Genome wide association studies on three-dimensional aortic geometric indices measured with magnetic resonance imaging

Abstract Introduction To date, three-dimensional properties of the aorta have not been comprehensively quantified in a population level dataset, nor has there been a genome-wide association study (GWAS) to identify significant single nucleotide polymorphisms (SNPs) associated with geometric properties of the aorta. Methods We segmented the thoracic aorta of 48,045 subjects in the UK Biobank using a U-Net convolutional neural network. We then computed geometric measurements (centerline length, arch height, arch width, arch height-width ratio, curvature, and mean, minimum and maximum diameter) across six subsegments of the aorta. Centerline length was allometrically indexed to body height. Using generalized linear models, we performed GWAS on each geometric phenotype adjusted for the following covariates: (1) age, (2) sex, and (3) ten genetic principal components. SNPs with a minor allele frequency < 0.05 were excluded. We performed LD-based clumping with an r2 threshold of 0.001 to identify significant loci and used LDSC to compute the genetic heritability of each geometric phenotype. Results After performing LD clumping, we identified 457 significant(P