The clinical and genetic determinants of aortic stenosis

Shinwan Kany (Hamburg)1, J. Ramo (Cambridge)2, C. Hou (Cambridge)2, S. Jurgens (Cambridge)2, V. Nauffal (Cambridge)2, J. Cunningham (Cambridge)2, E. Lau (Cambridge)2, J. Olgin (San Francisco)3, J. Ho (Cambridge)2, S. Elmariah (San Francisco)3, M. Lindsay (Cambridge)2, P. Ellinor (Cambridge)2, J. Pirruccello (San Francisco)3

1Universitäres Herz- und Gefäßzentrum Hamburg Klinik für Kardiologie Hamburg, Deutschland; 2Broad Institute of MIT and Harvard Cardiovascular Disease Initiative Cambridge, USA; 3University of California San Francisco San Francisco, USA


Background and Aims: Aortic stenosis (AS) represents a significant global health burden. Understanding the clinical and genetic influences on aortic valve (AV) function is essential for managing this condition effectively. Currently, the comprehension of AV function in the general population is limited, and the genetic determinants of AV function variations, as well as their impact on AS risk, are not well understood.


Methods: Utilizing deep learning, we analyzed velocity-encoded MRI data from over 47,000 UK Biobank participants. We measured aortic valve area (AVA), peak velocity (PV), and mean gradient (MG), conducting epidemiological and survival analyses, as well as genome-wide association studies (GWAS) in 44,780 participants without cardiovascular disease. We integrated data using multi-trait analysis of GWAS (MTAG) from the UK Biobank and FinnGen, constructed polygenic scores (PGS) using PRScs, and tested them in the All of Us cohort.


Results: We observed an annual decrease of 0.03cm² in AVA in healthy individuals. Both severe AS (15/31 [48%] clinically diagnosed) and moderate AS (56/313 [18%]) were underdiagnosed. Biomarkers measured approximately 10 years prior to MRI revealed that higher levels of ApoB, triglycerides, Lp(a), and inflammatory markers were associated with increased gradients across the AV. After conducting the AS GWAS using MTAG, we identified 81 distinct loci (62 with AV traits, 54 with AS, 35 overlapping), including PCSK9 (β=0.032, P=5.8E-09) and LDLR (β=0.018, P=2.3E-10). Other identified loci were linked to lipid metabolism (LPA, FADS2, SORT1), atherosclerosis (CDKN2B, ARHGEF26, OTUD7B, PRDM16), inflammation (IL6, KLF2, TRAF1), and phosphate hemostasis (FGF23). Using a PGS derived from this data, participants in the top 5% of the MTAG-adjusted peak velocity PGS from the All of Us cohort had a hazard ratio (HR) of 2.5 for incident AS (P=4.4E-08). Mendelian randomization provided evidence supporting a causal role for Lp(a) and LDL—and elevated phosphate (P=6.0E-10)—on the risk for AS.


Conclusion: This study, employing machine learning and genetic analysis, highlights the underdiagnosis of AS and illuminates its early pathogenesis. By identifying genetic loci associated with AV function, including PCSK9 and LDLR for the first time, our findings contribute to earlier surveillance strategies and potential targets for preventive therapy, thereby improving the management of AS.

Study overview: Deep learning was used to create segmentation masks of the ascending aorta in the aortic flow image series in the UK Biobank. The masks were used to create velocity maps that were used to construct MRI-derived aortic valve measurements. The measurements were then used for common variant GWAS. The summary statistics were meta-analyzed with aortic stenosis summary statistics in the MTAG framework to enhance discovery. Further analyses were performed using Mendelian randomization and polygenic scores. The MTAG-derived polygenic scores were used to predict incident aortic stenosis in the external AllofUS cohort.
Diese Seite teilen