Genetic Relationship Between Atrial Fibrillation Risk and Left Atrial Dysfunction

© 2025 HMP Global. All Rights Reserved.

Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of EP Lab Digest or HMP Global, their employees, and affiliates. 

Featured is the presentation entitled "Genetic Relationship Between Atrial Fibrillation (AF) Risk and Left Atrial (LA) Dysfunction" by Patrick T Ellinor, MD, PhD, from Session 6 at Western AF 2025.

It’s great to see everyone and nice to be back in Salt Lake City. I have grant support from the NIH, AHA, Leducq Foundation in the EU, sponsored research support from Pfizer, Bristol Myers Squibb, Novo Nordisk, and Bayer, none of which is relevant to this talk, and I have no personal consulting income.

So, most of my work to date has been focused on the genetics of atrial fibrillation (AF) and that has been in partnership with many of the folks in this room. We have a few of those large-scale studies coming out next week. But this was a side project in the lab. The lab in the years since has gone in many directions, combining both the genetics and some of the machine learning (ML) and deep learning (DL) work that we're doing. Our goal was to ask whether larger left atrial volumes causally increase AF stroke risk. It is a pretty basic question that has been kicking around for decades.

Before I start, I want to acknowledge that this was led by a really talented fellow, James Pirruccello, MD, who is now established as assistant professor at the University of California-San Francisco. James sought to answer this question. We're all well familiar with left atrial enlargement, which we know is prospectively associated with AF, but it's a chicken and egg problem. Is it a larger left atrium that leads to AF, or vice versa? Is that larger left atrial volume causal or just an association? That's what we set out to try and answer. 

To do that, we turned to one of our favorite playgrounds, which is the UK Biobank, which as you heard from Uli Schotten and others, is a prospective study of half a million individuals, middle-aged to elderly, with clinical, genetic, and laboratory data, and magnetic resonance imaging (MRI) data that we've mined for this and other projects. They have about 69,000 studies that are available currently—they're going to 100,000—and it's all publicly available, so it's a great resource for us. 

We set out to measure left atrial structure and function by building DL models to identify the left atrium size and structure. Before medical school, James started, founded, and sold an internet company, so he's gifted enough to move between the science of medicine and the science of artificial intelligence and ML. The first thing that he did was to extract all the images. Now, these are just a couple of representative cine loops from the 2-chamber or 4-chamber view; there are 100 images in each of these loops. If you do 40,000 individuals, you'll have 4 million individual frames to measure. You could try to do this manually, but I wouldn't advise it.

The first thing that we do is called semantic segmentation. When you're building a DL model, you need to tell it: “what is the atrium, what is the ventricle, what is the lungs, or an area to ignore?” That's what he's done here. This is with a program where you literally color on an iPad in the different areas and you say, “this is the left atrium and then go find this in the other 4 million images.” These studies were not specifically designed to study the atrium. The UK Biobank takes an MRI from head to toe, and they do a very dense set through the brain and through the ventricle, but to save time, they did not spend much time on the atrium, which is obviously our favorite part of the heart. We also added the 3-chamber view and the short axis whenever it was available, and I'm going to skip a lot of the methods, but that provided quite a bit of additional information. 

Once we had trained and told the models, this is the atrium, we then were able to extract that information, visually reconstruct it, and make surface projections of it both in systole and diastole. From there, you have both volumes, atrial ejection fractions, minimum volumes, and maximum volumes. So, you now have a series of quantitative traits. With those quantitative traits, the first thing we wanted to do was ask, does the left atrial size and volume correlate with diseases? This is essentially a positive control. If this didn't work, we were going to pretty much call the study right there. As you'd expect, with prevalent diseases, the larger your left atrial volume is, the greater the association with AF, heart failure, hypertension, and stroke.

The same was true with incident diseases as well, as one would expect. Once we had this confirmed that we had these disease associations, we then went on to study the genetics of left atrial size and structure. First, we excluded everyone with AF. We wanted the genetics of normal people rather than those with AF or any structural abnormalities. That left us with about 35,000 folks. For us, running the genetics is pretty straightforward. College kids can do it these days or even high school students can run quantitative genetic studies. We ran standard GWAS. I'm going to skip all of the details of that other than to say that out of the 35,000 individuals, we had 20 different loci, 8 of which as one might expect, overlapped with the genetics of AF. We did see a few old favorites. PITX2, which is the strongest genetic variant associated with AF, came up in a number of these associations, and there were other genes like TTN and a few other favorites as well.

We can sum up the genetic basis of left atrial size and structure in a polygenic risk score and apply that polygenic risk score. We validated it in a separate study called the AllOfUs study, a similar longitudinal study in the United States like the UK Biobank, but with some important differences. In that large study, we had 21,000 AF cases and 400,000+ controls, and what you see is a modest association. So, there's a modest association between left atrial volume, the genetic risk for left atrial volume, and the risk of AF. Obviously, if you want to predict AF, you'd better use a polygenic risk score for AF itself than an intermediate trait. If we compare the AF and the left atrial polygenic scores, there was only a modest correlation between them and R squared to 0.01, so really quite modest. The left atrial PRS is not significantly associated with AF once you adjust for the AF polygenic risk score. 

So, having the genetic basis of left atrial volume allows us to go back to the original question, which is to say, does a larger left atrial size increase AF risk? The problem is epidemiologic associations can be confounded or reversed, and we don't want reverse causality here. Once you have the genetics, the genetics can't go backwards. You can't go from a left atrial size to modifying your genetics. So, that allows you to ask, are these genetic variants a proxy for increased left atrial volume, and ultimately AF? 

Seung Hoan Choi, who now has a lab at Boston University Medical Center and the Framingham Heart Study, did this part of the study and showed that there was a positive correlation and that MR supports that there is a causal effect with left atrial volume on AF risk. 

To sum up, we saw that left atrial volume can be measured in MRI data, surface reconstruction allows the atrial volume measurements even when it's not dedicated for atrial imaging, and left atrial structure and function are heritable, causal risk factors for AF. 

This is the full reference for background. It was a short 3 years in the publication process, and I get to work with a lot of really talented folks. Thanks so much for your time and attention.

The transcripts have been edited for clarity and length.