Title: “A Metabolomic Approach to Phosphorylated Tau 217”
Introduction: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by accumulation of two biologic classifications, amyloid-beta plaques and tau neurofibrillary tangles. Early detection of AD pathology is critical, and blood-based biomarkers offer a minimally invasive path toward that goal. Phosphorylated tau 217 (pTau217) is a “Core 1” biomarker that tracks amyloid pathology in early disease stages and tau pathology in later stages, making it a promising indicator of preclinical AD. Metabolomics is the measurement of small molecules along metabolic pathways that offers an additional window into disease processes and may complement existing biomarker approaches. Plasma metabolites are particularly attractive because they are easily and relatively non-invasively obtained.
Objective: This study aims to use plasma metabolomic data to understand plasma pTau217 positivity in cognitively unimpaired and at-risk adults, evaluating select metabolites for a pathway analysis.
Methods: Data will be drawn from two Wisconsin cohorts: the Wisconsin Registry for Alzheimer’s Prevention (WRAP) and the Wisconsin Alzheimer’s Disease Research Center (ADRC). Fasting plasma samples underwent untargeted metabolomics analysis. After quality control, batch correction, and imputation, data will be split into training and test sets. A LASSO Cox regression framework with cross-validation will be used to identify “important” metabolites. A subsequent analysis is planned to better understand if these metabolites hold any meaningful relationships.
Results: AUROC value of the training model initially showed 0.798, when applied to the test data, showed only 0.159. After digging into the model, an internal, initially unaccounted for cross validation was leaking repeated measures data at the individual level across folds, thus over inflating the results. Correction of this lead to null results with the current parameters due to a severe decrease in sample size per fold and subsequently drop in statistical power. Although the training model showed promise, it appears to be artifact rather than true results. This project is in the results interpretation phase.
