An integrated computational and in vitro framework identifies CTSB and MMP7 as DEHP-responsive candidates in gastric adenocarcinoma

Abstract

Di(2-ethylhexyl) phthalate (DEHP) is a widespread endocrine-disrupting chemical, but its molecular effects in gastric adenocarcinoma remain incompletely understood. This study integrated network toxicology, multi-cohort transcriptomics, weighted gene co-expression network analysis, protein–protein interaction analysis, functional enrichment, machine learning, SHAP interpretation, molecular docking, molecular dynamics simulation, single-cell analysis and in vitro evaluation. Three GEO datasets comprising 252 samples were used for transcriptomic discovery and model development, three independent datasets comprising 317 samples were used for model selection, and GSE66229 comprising 400 samples was retained as a locked external test cohort. MKN-45 cells were exposed to vehicle, 100 or 200 µM DEHP for 24 h and evaluated using CCK-8, RT–qPCR and western blotting. Differential expression analysis identified 541 significant genes, and WGCNA prioritised 172 tumour-associated genes. Intersection with 189 predicted DEHP-related targets yielded seven candidate genes: BCL2A1, CCR1, CTSB, CTSK, MMP7, MMP9 and PDE4B. Among 127 prespecified machine-learning configurations, Gaussian naive Bayes achieved a development cross-validation AUC of 0.873 and a model-selection macro-average AUC of 0.853. In the locked GSE66229 cohort, the frozen model retained favourable discrimination, with an AUC of 0.866 (95% CI 0.828–0.899), although probability calibration and threshold transport were limited. CTSB showed the highest locked single-gene AUC of 0.906, whereas SHAP analysis ranked PDE4B, CTSK and CTSB as the leading contributors to multigene predictions. Molecular docking and 50-ns molecular dynamics simulations indicated structurally plausible interaction modes, while single-cell mapping placed the candidate genes in distinct tumour-microenvironment compartments. In MKN-45 cells, CCK-8 analysis showed a descriptive decrease in normalized viability under the tested exposure conditions, while RT-qPCR and western blotting showed increased CTSB and MMP7 expression at both the mRNA and protein levels; PDE4B remained unchanged. These findings support CTSB and MMP7 as preliminary DEHP-responsive candidates in an MKN-45 gastric adenocarcinoma cell model under high-dose, short-term in vitro exposure conditions.