Aberrant regulation of a poison exon caused by a non-coding variant in Scn1a-associated epileptic encephalopathy

Dravet syndrome (DS) is a developmental and epileptic encephalopathy that results from mutations in the Nav1.1 sodium channel encoded by SCN1A. Most known DS-causing mutations are in coding regions of SCN1A, but we recently identified several disease-associated SCN1A mutations in intron 20 that are within or near to a cryptic and evolutionarily conserved “poison” exon, 20N, whose inclusion leads to transcript degradation. However, it is not clear how these intron 20 variants alter SCN1A transcript processing or DS pathophysiology in an organismal context, nor is it clear how exon 20N is regulated in a tissue-specific and developmental context. We address those questions here by generating an animal model of our index case, NM_006920.4(SCN1A):c.3969+2451G>C, using gene editing to create the orthologous mutation in laboratory mice. Scn1a heterozygous knock-in (+/KI) mice exhibited an ~50% reduction in brain Scn1a mRNA and Nav1.1 protein levels, together with characteristics observed in other DS mouse models, including premature mortality, seizures, and hyperactivity. In brain tissue from adult Scn1a +/+ animals, quantitative RT-PCR assays indicated that ~1% of Scn1a mRNA included exon 20N, while brain tissue from Scn1a +/KI mice exhibited an ~5-fold increase in the extent of exon 20N inclusion. We investigated the extent of exon 20N inclusion in brain during normal fetal development in RNA-seq data and discovered that levels of inclusion were ~70% at E14.5, declining progressively to ~10% postnatally. A similar pattern exists for the homologous sodium channel Nav1.6, encoded by Scn8a. For both genes, there is an inverse relationship between the level of functional transcript and the extent of poison exon inclusion. Taken together, our findings suggest that poison exon usage by Scn1a and Scn8a is a strategy to regulate channel expression during normal brain development, and that mutations recapitulating a fetal-like pattern of splicing cause reduced channel expression and epileptic encephalopathy. Author Summary Dravet syndrome (DS) is a neurological disorder affecting approximately 1:15,700 Americans[1]. While most patients have a mutation in the SCN1A gene encoding Nav1.1 sodium channels, about 20% do not have a mutation identified by exome sequencing. Recently, we identified variants in intron 20N, a noncoding region of SCN1A, in some DS patients [2]. We predicted that these variants alter SCN1A transcript processing, decrease Nav1.1 function, and lead to DS pathophysiology via inclusion of exon 20N, a “poison” exon that leads to a premature stop codon. In this study, we generated a knock-in mouse model, Scn1a+/KI, of one of these variants, NM_006920.4(SCN1A):c.3969+2451G>C, which resides in a genomic region that is extremely conserved across vertebrate species. We found that Scn1a+/KI mice have reduced levels of Scn1a transcript and Nav1.1 protein and develop DS-related phenotypes. Consistent with the poison exon hypothesis, transcripts from brains of Scn1a+/KI mice showed elevated rates of Scn1a exon 20N inclusion. Since Scn1a expression in the brain is regulated developmentally, we next explored the developmental relationship between exon 20N inclusion and Scn1a expression. During normal embryogenesis, when Scn1a expression was low, exon 20N inclusion was high; postnatally, as Scn1a expression increased, there was a corresponding decrease in exon 20N usage. Expression of another voltage-gated sodium channel transcript, Scn8a (Nav1.6), was similarly regulated, with inclusion of a poison exon termed as 18N early in development when Scn8a expression was low, followed by a postnatal decrease in exon 18N inclusion and corresponding increase in Scn8a expression. Together, these data demonstrate that poison exon inclusion is a conserved mechanism to control sodium channel expression in the brain, and that an intronic mutation that disrupts the normal developmental regulation of poison exon inclusion leads to reduced Nav1.1 and DS pathophysiology.


Introduction
Dravet syndrome (DS) is a developmental and epileptic encephalopathy (DEE) characterized by intractable seizures, developmental delay, speech impairment, ataxia, hypotonia, sleep disturbances, and other health problems [3]. In the U.S., DS incidence is 1 per 15,700 [1], and 73% of patients die before the age of 10 years [4].
The most frequent cause of DS are loss-of-function mutations of SCN1A, which encodes the type I voltage-gated sodium channel (Na v 1.1) alpha subunit, part of a larger family of nine sodium channel proteins (Na v 1.1 -Na v 1.9) that control neuronal excitability [5]. Pathogenic SCN1A mutations are generally heterozygous and often occur de novo in DS. DS-associated SCN1A mutations lead to a loss of Na v 1.1, which is predominantly expressed in inhibitory GABAergic interneurons, so loss of function leads to network disinhibition [6]. Importantly, the molecular mechanisms for Na v 1.1 loss of function differ between various SCN1A mutations; many cause nonsense-mediated RNA decay, while other missense mutations affect Na v 1.1 stability or function [5].
Only 80% of DS patients have pathogenic SCN1A variants detectable within coding exons [7], suggesting that variants in noncoding regions near SCN1A may contribute to disease in some patients. A genomic analysis of 640 DEE patients found that five patients harbored rare variants predicted to be deleterious within a highly conserved region deep within SCN1A intron 20 [2]. A 64-bp segment within this region can be alternatively spliced and included as an exon termed 20N [2]. Exon 20N is known as a poison exon because it leads to a truncated and presumably nonfunctional SCN1A isoform due to a stop codon that arises with the frameshift caused by the 64-bp inclusion [8]. Several of the intron 20 variants identified in DEE patients increased inclusion of poison exon 20N in splice reporter assays in non-neuronal cells [2]. As a result, variant-induced aberrant inclusion of SCN1A poison exon 20N has been proposed as a mechanism for Na v 1.1 loss of function in DEE patients [2].
Our prior work on non-coding variation and poison exon inclusion in DS was carried out in non-neuronal cultured cells with artificial constructs and did not determine if any of the non-coding variants could recapitulate the phenotype of DS in an organismal context. Here, we report the construction and analysis of a mouse model for a SCN1A variant, NM_006920.4(SCN1A):c.3969+2451G>C (hereafter, c.3969+2451G>C), that we identified in our index patient and that lies within the alternatively spliced poison exon, 20N. [2]. Our results provide rigorous evidence of causality for a non-coding variant, allow direct measurement of poison exon usage in vivo, and give new insight into the normal function of poison exons for sodium channel genes and the consequent relationship to human genetic disease.

Results
Evolutionary conservation in intronic regions harboring a DS-causing variant. We first examined evolutionary conservation in the region surrounding the SCN1A intron 20 pathogenic variant that we studied here, c.3969+2451G>C, as a prerequisite to identifying the orthologous variant in mice. Human intron 20 is ~ 8 kb, within which there exist three highly conserved segments of several hundred nucleotides in length (Fig. 1A). Exon 20N and the surrounding region is highly conserved as indicated by quantitative assessment with genomic evolutionary rate profiling (GERP) (Fig. 1B) and alignment across 77 vertebrates (S1 Fig.). The G>C substitution in our index patient lies within exon 20N and is perfectly conserved along with neighboring nucleotides in the mouse. SCN1A transcripts that contain this exon 20N are "poisoned" due to a frameshift and consequent premature termination codon in exon 21 ( Fig. 1C-D); the same is true for mouse Scn1a.
Scn1a mRNA and protein levels are reduced in the brains of Scn1a +/KI mice. We used CRISPR/Cas9 gene editing [9] to generate mice harboring the c.3969+2451G>C variant on a C57BL/6J background. A guide RNA located upstream of the variant position ( Fig. 1B) was used to produce a G>C single nucleotide substitution at the corresponding position in mouse Scn1a, NC_000068.7:g.66293870C>G (GRCm38.p6) ( Fig. 2A). All genotypes were confirmed by Sanger sequencing. Animals carrying one allele of the edited variant are termed Scn1a +/KI and compared to nonmutant Scn1a +/+ littermates.
In cortical tissue from postnatal Scn1a +/KI mice, qRT-PCR for an amplicon between exons 19 and 20 (Materials and Methods) indicated an ~50% reduction in levels of Scn1a mRNA (Fig. 2B). Analysis of RNA-seq data from +/+ and +/KI animals yielded a similar result (Fig. 2C). We assessed Na v 1.1 protein levels with antisera targeting C-terminal or N-terminal epitopes; in both cases, levels of full-length protein (260 kDa) was reduced by ~ 50% (Fig. 2D-I), and there was no evidence of a truncated protein (157 kDa) that would otherwise correspond to the protein predicted from a transcript that contains exon 20N (S2 Fig. A). Taken together, these results indicate that the variant we introduced into Scn1a leads to the absence of Na v 1.1, likely due to nonsense-mediated decay of a transcript that contains exon 20N and a downstream premature termination codon.
Scn1a +/KI mice exhibit Dravet syndrome-like phenotypes. A number of SCN1A mutant mouse models have been described previously as models for DS [10][11][12][13][14][15]. As in patients with DS [16] and prior reports in other DS mouse models [10,11,[17][18][19][20][21], Scn1a+/KI mice exhibited increased premature mortality compared to Scn1a+/+ littermates, with about 40% mortality at 3 months and 55% mortality between birth and 18 months (Fig. 3A). These figures include only deaths after weaning at P21, when genotypes were obtained. We also noted high mortality rates in these litters prior to weaning and genotyping, so overall premature mortality rates C Scn1a, counts Scn1a, % Scn1a+/+ We tested Scn1a +/KI mice that survived to adulthood in a battery of behavior assays [22][23][24][25] to investigate if they developed behavioral deficits reported in other DS mouse models [26,27]. Consistent with phenotypes of other DS models [22,28], Scn1a +/KI mice exhibited hyperactivity in the open field ( Fig. 3B-F), including increased distance travelled ( Fig. 3B-C) and jumps (Fig. 3D). Notably, percent time in the center of the open field and stereotypic counts were similar in Scn1a +/KI mice compared to Scn1a +/+ mice ( Fig. 3E-F indicating no apparent deficits in short-term memory. Lastly, Scn1a +/KI mice did not show any behavioral deficits in the tube test of social dominance [31] (S4A Fig.) or threechamber sociability test [32] (S4B-D Fig.). Overall, our results on DS-associated behavioral phenotypes in Scn1a +/KI are similar to what has been reported previously in other DS mouse models. Taken together with the molecular characterization (Fig. 2), these results demonstrate that the mouse models the molecular pathophysiology of a conserved non-coding mutation in exon 20N and provides compelling evidence of its pathogenicity in DS.

Retention of exon 20N in Scn1a +/KI mice.
We designed several qPCR primer sets to detect mRNA transcripts either containing or excluding exon 20N (Fig. 4A), after reverse transcription with random primers (Materials and Methods). Amplicon 1 spans from exon 20 to exon 21 and generates a 56-bp product without exon 20N or a 120-bp product when exon 20N is included. In RNA from cortex of animals aged 1.9 mo -19 mo, levels of the larger transcript reflecting exon 20N inclusion were undetectable in Scn1a +/+ mice, but easily detectable in Scn1a +/KI mice ( Fig.  4B-C). A second set of primers spans the introns between exons 20 and 20N, and between exons 20N and 21, allowing measurement of exon 20N-containing transcripts as a 96 bp product, amplicon 2, that can be directly compared to a 111 bp product, amplicon 3, that spans exons 19 and 20 (Fig.  4A). Expressed as a percentage of amplicon 2/amplicon 3, exon 20N is included in 0.97% of Scn1a transcripts in +/+ mice, and in 4.8% of Scn1a transcripts in +/KI mice (Fig.  4D). Similar results were obtained after reverse transcription with oligodT. Assuming an additive model in which the presence of the KI allele does not influence activity of the + allele, and vice versa, we conclude that gene-edited variant leads to a ~9 to 10-fold increase in Scn1a transcripts that contain exon 20N. We also note that the levels of normal Scn1a mRNA and protein are reduced ~50% in +/KI compared to +/+ mice (Fig. 2), which implies that nearly all transcription from the KI allele contains exon 20N, and that ~95% of exon 20N-containing transcripts are degraded, likely by nonsense-mediated decay.
As an alternative approach to evaluating usage of exon 20N, we constructed and analyzed RNA-seq libraries from brain tissue of four Scn1a +/KI mice and four Scn1a+/+ littermates. The number of reads that aligned to all Scn1a exons was 387.83 ± 56.49 and 223.24 ± 27.58 in +/+ and +/ KI mice, respectively (Fig. 2C), but the number of reads that aligned to exon 20N were (0,0,1,0) and (0,0,0,0) in +/+ and +/KI mice, indicating that most transcripts that contain exon 20N are degraded.

Potential function of poison exons in Scn1a and Scn8a.
Aberrant regulation of poison exons as a pathogenetic mechanism, and conservation of that mechanism in humans and mice, raises the more general question of how and why Scn1a poison exons are used normally during development and differentiation. Voltage-gated sodium channels (Na v 1.1-Na v 1.9) have distinct developmental and regional patterns of expression [33][34][35][36]. Na v 1.1, the alpha subunit encoded by Scn1a, rises after a lag phase to adult levels during the second to fourth postnatal weeks in both mouse and rat brains, and is expressed primarily in a subset of GABAergic interneurons [33 -35]. The homologous protein Na v 1.6, encoded by Scn8a, has a similar pattern of developmental expression to Na v 1.1, but is expressed primarily in glutamatergic neurons. Previous work from Meisler and colleagues on Scn8a has identified a poison exon, 18N, whose expression is highest in fetal brain [37]. Inclusion of exon 18N in Scn8a is regulated by several RNA binding proteins [38,39], and for which a "fail-safe" mechanism has been proposed to prevent the synthesis of active protein in cells or tissues where it would be deleterious. We explored that idea for Scn1a by first measuring expression of alternative isoforms in different tissues of +/+ and +/KI mice using isoform-specific amplicons as shown in Fig. 4A. In heart, kidney, liver, and lung of +/+ and +/KI mice, Scn1a was expressed at very low levels as detected by qRT-PCR ( Table 1).
We did not have access to fetal tissues from Scn1a +/KI mice, but we analyzed usage of Scn1a exon 20N in non-mutant mice by analyzing a previously generated RNA-Seq dataset of mouse cortex at multiple developmental timepoints [40].  Table 1. Scn1a expression in various tissues. Quantification of total Scn1a mRNA levels using amplicon "1" (Figure 3) in Scn1a +/+ and Scn1a +/KI mice (n=1). Low levels of Scn1a mRNA were expressed in lung, liver, kidney, and heart as evidenced by the high DCt values. Tbp was used to normalize Scn1a expression. UD = undetectable. Transcripts containing 20N (Amplicon "2" in Figure 3) were undetectable or barely detectable in tissues other than brain. Expressed as a proportion of reads that align to exon 20N compared to all other exons, ~70% of Scn1a transcripts include 20N at E14.5, gradually decreasing to <10% by P30, and remaining minimal throughout adult life (Fig. 5A). This pattern is inversely correlated with the overall level of Scn1a mRNA, inferred from the total number of reads (Fig. 5A). Thus, as exon 20N usage decreased, more Scn1a mRNA was produced, consistent with the poison exon inclusion being used to reduce Scn1a levels during development. We used the same dataset to evaluate usage of poison exon 18N in Scn8a and observed a very similar pattern (Fig. 5B). This confirms the results of Meisler and colleagues [37], and suggests that poison exons for both sodium channel genes serve a similar function.

Discussion
Here, we generated a knock-in mouse model of an intronic variant, previously identified as a de novo mutation in a patient with DS, to explore its effects on Scn1a expression and function in vivo.  reduction in brain Scn1a mRNA and Na v 1.1 protein levels, resulting in expression of DS-related phenotypes (Figs. 2-3, Supplementary File 1). Inclusion of poison exon 20N in adult brain from Scn1a +/+ mice was ~1% and increased ~ fivefold in +/KI mice. Together with additional analyses and earlier work [37], our results suggest that poison exons in at least two neuronal sodium channels serve an important function in developmental regulation, suppressing expression of functional sodium channels until later stages of brain development. The mechanism is evolutionarily conserved and represents a previously unrecognized potential source of Mendelian disease. Mutations in the SCN1A gene are the most common cause of DS, accounting for 80% of patients. While 20% of patients still do not have a definitive molecular diagnosis after exome sequencing, our previous results identified a number of variants in and around exon 20N in multiple patients with DS/DEE [2]. Our results provide compelling evidence for pathogenicity of the variant reported here and, by extension, additional variants that may enhance inclusion of exon 20N. Our in vivo finding that poison exon 20N inclusion leads to Scn1a loss of function explains how the phenotype of Scn1a+/KI mice mimics that of other DS models, since reduced Scn1a expression is the common feature underlying DS. Scn1a loss of function results in decreased expression of Na v 1.1, a voltage gated sodium channel responsible for promoting electrical excitability of neurons [41]. Loss-offunction Scn1a mutations would decrease neuronal activity. Scn1a is predominantly expressed in inhibitory GABAergic interneurons [12], so DS mutations decrease activity of the inhibitory circuitry and resulting disinhibition contributes to seizure generation [41].
Since there is no disease-modifying therapy for DS patients, better understanding the molecular pathogenesis in different families is critical. In terms of the intronic variant we studied, there is a growing list of therapeutic strategies that target mechanisms related to alternative splicing. An RNA-based therapeutic triggering poison exon inclusion in DHX9, a gene involved in Ewing sarcoma, has been exploited to enhance the efficacy of chemotherapy in cancer patients [42]. Additionally, targeting RNA splicing of SMN2, haploinsufficiency of which leads to spinal muscular atrophy (SMA), has led to the development of the first FDA-approved drug to treat SMA [43]. The Scn1a+/KI mouse model we have developed based on patients with non-coding variants may provide support to assess poison exon 20N targeting therapeutics postnatally and other methods that may lead to an increased expression of Na v 1.1.
In summary, our data indicate that poison exon inclusion is a conserved mechanism to suppress gene expression that is induced by an intronic mutation in SCN1A leading to DS. These findings deepen our understanding of the molecular genetic mechanisms leading to DS and provide a new mouse model for studying the effects of a novel intronic mutation. Further, confirmation of the relevance of poison exon inclusion to a Mendelian disorder, coupled to the observation that multiple genes are regulated by this mechanism during development, suggests that variation affecting poison exons may be more broadly relevant to human disease.

Materials and Methods
Conservation Assessment GERP Analysis. The pre-mRNA diagram constructed using the R package ggbio [44] [45] with coordinates obtained from the UCSC Genome Table Browser [46]. Conservation was analyzed across the region using GERP (Genomic Evolutionary Rate Profiling) conservation scores for each position [47]. Positive GERP scores reflect a high level of conservation, while negative GERP scores reflect a neutral rate of substitution at the region.
Cross-species conservation analysis. Multiple sequence alignments were retrieved from the Vertebrate Multiz Alignment & Conservation [48] (100 Species) track of the UCSC Genome Table Browser [46].

Animals.
Mice were on a congenic C57BL/6J background. Mice were housed in a pathogen-free barrier facility on a 12hour light/dark cycle with ad libitum access to water and food (NIH-31 Open Formula Diet, #7917, Harlan). Mice were genotyped by PCR from tail tissue collected at weaning and at death. Both male and female mice were used in experiments. Littermate siblings were used as controls in each experiment. Experiments were completed by blinded investigators. For postmortem analyses, mice were anesthetized by Fatal-plus (Vortech) and perfused with 0.9 % saline. Tissues were then removed, weighed, and dissected for processing as described below. All experimental protocols were approved by the Institutional Animal Care and Use Committee of the University of Alabama at Birmingham.
Targeted CRISPR Scn1a +/KI Generation. The CRISPR guide was generated using an online tool (http://crispr.mit. edu). A mouse carrying the NC_000068.7:g.66293870C>G (GRCm38.p6) mutation was generated at the University of The Cas9 protein-crRNA-tracrRNA RNPs were assembled according to the manufacturer supplied instructions. Injections, manipulations of C57BL/6 embryos and subsequent maintenance of mice lines were carried out at the TGEMs facility using protocols and methods compliant with the University of Alabama at Birmingham Institutional Animal Care and Use Committee and the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health.
About 150 blastocysts were injected. Of these 28 F0 pups survived up to weaning and were genotyped for the engineered mutation. Six of these carried the mutation. Three of these were chosen for further crosses as these had the strongest peak for the changed base and therefore likely to have less mosaicism.
Genotyping. DNA from weaned mouse tails was prepared using the UAB Transgenic & Genetically Engineered Models Core Facility protocol: Tail clips were incubated at 65°C for 12 hours in lysis buffer (10mM Tris, 75mM NaCl, 25mM EDTA, 1% SDS, 0.5mg/ml Proteinase K) with intermittent shaking. DNA was extracted using Phenol:Chloroform, precipitated with ethanol, dried and resuspended in TE buffer. DNA was quantified using the Thermofisher Qubit™ dsDNA HS Assay Kit (Q32854).
Mouse tissue RNA extraction. RNA was extracted from frozen Scn1a+/+ and Scn1a+/KI mouse tissues using the Norgen Biotek Corp Animal Tissue RNA purification kit (Cat # 25700) using the manufacturer supplied protocol. Genomic DNA contamination was reduced/eliminated using Thermofisher TURBO DNA-free™ Kit (AM1907). RNA was quantified using Thermofisher Qubit™ RNA BR Assay Kit (Q10211).
cDNA Synthesis. Random primed and oligo-dT primed cDNA was synthesized from total RNA using Thermofisher SuperScript™ IV First-Strand Synthesis System (18091050) using the manufacturer supplied protocol. The Random primed cDNA was used for qPCR analyses.
Oligos for the housekeeping gene Tbp were ordered from IDT Technologies. Product # Mm.PT.39a.22214839.
Selected amplified DNA from the qPCR was analyzed on the Agilent 2100 Bioanalyzer using the Agilent DNA 1000 Kit (5067-1504).

RNA-seq.
RNA-seq libraries were prepared from whole brain total RNA from 4 Scn1a +/+ and 4 Scn1a +/KI mice using the Lexogen QuantSeq 3' mRNA-Seq Library Prep Kit FWD for Illumina. Briefly, first strand cDNA was synthesized from total RNA using oligo dT oligo containing an Illumina Read 2 linker. Then, the template RNA was digested and 2nd strand synthesized using random primer containing Illumina Read 1 linker and Unique Molecular Identifiers (UMIs). The libraries were then amplified using Illumina index containing oligos. Single-end sequencing was done on an Illumina NextSeq. The reads generated were deduplicated using the UMIs. The RNA-seq reads were trimmed using Trim Galore (https://github.com/FelixKrueger/TrimGalore). The trimmed reads were aligned to the mouse genome build GRCm38.p6 using STAR aligner [49]. HTseq was used to generate counts for genes from the alignments [50]. Scn1a and Gapdh counts were extracted from the normalized table of counts in the R package DESeq2 [51].
Behavioral Assessment. For all behavioral tests, experiments were carried out during light cycle at least one hour after the lights came on. All mice were transferred to testing room for acclimation at least one hour prior to experiments. Testing apparatuses were cleaned by 75% ethanol between experiments and disinfected by 2% chlorohexidine after experiments were finished each day. All mice were tested in all the behavioral tests in the same order. Investigators were blind to the genotype of individual mouse at the time of experiment.
Open Field. Each mouse was placed into the corner of an open field apparatus (Med Associates) and allowed to walk freely for 10 minutes. Total and minute by minute ambulatory distance, jumps, stereotypic behavior counts, and percent time in center of each mouse were determined using the manufacturer's software.
Elevated Plus Maze. Elevated Plus Maze (Med Associates) has two open arms and two closed arms. Mice were placed in the hub of the maze and allowed to explore for five minutes. The time in each arm, as well as entrances to each arm, explorations, and head dips over the edge of the maze, were monitored by video tracking software (Med Associates).
Y Maze. The Y-Maze apparatus consisted of three 15-inch long, 3.5-inch wide and 5-inch high arms made of white opaque plexiglass placed on a table. Each mouse was placed into the hub and allowed to freely explore for 6 minutes, with video recording. An entry was defined as the center of mouse body extending 2 inches into an arm, using tracking software (CleverSys). The chronological order of entries into respective arms was determined. Each time the mouse entered all three arms successively (e.g. A-B-C or A-C-B) was considered a set. Percent alternation was calculated by dividing the number of sets by the total number of entries minus two (since the first two entries cannot meet criteria for a set). Mice with 12 or fewer total entries were excluded from spontaneous alternation calculations due to insufficient sample size.
Tube Test for Social Dominance. The tube test for social dominance was conducted as previously described [31].Mice of the same sex, but opposite genotype, were released into opposite ends of a clear plastic tube and allowed to freely interact. Under these conditions, one mouse will force the other out of the tube. The first mouse with two feet out of the tube was considered to have lost the match. Each mouse was paired with three different opponents of the opposite genotype, and the winning percentage was calculated for each mouse by dividing the number of wins by the total number of matches.
Three-Chamber Sociability Test. The threechamber sociability test was conducted as previously described [32]. Mice were allowed to freely explore a threechambered testing apparatus for 10 min prior the introduction of wire cages containing a novel mouse (adult sex-matched C57Bl/6J) or a novel object (Lego block). Investigation of the novel mouse and object was then monitored for 10 min using video tracking software (CleverSys).
Scn1a and Scn8a in Developing Mouse Brain. Publicly available RNA-seq data (SRA Accession # SRP055008) [40] was used to check the expression of Scn1a and Scn8a in the developing mouse cortex. The RNA-seq reads were trimmed using Trim Galore (https://github.com/FelixKrueger/ TrimGalore). The trimmed reads were aligned to the mouse genome build GRCm38.p6 using STAR aligner. HTSeq was used to generate counts for genes from the alignments. Scn1a and Scn8a counts were extracted from the normalized table of counts in R package DESeq2. To calculate the proportion of 20N containing transcripts, samtools depth was used to extract coverage across each base of each exon of Scn1a and Scn8a [52]. The average read count per base was calculated by dividing the total read count by the size in bp of the exons. The averages for the 20N exon and all the exons were calculated separately. After normalizing to the number of mapped reads in each sample, percent poison exon usage was calculated using the formula: (average depth of coverage of the poison exon) / (average depth of coverage of all exons) *100.
Statistics. mRNA and protein levels were analyzed by Student's t-test. Behavioral tests were analyzed by Student's t-test or two-way RM-ANOVA specified in the figures dependent on the outcome measure. The survival data were analyzed by Kaplan-Meier statistics and post-hoc Log-rank (Mantel-Cox) test.
Two-tailed p-values were calculated for all analyses, and the cut-off for statistical significance was set at 0.05. GraphPad Prism 7 was used for all analyses. Data are presented as mean ± SEM (Standard Error of the Mean). Significance denoted as *p<0.05, **p<0.01, ***p<0.001, ****p<0001.