Serum metabolic signatures for Alzheimer’s Disease reveal alterations in amino acid composition and energy metabolism – A validation study

Abstract

Abstract Background: Alzheimer’s Disease (AD) is complex and novel approaches are urgently needed to characterise disease pathology and to aid in diagnosis. Metabolites are the end-products of upstream molecular alterations, whereby small changes at the genetic or protein level result in substantial changes at the metabolite level. Blood is frequently used as a source for biomarkers; however, its complexity prevents proper detection. The analytical power of metabolomics, coupled with statistical tools, can assist in reducing this complexity. Furthermore, the current bottleneck in biomarker research is reproducibility and appropriate validation. Thus, we sought to validate a previously proposed panel of metabolic blood-based biomarkers for AD and expand our understanding of the pathological mechanisms involved in AD that are reflected in the blood. Methods: In the validation cohort serum and plasma were collected from 25 AD patients and 25 healthy controls. Serum was analysed for metabolites using nuclear magnetic resonance (NMR) spectroscopy, while plasma was tested for markers of neuronal damage and AD hallmark proteins using single molecule array (SIMOA). A combination of multivariate and univariate statistics were utilized to validate established biomarkers and uncover new disease-related evidence. Results: The diagnostic performance of the proposed metabolite biomarker panel was confirmed using sparse-partial least squares discriminant analysis (sPLS-DA) with an area under the curve (AUC) of 0.89 (95 % confidence interval: 0.79 – 0.98). Five metabolites (pyruvic acid, valine, leucine, histidine, and isoleucine) were consistently reduced in both the discovery and validation cohorts. Pathway analysis of significantly altered metabolites in the validation set revealed that they are involved in branched-chain amino acids (BCAAs) and energy metabolism (glycolysis and gluconeogenesis). Additionally, a moderate correlation was observed between valine and the proteins neurofilament light and glialfibrillary acidic protein. By combining the significant protein expression levels measured by SIMOA with the sPLS-DA model, the AUC increased to 0.97 (95 % CI: 0.93 – 1.00). Conclusions: Our proposed panel of metabolites was successfully validated using a combined approach of NMR and sPLS-DA. It was discovered that cognitive-impairment-related metabolites belong to BCAAs and are involved in energy metabolism.