SOBRE ESTE EPISÓDIO
Jane: Hi. Welcome to episode three of Getting Personal, Omics of the Heart. I'm Jane Ferguson and this podcast is brought to you by The Functional Genomics and Translational Biology Council of The American Heart Association. In this episode, I talk to Jonathan Mosley about an interesting genetic method he has developed to look at shared genetic contributors that influence risk phenotypes, as well as disease risk, which can be used to integrate data from prospective studies with large scale electronic health record data. We also highlight a recent AHA scientific statement on genetic literacy and Nevine and I discuss the latest in precision medicine.
Our large hurdle in implementing precision medicine will be to increase understanding of genetics and genomics amongst healthcare providers. Sima Mittal, Karen [inaudible 00:01:11] and colleagues tackled this issue as part of a recent AHA scientific statement published on behalf of The Council on Functional Genomics and Translational Biology, The Council on Cardiovascular Disease in the Young, The Council on Cardiovascular and Stroke Nursing, The Stroke Council, The Council on Lifestyle and Cardiometabolic Health and The Council on Quality of Care and Outcomes Research.
As the promise of genetics guided treatments is becoming a reality, cardiovascular healthcare providers often struggle to stay up to date on this large and rapidly advancing field. Although there is a need for more dedicated genetics professionals, such as genetic counselors, it is also important that all cardiovascular practitioners maintain core competencies in cardiovascular genetics. This statement entitled, "Enhancing Literacy in Cardiovascular Genetics" was published in the October 2016 issue of Circulation Cardiovascular Genetics and outlines useful information for the cardiovascular practitioner when considering genetic and pharmacogenetics testing and includes pointers to resources for enhancing knowledge and genetics and genomics. As always, this and the other papers mentioned in this episode are linked on the podcast website at fgtbcouncil.workpress.com.
Hi, Jonathan. Thank you for joining.
Jonathan: Thank you, Jane. I'm happy to be here.
Jane: Maybe first I'll give you a chance to introduce yourself.
Jonathan: Sure. My name is Jonathan Mosley. I'm up here at Vanderbilt University. I'm an instructor in the department of medicine and do both a little bit of clinical work related to hypertension but spend most of my time doing research, and in particular genetic research, using Vanderbilt's integrated electronic health record and biobank data system we call Bioview.
Jane: Which is a really fantastic resource and you've been doing some really interesting things with it. Today, I wanted to talk about sort of an interesting method and the application of that method that you've developed. For our listeners, there's a link to these papers on the website but you can find them. We're going to be talking about two different papers. One of them called Defining a Contemporary Ischemic Heart Disease Genetic Risk Profile Using Historical Data and was published in the December 2016 issue of Circ CV Genetics and then a second sort of related paper entitled Investigating the Genetic Architecture of the PR Interval Using Clinical Phenotypes is in the current issue of Circ CV Genetics, which is the April 2017 issue.
Jonathan, maybe to start you could tell us a little bit about sort of your thinking behind the development of this new method.
Jonathan: Yeah. I can kind of give you a rationale. I'll focus on that first paper initially. The content area really relates to ischemic heart disease or coronary heart disease, so disease due to blockages of your coronary arteries. This disease in particular it's really I think an epidemiologic success story. Really over the last several decades there's been a pretty marked decline in death due to coronary heart disease. A good portion of that can be attributed to the fact that we've really started to understand this disease and understand risk factors. I think the framing in studies is always brought up at the prototype but study these prospective studies like Framingham. Help us identify risk factors that modulate risk of coronary heart disease. By changing behaviors like smoking and treating cholesterol levels, we've been able to now really change the epidemiology of this disease.
There's also been some other changes in society I should add. We've become a much heavier society and so this also modulates our risk of coronary heart disease. If it were cheap and easy to long perspective studies, you could imagine that it would be desirable to start a new Framingham today kind of under the notion that the epidemiology of this disease has changed. Let's create a new cohort that reflects today's epidemiology so we can really target the prevalent risk factors that we're seeing in our population right now and really try to target those risk factors.
It's not really a feasible approach. Starting a perspective study is constrained for quite a few reasons.
Jane: Right. None of us have the funding for that.
Jonathan: Yeah. In terms of cost and then there's always an inherent lag also in these perspective studies in terms of you have to wait for outcomes to occur too. Kind of a rationale underlying this first paper is is there a quicker way which we could try to get the outcomes that we might otherwise expect from a perspective study? At Vanderbilt, we have this data resource that I mentioned earlier called Bioview. A large amount of the health information and data related to clinical encounters are captured in electronic form and this is available in a de-identified form for research. We have a dataset of over two million clinical records or records on two million individuals that we can use for research. About 235,000 of these individuals also have DNA available. In particular, we can do research looking at genetic modulators of disease.
In this dataset that I have, I can quickly identify thousands of cases of various outcomes like myocardial infarctions or ischemic heart disease or diabetes. We can identify far more of those outcomes than you'd except really in most prospective studies. They're not large enough to observe that many outcomes but really a limitation of these is that we don't collect standardized baseline measurements within individuals who are captured in electronic health records. In other words, they're just encountering the health systems for different reasons but there's no protocol that describes specific baseline data that should be captured. It's really not feasible to do a perspective study or even a retrospective study despite having all these great outcomes.
The rationale for the current study was really can we take all these outcomes that we're observing in institutions like Vanderbilt and others and perhaps get the baseline risk factors from another source. In this kind of imagined study design, you take risk factors that you're able to measure under epidemiologic conditions and then see whether they're associated then with these outcomes that we're observing in our EHR systems. That's really the study design. That's really the hypothesis that I was exploring. Could you implement this particular study?
Jane: That's really interesting. I think it's a great way to sort of maximize the availability of data from different sources to sort of take what you need from different studies and combine it. For this kind of study, if you're combining data from two different studies, how similar do the demographics have to be? Do these have to be from people in the same country, the same racial group? How important are those sort of factors?
Jonathan: Using standard epidemiologic methods, it's not possible obviously to link kind of risk measured in one population and outcomes measured in another population. Really, emerging genetic methods allow us to do this. What we're doing is we're capturing or using genetics really to measure or to link these risk factors in one population into another population. Issues like race are important. We certainly know that there are racial differences that if you don't account for these they can really lead to a lot of confounding and unexpected findings. It's important that you can get populations that are genetically similar for the method that we used.
Jane: Yeah. Maybe we can go into the results of what you found in the first study looking at ischemic heart disease genetic risk.
Jonathan: Yeah. Let me just kind of tell them a little bit about the genetic method that we used to associate these. What we did is we used a method measure the effects of large numbers of snips on a phenotype. This method is fairly similar to or has a lot of things in common with kind of very old genetic methods where you'd estimate heritability. Often, you'd do this by you'd collect related individuals and you wouldn't directly measure their genetics or how genetically similar they were, but you could infer how much genetics they shared by knowing their relationship. For instance, a mother and a child share on average half their genetics. Then what you can do by kind of inferring how genetically similar individuals are, you can measure a phenotype and then estimate the contribution of genetics to that phenotype based on the shared.
With kind of current genotyping methods and with appropriate sample sizes, now you can do similar types of analysis, but instead of inferring how genetically similar people are, you can actually measure how similar they are based on common snips. There's basically computations you can do across large numbers of snips to estimate how genetically similar they are. Once you have the similarities, then you can quantitate really how much of the variability in the phenotype can be modulated by genetics. That quantity is often called the chip heritability or it's really the heritability that can be due to large numbers of snips. You can estimate the heritability of two phenotypes but you can also measure the extent or quantitate the extent to which those genetics are shared between two phenotypes. This quantity is called the genetic correlation.
What we did in this particular study is that we took baseline risk factors that were measured in the Eric population. There's about 8,000 unrelated individuals of European ancestry. We took baseline risk factors that had been measured in their first visit and then we took two different outcomes that we curated from our electronic health record dataset. This was the dataset that actually came from the emerged population, which is a collection of institutions like Vanderbilt that have come together to pull their resources in order to create larger datasets for study.
What I did was measure then the genetic correlation between these 37 baseline factors measured in the Eric study and these two phenotypes. The first one I looked at was type II diabetes. I kind of think of that as a positive control. It's fairly easy to clinically diagnose or create a definition of diabetes. Basically, if you have a glucose above a certain threshold then you have diabetes. It's fairly easy to standardize. I expected that this phenotype would work well. What I did was I measured the genetic correlation between each of those 37 risk factors and type II diabetes and then I also did a longitudinal analysis within the Eric study where I measured hazard ratios, so a standard epidemiologic measure of risk, basically for each of these 37 risk factors to find out kind of the epidemiologic association between these risk factors.
When I compared them, what I found was that genetic correlations or measures of genetic risk across populations very closely corresponded to the hazard ratios or a standard epidemiologic measure of association. Suggesting for type II diabetes, really the epidemiologic risk seems to be measuring the genetic risk of the disease.
Jane: Yeah, which is interesting. I guess for diabetes, we know a lot of the risk factors but then can your method say how many sort of the risk factors are still missing for example? Would you conclude from this that we know most of the things that cause diabetes or could this method be used for finding novel biomarkers for example? Instead of looking at only 37 risk factors, if you looked at more do you think you would add a lot of additional information or do you think we already know sort of the bulk of what's increasing risk of diabetes?
Jonathan: I think that's really kind of ... You hit on the important point. This analysis was done using well-known risk factors but really I think the ultimate goal would be discovery. I think there is a lot more discovery to be had, and in particular, whether a risk factor may or may not account for a lot of risk, you might identify new biomarkers that may help you make an earlier diagnosis that can allow you to then perhaps modify the course of the disease.
Jane: I know in this paper, as well as looking at diabetes, you were looking at ischemic heart disease and finding a different pattern to what you saw for diabetes.
Jonathan: Yeah. The conclusion from the diabetes was that diabetes that we're diagnosing clinically, that seems to be very similar to the diabetes phenotype that was identified in the Eric population. That didn't seem to be true of the ischemic heart disease where we actually saw that the risk factors, again, these 37 factors that we looked at, really the genetic measure didn't follow the epidemiologic measure nearly so well. We do see some important risk factors where they did follow each other well. For instance, high systolic blood pressure, high triglycerides increase your risk. Low HDL increasing your risk. Even smoking, which is interesting, how you smoke is actually genetically modulated, so we can measure smoking as a risk factor and see that smoking increases risk.
There are some differences that caught us by surprise. For instance, LDL cholesterol, which is a strong known risk factor for ischemic heart disease, showed a genetic correlation of close to zero, suggesting that there is not a lot of shared genetics going on between modulators of LDL levels and ischemic heart disease in our population. That really caught us by surprise.
Jane: Yeah, it's interesting. Of course, LDL is one of the risk factors that's most treated and that we're most aware of and we have a lot of different therapeutic options for treating LDL cholesterol. Do you think have we sort of reached the end of how much additional effect we can get from treating LDL? Are we already so good at treating LDL that any sort of additional therapeutic options down that avenue may not give us any additional gains in preventing disease or do you think there's some other explanation for why LDL did not appear to have a strong correlation?
Jonathan: I'd certainly like to think that we're great at treating LDL here at Vanderbilt and it certainly could be one contributor is that we've perhaps attenuated the genetic effect of LDL. Unfortunately, kind of with the nature of the dataset that I used, I didn't have information on use of statins and other drugs that could give a sense of whether that was an important effect modifier. What the patterns of associations also shows in our dataset, again, this association with low HDL, high systolic blood pressure, high triglycerides seem to be a big driver, is that we really might have a population where the metabolic, this syndrome really driven being overweight and obesity, is really an important driver of risk in our population. It's possible that we might represent this changing epidemiology of the disease.
Our other thought was that maybe just LDL doesn't work as a phenotype but we actually looked at another outcome, peripheral artery disease, and actually found a pretty strong association with peripheral artery disease. I don't think that there's an inherent problem with the phenotype. I think it's an excellent question and I think it's something that we're still trying to figure out the answer to.
Jane: Maybe for ischemic heart disease, maybe treating the sort of obesity, metabolic aspect may be more important for helping these individuals.
Jonathan: Yeah and I don't think our data supports stopping treating LDL but maybe it's possible that we can say we're doing a good job, at least in this particular population that we studied.
Jane: Right, right. You've used this method and you sort of showed it really nicely with the ischemic heart disease and then the type II diabetes, and then in your more recent publication, you've shown how this can be applied to other phenotypes sort of in a more directed way. I'd love to chat about that a little bit.
Jonathan: Yeah. In this paper, it's really running that same experiment backwards. Again, because we're measuring risk across populations and we're doing it based on underlying genetics, which your genetic risk is determined at birth, so we can either start with risk factors or diseases and there's kind of no temporality in this particular study design. In the second paper, what we did is somewhat run the experiment backwards where we took a risk factor or a biomarker really, in this case it was the PR interval derived from the cardiac electrocardiogram, and we asked the question for what diseases does the genetic risk driving that disease did those same genetics also modulate the length of your PR interval?
What we found is one, with atrial fibrillation, and one was with measures really related to adiposity. Genetic factors which tend to make you heavier prolong your PR interval. The surprising finding here was that actually genetic factors which tend to shorten your PR interval increase your risk of atrial fibrillation. This seemed to at least conflict with a little bit of what's been published in the epidemiologic literature where the association has gone in the other direction. That was an interesting observation. Kind of our bottom line that we came to is actually if you go through the literature as a whole, there's really a U shaped relationship between PR interval duration and atrial fibrillation risk. We think that the genetics might be contributing to that lower end of the U or the inverse relationship.
Jane: Yeah, which is really interesting. You have to be in that sweet spot right in the middle of not too long and not too short.
Jonathan: Lifestyle factors. Again, obesity is always such a big driver of these things and metabolic phenotypes just tend to modulate a lot of these biomarkers.
Jane: Right, right. Really interesting. For your next studies, maybe you'll be looking at other obesity related phenotypes I guess.
Jonathan: Yeah but also, as you alluded to earlier, really the next step is to now start exploring more novel biomarkers. These studies allowed us to use pretty well described phenotypes and biomarkers to give us a sense of expectation of the results that we might see but now really we hope to move on to more discovery and novel discovery.
Jane: I think it's a really exciting method. It has a lot of promise and it'll be really interesting to see where you go next with this.
Jonathan: Yeah. Well, thank you for the opportunity to talk about it.
Jane: Yeah, thank you so much.
Hi, Navine. How are you doing?
Navine: I'm doing well, Jane. A couple of exciting papers have been published in both Circulation and Circulation Cardiovascular Genetics I see.
Jane: Absolutely. There's two that were published in Circulation in this month, so April 4th issue of Circulation that I thought were pretty interesting. They were sort of related to each other thematically. The first one is called Genetic Risk Prediction of Atrial Fibrillation and this was published by Steven Lubitz, Xiaoyan Yin, Emilia Benjamin and colleagues on behalf of the AFGen Consortium. What they did was they looked at variance associated with atrial fibrillation and they generated this AF genetic risk scores and then they looked at the association between this genetic risk score and incidence of atrial fibrillation in five prospective studies. That was almost 19,000 individuals. As well as looking at the relationship between this risk score and AF, they also looked at the relationship between the risk for stroke in a separate study of 500 stroke cases.
What was sort of interesting is they were able to find that this genetic risk score was associated with the incidence of atrial fibrillation and it did add a little bit of additional discriminative power beyond just looking at clinical risk factors for AF but it was moderate. The addition of this genetic risk score didn't add a huge amount of additional predictive capacity.
They did also find that this genetic risk score was associated with stroke. It's sort of an interesting example of using a genetic risk score to look at prediction of incidence disease and also sort of related diseases such as stroke.
Navine: Sure. I think genetic risk scores are useful when the effects of each individual variant are very low. Compiling comprehensive genetic risk score may add incremental value, and of course being able to predict the onset of atrial fibrillation is very valuable for many patients, as it's a common cardiovascular condition. Hopefully, this will be refined as we move forward.
Jane: Absolutely. Actually, the second paper was sort of a similar application. This paper is called Common Genetic Variant Risk Scores Associated with Drug Induced QT Prolongation and Torsade de Pointes Risk. The authors here were David Strauss and Christopher Newton-Cheh and colleagues. They generated a genetic risk score variants that have been associated with QT intervals through sort of various genome [inaudible 00:27:54] association studies and then they did genetic analysis in a relatively small number of subjects, 22 subjects, but they looked at the association between this genetic risk score and the drug induced QT prolongation. They found that it actually explained quite a significant proportion of the variability in drug induced QT prolongation and it was a significant predictor of drug induced Torsade De Pointes.
In this case, compared to the first paper where it had a relatively modest effect, they actually saw quite a good effect here where the addition of the genetic risk score was able to predict the reaction to the drugs.
Navine: Yeah, Jane. I think as genetic information for each individual patient is going to be increasingly available, as whole genome sequencing or whole [inaudible 00:28:52] sequencing or even doing [inaudible 00:28:54], the costs seem to be going down. If patients have a digenetic profile available, then compiling such genetic risk scores and then being able to apply them for individual situations would make sense. Then I think it could be more widely clinically applicable because people should realize getting genetic testing is just not getting genetic testing but it's a lifetime of information that's available as we are able to use this genetic information in various clinical conditions as we are seeing in these two papers.
Jane: Absolutely and it actually relates back to the AHA statement that we highlighted in this episode where genetic literacy is so important. It emphasizes the fact that practitioners and healthcare providers need to be aware of genetic testing an be aware of the potential that it has so that when genetic data is available for patients it can be used throughout the whole lifetime of that patient.
Navine: That's great, Jane. I found an interesting paper that was published in this month, April 2017 Circulation Cardiovascular Genetics. It's volume 10 and it's titled Prevalence and Clinical Implications of Double Mutations in Hypertrophic Cardiomyopathy. The first author is Dana Fourey and the senior author is Arnon Adle. Essentially, what this study did was looked at all the hypertrophic cardiomyopathy in gene panel results that were available in 1,411 patients over a 12 year period and try to discern how many of these patients were genotype positive. It turns out that 19% of these patients, or 272 of the 1411, had pathogenic or likely pathogenic variance.
The purpose of this paper was to see how many of these patients have so-called double mutations because having two pathogenic or likely pathogenic mutations in earlier studies have shown an earlier disease onset, more severe left ventricle hypertrophy, higher prevalence of advanced heart failure and increased risk of sudden cardiac death. They wanted to see if this was true in this large cohort of patients. It turns out that actually just 1.8% of that total population, or 25 patients, had such double mutations; meaning, two likely pathogenic or pathogenic mutations. Fairly small number of the total population.
What was interesting was as they applied the latest American College of Medical Genetics and Genomics criteria, and remember this was done over a 12 year period, and what they decided to do was look at all these 25 variants and apply the latest criteria and see after this stringent overview whether these likely pathogenic or pathogenic variants held up. It turns out that of the 25, only one genotype actually held up to these criteria. Only one patient had these double mutations that were pathogenic or likely pathogenic, so 0.07% of the total population. The bottom line is, though it sounds interesting that if people have two likely pathogenic or pathogenic mutations they should theoretically have a worse prognosis and more severe disease, the reality is that this is unusual, and furthermore, when they compared these patients who had double mutations with those who had single mutations, they found no difference in these high risk features or premature death.
It turns out that our knowledge is still evolving regarding having double mutations in hypertrophic cardiomyopathy. There's a nice accompanying editorial in Circulation Genetics April issue of this year if people want to read further into this article.
Jane: It sounds really interesting. I'm thinking that that is a little related to the other paper that I wanted to talk about, which is not specifically cardiovascular related but I think they used an interesting approach that could be applied to cardiovascular conditions. The first Author is [inaudible 00:33:56] Cummings, last author Daniel MacArthur and this was published this month in Science Translational Medicine the 19th of April. The title is Improving Genetic Diagnosis in Mendelian Disease With Transcriptome Sequencing.
What they did was they took muscle samples that had been collected from patients with a variety of rare mendelian conditions such as muscular dystrophies and myopathies. They decided to do RNA sequencing in these samples. These samples were from subjects and families that had previously been very difficult to diagnose. A lot of these subjects had already been subjected to whole exome sequencing or whole genome sequencing, and in many cases, they had been unable to find the [inaudible 00:34:43] mutation.
They sequenced the RNA from the muscle and they compared it to control muscle RNA from [inaudible 00:34:53]. [inaudible 00:34:54] contains RNA data from hundreds of different individuals with healthy controls and they were able to filter the data from [inaudible 00:35:01] to get high quality RNA samples matched to the patient's age and BMI to some degree. They then compared the sequences to see if they could find aberrant splice variation or aberrant expression in the RNA samples from the patient samples compared with the controlled. They actually were able to find a lot of additional interesting causal mutations that were able to explain the diagnosis in subjects who were previously undiagnosed.
Actually, in this sample of very challenging rare cases, they had an overall diagnosis rate of 35%. By identifying sort of aberrant splice patterns in these patient samples, they identified multiple causal variants that were not able to be identified through the usual means through whole genome sequencing or whole exome sequencing. They actually found 17 families where they were able to make a diagnosis where there had previously been none.
Although this is in different conditions, not cardiovascular, but I think it highlights how sometimes using a different approach, for example doing transcriptome instead of genomic profiling in the disease relevant sample can give really interesting insight that you wouldn't get just from looking at the DNA sequence.
Navine: That's a fascinating approach. This is more like a genotype transcriptome correlative study after the transcriptome has been further refined based on so-called normal transcriptome and this way they were able to identify the functional significance of certain genetic variants based on what the transcriptome looks like in disease states.
Jane: Yeah, absolutely. They were able to sort of actually link variants of unknown significance with the actual transcriptomic pattern and then highlight the variants that actually did have a causal effect on gene transcription.
Navine: Thanks for pointing that out, Jane. It could be easily applicable to cardiovascular disease. It'll be interesting to see if papers come out based on this study design.
Jane: Yeah. I think so.
Navine: All right. We look forward to reviewing some more exciting papers next month. We'll be well into summer soon.
Jane: We will. Great. Well, thank you, Navine.
Navine: Thank you.
Jane: That's all for this month. Thanks to Rick Andreasen at the Mayo Clinic Media Support Services for production assistance. Thanks, everyone, for listening. We look forward to bringing you another episode of Getting Personal Omics of the Heart next month.
