Cerebrospinal fluid metabolomics identifies a key role of isocitrate dehydrogenase in bipolar disorder: evidence in support of mitochondrial dysfunction hypothesis

Abstract

Although evidence for mitochondrial dysfunction in the pathogenesis of bipolar disorder (BD) has been reported, the precise biological basis remains unknown, hampering the search for novel biomarkers. In this study, we performed metabolomics of cerebrospinal fluid (CSF) from male BD patients (n=54) and age-matched male healthy controls (n=40). Subsequently, post-mortem brain analyses, genetic analyses, metabolomics of CSF samples from rats treated with lithium or valproic acid were also performed. After multivariate logistic regression, isocitric acid (isocitrate) levels were significantly higher in the CSF from BD patients than healthy controls. Furthermore, gene expression of two subtypes (IDH3A and IDH3B) of isocitrate dehydrogenase (IDH) in the dorsolateral prefrontal cortex from BD patients was significantly lower than that of controls, although the expression of other genes including, aconitase (ACO1, ACO2), IDH1, IDH2 and IDH3G, were not altered. Moreover, protein expression of IDH3A in the cerebellum from BD patients was higher than that of controls. Genetic analyses showed that IDH genes (IDH1, IDH2, IDH3A, IDH3B) and ACO genes (ACO1, ACO2) were not associated with BD. Chronic (4 weeks) treatment with lithium or valproic acid in rats did not alter CSF levels of isocitrate, and mRNA levels of Idh3a, Idh3b, Aco1 and Aco2 genes in the rat brain. These findings suggest that abnormality in the metabolism of isocitrate by IDH3A in the mitochondria plays a key role in the pathogenesis of BD, supporting the mitochondrial dysfunction hypothesis of BD. Therefore, IDH3 in the citric acid cycle could potentially be a novel therapeutic target for BD.

Comparison of Resampling Methods for Bias-Reduced Estimation of Prediction Error: A Simulation Study Based on Real Datasets from Biomarker Discovery Studies

“…The coefficient β of F in the logistic distribution, the dimension r of the covariates, the covariance matrix Σ of F , and the sample size n are set according to the cases detailed in Table 1. The intercept coefficient β0 of F equals −1 in the logistic distribution, similar to previous studies (Ogata et al, 2012;Yoshimi et al, 2016). Each independent Bernoulli variable yi of illness follows the logistic distribution.…”

“…In order to simulate conditions in specific biomarker discovery studies, the simulation was based on previously published cases (Ogata et al, 2012;Yoshimi et al, 2016).…”

“…Although evidence of mitochondrial dysfunction in bipolar disorder (BD) pathogenesis has been reported, the precise biological basis remains unknown, thus hampering the identification of novel biomarkers (Yoshimi et al, 2016). In that study, metabolomics was performed on the cerebrospinal fluid (CSF) of male BD patients (n = 54) and age-matched male healthy controls (n = 40).…”

“…In previous studies, we applied LOOCV for the analysis of specific biomarker discovery studies (Ogata et al, 2012;Yoshimi et al, 2016) in which the prediction variables are binary outcomes (see Sections 3.1 and 3.2 for further information regarding these studies). However, the relative performance of the resampling methods, including LOOCV, was not completely clear for the datasets from these studies (involving approximately 10 biomarkers with or without binary biomarkers and various correlations between the biomarkers).…”

“…In this study, we conducted a simulation study to evaluate the performance of the resampling methods in stepwise logistic regression analysis based on simulation settings (including statistical models) derived from two real biomarker discovery studies (Ogata et al, 2012;Yoshimi et al, 2016). In order to incorporate unequal prior relative frequencies between binary outcomes (e.g., disease and non-disease) we considered various cut-off values for the posterior probability in the fitted logistic models when classifying the samples.…”