Abstract TH813: Multi-metabolite signatures of continuous glucose monitoring phenotypes are associated with future diabetes risk

Abstract

Background: Despite the popularity of continuous glucose monitoring (CGM) usage, especially in diabetes, prognostic and physiological implications of CGM phenotypes in individuals without diabetes remain uncertain. We investigated associations of comprehensive CGM phenotypes with a broad circulating metabolome in community-dwelling individuals and assessed their relations with future diabetes risk. Methods: We included Framingham Heart Study (FHS) participants without diabetes who had up to 10 days of Dexcom G6 pro CGM data and fasting blood sampling for metabolite profiling. We selected 8 CGM variables representing distinct physiologies a priori and conducted principal component (PC) analysis of all 61 CGM variables. Associations of circulating metabolites (N=449) with each of the key CGM phenotypes (selected variables and PCs) were examined using LASSO regression to account for collinearity. In Coronary Artery Risk Development in Young Adults (CARDIA) study participants without diabetes who completed metabolite profiling at Year 7, we applied FHS-derived LASSO coefficients to calculate multi-metabolite scores for each CGM phenotype, and tested their associations with incident diabetes through Year 35 in multivariable Cox models adjusted for age, sex, race, fasting glucose, and body mass index (BMI). Results: In 571 FHS participants (55% women, median age 61 years, fasting glucose 97 mg/dL, BMI 27.3 kg/m 2 ), 3 PCs explained >80% of variation, with loadings weighted on measures of high glucose (PC1), glycemic variability (PC2), and hypoglycemia (PC3; excluded due to null metabolite associations in linear models). There were 35-67 LASSO metabolite coefficients per score representing diverse pathways ( Figure A ). The metabolites with the highest absolute loading values for PC1 and PC2 reflected metabolites with known (e.g., amino acids, glyceropholipids, bile acids) and novel (e.g., ketoisovaleric acid, 3-methylhistidine) relations to diabetes. In 2332 CARDIA participants (45% women, 45% Black individuals, median age 33 years, fasting glucose 87 mg/dL, BMI 24.8 kg/m 2 ), 343 incident diabetes events occurred between Years 7 and 35. All multi-metabolite scores were associated with diabetes incidence in multivariable-adjusted models (Figure B) . Conclusion: Multi-metabolite signatures of CGM phenotypes are associated with future risk of developing diabetes independent of fasting glucose, identifying novel potential metabolic pathways underlying cardiometabolic risk.