Plasma metabolomic and lipidomic alterations associated with anti-tuberculosis drug-induced liver injury

Background: Anti-tuberculosis drug-induced liver injury (ATB-DILI) is an adverse reaction with a high incidence and the greatest impact on tuberculosis treatment. However, there is a lack of effective biomarkers for the early prediction of ATB-DILI. Herein, this study uses UPLC‒MS/MS to reveal the plasma metabolic profile and lipid profile of ATB-DILI patients before drug administration and screen new biomarkers for predicting ATB-DILI. Methods: A total of 60 TB patients were enrolled, and plasma was collected before antituberculosis drug administration. The untargeted metabolomics and lipidomics analyses were performed using UPLC‒MS/MS, and the high-resolution mass spectrometer Q Exactive was used for data acquisition in both positive and negative ion modes. The random forest package of R software was used for data screening and model building. Results: A total of 60 TB patients, including 30 ATB-DILI patients and 30 non-ATB-DILI subjects, were enrolled. There were no significant differences between the ATB-DILI and control groups in age, sex, smoking, drinking or body mass index (p > 0.05). Twenty-two differential metabolites were selected. According to KEGG pathway analysis, 9 significantly enriched metabolic pathways were found, and both drug metabolism-other enzymes and niacin and nicotinamide metabolic pathways were found in both positive and negative ion models. A total of 7 differential lipid molecules were identified between the two groups. Ferroptosis and biosynthesis of unsaturated fatty acids were involved in the occurrence of ATB-DILI. Random forest analysis showed that the model built with the top 30 important variables had an area under the ROC curve of 0.79 (0.65–0.93) for the training set and 0.79 (0.55–1.00) for the validation set. Conclusion: This study demonstrated that potential markers for the early prediction of ATB-DILI can be found through plasma metabolomics and lipidomics. The random forest model showed good clinical predictive value for ATB-DILI.