The Promise and Peril of Population Genomics for the Development of Genome-first Approaches in Mendelian Cardiovascular Disease

Abstract

The rich tradition of cardiovascular genomics has placed the field in prime position to extend our knowledge toward a genome-first approach to diagnosis and therapy. Population-scale genomic data has enabled exponential improvements in our ability to adjudicate variant pathogenicity based on allele rarity, and there has been a significant effort to employ these sizeable data in the investigation of rare disease. Certainly, population genomics data has great potential to aid the development of a genome-first approach to Mendelian cardiovascular disease, but its use in the clinical and investigative decision making are limited by the characteristics of the populations studied, and the evolutionary constraints on human Mendelian variation. To truly empower clinicians and patients, the successful implementation of a genome-first approach to rare cardiovascular disease will require the nuanced incorporation of population-based discovery with detailed investigation of rare disease cohorts and prospective variant evaluation.

The aspiration of genome-first cardiovascular medicine includes the interpretation of a single, otherwise asymptomatic patient’s genome to determine risk of silent or future disease. This is of course an exciting prospect, but the challenges of evidence-based implementation of such a paradigm are broad-ranging. One approach to understanding the application of genomic findings in otherwise asymptomatic patients has been to study these patients in the context in which they might present to the clinic: from the general population.

This approach has been catalyzed by a recent exponential increase in direct-to-consumer and clinical provision of polygenic risk scores, pharmacogenomic reports, and whole genome sequencing, which has been fueled by a reduction in the cost of sequencing. While this rapid rise in production has vastly increased patient, professional and investigator access to such data, what we do with it remains a significant challenge.

Certainly, understanding a priori genetic risk of asymptomatic patients in the population will inform our practice of genome-first cardiovascular medicin. For some problems, this is an appropriate approach: certainly the study of common variants, pharmacogenetics and polygenic risk in the general population can be performed with reasonable statistical power and rigor. But for Mendelian disease, the number of patients observed in the population declines exponentially from those with a rare disease, to even fewer with that rare disease caused by a specific gene, to the vanishingly small number of patients, especially unrelated, who share the same pathogenic variant. To inform the care of patients with Mendlian cardiovascular disease, population genetics can be a rich and fruitful tool, but by itself, will not be adequate to enable the full potential of genome-first medicine. Rather, it will also be critical to bring the years of experience we have in studying these rare diseases in the context of specific patients and their families to bear, and to expand our prospective adjudication of novel and rare variants as well as our understanding of gene-specific disease phenotypes.

Cardiovascular genetics and the charge of genome-first medicine

Historically, cardiovascular medicine has been at the forefront of diagnosis and therapy for genetic disease. The field has identified causative variants in a large number of genes related to Mendelian cardiomyopathy, aortopathy, arrhythmia and hypercholesterolemia. Detailed study of these variants has helped us understand how these genetic etiologies differ from other causes of cardiovascular disease, and has led to guideline-driven changes in management,^1–4 including novel therapies.^5–7 Pathogenic variant identification has enabled cascade genetic screening in families with the goal of reducing the clinical screening burden to identify those at risk of myocardial infarction and sudden death. This proved to be a cost effective strategy in familial hypercholesterolemia and hypertrophic cardiomyopathy even prior to recent decreases in the cost of genetic testing.^8,9 Specifically, the incremental cost per quality adjusted life year was well within acceptable range and in fact it was estimated for HCM that below a threshold of ~$200USD for proband testing, the cascade screening, would be cost saving.⁹ Notably, this is a benchmark that has either been reached or is quickly being approached by commercial clinical genetic testing.^8–10 Lastly, and highly pertinent to a genome-first approach, is the specific success of cardiovascular pharmacogenomics: the discovery and investigation of variants in Cytochrome P450 enzymes modulating effectiveness and reactions to warfarin, clopidogrel and statins have lead to change in practice and first-of-their-kind clinical trials.^11–13 It is because of this success in exploiting the power of genetics that cardiovascular medicine is now uniquely challenged with maximizing the power of the knowledge we have gained to realize the full potential of genome-first medicine for Mendelian disease.

The centrality of population genomics to genetic medicine

To discuss the utility and drawbacks of population genomics data to genome-first cardiovascular medicine, it is important to understand the context through which these resources have emerged.The publication of the first human reference genome in 2003 allowed, for the first time, comparison of sequencing against a full genome sequence.¹⁴ However, the limited number of genomes combined to create this reference did not adequately capture the significant genetic variation in the population. The advent of larger population-genomics datasets including The 1000 Genomes Project¹⁵ and the National Heart Lung and Blood Institute’s Exome Sequencing Project (ESP)¹⁶ made it clear just how much genetic variation there is within the human population, highlighting the crucial nature of population allele frequency to the interpretation of clinical genetic findings. Shortly thereafter, as the use of next generation sequencing grew exponentially, so did the size of available population datasets: the Exome Aggregation Consortium (ExAC) and the Genome Aggregation Database (gnomAD)¹⁷ most recently reported whole exome and genome sequences of over 140,000 patients from a range of geographic populations. These population-scale genomic data have certainly revolutionized Mendelian variant interpretation and the study of genetic architecture, as minor allele frequencies (MAF) in the population are a critical indicator of likelihood of Mendelian pathogenicity. Comparison of variant prevalence in large rare disease cohorts compared to population genetic data has led to significant discovery regarding the genetic architecture of inherited cardiomyopathy at the level of gene and variant causality.^18–22 Because these population-scale datasets contain an unselected cross-section of the population, they were initially most informative in defining background allele frequency, partially enabling variant interpretation based on more accurate estimate of MAF (i.e., population rarity).

The promise of population genomics data in clinical cardiovascular genetics

However, the use of population-level datasets such as ExAC and gomAD as controls in comparison to rare genetic cardiovascular disease is fraught by the inevitable presence of these diseases in the population cross section included. Until recently, very little associated phenotypic data was available for these large scale genomic data sets, limiting their use in disease-related discovery without a well-phenotyped comparison group. To answer this need, several large-scale efforts to unite phenotype and genotype in population-scale datasets began. Some of the largest include the Million Veteran Program, UK Biobank, deCODE genetics, the eMERGE network, BioVU at Vanderbilt University, Geisinger Health and the Penn Medicine Biobank. These prospective cohorts offer not only associated diagnostic codes, but some also include other rich phenotypic electronic health record data. Recent advances in deep learning have allowed high throughput computer vision interpretation of studies including electrocardiograms²³, magnetic resonance imaging²⁴ and echocardiography²⁵ to better define disease phenotypes in these population-scale datasets. Indeed, these rich datasets have already resulted in critical observations, in particular with respect to the common genetic underpinnings of cardiovascular disease. Perhaps some of the most powerful examples of discovery in population genomics datasets have been in type 2 diabetes and hyperlipidemia associated with CAD and MI. The first of its scale combined exome capture, lipid level and outcomes data from over 70 cohorts to assess over 300,000 individuals for variants associated with hyperlipidemia and stroke and identified multiple loci associated with these outcomes.²⁶ The Million Veteran Program later used a similar method in a comparable number of highly phenotyped patients to identify novel therapeutic targets for lowering CAD risk.²⁷ Resultant of these efforts, cardiovascular medicine has been one of the first fields to implement polygenic risk scores into practice, ^28,29 and recent data combines monogenic disease with polygenic risk to refine disease outcome predictions in familial hypercholesterolemia and TTNtv associated dilated cardiomyopathy.^30,31

Some groups have explored the use of these population level data sets to study monogenic diseases. In a large effort combining the Geisinger and Penn Medicine Biobanks, Titin truncating variant (TTNtv) carriers were found to have lower left ventricular ejection fraction and increased incidence of atrial arrhythmia in those of European, but not African, ancestry.³² These findings were corroborated in the UK Biobank.³³ In another example of rare disease, gene burden analysis has demonstrated that loss of function variants in Lamin A/C (LMNA, loss of function defined either by truncation or high predicted damaging variation), are associated with cardiomyopathy, conduction disease and renal disease.³⁴ Arrhythmogenic right ventricular cardiomyopathy (ARVC) -associated variants in Plakophilin 2 (PKP2), which are exceedingly rare in the population, were also examined in the large and well-phenotyped Geisinger database, and were found to display very low penetrance of ARVC in a middle-aged population.³⁵ Lastly, a striking example of the value of familial relationships encoded within population data: a genome wide association study performed by deCODE genetics revealed two rare variants in Filamin C and Nkx2.5 that were associated with dilated cardiomyopathy and sudden cardiac death in Iceland.³⁶

Limitations of population-scale data for the study of Mendelian disease

The promise of these population datasets is clear for understanding the effects of gene-level pathogenic variation on phenotype, determining polygenic risk, and quantifying population penetrance of genetic disease. However, clinicians and scientists must be aware of their limitations in specific settings: First, the population rarity of the majority of gene-specific pathogenic variants makes small sample size likely, even in the largest of population datasets. Most pathogenic variants in monogenic disease are unique to an individual or family, ³⁷ and the site frequency of variation in the human genome decays exponentially to one, with more than fifty percent of all variants in ExAC seen in only one individual.¹⁷ This not only makes statistically meaningful comparisons difficult, but when training classifiers and developing risk prediction models, the rarity of these samples must be accounted for to avoid bias in estimates of accuracy.³⁸ It also limits the ability of population datasets to evaluate variant pathogenicity at the level of unique variants, though novel variant assessment may be informed by location in domains enriched for known pathogenic variation (e.g. in MYH7 and RBM20), or specific variant class effects (e.g., missense versus truncating variants in TTN). ^18,20–22 Creative methods to address this issue by prospective cellular modeling in hypertrophic cardiomyopathy³⁹ and Brugada syndrome⁴⁰, and high throughput phenotyping of deep mutational scans in yeast for catecholaminergic polymorphic ventricular tachycardia⁴¹ have been reported and are on track to become a staple of future variant adjudication.

Second, researchers should be aware of the relative strengths and weaknesses of different population datasets for addressing specific hypotheses. For example, many of these datasets do not include children and may be skewed toward older patients. This indicates that patients perhaps most highly affected by classically monogenic disease may not be captured due either to survival bias or early sequestration in highly specialized medical care. It is important to note as well that the recruitment practices for each of these datasets make them specifically suited to answer some questions and not others. For example, some are enriched for asymptomatic, healthy patients (e.g., Project Baseline⁴²) while others are enrolled from populations interacting with the medical system. Further, ancestry and genetic background contribute significantly to the effect size and expressivity of variants causative of monogenic disease.^32,43 Some databases are of particular value for examining these interactions because they are enriched for patients of multiple ancestries (e.g., the Penn Medicine Biobank⁴⁴ and the Million Veteran Program⁴⁵) while others may represent more homogeneous populations that are enriched for other characteristics of interest (e.g., familial relationships in deCODE genetics³⁶). Of course, the type of genomic data included in each dataset varies from whole genomes to array-based genotyping, which are differentially suited for discovery. Lastly, it is important to reiterate that population cross-sections such as gnomAD and ExAC will include patients with monogenic disease, and so their use in case-control studies must account for this significant limitation.

Lastly, estimates of population penetrance for monogenic diseases are also quite different than what we have observed in published kindreds, and may not be applicable to specific patients. For example, pathogenic and likely pathogenic cardiovascular disease variants found in a large prospective population cohort were associated with a 3-fold elevated risk of cardiovascular death.⁴⁶ Though this risk is certainly substantial, it is lower than published familial penetrance reports. The same was observed with respect to the penetrance of ARVC in PKP2 variant carriers as described above.³⁵ At the same time, in a cohort of unexplained sudden death cases, a series of patients with structurally normal hearts were found to carry pathogenic variants in PKP2,⁴⁷ and in a national registry of cardiac arrest survivors, 21% of otherwise phenotype-negative patients were found to have a pathogenic variant in a known arrhythmia- or cardiomyopathy-associated gene.⁴⁸ These reports remind us that monogenic disease expressivity and penetrance is dependent on genomic context (e.g. as indicated by family history or the effect of polygenic risk of LV dilation and myocardial infarction),^30,31 as well as environment (e.g., alcohol and pregnancy in TTNtv dilated cardiomyopathy).^49,50 And, at the same time, pathogenic variants in some genes may confer greater risk of poor outcome than others.^21,51–53 As we enter an era of genome-first medicine, we must be aware of the limitations of under-estimating a patient’s risk based on population penetrance alone. The presence or absence of associated disease phenotype, family history, and known risk factors and calculators for specific disease (e.g. those for hypertrophic cardiomyopathy⁵⁴) remain powerfully informative even in the context of precision genetic diagnosis, and are certainly critical risk modifiers that can drastically increase a patient’s predicted risk as compared to population penetrance based estimates.

A framework for incorporation of population genomics data to enable genome-first cardiovascular medicine

Population based genomics data can and should be integrated both into our current clinical practice and into the major lanes or research to further enable genome-first cardiovascular medicine. With respect to diagnosis, in the clinic, population genomics data has enabled variant interpretation with respect to population rarity (MAF) per American College of Medical Genetics guidelines for years.⁵⁵ With respect to prognosis, available datasets have shown us that for some diseases (e.g. ARVC), middle aged populations carrying pathogenic alleles are unlikely to develop disease.³⁵ They have also shown us that disease severity may be higher in patients with specific ancestries,³² and that the risk of some more common diseases (e.g., atrial fibrillation) may be higher in otherwise asymptomatic pathogenic variant carriers.³⁴

But, at this time, population data by themselves fall short of accurate estimation of variant-specific disease risk in patients not highly represented in available datasets (e.g. young patients). As is described above, this is largely due to the lack of representation of all possible variants in the general population due both to evolutionary constraint and random mutation rate. Emerging methods for in silico ^21,22,56, in vitro ⁴¹ and in cellulo ^40,57 prospective adjudication of regional and variant-specific effects on protein function can be integrated with population level genomics to yield better estimates of disease causality. As well, with respect to prognostic data for patients identified through genome-first or family screening approaches, population level data often fall short of detailed outcomes predictions due to low statistical power. An important solution here is to continue international efforts to combine gene-specific cohorts from international sources to complete well-designed case-control comparisons.^21,53,58,59 These comparisons must of course account for sampling bias, especially when using large, poorly phenotyped population data sets as controls. Fortunately, by the nature of the expert genomics centers at which these studies are often conducted, they can identify genotype-negative controls, even those from the same kindreds, in order to examine the role of environment, disease phenotype and genetic background on the risk of eventual development of disease and its severity.

Conclusions

At this inflection point in cardiovascular genetics, our field is poised to bring decades of work to bear for a large proportion of the population affected by and at risk for Mendelian cardiovascular disease. Population datasets have been an integral part of the work that has brought us to this point, and will continue to contribute enormously as we move forward. Many critical questions remain to be answered, to some of which population datasets will be central, and others for which these rich datasets can offer complimentary findings. To fully realize the potential of cardiovascular genetics and genomics to change patient lives, merging and reconciliation of observations from familial and disease cohort genetics, molecular and cellular investigations, and population genomic datasets will continue to be required.

Nonstandard Abbreviations and Acronyms:

(MAF)

Minor Allele Frequency

(ExAC)

Exome Aggregation Consortium

(gnomAD)

Genome Aggregation Database

(ARVC)

Arrhythmogenic Right Ventricular Cardiomyopathy

(HCM)

Hypertrophic Cardiomyopathy

(FH)

Familial Hypercholesterolemia