Summary

Introduction

In monogenic disease, incomplete penetrance and variable phenotype expressivity are typically observed. This is true even when individuals have inherited the same rare pathogenic variant of major effect. Epilepsy is no exception.

Whilst the overall genetic architecture of the epilepsies is complex, the number of over 1000 established monogenic causes, sets epilepsy apart from many other complex diseases.1^,2 Most epilepsy genes are associated with a broad phenotypic spectrum, even in families segregating a pathogenic variant.3 This creates confusion and concern for patients and clinicians, particularly with regards to disease prognostication.

Genetic modifiers have long been proposed to at least partially explain phenotypic heterogeneity of monogenic diseases. Therefore, modifiers are important to detect, not just for disease prognosis, but because they may reveal novel therapeutic avenues. Despite promising findings in murine models, limited progress towards identifying human modifier genes has been made.4 Recently, however, multiple studies of specific human diseases, such as coronary artery disease and cancer, highlighted that common polygenic background could modify the penetrance of monogenic variants.5^,6

In the epilepsies, interrogation of common variant burden has recently become feasible, through the application of polygenic risk scores (PRSs) derived from large genome-wide association studies (GWASs).7 The common focal and generalised epilepsies have largely been regarded as following complex inheritance, invoking a polygenic basis with possible environmental contributions. It was unsurprising, therefore, to find that epilepsy PRSs were enriched in independently sampled patients with common focal and generalised epilepsies.8 Intriguingly, however, cohorts with rare epilepsies of known or presumed monogenic cause, such as the developmental and epileptic encephalopathies (DEEs)9 and large families with epilepsy,10 were also enriched for epilepsy PRSs.

With common genetic variation playing a role in monogenic epilepsies, we sought to determine whether common variant burden could explain (at least partially) the elusive genetic modifiers influencing phenotype in the epilepsies. Specifically, we asked whether increased epilepsy polygenic risk is associated with epilepsy diagnosis (penetrance) and increasing phenotype severity (variable expressivity) in families with monogenic epilepsies.

Methods

Results

Polygenic risk in GEFS+ families

Within families, higher ‘all’ epilepsy PRSs were associated with epilepsy status. For every 1-SD increase in epilepsy PRS, relatives were approximately 40% more likely to be diagnosed with epilepsy than not (OR = 1.39, 95% CI 1.08, 1.80, p_adj = 0.040, Wald test), where the mean epilepsy PRS for unaffected relatives was −0.0094 (SD 0.96) compared to 0.31 (SD 1.05) for affected relatives (Fig. 1). We note, however, that despite the positive shift in PRS distribution for family members with epilepsy compared to those without, the distributions do largely overlap limiting current clinical application (ROC AUC = 74.9%, 95% CI 69.4%, 80.4%; Supplementary Figure S6).

Open in a separate window

Fig. 1

a: Normalised distributions of epilepsy PRS in GEFS+ cohort stratified by epilepsy status, b: Mean epilepsy PRS values with 95% confidence intervals for cohort stratified by epilepsy status relative to the control population sample (bold horizontal line at y = 0). Note, the unaffected relative group without epilepsy includes individuals with no seizures and those with febrile seizures only. The epilepsy PRS model was comprised of 39,074 SNPs. The logistic mixed-effects regression model used PRS as exposure variable, sex assigned at birth, cohort, RV, FS and the first five ancestry PCs as fixed-effects covariates, and family identifier as random-effects covariate. The reported p-value was adjusted for the four PRS models tested (see methods; Supplementary Table S1).

Relatives in the highest decile of ‘all’ epilepsy polygenic risk were 3.5 times more likely to have an epilepsy diagnosis compared to all other relatives (OR = 3.52, 95% CI 1.50, 8.22, p = 0.0037, Wald test; Supplementary Table S4a). When using the lowest decile of the ‘all’ epilepsy PRS as the reference compared to all other deciles of epilepsy polygenic risk, it was only the highest decile that was significantly enriched for relatives with epilepsy compared to those without (OR = 4.44, 95% CI 1.05, 18.7, p = 0.041, Wald test; Supplementary Table S4b).

When families were stratified by RV status, the positive epilepsy PRS association with epilepsy status was not significant; likely due to the smaller sample sizes (Supplementary Figure S7).

Pairwise phenotype severity analyses

Our cohort comprised a total of 2399 relative pairs. Of these, 784 (33%) relatives had concordant phenotypes (Supplementary Table S5) and a mean PRS difference of zero (Supplementary Figure S3).

For discordant pairs, relatives with the more severe phenotype had a higher epilepsy PRS compared to their less severely affected counterpart; the mean PRS difference was +0.19 (p_adj = 0.010, permutation-derived, Fig. 2a). The difference increased in an additive trend with degree of phenotype discordance (Fig. 2b). Whilst the mean PRS difference was zero for discordant relatives separated by only 1-grade of severity, the difference was +0.27 for relatives separated by 2-grades, and then +0.41 for relative pairs with a 3-grade difference in phenotype severity.

Open in a separate window

Fig. 2

Mean epilepsy PRS difference with 95% confidence intervals between pairs of relatives within GEFS+ families. a: Comparison between relative pairs with concordant versus discordant phenotypes. b: Comparison between relative pairs grouped by degree of phenotype discordance where 0-grade difference captures concordant phenotypes. Statistical comparisons were made using linear mixed models with kinship included as a fixed-effect covariate and family identifier (FID) as a random-effects covariate. Permutation derived p-values were adjusted for five test comparisons.

The positive trend did not hold for the group of relative pairs with greatest phenotypic difference (4-grade). This group comprised a much smaller sample size of only 36 relative pairs, from just 12 families, where one family member had no seizures whilst the other had a DEE (Supplementary Table S5). More than half of these phenotypically disparate pairs (20/36, 56%) are carriers of the SCN1B p.Cys121Trp founder variant. Notably, one of the SCN1B p.Cys121Trp relatives with DEE is bi-allelic with a known second RV in SCN1B (c.449-3C>A), thus representing a different genetic model of recessive inheritance.

Variant-specific analyses

We repeated the pairwise analyses in experiments comparing relatives carrying one of two RVs: SCN1B p.Cys121Trp and GABRG2 p.Arg82Gln. These two pathogenic variants accounted for the largest number of RV positive samples (n = 41 SCN1B p.Cys121Trp; n = 32 GABRG2 p.Arg82Gln) (Supplementary Table S3), thus providing sample sizes sufficient for independent analysis.

Relatives sharing the GABRG2 p.Arg82Gln variant (496 relative pairs; 128 concordant, 368 discordant) demonstrated a strong positive association between epilepsy PRS and phenotype severity (Fig. 3). On average, relatives with the more severe phenotype had a higher epilepsy PRS compared to those with a less severe phenotype (+0.54 mean PRS difference; p_adj = 0.0010, permutation-derived). Pairwise epilepsy PRS differences for the SCN1B p.Cys121Trp founder variant (820 pairs; 176 concordant, 644 discordant) showed a similar but non-significant trend (Table 2).

Open in a separate window

Fig. 3

Pedigree of family with the GABRG2 p.Arg82Gln rare variant. Modified from version published in Marini et al., 2003 to only include relatives with both the RV and DNA available for PRS generation. Individual pedigree identifiers have been kept the same for all relatives that are included in both papers. Eight of the eleven relatives coded as severity scale 4 (orange symbol) have childhood absence epilepsy (a generalised epilepsy syndrome). The exceptions are individuals II-6 and IV-9 with temporal lobe epilepsy (a focal epilepsy syndrome) and III-15 with unclassified epilepsy. Individual II-2 coded as severity scale 5 (red symbol) has a diagnosis of epilepsy with myoclonic-atonic seizures (EMAtS).

Table 2

Variant-specific mean pairwise PRS difference result comparisons for three epilepsy PRS models.

Rare-variant	PRS model	Mean PRS difference ±SD		Unadjusted p-values	Adjusted p-values
		Concordant pairs	Discordant pairs
GABRG2 p.Arg82Gln	‘All’ epilepsy PRS	0.00 ± 1.24	0.54 ± 1.38	0.00017	0.0010
	Generalised epilepsy PRS	0.00 ± 1.20	0.52 ± 1.33	0.00017	0.0010
	Focal epilepsy PRS	0.00 ± 1.16	0.25 ± 1.34	0.047	0.28
SCN1B p.Cys121Trp	‘All’ epilepsy PRS	0.00 ± 1.22	0.14 ± 1.34	0.10	0.60
	Generalised epilepsy PRS	0.00 ± 1.43	0.00 ± 1.46	0.46	1.0
	Focal epilepsy PRS	0.00 ± 1.65	0.27 ± 1.66	0.029	0.17

Open in a separate window

The GABRG2 p.Arg82Gln are based on 128 relative pairs concordant for phenotype and 368 discordant. The SCN1B p.Cys121Trp are based on 176 relative pairs concordant for phenotype and 644 discordant. Statistical comparisons were made using linear models with kinship included as a fixed-effect covariate. Permutation derived p-values were adjusted for six test comparisons.

Stratifying the variant-specific discordant pairs by degree of phenotype disparity, a similar trend to the overall cohort result was seen for SCN1B p.Cys121Trp where the positive association failed to hold for the most disparate relatives (i.e., 4-grade pairs). This was not surprising as SCN1B p.Cys121Trp relatives contributed the majority of pairs to this group. In contrast, however, for GABRG2 p.Arg82Gln pairs, the positive trend held across all four degrees of phenotypic disparity, albeit with even smaller numbers (Supplementary Figure S8).

Next, noting that the epilepsy phenotypes associated with the two RVs differed, we also performed variant-specific analyses using the generalised and focal epilepsy PRS models. GABRG2 p.Arg82Gln relatives were more likely to have a generalised epilepsy syndrome diagnosis (specifically childhood absence epilepsy)32 compared to relatives with the SCN1B p.Cys121Trp RV who were more likely to have a focal epilepsy syndrome diagnosis (specifically temporal lobe epilepsy)14 (Supplementary Table S6).

For the GABRG2 family, the generalised epilepsy PRS model performed almost as well as the ‘all’ epilepsy model, however, the focal PRS model did not. For the SCN1B relatives, we saw the opposite with the focal epilepsy PRS model performing better than the ‘all’ epilepsy model whilst the generalised model showed no effect (Table 2).

Discussion

Epilepsy is highly genetically and phenotypically heterogeneous.2 Epilepsy genes often have a broad phenotypic spectrum, ranging from self-limited epilepsies to DEEs, which may be due to different functional effects (e.g., loss-of-function versus gain-of-function).33 To account for allelic heterogeneity in the search for genetic modifiers, we took a familial approach, allowing us to control for the monogenic cause of major effect, at both the gene and variant level, to determine whether polygenic background plays a role in phenotypic variation.

We explored whether common variant burden explained phenotypic penetrance and severity in families with GEFS+ with a known or presumed pathogenic variant of major effect. Family members with higher epilepsy PRSs were more likely to have epilepsy and greater phenotypic severity. For every 1-SD increase in epilepsy PRS, relatives were approximately 40% more likely to have epilepsy; and in the top decile of the epilepsy PRS distribution, we observed a 4-fold enrichment of relatives with an epilepsy diagnosis. On average, relatives with a more severe phenotype had higher epilepsy PRSs compared to less severely affected relatives and the PRS difference increased with severity disparity.

We specifically targeted GEFS+, a syndrome characterised by phenotypic heterogeneity.3 GEFS+ families typically show autosomal dominant inheritance patterns, with several genes identified. Our cohort included two large families with a known cause; SCN1B p.Cys121Trp and GABRG2 p.Arg82Gln. Due to the large number of relatives in each, we also performed variant-specific analyses and found that higher ‘all’ epilepsy polygenic risk was associated with greater phenotype severity for carriers of GABRG2 p.Arg82Gln.

Interestingly, the pattern of epilepsy phenotypes within each family differed. Individuals with the SCN1B RV were more likely to have temporal lobe epilepsy (focal epilepsy), whilst the GABRG2 RV carriers were more likely to have childhood absence epilepsy (generalised epilepsy). This led us to explore whether common risk variants specific to generalised or focal epilepsy had differing phenotypic impacts. We found an enrichment for common focal epilepsy risk variants in SCN1B p.Cys121Trp carriers with more severe phenotypes. One exception was an individual who had a second RV in SCN1B contributing to a rare recessive SCN1B DEE phenotype.34 Taken together, common variants, specific to focal epilepsy risk, and rare variants are likely to contribute to the phenotypic heterogeneity in GEFS+ associated with SCN1B p.Cys121Trp.

The phenotypic heterogeneity in GEFS+ caused by GABRG2 p.Arg82Gln is strongly associated with common risk variants for both ‘all’ epilepsy and generalised epilepsy. This variant, previously annotated as p.Arg43Gln, segregates in a GEFS+ family with predominantly FS and childhood absence epilepsy.35 In 2003, an attempt to identify genetic modifiers by linkage analysis to explain this phenotypic heterogeneity revealed putative loci on chromosomes 10q21, 13p11-q12, 14q22-q23 and 15q11-q13,32 however, no specific variants were identified. A knock-in GABRG2 p.Arg82Gln mouse model (known as Gabrg2 R43Q) recapitulated the FS and childhood absence epilepsy phenotypes observed in the GEFS+ family.36 Subsequently the Gabrg2 R43Q mouse was back-crossed to strains with different seizure susceptibility. This revealed that only the knock-in mice on a DBA/2J “seizure susceptible” background developed spike-and-wave discharges; knock-in mice on a C57BL/6J “seizure resistant” background had no spontaneous spike-and-wave activity.37 This result mirrors our observation that the penetrance of the RV depends on epilepsy risk conferred by genetic background as captured by the ‘all’ epilepsy PRS.

The ‘all’ epilepsy PRS model utlised in this study incorporated >39,000 common SNPs. Determining which of these SNPs, or associated genes, are contributing to the observed genetic modification is an important, though challenging, next step. For GABRG2 p.Arg82Gln, we hypothesised an interplay between one or more PRS variants involved in the same biological pathway. Thus, our exploratory analysis focussed on SNPs in the epilepsy PRS associated with “GABA receptor activity” genes. This revealed a single SNP in GABRB3 as a top interaction candidate. The plausibility of a rare GABRB3 modifier variant (in linkage disequilibrium with rs12898904) for generalised epilepsy is supported by reports of rare GABRB3 variants being associated with absence epilepsies and more severe epilepsy syndromes.38^,39 Furthermore, GABRB3 was included in one of the four “childhood absence epilepsy linkage” regions (15q11-q13) reported in this family in 2003.32 Systematic analysis of gene–gene interaction or epistasis in complex disease has been limited, largely due to the substantial multiple-testing burden intrinsic to genome-wide epistasis screening.40 Leveraging our results to target specific SNPs will help navigate the multiple-testing burden of any hypothesis-free approach.

The correlation of higher epilepsy PRSs and more severe phenotype did not hold true in all cases. For example, two individuals with the GABRG2 p.Arg82Gln variant coupled with a high PRS only had FS (Fig. 3; IV-10 and IV-26). These ‘genomically resilient’ individuals may harbor protective variants and represent another avenue for future research. In one reported GEFS+ family, a second variant in SCN1A modified pathogenicity of the pathogenic SCN1A variant shared by all family members. The second rare variant, exclusively shared by unaffected carriers, counteracted the loss-of-function effect of the pathogenic variant.41 Mouse models have also revealed other potential modifier genes for GEFS+. Specifically, knock-in mice, heterozygous for an Scn1A GEFS+ missense variant (p.Arg1648His), demonstrated dramatic phenotypic variability when combined with variants in Scn2a, Kcnq2, and Scn8a.42

Our study has limitations. Overall, our sample size was small and we cannot rule out result bias from unmeasured confounders. As with most PRS analyses, we remained restricted to individuals of European ancestry. This is because the epilepsy GWAS from which the PRSs were derived comprised >90% European ancestry cases.7 Removal of non-European ancestry families further reduced our cohort's size. To counter this, we included families with both known and presumed RVs of major effect to ensure our sample size was as large as possible. As a result, the families without a known RV, may not be “monogenic”. In this case, the enrichment for epilepsy PRSs potentially reflects the contribution of common epilepsy risk variants to complex inheritance as opposed to acting as a genetic modifier on a gene of major effect. However, by investigating two specific RVs, our results are shown to support epilepsy PRSs playing a role in modifying RV penetrance and possibly epilepsy phenotype. Our differing results for SCN1B and GABRG2 suggests that certain genes and variants may be more susceptible to common variant genetic modification, however, these were also the two largest families and may not reflect effects in smaller families with different RVs. As such, it is important to note that epilepsy PRSs are not yet at the stage of clinical application, even for families with GEFS+; they remain underpowered for the purposes of disease prediction and for determining who will develop which type of epilepsy.

The mechanisms underlying phenotypic variability of most monogenic disorders remain poorly understood. Whilst epilepsy modifying genes have been hypothesised, they have been elusive. We provide evidence that, collectively, common epilepsy risk variants play a role, although specific variants have not been identified. This work shifts the current paradigm of epilepsy genetics – no rare pathogenic variant is likely to be acting alone. Interpretation of rare pathogenic variants in the context of polygenic background promises to provide an additional level of stratification for epilepsy risk in patients and their relatives. Furthermore, identifying specific modifier genes or pathways in the future has the potential to inform prognostic and genetic counselling, development of novel therapeutics and potentially disease prevention.

Contributors

K.L.O, I.E.S, C.A.E., B.E.G., S.F.B., M.B. contributed to the conception and design of the study. K.L.O., C.A.E., B.E.G. contributed to the curation, quality control and analysis of genetic data. K.L.O, I.E.S, B.E.G., C.A.E., S.F.B., Epi4K Consortium contributed to participant recruitment and phenotyping. K.L.O, C.A.E., M.B. performed and interpreted statistical analyses. K.L.O, C.A.E., M.B. have directly accessed and verified the underlying data reported in the manuscript. K.L.O. wrote the first manuscript draft. All co-authors contributed to manuscript revisions and read and approved the final version.

Data sharing statement

Data used in this study is not available due to privacy and ethical restrictions.

Declaration of interests

M.B. has received payment for thesis examination from University of Sydney, Australian National University, University of Melbourne; has received support for registration to the Genetics Society of AustralAsia and GeneMappers conferences; is/was a member of the Australian Academy of Health and Medical Sciences Reports Committee, the GenV Steering Committee, the Australian Academy of Health and Medical Sciences Australian Learned Academies Data Internetworking Network (ALADIN) Project Steering Committee, American Epilepsy Society Basic Sciences Committee, the Viertel Foundation Medical Advisory Board and a board member for the Australian Genome Research Facility. S.F.B has received educational grants from UCB Pharma, Eisai, SEER, Chiesi, LivaNova; consulting fees from Praxis Precision Medicines, Sequiris; speaker honoraria from Eisai, DeltaMed; serves as the Chief Medical Officer for the Epilepsy Foundation (Victoria); and has a patent for SCN1A testing held by Bionomics Inc and licensed to various diagnostic companies. I.E.S has served on scientific advisory boards for Bellberry Ltd, BioMarin, Chiesi, Eisai, Encoded Therapeutics, Garvan Institute of Medical Research, Knopp Biosciences, Longboard Pharmaceuticals, UCB, Takeda Pharmaceuticals; has received speaker honoraria from Akumentis Pharma, Biocodex, BioMarin, Chiesi, Eisai, GlaxoSmithKline, Liva Nova, Nutricia, Stoke Therapeutics, UCB, Zuellig Pharma; has received funding for travel from Biomarin, Eisai, GlaxoSmithKline, Stoke Therapeutics, UCB; has served as an investigator for Anavex Life Sciences, Cerebral Therapeutics, Cerecin Inc, Cereval Therapeutics, Eisai, Encoded Therapeutics, EpiMinder Inc, ES-Therapeutics, GW Pharmaceuticals, Marinus Pharmaceuticals, Neuren Pharmaceuticals, Neurocrine BioSciences, Ovid Therapeutics, Takeda Pharmaceuticals, UCB, Ultragenyx, Xenon Pharmaceuticals, Zogenix, Zynerba Pharmaceuticals; and has consulted for Biohaven Pharmaceuticals, Care Beyond Diagnosis, Cerecin Inc, Eisai, Epilepsy Consortium, Longboard Pharmaceuticals, UCB, Zynerba Pharmaceuticals; and is a Non-Executive Director of Bellberry Ltd and a Director of the Australian Academy of Health and Medical Sciences and the Australian Council of Learned Academies Ltd. She may accrue future revenue on pending patent WO2009/086591: Diagnostic And Therapeutic Methods For EFMR (Epilepsy And Mental Retardation Limited To Females); has a patent for SCN1A testing held by Bionomics Inc and licensed to various diagnostic companies; has a patent molecular diagnostic/theranostic target for benign familial infantile epilepsy (BFIE) [PRRT2] with royalties paid. C.A.E has received consulting fees from Epiminder Pty Ltd. The Epi4K Consortium was supported by an NINDS National Institute of Health Grant [U01NS077367].

Acknowledgements

Footnotes

Contributor Information

Epi4K Consortium:

References