Epilepsy, hippocampal sclerosis and febrile seizures linked by common genetic variation around SCN1A

Subjects

Patients were recruited during clinical appointments. MTLEHS was defined as in Wieser (2004). The diagnosis was made and/or reviewed by a consultant epileptologist who was part of this study, with access to history and investigation results. Patients with bilateral hippocampal sclerosis or dual pathology were excluded. One thousand and eighteen patients were included in the discovery stage and 959 patients in the replication. The number of patients by country is shown in Table 1, with further details in Supplementary Table 1. A history of presence or absence of febrile seizures was accepted only if contemporary medical records or a parental account was available; otherwise it was considered unknown, and not eligible for analysis. Population-based controls (n = 7552) were included in the discovery stage, and 3591 in the replication (Table 1 and Supplementary Table 1).

We also studied 542 individuals who had had febrile seizures but by the last follow-up had not had unprovoked seizures. These came from three groups: a German group; an Austrian group and the ALSPAC (Avon Longitudinal Study of Parents and Children) cohort, the latter followed to age 13 years (Supplementary material); MTLEHS after febrile seizures almost always develops by the age of 15 (Neligan et al., 2012). These cases were compared with 7387 control subjects from three relevant populations (Table 1). For the German and Austrian samples, the same controls as in the MTLEHS study were used.

To minimize population stratification, only individuals of white European ancestry were included. In the discovery stage, a combination of self-identified ancestry and EIGENSTRAT principal component methods was used to determine European ancestry. In the replication and febrile seizures analyses, only self-reported white individuals of European ancestry were included. More detailed ancestry data were available from all sources except Austria, allowing exclusion of individuals self-reported as coming from countries other than those where they were recruited.

Genotyping and quality control

In the discovery stage, all but the Austrian samples and Belgian controls comprised a subset of a previously described data set (Kasperaviciūte et al., 2010), genotyped on Illumina genome-wide genotyping chips, mostly on Illumina Human610-Quadv1/Human1-2M-DuoCustom. One hundred and fifty-seven Austrian patients and 332 controls were genotyped on Illumina HumanCNV370duo, and 285 Belgian controls were genotyped on Illumina HumanHap300 genotyping chips. Gender and relatedness checks were performed on all samples. The cluster plots of the top-associated single nucleotide polymorphisms were inspected manually. Details are given in Kasperaviciūte et al. (2010) and in the online Supplementary material. For replication analysis, several methods were used for genotyping.

Statistical analysis

In the discovery stage, genome-wide association analysis was performed using PLINK. Only single nucleotide polymorphisms present on both Illumina Human610-Quadv1 and Human1-2M-DuoCustom were analysed. In the discovery stage, we performed logistic regression using an additive model, including all significant EIGENSTRAT axes (assessed using the Tracy-Widom statistic with P < 0.05) as covariates. Only single nucleotide polymorphisms with minor allele frequency of ≥ 1% were analysed. Since the replication samples did not have genome-wide data available to calculate EIGENSTRAT axes, we performed stratified analysis using the Cochran-Mantel-Haenszel test for 2 × 2 × 8 stratified case-control subsamples deriving from eight different recruitment countries and self-identified ancestry, using R. The Woolf test was used to assess effect heterogeneity. Meta-analysis of discovery and replication studies was performed using the inverse variance-weighted fixed-effects model as implemented in the GWAMA software (Mägi and Morris, 2010). We considered an association to be genome-wide significant at P < 5 × 10⁻⁸.

To fine map the association signal in the discovery stage, we imputed single nucleotide polymorphisms in the 10 Mb region surrounding rs7587026. Imputation was performed using MINIMAC (Howie et al., 2012), and 1000 Genomes Project data (1000 Genomes Project Consortium et al., 2010) as the reference data set. Subsequent association analysis was performed using MACH2DAT (Li et al., 2010) using significant EIGENSTRAT axes as covariates.

Power calculations were performed using Genetic Power Calculator (Purcell et al., 2003).

Expression analysis

We tested association between genotypes of the two top single nucleotide polymorphisms rs7587026 and rs11692675 and SCN1A exons and gene expression in the middle temporal cortex (Brodmann areas 20 and 21) from 78 patients with MTLEHS who had undergone surgical resection, compared with 78 neurologically normal individuals from the MRC Sudden Death Brain and Tissue Bank. We specifically chose not to study the hippocampus to avoid confounding due to tissue changes such as cell loss and gliosis. All samples were randomly hybridized to Affymetrix Human Exon 1.0 ST arrays. Differential expression of SCN1A transcripts incorporating the ‘neonatal’ or ‘adult’ exon 5 form (5N or 5A exon, respectively), and expression of non-coding exons 1a and 1b (GenBank accession numbers DQ993522 and DQ993523, respectively) (Martin et al., 2007) in the 5’ region of SCN1A, were tested by quantitative RT-PCR as they are not covered by the array. Details are provided in the Supplementary material.

Further, we tested whether the associated single nucleotide polymorphisms have an effect on expression or splicing of any genes in the genome in post-mortem tissue of nine brain regions from 134 control individuals (Supplementary material).

Genome-wide association analyses

We performed a two-stage study. For discovery, we first investigated genome-wide association between all MTLEHS and 531 164 single nucleotide polymorphisms in 1018 MTLEHS cases and 7552 controls from seven populations of European descent (Table 1 and Supplementary Table 1). Using logistic regression analysis and correcting for population stratification, suggestive association emerged for three single nucleotide polymorphisms in a region of strong linkage disequilibrium on chromosome 2q24.3 encompassing SCN1A and other sodium channel genes (Supplementary Fig. 1). The most strongly associated single nucleotide polymorphism, rs11692675, is within intron 3 of the SCN1A full-length transcript variant (NM_001202435.1) {P = 5.26 × 10⁻⁸, odds ratio for G allele [OR(G)] 1.31, 95% confidence interval (CI) 1.19–1.44; Table 2}. Two other single nucleotide polymorphisms within SCN1A intron 1, had similarly low P-values: rs7587026 (r² = 0.806 with rs11692675 in CEU population based on 1000 Genomes data set), P = 1.19 × 10⁻⁷ [OR(A) = 1.31, 95% CI: 1.19–1.45]; and rs580041 (r² = 0.806 with rs11692675), P = 5.74 × 10⁻⁷ [OR(A) = 1.29, 95% CI: 1.17–1.43].

SCN1A encodes brain-expressed voltage-gated sodium channel type I, alpha subunit. It bears the largest number of known epilepsy-related mutations, some associated with febrile seizures (Oliva et al., 2012). The common SCN1A single nucleotide polymorphism rs3812718, affecting splicing (Heinzen et al., 2007), has also been associated with febrile seizures (Schlachter et al., 2009), though replication has failed (Petrovski et al., 2009). Retrospective studies show association between MTLE and febrile seizures (Pittau et al., 2009; O’Dell et al., 2012). Whether febrile seizures cause MTLEHS (Koyama et al., 2012) or whether pre-existing hippocampal abnormalities predispose to febrile seizures (Cendes, 2004), which may then also be injurious, is unknown. Clinical differences between patients with and without a history of febrile seizures suggest MTLEHS is heterogeneous (Thom et al., 2010).

This evidence motivated our pre-analysis collection of febrile seizure data, and our previous study of febrile seizures (Petrovski et al., 2009). We performed analysis of patients in the discovery cohort with a known history of presence of childhood febrile seizures (MTLEHS + FS, n = 341) (Table 2 and Fig. 1). The strongest association was for rs7587026, P = 2.64 × 10⁻⁸ [OR(A) = 1.59, 95% CI: 1.35–1.87] and rs580041, P = 8.91 × 10⁻⁷ [OR(A) = 1.56, 95% CI: 1.33–1.84], whereas the signal for rs11692675 was slightly weaker, P = 1.25 × 10⁻⁶ [OR(G) = 1.49, 95% CI: 1.27–1.75]. No association was seen in patients with MTLEHS without febrile seizures (MTLEHS−FS), despite similar sample size.

To refine the association signal, we performed regional imputation in the discovery data set of a 10 Mb region surrounding rs7587026 using the 1000 Genomes reference panel. Two single nucleotide polymorphisms had slightly lower P-values in the MTLEHS+FS analysis than the original single nucleotide polymorphisms [rs16851603 (P = 2.23 × 10⁻⁸) and rs3919196 (P = 2.26 × 10⁻⁸)], but neither were significantly stronger than the original associations, and these signals reflected regional linkage disequilibrium structure (Supplementary Fig. 2). No known functional variants in SCN1A, nor in other genes in the region, were in high linkage disequilibrium with rs7587026. The association signal is localized within one linkage disequilibrium block that also spans the promoter and 5’ UTR region of SCN1A (Supplementary Fig. 2).

Replication and combined analyses

We selected the two top single nucleotide polymorphisms, rs7587026 and rs11692675, for replication in an independent sample of 959 patients with MTLEHS, of whom 416 had MTLEHS+FS, and 3591 population-matched controls of European descent from eight populations (Table 1 and Supplementary Table 1). We did not study rs580041 because of its perfect linkage disequilibrium with rs7587026 in white Europeans (r² = 1). We detected an association between rs7587026 and MTLEHS + FS, P = 5.88 × 10⁻³ [OR(A) = 1.26, 95% CI: 1.07–1.48; Table 2]; this value remains significant at a revised alpha threshold of 6.3 × 10⁻³ after Bonferroni correction for multiple comparisons in the replication cohort. No association was detected for MTLEHS−FS.

Meta-analysis of the discovery and replication samples confirmed the association of the 2q24.3 locus with MTLEHS+FS at genome-wide significant level for rs7587026 [P_meta = 3.36 × 10⁻⁹, OR(A) = 1.42, 95% CI: 1.26–1.59]; the signal for rs11692675 did not reach genome-wide significance [P_meta = 4.78 × 10⁻⁶, OR(G) = 1.30, 95% CI: 1.16–1.46]. No significant heterogeneity in effect sizes was detected among different populations (Fig. 2, Woolf’s test for heterogeneity P = 0.45, see Supplementary Tables 7–10 for allele frequencies in all populations).

Febrile seizure analysis

To explore whether the observed association with rs7587026 is specific for MTLEHS+FS, or is specific for febrile seizures in general, we examined a total of 7387 controls and three data sets of patients (totalling 542) who had had febrile seizures but had not developed epilepsy by the time of the latest follow-up. It has been shown that almost all children who go on to develop any type of epilepsy after febrile seizures have done so by the age of 15 years (Neligan et al., 2012). Therefore, to address the specificity of the association for MTLEHS+FS rather than febrile seizures alone, the ideal febrile seizures cohort would have been followed to age 15 at least. Of the three data sets available to us, only one met this criterion closely. The ALSPAC prospective cohort, which has the most comprehensive phenotypic data of the three data sets, followed children to age 13: there was no association of rs7587026 with febrile seizures in 171 individuals who did not go on to develop epilepsy (Table 3) in comparison with 6443 controls from the same cohort. The two other cohorts of children with febrile seizures, from Austria [samples partially overlapping with those reported in Schlachter et al. (2009)] and Germany, were ascertained at a young age (to six years of age only) and had no follow-up to establish whether the children had febrile seizures only, or febrile seizures in the context of subsequent epilepsy including MTLEHS, and are therefore not best suited to address the question, but were examined as few cohorts overall are available. Bearing this key caveat in mind, in these two data sets there was an observed association of febrile seizures with rs7587026 (Table 3). The previously reported association in the Austrian population with an SCN1A functional splice-site single nucleotide polymorphism, rs3812718, was also seen in our Austrian sample [which is not unexpected as there is partial overlap of cases with those in the original report (Schlachter et al., 2009)] and was present in the German sample. The observed association of rs7587026 with febrile seizures disappeared in both Austrian and German data sets when analysis was conditioned on rs3812718 (P > 0.19; Table 3). Moreover, although the association of febrile seizures with rs3812718 may be thought to be of interest for pure febrile seizures alone, we note there is no association of rs3812718 with febrile seizures in the best characterized cohort, from ALSPAC (Table 3), nor in a published sample (Petrovski et al., 2009).

Thus, although other single nucleotide polymorphisms in or near SCN1A may predispose to pure febrile seizures, the signal we observed in MTLEHS+FS is very unlikely to be due to the history of febrile seizures alone. Moreover, no significant association was detected in a group of patients with other partial epilepsies with a history of febrile seizures [data set from Kasperaviciute et al. (2010); rs7587026, P = 0.24, OR(A) = 1.15, 95% CI: 0.91–1.45]. The sample for this analysis was smaller (177 patients; 7552 controls), but had 81% power to detect association of OR ≥ 1.42 (as seen in MTLEHS+FS group combined analysis) under 0.05 significance level. Collectively, we found no evidence that the MTLEHS+FS association was due to febrile seizures, or that it holds for all partial epilepsies with febrile seizures.

SCN1A expression in the human brain

The observed association could act by modulating SCN1A gene expression. The associated region harbours several alternative untranslated SCN1A exons (Martin et al., 2007; Nakayama et al., 2010). We did not detect association between rs7587026 and any protein-coding exon except one (see below) or total SCN1A expression, or with expression of untranslated 5’ exons 1a and 1b (Martin et al., 2007) (data not shown) in 78 patients and 78 control subjects.

The presence or absence of transcripts incorporating the ‘neonatal’ SCN1A exon 5 (‘5N’) was significantly different according to genotype of the two top single nucleotide polymorphisms (rs11692675 and rs7587026, P-values 1.08 × 10⁻⁹ and 1.17 × 10⁻⁶, respectively; Supplementary material). For rs11692675 and rs7587026, respectively, none and 1% of the individuals with the GG and AA genotype showed SCN1A transcripts in the neonatal form, compared with 83% and 81% with the genotype AA or CC. This alternative splicing event is influenced by rs3812718 (Heinzen et al., 2007). The association of alternative splicing with rs922224 (r² = 1 with rs3812718) was stronger, P = 2.33 × 10⁻³¹. The level of expression of SCN1A exon 5N was also significantly different according to genotype (P = 1.62 × 10⁻¹¹ for rs11692675, 2.70 × 10⁻⁶ for rs7587026, 7.40 × 10⁻³⁴ for rs922224). In conditional analyses including all three single nucleotide polymorphisms, only rs922224 remained significant (P = 1.08 × 10⁻²⁵). Finally, expression quantitative trait loci analyses for subsets of patients according to a known history of presence (n = 46) or absence (n = 27) of febrile seizures in childhood for rs11692675 or rs7587026 showed significant differences in the level of expression of 5N exon according to genotype in both MTLEHS+FS and MTLEHS−FS. Including both rs11692675 or rs11692675 and rs922224 in the regression models, only rs922224 remained significant in both MTLEHS+FS and MTLEHS−FS groups (Supplementary material).

We cannot exclude the possibility that rs7587026 (or another single nucleotide polymorphism in the high linkage disequilibrium region) may act as an additional splicing controller to rs3812718, but our data are consistent with rs7587026 having no solo effect on 5N splicing. We also did not detect any correlation using a significance level of P < 5 × 10⁻⁵ between rs7587026 and expression/splicing of any other genes across the genome (Supplementary material).