High-resolution NMR metabolomics of patients with subjective cognitive decline plus: Perturbations in the metabolism of glucose and branched-chain amino acids

Background: Subjective cognitive decline plus (SCD plus) increases the risk of Alzheimer's disease (AD), and this may be an early stage of AD that precedes amnestic mild cognitive impairment (aMCI). We examined alterations of serum metabolites and metabolic pathways in SCD plus subjects using H-1-magnetic resonance spectroscopy (H-1 NMR) metabolomics. Methods: Serum samples from subjects with SCD plus (n = 32), aMCI (n = 33), and elderly controls (ECs, n = 41) were analyzed using an 800 MHz NMR spectrometer. Multivariate analyses were used to identify serum metabolites, and two machine-learning methods were used to evaluate the diagnostic power of these metabolites in distinguishing SCD plus subjects, aMCI subjects, and ECs. Results: Eight metabolites differentiated SCD plus from EC subjects. A random forest (RF) model discriminated SCD plus from EC subjects with an accuracy of 0.883 and an area under the receiver operating characteristic curve (AUROC) of 0.951. A support vector machine (SVM) model had an accuracy of 0.857 and an AUROC of 0.946. Nine other metabolites distinguished SCD plus from aMCI subjects. An RF model discriminated SCD plus from aMCI subjects (accuracy: 0.975, AUROC: 0.998) and an SVM model also discriminated these two groups (accuracy: 0.955, AUROC: 0.991). Disturbances of glucose and branched-chain amino acid (BCAA) metabolism were the most striking features of SCD plus subjects, and valine was positively correlated with Auditory Verbal Learning Test delayed-recall score. Conclusions: Serum metabolomics using 1H NMR provided noninvasive identification of perturbations in glucose and BCAA metabolism in subjects with SCD plus.

Automated summary

Serum metabolomics using 1H NMR provided noninvasive identification of perturbations in glucose and branched-chain amino acid metabolism in subjects with SCD plus.