Integration of vitro models with machine learning and epidemiological data reveals PCB-induced glucose metabolism disruption linked to mitochondrial dysfunction

Polychlorinated biphenyls (PCBs) have been reported to be associated with type 2 diabetes mellitus (T2DM); thus, the knowledge of their endocrine disruption mechanisms would be vital for assessing health risks. This study revealed the potential mechanism of abnormal glucose metabolism due to acute PCB-153 exposure in HepG2 cells through integrated transcriptome and DNA methylation analysis. Based on a joint analysis of two omics, the random forest machine learning model was established with 200 trees and cross-validated five times. Potential biomarkers identified by machine learning pointed to impaired mitochondrial function. Subsequent validation confirmed PCB-153-induced mitochondrial dysfunction, evidenced by reduced mitochondrial DNA copy number (mtDNAcn), adenosine triphosphate (ATP) production, mitochondrial membrane potential, and ATPase activity, alongside altered morphology and elevated Reactive Oxygen Species (ROS). Critically, abnormal glucose metabolism was significantly attenuated and even recovered to control levels after enhancement of mitochondrial function, suggesting that PCB-153 promoted glucose metabolic defects in relation to mitochondrial dysfunction. The decline of mtDNAcn in the T2DM nested case-control population provided further evidence for long-term PCBs exposure inducing mitochondrial dysfunction. In addition, significant multiplicative and additive interactions were observed between mtDNAcn and PCB-138, PCB-153, lowly chlorinated PCBs, highly chlorinated PCBs, ΣNDL-PCBs on the 5-year FBG levels changes (P_interaction: 0.004-0.03; RERI: -0.44 to -0.31; AP: -0.39 to -0.21). Our findings highlighted the importance of maintaining normal mitochondrial function in glucose metabolism of non-dioxin-like PCBs exposure and provided new insights into T2DM pathogenesis caused by PCBs exposure.

DNA methylation; PCBs; glucose metabolism; machine learning; mitochondrial function; mtDNAcn.