Proteomic Predictors of Incident Diabetes: Results From the Atherosclerosis Risk in Communities (ARIC) Study

Abstract

OBJECTIVEThe plasma proteome preceding diabetes can improve our understanding of diabetes pathogenesis.RESEARCH DESIGN AND METHODSIn 8,923 Atherosclerosis Risk in Communities (ARIC) Study participants (aged 47–70 years, 57% women, 19% Black), we conducted discovery and internal validation for associations of 4,955 plasma proteins with incident diabetes. We externally validated results in the Singapore Multi-Ethnic Cohort (MEC) nested case-control (624 case subjects, 1,214 control subjects). We used Cox regression to discover and validate protein associations and risk-prediction models (elastic net regression with cardiometabolic risk factors and proteins) for incident diabetes. We conducted a pathway analysis and examined causality using genetic instruments.RESULTSThere were 2,147 new diabetes cases over a median of 19 years. In the discovery sample (n = 6,010), 140 proteins were associated with incident diabetes after adjustment for 11 risk factors (P < 10−5). Internal validation (n = 2,913) showed 64 of the 140 proteins remained significant (P < 0.05/140). Of the 63 available proteins, 47 (75%) were validated in MEC. Novel associations with diabetes were found for 22 the 47 proteins. Prediction models (27 proteins selected by elastic net) developed in discovery had a C statistic of 0.731 in internal validation, with ΔC statistic of 0.011 (P = 0.04) beyond 13 risk factors, including fasting glucose and HbA1c. Inflammation and lipid metabolism pathways were overrepresented among the diabetes-associated proteins. Genetic instrument analyses suggested plasma SHBG, ATP1B2, and GSTA1 play causal roles in diabetes risk.CONCLUSIONSWe identified 47 plasma proteins predictive of incident diabetes, established causal effects for 3 proteins, and identified diabetes-associated inflammation and lipid pathways with potential implications for diagnosis and therapy.