UC San Diego

Type 2 Diabetes (T2D) is a heterogeneous and complex disease that benefits from personalized treatment. Although many individuals are diagnosed with diabetes, both the types of complications and the onset and rate of development may differ. Improved ability to discern which individuals are at higher risk of developing complications could help focus treatment. In this dissertation, I utilize clinical, metabolomic, and proteomic data to distinguish individuals at risk of metabolic complications using the Diabetes Prevention Program (DPP) and Outcomes Study (DPPOS). Chapter 1 describes a tensor decomposition and clustering methodology to establish four longitudinal clusters using 12 clinical phenotypes from 1,732 individuals over 19 years. Clusters 1 and 2 maintained metabolically stable trajectories with low metabolic complication incidence, whereas clusters 3 and 4 demonstrated microvascular and macrovascular complications, respectively. Chapter 2 further investigates the clusters found in Chapter 1, identifying differences in pairwise comparisons of metabolite profiles. Although cluster creation was independent of metabolomics data, we found a strong negative correlation between clusters 1 and 3, suggesting these clusters are metabolic opposites. Different metabolite communities were enriched for significant metabolites in the cluster 1 vs 3 analysis compared to the cluster 2 vs 4 analysis. The metabolomic differences between clusters support the potential of metabolomic communities as markers capable of distinguishing groups with metabolic complications. Chapter 3 focuses on nonproteinuric kidney dysfunction, where the traditional albuminuria biomarker does not increase before disease onset. In the metabolomics dataset, we found a novel metabolite that predicts kidney dysfunction independently of the urine albumin creatinine ratio. In the proteomics dataset, we prioritized ten proteins, including several supported by genetic enrichment or replicated in an external proteomics cohort. Together, this work demonstrates complementary frameworks for characterizing metabolic complications using multimodal datasets. These findings highlight the value of integrating clusters derived from longitudinal clinical data with metabolomics to provide insights into metabolic complications as well as the potential of metabolomics and proteomics to uncover novel biomarkers for early detection of renal dysfunction.