Abstract Body: Background. The proteome is altered in obesity and proteomic signatures of BMI have been identified. However, few studies have examined differences in clinical profiles between actual and protein-predicted BMI.
Hypothesis. Plasma protein-based BMI classification will better capture cardiometabolic risk compared to actual BMI classification.
Methods. Data on cardiometabolic phenotypes and plasma proteins (4979 proteins via SomaScan) were available in 645 adults (56% Female, 35% Black, 17-65 yrs) from the HERITAGE Family Study. LASSO regression models with 10-fold-cross-validation were used to create a BMI proteomic signature. Protein-predicted BMI was classified into weight classes and compared to actual BMI classification to create 7 groups (underpredicted, matched, or overpredicted) (Fig 1A). General linear models adjusted for age, sex, race, and measured BMI were used to examine differences in cardiometabolic traits across groups.
Results. The LASSO model (R²=0.82, RMSE=2.2) included 208 proteins. Protein-predicted BMI correlated with actual BMI at r=0.95 (p<0.0001), with a 16% misclassification rate (Fig 1A). Participants whose proteins underpredicted BMI class generally had more favorable cardiometabolic profiles than matched groups of the same class, while overpredicted BMI class groups showed worse cardiometabolic profiles than matched groups of the same class (Fig 1B).
Conclusions. We identified individuals matched for BMI but with opposing proteomic signatures that differed in cardiometabolic risk profiles. Proteomic profiling may identify clinically meaningful heterogeneity in cardiometabolic health not fully captured by BMI that could potentially be used as biomarkers and/or targets of therapeutic responsiveness.