Epilepsy gene variants differ sharply between East Asian and European populations

Epilepsy touches the lives of roughly 50 million people worldwide, and for a substantial share of those diagnosed early in life the disorder traces back to a single faulty gene. Genetic testing for epilepsy is now among the most frequently ordered genomic diagnostics on the planet, yet the reference databases that laboratories rely on to interpret results have never been systematically probed for ancestry-specific quirks in epilepsy genes. A team of neurologists and geneticists from Guangzhou Medical University and Southern Medical University in China has now filled that gap with an unusually rigorous analysis, published in Frontiers in Genetics. By resolving every variant down to the exact nucleotide change, the researchers uncovered seventeen pathogenic or likely pathogenic epilepsy variants that are dramatically more common in East Asian populations than in non-Finnish European populations — and, in the process, exposed a methodological flaw that they estimate corrupts three out of every four candidate signals produced by the standard database-mining approach.

The logic of population frequency sits at the heart of clinical variant interpretation. Under the ACMG/AMP guidelines that govern genetic testing worldwide, a variant that is too common in reference populations cannot cause a fully penetrant, severe early-onset disease, so frequency thresholds feed directly into benign evidence criteria such as PM2 and BS1. ClinVar, the primary public archive of variant classifications, and gnomAD, a frequency catalogue now spanning nearly one million individuals across eight ancestry groups, are the twin pillars of this reasoning. Database-mining studies routinely combine the two resources at the level of dbSNP reference identifiers, or rsIDs, a shortcut that has characterized population-specific pathogenic variants in monogenic Parkinson disease, cardiomyopathy and Brugada syndrome. But a single rsID can aggregate several distinct nucleotide changes at one genomic position, and programmatic lookups typically return the frequency of the most common allele without specifying which change it represents. How much this conflation distorts population enrichment analyses had never been quantified.

The Chinese team, led by co-first authors Jin-Xing Lai and Ming-Feng He, assembled a catalogue of 199 epilepsy-associated genes: 168 in which epilepsy is the core or pure symptom, as defined by a recent comprehensive classification, plus 31 neurodevelopmental disorder genes routinely included in clinical epilepsy panels. From ClinVar they extracted 46,746 unique pathogenic and likely pathogenic variants across 191 genes, of which 23,316 were single nucleotide variants corresponding to 9,847 unique rsIDs. The heaviest burdens sat in NF1 with 7,461 variants, SCN1A with 3,129 and TSC2 with 2,355. Rather than trusting rsIDs, the researchers aligned the ClinVar protein and cDNA notation of every variant to the HGVS consequences recorded in gnomAD v4.1.1 gene-level tables, then confirmed each match against the full GRCh38 coordinates, reference allele and alternate allele. This nucleotide-resolved procedure assigned population-specific allele frequencies to 805 informative variants across 131 genes, drawn from East Asian and non-Finnish European ancestry groups.

Fisher’s exact tests with Benjamini-Hochberg correction identified nine variants significantly enriched in East Asian relative to non-Finnish European populations, with odds ratios ranging from 36.3 to 867.9. The strongest signals were the ST3GAL5 splice donor variant c.13+1G>A, found in 28 of 41,244 East Asian alleles but only 2 of 418,078 European alleles, and ALDH7A1 p.Tyr516Cys, present in 16 of 44,882 East Asian alleles and completely absent from Europeans. The remaining significant variants mapped to GOSR2 p.Arg107Ter, SETD1A p.Arg913Cys, a second ALDH7A1 splice-region variant, AARS1 p.Arg840Ter, KCNT1 p.Arg25Trp, the WWOX start-loss variant c.2T>C and BRAT1 p.Leu59Pro. Four of the nine were absent from the European reference sample altogether; the other five showed East Asian frequencies 38- to 158-fold higher than their European counterparts. The ALDH7A1 missense variant carried the largest effect size of the set, with a Haldane-Anscombe-corrected odds ratio of 867.9 — an estimate of enrichment for a variant with no observed European carriers, so no direct frequency ratio can be computed.

Thirteen of the seventeen confirmed variants — nine significant plus eight exploratory low-frequency signals — resided in just seven autosomal recessive genes: ALDH7A1, ST3GAL5, GOSR2, AARS1, BRAT1, WWOX and UBA5. That concentration has direct clinical relevance because recessive genes are exactly where carrier screening operates. The two ALDH7A1 variants together account for 22 of 44,882 East Asian alleles, a combined carrier frequency of roughly 0.05 percent, while the ST3GAL5 splice donor variant reaches 0.07 percent. Each of these genes is an established cause of autosomal recessive epilepsy: ALDH7A1 causes pyridoxine-dependent epilepsy, ST3GAL5 underlies GM3 synthase deficiency, GOSR2 drives progressive myoclonus epilepsy, and WWOX is implicated in epileptic encephalopathy. The study quantifies East Asian population frequencies of these known pathogenic variants rather than claiming their discovery. To define the ancestry boundary of the enrichment, the team also extracted South Asian, African and Admixed American frequencies: fifteen of the seventeen variants were absent or singletons in those groups, supporting East Asian-specific or East Asian-dominant enrichment, with one WWOX variant showing a shared East and South Asian pattern.

The four variants mapping to dominant genes demand more caution. The SETD1A p.Arg913Cys variant, at 8 of 44,876 East Asian alleles, approaches the 0.02 percent threshold commonly used for the BS1 benign-frequency criterion and may reflect incomplete penetrance rather than a fully penetrant epilepsy allele. KCNT1 p.Arg25Trp, ADGRV1 p.Trp1139Leu and KCNC1 p.Arg164Trp were present at low frequency, and two exploratory signals had raw East Asian counts that did not exceed their European counts, with their odds ratios reflecting the smaller East Asian reference sample. Three variants — GOSR2 p.Arg107Ter, BRAT1 p.Leu59Pro and UBA5 p.Arg246Ter — carried no assertion criteria in ClinVar, meaning zero review stars, and their classifications should be reassessed. The recessive-gene concentration was nevertheless not an artifact of low-confidence classifications: restricting to variants with at least one review star left fourteen signals, ten of them recessive, and stricter exploratory thresholds left the recessive proportion above 80 percent. Molecularly, ten of the seventeen variants were loss-of-function changes — six nonsense, three canonical splice-donor and one start-loss — the subset carrying the strongest prior for pathogenicity under the PVS1 criterion.

The methodological comparison is where the study turns explosive. The researchers repeated the entire enrichment analysis using the standard rsID-level workflow, assigning each identifier the frequency of its most common allele without verifying which nucleotide change it represented. That shortcut yielded 60 candidate East Asian-enriched variants. Nucleotide-resolved matching confirmed only fifteen as genuine pathogenic signals — a false-positive rate of 75.0 percent — and it additionally recovered two genuine signals, UBA5 p.Arg246Ter and KCNC1 p.Arg164Trp, that the rsID workflow missed entirely. Among the 45 false positives, 20 arose when a synonymous allele was the most common at the position, 20 when a distinct protein-altering allele was the most common, 4 from splice-region conflation and 1 from a tri-allelic start-loss position. In every case the ClinVar pathogenic allele was a different nucleotide change at the same spot. Although only 12.9 percent of rsIDs with gnomAD data were multi-allelic, the error rate in the candidate set was nearly six times higher, because a common benign allele is precisely what marks a position as enriched.

The rsID workflow also assigned an informative frequency to 1,707 variants, compared with 805 under nucleotide-resolved matching, a gap reflecting positions where the ClinVar pathogenic allele is absent from gnomAD but another allele at the same position carries frequency data. The authors argue that published population-specific variant lists derived from rsID-level queries — including those in monogenic Parkinson disease and Brugada syndrome — warrant nucleotide-level revalidation, and they call for nucleotide-resolved matching by HGVS notation or full genomic coordinates with reference and alternate alleles to become a standard step in ClinVar-gnomAD integration. The findings extend earlier observations of ancestry bias in ClinVar submissions, where European-ancestry ascertainment has long skewed the archive. The predominance of NFE-only variants in the frequency distribution — 720 of 805 informative variants had nonzero European frequency only, versus 28 with East Asian frequency only — mirrors both the larger European sample in gnomAD and the European skew of ClinVar submissions, underscoring how reference data themselves shape which population differences can even be detected.

The study has clear boundaries. It was restricted to single nucleotide variants, so genes with prominent insertion, deletion or copy-number mechanisms, such as NF1 microdeletions and MECP2 duplications, are incompletely captured. The eight exploratory signals did not survive multiple-testing correction, and their allele counts of two or three are sensitive to sequencing artifacts and related individuals in gnomAD. With roughly 44,900 East Asian alleles against about 1,180,000 European alleles, the analysis is insensitive to East Asian-specific enrichment below an allele frequency of about 4.5 × 10⁻⁵, so negative findings below that floor do not imply equal frequencies. No independent ACMG/AMP re-curation was performed. Still, the core conclusions held through every sensitivity check, including restricting the gene set to the 168 core epilepsy genes. For clinicians and laboratories, the message is actionable: ancestry-matched frequency interpretation and population-aware carrier frequency estimation are priorities for equitable epilepsy genetic testing, and the database shortcut that many pipelines still trust is quietly inflating the very signals they are meant to find.

Subject of Research: Population-specific frequencies of ClinVar pathogenic variants in epilepsy-associated genes

Article Title: Population-specific differences in ClinVar pathogenic and likely pathogenic variants of epilepsy-associated genes between East Asian and non-finnish European populations: a nucleotide-resolved gnomAD v4.1.1 analysis

Article References: Lai, J.-X., He, M.-F., Fan, Z.-C., Yan, F.-J., Guan, B.-Z., & Liu, X.-H. (2026). Population-specific differences in ClinVar pathogenic and likely pathogenic variants of epilepsy-associated genes between East Asian and non-finnish European populations: a nucleotide-resolved gnomAD v4.1.1 analysis. Frontiers in Genetics, 17, Article 1973684. https://doi.org/10.3389/fgene.2026.1973684

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1973684

Keywords: epilepsy, ClinVar, gnomAD, variant interpretation, population frequency, carrier screening, East Asian ancestry, autosomal recessive, rsID conflation, genetic testing, ACMG guidelines, neurogenetics