Metabolic perturbation has been associated with depression. An untargeted metabolomics approach using liquid chromatography-high resolution mass spectrometry was employed to detect and measure the rat serum metabolic changes following chronic social isolation (CSIS), an animal model of depression, and effective antidepressant fluoxetine (Flx) treatment. Univariate and multivariate statistics were used for metabolic data analysis and differentially expressed metabolites (DEMs) determination. Potential markers and predictive metabolites of CSIS-induced depressive-like behavior and Flx efficacy in CSIS were evaluated by the receiver operating characteristic (ROC) curve, and machine learning (ML) algorithms, such as support vector machine with linear kernel (SVM-LK) and random forest (RF). Upregulated choline following CSIS may represent a potential marker of depressive-like behavior. Succinate, stachydrine, guanidinoacetate, kynurenic acid, and 7-methylguanine were revealed as potential markers of effective Flx treatment in CSIS rats. RF yielded better accuracy than SVM-LK (98.50% vs. 85.70%, respectively) in predicting Flx efficacy in CSIS vs. CSIS, however, it performed almost identically in classifying CSIS vs. control (75.83% and 75%, respectively). Obtained DEMs combined with ROC curve and ML algorithms provide a research strategy for assessing potential markers or predictive metabolites for the designation or classification of stress-induced depressive phenotype and mode of drug action.
Read full abstract