PhD Student McGill University Montreal, Quebec, Canada
Background: Coronary artery disease (CAD) is clinically heterogeneous, reflecting diverse biological mechanisms and underlying endotypes that traditional risk scores and genetic predictors capture only imperfectly. Appreciating this heterogeneity is clinically critical, as CAD can diverge markedly in metabolic burden, prognosis, and response to treatments. Circulating proteins represent a central layer in the omics cascade—bridging genome and phenotype, reflecting pathophysiological change, and now measurable at scale via high-throughput assays. We hypothesized that individual-level plasma proteomic profiles would offer a clinically relevant, mechanistically informative lens on CAD heterogeneity, enabling the identification of biologically coherent patient subgroups and continuous gradients of clinical characteristics, including risk of comorbidities.
METHODS AND RESULTS: We leveraged plasma proteomic signatures in 42,803 UK Biobank participants (3,713 with incident CAD) to dissect CAD heterogeneity. First, we identified 320 proteins out of 2,923 that improved incident CAD risk prediction beyond conventional clinical scores (QRISK3 and ASCVD). Using these selected proteins, we applied DDRTree-based dimensionality reduction in incident CAD cases, classifying them into nine discrete clusters defined by two proteomic dimensions exhibiting unique metabolic signatures. In addition to discrete clusters, these dimension values also captured a continuous phenotypic spectrum of CAD severity, reflected by increasing body mass index, hemoglobin A1c, triglycerides, and other metabolic markers (Figure). The top panel shows each incident CAD case in proteomic dimension space, while the bottom panel illustrates nine distinct clusters with varying clinical characteristics, where all clinical variables are normalized. Phenome-wide association analyses across multiple ICD-coded outcomes in the UK Biobank further linked the proteomic dimensions to increased risks of type 2 diabetes, obesity, renal failure, and other metabolic-renal diseases, suggesting that these dimensions encapsulate broad metabolic and renal dysfunction. Notably, individuals with prevalent CAD clustered disproportionately in the most metabolically adverse region defined by the two proteomic dimensions, indicating that these dimensions may capture underlying metabolic trajectories and more advanced disease progression.
Conclusion: Our findings underscore the utility of plasma proteomic signatures in illuminating both distinct clusters and continuous gradients of CAD phenotypes with varying comorbidity risk. We highlight a framework for refining disease classification and uncovering biologically meaningful patient subtypes. These insights set the stage for personalized risk stratification and targeted therapeutic strategies, ultimately advancing precision medicine approaches for patients with CAD.
