Widespread genomic influences on phenotype in Dravet syndrome, a ‘monogenic’ condition

Abstract Dravet syndrome is an archetypal rare severe epilepsy, considered ‘monogenic’, typically caused by loss-of-function SCN1A variants. Despite a recognizable core phenotype, its marked phenotypic heterogeneity is incompletely explained by differences in the causal SCN1A variant or clinical factors. In 34 adults with SCN1A-related Dravet syndrome, we show additional genomic variation beyond SCN1A contributes to phenotype and its diversity, with an excess of rare variants in epilepsy-related genes as a set and examples of blended phenotypes, including one individual with an ultra-rare DEPDC5 variant and focal cortical dysplasia. The polygenic risk score for intelligence was lower, and for longevity, higher, in Dravet syndrome than in epilepsy controls. The causal, major-effect, SCN1A variant may need to act against a broadly compromised genomic background to generate the full Dravet syndrome phenotype, whilst genomic resilience may help to ameliorate the risk of premature mortality in adult Dravet syndrome survivors.


Introduction
With the discovery of numerous monogenic epilepsies, our understanding of the genetic architecture underlying developmental and epileptic encephalopathies (DEEs) has grown immensely. 1The initial identification of monogenic epilepsies is usually made through genetic studies of individuals with relatively homogeneous phenotypes.Subsequent characterization of additional cases with pathogenic variants in the same gene typically broadens the phenotypic spectrum. 2,3his evolving breadth of clinical presentations, even with a core defining phenotype, can become surprisingly wide and unexplained.One potential source of such phenotypic diversity within a single monogenic epilepsy may be variation across the rest of the genome.This possibility is rarely explored; typically, genetic investigations cease with the discovery of the first plausibly culpable variant.
Pathogenic variants in the voltage-gated sodium channel alpha subunit 1 gene (SCN1A) are one of the most frequent causes of monogenic epilepsies, though all are rare. 4The archetypal phenotype associated with pathogenic SCN1A variants is Dravet syndrome.The spectrum also includes familial febrile seizures, genetic epilepsy with febrile seizures plus (GEFS+), and other SCN1A-related epilepsies that do not obviously fit these categories but may share some core features, such as fever-provoked seizures. 5Further, people with pathogenic variants in SCN1A may also present with features beyond epilepsy, including mild to severe intellectual disability, behavioural problems and movement disorders. 5Within SCN1A-related conditions, and even for a given pathogenic variant, phenotypic heterogeneity can be observed: a given SCN1A variant may segregate with epilepsy in a family, and cause GEFS+ in one individual, and Dravet syndrome in another; individuals meeting a tight clinical definition for Dravet syndrome, harbouring identical SCN1A variants, may show divergent phenotypes.This wide range of associated phenotypes confounds prognostication for infants with SCN1A-related epilepsies and makes treatment challenging.As a prototypic monogenic disorder, SCN1A-related epilepsies provide a model for elucidating the potential contribution of background genetic architecture to the disease phenotype.
Additional genetic factors have been implicated in the phenotypic diversity seen in SCN1A-related epilepsies.Disease severity could be modulated by genomic factors directly related to SCN1A, such as variant class, mosaicism of the pathogenic SCN1A variant, or variants in non-coding regulatory regions affecting the expression of the mutated or wild-type SCN1A allele. 6,7Alternatively, variants in other genes may influence SCN1A-related epilepsy phenotypes, constituting blended phenotypes that reflect an aggregation of distinct or overlapping features, depending on the pathway or function of the gene(s) harbouring the additional variant(s). 8The poly-genetic 'background' of each individual may act as a phenotypic modifier.Evidence from animal models suggests that genetic background may modulate Dravet-like phenotypes, whilst an enrichment of rare variants in neuronal excitability genes has been reported in severe Dravet syndrome compared to mild Dravet syndrome. 9,10Beyond genomic influences, clinical management, including medication choices, may also affect outcomes, 11 potentially through interactions with individual genetic features.
To test the hypothesis that the background genetic architecture influences the phenotypic presentation of individuals with monogenic epilepsy, we used whole-genome sequencing (WGS) across a cohort of adults with clinically well-characterized SCN1A-related Dravet syndrome.We studied several features of background genomic variation, including the contribution of rare variants in epilepsy-related genes, and common variation across the genome, including polygenic risk scores (PRS), aiming to elucidate whether these features influence Dravet syndrome phenotypes.

Ethics statement
This research was approved by the relevant ethics committee.For all cases, written informed consent for research use of clinical and genetic data was obtained from patients, their parents, or legal guardians in the case of those with intellectual disability.All individuals for whom detailed phenotypic information is provided were recruited through a REC-approved study (REC 11/LO/2016), and all phenotypic and genetic information was gathered under this approval.

SCN1A-related Dravet syndrome cohort
Thirty-four adults with SCN1A-related Dravet syndrome were recruited from epilepsy clinics at the National Hospital for Neurology and Neurosurgery, London, UK through a REC-approved study (REC 11/LO/2016).WGS was performed on DNA extracted from peripheral blood (Supplementary material 1).Detailed clinical phenotyping was undertaken by L.M.C. after comprehensive review of the medical records.The Dravet syndrome phenotype was re-evaluated independently by L.M.C., S.B. and S.M.S. with reference to the diagnostic criteria for Dravet syndrome recently proposed by the International League Against Epilepsy (ILAE) 12 (Supplementary Table 1 and Supplementary material 2).
The full cohort of 34 individuals with Dravet syndrome was used for the blended phenotype analysis.For PRS and burden analyses, only individuals of European ancestry (28/34) were included (Supplementary Fig. 1 and Supplementary material 3).A cohort including 13 individuals with Dravet syndrome of European ancestry who have missense SCN1A variants was used for post hoc analyses.

Control cohorts
All control cohorts were compiled from participants recruited to the Genomics England (GEL) 100 000 genomes project (Supplementary Fig. 2).Only individuals of European ancestry were considered in the control cohorts (Supplementary Fig. 1 and Supplementary material 3).

Genomics England epilepsy controls
The GEL epilepsy control cohort consisted of 772 adults with epilepsy recruited from clinics at the National Hospital for Neurology and Neurosurgery, London, UK, through a REC-approved study (REC 11/ LO/2016) and genotyped by the GEL 100 000 genomes project.All individuals fell within the GEL 'epilepsy and other features' disease group.The human phenotype ontology (HPO) terms used for these individuals when recruited to the GEL 100 000 genomes project can be found in Supplementary Table 2. To minimize the possibility that individuals within this cohort had SCN1A-related epilepsies, individuals with unique variants in SCN1A (i.e.not present in gnomAD, version 3.1.1)were excluded (Supplementary Fig. 2).

Genomics England controls
The GEL control cohort consisted of 1187 unaffected relatives of probands from GEL disease categories considered to be unrelated to epilepsy (Supplementary Table 3). 13,14Medical information regarding these individuals is unknown, and a proportion, likely reflective of the prevalence of active epilepsy in the UK (5-10 per 1000), may have epilepsy, which would serve only to reduce the power of our comparisons.To minimize the number of individuals with potential 'monogenic' epilepsies in this cohort, individuals with unique variants (i.e.not present in gnomAD) in epilepsy-related genes were excluded (Supplementary Fig. 2).

Genomics England SCN1A controls
Following testing of the primary hypotheses, it became clear that a further post hoc investigation would be useful, examining individuals bearing ultra-rare SCN1A variants, but without epilepsy.The GEL SCN1A control cohort consisted of 45 GEL probands of European ancestry [median age at recruitment 37 years (range 4-71)] from disease categories considered to be unrelated to epilepsy (Supplementary Table 3), 13 who were also identified as having unique/ultra-rare SCN1A missense variants (i.e.not present in gnomAD) (Supplementary Fig. 2).No individuals in the disease categories considered to be unrelated to epilepsy had truncating SCN1A variants.HPO terms and medical history timelines were reviewed for all identified cases and no individuals were found to have phenotypes that are known to be associated with SCN1A variants (Supplementary material 4 and Supplementary Table 4).

Epilepsy-related gene selection and annotation
To test the hypothesis that the phenotypic heterogeneity seen in Dravet syndrome could be partly explained by variation in other epilepsy-related genes, in addition to SCN1A, samples were screened for rare variants across the canonical coding sequences of 190 monoallelic or X-linked epilepsy-related genes in the GEL Genetic Epilepsy Syndromes (version 2.489) panel (Supplementary Table 5 and Supplementary material 5).Only genes designated by GEL with a 'green' rating, (i.e.those in which there is a high level of evidence for gene-disease association), were included and are referred to as 'epilepsy-related genes'. 13,15Rare variants were defined as those with an allele frequency in gnomAD ≤0.0005, which is in line with previously defined 'rare' variant allele frequencies. 16,17he region of each epilepsy-related gene was extracted from variant call format and annotated using ANNOtate VARiation (ANNOVAR, version 2019Oct24).Stop-gains, frameshift-deletion, frameshift-insertion, in-frame-deletion, in-frame-insertion, splicing, and missense variants with a read coverage ≥8 were selected as qualifying variants.All variants were confirmed manually using the Integrative Genomics Viewer (IGV, version 2.9.4).

Gene and gene-set based collapsing analyses of rare variants
An enrichment of rare variants in known epilepsy-related genes confers risk for common and rare epilepsies. 16To test the hypothesis that there was an excess of rare variants in epilepsy-related genes in individuals with Dravet syndrome compared with GEL Epilepsy controls, we performed a gene-based and gene-set collapsing analyses for rare variants across 190 epilepsy-related genes. 13,15The optimal sequence kernel association test (SKAT-O) as implemented in SKAT R package version 2.0.1 was used. 18CN1A variants were excluded in both gene-based and gene-set collapsing analyses, to avoid the overestimation of enrichment of rare variants.The variants in these 190 genes were identified using region extraction and Ensembl Variant Effect Predictor (VEP) annotation. 19Variants that were observed less than three times in each cohort were included in the SKAT-O analysis.Gender was included as a covariate.A small sample size adjustment by SKAT-O was used.To determine if X chromosome gene variants were driving enrichment of rare variants in Dravet syndrome cases, we performed a rare variant collapsing analysis for the 153 epilepsy-related genes on autosomal chromosomes.To explore whether the burden of rare variants in epilepsy-related genes may influence the expressed phenotype in the setting of a unique SCN1A variant, a post hoc analysis was performed estimating the gene and gene-set based rare variant enrichment across the Dravet syndrome and GEL SCN1A control cohorts. 20Bonferroni correction was applied to P-values to correct for multiple testing.

Blended phenotypes
Several large patient series have shown that 3.2-7.2% of those in whom a molecular diagnosis has been identified have multiple molecular diagnoses, i.e. a pathogenic variant at more than one genetic locus, each associated with a distinct clinical disease, and each segregating independently. 8Each independent clinical-molecular diagnosis may have distinct or overlapping phenotypic features which together result in a 'blended phenotype', representing the complex interaction between effects of pathogenic variants in multiple genes within one individual. 8To test the hypothesis that phenotypic heterogeneity could be explained by 'blended phenotypes' in some individuals with Dravet syndrome, rare variants in additional epilepsy-related genes were evaluated for 'potential clinical relevance' (Fig. 1 and Supplementary material 6).All variants that met the 'potential clinical relevance' criteria were evaluated by three clinicians (L.M.C., S.B. and S.M.S.), and the published phenotypes associated with each epilepsy-related gene were compared with the phenotype of the individual harbouring that gene variant, to determine its potential contribution.Additional variants were determined to potentially contribute to blended phenotypes when aspects of the individual's phenotype were better explained by the additional epilepsy-related gene variant than the SCN1A variant (Fig. 1).Variants that were deemed to contribute to blended phenotypes were subsequently classified using American College of Medical Genetics and Genomics-Association for Molecular Pathology (ACMG-AMP) criteria, excluding the criterion 'BP5 alternate locus observations' due to the known presence of the SCN1A variant, 21 and were included if they were classified as pathogenic, likely pathogenic or variants of uncertain significance (VUS).

PRS
To test the hypothesis that common genetic variation also influences the phenotype, PRS were calculated for epilepsy, intelligence and longevity in the Dravet syndrome, GEL Epilepsy and GEL control cohorts.PRS for intelligence, longevity and epilepsy were estimated using genome-wide association study (GWAS) summary statistics generated by the ILAE Consortium on Complex Epilepsies, 22 Savage et al. 23 and Deelen et al., 24 respectively.To investigate the formal genetic correlation between intelligence, longevity and epilepsy, we performed linkage disequilibrium score regression (LDSC) comparing the GWASs used for each PRS estimation (Supplementary Fig. 3).Genetic correlation rates were calculated using the LDSC tool 25 (Supplementary material 7).
Following quality control steps (Supplementary material 8), we calculated PRS based on the overlap of the study groups' remaining quality-controlled single nucleotide polymorphisms (SNPs). 26PRS for each individual was obtained using the clumping and thresholding method implemented by PRSice-v2.3.3 across a set of P-value thresholds (PT = 10 −4 , 10 −3 , 10 −2 , 5 × 10 −2 , 10 −1 , 0.5, 1). 27PT with the best fit for the target trait across the thresholds was identified (Supplementary material 9, and Supplementary Figs 4-10).R 2 was used to measure the variance explained by the PRS and was produced directly from PRSice. 27o compare PRS between the three cohorts for the selected bestfit PT, a one-way ANOVA was applied (Supplementary material 10).The analysis of variance model was adjusted for sex and the first four principal components of ancestry, which further controls for ancestry bias. 28Differences in the means between each pair of groups were assessed for significance using a post hoc multiple pairwise comparison (Tukey's test).To correct for multiple testing across three PRS analyses Bonferroni correction was applied to P-values and the significance set to α = 0.05 / 3.
To further demonstrate that a potentially 'causal' SCN1A variant is acting against a genomic background that may influence the expressed phenotype, we performed a set of post hoc analyses.We estimated the same three PRS across the Dravet syndrome and GEL SCN1A control cohorts.Differences in the PRS between cohorts were calculated as above.There is evidence that the most significantly associated SNP from the epilepsy GWAS may exert regulatory control over SCN1A 22 and, therefore, may influence the outcome of PRS for epilepsy in Dravet syndrome.Therefore, we also performed a localized PRS for epilepsy, intelligence and longevity, where we separated out from the GWAS of common epilepsies the genome-wide significant SNPs which mapped to 2q24.3 and corresponded to the SCN1A-related locus.Although the 2q24.3signal consisted of two independent subsignals, as shown in 2018 by the ILAE Consortium on Complex Epilepsies, 22 the insufficient number of genome-wide significant SNPs corresponding to the two subsignals made performing separate PRS analyses for the two signals impossible; therefore, the genomewide significant 2q24.3SNPs across the two regions were considered as a single SCN1A-related signal.Localized PRS for epilepsy, intelligence and longevity were performed both for only the 2q24.3SNPs and excluding the 2q24.3SNPs and compared across the three cohorts.

Data and code availability
The data can be requested by emailing the corresponding author.Data will be shared with bona fide researchers after approval of proposals with signed data access agreements as required by, and subject to, institutional and national regulations.
No bespoke code was used for this study.All code used in the manuscript is in the public domain already and has been appropriately referenced.
All pre-identified SCN1A variants were validated in the WGS data.Across the 34 individuals, 34 unique SCN1A variants were identified including one whole gene deletion.Details of the SCN1A variants can be found in Fig. 2, Supplementary material 11, and Supplementary Table 1.The variant distribution is comparable to published cohorts of individuals with SCN1A-related syndromes. 4,29,30No obvious association between variant class (i.e.missense or null) and specific phenotypes was observed (Supplementary Table 1).In addition, divergent phenotypes were seen in two unrelated individuals (Cases 1-105287 and 1-105683) who shared the same SCN1A variant (Supplementary Table 6).The WGS mean read coverage of the SCN1A gene region across the samples was 43.5 (excluding the SCN1A gene deletion).Visual inspection of the aligned reads using IGV showed an average alternate allele fraction of the known pathogenic SCN1A variants of 47.81%, confirming heterozygosity (excluding the homozygous SCN1A variant and whole gene deletion).None of the individuals showed evidence for mosaicism of the pathogenic SCN1A variant (P-value > 0.05; chi-squared test) (Supplementary Table 1 and Supplementary material 12).
We explored whether particular differences between ultra-rare SCN1A missense variants identified in the Dravet syndrome and GEL SCN1A control cohorts might explain differences in phenotype between these groups.No difference in the SCN1A-encoded variant residue location within the protein sequence was seen between missense variants identified in the Dravet syndrome cohort compared with the GEL SCN1A control cohort (Supplementary Table 1, Supplementary Table 4 and Supplementary material 13).Five GEL SCN1A controls carried SCN1A missense variants that have previously been reported in association with epilepsy syndromes, including Dravet syndrome, [31][32][33][34][35] or sudden unexpected death 36,37 (Supplementary Table 4).

Collapsing analyses of rare variants: enrichment of rare variants in Dravet syndrome
All individuals with Dravet syndrome were first assessed for the presence of additional rare variants, meeting a frequency cut-off of ≤0.0005 in gnomAD, across 190 epilepsy-related genes: 95 additional rare variants across 59 epilepsy-related genes were identified (Supplementary Table 7).Individuals had a median of 3 (range 0-7; interquartile range 2-3) additional rare variants (Supplementary Table 1).
To evaluate if individuals with Dravet syndrome harbour a higher burden of additional rare variants compared to the control cohorts, we performed gene-based and gene-set collapsing analyses for rare variants across 190 epilepsy-related genes, excluding SCN1A. 13,15Each gene was considered individually for the genebased analysis, while all 190 genes were considered as a set for the gene-set collapsing analysis.In the gene-set collapsing analysis, there was an enrichment (P = 0.0006) of rare variants in epilepsy-related genes in Dravet syndrome (78 qualifying rare variants in 28 cases; 2.78 variants per individual) compared to the GEL Epilepsy controls (1251 qualifying rare variants in 772 cases; 1.62 variants per individual), in concordance with a previous study reporting an excess of rare variants in (different but overlapping) epilepsy-related genes in individuals with Dravet syndrome. 38he gene-based collapsing analyses suggested a higher rare variant burden in the genes EHMT1, CHD2, FLNA, TSC1, PRICKLE1, SETBP1, NRXN1, SPTAN1 and ARID1B (P < 0.05) in Dravet syndrome compared to GEL Epilepsy controls (Supplementary Fig. 11A), but after correction for multiple comparisons, none of the adjusted P-values were significant.Of the 78 rare variants identified in these individuals with Dravet syndrome, a significant proportion (11/78 variants; 14.10%) overlapped with the 1251 rare variants identified in the GEL Epilepsy controls (P = 0.0001, Fisher's exact test).The results of the collapsing analysis for rare variants across 153 autosomal genes showed the same direction of enrichment as in the main analysis for rare variants across all 190 genes (Supplementary material 14).Though we investigated whether the observed variant enrichment in Dravet syndrome was driven by individuals with missense SCN1A variants but were underpowered to formally report this outcome (Supplementary material 15 and 16).

Rare variants in additional epilepsy-related genes: blended phenotypes and phenotypic heterogeneity
Across all individuals with Dravet syndrome, 51 rare variants in 38 epilepsy-related genes met pre-specified 'potential clinical relevance' criteria and underwent a detailed phenotype-genotype review (Supplementary Table 7).Five variants across four epilepsy-related genes (DEPDC5, CHD2, SCN8A and IQSEC2), all VUS by ACMG-AMP criteria alone, were considered to offer an independent molecular diagnosis, alongside the known SCN1A variant, resulting in blended phenotypes including features of both Dravet syndrome and the additional epilepsy-related genetic disorder.Parental samples were not available for these five adults, so we were unable to determine if the additional variants were de novo.For each of the five individuals, the variant and phenotype are discussed in detail (see Case 1 below and Supplementary material 17).
The identified DEPDC5 missense variant replaces a highly conserved alanine with threonine at codon 1395 of the DEPDC5 protein (Fig. 3B and C), with a Genomic Evolutionary Rate Profiling (GERP) score of 4.1, indicating the site is under evolutionary constraint. 40omputational evidence (SIFT, PolyPhen-2, MutationTaster) suggests the variant is damaging (Supplementary Table 7).2][43][44] This variant is encountered in seven individuals in gnomAD, corresponding to an allele frequency of 0.00005, considered to be within the pathogenic range, 45 and is absent from an ancestry-matched population database (n = 800). 46The penetrance of DEPDC5-related epilepsies is estimated to be around 60%, 47 and therefore the presence of this variant at low numbers within a population database would not be unexpected.This variant is considered a VUS according to a classification framework specifically adapted to GATOR1 genes, 48 by ACMG-AMP criteria, and reported as a VUS in ClinVar.To further explore its potential pathogenicity, in silico modelling was undertaken.Ala1395 lies at an internal inter-domain interface between the N-terminal, Structural Axis for Binding Arrangement (SABA) and C-terminal domains of DEPDC5 (domains as defined by Shen et al. 39 ), in close proximity to residues within those domains (Fig. 3D-G and Supplementary Fig. 12A-C).The effect of the variant was examined in both published structures for DEPDC5, protein data bank (PDB) 6ces (GATOR1 complex bound to Rag GTPases) and 6cet (GATOR1 complex alone), with similar, though not identical, results (for details, see Fig. 3H, Supplementary Fig. 12D and Supplementary material 18).In summary, the Ala1395Thr variant has a deleterious impact either on the folding and/or stability of DEPDC5, or impairs the ability of the GATOR1 complex to respond to Rag GTPases, in both cases likely leading to loss of function, the most commonly recognized mechanism of disease causation associated with DEPDC5 variants.
FCD is a malformation of cortical development.We explored the potential contribution of the SCN1A and DEPDC5 variants to the FCD by examining the dynamic expression patterns of those genes in the human temporal neocortex.FCD is thought to arise at 8-20 weeks post-conception, 49 the time frame in which DEPDC5 has a peak in expression; conversely, at this time expression of SCN1A is minimal (Supplementary Fig. 13 and Supplementary material 19).Therefore, the variant in DEPDC5 is temporally more likely to be causative of the FCD, in keeping with known consequences of DEPDC5 loss of function variants. 44,50However, we acknowledge that this finding is an association only, that is, we do not know and cannot establish when the FCD arose in the individual.2][53][54][55] To our knowledge, in these reports, only SCN1A sequencing was undertaken.
Overall, in the context of the visualized FCD, concordant electroclinical onset for many of his seizures, the in silico analysis and the temporal expression, we consider this variant to likely be contributory, thus potentially responsible for generating a blended phenotype in this individual.To confirm this finding a full exploration with model systems would be required.

PRS analyses
In Dravet syndrome, phenotypic heterogeneity encompasses many elements, including seizure severity and type, degree of intellectual disability, risk of sudden unexpected death in epilepsy (SUDEP) and comorbidities.Common genetic variation that confers risks for these traits may influence the phenotypic expression.We used two PRS analyses to explore key characteristics of Dravet syndrome for which there is known phenotypic heterogeneity: 'epilepsy' and 'intelligence'.In addition, recognizing that our adult Dravet syndrome cohort represents self-selected survivors, we also performed a PRS for 'longevity'.All PRS were performed on individuals of European ancestry only.

PRS for intelligence: common genetic variation may influence severity of intellectual disability in Dravet syndrome
7][58] Identical SCN1A variants can present with a range of cognitive phenotypes even within families. 591][62] We hypothesized that the common variant load for intelligence would be lower in individuals with Dravet syndrome compared with GEL Epilepsy and GEL controls.PRS for intelligence was significantly lower in the Dravet syndrome cohort than in GEL Epilepsy (adjusted P = 0.0024, at PT = 10 −4 , Tukey's test), and GEL controls (adjusted P = 0.003, at PT = 10 −4 , Tukey's test).There was no significant difference in the intelligence PRS between GEL epilepsy and GEL controls (adjusted P = 0.69, at PT = 10 −4 , Tukey's test) (Fig. 4A, Supplementary material 9, and Supplementary Figs 4 and 5).The intelligence PRS explained approximately 3% (R 2 = 0.03) of the total phenotypic variance in the Dravet syndrome group (derived from PRSice; Supplementary Fig. 6A).

PRS for longevity: common genetic variation may contribute to survival in Dravet syndrome
An estimated 10-20% of children with Dravet syndrome die before reaching adulthood, mostly due to SUDEP and status  Quantitative results are given in Supplementary material 18. Analysis of DEPDC5 from PDB 6cet is shown in Supplementary Fig. 1 epilepticus. 63,64We hypothesized that the longevity PRS would be higher in this cohort of individuals with Dravet syndrome who have survived into adulthood (mean age 32.5 years), especially as many had received a late diagnosis and had unknowingly had what in retrospect was suboptimal antiseizure medication (e.g.sodium channel-blocking medications) (Supplementary Table 1).PRS for longevity was significantly higher in the Dravet syndrome cohort than in GEL Epilepsy controls (adjusted P = 0.011, at PT = 10 −2 , Tukey's test), and higher than, but not significant, in GEL controls (adjusted P = 0.024, at PT = 10 −2 , Tukey's test).No significant difference was seen in the longevity PRS comparing GEL controls with GEL Epilepsy controls (adjusted P = 0.68, at PT = 10 −2 , Tukey's test) (Fig. 4B, Supplementary material 9, and Supplementary Figs 7 and  8).The longevity PRS explained around 2% (R 2 = 0.02) of the total phenotypic variance in the Dravet syndrome cohort (Supplementary Fig. 6B).

PRS for epilepsy: no common genetic variant contribution to the epilepsy phenotype in Dravet syndrome
Variants in SCN1A are associated with a spectrum of disorders in which the seizure phenotype is variable, from simple, self-remitting febrile seizures at the mild end, to drug-resistant epilepsy in people with Dravet syndrome at the severe end.Even amongst family members segregating one pathogenic SCN1A variant, the severity of the seizure phenotype can be wide-ranging, suggesting a contribution of additional genetic variation to the phenotype. 65Therefore, we hypothesized that the PRS for epilepsy would be higher in individuals with Dravet syndrome compared to GEL epilepsy and GEL controls.The epilepsy PRS was higher in the Dravet syndrome cohort compared with the GEL epilepsy and GEL controls, although this did not reach statistical significance (adjusted P = 0.89, at PT = 10 −2 , and adjusted P = 0.11, at PT = 10 −2 , Tukey's test, respectively).As expected, the epilepsy PRS was significantly higher in GEL epilepsy compared with GEL controls (adjusted P < 2.22 × 10 −16 , at PT = 10 −2 , Tukey's test) (Fig. 4C, Supplementary material 9, and Supplementary Figs 9 and 10).The epilepsy PRS explained around 0.05% (R 2 = 0.0005) of the total phenotypic variance in the Dravet syndrome cohort (Supplementary Fig. 6C).

Variation in SCN1A does not influence difference in PRS for intelligence and longevity
To further investigate the influence of SCN1A-related common variation on the PRS results, we selected the genome-wide Figure 4 PRS applied across the cohorts.(A) Polygenic risk score (PRS) for intelligence was lower in the Dravet syndrome cohort than in GEL Epilepsy (adjusted P = 0.0024) and GEL control cohorts (adjusted P = 0.003).The difference between GEL Epilepsy and GEL controls was not significant (adjusted P = 0.69).(B) PRS for longevity was significantly higher in the Dravet syndrome cohort than in GEL Epilepsy controls (adjusted P = 0.011), and higher than, but not significant, in GEL controls (adjusted P = 0.024) and not significantly different in GEL Epilepsy controls compared to GEL controls (adjusted P = 0.68).(C) PRS for epilepsy was not significantly different in the Dravet syndrome cohort compared with the GEL controls (adjusted P = 0.89) and GEL Epilepsy controls (adjusted P = 0.11).PRS for epilepsy was significantly higher in the GEL Epilepsy controls than in the GEL controls (adjusted P < 2.22 ×10 −16 ).The per-PRS P-values shown in the graphics are estimated using a post hoc multiple pairwise comparison (Tukey's test).As multiple PRS analyses were performed, the final adjusted P-value significance threshold was set to α = 0.05/3.GEL = Genomics England.
significant SNPs from the largest published GWAS of common epilepsies, which mapped to 2q24.3, corresponding to the SCN1A-related locus. 22We then performed a localized PRS for intelligence, longevity and epilepsy first excluding the 2q24.3SNPs, and then evaluating only the 2q24.3SNPs. 22Exclusion of the SCN1A signal did not modify the findings from the full PRS analysis, confirming that common variation in SCN1A is not driving the lower PRS for intelligence and higher PRS for longevity in the Dravet syndrome cohort compared with GEL Epilepsy and GEL control cohorts (Supplementary Fig. 14).PRS performed considering only the 2q24.3SCN1A-related SNPs did not show a significant difference across the cohorts, further supporting the finding that the SCN1A signal is not driving differences in PRS (Supplementary Fig. 15).

PRS and burden analyses of GEL SCN1A control cohort
To further evaluate the hypothesis that additional rare and common genetic variation may be necessary for the Dravet syndrome phenotype in some individuals with SCN1A variants, a post hoc exploration with PRS and burden analysis was undertaken, comparing individuals with Dravet syndrome with a GEL SCN1A control cohort composed of 45 GEL probands with unique SCN1A missense variants, but without epilepsy (Supplementary Table 4).Five GEL SCN1A controls carried unique SCN1A variants that have previously been reported in association with epilepsy syndromes [31][32][33][34][35] or sudden unexpected death 36 (Supplementary Table 4).
PRS for intelligence was lower but not significant (adjusted P = 0.033, at PT = 10 −4 , Tukey's test) (Fig. 5A), PRS for longevity was higher but not significant (adjusted P = 0.049, at PT = 10 −2 , Tukey's test) (Fig. 5B), and PRS for epilepsy was higher but not significant (adjusted P = 0.28, at PT = 10 −1 , Tukey's test) in the Dravet syndrome cohort compared with the GEL SCN1A controls (Fig. 5C).We also compared PRS for intelligence, longevity, and epilepsy between GEL SCN1A controls and the 13 Dravet syndrome cases with SCN1A missense variants.No significant difference was identified, though the direction of effect was maintained in comparison to the main analysis (Supplementary Fig. 16).
The gene-set collapsing analysis revealed an enrichment (P = 0.010) of rare variants in Dravet syndrome (78 variants in 28 individuals; 2.78 variants per individual) compared with GEL SCN1A controls (81 variants in 45 individuals; 1.8 variants per individual).None of the variants identified in Dravet syndrome overlapped with variants in the GEL SCN1A controls.A gene-based collapsing analysis highlighted an increased variant burden in CHD2, FLNA and TSC1 (P < 0.05) in Dravet syndrome compared with GEL SCN1A controls (Supplementary Fig. 11B) that was not significant after correction for multiple comparisons.

Discussion
Dravet syndrome is the archetypal DEE and amongst the most common of the rare epilepsies. 1,4Understanding of Dravet syndrome pathophysiology is amongst the most advanced for any DEE, reflected in the range of targeted therapies now in development. 66- 68The core phenotype is sufficiently distinct that the diagnosis is usually made clinically, followed by genetic testing anticipating a causal SCN1A variant, reflecting the very strong association between phenotype and causal gene.Nevertheless, the currently understood full phenotypic spectrum of Dravet syndrome is very broad, to the extent that in the absence of the telling early clinical history, the diagnosis may be missed clinically, especially in adulthood, and only considered on revelation of a putatively pathogenic SCN1A variant. 69Moreover, even given the distinct core phenotype, there is marked phenotypic heterogeneity within the syndrome, 30 which is not fully explained by differences between causal pathogenic variants, 29,70 and unexplained heterogeneity (not always due to mosaicism) within families segregating one pathogenic variant 65 and between unrelated individuals carrying the same variant. 71'Incomplete penetrance' and 'variable expressivity' are useful operational constructs in clinical practice to accommodate such heterogeneity.As with the concept of a 'syndrome', the undoubted utility of the terms 'penetrance' and 'expressivity' presumably reflects their basis in biology and pathophysiology.Some of the heterogeneity captured by these terms is probably due to genetic variation beyond the causal SCN1A variant.Digenic, oligogenic, polygenic, dual molecular diagnoses, mutational burden and double-hit contributions to disease phenotypes are well established as concepts. 8Discovering real examples in epilepsy is complicated both by the many syndromes and conditions that constitute this umbrella term, and by the known common variant contribution to the epilepsies overall.Controlling for the main genetic contributor of a genetic condition can allow additional genetic contributions to the phenotype to be discovered, as has been shown for example in Huntington's disease. 72,73Here, we adopted the same approach to Dravet syndrome, exploring WGS from a small group of adults with Dravet syndrome due to variation in SCN1A.We show that in clinically-distinct cases of Dravet syndrome, with a known SCN1A variant (classified as pathogenic or likely pathogenic in 33/34 cases, and published as pathogenic in the remaining case 74 ), there are examples of blended phenotypes, an excess of rare variants in epilepsy-related genes, and polygenic contributions to the overall phenotype, with additional evidence for genomic resilience (significantly elevated PRS for longevity).We show that beyond the causal coding or genic SCN1A variant, enrichment of rare variants in epilepsy-related genes and common variation in both SCN1A and across the genome are present and may have an impact.The presence of two disease-causing rare variants can lead to blended phenotypes, as shown by the presence of symptomatic FCD and a DEPDC5 variant in one individual with a clear Dravet syndrome phenotype due to a causal variant in SCN1A, with additional examples in other genes (CHD2, IQSEC2 and SCN8A).PRS analyses demonstrate that the causal SCN1A variant is acting against particular backgrounds.The effect size (as demonstrated by the explained variance) is limited, a common observation in studies of polygenic risk using current tools.However, evidence shows that the polygenic background may have a more substantial and clinically relevant effect in individuals with a monogenic disease, 75,76 demonstrating the principle that the rest of the genome is not inert in monogenic epilepsies, as recently demonstrated in unselected DEEs. 77or example, in two unrelated individuals with Dravet syndrome from this cohort, who share the same SCN1A splicing variant, the milder seizure and cognitive phenotype in one may in small part be explained by their lower epilepsy, and higher intelligence, PRS, respectively (Supplementary Table 6), demonstrating how a more (or less) favourable genetic background may contribute to explaining intra-familial and variant-specific phenotypic heterogeneity, and have bearing on our understanding of disease biology in 'monogenic' epilepsies.Of particular interest, the significantly lowered PRS for intelligence in our cohort could imply that even with symptomatic treatment leading to seizure freedom, or with disease-modifying treatment increasing SCN1A expression, the full phenotype of Dravet syndrome may not be entirely reversible.All these additional rare and common variants are obviously present independently of the observed SCN1A variant.Our results demonstrate that there is value in exploring additional genomic variation even when a 'causal', plausible and compatible pathogenic variant is identified, but clearly challenges remain in such work.Gathering and sequencing a cohort large enough to explore additional genomic variation, such as SCN1A-independent common (for example, through a genome-wide SNP-based association study) and rare variation (for example, through gene burden testing) is challenging.Functional validation for multiple variants will be complex, especially when, in most cases, there is no functional validation in clinical practice for the SCN1A variant itself found in an individual with Dravet syndrome: individual-based induced programmable stem cells and organoids may offer a way forward. 37ore tools are being developed that will allow integration and joint analysis of the contributions of different types of variation (e.g.category-wise association studies), but many potentially useful existing tools, especially those devised for clinical application, such as the ACMG-AMP system, are not intended to be used for additional variants 21 : our mindset is still largely centred on monogenic causation.
Nevertheless, we demonstrate that pathogenic variants in SCN1A do not necessarily act alone to produce the final phenotype: SCN1A may be the gene of major effect in Dravet syndrome, but it is not always the only gene, or only variant, of relevance.Moreover, Dravet syndrome-causing pathogenic variants may need to act against a broadly compromised genomic background (with, for example, a lower PRS for intelligence) to generate the full Dravet syndrome phenotype, whilst on the other hand genomic resilience may ameliorate some serious outcomes, such as premature mortality in Dravet syndrome, as shown by the elevated PRS for longevity in our adult Dravet syndrome survivors, most of whom had received a diagnosis in adulthood, and had been exposed to contraindicated medication.That a causal SCN1A variant inevitably acts within the context of the rest of the genome, some variation within which is relevant to the final phenotype, is perhaps unsurprising, but has not been demonstrated across a range of SCN1A variants before, and has not been addressed using the range of variation that can be examined using WGS data.Such work may help define the true phenotypic breadth of DS and other 'monogenic' conditions, and constrain the often bewildering expansion of phenotype in any given condition.Finally, the revelation of additional influential genomic variation in individual cases may have relevance to individual prognostication, and to treatments currently in development (e.g.gene-based therapies), informing realistic outcomes to be expected from new and existing treatments, and point the way to novel treatments, for example by using information from genomic variants in individuals with mild phenotypes to generate therapies to lessen severity in those with more severe phenotypes.
There are limitations to this study, primarily the limited size of the cohort, the cohort only consisting of adults and the lack of experimental validation using appropriate model systems.Despite these limitations, the results suggest that there may be occasions when stopping at the first plausible causal variant is premature, 8 with additional biological information of value identifiable by more extensive interrogation of the rest of an individual's genome.Non-genomic factors will undoubtedly also modulate phenotype, but genomic variation may contribute more than is currently believed.

Figure 1
Figure 1 Method for selection of variants in epilepsy-related genes.Method for selection of variants in epilepsy-related genes with 'potential clinical relevance' that may contribute to blended phenotypes.GEL = Genomics England.

Figure 2
Figure 2 Distribution of SCN1A variants found in the Dravet syndrome cohort.A schematic diagram of the SCN1A gene.Exons are indicated by vertical black boxes (1-29) and introns by the horizontal black line (not to scale).Missense (purple), splicing (dark blue), frameshift insertion (light blue), frameshift deletion (green) and stop-gain (red) variants are shown.The whole gene deletion is not shown.Variants are shown according to the NM_001165963.4reference sequence.

Figure 3 FCD
Figure 3 FCD and details of DEPDC5 variant.(A) Brain MRI showing FCD.Coronal T 1 -weighted brain MRI from Case 1-102398, with DEPDC5 variant NM_001242896.3:c.G4183A:p.A1395T, showing left temporal lobe FCD (right of patient is on the left of the image, following radiological convention), with blurred grey-white interface and cortical thickening apparent in the left temporal lobe across several consecutive slices.(B) MetaDome map of regional constraint in DEPDC5.Grey bar below the graph represents the protein, pink bars showing Pfam domains: PF12257, Vacuolar membrane-associated protein Iml1 domain; PF00610, Domain found in Dishevelled, Egl-10, and Pleckstrin (DEP); A1395 is marked by a vertical green line, with a reported tolerance score of 0.28 ('intolerant').(C) VarSite sequence logo for DEPDC5 residues 1375-1414, based on alignment of structural homologues; below the logo is the sequence of DEPDC5 itself, with A1395 boxed; sequence conservation score for this residue was 0.92 [range 0 (low)-1 (high)]; alanine was observed at this position in 31/33 aligned sequences.(D) Structure of the GATOR1-Rag GTPases complex and context of DEPDC5 Ala1395.PDB 6ces, the structure of the heterotrimeric GATOR1 complex (DEPDC5:NPRL2:NPRL3) bound to RagA and RagC GTPases; protein surfaces shown by colour as indicated (except DEPDC5, shown as a ribbon and coloured by structural domains as annotated by Shen et al. 39 : bright green = N-terminal domain (NTD) (residues 38-165); cyan = SABA domain (166-425); orange = steric hindrance for enhancement of nucleotidase activity (SHEN) domain (721-1010); dark green = DEP domain (1175-1270); violet = C-terminal domain (CTD) (1271-1600); Ala1395 is pink with sidechain atoms shown as spheres.(E and F) Ala1395 lies at an inter-domain interface in DEPDC5.The figure shows selected residues of DEPDC5 from PDB 6ces (chain D); residues of the NTD, SABA domain and CTD are shown as separate surfaces; residues of the SHEN domain and DEP domain are shown as ribbons.F shows the same structure as E with SHEN and DEP domains removed; residues Tyr108 (bright green), Phe326 (blue) and Ala1395 (rose pink) lie in close proximity at a three-way interface between the NTD, SABA and CTD.(G) Enlarged image of the DEPDC5 structure (PDB 6ces, chain D) as in E and F, zoomed to show detail around the three-way interface between the NTD, SABA and CTD; (H) The Ala1395Thr substitution results in reduced space at the inter-domain interface in 6cesD.This figure shows the same structure as G after introduction of the Ala1395Thr variant by in silico mutagenesis.Quantitative results are given in Supplementary material 18. Analysis of DEPDC5 from PDB 6cet is shown in Supplementary Fig.1

Figure 5
Figure5PRS applied across the GEL SCN1A control and Dravet syndrome cohorts.(A) Polygenic risk score (PRS) for intelligence was lower, but not significant, in the Dravet syndrome cohort than in GEL SCN1A controls (adjusted P = 0.033).(B) PRS for longevity was higher, but not significant, in the Dravet syndrome cohort than in GEL SCN1A controls (adjusted P = 0.049).(C) PRS for epilepsy was not significantly different between the Dravet syndrome cohort and GEL SCN1A controls (adjusted P = 0.28).Black circles = individuals from the GEL SCN1A control cohort with variants previously reported to be associated with disease.The per-PRS P-values shown in the graphics are estimated using a post hoc multiple pairwise comparison (Tukey's test).As multiple PRS analyses were performed, the adjusted P-value significance threshold was set to α = 0.05/3.GEL = Genomics England.