Plasma Proteomics to Predict Insulin-mediated Glucose Disposal/Uptake

Abstract

Background Direct measurement of insulin sensitivity (IS) by the hyperinsulinemic euglycemic clamp (HEC) technique is considered a ‘gold-standard’ method, but is laborious and expensive. High-throughput plasma proteomics might provide an opportunity to improve and streamline the prediction of IS. Methods We measured 828 proteins using the OLINK PEA-technique in baseline fasting plasma from 968 participants of the Relationship between Insulin Sensitivity and Cardiovascular disease (RISC) study. We estimated the variance explained of the HEC-derived insulin- mediated glucose uptake (M-value, per kilogram of lean body mass) by clinical covariates alone, and applied LASSO regression to develop prediction models incorporating protein levels. Models were trained in 70% of the cohort and tested in the remaining 30%. Results Clinical covariates alone (age, sex, BMI, lipids, and blood pressure) explained 19.8% (95% confidence interval 15.3-24.2%) of the variance of the M-value. LASSO selected 50 proteins, which increased the variance explained in the testing set to 44.8% (95% CI 38.1-51.3%). A separate protein-only LASSO model of 46 proteins explained a similar proportion of variance (44.1%,95% CI 37.5-50.7%) as the clinical covariates and proteins combined. Conclusion & future directions Plasma proteomic profiling substantially improved the prediction of the M-value, which may allow for accurate assessment of IS without the need of HEC. We are currently evaluating the RISC based models in a replication cohort (the Uppsala Longitudinal Study of Adult Men, ULSAM).