CRISPR activation for SCN2A-related neurodevelopmental disorders

Abstract

The majority of neurodevelopmental disorders (NDDs) single gene diagnoses act via haploinsufficiency, where only one of the two gene copies is functional¹. SCN2A haploinsufficiency is one of the most frequent causes of NDD, often presenting with autism spectrum disorder (ASD), intellectual disability (ID), and, in a subset of children, refractory epilepsy². Here, using SCN2A haploinsufficiency as a proof-of-concept, we show that upregulation of the existing functional gene copy via CRISPR activation (CRISPRa) can rescue neurological-associated phenotypes in Scn2a haploinsufficient mice. We first show that restoring Scn2a expression in adolescent heterozygous Scn2a conditional knock-in mice rescues electrophysiological deficits associated with Scn2a haploinsufficiency (Scn2a^+/−). Next, using an adeno-associated virus (AAV) CRISPRa-based treatment in adolescent mice, we show that we can correct intrinsic and synaptic deficits in neocortical pyramidal cells, a major cell type that contributes to NDDs and seizure etiology in SCN2A haploinsufficiency. Furthermore, we find that systemic delivery of CRISPRa protects Scn2a^+/− mice against chemoconvulsant-induced seizures. Finally, we also show that AAV-CRISPRa treatment rescues excitability in SCN2A haploinsufficient human stem-cell-derived neurons. Our results showcase the potential of this therapeutic approach to rescue SCN2A haploinsufficiency and demonstrates that rescue even at adolescent stages can ameliorate neurodevelopmental phenotypes.

Neurodevelopmental disorders (NDDs), including intellectual disability, autism spectrum disorder (ASD), and early-onset epilepsy, affect over 3% of the general population in the US³. Recent advances in gene discovery have facilitated the identification of 387 high confidence genes associated with NDDs from autosomal germline de novo variants^4–6. Two hundred and twenty-seven of these NDD genes are either known or predicted to cause ASD due to haploinsufficiency, where the mRNA/protein levels from the residual gene copy are insufficient to enable the typical function of the gene¹. Gene replacement therapy could potentially restore transcriptional balance and rectify deficits in haploinsufficient diseases by delivering functional gene copies. However, in a majority of NDD-related genes, traditional gene therapy vectors cannot be used. Gene therapy relies primarily on using recombinant adeno-associated virus (rAAV) for transgene delivery due to the limited pathogenicity and long-term transgene expression of the virus⁷. However, rAAV has a limited packaging capacity (4,700 base pairs optimal packaging capacity minus ~1,700 base pairs bp needed for transgene expression)⁸. Examination of the coding sequence (CDS) length of the 227 haploinsufficient NDD associated genes shows that 117 exceed rAAV vector capacity (Supplemental Table 1), excluding them from traditional gene replacement therapy approaches⁹.

Here, we tested whether cis-regulation therapy (CRT) could be a viable approach for treating NDD-associated haploinsufficiency. CRT utilizes a nuclease-deficient gene-editing system, such as a “dead” Cas9 (dCas9), fused to a transcriptional modulator to target gene regulatory elements (i.e. promoter or enhancer) and alter gene expression¹⁰. For this study, we used dCas9 fused to a transcriptional activator, termed CRISPR activation (CRISPRa), to upregulate gene expression and correct haploinsufficiency by targeting a gene promoter. For our proof-of-concept, we utilized NDD-associated haploinsufficiency of the sodium voltage-gated channel alpha subunit 2 (SCN2A), a gene with a six kilobase-long CDS⁹. Variants in this gene can enhance channel function, resulting in neonatal-onset seizure phenotypes, or inhibit channel function, resulting in SCN2A-related NDDs². Loss-of-function (LoF) variants in SCN2A are one of the most prevalent genetically-defined subsets of NDD, with an estimated incidence of ~300 births per year in the US (about 1 in 12,000 births)¹¹. SCN2A-related NDDs are characterized by severe intellectual disability, global developmental delay, ASD, and seizures that can be refractory to traditional treatment methods².

SCN2A encodes the neuronal voltage-gated sodium channel, Na_V1.2, which is broadly expressed in the central nervous system, including in neocortical excitatory pyramidal cells whose dysfunction is implicated in NDD etiology¹². Within the first year of life in humans and before postnatal (P) day 7 in mice, Na_V1.2 plays an essential role as the primary Na_V responsible for action potential (AP) initiation in pyramidal cell axons¹³. After this period, Na_V1.2 is enriched throughout the somatic and dendritic domains in addition to the most proximal region of the axon initial segment (AIS)¹⁴. Scn2a haploinsufficiency in mice—either from birth (constitutive) or conditionally induced after P7—results in impairments in dendritic intrinsic excitability and excitatory synapse function^14,15 (Fig. 1a). This suggests that Na_V1.2 actively helps maintain dendritic function throughout life and raises the possibility that rescue from Scn2a haploinsufficiency can be therapeutic, even if such a rescue is administered later in life.

Fig 1. Genetic rescue of Scn2a haploinsufficiency in adolescent mice rescues electrophysiological deficits.

a. Summary of Scn2a haploinsufficiency effects on mouse mPFC layer 5b pyramidal neurons (L5PNs).

b. Strategy for genetic restoration of Scn2a in Scn2a^+/KI conditional knock-in model.

c. 2PLSM z-stack of layer 5 mPFC in acute slice from Scn2a^+/KI mouse without Cre injection. Arrows highlight GFP-negative, presumptive interneurons.

d. Schematic of rAAV-EF1α-Cre-mCherry injection into the mPFC of P30 Scn2a^+/KI mice. Electrophysiological and imaging experiments in Scn2a^+/KI and control mice were performed between P60–70.

e. APs per 300 ms stimulation epoch across a range of current amplitudes in Scn2a^+/KI Cre-negative (green) and Scn2a^+/KI Cre positive (red) L5PN.

f. Left: Representative phase-plane plots (dV/dt vs. voltage) of somatic APs across conditions. Right: Peak dV/dt of the first AP evoked by a near-rheobase current. n = 44 Scn2a^+/+, 37 Scn2a^+/−, 30 Scn2a^+/KI Cre−, 21 Scn2a^+/KI Cre+ cells. ****p < 0.0001, *p = 0.02, Holm-Šídák multiple comparisons test.

g. Left: 2PLSM imaging of bAP-evoked Ca transients (doublet bursts) in L5PN apical dendrites (>400 μm from soma) across conditions. Right: Ca transient amplitude of the first of 5 bursts (top) and area under the curve (AUC) (bottom). n = 49 Scn2a^+/+, 23 Scn2a^+/−, 29 Scn2a^+/KI Cre−, 26 Scn2a^+/KI Cre+ imaging sites across ≥7 L5PNs per condition from ≥2 animals. ****p < 0.0001. Holm-Šídák multiple comparisons test.

h. Left: AMPA and mixed AMPA/NMDA receptor-mediated EPSCs at −80 and +30 mV, respectively. NMDA receptor-mediated component measured 50 ms after stimulation (dotted line). Right: Quantification of AMPA:NMDA ratio. n = 8 Scn2a^+/+, 9 Scn2a^+/−, 8 Scn2a^+/KI Cre−, 9 Scn2a^+/KI Cre+ cells. *p = 0.02, **p = 0.004, ***p = 0.009, ****p = 0.002, Holm-Šídák multiple comparisons test. Circles are single cells (f, h) or single imaging sites (g); box plots are medians, quartiles, and 100% tails.

As the intrinsic and synaptic deficits of SCN2A haploinsufficiency can be readily measured via electrophysiology, we hypothesized that SCN2A could provide an efficient and quantifiable model for CRT-based interventions. We administered rAAV-based CRISPRa targeting the Scn2a promoter to Scn2a heterozygous null mice and neurons differentiated from human embryonic stem cells (hESCs)¹⁶ and demonstrate that this approach can rescue excitability deficits in both models. In mice, we also observed that CRISPRa can rescue these deficits at adolescent stages and protect Scn2a heterozygous mice against chemoconvulsant-induced seizures. Combined, our work showcases a potential therapeutic approach that could be applied to many genes that, when haploinsufficient, are associated with NDDs and suggests that rescue later in life could ameliorate electrophysiological phenotypes associated with Scn2a haploinsufficiency.

Adolescent Scn2a rescue is therapeutic

To test the feasibility of increasing functional gene copy number to rescue the electrophysiological deficits associated with Scn2a haploinsufficiency, we first developed a conditional knock-in mouse line. In this mouse, termed Scn2a^+/KI, exons 3–5 of one Scn2a allele were flipped and flanked by LoxP sites, with an eGFP sequence in frame to visualize cells that express Scn2a (Fig. 1b). Acute slices were made from these animals, and neurons were visualized with 2-photon laser scanning microscopy (2PLSM) of GFP fluorescence against a background image of neuropil imaged with 2PSLM scanning differential interference contrast (DIC) imaging (Fig. 1c). GFP-positive neurons targeted for whole-cell recordings had electrophysiological features indicative of Scn2a^+/− pyramidal cells, including a decrease in the peak change in voltage during AP depolarization (peak dV/dt) relative to Scn2a^+/+ controls (Fig. 1c–f). Moreover, parvalbumin-positive interneurons were GFP negative, consistent with known Na_V1.2 functional expression patterns (Extended Data Fig. 1)^14,17.

Previously, we showed that inducing Scn2a haploinsufficiency during adolescence impairs intrinsic and synaptic function¹⁵. To test whether these deficits can be rescued by gene reinstatement during adolescence, we injected rAAV-EF1α-Cre-mCherry into the medial prefrontal cortex (mPFC) of these Scn2a^+/KI mice around P30 (Fig. 1d). After at least four weeks to allow for transgene expression, mCherry-positive neurons were targeted for whole-cell recordings in acute slices. In the presence of Cre, multiple measures of neuronal and synaptic function were comparable to those found in wild-type (WT) mice, including peak AP dV/dt, AP-burst evoked calcium influx in the apical dendritic tuft, and AMPA:NMDA ratio, which is a common proxy for synapse strength and is abnormally low in Scn2a^+/− cells¹⁴ (Fig. 1f–h). Together, these data show that features of intrinsic and synaptic function that depend directly and indirectly on proper Na_V1.2 function can be restored to WT levels upon reactivation of the second allele in adolescent stages. They also suggest that the development of a CRT approach to increase the expression of the functional SCN2A allele could rescue deficits caused by haploinsufficiency.

Scn2a-CRISPRa optimization in vitro

To optimize an rAAV-based CRISPRa approach to rescue Scn2a haploinsufficiency, nine different sgRNAs targeting the Scn2a promoter were screened for their ability to upregulate Scn2a in mouse neuroblastoma cells (Neuro-2a). Guides were co-transfected with a Staphylococcus aureus (sa) dCas9 fused to the transcriptional activator VP64 (Supplemental Table 2). sadCas9-VP64 is compact in size (3.3 kb), allowing it to fit into an rAAV vector, and has been shown to promote upregulation to therapeutic levels in other forms of haploinsufficiency¹⁸. Three sgRNAs increased Scn2a mRNA levels and were packaged into an rAAV-DJ serotype vector, which provides high expression levels in many cell types and tissues, including neurons¹⁹. rAAV-sgRNA viruses were then co-transduced with the rAAV-sadCas9-VP64 activator, both using the AAV-DJ serotype, into Neuro-2a cells for five days. One of the three rAAV-sgRNA was found to significantly increase Scn2a mRNA expression by two-fold and was selected for subsequent mouse studies (Extended Data Fig. 2a–b).

To test for off-target effects, we carried out RNA-seq using two biological replicates and three technical replicates for each condition (six samples per condition) and analyzed the mRNA expression of all genes within the same topologically associated domain (TAD) and all Scn sodium channel genes following plasmid CRISPRa transfection into Neuro-2a cells. We observed that Scn2a was the only upregulated gene in its TAD and also the only upregulated Scn sodium channel compared to negative control using RNA-seq (Extended Data Fig. 3a–c). We observed additional genes to be upregulated compared to the negative control, with gene ontology (GO) analysis showing an enrichment for viral response (Extended Data Fig. 3d–e) likely from our AAV infection. In addition, we analyzed differentially expressed (DE) genes that are close to (10kb) sgRNA sequence off-targets using Cas-OFFinder²⁰ with sequences up to 3 mismatches and 2 bulges. With these criteria, we found six DE genes from our RNA-seq (Extended Data Fig. 3f). These include Ptpn14 and one transcript with unknown function (ENSMUSG00000125254) which have a sequence hit with 3 mismatches and 1 bulge near the gene. We also found three genes (Zfp697, Ifi44l, Csmd1) and one transcript (Gm17131) with unknown function that have 3 mismatches and 2 bulges. Thus, the odds of the Scn2a-targeted sgRNA binding are low. Furthermore, Ptpn14 regulates cell proliferation and is not known to have a major functional role in the brain²¹. The three other genes are related to viral response: Zfp697 is involved in the mediation of interferon-gamma response in muscle cells²², Ifi44l is interferon-induced and involved in anti-viral response²³, and Csmd1 is a complement inhibitor²⁴ that was downregulated. In summary, our results show that Scn2a is significantly upregulated with minimal off-target effects using our CRISPRa approach.

Cell-autonomous CRT in Scn2a^+/− mice

To test whether Scn2a-rAAV-CRISPRa (rAAV-sadCas9-VP64 + rAAV-Scn2a-sgRNA) could be used as a therapeutic intervention for Scn2a haploinsufficiency in adolescent mice, we stereotaxically injected Scn2a-rAAV-CRISPRa into the mPFC of constitutive Scn2a^+/− mice at ~P30 (Supplemental Table 3). Mice were injected unilaterally in one hemisphere and the opposite hemisphere was used as a biologically matched uninjected control (Fig. 2a–b). As a negative (no sgRNA) control, Scn2a^+/− mice were co-injected with rAAV-sadCas9-VP64 along with an rAAV-mCherry (termed Empty Vector). At least four weeks post-injection, the mPFC of the injected and uninjected hemispheres were dissected and extracted for qPCR analysis and electrophysiological recordings. Scn2a^+/− mice injected with Scn2a-rAAV-CRISPRa showed an upregulation of Scn2a mRNA of around 1.5-fold via qRT-PCR compared to the uninjected hemisphere, while rAAV-empty-mCherry injected mice showed no upregulation of Scn2a mRNA expression (Fig. 2c).

Fig. 2. Intracranial Scn2a-rAAV-CRISPRa in mPFC rescues excitability deficits in Scn2a^+/− neurons.

a. CRT design to rescue Scn2a haploinsufficiency.

b. Local injection of Scn2a-rAAV-CRISPRa (viruses injected in brackets) and coronal section detailing injection site, stained for mCherry and DAPI.

c. qPCR of Scn2a mRNA across conditions. n = 9 Scn2a^+/+ + CRISPRa, 6 Scn2a^+/+ + empty, 10 Scn2a^+/− + CRISPRa, 5 Scn2a^+/− + empty: 1.0 ± 0.06, n = 5 mice. **p = 0.0016; ***p = 0.0007. Mann-Whitney test.

d. AP phase-plane plots CRT-treated and empty-treated Scn2a^+/+ and Scn2a^+/− mice. Populations also compared to interleaved, uninjected controls. n = 16 Scn2a^+/+, 12 Scn2a^+/−, 24 Scn2a^+/++CRISPRa, 19 Scn2a^+/−+CRISPRa: 17 Scn2a^+/++empty, and 27 Scn2a^+/−+empty cells. ****p < 0.0001. n.s., not significant. Holm-Šídák multiple comparisons test.

e. AMPA:NMDA ratio, as in Fig. 1h, for Scn2a^+/−+CRISPRa (mCherry+, purple, n=7) and Scn2a^+/− (mCherry–, cyan, n=4) internal controls. Gray bar: 95% confidence interval of Scn2a^+/+ control data from¹⁴. **p = 0.006. Mann-Whitney Wilcoxon test.

f. bAP-evoked Ca imaging, as in Fig. 1g, in L5PNs from P55–84 Scn2a^+/+ (black, n=38 sites, 11 cells, 5 mice), Scn2a^+/− (cyan, n = 40 sites, 11 cells, 5 mice), Scn2a^+/− + CRISPRa (purple, n=18 sites, 5 cells, 5 mice), and Scn2a^+/+ + CRISPRa (gray, n = 29 sites 8 cells, 8 mice). **p = 0.0002, ***p = 0.0004, ****p < 0.0001, Mixed-effect model, Holm-Šídák multiple comparisons test.

Circles are animals (c), single cells (d-e) or imaging sites (f); box plots are medians, quartiles, and 100% tails.

To assess the phenotypic effects of Scn2a-rAAV-CRISPRa on neuronal excitability, acute slices from the mPFC were dissected four weeks post-injection and electrophysiological recordings were carried out from neurons in these slices. AP waveform, measured with somatic current-clamp recordings, are sensitive to Scn2a expression levels in single cells, with peak dV/dt being directly proportional to Na_V1.2 density at the somatic membrane^14,25. We found that peak dV/dt was restored to WT levels in mCherry-positive neurons in CRISPRa-treated Scn2a^+/− mice (Fig. 2d). mCherry-negative neurons, recorded either from the same slices or in interleaved constitutive controls, displayed dV/dt values expected for Scn2a^+/+ and Scn2a^+/− neurons respectively. Similarly, no upregulation in dV/dt was noted in mCherry-positive empty vector controls (Fig. 2D). Other aspects of neuronal excitability were stable (Extended Data Fig. 4).

Scn2a haploinsufficiency is also known to impair dendritic excitability and excitatory synapse function. We found that P30 Scn2a-rAAV-CRISPRa administration was able to rescue these impairments. AMPA:NMDA ratio was rescued to WT levels (Fig. 2e). Similarly, AP-evoked calcium transients in distal apical dendrites, which are readily observed in WT but not Scn2a^+/− cells, were also restored (Fig. 2f). This suggests that CRT may be effective for Scn2a to help re-engage such processes in the mPFC.

Systemic CRT in Scn2a^+/− mice

We next explored the feasibility of intravenous systemic rAAV-CRISPRa delivery, as it is a more clinically translatable delivery approach. Scn2a^+/− mice were injected, either via the tail vein or the retro-orbital sinus, with Scn2a-rAAV-CRISPRa constructs (sadCas9-VP64 and Scn2a-sgRNA) packaged using the PHP.eB rAAV serotype, which can readily pass through the blood brain barrier in C57BL/6 mice and provide robust brain transgene expression. Mice injected at P30 via the tail vein showed widespread expression throughout the brain when harvested >4 weeks later (Fig. 3a–b). To test for off-target effects, we carried out RNA-seq on nuclei from cortical neurons sorted from WT, Scn2a^+/− and Scn2a-rAAV-CRISPRa Scn2a^+/− PHP.eB rAAV serotype injected mice using three mice per condition (see Methods). We then analyzed the mRNA expression of all genes within the same TAD and all Scn sodium channel genes. We observed that Scn2a showed an upregulated trend that was not-significant after correcting for multiple testing in the Scn2a-rAAV-CRISPRa Scn2a^+/− injected mice compared to Scn2a^+/− uninjected mice and no other gene in its TAD nor sodium channel gene showed significant differential expression compared to the WT and Scn2a^+/− mice (Extended Data Fig. 5a). In contrast to our in vitro RNA-seq, we did not observe any viral response associated DE genes (Extended Data Fig. 5b–e). DE genes between Scn2a-rAAV-CRISPRa Scn2a^+/− injected and Scn2a^+/− uninjected mice included dCas9 and mCherry along with a handful of transcripts that represent genes of unknown function. In addition, we analyzed sgRNA sequence off-targets with sequences up to 3 mismatches and 2 bulges. With these criteria, we found only one putative off-target with 3 mismatches and 2 bulges, nearby a downregulated gene, Ptchd4, a negative regulator of the hedgehog pathway²⁶ (Extended Data Fig. 5f). This gene did not have a significant expression difference when comparing Scn2a-rAAV-CRISPRa Scn2a^+/− injected mice to the WT. Our inability to observe significant upregulation for Scn2a in the RNA-seq from Scn2a-rAAV-CRISPRa RNA-seq could be due to its low expression levels and manifestation in specific neuronal cell types²⁷ and the last Scn2a exon having high similarity to Scn1a, Scn3a and Scn9a (ranging from 92–96% identity). Both attributes likely ‘dilute’ its expression levels in RNA-seq despite sorting neurons from the cortex. However, we should note that we observed significant upregulation via qRT-PCR for Scn2a for numerous mice injected with Scn2a-rAAV-CRISPRa, either via local or systemic injection (Fig. 2c, Extended Data Fig. 2d) and Scn2a-rAAV-CRISPRa dual vector components were detected in the RNA-seq and through 460 days (16 months) via qRT-PCR demonstrating the long-term, persistent AAV transgene expression (Extended Data Fig. 6).

Fig. 3. Systemic CRT rescues electrophysiological deficits and is protective against chemoconvulsant-induced seizure.

a. Schematic of Scn2a-rAAV-CRISPRa-PHP.eB intravenous tail vein delivery. Viruses injected described in brackets.

b. Sagittal brain section from P80 Scn2a^+/−+CRISPRa mouse following tail vein injection of Scn2a-rAAV-CRISPRa-PHP.eB at P30, stained for mCherry and DAPI.

c. Peak AP dV/dt from P82–241 Scn2a^+/+ (black, n=44), Scn2a^+/− (cyan, n=37), and tail vein Scn2a-rAAV-CRISPRa-PHP.eB injected Scn2a^+/++CRISPRa (gray, n=23) and Scn2a^+/−+CRISPRa (purple, n=20) L5PNs. ****p<0.0001, Holm-Šídák multiple comparisons test.

d. AMPA:NMDA ratio from P79–85 Scn2a^+/+ and Scn2a^+/− uninjected controls (left) and P82–241 Scn2a-CRISPRa-PHP.eB treated Scn2a^+/+ and Scn2a^+/− mice (right). n = 7, 14, 11, 13 cells, respectively. *p<0.05. Holm-Šídák multiple comparisons test vs. Scn2a^+/−.

e. Left: Modeled distribution of Na_V1.2 and Na_V1.6 across axon initial segment (AIS), soma, and proximal apical dendrite. Middle: AP phase-plane plots with varying Na_V1.2 plasma membrane densities. Right: Predicted peak AP dV/dt as a function of Na_V1.2 expression levels.

f. Empirical measurements of peak AP dV/dt in Scn2a^+/+, Scn2a^+/−, and Scn2a^−/− L5PNs compared to Scn2a^+/+ + CRISPRa. Scn2a^+/+, Scn2a^+/−, and Scn2a^−/− values are from²⁵. Scn2a^+/++CRISPRa are merged local and tail vein peak dV/dt values. Data placed at 175% of WT based on PCR upregulation (Fig. 2c). Scn2a^+/+ vs. Scn2a^+/++CRISPRa; p=0.5, Mann-Whitney test.

g. EEG from screws placed above prefrontal cortex in naïve or CRISRPa-treated WT and Scn2a^+/− mice receiving 8 mg/kg 4-AP at onset of recording.

h. Expansion of grey bar area in g. Red marks denote automated identification of ictal-like activity.

i. Ictal spikes, binned per 5 minutes, across groups. Bars and shading are mean±SEM.

j. Events per animal at the peak of ictal event period. * p<0.03, ** p<0.004. Holm-Šídák multiple comparisons test.

Circles are single cells (c, d, f) or animals (k); box plots are medians, quartiles, and 100% tails.

Electrophysiological analysis of Scn2a-rAAV-CRISPRa-PHP.eB injected Scn2a^+/− mice showed a rescue of AP peak dV/dt and AMPA:NMDA ratio, comparable to that of local stereotactic Scn2a-rAAV-CRISPRa treatment directly into the mPFC (Fig. 3c–d). Together, these data demonstrate the feasibility of systemic administration of rAAV-CRISPRa to rescue Scn2a haploinsufficiency-associated electrophysiological defects.

Scn2a upregulation is well-tolerated

Gain-of-function missense variants in SCN2A are associated with multiple forms of epilepsy². Thus, a major concern with a CRISPRa approach to rescue SCN2A haploinsufficiency is overexpression beyond normal physiological levels, as this may result in hyperexcitability, similar to that observed with gain-of-function missense variants. To assess this, we administered CRISPRa intracranially to WT mice and assessed their cellular excitability. Intracranial Scn2a-rAAV-CRISPRa injection into mPFC upregulated Scn2a mRNA ~1.75-fold in WT mice (Fig. 2c). Compartmental models of pyramidal cells predicted that such increases in mRNA, if translated into Na_V1.2 channels that incorporate into neuronal membranes, should increase peak dV/dt of APs (Fig. 3e). Indeed, AP velocity in such models was proportional to Na_V1.2 membrane density from 0 to 200% of normal levels. In empirical studies, we observed a similar linear relationship between AP peak dV/dt and Scn2a copy number between 0 and 100% of normal levels; however, Scn2a-rAAV-CRISPRa-treated Scn2a^+/+ cells had peak dV/dt values that were no different than untreated controls (Fig. 3f). This suggests that maximal Na_V1.2 function is regulated, perhaps by auxiliary subunits, scaffolding partners, and post-translational modifications, and cannot exceed WT levels¹³, mirroring similar observations made with other approaches that increase expression of SCN1A²⁸.

CRT suppresses induced seizures

Children with SCN2A haploinsufficiency can experience seizures after the first few months of life that are often refractory to treatment². As such, there is a clear unmet need for treatment options that can reduce seizure burden in SCN2A haploinsufficiency. At the cellular level, these seizures may be due to loss of Na_V1.2 expression in neocortical pyramidal cell dendrites, as marked reductions in Scn2a expression can increase intrinsic excitability in these cells through the loss of Na_V1.2-supported activation of dendritic potassium channels that contribute to membrane repolarization during APs^25,29. Furthermore, complete loss of Na_V1.2 in layer 5 pyramidal cells is sufficient to generate seizure-like EEG activity in mice³⁰, possibly via their interactions with potassium channels. Given these observations, we tested whether Scn2a haploinsufficient mice had increased sensitivity to the chemoconvulsant 4-aminopyridine (4-AP), a nonselective blocker of delayed rectifier potassium channels that drive AP repolarization. Scn2a^+/− mice exhibited a bias towards seizure-related behavioral activity at earlier times and at lower doses of 4-AP compared to WT (Extended Data Fig. 7). Simultaneous electroencephalography (EEG) in these mice showed that 4-AP administration, at an intermediate dose (8 mg/kg), evoked ictal spikes in frontal cortex that were associated with myoclonic jerking behavior (Fig. 3g–j, Extended Data Fig. 7). These events peaked in frequency 30–40 minutes after 4-AP administration, but then subsided within an hour of dosing. At higher doses (15 mg/kg), ictal spikes often preceded tonic-clonic seizures. To determine whether these seizures could be prevented with our CRT approach, we injected PHP.eB-based Scn2a-rAAV-CRISPRa systemically via the retro-orbital sinus at P30 and challenged mice with 8 mg/kg 4-AP 42–65 days later. Remarkably, ictal spike frequency was reduced to levels observed in naïve WT animals in CRT-treated Scn2a^+/− mice (Fig. 3g–j).

Using whole-cell electrophysiology, we found that CRISPRa-based Scn2a mRNA overexpression in WT neurons does not lead to increased Na_V1.2 expression in membranes beyond normal WT levels (Fig. 3). Thus, mice should tolerate systemic Scn2a over-expression without increased seizure burden. To test this, we injected PHP.eB-based Scn2a-rAAV-CRISPRa into WT mice and assessed animal response to 4-AP as in Scn2a^+/− mice. Ictal spike rates in these mice were comparable to those in untreated WT mice, suggesting that the ceiling placed on Na_V1.2 density observed at the single cell level can help ensure that CRT-based treatments do not inadvertently increase seizure burden.

CRT in human SCN2A^+/− excitatory neurons

To further investigate the translational potential of this approach, we tested the ability of SCN2A-rAAV-CRISPRa to rescue the electrophysiological phenotypes of SCN2A^+/− neurons differentiated from human embryonic stem cells (hESCs). sgRNA constructs that target the human SCN2A promoter were designed in a similar strategy to those for mouse and screened for upregulation of SCN2A in a human neuroblastoma cell line (SH-SY5Y). One sgRNA out of the ten tested increased SCN2A mRNA by 1.5- and 1.3-fold following transient co-transfection with dCas9-VP64 or transduced via a rAAV-DJ serotype, respectively (Extended Data Fig. 8a–b). SCN2A^+/+ and SCN2A^+/− hESCs¹⁶ were differentiated into excitatory neurons using the SMADi STEMDiff forebrain differentiation and maturation protocol (Fig. 4a; see Methods). Immunostaining of cells at day in vitro (DIV) 65 with MAP2 and GFAP suggested efficient neuronal differentiation alongside supporting glial cells (Fig. 4b).

Fig. 4. SCN2A-rAAV-CRISPRa rescues spiking properties of SCN2A^+/− human embryonic stem cell-derived neurons.

a. Schematic of electrophysiological recording of CRISPRa-treated differentiated hESC-derived neurons. Cells were transduced at DIV 30, and experiments were performed at DIV 65–66.

b. Differentiated SCN2A^+/− hESC-derived neurons at DIV 65 immunostained with antibodies against the neuronal marker MAP2, the glial marker GFAP, and DAPI.

c. APs generated from a range of current amplitudes (0–340 pA, 10 pA intervals, 300 ms) in DIV 65–66 SCN2A^+/+ (black), SCN2A^+/− (cyan), and SCN2A-rAAV-CRISPRa treated SCN2A^+/− (purple) hESC-derived neurons. Arrows highlight spontaneous excitatory postsynaptic potentials (EPSPs) that occurred within recording epoch.

d. Left: DIC image of SCN2A^+/− hESC-derived neuron and recording electrode. Right: AP (spikes) per 300 ms stimulation epoch versus current injection amplitude. SCN2A^+/+ vs. SCN2A^+/−: ****p < 0.0001. SCN2A^+/− vs. SCN2A^+/− + CRISPRa ****p < 0.0001. Holm-Šídák multiple comparisons test.

e. Representative phase-plane plots and quantification of peak dV/dt in SCN2A^+/+, SCN2A^+/−, and SCN2A-rAAV CRISPRa treated SCN2A^+/− hESC-derived neurons. n = 18 SCN2A^+/+, 18 SCN2A^+/−, 13 SCN2A^+/− + CRISPRa neurons. ** p = 0.0014, ***p = 0.0004, ****p < 0.0001, Holm-Šídák multiple comparisons test.

Circles are single cells; box plots are medians, quartiles, and 100% tails.

Early in life, both human and rodent excitatory pyramidal cells rely on Na_V1.2 expression in the AIS to support AP electrogenesis¹³. During this period, which persists over the first year in of life in humans, Scn2a haploinsufficiency suppresses AP output¹⁴. As such, CRT can be evaluated in hESCs by examining both AP waveform and AP output. At DIV 65–66, SCN2A^+/+ neurons were mature enough to support repetitive AP activity (Fig. 4c). By contrast, AP electrogenesis was markedly blunted in SCN2A^+/− neurons, with few evoked APs that had very low peak dV/dt values compared to interleaved controls (Fig. 4d–e). Furthermore, the axon initial segments of these SCN2A^+/− neurons were ~19% longer than those from SCN2A^+/+ counterparts (Extended Data Fig. 9a–b), suggesting that these neurons are responding to reduced excitability by expanding this region associated with AP initiation¹³. SCN2A^+/− neurons treated with SCN2A-rAAV-DJ-CRISPRa at DIV 30 had mRNA levels that were comparable to WT levels (Extended Data Fig. 8c) and at DIV 65–66 exhibited a full recovery in AP output with a concomitant increase in peak dV/dt (Fig. 4c–e). Moreover, CRISPRa-infected SCN2A^+/− initial segments were 19% shorter than those of uninfected SCN2A^+/− neurons visualized in the same preparation, suggesting that axon initial segment structural plasticity was reversed by CRISPRa-treatment (Extended Data Fig. 9c–d). In conclusion, these data demonstrate that CRISPRa can provide a potential therapeutic benefit for SCN2A haploinsufficiency in human neuronal cell types, rescuing both electrical and structural aspects of neuronal excitability.

Discussion

Two hundred and twenty-seven high-confidence NDD-associated genes are thought to cause NDDs due to haploinsufficiency, with SCN2A haploinsufficiency being one of the most prevalent⁴. Here, we showed how a CRISPRa-based CRT approach could be used to rescue the electrophysiological deficits associated with SCN2A haploinsufficiency. First, using a Cre-induced rescue of Scn2a haploinsufficiency in adolescent mice, we showed the feasibility of this approach to provide a potential therapeutic during adolescence. Next, by direct stereotaxic mPFC and intravenous tail vein injections of Scn2a-rAAV-CRISPRa, we upregulated Scn2a in adolescent Scn2a^+/− mice and demonstrated rescue of multiple features of intrinsic excitability, including AP electrogenesis and associated activation of dendritic calcium channels, as well as synaptic efficacy, as measured by AMPA:NMDA ratio. Of note, these synaptic features were rescued with CRISPRa administration in early adolescence, suggesting that both Scn2a haploinsufficiency interrupts normal circuit refinement and that these processes can be reinitiated if normal levels of Na_V1.2 are restored. Furthermore, we found that Scn2a^+/− mice are susceptible to seizure induced by a potassium channel antagonist, and that such seizures can be prevented by CRISPRa treatment. Together, these results suggest that CRT is a viable therapeutic approach for SCN2A haploinsufficiency and could be a viable approach to mitigate treatment-refractory seizures observed in this patient population³¹.

CRT has been successful in treating rodent models of both CNS and non-CNS haploinsufficiency-related disorders^18,32–39. In the case of sodium channel haploinsufficiency, CRT also overcomes two of the main limitations of traditional gene replacement therapy, rAAV packaging capacity and ectopic expression. Sodium channels are long genes whose cDNA length exceeds the rAAV payload capacity. Classical gene replacement therapy would require splitting the channel into two components using dual AAVs and placing a linker to connect them as was recently used for Scn1a⁴⁰. CRT overcomes this limitation by packaging CRISPRa components that target the gene’s endogenous regulatory elements to upregulate its expression. In addition, results from a transgenic-based CRISPRa suggest that upregulation occurs only in the tissue/cell type where the targeted regulatory element is active¹⁸, providing additional specificity to the rAAV serotype or promoter used to drive expression. In particular, this could be extremely beneficial for sodium channels that have similar protein structures yet have distinct functions in skeletal, cardiac, central, and peripheral nervous systems, which makes it difficult to target specific isoforms pharmacologically without off-target effects^41,42. An additional advantage is that CRT uses a nuclease-deficient DNA targeting molecule that does not edit the DNA, and thus off-target effects will not lead to ‘DNA scars’. The use of a CRISPR-based CRT approach could lead to immunogenicity due to the dCas9. However, there are many efforts to engineer Cas proteins with reduced immunogenicity by epitope masking^43,44, or use of alternate Cas proteins, such as those from non-pathogenic bacteria that have not been exposed to humans¹⁰. Alternatively, other DNA targeting molecules like zinc fingers or Transcription Activator-Like Effector Nucleases (TALENS), that should be less immunogenic and smaller in size⁴⁵, could be used instead.

A major goal of CRT is to achieve physiological levels of the modulated gene that provide therapeutic benefit. Obtaining levels that are too high could also have deleterious physiological impacts whereas levels too low may not be within the therapeutic window of efficacy. For SCN2A, this is of particular importance as gain-of-function variants are associated with epileptic encephalopathies². Here, we intentionally used VP64 as the transcriptional activator to avoid excess Na_V1.2 production, as it has modest upregulating potential. Furthermore, we assessed whether Scn2a over-expression via CRISPRa above normal physiological conditions affected brain function by upregulating Scn2a mRNA in WT mice. Heightened levels of Scn2a mRNA did not correspond to elevated intrinsic or network neuronal hyperexcitability. We speculate that this could be due to a potential ceiling on total Na_V1.2 membrane density, perhaps imposed by ancillary subunits, scaffolding partners, or posttranslational modifications¹³, whose regulation remains unchanged. Furthermore, we observed no electrophysiological effects or gene expression changes that would indicate that CRT altered the expression of other CNS sodium channel genes or genes that are within the Scn2a TAD region (Extended Data Fig. 3a–b).

Treatments for NDDs would likely be most beneficial when administered early, ideally before symptom onset. Nevertheless, brain development is a dynamic process that spans multiple decades of life, and interventions administered even later in life could have some therapeutic benefit⁴⁶. One example is an antisense oligonucleotide therapy for Angelman Syndrome that improved communication and motor skills and reduced epileptiform discharges when administered to 4–17-year-old children⁴⁷, despite work in rodent models suggesting that therapies should ideally be administered prenatally^48,49. Scn2a haploinsufficiency in mice results in life-long, cell-autonomous impairments in neocortical pyramidal cell dendritic excitability and synapse function¹⁴. Here, we find that these impairments can be rescued with adolescent reactivation of a loss-of-function allele (Fig. 1) or by CRT-based over-expression via the residual, functional allele (Fig. 2–3), suggesting that rescue of normal dendritic excitability can restore many aspects of neuronal function. Future work will helps clarify other Scn2a-dependent processes and the critical therapeutic periods needed to address various treatment outcomes across development.

Here, we focused on seizure susceptibility as a primary behavioral outcome measure for CRT. Seizures are a major health concern within the SCN2A community, affecting an estimated 20–30% of children with loss-of-function variants, and, by definition, all gain-of-function cases⁵⁰. We found that 4-AP application resulted in a robust, sex-independent increase in ictal spiking in PFC EEG records that allowed for CRT evaluation in Scn2a^+/− and safety in WT littermates (Fig. 3g–j). CRT treatment did not lead to any change in seizure burden in 4-AP-treated mice. This is critical, as it suggests that overexpression of Scn2a genes can be regulated, avoiding overexpression of Na_V1.2 function at neuronal membranes. This places an intrinsic safety factor for Scn2a CRT implementation.

In the future, it will be critical to evaluate CRT approaches further with other behaviors that better reflect NDD aspects of Scn2a haploinsufficiency. Unfortunately, the field has struggled to identify robust phenotypes in Scn2a^+/− mice in commonly assessed behaviors^14,51–55. Leveraging other model systems will likely be critical for assessing proper therapeutic developmental windows and safety profiles. Furthermore, several aspects of this approach will need to be optimized, including safety assessments of rAAV components, off-target CRISPRa effects, the use of non-human primates (NHPs) to assess efficient delivery, upregulation levels, cytotoxicity, immunogenicity and AAV toxicity and serotypes^56,57. For the latter point, the PHP.eB serotype used here is only effective in specific mouse strains⁵⁸, but recent serotypes that pass the blood-brain barrier in NHPs⁵⁹ or use of low-intensity focused ultrasound⁶⁰ to increase infiltration could potentially be used. In addition, it will be vital to confirm that the SCN2A variant to be treated causes a non-functional transcript, as this current approach upregulates both alleles. Lastly, it is important to note that this therapeutic approach is useful only for variants that lead to a complete loss-of-function allele. Approaches that utilize allele-specific targeting could potentially be used to overcome this hurdle in cases where the variant allele does not promote gain-of-function.

In summary, this study shows how CRT using CRISPRa can rescue phenotypes in a genetic mouse model of a prevalent cause of NDDs. CRT is a customizable platform technology that can modify gene expression without directly editing the genome and can be tailored to rescue other disorders caused by haploinsufficiency. The application of CRT to treat SCN2A haploinsufficiency leverages a growing understanding of the genetic etiology of NDDs and demonstrates a potential path forward for treating complex neurodevelopmental conditions through modern genetic medicine approaches.

Methods

CRISPRa in vitro optimization

Ten sgRNAs targeting the mouse Scn2a or human SCN2A promoters were designed using the Broad Institute’s GPP sgRNA Design Tool (Genetic Perturbation Platform, Broad Institute). These guides were individually cloned into pAAV-U6-sasgRNA-CMV-mCherry-WPREpA at the BstXI and XhoI restriction enzyme sites using the In-Fusion HD cloning kit (Clontech). rAAV vectors were generated using similar plasmids and cloning methods as was referenced in¹. Mouse sgRNAs were tested in the neuroblastoma cell line Neuro-2a (ATCC CCL-131). Cells were grown in Dulbecco’s Modified Eagle’s Medium (DMEM) with 10% FBS and 1% Penicillin-Streptomycin, and SH-SY5Y cells were grown in Eagle’s Minimum Essential Medium (EMEM) with 10% FBS and 1% Penicillin-Streptomycin following ATCC guidelines. Cells were transiently co-transfected with individual sgRNA cloned into pAAV-U6-sasgRNA- CMV-mCherry-WPREpA along with pCMV-sadCas9-VP64 for 48 hours using Opti-MEM Reduced Serum Medium (Thermo Fisher) and X-tremeGENE HP (Sigma-Aldrich). RNA was isolated using the RNeasy Mini Kit (Qiagen) following the manufacturer’s protocol. cDNA was synthesized using SuperScript III First-Strand Synthesis System (Invitrogen) and qPCR was conducted using SsoFast EvaGreen Supermix (Bio-Rad) and analyzed using the ΔΔCT methods comparing to a no-sgRNA transfection and normalized to Actb as a housekeeping gene. rAAVs were produced at the Stanford Gene Vector and Viral Core (see Supplemental Table 3 for genomic titers).

RNA-seq

For cell culture RNA-seq, we used two biological replicates and three technical replicates for each condition (six samples per condition) from Neuro-2a treated either with rAAV-sadCas9-VP64 (negative control) or rAAV-sadCas9-VP64 + rAAV-Scn2a-sgRNA. For mouse RNA-seq, cortices were dissected from wild-type and heterozygous mice, with or without rAAV-sadCas9-VP64 + rAAV-Scn2a-sgRNA injection, three mice per condition. For each sample, nuclei were isolated from approximately 200 mg of cortex using the Singulator 100 System (S2 Genomics) with the “Single Shot Standard Nuclei Isolation v2” program. The isolation was performed in 3 mL of lysis buffer (10 mM Tris-HCl, pH 7.4; 10 mM NaCl; 3 mM MgCl₂·6H₂O; 0.05% IGEPAL CA-630; 0.5 mM DTT; nuclease-free water) and 3 mL of wash buffer (5% bovine serum albumin, 0.25% glycerol in 1× PBS). To preserve RNA integrity, 50 μL of Protector RNase Inhibitor (Sigma-Aldrich, #3335402001) was added directly to the Singulator cartridge. Following isolation, nuclei were centrifuged, and debris were removed by centrifugation in wash buffer supplemented with 1× Nuclei Debris Removal Stock Reagent (S2 genomics #100253628) and 37.5 μL of RNase inhibitor at 600 × g for 15 min at 4°C, with deceleration set to 1. Purified nuclei were then washed once in wash buffer and resuspended in 500 μL of wash buffer supplemented with 6.5 μL of Protector RNase Inhibitor, DAPI (Biotium, #40043; 1:1000), and anti-NeuN-488 antibody (Sigma-Aldrich, #MAB377X; 1:500). The suspension was incubated for 1 hour at 4°C in the dark to allow staining. Finally, nuclei were washed once with wash buffer, resuspended in 1 mL of wash buffer supplemented with 12.5 μL of Protector RNase Inhibitor, and transferred to a FACS tube. Neuronal nuclei were sorted based on DAPI and NeuN positivity using a BD FACSAria Fusion (BD Biosciences) (Extended Data Figure 5g–i). Sorted nuclei were then resuspended in 700 μL of QIAzol lysis reagent (Qiagen, #79306), and total RNA was extracted using the miRNeasy kit (Qiagen, #217084) following the manufacturer’s instructions. Extracted RNA was sent to Novogene for mRNA library preparation (polyA enrichment) and sequenced using a NovaSeq 6000 generating 150 base pair end reads.

RNA sequencing reads were trimmed on the 3’ ends of reads where the quality score was less than 20 using trim galore 0.6.10 through nf-core RNA-seq pipeline version 3.18⁶¹. Briefly, sequences were mapped to mm39 with STAR 2.7.8a⁶² and counts were computed using featureCounts with gtf provided from Ensembl (Mus_musculus.GRCm39.113). Post-alignment, gene annotation was carried out using Ensembl Transcripts release 113. Genes exhibiting a maximum count of one or less were removed. The remaining gene counts were normalized to counts per million (CPM). Differential expression was characterized using DESeq2 version 1.44 (2024-05-01)⁶² and R version 4.4. Genes with an average normalized count of less than one across all samples were considered not significantly different. Gene Ontology (GO) over-representation analysis using GSEApy version 1.1.3⁶³ was performed on significantly upregulated genes. Upregulated genes were tested for GO term enrichment against a background of genes expressed in the RNA-seq samples using Fisher’s exact test for statistical significance. Benjamini-Hochberg Multiple testing correction was performed and GO terms with an adjusted p-value less than 0.05 were considered significantly enriched. All RNA-seq data is available on the NCBI Gene Expression Omnibus as Bioproject PRJNA1292778.

Mouse generation, husbandry, and genotyping

All mouse work was approved by the UCSF Institutional Animal Care and Use Committee. All mice were maintained on a C57BL/6J background on a 12:12 light-dark cycle (light on at 6 AM) with ad libitum food and water, with daily health monitoring. Genotyping was performed via tail clips using the KAPA Mouse Genotyping Kit (Roche #07961766001). See Supplemental Table 2 for primer sequences. Scn2a^+/− were as described in⁶⁴, conditional knockout (Scn2a^+/fl) were as described in¹⁴. Scn2a^+/KI mice were created at Cyagen Biosciences by inverting exons 3–5 of the Scn2a gene and inserting a SA-2A-eGFP-polyA cassette upstream of the inverted exons. The inverted exons and the SA-2A-eGFP-polyA cassette were flanked with loxP and Lox2272 sites to ensure the excision of the SA-2A-eGFP-polyA cassette and the reversion of exons 3–5 to the correct orientation following Cre recombinase induced recombination. A targeting vector containing the inverted exons, SA-2A-eGFP-polyA cassette, recombination sites, selection markers, and homology arms was assembled from mouse genomic fragments amplified from BAC clone RP23–332C13 and RP23–55C23. The targeting vector was linearized by restriction enzyme digestion and transfected into C57BL/6 embryonic stem cells via electroporation, and successfully transfected cells were identified by drug selection, PCR verification, and Southern blot confirmation. Confirmed clones were introduced into host embryos and transferred to surrogate mothers. Chimerism in the resulting pups was identified via coat color. F0 male chimeras were bred with C57BL/6J females to generate F1 heterozygous mutants that were identified by PCR.

In vivo AAV administration

For local AAV administration, mice P30-P40 of age were kept under live anesthetic isoflurane at 0.5–2.0% and mounted onto the stereotaxic machine (Kopf 1900). 500 nL of rAAV at a 1:1 ratio of sgRNA and sadCas9-VP64 or 1 μL of EF1α-Cre-mCerry was injected into the mPFC at stereotaxic coordinates [mm]: anterior-posterior [AP], +1.7, mediolateral [ML] −0.35; dorsoventral [DV]: −2.6 at a viral infusion rate of 0.1 μl min−1. For systemic AAV administration, mice age P30-P40 were kept on a 37° C warm pad and harnessed using a brass mouse restrainer (SAI Infusion Tech). Lateral tail vein injections were carried out with rAAV at a 1:1 ratio of sgRNA and sadCas9-VP64 (1×10^11 vg/mouse) suspended in 200 uL saline using a 30G needle. For either injection, mice were used for subsequent experiments four weeks post-injection. Tail vein injected animals used for EEG recording experiments, were subsequently divided across experiments for electrophysiological recordings, qPCR, and immunofluorescence.

In vivo dissections

Four weeks post-injection, mouse brains were removed and 250 μm thick coronal slices containing the mPFC were dissected in artificial cerebrospinal solution containing (in mM): 87 NaCl, 25 NaHCO₃, 25 glucose, 75 sucrose, 2.5 KCl, 1.25 NaH₂PO₄, 0.5 CaCl₂ and 7 MgCl₂; bubbled with 5%CO₂/95%O₂; 4°C. Fluorescence mCherry expression in the mPFC of the injected hemisphere was validated using a fluorescent stereomicroscope (Nikon SMZ1500). For mRNA expression analyses, both the injected (mCherry-positive) and uninjected hemisphere (mCherry-negative) were dissected using a sterile miltex disposable punch biopsy (Medline MIL3332P25) and flash frozen in RLT lysis buffer (Qiagen). RNA was extracted using the RNAEasy Micro Kit (Qiagen) following the manufacturer’s protocol. cDNA and downstream qPCR were conducted using SuperScript III (Invitrogen) and SsoFast EvaGreen Supermix (Bio-Rad) on a QuantStudio 6 Flex Real Time PCR system (Applied Biosystems). qPCR results comparing the injected and uninjected hemispheres were analyzed using the ΔΔCT methods and normalized to Actb, a housekeeping gene.

Ex vivo electrophysiology and two-photon imaging

All ex vivo electrophysiology and two-photon imaging were acquired and performed using the same methods as described previously^14,25. Mice were anesthetized using isoflurane and 250 μm coronal slices were prepared. Cutting solution contained (in mM): 87 NaCl, 25 NaHCO₃, 25 glucose, 75 sucrose, 2.5 KCl, 1.25 NaH₂PO₄, 0.5 CaCl₂ and 7 MgCl₂; bubbled with 5%CO₂/95%O₂; 4°C. Following cutting, slices were either incubated in the same solution or in the recording solution for 30 min at 33 °C, then at room temperature until recording. Recording solution contained (in mM): 125 NaCl, 2.5 KCl, 2 CaCl₂, 1 MgCl₂, 25 NaHCO₃, 1.25 NaH₂PO₄, 25 glucose; bubbled with 5%CO₂/95%O₂; 32–34°C, ~310 mOsm. Neurons were visualized with differential interference contrast (DIC) optics for conventional visually guided whole-cell recording, Dodt contrast imaging, or with 2-photon-guided imaging of reporter-driven mCherry or GFP fluorescence overlaid on an image of the slice (scanning DIC). Cells of interest were chosen based on fluorescence expression in all experiments in Scn2a^+/KI mice and in all experiments utilizing CRISPRa constructs. For current-clamp recordings, patch electrodes (Schott 8250 glass, 3–4 MΩ tip resistance) were filled with a solution containing (in mM): 113 K-Gluconate, 9 HEPES, 4.5 MgCl₂, 0.1 EGTA, 14 Tris₂-phosphocreatine, 4 Na₂-ATP, 0.3 tris-GTP; ~290 mOsm, pH: 7.2–7.25. For Ca²⁺ imaging, EGTA was replaced with 250 μM Fluo-5F and 20 μM Alexa 594. For voltage-clamp recordings of and synaptic activity, internal solution contained (in mM): 110 CsMeSO₃, 40 HEPES, 1 KCl, 4 NaCl, 4 Mg-ATP, 10 Na-phosphocreatine, 0.4 Na₂-GTP, 0.1 EGTA; ~290 mOsm, pH: 7.22. All data were corrected for measured junction potentials of 12 and 11 mV in K- and Cs-based internals, respectively.

Electrophysiological data were acquired using Multiclamp 700A or 700B amplifiers (Molecular Devices) via custom routines in IgorPro (Wavemetrics). For measurements of action potential waveform, data were acquired at 50 kHz and filtered at 20 kHz. For all other measurements, data were acquired at 10–20 kHz and filtered at 3–10 kHz. For current-clamp recordings, pipette capacitance was compensated by 50% of the fast capacitance measured under gigaohm seal conditions in voltage-clamp prior to establishing a whole-cell configuration, and the bridge was balanced. For voltage-clamp recordings, pipette capacitance was compensated completely, and series resistance was compensated 50%. Series resistance was <15 MΩ in all recordings. Experiments were omitted if input resistance changed by > ±15%.

Two-photon laser scanning microscopy (2PLSM) was performed as previously described¹⁴. A two-photon source (Coherent Ultra II) was tuned to 810 nm for morphology and calcium imaging. Epi- and transfluorescence signals were captured either through a 40×, 0.8 NA objective for calcium imaging or a 60×, 1.0 NA objective for spine morphology imaging, paired with a 1.4 NA oil immersion condenser (Olympus). Fluorescence was split into red and green channels using dichroic mirrors and band-pass filters (575 DCXR, ET525/70m-2p, ET620/60m-2p, Chroma). Green fluorescence (Fluo-5F) was captured with 10770–40 photomultiplier tubes selected for high quantum efficiency and low dark counts (PMTs, Hamamatsu). Red fluorescence (Alexa 594) was captured with R9110 PMTs. Data were collected in linescan mode (2–2.4 ms/line, including mirror flyback). For calcium imaging, data were presented as averages of 10–20 events per site and expressed as Δ(G/R)/(G/R)max*100, where (G/R)max was the maximal fluorescence in saturating Ca²⁺ (2 mM) (Yasuda et al., 2004). Action potential backpropagation experiments were performed in 25 μM picrotoxin, 10 μM NBQX and 10 μM R-CPP.

EEG implant surgeries

Mice were anesthetized with isoflurane and placed on a stereotaxic apparatus. Screws with wire leads were implanted in 2–3 turns into five burr holes at (stereotaxic coordinates relative to bregma [mm]) PFC: 1.7 anterior-posterior (AP), −0.3 mediolateral (ML); S1: −1.8 AP, 2.5 ML; reference and ground: −5 AP, 0.9 ML (Pinnacle Technology). A head mount was then attached to the wire leads and secured in adhesive dental cement. Mice were given at least one week for post-surgery recovery prior to EEG recordings.

EEG recordings and analysis

EEG recordings were performed using the Sirenia Acquisition Software (Pinnacle Technology) at 2 kHz sampling frequency with simultaneous video capture. 4-aminopyridine (4-AP) doses at 6 and 8 mg/kg were delivered in sessions spaced at least 2 days apart. 15 mg/kg was administered at least 1 week thereafter. 4-AP was prepared fresh at the start of every recording day. Recordings were performed between 10 AM and 6 PM to control for the circadian light-dark cycle. EEG analyses were performed with custom routines in IgorPro based on deconvolution-based event detection for spontaneous synaptic activity as described in⁶. A series of ictal spike events from EEG screws placed above frontal cortex were averaged to create a 100-ms template event for all subsequent analysis. Ictal spikes were identified as all events with a deconvolved signal 3.5x above background noise.

Behavioral seizure scoring was made from simultaneous video and EEG monitoring. Behavioral classification of increasing seizure severity was based on the Racine scale and a previous report of seizures in Scn2a^+/− mice^30,65. Latency to seizure activity was considered the time from 4-AP administration to the onset of the first seizure. Behavioral arrest was defined as a sudden immobilization of the animal. A myoclonic jerk was defined as a neck or body twitch accompanied by a single sharp spike in the EEG signal. A tonic-clonic seizure was defined as clonic convulsions leading to wild jumping also with high amplitude spiking in the EEG signal. The animal was considered deceased when the tonic-clonic seizure led to tonic extension of hind limbs followed by loss of body tone and close to flat EEG signal.

Computational compartmental modeling

A pyramidal cell compartmental model was implemented in the NEURON environment (v7.7) based on the Blue Brain Project thick-tufted layer 5b pyramidal cell (TTPC1) model used in our previous study^25,66,67. The TTPC1 model was adjusted to include an AIS, and the original Na channels in the TTPC1 model were replaced with Na_V1.2 and Na_V1.6 channels in compartments with densities as described previously²⁵.

hESC differentiation, maturation, and electrophysiology

Wildtype and SCN2A^+/− HUES66 (NIH Registration #0057) hESC cell lines were obtained from the Harvard Stem Cell Institute and plated on Matrigel (Corning) coated standard tissue culture plates maintained in mTESR (STEMcell technologies). NPCs hESCs were differentiated following the manufacturer’s instructions using the STEMDiff SMADi Embryoid Body Neural Induction protocol (STEMcell Tech) following a similar strategy documented in⁶⁸. Neural progenitor cellsNPCs were further differentiated into neuronal forebrain precursors using the STEMdiff ForeBrain Neuron Differentiation protocol (Document #10000005464, Stemcell Technologies, Vancouver, CanadaSTEMcell Tech), and neuronal precursors were matured into forebrain neurons using the STEMdiff Forebrain Maturation Kit with BrainPhys (STEMCell Tech). Neuronal precursors were maturated on Poly-L-Ornithine Laminin coated German glass coverslips (Neuvitro GG-25–1.5-Laminin) for 65 days. Whole-cell current-clamp recordings were done as in mouse acute slices using identical solutions.

Immunofluorescence of fixed cells or tissue

For hESCs, differentiated neurons were fixed at DIV 65 with 4% (paraformaldehyde) PFA and 4% sucrose and blocked with 10% Normal Goat Serum in phosphate buffered saline (PBS) in 0.2% Triton-X (PBST). All coverslips within a set of experiments, including those used previously that day for electrophysiology, were fixed within 1 hr of each other to minimize developmental differences across coverslips. Following fixation, cells were incubated with primary antibodies against MAP2 (Invitrogen PA1–10005) at 1:5000 and Ankyrin-G (Neuromab 75–146) at 1:1000 overnight at 4 degrees. Cells were then incubated with the secondary antibodies Goat anti-Chicken IgY (H+L) Cross-Absorbed Secondary Antibody, Alexa Fluor Plus 647 (Invitrogen A32933) at 1:500 and Goat anti-Mouse IgG (H+L) Highly Cross Adsorbed Secondary Antibody, Alexa Fluor 488 (Thermo Fisher A-11029) at 1:500. They were then mounted on coverslips with ProLong^™ Diamond Antifade Mountant with DAPI (Thermo Fisher). Images were captured with a 40× 1.4NA objective using an Olympus Fluoview FV3000 confocal microscope. AIS length was measured from maximum intensity projections generated from z-stacks that contained the entire bounds of the neuron based on ankyrin-G and MAP2 immunofluorescence. Only initial segments with clearly detectable start and end points were quantified using the segmented line tool in FIJI (ImageJ). Start and end points were determined with a threshold of 50% peak ankyrin-G fluorescence intensity using plot profiles (pixel intensity/distance) in FIJI.

For whole-brain mouse immunohistochemistry, mice were perfused with 4% PFA in PBS and brains were removed and sectioned coronally or sagittally on vibratome (Leica VT1000) at 50–75 μm and mounted on slides with ProLong^™ Gold Antifade Mountant with DAPI. For imaging of Scn2a^+/KI sections, tissue was permeabilized, blocked, and immunostained for GFP (anti-GFP AlexaFluor 488, 1:500, ThermoFisher A-21311) and parvalbumin (PV27, 1:500, Swant; AlexaFluor 564 secondary, 1:500, ThermoFisher). Sections were imaged either with an Olympus Fluoview FV3000 (coronal sections) or Keyence BZ-X widefield microscope (sagittal sections).

Quantification and statistical analysis

Unless otherwise noted, all data are shown in figures as box plots with median and quartiles and min. and max. tails with individual data points overlaid, or mean ± standard error with individual data points overlaid. A Kruskall Wallis nonparametric test, with Holm-Šídák multiple comparisons testing was performed unless otherwise noted. Statistical tests are noted in figure legends. All tests used were two-sided. Results were considered significant at alpha value of p < 0.05. For multiple comparisons tests used to compare several conditions, no indication of significance in the figure means not significant. Statistical analysis was performed using Prism 9 (Graphpad software). For electrophysiology and imaging experiments, acute slices were typically generated blind to genotype and experiments were interleaved between two or three genotypes/injection conditions to control for recording conditions. Group sample sizes were chosen based on standards in the field and previous similar experiments conducted by our group^{14, 25}. No statistical methods were used to predetermine sample size. A minimum of 2 biological replicates were tested in all experiments.

Extended Data

Extended Data Figure 1: Excitatory pyramidal neurons in the mPFC are GFP+ in Cre-negative Scn2a^+/KI animals.

a. Coronal brain sections from P60 Scn2a^+/KI mouse (Cre−) immunostained with anti-GFP and anti-parvalbumin (PV).

b. Zoom of area highlighted by dashed box in panel a, with GFP and Parvalbumin channels separated then merged at bottom. Parv+ somata are circled in the GFP panel.

c. Further zoom of region in layer 5b in panel b. Parv+ somata are circled as in panel c.

d. Quantification of mean fluorescence intensify of GFP in PV-negative cells, PV-positive cells, and neuropil (area without somata as a measure of background fluorescence). Data are from 2 mice. Circles are mean GFP intensity values in ROIs of Parv− somata, Parv+ somata, or neighboring neuropil in single optical sections. Box plots are medians, quartiles, and 100% tails. n = 67 Parv− and 37 Parv+ cells analyzed; Parv− vs. Parv+: ****p < 0.0001. Parv− vs. neuropil: ****p < 0.0001. Holm-Šídák multiple comparisons test.

Extended Data Figure 2: In vitro optimization of CRISPRa constructs in mouse Neuroblastoma-2A cells.

a. Fold change of Scn2a expression in Neuro-2a cells transfected with plasmids containing sgRNAs targeting the promoter of mouse Scn2a compared to a no-sgRNA VP64 control. Blue bars represent plasmids with largest increase in Scn2a expression. Circles are replicates, overlaid on mean ± SD.

b. Fold change of Scn2a transduced with rAAV-DJ virus in Neuro-2a cells. Circles are replicates, overlaid on mean ± SD.

c. Total animal weight at time of 8 mg/kg 4-AP administration (animals aged 69–119 days). Circles are animals. Box plots are medians, quartiles and 100% tails.

d. qPCR analysis of dCas9 and mCherry mRNA within the mPFC of tail vein injected Scn2a^+/+ + CRISPRa (light gray) and Scn2a^+/− + CRISPRa (purple) versus uninjected controls Scn2a^+/+ (dark gray) or Scn2a^+/− (cyan). Injected animals with at least a 10-fold increase in expression levels of both dCas9 and mCherry to the average Scn2a^+/+ uninjected controls were included in EEG datasets in Fig 3.

Extended Data Figure 3: Scn2a-CRISPRa off-target analysis in in vitro cell lines.

a. RNA-seq expression levels of Scn2a TAD-domain genes from Scn2a-rAAV-CRISPRa treated Neuro-2a cells compared to VP64-only. Circles are individual replicates, box plots are medians, quartiles and 100% tails. p-values noted above data. Wald-log test.

b. RNA-seq expression levels of sodium channels from Scn2a-rAAV-CRISPRa treated Neuro-2a cells compared to VP64-only. Circles are individual replicates, box plots are medians, quartiles and 100% tails. p-values noted above data. Wald-log test.

c. Volcano plot representing Log₂ fold-change in expression levels for each gene. Wald-log test. P-values noted above each comparison. Grey dots represent no significant DEGs for CRISPRa, purple dots signify Scn2a and nearby genes, and orange dots denote upregulated genes.

d. Gene ontology (GO) analysis showing enrichment of molecular functions.

e. Gene ontology (GO) analysis showing enrichment of biological processes.

f. Analysis of sgRNA sequence off-targeting using Cas-OFFinder.

Extended Data Figure 4: Intrinsic electrophysiology of Scn2a^+/KI and CRISPRa treated neurons.

a. AP threshold from P57–85 Scn2a-rAAV-empty vector Scn2a^+/+ (light gray) and Scn2a^+/− (magenta) neurons, Scn2a-rAAV-CRISPRa treated Scn2a^+/+ (dark gray) and Scn2a^+/− (purple) neurons, and Scn2a^+/KI Cre− (green) and Scn2a^+/KI Cre+ (gray) neurons. Threshold of the first AP evoked by a near-rheobase current. Scn2a^+/+ + empty: n = 17 cells; Scn2a^+/− + empty: n = 27 cells; Scn2a^+/+ + CRISPRa: n = 24 cells; Scn2a^+/− + CRISPRa: n = 19 cells; Scn2a^+/KI Cre−: n = 29 cells; Scn2a^+/KI Cre+: n = 21 cells. No significant differences. Holm-Šídák multiple comparisons test. Circles are individual cells, box plots are medians, quartiles and 100% tails.

b. Input resistance (MΩ). Scn2a^+/+ + empty: n = 17 cells; Scn2a^+/− + empty: n = 27 cells; Scn2a^+/+ + CRISPRa: n = 17 cells; Scn2a^+/− + CRISPRa: n = 11 cells; Scn2a^+/KI Cre−: n = 26 cells; Scn2a^+/KI Cre+: n = 19 cells. No significant differences. Holm-Šídák multiple comparisons test. Circles are individual cells, box plots are medians, quartiles and 100% tails.

c. Rheobase current (pA) to generate first spike. Scn2a^+/+ + empty: n = 16 cells; Scn2a^+/− + empty: n = 27 cells; Scn2a^+/+ + CRISPRa: n = 17 cells; Scn2a^+/− + CRISPRa: n = 12 cells; Scn2a^+/KI Cre−: n = 25 cells; Scn2a^+/KI Cre+: n = 19 cells. No significant differences. Holm-Šídák multiple comparisons test. Circles are individual cells, box plots are medians, quartiles and 100% tails.

d. APs per 300 ms stimulation epoch for each current amplitude. Left: Quantification of firing rate slope of data on Right. Scn2a^+/+ + empty: n = 16 cells; Scn2a^+/− + empty: n = 27 cells; Scn2a^+/+ + CRISPRa: n = 17 cells; Scn2a^+/− + CRISPRa: n = 12 cells; Scn2a^+/KI Cre−: n = 25 cells; Scn2a^+/KI Cre+: n = 19 cells. No significant differences. Holm-Šídák multiple comparisons test. Circles are individual cells, box plots are medians, quartiles and 100% tails.Right: Number of APs versus current amplitude injected. Circles and bars are mean ± SEM at each stimulus intensity.

Extended Data Figure 5: Scn2a-CRISPRa off-target analysis in in vivo mouse neocortex.

a. RNA-seq expression levels of Scn2a TAD-domain genes from Scn2a-rAAV-CRISPRa treated Neuro-2a cells compared to VP64-only. Circles are individual mice, box plots are medians, quartiles and 100% tails.

b. RNA-seq expression levels of sodium channels from Scn2a-rAAV-CRISPRa treated Neuro-2a cells compared to VP64-only. Circles are individual mice, box plots are medians, quartiles and 100% tails.

c. Volcano plot representing Log₂ fold-change in expression levels for each gene for WT vs Scn2a^+/− “Het”. Wald-log test. P-values noted above each comparison. Grey dots represent no significant DEGs for CRISPRa, purple dots signify Scn2a and nearby genes, and orange dots denote upregulated genes.

d. Same as c, but for CRISPRa-treated Scn2a^+/− “CRISPRa” vs WT.

e. Same as c, but for CRISPRa-treated Scn2a^+/− “CRISPRa” vs Scn2a^+/− “Het”.

f. Analysis of sgRNA sequence off-targeting using Cas-OFFinder.

g. Data representation of NeuN-positive neuronal nuclei isolated from cortical tissue for FACS sorting. Representative FACS plots visualized with FlowJo V10 with percentage of parent gates for each population. Singlets were selected based on FSC-H versus FSC-A

h. From singlets, DAPI-positive events were gated to identify nuclei.

i. NeuN-positive neuronal nuclei were selected using an anti-NeuN antibody conjugated to Alexa Fluor 488.

Extended Data Figure 6: CRISPRa expression persists through 16 months post-systemic injection.

Peak AP dV/dt (top) at 30-, 56-, 107-, and 210-days following tail-vein injection of Scn2a-rAAV-CRISPRa-PHP.eB in Scn2a^+/−mice (purple). Circles represent individual neurons. Lines are linear regression. qPCR of mCherry (middle) or dCas9 (bottom) mRNA from neocortical samples collected from tail-vein or retro-orbital Scn2a-rAAV-CRISPRa-PHP.eB injected Scn2a^+/+ (gray) or Scn2a^+/− mice (purple) across time. Circles represent single mice. Lines are linear regression.

Extended Data Figure 7: Behavioral and EEG response to increasing doses of 4-AP in Scn2a^+/− mice.

a. Example PFC EEGs from animals receiving 8 and 15 mg/kg 4-AP. 4-AP administered at onset of all recordings. Dashed box denotes a tonic clonic seizure and subsequent mortality occurring with 15 mg/kg dosing in Scn2a^+/− animal. EEG data within boxed area is shown at higher resolution on right.

b. Survival curves for all WT and Scn2a^+/− animals for 6, 8 and 15 mg/kg 4-AP. WT: black, Scn2a^+/−: cyan.

Extended Data Figure 8: In vitro optimization of CRISPRa constructs in human SH-SY5Ycells.

a. Fold change of Scn2a expression in SH-SY5Y cells transfected with plasmids containing sgRNAs targeting the promoter of human Scn2a compared to a no sgRNA VP64 control.

b. Fold change of Scn2a transduced with rAAV-DJ virus in human SH-SY5Y cells.

c. SCN2A mRNA expression from SCN2A^+/+ (black), SCN2A^+/− (cyan), and SCN2A-rAAV-CRISPRa treated SCN2A^+/− (purple) hESC-derived neurons normalized to wild type average. SCN2A^+/+ vs. SCN2A^+/−: *p = 0.03. Holm-Šídák multiple comparisons test.

n = 3 replicates for all conditions. circles are individual replicates. Bars are mean ± SD.

Extended Data Figure 9: Axon initial segment structural plasticity occurs in SCN2A^+/− neurons and is rescued by CRISPRa.

a. Representative images of SCN2A^+/+ (black) and SCN2A^+/− (cyan) human stem-cell-derived neurons immunostained with antibodies against ankyrin-G (green) and MAP2 (magenta). Arrowheads denote start and end points used to quantify AIS length.

b. Quantification of AIS length. SCN2A^+/+: n = 56 cells, 3 dishes. SCN2A^+/−: n = 56 cells, 3 dishes. ****p < 0.0001. Mann-Whitney test.

c. Representative images of SCN2A^+/− neurons expressing Scn2a-rAAV-CRISPRa-mCherry (purple) and mCherry-negative internal SCN2A^+/− controls (cyan). Immunostaining against ankyrin-G and MAP2.

d. Quantification of AIS length. SCN2A^+/−: n = 122 cells, 3 dishes. SCN2A^+/− + CRISPRa: n = 23 cells, 3 dishes. ****p < 0.0001. Mann-Whitney test.

Circles are individual AIS calculations. Box plots are medians, quartiles, and 100% tails.

Data availability

All RNA-seq data is available on the NCBI Gene Expression Omnibus as Bioproject GSE193605. Source data for all statistical analyses are provided with this article. Requests for materials can be directed to NA (CRT-based approaches) or KJB (mouse models).

Code availability

No custom code was generated for this work.