Prime editing of a pathogenic Scn1a allele ameliorates seizure phenotypes in a GEFS⁺ mouse model

Abstract

Generalized epilepsy with febrile seizures plus (GEFS⁺) is an inherited epileptic disorder predominantly linked to autosomal dominant, loss-of-function mutations in the sodium voltage-gated channel alpha subunit 1 (SCN1A) gene, which encodes the α-subunit of the neuronal voltage-gated sodium ion channel type 1 (Na_V1.1). Reduced Na_V1.1 function in γ-aminobutyric acid (GABA)–ergic inhibitory interneurons impairs inhibitory signaling and leads to neuronal hyperexcitability. Clinically, GEFS⁺ is characterized by a spectrum of seizure types, often beginning with febrile seizures in early childhood and progressing to generalized tonic-clonic seizures later in life. Here, we used prime editing to correct the pathogenic SCN1A-K1270T mutation in the Scn1a^KT/+ mouse model of GEFS⁺. Adeno-associated viral (AAV) vectors were employed to deliver an intein-split prime editor under the control of a neuron-specific promoter into the cerebral ventricles of neonatal mice. This enabled efficient in vivo editing, achieving 34.7 ± 14.5% correction of the mutant allele in cortical bulk DNA, 81.2 ± 5.9% correction of mRNA, and improved multiple disease-relevant phenotypes. Survival increased from 80% in control-treated animals to 100% in treated mice, cortical inhibitory neuron transmission was improved (frequencies of inhibitory postsynaptic currents were increased from 0.32 to 1.32 hertz), and the frequency of induced febrile seizures decreased from 78.6% to 13.3%, approaching the frequency seen in wild-type mice (8%). These findings suggest the therapeutic potential of prime editing for the treatment of patients with SCN1A-associated GEFS+..

Introduction

Generalized epilepsy with febrile seizures plus (GEFS⁺) is a genetically diverse epilepsy syndrome characterized by a broad clinical spectrum. It ranges from mild febrile seizures and febrile seizures plus, where episodes extend beyond early childhood or progress to afebrile seizures, to severe seizures with more complex phenotypes, as is typically observed in Dravet syndrome (1, 2). In most patients with GEFS⁺, febrile seizures occur for the first time between the age of 6 months and 5 years (3–5), triggered by fever-induced hyperthermia. Roughly 30-40% of affected children continue to experience recurrent seizures after their first episode, which is frequently accompanied by impaired learning as well as motor and cognitive development (6–9).

Disease management is currently limited to symptomatic administration of broad-spectrum antiepileptic drugs that often produce adverse effects and rarely achieve complete seizure control (10, 11). Autosomal-dominant mutations in SCN1A, encoding the α-subunit of the voltage-gated sodium channel Na_V1.1, account for about 10% of all GEFS⁺ cases, with missense mutations constituting over 50% of all SCN1A variants (12–14). These mutations primarily reduce sodium currents in gamma-aminobutyric acid (GABA)-ergic interneurons, impairing inhibitory signaling and driving neuronal hyperexcitability that leads to seizures (15–19). The severity of seizures and comorbidities scale with the functional impact of the SCN1A mutations, with a complete loss-of-function of one of the alleles typically manifesting as Dravet syndrome (20).

Because SCN1A-associated GEFS⁺ is monogenic, correction of the pathogenic allele through genome editing should restore normal Na_V1.1 function and provide durable cure. Prime editing is a versatile genome editing technology that is based on the fusion of a Cas9 nickase with a reverse transcriptase, which transcribes a synthetic RNA template that subsequently gets incorporated into the nicked DNA strand. Prime editing works independently of DNA double-strand break formation and homology-directed repair (HDR), enabling precise gene correction in postmitotic cells such as neurons (21–25).

Here, we assessed whether in vivo prime editing in the brain could be used to correct the human pathogenic SCN1A-K1270T mutation in the Scn1a^KT/+ mouse model of GEFS⁺ and thereby alleviate GEFS⁺ associated phenotypes (18). The autosomal dominant K1270T variant was originally identified in a multigenerational family with GEFS⁺ and is clinically characterized by febrile seizures beginning in infancy that, in most affected individuals, persist beyond early childhood and progress to afebrile generalized or focal seizures with variable expressivity (26, 27). The mutation has also been functionally characterized across multiple systems and shown to reduce excitability of inhibitory neurons, including in a Drosophila knock-in model and in human induced pluripotent stem cell (hiPSC)–derived neurons (27, 28). Consistent with these observations, the Scn1a^KT/+ mouse model recapitulates key features of GEFS⁺, including susceptibility to heat-induced seizures (18). Moreover, electrophysiological recordings from hippocampal slices demonstrated reduced action potential amplitude in parvalbumin (PV)-expressing inhibitory Cornu Ammonis (CA1) neurons, supporting a loss-of-function mechanism in inhibitory circuits.

In this study, intracerebroventricular (ICV) delivery of adeno-associated viral (AAV) vectors encoding a prime editing system resulted in efficient correction of the pathogenic variant in the brains of Scn1a^KT/+ mice. This molecular correction restored inhibitory synaptic transmission and markedly lowered the incidence of hyperthermia-induced febrile seizures. Together, these findings establish prime editing as a promising therapeutic strategy for SCN1A related GEFS⁺ and potentially also other genetic epilepsies.

Results

In vitro prime editing of the pathogenic Scn1a^KT variant

To explore the therapeutic potential of prime editing for SCN1A-associated epilepsies, we employed the Scn1a^KT mouse model, in which one copy of the Scn1a^KT allele triggers febrile seizures (18).

We first applied the in silico prediction tool PRIDICT (29, 30) to design prime editing guide RNAs (pegRNAs) predicted to revert the mutant ACC codon back to wild-type AAA (fig. S1). We selected twelve top-ranked pegRNA designs targeting three different protospacers with varying primer binding site and reverse transcriptase template lengths (Fig. 1A). To evaluate the editing efficiency of these pegRNAs, we co-transfected them along with the PEmax prime editor (PE) into K562 reporter cells harboring part of the murine Scn1a^KT locus. From the tested pegRNA designs, pegRNA_c1 and pegRNA_c3 (referred to as PEmax_c1 and PEmax_c3) showed the highest Scn1a-KT correction efficiencies, with 8 ± 1.5% and 7.5 ± 0.9% editing, respectively (Fig. 1B, Data files S1 and S2). To further enhance editing, we paired both pegRNAs with three different PE3b nicking single guide RNAs (sgRNAs), which introduce a nick in the non-edited DNA strand once the newly synthesized prime-edited strand is successfully integrated (21) (Fig. 1A). This strategy boosted correction up to 15.4 ± 2.5% for pegRNA_c1 and 14.1 ± 2.6% for pegRNA_c3 (Fig. 1C), without increasing the occurrence of undesired byproducts such as indels or scaffold integrations for pegRNA_c3 (Fig. 1, B and C). We therefore selected PEmax_c1, PEmax_c3 and PE3b_c3 (PE3b approach employing PEmax with pegRNA_c3 and the ngRNA_n2) for subsequent in vivo experiments.

Fig. 1. Prime editing corrects the pathogenic Scn1a-KT variant in vitro.

(A) Schematic representation of the Scn1a-KT allele and the corrected sequence. pegRNA spacers and PAM sequences are shown in blue, PE3b ngRNA spacers and PAM sequences, which bind to the corrected allele, are shown in yellow. The pathogenic mutations are marked in purple, corrected bases in green. The corresponding protein sequences are displayed below the DNA sequences. Spacer 1 corresponds to pegRNAs c1–c4, Spacer 2 to c5–c8, and Spacer 3 to c9–c12. Spacer 1 (highlighted in darker blue) was used for the in vivo studies. (B) Intended editing efficiency (upper panel), indels (middle panel) and scaffold integrations (lower panel) for various pegRNA designs in K562 cell lines. pegRNA designs used for in vivo studies are marked in dark blue. (C) Intended editing efficiency (upper panel), indels (middle panel) and scaffold integrations (lower panel) for various PE3b designs in K562 cell lines. (B and C) Data are expressed as mean ± SD (n=3). *P < 0.05, **P < 0.01, and ****P < 0.0001. Ordinary one-way ANOVA with Dunnett’s multiple comparisons (two-sided) comparing each design vs c3 or nNT, respectively (adjusted P values and 95% CI of mean differences in Data file S2).

In vivo correction of the Scn1a^KT mutation

To target the GABAergic inhibitory interneurons that are predominantly impaired by SCN1A mutations, we placed the PE coding sequence under control of the human synapsin 1 (hSyn1) promoter, which drives broad pan-neuronal expression (31, 32), and packaged the constructs into neurotropic adeno-associated virus (AAV) PHP.eB capsids (33). Because the full length PE exceeds the ~5 kb AAV packaging limit (34), we employed the Npu intein-mediated protein trans-splicing system to split the PE at residue 713 of Cas9 (22). The N-terminal part of the PE was co-expressed with the pegRNA and ngRNA from one AAV vector, whereas the larger C-terminal part was supplied by a second vector (Fig. 2A). Both AAVs were mixed in an equimolar ratio and administered by intracerebroventricular (ICV) injection into one-day-old heterozygous (Scn1a^KT/+) or homozygous (Scn1a^KT/KT) pups at a total dose of 2.5 × 10¹³ vg/kg (Fig. 2B).

Fig. 2. In vivo prime editing corrects the pathogenic Scn1a-KT variant in mice.

(A) Schematic representation of the AAV-PHP.eB PE dual vector split intein design. Constructs are not depicted to scale. ITR: inverted terminal repeat; hSyn1: human synapsin 1 promoter; hU6: human U6 promoter; mU6: mouse U6 promoter; M-MLV-RT: Moloney murine leukemia virus reverse transcriptase. ngRNA: nicking guide RNA; pegRNA: prime editing guide RNA (B) Schematic representation of the experimental setup and timeline of the in vivo studies. Total AAV doses are given for each condition in vg/kg (vector genomes per kilogram bodyweight). NGS: next generation sequencing. AAV control: c-terminal of PEmax only. (C) Scn1a-KT correction efficiency in the cortex of Scn1a^KT/+ mice following ICV injections of AAV particles at PND1 (AAV control n=9; PEmax_c1 n=14; PEmax_c3 n=26; PE3b_c3 n=30). (D) Scn1a-KT correction efficiency in cortex of Scn1a^KT/KT mice following ICV injections of AAV particles at PND1 (AAV control n=10; PEmax_c1 n=12; PEmax_c3 n=22; PE3b_c3 n=14). (E) Scn1a-KT correction efficiency in mRNA measured in cDNA by reverse transcription. RNA was isolated from the cortex and hippocampus of Scn1a^KT/+ mice following ICV injections of AAV particles at PND1 (PEmax_NT n=3; PE3b_c3 n=3). (F) Scn1a-KT correction efficiency across different brain regions of Scn1a^KT/+ mice 10 weeks post ICV injections of AAV particles encoding for PE3b_c3 (n=3). (G) Indel formation in cortex of Scn1a^KT/+ mice following ICV injections of AAV particles at PND1 (AAV control n=17; PEmax_c1 n=12; PEmax_c3 n=26; PE3b_c3 n=28). (H) Survival curves spanning from 0 to 50 days of Scn1a^KT/KT mice following PND1 AAV injections (Kaplan-Meier with Log-rank (Mantel-Cox) test: PEmax_c1 (n=14, P = 0.77), PEmax_c3 (n=30, P = 0.0023), PE3b_c3 (n=14, P=0.0001)). (I) Survival curves spanning from 0 to 50 days of Scn1a^KT/+ mice following PND1 AAV injections (Kaplan-Meier with Log-rank (Mantel-Cox) test: PEmax_c1 (n=15, P = 0.57), PEmax_c3 (n=27, P = 0.85), PE3b_c3 (n=22, P = 0.029). (C,D,G) Mean ± SD; *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Ordinary one-way ANOVA with Dunnett’s multiple comparisons (two-sided); adjusted P and 95% CI are stated in Data file S2.

To quantify correction efficiency, genomic DNA was isolated from the cerebral cortex of Scn1a^KT/+ and Scn1a^KT/KT mice and analyzed by deep sequencing. In heterozygous animals we observed 10.4 ± 5.3% correction of the humanized Scn1a-KT allele with PEmax_c1, 21.4 ± 11% with PEmax_c3, and 34.7 ± 14.5% with PE3b_c3 (Fig. 2C), with no editing being detected in control animals injected only with the C-terminal PE AAV (AAV control) or the PE in combination with a non-targeting pegRNA (PEmax_NT, fig. S2A). In homozygous mice, editing rates were in a similar range, reaching 8.7 ± 3.5% for PEmax_c1, 18.6 ± 8.4% for PEmax_c3, and 36.4 ± 8.1% for PE3b_c3 (Fig. 2D). Because PE expression was restricted to neurons whereas bulk DNA sequencing analyzed lysates that also included glial cells, the measured editing rates were expected to underestimate neuronal Scn1a-KT correction efficiencies. To address this, we quantified editing at the RNA level, thereby selectively assessing Scn1a transcripts expressed in neurons. Consistent with our expectation, correction rates in mRNA isolated from the cortex and hippocampus of PE3b_c3-treated Scn1a^KT/+ mice were substantially higher, with 81.2 ± 5.9% and 82.4 ± 5.4%, respectively (Fig. 2E).

We next assessed whether editing efficiencies differed across brain regions and isolated DNA from multiple brain areas of PEmax_NT, PEmax_c3 and PE3b_c3 treated Scn1a^KT/+ mice. Consistent with the biodistribution of AAV-PHP.eB after neonatal ICV delivery (35), editing rates were highest in the cortex (37.4 ± 0.8% with PE3b_c3; Fig. 2F), with substantial editing also detected in other telencephalic regions implicated in GEFS⁺, including the striatum (20.2 ± 2.5%) and hippocampus (33.8 ± 2.3%). In contrast, hindbrain regions showed minimal editing, with only 0.5 ± 0.7% correction in the cerebellum. A similar distribution in editing rates was observed in PEmax_c3-treated animals (fig. S2B), whereas no editing was observed in control animals treated with PEmax in combination with a non-targeting pegRNA (PEmax_NT; fig. S2C).

Lastly, we evaluated the specificity and safety of the prime editing treatment. At the targeted Scn1a^KT locus, no indels or scaffold integrations above background were observed with any of the tested pegRNAs (Fig. 2G; fig. S2A, D to I). In addition, no adverse effects on body weight (fig. S2, J to M) or survival were observed in wild-type (Scn1a^+/+) animals expressing PEmax under the hSyn1 promoter (fig. S2, N). Deep sequencing of GUIDE-seq (36, 37)-identified off-target binding sites of the pegRNA, together with selected CRISPOR(38)-predicted off-target binding sites, revealed no detectable off-target editing in PEmax_c3 and PE3b_c3 treated animals compared to PEmax_NT treated animals (fig. S3, Data file S3). Longitudinal analysis over 6 months further demonstrated stable on-target editing in PE3b_c3-treated mice, with no increase in off-target activity compared to animals analyzed at two months of age (fig. S4, A to C). Consistent with this, no significant differences in body weight were observed between PE3b_c3-treated and untreated mice over the observed period (P = 0,95 female; P = 0,80 male, fig. S4, D and E).

Together, these results demonstrated that AAV-mediated delivery of hSyn1-driven, intein-split PEmax enabled efficient and precise correction of the Scn1a-KT allele across multiple GEFS⁺-relevant brain regions, without inducing adverse effects or off-target mutations.

Prime editing extends the lifespan of Scn1a^KT/KT and Scn1a^KT/+ mice

Although no patients with homozygous SCN1A-K1270T mutations have been reported, Scn1a^KT/KT mice are viable at birth but develop seizure-associated death around postnatal day 17 (18). We therefore also performed neonatal ICV injection of AAV-PHP.eB encoding for the different PE systems into Scn1a^KT/KT mice to assess treatment efficacy in this more severe disease context. Treatment resulted in a median lifespan increase from 17 days in controls to 18.5 days (+8.8%) with PEmax_c1, 20.5 days (+20.6%) with PEmax_c3, and 23 days (+35.3%) with PE3b_c3 (Fig. 2H), with a strong correlation between editing rates and survival time across individual animals (Spearman = 0.634, Pearson r = 0.636, P = 0.0001; fig. S4F).

We next evaluated whether prime editing also improved survival in heterozygous Scn1a^KT/+ mice, which occasionally experience sudden unexpected death in epilepsy (SUDEP) within the first two months in untreated or control-treated groups (Fig. 2I) (18). Although no significant improvement in survival was observed with PEmax_c1 or PEmax_c3 (P = 0,57; P = 0,85), treatment with PE3b_c3 resulted in 100% survival during the observed period (Fig. 2I).

To assess whether an earlier intervention could further improve survival of Scn1a^KT/KT mice, we next delivered PEmax_c3 in utero at embryonic day E14-16 (fig. S5A). However, in contrast to neonatal ICV delivery, this approach resulted in reduced transduction efficiency and altered biodistribution of AAV-PHP.eB (fig. S5, B to D), leading to lower editing efficiencies in the cortex (6.3 ± 2.3%) and hippocampus (2.7 ± 1.4%) and no improvement in survival (fig. S5, E to I).

Prime editing rescues cortical inhibitory transmission in Scn1a^KT/+ mice

To assess whether PE treatment restored deficits in sodium channel function in Scn1a^KT/+ mice, we first performed a biophysical characterization of cortical fast-spiking neurons of Scn1a^+/+, Scn1a^KT/+, and PE3b_c3-treated Scn1a^KT/+ mice. To specifically target inhibitory neurons in cortical slices, Scn1a^KT/+ mice were crossed with GAD67-EGFP mice. Only green fluorescent neurons in cortical layer V were recorded (Fig. 3, A and B). Neurons were considered fast-spiking if their action potentials evoked by a 500 pA depolarizing current injection had half-width of ≤ 0.6 ms and action potential frequencies ≥ 100 Hz (Fig. 3,C and D). Compared to wild-type controls, action potentials of Scn1a^KT/+ mice displayed smaller amplitudes (mean amplitude 73.5 ± 8.3 mV versus 81.6 ± 2.7 mV in Scn1a^+/+ mice, Fig. 3E left) and a reduced maximum slope (mean slope 273.9 ± 57.2 mV/ms versus 356 ± 69.0 mV/ms in Scn1a^+/+ mice; Fig. 3E right). Both action potential amplitude and maximum slope were restored to near wild-type levels in PE3b_c3-treated Scn1a^KT/+ mice (Fig. 3E). Other biophysical properties (including capacitance, resting membrane potential, action potential threshold, rheobase, input resistance as well as input-output relationship) did not differ significantly between the three groups of mice (fig.S6, A and B, see Data file S2 for p values). To further investigate whether Scn1a^KT/+ mice would exhibit a deficit in cortical synaptic inhibition, we recorded spontaneously occurring inhibitory postsynaptic currents (sIPSCs) from layer V pyramidal neurons. These neurons receive the bulk of GABAergic input from local Scn1a-positive interneurons and therefore provide a sensitive read-out of the inhibitory tone in the cerebral cortex (Fig. 3, F to H). Average sIPSC amplitudes were not different between the three groups of mice (29.7 ± 17.3 pA versus 25.06 ± 14.35 pA and 33.79 ± 15.14 pA, for Scn1a^+/+, Scn1a^KT/+ and PE3b_c3 treated mice, respectively; Fig. 3I). However, sIPSCs of Scn1a^KT/+ mice occurred at significantly longer interevent intervals (1.56 ± 0.81 s versus 3.10 ± 2.10 s, for Scn1a^+/+ and Scn1a^KT/+ respectively; P = 0,0026 Fig. 3J), indicating a reduced frequency of inhibitory synaptic input in Scn1a^KT/+ mice, which was reversed by PE3b_c3 treatment (0.76 ± 0.54 s). Together, these results demonstrated that correction of the Scn1a-KT allele by prime editing in the brain restored synaptic inhibition in Scn1a^KT/+ mice.

Fig. 3. In vivo prime editing restores cortical inhibitory transmission in Scn1a^KT/+ mice.

(A) Schematic representation of recordings from green fluorescent cortical layer V interneurons. (B) Triple immunostaining showing overlap between recorded parvalbumin (PV)-positive cells (light blue) and enhanced geen fluorescent protein (EGFP)-expressing cells (green), stained using biocytin-streptavidin interaction (red). Scale bar: 100 µm. (C) Representative recordings of action potential firing from fast spiking interneurons in Scn1a^+/+ and Scn1a^KT/+ mice, and Scn1a^KT/+ mice treated with PE3b_c3. (D) Overlap of a single representative action potential from each group.. (E) Average amplitudes (left) and slope (right) of action potential upstroke (10 cells from 4 Scn1a^+/+ mice (grey), 9 cells from 4 Scn1a^KT/+ mice (yellow) and 9 cells from 4 PE3b_c3-treated Scn1a^KT/+ mice (blue)). One-way ANOVA followed by Tuckey’s post-hoc test (Fig. 3E-left F(2,25)=5,767, P=0,0087; Fig. 3E-right F(2,25)=10,48, P=0,0005). (F) Schematic representation of recordings from a pyramidal neuron in cortical layer V. (G) Pyramidal neuron from Scn1a^+/+ filled with biocytin and stained with streptavidin A647. (H) Representative recordings of sIPSCs from Scn1a^+/+ and Scn1a^KT/+ mice and Scn1a^KT/+ mice treated with PE3b_c3. (I) (left) Cumulative frequency distributions of amplitudes of sIPSCs from Scn1a^+/+ (grey), Scn1a^KT/+ (yellow) and PE3b_c3-treated (blue) Scn1a^KT/+ mice (10 cells from 4 mice for each group). (right) Average amplitudes of sIPSCs of each condition (each dot represents the average of the amplitudes of the first 21 sIPSCs over 3 min of recording). Scale bar: 100 µm. (J) (left) Cumulative frequency distributions of inter-events interval (IEI) of Scn1a^+/+ (grey), Scn1a^KT/+ (yellow) and PE3b_c3-treated Scn1a^KT/+ animals (blue) (10 cells from 4 mice for each group). (right) Average IEIs of sIPSCs of each condition (each dot represents the average of the IEIs of the first 21 sIPSCs over the first 3 min of recording). One-way nested ANOVA followed by Tuckey’s post-hoc test (Fig. 3I-right F(2,9)= 0.36, P = 0,71; Fig. 3K-right F(2,27)= 7.46, P = 0,0026). Values are mean ± SD. *P < 0.05, **P < 0.01 and ***P < 0.001

Scn1a^KT/+ mice show elevated locomotor activity

To assess potential GEFS⁺ associated comorbidities (39), we performed a battery of behavioral tests including the open field test, novel object exploration test, three-chamber sociability test (3-CST) and rotarod testing (Fig. 4 and fig. S7). In general, Scn1a^KT/+ mice displayed increased locomotor activity, reflected by increased total distance travelled and higher locomotor velocity in the open field test and the novel object exploration test (Fig. 4, B to E). Neither test revealed signs of altered anxiety-like behavior, as time spent in the center and periphery of the arena was comparable between Scn1a^+/+ and Scn1a^KT/+ mice (Fig. 4, F to L). Sociability assessed in the 3-CST was also unaltered (Fig. 4, M to P). In general, altered locomotion was also present in PE3b_c3-treated Scn1a^KT/+ mice, suggesting that it may arise from prenatal circuit rearrangements not readily reversed by the PE3b_c3 treatment (Fig. 4). No significant interactions between genotype/treatment and sex were observed in the open field test and novel exploration test (fig. S7, see Data file S2 for p values). Lastly, no deficits in motor coordination assessed in the rotarod test were observed (Fig. 4, Q and R).

Fig. 4. Some behavioral phenotypes improve with prime editing treatment in Scn1a^KT/+ mice.

(A to R) Open field test (OFT), novel object exploration (NOE) test, three-chamber sociability test (3-CST) and rotarod testing were performed in Scn1a^+/+, Scn1a^KT/+, and Scn1a^KT/+ mice treated with PEmax_c3. (A) Schematic representation of the experiment setup (blue square = periphery; yellow square = center; red circle = new object) (B and C, F to H) OFT results: (B) total distance traveled (cm), (C) mean velocity (cm/s), cumulative duration (s) spent in the center (F) and periphery (G), and (H) ratio of time spent in the center versus periphery (Tc/Tp). Scn1a^+/+ (n=21), Scn1a^KT/+(n=28), and Scn1a^KT/+ mice treated with PEmax_c3 (n=26). (D and E, I to L) NOE test results: (D) total distance traveled (cm). (E) mean velocity (cm/s). (I to K) cumulative duration (s) spent in the center (I), periphery (J), and interacting with the object (object exploration, K). (L) Ratio of time spent in the center versus periphery (Tc/Tp). (M) Schematic of the 3-CST experimental setup: Phase 1 (habituation), Phase 2 (sociability) and Phase 3 (social novelty); 10 min per phase. (N to P) 3-CST results: (N) Total distance traveled across the three phases. (O and P) Cumulative duration (%) spent sniffing the object (empty cage) or an unfamiliar mouse (O), and with the familiar or unfamiliar mouse (P). Scn1a^+/+(n=17), Scn1a^KT/+(n=23), and Scn1a^KT/+ mice treated with PEmax_c3 (n=10). (Q) Schematic of the rotarod test experimental setup: 4 days of habituation, 1 test day (5 trials T1-T5). (R) Average latency to fall(s) across the five test trials. Scn1a^+/+(n=8), Scn1a^KT/+(n=7), and Scn1a^KT/+ mice treated with PEmax_c3 (n=10). Behavioral tracking was generated using EthoVision XT. For OFT, NOE and rotarod tests, one-way ANOVA followed by Tukey’s multiple comparisons test was applied. For 3-CST, two-way ANOVA with Tukey’s multiple comparisons test was applied. Values are mean ± SD. Statistical differences indicated at *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Fig. 4O (F(2,47) = 0.66, P = 0.52) and Fig. 4P (F(2,47) = 1.051, P = 0.36). Full statistical analysis stated in Data file S2. Fig. 4, A, M and Q were created in BioRender. Pietrafesa, F. (2026) https://BioRender.com/djnwp09.

Prime editing markedly lowers the incidence of febrile seizures in Scn1a^KT/+ mice

We next investigated whether PE-mediated restoration of the cortical inhibitory tone in Scn1a^KT/+ brains could protect animals from febrile seizures, the hallmark clinical feature of GEFS⁺. To this end, mice were exposed to a temperature-controlled chamber that gradually raised their body temperature over 30 min to a maximum of 44 °C (Fig. 5A). Ictal events, which displayed characteristic EEG signatures of heat-evoked seizures (fig. S8, A to D), were recorded by continuous video monitoring.

Fig. 5. In vivo prime editing reduces frequency of febrile seizures in Scn1a^KT/+ mice.

(A) Schematic representation of the heating protocol used for inducing febrile seizures in a custom-built heating chamber. (B) Percentage of Scn1a^KT/+ mice experiencing heat-induced febrile seizures after treatment with different AAV vectors. AAV Scn1a^+/+ refers to Scn1a^+/+ mice injected with the AAV control vector. Numbers above bars indicate animals with seizures and total animals. Statistical comparisons were performed using Fisher’s exact test (two-sided): odds ratio and 95% Cl are stated in Data file S2. (C) Editing efficiency plotted for mice with febrile seizures and without febrile seizures. All ICV-treated mice were pooled together for this analysis. (D to G) Characterization of heat-induced febrile seizures in Scn1a^KT/+ mice after ICV treatment with different AAV vectors. Seizures were scored for threshold temperature (D), latency (E), duration (F) and severity (G). A revised Racine scale (40) was used to measure severity. (D to G) Ordinary one-way ANOVA with Dunnett’s multiple comparisons (two-sided); ** P< 0.01 ; adjusted P and 95% CI are included in Data file S2.

As expected from previous studies (18), untreated Scn1a^KT/+ mice were highly susceptible to febrile seizures, which occurred in 78.6% of AAV-control animals and 76.9% of PBS-treated animals. In contrast, only 8.3% of wild-type littermates experienced heat-evoked seizures (Fig. 5B). Treatment with PEmax_c1, which produced relatively low correction rates, resulted in only a modest reduction in heat-evoked seizure incidence (69.2%; Fig. 5B). However, treatment with PEmax_c3 or PE3b_c3, which resulted in substantially higher editing efficiencies, led to a marked reduction of febrile seizures to 13.3% or 18.75%, respectively (Fig. 5B). Although there was a trend towards an inverse correlation between cortical editing rates and seizure frequency when analyzing individual animals, this did not reach significance (p = 0.14; Fig. 5C). Supporting that high prime editing rates may be critical for a therapeutic effect, in utero delivery, which yielded 6.3 ± 2.3% and 2.7 ± 1.4% of cortical and hippocampal editing, respectively, failed to lower seizure prevalence or characteristics of the seizures (fig. S9, A to E).

Next, we analyzed characteristics of febrile seizures in untreated and PE-treated Scn1a^KT/+ mice. Seizure threshold temperature and latency did not differ among groups (Fig. 5, D and E), and although we observed a trend towards milder seizures in mice treated with PEmax_c3 and PE3b_c3, as indicated by shorter seizure durations and lower scores on the Racine Scale (40), these did not reach significance (Figs. 5, F and G; fig. S8E). No differences in febrile characteristics were observed between untreated, PBS-treated, AAV control-treated and PEmax_NT-treated animals (fig. S8, F to J). Lastly, we also analyzed a group of PE3b_c3-treated animals at 6 months of age. This group showed a similar reduction in febrile seizures (30%) and comparable seizure characteristics to PE3b_c3 treated animals analyzed at 2 months of age (fig. S9, F to J), indicating that the treatment effect was durable. In summary, these findings demonstrated that correction of the pathogenic Scn1a^KT allele via prime editing markedly reduced the frequency of febrile seizures but did not reduce severity once they occurred.

Discussion

Correcting the root cause of SCN1A-associated epilepsies has long represented a major therapeutic objective, yet conventional gene-addition strategies face two major hurdles. First, dose sensitivity, because elevating Na_V1.1 beyond its physiological range can provoke hyperexcitability of fast-spiking inhibitory neurons (16). Second, cell-type specificity, because Na_V1.1 expression is naturally confined to GABAergic interneurons, and ectopic expression in excitatory neurons risks mistimed or aberrant currents that disrupt network balance (41). Precise in situ correction of the pathogenic allele would bypass these limitations; however, the low HDR activity in post-mitotic neurons have thus far precluded applications of clustered regularly interspaced short palindromic repeats (CRISPR)-CRISPR-associated (Cas) nuclease systems.

Here, we employed prime editing to correct the Scn1a^KT mutation. Like base editing, prime editing does not rely on DNA double strand break formation and HDR, thereby reducing the risk of unintended indels or chromosomal rearrangements. Moreover, prime editing offers greater versatility; whereas base editing is mainly limited to inducing transition mutations (C•G to T•A or A•T to G•C), prime editors can mediate a broad spectrum of small genetic alterations, including all types of point mutations, small insertions, and deletions. In this study we demonstrated an average Scn1a-KT correction rate in the cortex of 10.4% and 21.4% when combining PEmax with two different pegRNAs, and 34.7% when employing the PE3b_c3 system that co-introduces an additional nick opposite of the newly synthesized strand. These values underestimate neuron-specific Scn1a-KT correction, because deep sequencing was performed on DNA isolated from whole-cortex lysates, of which 34% are non-neuronal cells that do not express the hSyn1-driven editor (42). In line with this reasoning, cortical mRNA sequencing revealed 81.2% editing in PE3b_c3 treated mice.

Although therapeutic effects in Scn1a^KT/+ mice were observed with all tested PE systems, only PE3b_c3 conferred full protection from mortality and robust reduction in febrile seizure frequency. This suggests that achieving correction above a certain threshold within the AAV-PHP.eB-transduced brain regions is necessary to restore network stability in GEFS⁺. Consistent with this notion, Mich et al. recently reported a dose-dependent phenotypic rescue in a haploinsufficient Dravet model by delivering an interneuronally-expressed SCN1A copy via the same route and capsid - neonatal ICV injection of AAV-PHP.eB (43). Interestingly, although PE3b_c3 treatment reduced febrile seizure incidence to near wild-type levels, seizure threshold, latency, duration, and severity were not measurably altered among seizures that still occurred. Thus, in this model, the primary therapeutic benefit of prime editing appears to be lowering the probability of febrile seizure occurrence rather than modifying seizure dynamics after onset.

In addition to treating heterozygous Scn1a^KT/+ mice, we further analyzed homozygous Scn1a^KT/KT animals. These mice exhibit pronounced premature SUDEP-like mortality, more closely resembling the phenotype observed in Dravet syndrome mouse models carrying heterozygous null mutations in Scn1a. Even the most efficient prime editing construct tested, PE3b_c3, conferred only a modest survival benefit in this genetic background, suggesting that higher editing rates may be required to achieve therapeutic efficacy in Dravet syndrome compared to GEFS⁺. However, to accurately define the necessary editing threshold for full phenotypic rescue, future studies in mouse models harboring heterozygous null alleles—thereby more faithfully recapitulating the genetic basis of Dravet syndrome—will be essential.

For clinical translation, specificity is a critical consideration, particularly when the editor is delivered via AAV and expressed over a long timeframe. The pegRNA design used in this study destroys the PAM upon correction of the Scn1a-KT mutation, preventing re-targeting of the edited allele. Accordingly, we observed no accumulation of indels or scaffold integrations at the on-target locus. Further supporting the safety of our approach, deep amplicon sequencing of GUIDE-seq-identified and selected CRISPOR-predicted off-target sites revealed negligible unintended edits at other genomic regions. Lastly, restricting editor expression to neurons via the hSyn1 promoter (31) - combined with the neurotropic AAV-PHP.eB capsid - suppressed editing in non-neural tissues.

Important translational challenges remain before clinical implementation can be envisioned. First, immune responses against non-self components such as SpCas9, the Moloney murine leukemia virus reverse transciptase (MLV-RT) and intein-split fragments, as well as pre-existing immunity to AAV capsids pose a critical obstacle and may limit the effectiveness of in vivo prime editing in primates (44, 45). Second, although ICV injections enable efficient and widespread brain transduction in neonatal mice, this approach does not directly translate to clinical practice. In humans, achieving a comparable central nervous system (CNS)-wide delivery would require invasive intrathecal and intraventricular administration, which carry substantial procedural risks, particularly in very young patients. A clinically more attractive alternative would be systemic administration of AAV vectors capable of efficiently crossing the human blood–brain barrier (BBB). However, although several AAV capsids have been developed that efficiently cross the BBB in mice, their efficacy is more variable in non-human primates (46). Third, the dual-vector system required for the delivery of the full-length PE imposes constraints on dosing, as titers that ensure neuronal co-transduction with two AAVs need to be applied. The development of smaller PE systems that fit on single AAV vector would therefore be critical for future clinical translation. Lastly, while high editing rates were achieved in cortical and other forebrain regions, including the hippocampus and striatum, editing efficiencies were markedly lower in hindbrain structures, with minimal editing detected in the cerebellum. Given that SCN1A-related epilepsies are thought to involve brainstem circuits leading to respiratory dysfunction and SUDEP, limited editing in these regions may restrict the ability to ameliorate these complications (47–49). Consistent with this, in homozygous Scn1a^KT mice, which exhibit a markedly severe phenotype more resembling Dravet-like epilepsies, survival was only slightly improved, suggesting that higher and more uniform editing efficiencies across multiple brain regions may be required to achieve effective therapeutic benefit in such severe disease contexts.

In conclusion, these results highlight prime editing as a potential therapeutic modality for correcting SCN1A‐related GEFS⁺. As prime‐editor designs and delivery vectors continue to improve, this strategy holds promise not only for a broad spectrum of SCN1A‐driven epilepsies, but potentially also for other monogenic neurological disorders.

Materials and Methods

Study Design

The objective of this study was to evaluate the efficiency, specificity, and functional impact of prime editing mediated correction of the pathogenic SCN1A-K1270T mutation associated with febrile seizures and epilepsy. The primary hypothesis was that in vivo prime editing would result in molecular correction and functional phenotypic improvement. All experiments were conducted as controlled laboratory studies in cellular systems and a disease-relevant mouse model.

In vitro experiments were performed in engineered reporter cell to optimize pegRNA design and prime editor configurations by assessing on-target editing efficiency and specificity. All animal experiments were approved by the Kantonales Veterinäramt Zürich under approval number ZH022/2022 and were performed in accordance with institutional and national guidelines. In vivo experiments were conducted in Scn1a-K1259T mice following adeno-associated virus–mediated delivery of prime editor components by neonatal intracerebroventricular injection. Molecular, electrophysiological, behavioral, and seizure-related endpoints were selected prospectively to assess editing outcomes across multiple biological scales. Sample sizes for in vivo experiments were based on handling capacity using Fermi approximations. No formal power calculations were performed. Sample sizes were not altered during the course of the study. No predefined rules for early stopping of data collection were applied. Inclusion and exclusion criteria were established prospectively. Animals were excluded only for medical reasons unrelated to the expected phenotype. No biological outliers were excluded from analyses.

Primary endpoints included molecular quantification of prime editing outcomes and seizure susceptibility in febrile seizure assays. Secondary endpoints included EEG/EMG recordings, behavioral assessments, and ex vivo slice electrophysiology. Multiple endpoints were analyzed using appropriate statistical tests, with corrections applied where indicated, as described in the Statistical Analysis section.

All in vitro experiments were independently repeated at least three times unless otherwise stated. For in vivo studies, individual animals were treated as biological replicates, and replicate numbers are reported in the figure legends. Results were reproduced across independent cohorts where applicable. Animals were randomly assigned to treatment groups when feasible. Investigators were blinded to genotype and treatment during behavioral testing, seizure induction, EEG/EMG acquisition and analysis, and electrophysiological recordings. Data analysis was performed in a blinded manner until completion of primary analyses.

Animal studies

Animal experiments were performed in accordance with protocols approved by the Kantonales Veterinäramt Zürich and in compliance with all relevant ethical regulations. C57BL/6J mice and Scn1a^em10dwd/J (Scn1a^KT, JAX stock #036177(18)) were housed in a pathogen-free animal facility at the Institute of Pharmacology and Toxicology of the University of Zurich under temperature- and humidity-controlled conditions on a 12-hour light/dark cycle, with food and water provided ad libitum. Mice were group-housed and provided with environmental enrichment unless experimental procedures required otherwise. Scn1a^KT mice carry a missense mutation (Scn1a-K1259T, in Ensembl transcript Scn1a-201), corresponding to the human SCN1A-K1270T variant. Mice were kept in a temperature- and humidity-controlled room on a 12-hour light/dark cycle. Mice were fed a standard laboratory chow (Kliba Nafag no. 3437 with 18.5% crude protein). Both sexes were used in the experiments and pooled for subsequent analysis, unless noted otherwise.

Neonatal injections and in utero injections

The average weight of neonatal (1 day) mice was 1.5 g. For intracerebroventricular injections (ICV) new-born animals (PND1) received 1 × 10¹¹ (2 µL per injection site) AAV vector genomes (vg) per animal and construct. For in utero injections (50), embryos at E14-16 were pulled out through a surgical incision from the anesthetized dam (mother) and injected via ICV route with a maximal dose of 2 × 10¹⁰ vg. Injected embryos were put back and the surgical incision was sutured.

Heat-induced seizure experiments

To assess the occurrence of heat-induced behavioral seizures, wild-type and Scn1a^KT/+ mice from both sexes (weigh ≥ 15 g), were included in the study. Prior to testing, mice were anesthetized with 1–2% isoflurane for 30–45 seconds. A rectal temperature probe (RET-3, Braintree Scientific) was then inserted and secured to the tail to monitor body temperature throughout the procedure. Following anesthesia, the animals were allowed to recover in their home cages for approximately 15 min to resume normal activity and stabilize their core body temperature at 36–37°C. The heat-induced seizure test was conducted in a in house custom-built forced air chamber (18), preheated to 45°C one hour before the experiment. The chamber was equipped with a digital temperature controller and a thermocouple device to precisely control and maintain the desired temperature. The temperature was gradually increased every 7 min until a maximum of 60°C was reached. A Plexiglas front panel was installed to facilitate video recording of the entire heating session for subsequent analysis. Seizure events were scored by an experimenter blind to the mouse genotype, using a revised Racine scale (40): 2 facial jerking; 3 neck jerks; 4 clonic seizure (lying on side/belly); 5 clonic, tonic-clonic seizure (lying on back); 6 clonic, tonic-clonic seizure (wild jumping); 7 tonic extension, leading to respiratory arrest and death.

For each mouse, several parameters were recorded: seizure threshold temperature, latency to seizure onset, seizure severity score and jumping during the heating phase. The seizure threshold was defined as the temperature at which the first observable behavioral seizure of any severity occurred. Seizure latency was measured as the time (in min) from the introduction of the mouse into the heated chamber (time = 0) to the onset of the first seizure at the seizure threshold. The experiment was terminated under either of the following conditions: (i) upon completion of a seizure episode, (ii) when the mouse’s body temperature reached 44°C or (iii) the time limit of 30 min was reached.

EEG electrode implantation, EEG-EMG recording and analysis

Surgeries for EEG electrode implantation were performed on anaesthetized adult male and female mice (6-8 weeks). At positions in the frontal cortex (ML: ± 1.5 mm, AP + 1.5 mm), parietal cortex (ML: ± 1.8 mm, AP – 2.5 mm), and the cerebellum (ML: 0 mm, AP - 6 mm) a total of five gold-plated miniature screws (0.9-mm diameter) were implanted. The screw in the right frontal cortex and the left paretal cortex were connected via copper wires to the EEG recording system, while the screw in the cerebellum was connected and utilized as a reference. The implants and copper wires were fixed using dental cement (Paladur 2-component system and TetricEvoFlow). Additionally, for electromyography (EMG) measurement, two gold wires (0.2-mm diameter), were implanted in the left and right neck muscles. After the surgery, mice had a 1-week recovery period before being subjected to EEG recording conditions. EEG and EMG signals were continuously recorded for 72 hours on a 12h/12h day/night cycle, starting recordings at light onset (ZT0). After the initial continuous recording session, mice were subjected to heat induced seizure experiments as described here.

Signal acquisition parameters and processing was performed as previously described (51). In short, parameters were set using Signal Express NI 2015 controlled via LABVIEW. Processing was divided in signal amplification (factor ~2000), analogue filtered (high-pass filter: −3 dB at 0.016 Hz; low-pass filtered: −3 dB at 40 Hz) and sampled with 512 Hz. The processed signal was decimated and stored in European Data Format (EDF) files at 128-Hz resolution. Lastly, EEG and EMG signals were digitally filtered (between 0.1 and 40 Hz for EEG and 10 to 30 Hz for EMG) and stored with a final resolution of 128 Hz.

For heat-induced seizure analysis, processed EEG data was loaded into MNE-Python, signals from heat-induced seizure experiments were separated for each mouse and further filtered for exact start and end timepoints. The line-length across the EEG signal was calculated, which was further used as threshold to detect individual peaks of epileptic spikes (factor 3). A seizure was further defined, if 30 epileptic spikes were detected within a sliding window of 10 seconds. Finally, detected seizures were analyzed individually to avoid false positive seizure events due to complications in the measurement setup(52).

Slice preparation and electrophysiology

Coronal brain slices were prepared from 3 - 6 week old Scn1a^+/+, Scn1a^KT/+ and PE3b_c3 treated mice of both sexes. Mice were deeply anesthetized with 5% isofluorane and decapitated. After removal of the skull, brains were transferred to ice-cold cutting solution containing (in mM): 65 NaCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 2.5 KCl, 105 sucrose, 25 glucose, 7 MgCl₂, 0.5 CaCl₂, pH 7.4 continuously bubbled with 95% O₂ and 5% CO₂. Cerebellum and hindbrain were removed and the remaining brain was glued on a metal block mounted on a vibratome slicer (Microm, model HM 650 V) to prepare 300-μm thick coronal sections. Slices were transferred to 37°C artificial cerebrospinal fluid (aCSF) containing (in mM): 120 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 26 NaHCO₃, 5 HEPES, 1 MgCl₂, 2 CaCl₂ and 14.6 glucose, equilibrated with 95% O₂, 5% CO₂ (pH 7.40). Slices were superfused continuously with aCSF at a rate of 1 - 2 ml / min. Whole cell recordings were made at room temperature using borosilicate glass patch pipettes (resistance 4 – 8 MΩ) filled with intracellular solution containing (in mM): 130 K⁺-gluconate, 5 NaCl, 1 EGTA, 10 HEPES, 5 Mg-ATP, 0.5 Na-GTP (pH 7.35, 280 – 290 mosm / l).

Intrinsic biophysical properties in fast-spiking GAD67-EGFP positive neurons of cortical layer V were investigated. Fast-spiking neurons were identified based on their action potential firing evoked by 1 s long current injections of 500 pA. A neuron was classified as fast spiking if its action potential frequency was > 100 Hz and its action potential half-width was ≤ 0.6 ms. Capacitance (C) was determined under voltage-clamp from the charge transfer carried by the capacitative current elicited by a 100-ms hyperpolarizing voltage step of -10 mV. Resting membrane potentials (RMP) were recorded immediately after switching from voltage-clamp to current-clamp. Input resistance (R_input) was determined from a sequence of 2-s hyperpolarizing current steps of increasing amplitude (−5 pA increment, delivered every 10 s). Rheobase was determined through a ramp depolarization protocol (200 mV / s slope). Action potential (AP) thresholds were calculated at rheobase and defined as the voltage at which dV/dt first reached ≥ 5 mV / ms.

Inhibitory network activity was assessed in whole-cell recordings made from layer V pyramidal neurons using the following internal solution containing (in mM): 130 CsMeS, 3 NaCl, 10 HEPES, 0.6 EGTA, 4 Mg-ATP, 0.3 Na-GTP, 5 QX-314 (pH 7.35, 280 - 290 mosm / l). Biocytin (0.1%) was included in all internal solutions for subsequent staining of the recorded neuron. Neurons were voltage clamped at 0 mV (i.e., close to the reversal potential of AMPA receptor mediated synaptic currents) to record sIPSCs in isolation. Data were acquired using an EPC9 amplifier (HEKA Elektronik, Lambrecht, Germany) controlled with Patchmaster, version 2×80 acquisition software, dual-stage filtered with Bessel filter 1 at 10 kHz + Bessel filter 2 at 2.9 kHz and sampled at 50 kHz. After finishing the biophysical characterization, recording pipettes were left attached to the recorded neuron for 20 min. Slices were then briefly fixed for 30 min at room temperature in 4% PFA and subsequently washed with PBS. Biocytin was stained with streptavidin conjugated with Alexa fluor-647 (1 : 500). The first 21 IPSCs were used for analysis using the Mini Analysis program (Version 6.0.7). This number [21] was chosen because it was the minimum number of events per cell recorded during three minutes.

Statistical analysis

All statistical analyses were performed using GraphPad Prism 10.3.0. Unless stated otherwise, data are presented as biological replicates with means ± standard deviation (SD). Sample sizes, exact statistical tests, and numbers of replicates are reported in the corresponding figure legendsor in Data file S2.

Comparisons between two groups were performed using two-tailed Student’s t tests or Welch t tests, as appropriate. If not stated otherwise, comparisons involving more than two groups were analyzed using one-way, including nested analysis of variance (ANOVA) when data contained hierarchical structure, or two-way ANOVA, followed by Tukey’s or Dunnett’s multiple-comparisons tests. Categorical data were analyzed using Fisher’s exact test, and survival data were analyzed using Kaplan–Meier estimates with log-rank (Mantel–Cox) tests.

For parametric analyses, data distribution and variance were assessed to ensure that test assumptions were met. When assumptions were not met, alternative statistical approaches were applied as indicated in the figure legends. All statistical tests were two-sided, and P < 0.05 was considered statistically significant.

Descriptions of plasmid generation, cell culture, transfection and genomic DNA preparation, reporter cell line generation, AAV production, brain region dissection, next generation sequencing, GUIDE-seq, transcardiac perfusion, immunohistochemistry, and behavioral assays are provided in the Supplementary Materials and Methods.

One Sentence Summary.

OVERLINE: GENE THERAPY

Prime editing corrects a pathogenic human SCN1A mutation in mice, reducing seizure frequenciesand improving survival in this GEFS⁺ model.

Data, code and materials availability

All data associated with this study can be found in the paper or supplementary materials. DNA sequencing data have been deposited under accession number PRJNA1293326 (NCBI Sequence Read Archive). Code for analysis of EEG data is provided on Zenodo (52). All materials used in this study are commercially available or will be supplied upon reasonable request.