Targeted proteomics and the presence of carotid plaques in subjects at low cardiovascular disease risk

Background and Aims: We demonstrated that targeted proteomics predicts the occurrence of Subclinical Carotid Atherosclerosis (SCA) in apparently healthy subjects in primary prevention for cardiovascular disease (CVD). SCA can be present, as multifocal and/or stable/vulnerable, also in primary prevention. We now aim at addressing whether proteomics can help to characterize the presence of multifocal, stable/vulnerable SCA in low CVD risk subjects. Methods: A machine learning (ML) model, trained on the ultrasound frames from 317 subjects (118 women) at low CVD risk (SCORE) but with SCA (i) counted the number of plaques (1, 2, 3+) in both carotids and (ii) characterized the vulnerability of SCA (“grayscale”, where lower=vulnerable and higher=stable). The same model classified the best set of plasmatic proteins (368 measured with OlinkTM) associated with number of plaques and the vulnerability of SCA. Results: The higher number of plaques associated with factors included in clinical algorithms (age, hypertension, low cholesterol in High Density Lipoproteins). By contrast, lower grayscale did not associate with any factor. The ML classified 40 proteins, outperforming the identification of subject with 3+ plaques as compared to those with lower number (AUC=0.659 (0.525-0.777), p=0.035). Also, the same model found 16 proteins, chemokines and inflammatory markers (none in common with those identifying number of plaques) outperforming the identification of subject with more vulnerable vs those with more stable plaque (Area Under the Curve, AUC=0.647 (0.514-0.765), p=0.042). Conclusions: We provide a first-in-class evidence that combining targeted proteomics with ML can identify subjects with advanced atherosclerosis that cannot be captured by clinical algorithms.