PFOS is a ubiquitous environmental pollutant with suspected lung carcinogenic potential. Perfluorooctanesulfonate (PFOS) is among the most extensively produced per- and polyfluoroalkyl substances (PFAS), often called forever chemicals due to their extreme environmental persistence and resistance to metabolic clearance. PFOS bioaccumulates in human tissues and has documented hepatotoxicity, endocrine-disrupting effects, and immunomodulatory potential. Epidemiological evidence links PFAS mixtures to multiple malignancies, and the lung, as a highly vascularized organ and primary interface for environmental interaction, is a plausible site of PFOS bioaccumulation and toxicity.
The molecular mechanisms linking PFOS to NSCLC are poorly defined. While suggestive epidemiological associations between PFOS exposure and lung cancer risk have been reported, the evidence is heterogeneous across cohorts and exposure metrics, partly because real-world PFAS research is complicated by co-exposure mixtures, residual confounding from smoking, and inconsistent exposure assessment. Prior toxicological research has disproportionately focused on hepatic and metabolic endpoints, leaving the PFOS-associated molecular landscape in lung tissue largely uncharted.
PFOS mimics fatty acids and perturbs PPAR and mTOR signaling. Mechanistically, PFOS acts as a structural analogue to fatty acids, disrupting lipid metabolism and peroxisome proliferator-activated receptor (PPAR) signaling. Emerging evidence indicates that such metabolic perturbations crosstalk with the PI3K-Akt-mTOR axis, a master regulator of cell growth and translational control that is frequently hyperactivated in lung adenocarcinoma. Downstream effectors of this metabolic-translational interface, particularly those governing cap-dependent translation, represent potential unverified convergence points where PFOS-driven environmental stress synergizes with intrinsic oncogenic signaling.
An integrated computational framework was needed to triangulate molecular mechanisms. This study combined network toxicology, multi-omics transcriptomics, and genetic causal inference to systematically identify the molecular hub genes connecting PFOS exposure to NSCLC pathogenesis. The framework hypothesized that PFOS perturbs LUAD-relevant metabolic-translation-immune axes, converging on a limited set of hub genes that could be prioritized for experimental validation and risk assessment.
Multi-database chemical-protein interaction screening identified 256 PFOS targets. The canonical SMILES representation of PFOS was queried against three independent toxicological databases: ChEMBL, STITCH (using a combined score threshold of 0.4 for medium confidence), and SwissTargetPrediction (retaining all targets with non-zero probability scores). The resulting target lists were merged and deduplicated to generate 256 non-redundant PFOS-associated molecular targets in Homo sapiens.
Cross-referencing PFOS targets with 2,173 NSCLC-associated genes yielded 41 shared targets. NSCLC-associated genes were compiled from OMIM and GeneCards databases and deduplicated to 2,173 non-redundant entries. Intersection analysis identified 41 shared targets fulfilling both criteria of documented chemical interaction with PFOS and established disease relevance in NSCLC, specifically enriched in PPAR signaling and lipid-sensing modules that are canonical hallmarks of PFAS-mediated metabolic disruption.
LASSO and SVM-RFE machine learning algorithms independently selected three core targets. Both algorithms were applied to the 41 shared targets using GSE33532 transcriptomic data from paired NSCLC and normal lung tissues. LASSO regression with 10-fold cross-validation identified five candidate genes, while support vector machine with recursive feature elimination and a radial basis function kernel selected three. The intersection of both independent methods identified three core targets: CD36, ABCB1, and EIF4EBP1, a consensus strategy designed to minimize overfitting and enhance reproducibility.
Two-sample Mendelian randomization and molecular docking tested causal plausibility. Mendelian randomization used cis-expression quantitative trait loci from the eQTLGen Consortium as genetic instruments for EIF4EBP1 expression, with outcome data for LUAD and LUSC from the FinnGen R12 release. Linkage disequilibrium clumping used a European reference panel with stringent independence thresholds. Molecular docking simulations were performed using AutoDock Vina to assess the structural basis of direct PFOS-EIF4EBP1 interaction.
EIF4EBP1 achieved an AUC of 0.936 in the discovery cohort. Differential expression analysis in the GSE33532 cohort confirmed that EIF4EBP1 was significantly upregulated in NSCLC tissues compared to normal lung, while CD36 and ABCB1 were markedly downregulated. ROC curve analysis confirmed strong diagnostic utility for all three core genes, with EIF4EBP1 achieving the highest AUC of 0.936, indicating highly accurate discrimination of tumor from normal tissue in the discovery dataset.
TCGA validation confirmed EIF4EBP1 upregulation across both NSCLC subtypes. Independent external validation in the TCGA cohort confirmed that EIF4EBP1 was significantly elevated in both lung adenocarcinoma (LUAD, n=483) and lung squamous cell carcinoma (LUSC, n=486) compared to matched normal controls. This cross-cohort and cross-platform consistency reinforced the biological significance of EIF4EBP1 upregulation as a robust feature of NSCLC rather than a dataset-specific artifact.
Survival analysis revealed striking subtype-specific prognostic divergence. Kaplan-Meier analysis in the TCGA cohort demonstrated that high EIF4EBP1 expression significantly correlated with worse overall survival in LUAD patients (HR = 1.9, 95% CI: 1.2-2.9, p = 0.0052), identifying it as a risk factor in this subtype. Paradoxically, high EIF4EBP1 expression correlated with favorable prognosis in LUSC (HR = 0.67, p = 0.047), indicating that the same molecule plays fundamentally different roles in the two NSCLC subtypes.
EIF4EBP1 was prioritized over CD36 and ABCB1 as a translational switch. CD36 and ABCB1 function primarily in xenobiotic transport and lipid uptake, respectively. EIF4EBP1 functions as a critical convergence point downstream of mTOR where oncogenic signaling and environmental metabolic stress intersect to regulate cap-dependent protein translation. This biological centrality at the mTOR-PI3K-Akt signaling axis made EIF4EBP1 the most plausible candidate molecular initiating event connecting PFOS exposure to malignant transformation.
NSCLC displays an adaptive immune-skewed tumor microenvironment. CIBERSORT immune deconvolution of NSCLC samples revealed significantly increased infiltration of plasma cells, activated CD4+ memory T cells, follicular helper T cells, gamma-delta T cells, and M1 macrophages compared to normal lung. Conversely, innate effectors including monocytes, NK cells, neutrophils, and eosinophils were substantially depleted, reflecting a shift toward a dysregulated adaptive immune profile within the tumor microenvironment.
EIF4EBP1 expression specifically tracks the adaptive immune infiltration signature. Among the three core genes, EIF4EBP1 exhibited a distinct positive correlation with adaptive immune components including plasma cells, activated CD4+ memory T cells, and M1 macrophages, while correlating negatively with monocytes and M2 macrophages. CD36 and ABCB1 instead correlated with myeloid and resting lymphocyte subsets. This EIF4EBP1-specific immune correlation pattern suggests that its overexpression in the tumor milieu is mechanistically linked to adaptive immune remodeling in NSCLC.
Mendelian randomization demonstrated causal effect on LUAD but not LUSC risk. Using genetic instrument rs28565141 (F-statistic = 28.622 confirming instrument strength), genetically predicted higher EIF4EBP1 expression was causally associated with increased LUAD risk (OR = 4.196, 95% CI: 1.209-14.565, p = 0.024). In contrast, no significant causal effect was detected for LUSC (OR = 0.878, p = 0.851). This subtype-specific causal signal from genetic epidemiology is consistent with the known predominance of mTOR pathway dysregulation in adenocarcinoma.
LUAD-specific vulnerability is grounded in its alveolar type II cell of origin. The LUAD-specific causal association is biologically plausible because LUAD originates from alveolar type II cells, which possess specialized metabolic and translational machinery required for surfactant production that is heavily reliant on lipid homeostasis. PFOS, by mimicking fatty acids and disrupting PPAR-mediated lipid signaling, would preferentially perturb this metabolic machinery, synergizing with EIF4EBP1-mediated mTOR dysregulation to drive adenocarcinoma rather than squamous cell carcinoma pathogenesis.
PFOS forms a stable non-covalent complex with EIF4EBP1 at -7.2 kcal/mol. Molecular docking simulations predicted that PFOS occupies a shallow groove on the EIF4EBP1 protein surface in an extended conformation. The binding interface is stabilized by a cooperative network of interactions: hydrogen bonds with the side-chain amide of Gln187, the carboxylate of Asp44A, and the amine of Lys88A (with H-bond distances of 1.9-3.1 Angstroms); hydrophobic van der Waals contacts from the fluoroalkyl tail; and parallel pi-pi stacking with the indole rings of Trp43A and Trp89A. This favorable docking energy supports structural plausibility for PFOS as a direct modifier of EIF4EBP1 function.
An adverse outcome pathway maps the mechanistic sequence from PFOS to NSCLC. The adverse outcome pathway (AOP) framework constructed from the multi-omic findings defines: the molecular initiating event as PFOS directly binding EIF4EBP1 and nuclear receptors such as PPARs; intermediate key events including activation of PI3K-Akt-mTOR signaling, EIF4EBP1-centered translational reprogramming, and immune microenvironment remodeling toward a dysregulated adaptive state; and the adverse outcome of NSCLC initiation, progression, and potential therapy resistance.
The translational switch model explains PFOS-to-LUAD carcinogenesis. In this mechanistic model, PFOS-induced PPAR disruption and lipid metabolic stress synergize with mTOR hyperactivation already present in LUAD precursor cells. EIF4EBP1, as a downstream mTOR effector regulating cap-dependent translation, becomes hyperactivated and promotes synthesis of oncogenic proteins that drive proliferation, immune evasion, and disease progression. This hypothesis positions EIF4EBP1 as the molecular node where environmental chemical exposure converges with intrinsic cancer signaling.
Experimental validation in toxicological models is required before clinical translation. The authors identified essential next steps for validating the computational findings: biophysical assays such as surface plasmon resonance to quantify PFOS-EIF4EBP1 binding affinity; in vitro studies in lung adenocarcinoma cell lines measuring EIF4EBP1 phosphorylation state and cap-dependent protein synthesis; and EIF4EBP1-deficient animal models in lung carcinogenesis studies to confirm whether the adverse outcomes are mechanistically driven by this hub. Large-scale human biomonitoring cohorts linking PFOS exposure measurements to EIF4EBP1 expression are also needed.
EIF4EBP1 links PFOS exposure to LUAD through a lipid-metabolic-translational axis. This study constructs a novel EIF4EBP1-centered mechanistic narrative for PFOS-induced lung toxicity, proposing that PFOS promotes LUAD specifically by exploiting the intersection of lipid metabolic signaling and mTOR-mediated translational control. The converging evidence from network toxicology, transcriptomics, Mendelian randomization, and structural docking provides a data-driven rationale for considering PFOS as a subtype-specific risk factor for lung adenocarcinoma.
Subtype specificity has important implications for risk assessment frameworks. The finding that EIF4EBP1 drives opposing prognostic outcomes in LUAD versus LUSC, with causal genetic support specifically for adenocarcinoma, underscores the importance of distinguishing NSCLC subtypes when evaluating carcinogenic risks of environmental chemical exposures. Current PFAS risk assessment frameworks that treat NSCLC as a uniform endpoint may substantially underestimate or mischaracterize subtype-specific carcinogenic risks.
Limitations require cautious interpretation of the computational findings. Key limitations include: the use of a single genetic instrument for Mendelian randomization requiring caution regarding horizontal pleiotropy; reliance on bulk transcriptomic data that cannot resolve EIF4EBP1 expression within specific tumor subclones or distinct immune niches; the fact that immune patterns observed likely reflect general tumor-associated remodeling rather than confirmed PFOS-specific effects; and the inherently computational nature of all molecular interaction findings pending biophysical validation. The proposed AOP framework should be interpreted as a hypothesis-generating conceptual model rather than a definitive mechanistic map.