Introduction

Impulsivity is a heritable behavioural trait leading to actions that are “poorly conceived, prematurely expressed, unduly risky, or inappropriate to the situation and that often result in undesirable consequences”1. Estimates of heritability for impulsivity from a study of twins were between 33% and 56% at ages 11–13 years, and between 19% and 44% at ages 14–162. Raised impulsivity is a key endophenotype of attention-deficit hyperactivity disorder (ADHD)3, bipolar disorder4 and juvenile myoclonic epilepsy (JME)5,6,7. ADHD is characterised by inattention, hyperactivity and impulsivity. Individuals with ADHD show more signs of impulsivity (attentional, non-planning and motor) compared to controls8. A previous genome-wide association study (GWAS) of impulsive personality traits (UPPS-P Sensation Seeking, Drug Experimentation and UPPS-P Negative Urgency) in 22,861 healthy individuals of European ancestry demonstrated two significant associated loci at 3p12.1 and 22q13.19. Variants at the 3p12.1 locus correlated with predicted Cell Adhesion Molecule–2 (CADM2) expression, in the putamen10, and the 22q13.1 locus near CACNA1I has been previously implicated in schizophrenia11. CADM2 mediates synaptic signalling and is highly expressed in the human cerebral cortex and cerebellum12. Given impulsivity is elevated in neuropsychiatric disorders, there may be shared genetic mechanisms across disorders and/or with impulsivity in the general population, however to our knowledge there has been no GWAS of impulsivity in any neuropsychiatric disorder.

Impulsivity is elevated in different epilepsies, but the evidence across multiple dimensions of impulsivity is strongest in JME5,6,7. JME is a common adolescent-onset syndrome characterised by awakening myoclonic, generalised tonic-clonic and absence seizures, often triggered by sleep deprivation. Trait impulsivity in JME is associated with the frequency of both myoclonic and absence seizures6, but it is not clear if this indicates a causal relationship or a common mechanism regulating both impulsivity and seizures, though convergent lines of evidence suggest the involvement of overlapping prefrontal-striatal networks in both JME and impulsivity13,14,15,16,17,18,19,20. Finding a shared aetiology would offer new therapeutic approaches for drug-resistant epilepsy.

The overall syndrome of JME has complex inheritance with few replicated susceptibility loci21,22, and other loci with less support22,23,24. A major challenge for epilepsies of complex inheritance is to explain the wide variation in phenotypic expression and treatment response between individuals. Forty-percent experience antiseizure medication (ASM) resistance or intolerance25. In addition, no current ASM modifies the lifelong disease course of JME and the pharmacological options are sparse, especially for women25. Hence novel treatments based on genetic disease mechanisms, such as those emerging for monogenic channelopathy and mTOR pathway epilepsies, are urgently needed26,27. Our methodological approach is to carry out genome-wide analysis of endophenotypes in JME such as impulsivity and clinically relevant outcomes such as ASM resistance, a strategy with predicted advantages for reducing heterogeneity, increasing statistical power28,29 and improving direct clinical translation for precision medicine.

Results

Genome-wide association analysis with BIS-Brief

We investigated the influence of 8,950,360 variants on impulsivity in European ancestry patients with JME (n = 324) and a mega-analysis including all ancestries (n = 372), who self-rated their trait impulsivity using the Barratt Impulsivity Scale, eight-item BIS-Brief version30. We conducted a GWAS of BIS-Brief score (Supplementary Fig. 2) in the European subset, adjusted for sex, genotyping batch, age at consent, population stratification, and seizure frequency (Supplementary Table 2). We discovered two genome-wide significant loci, one on chromosome 8 (rs73293634 (G/T)) and one on chromosome 10 (rs75042057 (T/G) (Fig. 1, Table 1, Supplementary Figs. 3 and 4). Given the distribution of BIS-Brief was slightly right skewed, for sensitivity analysis we tested the SNP associations using an inverse rank normal transformed BIS-Brief phenotype as well. Qualitatively similar results were obtained with rs73293634 and rs75042057 demonstrating association with p = 3.1 × 10−8 and p = 1.4 × 10−7, respectively (Supplementary Table 3). The distribution of BIS-Brief by rs73293634 and rs75042057 genotypes are provided in Supplementary Fig. 5. In a mega-analysis comprised of all ancestral groups (Supplementary Fig. 6), these loci were further supported including by a nearby chromosome 8 SNP (rs146866040, r2 = 0.89) with stronger evidence of association in the combined ancestry mega-analysis as measured by the p value (P = 1.57 × 10−9; Table 1), providing cross-ancestral support for the locus.

Fig. 1: Manhattan plot showing GWAS with BIS-Brief score.
figure 1

Linear regression was used to test association of each SNP with BIS-Brief. Sex, genotyping batch, age at consent, first three PCs, and the frequency of myoclonus or absence seizures were included as covariates in the model. We found two significant genome-wide associations on chromosome 8 (rs73293634 (G/T)) and 10 (rs75042057 (T/G)) in the analysis of 324 European individuals with JME. Variants below −log10P < 1 were omitted in the plot.

Full size image
Table 1 Summary of genome-wide associated variants for the GWAS of BIS scores in JME (n = 324).
Full size table

rs73293634 falls in an intergenic region near SLCO5A1. The phenotypic variation explained (PVE) for rs73293634 was 10.1% in the European analysis. Although a second JME cohort with impulsivity measured is not available for replication, Watanabe et al. 31 reported a rs73293634 association with risk taking in the UK Biobank, where they asked the question “Would you describe yourself as someone who takes risks?” (OR (95% CI) = 1.032 (1.001–1.065), p = 0.04, minor allele frequency (MAF) = 0.03, N = 371,049). Association results posted on the same data by the Neale Lab32 with ~23 K fewer participants, provides a similar qualitative conclusion (β (SE) = 0.005 (0.003), p = 0.09, MAF = 0.03, N = 348,549). Two individuals with large structural deletions that include SLCO5A1 are reported in the Decipher Genomics database with seizures and neurodevelopmental disorder (www.deciphergenomics.org/gene/SLCO5A1/patient-overlap/cnvs).

The significant genome-wide association on chromosome 10 (rs75042057) falls in intron 22 of PARD3 (NM_001184785.2). The PVE by the SNP is 9.3%, although there are no variants in linkage disequilibrium with this SNP so further interrogation and confirmation of this locus is required. We note, however, that significant linkage (multipoint max LOD 4.23, alpha 0.34) was previously reported to this locus in French-Canadian families with idiopathic generalised epilepsy (IGE)33, of which JME is a common subtype. As well, rs75042057 was also associated with risk-taking in the UK Biobank (OR (95% CI) = 1.067 (1.029–1.106), p = 4.79E-4, MAF = 0.02, N = 371,049)31.

Colocalization analysis with gene expression

Since the GWAS-associated variants are not exonic, we next assessed whether the variants impact gene expression, and for which gene in which tissue of origin, by assessing colocalization of the genome-wide significant peaks with expression quantitative trait loci (eQTL) in brain tissues. We used eQTLs from the Genotype-Tissue Expression project (GTEx) v812, PsychENCODE34, and human fetal brains35 and combined them with the GWAS summary statistics from the mega-analysis, for colocalization analysis adjusting for multiple hypothesis testing36. Colocalization analysis with eQTLs from GTEx brain and tibial nerve tissues for genes at the locus (chr8:69,650,000–70,000,000, hg38) shows significant colocalization with SLCO5A1 in the cerebral cortex, and no colocalization with other genes in the region (Fig. 2a and Supplementary Fig. 7; Simple Sum 2 colocalization P = 9.5 × 10−3). The minor allele for the lead SNP rs73293634 (T) decreases expression in GTEx cerebral cortex (Fig. 2c). We found no significant colocalization with eQTLs from PsychENCODE34 and fetal brains35, although nearby variants in the locus in adult brains in PsychENCODE have, in general, a clear influence on SLCO5A1 expression (Fig. 2b). According to BrainSpan37,38, SLCO5A1 is highly expressed prenatally, with expression dropping after birth but remains detectable throughout adulthood (Fig. 2d). We did not observe significant colocalization at the chromosome 10 locus with eQTLs from adult brains in GTEx12, PsychENCODE34 or fetal brains35.

Fig. 2: LocusFocus70 plot for the GWAS with BIS-Brief in JME (circles) and eQTLs in GTEx12 brain and tibial nerve tissues for the SLCO5A1 gene (lines).
figure 2

The Simple Sum 236 and COLOC269 colocalization methods implemented in LocusFocus (v1.4.9)70 were used to test for colocalization of the BIS-Brief genome-wide peaks with eQTL analyses brain tissues from GTEx v812, PsychENCODE34, and fetal brain35. a Colocalization figure from LocusFocus for the SLCO5A1 gene. Lines depict the minimum P value trace in a sliding window for SLCO5A1 eQTLs from GTEx, one line per tissue. Circles depict the GWAS with BIS-Brief, with the lead SNP in purple and pairwise LD with the lead SNP marked as shown in the legend, calculated using the 1000 Genomes Project60 European subset. Significant colocalization is observed for SLCO5A1 eQTLs in GTEx v8 for the cerebral cortex after increasing sample size in a mega-GWAS (n = 367, −log10 Simple Sum 236 P = 9.5 × 10−3). Colocalization analysis with only the Europeans is provided in Supplementary Fig. 7. Colocalization was also tested for all other nearby genes shown in the figure, but no other genes’ eQTLs colocalized with BIS-Brief GWAS (not shown). b Colocalization analysis with PsychENCODE eQTLs in the dorsolateral prefrontal cortex (DLPFC) (n = 1866)34, and eQTLs derived from second trimester fetal brains (n = 120)35, with GTEx’s brain cortex eQTL as in A provided for reference. Colocalization analysis results suggest no colocalization with either PsychENCODE (Simple Sum 2 P = 0.985) or fetal brain eQTLs (does not pass first stage test in Simple Sum 2 for having significant eQTLs in the region). c Violin plot for the eQTL effect of rs73293634 SNP on SLCO5A1 expression in the cerebral cortex from GTEx v8. d Expression change of SLCO5A1 from brains in various developmental stages from BrainSpan37,38. pcw, post conception weeks; preadolescence, 2–12 years old (inclusive); adolescence, 13–19 years old; adult, ≥20 years old (oldest samples are 40 years old). The centre lines represent the 50th percentile (median) and the bounds of the boxes are the 75th and 25th percentiles (interquartile range) with the whiskers being the largest value within 1.5 times the interquartile range above the 75th percentile and smallest values within 1.5 times the interquartile range below the 25th percentile.

Full size image

Functional characterisation of SLCO5A1

SLCO5A1 is a membrane-bound organic anion transporter with no known substrate39 (Fig. 3). We performed a full protein BLAST (BLASTp) search of the SLCO5A1 polypeptide sequence (NP_112220.2) on Drosophila melanogaster to identify the closest matching sequence alignment. While several members of the Oatp family were found to have significant homology, Oatp30B was the family member with the highest homology and a 37.66% identity and E-value of 2 × 10−150 (NP_995667.1). SLCO5A1 was the closest human analogue of Oatp30B also in a reverse BLASTp. Indeed, BLASTp of Oatp30B polypeptide sequence (Q9VLB3) across all species for conserved domains reveals this gene has conserved major facilitator superfamily (MFS), OATP, and Kazal domains (Fig. 3 and Supplementary Fig. 8). We therefore used an effective RNAi transgenic line (Supplementary Fig. 9A) to assess the effect of pan-neuronal adult knockdown of Oatp30B/SLCO5A1. Flies with reduced Oatp30B levels displayed a small but significant shortening of their lifespan (Supplementary Fig. 9B) and a striking over-reaction to vibration stimuli applied through the automated Drosophila Arousal Tracking (DART) system40, which elicit an otherwise modest activity response in two separate control fly genotypes (Fig. 4a). Additional analysis of locomotor behaviour clarifies that Oatp30B knockdown did not alter the speed of flies or the duration of each activity bout or the interval in between bouts of action (Supplementary Fig. 9C–E), indicating a specific defect in excessive response to stimuli. Furthermore, Oatp30B knockdown led to a dramatic increase in the frequency of seizure-like events (Fig. 4b) when exposed to hyperthermia, a trigger for seizures in Drosophila41. Recovery to full motility after seizure-like events was also significantly slower in flies with Oatp30B knockdown (Fig. 4c). These data establish a common causal link between Oatp30B/SLCO5A1 downregulation, startling behaviour, and susceptibility to seizure-like events.

Fig. 3: Domain architecture of human SLCO5A1.
figure 3

a Schematic representation of the protein with the indication of recognised domains. A SMART analysis to identify structural domains confirmed the presence of two modules, Major Facilitator Superfamily (MFS) and a Kazal domain, interspaced with potentially unstructured sequences. The MFS transporters are membrane proteins capable of transporting small solutes in response to chemiosmotic ion gradients72,73. They are represented in many organisms from Archaea to Homo sapiens. MFS proteins target a wide range of substrates, including ions, carbohydrates, lipids, amino acids and peptides, nucleosides and other small molecules and transport them in both directions across the membrane74. The Kazal domain is an evolutionary conserved module usually acting as a serine-protease inhibitor. b Predicted model of the monomeric form of SLCO5A1 from amino acids 115–766, built using the SwissModel homology server (https://swissmodel.expasy.org) and utilising the template structure pdb:7eeb. Red: alpha helices; Yellow: Beta strands; Green: Loops.

Full size image
Fig. 4: Startling reaction to trains of vibrations, increased seizure prevalence and increased post-seizure recovery time in flies with Oatp30B knockdown.
figure 4

a Startling reaction to trains of vibrations. The UAS-Oatp30BIR (GD12775) transgenic or the control UAS-GFPIR were driven with nSyb-Gal4 and Ubi-Gal80ts. The w1118 strain is a control for the genetic background in absence of transgenes. Mean ±SEM **P < 0.01, One Way ANOVA, Tukey’s post-hoc test. Units are the vibration events experienced 6 times for each fly, n = 174–210. b Increased seizure prevalence. The UAS-Oatp30BIR (GD12775) transgenic or the control UAS-GFPIR were driven with nSyb-Gal4 and Ubi-Gal80ts. Percent ±SE ****P < 0.0001, Log-rank (Mantel-Cox) test, X2 24.68 for 1 df, n = 34–36. c Increased post-seizure recovery time. The UAS-Oatp30BIR (GD12775) transgenic or the control UAS-GFPIR were driven with nSyb-Gal4 and Ubi-Gal80ts. Mean ±SEM *P < 0.05, Mann Whitney non-parametric test, two tails, n = 10–26. Only flies that displayed a seizure within 120 s as in b have been included in the analysis.

Full size image

Gene enrichment analyses

We next sought to assess whether there was additional signal in the GWAS where sub-GWAS significant variants could inform additional contributing genes or pathways and whether there were shared genetic contributions with other psychiatric or epilepsy phenotypes. We selected all variants displaying P ≤ 5 × 10−4 and annotated these variants to the transcription start site of the nearest gene resulting in 810 unique genes. Gene enrichment analysis using one-sided hypergeometric tests42 identified a 12.7-fold enrichment of associated genes from the presynaptic membrane organisation gene set (five out of nine genes; gene ontology (GO):0097090) and a 14.3-fold enrichment of associated genes from the presynaptic membrane assembly gene set (five out of eight genes; GO:0097105). These genes were PTEN, NLGN1, PTPRD, IL1RAPL1 and NLGN4X (Table 2). The combined PVE for the lead variants annotated to these five genes was 15.6% (25.8% with the addition of rs73293634 from the SLCO5A1 locus and rs75042057 from the PARD3 locus).

Table 2 List of top variants annotated to the five presynaptic assembly genes enriched in the European GWAS of BIS in JME (n = 324).
Full size table

The permutation tests of presynaptic membrane organisation (GO:0097090) over-enrichment and of presynaptic membrane assembly (GO:0097104) over-enrichment both produced permutation-based p values of 0.0005.

Investigation of these 810 genes revealed further43,44,45 that there was a significant overlap with genes reported in the GWAS catalogue that contribute to phenotypes relevant to the predominance of JME seizures on awakening, impulsivity and metabolism: chronotype (66 out of 522 genes overlap, P = 2.92 × 10−12), obesity-related traits (77 out of 662 overlap, P = 2.69 × 10−12), general risk tolerance (30 out of 238 overlap, P = 2.30 × 10−5), and adventurousness (21/134, P = 3.70 × 10−5).

Polygenic risk score analysis

Given impulsivity is a major component of ADHD, risk taking, bipolar disorder and epilepsy, we tested and found that a higher ADHD polygenic risk score (PRS) was significantly associated with a higher BIS-Brief score (p = 1.60 × 10−3) (Supplementary Fig. 10). It should be noted that the lead SLCO5A1 SNP, rs73293634, was not present in the ADHD GWAS from which the PRS was calculated, but rs146866040 which is in high LD did not show evidence of association itself with ADHD (OR (SE) = 0.9481(0.0562), p = 0.34)46. The rs75042057 SNP on Chr10 was also not present in the ADHD dataset nor was there a proxy with R2 > 0.6 available. The risk-taking PRS was also nominally associated with a higher BIS-Brief score (p = 0.018). PRSs for bipolar disorder, generalised and focal epilepsy did not reach statistical significance for association with BIS-Brief score at the 5% or Bonferroni-corrected level of 1% (P = 0.08, 0.33 and 0.96, respectively) (Supplementary Table 4). Altogether this suggests that the impulsive trait seen in JME is an endophenotype that shares genetic architecture with impulsivity in the general population as well as with individuals diagnosed with ADHD.

Discussion

This is a GWAS of trait impulsivity in a neuropsychiatric disorder and we present convergent evidence for the role of SLCO5A1 in impulsivity and seizure susceptibility through triangulation47 with GWAS, independent replication, colocalization with gene expression and functional evaluation in Drosophila48. While several Oatp family members display significant homology to SLCO5A1, the identified Oatp30B was the closest polypeptide in a BLASTp search and SLCO5A1 was the human polypeptide with the highest homology in a reverse BLASTp search. Therefore, whereas our analysis does not rule out some contribution by other closely related Oatp genes, for instance Oatp26F, it has identified a major role of Oatp30B in regulating startling and seizure-like behaviour in Drosophila. In contrast to human SLCO5A1, Oatp30B is expressed in the nervous system at constant low to moderate levels throughout fly stages, from development to adulthood. This enables investigation of gene function in vivo, in adult flies, although it limits generalisation as an SLCO5A1-linked disease model.

One GWAS of impulsive traits in the general population identified genome-wide significant association with variants in the CADM2 gene. CADM2 encodes a cell adhesion protein from the SynCam Immunoglobulin superfamily of recognition molecules, important for synaptic organisation and specificity; association of variants at the CACNA1I locus has been observed in previous studies with schizophrenia9. Our GWAS did not show significant association with these previously reported general population-associated variants at the CADM2 and CACNA1I loci9 (P = 0.152, beta = −0.52 for rs139528938; and P = 0.32, beta = −0.35 for rs4522708; the latter a SNP with r2 = 0.87 with the reported SNP, rs199694726, in our BIS-Brief dataset). Genome-wide summary statistics were not available to make additional comparisons. Genome-wide summary statistics were available for the risk-taking phenotype in the UK Biobank31, in which we observed replication of our lead genome-wide significant SLCO5A1 variant, rs73293634.

Previous expression studies show that SLCO5A1 upregulates gene sets implicated in cell adhesion, synapse assembly and organisation, principally belonging to the cadherin superfamily39; and the enrichment for presynaptic membrane assembly and organisation pathways in our dataset includes genes encoding trans-synaptically interacting proteins that are implicated in a wide range of neuropsychiatric disorders49,50. Genetic correlation between ADHD and the BIS-Brief score suggests converging genetic influences across ADHD and epilepsy. Taken together, these results support an important role for specific cell recognition molecules in the organisation of synaptic connections as a mechanism for variation in impulsivity across health and disease51.

While prefrontal-striatal inhibitory control networks are implicated in impulse control, specifically between mPFC and nucleus accumbens18,20, a role for these limbic networks has only been hinted at in epilepsy. Striato-nigral circuits, preferentially involving the ventral striatum, have long ago been implicated in the regulation of generalised seizures in rodent models of generalised epilepsy19. Recently, an initiating role for cortico-striatal networks in absence seizures with generalised spike-and-wave discharges has been shown in the mouse model of genetic epilepsy caused by haploinsufficiency of STXBP152, specifically by reduced cortical excitatory transmission onto striatal fast-spiking interneurons. The startling and the seizure-like phenotype of the SLCO5A1/Oatp30B knockdown in Drosophila suggests the genetic co-causality of startling and seizures. While it is not possible to define startling as the Drosophila equivalent of impulsivity, the two traits share some commonality in the lack of moderation in behaviour. This offers some additional support to the idea that excitatory-inhibitory imbalance in the prefrontal-striatal network may predispose simultaneously to epilepsy and impulsivity substrates and invites new approaches to neuromodulation of generalised seizures.

Methods

Human participants

We collected cross-sectional clinical and genetic data from the Biology of Juvenile Myoclonic Epilepsy (BIOJUME) consortium study, which focuses on gathering cases with JME (n = 864)25. Inclusion criteria have been discussed previously6. BIOJUME is a study across 50 sites in 10 countries (Appendix). Furthermore, all participants’ medical history was reviewed by a phenotyping committee to validate the diagnosis of JME. Written informed consent was obtained from all participants prior to inclusion in the study and ethical approval from the UK Health Research Authority, South Central Oxford C Research Ethics Committee (16/SC/0266) and all other collaborating sites was obtained. The SickKids Research Ethics Board of The Hospital for Sick Children (1000033784) also gave ethical approval for this work.

Barratt impulsivity scale-brief (BIS-Brief)

We collected self-rating of trait impulsivity through the BIS-brief6,30. The BIS-Brief is a short version of the BIS, one of the most commonly used measures of impulsiveness. The current version of BIS (BIS-11) includes 30-items measuring 3 theoretical subtraits: attentional, motor, and non-planning impulsiveness. BIS-Brief is a unidimensional scale including 8 of the original BIS-11 items generating a total score ranging from 8 to 32. BIS-Brief demonstrated similar indices of construct validity observed for the BIS-11 total score. Using BIS-Brief in large epidemiological studies of psychiatric disorders reduces the burden on respondents without loss of information29.

Genotyping quality control

DNA was extracted from blood by each consortium site and sent to The Centre for Applied Genomics at The Hospital for Sick Children in Toronto for genotyping. We genotyped participants’ DNA in four batches (n = 702) using the Illumina Omni 2.5 array. SNPs were called using the self-clustering method in Genome Studio. We performed quality control (QC) for each genotyped batch using PLINK v1.90b6.1853 and custom in-house scripts. Briefly, we removed individuals and variants with call rates below 90%; samples with sex mismatches and/or high heterozygosity; males with heterozygous calls for X chromosome markers (non-pseudoautosomal region); and females with non-missing calls for markers on the Y chromosome. We retained heterozygous calls for mitochondrial markers in both sexes (i.e., due to heteroplasmy). We obtained an unrelated sample by using KING v.2.2.4 software’s54 --unrelated option (that is, those with estimated kinship coefficient <0.088). We corrected and updated the ped file with all found relationships, and identified markers with Mendelian errors using PEDSTATS 0.6.1255. We flagged 399 markers but did not remove those out of Hardy–Weinberg Equilibrium (P < 10−4). We conducted principal component analysis adjusted using the kinship matrix output by KING using PC-AiR in the GENESIS v2.16.0 package56.

We performed quality control on each genotyping batch separately, followed by removal of ambiguous A/T, G/C SNPs, chr0 SNPs, indels, monomorphic variants, and duplicate variants; and performed strand alignment using Will Rayner’s alignment files (www.well.ox.ac.uk/~wrayner/strand/), then merged all batches. We re-analysed and removed cryptic relationships across batches. The final merged set contained 1,489,917 variants, 695 individuals (241 males, 454 females) including 23 related pairs (for association analyses however, an unrelated set was selected).

Genotype imputation

We used the McCarthy Tools v4.3.0 to prepare the genotype data for imputation (www.well.ox.ac.uk/~wrayner/tools/HRC-1000G-check-bim-v4.3.0.zip) using TOPMed as the reference panel (r2@1.0.0) on the TOPMed imputation server57,58,59. We converted coordinates from hg37 to hg38 coordinates using strand files (www.well.ox.ac.uk/~wrayner/strand/InfiniumOmni2-5-8v1-4_A1-b38-strand.zip). We merged the pseudoautosomal region (PAR) using PLINK’s --merge-x option and checked variants using the HRC checking tool. We removed a total of 282,660 variants due to no matches in the reference (but still analysed for association with BIS-Brief afterwards), and 1,739,329 variants remained for imputation on the server. We used Eagle v2.4 for phasing, and minimac v4 v1.0.2 for imputation. We kept variants with imputation quality score r2 > 0.4 and MAF > 1% for analysis. A total of 8,950,360 variants remained for association analysis.

Genome-wide association analysis

We included for analysis 381 individuals who passed phenotype QC with complete BIS-Brief rating. From these, four failed genotyping QC, and one individual was removed due to cryptic relatedness (n = 376). The mega-GWAS analysis consisted of a total of 372 unrelated individuals adjusted for sex, genotyping batch, and population stratification (Supplementary Fig. 1). The mega-GWAS was used for colocalization analysis of the genome-wide association peak on chromosome 8. We identified 329 patients as European ancestry (defined as within 6 standard deviations from the 1000 Genomes60 European cluster in a principal component analysis). Among these, five patients had missing information on seizure frequency, so we used 324 individuals for the genome-wide association analysis. The current sample size is sufficient to detect genetic variants that explain 12% of the variance in the BIS-Brief score with 80% power after adjusting for multiple hypothesis testing at the genome-wide significance level. We adjusted for sex, genotyping batch, age at consent, population stratification, and the frequency of myoclonus or absence seizures. The relationship of the frequency of myoclonus or absence seizures, and its relationship with ASM and sex with trait impulsivity in JME, has been described previously and was thus adjusted for in current regression analyses6,25. All analyses were conducted in the European subset unless noted otherwise. Chromosome X (non-pseudoautosomal region) was analysed with males coded as zero for the reference allele and two for the alternate allele, under the assumption of X-inactivation61. Genome-wide significant loci were further investigated for replication of association with risk-taking phenotypes in the general population using publicly available summary statistics31,32.

Gene enrichment analysis

Variants with P ≤ 5 × 10−4 were annotated to the gene with the nearest transcription start site using the Ensembl Variant Effect Predictor (v94)62. This gene set was used as input in a GO enrichment analysis63,64, to test for enrichment in annotated pathways. One-sided hypergeometric tests were completed to identify over-representation of pathways42. To reduce the risk of false positive results, a permutation procedure65 was employed by randomly shuffling GWAS p values 2000 times, each time re-applying the P ≤ 5 × 10−4 threshold and calculating the hypergeometric test statistics. For one pathway, the final permutation-based p value was calculated as the percentage of the 2000 permutations that produced a p value less than or equal to the p value calculated from the non-permuted data. A pseudo count was added during this calculation to prevent calculating p values equal to 0.

Phenome-wide association study (PheWAS) analysis

We queried the top associated genome-wide variant and the top associated variant for each of the nine presynaptic assembly enriched genes across PheWAS databases: GWAS Atlas (https://atlas.ctglab.nl/), Global Biobank Engine66, PheWeb67, and Gene Atlas68.

We used PheWeb portals:

PRS analysis

Clumping and thresholding were used to calculate ADHD, risk taking, bipolar disorder, generalised epilepsy, and focal epilepsy PRS in individuals of European ancestry using PLINK v1.953. Five PRS were calculated. A Bonferroni-corrected critical value for significance would therefore be p < 0.05/5 = 0.01. The source of summary statistics used, variant filtering, clumping and thresholding details are summarised in Supplementary Table 1. PRS values were generated by weighting selected SNPs after clumping and thresholding by the additive scale effect (log10(OR)/Beta), and then summing over the variants. The PRS values were then centred to the mean. Association of PRSs with BIS-Brief was tested using linear regression with age, sex, and frequency of absence/myoclonic seizure as covariates in the model.

Colocalization analysis

We used the Simple Sum 236 and COLOC269 colocalization methods as implemented in LocusFocus70 (v1.4.9) to test for colocalization of the genome-wide peaks with eQTL analyses in brain tissues in GTEx v812, PsychENCODE34, and fetal brain35. For the genome-wide associated locus on chromosome 8, we performed colocalization analysis using both the mega-GWAS and Europeans-only GWAS. The required significance threshold, after multiple testing of all colocalization datasets analysed was 0.01.

Domain architecture of SLCO5A1

A BLAST search against the entire Protein Data Base (PDB) identified only one hit with a convincingly high E-value (1e-55) that pointed to the Chain L of the Kazal-like domain containing mice protein (7EEB). The search had a 26% identity and a coverage of 74%. After this hit, the other four identified sequences had E values > 0.002, clearly distinguishing between significant and non-significant hits. 7EEB is a large complex containing several subunits, among which is SLCO6C1, which is the region scoring for SLCO5A1.

Phenotypic variance explained

To assess the PVE by a SNP or a group of SNPs, we calculated the partial r2 as the proportion of the residual sum of squares (RSS) reduced when adding the SNP (or group of SNPs) to the base regression model with all covariates.

siRNA probe design and knockdown of Oatp30B in Drosophila melanogaster

Drosophila

Flies were maintained and crossed at 18 °C. All ageing was done in a controlled environment of 29 °C and 60% humidity.

Stocks

ubiGal80ts // UAS-Oatp30BIR (GD12775) obtained from the VDRC // w1118, nSyb-Gal4, TubGal4 and UAS-GFPIR obtained from the BDSC.

Lifespan

Lifespan analysis was performed as previously reported41. All crosses were maintained at 18 °C during the developmental stages of the progeny. Newly eclosed adult flies were collected within 5 days at 18 °C. Females and males were pooled together and equally distributed within vials.

Motor behaviour assay

Single-fly tracking was carried out as previously described41. In each of three experiments, up to 12 flies per genotype, aged 15 days (adult stage) at 29 °C to allow RNAi expression and knockdown, were placed into individual round six-wells arenas. The protocol used consisted of 6 stimuli events equally split during a period of 2 h and 15 min, the first one starting after 30 min of recording, and the last one 30 min before the end of the protocol. Each stimuli event was composed of five vibrations of 200 ms spaced by 500 ms. The x/y position of each single fly was tracked and analysed using the DART software in order to evaluate the relative speed and activity before, during and after the stimuli event. The speed analysis was used for the “Stimuli Response Trace” and the general activity used to deduce “Active Speed”, “Mean Bout Length” and “Inter-Bout Interval”, using a custom-made modification of the DART software40.

Heat-induced seizure assay

Flies aged 15 days at 29 °C to allow RNAi expression and knockdown were isolated into new plastic vials without food for 10–20 min before immersion in a 42 °C water bath for 120 seconds. Each tube was video recorded during and post immersion and seizures were defined as a period of brief leg twitches, convulsions, and failure to maintain standing posture. Flies were, thereafter, allowed to recover at room temperature and the time to recover from seizure was calculated only for flies that had undergone seizures. All experiments were randomised and double-blinded.

RNA extraction and qPCR

RNA was extracted as previously reported71 from 15 adult flies of both sexes, aged 15 days at 29 °C to allow RNAi expression and knockdown, using TriZol (Thermo-Fischer). cDNA was generated using SuperScript III Reverse Transcriptase (Thermo-Fischer). Quantitative PCR was performed in combination with qPCRBIO SyGreen Blue mix (PCR Biosystems) on Quantstudio 7 from real-time PCR system (Thermo-Fischer). eIF4a was used as housekeeping control. The following oligos were used: Oatp30B Fw (GAATCCGACCAACCGCCTGA), Oatp30B Rv (ATGGATTCCTGCCGCCTGTG), eIF4a Fw (CGTGAAGCAGGAGAACTGG), eIF4a Rv (CATCTCCTGGGTCAGTTG).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.