Introductory
Epileptic encephalopathies are a devastating group of epilepsies with a poor prognosis,
for which the majority have unknown etiology. We perform targeted massively parallel
resequencing of 19 known and 46 candidate epileptic encephalopathy genes in 500 patients
to identify novel genes and investigate the phenotypic spectrum of known genes. Overall,
we identify pathogenic mutations in 10% of our cohort. Six of the 46 candidate genes
had one or more pathogenic variants, collectively accounting for 3% of our cohort.
We show that de novo CHD2 and SYNGAP1 mutations are novel causes of epileptic encephalopathies,
accounting for 1.2% and 1% of cases respectively. We also further expand the phenotypic
spectrum for SCN1A, SCN2A, and SCN8A mutations. To our knowledge, this is the largest
cohort of patients with epileptic encephalopathies to undergo targeted resequencing.
Implementation of this rapid and efficient method will change diagnosis and understanding
of the molecular etiologies of these disorders.
Epilepsy is one of the most common neurological disorders with a lifetime incidence
of 3%. Epileptic encephalopathies are a devastating group of epilepsies characterized
by refractory seizures and cognitive arrest or regression associated with ongoing
epileptic activity, and typically carry a poor prognosis
1
. De novo mutations in several known genes are responsible for some epileptic encephalopathies
2
. Furthermore, we and others have shown that rare, de novo copy number variants (CNVs)
account for up to ~8% of cases
3, 4
. Despite this recent progress, making a genetic diagnosis in a patient can be challenging
as there is both genetic heterogeneity for a given epilepsy syndrome and phenotypic
heterogeneity for a specific gene.
The full phenotypic spectrum associated with mutations in known epileptic encephalopathy
genes is not known. Very few studies have investigated the role of any given gene
across a wide spectrum of epileptic encephalopathy syndromes. This makes serial gene
testing in the clinical setting an inefficient and expensive process, after which
the vast majority of cases remain unexplained. Furthermore, it is clear that discovery
of additional genes that cause epileptic encephalopathies is needed to facilitate
genetic diagnosis. Here, we take advantage of a high-throughput targeted sequencing
approach to perform comprehensive sequence analysis of 65 genes (19 known genes and
46 candidate genes) (Supplementary Fig. 1) in 500 patients with a range of epileptic
encephalopathy phenotypes (Table 1). Candidate genes were selected from epilepsy-associated
CNVs (n=33) or because mutations cause associated neurodevelopmental disorders or
other epilepsy syndromes (n=13). Using this approach, we (i) identify novel epileptic
encephalopathy genes and (ii) delineate the phenotypic spectrum and mutation frequency
for both known and novel epileptic encephalopathy genes.
Overall, 91% of the target (65 genes) was sequenced at >25X coverage, required for
accurate variant calling (Supplementary Fig. 2). We achieved 91% sensitivity across
685 variants (161 loci) from 12 samples that had previously undergone exome sequencing
and 100% sensitivity for 24 known variants in previously tested patients; these patients
were not included in the discovery cohort.
We detected one or more pathogenic or likely pathogenic mutations in six of our 46
candidate genes, with multiple individuals carrying mutations in either of the two
novel epileptic encephalopathy genes, CHD2 (NM_001271.3, NP_001262.3) and SYNGAP1
(NM_006772.2, NP_006763.2) (Table 1, 2, Fig. 1).
Remarkably, we detected six de novo variants in the candidate gene, CHD2 (Fig. 1,2),
selected from within the critical interval of 15q26.1 deletions detected in patients
with a range of epileptic encephalopathies (Supplementary Fig. 3)
5, 6
. Four mutations lead to premature truncation of CHD2 (Table 2). Two de novo missense
variants disrupt highly conserved residues within the SNF2-related helicase/ATPase
domain (p.Trp548Arg and p.Leu823Pro), and are predicted to be damaging by both PolyPhen2
and SIFT. CHD2 codes for a member of the chromodomain helicase DNA-binding family
of proteins and is characterized by the presence of chromatin remodeling, chromo (chromatin
organization modifier) and SNF2-related helicase/ATPase domains. These domains suggest
function of this protein as a chromatin remodeler
7
. While functional studies in CHD2 are limited, studies of another CHD protein family
member, CHD7, have shown that the helicase domain is responsible for ATP-dependent
nucleosome remodeling, an integral process in target gene regulation. Furthermore,
in vivo studies of human CHD7 mutations within the helicase domain, which cause CHARGE
syndrome, resulted in decreased remodeling ability
8
. These results suggest that the two de novo missense mutations described here may
disrupt CHD2 function in a similar manner, while truncating mutations likely result
in haploinsufficiency.
The six patients with CHD2 mutations had distinctive features with a median seizure
onset of 18 months (range 1–3 years, Table 2): myoclonic seizures in all, photosensitivity
in three and all had ID, ranging from moderate to severe. A de novo CHD2 frameshift
mutation was reported in a proband with ID and absence seizures
9
and a de novo missense mutation in an individual with autism spectrum disorder (ASD)
10
. These results suggest that mutations in CHD2 contribute to a broad spectrum of neurodevelopmental
disorders. Notably, recent studies implicate de novo mutations in CHD8 in patients
with ASD
11
. Interestingly, three genes of the chromodomain family (CHD2, CHD7, CHD8) have now
been implicated in disorders that impact the neurodevelopmental system. Further studies
of this nine-member gene family will determine the role of each across the spectrum
of neurodevelopmental disorders, and provide exciting new avenues of research.
We identified nine pathogenic or likely pathogenic variants in four of the 13 ‘epilepsy-associated’
genes (Fig. 1). We found five truncating variants in SYNGAP1 (Fig. 2). Patients with
SYNGAP1 mutations had median seizure onset of 14 months (mean 14 months, range 6 months
to 3 years) (Table 2). They had multiple seizure types, early developmental delay
and subsequent regression. Outcome was poor with moderate to severe ID. SYNGAP1 mutations
have been associated with ID and, although most patients have epilepsy, seizures are
typically well controlled
9, 12–18
. Our study represents the first cases of epileptic encephalopathies with SYNGAP1
mutations. These observations suggest that epilepsy is a core feature of both static
and progressive encephalopathies associated with SYNGAP1 mutations, and carry important
implications for diagnostic testing.
Variants were identified in three additional ‘epilepsy associated genes’. There were
two de novo variants in MEF2C (NM_002397.4, NP_002388.2), a missense variant and a
stop-loss variant (p.*464SerExt*?). Furthermore, we found de novo pathogenic variants
in MBD5 (NM_018328.4, NP_060798.2) (Thr157Glnfs*4) and GABRG2 (NM_000816.3, NP_000807.2)(p.Arg323Gln)
(Table 2).
We detected a premature truncation mutation (p.Tyr805*) in the CNV candidate gene,
HNRNPU (NM_031844.2, NP_114032.2). The p.Tyr805* change arose as a result of two consecutive
single nucleotide changes c.471T>C and c.472A>T (Supplementary Fig. 4) that occur
two amino acids upstream of the termination codon. Neither variant was maternally
inherited; paternal DNA was not available. A recent report identified HNRNPU as a
candidate for the ID and seizure phenotypes of probands with 1q44 microdeletions
19
. In addition, a de novo splice-site variant was identified in a proband with a complex
neurodevelopmental phenotype including epilepsy
20
. Collectively, these data suggest that haploinsufficiency of HNRNPU is associated
with epileptic encephalopathy as well as ID, though further phenotype-genotype correlation
will improve our understanding of the HNRNPU phenotypic spectrum.
We identified 32 variants fulfilling our criteria for pathogenicity and an additional
four variants that are likely pathogenic in ten of 19 known epileptic encephalopathy
genes (Fig. 1, Table 1, Table 3). We identified multiple patients with mutations in
STXBP1, CDKL5, SCN1A, SCN2A, PCDH19 and KCNQ2, accounting for 69% (36/52) of all mutation-positive
individuals in our cohort. We detected an additional 16 rare variants in six of these
19 known genes for which we were unable to conduct segregation analysis; it is probable
that a number of these variants are also pathogenic (Supplementary Table 1).
The phenotypes identified in patients with mutations in known genes are provided (Table
3), and for some we expand the known phenotypic spectrum. For example, we identified
a homozygous recessive missense mutation in PNKP in a single proband with unclassified
epileptic encephalopathy. PNKP mutations are associated with early infantile epileptic
encephalopathy comprising microcephaly, early-onset intractable seizures and developmental
delay
21
. By contrast, our patient did not have microcephaly (head circumference 50th centile)
or developmental delay but had normal cognition despite refractory epilepsy with multiple
seizure types. Also, three patients with SCN1A mutations presented with an epilepsy-aphasia
phenotype, of which two also had FS+. SCN1A mutations are well known to be associated
with genetic epilepsy with febrile seizures plus (GEFS+) but have not previously been
reported with epilepsy-aphasia syndromes
22, 23
. It is possible that the SCN1A mutation is not responsible for the epilepsy-aphasia
syndrome but equally it could be a modifier predisposing the individual to this group
of epileptic encephalopathies. Further work is warranted to clarify this association,
perhaps most effectively with exome-sequencing in these patients.
We detected five variants in SCN2A, which encodes the α2 subunit of the voltage gated
sodium channel. To date, the majority of SCN2A mutations have been associated with
the self-limited autosomal dominant syndrome of benign familial neonatal-infantile
seizures (BFNIS)
24
. Previously, only three de novo variants have been reported in patients with epileptic
encephalopathies
25, 26
. Interestingly our five cases show similar variability in the range of onset seen
in BFNIS with three beginning in the neonatal period (11 hours to 2 days) and two
in infancy (6 weeks, 13 months). Two had relatively early offset of seizures at 5
weeks and 7 months. The refractory nature of seizures did not correlate with intellectual
outcome, which ranged from mild (2) to severe (3) intellectual disability. We conclude
that SCN2A is an important contributor to the overall burden of epileptic encephalopathies,
accounting for 1% of cases.
We also identified a pathogenic missense mutation (p.Leu1290Val) in SCN8A. To date,
only a single de novo SCN8A mutation (p.Asn1768Asp) has been described in a proband
with severe epileptic encephalopathy and sudden unexplained death in epilepsy
27
. Here we describe a second patient presenting with an epileptic encephalopathy beginning
at 18 months. Interestingly, this variant was paternally inherited, though the father
was shown to have somatic mosaicism (13% mutant allele) supporting its pathogenic
effect as seen in other genetic encephalopathies with parental mosaicism
28
.
The findings in this large series of patients with hitherto unsolved epileptic encephalopathies
allows us to begin to frame the overall genetic architecture of this group of disorders.
We identified pathogenic or likely pathogenic mutations in 10% of our cohort, with
mutations in 16 genes. However, this mutation rate is likely to be an underestimation
of the true contribution of each gene to the overall burden of epileptic encephalopathies.
Our cohort excluded patients with previously identified mutations, and we were unable
to conduct segregation analysis for a subset of variants we identified, some of which
are likely to be pathogenic. Furthermore, as larger numbers of patients with mutations
of specific genes are identified, distinctive epileptic encephalopathy phenotypes
are likely to emerge. Taken together, with up to 8% rare CNVs in epileptic encephalopathy
patients in an earlier analysis of a subset of this series
3
, we can now collectively ascribe causality for ~18% of all epileptic encephalopathies
of unknown cause.
The genetic heterogeneity of epileptic encephalopathies is considerable; likely pathogenic
variants were found in nine known or novel genes (see Fig. 2). Even the most commonly
mutated genes in our study each account for only up to 1.6% of cases. Notably, we
elucidate new genes found to be commonly mutated in epileptic encephalopathies, with
CHD2, SYNGAP1 and SCN2A accounting for 1–1.2% of cases each, a frequency similar to
that of mutations in SCN1A, STXBP1 and CDKL5 in our cohort. However, no mutations
were seen in nine other known genes (ARX, FOXG1, KCNT1, MECP2, PLCB1, SLC25A22, SLC2A1,
SPTAN1, ARHGEF9) in 500 patients. These results suggest that pathogenic mutations
in these genes, while important, are rare causes of epileptic encephalopathies (<0.2%
each in our cohort), or cause only very distinct syndromes that were not prevalent
in our cohort. These findings support a clinical approach to genetic diagnosis that
employs large gene panels or whole exome sequencing, as it will remain difficult and
expensive to determine a priori the causative gene in a given patient.
Notably, mutations in SYNGAP1 and CHD2 have now been described in probands with epileptic
encephalopathy, ID and ASD phenotypes, highlighting the shared genetic basis of neurodevelopmental
disorders. Unbiased approaches such as exome or whole genome sequencing provide an
avenue to gene discovery, but large cohorts will be required to identify two or more
patients with de novo mutations in the same gene
9, 18
. Our results show the power of targeted resequencing to screen large numbers of patients
in a high-throughput and cost-effective manner. This approach is critical to identify
additional patients with mutations in genes where a single de novo mutation is identified
by exome sequencing approaches, to determine overall mutation frequency in a given
phenotype, and to describe genotype-phenotype correlations. Applying this approach
across various neurodevelopmental disorders will identify additional mutation positive
patients for a specific gene and enhance our understanding of disease mechanisms.
ONLINE METHODS
Patients
This study was approved by the Human Research Ethics Committees of Austin Health and
the University of Washington. Probands with epileptic encephalopathies were recruited
from the epilepsy clinic at Austin Health, the practices of the investigators and
by referral for epilepsy genetics research from around Australia and internationally
after informed consent. The cohort consisted of 500 patients with a diverse range
of epileptic encephalopathy phenotypes. An epileptic encephalopathy was defined as
refractory seizures and cognitive slowing or regression associated with frequent,
ongoing epileptiform activity
1
. Detailed epilepsy and medical histories were obtained together with the results
of investigations including EEG and MRI studies. Epilepsy syndromes were classified
according to the Organization of the International League Against Epilepsy Commission
on Classification
1
(Table 1). Some patients had already undergone mutation screening for specific epileptic
encephalopathy genes; none had been screened for all known genes. Patients with a
previously identified disease causing mutation were excluded from this study. Some
patients with known mutations were included as mutation positive controls but were
not included in the 500 cases in the discovery cohort. Furthermore, 369/500 patients
had been previously screened for pathogenic CNVs, either in the research setting (n=257)
3
, or by clinical testing (n=112). Probands with pathogenic CNVs were not included
in this study.
Gene Selection
We selected 65 genes for sequence analysis (Fig. 1). The “known” gene group included
genes in which mutations are known to cause one or more epileptic encephalopathy syndromes
(n=19)
2, 21, 25, 27, 28, 32–34
. We also selected two sets of candidate genes for epileptic encephalopathies. The
first set includes 13 ‘epilepsy-associated genes’, more commonly implicated in patients
with non-epileptic encephalopathy forms of epilepsy (CHRNA7, KCNQ3, GABRD, GABRG2,
PRICKLE1, CACNB4, SCN1B) or a related neurodevelopmental disorder with epilepsy as
a comorbid feature (GRIN2B, MBD5, MEF2C, SYNGAP1, SYN1, ATP1A2)
35–39
. None are an established cause of epileptic encephalopathies. We also selected 33
candidate genes, primarily from epilepsy-associated CNVs, either from published cohorts,
case reports or unpublished data
3, 4, 40
(see Supplementary Note and Supplementary Table 2 for candidate gene selection).
Controls
Sixteen samples that had been previously subject to exome sequencing were included
in all analyses and used to assess the sensitivity of variant calling. We also included
24 probands with a known variant in a known or candidate gene to further validate
our approach.
Target capture and sequencing
We used Molecular Inversion Probes (MIPs) to capture all exon and intron/exon boundaries
(5bp flanking) of target genes (Refseq, hg19 build) (Supplementary Table 3). Detailed
methodology is described elsewhere
11
. Briefly, pooled MIPs were used to capture target exons from 100ng of each proband’s
DNA. PCR was performed using universal primers with the introduction of unique eight-base
barcodes on the tagged reverse primer. Pooled libraries were subject to massively
parallel sequencing using a 101 paired-end protocol on a Hiseq. The libraries were
prepared and sequenced in two batches, comprising a total of 30 (target 1) and 35
(target 2) genes.
Data analysis and variant calling
Raw read data processing and mapping with BWA (See URLs) was performed as described
11
. Single nucleotide variant (SNV) and indel calling and filtering was performed using
the Genome Analysis Tool Kit (GATK) (see URLs). Variants that did not adhere to the
following criteria were excluded from further analysis: allele balance >0.70, QUAL<30,
QD<5, coverage<25X, clustered variants (window size-10). Variants were annotated with
Seattle seq (see URLS) and the ESP6500 dataset (see URLs) was used to assess variant
frequency in the control population. For dominant (or de novo) models we considered
only variants not present in this control sample set. For recessive candidates, we
considered variants with a frequency in controls of <1%. Only non-synonymous, splice-site
or frameshift (‘damaging’) variants were assessed further. The GATK Depth of Coverage
tool was used to calculate overall depth of coverage for each sample at a threshold
of 25X, as well as the mean percentage (across all samples) of bases covered >25X
for each gene.
Rare variant segregation analysis
Where family members were available, segregation analysis was carried out for all
rare (not present in ESP6500 controls), possibly damaging (non-synonymous, essential
splice-site or frameshift) variants for all 65 target genes. This analysis was performed
using a ‘MIP-pick’ strategy. We selected and re-pooled only the MIPs that captured
the genomic sequence harboring the rare variant of interest and performed target enrichment
PCR and sequencing as above for all relevant probands and family members. This approach
allowed us to sequence variants at very high depth and detect somatic mosaicism in
parents.
Criteria for pathogenicity of rare variants
For those rare, possibly damaging variants where segregation analysis could be performed,
we required the variant to meet one of the following criteria to constitute a novel
pathogenic variant. Pathogenic variants: (i) arose de novo, (ii) segregated with the
disorder, (iii) were inherited from a parent with somatic mosaicism, or (iv) adhered
to a recessive, X-linked or parent-of-origin mode of inheritance, where applicable
(Supplementary Fig. 1).
In certain instances we were unable to determine the inheritance of a rare variant
due to the unavailability of DNA from one or more parent. It is likely that a subset
of these variants also cause disease, though here we report only those variants that
are likely to lead to protein truncation (i.e. splice-site, nonsense, frameshift,
stop-loss) as being ‘likely pathogenic’. Additionally, two missense mutations in known
genes (STXBP1, SCN2A) were interpreted to be ‘likely pathogenic’ based on the high
incidence of pathogenic missense mutations in these genes, which was further supported
by the available parent not carrying the variant. We performed microsatellite analysis
using the PowerPlex S5 system [Promega] in all parents of probands with a de novo
mutation to confirm maternity and paternity.
Supplementary Material
1