Diagnostic implications of genetic copy number variation in epilepsy plus

Summary Objective Copy number variations (CNVs) represent a significant genetic risk for several neurodevelopmental disorders including epilepsy. As knowledge increases, reanalysis of existing data is essential. Reliable estimates of the contribution of CNVs to epilepsies from sizeable populations are not available. Methods We assembled a cohort of 1255 patients with preexisting array comparative genomic hybridization or single nucleotide polymorphism array based CNV data. All patients had “epilepsy plus,” defined as epilepsy with comorbid features, including intellectual disability, psychiatric symptoms, and other neurological and nonneurological features. CNV classification was conducted using a systematic filtering workflow adapted to epilepsy. Results Of 1097 patients remaining after genetic data quality control, 120 individuals (10.9%) carried at least one autosomal CNV classified as pathogenic; 19 individuals (1.7%) carried at least one autosomal CNV classified as possibly pathogenic. Eleven patients (1%) carried more than one (possibly) pathogenic CNV. We identified CNVs covering recently reported (HNRNPU) or emerging (RORB) epilepsy genes, and further delineated the phenotype associated with mutations of these genes. Additional novel epilepsy candidate genes emerge from our study. Comparing phenotypic features of pathogenic CNV carriers to those of noncarriers of pathogenic CNVs, we show that patients with nonneurological comorbidities, especially dysmorphism, were more likely to carry pathogenic CNVs (odds ratio = 4.09, confidence interval = 2.51‐6.68; P = 2.34 × 10−9). Meta‐analysis including data from published control groups showed that the presence or absence of epilepsy did not affect the detected frequency of CNVs. Significance The use of a specifically adapted workflow enabled identification of pathogenic autosomal CNVs in 10.9% of patients with epilepsy plus, which rose to 12.7% when we also considered possibly pathogenic CNVs. Our data indicate that epilepsy with comorbid features should be considered an indication for patients to be selected for a diagnostic algorithm including CNV detection. Collaborative large‐scale CNV reanalysis leads to novel declaration of pathogenicity in unexplained cases and can promote discovery of promising candidate epilepsy genes.


Summary
Objective: Copy number variations (CNVs) represent a significant genetic risk for several neurodevelopmental disorders including epilepsy. As knowledge increases, reanalysis of existing data is essential. Reliable estimates of the contribution of CNVs to epilepsies from sizeable populations are not available. Methods: We assembled a cohort of 1255 patients with preexisting array comparative genomic hybridization or single nucleotide polymorphism array based CNV data. All patients had "epilepsy plus," defined as epilepsy with comorbid features, including intellectual disability, psychiatric symptoms, and other neurological and nonneurological features. CNV classification was conducted using a systematic filtering workflow adapted to epilepsy. Results: Of 1097 patients remaining after genetic data quality control, 120 individuals (10.9%) carried at least one autosomal CNV classified as pathogenic; 19 individuals (1.7%) carried at least one autosomal CNV classified as possibly pathogenic. Eleven patients (1%) carried more than one (possibly) pathogenic CNV. We identified CNVs covering recently reported (HNRNPU) or emerging (RORB) epilepsy genes, and further delineated the phenotype associated with mutations of these genes. Additional novel epilepsy candidate genes emerge from our study. Comparing phenotypic features of pathogenic CNV carriers to those of noncarriers of pathogenic CNVs, we show that patients with nonneurological comorbidities, especially dysmorphism, were more likely to carry pathogenic CNVs (odds ratio = 4.09, confidence interval = 2.51-6.68; P = 2.34 × 10 −9 ). Meta-analysis including data from published control groups showed that the presence or absence of epilepsy did not affect the detected frequency of CNVs. Significance: The use of a specifically adapted workflow enabled identification of pathogenic autosomal CNVs in 10.9% of patients with epilepsy plus, which rose to 12.7% when we also considered possibly pathogenic CNVs. Our data indicate that epilepsy with comorbid features should be considered an indication for patients to be selected for a diagnostic algorithm including CNV detection. Collaborative largescale CNV reanalysis leads to novel declaration of pathogenicity in unexplained cases and can promote discovery of promising candidate epilepsy genes.

| INTRODUCTION
Current estimates suggest that genetics contribute to causation in 50%-70% of the epilepsies. 1 Copy number variations (CNVs) represent a prominent type of variant carrying risk for certain epilepsies. [2][3][4][5] Whole genome oligonucleotide array CGH or SNP array is routinely included in evaluation of patients with complex phenotypes with a suspected genetic cause. 6 CNVs, as a risk factor or cause, have been reported in ~5%-12% of patients with different types of epilepsies. 2,4,5,[7][8][9] The risk of a pathogenic CNV is reportedly increased with concurrent intellectual disability (ID), dysmorphic features, autism spectrum disorder (ASD), drug resistance, or other comorbidities, from a study of 222 patients. 10 Recurrent CNV "hotspots" predispose to different types of epilepsies. 5,11 CNV detection has pointed to novel epilepsy genes. 12 Robust estimates of the frequencies and types of putatively relevant CNVs in epilepsy are needed to determine whether CNV detection should be included in genetic evaluation of patients with various epilepsy phenotypes. As knowledge of epilepsy genetics increases, systematic, iterative reevaluation of genetic data becomes essential. This process requires large numbers of individuals to be corralled, and because such data will inevitably come from different centers using different technologies, a robust means of joint reevaluation is essential.
Epilepsy is often a feature of neurodevelopmental disorders (NDDs). A recent study on individuals with NDDs and epilepsy reported similar results for rare variant frequency for individuals ascertained to have epileptic encephalopathy (EE) and for individuals ascertained for NDDs with unspecified epilepsy, 13 suggesting that, genetically, epilepsy can be considered part of the spectrum of NDDs. Looking at this concept from the perspective of CNV, and to determine the frequency of CNVs in particular epilepsy phenotypes, we assembled a large international cohort of patients with the phenotype of "epilepsy plus," which we define as the occurrence of epilepsy and comorbid features, including ID and psychiatric, neurological, and nonneurological features. Preexisting array data were systematically investigated using a workflow based on current knowledge of CNV classification. The workflow enabled combination of multicenter CNV data to provide a robust, up-to-date reevaluation of the contribution of CNVs to epilepsy plus and identified new candidate pathogenic autosomal CNVs. The method can be applied iteratively with additional cohorts at future time points, making optimal use of existing data.

| Ethics
This study was approved by the ethics committees of the participating centers. Written informed consent was provided by the patient, or the parent or the guardian of each patient as appropriate.

| Data collection
Preexisting CNV data, derived from array CGH or SNP array conducted for clinical or research purposes, were collected from eight specialist epilepsy and/or genetic centers (Table  S1). All patients also had comorbid features including ID, autism, dysmorphic features, other neurological or nonneurological conditions, structural brain abnormalities, or multidrug resistance. 14 Clinical information was collected through referring clinicians. Seizure and epilepsy/syndrome types were classified according to the International League Against Epilepsy criteria when available. 15

| CNV analysis: Quality control and classification
All CNV calls were provided by the contributing centers (Table S1). Figure 1 shows the workflow we used to classify CNVs (Data S1). We focused only on autosomal CNVs due to higher quality of CNV calls from nonsex chromosomes. 16

K E Y W O R D S
array CGH, copy number variants, epilepsy genes, SNP array

Key Points
• CNV is an important contributor to the causation of epilepsy plus, with pathogenic and possibly pathogenic CNVs present in nearly 13% of cases • The use of a specifically adapted workflow to classify CNVs allows the analysis of data from retrospectively collected patients screened through different platforms • This study highlights CNVs covering recently reported (HNRNPU) or emerging (RORB) epilepsy genes, and further delineates the associated phenotype • Patients with nonneurological comorbidities, especially dysmorphism, were more likely to carry pathogenic CNVs To ensure high reliability, we considered only CNVs with high calling confidence according to the following criteria: (1) size ≥ 150 kb, (2) coverage of ≥30 consecutive probes for SNP arrays and ≥3 probes for array CGH, and (3) microdeletion/microduplication frequency < 1% in the entire study sample. Samples with a total number of deletion or duplication (or both) calls >2 SD from the mean number of any calls/sample across the entire dataset were excluded from the analysis. Further manual analysis used a bespoke workflow based on current understanding of classification, 17,18 including the American College of Medical Genetics guidelines 19 and additional literature. [2][3][4][5][6]11 CNVs were classified into four groups: pathogenic, possibly pathogenic, benign, or of unknown significance. Briefly, the workflow was as follows: First, common CNVs, present in the healthy population, 20 were classified as "benign." All remaining CNVs were then classified as "pathogenic" if they met the following criteria: ≥80% overlap of the study CNV with any CNVs known to be associated with epilepsy; or, CNV with a size ≥ 3 Mb; or, CNV with a size < 3 Mb and > 1 Mb, and with de novo occurrence. The remaining CNVs were further classified according to their gene content. A CNV was classified as pathogenic when it involved a gene known to be associated with epilepsy (Table S2), the phenotype was concordant with that in the literature, and the type of CNV (deletion/duplication) matched current knowledge on the pathogenic mechanism of the gene change (gain or loss of function). If a CNV contained a gene associated with epilepsy but the other conditions were not fulfilled, the CNV was considered pathogenic only if proven to be de novo and was otherwise classified as "possibly pathogenic." CNVs containing a brain-expressed gene, according to the published datasets 21 and the database GTEx (http://www.gtexportal.org/home/), were classified as "possibly pathogenic" only if de novo. Analysis of recessive F I G U R E 1 Workflow used to classify the copy number variations (CNVs) in our cohort of patients with epilepsy plus. Stepwise procedures are shown for CNV classification into benign, pathogenic, possibly pathogenic, and unknown significance groups. CGH, comparative genomic hybridization; CNS, central nervous system; SNP, single nucleotide polymorphism inheritance of epilepsy genes was not considered due to limitations of most CNV platforms on calling homozygous deletions or duplications. The remaining CNVs were classified as "of unknown significance."

| Phenotype enrichment analysis
Using Fisher's exact test, we investigated whether patients carrying a pathogenic autosomal CNV, compared to those not carrying a pathogenic CNV, had overrepresentation for specific phenotype components (nonneurological disorders, neurological or psychiatric disorder, ID, facial dysmorphism, brain abnormalities, epilepsy onset < 1 year of age, and EE). The analysis was conducted in two ways-(1) for any pathogenic CNV and (2) for only large (>1 Mb) pathogenic CNVs-and was corrected for multiple testing accordingly.

| Meta-analysis
To determine the impact of epilepsy on the probability of identifying pathogenic CNVs, we used the following strategy. First, we split our cohort into two subgroups including patients with (1) epilepsy and ID including autistic features and (2) epilepsy and other psychiatric/neurological disorders. We gathered two "historical control groups" through a systematic review of the literature and a meta-analysis and estimated the yield of pathogenic CNVs in patients with (1) ID including autistic features (without epilepsy) and (2) psychiatric/neurological disorders (without epilepsy). Then, we compared the yield of pathogenic CNVs between these groups with (1) epilepsy and ID including autistic features versus the historical-control group with only ID and autistic features and (2) epilepsy with other psychiatric/neurological disorders versus the corresponding control group from the literature manifesting only other psychiatric/neurological disorders. We used the Cochran Q test to assess heterogeneity across studies.
To evaluate whether epileptic encephalopathies might specifically contribute to the yield of pathogenic CNVs, we compared patients with epilepsy manifesting as EE (epilepsy-EE) from our cohort to those with epilepsy without EE (epilepsy-notEE) from a systematic review of the literature.
The full search strategy, inclusion criteria, and methods are available in Data S1.

| Pathogenic CNVs
To simplify presentation, we further divided pathogenic CNVs into four subgroups: (1) recurrent CNVs with welldocumented enrichment in epilepsy; (2) CNVs related to a genetic Online Mendelian Inheritance in Man database (OMIM) syndrome with neurological symptoms in which epilepsy can feature; (3) CNVs not known to be enriched in epilepsy and not associated with any other OMIM syndrome, but containing at least one gene that is already implicated in epilepsy; and (4) CNVs based on size combined with de novo occurrence.

| CNVs related to a genetic OMIM syndrome with neurological symptoms in which epilepsy can feature
Thirty-three individuals had pathogenic CNVs (33/120, 27.5%) mapping to regions for well-characterized genetic syndromes associated with neurological features including epilepsy (Table S4a) and consistent with the relevant syndrome. The most frequent were as follows: the Williams-Beuren 7q11.23 deletion syndrome (five patients), 15q11.2 duplication syndrome, distal (three patients), 16p11.2 duplication syndrome including PRRT2 (four patients), the Potocki-Lupski 17p11.2 duplication syndrome (two patients), and 17p13.3 deletion syndrome, also known as Miller-Dieker lissencephaly deletion syndrome (three patients). We also identified de novo duplications at 2q24.3 22 and at 4p16.3-p13, 23 for which regions both deletions and reciprocal duplications have been associated with epilepsy. 22,23

| CNVs including epilepsyrelated genes
Nineteen individuals had a CNV (19/120, 15.8%) including epilepsy-related genes ( Table 2). Five individuals had a CNV including HNRNPU (four de novo deletions and one duplication; two deletions and the duplication also contained the flanking AKT3 gene). The four probands carrying deletions presented with epilepsy classified as Lennox-Gastaut syndrome in one patient, genetic generalized epilepsy (GGE) in another, and early onset, drug-resistant epilepsy not otherwise classified in the remaining two. Moderate to severe ID was reported in four patients and one also had ASD. Three had microcephaly, congenital and severe (−4 SD) in one. Brain magnetic resonance imaging showed corpus callosum agenesis or hypoplasia in three of four patients. Facial dysmorphic features were observed in three patients. The patient carrying a large (>100 Mb) duplication involving, among many other genes, HNRNPU and AKT3, had a complex phenotype including neonatal seizure onset, polymicrogyria, and multiple cardiac defects. Three individuals had a 9q21.13 deletion, one de novo and two of unknown inheritance, including a gene with recently described association with epilepsy, RORB. All the patients presented with ID and generalized epilepsy with absences or atypical absences, with eyelid myoclonia in two cases and photosensitivity in one. Further clinical details of patients with CNVs including HNRNPU or RORB are provided in Tables S5a and S5b. Three deletions encompassed the ADGRV1 gene, two of which included MEF2C. Additional epilepsy genes that were found deleted or duplicated in single patients are listed in Table 2 and include GNAO1, NEDD4L, and SIK1.

| Possibly pathogenic CNVs
Nineteen individuals (19/1097, 1.73%) had a total of 20 CNVs classified as possibly pathogenic (one individual had two possibly pathogenic CNVs); 10 were de novo (Table 3). For 17 of 19 individuals (18/20 CNVs), DNA was available to check the CNV and/or inheritance using MAQ analysis. Eleven of the 18 analyzed CNVs were confirmed; in seven cases, the test was inconclusive (Table 3). These CNVs were classified as possibly pathogenic because they included an epilepsy gene but were inherited or the direction of the change was not concordant with the known disease mechanism (loss or gain of function) or phenotype, or because they included a brain-expressed gene and were de novo. CNVs falling in the first category were a maternally inherited 10q23 deletion including LGI1, and a maternally inherited 20q13 duplication including KCNQ2, CHRNA4, and EEF1A2. Four other inherited CNVs included recessive genes: PLCB1, TBC1D24, ABAT, and CNTNAP2. A possible additional single nucleotide variant (SNV) on the other allele cannot be excluded. Of note, the PLCB1 deletion was confirmed to be homozygous and would be considered pathogenic, but our flowchart was not developed for recessive analysis.
In the second category, we identified several interesting candidate genes including a de novo deletion including STAG1 and a de novo intragenic duplication in FGF12. In both genes, only recently have pathogenic SNVs been reported in patients with neurodevelopmental disorders including epilepsy. 24,25 We further identified a deletion including SETBP1 associated with Shinzel-Gieidon syndrome and ID (OMIM 611060) and a duplication including HCN2, a gene in which SNVs exerting a gain-of-function effect have recently been suggested as a risk factor for genetic generalized epilepsies. 26 HCN2 has also previously been associated with febrile epilepsy syndromes; interestingly, the patient carrying this CNV also had a history of febrile seizures. 27 Other interesting candidate genes located in identified de novo deletions or possibly disrupted by intragenic breakpoints of identified duplications included FMN2 (also associated with AR mental retardation MIM616193), CHRM3, CSNK1G3, and NMT1, all of which are brain-expressed and predicted to be intolerant to loss of function (probability of loss-of-function intolerance ≥ 0.99) according to the latest gnomAD (http:// gnomad.broadinstitute.org/about) constraint metrics (https:// www.nature.com/articles/nature19057).

| Meta-analysis
The search identified 4806 citations, of which 59 papers met the inclusion criteria and were included in the systematic review. Overall meta-analysis showed that in patients with ID without epilepsy, the yield of pathogenic CNVs was 15% (95% confidence interval [CI] = [14][15][16][17], and in patients with psychiatric/neurological disorders without epilepsy, the yield was 8% (95% CI = 5-12; Figure S3, Table 4). These data were compared with the two subgroups from our cohort: (1) patients with epilepsy and intellectual disabilities, including autistic features, with a yield of 13.5% (95% CI = 9.2-18.9); and (2) patients with epilepsy and psychiatric/neurological comorbidities, with a yield of 10% (95% CI = 7.9-11.7). We did not find statistically significant differences for any of these comparisons (P values from heterogeneity test were >0.05).

| DISCUSSION
Most epilepsies, especially when beginning in infancy and childhood, have a prominent genetic contribution. Numerous next generation sequencing, whole exome sequencing, and whole genome sequencing studies have been published in recent years uncovering single gene mutations in many epilepsies and epilepsy syndromes. Yet, the contribution of CNVs to the epilepsies, especially those complicated by comorbidities, has been less explored. Most published reports are single-center studies. The largest sample size was 2454 patients including a large cohort of 1366 patients with genetic generalized epilepsy in addition to 281 patients with rolandic epilepsy and 807 patients with adult focal epilepsy 28,29 ; the biggest cohort specifically addressing the epilepsy plus phenotype studied 222 individuals. 4 The maximum frequency of pathogenic CNVs reported in any of these series was 12%, with a range of 5%-12%. These studies tended to focus on individuals who were children at the time of testing. 2,5,9 The importance of rare CNVs has been well recognized in patients with neuropsychiatric disorders including unexplained ID, congenital anomalies, and seizures. Thus, clinical geneticists, pediatric neurologists, and epileptologists commonly request chromosomal array CGH to obtain a genetic diagnosis for patients with such clinical features.
However, CNVs may be seen in healthy control individuals, and determination of the pathogenicity of newly identified CNVs can be challenging. To evaluate the role of pathogenic CNVs and identify possible candidate genes, we investigated the occurrence of CNVs in epilepsy plus, in a cohort among the largest reported to date. [2][3][4][5]9 Data were collected from eight centers and included both adults and children. Autosomal CNV classification was conducted using a systematic filtering procedure specifically adapted to epilepsy. The workflow was an essential tool to identify, reanalyze, and reinterpret CNVs in this retrospectively collected cohort, in which CNV testing had been performed using different platforms in different laboratories. About 11% of patients with epilepsy plus harbored a pathogenic autosomal CNV. This number reaches 12.7% when we also consider the possibly pathogenic CNVs. Previous similar studies report a diagnostic yield ranging from ~5% to 12%. [2][3][4][5]8,9 Thus, our result fits at the upper limit of this range, probably mainly due to the "epilepsy plus" phenotype of our cohort and to the application of a standardized workflow. Previously published studies 3,4,8 that reported similar yields of pathogenic CNVs (9.3%, 8.1%, and 12%, respectively) also examined patients with complex epilepsy including ID. Overall, results from both our and similar previous studies indicate that within the complex phenotype of neurodevelopmental disorders, when seizures are associated with ID or with other neurological and nonneurological comorbidities, there is a higher probability of identifying a pathogenic CNV than in epilepsy alone. We checked the original classification, where available (138/142), of pathogenic and possibly pathogenic CNVs before and after applying the workflow method we propose here. We found that 7.2% (10/138) of cases were discrepant. The main direction of change was from CNVs (8/10 CNVs) originally classified as of "unknown significance" to "pathogenic" and "possibly pathogenic" (Table S7). This is expected as information about brain-expressed genes or gene regions associated with epilepsy increases. We have confirmed that reanalysis of existing data over time is essential.
Our study confirms the importance of specific CNVs in epilepsy and broadens some of the associated phenotypic spectra.
We also found several CNVs that included the genes HNRNPU (1q44) and RORB (9p21.13), both recently associated with epilepsy. 28,[30][31][32] Microdeletions of the 1q43q44 critical region have been associated with ID, dysmorphism,  (Table S6a). In our cohort, five patients carried CNVs mapping to the 1q43q44 critical region, and in addition to the HNRNPU gene, in two duplications and one deletion, the chromosomal rearrangement included also the AKT3 gene, which might contribute to brain abnormalities observed in these patients. Patients with deletions showed dysmorphic features, early onset psychomotor delay, and early onset epilepsy. These data confirm the role of HNRNPU in neurodevelopment and epileptogenesis. Mutations in RORB were first reported in a patient with mild ID and partial epilepsy. 31 More recently, other mutations were identified in patients with neurodevelopmental disorders and mostly GGE, including absence seizures (Table  S6b). In our cohort, three patients carried deletions including RORB and exhibited ID and generalized epilepsy, including absence seizures with eyelid myoclonia, and autistic features in one patient, supporting a role for RORB in GGE and, more broadly, in several neurodevelopmental disorders.
Among the syndromic pathogenic autosomal CNVs, we identified three patients with duplications mapping to the 17p11.2 Potocki-Lupski syndromic region, which is reciprocal to the Smith-Magenis deletion syndrome in which epilepsy is often seen. 33 These three patients had a phenotype consistent with Potocki-Lupski syndrome; the occurrence of epilepsy supports previous evidence of its presence as a rare feature of 17p11.2 duplications. 5 Interestingly, we identified five patients with a 7q11.23 deletion containing the Williams-Beuren region; four of these individuals had Lennox-Gastaut syndrome, and the fifth (previously reported by Ramocki et al 34 ) had a generalized drug-resistant epilepsy.
CNVs classified as pathogenic only because of large size (Table S4b) represented 27% (33/122) of all the pathogenic CNVs. These CNVs included a large number of genes, but the phenotype of affected individuals was complex and we were unable to identify an association with known genetic syndromes or with candidate epilepsy genes. However, for one individual with EE and a large de novo 13q13.1-q13.3 deletion, we can suggest that a key gene is NBEA, which was reported as a possible EE gene through an in silico prioritization approach 35 and was recently associated with neurodevelopmental disease with epilepsy. 36 Four of the CNVs we classified as large and pathogenic were inherited. Interestingly, a duplication on 12q21.31 was inherited from a mother with a family history of autism. Autism has been reported in Decipher in a patient carrying an overlapping duplication (Table S4b). Following our algorithm, we consider these CNVs pathogenic, noting the incomplete penetrance often characterizing neurological and epileptic disorders and because we could not exclude related neurological traits in the transmitting parent.
A possibly pathogenic autosomal CNV was identified in 1.7% of the patients. As well as some known epilepsy genes, discussed in the results section, we propose other genes in these regions that can be considered potential candidates for causing epilepsy, but need further validation. We found a de novo 18q12.3 deletion, which only encompassed the gene SETBP1. Heterozygous missense mutations in SETBP1 cause Schinzel-Giedion syndrome (OMIM #269150), characterized by severe ID and specific craniofacial features, 37 wherein seizures also occur. 38,39 Mutations leading to haploinsufficiency, such as the deletion in our patient, have been reported in association with a distinct neurological syndrome, which includes mild to moderate ID without the typical syndromic craniofacial features. 17,40,41 The patient in this study only showed severe epilepsy and ID, suggesting that the SETBP1-mutation phenotype may be broader than previously described. One patient had a microdeletion, classified here as possibly pathogenic, which includes STAG1, now linked with epilepsy as a cohesinopathy, 24 and one patient carried a de novo intragenic duplication in FGF12 in which SNVs have recently been reported in patients with epileptic encephalopathies. 25 Other interesting candidate genes are highlighted in Table  3 and include HCN2, FMN2, CHRM3, CSNK1G3, and NMT1.
Eleven patients (1%) in our study cohort had a double hit (including pathogenic and possibly pathogenic CNVs). Here, the CNV burden alone could contribute to the neurodevelopmental phenotype; as shown by Girirajan and colleagues, 42 children with two or more rare and large CNVs of unknown significance were eight times more likely to have developmental delay compared to controls, possibly by disruption of dosage-sensitive genes. 42 We note, however, that our analysis focused only on CNVs with a certain pathogenic meaning and as such gives no insight into the general burden of CNVs per patient. The enrichment analysis showed a significant association of pathogenic autosomal CNVs with nonneurological disorders and dysmorphism (for both large pathogenic and any pathogenic CNV); large pathogenic CNVs showed a more profound and significant association with dysmorphism and non-neurological disorders only. An enrichment of CNVs in patients with dysmorphism has been observed in previous studies, 4 underscoring the importance of testing for CNVs in patients with epilepsy and associated comorbidities. Likewise, results from our data compared to historical controls, from a systematic literature review and meta-analysis, confirm that the percentages of pathogenic CNVs, when the phenotype includes or excludes epilepsy, do not vary significantly. Thus, although a search for CNV is undoubtedly worthwhile in people with epilepsy plus, it may not be that such CNVs drive only epilepsy, but for patients ascertained through their epilepsy, the presence of additional features points to an elevated likelihood of finding an underlying pathogenic CNV. We hypothesize that although epilepsy as a phenotype does not add a quantitative contribution to the diagnostic yield, its presence could be related to the type, location, and gene content of an underlying pathogenic CNV. Results from our data, comparing patients with epilepsy-EE versus historical controls with epilepsy-notEE, showed a nonsignificantly lower yield of pathogenic CNVs in patients with EE, raising a possible hypothesis that when epilepsy manifests as EE, the likelihood of finding a pathogenic CNV decreases and that EE is more often the consequence of single gene mutations.
Our study has limitations beyond its retrospective structure. The filtering workflow used allowed us to obtain a systematic classification of the large number of CNVs examined, but we recognize it is not perfect and might not accurately classify CNV mapping to hypervariable chromosomal regions. Pathogenic CNVs could be missed due to filtering out of small CNVs, misclassification of abnormalities, or an incomplete list of genes associated with epilepsy (new epilepsy-related genes continue to be reported). We excluded the sex chromosomes from the CNV calling and subsequent analysis, because copy number calling from these chromosomes is prone to falsepositive calls and might inflate the reported frequencies of diagnostically relevant CNVs as the X chromosome in particular has been associated with neurodevelopmental disorders. Also, recessive disease cannot be ruled out with this type of analysis unless the second allele is studied with another approach.
In conclusion, we highlight the pathogenic causative role of autosomal CNVs in almost 11% of patients (and up to 12.7% when also considering the possibly pathogenic) with unexplained epilepsy with comorbidities and reiterate the concept that CNVs should be sought in patients with seizures especially when associated with other neurological and nonneurological conditions. This study opens new perspectives for a better understanding and evaluation of CNVs identified in patients with epilepsy plus. We show that the reinterpretation of preexisting data using an adapted workflow can highlight new findings, and we recommend periodic systematic review of preacquired genetic data, as new methods and data become available. The workflow used here, specifically designed for epilepsy, can be used to homogenize data from different cohorts often collected at different times. Establishing the causative role of some CNVs can be challenging, especially when the CNV is not associated with a known syndrome, or similar CNVs may not be of the same size, might include different genes, and not have familial segregation data available to help interpretation. Bespoke, disease-specific algorithms may assist in assignment of CNVs to diagnostic categories that are more definitive than either "possibly pathogenic" or "of unknown significance." There remain CNVs whose role will only be clarified by increasing the number of cases studied, functional studies, and continued exchange between clinicians and laboratory scientists. This study represents the first project of a newly formed and growing international consortium for CNVs in epilepsy (EpiCNV), in which large-scale data aggregation and sharing will be utilized as a new tool for CNV and gene identification in the epilepsies.