Deep learning enables genetic analysis of the human thoracic aorta

James P. Pirruccello 1,2,3,4,5, Mark D. Chaffin 3,4, Elizabeth L. Chou 2,6, Stephen J. Fleming 4,7, Honghuang Lin 8,9, Mahan Nekoui 3,5, Shaan Khurshid 1,2,3, Samuel N. Friedman 7, Alexander G. Bick 3,10, Alessandro Arduini 3,4, Lu-Chen Weng 3, Seung Hoan Choi 3, Amer-Denis Akkad 4, Puneet Batra 7, Nathan R. Tucker 11, Amelia W. Hall 3, Carolina Roselli 3,12, Emelia J. Benjamin 8,13,14, Shamsudheen K. Vellarikkal 3, Rajat M. Gupta 15, Christian M. Stegmann 4 , Dejan Juric 16,5, James R. Stone 17,5, Ramachandran S. Vasan 8,13,14, Jennifer E. Ho 1,2,5, Udo Hoffmann 18,19, Steven A. Lubitz 1,2,3,5, Anthony Philippakis 7,20, Mark E. Lindsay 1,2,3,5,21, Patrick T. Ellinor 1,2,3,4,5 

  1. Division of Cardiology, Massachusetts General Hospital, Boston, Massachusetts, USA
  2. Cardiovascular Research Center, Massachusetts General Hospital, Boston, Massachusetts, USA
  3. Cardiovascular Disease Initiative, Broad Institute, Cambridge, Massachusetts, USA
  4. Precision Cardiology Laboratory, The Broad Institute & Bayer US LLC, Cambridge, Massachusetts, USA
  5. Harvard Medical School, Boston, Massachusetts, USA
  6. Division of Vascular and Endovascular Surgery, Massachusetts General Hospital, Boston, Massachusetts, USA
  7. Data Sciences Platform, Broad Institute, Cambridge, Massachusetts, USA
  8. Framingham Heart Study, Boston University and National Heart, Lung, and Blood Institute, Framingham, Massachusetts, USA
  9. Department of Medicine, Section of Computational Biomedicine, Boston University School of Medicine, Boston, Massachusetts, USA
  10. Department of Medicine, Center for Genomic Medicine, Massachusetts General Hospital, Boston, Massachusetts, USA
  11. Masonic Medical Research Institute, Utica, New York, USA
  12. University Medical Center Groningen, University of Groningen, Groningen, 9712 CP, NL
  13. Department of Medicine, Cardiology and Preventive Medicine Sections, Boston University 32 School of Medicine, Boston, Massachusetts, USA
  14. Epidemiology Department, Boston University School of Public Health, Boston, Massachusetts, USA
  15. Department of Medicine, Divisions of Cardiovascular Medicine and Genetics, Brigham and Women's Hospital, Boston, Massachusetts, USA
  16. Cancer Center, Massachusetts General Hospital, Boston, Massachusetts, USA
  17. Department of Pathology, Massachusetts General Hospital, Boston, Massachusetts, USA
  18. Department of Radiology, Harvard Medical School, Boston, Massachusetts, USA
  19. Cardiovascular Imaging Research Center, Massachusetts General Hospital, Boston,  Massachusetts, USA
  20. GV, Mountain View, California, USA
  21. Thoracic Aortic Center, Massachusetts General Hospital, Boston, Massachusetts, USA

Abstract: The aorta is the largest blood vessel in the body, and enlargement or aneurysm of the aorta can predispose to dissection, an important cause of sudden death. By leveraging a deep learning architecture that was originally developed to recognize natural images, we trained a model to evaluate the dimensions of the ascending and descending thoracic aorta in cardiac magnetic resonance imaging. After manual annotation of just 116 samples, we applied this model to over 4.6 million images from the UK Biobank. We then conducted genome-wide association studies in up to 39,688 individuals, revealing 82 loci associated with ascending and 47 with descending thoracic aortic diameter, of which 14 loci overlapped. Integration of common variation with transcriptome-wide analyses, rare-variant burden tests, and single nucleus RNA sequencing prioritized genes including SVIL, a gene highly expressed in vascular smooth muscle and significantly associated with the diameter of the descending aorta. A polygenic score for ascending aortic diameter was associated with a diagnosis of thoracic aortic aneurysm in the remaining 385,621 UK Biobank participants who did not undergo imaging (HR = 1.43 per standard deviation; CI 1.32-1.54; P = 3.3·10-20). Defining the genetic basis of the diameter of the aorta may enable the identification of asymptomatic individuals at risk for aneurysm or dissection and facilitate the prioritization of potential therapeutic targets for the prevention or treatment of aortic aneurysm. Finally, our results illustrate the potential for rapidly defining novel quantitative traits derived from a deep learning model, an approach that can be more broadly applied to biomedical imaging data.

 

Resources