ExACtly zero or once

DISCUSSION

The interpretation of putative epilepsy genetic variants remains complex for researchers and clinicians alike. Here, we assessed the allele count of literature-reported epilepsy variants in the ExAC database and found that variants with strong segregation evidence for pathogenicity were present no more than once in the ExAC collection of over 60,000 individuals, illustrating that these variants are ultra-rare in the general population. This pattern was also seen in the variants found in other neurologic disorders that were reported based on strong segregation support. However, the literature-reported epilepsy variants without strong segregation support were more likely to be present in more than 1 carrier among the ExAC collection (figure), suggesting that many of these variants (table 2) are unlikely to be penetrant disease-causing alleles, but rather background variation in established epilepsy genes or possibly risk alleles contributing to a more complex genetic architecture that is not currently well understood.

WES in patients with epilepsy can be a powerful diagnostic tool, but interpretation of the data remains complex. In the field of epilepsy genetics, this complexity is highlighted by the likely false attribution of pathogenicity to 2 genes, EFHC1 and SRPX2. SRPX2 was deemed a pathogenic gene in 2006 in 2 families with epilepsy, speech dyspraxia, and intellectual impairment, including one from our centre.⁵ The key family was subsequently shown to have a GRIN2A variant, and in fact the SRPX2 variant from this family appears in ExAC 47 times. Similarly, the pathogenicity of EFHC1 as a gene for juvenile myoclonic epilepsy has been disputed.⁶

Even for widely accepted epilepsy genes, the interpretation of VOUS is complex. A large contribution to false-positive epilepsy variants comes from variants identified through “secondary” studies, wherein pathogenicity is largely inferred based on the knowledge of prior disease-causing variants within the gene. These variants are often not subject to the same rigorous evaluation of pathogenicity as variants from the original discovery,⁶ resulting in more permissive pathogenicity assignments. Indeed, we know that epilepsy genes carry background variation in the general population. For example, the variants in table 2, which are in established epilepsy genes, were published as pathogenic and some of these are present in ExAC at relatively high frequencies, emphasizing the caution necessary when assessing pathogenicity solely on the basis of a known epilepsy gene. This is one of the most important considerations in the interpretation of results from clinical gene panels.

The issue of variant misclassification extends well beyond the field of epilepsy to other monogenic diseases. It has recently been demonstrated that each ExAC participant carries, on average, ∼54 variants that have previously been reported as disease causing.³ Of these, ∼41 variants have a minor allele frequency of >1%.³ It is therefore likely that many of these variants are either benign variants or contributing to complex inheritance.

The frequency of VOUS in control samples is thus an invaluable line of evidence in interpretation, and the ExAC database offers a control cohort of unprecedented size. ExAC is not enriched for pathogenic variants in genes that are commonly tested in a clinical diagnostics setting, thus supporting its utility in classifying variants.⁷ In a recent study, the ExAC database was found to have an excess of ostensibly pathogenic variants for prion disease but found that the variants with the strongest independent evidence of pathogenicity were absent from ExAC, which is consistent with our findings.⁸ The authors attributed the excess allele frequency for variants with weak evidence for pathogenicity to the incorrect attribution of pathogenicity for certain variants or reduced penetrance of these alleles, an interpretation shared here.

Thus, the interrogation of ExAC provides a highly useful way of assessing the relevance of VOUS for patients with epilepsy and is likely to be widely applicable to other neurologic disorders. We have shown that disease-causing variants in epilepsy are very rare, and thus variants that are present more than once in the current ExAC cohort are unlikely to contribute to disease in a large way. It is important that the inverse is definitely not true—we cannot infer that a variant's absence in ExAC is sufficient evidence of pathogenicity.

The use of large reference cohorts as a clinical tool is not without limitations. As the minor allele frequency resolution improves with additional samples added to large reference cohorts (like ExAC), it is possible that clinically relevant variants for the milder epilepsies might begin to appear more than once. Indeed, our results with ExAC were supported with ExAC V2 plus gnomAD where the sample size was doubled. The only exception was 2 variants in SCN1B associated with febrile seizures and GEFS+ which are present 3 and 4 times (table e-1). A possible explanation for this slightly elevated frequency is that variants causing febrile seizures (a mild phenotype often resolving with age) are expected to have a subtler impact on reproductive fitness than variants causing other common genetic generalized and nonacquired focal epilepsies, and this is also coupled with the higher rates of febrile seizures in the general population than other common epilepsies. Thus, it is unlikely that variants with a minor allele frequency >0.0005% (approximate current ExAC V2 resolution) have a major contribution to autosomal dominant forms of mild epilepsies.

Our study and the message presented here are independently supported by our recent Epi4K study, where we found that when comparing to a control population, the epilepsy risk signals identified in epilepsy genes among large collections of epilepsy samples were driven by ultra-rare variants absent in ExAC.⁹ Our current study takes an alternative approach, comparing the frequency in ExAC of variants with and without strong segregation support, to reach the same conclusion. It is important to note that although European and some other genetic ancestries are well represented in ExAC, many ethnic minorities are underrepresented or not represented at all, posing additional issues in interpretation when assessing data from a patient from an ethnic background not well represented in these databases.¹⁰