Abstract Body: Background: Disordered lipid metabolism is highly prevalent in diabetes. Mechanisms have been identified through which lipids can impact glucose metabolism. Previous cohort studies of lipidomics and diabetes often have been limited to narrow lipidomic coverage or have lacked sample diversity in sociodemographic characteristics.
Methods: Data were from CARDIA, a population-based cohort of self-identified Black and White race adults, recruited from 4 U.S. urban centers in 1985-86 (n=5,115). On a sample subset, lipidomics data were generated from fasting plasma collected at the Yr 20 exam, using infusion-mass-spectrometry, yielding n=14 lipid classes and n=756 molecular species. For analysis, molecular species within classes that contained >1 fatty acid were grouped by total carbon length and number of double bonds (n=300 total species). Participants who had prevalent diabetes or were pregnant at Year 20 were excluded from the analysis. Associations between distinct lipid classes/species and 10-year incident diabetes [n (cases) = 1,084 (104)] were quantified with Cox proportional hazards regression, adjusting for sociodemographics, health behaviors, and clinical variables, and accounting for multiple comparisons.
Results: Ten of the 14 classes were significantly associated with 10-year incident diabetes (Fig 1). Findings for species within 8 classes were consistent with class-level associations, (direction and significance); other class-species associations displayed discordance (Fig 2). For example, the ceramide class was not, but a ceramide species (22:2) was, associated with incident diabetes.
Conclusions: Class- and species-level analysis revealed unique patterns of associations, motivating analysis of both. Lipidomic signatures of diabetes may elucidate metabolic pathways implicated in diabetes risk that are not captured in standard lipid panels. These findings may inform clinical evaluation of diabetes risk in early-middle-age, when preventive measures may be more effective.