1C3 (-8.4 kcal/mol), and FADS1 (-8.5 kcal/mol). A multivariable Cox risk model based on nine key genes served as an independent prognostic predictor for HCC (HR = 1.746, 95% CI: 1.477-2.064, P < 0.001). The nomogram achieved AUCs of 0.836, 0.810, and 0.788 at 1, 3, and 5 years, respectively, indicating good predictive performance. The high-risk group was significantly associated with high tumor mutational burden (TMB), TP53 mutations, and low immune evasion scores. Regarding the tumor immune microenvironment, CIBERSORT and ssGSEA analyses showed marked enrichment of Tregs and M0 macrophages, while most effector immune cells and functions were suppressed. Single-cell transcriptomics further showed enrichment of endothelial cells, fibroblasts, hepatocytes, and macrophages in HCC tissues, with notable reductions in T cells, B cells, NK cells, and neutrophils, indicating an immunosuppressive microenvironment with stromal remodeling. Cell-cell communication analysis indicated that the MIF-CD74 receptor axis is central in immune-cell interactions. Drug-sensitivity analysis suggested that the high-risk group was more sensitive to GDC0810, BPD-00008900, and Fulvestrant, indicating potential beneficiary populations. Pan-cancer analysis showed that the risk model also had diagnostic and prognostic value in LUAD, KIRP, KIRC, and KICH, suggesting cross-cancer generalizability.

This study systematically reveals that EDCs promote HCC initiation and progression by perturbing cell cycle, metabolic, and immune homeostasis through multi-target, multi-pathway mechanisms. The nine-gene risk model demonstrates superior performance in HCC diagnosis and prognosis and shows potential clinical translational value in drug-sensitivity prediction and pan-cancer analyses. This work provides a new perspective at the intersection of environmental toxicolology and precision oncology and informs individualized therapeutic strategies.