The epidemiological component drew on data from the National Health and Nutrition Examination Survey, or NHANES, a nationally representative program that collects serum chemical measurements from thousands of Americans. The team analyzed records from 1,834 participants, comparing serum PFOS concentrations against MASLD status as defined by the fatty liver index, a validated algorithm that integrates body mass index, waist circumference, triglycerides, and gamma-glutamyl transferase. Their analysis identified a statistically significant nonlinear association between serum PFOS and the odds of MASLD, with a p-value below 0.001. Strikingly, the signal was detectable at what the authors describe as background exposure levels, around 7.76 nanograms per milliliter of serum, a concentration typical of the general population rather than heavily contaminated communities. This suggests that even ordinary, everyday exposures may be relevant to liver metabolic health.

Epidemiological associations alone cannot reveal mechanism, so the researchers turned to computational biology to map the possible molecular landscape. They first compiled a set of 874 genes that overlap between PFOS toxicity signatures and MASLD-associated genes, then applied two independent machine-learning feature selection algorithms, LASSO regression and support vector machine recursive feature elimination, or SVM-RFE, to prioritize the most discriminating candidates. The convergence of both methods on the same small set of genes is what makes the result noteworthy: five genes, CYP7A1, GRIA3, PHLDA1, SOCS2, and WNT5A, emerged as the core of the study’s proposed immune-metabolic signature. Each of these genes occupies a distinct niche in liver and immune biology, and together they span bile acid synthesis, glutamatergic signaling, inflammatory regulation, and developmental Wnt pathways.

An exploratory classification model built on these five genes achieved an apparent area under the curve of 0.998, with a 95 percent confidence interval of 0.993 to 0.998, within the transcriptomic dataset used to train it. The authors are careful to frame this as an apparent performance measure within the analyzed data rather than a clinically validated diagnostic, a caution that is standard and appropriate for machine-learning models trained and evaluated on the same dataset. Still, the near-perfect separation hints that PFOS and MASLD leave convergent transcriptional fingerprints that such algorithms can detect, and that the five-gene panel could serve as a hypothesis-generating tool for future biomarker studies rather than a ready-made test.

To understand where in the liver these genes act, the team turned to single-cell RNA sequencing data from human liver tissue. The cell-type resolution revealed a striking division of labor. CYP7A1, the rate-limiting enzyme of the classical bile acid synthesis pathway, and PHLDA1, a pleckstrin homology-like domain gene previously implicated as a suppressor of fatty liver progression, were both enriched in hepatocytes, the metabolic workhorses of the liver. SOCS2, a suppressor of cytokine signaling, and WNT5A, a secreted ligand central to fibrogenic and inflammatory crosstalk, mapped primarily to hepatic stellate cells, the fibrosis-driving cells of the liver sinusoid. Meanwhile GRIA3, an ionotropic glutamate receptor subunit more familiar from neuroscience than hepatology, appeared in T cells, pointing to a possible immune dimension in which the chemical’s effects extend beyond hepatocyte lipid handling into adaptive immune signaling.

The researchers next probed whether PFOS could physically interact with the protein products of these five genes using molecular docking, a computational technique that predicts binding poses and affinities between small molecules and protein structures. The docking scores ranged from minus 6.3 to minus 9.3 kilocalories per mole, values in the range typically associated with plausible binding interactions. These are in silico predictions, and docking cannot confirm that such binding occurs in living tissue, but the results are consistent with the idea that PFOS, which is known to bind serum albumin and to activate nuclear receptors, may also engage specific signaling proteins relevant to liver metabolism and inflammation. The docking results provide a structural scaffold for subsequent biochemical validation rather than proof of direct causation.

The in vivo component of the study involved exposing mice to PFOS for twelve weeks and then examining liver tissue for molecular and pathological changes. Reverse transcription quantitative polymerase chain reaction, or RT-qPCR, was used to measure the expression of the candidate genes in PFOS-treated mouse liver, and the observed transcriptional changes were directionally consistent with the computational predictions, providing a modest but meaningful cross-species validation of the gene signature. The exposed mice also developed measurable hepatic lipid accumulation, elevated markers of liver injury, and broader dysregulation of lipid metabolism, phenotypes that echo earlier rodent studies in which PFOS exposure produced steatosis, altered triglyceride handling, and inflammatory stress in liver tissue.

These findings arrive amid a rapidly expanding body of literature on the hepatic effects of per- and polyfluoroalkyl substances, or PFAS, the chemical class to which PFOS belongs. Previous studies have linked PFOS and related compounds to altered liver function biomarkers in exposed human populations, to steatosis in rodents and in human liver spheroids, to disruption of hepatic transporters, and to activation of innate immune pathways such as the AIM2 inflammasome. A recent study of human liver samples found that perfluorooctane sulfonate associates with steatotic liver disease in a sex-dependent manner, and prenatal PFAS exposure has been tied to increased susceptibility to liver injury in children. The new study distinguishes itself not by discovering the association, which was already suspected, but by attempting to trace a mechanistic thread from population statistics down to individual cell types and protein binding sites.

The authors are explicit that their framework is hypothesis-generating rather than definitive. The epidemiological analysis is cross-sectional and cannot establish temporality, the transcriptomic datasets were analyzed retrospectively, the machine-learning performance figures are apparent rather than independently validated, and the docking scores await direct biochemical confirmation. Prospective cohort studies with longitudinal PFOS measurements, experimental studies that directly test binding of the chemical to the candidate proteins, and intervention studies that block or rescue the proposed pathways would all be needed to confirm causation. Nevertheless, the convergence of evidence from a nationally representative human survey, two independent machine-learning approaches, single-cell mapping, docking, and animal experiments creates a coherent and testable immune-metabolic model of how a persistent environmental chemical may contribute to one of the world’s most common liver diseases.

The broader public health implications are considerable. MASLD currently affects hundreds of millions of people and is projected to impose a rising global burden through 2045, driven largely by obesity and diabetes but increasingly recognized to involve environmental contributors, including endocrine-disrupting chemicals. PFOS, though phased out of production in many countries, persists in soil, water, and human serum for years, and exposure continues through contaminated drinking water, food packaging legacy, and dietary routes. If low-level PFOS exposure genuinely contributes to hepatic immune-metabolic disruption, then environmental remediation and exposure limits take on added urgency as liver disease prevention strategies. The five-gene signature identified here may also point toward biomarkers capable of identifying individuals whose liver health is being silently influenced by chemical exposures, long before overt disease develops.

Subject of Research: Association of PFOS exposure with immune-metabolic disruption in metabolic dysfunction-associated steatotic liver disease using integrated epidemiological, computational, and experimental evidence

Article Title: PFOS exposure is linked to immune-metabolic disruption in MASLD: integrated population, computational, and experimental evidence

Article References: PFOS exposure is linked to immune-metabolic disruption in MASLD: integrated population, computational, and experimental evidence. (n.d.). https://doi.org/10.1007/s11030-026-11727-8

Image Credits: AI Generated

DOI: 10.1007/s11030-026-11727-8

Keywords: PFOS, PFAS, MASLD, fatty liver disease, forever chemicals, NHANES, machine learning, single-cell transcriptomics, molecular docking, network toxicology, hepatotoxicity, environmental pollution