Yanan Ma

Accelerated brain aging is increasingly recognized as a transdiagnostic risk factor for neuropsychiatric and neurodegenerative disorders, yet its metabolic underpinnings remain poorly understood. Here we integrated multimodal neuroimaging (MRI), plasma metabolomics, and genomic data from the UK Biobank to identify metabolic markers of brain aging and evaluate their causal relevance. Using 1079 imaging-derived phenotypes (IDPs) from 4333 healthy participants, we trained and validated machine learning models for brain age prediction, with a least absolute shrinkage and selection operator (LASSO) regression model achieving the best performance (mean absolute error = 3.26 years, R² = 0.68). Brain age gap (BAG) was then estimated in 37,458 participants. Association analyses in 21,780 individuals identified nine plasma metabolites significantly linked to BAG after Bonferroni correction, with glucose showing the strongest effect (β = 0.32, P = 9.90 × 10⁻¹²). Genome-wide association studies (GWAS) identified 392 BAG-associated single-nucleotide polymorphisms (SNPs) (P < 5 × 10⁻⁸), and two-sample Mendelian randomization (MR) provided evidence supporting a potential causal role of glucose in accelerating brain aging. Clinically, elevated plasma glucose was positively associated with seven brain disorders, including all-cause dementia, Alzheimer's disease, vascular dementia, Parkinson's disease, stroke, depression, and anxiety, and negatively associated with cognitive performance, movement function, and mental health outcomes. Higher glucose concentrations were also associated with reduced regional brain volumes across 80 cortical, subcortical, and cerebellar regions. These findings implicate glucose metabolism as a modifiable pathway in brain aging, with implications for early intervention strategies aimed at preserving brain health across the lifespan.

Large-scale metabolomics enables systematic profiling of circulating small-molecule metabolites and provides an integrative view of genetic, environmental, lifestyle, and microbial influences on human health, yet comprehensive evidence linking metabolites to diverse traits, diseases, comorbidity, prediction, and causality remains limited. Here, we analyze 251 Nuclear magnetic resonance (NMR)-derived plasma metabolites in 212,751 UK Biobank participants, with independent validation in 177,013 individuals from a newly released data wave. Testing across 884 health-related traits, 722 prevalent diseases, and 1137 incident diseases using multivariable regression and stringent Bonferroni correction, we identified 67,505 metabolite-trait, 21,982 metabolite-prevalent disease, and 41,214 metabolite-incident disease associations, with replication rates of 76.1%, 83.3%, and 74.0%, respectively. Associations were most concentrated in endocrine, metabolic, and circulatory diseases. Unsupervised clustering of disease-metabolite profiles revealed 10 and 12 disease clusters (for prevalent and incident diseases, respectively) that span International Classification of Diseases (ICD)-10 chapters yet share convergent metabolomic signatures, providing a metabolic basis for clinical comorbidity. Metabolite-based predictive models showed stronger performance for short-term than long-term outcomes, with creatinine as the most frequently selected predictor across prevalent disease models. Bidirectional Mendelian randomisation (MR) identified 61 putative metabolite-to-disease causal effects and 558 disease-driven metabolic alterations. This study provides a systematically validated plasma metabolomic atlas (https://net-matebolic.vercel.app/) linking metabolites with human traits and diseases, offering insights into comorbidity, risk prediction, and potential causal pathways.

Depression is a leading cause of mental and physical disability globally, with its onset and progression influenced by a complex interplay of dietary, psychological, and biological factors. Recent research suggests a link between sugar-sweetened beverage (SSB) consumption and depression risk, although the potential biological pathways underlying this association remain poorly understood. This study utilized data from 192,045 participants in the UK Biobank to examine the prospective association between SSB consumption and incident depression using Cox proportional hazards models. SSBs were defined as the sum of five beverage categories assessed via the Oxford WebQ 24-hour dietary recall. Directional consistency of the association was further examined across three external supporting datasets encompassing diverse populations: NHANES, YRBSS, and the Jiangsu Provincial Database. We further investigated whether proteins, metabolites, inflammatory markers, and brain imaging phenotypes may serve as candidate mediators statistically consistent with mediation of the SSB-depression association. High SSB consumption was associated with an 18% higher risk of incident depression compared with non-consumers (HR = 1.18; 95% CI: 1.11-1.25), with consistent directional associations observed across external supporting datasets. A plasma proteomic signature comprising 229 proteins was constructed using elastic net regularization and was associated with an increased risk of incident depression. Exploratory mediation analyses identified 72 proteins, 36 metabolites, and 5 inflammatory markers as candidate mediators, with IL1RN showing the strongest protein-level candidate mediating effect (9.6%), and Unsaturation and neutrophil count showing the strongest metabolite- and inflammatory marker-level effects, respectively. This study provides preliminary evidence that proteins, metabolites, and inflammatory markers may serve as candidate mediators statistically consistent with mediation of the association between SSB consumption and incident depression. These findings are exploratory and hypothesis-generating, and future experimental studies are needed to validate these candidate pathways and assess their potential as targets for dietary interventions in depression prevention.

Supplementary Figure S3: Sensitivity analysis for receiver operating characteristics (ROC) curves of selected metabolites at 5-year follow-up for five gastrointestinal cancers prediction models in the training set. C-statistics are shown for CRC (A), PC (B), EC (C), GC (D), and HCC (E). Abbreviations: CRC, colorectal cancer; EC, esophageal cancer; GC, gastric cancer; HCC, hepatocellular carcinoma; PC, pancreatic cancer.

Supplementary Table S14: Sensitivity analysis for the associations between metabolite risk score and the risk of gastrointestinal cancers

Supplementary Table S3: HR (95% CI) of the association between plasma metabolites and incident gastrointestinal cancer events in model 2

Background:Emerging evidence indicates that lipid metabolism plays a crucial role in gastrointestinal (GI) cancers.Methods:This prospective cohort study included 112,655 cancer-free adults from the UK Biobank with metabolomics data at baseline. Multivariate Cox proportional hazards regression models were used to evaluate the associations between metabolites and GI cancers. LASSO regression was used to create metabolite risk scores, and their predictive performance for GI cancers was assessed using the C-statistic derived from the time-dependent ROC curves.Results:A total of 93, 95, 33, 97, 129, and 29 metabolites were associated with colorectal, pancreatic, esophageal, gastric, hepatocellular, and gallbladder/biliary tract cancers, respectively. Higher levels of phospholipids to total lipids in large high-density lipoprotein percentage and monounsaturated fatty acids to total fatty acids percentage were associated with increased risk of all GI cancers (HRs from 1.09 to 1.34). Conversely, higher levels of unsaturated fatty acids, polyunsaturated fatty acids, ratio of polyunsaturated to monounsaturated fatty acids, and docosahexaenoic acid were associated with lower risk of six GI cancers (HRs from 0.69 to 0.93). The highest tertile of metabolic risk scores were positively associated with all GI cancers (HRs from 1.86 to 6.42), and new prediction models showed moderate accuracy for five-year cancer incidence (C-statistics from 0.713 to 0.793).Conclusions:Our findings highlight the significant role of lipid metabolism in GI cancers and provide potential noninvasive biomarkers for enhancing precise prevention.Impact:Lipid metabolism holds clinical potential for precise prevention and early intervention of GI cancers.

Supplementary Table S11: Sensitivity analysis for association between selected metabolic biomarkers and gastrointestinal cancers by LASSO among White participants