CRISPRa-mediated Kcna1 upregulation decreases neuronal excitability and suppresses seizures in a rodent model of temporal lobe epilepsy

Epilepsy is a major health burden, calling for new mechanistic and therapeutic insights. CRISPR–mediated gene editing shows promise to cure genetic pathologies, although hitherto it has mostly been applied ex-vivo. Its translational potential for treating non-genetic pathologies is still unexplored. Furthermore, neurological diseases represent an important challenge for the application of CRISPR, because of the need in many cases to manipulate gene function of neurons in situ. A variant of CRISPR, CRISPRa, offers the possibility to modulate the expression of endogenous genes by directly targeting their promoters. We asked if this strategy can be effective to treat acquired focal epilepsy. We applied a doxycycline-inducible CRISPRa technology to increase the expression of the potassium channel gene Kcna1 (encoding Kv1.1) in mouse hippocampal excitatory neurons. CRISPRa-mediated Kv1.1 upregulation led to a substantial decrease in neuronal excitability. Continuous video-EEG telemetry showed that AAV9-mediated delivery of CRISPRa, upon doxycycline administration, decreased spontaneous generalized tonic-clonic seizures in a model of temporal lobe epilepsy. The focal treatment minimizes concerns about off-targets effects in other organs and brain areas. This study provides the proof of principle for a translational CRISPR-based approach to treat neurological diseases characterized by abnormal circuit excitability.


Introduction
Epilepsy affects up to 1% of the population, and 30% of patients continue to experience seizures despite optimal medication (1, 2). Although the majority of drug-resistant epilepsies are focal, targeting drugs to a restricted brain region presents major challenges, and potentially curative surgery is limited to a minority of cases where the seizure focus is remote from eloquent cortex (3). Gene therapy holds promise as a rational replacement for surgery for intractable pharmaco-resistant epilepsy, and could in principle improve the prospect for seizure freedom in many people (3,4). Several approaches have been proposed to interfere with epileptogenesis or to decrease seizure frequency in chronic epilepsy (5). Current experimental gene therapies mainly rely on viral vector-mediated expression of genes encoding normal CNS proteins or exogenous non-mammalian proteins (4,(6)(7)(8)(9). This approach has several potential limitations, including a finite packaging capacity of viral vectors, difficulty in ensuring normal splicing and post-transcriptional processing, and, in the case of non-mammalian proteins, concerns about immunogenicity. Modulating the expression of endogenous genes, in contrast, would represent an important step toward safe and rational treatment of intractable epilepsy and other neurological diseases.
The DNA editor/regulator CRISPR/Cas9 (10-12) represents a powerful tool to modify endogenous genes, not only in somatic cells but also in mammalian neurons (13,14). In addition to permanently altering endogenous gene sequences, CRISPR/Cas can regulate the activity of genes through promoter modulation, an approach known as CRISPR activation (CRISPRa) (10,15). CRISPRa is therefore a promising tuneable tool to increase the expression of genes encoding, for instance, ion channels, in chronic epilepsy in order to restore physiological levels of network activity (6,16). The CRISPRa system is composed by a nuclease-defective Cas9 (dCas9) fused to a transcription activator and a small guide RNA (sgRNA) that targets dCas9 to the promoter of the gene of interest (10). The advantages of this system are multiple. First, it is versatile because the targeted gene can be switched simply by changing the sgRNA. Second, CRISPRa preserves the full range of native splice variants and protein biogenesis mechanisms (15). Third, CRISPRa is, in principle, safe because it only alters the promoter activity of genes that are already transcribed in targeted neurons. It can be targeted to specific neurons in the epileptic focus using established viral vectors (17).
Here, we report the use of CRISPRa to treat a mouse model of temporal lobe epilepsy, from in vitro validation to demonstration of efficacy in reducing seizure frequency in vivo.

A CRISPRa system targeting the Kcna1 promoter region increases Kv1.1 expression and decreases neuronal excitability
We first asked if CRISPRa can be exploited to increase endogenous gene expression in glutamatergic neurons and decrease their excitability. As a proof of principle, we chose the Kcna1 gene. Lentivirus-mediated overexpression of Kcna1 reduces neuronal excitability and, when targeted to principal cells, suppresses seizures in neocortical models of epilepsy (6,18). We first conducted a bioinformatic analysis to identify its promoter region. Alignment of datasets of gene expression and epigenetic markers of actively transcribed genes in perinatal and adult mouse brain identified peaks of enrichment for RNA PolII, mono-and trimethylation of lys4 and acetylation of lys27 of H3 histone along the gene (Supplementary Figure 1A). One of these regions was located immediately upstream to the annotated Kcna1 transcription start site (TSS) and identified as a suitable target for CRISPRa. We submitted 200 bps from this region to the CRISPOR web tool (http://crispor.tefor.net) for sgRNA design, and selected four candidate guides (sg4, sg14, sg19 and sg30) for validation, initially in the P19 cell line. sgRNAs were lipofected individually or in combinations, together with a construct carrying dCas9 fused to the transcriptional activator VP160 (dCas9-VP160) and a Puromycin resistance cassette. dCas9 with sg4, sg14 or sg19, but not with sg30, significantly upregulated the expression of the Kcna1 gene. We focused on sg4 and sg19, which induced the highest levels of Kcna1 expression (Supplementary Figure 1B). When tested in combination, sg4 and sg19 together were also efficacious, but not sg4 and sg30 (Supplementary Figure 1C). We confirmed that the highest efficiency of upregulation of Kcna1 in primary neurons was achieved with sg19 (Figure 1 B). Consequently, we generated a construct carrying dCAS9-VP160 driven by the Ef1a promoter and either the sg19 targeting the Kcna1 promoter (Kcna1-dCAS9A) or a control sgRNA targeting LacZ (Ctrl-dCAS9A). Western Blot analysis confirmed increased Kv1.1 protein levels in sg19treated neurons when compared to the sgLacZ control. Importantly, we detected increased levels of glycosylated Kv1.1, corresponding to mature protein, implying normal processing of the upregulated potassium channel ( Figure 1C, D).
The CRISPOR tool predicted 250 putative off-target genes for sg19, mostly with a low likelihood score. To evaluate the specificity of CRISPRa we performed a gene expression profile analysis in primary neurons treated with Kcna1-dCAS9A and compared this with Ctrl-dCAS9A transduced neurons. No consistent alteration in the transcriptome of sg19 treated neurons was observed, except for a significant increase in Kcna1 (red dot, Figure 1E). Six out of 250 predicted off-targets for sg19 were located close to promoters of Mylpf, Efcab4a, Nudcd2, Pde4b, Gc and Vps16 genes. However, none of these genes showed a significant change in expression in either the transcriptome analysis (green dot, Figure 1E) or in quantitative RT-PCR assays ( Figure 1F).
Exogenous Kcna1 overexpression results in a decrease in neuronal excitability (6). To test the efficiency of the CRISPRa system, primary neurons were transduced at 1DIV with a lentivirus expressing Kcna1-dCAS9A or Ctrl-dCAS9A. After 14-16DIV we used whole-cell patch clamp recordings to analyse neuronal excitability of both experimental groups ( Figure   1G). The maximal firing frequency was significantly decreased in neurons transduced with Kcna1-dCAS9A when compared to Ctrl-dCAS9A ( Figure 1H). Other excitability parameters sensitive to Kv1.1 were also changed in neurons transduced with Kcna1-dCAS9A in comparison with Ctrl-dCAS9A: the current threshold was increased, and action potential width was decreased ( Figure 1I). Passive membrane properties and other AP properties were however unchanged (Supplementary Figure 2).

Kcna1-dCas9A decreases CA1 pyramidal cell excitability
In order to test the efficacy of CRISPRa in vivo, we subcloned the CRISPRa elements in two separate AAV9 vectors. One AAV vector carried the dCAS9-VP64 under the control of a rtTA responsive element (TRE), while the other vector included sg19 (or sgLacZ as a control) element and a human Synapsin promoter (hSyn) upstream to a floxed rtTA-t2a-tdTomato cassette. This experimental design allowed the Kcna1-dCas9A system to be activated in forebrain excitatory neurons of Camk2a-cre mice transduced with both AAVs, but only after doxycycline administration (Figure 2A, B). We co-injected both AAVs in the hippocampus of 2-3 month old Camk2a-CRE mice, which were subsequently fed with a doxycycline diet for 3 weeks and then sacrificed for preparation of acute brain slices. we applied activity clamp, a method to assess neuronal excitability in the face of epileptiform barrages of excitation (19). Neurons expressing Kcna1-dCAS9A fired less than neurons expressing Ctrl-dCAS9A when exposed to the same simulated synaptic input. Taken together, these results support using Kcna1-dCAS9A as a candidate antiepileptic gene therapy (Figure 2 G).

Kcna1-dCas9A decreases seizure frequency in a mouse model of temporal lobe epilepsy
We administered Kcna1-dCAS9A in a mouse model of acquired epilepsy. C57BL/6J WT animals were injected with kainic acid (KA) in the right amygdala (20). This induced a period of status epilepticus (SE), which was quantified by video recording to monitor seizure severity (Supplementary Figure 5 and Video 1). One week later, we injected either Kcna1-dCAS9A or Ctrl-dCAS9A AAVs in the right ventral hippocampus, and at the same time we Video 2). The baseline-normalized seizure count in animals injected with Ctrl-dCAS9A was significantly greater than in animals injected with Kcna1-dCAS9A. All the animals treated with Kcna1-dCAS9A showed a reduction of the number of seizures after doxycycline was added to the food (Figure 3 D). Furthermore, whilst the median total seizure count for Ctrl-dCAS9A animals was similar before and after doxycycline, Kcna1-dCAS9A animals exhibited a median reduction of approximately 50%. Other EEG parameters such as broadband power and seizure duration were unchanged (Figure 3 E). Taken together, these results suggest that Kcna1-dCAS9A increases the threshold for triggering a generalized tonic-clinic seizure but otherwise does not change general network properties. Histogram of the number of seizures after compared to before treatment. Mann-Witney nonparametric test. F. Seizure duration, coastline, power (12-30Hz and 50-70Hz) after the treatment normalised to before the treatment. Student's t test.

Discussion
Although CRISPR has attracted intense interest as a possible treatment for inherited or acquired genetic disorders, it has, hitherto, received much less attention as a potential tool to treat acquired non-genetic diseases. The overwhelming majority of epilepsy cases, which represent an enormous disease burden, are not thought to be due to single gene mutations but are acquired during life, often secondary to a variety of brain insults such as infections, strokes and injuries (1, 2). Here we have shown that CRISPRa can be used to increase endogenous Kcna1 expression to modulate neuronal activity, and thereby decrease the frequency of seizures in a mouse model of chronic temporal lobe epilepsy. This approach can potentially be used to regulate the expression of any gene, opening the way to treating many other neurological diseases associated with altered transcription. At present the main obstacles to translation for the CRISPR/Cas9 toolbox are absence of long-term data on potential immunogenicity of the bacterial nuclease in humans and possible off-targets that have not been detected by transcriptomic analysis (21). CRISPRa is, in principle, less likely to have deleterious off-target effects than gene editing because it does not cleave DNA (17,22), but further research is necessary.
Among distinct advantages of CRISPRa over exogenous gene delivery is the possibility to select one or more sgRNAs to tailor the exact level of gene expression independently by the number of viral copies effectively entered within each neuron. In addition, several sgRNAs can be designed to control the transcription of heteromultimeric proteins such as GABAA or NMDA receptors, or multiple genes in a signaling pathway. Finally, if gene supplementation therapy is strongly biased by the overall gene size which should not exceed the viral capacity cargo, the CRISPRa can be potentially applied to control the activity of any gene irrespective of its length (23). Although the present study made use of two AAVs to allow inducible activation of CRISPRa and expression of a fluorescent reporter protein, for clinical translation these features would not be required, and so it should be possible to package both the dCAS9 and the sgRNA in a single AAV to simplify clinical delivery. Further refinements can be considered, such as the use of an inducible promoter to allow the therapy to be switched off (16), which would not be possible with a gene editing strategy.
These results demonstrate that CRISPR-mediated control of gene expression can be successfully exploited to modulate neuronal activity to obtain a significant and long-term clinical management of chronic seizures in an experimental model of intractable temporal lobe epilepsy.

Study Design
This study aimed to test the hypothesis that upregulating endogenous genes (e.g. Kcna1) with CRISPRa can treat chemoconvulsant-induced temporal lobe epilepsy. The experiments were designed to achieve a power >0.8 with an a = 0.05. For in vivo experiments the 3Rs guidelines for animal welfare were also followed. Outliers were not excluded and at least 3 independent repetitions were performed. Exclusion criteria were applied for all the recordings (see methods below). All the experiments were randomized and researchers were blinded during recordings and analysis.

Animals and ethics.
All experimental procedures were carried out in accordance with the UK Animals (Scientific Procedures) Act 1986. C57BL/6J and Camk2a-CRE mice were used for the experiments.

RNA isolation and quantitative RT-PCR. RNA was extracted using TRI Reagent (Sigma)
according to the manufacturer's instructions. For quantitative RT-PCR (RT-qPCR), cDNA synthesis was obtained using the ImProm-II Reverse Transcription System (Promega) and RT-qPCR were performed with custom designed oligos (Table 1)

Off-targets.
Employing the free Galaxy web-tool (https://usegalaxy.org/) we generated two datasets: one containing sg19 off-target sequences predicted by CRISPOR web tool (http://crispor.tefor.net) and one containing all the 500 bp genomic regions (NCBI37/mm9) upstream to transcription start sites (TSS) of annotated transcripts. Intersecting the two datasets, all sg19 off-target sequences in putative gene promoters were derived. To identify genes regulated by putative promoters, the sequence of the predicted off-targets was aligned by IGV to the reference genome and to transcripts annotated in ENSEMBL database.
Validation of expression levels of putative off-target genes was performed by RT-qPCR. In vitro and ex-vivo electrophysiology analysis. All the electrophysiology analysis was performed with an automated Python script. Passive properties were calculated from the hyperpolarizing steps of the current clamp steps protocol. Input resistance was averaged from three current steps (2 negative and one positive). Capacitance was calculated from the hyperpolarizing current step as follows. Firstly, the input resistance was determined as the steady-state DV/DI (voltage/current), then the cell time constant (tau) was obtained by fitting the voltage relaxation between the baseline and the hyperpolarizing plateau. Capacitance was then calculated as tau/resistance. Single action potential parameters were calculated as previously described (29). An event was detected as an action potential if it crossed 0mV and if the rising slope was >20mV/ms in a range of injected currents from 0pA to 500pA. All the experiments were performed at room temperature (22-24°C). All recordings and analysis were carried out blind to vector transduced.

RNA-seq
Activity clamp. The template simulating the barrage of synaptic conductances during epileptiform bursts was previously described (19). Dynamic clamp software (Signal 6.0, Cambridge Electronic Design, Cambridge, UK) and a Power3 1401 (CED) were used to inject both excitatory and inhibitory conductance templates simultaneously in a neuron recorded in current clamp configuration (iteration frequency 15 kHz). Erev was set to 0 mV and -75 mV for excitatory and inhibitory conductances respectively, and corrected for a liquid junction potential of 14.9 mV. Incrementing synaptic conductances were injected in recorded neurons to establish the conductance threshold for action potential generation. Current clamp recordings for activity clamp were performed with the same external and internal solutions as given above.
Surgical procedures. All the surgery procedures were performed in adult mice (2-3 months) anesthetized and placed in a stereotaxic frame (Kopf). Exclusion criteria. Only animals recorded for the entire period of the experiment (6 weeks after KA) were used in the analysis. At the end of the experiments some animal tissues were analysed with qRT-PCR and others were verified with immunofluorescence. On total of 24 mice injected with kainic acid, 20 animals (80%) were implanted and injected. 17 were recorded for entire duration of the experiment. 2 did not express dCAS9 and for this reason were excluded from the analysis. 15 mice were used for the analysis (9 Ctrl-dCAS9A and 6 Kcna1-dCAs9A).
EEG (or ECoG) recordings. The ECoG was acquired wirelessly using hardware and software from Open Source Instruments, Inc. The ECoG was sampled at a frequency of 256Hz, band-pass filtered between 1 and 160Hz, and recorded continuously for the duration of the experiments. The animals were housed independently in a Faraday cage. EEG analysis. Spontaneous seizures were detected from chronic recordings using a semiautomated supervised learning approach (suppl. figure 7). First, a library containing examples of epileptiform activity was built using seizures identified from visual inspection of ECoG data. The recordings were saved in hour-long files, and for each seizure this full hour was included in the library. Recordings were chunked into 5 second blocks that were labelled as either "ictal" or "interictal" if they contained epileptiform-labelled activity or not, respectively. For each five second chunk of recording, 15 features were extracted (suppl. unlabelled recordings, the discriminative classifier was first used to predict the class of consecutive five second chunks. We then applied the forward-backward algorithm to obtain the marginal probability of being in seizure state for each recording chunk given the surrounding classifier predictions. The smoothed predictions were then manually verified, false positives removed from the analysis and start and end locations adjusted. In order to quantify the performance of our approach, we randomly selected four 2 week chunks of recordings and visually examined the traces for seizures and compared to classifier predictions (blinded). During the 8 weeks, we did not detect visually any seizures that were not marked by the classifier -as such, for this model of epilepsy, our false negative rate was less than 1/300. False positives were less of a concern, but in general we observed << n seizures for a given period of time. For further information and code, please see: https://github.com/jcornford/pyecog. Video recordings. IP cameras from Microseven (https://www.microseven.com/index.html) were used and synchronised via the Windows time server to the same machine as the ECoG was acquired. Continuous video recordings produced 6 videos/hour.
Statistics. Data are plotted as box and whiskers, representing interquartile range (box), median (horizontal line), and max and min (whiskers), together with all the points. The mean is further shown as "+". The statistical analysis performed is shown in each figure legend.
Deviation from normal distributions was assessed using D'Agostino-Pearson's test, and the F-test was used to compare variances between two sample groups. Student's two-tailed ttest (parametric) or the Mann-Whitney test (non-parametric) were used as appropriate to compare means and medians. Fisher's exact test was used to analyze the contingency  and AP (C) parameters. Student's t test. Only the change in input resistance was significant for Ctrl-dCAS9A compared to Kcna1-dCAS9A neurons.