Parkinson's disease (PD) is an aging-associated neurodegenerative movement disorder with increasing morbidity and mortality rates. The current gold standard for diagnosing PD is clinical evaluation, which is often challenging and inaccurate. Metabolomics and lipidomics approaches have been extensively applied because of their potential in discovering valuable biomarkers for medical diagnostics. Here, we used comprehensive untargeted metabolomics and lipidomics methodologies based on liquid chromatography-mass spectrometry to evaluate metabolic abnormalities linked with PD. Two well-characterized cohorts of 288 plasma samples (143 PD patients and 145 control subjects in total) were used to examine metabolic alterations and identify diagnostic biomarkers. Unbiased multivariate and univariate studies were combined to identify the promising metabolic signatures, based on which the discriminant models for PD were established by integrating multiple machine learning algorithms. A 6-biomarker predictive model was constructed based on the omics profile in the discovery cohort, and the discriminant performance of the biomarker panel was evaluated with an accuracy over 81.6% both in the discovery cohort and validation cohort. The results indicated that PC (40:7), eicosatrienoic acid were negatively correlated with severity of PD, and pentalenic acid, PC (40:6p) and aspartic acid were positively correlated with severity of PD. In summary, we developed a multi-metabolite predictive model which can diagnose PD with over 81.6% accuracy based on this unique metabolic signature. Future clinical diagnosis of PD may benefit from the biomarker panel reported in this study.
Read full abstract