Growing pains in cardiovascular genetics

Abstract

Over half a century ago, heart specialists – there were no board certified cardiologists yet – began to report unusual phenotypes like very long QT intervals or extraordinary left ventricular hypertrophy that ran in families and caused sudden death in the young. In the 1980s and 90s, linkage analysis in large families identified the first disease genes, landmark discoveries that have been critical for understanding basic physiologic and pathophysiologic processes in the heart, counselling families, defining penetrance that is so often incomplete, and identifying phenotype positive patients in whom coding region variants in these first disease genes were absent. In those patients, disease pathways defined by early rigorous linkage-based studies implicated new candidate genes for mediating these cardiovascular genetic phenotypes. So, for example, once beta-myosin heavy chain mutations were identified in hypertrophic cardiomyopathy (HCM), genes encoding other contractile proteins became logical candidates; similarly, ion channel genes or modifiers were candidates in channelopathies and desmosomal proteins in arrhythmogenic right ventricular cardiomyopathy (ARVC). As a result, we now have long lists of disease genes for major cardiovascular genetic diseases, and dramatic improvements in sequencing costs and efficiencies have enabled widespread application of panel-based sequencing. While these advances have improved care of affected patients and their families, two papers in the current issue of Circulation highlight potential flaws in the logic that underlies increasing use of these panels across large patient populations.^1,2

What constitutes a disease gene?

Patients with an obvious phenotype but no variant in a recognized disease gene and no large kindred for formal linkage present the community with a problem (or opportunity) of discovering new disease genes. In such patients, identification of a rare variant in a disease pathway-related candidate gene has been used to label the variant as causative. Further evidence to support causality can be generated by segregation of a disease phenotype with a rare variant in multiple members of a kindred and, where in vitro assays were available (notably for ion channels), the finding that a rare variant altered function. This disease pathway-based logic has generated large numbers of putative disease genes, and these are routinely screened in sequence-based arrhythmia or cardiomyopathy panels. However, the evidence base linking these genes to disease is weaker than that derived from statistically-driven linkage in large kindreds. The NIH’s Clinical Genomic Resource (ClinGen) has proposed a framework for re-evaluation of these disease-gene associations,³ and Hosseini et al.¹ have now used the ClinGen approach to evaluate the evidence that relates variants in 21 genes reported to the Brugada Syndrome (BrS).

The process, which includes multiple independent teams reviewing source literature using standardized ClinGen criteria, concluded that available evidence met current standards for designating a disease gene in only one instance, the cardiac sodium channel gene SCN5A. The other 20 genes were classified as “disputed”, based largely on very limited or no segregation and/or no evidence that the rare variant was more common in cases than controls. Interestingly, the initial description of SCN5A as a BrS disease gene was based exclusively on biologic plausibility and not linkage,⁴ although subsequent work in a large kindred has provided evidence of linkage to the SCN5A locus.⁵ Conversely, one gene, GPD1L, previously associated with BrS by linkage in a large kindred, was downgraded to “disputed” for two reasons: first, the linked region was large and contained multiple genes that were not evaluated, and second, the putative causative variant is now known to be not very rare, with a minor allele frequency of ~1/6000 in Europeans and South Asians in the Genomic Aggregation Database (http://gnomad.broadinstitute.org/), which reports sequence variants discovered by whole exome and whole genome sequences across >140,000 subjects of diverse ancestries.

The Hosseini study does a service to the community by identifying disease genes that do not meet contemporary standards for pathogenicity, and thus, for example, should not be included in panel testing. One can quibble with some of the designations; for example, GPD1L was downgraded in part because the variant was considered too common to be pathogenic. However, the whole question of how rare a rare variant has to be in order to be designated pathogenic is increasingly muddy. One approach uses population frequencies, as in gnomAD, and in this case an allele frequency <3/100,000 for long QT syndrome variants has been proposed.⁶ However, in founder populations, disease-associated variants can in fact be much commoner; in a Quebec founder population, a pathogenic LDLR variant (a 10 kb deletion) causing familial hypercholesterolemia is present in 1/200 individuals.⁷ The converse also appears to be true: screening of monogenic disease genes across large populations not selected for any particular phenotype often turns up patients with no phenotype but nevertheless carrying rare variants designated pathogenic.^8,9 Just as ClinGen is re-evaluating reported associations between Mendelian disease and specific genes, criteria have also been promulgated to evaluate pathogenicity of individual variants in disease genes.¹⁰ Application of these criteria, which include linkage, population frequency, and in vitro data, have resulted in proposed “demotion” of previously-designated pathogenic variants to uncertain significance or benign.¹¹ We have also emphasized that while variants may be pathogenic, penetrance is often incomplete, further complicating rare variant interprtation.¹²

The extent to which the BrS is a monogenic disease also muddies these waters. As the Hosseini report highlights, many kindreds have only single affected individual, and genetic testing, whether in SCN5A or all 21 genes, still yields a likely variant in well under half of cases. There is now very good evidence that common variants, at least some of which are known to modulate ion channel gene expression, contribute to variability in the Brugada phenotype,¹³ so this disease is a nice example of how autosomal dominant “monogenic” diseases may actually be multigenic, with common and rare variants contributing to the observed clinical phenotypes. The challenges in validating the role of commoner variants alone, or in combination as “genetic risk scores”, in these diseases are at least as large as identifying rarer variants with larger effect sizes.

Evaluating the potential genetic basis for unusual ECGs

Widespread genetic testing in individuals with no manifest phenotype runs the risk of labeling patients with a disease they do not have.^8,9 Sheikh and colleagues² emphasize this point in their current study of 100 athletes that did have a possibly genetically-mediated phenotype, T wave inversion. This highly selected referral population underwent extensive clinical evaluation to establish whether a possible genetic disease was present, and were also sequenced in 311 candidate cardiomyopathy or channelopathy genes. There were 21 with a clinical diagnosis, primarily HCM, and a genetic diagnosis was found in 8/21 (in undisputed disease genes), while genetic variants, likely unrelated to the presenting phenotype, were found in 2 of the 79 athletes with no clinical diagnosis. The study also raises the interesting question of the extent to which athleticism per se elicits a phenotype in the presence of genetic disease, as is well-recognized in ARVC.¹⁴ The manuscript also contains the charming phrase “pathogenic polymorphism” describing a TTR variant found in 4% of subjects of African origin but nevertheless designated pathogenic. While conventional wisdom argues this is “too common” to be pathogenic, the founder examples above, the idea that “monogenic” disease phenotypes may be multigenic, and the precedent that common variants can elicit important phenotypes in the face of environmental inputs (e.g. on exposure to certain drugs¹⁵) reinforce the sense that strict cut-offs for what is and is not pathogenic have to be taken with a grain of salt.

Are we now smarter?

One expansive view of personalized medicine incorporates whole genome sequencing at an early age with appropriate counselling for disease risk and management and future therapies through life. The studies discussed here emphasize that at least at this point, we do not have good examples in the cardiovascular domain of situations in which genetic testing in the absence of a phenotype or a family history yields actionable information. It may be that such a preemptive sequencing approach is reasonably applied in selected settings, such as screening for cancer gene or FH variants in high risk ancestries. The Sheikh report strengthens the argument for the primacy of clinical evaluation, and supports the contention that genetic screening in the absence of a phenotype is a fool’s errand. The Hosseini report emphasizes that the field and the tools available to us are rapidly evolving so we need to be open to reinterpreting conventional wisdom. Both studies highlight the increasing complexities of deploying genetic testing of any type in a routine clinical environment. Medical geneticists and genetic counsellors are vital partners in this adventure but often lack content expertise relevant to cardiovascular diseases. Just as the first discoveries in the field were made before there were certified cardiologists, it may be time to consider creation of certification of a specialty in cardiovascular genomic medicine, as technologies, thinking, and knowledge in the field continue to evolve.