Metabolomic aging and its association with depression in a UK cohort

Abstract

Introduction Depression is associated with both reduced lifespan and cardiovascular disease risk. Accelerated biological aging may underlie this association. Modern ‘omics’ platforms provide new opportunities for the systematic assessment of biological aging. Genome-wide DNA methylation data has been used to predict chronological age with high accuracy, and ‘age acceleration’ - the difference between DNA methylation age and chronological age - is predictive of mortality. Few studies have employed metabolomics (small molecule profiling) to model the aging process. We have employed untargeted metabolomics across multiple nuclear magnetic resonance spectroscopy (NMR) and mass spectrometry (MS) based platforms, to develop a highly predictive model of age, within a large sample from the Airwave Health Monitoring Study (AIRWAVE). We investigate the correlation between metabolomic age and DNA methylation age and show that metabolic age acceleration is associated with depression and other disease risk factors. Methods Full metabolomic data was acquired for 2238 participants (age range: 19.2–65.2 years, 61% male) in the AIRWAVE cohort of employees of police forces from across Great Britain. Nine metabolomic analyses were applied: NMR, hydrophilic interaction chromatography-MS and reversed-phase chromatography-MS (both positive and negative modes) in both serum and urine, and NMR-based Bruker IVDr Lipoprotein Subclass Analysis in serum only. NMR profiles were aligned and interpolated onto a common 20,000-point grid while MS data was processed into retention time-m/z pairs. All platforms were combined into one dataset providing 98,824 metabolic features. A metabolomic age prediction model was constructed using elastic net modelling, with penalisation parameters found following 10-fold cross validation, in a training portion of the data (80% of participants) and then validated in a test set (remaining 20%). Metabolic age acceleration (AA) was defined as the difference between predicted metabolomic and chronological age. We computed the ‘intrinsic’ AA, defined as the residuals from the linear regression of AA with chronological age, hereafter referred to as ‘metAA’. DNA methylation was measured in leukocytes in 1102 participants using the Illumina 850 K methylationEPIC array and Hannum's DNA methylation age calculated. Intrinsic DNA methylation AA was calculated as for metAA, additionally accounting for blood cell proportion, and referred to as ‘DNAmethAA’. Depression was assessed through the Patient Health Questionnaire (PHQ-9) and associations with metAA and DNA methAA were assessed through linear regression. Results Metabolomic age, based on 472 features, was highly correlated both with chronological age in the independent validation set (Pearson's correlation r = 0.85) and with DNA methylation age ( r = 0.85). There was no relationship between DNAmethAA and metAA ( r = 0.02). In analyses adjusted for sex, disease risk factors and diet, metAA was associated with both having some depressive symptoms [β, interpretable as years of increase in metabolic age, = 0.70, 95% confidence interval (CI): 0.38, 1.01] and being a depression case (0.54, 95% CI: 0.08, 0.99) compared to those with no symptoms of depression. Obesity (1.35, 95% CI: 0.95, 1.69); low income (0.33, 95% CI: 0.00, 0.65); and heavy drinking (0.72, 95% CI: 0.08, 1.36) were also associated with metAA. Only being male was significantly associated with DNAmethAA (0.89, 95% CI: 0.47, 1.30). Participants with low income had higher DNAmethAA than those with high income, although this was not statistically significant [0.41 (−0.12, 0.94) after adjustment for sex]. Conclusion Metabolomic analysis can be used to predict age with high accuracy. Accelerated metabolic aging appears to capture a different dimension of the aging process to DNA methylation age acceleration, and may contribute to associations between depression and mortality.