Plasticizers and prostate cancer: unraveling the link through network toxicology and machine learning
Background Plasticizers, as widespread environmental endocrine disruptors, are increasingly linked to an elevated risk of prostate cancer (PCa). However, the specific molecular mechanisms by which they drive PCa initiation and progression remain incompletely elucidated. Addressing this knowledge gap is crucial for assessing environmental health risks and identifying potential intervention targets. Methods This study employed a multi-level integrated research strategy. First, the toxicological profiles of target plasticizers were predicted using ADMETlab and ProTox platforms. Second, plasticizer-related targets were identified by integrating multiple databases and then cross-referenced with differentially expressed genes in PCa from TCGA and GEO cohorts to obtain shared targets. Subsequently, a protein-protein interaction (PPI) network was constructed and analyzed topologically. GO and KEGG enrichment analyses were performed to explore underlying biological processes and pathways. A total of 98 combination prediction models based on 10 machine learning algorithms were developed and evaluated to identify core prognostic genes. Furthermore, single-cell and spatial transcriptomics data were utilized to examine the expression localization of core genes within the tumor microenvironment. Molecular docking simulations were conducted to validate the binding affinity between plasticizers and core target proteins. Finally, in vitro experiments demonstrated the pro-tumorigenic effects of DMP and its regulatory role in PLK1 expression in prostate cancer cells. Results Toxicity predictions confirmed the carcinogenic potential of DEP, DMP, and DOP. A total of 183 bridging genes connecting plasticizers and PCa were identified. Enrichment analysis revealed their significant involvement in key pathways including inflammatory response, cell cycle, p53 signaling, and chemical carcinogenesis. PPI network analysis preliminarily screened hub genes such as ALB and MMP9. Through systematic machine learning modeling and prognostic analysis, the core targets were further narrowed down to PLK1, ALB, and CCNA2. Among these, high expression of PLK1 was significantly associated with shorter disease-free survival in multiple independent cohorts. Molecular docking results indicated that all three plasticizers could bind stably to the PLK1 protein with high affinity (binding free energy < -5.0 kcal/mol). Single-cell and spatial transcriptomic analyses showed high expression of PLK1 in tumor epithelial cells. In vitro experiments confirmed that DMP promotes the proliferation, migration, and invasion of PCa cells, as well as upregulates PLK1 expression. Pan-cancer analysis further indicated that PLK1 is commonly overexpressed in various cancers and associated with poor prognosis. Conclusion This study integrates computational toxicology, bioinformatics, machine learning, and experimental validation to reveal that common plasticizer exposure may promote PCa progression through dysregulation of cell cycle and inflammatory pathways, with PLK1 identified as a central molecular target. These findings establish a multi-omics evidence chain supporting the carcinogenic potential of environmental endocrine disruptors and provide a scientific basis for considering PLK1 as both a biomarker for risk assessment and a therapeutic target in plasticizer-associated PCa.