Transcriptomics-informed systems toxicology prioritizes candidate links between endocrine-disrupting chemicals and chronic kidney disease.

Abstract

Endocrine-disrupting chemicals (EDCs) are environmentally relevant chemicals that may contribute to chronic kidney disease (CKD), but their molecular links to CKD remain difficult to infer from exposure-independent transcriptomic data. We developed a transcriptomics-informed systems toxicology workflow to prioritize CKD-relevant candidate genes associated with predicted human protein targets of 12 selected representative EDCs compiled from ChEMBL, SwissTargetPrediction and PharmMapper. GEO datasets GSE37171 and GSE66494 were used to identify CKD-associated transcriptomic changes. Differential expression analysis identified 1,080 significant genes, and WGCNA selected three disease-associated modules. Their intersection yielded 813 CKD-related genes, of which 78 overlapped predicted targets of the selected EDCs. Functional annotation of the 78-gene candidate pool highlighted xenobiotic-response and leukocyte-migration processes, together with MAPK-, NF-kB- and HIF-1-related pathway annotations. A nested-cross-validation Elastic Net workflow with training-fold-restricted preprocessing and feature selection was followed by discovery-stage stability assessment, dataset-stratified expression evaluation and additive contribution analysis. This refinement retained FOS and MDK as the final candidate panel. The FOS/MDK model achieved a combined external-evaluation AUC of 0.797 (95% CI, 0.721-0.874), with external AUCs of 0.787 in GSE104948 and 0.974 in GSE104954. FOS was consistently lower and MDK consistently higher in CKD across all four GEO datasets, and complementary Nephroseq V5 cohort-level summaries supported higher MDK and lower FOS expression across additional human kidney-disease comparisons. Exploratory molecular docking provided in silico pose-prediction context for the prioritized proteins and selected EDC ligands; among all tested protein-ligand pairs, MDK showed the most favorable Vina scores with benzo[a]pyrene, PFOA and bisphenol A, whereas FOS showed its most favorable predicted pose with benzo[a]pyrene. Overall, this workflow prioritizes an externally evaluated FOS/MDK diagnostic-candidate panel and testable hypotheses for future exposure-aware experimental and epidemiological validation.