A Longitudinal Metabolomics Analysis of Geroprotective Interventions in Aging Mice

Abstract

Metabolism is a fundamental process which provides energy to living organisms, and its dysfunction is a shared feature of many noncommunicable diseases and of the degenerative process of aging. Identifying the metabolic hallmarks of aging is an ongoing pursuit, particularly in mice, which are commonly used for modeling human aging. Multiple dietary and pharmacological interventions have been shown to modulate age-related declines in metabolic function and increase lifespan in mice, although the precise mechanism by which this occurs is unknown. Here, we perform a longitudinal metabolomics analysis of aging mice on distinct dietary paradigms: control, methionine restriction (MetR), 15% caloric restriction (PF), and two drug supplemented diets: ZGN1062, and ZGN201, to better understand how age, sex, and dietary interventions modulate the serum metabolome. We report 607 known metabolites and 1004 unknown metabolites with significant associations with age, sex, diet, or the combination of these variables. Moreover, we use decision trees, gradient boosting, and Long Short-Term Memory (LSTM) machine learning models to predict lifespan from metabolomic profiles, finding that gradient boosting is particularly effective at predicting days of life remaining in mice. Overall, this thesis employs a range of computational tools to identify metabolites that are age-associated, sexually dimorphic, lifespan-predictive, or modulated by dietary interventions, allowing us to generate hypotheses about age-induced metabolic dysregulation and how geroprotective interventions may slow this process.