Integrated Adverse Outcome Pathway and Machine Learning Reveal the Mode of Action and Biomarkers of Lithium-Induced Risk of Chronic Kidney Disease

Abstract

Long-term lithium exposure may lead to chronic kidney disease (CKD), but its mode of action (MOA) and biomarkers remain unclear. This study aimed to explore the potential mechanisms by integrating an adverse outcome pathway (AOP) framework and machine learning. A chemical-gene-phenotype-disease network (CGPDN) was used to construct an AOP framework through network toxicology. Drug enrichment analysis, molecular docking, and in vitro experiments were conducted to identify potential targeted drugs. Moreover, machine learning was used to identify potential noninvasive biomarkers. The AOP network was constructed from 14 key targets and 12 key phenotypes that were screened via CGPDN. In this AOP framework, “response to metal ion” was characterized as the molecular initiating event. Furthermore, “oxidative stress” and “Increase, Cell death” were identified as central key events. Drug enrichment analysis and molecular docking revealed that simvastatin is a potential target drug. In addition, lithium exposure induced abnormal expression of 7 key targets in HK-2 cells, while simvastatin pretreatment reversed the changes in these targets and reduced renal toxicity. The expression patterns of FAS, HSP90B1, and CASP1 were consistent with those observed in the CKD kidney and PBMC transcriptomic data sets. Further, a machine learning model and Shapley Additive exPlanations analysis confirmed that the three targets were potential blood biomarkers of lithium-induced CKD. This study developed a novel paradigm integrating AOP with machine learning, which elucidated the MOA of lithium-induced CKD and filled the gap in early-warning biomarkers. Our findings reveal simvastatin as a promising intervention drug and provide a feasible framework for early risk assessment in lithium-exposed populations.