Numerous studies and case reports show comorbidity of autism and epilepsy, suggesting some common molecular underpinnings of the two phenotypes. However, the relationship between the two, on the molecular level, remains unclear. Here, whole exome sequencing was performed on a family with identical twins affected with autism and severe, intractable seizures. A de novo variant was identified in the KCND2 gene, which encodes the Kv4.2 potassium channel. Kv4.2 is a major pore-forming subunit in somatodendritic subthreshold A-type potassium current (ISA) channels. The de novo mutation p.Val404Met is novel and occurs at a highly conserved residue within the C-terminal end of the transmembrane helix S6 region of the ion permeation pathway. Functional analysis revealed the likely pathogenicity of the variant in that the p.Val404Met mutant construct showed significantly slowed inactivation, either by itself or after equimolar coexpression with the wild-type Kv4.2 channel construct consistent with a dominant effect. Further, the effect of the mutation on closed-state inactivation was evident in the presence of auxiliary subunits that associate with Kv4 subunits to form ISA channels in vivo. Discovery of a functionally relevant novel de novo variant, coupled with physiological evidence that the mutant protein disrupts potassium current inactivation, strongly supports KCND2 as the causal gene for epilepsy in this family. Interaction of KCND2 with other genes implicated in autism and the role of KCND2 in synaptic plasticity provide suggestive evidence of an etiological role in autism.
Inherited seizure disorders are caused by abnormal regulation of membrane excitability. While the genetic basis of epilepsy is highly heterogeneous, the most common genes in which mutations lead to a seizure disorder encode ion channels that mediate the axonal conduction of action potentials and signal transduction through synaptic transmission (1,2). To date, autosomal dominant mutations in KCND2 have not been identified to cause epilepsy. Autism spectrum disorder (ASD), another early-onset disorder with a high prevalence of ∼1.1% (3), has been widely demonstrated to be heritable, but is also highly complex. Several large-scale genome-wide association studies have resulted in few strong common allele associations. Thus, with the advent of more powerful sequencing technology, attention has shifted to considering ASD as many independent rare monogenic disorders, and much effort is being made to identify de novo variants in ASD population (4–8). In many of the cases, rare de novo variants in ASD cases are identified in a single subject, and independent mutations in additional ASD cases are needed to determine causality. Both epilepsy and ASD are commonly associated with other phenotypes and co-occurrence of autism and epilepsy has been well documented. Recent reports show that 15–35% of children with epilepsy also present with ASDs and 7–46% of ASD patients are affected with epilepsy (9–14). Several candidate genes have been implicated as causal for the observed comorbidity of ASD and epilepsy, but none are conclusive (15).
Potassium channel genes are one of the largest groups of genes that have been implicated for epilepsy and seizures and also one of the most widely distributed types of ion channels that shapes the action potential and sets the resting membrane potential. Shal-type K+ channels, Kv4.x, are expressed throughout the brain, heart and smooth muscles. Each isoform, Kv4.1, Kv4.2 and Kv4.3, has different expression patterns, with Kv4.2 being the predominant isoform in some of the specific brain areas such as hippocampus CA1 pyramidal cells, dentate gyrus granule cells, habenula and cerebellum granule cell layer anterior lobules (2). In neurons, Kv4 proteins serve as the pore-forming subunits in inactivating somatodendritic subthreshold A-type potassium current (ISA) channels that regulate the rate of low-frequency firing and control backpropagation of action potentials into the dendritic tree (16). The availability of ISA channels, which is determined by the extent of inactivation, reflects the recent history of activity in the cell. Inactivation is therefore essential for regulating the physiological functions of ISA (16). ISA channels exhibit at least two types of inactivation (16,17). Kv4 channels can inactivate from the open state using an N-type mechanism in which the open pore is plugged by an N-terminal inactivation domain (18). However, this form of inactivation is unstable in Kv4 channels and probably does not contribute significantly to the steady-state availability of ISA channels in neurons. Kv4 channels also inactivate from preopen closed states when the voltage sensor domains activate but the pore fails to open (17,19,20). This is the predominant form of Kv4 inactivation.
Here we report a whole exome sequencing discovery of a novel de novo heterozygous variant in KCND2 gene in identical twin patients with autism and epilepsy (Fig. 1A) and demonstrate that this variant leads to altered electrophysiology in vitro with impaired closed-state inactivation. Based on this finding, taken together with previous reports on the role of Kv4.2 channel in synaptic plasticity, and that other components of the Kv4.2 channel are associated with autism, we propose that KCND2 is the causal gene for epilepsy in this family and has a role in the etiology of autism.
Patient clinical information
The monozygotic twins affected with seizures and autism (Fig. 1A) were born at 35 weeks gestation without complication and began to have seizures at 2 months of age, consisting of brief jerks of the extremities. Over the course of the first decade of life, the seizures were uncontrolled, occurring up to many hundreds per day, despite trials of multiple medications and the ketogenic diet. Diagnostic evaluations included MRI scans and metabolic testing, which were unremarkable. An electroencephalography performed in the first year of life demonstrated 3–4 Hz poly spike and wave by report. As the patients grew older, the seizures consisted of both staring spells as well as generalized convulsions. Between 8 and 10 years of age, the seizures decreased in frequency, with the parents reporting that generalized convulsions were occurring less than weekly, and staring spells occurring somewhat more frequently. The subjects have had no significant other medical problems with repeated normal physical examinations. Their head circumference is approximately the 95th percentile for age. At the age of 15, the subjects underwent detailed evaluation to determine whether the diagnosis of autism was appropriate and to evaluate their cognitive development. To assess autism, they (and parents) were administered Autism Diagnostic Interview–Revised (21) and the Autism Diagnostic Observation Schedule–2, Module 1 (22). Both subjects met criteria for ASD, based on the DSM-V (Diagnostic and Statistical Manual of Mental Disorders, 5th Edition) (23). Owing to their limited verbal abilities, the subjects were administered the Mullen Scales of Early Learning (24) to obtain an estimate of their cognitive functioning. Both exhibited overall cognitive functioning in the ‘Very Low’ range. Twin 1 had expressive language estimated at the level of 4 months of age, and receptive language was equivalent to 11 months. Twin 2 had expressive language estimated at 9 months of age, and receptive language was equivalent to 16 months.
Whole exome sequencing
Two de novo variants in the KCND2 and BICC1 genes were identified, and two sets of compound heterozygous variants were identified in the SLC8A2 gene and the GPR124 gene (Table 1). No rare homozygous or hemizygous variants were identified, and no large intervals of homozygosity were identified. Except for one variant in the GPR124 gene, all variants were novel, not observed in the ∼6500 exome variant server (EVS, http://eversusgs.washington.edu/EVS/), or in an internal database from the UCLA Clinical Genomics Center consisting of 700 exomes.
|Gene||Genomic position (hg19)||Inheritance||Zygosity||DNA change||Protein change||MAF in EVSa|
|Gene||Genomic position (hg19)||Inheritance||Zygosity||DNA change||Protein change||MAF in EVSa|
aMAF in EVS: Minor allele frequency in exome variant server.
SLC8A2 gene, in which the affected twins have compound heterozygous variants, encodes one of the three isoforms of the plasma membrane Na+/Ca+ exchanger (NCX), NCX2. SLC8A2 has not been associated with any human disease or phenotype yet. However, NCX2 is reported to be restrictively expressed in hippocampus, cerebellum and cortex, and knockout mice showed enhanced learning and memory capacity (25). Variants in BICC1 are associated with cystic renal dysplasia in humans (26), and mouse mutations in Bicc1 result in polycystic kidneys (27), which our patients do not manifest. The de novo variant found in KCND2 gene replaces valine for methionine at amino acid 404 (p.Val404Met) and is predicted to be ‘deleterious’, ‘probably damaging’ and ‘deleterious’ by SIFT, PolyPhen-2 and Condel, respectively. The conservation score measured by GERP was 5.41, and valine is highly conserved across 41 vertebrates. This variant is found at the C-terminal end of the transmembrane helix S6 region (amino acid 385–405, Fig. 1B) that makes up the ion permeation pathway along with the S5 domain and the P-loop (28). The six transmembrane domains of Kv4.2 are in general depleted of predicted-to-be-damaging variants in EVS, suggesting that variation is not well tolerated in these regions (Fig. 1B). One prior frameshift mutation (p.Asn587fs*1) in KCND2 was reported in a child with temporal lobe epilepsy. However, the unaffected father of the patient was a carrier of the same frameshift mutation. The affected proband was not reported to have autism.
P.Val404Met mutation dominantly impairs closed-state inactivation
As the de novo heterozygous variant in KCND2 was the most plausible variant related to the seizure disorder, we further investigated the functional consequence of the p.Val404Met mutation on Kv4.2 channels in vitro by expressing wild-type and mutant proteins in Xenopus oocytes (Fig. 2) (29,30). Ionic currents were evoked by depolarizing from a holding potential of −100 mV to voltages ranging from −80 to +70 mV. Potassium current through wild-type Kv4.2 channels was first detected at −50 mV (Fig. 2A). At more depolarized voltages, the rate of opening and peak current amplitude increased, and inactivation became more obvious. Similarly, K+ current through mutant p.Val404Met channels was first detected at −50 mV (Fig. 2A). However, at more depolarized voltages, p.Met404 currents reached their peak amplitude significantly later than p.Val404 (Figs 2A and B and 3A), and the decay of the current was significantly slower and less complete. As a result, the ratio of current amplitude at the end of the pulse relative to the peak current amplitude was significantly increased in mutant channels compared with wild-type channels (Figs 2A and B and 3A). Because the time to peak reflects the competing processes of opening and inactivation, changes in the time course of inactivation are likely responsible for the effect of the p.Val404Met mutation on the time to peak.
The genomes of the affected twins contain one wild-type and one mutant KCND2 allele. Because potassium channels are tetramers of pore-forming subunits (31), p.Val404 and p.Met404 subunits are expected to coassemble into functional heterotetramers. To characterize the functional properties of channels containing both types of subunits, p.Val404 and p.Met404 were coexpressed at a 1:1 ratio (Fig. 2A). Both RNAs made active channel proteins with no significant difference in expression level. The resulting currents resembled those obtained when the p.Met404 protein was expressed alone, reaching peak current amplitude significantly later and inactivating more slowly and less completely than p.Val404 channels (Fig. 2A and B). These results indicate that the functional effects of the p.Val404Met mutation are dominant, consistent with the family pedigree and heterozygous genotype.
In vivo, Kv4 pore-forming subunits associate with auxiliary subunits, including members of the K+ channel interacting protein (KChIP) and dipeptidyl-peptidase-like protein (DPP) families (32,33). These auxiliary subunits modulate the functional properties of Kv4 channels to more closely resemble native ISA currents. KChIP and DPP subunits have differential effects on Kv4 inactivation. KChIP subunits bind to the cytoplasmic N-terminal domain of Kv4 subunits and prevent or greatly inhibit N-type open-state inactivation (34). In contrast, DPP10a and DPP6a subunits confer fast, open-state inactivation via their N-terminal domains, which insert into the cytoplasmic end of the open pore, preventing conductance (35). To determine the effect of the p.Val404Met mutation in the presence of auxiliary subunits, p.Val404 or p.Met404 were coexpressed with KChIP3a or DPP10a by injecting RNA encoding KChIP3a or DPP10a at a 1:1 molar ratio (32) (Fig. 2C–F). In addition, a 1:1 mixture of p.Val404 and p.Met404 was coexpressed with KChIP3a (Fig. 2C and D). As previously reported, KChIP3a slowed the rate and decreased the extent of current decay in wild-type channels (Figs 2C, D and 3B) (32,36). When KChIP3a was incorporated with p.Met404 channels, little inactivation occurred during pulses lasting for 150 ms or 1 s (Figs 2C, D and 3B). When KChIP3a was incorporated into channels formed from a mixture of wild-type and mutant subunits, macroscopic current decay was somewhat faster than in mutant channels, resulting in a greater extent of inactivation (Figs 2C and D and 3B). The time to reach the peak current amplitude was correlated with the rate and extent of inactivation (Figs 2D and 3B). Wild-type channels reached their peak current amplitude significantly faster than p.Met404 channels, whereas mixed p.Val404/p.Met404 channels had an intermediate value (Fig. 3B). These results indicate that the p.Val404Met mutation significantly impairs closed-state inactivation, which is the main form of inactivation that occurs in the presence of KChIP3a (19).
When p.Val404 or p.Met404 was coexpressed with DPP10a, a fast component of inactivation conferred by the auxiliary subunit was observed (Fig. 2E and F) (35). Fast inactivation decreased the time to peak current amplitude, which did not differ significantly between wild-type and mutant channels (Fig. 3C). However, the effect of the p.Val404Met mutation on closed-state inactivation was still evident because the ratio of current amplitude at the end of the pulse relative to the peak current amplitude was significantly greater than in wild-type channels (Figs 2F and 3C). These results indicate that the p.Val404Met mutation does not interfere with open-state inactivation conferred by the DPP10a subunit and that DPP10a does not mask the aberrant effect of p.Met404 on closed-state inactivation kinetics.
Our results indicate that the substitution mutation, p.Val404Met, profoundly impairs closed-state inactivation when the mutant protein is expressed alone or in combination with wild-type Kv4.2. The effect of the mutation on closed-state inactivation is evident in the presence of the KChIP3a or DPP10a auxiliary subunits, which associate with Kv4 subunits to form ISA channels in vivo (33). Our results are compatible with previous reports implicating p.Val404 in Kv4.2 and the equivalent residues p.Val406 and p.Val401 in Kv4.1 and Kv4.3, respectively, in the mechanism of closed-state inactivation (20,34,37). Closed-state inactivation in Kv4 channels is thought to involve uncoupling between the voltage sensor domains and the cytoplasmic gate that closes the pore. Mechanical coupling is provided by residues in the linker between transmembrane segments S4 and S5. In response to depolarization, conformational changes in the voltage sensor domain pull on the gate, causing the S6 helices to splay apart by bending at a well-conserved sequence Pro–Val–Pro–Val corresponding to residues 401–404 in Kv4.2 (38). In Kv4.2 channels, interactions between p.Val404 in S6 and residues in the S4–S5 linker, particularly Glu323 at the cytoplasmic end of S5, maintain the coupling between the gate and the sensor (20). Because p.Val404 is a pivotal residue in gate–sensor interactions, mutations at this position profoundly affect closed-state inactivation, as we observe in channels bearing the mutation found in this family (19).
Our results taken together with previous reports strongly suggest that the size of the residue at position 404 is a major factor in determining the likelihood of inactivation. For instance, mutations p.Val404Ala and p.Val406Ile at the equivalent position in Kv4.1 have opposite effects on closed-state inactivation rate (20,37). Replacing valine (mean volume in protein = 139.1 Ǻ3) (39) by the smaller alanine residue (90.1 Ǻ3) enhances closed-state inactivation, perhaps because the smaller residue weakens the (already-weak) coupling between position 404 and residues in the S4–S5 linker, increasing the likelihood of sensor–gate uncoupling (20). In contrast, replacing valine by the larger isoleucine residue (164.9 Ǻ3) impairs closed-state inactivation perhaps because the larger residue is better able to maintain the physical interaction between the gate and the voltage sensor domain so that opening is preferred over closed-state inactivation (37). Methionine (167.7 Ǻ3) is similar in size to isoleucine, consistent with our results showing greatly decreased closed-state inactivation in the presence of the p.Val404Met mutation (39). Given the key role of closed-state inactivation in determining the steady-state availability of ISA channels in neurons, the dominant phenotype of p.Val404Met is likely to disrupt the normal physiological functions of ISA.Reduced inactivation (increased K+ currents) because of the p.Val404Met mutation would be expected to decrease excitability, which, in GABAergic neurons, could lead to the observed seizure disorder through disinhibition.
While the specific KCND2 mutation identified here is not clearly demonstrated to be causal of ASD, we note that several links can be made between Kv4.2 mutations and ASD. First, rare variants in KCND2 have been previously identified in individuals with autism including a submicroscopic de novo deletion encompassing KCND2 (40), three translocation breakpoints at 7q22.1, 7q31.2 and 7q31.3 potentially disrupting KCND2 (41), and three substitution variants, p.Asn544Ser, p.Phe538Ser and p.Arg539Leu. The three substitution variants were found in the C-terminal intracellular portion of KCND2 and were reported in three independent autism cases (42). Based on PolyPhen-2/SIFT, these three variants are predicted to be benign/tolerated, benign/tolerated and possibly damaging/tolerated, respectively. However, all three amino acid positions are completely conserved in primates, mouse and rats. KCND2 is located within a recently identified interval of enriched homozygous haplotype shared within a large ASD cohort (43). Secondly, DPP6 and DPP10 proteins are auxiliary subunits of Kv4.2. These DPP proteins interact with the transmembrane domains of Kv4 proteins, and confer fast inactivation of the channel. Although no point mutations have been reported for DPP6 and DPP10 in ASD or epilepsy cases, each of these genes has been repeatedly reported within copy number variants in autism cases (44,45). Further, loss of function variants in DPP proteins are predicted to slow channel inactivation similar to the functional consequence of V404M observed here. Thirdly, Fragile X syndrome is one of the leading single-gene disorders associated with ASD, accounting for up to 5% of the ASD cases (46). Kv4.2 is highly expressed in the cell bodies and along the dendrites of hippocampal neurons, and the fragile X mental retardation protein (FMRP), which is itself regulated by a transmitter receptor metabotropic glutamate receptor (mGluR), interacts directly with Kv4.2–3′UTR to suppress Kv4.2 translation (47). When FMRP is knocked out, the expression level of Kv4.2 increases compromising synaptic plasticity (47). Notably, a significant overlap between FMRP target genes and genes mutated in ASD individuals has been observed (48). In hippocampus CA1 pyramidal cells, Kv4.2 expression determines synapse maturational state (49). Lastly, recent data and network-based gene expression analysis indicate that KCND2 mRNA is coexpressed in developing human cortex with other autism-implicated genes within a module related to synaptic development (M17 in Parikshak et al.) (50). Further, data from fetal human and adult primate laminae show that KCND2 is preferentially expressed in the outer cortical plate of fetus and adult L2, suggesting Kv4.2 dysfunction may preferentially affect upper cortical layer neurons further linking the variant to autism phenotype. KCND2 expression pattern in adult correlates highly with GABAergic markers, further suggesting that it may play a role in this cell type. From the combination of all of the above data, it is highly plausible that the specific mutation identified here has directly contributed to a defect in synaptic homeostasis and the severe autism phenotype.
In conclusion, we identified a de novo variant p.Val404Met in KCND2 using whole exome sequencing in a family with atypical autism and severe, intractable epilepsy. The non-synonymous de novo variant p.Val404Met identified in KCND2 gene is a previously studied residue that plays a critical role in both voltage-dependent gate opening and closed-state inactivation, and has been suggested to be involved in a dynamic coupling between the voltage sensor and the cytoplasmic gate to mediate the inactivation of Kv4.2 channels (20). Although without an in vivo model of KCND2 mutation, we cannot exclude that other variants, especially the ones that were identified in de novo or compound heterozygous states in BICC1, SLC8A2 and GPR124, may contribute some effect to the complex phenotype of epilepsy and autism, we believe that these data in aggregate strongly implicate that mutations in KCND2 can cause epilepsy and autism. Also, a number of reports showing essential proteins interacting with Kv4.2 being implicated in autistic features provide suggestive evidence of Kv4.2 playing a role in autism even though KCND2 has not been reported in any of the recent exome sequencing studies (4–6,8) implying that autism cases explained by KCND2 mutations will be extremely rare.
MATERIALS AND METHODS
Whole exome sequencing
After institutional review board-approved informed consent, blood was obtained from participating family members: the unaffected parents and the affected identical twins. Genomic DNA was extracted using standard methods (QIAcube Qiagen, Valencia, CA, USA) at the UCLA Orphan Disease Testing Center. Library preparation, sequencing and data analysis were performed at the UCLA Clinical Genomics Center using the CLIA- (Clinical Laboratory Improvement Amendments) and CAP (College of American Pathologists)-validated protocols. An amount of 3 μg of high-molecular-weight genomic DNA from each participant was subjected to library preparation and exome capture following the Agilent SureSelect Human All Exon 50 Mb Illumina Paired-End Sequencing Library Prep Protocol. Sequencing was performed on an Illumina HiSeq2000 as a 50 bp paired-end run. For each sample, ∼200 million independent paired reads were generated for an average coverage of 140× of RefSeq protein coding exons and flanking introns (±2 bp) with ∼95% of these bases covered at ≥10×. The sequences were aligned to the hg19/b37 genome release using Novoalign. PCR duplicates were marked using Picard. GATK was used for indel realignment and base quality recalibration. Both SNVs (single nucleotide variants) and small INDELs (insertions and deletions) were called using GATK unified genotyper (51,52). All variants were annotated using the customized VEP (variant effect predictor) engine from Ensembl. Linkdatagen (53,54) and PLINK (55) determined regions of homozygosity by descent. A total of 21 890 variants were shared between the two affected identical twins, and rare protein-altering variants were examined under different inheritance models.
We obtained the full-length human KCND2 cDNA in pCR-Blunt II-TOPO (clone Id 40012007) from Thermo Scientific Open Biosystems (Waltham, MA, USA). The mutation was introduced into the reference sequence construct using QuikChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA, USA).
Plasmid cDNA clones of human KChIP3a (pT3T7–hKChIP3a) and DPP10a (pBSR–hDPP10a) were kindly provided by Dr Paul J. Pfaffinger, Baylor College of Medicine (32). RNA was transcribed in vitro using the mMessage mMachine T7 Ultra kit (Ambion, Life Technologies, Grand Island, NY, USA) and injected into Stage V–VI oocytes from Xenopus laevis. All animal procedures were approved by the Chancellor's Animal Research Committee at the University of California, Los Angeles.
Ionic currents were recorded in oocytes 1–3 days after RNA injection using a Warner OC-725 two-electrode voltage clamp (29,30). Electrodes were filled with 3 m KCl and had resistances ranging from 0.3–1.0 MΩ. The bath solution contained 2 mm KCl, 96 mm NaCl, 0.5 mm CaCl2, and 5 mm HEPES, pH 7.5. The functional properties of wild-type (p.Val404), mutant (p.Met404), and mixed wild-type/mutant channels were investigated in the presence or absence of KChIP3a or DPP10a using standard pulse protocols as described in the figure legends. Linear leak and capacitive currents were subtracted using a P/−4 protocol.
Online Mendelian Inheritance in Man (OMIM), http://www.ncbi.nlm.nih.gov/Omim (accessed on 10 February 2014).
Novocraft Short Read Alignment Package, http://www.novocraft.com (accessed on 10 February 2014).
Genome Analysis Toolkit, ftp://firstname.lastname@example.org (accessed on 10 February 2014).
SAMtools, http://samtools.sourceforge.net/ (accessed on 10 February 2014).
Picard, http://picard.sourceforge.net/ (accessed on 10 February 2014).
PolyPhen-2, http://genetics.bwh.harvard.edu/pph2/bgi.shtml (accessed on 10 February 2014).
SIFT, http://sift.jcvi.org/ (accessed on 10 February 2014).
Exome Variant Server, NHLBI Exome Sequencing Project (ESP), Seattle, WA, http://evs.gs.washington.edu/EVS/ (accessed on 10 February 2014).
Human Reference Genome: ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/reference/ (accessed on 3 September 2013).
The work was performed within the UCLA Clinical Genomics Center with support from the P30 UCLA Muscular Dystrophy Core Center grant to S.N., National Institutes of Health grant 5R01NS073871-03 to S.N. and National Institutes of Health grant 5R01GM043459-21 to D.M.P.
Technical work for sequencing was performed by Jean Reiss and Traci Lyn Toy. Computational support for sequence data analysis was provided by Bret Harry. Variant annotation was assisted by programs developed by Paige Taylor. We acknowledge Jieun Hong and Dr Joanna Jen for assisting with the mutagenesis and subcloning. We thank Dr Paul J. Pfaffinger (Baylor College of Medicine) for human KChIP3a and DPP10a clones.
Conflict of Interest statement. None declared.