# Unraveling the PFOS-NSCLC axis: integrated network toxicology, machine learning, and causal inference identify EIF4EBP1 as a key molecular hub

## TL;DR

This study identifies EIF4EBP1 as a key gene linking PFOS exposure to lung cancer, using machine learning and causal analysis to explore how this pollutant may contribute to cancer development.

## Contribution

The study introduces a novel integration of network toxicology, machine learning, and causal inference to identify EIF4EBP1 as a central hub in PFOS-related non-small cell lung cancer.

## Key findings

- EIF4EBP1 is significantly upregulated in both discovery and validation cohorts, with subtype-specific prognostic value in lung cancer.
- Mendelian randomization suggests a causal link between EIF4EBP1 expression and increased risk of lung adenocarcinoma.
- Molecular docking confirms a stable interaction between PFOS and the EIF4EBP1 binding pocket.

## Abstract

Perfluorooctanesulfonic acid (PFOS) is a persistent environmental pollutant with suspected carcinogenic potential; however, the molecular mechanisms driving PFOS-associated non-small cell lung cancer (NSCLC) remain obscure. In particular, the interplay between chemical exposure, oncogenic signaling nodes, and tumor microenvironment (TME) remodeling is poorly defined. This study integrates systems toxicology with multi-omics to elucidate the role of EIF4EBP1 as a mechanistic bridge connecting PFOS exposure to NSCLC pathogenesis.

We synthesized chemical–protein interactions from toxicological databases (ChEMBL, STITCH, and SwissTargetPrediction) and disease-associated genes to map the PFOS–NSCLC intersection. Robust feature selection, utilizing LASSO and SVM-RFE algorithms, was applied to transcriptomic data from the GSE33532 discovery cohort to identify core targets. Key findings were substantiated through external validation in The Cancer Genome Atlas (TCGA) dataset, including differential expression and survival analyses. Causal associations were investigated via two-sample Mendelian randomization (MR), and the immune landscape was characterized using the CIBERSORT algorithm. Molecular docking simulations and an adverse outcome pathway (AOP) framework were further employed to assess mechanistic plausibility.

Network analysis identified 41 shared targets significantly enriched in PPAR signaling and xenobiotic metabolism. Machine learning consensus prioritized EIF4EBP1 as a critical hub gene. EIF4EBP1 was significantly upregulated in both the discovery (AUC = 0.936) and TCGA validation cohorts. Clinical analysis revealed subtype-specific prognostic value, where high EIF4EBP1 expression correlated with poor survival in lung adenocarcinoma (LUAD) but favorable outcomes in squamous cell carcinoma (LUSC). Immunologically, EIF4EBP1 expression tracked with an adaptive immune-skewed profile, characterized by increased plasma cell and activated CD4 + memory T cell infiltration. MR analysis indicated a potential causal effect of genetically predicted EIF4EBP1 expression on increased LUAD risk (OR = 4.196, 95% CI: 1.209–14.565), but not squamous cell carcinoma. Structural docking confirmed a stable, non-covalent interaction between PFOS and the EIF4EBP1 binding pocket (−7.2 kcal/mol).

This study identifies EIF4EBP1 as a putative molecular initiating node linking PFOS exposure to LUAD susceptibility and immune modulation. The constructed AOP framework suggests a mechanism wherein PFOS-mediated translational dysregulation contributes to subtype-specific carcinogenesis. These findings provide a data-driven rationale for risk assessment and warrant further experimental verification in toxicological models.