A total of 320 cases (trios) with epilepsy without acquired causes (idiopathic epilepsies) were recruited in this study from the Epilepsy Center of the Second Affiliated Hospital of Guangzhou Medical University in China between January 2013 and December 2020. The complete pedigree and clinical data of the probands were collected, including age, gender, age of seizure onset, type, course, and frequency of seizure, family history, therapy, prognosis, general and neurological examination, long-term video EEG, and brain magnetic resonance imaging (MRI). Epilepsy syndromes and epileptic seizures were diagnosed according to the criteria of the Commission on Classification and Terminology of the ILAE (1981, 1989, 2001, 2010, and 2017). Idiopathic generalized epilepsy were diagnosed based on a range of seizure types including absence, spasms, myoclonus, clonic, atonic, tonic, and tonic-clonic seizures, supported by typically generalized discharges on EEG. Idiopathic partial epilepsy had partial seizures and features of bilateral discharge or tendency of generalized discharge. EEG examinations showed focal abnormalities with features of idiopathic epilepsies, including shifting, bilateral, or multiple focal discharges with normal backgrounds. Patients with acquired causes were excluded. All subjects were followed up for at least one year.

This project was approved by the Ethics Committee of the Second Affiliated Hospital of Guangzhou Medical University and was conducted according to the guidelines of the International Committee of Medical Journal Editors regarding patients’ consent for research or participation. Written informed consent was obtained from the patients and their legal guardians.

Peripheral blood samples were obtained from the probands, parents, and other available family members to determine the origin of the identified genetic variants. Genomic DNA was extracted as previously reported (Wang et al., 2018; Shi et al., 2019). Trio-based Whole-Exome Sequencing (WES) was performed with the Illumina HiSeq 2500/4000 platform by BGI-Shenzhen (Shenzhen, China). Deep sequencing data were aligned to the reference GRCh37 build (hg19) and variants were called according to the standard procedures as previously reported (Shi et al., 2019). A case-by-case analytical approach was used to identify candidate causative variants in each trio. Generally, the rare variants with a minor allele frequency < 0.005 were first prioritized in the 1000 Genomes Projects, Exome Aggregation Consortium, and Genome Aggregation Database (gnomAD) (Genomes Project et al., 2015; Karczewski et al., 2020). Next, potentially pathogenic mutations were retained, including frameshift, nonsense, canonical splice site, initiation codon, and missense mutations predicted as being damaging by in silico tools¹. Finally, potential disease-causing variants were screened under following five models: (1) epilepsy-associated gene model; (2) de novo dominant inheritance model; (3) autosomal recessive inheritance model, including compound heterozygous and homozygous variants; (4) X-linked inheritance model; (5) co-segregation model. To identify novel epilepsy-associated genes, the known epilepsy-associated genes (Wang et al., 2017) were not analyzed in the present study. Genes with repetitively identified de novo variants, bi-allelic variants, hemizygous variants, or variants with segregations, were selected for further studies to define the gene-disease association. SHROOM4 appeared as one of the candidate genes with recurrent hemizygous variants in this cohort. All variants in SHROOM4 were annotated based on transcript NM_020717.5. Positive findings and the variant origination were validated by Sanger sequencing.

Protein modeling was conducted via Iterative Threading ASSEmbly Refinement (I-TASSER) software to predict the effect of candidate variants on molecular structure (Zhang, 2008; Yang and Zhang, 2015). Three-dimensional protein structure and hydrogen bonds alteration were visualized and analyzed by using PyMOL 1.7.

I-Mutant server was used to predict protein stability alteration caused by single nucleotide mutations-related amino acid change (Capriotti et al., 2005). Protein stability was measured with free energy change value (DDG, kcal/mol).

To evaluate the genotype–phenotype correlation, SHROOM4 variants were systematically reviewed through PubMed database and human gene mutation database up to December 2021.

IBM SPSS Statistics 19 was used for statistical analysis. Fisher’s exact test was applied to access the allele frequencies of SHROOM4 variants in the cohort of this study and the control populations, and the proportions of missense variants between epilepsy and intellectual disability. A p value of < 0.05 was considered to be statistically significant.

Six hemizygous missense variants in SHROOM4, including c.13C > A/p.Pro5Thr, c.3236C > T/p.Glu1079Ala, c.3581C > T/p.Ser1194Leu, c.4288C > T/p.Arg1430Cys, c.4303G > A/p.Val1435Met, and c.4331C > T/p.Pro1444Leu, were identified in six unrelated cases with idiopathic epilepsy (Figure 1 and Table 1). All of the hemizygous missense variants were inherited from their asymptomatic mothers, consisted with a classical X-linked recessive (XLR) inheritance pattern (Figures 1A,B).

No hemizygote of these variants were found in the controls of gnomAD-all populations, except the variant Ser1194Leu with an extremely low frequency (1/87,776) (Table 2). All of the hemizygous variants were not found in the controls of gnomAD-East Asian populations.

The aggregate frequency of the hemizygous variants in this cohort was significantly higher than that in controls (Table 2), including the gnomAD-all population (6/448 vs. 2/180,876; p = 6.134 × 10^–15), the controls of gnomAD-all population (vs. 1/79,656; p = 4.273 × 10^–13), the gnomAD-East Asian population (vs. 0/13,792, p = 9.386 × 10^–10), and the controls of the gnomAD-East Asian population (vs. 0/6,901, p = 9.462 × 10^–8).

All SHROOM4 variants identified in this study were predicted to be damaging by one of the silico tools (Supplementary Table 1). Protein sequence alignment indicated that five of six variants are located at residues that are highly conserved across species (Figure 1C). The Arg1430 was less conserved but was predicted to be conserved by GERP (score = 4.15) and phyloP (score = 2.907) (Supplementary Table 2). None of the six affected patients had pathogenic or likely pathogenic variants in genes known to be associated with epileptic phenotypes (Wang et al., 2017).

The clinical features of the six cases with SHROOM4 variants were summarized in Table 1. The onset age of seizures ranged from 3 years to 16 years old, with a median age of onset of 5.5 years. The patient of case 1 was diagnosed as childhood absence epilepsy (CAE) characterized by absence seizure and 3 Hz generalized spike-slow waves on EEGs (Figure 2A). The patients of case 3, case 4, and case 5 were diagnosed as generalized epilepsy, including Lennox-Gastaut syndrome (LGS), idiopathic generalized epilepsy (IGE), and childhood myoclonic epilepsy (CME); and the three cases had both generalized and focal discharge features on EEGs (Figures 2C−E). The patients of case 2 and case 6 were diagnosed as partial epilepsy, i.e., benign childhood epilepsy with centrotemporal EEG spikes (BECTS) and idiopathic partial epilepsy (IPE), but their EEGs had generalized discharges (asymmetric) (Figures 2B,F). In a word, these patients mainly present with generalized epilepsy, or idiopathic partial epilepsy with generalized seizures or generalized discharge on EEGs. All patients showed normal development. Their brain MRI were normal. These patients all presented good responses to antiepileptic drug and achieved seizure-free.

The SHROOM4 protein contains two evolutionarily conserved domains, i.e., an N-terminal PDZ domain, and a C-terminal ASD2 domain (Hagens et al., 2006; Yoder and Hildebrand, 2007). In the present study, variant Pro5Thr was located in the N-terminal PDZ domain, while Glu1079Ala, Ser1194Leu, Arg1430Cys, Val1435Met, and Pro1444Leu were located in or near to the C-terminal ASD2 domain (Figure 3A).

The molecular effect of the missense variants was analyzed by using I-TASSER for protein modeling and PyMOL 1.7 for visualization (Figure 3B). Three of the variants, including Glu1079Ala, Ser1194Leu, and Arg1430Cys, changed hydrogen bonds with the surrounding residues. Originally, Glu1079 formed two hydrogen bonds with residues Leu1080 and Lys1083, respectively. When glutamine was replaced by alanine, the hydrogen bonds with Leu1080 was destroyed. Ser1194 formed two hydrogen bonds with His1190. When serine was replaced by leucine, one of the hydrogen bond was destroyed. Arg1430 formed seven hydrogen bonds with residues Asp1371, Glu1423, His1427, Asp1429, and Arg1431, respectively. When arginine was replaced by cystine, five hydrogen bonds with Asp1371, Glu1423, and Asp1429 were destroyed, and only the hydrogen bonds with His1427 and Arg1431 were kept.

The other three variants, including Pro5Thr, Pro1444Leu, and Val1435Met, have no changed in hydrogen bonds, but they were predicted to decrease the protein stability significantly (ΔΔG values < −0.5 kcal/mol) (Figure 3B).

Previously, 12 SHROOM4 variants have been reported, including 3 missense (Hagens et al., 2006; Farwell et al., 2015; Routier et al., 2019), 2 duplications (Froyen et al., 2007; Isrie et al., 2012), and 7 destructive variants (1 nonsense, 3 gross deletion variants (Honda et al., 2010; Armanet et al., 2015; Danyel et al., 2019; Heide et al., 2020), and 3 complex rearrangements (Hagens et al., 2006; Dong et al., 2021)) (Supplementary Table 2). Eleven of the variants associated with intellectual disability (ID), of which 9 were destructive mutations or duplications (Figure 3C). Among the three missense variants, one missense variant was associated with ID (Farwell et al., 2015) and one missense variant was associated with ID and seizures (Hagens et al., 2006). Additional one missense variant (His401Tyr) was reported to be associated with myoclonic-atonic epilepsy (MAE) (Routier et al., 2019). These variants are all hemizygous. In this study, all hemizygous variants were missense variants, which were associated with idiopathic epilepsy without ID (Figure 3A). The proportion of missense variants in epilepsy (7/7) was significantly higher than that in ID (1/10) (p < 0.001) (Figure 3C), suggesting a genotype-phenotype correlation.