Plasma proteomic architecture of incident valvular heart disease in the UK Biobank: discovery, genetic prioritization, and exploratory risk discrimination

Background Valvular heart disease (VHD) is increasingly prevalent in aging populations, but the circulating proteomic features that precede incident disease remain incompletely characterized. We aimed to identify plasma proteins associated with incident VHD, prioritize genetically supported candidates, and assess the incremental value of proteomic information for risk discrimination. Methods After phenotype-proteome alignment, sample- and protein-level quality control, and removal of duplicate participant identifiers, we included 40,435 UK Biobank participants free of prevalent VHD at baseline. Associations of 2,745 Olink Explore proteins with incident broad VHD, mitral regurgitation, aortic stenosis, and aortic regurgitation were examined using multivariable Cox proportional hazards models adjusted for demographic and clinical covariates; in subtype-specific analyses, the remaining VHD subtypes were modeled as time-varying covariates. Genetic prioritization was performed using two-sample cis-Mendelian randomization (cis-MR) with deCODE cis-pQTL instruments and FinnGen R12 summary statistics for broad VHD and operated aortic stenosis. Secondary analyses included sex-stratified association testing and exploratory internally validated prediction modeling using logistic LASSO after an 80/20 train-test split. Results During a median follow-up of 11.0 years, 1,250 participants developed broad VHD, 391 developed mitral regurgitation, 303 developed aortic stenosis, and 105 developed aortic regurgitation. Across the four endpoints, 421 Bonferroni-significant protein-disease associations were identified, including 316 for broad VHD, 81 for mitral regurgitation, and 24 for aortic stenosis; none met the study-wide significance threshold for aortic regurgitation. The strongest broad-VHD signals included NPPB, ANGPT2, LTBP2, GDF15, and CRIM1. Cis-MR prioritized 20 proteins for broad VHD and 17 for aortic stenosis, with EPHB4, GHR, and SCARF2 supported by both prospective and genetic evidence. Broad-VHD effect estimates were highly concordant between women and men (Pearson r = 0.75), and no protein-by-sex interaction remained significant after false discovery rate correction. In exploratory prediction analyses, a 19-protein score improved internal discrimination for broad VHD beyond clinical variables alone (AUC, 0.772 vs 0.723; DeLong P = 5.1 × 10 − 5). Conclusions Large-scale plasma proteomics identified molecular features preceding incident VHD and prioritized a subset of genetically supported candidate proteins. Proteomic information may improve VHD risk discrimination, although predictive performance and clinical utility require validation in independent cohorts.