Effective knockdown-replace gene therapy in a novel mouse model of DNM1 developmental and epileptic encephalopathy

Abstract

Effective gene therapy for gain-of-function or dominant-negative disease mutations may require eliminating expression of the mutant copy together with wild-type replacement. We evaluated such a knockdown-replace strategy in a mouse model of DNM1 disease, a debilitating and intractable neurodevelopmental epilepsy. To challenge the approach robustly, we expressed a patient-based variant in GABAergic neurons—which resulted in growth delay and lethal seizures evident by postnatal week three—and delivered to newborn pups an AAV9-based vector encoding a ubiquitously expressed, Dnm1-specific interfering RNA (RNAi) bivalently in tail-to-tail configuration with a neuron-specific, RNAi-resistant, codon-optimized Dnm1 cDNA. Pups receiving RNAi or cDNA alone fared no better than untreated pups, whereas the vast majority of mutants receiving modest doses survived with almost full growth recovery. Synaptic recordings of cortical neurons derived from treated pups revealed that significant alterations in transmission from inhibitory to excitatory neurons were rectified by bivalent vector application. To examine the mutant transcriptome and impact of treatment, we used RNA sequencing and functional annotation clustering. Mutants displayed abnormal expression of more than 1,000 genes in highly significant and relevant functional clusters, clusters that were abrogated by treatment. Together these results suggest knockdown-replace as a potentially effective strategy for treating DNM1 and related genetic neurodevelopmental disease.

Graphical abstract

Jones, Soundararajan and colleagues developed a knockdown-replace gene therapy strategy in a mouse model of an intractable neurodevelopmental disease caused by DNM1 mutations. Delivery of a single neonatal dose of a recombinant virus virtually eliminated the most severe symptoms and is potentially applicable to any variant in this gene.

Introduction

DNM1 encodes dynamin-1, a large GTPase that catalyzes endocytosis and synaptic vesicle recycling.^1,2,3 DNM1 is expressed exclusively in the CNS, localizing to the neuron presynaptic terminal.^1,4,5,6 To date, more than 50 patients have been identified with de novo pathogenic variants in DNM1. Mutations reside in the GTPase and middle domains of the protein, driving severe developmental and epileptic encephalopathy (DEE).^{7,8,9,10,11,12} DEE is primarily attributed to genetic causes, encompassing a clinically complex and therapeutically challenging diagnosis, including and beyond seizures in children.^13,14,15 Clinical features are relatively homogeneous with affected children exhibiting intractable seizures starting within the first year of life, severe to profound intellectual disability, developmental delay, and muscular hypotonia.^8,9,10 As is the case with many DEEs, patients typically have intractable epilepsy with limited, if any, efficacy of antiepileptic medications.⁸

Prior evidence for a direct role for dynamin-1 in genetic epilepsy came in 2010 from the spontaneous mouse fitful (Dnm1^Ftfl) mutation,¹⁶ Although heterozygous fitful mice have recurrent non-lethal seizures without other overt features, homozygotes suffer a more severe, earlier DEE-like phenotype characterized by ataxia, neurosensory deficits, and severe seizures that result in death.^7,16,17 The underlying missense mutation is exclusive to an alternate exon (exon 10a, encoding Dnm1a) leaving intact a mutually alternative spliced exon (exon 10b, Dnm1b). Although there is overlap in expression, Dnm1b is expressed highest during early neuronal development and Dnm1a increases postnatally to become the predominant adult isoform.¹⁶ However, homozygous mice lacking Dnm1a or Dnm1b isoforms exhibit neither seizures nor other overt abnormalities associated with the Dnm1^Ftfl allele,^7,17 reflecting both functional redundancy and the dominant-negative nature of the pathogenic variant. More recently, in 2020 Aimiuwu and colleagues¹⁸ used homozygous fitful mice to demonstrate successful interfering RNA (RNAi) mitigation of key disease features whereby a single neonatal dose delivered converted a 100% lethal, severe seizure phenotype to a milder disease, including 75% survival until at least 4 weeks of age. These results showed the potential for treatment of heretofore intractable DNM1 DEE and possibly broader application.

Here we describe a new mouse model of DNM1 DEE, encoding the G359A variant that was identified in at least two DEE patients.¹⁹ Like Dnm1^Ftfl, G359A maps to the DNM1 middle domain and also encodes a dominant-negative effect that impairs endocytosis.¹ However, as with other pathogenic DNM1 human variants, G359A resides on an exon common to both Dnm1a and Dnm1b isoforms. Anticipating that mutant heterozygotes would be more severe than Dnm1^Ftfl including compromised husbandry, in this study we generated a conditional knock-in mutation to both examine cellular etiology and explore a new approach to therapy that is potentially applicable to any DNM1 pathogenic variant.

Results

Growth delay, poor survival, and severe seizures in conditional knock-in Dnm1 G359A mice

We developed a new mouse model of DNM1 DEE based on the G359A variant. Because heterozygous null Dnm1 mice are not impaired, to circumvent the anticipated poor husbandry and survival of Dnm1^G359A/+ heterozygotes, a Dnm1^G359A conditional mutation was generated using the knockout-first approach by inserting a gene trap cassette flanked by loxP sites engineered to carry a point mutation (c.1076G>C) in exon 8. The targeted allele is converted to mutant upon Cre-mediated deletion of the gene trap cassette, resulting in a glycine-to-alanine substitution at dynamin-1 amino acid residue 359 (G359A) (Figure S1).

Modeling G359A conditionally also allowed mutant allele expression both pan-cellularly and in neuron subtypes, providing an opportunity to select relevant phenotypic features that would present a rigorous challenge to new therapies. We initially expressed G359A broadly from the early embryo stage by crossing to Sox2-Cre driver mice. Sox2-Cre:G359A mice were weighed and observed for overt behavioral abnormalities and survival until postnatal day (PND) 30. Almost 50% of the mice died of indeterminate cause as early as a few days postnatal. However, while surviving heterozygotes showed very significant growth deficits from PND 5 (p = 1.1 × 10⁻¹⁰; n = 15 G359A/+, n = 20 −/+) (Figure 1), no overt abnormal behaviors or seizures were observed in the surviving mutants. Nestin-Cre:G359A mice were next examined to determine if mutant Dnm1 expression in all neurons yielded a better model for testing. Nestin-Cre G359A mutants showed some lethality later in adulthood, with 75% having non-lethal handling-associated seizures. They also had significant growth deficits (p = 3.3 × 10⁻⁸; n = 14 G359A/+, n = 14 −/+) (Figure 1). While more suitable, the incomplete penetrance and delayed seizure onset of Nestin-Cre G359A mice were not pragmatic for therapy testing.

Figure 1.

Visualization of 4-week growth curves for mutant and control littermates by Cre driver strain

Dnm1 genotype indicated by solid line (heterozygous: G359A/+) or dotted line (hemizygous, −/+). Cre driver strain indicated by color (Gad-Cre, black; Nestin-Cre, blue; Nkx2.1-Cre, green; Pvalb-Cre, purple; Sox2-Cre, red). The shading indicates standard error, and Xs are shown to indicate (for Sox2-Cre and Gad2-Cre) when mice were found dead. Weighing was terminated for Gad2-Cre mice after 18 days as most of the G359A/+ mice had succumbed to lethal seizure.

Following a prior study of Dnm1^Ftfl mice where it was shown that severe seizures relied DNM1 depletion from GABAergic neurons,⁷ we then examined G359A expression using Gad2-Cre driver mice, resulting in 100% lethal seizures starting as early as PND 12, fully penetrant by 3 weeks (n = 9) (Figure 1) with less severe growth defects than Sox2-Cre but still significant. Use of Nkx2.1-Cre, which is expressed in a subset of GABAergic neurons, resulted in no significant growth deficits (G359A/+ n = 11'; −/+ n = 19) (Figure 1) and few overt seizures until adulthood. Pvalb-Cre:G359A mutant mice did not show any significant early growth deficits (G359A/+ n = 20; −/+ n = 27) (Figure 1), but experience a low incidence of lethal seizures late in adulthood (not shown). Because Gad2-Cre conferred the most robust phenotypes, namely, fully penetrant severe seizures and delayed growth and each readily assessed by 4 weeks of age, we chose the Gad2-Cre:G359A model for testing gene therapy.

Bivalent adeno-associated virus vector to deliver RNAi and cDNA gene therapy

Given the challenges of efficiently and selectively targeting the G359A G>C transversion mutation we employed a knockdown-replace strategy. Specifically, we generated an adeno-associated virus (AAV)9 vector co-expressing an artificial miRNA engineered to direct RNAi against Dnm1, and a Dnm1 cDNA. This cDNA was synthesized to encode wild-type protein, but the nucleotide sequence was modified to optimize transfer RNA use and to be resistant to RNAi (Figure S2). We first used a bioinformatic screen^20,21 to identify effective artificial miRNAs targeting mouse Dnm1 and human DNM1 mRNAs, selecting 17 sequences each containing 22 nucleotides of antisense base pairing with human and mouse mRNAs (including G:U RNA base pairing) and cloned each as a DNA expression template into a U6 promoter plasmid (U6T6).²¹ We then screened potential miDnm1 candidates using a dual luciferase assay in HEK293 cells transfected with U6.miDNM1 plasmids, or controls, and a dual luciferase reporter plasmid (psiCheck2-DNM1) containing full-length DNM1 cDNA as the 3′ UTR of Renilla luciferase and a separate firefly luciferase cassette as a transfection control (Figure 2A). Compared with the U6T6 empty control, all 17 U6.DNM1 constructs triggered silencing of the heterologous Renilla luciferase-DNM1 reporter and we selected five constructs (mi249, mi1156, mi1505, mi1869, and mi2186) for confirmatory studies against mouse Dnm1 targets. Specifically, we generated two additional psiCheck2 plasmids containing a wild-type mouse Dnm1 cDNA or a codon-optimized engineered mouse Dnm1 cDNA with wobble mutations in miRNA binding sites, and tested our five lead sequences against each reporter construct. All five lead miDnm1s directed silencing of the wild-type mouse Dnm1 reporter, and four showed reduced silencing against the codon-optimized Dnm1 target (Figure 2B). We selected mi1869 as our final lead construct for in vivo studies, because it silenced both human DNM1 and mouse Dnm1 by more than 60% (N = 3; p < 0.0001; one-way ANOVA; Dunnett’s multiple comparison test) and was ineffective against the engineered Dnm1 cDNA (N = 3; p = 0.91; one-way ANOVA, Dunnett’s multiple comparison test) (Figure 2B). miDnm1-1869 and the RNAi-resistant, codon-optimized Dnm1 cDNA were synthesized in an AAV9 vector, respectively driven by a U6 promoter or human synapsin 1 promoter (Figure 2C). For a knockdown-only control, we also constructed a version of the vector that lacks the Syn1 promoter; for a replacement-only control, we included only the modified cDNA (Table S1).

Figure 2.

Bivalent AAV vector development design

(A) Renilla-Firefly dual-luciferase assay for testing efficacy of microRNA shuttle designed to interfere with mouse (Dnm1) or human (DNM1) dynamin-1 mRNA in cell culture. Graph shows efficacy luciferase expression screen of fifteen test inserts compared with empty or miGFP vector against DNM1 expressed in HEK cells. (B) Five test inserts were evaluated against wild-type (left) mouse Dnm1, all but one of which did not target codon-optimized, RNAi-resistant mouse Dnm1 (right). (C) Design of bivalent AAV vector showing tail-to-tail configuration of SYN1-coDNM1-V5 and U6-mi1869 inserts. Error bars represent standard error of the mean.

We delivered 7.2 × 10¹¹ viral genomes (vg) of ssAAV9-U6-miDnm1-hSyn1-CO-mDnm1-V5 to PND0-PND1 Gad2-Cre:G359A pups by intracerebroventricular injection. We considered this to be a maximum dose based on the virus concentration and the upper practical volume limit of approximately 10 μL per newborn pup. Thereafter, every 2–3 days until at least 4 weeks of age we monitored body weight, survival, and any seizures observed during handling. At this high dose, while only one-half of the animals eventually succumbed to a terminal seizure by 4 weeks, this was irrespective of Dnm1 genotype, i.e., the rate of lethality was similar for −/+ littermates (Figures 3A and 3B). At this dose, for animals that reached 4 weeks body weight never reached that of untreated −/+. Next, we applied a half dose (3.6 × 10¹¹ vg) and a one-tenth dose (7.2 × 10¹⁰ vg) to additional pups. At the half dose, body weight was further improved although still lagged behind that of untreated −/+, and all the 10 heterozygotes survived (Figure 3C). At 7.2 × 10¹⁰ vg, only 1 of 15 treated heterozygotes died by 4 weeks, but after an initial lag the growth curve neared that of untreated −/+ by 4 weeks (p = 0.14; repeated measures multivariate ANOVA [MANOVA]) (Figures 3D and 4A).

Figure 3.

Visualization of individual growth curves and survival across gene therapy vectors

Gad2-Cre, heterozygous G359A/+ are indicated by solid lines and hemizygous −/+ littermates by dotted lines. All treatments were at PND 1 with the noted condition by (A, B, C, D, E, or F) as in the text. Pups were weighed approximately every 3-4 days, with intervening weights interpolated linearly, before analysis. Lines that terminate before PND 28 mark the last day a pup was seen alive. The numbers at the top of each panel show the number of pups found dead over the total number for each group. nd, not done.

Figure 4.

Repeated measures MANOVA analysis of body weight

First 4 weeks (A) and weeks 5–12 (B). In (A), pup body weights were measured approximately every 3-4 days; intervening day weights were interpolated linearly before analysis. The Bonferroni adjustment was applied to p values for pairwise comparisons in (A).

Although the nominal survival endpoint in our study was 4 weeks of age, and some survivors were thereafter used for molecular validations, the remaining treated G359A/+ mice survived for as long as we were able to keep them. Thus, after 4 weeks, at 7.2 × 10¹⁰ vg/pup, three mice lived to at least 60 days, four further mice lived to at least 90 days, and four more mice lived until at least 6 months; at the 3.6 × 10¹⁰ vg/pup dose six mice lived until at least 40 days and five mice lived to at least 7 months; at the 7.2 × 10¹¹ vg/pup dose, three mice lived until at least 70 days, four mice lived to at least 100 days, and one mouse lived to at least 150 days.

The 7.2 × 10¹⁰ vg dose was very effective in preventing severe seizures and prolonging survival, but post hoc observations suggested the survivors were not completely normal. First, by 12 weeks postnatal, the body weight of treated heterozygotes still lagged behind that of treated −/+ by more than 4 g (p = 0.02, repeated measures MANOVA) (Figure 4B). In addition, older treated G359A/+ male mice, but not untreated mutants or treated −/+, exhibited aggression, e.g., fighting with male littermates. Also, although each of four treated G359A/+ females that were bred with a wild-type partner produced several litters each, only two successfully and consistently raised their pups. We also note that the rescue was less effective when Gad2-Cre: G359A was tested on the inbred C57BL/6J strain background, compared with the hybrid background, which had been chosen to maximize litter size, pup size, and overall health. Thus, while G359A/+ pups treated with 7.2 × 10¹⁰ vg bivalent vector survived significantly longer than untreated (p < 0.002, Wilcoxon log rank test) (Figure S3), with two treated G359A/+ mice living to at least 160 days, most succumbed to a lethal seizure before 4 weeks of age.

To determine whether the RNAi and the cDNA replacement features are both necessary and sufficient for these significant improvements in phenotype, in Gad2-Cre:G359A mice we tested the effect of RNAi knockdown without replacement, and vice versa. Although one of the eight Gad2-Cre: G359A/+ mice treated with only the replacement co-Dnm1 cDNA did survive until at least 80 days, neither knockdown- nor replacement-only control came close to the success of the bivalent vector (Figures 3E and 3F), indicating that both RNAi and cDNA are required. The 7.2 × 10¹⁰ vg/pup dose of the bivalent vector was chosen for further cellular and molecular assessments.

Molecular assessment of transduction

Bulk RNA sequencing (RNA-seq) was performed on combined cortex-hippocampus tissue RNA from PND15-PND17 pups. Endogenous Dnm1 mRNA counts in treated G359A/+ pups were at 78% of the level of untreated −/+, representing 22% knockdown of Dnm1 across these brain regions (Figure 5A and Supplemental File 1). Knockdown was greater in −/+ mice than in G359A/+ mice (p = 0.0004 vs. p = 0.02, respectively, generalized linear mixed model), presumably because the −/+ mice have only one expressed Dnm1 copy. We also observed that exogenous, codon optimized, RNAi-resistant Dnm1 mRNA in treated G359A/+ pups was expressed at 5.9% of the level of endogenous Dnm1 mRNA of untreated (Figure 5A and Supplemental File 1). Nevertheless, assessment of protein levels at the same age shows that an ample amount of total DNM1 and DNM1-V5 was expressed in G359A/+ treated tissue (Figures 5B and 5C), after normalization, almost 30% of that of untreated G359A/+ (Figure S4); the increase presumably represents the effect of codon optimization of the replacement mRNA.

Figure 5.

Quantification of endogenous and viral-delivered dynamin-1 mRNA and protein in PND15-PND17 mouse pups

(A) Normalized RNA-seq transcript counts (also see Table S3) in bivalent treated or control heterozygous G359A/+ and hemizygous −/+ pups for endogenous Dnm1 mRNA (Generalized Linear Mixed Model: p < 0.01 genotype effect, p < 0.0001 treatment effect) and exogenous codon-optimized, RNAi resistant, V5 epitope-tagged virally transduced Dnm1 mRNA (CO-Dnm1-v5). (B and C) Western blot showing total DNM1 protein (detected with DNM1-specific antibody; least-squares regression: p < 0.05 genotype effect, p < 0.05 treatment effect) and exogenous DNM1 protein detected with antibody to the V5 epitope tag. Error bars represent standard effort of the mean.

We recognize that these levels reflect tissue-wide, not single cell, averages. To estimate the fraction of neurons transduced, we visualized AAV9 transduction by co-staining of exogenous, modified DNM1 protein and for parvalbumin neurons, the latter chosen because they are sparse enough to readily count cell bodies but represent a large fraction of the inhibitory neuron population in the forebrain (Figures 6A and 6B and Table S2). In the cerebral cortex, across genotypes and replicates an estimated 7.4% of all transduced neurons were parvalbumin positive neurons, and an estimated 10.5% of all parvalbumin positive neurons were transduced. In the hippocampus, across both genotypes and replicates an estimated 2.4% of all transduced neurons were parvalbumin-positive neurons, and an estimated 12.5% of all parvalbumin-positive neurons were transduced. These results show that only a fraction of neurons need be transduced to produce the phenotypic improvements we observed and, together with molecular assessments, suggest that endogenous Dnm1 was fully depleted on a per-cell basis and only a modest amount (approaching haploid levels of) Dnm1 replacement suffices for this level of rescue.

Figure 6.

Viral transduction at 2 weeks postnatal

Shown at the left are z stack images from Dnm1 heterozygous G359/+ and hemizygous −/+ brain from treated and control pups. scAAV9-U6-miDnm1-hSYN1-coDnm1 is visualized with an antibody to the V5 epitope tag (green). Inset is higher magnification of stippled boxes, showing examples from single layer of V5 and parvalbumin (PV, red) and colocalization (yellow-orange). Shown at the right are neuron counts of V5 (green), PV (red) and colocalized cells (yellow) and their mean values (V5, dark gray; PV, medium gray; colocalized, light gray).

Impact of G359A and gene therapy on the transcriptome

Although dynamin-1 protein is not known to have a direct role in gene expression, given the severe condition of Gad2-Cre:G359A/+ pups, we nevertheless anticipated a measurable impact on the transcriptome. If so, it follows that some correction should accompany successful gene therapy. From the RNA-seq experiment we detected 539 genes significantly downregulated in G359A/+ mutant compared with −/+, and 472 genes significantly upregulated (false discovery rate-adjusted p ≤ 0.05), with the largest change being 4.6-fold in either direction (Supplemental File 2). In contrast, treatment alone on −/+ pups yielded only 19 downregulated genes, including Dnm1, and 29 upregulated with the greatest change being 3.5-fold in either direction. Importantly, the latter comparatively modest changes in −/+ suggest that treatment per se does not lead to a significantly altered transcriptome.

To assess the impact of the bivalent vector on the transcriptome of Dnm1 G359A/+ and −/+ pups, in a way that highlights the relevant biological effect, we examined functional annotation clustering based on Gene Ontology (GO) terms. Down-regulated genes and up-regulated genes were analyzed separately to optimize the power of this approach. For the 539 down-regulated genes, highly significant clustering (over 7 orders of magnitude) was observed for GO Biological Process terms that would be expected mechanistically for impaired dynamin-1 function; groups such as synaptic signaling, modulation of chemical synaptic transmission, regulation of trans-synaptic signaling, and others (Figure 7A). For GO Cell Compartment terms even more significant clustering (15 orders of magnitude) was seen for terms such as synapse, neuron projection, and others (Figure 7B). These results served as a baseline against which to compare the effect of treatment in mutant pups. Indeed, when comparing treated G359A/+ with untreated −/+, the clustering for the most affected categories was effectively flattened, decreasing by 5.5 orders of magnitude (GO Biological Process) or 12 orders of magnitude (GO Cell Compartment), suggesting significant correction in functions reflecting the known mechanism of dynamin-1. The 472 up-regulated genes also showed significant clustering (Figures 7C and 7D), but a lesser extent and for very different GO Biological Process terms, perhaps reflecting physiological response to the pathogenic mechanism. But even in these categories clustering was significantly decreased after treatment, although to a lesser extent than it was for down-regulated genes.

Figure 7.

Functional annotation clustering using GO terms of treated and untreated mutant and control mouse pup cortex and hippocampus RNA combined harvested at PND15-PND17

(A and B) Reduced expression in untreated heterozygous G359A/+ compared with hemizygous −/+. (C and D) Increased expression in untreated G359A/+ compared with −/+. Clustering significance is indicated by the magnitude of the log(Q) value (logarithm of the false discovery rate-adjusted p-value) on the Y axis. GO terms were ordered left to right on the X axis based on the significance of clustering in untreated mutant compared with control (green line). More nominal clustering is observed in the other pairwise comparisons (−/+ untreated vs. treated: purple line; G359A/+ untreated vs. treated: red line; G359A/+ treated vs. −/+ untreated: blue line).

Interestingly, a greater number of gene expression differences were observed between treated and untreated G359A/+ pups compared with those between treated and untreated −/+ (down-regulated 364 genes vs. 19 genes; up-regulated 408 genes vs. 29 genes, respectively). These differences are comparably reflected in GO Cell Compartment clustering as clustering spikes in groups representing neuronal projections and synapses (Figures 7B and 7D). Further studies will be needed to determine the degree to which these differences represent residual Gad2-Cre:G359A pathologies or features that emerge from the treatment.

Impact on synaptic inhibition

Previous data from Dnm1 knockout mice and mice with pathogenic Dnm1 variants suggest that the major cellular effect of Dnm1 dysfunction is disruption of presynaptic neurotransmission.^4,16,22,23 We, therefore, set out to determine how the G359A variant affects excitatory and inhibitory neurotransmission and to what degree these effects are corrected by bivalent vector treatment. To do this, we performed electrophysiological recordings from pairs of cultured mouse cortical neurons, a system that has been previously used to characterize the physiological roles of Dnm1.^4,22,23,24

We first measured action potential evoked responses in connected pairs of excitatory-inhibitory neurons from Gad2-Cre: G359A/+ and Gad2-Cre: +/+ mice, and in the presence or absence of transduced bivalent vector. We found that the amplitude and paired pulse ratios of evoked excitatory currents onto inhibitory neurons were unaltered by the mutation or by bivalent vector treatment (Figure 8A). The amplitude of evoked inhibitory responses onto excitatory neurons, however, was decreased in mutant cells, and the paired pulse ratio was increased, indicating an impairment in presynaptic GABA release. This deficit is similar to previous findings in both Dnm1 null and Dnm1fitful neurons.^4,25 Both of these effects were rescued by the bivalent vector (Figure 8B). We also observed enhanced facilitation in evoked induced pluripotent stem cells (IPSCs) at 10 Hz in mutant neurons, similar to a previous observation in Dnm1, Dnm3 double null neurons,²⁶ and this was also rescued (Figure 8C). Finally, we measured miniature IPSCs (mIPSCs) in the presence of tetrodotoxin (TTX). Although there was no change in the frequency of mIPSCs (not shown), the size of IPSCs was increased in mutant neurons, and reversed by treatment. Together, these experiments show that the G359A variant impairs several aspects of presynaptic GABAergic neurotransmission, which are likely key mediators of the on vivo phenotypes, and that the AAV9-hSYN1-coDnm1-U6-miDnm1 viral vector fully protects GABAergic neurons from each of these functional deficits.

Figure 8.

Neurons from Dnm1 G359A; Gad-Cre mice show altered inhibitory synaptic inhibition that is rescued by the bivalent gene therapy

(A) Representative traces and summary data showing that the strength of evoked excitatory transmission and paired pulse ratios onto inhibitory neurons are unaltered by G359A expression or the bivalent treatment. (B) Conversely, the amplitude of evoked inhibitory responses onto excitatory neurons are reduced by G359A expression and the paired pulse ratio is increased. Both of these changes are rescued by the bivalent treatment. (C) G359A neurons also show enhanced facilitation of evoked IPSCs that is rescued by treatment. (D) The size of mIPSCs recorded in excitatory neurons is increased in G359A neurons and rescued by bivalent treatment. n.s. = p > 0.05, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 as tested with generalized estimating equations. Error bars represent standard error of the mean.

Discussion

In this study, we developed a mouse model that is representative of dominant-negative variants causing DNM1 DEE, an ultra-rare and devastating childhood disease causing severe symptoms that do not usually respond to medicines. The model, designed to encode the pathogenic amino acid substitution G359A residing in the DNM1 middle domain responsible for molecular assembly, was then used to explore a powerful approach to gene therapy—knockdown-replace—which is potentially broadly applicable to any pathogenic variant of a gene.

Genetically engineered mouse strains for studying neurodevelopmental disease are known to vary in the degree to which they model human clinical features, and DEE models are no exception, collectively spanning the full range of expectations from mild to severe.²⁷ We have speculated that interspecies differences in neurodevelopmental timing—both relative and absolute—and in the magnitude of expression and impact in different neuron types are major reasons behind the mismatches. To that end, we first examined G359A in a variety of cellular contexts to identify a form of the model that is suitable for a rigorous test of the therapy. Although the genetically most accurate mouse model would be a mutation that is constitutive from conception, we generated a conditional model because we anticipated impaired husbandry and survival. The conditional model further allowed both approximation of constitutive expression - the Sox2-Cre driver conveys very early pan-cellular expression—and expression in a variety of neuron types. While approximately one-half of the Sox2-Cre:G359A heterozygotes died before PND 10, which we judged to be too soon after vector transduction to provide a fair test of rescue, expression in neuron subtypes proved to be more useful. Thus, led by a prior study of Dnm1^Ftfl mice showing that most severe phenotypes were associated with Dnm1 deficiency in inhibitory neurons,⁷ we turned to interneuron Cre drivers. Thus, G359A expression in all GABAergic inhibitory neurons via Gad2-Cre led to significant growth delay without early postnatal lethality, and severe seizures by 3 weeks of age. Although crossing to inhibitory neuron subtype drivers such as Nkx2.1-Cre and Pvalb-Cre also produced seizures, we chose Gad2-Cre for testing because of its strong, highly penetrant features, even on a mixed strain background. The success of phenotypic rescue in the Gad2-Cre:G359A model gives more confidence about testing effectiveness against more subtle DNM1 DEE behavioral and neurosensory phenotypes in the future, where experimental design with further attention to interspecies and strain differences will be essential.

Knockdown-replace is a strategy that could apply to any molecular genetic mechanism, but is particularly relevant heterozygous dominant-negative variants, which inherently represent imbalanced expression between mutant and wild-type alleles. The bivalent vector design used tried-and-true promoter-enhancers suitable for the respective RNA cargo. Because of the cargo size limit of AAV, to include both these genes and promoters in a single efficient vector required the single-strand AAV genome for expression that, after cell transduction, can take 1 week or longer than self-complementary AAV to produce a double-stranded DNA template suitable for transcription.²⁸ In this respect, it was surprising to us that transduction of mouse pups even within 1 day of birth was early enough for a disease that, in mouse, has a readily detectable phenotype—growth delay—before the end of the first week. Indeed, across the range of doses we tried, around the beginning of week 2, there was a noticeable growth rate depression that preceded a modest acceleration. We think that this lag-and-recovery corresponds with the robust early knockdown of endogenous Dnm1 mRNA (due to miDnm1 driven by the powerful U6 promoter) followed by eventual full expression of the exogenous Dnm1 mRNA (driven by the SYN1 promoter). We further speculate that mutants on the less successful C57BL/6J inbred strain background, which are smaller to begin with, are more difficult to successfully treat because they suffer more from this expression lag. It is encouraging that the transduction of primary neurons normalized the synaptic changes caused by G359A, as these synaptic phenotypes are likely primary consequences of Dnm1 dysfunction,²⁵ and previous work had indicated that improving synaptic dysfunction caused by Dnm1 variant can have a disease-modifying effect.²⁹ Because primary neurons are easier to transduce efficiently than in vivo and our in vivo transduction rates were modest, improvements in viral delivery efficiency in vivo may further improve the effectiveness of the knockdown-replace strategy. Although the effective window for therapy is, in absolute terms, vastly wider in children than in mice, further optimization of expression level and timing in viral vectors may lead to improved pre-clinical effectiveness in severely impaired mouse models.

Gene therapies for genetic neurodevelopmental diseases like DEEs are in their infancy, but there are already signs of success. Oligonucleotide-based therapy tested in mouse models for DEE caused by mutations in SCN1A³⁰ was shown to confer significant resistance to severe phenotype features in mice and is presently in clinical trial (ClinicalTrials.gov ID NCT04740476). Also in clinical trial is an AAV-based mini-cDNA replacement of MECP2 RNA for Rett syndrome (ClinicalTrials.gov ID NCT05898620), also developed in mouse model studies.³¹ The preclinical studies clearly show these approaches can be very effective in the mitigation of seizures, a likely result given that seizures rely on connectivity across often widely distributed spatial cellular networks that may only be sparsely transduced by vector delivery. Although merely preventing seizures alone may be highly desirable or even transformative for families, further studies may be required to determine whether more focal pathophysiology is similarly improved.

Materials and methods

Animals and identification

All procedures involving mice were approved by Columbia University Institutional Animal Care and Use Committee and carried out in accordance with the National Institute of Health Guide for the care and use of laboratory animals. Both sexes were used throughout this study. Dnm1^G359A-cKI mice were generated by Leveragen Inc. (Boston, MA, USA) using a knockout-first approach. Briefly, a floxed gene trap cassette was inserted in front of the target exon while replacing the target exon with engineered mutant exon containing the desired point mutation. This orientation created a knockout allele that can be turned into the point mutation allele by Cre-mediated deletion of the gene trap cassette. Homologous recombination was used to target this construct to the Dnm1 locus in mouse embryonic stem cells derived from 129S6 and B6NTac inbred strains. Founder Dnm1^G359A/+ mice were continually crossed to C57BL/6J (JAX Stock No. 664; Bar Harbor, ME, USA) to produce congenic B6J-Dnm1^G359A/+ mice, or crossed to FVB/NJ (JAX Stock No. 1800) to produce heterozygous F₁ hybrid progeny for mating to B6J-Cre driver strains to generate N₂ hybrids used in most experiments.

Newborn mice paw pads were tattooed for identification from PND 0 to PND 10 by the AIMS pup tattoo identification system using Ketchum Animal Tattoo Ink (#329AA) and ear notched at PND 10 for subsequent identification. Mice were weighed approximately every 3–4 days until 4 weeks of age, and approximately weekly thereafter, and never separated from their home cage for extended amounts of time. During handling, mice were observed for handling-induced tonic-clonic seizures and monitored for recovery before being returned to their home cage.

N₁ hybrid mice carrying the Dnm1^G359A conditional mutation were crossed to the following Cre driver strains purchased from JAX for experiments: Sox2-Cre (JAX Strain # 008454), Nestin-Cre (JAX Strain # 003771), Emx1-Cre (JAX Strain # 005628, Gad2-Cre (JAX Strain # 028867), Nkx2.1-Cre (JAX Strain #008661), and Pvalb-Cre (JAX Strain # 017320). Genotyping for the G359A encoding mutant allele was achieved by PCR using primers 5′-ACG AAG TTA TTA GGT CTG AAG AGG-3′ and 5′-CTT GTA GTT GCC GTC GTC CT-3′ to produce a 562-bp band.

Production of AAV vectors

Cloning and production of ssAAV9-U6-miDnm1-1869 -Syn promoter-co-Dnm1-V5 was carried out according to previously described protocols in.³² Briefly, full-length human DNM1 and mouse Dnm1 sequences were input into the Harper miRNA shuttle predictor version 1.0.³² Perfect 22-nt sequence-matched miRNAs (including G:U wobble base pairs) were selected for in vitro testing. Predicted sequences were cloned into the U6T6 backbone (Boudreau et al.²¹). The AAV vector expressing the codon-optimized Dnm1-V5 alone was generated by replacing a human DNM1 cDNA driven by Syn1 with the codon optimized Dnm1 sequence by cloning into NheI/KpnI flanking sites. For the purpose of quantifying transgenic expression in vivo, a V5 tag was added onto the end of the codon-optimized Dnm1 with PCR using primers 5′-AAAAGGATCCTCAGCTCGAAAGAC and 3′- TTTTGGTACCTTACGTAGAATCGAGCCCGAGGAGAGGGTTAGGGATAGGCTTCCCGGGATCGGAGATGGTGATG and then cloning into the BamHI/KpnI sites. Lead candidate U6.miDnm1-1869 and the human SYN1 promoter-coDnm1 cDNA cassettes were inserted into a ssAAV9 proviral backbone in toe-to-toe orientation so that the miRNA T6 termination sequence was positioned adjacent to the Dnm1 poly A signal. The miRNA knockdown-only control vector was generated by removing the SYN1 promoter sequence with SpeI and NheI. AAV9 vectors were generated and quantified for titer by Andelyn Biosciences (Columbus, OH, USA). miDnm1-1869 mature structure and sequence is provided in Figure S2.

Dual-luciferase assay screening of DNM1-targeting miRNAs

HEK293 cells were cultured according to manufacturer’s instructions (ATCC, Manassas, VA, USA). Two hours before transfection, 65,000 cells were cultured in a white 96-well plate and later co-transfected (Lipofectamine 2000, Invitrogen, Carlsbad, CA, USA) with a dual-luciferase expression plasmid containing either human DNM1, mouse DNM1, or DNM1-co placed in the 3′ UTR of Renilla Luciferase and a U6T6 plasmid expressing each artificial miRNA. After 24 h, cells were lysed with 30 mL 1× Passive Lysis Buffer (Dual-Luciferase Reporter Assay, Promega, Madison, WI, USA). Cells were then vortexed on a plate shaker at 425 rpm for 20 min. A GloMax Discover Machine (Promega) was used to quantify luciferase expression according to the Dual-Glo protocol generated by Promega. Percent knockdown by each miRNA was calculated by determination of the ratio of Renilla:Firefly expression and normalizing to the empty U6T6 vector control.

Intracerebroventricular injection

Intracerebroventricular delivery of vector was carried out at PND 0 according to methods described in.³³ Briefly, mice were anesthetized by hypothermia and injected with 5–10 μL depending on the dose. Injections were carried out free-hand using 10 μL Hamilton Neuros Syringe (#65460-06) at approximately two-fifths the distance from the lambda suture to each eye. After treatment pups were monitored for growth, overall survival, and any over features such as handling-associated seizures.

Bulk RNA-seq and analysis

Cortex and hippocampus hemispheres were sampled together from PND15-PND16 mice in triplicate, snap-frozen and delivered to Columbia’s Molecular Pathology Shared Resource core for RNA preparation (Qiagan [Hilden, Germany] miRNeasy micro kit – extracts RNA and miRNA over 200 nt). The purified RNA was provided to Columbia’s Sulzberger Genomics Core for paired-end sequencing (2 × 75-bp paired-end sequencing on the Aviti Element) Between 47 million and 70 million reads per sample were aligned to the reference mouse genome GRCm39.110 to which a sequence was added for the codon-optimized, RNAi-resistant Dnm1 cDNA. Count estimates were obtained after reads were mapped to the reference genome using the STAR aligner³⁴ and differential gene expression analysis was performed using DESeq2³⁵ at the GenePattern server.³⁶ Results were filtered at adjusted p value of ≤0.05 and subsequent GO functional annotation clustering analysis was done using VLAD.³⁷

Western blot

Dissected mouse brain tissue was snap frozen in liquid nitrogen and stored at −80°C until the time of extraction. Tissue was thawed on ice and homogenized using a motorized pestle in RIPA buffer containing both protease and phosphatase inhibitor cocktail (Roche, Basel, Switzerland). Samples were centrifuged and the resulting supernatant was collected and quantified using BCA method (Pierce, Appleton, WI, USA) with BSA as standard. Using Xcell Surelock Mini Cell system, a total of 15 μg of protein lysates were loaded onto a 4%–12% SDS-PAGE gels and subsequently transferred to PVDF membrane. The membranes were incubated with primary antibodies: Dynamin1- 1:200 (Invitrogen #PA1-660); V5-Tag (D3H8Q), 1:1,000 (CST #13202); ACTB, 1:15,000 (Santa Cruz Biotechnology, Dallas, TX, USA: sc-47778) overnight at 4°C, followed by incubation with secondary horseradish peroxidase-conjugated goat anti-rabbit (1:10,000) (Proteintech, Rosemont, IL, USA; SA00001-2) for 1 h at room temperature. Signals were developed using Amersham ECL Western Blotting Detection Reagent (GE Healthcare, Chicago, IL, USA; RPN2106) and visualized using western blot imaging system (Azure Biosystems, Dublin, CA, USA; Azure C400).

Immunofluorescence and image analysis

Samples and processing

Three mice from each group (Dnm1 genotype, treatment) were perfused with 4% paraformaldehyde (PFA) for immunohistochemical assessment at PND 21–30. Brains were dissected from the skull and postfixed in 4% PFA overnight at 4°C. Free-floating 100-μm sections were collected using a vibratome (Leica, Wetzlar, Germany; VT1000S). The sections were permeabilized and blocked with 5% normal goat serum and 0.3% Triton X- in PBS for 1 h at room temperature. Slices were incubated with either NeuN antibody (1:500, Sigma, St. Louis, MO, USA; Cat. No. #MAB 377) or parvalbumin antibody (1:100, Synaptic Systems, Gothenburg, Germany; cat no #195004) with V5 tag antibody (1:200, Cell Signaling Technologies, Danvers, MA; Cat. No. #13202) in blocking buffer (5% normal donkey serum in 0.01% Triton X- in PBS) for 2 h in room temperature. Afterward, sections were washed in PBS for 10 min three times and incubated in Alexa Fluor secondary 555 (1:1,000, Thermo Fisher Scientific, Waltham, MA, USA; Ref. No. A21428), Alex Fluor secondary 595 (1:1,000, Thermo Fisher Scientific; Ref. No. A11076), Alex Fluor secondary 488 (1:1,000, Thermo Fisher Scientific; Ref. No. A11008, A28175) with DAPI (1:1,000) and for 2 h at room temperature. Sections were mounted on slides and cover slipped with fluoromount-G (Southern Biotech, Birmingham, AL, USA; Cat. No. 0100-01).

Image quantification

Slides were imaged using a Zeiss LSM-800 confocal microscope and Zen v2.3. Tiled images of 20× magnification was acquired of the hippocampus and the somatosensory cortex region for each section keeping the laser and gain settings constant. After image acquisition, cell counting was manually carried out using the cell counter plugin within Zen Lite Software. This involved quantifying PV-positive, V5-positive, and PV-V5 co-localized cells across the entirety of the hippocampal and somatosensory regions. Postprocessing of images was carried out with Adobe Photoshop and the graphical representation were generated using JMP software.

Electrophysiology: Cell culture

To generate astrocyte feeder layers, cortical hemispheres from PND 0 to PND 1 wild-type C57BL/6J mice of either sex were dissected in cold HBSS (Gibco, Billings, MT, USA). For primary neuron culture, cortex from PND0-1 mice were dissected in cold HBSS and cell culture carried out as described.²⁵ The day after plating, approximately 4 × 10¹⁰ genome copies of AAV8-CaMKII-GFP virus (UNC Vector Core), 4 × 10¹⁰ of AAV9-hSyn-DIO-mCherry (Addgene #50459) virus, and 4 × 10¹⁰ of the bivalent AAV were added to each well.

Electrophysiology: Recordings and data acquisition

Whole-cell recordings were performed with patch-clamp amplifiers (MultiClamp 700B amplifier; Molecular Devices, San Jose, CA, USA) under the control of Clampex 10.5 (Molecular Devices, pClamp, RRID:SCR_011323). Data were acquired at 10 kHz and low-pass filtered at 6 kHz. The series resistance was compensated at 70%, and only cells with series resistances maintained at less than 15 MΩ were analyzed. Patch electrodes were pulled from 1.5-mm o.d. thin-walled glass capillaries (Sutter Instruments) in five stages on a Flaming-Brown micropipette puller (model P-1000; Sutter Instruments). Internal solution contained 136 mM KCl, 17.8 mM HEPES, 1 mM EGTA, 0.6 mM MgCl², 4 mM Na⁺-ATP, 0.3 mM Mg²⁺-GTP, 12 mM creatine phosphate-2K⁺, and 50 U/mL phosphocreatine kinase. The pipette resistance was between 2 and 4 MΩ. Standard extracellular solution contained the following (in mM): 140 NaCl, 2.4 KCl, 10 HEPES, 10 glucose, 4 MgCl₂, and 2 CaCl₂ (pH 7.3, 305 mOsm). All experiments were performed at 34°C–36°C with temperature maintained by a ThermoClamp system (Automate Scientific, Berkeley, CA, USA). Whole-cell recordings were performed on neurons from Dnm1^G359A/⁺ and wild-type groups in parallel on the same day (14–16 in vitro). On each day, and for each group, an attempt was made to record from equal numbers of glutamatergic (GFP⁺) and GABAergic (mCherry⁺) neuron pairs. All electrophysiology experiments were performed by two independent investigators masked to the genotypes, and all data were analyzed offline with AxoGraph X software (AxoGraph Scientific, Union City, CA, USA; RRID: SCR_014284).

For voltage-clamp experiments, neurons were held at −70 mV. Miniature synaptic potentials were recorded for 70–90 s in 500 nM TTX (Enzo Life Sciences, Farmingdale, NY, USA) to block action potential (AP)-evoked release, and either the GABAA receptor antagonist bicuculine methiodide (30 μM; Hello Bio, Bristol, UK) to isolate miniature excitatory postsynaptic currents (mEPSCs), or the AMPA receptor antagonist NBQX disodium salt (10 μM; TOCRIS Bioscience, Bristol, UK) to isolate mIPSCs. Data were filtered at 1 kHz and analyzed using template-based miniature event detection algorithms implemented in the AxoGraph. The threshold for detection was set at 3.5 times the baseline SD from a template of 0.5 ms rise time and 3 ms day for mEPSCs, and 20 ms day for mIPSCs. For paired neuron recordings, AP-evoked EPSCs and IPSCs were triggered by a 2-ms somatic depolarization to 0 mV at 0.1 Hz. The shape of the evoked response and the effect of receptor antagonists (bicuculline and NBQX) were analyzed to verify the glutamatergic or GABAergic identities of the responses.

Data and code availability

Any materials or information developed for this paper not in the paper itself will be available upon request to qualified researchers. Newly generated mouse strains will be made available from a public repository or vendor.