TL;DR
This study introduces five "metabolome-based biological age gaps" (MetBAGs) derived from plasma metabolites, offering a new multi-organ perspective on aging. These MetBAGs not only predict chronological age with reasonable accuracy but also reveal potential causal links to cardiometabolic diseases and mortality risk, suggesting their utility in clinical risk stratification and personalized health strategies.
What Problem Does This Paper Address?
While biological aging clocks based on clinical phenotypes, neuroimaging, and proteomics have advanced our understanding of aging and disease, the role of plasma metabolomics has remained relatively underexplored. Metabolomics offers a dynamic, real-time snapshot of an individual's metabolic state, influenced by factors like diet and environment, making it a promising but complex layer of data for aging research. The challenge lies in integrating this complex, multi-organ metabolic data into robust aging models that can predict health outcomes and disease risk.
Methods
The researchers developed five organ-specific MetBAGs (digestive, hepatic, immune, endocrine, and metabolic) using 107 plasma non-derivatized metabolites from a large UK Biobank cohort. They employed two AI/ML models (Lasso regression and neural networks) for age prediction, selecting the best-performing model for each organ based on generalizability. Crucially, they found that using composite metabolites (sums or ratios) led to poor generalizability, highlighting the importance of using raw metabolite data. They then investigated the genetic architecture of these MetBAGs using genome-wide association studies (GWAS) and a Bayesian method (SBayesS) to estimate SNP-based heritability, selection signatures, and polygenicity. Genetic correlation (LDSC) and Mendelian randomization were used to explore causal relationships between MetBAGs and various disease endpoints. Finally, they assessed the clinical utility of MetBAGs and their polygenic risk scores (PRS) in predicting disease categories and mortality using neural networks and Cox proportional hazard models.
Datasets
- UK Biobank (UKBB): The primary dataset used for developing and training the MetBAG models. It included plasma NMR metabolic biomarkers from approximately 274,247 participants. Genotype data from UKBB was used for genetic analyses.
- FinnGen: A Finnish population-based study providing GWAS summary statistics for disease endpoints (DEs).
- Psychiatric Genomics Consortium (PGC): Provided GWAS summary statistics for psychiatric disorders.
Key Insights and Ideas
- Metabolites as a Dynamic Aging Clock: Plasma metabolites, reflecting real-time metabolic processes, can serve as a dynamic and sensitive indicator of biological aging, complementing more static measures like neuroimaging or proteomics.
- Organ-Specific Metabolic Signatures: Different organ systems have distinct metabolic fingerprints that can be captured by specific sets of plasma metabolites, allowing for the development of organ-specific aging clocks.
- Composite Metabolites Harm Generalizability: The study found that combining raw metabolites into composite measures (sums/ratios) significantly reduced the generalizability of age prediction models, likely due to multicollinearity. This emphasizes the value of analyzing individual metabolites.
- Genetic Underpinnings of Metabolic Aging: The MetBAGs exhibit moderate heritability, suggesting a significant genetic contribution to metabolic aging. Negative selection signatures indicate evolutionary pressure to maintain stable metabolic processes.
- Metabolic Aging and Cardiometabolic Health: There are strong genetic correlations and potential causal links between MetBAGs and cardiometabolic conditions like metabolic disorders and hypertension, underscoring the interconnectedness of metabolic health and aging.
- Multi-Organ and Multi-Omics Synergy: Integrating information from multiple organs and different data types (metabolomics, genetics) significantly improves the prediction of disease categories and mortality risk.
Key Results
- Age Prediction Accuracy: The MetBAG models achieved a mean absolute error of approximately 6 years, with Pearson's r coefficients ranging from 0.25 to 0.42 on independent test data.
- Genetic Architecture: 405 MetBAG-locus pairs were identified through GWAS. SNP-based heritability ranged from 0.09 to 0.18, negative selection signatures were strong (-0.93 to -0.76), and polygenicity was low (0.001 to 0.003).
- Causal Links to Disease: Mendelian randomization revealed potential causal links between MetBAGs and cardiometabolic conditions, including metabolic diseases and hypertension. For example, the digestive MetBAG was causally linked to heart valve disease.
- Predictive Power for Mortality: All five MetBAGs and their polygenic risk scores (PRS) were significantly associated with mortality risk. The metabolic MetBAG showed the highest association (HR = 1.16). Combining MetBAGs and PRS improved mortality prediction, achieving an average concordance index of 0.72.
- Disease Category Prediction: While modest, combining MetBAGs and PRS showed some improvement in classifying 14 systemic disease categories, particularly for genitourinary and circulatory diseases.
Limitations and Future Work
- UK Biobank Reliance: The study primarily used data from the UK Biobank, which has a specific demographic profile (largely European ancestry). Validation in diverse populations is needed to assess broader applicability.
- Organ Specificity of Metabolites: Plasma metabolites are inherently systemic, making it challenging to assign them to a single organ. The study used a pragmatic approach to annotate metabolites to organs, but this may not fully capture complex interactions.
- Causality Interpretation: While Mendelian randomization provides evidence for causal relationships, it relies on stringent assumptions that can sometimes be violated. The authors conducted sensitivity analyses to address this.
- Future Directions: The authors suggest using MetBAGs for clinical risk stratification, personalized intervention strategies, and optimizing clinical trial designs. They also emphasize the importance of integrating multi-omics data and investigating dynamic shifts in metabolomes throughout the lifespan. Further methodological advancements are needed to enhance the organ specificity of plasma metabolomics data.
Summary updated