Novel and de novo mutations in pediatric refractory epilepsy

Pediatric refractory epilepsy is a broad phenotypic spectrum with great genetic heterogeneity. Next-generation sequencing (NGS) combined with Sanger sequencing could help to understand the genetic diversity and underlying disease mechanisms in pediatric epilepsy. Here, we report sequencing results from a cohort of 172 refractory epilepsy patients aged 0–14 years. The pathogenicity of identified variants was evaluated in accordance with the American College of Medical Genetics and Genomics (ACMG) criteria. We identified 43 pathogenic or likely pathogenic variants in 40 patients (23.3%). Among these variants, 74.4% mutations (32/43) were de novo and 60.5% mutations (26/43) were novel. Patients with onset age of seizures ≤12 months had higher yields of deleterious variants compared to those with onset age of seizures > 12 months (P = 0.006). Variants in ion channel genes accounted for the greatest functional gene category (55.8%), with SCN1A coming first (16/43). 81.25% (13/16) of SCN1A mutations were de novo and 68.8% (11/16) were novel in Dravet syndrome. Pathogenic or likely pathogenic variants were found in the KCNQ2, STXBP1, SCN2A genes in Ohtahara syndrome. Novel deleterious variants were also found in West syndrome, Doose syndrome and glucose transporter type 1 deficiency syndrome patients. One de novo MECP2 mutation were found in a Rett syndrome patient. TSC1/TSC2 variants were found in 60% patients with tuberous sclerosis complex patients. Other novel mutations detected in unclassified epilepsy patients involve the SCN8A, CACNA1A, GABRB3, GABRA1, IQSEC2, TSC1, VRK2, ATP1A2, PCDH19, SLC9A6 and CHD2 genes. Our study provides novel insights into the genetic origins of pediatric epilepsy and represents a starting-point for further investigations into the molecular pathophysiology of pediatric epilepsy that could eventually lead to better treatments. Electronic supplementary material The online version of this article (10.1186/s13041-018-0392-5) contains supplementary material, which is available to authorized users.


Introduction
Epilepsy is a complex group of chronic brain disorders that are characterized by recurrent spontaneous seizures, and these can often begin in childhood. Repeated and refractory seizures can cause long-term cognitive impairment, decreased social participation and significantly lower quality of life [1,2]. Epilepsy is one of the most common neurological disorders with 50 to 100 million affected worldwide, and 2 to 4 million new cases diagnosed each year [3].
Epilepsy is a heterogeneous disease with diverse clinical manifestations and causes, including altered ion channel expression, neurotransmitter signaling, synaptic structure, gliosis, and inflammation [1]. Estimates of heritability from twin studies range from 25% to 70% [4,5]. Although the range of heritability estimates is quite large, disparate studies using varied methods and studying divergent populations are all consistent in concluding that there is a substantial inherited component to epilepsy [6]. Because of this, we sought to investigate the genome in a heterogeneous set of patients with epilepsy and their parents, with the hope that we would identify novel mutations and confirm existing reports of genetic associations with epilepsy. This type of genetic information can provide an entry point into the biology of epilepsy that could eventually lead to new molecular treatment targets.
With the rapid progress of next-generation sequencing (NGS) techniques, our knowledge of the genetic etiology in many brain disorders such as epilepsy, autism and intellectual disability has expanded greatly [7,8]. NGS is now capable of efficient and accurate sequencing of entire genomes with small amounts of tissue at ever decreasing costs and has required new approaches to analysing the very large amount of data obtained. For this study, our priority was to separate common and benign genetic variants from those that are likely to be related to the cause of epilepsy, and we chose to apply the American College of Medical Genetics and Genomics (ACMG) guidelines [9][10][11]. The ACMG guidelines classify variants into pathogenic, likely pathogenic, uncertain significance, likely benign, and benign categories based on genetic information that includes population, functional, computational and segregation data. In this study, we investigated 153 epilepsy candidate genes in a cohort of 172 refractory epilepsy pediatric patients. We aimed to provide genetic diagnoses of this patient cohort and explore the genetic etiology of pediatric refractory epilepsy.

Participants
We retrospectively collected and analyzed 172 cases of pediatric refractory epilepsy patients between the ages of 1 day to 14 years old in the Department of Pediatrics of Qilu Hospital, China. The program adhered to guidelines of patients' consent for participation and research was supported by the Ethics Committee of Qilu hospital, Shandong University (No. 2016(027)).
All patients were examined and diagnosed at the Pediatric Department in Qilu Hospital using a combination of patients' illness history, previous history, family history, physical examinations, developmental evaluation, hematological examination, ambulatory or video electroencephalography (AEEG/VEEG) monitoring, magnetic resonance imaging (MRI) or computed tomography (CT), and genetic sequencing. Developmental evaluation included gross motor, fine motor, language, and personalsocial skills. The above information was reviewed by two qualified pediatric epileptologists. Seizure types and epilepsy syndromes were diagnosed and classified according to the guidelines of International League Against Epilepsy (2014, 2017) [12,13].
Next-generation sequencing Targeted gene capture and sequencing Blood samples of the patients and their biological parents were collected to test if the mutations were de novo or inherited. Genomic DNA was extracted from peripheral blood using the QIAamp DNA Mini Kit (Qiagen, China).
One hundred fifty-three genes (Table 1) associated with epilepsy were selected by a gene capture strategy, using the GenCap custom enrichment kit (MyGenostics, China) following the manufacturer's protocol. The biotinylated capture probes were designed to tile all of the exons without repeated regions. The captured DNAs were eluted, amplified and then their polymerase chain reaction (PCR) products were purified with SPRI beads (Beckman, USA). The enriched libraries were sequenced for paired-end reads of 150 bp by Illumina HiSeq X Ten.

Data analysis and pathogenicity of candidate variants
After sequencing, raw data were saved in FASTQ format. Illumina sequencing adapters and low quality reads (< 80 bp) were filtered by Cutadapt [14]. Clean reads were aligned to UCSC hg19 human reference genome using the Burrows-Wheeler Alignment [15] tool. Duplicated reads were removed using Picard (http://broadinstitute.github.io/picard). Insertions, deletions and SNP variants were detected and filtered using the Genome Analysis Toolkit [16]. Then the identified variants were annotated using ANNOVAR [17] and associated with the following databases: 1000 genomes, Exome Aggregation Consortium, The Human Gene Mutation Database, and predicted by Mutation Taster (MT) [18], Sorting Intolerant From Tolerant (SIFT) [19], PolyPhen-2 (PP2) [20] and Genomic Evolutionary Rate Profiling (GERP++) [21,22]. Splice-site were predicted by Human Splicing Finder [23]. All variants identified by the Illumina HiSeq X Ten sequencer were confirmed by Sanger sequencing. The pathogenicity of mutations was assessed in accordance with American College of Medical Genetics and Genomics guideline (ACMG) [9][10][11].

Statistical analysis
Statistical analysis was performed using SPSS19. The yields of deleterious variants in patients with different onset age or family history were compared using the chisquared test.

Results
In the current study, we recruited 172 epilepsy pediatric patients, including 23 with Dravet syndrome, ten with Ohtahara syndrome, two with Ohtahara syndrome evolving to West syndrome, ten with West syndrome, two with West syndrome evolving to Lennox-Gastaut syndrome, five with Lennox-Gastaut syndrome, four with Doose syndrome, two with epilepsy of infancy with migrating focal seizures, two with epileptic encephalopathy with continuous spike and wave during sleep, and one each with temporal lobe epilepsy, early myoclonic encephalopathy, Landau-Kleffner syndrome, and glucose transporter type 1 deficiency syndrome. Three patients had Rett syndrome, five had tuberous sclerosis complex, and one had Sturge-Weber syndrome. Forty-two patients were diagnosed as unclassified epileptic encephalopathy and 57 patients were diagnosed as unclassified refractory epilepsy due to nonspecific manifestations ( Table 2).
The DNA samples of patients were analyzed by using NGS and the variants were validated by Sanger Sequencing. For the samples subjected to targeted sequencing, the quality assurance (QA) /quality control (QC) file are provided in Additional file 1: Table S2.
More recent studies suggest that many severe epilepsy types begin in infancy or childhood, especially those with psychomotor retardation and epileptic encephalopathies are often due to de novo mutations [30,31]. In our study, 32/43 (74.4%) pathogenic or likely pathogenic variants were de novo, five (11.6%) were paternal, one (2.3%) was maternal, and five (11.6%) were unknown due to blood samples from parents were unavailable (Table 3).
To further explore the genetic pathogenesis of epilepsy, we subdivided the mutated genes into nine groups according to  (Fig. 1b). Ion channels (voltage-gated and ligandgated) accounted for 55.8% in total, suggesting that dysfunction of ion channels plays critical roles in the pathogenesis of epilepsy.
There were 16 mutations in SCN1A gene, of which six (37.5%) were missense mutations, one (6.25%) was a splice site mutation, four (25%) were nonsense mutations, four (25%) were frame-shifts, and one (6.25%) was deletion mutation. Thirteen of the 16 (81.3%) SCN1A mutations were de novo and 11 (68.8%) were novel. We further analysed the positions of the mutations in the affected proteins corresponding to gene mutations and found that 43.8% (7/16) of protein changes are in the intracellular loop of sodium channel protein type 1 subunit alpha, 31.3% (5/16) are in the extracellular loop, 18.8% (3/16) are in the transmembrane region, and 6.25% (1/16) are in the pore forming area (Fig. 2).
There has been a marked increase in genetic diagnoses of a number of key childhood-onset epilepsy syndromes, such as Dravet syndrome, which has been mainly linked to SCN1A [17]. In our 16 patients diagnosed as Dravet syndrome with pathogenic or likely pathogenic variants, all identified mutations were in the SCN1A gene. These 16 Dravet syndrome patients had typical manifestations: onset between 3 to 8 months of age, fever-sensitive, multiple seizure types, and developmental delay after seizure onset. 81.25% (13/16) SCN1A mutations were de novo in Dravet syndrome patients and one was inherited from the father who had a history of febrile seizures (FS). 12.5% (2/ 16) SCN1A mutations were unknown. Pathogenic and likely pathogenic mutations each accounted for 50% (Table 4). 50% (8/16) of the Dravet syndrome variants cause nonsense or frameshift mutations that result in truncated proteins, which was consistent with a previous study [57]. We evaluated whether different seizure types, family history, abnormal brain MRI, or developmental delay were associated with specific SCN1A mutation types or locations within the gene. We did not detect any bias towards particular regions of the gene or in the type of mutation, although our small sample size did not provide substantial power (Additional file 1: Tables S3 and S4).
One of the West syndrome patients were found to carry two mutations: a nonsense ADSL (c.253C > T, p.R85X) mutation was scored as likely pathogenic and was inherited from her unaffected mother. Another reported missense ADSL (c.71C > T, p.P24L) [58] mutation which was inherited from her unaffected father were scored as uncertain pathogenicity. ADSL has been reported to be related       to adenylosuccinate lyase deficiency, which is an autosomal recessive defect of purine metabolism [59,60]. The patient presented with spasms 2 months after birth. Brain MRI showed cerebral dysplasia and EEG showed hypsarrhythmia and multifocal discharges. The patient also had developmental delay and lack of eye contact. A definitive diagnosis can be made with high performance liquid chromatography examination of the urine to detect the ratio of succinyladenosine and succinyl-aminoimidazole carboximide riboside, but this was not available for the patient in question. Thus, this patient was diagnosed clinically as having West syndrome. A novel frame-shift mutation in SYNGAP5 (c.274_277del, p.G92fs) was detected in a patient with Doose syndrome. This patient presented with myoclonic and myoclonicastatic seizures, as well as having atypical absence seizures. SYNGAP5 had been reported to be associated with Doose syndrome and mental retardation, autosomal dominant 5 (MRD5) [51,61,62]. This mutation, which is very rare, was de novo, and caused frameshift changes in Ras/Rap GTPaseactivating protein SynGAP, was therefore scored as pathogenic (Table 5).
One glucose transporter type 1 deficiency syndrome patient presented with seizures at age 28 months. The patient has alopecia and was almost bald at 4 years old. The child did not have other abnormalities in blood tests, brain MRI, or neurological exam. Her cerebrospinal fluid glucose value was 2.04 mmol/L (blood glucose value was 7.2 mmol/L before lumbar puncture; fasting blood glucose value was 5.2 mmol/L). NGS identified a missense mutation in SLC2A1 (c.296T > G, p.M99R). The mutation was de novo and novel. The patient's parents and sister were normal, which is consistent with the sequencing results. Symptoms improved with a ketogenic diet, with seizures controlled for more than 6 months.
One MECP2 mutation (c.799C > T, p.R267X) was detected in a girl diagnosed as Rett syndrome. The girl developed normally for the first 18 months, gradually lost speech ability while developing repetitive hand-wringing. Seizures began at age 3 years. The MECP2 gene is located on the X-chromosome, and Rett syndrome is inherited through this gene in a dominant fashion [63]. This patient had a de novo MECP2 nonsense mutation, consistent with her parents being unaffected. 40% (2/5) of tuberous sclerosis complex patients were diagnosed with West syndrome associated with tuberous sclerosis complex in our study. Tuberous sclerosis complex is closely related to the TSC1/TSC2 genes [64][65][66][67].
In our study, all of the tuberous sclerosis complex patients' initial presentations were seizures, of which 80% (4/5) presented in the first year of life. 60% (3/5) had hypomelanotic macules and 40% (2/5) had multi nodules. One patient's only clinical manifestation was seizures and three (60%) patients with seizures had only one major feature of tuberous sclerosis complex. After sequencing, 60% (3/5) patients were found to have deleterious TSC1 or TSC2 mutations.
We identified more than one TSC1/2 mutations in 2 patients. One patient has two TSC2 mutations inherited from his affected father. Facial angiofibromas appeared by age 3-4 years in 60% (3/5) patients in the follow-up period. Gilboa et al. [68] reported four patients with the same TSC1 genomic deletion (9q34.13q34.2) in a family   and none of them fulfilled the clinical criteria for tuberous sclerosis complex. In our study, one patient with pathogenic TSC1 (c.2768_2769insC, p.L924Ffs*26) mutation presented with focal seizures beginning at age four. There were two hypopigmented macules on the patient's abdomen. The brain MRI results were normal and there are no other features of tuberous sclerosis complex. This de novo mutation causes a frame-shift in hamartin and has not been reported previously. Thus, this patient was considered to have unclassified refractory epilepsy. One unclassified epileptic encephalopathy patient had two deleterious mutations: SCN8A inherited from his affected father (c.4324G > A, p.E1442K) and IQSEC2 (c.4246 _4247insG, p.S1416fs). Early-infantile epileptic encephalopathies (EIEE) caused by SCN8A mutations are designated as EIEE13 (OMIM #614558) [69]. The missense mutation in SCN8A is very rare in the general population, and had been previously predicted to be damaging by SIFT, MT and PP2. IQSEC2 is an X-linked gene that has been reported to be related to intellectual disability and epilepsy, and it encodes the IQ motif and SEC7 domain-containing protein 2 [70]. The identified novel IQSEC2 mutation was de novo and was scored as being pathogenic.
Other pathogenic or likely pathogenic mutations found in patients with unclassified epileptic encephalopathy included CACNA1A, GABRA1, GABRB3, PCDH19, and CHD2. Epileptic encephalopathies with the above mutations had been designated as EIEE42, EIEE19, EIEE43, EIEE9 and EEOC (childhood-onset epileptic encephalopathy) according to Online Mendelian Inheritance in Man (OMIM). Other deleterious variants found in patients with unclassified refractory epilepsy were in VRK2, ATP1A2, and SLC9A6. Taking these unclassified epileptic encephalopathies and unclassified refractory epilepsy patients' clinical manifestations into consideration, we found that all patients with deleterious mutations in genes encoding ion channels (SCN8A, CACNA1A, GABRB3, GABRA1) had similar clinical symptoms: onset age of seizures within the first year, epileptic encephalopathy and developmental delay. In contrast, patients with mutations in VRK2, ATP1A2, and SLC9A6, had relatively later onset age of seizures.
We then assessed the clinical benefit of genetic testing in those patients with identified deleterious variants. NGS  (Table 6).

Discussion
Epilepsy is highly heterogeneous and can be primarily genetic in origin, or be secondary to structural or metabolic disorders of the central nervous system [71,72]. To date, over 500 genes have been implicated in epilepsy [73][74][75][76]. However, the overlapping clinical features of different epilepsy syndromes and non-specific phenotypes can hamper clinical and genetic diagnosis [53]. The correct genetic diagnosis can help to guide treatment and prognosis. In addition to genetic origins, pediatric epilepsy may also arise from epigenetic mechanisms mediating gene-environment interactions during neurodevelopment. In this study, we used NGS to investigate 153 epilepsy related genes in a cohort of 172 refractory epilepsy children.
An SCN1A mutation was first discovered in epilepsy in 2000 [72], and now hundreds of new SCN1A mutations have been described in epilepsy patients, making it the most common epilepsy-related gene [78]. In our study, we found SCN1A mutations in 16/44 deleterious variants, making it the most common gene to show variation in our study. SCN1A encodes the Nav1.1 pore-forming α-subunit, expressed mainly in inhibitory GABAergic neurons. The α-subunit comprises four homologous domains (I-IV), forming a tetrameric structure. Each domain is composed of six transmembrane segments (S1-S6) [77]. The S4, voltage-sensing segment has multiple positively charged amino acids. The intracellular loop between III and IV domain functions as the inactivation gate. The α-subunit is usually associated with two β-subunits that influence α-subunit localization and function [77]. Among α-subunit of sodium channel genetic variants in our study, 43.8% (7/16) are within the intracellular loop, 31.3% (5/16) in the extracellular loop, 18.8% (3/16) in the transmembrane area, and 6.25% (1/16) in the pore forming area. All the extracellular mutations are between S5 and S6, which is very close to the pore forming area. These variants may influence the initiation and propagation of action potentials, making these inhibitory GABAergic neurons less excitable. Some antiepileptic drugs (AEDs) bind to the inner cavity of the pore of the sodium channel (IS6, IIIS6 and IVS6) [77,79]. The pore forming area or internal/external loop could be promising targets for new seizure prophylaxis medications.
Patients harboring SCN1A mutations can have with Dravet syndrome or generalized epilepsy with febrile seizures plus. One Dravet syndrome patient inherited the SCN1A mutations from his father only had febrile seizures. This could be due to somatic mosaicism [72,80,81]. A Dravet syndrome mouse model (Nav1.1 knockout-based) responded well to stiripentol and clobazam, which are commonly used to treat Dravet syndrome [82][83][84][85]. One of the patients in our study was treated with oxcarbazepine, which blocks sodium channels and worsened seizures, before the diagnosis of Dravet syndrome was made. This case illustrates the importance of correct molecular diagnosis in selecting the best anticonvulsant.
Approximately half of Ohtahara syndrome patients with STXBP1 mutations evolve to West syndrome [86]. In our study, there was one such patient with a nonsense mutation in STXBP1, suggesting that this gene could play a role in the etiology of West syndrome. Our findings also suggest that STXBP1 is related to both Ohtahara syndrome and West syndrome.
Genetic studies of neuropsychiatric disease have led to the discovery of molecular etiology and pathophysiology. For example, most cases of Rett syndrome are now known to arise from mutations in the MECP2 gene, which codes for a methyl-CpG-binding protein 2 [90]. Another example is glucose transporter type 1 deficiency syndrome, which has been attributed to variants in SLC2A1, SLC2A2, and GLUT1. In our study, the glucose transporter type 1 deficiency syndrome patient did not have cerebrospinal fluid analysis as part of their diagnostic work-up until the genetic data suggested the diagnosis. This example illustrates the utility of NGS in clinical  [91,92]. In our study, SCN1A was the main deleterious variant in Dravet syndrome and KCNQ2, STXBP1, SCN2A were found in Ohtahara syndrome. Deleterious variants in STXBP1, KCNT1, CDKL5, ADSL genes were found in West syndrome. Novel mutations in SYNGAP1 were found in Doose syndrome, a SLC2A1 mutation was found in GLUT1-DS and a de novo MECP2 mutation were found in Rett syndrome. TSC1/TSC2 variants were found in 60% of patients with tuberous sclerosis complex. Mutations found in unclassified epileptic encephalopathy were mainly in ion-channel genes. Thus, our study reinforces previous observations that the clinical syndrome and genetic etiology do not always match.
We tested 153 epilepsy genes and found 43 pathogenic and likely pathogenic variants in this study. Considering that over 500 epilepsy genes have been reported [73][74][75][76], our work was not comprehensive, which is a limitation of this study. With the decreasing cost of whole genome sequencing, the interrogation of the entire genome is now feasible for larger samples of epilepsy patients, and this approach has already been fruitful in other neuropsychiatric disorders such as autism, Kabuki syndrome, Bohring-Opitz syndrome and others [93,94].
For genetic testing, it is proposed to conduct the strong candidate gene sequencing first (SCN1A for Dravet syndrome, MECP2 for Rett syndrome and TSC1/2 for tuberous sclerosis complex) before a NGS multi-gene panel testing [95][96][97]. In our study, we conducted targeted panel sequencing on Dravet syndrome and Rett syndrome patients before screening the strongest candidate gene for the following reasons. First, the correct clinical diagnosis of these syndromes can be difficult, especially in some of the younger patients in our sample, and often requires longitudinal assessment, which delays the correct diagnosis. Thus, we elected to perform NGS on our subjects before knowing the clinical diagnosis in some cases, such as these syndromes. Since our NGS panel that contains 153 epilepsy genes, our approach could facilitate the correct diagnosis in some cases. Second, it is now apparent that while 70-80% Dravet syndrome patients have SCN1A mutations, mutations in other genes such as SCN1B, SCN2A, SCN8A, PCDH19, GABRA1, GABRG2, STXBP1, CHD2 genes can cause Dravet syndrome like phenotypes [98], which would be missed if only SCN1A was sequenced. Similarly, CDKL5 and FOXG1 have been associated with atypical Rett syndrome [99], in addition to MECP2.
In tuberous sclerosis complex patients, we have a similar clinical scenario in which most features of tuberous sclerosis complex become evident only after 3 years of age, limiting their usefulness for early diagnosis [100]. In our study, all of the tuberous sclerosis complex patients' initial presentations were seizures, of which 80% presented in the first year of life. 60% had hypomelanotic macules and 40% had multi nodules. 20% patient's only clinical manifestation was seizures and 60% patients with seizures had only one major feature of tuberous sclerosis complex. 60% patients were found to have deleterious TSC1 or TSC2 mutations by NGS sequencing. Facial angiofibromas appeared by age 3-4 years in 60% patients in the follow-up period.
In summary, we identified 43 pathogenic or likely pathogenic variants, of which 26 mutations were novel and 32 were de novo. Variants in ion channel genes accounted for the largest category of gene in children with refractory epilepsy. Dravet syndrome is closely related to the SCN1A gene, which was the most frequently-appearing gene showing variants in our study. Novel and de novo mutations were found in Ohtahara syndrome, West syndrome, Doose syndrome and tuberous sclerosis complex pediatric patients. We also found a novel mutation in glucose transporter type 1 deficiency. Our results reinforce the importance and feasibility of precise genetic diagnosis for epilepsy, with the hope that in future, this will both aid in understanding the molecular pathophysiology and lead to new treatment targets.

Additional file
Additional file 1: Table S1. The expression levels of the 153 targeted genes in brain. Table S2. The quality assurance (QA) /quality control (QC) of targeted sequencing. Table S3. The frequencies of different mutation locations in SCN1A gene and their corresponding phenotypes in Dravet syndrome patients. Table S4