Comparison of multi-miRNA models from donor-matched plasma and cerebrospinal fluid as candidate biomarkers for Alzheimer’s disease prediction

We discovered and validated miRNA biomarkers for Alzheimer's disease (AD) in human cerebrospinal fluid (CSF). However, as more easily accessible biofluids are preferred for clinical assays, we compared the performance of the AD miRNAs in CSF to plasma to determine their potential use as AD biomarkers in this readily available biofluid. We obtained 320 donor- and date-matched normal control (NC) and AD CSF and plasma samples from the AD Neuroimaging Initiative, then analyzed 57 candidate AD miRNAs in both biofluids by RT-qPCR. For analysis, we divided the sample sets into 80% for discovery and 20% for validation. We then used predictive modeling of the 57 candidate AD miRNAs in the discovery phase to develop AD classifiers for each biofluid that showed similar performance in both CSF and plasma. However, in the validation phase AD classification performance was only maintained for the plasma models. When the plasma miRNA models were combined with clinical predictors (APOE genotype, age, sex, years of education) there was a boost in classification performance to a level comparable to the CSF proteins (Aβ₄₂, t-tau, p-tau181) alone, which suggests that plasma miRNA assays combined with clinical information could serve as an alternative to more invasive CSF-based measurements. We also developed a parsimonious model based on two miRNAs that were top ranked and had increased (miR-23a-3p) and decreased (miR-423-5p) expression in AD. The ratio of these top two miRNAs classified AD with similar performance to all 57 miRNAs jointly and was sensitive to changes in cognitive score. These results support that an assay consisting of only two plasma miRNAs can classify AD from NC, and is informative about the disease stage reflected in cognitive scores.