Epilepsy is a common neurological disease characterized with recurrent unprovoked seizures. As one of the most prevalent brain disorders, epilepsy imposes a huge economic burden on society and affects about 70 million people worldwide (Thijs et al., 2019). Accumulating evidence indicate that epilepsy has a strong genetic component (Kjeldsen et al., 2003; Speed et al., 2014; Koeleman, 2018). Twin studies showed that genetic factors (total heritability) account for about 80% of the liability to epilepsy (Famula et al., 1997; Kjeldsen et al., 2003). Recent genome-wide association study (GWAS) estimated that the single nucleotide polymorphism (SNP) heritability (the proportion of variance in liability that can be attributed to SNPs, h²_SNP) of genetic generalized epilepsy and focal epilepsy is about 32.1 and 9.2%, respectively (The International League Against Epilepsy Consortium on Complex Epilepsies, 2018), further indicating the pivotal role of common genetic variation in epilepsy. A recent study by Speed et al. (2014) also estimated that common variants collectively explain about 26% of phenotypic variation for all epilepsy. Despite the high heritability of epilepsy, only limited risk variants and loci have been identified by large-scale genetic studies to date (International League Against Epilepsy Consortium on Complex Epilepsies, 2014; The International League Against Epilepsy Consortium on Complex Epilepsies, 2018).

To further identify risk variants and to uncover the missing heritability of epilepsy, in this study, we report the largest genome-wide meta-analysis of epilepsy (N = 8,00,869 subjects) by integrating four large-scale GWASs of epilepsy. The first GWAS (14,534 cases and 24,218 controls) was from a recent study by the International League Against Epilepsy Consortium on Complex Epilepsies (ILAE Consortium) (The International League Against Epilepsy Consortium on Complex Epilepsies, 2018). The second GWAS was from the UK Biobank (Sudlow et al., 2015), genome-wide summary statistics of epilepsy in UK Biobank [5,087 epilepsy cases (Phecode:X345) and 3,95,209 controls] generated by the scalable and accurate implementation of generalized mixed model (SAIGE) (Zhou et al., 2018) were used in this study. The third GWAS was from a recent study by Ishigaki et al. (2020) (2,143 epilepsy cases and 2,10,310 controls). The fourth GWAS dataset was from the FINNGEN¹ [4,588 epilepsy cases (phenocode: G6_EPLEPSY) and 1,44,780 controls].

We firstly examined the genome-wide significant (GWS) loci in each GWAS. For epilepsy GWAS from the ILAE Consortium, two loci reached GWS level (of note, this dataset contains three subtypes and we only included GWAS of all epilepsy) (The International League Against Epilepsy Consortium on Complex Epilepsies, 2018). For epilepsy GWAS from UK Biobank, Japanese population, and FINNGEN, no loci showed GWS associations with epilepsy (Supplementary Figures 1–3). We then conducted a genome-wide meta-analysis by combining the genome-wide association results of the four studies. The QQ plot is shown in Supplementary Figure 4. The lambda_GC (λ_1,000, genomic control (GC) inflation lambda scaled for 1,000 cases and 1,000 controls) of our meta-analysis is 1.0027, indicating that the association signals were mainly driven by polynenicity rather than population structure. In this largest trans-ethnic meta-analysis (26,352 cases and 7,74,517 controls), we identified three GWS (p < 5 × 10^–8) risk loci for epilepsy (Figure 1 and Table 1). The risk locus on 2q24.3 has been reported in a previous GWAS (The International League Against Epilepsy Consortium on Complex Epilepsies, 2018). Of note, the most significant variant (lead SNP rs11890028) on 2q24.3 resides in the intron 7 of SCN1A, a well-characterized risk gene for epilepsy (Claes et al., 2001; Parihar and Ganesh, 2013; International League Against Epilepsy Consortium on Complex Epilepsies, 2014; Figure 2B). However, the risk loci on 7q21.11 (lead SNP rs11978015, p = 9.26 × 10^–9) and 8p23.1 (lead SNP rs28634186, p = 4.39 × 10^–8) are newly identified in the present study (Figures 2B,C). Of note, rs11978015 resides in upstream of GRM3 (Figure 2B), which encodes glutamate metabotropic receptor 3. GRM3 has pivotal roles in neurotransmission and is involved in most aspects of normal brain function (Blacker et al., 2017; Chaki, 2017; Jin et al., 2018; Neale and Olszewski, 2019; Joffe et al., 2021; Kellner et al., 2021). Another GWS risk variant rs28634186 is located in intergenic region and the nearest gene for rs28634186 is TNKS (Figure 2C). Genetic correlation between epilepsy and other diseases [using LD Hub (Zheng et al., 2017)]⁸ showed significant correlations with amyotrophic lateral sclerosis (ALS), schizophrenia, and bipolar disorder (Supplementary Table 1).

To identify the tissues and cell types that risk genes may exert their biological effects on epilepsy, we further performed tissue- and cell-type-specific enrichment analysis using MAGMA (see text footnote 5). As expected, the GWAS associations were significantly enriched in brain tissues (Figure 3A), with the highest enrichment in the cerebellar hemisphere and frontal cortex (Figure 3A). Cell-type-specific enrichment analysis showed significant enrichment of GWAS associations in telencephalon projecting inhibitory and excitatory neurons (Figure 3B).

To identify risk genes whose genetically regulated expression change are associated with epilepsy, we further conducted a TWAS by integrating genome-wide summary statistics of epilepsy (from meta-analysis) and SNP-gene expression weights from the PsychENCODE (N = 1,371).⁹ Three genes (TTC21B, RP11-375N15.2, and TNKS) showed transcriptome-wide significant (TWS) associations (Bonferroni corrected p < 0.05) with epilepsy (Figure 4), indicating that risk variants may confer epilepsy risk through regulating the expression of these genes.

Finally, to explore if the risk identified may be targeted for epilepsy treatment, we further examined the interactions between the identified risk genes and drugs using DGIdb (Freshour et al., 2020). Our analysis showed that SCN1A interacts with many drugs (Supplementary Table 2), suggesting this gene may be targeted as a potential therapeutic target. In addition, GRM3 and TNKS also show interactions with several drugs (Supplementary Tables 3, 4). These data suggest that these three genes may be targeted for epilepsy treatment.

In summary, we performed the largest meta-analysis of epilepsy GWAS in this study and identified two new risk loci for epilepsy. We also showed that genetic associations of epilepsy are enriched in brain tissues and telencephalon projecting inhibitory and excitatory neurons. Of note, we identified three risk genes (TTC21B, RP11-375N15.2, and TNKS) whose expression perturbation may have a role in epilepsy. Interestingly, TTC21B resides in the 2q24.3 locus, a region that contains the well-characterized epilepsy risk gene SCN1A (Claes et al., 2001, 2009; Meng et al., 2013; Parihar and Ganesh, 2013; International League Against Epilepsy Consortium on Complex Epilepsies, 2014; Haigh et al., 2021). Although the genome-wide association signal (Figure 2A) and previous studies (Claes et al., 2001, 2009; Meng et al., 2013; Parihar and Ganesh, 2013; International League Against Epilepsy Consortium on Complex Epilepsies, 2014; Haigh et al., 2021) have clearly showed that SCN1A represents the most possible causal gene for this risk locus, our TWAS suggested that TTC21B may also have a potential role in epilepsy (Figure 4). In fact, previous studies also have revealed the potential role of TTC21B in epilepsy (Mirza et al., 2017; The International League Against Epilepsy Consortium on Complex Epilepsies, 2018). More work is needed to elucidate the role of TTC21B in epilepsy.

We noticed a second peak which is not in linkage disequilibrium (LD) with the main signal at 2q24.3 (located at the right of the top hit) (Figure 2A), suggesting two independent genetic signals at this locus. We thus performed a conditional analysis using genome-wide complex trait analysis (GCTA) (Yang et al., 2011). Conditional analysis on the top hit SNP rs11890028 suggested that the second peak might not be an independent signal. The p-value of rs11896706 (located at the right of the top hit, Figure 2A) conditioned on rs11890028 (the top hit) is 5.01 × 10^–5. In addition, we also checked the LD pattern of genetic variants spanning this genomic region [using genotype data of Europeans from the 1,000 Genomes project (The 1000 Genomes Project Consortium, 2015)]. We found that the lead SNP (rs11890028) showed weak LD (r² = 0.13) with SNP rs11896706 in this region, suggesting that the second signal peak in this locus was likely due to LD between the lead SNP rs11890028 and nearby variants. More work is needed to further investigate if two independent GWAS signals at this locus.

Intriguingly, our findings suggested that GRM3 may be a potential risk gene for epilepsy. GRM3 encodes glutamate metabotropic receptor 3 (mGluR3), a member of the family of G protein-coupled receptors. As a major receptor of glutamate, postsynaptic GRM3 is crucial for mGluR3-dependent long-term depression (LTD) (Joffe et al., 2021) and cognitive function (Jin et al., 2018; Neale and Olszewski, 2019). The binding of glutamate to mGluR3 results in activation of G protein-coupled receptor, which in turn regulates gene transcription, release of neurotransmitter, neuron activity, and synaptic transmission (Blacker et al., 2017; Chaki, 2017; Kellner et al., 2021). In fact, a recent study has proposed that modulation of astrocyte glutamate uptake (and/or mGluR activation) may represent a potential therapeutic approach for epilepsy treatment (Peterson and Binder, 2020). These lines of evidence suggest that GRM3 may have a role in epilepsy.

Our GWAS meta-analysis suggested that 8p23.1 is a risk locus for epilepsy. To further explore the potential risk gene in this locus, we searched literatures and performed additional analysis. Based on the results of GWAS and TWAS, we speculated that TNKS (also known as PARP5A) may be the potential risk gene at this locus. First, most of the significant variants identified in our GWAS meta-analysis are located in genomic region (including gene body and upstream of TNKS) containing TNKS (Figure 2C). Second, our TWAS results suggested that TNKS is a gene whose genetically regulated expression may have a role in epilepsy (Figure 4). Third, expression analysis using single-cell RNA-seq data showed that TNKS is highly expressed in different cell types of the brain (Supplementary Figures 5, 6). Fourth, the biological function also supports TNKS as a potential risk gene. TNKS encodes tankyrases [members of the poly(ADP-ribose) polymerase (PARP) family] that are involved in the regulation of Wnt/beta-catenin signaling (Bao et al., 2012; Kulak et al., 2015; Yang et al., 2019) and PTEN (Li et al., 2015). Considering the important role of Wnt/beta-catenin signaling (Bengoa-Vergniory and Kypta, 2015; Arnes and Casas Tinto, 2017; Noelanders and Vleminckx, 2017; Bem et al., 2019) and PTEN (Endersby and Baker, 2008; Zhou and Parada, 2012; Spina Nagy et al., 2021) in the brain, it is possible that TNKS confers risk of epilepsy by regulating Wnt/beta-catenin signaling pathway. In fact, recent studies also have showed that TNKS modulates TDP-43, a protein with a central role in ALS and frontotemporal degeneration (FTD) (McGurk et al., 2020; Tanji et al., 2021), further suggesting the pivotal role of TNKS in the human brain. Finally, Wnt/β-catenin signaling was proposed to be a potential target for epilepsy therapy (Huang et al., 2015; Hodges and Lugo, 2018). Collectively, these evidence suggest that TNKS may represent the potential risk gene at 8p23.1. However, further genetic studies and functional characterization are needed to validate if TNKS is a risk gene for epilepsy.

It should be noted that biobank data as a resource for epilepsy GWAS has weakness and limitations. Compared with traditional GWAS (GWASs usually use well-phenotyped cohort), the subjects included in biobank data are usually based on electronic records or questionnaire, which may affect the accurate phenotyping of the included individuals. In addition, the number of cases included in biobank data is usually much smaller than controls (case-control imbalance), which influences the statistic power of biobank data. Finally, population relatedness or structure also needs to be carefully considered and controlled in biobank data.

In conclusion, our study not only provides new insights into genetic architecture of epilepsy but also prioritizes potential molecular targets (including GRM3 and TNKS) for development of new drugs and therapeutics for epilepsy.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s. Custom codes used (including Perl, R, and Unix) were used for data processing. These custom codes can be made available from the corresponding author upon request.

The studies involving human participants were reviewed and approved by the ILAE Consortium, UK Biobank, JENGER, and FINNGEN. The patients/participants provided their written informed consent to participate in this study.

X-JL conceived, designed, and supervised the whole study and wrote and revised the manuscript. MS, X-JL, YY, LL, and WL collected the GWAS summary statistics, performed the data processing and transformation, and conducted the GWAS meta-analysis. JL performed the tissue and cell type-specific enrichments analyses, TWAS, and generated the Manhattan and locus zoom plots. LL provided critical comments for the manuscript improvement. All authors read this manuscript carefully, provided critical comments, and approved the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.