Base editing may reduce seizures and sudden death in a genetic epilepsy caused by SCN8A mutation

Thevariant R1872W (c.5614C>T) () results from a CGG to TGG substitution on the coding DNA strand. We targeted the negative (complementary) strand to enable an adenine to guanine (A-to-G) conversion. This ultimately resulted in a TGG to CGG correction upon completion of the intermediate base editing conversion of adenine to inosine () (). Additionally, there is no canonical NGG PAM, the typical recognition site for Cas9 (), near the R1872W loci in a utilizable position for single guide RNA(sgRNA) constructs. Therefore, we explored base editors with fluid PAM sequences in cell screens to identify the most efficient constructs. We designed a set of sgRNA sequences capable of utilizing 23 published ABE deaminase constructs with diverse PAM recognition profiles to revert R1872W. We chose to pair these into 16 testable constructs (and; supplemental material available online with this article;) based on reported PAM preferences (–) and high predictive accuracy from BE-HIVE (), CRISPOR (), and CRISPR RGEN tools (). Each construct was designed to target the same adenine at the R1872Wlocus and took into consideration the available PAM sequences in the area. SCN8A SCN8A Figure 1A Figure 1, B and C ^{17 19} Supplemental Table 1 Supplemental Figure 1 ^{20 23 24 25 26} https://doi.org/10.1172/JCI196402DS1↗

To screen for efficient base editing, sgRNA constructs were initially tested in HEK293 cells engineered to harbor a mouse codon-optimizedR1872W allele (HEK293-R1872W). These cells were selected for their ease of transfection and editing efficiencies, which were determined in our experiment by genomic DNA extraction and Sanger sequencing without the need for antibiotic selection or flow cytometry GFP sorting. Among the 16 tested constructs, 4 constructs achieved acceptable A-to-G conversion (≥ 15%) without A-to-G bystander editing (≤ 1%). Another 4 constructs exhibited bystander adenine editing, and 8 constructs showed no evident on-target A-to-G conversion, disqualifying guides under both categories from further in vitro studies. (). To mitigate the effects of HEK293 cell triploidy and account for species-specific genomic codon differences influencing guide RNA efficiency, the most effective constructs with no bystander A-to-G activity were evaluated in CHO Flp-In cells (CHO-R1872W). The CHO line provides a diploid genomic context more representative of typical mammalian somatic karyotypes in mice and humans, unlike the multiallelic integration in HEK293 cells (,). These cells, engineered to carry the sameR1872W allele, provide a diploid genomic context and serve as a rodent model proxy for in vivo applications. Editing efficiencies in CHO-R1872W cells were similar to those observed in HEK293 cells with 4 constructs achieving acceptable A-to-G conversion (≥ 15%) with only 1 construct exhibiting potential bystander adenine editing (). Guide C2 and ABE8e-NRCH exhibited robust activity in the murine genome. We selected C2 for in vivo analyses due to the absence of nearby bystander adenines within the immediate target window (4–7 nucleotides (nt) from 5' start of guide) (). Additionally, its paired editor demonstrated enhanced fidelity compared with prior ABEs () along with compatibility to undergo evolution to recognize PAM sequences such as NRCH (), whereany nucleotide, R = A or G, and H = A, C, or T. SCN8A SCN8A N = Supplemental Figure 1 ^{27 28} Figure 1D ^{19 19 21}

The introduction of a V106W mutation into the deaminase domain of ABE8e achieves significantly reduced DNA off-target effects (,). The V106W modified deaminase domain has shown to be particularly advantageous in gene editing scenarios in which minimizing off-target and bystander activity is crucial. This modification would be advantageous for the R1872W mutation, where the presence of multiple adenines in the target sequence poses a heightened risk in a sensitive pediatric population. We introduced the V106W mutation into the base editor; the construct was then divided into N-terminal and C-terminal pAAV backbones, incorporating NpuN and NpuC intein proteins as previously described (,) to form the C2 and ABE8e-NRCH V106W split intein construct (C2 Split-V106W) (). To assess whether these modifications affected construct activity, we screened them in the same engineered HEK293-R1872W and CHO-R1872W cell lines, comparing results to C2 paired with unmodified ABE8e-NRCH. C2 Split-V106W exhibited no significant on-target reduction at the R1872W locus editing compared with C2 paired with unmodified ABE8e-NRCH in both cell lines (). ^{19 29 30 31} Figure 1E Figure 1E

To assess the efficacy of our base editors in vivo, we used a mouse model in which the corresponding site in mousehas been mutated to create the human R1872W missense mutation in a Cre-dependent manner () (,). We used EMX1-Cre to generate aDEE mouse model expressing the R1872W variant exclusively in forebrain excitatory neurons (-EMX1) () ().-EMX1 mice exhibit a typical seizure onset around P20 (,), allowing time for peak expression of our base editing AAVs and correction of the variant (,). SCN8A Scn8a SCN8A Scn8a Scn8a W/+ W/+ W/+ ^{10 32} Figure 2A ^{10 10 33 34 35}

To achieve in vivo base editing of the mutant R1872W allele, we designed a dual intein adeno-associated virus (AAV) delivery strategy, referred to as-ABE, to package the split ABE8e-NRCH (V106W) base editor paired with the C2 sgRNA (C2 Split-V106W) into 2 AAV capsids, as previously described () (). The-ABE treatment of dual intein vectors was paired with a third AAV directing GFP expression to visualize treated cells in later experiments (). For packaging, we selected the AAV.PhP.eB capsid, an engineered variant of AAV9 (–). We utilized PhP.eB to deliver the-ABE treatment along with the partnered GFP virus. As a control, we used the partnered GFP virus in the absence of-ABE, referred to as sham control (). SCN8A SCN8A SCN8A SCN8A ³¹ Figure 2B ^{31 36 38 31}

We delivered either-ABE or sham virus via ICV injection at P2 in-EMX1 mice, ().-ABE treatment significantly prolonged the survival of approximately 87% of-ABE–treated mice (23) compared with sham control-EMX1 mice (25) (). Early mortality was observed in 3-ABE–treated-EMX1 mice (< P65). An additional 2 ABE-treated-EMX1 mice survived until P106 and P110. For sequencing purposes, we euthanized 16-ABE–treated mice at varying timepoints after P150, making the true survival rates for-ABE–treated-EMX1 mice impossible to accurately calculate (and, C and D). SCN8A Scn8a SCN8A SCN8A n = Scn8a n = SCN8A Scn8a Scn8a SCN8A SCN8A Scn8a W/+ W/+ W/+ W/+ W/+ Figure 2B Figure 2C Figure 2C Supplemental Figure 2

Increased survival in-ABE–treated mice was likely attributed to a decrease in seizure burden. To assess this, we evaluated seizures using concurrent video and electroencephalogram (EEG) recordings in-EMX1 mice treated either with-ABE (11) or the sham virus (10) (and). Spontaneous seizures were detected in all sham control-EMX1 mice, and 8 of 10 mice succumbed to seizure-induced death before P40. The 2 remaining sham control-EMX1 mice succumbed to death before P65. We evaluated seizure severity using the standard Racine Scale. Sham control-EMX1 mice experienced 1–5 Racine scale seizures (). In contrast, spontaneous seizures were completely inhibited in 7 of the 11-ABE–treated-EMX1 mice (). In another 3 mice, seizure frequency was significantly reduced, and seizure severity was reduced to only stage 1–2 Racine scale seizures (). SCN8A Scn8a SCN8A n = n = Scn8a Scn8a Scn8a SCN8A Scn8a W/+ W/+ W/+ W/+ W/+ Figure 2, D and E Supplemental Video 1 Figure 2G Figure 2, D and E Figure 2, F and G

Global expression of the R1872W variant via EIIa-Cre (-EIIa) leads to premature death at approximately P15 (). We also evaluated our base editor in these mice (12;). Peak expression of AAVs, and therefore editing activity, occurs around 3 weeks (,,). Despite the severity and timeline of seizure-related death in this expression model,-ABE treatment significantly increased the average survival of approximately half of all-ABE–treated-EIIa mice. Two-ABE–treated mice were euthanized for sequencing at P163 and P283, therefore also underestimating the true survival time (, B, D, and E). Sham control-EIIa mice (13) survived to an average age of P14.5, typically following a single seizure event. Scn8a n = SCN8A SCN8A Scn8a SCN8A Scn8a n = W/+ W/+ W/+ ¹⁰ Supplemental Figure 4 ^{34 35 39} Supplemental Figure 4

We used Next-Generation targeted amplicon Sequencing (NGS) to assess the extent of correction of R1872W variant alleles from mice used to generate the survival curves. This was detected via PCR primers spanning the TALEN-silent mutations inserted within the mutant R1872W allele only (,,) (). For this sequencing process, we leveraged these engineered silent substitutions, introduced originally to prevent TALEN recleavage, to enabled selective amplification and quantification of exon 26b, the only locus harboring the pathogenic R1872W variant (CGG to TGG on the + strand; CCA on the – strand) (,). Notably, while exon 26a and the WT allele contain adenines at positions A1 and A12 (corresponding to A2 and A13 in guide RNA alignment), they lack the critical A4 (A5 in chosen guide coordinates), which defines the targetable mutant site. Thus, A4 editing serves as a unique marker for the pathogenic exon 26b, while A1 and A12 represent potential bystander edits across all three alleles. ^{10 32 40} Supplemental Figure 2A ^{10 32}

We dissected hippocampal and cortical regions from 21 total-ABE treated-EMX1 mice and 19 sham control-EMX1 mice assessed for survival inC. In-ABE–treated-EMX1 mice that did not exhibit premature death and survived past timepoint P150 (16), we observed a reversion of 18.5% of the mutant allele T/A to WT C/G in genomic DNA compared with sham control-EMX1 mice (from 72.5% ± 1.7% (19) in sham control to 54.0% ± 0.9% (16) in-ABE–treated mice). The percentage of reads from all cells in these regions with the variant T/A at the R1872W locus ofwas reduced and WT base pairing of C/G at the R1872W locus increased with an equivalent percentage (). Additional sequencing revealed less than 5% editing within the hippocampus and cortex of 3-ABE–treated-EMX1 mice that experienced premature death (P26, P40, and P44;, and, C and D). Thus, the observed level of editing in these mice was insufficient to support survival. On target Na1.6 A4 mRNA sequencing of-EMX1-ABE–treated mice (13) that were euthanized after P150 shows approximately 32% on-target A-to-G reversion in RNA of mutant transcript (from 43.4% ± 1.9% in sham controls (5) to 11.6% ± 1.5% in-ABE–treated mice (13);). The correlation between transcript expression and seizures is shown in. DNA editing efficiencies above 15% were associated with complete seizure cessation and the prevention of seizure-induced death. SCN8A Scn8a Scn8a SCN8A Scn8a n = Scn8a n = n = SCN8A SCN8A SCN8A Scn8a Scn8a SCN8A n = n = SCN8A n = W/+ W/+ W/+ W/+ W/+ W/+ Figure 2 Supplemental Figure 2B Figure 2, C and H Supplemental Figure 2 Supplemental Figure 2E Supplemental Table 2 v

We also assessed the extent of correction of the R1872W mutation in dissected hippocampal and cortical regions from 11-ABE–treated-EIIa mice and 9 sham control-EIIa mice used to generate the survival curve shown in. We observed approximately 20% reversion in the mutant allele of T/A to WT C/G in genomic DNA in dissected hippocampus and cortex of 6-ABE–treated-EIIa mice that survived past P21, a timepoint where all sham control-EIIa mice had succumbed to death (, B and C). The percentage of reads from all cells in these regions with the variant T/A at the R1872W locus ofwas reduced from an average of 93% ± 0.8% (with calculated 7% failure of heterozygous EIIa-Cre activation of the R1872W variant) to 73% ± 1.8%. WT base pairing of C/G at the R1872W locus increased with an equivalent percentage to the decrease in mutant T/A in-ABE–treated-EIIa mice (). SCN8A Scn8a Scn8a SCN8A Scn8a Scn8a SCN8A SCN8A Scn8a W/+ W/+ W/+ W/+ W/+ Supplemental Figure 4B Supplemental Figure 4 Supplemental Figure 4C

In both-EMX1 and-EIIa mice,-ABE treatment effectively converted the tryptophan-encoding TGG sequence of the mutant allele at the R1872W locus to its WT arginine-encoding CGG counterpart, thereby inducing increasing levels of healthy WT transcript. Scn8a Scn8a SCN8A W/+ W/+

To assess the A-to-G off-target activity of-ABE, we conducted RNA sequencing on brain tissue from-ABE–treated-EMX1 mice (5; P345) and-EMX1 sham control mice (5; P25-30). Our objective was to quantify A-to-G editing and transcript reversion events both at the on-target site (A4) and at any other potential off-target locations across the transcriptome. Utilizing the computational tools CRISPOR (UCSC) (), CRISPR RGEN tools (), and COSMID (), we identified all adenines at a total of 290 potential off-target sites across platforms harboring less than 5 nucleotide mismatches with the target sequence or guide RNA. Whole transcriptome RNA sequencing was performed using-ABE treated-EMX1 and-EMX1 sham control mice.demonstrates A-to-G percentage editing outcomes for all adenines at 177 predicted off-target exon, protein coding sites showing percent A-to-G editing visible as percentage transcript reversions, comparing-ABE–treated-EMX1 mice with-EMX1 sham control mice. This comparison controls for sequencing noise inherent to the brain transcriptome, which we observed at less than or equal to 1%, consistent with no biologically meaningful off target activity (–). We observed less than or equal to 1% average transcript reversion of 2 adenines at Ap3d1 and 1 adenine each at off-target Agpat3, Cables2, and Inst3 loci in the CNS across the-EMX1-ABE–treated mice compared with-EMX1 sham control mice. There was no A-to-G base editing at any of the other adenines at the additional assayed potential off-target loci. In comparison, editing of the target A4 adenine resulted in an average of 32% on-target transcript reversion of mutant-to-normal WT transcripts with less than or equal to 1% bystander A1 and A12, consistent with no biologically meaningful bystander activity (). Absence of significant bystander adenines in-EMX1-ABE–treated mice was confirmed via NGS amplicon sequencing validations (). These results indicate high genomic target specificity of-ABE for the on-target R1872W locus. SCN8A SCN8A Scn8a n = Scn8a n = SCN8A Scn8a Scn8a SCN8A Scn8a Scn8a Scn8a SCN8A Scn8a Scn8a SCN8A SCN8A W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+ ^{25 26 41} Figure 3A ^{42 45} Figure 3A Supplemental Figure 3

To further characterize off-target editing, we performed 30 × short read WGS on matched pairs of-ABE–treated-EMX1 (2; P345) and sham control-EMX1 mice (2; P25–P30) (). We evaluated 977 adenines at a total of 290 potential off-target sites harboring less than 5 nucleotide mismatches with the target or guide sequence, including all sites evaluated in the RNA sequencing analysis. WGS analysis revealed no significant off-target reads of A-to-G altered reads from the 2 pairs of-ABE–treated mice compared with their matched sham controls at all 977 adenines in all 290 potential off-target loci (, ≤ 1%), consistent with no biologically meaningful off-target activity (–). In comparison, an average of greater than 21% on-target editing at the R1872W target A4 adenine was observed with no significant bystander A-to-G editing (≤ 1%), at A1 and A12 (). The results of the comprehensive off-target analysis indicate high genomic target specificity of the-ABE base editing strategy for the on-target R1872W locus. To further support the specificity of our construct for bothand the R1872W mutant allele, we evaluated all members of thegene family at genomic locations most likely to exhibit off-target activity following-ABE treatment. These evaluations, based on transcriptome-wide RNA-seq and WGS data, confirmed that no detectable A-to-G editing occurred at any of these loci (). SCN8A Scn8a n = Scn8a n = SCN8A SCN8A SCN8A SCNX SCN8A W/+ W/+ Figure 3B Figure 3B ^{42 45} Figure 3B Supplemental Table 3

Increases in neuronal excitability are a hallmark of seizures inDEE (,), and hyperexcitability of excitatory cortical and hippocampal pyramidal cells was previously reported in the-EIIa mouse model (). We next determined whether-ABE treatment could rescue neuronal hyperexcitability (). As an additional control, we examinedmice in the absence of Cre, where the R1872W variant is not activated (-control). Electrophysiological recordings were conducted at a time point corresponding to seizure onset via EEG monitoring (). Transduced neurons were identified by GFP expression. Transduction of GFP in-EMX1 mice was widespread, with high staining in the hippocampus and cortex (). We examined neuronal excitability of GFP-positive pyramidal neurons in the somatosensory cortex layer IV/V of P17–25 sham control (25, 4 mice) and-ABE–treated-EMX1 mice (33 cells, 7 mice). Sham control-EMX1 neurons were hyperexcitable when compared with-control neurons (21, 4 mice), which was attenuated in-ABE–treated-EMX1 neurons.-ABE treatment did not fully restore firing frequencies to levels observed in-control mice at higher current injection steps (). Analysis of membrane and action potential (AP) properties showed a significant increase in downstroke velocity in-ABE–treated-EMX1 neurons compared with sham controls (). All other parameters were unchanged between ABE and sham control-EMX1 neurons. SCN8A Scn8a SCN8A Scn8a Scn8a Scn8a n = SCN8A Scn8a n = Scn8a Scn8a n = SCN8A Scn8a SCN8A Scn8a SCN8A Scn8a Scn8a ^{10 11 10} Figure 4 Figure 2D Figure 4, A and B Figure 4, C and D Supplemental Table 4 W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+

We also observed heightened neuronal excitability in P13–17-EIIa sham control mice (17, 5 mice) when compared with-control mice (29, 5 mice), which was completely rescued by-ABE treatment (26, 8 mice) (, A and B). Analysis of membrane and action potential properties revealed a decrease in rheobase and downstroke velocity and an increase in AP width in-EIIa sham control mice when compared with-EIIa-ABE mice. We also observed a decrease in AP threshold in both sham control and-ABE–treated-EIIa mice when compared with-controls. (). Scn8a n = Scn8a n = SCN8A n = Scn8a Scn8a SCN8A SCN8A Scn8a Scn8a W/+ W/+ W/+ W/+ W/+ W/+ Supplemental Figure 5 Supplemental Table 5

The most prominent biophysical feature ofGOF variants is an elevated persistent sodium current (I) (,,). Elevated Iis a major contributor to neuronal hyperexcitability in epilepsy (). Thus, we sought to determine whether-ABE treatment eliminated the pathological heightened I. We used slow voltage ramps in pyramidal neurons from layer IV/V of the somatosensory cortex to measure Iin sham control (24; 8 mice) and-ABE–treated-EMX1 mice (18; 8 mice) and compared them with-control mice (P17-P25:17; 6 mice). ABE treatment significantly decreased the amplitude of Icompared with sham control-EMX1 mice to levels recorded in-control neurons (). Half maximal activation voltages (V) were not different between the 3 groups (). SCN8A SCN8A n = SCN8A Scn8a n = Scn8a n = Scn8a Scn8a NaP NaP NaP NaP NaP 1/2 ^{3 11 13 13} Figure 4, E and F Supplemental Table 4 W/+ W/+ W/+ W/+

In addition to seizures, patients withDEE suffer from severe comorbidities, including movement disorders and cognitive impairment (). To examine locomotor activity and anxiety-like behavior, we evaluated both sham control (15) and-ABE–treated (22)-EMX1 mice along with-control mice (25) in the open field test. We used total distance travelled to assess locomotor activity and thigmotaxis as a measure of anxiety-like behavior (). Mice were tested between 4 and 8 weeks of age, a time point after the onset of seizures in-EMX1 mice (). We generated quantifications of locomotor activity to analyze behavior in all groups ().-EMX1 sham control mice had reduced locomotor activity and spent significantly less time in the open areas of the field compared with-control mice, indicating diminished mobility and increased anxiety (). Treatment with-ABE significantly improved both comorbidities, with locomotor activity completely restored to levels observed in-control mice (). Thigmotaxis was significantly reduced in-ABE–treated-EMX1 mice but did not reach-control levels (). We also tested general cognition and memory. We observed no significant differences in recognition memory (more time in the novel arm), spatial reference memory (more visits to the novel arm), or working memory (spontaneous alterations recorded) between the groups (). These findings collectively suggest that-EMX1 mice have underlying motor discrepancies and anxiety that can be ameliorated by early-ABE treatment. SCN8A n = Scn8a n = Scn8a Scn8a n = Scn8a Scn8a Scn8a SCN8A Scn8a SCN8A Scn8a Scn8a Scn8a SCN8A ^{5 46 10} Figure 5 Figure 5, A–C Figure 5B Figure 5, A and C Supplemental Figure 6 W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+ W/+

Both sexes experienceDEEs with equivalent proportions, there is no known sex-linked interaction inprimary function. Both male and female mice were used for this study with no discrimination to sex. SCN8A SCN8A

Details regarding cell culture protocol and conditions (,) can be found in Supplemental Material. ^{67 68}

Details regarding plasmids used in this study and subsequent cloning and viral packaging can be found in Supplemental Material.

For characterization of base editing constructs, mutation-integrated cells were seeded 16–18 h before transfection. Both cell types (HEK293T-R1872W and CHO-R1872W) were subjected to transfection at approximately 70% confluency with a plasmid encoding the base editor and a separate plasmid harboring the sgRNA-P2A-GFP construct. Transfection of base editor and guide construct occurred at a 3:1 mass ratio using Lipofectamine 3000 (ThermoFisher Scientific) as previously described () and in accordance with the manufacturer's protocols, but scaled up into a 6-well plate (Corning). This scale-up was done to evaluate high cell density conditions in either cell line. Cells did not undergo antibiotic selection or GFP cell sorting after base editor exposure during transfection and were cultured for 72 hours before harvesting. We collected gDNA from cells 72 hours after transfection and Genomic DNA was isolated using PurelinkDNA mini Kit (Invitrogen). PCR conditions and primers are detailed in Supplemental Material. PureLink PCR Purification Kit was utilized following completion of PCR. Sanger sequencing of the purified product of editedin HEK293T-R1872W and CHO-R1872W cells was performed at Eurofins. Genomics and analysis of files were conducted with EditR1.0.10 as previously described () and verified on SnapGene v.10.0.3. ^{19 69} SCN8A

Husbandry details can be found in Supplemental Material. Homozygous Cre females were used for breeding to ensure minimal germline recombination due to Cre, as shown previously (,). For all experiments, we usedmice as controls that contained the same allele encoding thevariant but lacked Cre activation and therefore only expressed WT Na1.6 from both alleles (). ^{70 71 32} Scn8a SCN8A W/+ v

AAV vector backbone plasmids were cloned to insert the sgRNA sequence and C-terminal base editor half of ABE8e-NRCH (V106W) into v5 Cbh-AAV-ABE-NpuC+U6-sgRNA (Addgene 137177), and the N-terminal base editor half v5 Cbh-AAV-ABE-NpuN (Addgene 137178) (Genscript Inc.) (). Viral capsid preparations, packaging, and purification of PhP.eB -AAV–packaged base editor plasmids were performed by Salk Institute's Gene Transfer, Targeting and Therapeutics Viral Vector Core. High-titer qualified AAV was stored at –80°C until use. Neonatal ICV injections were performed as previously described ().-EIIa and-EMX1 mice were given a single unilateral ICV (2 μL volume) of either 2.0 × 10viral genomes (vg) PhP.eB-CMV-GFP in PBS (Sham) or 1.0 × 10vg of the-ABE treatment in PBS (5.0 × 10vg of each dual intein PhP.eB-ABE vectors) along with 2.0 × 10vg PhP.eB-GFP at P1 using a 33-gauge needle attached to a 5 μL microvolume syringe as previously described (). This dosage aligns with high therapeutic doses used for P0 ICV AAV administration of base editors for rescuing Hutchinson-Gilford Progeria syndrome and Niemann-Pick disease in mice (,). ^{31 10 72 31 73} Scn8a Scn8a SCN8A W/+ W/+ 10 11 10 10

Brain tissue for immunohistochemistry was processed as follows: Mice were anesthetized and transcardially perfused with 10 ml ice-cold Dulbecco's PBS (DPBS, Gibco, 14200-075) followed by 10 ml ice-cold 4% PFA. Brains were fixed in 4% paraformaldehyde (PFA) and 30 μm coronal brain sections were obtained using a cryostat. Floating sections were fixed and stored in DPBS. Sections were permeabilized with 0.1% Triton X-100 in DPBS for 30 minutes and blocked with 2% bovine serum albumin (BSA) for 2 hours. Sections were incubated with the primary antibody, polyclonal rabbit anti-GFP (Abcam, ab290), diluted in DPBS containing 2% BSA at a concentration of 1:1000. The secondary antibody, AlexaFluor-488 goat anti-rabbit (A11034; Invitrogen), was diluted 1:500 in DPBS containing 2% BSA. Sections were stained free-floating in primary antibody on a shaker at 4°C overnight and with secondary antibody for 2 h at room temperature the following day. Sections were mounted using Prolong Gold antifade reagent with DAPI (Invitrogen, P36935) on a microscope slide. Images were gathered using Zeiss Zen software on a LSM700 or LSM880. Whole brain images were stitched following acquisition with a 20× objective (Plan-Apochromat 20 ×/0.8) and z-stack images were gained by use of a 63x objective (Plan-Apochromat 63 ×/1.4).

Preparation of acute brain slices for patch-clamp electrophysiology experiments was modified sparingly from standard protocols previously described (). Mice were anesthetized with isoflurane and decapitated. The brains were rapidly removed and kept in chilled slicing solution (1°C) containing (in mM): 93 N-Methyl-D-glutamine (NMDG), 2.5 KCl, 1.25 NaHPO, 20 HEPES, 5 L-ascorbic acid (sodium salt), 2 thiourea, 3 sodium pyruvate, 0.5 CaCl, 10 MgSO, 25 D-glucose, 12 N-acetyl-L-cysteine, 30 NaHCO; pH adjusted to 7.4 using HCl (osmolarity 310 mOsm). Slices were continuously oxygenated with 95% Oand 5% COthroughout the preparation. 300 μm coronal brain sections were prepared using a Leica Microsystems VT1200 vibratome. Slices were collected and placed in warmed (37°C) ACSF containing (in mM): 125 NaCl, 2.5 KCl, 1.25 NaHPO, 2 CaCl, 1 MgCl, 0.5 L-ascorbic acid, 10 glucose, 25 NaHCO, and 2 Na-pyruvate. Slices were incubated for approximately 30 min and then kept at room temperature for up to 6 h. ¹¹ 2 4 2 4 3 2 2 2 4 2 2 3

Brain slices were placed in a chamber superfused (2 ml/min) with continuously oxygenated recording solution warmed to 32 ± 1°C. Whole-cell recordings were performed using a Multiclamp 700B amplifier with signals digitized by a Digidata 1550B digitizer. Currents were amplified, lowpass filtered at 2 kHz, and sampled at 100 kHz. Borosilicate electrodes were fabricated using a Brown-Flaming puller (model P1000, Sutter Instruments) to have pipette resistances between 3 and 5 mΩ. Current-clamp recordings of neuronal excitability were collected in ACSF solution. The internal solution contained the following (in mM): 120 K-gluconate, 10 NaCl, 2 MgCl, 0.5 KEGTA, 10 HEPES, 4 NaATP, 0.3 NaGTP, pH 7.2 (osmolarity 290 mOsm). Intrinsic excitability was assessed using methods adapted from those previously described (). Resting membrane potential was manually recorded from the neuron at rest. The first AP at rheobase, defined as the maximum amount of depolarizing current that could be injected into neurons before eliciting an AP, was used to calculate passive membrane and AP properties, including threshold; upstroke and downstroke velocity, which are the maximum and minimum slopes on the AP, respectively; amplitude, which was defined as the voltage range between AP peak and threshold; APD, which is the duration of the AP at the midpoint between threshold and peak; and input resistance, which was calculated using a −20 pA pulse in current-clamp recordings. AP frequency–current relationships were determined using 1 s current injections from −140 to 600 pA. All patch-clamp electrophysiology data were analyzed using custom MATLAB R24.a scripts and/or ClampFit 11.2. 2 2 2 50 ¹¹

The recording solution has been previously described () and contained (in mM): 100 NaCl, 40 TEACl, 10 HEPES, 3.5 KCl, 2 CaCl, 2 MgCl, 0.2 CdCl, 4 4-aminopyridine (4-AP), 25 D-glucose. The internal solution for sodium channel recordings contained the following (in mM): 140 CsF, 2 MgCl, 1 EGTA, 10 HEPES, 4 NaATP, and 0.3 NaGTP with the pH adjusted to 7.3 and osmolality to 300 mOsm. Steady-state persistent sodium currents (I) were elicited using a voltage ramp (20 mV/s) from −80 to −20 mV. After collecting recordings at baseline, protocols were repeated in the presence of 500 nM tetrodotoxin (TTX; Alomone Labs) to completely isolate I. TTX-subtracted traces were analyzed by subtracting the current recorded in the presence of TTX from the current recorded in its absence. The half-maximal voltage for activation was calculated as previously described (). ^{11 74} 2 2 2 2 2 NaP NaP

Custom electroencephalogram (EEG) headsets (Digikey) were implanted in P18-EMX1 mice using standard surgical techniques as previously described (). Anesthesia was induced with 5% and maintained with 0.5%–3% isoflurane. Adequacy of anesthesia was assessed by lack of toe-pinch reflex. A midline skin incision was made over the skull and connective tissue was removed. Burr holes were made at the lateral/anterior end of the left and right parietal bones to place EEG leads and at the interparietal bone for ground electrodes. EEG leads were placed bilaterally in the somatosensory cortex and unilaterally placed in the occipital lobe. A headset was attached to the skull with dental acrylic (Jet Acrylic; Lang Dental). Mice received postoperative analgesia with ketoprofen (5 mg/kg, i.p.) and 0.9% saline (0.5 mL i.p.) and were allowed to recover a minimum of 2–4 days before seizure-monitoring experiments. Mice were then individually housed in custom-fabricated chambers and monitored for the duration of the experiment. The headsets were attached to a custom low-torque swivel cable, allowing mice to move freely in the chamber. EEG signals were amplified at 2000× and bandpass filtered between 0.3 and 100 Hz, with an analog amplifier (Neurodata model 12, Grass Instruments). Biosignals were digitized with a Powerlab 16/35 and recorded using LabChart 7 software at 1 kS/s. Video acquisition was performed by multiplexing four miniature night vision-enabled cameras and then digitizing the video feed with a Dazzle Video Capture Device and recording at 30 fps with LabChart 7 software in tandem with Biosignals. Modified Racine Scale was used to determine seizure severity in each epileptic indication. Behavioral responses were scored as: 0, normal behavior; 1, behavioral arrest; 2, orofacial automatisms; 3, unilateral forelimb clonus; 4, bilateral forelimb clonus and rearing; and 5, rearing with loss of balance (,). Scn8a W/+ ^{75 76 77}

Sequencing library preparation was performed according to previously published protocols (). We isolated genomic DNA (gDNA) with the Invitrogen PureLink Genomic DNA Mini Kit and approximately 25 mg of homogenized tissue was used for individual locus editing experiments. DNA from hippocampal and cortical tissue was microdissected and homogenized for genomic processing. Sequencing libraries were amplified in 3 steps, first to amplify the mutated locus of interest to confirm editing of the mutant allele without healthy read noise from the WT allele of the heterozygote mice (FP1/RP1). This was accomplished by detection of TALEN-induced silent mutations in exons 26b and 26a, which are placed into the mutant R1872W allele only and are not present in the WT allele (,,). The second primer set functioned to shorten the length of the amplicon to 150 bp (FP2/RP2); and third to add full-length Illumina sequencing adapters. PCR conditions and adapter primers are detailed in Supplemental Material. All subsequent library preparations and Illumina 2×150 bp sequencing was performed by Azenta (Genewiz) Inc. The samples were sequenced via alignment of fastq files and quantification of allele frequency for individual loci were performed using CRISPResso2 in batch mode (). The allele frequency for each site was calculated as the number of reads containing the full guide site along with nucleotides C/G at the R1872W locus and the total number of variants reads containing nucleotides T/A at the R1872W locus in the application. Default base editing parameters were used for both experimental sham control and-ABE groups on batch mode in CRISPResso2 to measure substitution presence in guide sequence of the amplicon in sham control and-ABE mice. Default parameters and direct comparisons of experimental groups were used to detect the equivalent present noise and nonfull alignments to the target sequence. ^{24 10 32 41 78} SCN8A SCN8A

5-EMX1 sham control and a total 13-EMX1-ABE treated mice (P345) were taken down; hippocampal and cortical tissues were immediately dissected and collected for evaluation of RNA transcripts. Total RNA was harvested from cells in the hippocampus and cortex using the RNeasy Mini kit (Qiagen). Library preparation and sequencing were performed by Azenta "Genewiz". Sequencing and subsequent FASTQs were performed and generated by Azenta "Genewiz" in paired-end mode on Illumina Novaseq. FastQC (v0.11.5) was used to assess the quality of each sample. Cutadapt (v3.4) was used to trim adapters. QC controlled FATSQ reads were then converted to BAM/BAI sequencing files via SAMTools and aligned to the mm39 murine reference genome. Mice (5) from each group's trimmed reads were then aligned to the GENCODE mouse reference genomeGRCm39 using IGV genome analysis browser and refined to canonical coding sequences using CCDS release 21 () to confirm the absence of editing at all adenines at 177 potential off-target sites harboring less than 5 mismatches with the target sequence or guide RNA within coding regions of DNA identified with CRISPOR (UCSC) (), CRISPR RGEN tools (), and COSMID (). Default parameters were used to remove low-quality bases, adapter sequences, low reads counts (total target reads ≤ 3, and variant reads ≤ 1 in 1 out of 10 experimental mice), and unpaired sequences (–). 5-EMX1 sham control and 5-EMX1-ABE–treated mice (P345) were compared at all adenines. Expression of the mutant and WT alleles in heterozygous-EMX1 mice confirmed equal allelic expression of both sham control and-ABE–treated groups. This was accomplished by comparing full WT transcripts to the detection of TALEN-induced silent mutations in exons 26b and 26a (,,) on the same platforms as described above. Scn8a Scn8a SCN8A n = Scn8a Scn8a SCN8A Scn8a SCN8A W/+ W/+ W/+ W/+ W/+ ^{79 25 26 41 42 45 10 32 40}

2 sets of paired-EMX1 sham control and-ABE–treated mice were euthanized; hippocampal and cortical tissues were immediately dissected and collected for evaluation of WGS. Total DNA was harvested from cells in the hippocampus and cortex using the Purelink Genomic DNA Mini Kit (Invitrogen). Library preparation and sequencing were performed by Azenta "Genewiz". Sequencing and subsequent FASTQs were performed and generated by Azenta "Genewiz" on Illumina MiSeq. FastQC (v0.11.5) was used to assess the quality of each sample. Cutadapt (v3.4) was used to trim adapters. QC controlled fastq reads were then converted to BAM/BAI sequencing files via SAMTools and aligned to the GENCODE mouse reference genomeGRCm38/mm10 using IGV genome analysis browser to confirm the absence of editing at all adenines at a total 290 potential off-target sites harboring less than 5 mismatches with the target sequence or guide RNA identified with CRISPOR (UCSC) (), CRISPR RGEN tools (), and COSMID (). Default parameters were used to remove low-quality bases, adapter sequences, low reads counts (variant reads ≤1 in 1 out of 4 experimental mice) and unpaired sequences (–). Scn8a SCN8A W/+ ^{25 26 41 42 45}

All behavioral tests were conducted in behavioral testing rooms between 10:00 and 14:00 hours of the day during the light phase of the light/dark cycle. Behavioral tests were performed in mice between the ages of 4 and 8 weeks. Mice were tested in a random order. After the tests, the equipment was cleaned with 70% ethanol to eliminate olfactory cues. Exploratory behavior, anxiety-like behavior, and general locomotor activity were examined using the open-field test according to previous studies (). Each mouse was placed in the center of the apparatus consisting of a square area surrounded by white acrylic walls (45 × 45 × 40 cm). The total distance travelled (m) and time spent in the central area (s) were recorded. The central area was defined as the middle 20 × 20 cm area of the field. Data was collected over a 5-min period using the ANY-MAZE software. Data analysis was performed in Microsoft Excel. All statistical comparisons were made using the appropriate test in GraphPad Prism 10. ⁴⁶

Spatial working memory was assessed using a white acrylic Y-maze apparatus (arm length: 40 cm, arm bottom width: 3 cm, arm upper width: 10 cm, wall height: 12 cm). Mice were placed at the center of the Y-maze for 5 min. Mice were tested with no previous exposure or habituation to the maze. A spontaneous alternation was defined as an entry into three different arms on consecutive choices. The percentage of alternation was calculated as the ratio of actual to maximum number of alternations. Time in novel arm, novel arm entries, and number of alternations were recorded and analyzed using the ANY-MAZE software exported to Excel.

Survival analysis was compared by Mantel-Cox comparison test. Seizure incidence was compared by Mann-Whitney comparisons test. Comparisons of 2 groups, namely genomic data, were compared by an unpairedtest when the data were normally distributed with equal variances, by Welch'stest (2-tailed) when the data were normally distributed with unequal variances, and by Mann-Whitney test when the data were not normally distributed. Comparisons of 3 groups, namely, membrane and AP properties, peak sodium currents, and behavioral studies, were compared by 1-way ANOVA followed by Dunnett's multiple comparisons test when the data were normally distributed with equal variances, by Brown-Forsythe ANOVA with Dunnett's multiple comparisons test when the data were normally distributed with unequal variances, and by the nonparametric Kruskal-Wallis test followed by Dunn's multiple comparisons test when the data were not normally distributed. Data were assessed for normality using the Shapiro-Wilk test. Data were tested for outliers using the ROUT method to identify outliers. A nested 1-way ANOVA followed by Tukey's test with multiple comparisons was used to compare groups in experiments in which repetitive measures were made from a single cell over various voltage commands or current injections. Data are presented as individual data points and/or mean ± SEM. Exactandvalues are reported in figure legends;values less than 0.05 were considered statistically significant. t t n P P

All experiments using live animals were approved by the University of Virginia Animal Care and Use Committees.

Allof each figure can be accessed infile. Plasmids are obtainable by request of corresponding authors. All NGS and high-throughput sequencing data has been deposited in the National Institute of Health Sequence Read Archive (SRA) under submission: Supporting data values Supporting Data Values https://www.ncbi.nlm.nih.gov/sra/PRJNA1356903↗