Clinical whole genome sequencing in pediatric epilepsy: Genetic and phenotypic spectrum of 733 individuals

Abstract

Objective

A large proportion of pediatric epilepsies have an underlying genetic etiology. Limited studies have explored the efficacy of whole genome sequencing (WGS) in a clinical setting. Our academic–clinical center implemented clinical whole exome sequencing (WES) in 2014, then transitioned to WGS from 2015. We report the diagnostic yield, genetic and phenotypic findings, and prognostic factors following WGS/WES in pediatric epilepsy.

Methods

The cohort included 733 families with pediatric epilepsy who received clinical WGS/WES between 2014 and 2022. WGS/WES was performed at the Genomic Medicine Center Karolinska for Rare Diseases and analyzed at the Center for Inherited Metabolic Diseases at Karolinska University Hospital. Phenotypic information was extracted from referrals and medical records. Genetic and phenotypic data were analyzed using descriptive statistics, and univariable and multivariable analyses.

Results

The median age at seizure onset was 9 months. Developmental delay and/or intellectual disability (DD/ID) was observed in 61.3% of the cohort; 38.1% of individuals received an International League Against Epilepsy epilepsy syndrome diagnosis. WGS/WES was performed in 640 (87.3%) and 143 (19.5%) families, respectively, totaling 2029 individuals. A molecular diagnosis was identified in 278 of 733 individuals (37.9%), including 51 of 211 individuals analyzed more than once (24.2% of reanalyzed cases). Independent predictors for receiving a genetic diagnosis included female sex (adjusted odds ratio [aOR] = 1.8, 95% confidence interval [CI] = 1.3–2.4, p < .001), neonatal seizure onset (aOR = 2.5, 95% CI = 1.6–4, p < .001), mortality (aOR = 2.2, 95% CI = 1.3–4.0, p = .0048), and an ID/DD/developmental and epileptic encephalopathy (DEE) diagnosis (aOR = 1.8, 95% CI = 1.2–2.5, p = .0019). The strongest independent predictor of ID/DD/DEE was microcephaly (aOR = 7.8, 95% CI = 2–53, p = .0099). In the solved cohort, gene group did not predict cognitive outcome.

Significance

Clinical WGS is an effective diagnostic tool in pediatric epilepsy. We identified female sex as a novel prognostic factor for receiving a genetic diagnosis and highlight the value of reanalyzing previously unsolved cases to improve diagnostic yield.

Key points.

• We identify female sex as a novel potential prognostic factor for receiving a genetic diagnosis.

• Reanalysis of WGS data in a subset of unsolved individuals yielded a diagnosis in nearly one quarter of those reanalyzed.

• A multidisciplinary and collaborative workflow is central to successfully implement WGS in a clinical setting.

• Suspected genetic epilepsies frequently present with complex and broad phenotypes.

• Distinct phenotypic trends were identified through grouping etiologies according to pathophysiological mechanisms, denoted "gene groups".

1. INTRODUCTION

Approximately 60%–70% of epilepsies have a genetic etiology. ¹ Whole exome sequencing (WES) and whole genome sequencing (WGS) have revolutionized the capacity to detect rare monogenic causes of epilepsy, with yields of up to 50%. ² , ³ Monogenic epilepsies give rise to syndromes that fall under the rare or ultrarare (affecting fewer than 1 in 2000 ⁴ /1 in 50 000, ⁵ respectively) disease categories; however, collectively, monogenic pediatric epilepsies present in one in 2120 live births,⁶ representing one of the more “common” rare disease groups.

An accurate and timely genetic diagnosis puts an end to the diagnostic odyssey, can inform prognosis and treatment ⁷ , ⁸ and is cost‐effective. ⁹ , ¹⁰ WGS enables a comprehensive genomic evaluation, where data can be reinterrogated at later dates to investigate novel epilepsy genes and etiologies. Identifying and disentangling the pathogenicity of variants of interest from WGS/WES requires collaboration across multiple disciplines. Prior studies have reported the outcomes of WGS/WES in pediatric epilepsy in a research setting; however, few report findings from a purely clinical workflow, and even fewer where WGS is the primary method applied. At Karolinska University Hospital in Stockholm, WGS was integrated into clinical practice in 2015 through an academic–clinical partnership at Genomic Medicine Center Karolinska for Rare Diseases (GMCK‐RD), ¹¹ which fosters a collaborative, multidisciplinary workflow. A large number of individuals with pediatric epilepsy have received WGS through GMCK‐RD from 2015 to 2022, offering the unique opportunity to retrospectively investigate the genetic and phenotypic spectrum in this cohort using real‐world data.

2. MATERIALS AND METHODS

2.1. Cohort

A total of 733 probands from 710 families (total = 2029 individuals) who received WGS and/or WES at GMCK‐RD and were analyzed at the Center for Inherited Metabolic Diseases (CMMS) at Karolinska University Hospital, Stockholm, Sweden, from 2014 to 2022 were included in the study. WGS referrals from the treating clinician had to meet one or more of the following clinical criteria: suspicion of epileptic encephalopathy, drug‐resistant seizures, abnormal development, or other associated neurological symptoms. Yet, some cases turned out to have self‐limited epilepsy. CMMS clinicians reviewed referrals to limit WGS to individuals with suspected genetic etiology. All individuals had childhood seizures of unknown etiology following clinical evaluation and fulfilled an epilepsy diagnosis in accordance with International League Against Epilepsy classifications. ¹² , ¹³ , ¹⁴ , ¹⁵ A small number of patients who received WES prior to 2014 during the pilot phase of clinical next generation sequencing integration were included in the study. Referrals were received from all over Sweden, and consent for WGS/WES was obtained according to clinical protocol. Ethical approval was obtained from the Swedish ethical review board (Dnr 2022‐06033‐01 and Dnr 2008/351‐31). A subset of the current cohort was previously published without detailed phenotypic analyses, in a broader study of GMCK‐RD patients. ¹¹

2.2. Clinical WGS workflow

WGS has been offered as a clinical test from 2015 through GMCK‐RD, in an academic–clinical partnership between the Clinical Genomics facility at the Science for Life Laboratory and three specialized clinics, including CMMS, Clinical Genetics and Genomics, and Clinical Immunology at Karolinska University Hospital. The multidisciplinary and specialized working environment at GMCK‐RD promotes effective integration of WGS into the clinical workflow across a broad range of disease groups. Epilepsy patients are referred to CMMS for WGS investigations. Samples are processed at CMMS, then genomic DNA is sent to the Clinical Genomics facility at the Science for Life Laboratory for WGS. The raw WGS data are processed using the Mutation Identification Pipeline (MIP), a bioinformatic pipeline composed of custom‐developed, open‐source tools, as well as pre‐existing tools, for variant calling, annotation and prioritization. ¹¹ MIP generates a rank score for each variant, which summates multiple parameters including Mendelian inheritance pattern, rarity, conservation, and predicted effect on protein. The prioritized/ranked WGS data are uploaded to our in‐house‐developed, user‐friendly web browser interface Scout. ¹⁶ WGS data are then reviewed by our multidisciplinary team including clinicians specialized in epileptology, neurology, and metabolic disorders, clinical geneticists, molecular biologists, biochemists, and bioinformaticians. Analysis follows an individualized, multitiered approach based on phenotypic and molecular findings. MIP is continually optimized with updated versions; thus, clinical investigations have expanded from assessing single nucleotide variants (SNVs) and insertion deletions to also include copy number variants and other structural variants (SVs), uniparental disomy, and repeat expansions. Clinical filters are applied to reduce secondary findings and difficult to interpret variants. Clinical filters include disease gene panels, exonic, splice junction, and disease‐associated RNA variants, as well as noncoding variants in compatible disease genes. All filtered variants are assessed by a specialist clinician and molecular biologist. All potentially disease‐causing variants are reviewed at weekly multidisciplinary team meetings. If they have not already been performed, individuals lacking molecular findings may then be investigated with customized Human Phenotype Ontology (HPO) panels. In persisting negative cases where the patient/legal guardian has provided consent, clinical filters are removed and the entire genome, including noncoding variants, is assessed in the research track. Reanalysis using updated clinical panels (which may include additional disease panels if the individual's phenotypic presentation has evolved since the prior analysis) may be performed upon request from the referring clinician or from CMMS clinicians. Results are reported back to the referring clinician. For positive results, including research findings with sufficient evidence of pathogenicity, genetic counseling is offered and, where appropriate, consultation with a CMMS clinician (Figure 1). For detailed methodology on the GMCK‐RD workflow and history, see Stranneheim et al. ¹¹ and Lindstrand et al. ¹⁷

FIGURE 1.

Center for Inherited Metabolic Diseases clinical whole genome sequencing (WGS) workflow. The multidisciplinary team is involved in all steps of the analytical process. Dashed arrows/steps 5B–C and 7 represent investigations performed in a subset of individuals where indicated. HPO, human phenotype ontology; IEM, inborn errors of metabolism; MIP, Mutation Identification Pipeline; MLPA, multiplex ligation‐dependent probe amplification. Created in BioRender (Henry [2025], https://BioRender.com/a05d584).

2.3. Data collection

Medical histories outlined in referrals and from clinician interactions with CMMS available at the time of WGS/WES investigations were retrospectively reviewed and transcribed to HPO ¹⁸ terms. Medical records corresponding to the period of analysis by CMMS were reviewed for intellectual disability (ID) or developmental delay (DD) and magnetic resonance imaging (MRI) reports, when this information was not explicitly stated at referral. See Appendix S1 for classifications for ID, DD, seizure onset ages, MRI abnormalities, and epilepsy syndromes. Mortality status was reviewed in all cases up until April 18, 2024. Other traits routinely asked for in clinical workup were regarded as absent when not stated at referral. Many cases were referred/investigated at an early age, when some epilepsy syndromes cannot be firmly recognized. Thus, medical records after the investigation period were reviewed for syndrome classification when an epilepsy syndrome diagnosis was suspected but could not be confirmed by the information provided at the time of investigation.

2.4. Statistics

Cohort characteristics were described using descriptive statistics. Univariable analyses with the chi‐squared test and odds ratio (OR) were performed across the following dependent variables: genetic diagnosis and ID/DD/developmental and epileptic encephalopathy (DEE). Fisher exact test and OR analyses were used when chi‐squared frequency assumptions were not met. Significant (p < .05 and confidence interval [CI] excluding 1) univariable variables, as well as geographic origin and sex to control for potential bias, were included in a two‐sided multivariable logistic regression model. If high multicollinearity was detected, only the variable with the highest OR on univariable analysis was included in the multivariable model. Individuals with promising variants of uncertain significance (VUS) were considered unsolved for descriptive statistics/analyses. All analyses were performed using R free software version 4.4.1.

3. RESULTS

3.1. Clinical characteristics

A total of 733 individuals with childhood onset epilepsy from 710 families underwent clinical WES/WGS at CMMS up until 2022; 45.3% of the cohort were female. The median age at referral was concentrated around early childhood at 4.0 years (range = 2 days–54.7 years; Table S1). The median seizure onset age was 9 months (range = 1.0 day–17.0 years). A family history of epilepsy or febrile seizures in first‐degree relatives (FDR) was described in 108 individuals (14.7%). Consanguineous unions were reported in 38 (5.2%) individuals' parents. A specific epilepsy syndrome diagnosis could be made in 280 (38.2%) individuals. Neurodevelopmental comorbidities were frequent, with ID/DD observed at referral in 449 individuals (61.3%), autistic behavior in 132 individuals (18.0%), and hyperactivity in 61 individuals (8.3%). Sixty‐eight individuals were deceased (9.3%) at the time of review. MRI abnormalities were identified in 260 of 680 individuals (38.2%) for whom MRI data were available (Figure 2A). Specific DEEs made up the majority of syndrome diagnoses (219 individuals, 29.9% of whole cohort; Figure 2B). The Stockholm (405/733, 55.3%) and external (328/733, 44.7%) cohorts were largely similar, but with some differences; the Stockholm cohort tended to be older at referral (median = 5.7 years vs. 3.9 years for external patients) and at seizure onset (median = 10 months vs. 7.0 months for external patients; Figure S1, Table S1).

FIGURE 2.

Whole cohort phenotypic traits and syndromes. (A) Human Phenotype Ontology (HPO) traits and sex, malformation of cortical development (MCD), any brain malformation, mortality, and epilepsy syndrome diagnosis. (B) Epilepsy syndromes. CAE, childhood absence epilepsy; DD, developmental delay; DEE, developmental and epileptic encephalopathy; EEM, epilepsy with eyelid myoclonia; EE‐SWAS, epileptic encephalopathy with spike‐and‐wave activation in sleep; EIDEE, early infantile DEE; EIMFS, epilepsy of infancy with migrating focal seizures; EMA, epilepsy with myoclonic absences; EMAtS, epilepsy with myoclonic atonic seizures; GEFS+, genetic epilepsy with febrile seizures plus; GGE, generalized epilepsy epilepsy; ID, intellectual disability; IESS, infantile epileptic spasms syndrome; JME, juvenile myoclonic epilepsy; LGS, Lennox–Gastaut syndrome; LKS, Landau–Kleffner syndrome; MEI, myoclonic epilepsy of infancy; MRI, magnetic resonance imaging; PME, progressive myoclonus epilepsy; SeLEAS, self‐limited epilepsy with autonomic seizures; SeLECTS, self‐limited epilepsy with centrotemporal spikes; SeLIE, self‐limited infantile epilepsy; SeLNE, self‐limited neonatal epilepsy; SHE, sleep‐related hypermotor epilepsy.

3.2. Genetic analysis/process

Of the 733 individuals with pediatric epilepsy who underwent WGS/WES at CMMS, 640 (87.3%) received WGS, 143 (19.5%) received WES, and 50 (6.8%) received both WGS and WES. A total of 2029 individuals, including biological parents and affected siblings, were sequenced. A total of 1804 received WGS, 380 received WES, and 155 received both WES and WGS. The majority of cases were sequenced as trios (566/710 families, 79.7%), followed by singletons (77/710, 10.8%), duos (42/710, 5.9%), quartets (25/710, 3.5%), and a quintet (1/710, .1%). All individuals except two (99.7%) received an epilepsy panel, where the two excepted individuals had a novel epilepsy gene identified during the pilot phase (previously published ¹⁹ ). A large number of individuals were also investigated with inborn errors of metabolism (IEM) panels (469, 64.0%) and individualized HPO panels (267, 36.4%). Most individuals were solved on the epilepsy panel (222/278, 79.9%), followed by HPO (19/278, 6.8%), research (16/278, 5.8%), IEM (12/278, 4.3%), and other (9/278, 3.2%) panels (Table S2).

The median time from seizure onset to a genetic diagnosis was 2.5 years (range = 35.9 days–53.1 years) and 4.1 months (range = 11 days–8.2 years) from referral to diagnosis. There was a positive trend of the total cases referred each year, nearly doubling from 56 individuals in 2014 to 109 in 2022 (Figure S2). The year 2014 had the highest yield, where 50.9% of all cases were solved. Yield decreases across the years; this likely reflects the ability to offer WGS to a broader selection of patients in more recent years, not just those with the most severe symptoms.

3.3. Diagnostic variants

A genetic diagnosis was identified in 278 of 733 individuals (37.9%; Table S2), including three individuals with digenic etiologies, totaling 281 etiologies. There were 302 disease‐causing variants identified across 130 unique genes (Table S3). SCN1A was the most enriched disease‐causing gene (Figure 3A). The majority of variants were SNVs or indels (280/302, 92.7%) and most frequently missense (162/302, 53.6%). Copy number and large SVs made up the remaining 22 (7.3%) causative variants (Figure 3B). The PRRT2 c.649dupC was the most recurrent variant in 12 individuals. Of the 281 etiologies, most variants were autosomal dominant (AD) de novo (129/281, 45.9%), followed by AD inherited (41/281, 14.6%) and autosomal recessive homozygous (32/281, 11.4%). Six variants (2.1%) were inherited from mosaic parents: four with AD inheritance and two X‐linked recessive (Figure 3C). Phenotype associations of the top recurrent genes revealed that KCNT1 had a high incidence of mortality, abnormal MRI findings, and ID/DD/DEE. STXBP1, MECP2, CDKL5, SCN1A, and NEXMIF also had a high ID/DD/DEE burden (Figure 4A). Deep intronic variants were identified as causative in two individuals. A deep intronic de novo SCN8A c.706 + 238G>A variant in the HUGO Gene Nomeculature Committee primary transcript NM_014191.3 led to a protein change in isoform 3 NM_001330260 (c.697G>A, p.(Val233Ile)). Fibroblast cDNA analysis confirmed expression of both transcript isoforms (previously reported as disease causing ²⁰ ). A homozygous WWOX c.107 + 119C>G variant was shown to lead to the inclusion of the first intron in transcript NM_016373, leading to a premature stop codon after cDNA analysis. Four large SVs were also identified: one de novo complex rearrangement in chromosome 2 including the genes SCN1A, SCN2A, and SCN3A; a de novo ~1.61‐Mb interstitial deletion on chromosome 16; a 3.98‐Mb de novo deletion on chromosome 8 corresponding to 8p32.1 microdeletion syndrome; and a 9.7‐Mb de novo duplication in chromosome 15 (Table S9).

FIGURE 3.

Genetic etiologies in solved individuals. (A) Present in ≥3 individuals across gene groups. (B) Variant types. (C) Mode of inheritance. AD, autosomal dominant; AR, autosomal recessive; comp het, compound heterozygous; homoz, homozygous; mat, maternally inherited; UPD, uniparental disomy; XD, X‐linked dominant; XR, X‐linked recessive. (D) Gene groups. Cell growth: cell growth, division, and proliferation related. Cell structure/homeostasis: cell structural integrity/homeostasis. DNA related: transcription, DNA repair, and chromatin remodeling. Other: other/multiple function proteins. CNV, copy number variant; NT, neurotransmitter; SV, structural variant.

FIGURE 4.

Human Phenotype Ontology (HPO) traits and specific developmental and epileptic encephalopathy (DEE) diagnosis, sex, and mortality. (A) Across the 10 highest yield solved genes (n = 106 etiologies). (B) Across gene groups in solved individuals. Darker red indicates higher proportion of specified trait. Cell growth: cell growth, division, and proliferation related. DNA related: transcription, DNA repair, and chromatin remodeling. Other: other/multiple function proteins. DD, developmental delay; ID, intellectual disability; MRI, magnetic resonance imaging; NT, neurotransmitter.

3.4. Gene groups

Genetic findings were subdivided into 11 different gene groups according to gene function. The most enriched gene group was the ion channel genes (24.8%), followed by synapse‐related genes (14.0%) then protein biosynthesis/degradation‐related genes (10.8%; Figure 3A,D, Table S4). Mitochondrial genes (1.4%) showed the highest rates of mortality and neonatal seizure onset, whereas ion channel and synapse genes had a lower burden of the displayed comorbidities (Figure 4B).

3.5. Candidate variants

Sixty‐four promising VUS were identified in 58 individuals and encompassed 50 genes. These VUS were reported back to the referring clinician, as they were rare, predicted to be damaging and/or affect a conserved amino acid, and were associated with a disease spectrum fitting the patient's phenotype. These patients were, however, considered unsolved, as these VUS lacked sufficient evidence to upgrade their status according to the American College of Medical Genetics and Genomics guidelines. The gene group distribution in the VUS was similar to that of the solved variants, but with a higher proportion of mitochondrial function genes (Figure S3, Table S10).

3.6. Diagnostic yield in subcohorts

The diagnostic yield of WGS, including individuals with prior negative WES investigations, was 35.0% (224/640). The diagnostic yield in the initial, highly prioritized cases who received WES was 37.8% (54/143). Fifty individuals received WGS after negative WES, resulting in 14 additional diagnoses. Four of these diagnoses would not have been identified/resolved with WES alone, representing an additional yield of 8% (4/50) of WGS over WES. The four cases included a large (>5 Mb) duplication, a complex SV, and two relatively large (~1800 and 4000 bp) deletions including 1–3 exons, respectively, with noncoding breakpoints. The other 10 individuals with negative WES had etiologies identified on WGS due to research track investigations (n = 5), an updated panel including the newly published gene (n = 4), or receiving a different panel (n = 1). All 10 individuals harbored exonic variants that could theoretically be resolved on WES, although this is assuming good coverage (Table S2).

Diagnostic yield was inversely related to seizure onset age: 60.4% (58/96) for neonatal, 46.4% (143/308) infantile, and 24.6% (72/293) childhood onset. Syndromes with the highest yield were myoclonic epilepsy in infancy (MEI; 1/1, 100%), Dravet syndrome (17/18, 94.4%), epilepsy of infancy with migrating focal seizures (EIMFS; 8/9, 88.9%), self‐limited neonatal epilepsy (SeLNE; 8/9, 88.9%), self‐limited infantile epilepsy (SeLIE; 15/17, 88.2%), and early infantile DEE (EIDEE; 12/15, 80%). The overall yield for specific DEE syndromes was 44.3% (97/219; Table 1).

TABLE 1.

Diagnostic yield ^a in individuals with an epilepsy syndrome diagnosis.

Syndrome	Yield	Genes (total individuals)
Early infantile DEE	12/15 (80%)	STXBP1 (5), BRAT1 (1), EIF2B4 (1), GNAO1 (1), GRIN1 (1), KCNQ2 (1), KCNT1 (1), PDK1 (1)
Infantile epileptic spasms syndrome	51/132 (38.6%)	STXBP1 (6), CDKL5 (4), WWOX (4), ALG6 (2), ELFN1 (2), KCNQ2 (2), SCN8A (2), UBA5 (2), UFSP2 (2), AKT3 (1), ALG13 (1), ATP7A (1), CACNA1E (1), CDK19 (1), CLTC (1), COL4A1 (1), COL4A2 (1), DNM1 (1), DPM1 (1), FGF13 (1), FOLR1 (1), FOXG1 (1), GABRB1 (1), GRIN2D (1), HSD17B10 (1), KIF1A (1), MT‐ATP6 (1), PPFIBP1 (1), PTPN23 (1), SMC1A (1), STAMBP (1), TBL1XR1 (1), TSC2 (1), TUBA1A (1)
Dravet syndrome	17/18 (94.4%)	SCN1A (16), CHD2 (1), b GABRG2 (1) b
Epilepsy of infancy with migrating focal seizures	8/9 (88.9%)	KCNT1 (2), SLC12A5 (2), ATP1A1 (1), GABRB3 (1), PIGQ (1), SCN2A (1)
Epilepsy with myoclonic atonic seizures	8/26 (30.8%)	NEXMIF (2), CHD2 (1), PPP3CA (1), SLC6A1 (1), SNAP25 (1), STXBP1 (1), SYNGAP1 (1)
Lennox–Gastaut syndrome	5/12 (41.7%)	STX1B (2), COL4A2 (1), DYNC1H1 (1), SCN8A (1)
Self‐limited epilepsy with centrotemporal spikes	0/2 (0%)	–
(D)EE‐SWAS	0/16 (0%)	–
Landau–Kleffner syndrome	0/2 (0%)	–
Self‐limited neonatal epilepsy	8/9 (88.9%)	KCNQ2 (7), SCN2A (1)
Self‐limited infantile epilepsy	15/17 (88.2%)	PRRT2 (11), SCN8A (3), SCN2A (1)
Myoclonic epilepsy in infancy	1/1 (100%)	CSNK2B (1)
Genetic epilepsy with febrile seizures plus	5/13 (38.5%)	SCN1A (3), HCN1 (1), STX1B (1)
Epilepsy with eyelid myoclonia	1/2 (50%)	SYNGAP1 (1)
Childhood absence epilepsy	1/6 (16.7%)	SLC2A1 (1)
Epilepsy with myoclonic absences	0/1 (0%)	–
Self‐limited epilepsy with autonomic seizures	1/3 (33.3%)	OPHN1 (1)
Sleep‐related hypermotor epilepsy	2/3 (66.7%)	CHRNA4 (2)
Juvenile myoclonic epilepsy	0/3 (0%)	–
Progressive myoclonic epilepsy	2/3 (66.7%)	EPM2A (1), GOSR2 (1)

Abbreviations: (D)EE‐SWAS, (developmental) epileptic encephalopathy with spike‐and‐wave activation in sleep; DEE, developmental and epileptic encephalopathy.

^a Total syndromes are per diagnoses (not individuals) and includes 11 individuals with a history of two epilepsy syndrome diagnoses and 1 individual with a history of three epilepsy syndrome diagnoses.

^b Genes identified in digenic disease.

A diagnosis was identified on first analysis in 227 of 278 solved individuals (81.7%).

Reanalysis was performed in 211 unsolved individuals from 2015 to June 28, 2024, resulting in an additional 51 diagnoses, or 24.2% of reanalyzed cases. Reanalyzed cases received a diagnosis a median of 2.8 years following initial referral (range = 83 days–8.2 years) and were most frequently solved due to research track investigations (12/51, 23.5%), receiving an updated panel containing a new disease gene (9/51, 17.6%), and receiving a different disease panel (6/51, 11.8%; Table 2).

TABLE 2.

Reason solved on reanalysis.

Reason solved on reanalysis	Count (%)
Research track	12/51 (23.5%)
Updated panel including newly described gene	9/51 (17.6%)
Different panel	6/51 (11.8%)
Unclear	5/51 (9.8%)
WGS following negative WES	4/51 (7.8%)
Updated panel a	4/51 (7.8%)
Experimental validation b	3/51 (5.9%)
Newly published phenotype	2/51 (3.9%)
Segregation validation	2/51 (3.9%)
Variant identified in another solved individual	2/51 (3.9%)
Long‐read WGS	1/51 (2%)
Updated MIP	1/51 (2%)

Abbreviations: MIP, Mutation Identification Pipeline; WES, whole exome sequencing; WGS, whole genome sequencing.

^a Panel expanded to include a gene that was published as disease causing at the time of previous analysis.

^b Variant validated on reanalysis with mRNA analysis/Sanger sequencing.

3.7. Independent predictors of genetic diagnosis and neurodevelopmental course

Independent predictors for receiving a genetic diagnosis and developing ID/DD/DEE on multivariable logistic regression analyses are presented in Table 3. Univariable analyses and sex‐related analyses are available in Tables S5–S7. Independent predictors for receiving a genetic diagnosis on multivariable analyses included neonatal seizure onset (adjusted OR [aOR] = 2.5, 95% CI = 1.6–4.0, p < .001), mortality (aOR 2.2; 95% CI = 1.3–4.0, p = .0048), female sex (aOR = 1.8, 95% CI = 1.3–2.4, p < .001,) and an ID/DD/DEE diagnosis (aOR = 1.8, 95% CI = 1.2–2.5, p = .0019). Independent positive predictors of an ID/DD/DEE diagnosis across the whole cohort included microcephaly (aOR = 7.8, 95% CI = 2–53, p = .0099), movement abnormality (aOR = 3.8, 95% CI = 1.9–8.7, p < .001), abnormal muscle tone (aOR = 2.3, 95% CI = 1.3–4.3, p = .0087), a genetic diagnosis (aOR = 1.7, 95% CI = 1.2–2.5, p = .0082), and infantile seizure onset (aOR = 1.6, 95% CI = 1.1–2.3, p = .016). A family history of epilepsy or febrile seizure in FDR was a significant negative predictor for ID/DD/DEE (aOR = .41, 95% CI = .25–.67, p < .001). The presence of a movement abnormality was a significant positive predictor for ID/DD/DEE in the solved cohort (aOR = 8.4, 95% CI = 1.6–156, p = .045), whereas negative predictors included a history of epilepsy or febrile seizure in FDR (aOR = .26, 95% CI = .1–.63, p = .0032) and dominant inherited variants (aOR = .36, 95% CI = .15–.88, p = .024). Univariable analysis identified the gene groups protein biosynthesis/degradation related and transporters as significant positive predictors for ID/DD/DEE and ion channels as a negative predictor; however, these associations did not persist on multivariable analysis (Table S6).

TABLE 3.

Multivariable logistic regression for dependent variables genetic diagnosis and ID/DD/DEE including all variables significant on univariable analysis.

Dependent variable ➔ Independent variable↓	Whole cohort: Genetic diagnosis, aOR (95% CI), p	Whole cohort: ID/DD/DEE, aOR (95% CI), p	Solved cohort: ID/DD/DEE, aOR (95% CI), p
Stockholm	1.3 (.95–1.8), .1	1.1 (.75–1.6), .66	1.2 (.58–2.4), .66
Female	1.8 (1.3–2.4), <.001 a	.98 (.68–1.4), .9	1.5 (.75–3), .25
Genetic diagnosis	NA	1.7 (1.2–2.5), .0082 a	NA
Family history in FDR	–	.41 (.25–.67), <.001 a	.26 (.1–.63), .0032 a
Neonatal seizure onset	2.5 (1.6–4), <.001 a	–	–
Infantile seizure onset	–	1.6 (1.1–2.3), .016 a	–
Microcephaly	–	7.8 (2–53), .0099 a	2.1 (.2–47), .57
Abnormal MRI	–	1.2 (.76–2.1), .38	1.1 (.44–2.8), .88
MCD	–	.58 (.08–2.7), .52	NA
Any brain malformation	–	2 (.46–14), .41	1.9 (.43–11), .42
ID/DD/DEE	1.8 (1.2–2.5), .0019 a	NA	NA
Movement abnormality	–	3.8 (1.9–8.7), <.001 a	8.4 (1.6–156), .045 a
Abnormal muscle tone	1.3 (.84–2.1), .23	2.3 (1.3–4.3), .0087 a	2.7 (1–8.9), .064
Deceased	2.2 (1.3–4), .0048 a	1.2 (.57–2.8), .62	–
Dominant inherited	NA	NA	.36 (.15–.88), .024 a
Recessive	NA	NA	1.9 (.64–6.3), .27
Ion channel	NA	NA	.75 (.35–1.6), .45
Protein biosynthesis/degradation related	NA	NA	2.5 (.44–21), .33
Transporters	NA	NA	2.4 (.42–46), .41

Note: Variables that were tested on univariable analysis that were nonsignificant and thus excluded from multivariable analysis are marked with an en dash.

Abbreviations: aOR, adjusted odds ratio; CI, confidence interval; DD, developmental delay; DEE, developmental and epileptic encephalopathy; FDR, first‐degree relatives; ID, intellectual disability; MCD, malformation of cortical development; MRI, magnetic resonance imaging; NA, not applicable/excluded due to high multicollinearity.

^a p < .05 is considered significant; however, if CI includes 1, the result is considered nonsignificant.

3.8. Treatment implications

Genes with the potential for targeted epilepsy treatment or nonspecific epilepsy treatment were annotated according to the American Academy of Neurology classification of evidence across classes I–IV, requiring varying degrees of evidence. Briefly, class I requires the most stringent evidence, including randomized controlled clinical trials and meeting five specific criteria to eliminate bias, whereas class IV includes noncontrolled studies. ²¹ Of the 281 etiologies, 144 (51.2%) were classifiable into one of the four classes, demonstrating the potential to be targeted with precision medicine. Genes that met the stringent class I criteria included 21 etiologies (7.5% of solved etiologies) followed by 21 of 281 (7.5%) in class II, 72 of 281 (25.6%) in class III, and 125 of 281 (44.5%) in class IV. See Appendix S2 for example cases from each class.

4. DISCUSSION

We present the largest clinical WGS cohort study on pediatric epilepsy to date and explore the phenotypic and genetic landscape of this cohort. Individuals presented across a broad phenotypic spectrum, with ID and/or DD as the most frequent comorbidity. A genetic diagnosis was identified in 37.9% of the cohort, including a significant proportion identified upon reanalysis of WGS data. Etiologies were heterogeneous and included a range of neurobiological functions. Female sex, ID/DD/DEE, mortality, and neonatal seizure onset were positive independent predictors for receiving a genetic diagnosis. Approximately 50% of individuals who received a genetic diagnosis may be eligible for precision medicine.

The broad and often complex phenotypes observed in our cohort are typical for early childhood epilepsies, but also reflect the clinical prioritization of more complex/severe cases for WGS. The median seizure onset age of our cohort was relatively young at 9 months, and the majority of specific syndromes were DEEs. Prior studies report an epilepsy syndrome diagnosis in approximately 50% of individuals, ²² , ²³ which is slightly higher than the 38% in our cohort. We did not allocate nonspecific DEE syndromes due to the difficulty of classifying these syndromes in patients who often present very young. However, the high incidence of ID/DD in our cohort indicates that a subset of unclassified individuals would likely fulfill a DEE diagnosis.

The successful implementation of WGS into the clinic is a result of many years of optimization with input from a range of specialists. The specialized team is closely involved in the entire analytical process, including the sharing of results and treatment recommendations, often meeting patients to discuss findings. We have seen a near doubling of individuals being referred over the years and have maintained a high diagnostic yield. A significant phase in the workflow evolution was the transition from WES to WGS in 2015, which provides the opportunity to compare the two modalities (although the WES cohort, n = 143, was considerably smaller than the WGS cohort, n = 640). WES had a higher diagnostic yield than WGS; however, this is likely due to the initial cases having more severe presentations with a high suspicion of genetic etiology. WGS offered a clear advantage in four individuals who received both WES and WGS who harbored SVs that were not detectable, or could not be resolved, on WES. The 8% increased diagnostic yield of WGS over WES identified in the current study closely approximates the 7.2% reported in a recent study of individuals with pediatric epilepsy and previously nondiagnostic WES. ²⁴ A further five individuals who had only WGS had deep intronic variants or SVs that would not be detected on WES.

There was considerable genetic heterogeneity in our cohort as exemplified by the 130 unique disease‐causing genes identified. The highest yield pathogenic genes largely matched those reported in other large pediatric epilepsy cohorts. ²⁵ , ²⁶ , ²⁷ TSC1/2 were underrepresented in our cohort, as individuals with tuberous sclerosis are often directly referred for sanger sequencing of TCS1/2. The pathogenic variant types and models of inheritance were similar to prior reports, with missense variants and AD de novo inheritance most frequently identified. ²² , ²⁷ , ²⁸ Unsurprisingly, ion‐channel genes were the most common gene group, making up one quarter of solved genes. However, most of the other 10 gene groups occurred in relatively similar proportions, highlighting the broad neurobiological functions implicated in epilepsy. We also found considerable phenotypic heterogeneity within specific genetic etiologies. Of the 15 KCNQ2 cases, seven had SeLNE, had two infantile epileptic spasms syndrome, one had EIDEE, and three of five without specific syndromes had comorbid ID. Syndromes associated with SCN2A variants included one individual with EIMFS, two individuals with comorbid ID and/or DD, one with SeLNE, and one with SeLIE. Of the 19 SCN1A variants, 16 were identified in individuals with Dravet syndrome, whereas three other individuals had genetic epilepsy with febrile seizures plus (GEFS+). Interestingly, four of six individuals with causative KCNT1 variants had MRI abnormalities. Three had bleeds/ischemic injuries, which highlights that acquired MRI lesions may not be contraindications to WGS. In a fourth case, MRI showed a migration disorder, which is in line with previously published cases of cerebral malformations in KCNT1 disorder. ²⁹ We also report the first causative genetic etiology in MEI due to a rare, de novo frameshift variant in CSNK2B (Table S9, ID 213). CSNK2B has previously been reported in individuals with infantile onset, predominantly generalized seizures, with neurodevelopmental trajectories ranging from normal to severe ID. ³⁰

Our diagnostic yield of 37.9% falls within the range of 32%–47.5% reported in four smaller comparable WGS studies on pediatric onset epilepsy. ² , ²⁷ , ³¹ , ³² The additional overall yield in our reanalyzed cases was 7.0%, or nearly one quarter of the reanalyzed individuals. Research track investigations and updated epilepsy panels were the most common reasons for uncovering previously obscured diagnoses, highlighting the value in continuously revisiting WGS data as the field advances. Two prior studies including DEEs and individuals with epilepsy and ID have reported similar yield gains of 5.8% ³³ and 10.5%, ³⁴ respectively, after reanalyzing WES data.

The current study validated previously known predictors/associations for receiving a genetic diagnosis: ID/DD/DEE, neonatal seizure onset, ³ , ²² , ²⁷ , ³¹ , ³⁵ and mortality, where an increased risk of mortality has been reported in genetic epilepsies, ³⁶ , ³⁷ especially in DEEs. ³⁸ Interestingly, we found females to be nearly twice as likely than males to receive a genetic diagnosis. Comparable large pediatric epilepsy cohort studies have not reported any differences between sexes. ²² , ²⁷ However, one smaller WES cohort study (n = 45) also found females to be significantly more likely to receive a genetic diagnosis, ³⁹ as did a recent report of 98 patients with DEEs who underwent genetic analyses. ⁴⁰ We hypothesized that X‐linked variants in our cohort may have been driving this significance; however, the effect persisted when X‐linked variants were excluded from the analysis. Sex‐based analyses revealed that solved males had higher mortality rates than females (Table S7, Figure S4). Population studies have reported increased male mortality rates in children with epilepsy ⁴¹ and patients with new onset status epilepticus. ⁴² Increased male mortality rates may result in inflated diagnoses in females. A Scn8a mouse model demonstrated greater mortality resilience in females than males, suggesting a potential neurobiological contribution to these differences. ⁴³

Independent predictors for ID/DD/DEE included features previously associated with poor cognitive outcomes: microcephaly, ⁴⁴ , ⁴⁵ movement abnormality, ⁴⁶ , ⁴⁷ abnormal muscle tone, ⁴⁸ receiving a genetic diagnosis, ³⁶ , ⁴⁹ and infantile seizure onset. ³⁶ The individuals in our cohort with SeLNE, SeLIE, and GEFS+ displayed high yields and often harbored AD inherited variants, which likely contributed to the negative associations we identified between family history of epilepsy/dominant inherited variants and ID/DD/DEE. The gene groups protein biosynthesis/degradation‐related and transporters were significant positive predictors, whereas ion channel genes were negative predictors, for developing ID/DD/DEE on univariable analysis. These variables were not, however, significant independent predictors on multivariable analysis, which may have been due to the small numbers underpowering the analysis.

Understanding the underlying pathophysiology paves the way for precision medicine‐based care and improved seizure control, as is exemplified for the individuals described in Appendix S2. Half our solved cohort showed potential to benefit from targeted therapies, which is in line with emerging evidence on precision medicine. Two recent large studies found a genetic diagnosis impacted treatment and resulted in clinical management changes in 45% ²⁵ and 50% ⁵⁰ of individuals with pediatric epilepsy, respectively.

The strength of this study reflecting the clinical process is also a limitation. Patients were referred for WGS from all over Sweden; thus, the protocol for classifying disease traits and access to medical histories varied by location. To account for some of this variability, we systematically reviewed entire medical records for variables of particular interest (e.g., ID/DD diagnosis); however, this was not feasible for all variables. Patients are increasingly referred for WGS at younger ages; thus, certain comorbidities or syndromes may not yet be evident in these young children. Consequently, these factors likely result in underestimations of the presence of certain disease traits, although predominant traits at referral are usually included. Although the large cohort size may offset these granularities when assessing the larger subcohorts, conclusions for the smaller subcohorts may be less reliable. A prospective study might have been able to account for some of these biases. However, the immense resources needed to implement the infrastructure for clinically integrated WGS meant the current retrospective design was the most feasible.

The current study demonstrates that WGS is an effective diagnostic tool when implemented with a multidisciplinary, integrated academic–clinical workflow. Furthermore, we highlight female sex as a novel predictor for receiving a genetic diagnosis and the importance of reanalyzing previously unsolved cases to improve diagnostic yield.

AUTHOR CONTRIBUTIONS

Olivia J. Henry conceptualized the study, collected and analyzed data, and drafted the manuscript for intellectual content. Tommy Stödberg conceptualized the study, collected and reviewed data, and revised the manuscript for intellectual content. Anna Wedell conceptualized the study, reviewed data, and revised the manuscript for intellectual content. Sofia Ygberg, Michela Barbaro, Nicole Lesko, and Leif Karlsson analyzed patient data and revised the manuscript for intellectual content. Lucía Peña‐Pérez provided bioinformatics support and revised the manuscript for intellectual content. Ann Båvner contributed to data analysis and revised the manuscript for intellectual content. Virpi Töhönen and Anna Lindstrand contributed to the planning of the study and revised the manuscript for intellectual content.

FUNDING INFORMATION

This work was supported by the Swedish Research Council (2023‐02388), the Knut & Alice Wallenberg Foundation (KAW2020.0228), and the Swedish state under the ALF agreement (FoUI‐955096).

CONFLICT OF INTEREST STATEMENT

None of the authors has any conflict of interest to disclose. 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

Data S1.

Table S9.

Table S10.

Henry OJ, Ygberg S, Barbaro M, Lesko N, Karlsson L, Peña‐Pérez L, et al. Clinical whole genome sequencing in pediatric epilepsy: Genetic and phenotypic spectrum of 733 individuals. Epilepsia. 2025;66:2966–2979. 10.1111/epi.18403

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.