GENOME-WIDE ASSOCIATION STUDY OF AORTIC STIFFNESS DERIVED FROM DEEP LEARNING IN CMR IMAGES OF 45 789 INDIVIDUALS IDENTIFIES LOCI LINKED WITH CELL-MATRIX STRUCTURE

Abstract Background Aortic stiffness is a key determinant of cardiovascular disease (CVD) risk. Advances in deep learning on cardiac magnetic resonance (CMR) imaging enable detailed phenotyping of cardiac traits, providing novel insights into CVD genetics. Aim We developed a novel CMR-based, pressure-independent measure of aortic stiffness (β0) and identified genetic determinants using genome-wide association studies (GWAS) of UK Biobank data. Methods A pre-trained neural network segmented CMR images of 45 789 European-ancestry UK Biobank participants to quantify ascending aorta diameter. Combined with contemporaneous blood pressure readings, this was used to calculate β0, assuming an exponential relationship between pressure and aortic diameter. GWAS excluded individuals with prior CVDs, identifying independent lead variants through conditional analysis. Genomic loci were defined by a 500 kb flanking region, and putative effector genes were prioritized using variant-to-gene mapping, polygenic priority scores, gene-based association tests, and proximity. Pathway enrichment analysis identified biological processes associated with β0. Results GWAS identified 17 independent lead variants (Fig. A). We identified putative effector genes at 16 of 17 genomic loci (Fig. B, one locus was not resolved). Among the putative effectors were ELN and LTBP4, genes implicated in elastin fiber formation pathway (P = 0.019); ELN, HAS2 and LTBP4 associated with extracellular matrix assembly pathway (P = 0.0075), and ULK4, ARHGAP24, ELN, HAS2, SVIL, ARHGAP22, CDH13, SMG6, and LTBP4 linked to regulation of cellular component organization (P = 0.0026). Conclusion Our findings link aortic stiffness to genetic pathways regulating extracellular matrix and cellular organization, providing a foundation for future mechanistic studies and potential therapeutic targets.