Genetic Background, Metabolomics, and the Mediterranean Diet Influence Dementia Risk
A new study published in Nat med on February 28, 2025 (DOI: 10.1038/s41591-025-03891-5) investigated the interplay between genetic predisposition, plasma metabolomic signatures, and adherence too a Mediterranean diet (MedDiet) in relation to dementia risk and cognitive function. Researchers analyzed data from two large, long-term cohorts: the Nurses’ Health Study (NHS) and the Health Professionals Follow-up Study (HPFS).
The study found that the association between the MedDiet and dementia risk varied based on APOE4 genotype. Specifically, an inverse relationship between the MedDiet and the amino acid asparagine was observed only in individuals homozygous for the APOE4 allele. A similar trend, though not statistically significant after correcting for multiple comparisons (FDR), was seen with the caffeine metabolite 1,7-dimethyluric acid in APOE4 carriers.
Mediation analysis revealed that 39.5% of the association between the MedDiet and dementia was statistically significant only in APOE4 carriers. This mediation was explained by seven metabolites: allantoin, C16:1 cholesteryl ester (CE), 1-methylguanine, C18:0 sphingomyelin (SM), 1,7-dimethyluric acid, C34:5 phosphatidylcholine plasmalogen, and piperine. No such mediation effect was observed in individuals without the APOE4 allele.
The researchers also assessed whether integrating omics data with lifestyle factors improved dementia prediction. In the NHS cohort, Harrell’s C-index – a measure of predictive accuracy – showed incremental improvements with the addition of APOE4 status, an Alzheimer’s Disease Risk (ADRD) polygenic risk score (PRS), and selected metabolites to a baseline model including age, education, family history, smoking, depression, and MedDiet adherence. similar incremental gains in predictive power were observed in the HPFS cohort using Cox models. Shapley Additive Explanations (SHAP) analysis identified age, APOE4 status, PRS, and metabolites as the most influential factors, with diet and profession also contributing to the model.Further investigation using Mendelian randomization (MR) and colocalization analyses suggested potential causal relationships between specific metabolites and cognitive function or Alzheimer’s Disease. Metabolites showing protective signals included 4-guanidinobutanoate (4-GBA), a metabolite related to gamma-aminobutyric acid (GABA); carotene diol (1) and carotene diol (2); and glutamine. These metabolites are linked to mechanisms involved in excitotoxicity control, redox defenses, and neurotransmission.
The study’s findings suggest that an individual’s genetic background influences which metabolic pathways are associated with dementia, and that the MedDiet may positively influence these pathways, particularly in APOE4 homozygotes.
Conclusions
This research establishes links between genetics, plasma metabolomic profiles, and MedDiet adherence in relation to dementia and cognitive performance. The data indicate that cholesteryl esters (CEs) and sphingomyelins (SMs) clustered as risk factors in APOE4 homozygotes, while the MedDiet correlated with favorable glyceride profiles.Selected metabolites were identified as mediating protective effects. While prediction accuracy improved only modestly, the biological insights suggest a potential for precision prevention strategies, including genotype-informed counseling and dietary guidance targeting lipid and one-carbon pathways.
The authors recommend replication of these findings in more diverse populations, longitudinal metabolomic sampling, and intervention trials to evaluate metabolite-guided MedDiet strategies. Such research could facilitate earlier intervention, personalized dietary choices, and improved clinical outcomes for individuals at risk of dementia.
The study acknowledges several limitations, including the predominantly european ancestry of the participants, the reliance on a single baseline metabolomic measurement, and the observational study design, which utilizes composite dementia outcomes.
The study was led by Y. Liu, X. Gu, Y. li, F. Wang,C.M. Vyas, C. Peng,D. Dong, Y. Li, Y.Zhang, Y.Zhang, O.A. Zeleznik, J.H. Kang, M. Wang, F.B. Hu, W.C. Willett, O.I. Okereke,A.H. Eliassen, P. Kraft, M.J. Stampfer, and D.D. Wang.
