Abstract 22: Novel Plasma Lipids Predict Risk of Diabetes: A Longitudinal Lipidomics Study in American Indians

Abstract

Background: American Indians (AIs) suffer disproportionately high rate of type 2 diabetes (T2D). Traditional biomarkers have limited value in predicting and tracking early onset and progression of T2D. There is an urgent need for early biomarkers in this high-risk but understudied minority population. Objective: To identify novel lipids predictive of T2D onset among American Indians, independent of standard risk factors. Methods: We studied 2,000 American Indians attending two clinical exams (2001-2003, 2006-2009, average 5-yr apart) in the Strong Heart Family Study (SHFS). Fasting plasma lipids were repeatedly measured by untargeted lipidomics using LC-MS/MS. All participants were free of overt CVD at baseline (2001-2003) and followed through 2017 (average 16-yr follow-up). Cox regression with frailty model was used to identify lipids predictive of T2D onset, adjusting for traditional risk factors including age, sex, smoking, BMI, triglyceride, HDL, insulin resistance, eGFR and dietary intake of protein. Longitudinal analysis was conducted by regressing changes in lipids on changes in T2D-related traits (e.g., fasting glucose, HbA1c, or insulin resistance) between baseline and 5-yr follow-up, adjusting for changes in BMI, triglyceride, HDL, and eGFR. Incremental prognostic value of lipids in diabetes risk prediction above traditional risk factors was estimated using area under the curve (AUC). Network analysis was conducted to examine the dynamic changes in lipid networks between baseline and 5-yr follow-up. Multiple testing was controlled by FDR. Results: Of 1,628 non-diabetic participants at baseline (mean age 39.8, 62% women), 189 and 359 individuals developed incident T2D after 5-yr and 16-yr follow-up, respectively. Our high-resolution lipidomics detected 1,826 lipids, of which 1,119 lipids (460 known, 659 unknown) passed stringent quality control. Seven lipids with known structures significantly predict over 30% increased (FA(18:1), FA(20:2), FA(22:2), SM(d34:0) ) or 19% decreased (PC(37:4), PC(37:5), PC(38:4)) risk of T2D onset at both follow-ups. Nine unknown lipids also predicted T2D onset. Longitudinal changes in CE(20:2), CE(22:6), PC(32:0) and two unknowns explain about 6.4% changes in fasting plasma glucose. These newly identified lipids significantly improve the performance of risk prediction over traditional risk factors (AUC increased from 0.787 to 0.803, P=0.002). Network topology analysis revealed that the network connectivity among lipids was much stronger in participants who developed new T2D compared to those who didn’t (degree of network connectivity, 4.2 vs 2.5, P<0.001). Conclusion: We identified several novel plasma lipids significantly predictive of incident T2D, independent of traditional risk factors. These findings highlight the importance of identifying novel lipid markers in early diabetes risk prediction.