The hidden effects of metabolism-disrupting agents on alcohol-related liver disease risk: an integrated epidemiological, machine learning, and network toxicology Analysis

Abstract Background Individual susceptibility to alcohol-related liver disease (ALD) varies substantially despite similar alcohol consumption patterns. Emerging evidence suggests metabolism-disrupting agents (MDAs) may synergistically amplify alcohol-induced hepatotoxicity, yet population-level evidence remains limited. This study investigated associations between 40 MDAs and ALD risk using integrated analytical approaches. Methods We analyzed 13,472 National Health and Nutrition Examination Survey participants (2005–2016). Forty MDAs spanning per- and polyfluoroalkyl substances, phenolic compounds, phthalate metabolites, polycyclic aromatic hydrocarbon metabolites, and volatile organic compound metabolites were measured using standardized protocols. Multivariable logistic regression assessed MDA-ALD associations. Restricted cubic splines characterized dose-response relationships. Subgroup analyses identified vulnerable populations by sex, age, race/ethnicity, body mass index, and hyperlipidemia status. Machine learning algorithms including LightGBM were developed with nested cross-validation (to prevent data leakage during feature selection) to identify predictive biomarkers. Network toxicology integrated computationally predicted MDA targets with ALD gene expression data (GEO GSE28619), followed by pathway enrichment analyses. Results Three MDAs demonstrated robust positive associations: benzylmercapturic acid (BMA; OR: 1.56, 95% CI: 1.31–1.86), perfluorohexanesulfonic acid (PFHxS; OR: 1.56, 95% CI: 1.28–1.90), and benzophenone-3 (BP-3; OR: 1.40, 95% CI: 1.15–1.72). BMA and PFHxS exhibited linear dose-response relationships without thresholds, while ATCA showed an inverted U-shaped pattern and BP-3 displayed a plateau pattern. Phenolic compounds and phthalates demonstrated stronger associations in females. LightGBM achieved optimal performance (cross-validation AUC: 0.760, test AUC: 0.706), identifying N-acetyl-S-(3-hydroxypropyl)-L-cysteine, N-acetyl-S-(N-methylcarbamoyl)-L-cysteine, and mono-carboxyoctyl phthalate as key predictive biomarkers, including compounds not significant in regression. Network analysis identified 192 shared genes with enrichment in neuroactive ligand-receptor interactions and nuclear receptor pathways (PPARα/γ, PXR, FXR). Conclusions Specific MDAs demonstrated significant ALD associations through complementary approaches. Evidence suggests MDAs may function as signal disruptors within neuro-endocrine-immune networks, with pathways paralleling alcohol-induced hepatotoxicity. Findings underscore incorporating environmental assessments into ALD risk stratification and highlight needs for prospective validation and mechanistic studies.