Integrating nontargeted metabolomics and machine learning to assess PFAS exposure associations with liver cancer risk.

Per- and polyfluoroalkyl substances (PFASs) are persistent synthetic chemicals, showing global presence in environment and human populations. Despite accumulating evidence implicating PFASs in liver toxicity, epidemiological data linking PFAS exposure to liver cancer risk remain limited, particularly regarding the underlying molecular mechanisms in humans. In this case-control study, we quantified concentrations of nine PFASs and performed nontargeted metabolomics in serum samples from 116 newly diagnosed liver cancer cases and 400 matched controls. Advanced supervised machine learning models were employed to identify PFAS exposure-associated metabolic features, followed by pathway enrichment and mediation analyses to elucidate biological mechanisms. Higher human serum levels of perfluorooctanoic acid (PFOA), perfluorononanoate, perfluorodecanoate, perfluorohexane sulfonate, perfluorooctanesulfonic acid, and 6:2 chlorinated polyfluoroether sulfonic acid were significantly correlated with the increased liver cancer risk. Weighted quantile sum index for the PFAS mixture was significantly correlated with an elevated liver cancer risk, with PFOA identified as the dominant contributor to this association. Least absolute shrinkage and selection operator (LASSO) approach based on metabolomic variables achieved the greatest predictive accuracy. Disruptions in methionine metabolism and phospholipid biosynthesis were identified as significant mediators in the association between serum PFAS levels and liver cancer risk, as revealed by LASSO modeling and mediation analyses. To our awareness, this study first demonstrates that multiple PFAS exposure may promote liver carcinogenesis through perturbations in amino acid and membrane lipid metabolism in humans.