Cardiovascular disease (CVD) is the leading cause of death worldwide, accounting for approximately 30% of global mortality and imposing a substantial economic burden on society. Due to the high heterogeneity of CVD, which involves diverse pathological mechanisms and clinical phenotypes, existing diagnostic tools such as cardiac magnetic resonance imaging and tissue biopsies face limitations including technical constraints, invasiveness, and high costs. Consequently, there is an urgent need to develop simple, cost-effective, and efficient methods for early screening and risk assessment.

Plasma proteins hold significant clinical value as disease biomarkers. This study leveraged the large prospective UK Biobank cohort, including 53,026 participants with an average follow-up of 14 years, to systematically analyze the associations between 2,920 baseline plasma proteins and 14 cardiovascular disease outcomes and mortality.

A multi-tiered analytical strategy was employed: First, Cox proportional hazards models were used to identify protein markers associated with CVD; second, linear regression analyses examined the relationship between protein levels and cardiac magnetic resonance imaging parameters; subsequently, machine learning prediction models were constructed to evaluate the predictive performance of proteins for disease; further, Mendelian randomization was applied to establish causal relationships; finally, functional enrichment analyses were conducted to elucidate underlying biological pathways.

The study identified 3,089 significant associations involving 892 unique protein analytes and 13 cardiovascular disease outcomes. NT-proBNP showed the strongest association with atrial fibrillation (P = 6.31 × 10⁻³¹³), while NPPB and GDF15 demonstrated robust associations with heart failure. Regarding cardiac structure and function, 445 proteins were significantly associated with 18 cardiovascular magnetic resonance parameters, with LEP and FABP4 showing the strongest associations with right ventricular end-diastolic volume.

A comprehensive prediction model based on 257 selected proteins demonstrated excellent performance for most CVD outcomes, achieving an AUC of 0.86 for abdominal aortic aneurysm. Compared to the conventional SCORE2 risk assessment, the protein model outperformed for 11 diseases, highlighting the complementary value of plasma proteins in enhancing CVD prediction accuracy.

Mendelian randomization analyses revealed causal relationships between 225 proteins and CVD, with LPA showing the strongest association with coronary artery disease (OR = 1.13). Drug target enrichment analysis validated several established therapeutic targets (e.g., PCSK9, VEGFA) while uncovering potential drug repositioning opportunities, providing clues for novel therapeutic strategies.

Mediation analyses indicated that modifiable risk factors such as smoking and body mass index primarily influence CVD risk through plasma protein mediation. Proteins including IGFBP4 and GDF15 functioned as broad-spectrum mediators, each affecting 9 cardiovascular outcomes, while other proteins exhibited disease-specific mediation effects.

Functional enrichment analyses revealed multiple CVD-related biological pathways, including inflammatory responses, endothelial function, and vascular remodeling. Transcription factor analysis identified known regulators such as NFKB1, RELA, and SP1, along with potential novel regulatory targets including JUND and SPDEF.


The authors systematically analyzed 2,920 plasma proteins in 53,026 individuals over 14-year follow-up, identifying 892 CVD-associated proteins, developing robust machine learning prediction models, and conducting comprehensive Mendelian randomization analysis.
DOI:10.1093/procel/pwaf072