A Large-Scale Serum Metabolite Panel for Baseline Detection of Alzheimer's Disease

INTRODUCTIONAccurate Alzheimers disease (AD) detection remains challenging and often requires invasive or costly procedures. Blood-based metabolomic signatures offer a promising non-invasive approach. This study aimed to identify a serum metabolite panel and evaluate its performance alone and in combination with apolipoprotein E (APOE) {varepsilon}4 genotype status for distinguishing AD from cognitively normal (CN) individuals. METHODSBaseline data from 594 participants in the Alzheimers Disease Neuroimaging Initiative (237 AD, 357 CN) were analyzed. High-resolution serum metabolomics (Biocrates MxP(R) Quant 500) and APOE genotype data were used for LASSO-based feature selection, followed by machine learning model training and evaluation on a held-out test set. RESULTSA panel of 151 metabolites distinguished AD from CN with high accuracy (test-set AUC=0.90). Adding APOE to the panel further improved model performance (AUC=0.91 versus AUC=0.75 for APOE alone; p<0.001), achieving strong sensitivity (0.92), specificity (0.84), and negative predictive value (0.94). Key predictive metabolites included bile acids, ether-linked phosphatidylcholines, and acylcarnitines, which are associated with pathways related to lipid metabolism, mitochondrial function, and the gut-liver-brain axis. CONCLUSIONIntegrating serum metabolomics with APOE enables accurate, non-invasive AD detection and offers a scalable screening approach with strong potential to rule out AD in primary care. ClinicalTrials.gov IdentifierNCT00106899 and related ADNI phases.