Abstract 39: Circulating Plasma Proteomics of Carotid Intima-Media Thickness in the UK Biobank Cohort

Introduction: Ultrasound carotid intima-media thickness (cIMT) is a well-established tool for improved cardiovascular risk stratification. The relative importance of traditional risk factors, genetic susceptibility, and plasma proteomics in predicting cIMT has not been previously explored among asymptomatic individuals. Hypothesis: We hypothesized that conventional atherosclerosis risk factors, polygenic scores (PRSs), and levels of plasma proteins each provide incremental value in predicting cIMT. Plasma proteomic profiles associated with cIMT may provide insight on the biological underpinnings of early atherogenesis in carotid arteries. Methods: We examined 6136 UK Biobank participants with 1461 proteins profiled in their plasma at their baseline visit using the proximity extension assay who subsequently returned to a second study visit to undergo a cIMT measure. With cIMT as outcome, we implemented linear regression models, stepwise Akaike information criterion (AIC)-based models, and the least absolute shrinkage and selection operator (LASSO) approach to screen for and select baseline demographic, genetic, physical, medical, and proteomic determinants of cIMT. We evaluated predictive performance of selected predictors by calculating proportion of variance explained (R 2 ) of cIMT. Lastly, we performed gene set enrichment analyses with significantly associated proteins. Results: The mean follow-up time between baseline and cIMT measurements was 9 years. A LASSO model incorporating 97 proteins, in addition to age, assessment center, genetic risk factors, smoking, blood pressure, trunk fat-free mass index, apolipoprotein B, and Townsend deprivation index, reached the highest R 2 value in the testing dataset (R 2 = 0.237, 95% C.I. 0.193, 0.281). After forcing minimal non-modifiable covariates such as age and sex, a LASSO model trained by proteins alone was almost as predictive as models trained by both covariates and proteins (R 2 = 0.218, 95% C.I. 0.173, 0.262). Non-proteomic covariates of robust predictive value included age at cIMT measurement as a proxy for follow up time (univariate screening R 2 = 19.4%), baseline age (R 2 = 18.8%), blood pressure and anthropometry measurements. Fat-free mass index in trunk was the strongest predictor to cIMT among adiposity measurements (R 2 = 3.6%). CHGB (Secretogranin-1), CST6 (Cystatin-M), and PRELP (Prolargin) were proteins consistently selected across all regression models. Gene set enrichment analysis found overrepresented pathways in hemostasis (MGLL), angiogenesis (IGFBP7), vesicle-mediated transport (VTA1), and glycosaminoglycan metabolisms (HAPLN1 and VCAN). Conclusions: Plasma protein profiles provide incremental value in predicting a cIMT measurement on average nearly a decade later. Our findings suggested the roles of vasculogenic signaling and extracellular matrix proteins in remodeling.