Despite the introduction of more than one dozen new antiepileptic drugs in the past 20 years, approximately one-third of people who develop epilepsy continue to have seizures on mono- or polytherapy1. Viral-vector-mediated gene transfer offers the opportunity to design a rational treatment that builds on mechanistic understanding of seizure generation and that can be targeted to specific neuronal populations in epileptogenic foci2. Several such strategies have shown encouraging results in different animal models, although clinical translation is limited by possible effects on circuits underlying cognitive, mnemonic, sensory or motor function. Here, we describe an autoregulatory antiepileptic gene therapy, which relies on neuronal inhibition in response to elevations in extracellular glutamate. It is effective in a rodent model of focal epilepsy and is well tolerated, thus lowering the barrier to clinical translation.
Approximately 70 million people worldwide are affected by epilepsy, of whom approximately 30% continue to have seizures despite optimal medical treatment3,4. Antiepileptic drugs have a narrow therapeutic window, mainly because they do not differentiate between neurons involved in seizure generation and those underlying normal brain function5. The most effective treatment option for refractory focal-onset epilepsy is surgical resection, but this treatment is restricted to cases in which the epileptogenic zone is relatively far from eloquent cortex6. Gene therapy to decrease neuronal excitability has shown promise in preclinical models but is also irreversible, thus limiting clinical translation. On-demand gene therapy with optogenetics7 or chemogenetics8 can address this issue but faces additional translational obstacles because of the need for continuous electroencephalogram monitoring and devices for light or ligand delivery. Efficient algorithms for seizure forecasting are available but are associated with an appreciable rate of false positives9,10. Furthermore, both electroencephalogram monitoring and light or ligand delivery require implanted devices, which are associated with surgical risks, postoperative complications, finite lifetime and interference with magnetic resonance imaging. Chemogenetics can potentially be used on a slower timescale, with exogenous ligand delivery by an oral or parenteral route, but this, too, may interfere with normal brain function for the duration of the therapeutic effect. We therefore aimed to develop a molecular tool to inhibit neurons in response to pathological accumulation of extracellular glutamate, a hallmark of excessive synchronous discharges of excitatory neurons in seizures11,12,13,14,15 (Fig. 1a).
For this purpose, we designed a viral plasmid bearing a fully codon-optimized sequence encoding a glutamate-gated Cl– channel (GluCl from Caenorhabditis elegans), in which the α subunit is linked via a self-cleaving viral 2A-peptide to the GluCl β subunit, with mYFP in the M3–M4 loops of both subunits16. The leucine residue at position 9 of the M2 pore-forming domain of the wild-type GluCl α subunit was mutated to phenylalanine to enhance the channel’s glutamate sensitivity16 (product denoted enhanced GluCl (eGluCl)). To bias expression of eGluCl to excitatory neurons, eGluCl was placed under the control of the human CaMKIIα promoter (hCamKII)17 (Fig. 1a). Subsequent full electrophysiological characterization in Neuro-2a cells confirmed that the half-maximal effective concentration (EC50) for glutamate decreased from 262 ± 53 μM (wild-type GluCl, n = 8 cells, mean ± s.e.m.) to 12 ± 3 μM (eGluCl, n = 8 cells) (P < 0.001, Student’s t test) (Fig. 1b), values comparable to the extracellular glutamate levels detected during seizures13,15.
After eGluCl lentivector packaging (titer 2.15 × 109 TU/ml), we assessed its transduction efficiency in vivo. We therefore injected adult male Sprague Dawley rats with either eGluCl or GFP control lentivector into the primary motor cortex (M1), then assessed expression after 7 d. eGluCl was detected mainly at extrasynaptic locations (Supplementary Fig. 1), as estimated by very limited colocalization with the synaptic marker PSD95: 0.29 ± 0.01% of eGluCl occurred at PSD95 puncta, whereas 13.50 ± 2.49% of PSD95 colocalized with eGluCl in transduced areas of the brain (Pearson’s correlation coefficient: 0.19 ± 0.03, n = 3 independent animal brain preparations) (Fig. 1c). Given that glutamate is efficiently cleared from the extrasynaptic space by excitatory amino acid transporters14,18,19, these results suggest that eGluCl should have a minimal effect on normal brain function. eGluCl was expressed at the earliest time point evaluated (day 3 postinjection) and persisted for at least 245 d, in agreement with results for other lentivectors20 (Fig. 1d).
We assessed the ability of eGluCl to attenuate seizures evoked by acute intracortical administration of the chemoconvulsant pilocarpine8. For this purpose, we implanted adult male Sprague Dawley rats with a wireless transmitter, with an electrode in M1, and secured a cannula for access to the same brain region. One week after surgery, we injected pilocarpine via the cannula and quantified the evoked seizure severity. Subsequently, the animals were treated with an equivalent dose of either eGluCl or a control lentivector expressing green fluorescent protein, hCaMKII-GFP (denoted GFP). The chemoconvulsant injection was then repeated in a randomized and blinded study comparing the effect of pilocarpine before (pre-Rx) and 14 d after (post-Rx) lentiviral treatment with either lentivector into the same region of the rat motor cortex (Fig. 2a). Animals treated with eGluCl exhibited a decrease in the electrocorticogram (ECoG) coastline, an aggregate of frequency and amplitude, between the two pilocarpine injections (absolute change in coastline: –1.08 ± 0.58 V at 20 min, n = 6 animals). In contrast, GFP-treated animals exhibited an increase in the coastline between the two pilocarpine injections (1.04 ± 0.34 V, n = 7 animals; comparison between eGluCl and GFP, P = 0.007, Student’s t test; Fig. 2b). A comparison of the number of large-amplitude ECoG spikes (>25% of the maximum amplitude evoked by pilocarpine) also revealed a pronounced effect of eGluCl (ratio of spikes in the second trial to the first trial: eGluCl, 0.60 ± 0.12; GFP, 1.27 ± 0.18; P = 0.012). Similarly, the ECoG power in the 4- to 14-Hz frequency band, which correlates with motor convulsion severity8 (Supplementary Video 1), decreased in eGluCl-treated animals (ratio of second trial to first: 0.44 ± 0.19) but increased in GFP-treated animals (1.47 ± 0.16; P = 0.002). Finally, the spike frequency decreased after eGluCl treatment (ratio: 0.65 ± 0.13) in contrast to that in GFP-treated animals (ratio: 1.29 ± 0.14; P = 0.007). The total duration of pilocarpine-evoked electrographic seizure activity, which is mainly determined by the clearance of the chemoconvulsant from the brain, showed a nonsignificant trend toward a decrease in eGluCl-treated animals (ratio: 0.79 ± 0.11) compared with GFP-treated animals (1.02 ± 0.10; P = 0.146) (Fig. 2c and Supplementary Fig. 2). eGluCl is thus effective at attenuating acute chemoconvulsant-evoked seizures.
We next determined the effect of eGluCl in a model of chronic focal neocortical epilepsy induced by tetanus-toxin injection into the visual cortex21. In this model, spontaneous seizures occur over several weeks (Fig. 3a,b, Supplementary Video 2 and Supplementary Fig. 3a). eGluCl or GFP lentivectors were injected 11 d after tetanus-toxin injection in a randomized and blinded study design. The number of seizures before treatment did not differ between the groups (eGluCl, 10.1 ± 1.5, n = 10 animals; GFP, 10.4 ± 1.2, n = 10 animals; P = 0.761, Student’s t test) (Fig. 3d). eGluCl significantly decreased the number of subsequent seizures in comparison to that in GFP-treated animals (generalized log linear mixed model: F(1, 59) = 20.66, treatment effect P < 0.001; F(8, 59) = 17.28, treatment × week interaction effect P < 0.001) (Fig. 3c). The cumulative number of seizures normalized to the number of pre-Rx seizures was also significantly decreased (eGluCl, 2.62 ± 0.26, n = 10; GFP, 3.79 ± 0.24, n = 10; P = 0.004, Student’s t test) (Fig. 3d), as was the absolute number of seizures experienced per animal post-Rx (eGluCl, 16.2 ± 2.9, n = 10 animals; GFP, 28.9 ± 4.4, n = 10 animals; P = 0.034, Mann–Whitney test) (Supplementary Fig. 3b). We did not, however, observe a significant difference in average seizure duration (eGluCl, 85.6 ± 53 s, n = 9 animals; GFP, 94.3 ± 4.7 s, n = 10 animals; P = 0.278, Mann–Whitney test; one eGluCl-treated animal did not experience any seizures post-Rx) or severity (Supplementary Fig. 3b,c), in agreement with evidence that seizure duration is mainly determined by the extent of the network involved rather than by activity at the focus22.
We examined whether eGluCl treatment might affect interictal activity by comparing the absolute ECoG coastline after discarding all seizures. The results revealed a steep increase after tetanus-toxin injection, and no difference between eGluCl- and GFP-treated animals (Fig. 4a). eGluCl is thus effective in decreasing the frequency of spontaneous seizures without altering the baseline ECoG.
To examine off-target effects of eGluCl on normal brain function, we injected another group of nonepileptic animals with either lentivector and determined whether the ECoG was affected. The average of 6-h ECoG coastline (0:00–6:00) 2 weeks post-Rx, expressed as a ratio relative to pre-Rx ECoG, did not differ between treatment groups (ratio: eGluCl, 1.01 ± 0.03, n = 7 animals; GFP, 1.06 ± 0.03, n = 6 animals; P = 0.293, Student’s t test). As a positive control, in an additional group of animals, we administered ivermectin (IVM, 5 mg/kg IP), an activator of GluCl23. This treatment significantly reduced the coastline in animals treated with eGluCl compared with GFP (ratio: eGluCl, 0.93 ± 0.04, n = 6 animals; GFP, 1.10 ± 0.03, n = 5 animals; P = 0.010, Student’s t test) (Fig. 4b).
To look for behavioral effects of eGluCl, we assessed performance in two tests of motor coordination sensitive to motor cortex lesions, again blind to treatment, after injection of eGluCl or GFP lentivector into layer 5 of the motor cortex. No significant difference was seen in the accelerating rotarod test between groups (average latency to fall in three rotarod sessions: eGluCl, 84.8 ± 12.0 s, n = 5 animals; GFP, 75.4 ± 15.0 s, n = 6 animals; P = 0.646, Student’s t test). We also observed no significant difference in the average number of steps taken while the same animals walked on an elevated grid during three sessions (eGluCl, 83.5 ± 12.7; GFP, 95.3 ± 17.5; P = 0.612, Student’s t test). Finally, we repeated the evaluation of motor coordination in the same animals after IVM. The results revealed a robust difference between eGluCl and GFP when the latency to fall post-IVM was expressed as a ratio of the earlier test sessions pre-IVM (eGluCl, 1.36 ± 0.12, n = 5 animals; GFP, 0.96 ± 0.05, n = 6 animals; P = 0.009, Student’s t test; Fig. 4c). Although unexpected, a similar paradoxical increase in latency to fall has previously been reported with M1 lesions24. IVM also led to a marked decrease in the number of steps taken on the elevated grid by eGluCl-treated animals but not by GFP-treated animals (expressed as a ratio of steps taken pre-IVM, eGluCl, 0.76 ± 0.09, n = 5 animals; GFP, 1.04 ± 0.08, n = 6 animals; P = 0.045; Fig. 4d). In a separate cohort of animals that did not receive IVM, we observed no significant difference in either test up to 25 d post-Rx (rotarod: two-way analysis of variance (ANOVA), F(1, 56) = 0.72; steps: two-way ANOVA, F(1, 56) = 0.37; eGluCl, n = 5 animals; GFP, n = 4 animals; Supplementary Fig. 4). We therefore conclude that eGluCl has no major effect on normal brain function, although an effect can be revealed by direct activation of the receptor by IVM, thus implying that the tests of motor coordination were sensitive to an increase in chloride conductance in transduced neurons.
Together, these results demonstrate that gene therapy with eGluCl is well tolerated and effective. Extrasynaptic glutamate is normally clamped to submicromolar concentrations by active transport19. Glutamate transporters nevertheless have a finite capacity to prevent extrasynaptic escape of the neurotransmitter25. eGluCl channels therefore open only in response to elevated extrasynaptic glutamate, as occurs during seizures26. Chloride-permeable channels inhibit both by hyperpolarizing neurons and by decreasing the effective membrane resistance (voltage and shunting inhibition, respectively). A collapse of transmembrane chloride gradients has been reported in seizure models27, thus compromising voltage inhibition, but the robust antiepileptic effect of eGluCl suggests that shunting inhibition persists. This mechanism of action is shared with benzodiazepines and barbiturates, although these drugs potentiate the action of GABAA receptors indiscriminately, thereby contributing to their narrow therapeutic window. Gene therapy with eGluCl thus represents a form of biochemical closed-loop chemogenetic treatment that does not require an exogenous agonist, and it is therefore a promising approach to treat refractory epilepsy in which the seizure focus is close to eloquent cortex.
Lentiviral transfer plasmids were constructed with standard molecular cloning techniques. Wild-type GluCl (GluCl α subunit, accession number G5EBR3; β subunit, accession number Q17328) including mutation RSR → AAA in the endoplasmic reticulum–retention motif16 was fully codon optimized for human expression and synthesized by Genscript. The L9′F mutation was inserted with a QuikChange II kit (Agilent). All plasmids were verified by sequencing before use (Source Bioscience). Sequences are available upon request. Lentivectors were produced by Cyagen Biosciences.
Voltage clamp recordings
To record the conductance of GluCl Cl– under the control of the human CamKII-α promoter, we used a mouse neuroblastoma cell line (Neuro-2a). Heterologous expression of GluCl or eGluCl was obtained with TurboFect transfection reagent (Thermo Fisher Scientific). Whole-cell patch-clamp recordings were performed with borosilicate-glass electrodes, which were pulled on a micropipette puller (Sutter Instruments), fire polished (Narishige) and had a final resistance of 2–3.5 MΩ. Data were filtered at 1 kHz, digitized at 5 kHz and recorded with WinEDR software (John Dempster, Glasgow, UK) and an Axopatch 1-D amplifier (Axon Instruments). A series resistance compensation of 60% was used throughout. Cells were held at –60 mV, and increasing concentrations of glutamate were applied through a custom-built single-cell perfusion system. All recordings were performed at room temperature. The extracellular recording solution contained 150 mM NaCl, 2.8 mM KCl, 2 mM CaCl2 and 10 mM HEPES, with the pH adjusted to 7.35 with NaOH, and with glutamate as indicated, and the intracellular recording solution contained 135 mM CsCl, 10 mM Cs-EGTA, 10 mM HEPES, 1 mM MgCl2 and 4 mM Na2-ATP, with the pH adjusted to 7.35 with CsOH.
All experiments were performed in accordance with the United Kingdom Animal (Scientific Procedures) Act of 1986 and were approved by the Home Office (license PPL70-7684). All animals were kept under a 12-h dark/light cycle (dark, 19:00–7:00) at a constant temperature of 21 °C and a humidity of 50%. Male Sprague Dawley rats (275–350 g) were anesthetized with isoflurane and placed in a stereotactic frame (Kopf). A cannula (Plastics One) and an ECoG electrode (Open Source Instruments) were placed in either the forelimb area of the right primary motor cortex (coordinates, 2.4 mm lateral, 1.0 mm anterior of bregma) or the right visual cortex (coordinates, 3.0 mm lateral, 7 mm posterior of bregma). An ECoG transmitter (A3028E, Open Source Instruments) was implanted subcutaneously to allow for wireless telemetry recordings. The reference electrode was implanted in the contralateral hemisphere. For the chronic visual cortex tetanus-toxin model, 15 ng of tetanus toxin was injected in a volume of 1.0 μl PBS at a rate of 200 nl min−1 in the layer 5 region. The Hamilton syringe (needle gauge 33) was held in place for 10 min after the injection.
Chemoconvulsant seizure model
After a 1-week recovery period from surgery, animals were injected on four consecutive days with increasing amounts (200, 450, 700 and 950 nl) of 3.5 mM pilocarpine solution in PBS into layer 5 of the right primary motor cortex at a rate of 100 nl min−1 (the Hamilton syringe was held in place for 2 min postinjection). Three days later, all animals were injected with 2.0 μl of lentivector at the same location and a rate of 200 nl s−1 (1.0 μl at –1.1 mm; 1.0 μl at –1.0 mm from pia; the Hamilton syringe was held in place for 5 min and 10 min, respectively, after virus injection). After waiting 2 weeks for transgene expression, we repeated the chemoconvulsant injection procedure. This experimental design allowed us to normalize every pilocarpine-evoked seizure to its corresponding pretreatment seizure. To minimize interanimal susceptibilities to different doses of pilocarpine, we averaged the normalized values across the different doses, rejecting any dose that was below the threshold for seizure induction. Virus injection and subsequent data analysis were performed by a researcher blinded to treatment.
Chronic tetanus-toxin epilepsy model
Eleven days after ECoG transmitter and tetanus-toxin injection, animals were injected with 2.0 μl of lentivector at a speed of 200 nl s−1 (1.0 μl at –1.1 mm; 1.0 μl at –1.0 mm from the pia; the Hamilton syringe was held in place for 5 min and 10 min, respectively, after virus injection) into layer 5 of the right visual cortex. Virus injection and subsequent data analysis were performed by a researcher blinded to treatment.
ECoG acquisition and analysis
The ECoG was acquired wirelessly with hardware and software from Open Source Instruments. The ECoG was sampled at a frequency of 512 Hz, band-pass filtered between 1 and 160 Hz, and recorded continuously for the duration of the experiments.
ECoG analysis in the pilocarpine-induced acute model of seizures was performed as previously described8. Briefly, the coastline, the number of spikes reaching at least 25% of the maximal amplitude recorded within the seizure, power in the 4- to 14-Hz band, seizure duration and interspike interval were calculated from raw traces imported into Python 3.5 (Supplementary Video 1). Values obtained for each dose of pilocarpine after lentivector treatment were compared to the corresponding values obtained for the same dose of pilocarpine before lentivector injection within each animal. The coastline analysis is shown as absolute difference, but other measures are shown as ratios for clarity.
The analysis of the chronic tetanus-toxin-induced model was performed as previously described7,8. Briefly, the ECoG was segmented into consecutive 1-s epochs, and six different metrics were calculated (coastline, power, intermittency, coherence, asymmetry and power between 20 and 50 Hz) (Supplementary Video 2). Each metric was mapped onto the interval 0–1, thus yielding a point in a six-dimensional hypercube of unit side. Each epoch was compared to a user-generated seizure library consisting of seizures validated from at least three different animals. If the Euclidean distance between the coordinates for the epoch fell within 0.2 of a validated seizure epoch, it was classed as a possible seizure. After successful detection of at least three seizures per animal, animal-specific seizure libraries were generated. Each seizure that was not qualified by at least six potential seizure-events was added to the animal-specific seizure library. For Fig. 3, the number of seizures in each animal was normalized to the number of seizures occuring in the week prior to treatment, to compensate for interanimal variability. Raw, non-normalized data are shown in Supplementary Fig. 3. For all datasets, the minimal duration to define a seizure was set to 10 s.
Rats injected with either eGluCl or GFP lentivector into the right motor cortex (long-term study), or both hemispheres (study involving IVM as a positive control) (M1 region layer 5, volume and coordinates as above) were assessed with the rotarod and elevated grid tests of motor coordination.
For the rotarod test (Ugo Basile), the rotation velocity increased from 3 to 30 r.p.m. over 5 min. The animals were habituated for two training sessions on the rotarod, each of which consisted of three trials. In the test sessions, the latency to fall was recorded for three consecutive trials, and the best performance was used for statistical analysis. For the elevated grid test, rats were placed on an elevated horizontal platform of 52 × 32 cm, consisting of a square painted steel wire array (4-mm diameter; 4-cm spacing) and were allowed to explore the arena for 2 min. All animals were allowed to habituate to the test environment for two consecutive days before the test sessions. Rats were video monitored from two different angles, and the number of steps was counted.
One group of animals was assessed on three consecutive days in both tests before lentivector injection, and then in eight subsequent sessions over the next 4 weeks. Another group of animals was treated with eGluCl or GFP lentivector 7 d before two habituation sessions on consecutive days, followed by three test sessions and three further sessions after IVM injection (5 mg/kg, each given 24 h before testing). All experiments and subsequent data analysis were performed by a researcher blinded to the treatment.
Staining was performed on free-floating 30- or 50-µm rat brain sections with the following antibodies: mouse anti-PSD95 (ab2723, Abcam), rabbit anti-GFP (ab6556, Abcam), guinea pig anti-MAP2 (188 004, Synaptic Systems), rabbit anti-neurofilament heavy polypeptide (ab8135, Abcam), Alexa Fluor 488 donkey anti-rabbit (A-21206, Thermo Fisher Scientific), Alexa Fluor 568 donkey anti-rabbit (A10042, Thermo Fisher Scientific), Alexa Fluor 405 goat anti-guinea pig (ab175678, Abcam) and CF568 donkey anti-mouse (20105, Biotium). Images were acquired with ZEN software (Zeiss) on an LSM710 confocal microscope (Zeiss). Colocalization of PSD95 and eGluCl analysis was performed via ImageJ 1.51n (Wayne Rasband, National Institutes of Health) plugin JACoP28.
All statistical analysis was performed with IBM SPSS 126.96.36.199 or Graph Pad Prism 5.01. Statistical significance was tested with Student’s two-tailed paired or unpaired t tests, two-tailed Mann–Whitney tests, repeated-measures ANOVA, or a generalized log linear mixed model with random effect of animal (autoregressive covariance) and fixed effect of treatment group, week and the interaction of treatment group and week. All data are shown as mean ± s.e.m., with data for individual animals also shown. The choice of parametric or nonparametric test followed a Kolmogorov–Smirnov test with the Dallal–Wilkinson–Lilliefor-corrected P value. The significance level was set to an α-error of P < 0.05.
Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.
Code is available upon request.
Data are available upon request.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank G. Schiavo (UCL Institute of Neurology) for the gift of tetanus toxin and S. Hart (UCL Institute of Child Health) for the mouse Neuro-2a cell line. We are grateful to J. Cornford for assistance with ECoG analysis and to K. Hashemi for help optimizing wireless-transmitter use. This project was supported by the European Union’s Horizon 2020 research and innovation program (Marie Skłodowska-Curie grant agreement no. 701411 to A.L.); the Medical Research Council (MR/L01095X/1 to D.M.K., S.S. and M.C.W.); and the Wellcome Trust (095580/Z/11/Z to D.M.K.; 104033/Z/14/Z to D.M.K. and S.S.).