Interpreting Incidentally Identified Variants in Genes Associated with Heritable Cardiovascular Disease

The Human Genome Project and a vision for Genomic Medicine

The first draft of a human genome – the product of the Human Genome Project (HGP) – was unveiled in the East Room of the White House just over 20 years ago, in spring 2000. It is instructive to recall that there were many naysayers when the HGP was started: it is not possible to overstate how the the fruits of that effort have revolutionized biomedical science, from our understanding of basic mechanisms to therapeutics (think PCSK9 inhibitors or covid-19 vaccines…) to, increasingly, clinical care. It is the cost of accurate genome sequencing that drives the idea of incorporating genetic information into the flow of clinical care: the first genome cost over $3 billion to produce, and whole genome sequencing now costs well under $1000 – less than the cost of a single cardiac MRI.

In 2010, the National Human Genome Research Institute (NHGRI, the successor to HGP) created a section of Genomic Medicine and a working group to think about how to integrate emerging genomic knowledge into clinical care. That working group proposed that "genomic medicine is an emerging medical discipline that involves using genomic information about an individual as part of their clinical care (e.g. for diagnostic or therapeutic decision-making) and the health outcomes and policy implications of that clinical use." The scientific statement that Andrew Landstrom and colleagues have generated addresses the "Variant of Uncertain Significance (VUS)" problem that currently represents an increasing obstacle to implementation of this vision of Genomic Medicine.

Personalized and Population-based Medicine

While the vision of delivering medical care based on a person's genome and other personal attributes such as environmental inputs (including the social determinants of health) has great appeal, it has also been viewed as undermining evidence-based approaches, often derived from studying large populations. However, that seems a false dichotomy since while it is true that most people are average for most traits like disease susceptibility or drug response, it is equally true that all people are at high (or low) risk for some traits. Thus an opportunity that a vision of Genomic Medicine presents is to identify individuals with those extreme risks and prevent disease.

How common are genetic variants?

Single nucleotide polymorphisms (SNPs), the commonest type of DNA variant, occur every 1000 basepairs or so; other variants such as insertion or deletions are less common. An average human harbors 84.7 million SNPs and 3-4 million other variants.¹ Only 1-3% of the genome is protein-coding so most efforts to date have focused on the "missense" SNPs and other variants that alter the amino acid sequence of the encoded proteins; variants in non-coding regions can alter protein function by modulating gene expression or splicing.

A few variants are common, many are rare but seen in many individuals, and most are very rare. In fact, most are so rare that every new genome sequence discovers new SNPs. Some are clearly disease-associated (designated P/LP, pathogenic or likely pathogenic), some are known to be benign (B/LB), and because most are so rare they have no well-understood function – they are VUS, and ultimately more VUS turn out to be benign than pathogenic. The problem that VUS present is that if every VUS were to trigger some clinical action (diagnostic testing or treatment), we would create generations of patients and families with needlessly heightened anxiety, we would bankrupt our healthcare system, and we would benefit only a few patients.

How do we know if a variant is pathogenic or benign?

The American College of Medical Genetics and Genomics (ACMG) has promulgated a series of criteria (and weights for each) to make this assessment.² These are

• Functional data: The results of a well-established functional assay, ideally calibrated with known pathogenic and known benign variants, can be used to assign pathogenicity.
• Segregation data: Whether or not a VUS is found in other affected family members and absent in those who appear unaffected.
• Population data: Large databases (such as the Genome Aggregation Database, gnomAD³) present SNP frequencies, across diverse ancestries, in hundreds of thousands of individuals. The assumption is that variants that are common are less likely to be pathogenic than those that are very rare.
• Computational data: In silico tools, although imperfect, can be used to predict mutant protein function. Further weight is added if a SNP results in an amino acid change at a protein position where a different SNP is known to be pathogenic.
• Other data, such as multiple literature reports, can add further weight.

• The scientific statement suggests a Bayesian approach to VUS evaluation – an assessment of pathogenicity based on known factors, followed by deep phenotyping of the variant carriers (ECG, imaging, procainamide challenge, etc), and reassessment of pathogenicity. While this is a somewhat looser framework than that proposed by ACMG, there is an appropriate emphasis on the importance of teams and national and international collaborations.

Understanding and managing the VUS problem

There are multiple scenarios under which sequencing identifies a variant in a known disease gene: (1) an established clinical diagnosis (hypertrophic cardiomyopathy, long QT syndrome, etc) and genetic testing is undertaken to inform subtype-specific care and to initiate cascade screening in the family; (2) a strong suspicion of an as-yet-undiagnosed genetic condition (cardiac arrest in a healthy person, a critically ill infant, a child with neurodevelopmental delay); (3) no clinical phenotype (e.g., direct to consumer or participation in a research study). In some of these scenarios, a P/LP variant in a cardiovascular disease gene may be identified when there was no clinical phenotype that prompted the sequencing. To deal with such "incidental findings", the ACMG has promulgated a list of 73 genes in which P/LP variants should be returned.⁴ Regardless of the scenario, the identification of a P/LP variant should prompt a clinical evaluation as described in the scientific statement. Notably, even if no clinical phenotype is detected, cascade screening should be undertaken since many cardiovascular (and other) genetic diseases are incompletely penetrant, i.e. not all P/LP carriers have disease. Similarly, identification of a clear B/PB variant should prompt no further evaluation.

Historically, many genes have been associated with Mendelian diseases on the basis of what we now recognize is weak evidence. For example, in a patient with long QT syndrome, the finding of a rare variant that in an in vitro system alters ion channel function has in the past led to the labeling of that gene as disease-causing. Importantly, while consensus expert review by the NHGRI's ClinGen consortium suggests that many of these designations are very weak,^5–7 many such genes are still routinely included on arrhythmia or cardiomyopathy panels. Many cardiovascular disease genes are large (e.g., TTN, RYR2, SCN5A) and so variants, most of which are VUS, are especially common.

Even for genes whose role in disease is well-established, acting on a VUS runs the risk of interventions that turn out to be inappropriate if the VUS is ultimately designated benign. We have seen ICD implants for KCNQ1 VUS in individuals with normal QT intervals,⁸ and cases of mastectomy for BRCA1 VUS have also been reported.⁹ As emphasized in the scientific statement, the finding of a VUS should not trigger any clinical action or cascade screening but rather should generate further evidence development. This can include efforts to accumulate phenotypes in other carriers of the same VUS (e.g. using web-based tools such as MatchMaker¹⁰) or further functional study as discussed below.

Glimpses into the future

Function at scale: Current estimates suggest that ~600 million SNPs have been identified to date. However, a "back of the envelope" calculation based on the known de novo mutation rate, the population of the earth, and the size of the human genome (3 billion basepairs) suggests that every missense variant that is compatible with life currently exists in ~51 individuals alive today.¹¹ Thus the VUS problem will grow rapidly, especially as non-European ancestry people (few of whom have been studied to date) are increasingly included in sequencing efforts. "One at a time" conventional approaches cannot meet the challenge of establishing in vitro function, and high throughput methods are now being deployed that hold the promise of delivering functional data for "all possible" missense and other variants.^12–14 These approaches not only provide a catalog of variant function, but offer insights into underlying structure and function which can then serve as starting points for the development of machine learning-based approaches to variant classification^15,16 and to understanding variable penetrance.¹⁷

Are common variants necessarily benign? An assumption in the ACMG classification scheme is that common variants cannot confer clinically important phenotypes. A well-recognized exception is pharmacogenetic (PGx) variants that are common and that can profoundly impact drug responses, often triggering severe adverse drug reactions. These are thought to have persisted in populations because drug exposure is a recent event in human history and so PGx variants have not been under selective evolutionary pressure.¹⁸ We have recently shown that common variants in SCN5A, encoding the cardiac sodium channel, may confer no baseline clinical phenotype but nevertheless increase risk for drug-induced arrhythmias.¹⁹ This is an example of how even common variants can have profound effects on human traits in the presence of drug or other environmental triggers.

Preemptive population screening: The Centers for Disease Control has proposed that sequencing genes associated with breast and colon cancer and with familial hypercholesterolemia can be an appropriate screening test in asymptomatic people, while the current ACMG list for return of "secondary findings'' is longer, 73 genes. Cost effectiveness analyses suggest that in certain age groups, screening these genes regardless of family history can be cost effective.²⁰ It seems likely that screening for unrecognized and serious Mendelian diseases will become part of routine clinical care. This approach will likely also include polygenic risk scores that combine hundreds (or millions) of common variants, each carrying only small risk, to identify individuals at unusually high risk for common diseases.^21,22

The challenge in Genomic Medicine is no longer acquisition of sequence data but developing a much clearer understanding of what is and is not useful to deliver to patients. The present scientific statement emphasizes the need for a nuanced, thoughtful, and team-based approach to evaluation of the increasing problem of VUS interpretation. Navigating to a Genomic Medicine-enabled future requires collaborations among large numbers of people of all ancestries and teams of providers, genome scientists, counselors, pharmacists and others to develop and validate that evidence; this is the vision driving the million person All of Us program.²³