Domain-specific associations of metabolomic features with cognition and neural networks among cognitively normal adults

Abstract

Background Alterations in systemic metabolism may reflect biological processes relevant to brain function and cognitive aging, yet the extent to which circulating metabolites relate to both cognitive performance and underlying neural network activity remains unclear. Methods The current study included 112 cognitively normal participants of the Reference Abilities Neural Network study at Columbia University who had metabolomic, cognitive, and imaging data available at baseline. Cognition was measured using a neuropsychological battery including three tests for each of the four cognitive domains of memory, reasoning, processing speed, and language. Participants underwent fMRI while performing 12 tasks, 3 per domain, to derive domain-specific neural network activation pattern scores such that patterns correlated positively with behavioral performance. Plasma collected from participants were analyzed using an untargeted liquid chromatography coupled high-resolution mass spectrometry-metabolomics assay to generate metabolic profiles. Linear regression was used to explore associations between metabolite features and outcomes including cognition and neural network activation pattern scores. Results Participants had a mean age of 56.6 years (standard deviation (SD) = 15.6), 50.0% were female, 71.4% were white, and the mean education level was 15.8 years (SD = 2.2). Of the 4,286 metabolite features analyzed, 8 were significantly associated with neural network activation pattern scores for reasoning, but no other domains after false discovery rate (FDR) adjustment. In contrast, no metabolite features were significantly associated with any cognitive domains of memory, reasoning, processing speed or language after FDR adjustment. Exploratory analyses using nominal p-values revealed minimal overlap in metabolite features associated with domain-specific cognitive performance and neural network measures. Six metabolite features were associated with all neural network domains, whereas only one metabolite feature was associated with all cognitive domains. Pathway-level analyses did not demonstrate consistent enrichment across domains, but suggested involvement of amino acid metabolism, lipid metabolism, energy production, cofactor and vitamin metabolism, and nitrogen handling pathways. Conclusion Our findings suggest limited overlap in significant metabolite associations across both cognitive performance and neural network activity, which may reflect some degree of domain specificity. Neural network measures may capture aspects of underlying biological processes that are not fully reflected in cognitive test performance.