Plasma multi-miRNA models classify Alzheimer’s, Parkinson’s, and Lewy body dementia

Le résumé fourni par la source

Introduction: Differentiating Alzheimer's disease (AD) from both Parkinson's disease (PD) and Lewy body dementia (DLB) is difficult due to their clinical similarities. Plasma biomarkers offer an alternative to neuroimaging and cerebrospinal fluid analysis; however, there are limitations with respect to differential diagnosis of AD, PD, and DLB with current clinical assays. Methods: Here we used machine learning to assess plasma miRNAs for their specificity in diagnosing AD vs. PD, and DLB. Our multi-center study assayed 57 AD-associated miRNAs in human plasma from 82 cognitively normal controls (NC), 87 AD, 100 PD, and 20 DLB. Predictive models generated by three independent machine learning methods were evaluated by cross-validated ROC curves [cvAUC (bootstrap bias-corrected 95% CI)]. We also used linear discriminant analysis with all 57 miRNAs to identify a model that best separates AD from PD + DLB participants and DLB from AD+PD participants. Further, we used Target prediction and Ingenuity Pathway Analysis to identify highly relevant mRNA targets of the miRNAs. Results: Individual assessment of the 57 miRNAs identified a subset of 10 miRNAs that were more AD-specific, and 23 miRNAs that were more PD and/or DLB associated. Ridge logistic regression predictive models with the 10 AD-specific miRNAs had good performance for separating AD vs. PD and DLB (cvAUC = 0.77 [0.70, 0.83]) and AD vs. PD (cvAUC = 0.79 [0.71, 0.85]), but not for AD vs. DLB (cvAUC = 0.58 [0.43, 0.79]). By developing a predictive model using data from all 57 miRNAs and elastic-net regression we achieved good separation of AD from PD (cvAUC = 0.80 [0.72, 0.86]) and DLB (cvAUC = 0.77 [0.64, 0.87]) with a subset of six miRNAs (miRs-19a-3p, 22-3p, 92b-3p, 101-3p, 143-3p, 423-5p) identified as the most important to these models. The linear discriminant analysis model achieved very good classification of AD vs. PD + DLB (cvAUC = 0.94 [0.87, 0.97]), PD from AD+DLB (cvAUC = 0.88 [0.80, 0.92]), and DLB from AD+PD (cvAUC = 0.85 [0.78, 0.89]) with miR-26a-5p and 146a-5p being most important for AD vs. PD + DLB, and miR-142-3p and 101-3p being most important for DLB vs. AD+PD. Target prediction and Ingenuity Pathway Analysis with miR-26a-5p, 146a-5p, 142-3p, 101-3p returned highly relevant mRNA targets associated with tauopathy, dementia, and movement disorders. Discussion: These data demonstrate that predictive modeling using plasma miRNA expression data may improve the differential diagnosis of AD from PD from DLB.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.