Penetrance of pathogenic epilepsy variants is low and shaped by common genetic background

Authors: Remi Stevelink, M. Martijn Piet, Yorgos Bos, Ruben van ’t Slot, Kees P. J. Braun, Bobby P. C. Koeleman

Abstract

Objective

The identification of pathogenic variants in developmental epileptic encephalopathy (DEE) genes can be vital for counseling and individualized treatment. Penetrance is usually considered to be high or full, although this has never been studied in population cohorts. Recent evidence shows that common polygenic risk factors (polygenic risk score [PRS]) are increased in DEE cases, suggesting they might modify risk.

Methods

We calculated the penetrance of autosomal dominant epilepsy variants that have previously been classified as (likely) pathogenic in ClinVar, in two large cohorts (n = 42 863 and n = 386 306) using whole genome sequencing data. Next, we calculated PRS to assess whether common variants could modify epilepsy risk among people carrying pathogenic variants.

Results

Most people carrying pathogenic DEE variants did not have epilepsy. Penetrance estimates suggested that the probability of epilepsy in pathogenic variant carriers ranges between 4.1% and 9.8%. Among people carrying epilepsy variants, PRS was predictive of an epilepsy diagnosis. A high PRS was associated with increased risk of severe epilepsy, whereas a low PRS seems protective in people carrying a pathogenic variant.

Significance

Our results suggest that average variant penetrance is lower than expected and modified by PRS. A high PRS combined with a pathogenic variant may be necessary to develop a severe epilepsy phenotype like DEE. Reconsidering penetrance assumptions could improve variant classification and diagnostic yield. Our findings may enhance genetic counseling by refining risk estimates and could extend to other diseases.

Key points Analysis of whole genome sequencing data from population cohorts showed that most carriers of (likely) pathogenic epilepsy variants did not develop epilepsy.Average penetrance estimates ranged from 4.1% to 9.8%, although it likely varies by gene and specific variant.Polygenic burden of common epilepsy risk variants modified epilepsy risk among carriers of pathogenic variants.Combining polygenic risk scores with rare variant data has potential to refine diagnostic accuracy and improve genetic counseling.

INTRODUCTION

Since 1995, more than 1000 epilepsy genes have been discovered. ^1^ Diagnosing a pathogenic variant as an etiological cause of epilepsy is important to determine prognosis and enable targeted treatment. ^2^ Finding a pathogenic variant in a child with developmental epileptic encephalopathy (DEE) is generally deemed sufficient to diagnose monogenic epilepsy. ^3^ , ^4^

Surprisingly, recent studies show that individuals with DEE and a causal pathogenic variant also carry an increased burden of common epilepsy risk variants, quantified as polygenic risk score (PRS). ^5^ , ^6^ , ^7^ , ^8^ However, it remains unclear whether people who carry a pathogenic variant and a low PRS remain healthy or show a milder phenotype.

Although reports of incomplete penetrance are documented (e.g., in SCN1A, SCN1B, and GABRG2), ^6^ , ^9^ autosomal dominant DEE variants are generally thought to be of high or full penetrance, as they are rarely inherited from a healthy parent. ^3^ , ^4^ Many DEE genes have been discovered by selecting for de novo variants that occur in multiple severely affected patients, but it is unknown whether these variants also occur in people with mild epilepsies or healthy individuals. Penetrance of recurrent variants is usually determined within large epilepsy families ^6^ , ^10^ ; however, these estimates could be inflated, because family members share environmental factors and polygenic background that are unrelated to the pathogenic variant itself.

Variants are often considered benign if they are reported in multiple healthy subjects, because individual pathogenic epilepsy variants are exceedingly rare in population databases such as GnomAD or ExAc. ^11^ , ^12^ However, there are thousands of established pathogenic epilepsy variants, and their collective population prevalence and penetrance remains elusive.

In this study, we first estimated the penetrance of autosomal dominant DEE variants by assessing their prevalence in large cohorts of unrelated people with and without epilepsy. Next, we calculated their PRS to assess whether common risk variants could modify penetrance.

MATERIALS AND METHODS

Study population

We used genomic and phenotypic data from two large Genomics England (GEL) and UK Biobank (UKB).

GEL served as a discovery cohort. Approximately 78 000 patients with rare diseases, cancer, and unaffected relatives were recruited via centers across the UK at a mean age of 30 years (range = 0–99). ^13^ , ^14^ Detailed standardized phenotypic information including disease status was collected for all subjects. Informed consent was obtained from all subjects, parents, or legal guardians.

UKB was used for replication. This population cohort included 500 000 individuals aged 40–69 years at recruitment. ^15^ Subjects were recruited irrespective of diseases, socioeconomic status, or ethnicity. Mental capacity to provide informed consent was assessed, likely resulting in exclusion of profound intellectual disability.

Subjects of both cohorts underwent germline whole genome sequencing (WGS) using Illumina next generation sequencing with a mean depth of 32×.

Phenotype definition

In GEL, epilepsy cases were identified if they were recruited under the category Early onset or syndromic epilepsy, which primarily constituted DEE and epilepsy plus other features like intellectual disability, diagnosed by a neurologist. To find additional subjects not specifically recruited for epilepsy, we also used International Classification of Diseases (ICD) codes to define cases with epilepsy (ICD G40), which we refined by excluding acquired structural causes (ICD codes C90–C96, C71, S06, I60–I63). Similar to a previous study, we defined severe epilepsy as when people were recruited for epileptic encephalopathy or epilepsy plus intellectual disability or had epilepsy plus an ICD diagnosis of intellectual disability (F70–F79) or a disorder of psychological development (F80–F89). ^5^ Controls had no epilepsy diagnosis (by recruitment or ICD code). No subjects were specifically recruited for epilepsy in UKB; therefore, only the above ICD definitions were used to classify cases and controls, restricted to age at epilepsy diagnosis prior to 18 years of age, under the assumption that monogenic epilepsy would manifest in childhood. ^16^ Year of diagnosis was inconsistently reported in GEL and therefore not used.

Selection of autosomal dominant pathogenic DEE variants

Similar to penetrance studies in other diseases, ^17^ , ^18^ we queried the ClinVar database (version 20240708) for coding or splicing variants in autosomal dominant DEE genes (Genes4Epilepsy v2024‐3), ^1^ retaining only variants classified as pathogenic, likely pathogenic, or pathogenic/likely pathogenic (hereafter referred to as pathogenic epilepsy variants for brevity). We removed variants with conflicting evidence or annotations suggesting incomplete penetrance (e.g., “low penetration,” “likely risk allele,” “risk factor”). Because variants in epilepsy genes can be associated with different diseases, we further filtered variants for a disease association with epilepsy, yielding a list of 4294 pathogenic variants across 92 DEE genes. Variants were filtered on a minor allele frequency < .0001 in each cohort. Finally, we performed manual curation of each identified variant to assess evidence of pathogenicity and removed variants where the predicted variant effect is not a known disease mechanism and/or where the specific variant is not robustly associated with epilepsy.

Subject quality control

To avoid inflation of PRS and penetrance estimates due to shared polygenic and environmental factors, we filtered both cohorts to include only unrelated subjects using genetically inferred relatedness (KING relatedness pruning using a threshold of >.0422, corresponding to 3rd degree relatedness or higher). European ancestry individuals were selected using genetic principal component analyses to increase homogeneity and enable PRS analyses using summary statistics from the largest European‐only epilepsy genome‐wide association study (GWAS). ^19^ Standard quality control measures were applied in both GEL and UKB, including exclusion of duplicates, outliers based on missingness, and genotyping heterozygosity, as described previously. ^13^ , ^15^

PRS analysis

PRSs were calculated with PRSice‐2 ^20^ using summary statistics from the largest European all epilepsy GWAS, containing 27 559 epilepsy cases. ^19^ WGS data were filtered on minor allele frequency > .01, Hardy–Weinberg disequilibrium (p < 1e‐10 in GEL and p < 1e‐54 in UKB), and >98% call rate per single nucleotide polymorphism (SNP). SNPs were pruned to a subset of uncorrelated SNPs (R ^2^ < .1), and SNPs with p < .5 were selected based on optimization in a previous study. ^21^ Individual PRSs were than calculated as the sum of weighted effect alleles, after which PRSs were standardized using Z‐score PRS−meanPRSSDPRS.

Statistical analysis

Logistic regressions were used to perform PRS group comparisons, corrected for the first four principal components of ancestry and sex, as described earlier. ^21^

We used linear regression analyses corrected for age and sex to compare fluid intelligence scores between groups in the UKB cohort.

Chi‐squared tests were used to compare pathogenic variant prevalence between the GEL and UKB cohorts.

To assess penetrance within the GEL cohort, we assessed the proportion of people with epilepsy among subjects who carried a pathogenic DEE variant. We corrected for ascertainment bias by correction penetrance estimates using the following formula as described previously ^22^ :PDG=PGD×PDPGD×PD+PGD¯×1−PDwhere D = disease, G = genotype (carrying a pathogenic variant), and D¯ = absence of disease. P(D|G) is the penetrance (probability of disease given a genotype); P(G|D) is the genotype frequency in cases; P(G|D¯) is the genotype frequency in controls, and P(D) is the general population prevalence of epilepsy. We assumed a .7% prevalence of broadly defined epilepsy and .29% prevalence of severe epilepsy phenotypes (DEE and epilepsy plus developmental impairment), based on recent studies. ^16^ , ^23^

All statistical tests were performed using R statistical software (version 4.3.2), and two‐tailed p‐values are reported.

RESULTS

Penetrance of pathogenic epilepsy variants in GEL

Among 42 863 individuals in the GEL discovery cohort, we identified 45 carriers (.10%) of a (likely) pathogenic autosomal dominant epilepsy variant. Of the 45 carriers, 14 (31%) had epilepsy, 13 of whom had a severe epilepsy phenotype; one subject had genetic generalized epilepsy with normal development. Among the 31 variant carriers without epilepsy, two (6%) had developmental impairment. The average age of people without epilepsy at last follow‐up was 50 years (range = 15–87), by which a monogenic epilepsy should have already manifested. ^16^

Variants were distributed across 15 genes; five variants were found in multiple people, whereas 32 (86%) were singletons (see Table S1 for a list of all variants and annotations). Regarding variant types, 23 (51%) of subjects carried a missense variant, 11 (24%) a nonsense variant, four (9%) a frameshift variant, and seven (16%) an intronic/splice variant. There was no difference in the proportion of missense variants between epilepsy cases and controls who carried pathogenic variants (odds ratio [OR] = .62, p = .59).

Using a broad definition of epilepsy, 1839 individuals in GEL (4.3%) had a diagnosis of epilepsy. This yields an ascertainment‐adjusted penetrance estimate of 6.8%, assuming a general population prevalence of .7%. ^23^ Penetrance of nonsense and frameshifts variants was 7.5%. Applying a stricter definition of severe epilepsy (DEE or epilepsy with developmental impairment), we found 1117 cases, corresponding to a penetrance of 4.1%, assuming a .29% population prevalence of severe epilepsies. ^16^ Thus, carrying a pathogenic variant was associated with a penetrance of 6.8% for any epilepsy and 4.1% for severe epilepsy, representing relative risks of 9.7 and 14.1, respectively.

Replication in UKB

In UKB, 297 of 392 123 individuals (.076%) carried a pathogenic epilepsy variant. This frequency was lower than in GEL (chi‐squared = 4.21, p = .040), but similar to GEL controls after excluding epilepsy cases (chi‐squared = .0001, p = .99). Nine of the people carrying a pathogenic variant had an ICD diagnosis of epilepsy (3.0%), none of whom had a severe epilepsy phenotype (DEE or epilepsy plus developmental impairment). None of the unaffected carriers had intellectual disability. Fluid intelligence scores were compared between unaffected carriers and controls to assess potential subclinical or undiagnosed intellectual disability; scores were not significantly different between the groups (beta = −.32, p = .054; Figure S1).

Pathogenic variants in UKB were distributed across 34 genes, 13 of which overlapped with those found in GEL. Thirty‐four variants were found in multiple individuals, whereas 78 (70%) were singletons (Table S2). When assessing variant types, 211 subjects carried a missense (71%), 44 a nonsense (15%), 11 a frameshift (4%), and 315 an intronic/splice (10%) variant. As in GEL, there was no difference in the proportion of missense variants between epilepsy cases and controls who carried pathogenic variants (OR = .80, p = .77).

Using a broad epilepsy definition, 796 individuals in UKB (.2%) had a diagnosis of epilepsy, yielding a penetrance estimate of 9.8% and a relative risk of 14 (vs. the general population prevalence of .7%). Among carriers of nonsense or frameshift variants, penetrance was 16.8% (three cases, 52 controls). Penetrance estimates represent the average across all variants; most variants were singletons, and only four identical variants were found in cases and controls across GEL and UKB.

PRS as modifier of penetrance and severity

We assessed whether common epilepsy risk variants, quantified as PRS, modify epilepsy variant penetrance. Among people carrying a pathogenic variant, individuals with epilepsy had significantly higher PRS compared to people without epilepsy (OR = 1.84, p = .047; Figure 1A). Subjects in the highest third of the PRS distribution had an eightfold higher risk of having epilepsy compared to the lowest quantile (OR = 8.1, p = .04; Figure 2). The area under the receiver operating characteristic curve for distinguishing epilepsy cases from noncases among pathogenic variant carriers was .80 (95% confidence interval = .67–.93; Figure 1B), suggesting fair predictive ability. We then compared PRS values across groups. Among individuals with epilepsy, those with a pathogenic variant had similar PRS to those without a pathogenic variant (OR = .80, p = .38). Unaffected variant carriers had a nonsignificantly lower PRS compared to controls without a pathogenic variant (OR = .72, p = .069). Epilepsy cases with and without a pathogenic variant had a higher PRS compared to controls without a variant (OR = 1.91, p = .019 and OR = 1.43, p < 2 × 10^−16^, respectively; Figure 3A; see Table S3 for statistics of all group comparisons).

Replication in the UKB cohort showed no significant PRS difference between epilepsy cases and unaffected controls who both carried a pathogenic variant (OR = .35, p = .086; Figure 3B). Within UKB, epilepsy cases with a pathogenic variant had a lower PRS than both controls (OR = .33, p = .032) and epilepsy cases without a pathogenic variant (OR = .20, p = .007). Unaffected variant carriers also had a nonsignificantly lower PRS compared to controls without a pathogenic variant (OR = .87, p = .12; see Table S4 for statistics of all group comparisons).

DISCUSSION

Our results suggest that the penetrance of pathogenic epilepsy variants may be lower than commonly assumed and modified by common risk variants. This has important implications for the interpretation of variant pathogenicity. Under the assumption of high or full penetrance, it is common practice to label variants as benign or of uncertain significance if they are present in multiple unaffected population controls, or in unaffected parents when using a trio whole exome sequencing approach. ^11^ , ^12^ However, our study's low average penetrance warrants caution in labeling variants as benign when unaffected family members or population controls harbor the same variant. Conversely, finding a pathogenic variant might be insufficient to diagnose a specific epilepsy syndrome, especially without considering the combined effect of common epilepsy risk variants. Potentially, a wide range of variants might be reclassified if assumptions about penetrance are reconsidered to improve diagnostic yield. Furthermore, our findings could have relevance for genetic counseling in carriers of established pathogenic variants, where incorporating polygenic risk might refine risk estimates for their children.

Our results suggest that PRS modifies penetrance of pathogenic epilepsy variants. In GEL, we found that an increased PRS was associated with higher epilepsy risk, whereas a decreased PRS was protective in people carrying a pathogenic variant. This is in line with a recent family‐based study, which found that PRS was associated with increased epilepsy risk and severity among family members carrying the same pathogenic variant. ^6^ Similarly, a recent study on hypertrophic cardiomyopathy demonstrated that PRS influenced rare variant penetrance and adverse outcomes, suggesting that the same interplay between rare and common variants might generalize to other diseases. In contrast to GEL, we observed a lower PRS in people with epilepsy and a pathogenic variant in UKB compared to people without a pathogenic variant, and also compared with those with a variant but without epilepsy. This discrepancy may reflect differences in ascertainment bias between the two cohorts. GEL specifically recruited people with early onset and syndromic epilepsy such as DEE, which could explain why 13 of 14 epilepsy cases with a pathogenic variant had severe epilepsies. In contrast, we found no severe epilepsy cases in UKB, which might be related to inclusion at 40–69 years of age and screening for capacity to understand informed consent, which likely excluded individuals with intellectual disability. Additionally, participation bias in UKB may have led to an overrepresentation of individuals with above‐average intelligence and health outcomes. ^24^ Consequently, severe epilepsy cases were likely excluded in UKB, and those remaining are most likely milder. The observed low PRS in UKB cases with a pathogenic variant suggest that a pathogenic variant could be sufficient to cause mild forms of epilepsy, whereas a combination of a pathogenic variant and a high PRS might result in severe epilepsies like DEE.

Our results should be interpreted in the light of several limitations. Due to the rare nature of individual pathogenic variants, we were not able to calculate penetrance estimates per specific variants. To increase robustness and statistical power, we grouped all variants in DEE genes together to assess average penetrance and influence of PRS. However, it is likely that different variants in different genes have a different penetrance. We defined epilepsy variants based on a curated list of DEE genes and established pathogenic epilepsy variants from the ClinVar database, which uses rigorous American College of Medical Genetics criteria for classification. However, we acknowledge that databases like ClinVar are not immune to errors, and some variants may later be reclassified as benign as new evidence emerges. ^25^ Variant misclassification could have resulted in decreased penetrance estimates. Additionally, we note that GEL and UKB have heterogeneity in phenotyping methods. Cases in the GEL cohort were diagnosed by neurologists using strictly defined criteria. However, epilepsy diagnosis in UKB is limited to retrospective ICD codes, which is prone to misclassification. The UKB recruitment of individuals aged 40–69 years may have limited the ability to capture childhood onset diseases. For example, childhood onset epilepsy that goes into remission by adulthood might have been missed at the time of inclusion. Conversely, there could be false‐positive labeling of an epilepsy ICD code in cases with provoked seizures and other paroxysmal events. This is in line with previous GWAS and PRS studies using ICD codes to classify epilepsy, which show a strong dilution of signal compared to clinical epilepsy cohorts. ^19^ , ^21^

CONCLUSIONS

In summary, although our study has inherent limitations, it suggests that the penetrance of pathogenic epilepsy variants is lower than previously assumed and influenced by polygenic background. These findings have implications for clinical variant interpretation and underscore the importance of integrating common genetic variation into the interpretation of variant pathogenicity and risk assessment. We hope this work encourages further studies to refine gene‐ and variant‐specific penetrance estimates and unravel how the full spectrum of genetic variation shapes epilepsy risk and severity. Furthermore, our findings may extend to other disorders in which both rare and common variants contribute to disease susceptibility, underscoring the broader relevance of integrating comprehensive genetic risk assessment into clinical practice.

AUTHOR CONTRIBUTIONS

Remi Stevelink and Bobby P. C. Koeleman designed and conceptualized the study. Remi Stevelink, M. Martijn Piet, Yorgos Bos, and Ruben van ‘t Slot analyzed the data and performed quality control procedures. Bobby P. C. Koeleman and Kees P. J. Braun supervised the study. Remi Stevelink wrote the first draft of the manuscript. All authors interpreted the data, read and revised the manuscript, and approved the final version of the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information