DISCUSSION
Variant interpretation is currently one of the most significant challenges in genomic medicine (Wright et al., 2019). Frequently, the causal relationship between the genotype, the identified variant, and the disease phenotype is not evident. This phenomenon results in ambiguous, erroneous, or incomplete interpretation of the genetic tests. An uncertain test result will not provide useful information to clarify the diagnosis, assist in treatment or management, or help in disease prevention (Richards et al., 2015). With this premise in mind, we revisited the criteria for interpretation of the clinical impact ofEFHC1 variants in a cohort of patients with GGEs. This information is likely to provide new information in the assessment of the clinical utility of EFHC1 testing.
Mutation screening in our cohort revealed that EFHC1 variants were almost exclusively in JME patients (10/11); only one of them was also identified in an individual with GTCS on awakening. However, we acknowledge the limited number of patients with other GGE phenotypes included in the present study. Few studies have investigatedEFHC1 variants in GGEs other than JME (Stogmann et al., 2006; Subaran et al., 2015). These studies also included a limited cohort of non-JME phenotypes (37 and 23 individuals, respectively), with variants considered potentially pathogenic reported in only two patients with JAE and one with unclassified GGE (Stogmann et al., 2006). Thus, the relatively low frequency of potentially pathogenic variants inEFHC1 in subjects with common GGEs suggests that it might not be the leading cause of these epilepsies.
All variants identified in our study are missense; they were found in patients with a family history of epilepsy (n = 20) as well as sporadic patients (n=25), a phenomenon that has also been observed in previous studies (Annesi et al., 2007; Bai et al., 2009; Jara-Prado et al., 2012; Ma et al., 2006; Medina et al., 2008; Raju et al., 2017; Subaran et al., 2015; Stogmann et al., 2006; Suzuki et al., 2004; Thounaojam et al., 2017; von Podewils et al., 2015). Furthermore, we found that overall, patients carrying variants in EFHC1 did not present distinct clinical characteristics compared to those without it. Previous studies also did not identify an apparent correlation between molecular findings and clinical features (Annesi et al., 2007; Thounaojam et al., 2017), although von Podewils et al. (2015) reported an association of the variants c.545G>A, c.685T>C and c.881G>A with early-onset GTCSs, a higher risk of status epilepticus, and a decreased risk of bilateral myoclonic seizures in series. Hence, we cannot exclude the possibility that additional studies in larger cohorts might highlight subtle phenotypic differences in JME patients with and without specific genetic variants.
We found that the amino acid residue changes that result from the missense variants are distributed throughout the EFHC1/myoclonin protein (Figure 3): in the three DM10 domains, in the region between the first two domains, in the EF-hand motif, and in the C-terminal region. Therefore, as described previously, we found no preferred region for the occurrence of EFHC1 variants in patients with JME and other GGEs (Bailey et al., 2017). Notably, our study is one of the few to report variants in the EF region: p.E589K and p.N607S, both of which are novel.
To further advance the classification of GGE-related EFHC1variants, we reviewed the ACMG/AMP guidelines, structured them into a comprehensive classification scheme, and applied them to our framework. We performed a literature review of variants that had already been established as either pathogenic or benign. Nine of the 11 variants found in the present study were previously reported in the literature. The variants c.662G>A and c.685T>C were reported as ‘pathogenic’ and found in patients with JME (Annesi et al., 2007; Bai et al., 2009; Bailey et al., 2017; Ma et al., 2006; Suzuki et al., 2004; Raju et al., 2017; von Podewils et al., 2015), although c.685T>C was not considered pathogenic in another report (Stogmann et al., 2006). Functional assays demonstrated that both variants reduce the effects of cell death, prevent the apoptosis of neurons with precarious calcium homeostasis during SNC development (Suzuki et al., 2004), induce defects in the mitotic spindle, and affect the radial morphology of glial and migratory neurons (de Nijs et al., 2009). In addition, both variants were found to co-segregate with affected individuals in families with JME (Suzuki et al., 2004).
The variants c.887G>A and c.896A>G were previously reported by Bailey et al. (2017) in patients from Brazil and classified as VUS, but no further information was provided. The variants c.475C>T, c.475C>G, c.545G>A, c.1343T>C, c.1855A>C are widely recognized as polymorphisms, and thus they are not considered pathogenic (Annesi et al., 2007; Ma et al., 2006; Stogmann et al., 2006; Subaran et al., 2015; Suzuki et al., 2004; Thounaojam et al., 2017). Functional studies showed that they do not affect cell death (de Nijs et al., 2009; Suzuki et al., 2004).
Moreover, we obtained the allele frequency of the EFHC1 variants in different population databases. We observed that the variation in the allelic frequency is dependent on the ethnic composition of the investigated population. Only the variants c.896A>G and c.1765G>A do not have allele frequencies higher than 1% in any of the investigated subpopulations (Table S1), and these two variants were absent from our group of 100 controls as well both databases of Brazilian individuals (BIPMed and AbraOM; Table 3). The allele frequencies of c.685T>C, c.887G>A, and c.1820A>G are only over 1% in specific subpopulations (c.685T>C: South Asian; c.887G>A and c.1820A>G: African; Table S1). The remaining six variants (c.475C>T, c.475C>G, c.545G>A, c.662G>A, c.1343T>C, and c.1855A>C) present allele frequencies higher than 1% in different populations.
To verify whether the frequency of EFHC1 variants in affected individuals is increased over controls, we screened 100 Brazilian individuals without a family history of epilepsy and performed an unconditional exact test. In addition, we performed the same test using two independent databases with genomic information on Brazilian individuals (BIPMed e ABraOM) to overcome the possibly insufficient sample size of our control group. Bailey et al. (2017) employed this approach with race-matched population groups from the ExAC database, but, unfortunately, the populations in the ExAC database do not match the Brazilian population.
We found a statistically significant association with JME for the variant c.685T>C when comparing the allele frequencies of our cohort to those in the ABraOM database (OR: 6.17; 95% confidence interval: 1.24–30.77; P value: 0.0245). Three variants (c.896A>G, c.1765G>A, and c.1820A>G) were not found in our control group or the Brazilian population databases; thus, the OR value was infinite (∞). For the remaining variants (c.475C>T, c.475C>G, c.545G>A, c.662G>A, c.887G>A, c.1343T>C, and c.1855A>C), we found no association between the variant and the risk for GGE.
To estimate the effect of potentially deleterious changes on protein function, we used the in silico prediction algorithms recommended by the ACMG/AMP (Richards et al., 2015). There was no consensus in these analyses—we did not observe concordance among the results of the 13 utilized algorithms for the investigated variants. Therefore, we were unable to use this evidence; the ACMG/AMP guidelines state that all of the in silico programs must agree on the prediction. This finding is in sharp contrast with what we have recently observed in the analysis of another epilepsy-related gene, SCN1A, in the context of Dravet syndrome (Gonsales et al., 2019). Indeed, as opposed to SCN1A ,EFHC1 is a gene that is tolerant to variation. This potential can be inferred by its probability of being loss-of-function intolerant (pLI), calculated as 0.000, and its Z scores (deviation of observed counts from the expected number) of 0.46 and 0.14 for synonymous and missense variants, respectively (http://gnomad.broadinstitute.org). ForSCN1A , a gene considered to be intolerant to variation, these values are much higher: pLI = 1.000, synonymous Z score = 0.88, and missense Z score = 5.22 (http://gnomad.broadinstitute.org).
The lack of consensus in the prediction analysis can be attributed to differences in the parameters used by the different algorithms (Sun & Yu, 2019; Walters-Sen et al., 2015). Also, we hypothesize that because the phenotype in most GGE patients is not as severe as in Dravet syndrome, it is possible that the putative genetic changes do not cause drastic disruptions in protein function. This factor would lead to inaccurate predictions. A recent study that evaluated the limitations in computational methods revealed that prediction models are usually excessively dependent on the conservation feature of the variants, a factor that results in predictive errors (Sun & Yu, 2019). The overlooked disease susceptibility of genes might also explain the failures of the computational tools. Interestingly, the limitations in the predictive power of the currently available in silicoalgorithms to analyze EFHC1 variants were seen even for amino acid substitutions that were previously shown to cause functional abnormalities in biological assays (Suzuki et al., 2004). Indeed, as mentioned above, the variants c.662G>A and c.685T>C, which was found in 1 and 3 of our patients with JME, respectively, were previously studied by Suzuki et al. (2004) and found to affect the apoptotic activity of neurons with precarious calcium homeostasis.
Applying the proposed modified classification guidelines to theEFHC1 variants from our cohort, only the variants c.662G>A and c.685T>C were classified as ‘pathogenic.’ However, c.662G>A also met the criteria to be classified as ‘benign.’ In cases when the criteria for benign and pathogenic variants are contradictory, the ACMG/AMP rules for combining criteria states that the variant should be classified as VUS (Richards et al., 2015). Thus, only variant c.685T>C can strictly be classified as ‘pathogenic’ (1/11, 9%).
It is noteworthy that the variant c.662G>A only met one benign criterion—the stand-alone BA1—due to an allele frequency of 5.3% in one African subpopulation. Indeed, all six variants classified as ‘benign’ (6/11, 55%) met the stand-alone criteria BA1 because they present an allele frequency greater than 5% in at least one database subpopulation. Therefore, the evidence-based population frequency promoted a substantial increase in the classification scores, which in some cases could have been overly weighted, especially when analyzing genes of minor effect. This phenomenon would induce an increased susceptibility rather than a major effect.
Ethnicity might have an important influence in defining which genetic factors are implicated in diseases with complex inheritance, including GGEs (Subaran et al., 2015). Thus, different genetic backgrounds would present distinct epilepsy susceptibility genes. In this scenario, one recently reported possibility is that EFHC1 variants might be pathogenic when they are found in specific genetic backgrounds (Subaran et al., 2015). Interestingly, we found variants in our patients originating from the Northeast part of Brazil that are common only in specific populations: c.662G>A, c.887G>A, and c.1820A>G, each with a higher allele frequency in populations of African ancestry. A study that investigated the global ancestry of Brazilians showed that the genetic composition of the populations from the specific region where the probands harbor these variants are originated (the Northeast region) is closer to the Europeans (Saloum de Neves Manta et al., 2013). It is important to highlight that the underrepresentation of non-European ancestry groups in population databases poses an additional challenge to the interpretation of genetic variants (Petrovski & Goldstein, 2016). A better population match would improve the application of population frequency criteria for underrepresented ethnicities.
In contrast to Mendelian disorders, common or complex inheritance diseases might not have one single major causative gene. Nevertheless, multiple genetic variants contribute to a small effect on disease risk. Hence, reduced penetrance and small effect size are possible explanations for why healthy individuals might harbor pathogenic variants (Cooper et al., 2013; Wright et al., 2019). Penetrance refers to the proportion of individuals in a population with a disease-related genotype that manifests the disease phenotype, while a small effect size implies that the variant has a low impact on the multifactorial etiology of the disease (Cooper et al., 2013). Therefore, some variants are insufficient to cause disease on their own and require interaction with other genetic and/or environmental factors to surpass an estimated threshold into a pathogenic phenotype (Cooper et al., 2013). In this context, EFHC1 might be considered a partially penetrant gene. Even among disorders for which this concept has been well-established, such as cancer syndromes, the implications of low penetrance or small effect size when considering a gene suitable for testing in the clinical setting has been discussed (Ellsworth, Turner, & Ellsworth, 2019; Wendt & Margolin, 2019). For instance, variants found in the highly penetrant susceptibility genes BRCA1 and BRCA2 for breast cancer are clinically actionable (Ellsworth et al., 2019). However, definitive clinical recommendations cannot be drawn for lower risk genes (Wendt & Margolin, 2019), and effective management and therapeutics strategies are still required for patients who harbor variants in other genes (Ellsworth et al., 2019). Although it is known that multiple gene variants are necessary to produce the GGE phenotype, there are still no accepted models on a presumed polygenic inheritance regarding the type of variants and the involved genes (Mullen, Berkovic, & Commission, 2018). Moreover, the clinical management of the GGEs is not altered by the result of the genetic test. This fact suggests that it should be considered only in the research context (Mullen et al., 2018).
Another crucial debate regarding genetic testing is the challenge that clinicians face when reporting the results to the patients or the parents, especially when a VUS is found. These variants have undefined clinical significance, and it is a consensus that they should not be used in clinical decision-making (Richards et al., 2015). Patients report anxiety symptoms, worries, and uncertainty in response to a VUS result (Makhnoon, Shirts, & Bowen, 2019). Moreover, in the case of genetic testing performed in children, there is the possibility that the parents will misinterpret the test results, a phenomenon that would lead to unnecessary anxiety due to excessive medical attention (Wynn et al., 2018). Thus, given that almost half of the variants found in our cohort are VUS (5/11, 45%), this potential represents a significant concern when considering EFHC1 for clinical genetic testing.