A new perspective on the neurotoxic mechanisms of six typical per- and polyfluoroalkyl substances (PFAS): insights from integrating network toxicology and random forest algorithm.

Abstract

Per- and polyfluoroalkyl substances (PFAS) are widely used in various industries but pose significant ecological and human health risks, particularly to the nervous system. However, the underlying neurotoxic mechanisms remain poorly understood. This study combines network toxicology and machine learning to explore these mechanisms. Using ADMETLAB 3.0, we assessed the environmental toxicity of six common PFAS and identified their potential targets using online tools. A compound-target interaction network was built, followed by protein-protein interaction (PPI) and KEGG pathway analyses to investigate toxicological pathways. Core targets were selected through machine learning, and differential gene expression was analyzed using transcriptomic data. Molecular docking simulations predicted binding affinities between PFAS and their core targets, while molecular dynamics simulations on key complexes were performed using Gromacs 2023.2 and the Charmm36 force field. PFDS showed the highest bioconcentration factors (BCF), while PFOA demonstrated the greatest toxicity. We identified 62 intersecting targets, with PTGS2, MMP9, and ESR1 being central in the PPI network. Transcriptomic analysis revealed 1,077 differentially expressed genes (DEGs), highlighting associated biological processes and pathways. The random forest model identified 20 core genes, with 9 significantly differentially expressed in the PFAS-treated group. Molecular docking suggested potential interactions between the compounds and core targets, and molecular dynamics simulations further supported the stability of the complexes under physiological conditions. This study provides valuable insights into the neurotoxic mechanisms of PFAS, enhancing our understanding of their impact on the nervous system.