Inhibition of glutamate decarboxylase (GAD) by ethyl ketopentenoate (EKP) induces treatment-resistant epileptic seizures in zebrafish

Epilepsy is a chronic brain disorder characterized by recurrent seizures due to abnormal, excessive and synchronous neuronal activities in the brain. It affects approximately 65 million people worldwide, one third of which are still estimated to suffer from refractory seizures. Glutamic acid decarboxylase (GAD) that converts glutamate into GABA is a key enzyme in the dynamic regulation of neural network excitability. Importantly, clinical evidence shows that lowered GAD activity is associated with several forms of epilepsy which are often treatment resistant. In the present study, we synthetized and explored the possibility of using ethyl ketopentenoate (EKP), a lipid-permeable GAD-inhibitor, to induce refractory seizures in zebrafish larvae. Our results demonstrate that EKP evoked robust convulsive locomotor activities, excessive epileptiform discharges and upregulated c-fos expression in zebrafish. Moreover, transgenic animals in which neuronal cells express apoaequorin, a Ca2+-sensitive bioluminescent photoprotein, displayed large luminescence signals indicating strong EKP-induced neuronal activation. Molecular docking data indicated that this proconvulsant activity resulted from the direct inhibition of both gad67 and gad65. Limited protective efficacy of tested anti-seizure drugs (ASDs) demonstrated a high level of treatment resistance of EKP-induced seizures. We conclude that the EKP zebrafish model can serve as a high-throughput platform for novel ASDs discovery.

SCIENTIfIC REPoRtS | 7: 7195 | DOI: 10.1038/s41598-017-06294-w VHC-treated larvae swam infrequently in a small dart-like manner and displayed baseline activities. Larvae exposed to EKP exhibited increased activity compared to VHC group in general with three different phases of seizure-related locomotion ( Fig. 2A, statistical analysis see Table S2).
A first phase was characterized by an increasing number of bursts of hyperactivity and agitation. At EKP concentrations ≥300 μM this was followed by a second phase in which hyperactivity reached a plateau. At this point larvae also tended to swim more rapidly in a corkscrew-like manner accompanied with entire body jerking and twitching. In a third phase, the period of increased motility was followed by an irreversible decline of the total larval movement at concentrations between 300-800 μM. This was due to loss of body posture and paralysis (as observed under a microscope), and finally larval death ( Fig. 2A). Onset and duration of each phase were clearly dose-dependent. As 400 μM EKP induced a robust increase of locomotor activity without inducing any larval death during the first 30 min (Fig. 2B,C), this concentration was used for further validation and pharmacological characterization of the model.

Local field potential (LFP) recordings and c-fos expression.
In order to confirm that 400 μM EKP-treated zebrafish larvae displayed abnormal brain activity, LFP recordings were performed in the optic tectum of both VHC-and EKP-treated 7 dpf larvae. 7 out of 35 VHC-treated larvae showed some epileptiform-like activity while multiple epileptiform events were found in all EKP-treated larvae (58/58) (Fig. 3A,B). In EKP-treated animals the mean frequency was 55.76 ± 6.87 events/10 min versus 0.48 ± 0.18 events/10 min in VHC-treated ones (Fig. 3B). Similarly to PTZ treated animals, c-fos expression was significantly increased in EKP-treated larvae compared to VHC-treated ones (Fig. 3C).
Neuroluminescence recordings. Next, neuronal activity was assayed in freely behaving zebrafish using the luminescent approach as reported by Naumann et al. 38 . Transgenic zebrafish Tg(elavl3:eGFP-apoAequorin(GA))expre ssing GFP-apoAequorin under the control of the elavl3 promoter were generated (Fig. 4A) and neuroluminescent signals were recorded for 30 min in VHC-and EKP-treated animals. In the VHC-treated group the averaged aequorin light signal remained relatively stable throughout the experiment. In EKP-treated larvae the averaged light signal was increased and displayed large fluctuations over time (Fig. 4B), resembling the LFP recordings in the same group. Hence, in vivo neuroluminescence clearly allows to detect and study abnormal brain activity occurring in freely behaving animals. SCIENTIfIC REPoRtS | 7: 7195 | DOI:10.1038/s41598-017-06294-w Pharmacology activity. Next, 14 commercially available ASDs were evaluated for their ability to prevent EKP-induced epileptiform activity. Larvae were pre-incubated for 18 hours (h) with the indicated ASD at its maximum tolerated concentration (MTC). Epileptiform activity was scored with behavior tracking, LFP and neuroluminescence recordings.
In case of behavior tracking, an ASD was considered to be active or slightly active if at least two time points or only one time point, respectively, showed a statistically significant decline in total movement in comparison with the EKP control group (Fig. S2) 11 . Based on these criteria, perampanel (PER) and zonisamide (ZSM) were identified as active compounds with PER as the only ASD that significantly prevented EKP-induced locomotion. Phenytoin (PHT) and topiramate (TPM) were identified as slightly active compounds (Fig. 5A).
LFP recordings revealed that ZSM, PER and TMP significantly attenuated the number of epileptiform discharges generated by EKP (Fig. 5B). Other ASDs were not active. For representative LFP recordings of EKP-treated larvae in combination with (or without) ASD pretreatment see Fig. S3.
For the neuroluminescent assay the cumulative amount of photons emitted during the 30-min EKP exposure period was quantified for VHC-and ASD-pretreated animals. ASDs were also tested at their MTC except for ETS and TGB which were evaluated at lower concentrations of 250 µM and 50 µM, respectively, due to toxic side-effects when co-incubated with CLZN-h. As shown in Fig. 5C, EKP-induced neuronal activity was significantly diminished in PHT, RGB, ZSM, PER and TPM pre-treated animals.
Homology modelling and ligand docking. The absence of 14 amino acids in the crystal structure for the human GAD65 (2OKK) is indicative of a flexible region in the protein. Therefore, the presence of a flexible loop was hypothesized by analyzing the superposition between the homology models and the human crystal structures. Final dimeric models after refinement of this flexible loop are shown in Fig. 6A and C for gad67 and gad65 respectively, demonstrating that the two subunits are intertwined as a result of an induced fit of the flexible loop (colored in pink).
For both final gad67 and gad65 dimeric models, the glutamate binding mode was compared to the ones of docked KPA and EKP. The binding affinity and the Ki values calculated by Autodock are reported in Table 1 for each of the complexes. Since values for the monomers are not within the range that is in agreement with observed Figure 2. Behavioral profile of zebrafish larvae exposed to EKP. (A) Locomotor behavior of 7 dpf larvae treated with EKP (200 μM-800 μM) as compared to VHC-treated larvae. The data on the Y-axis refer to the average total movement of larvae per 5-minute (min) intervals. Results shown were pooled from four independent experiments with 12 larvae per experiment that were individually analyzed. Error bars represent s.d. (B) Phase 2 seizure-related behavior observed in zebrafish exposed to EKP. Zebrafish larvae at 7 dpf were incubated with medium containing 200 μM-800 μM EKP for 30 min. The total number of larvae displaying phase 2 behavior was added up per minute interval. Each group represents a total of 12 larvae. (C) Kaplan-Meier cumulative survival curve of 7 dpf larvae during exposure to 200 µM-800 μM EKP or VHC (n = 36 for each group).
in vivo effects, the obtained Ki values confirm the hypothesis of dimerization. A close up of studied inhibitors and the native glutamate in the active site of gad67 and gad65 is shown in Fig. 6B and E respectively, demonstrating that the spatial binding mode of both ligands overlaps with the native binding of glutamate. It also indicates the importance of the carboxyl group (or ester group in EKP) that is engaged in close interactions with GLN84.B, LEU85.B, SER86.B and ARG457.B in both enzymes. The fact that EKP binds with slightly less affinity, could be explained by the presence of an ester group whereas KPA has a carboxyl group that has more favorable electrostatic interactions with ARG457.B.
The importance of the dynamic loop (ALA307.A -PHE354.A) is highlighted by the extra interactions with both inhibitors as shown in Fig. 6C and F. At position 325.A the glycine is in gad67 replaced by cysteine gad65 which displaces TYR356.A out of the active site in gad65. In addition, the active site contains a PHE184.B in gad65 instead of a more polar TYR184.B in gad67. Both of these relatively small differences account for more extensive hydrophobic interactions to occur in the active site of gad65 compared to gad67, that are slightly withdrawing ligands from the crucial ARG457.A towards the hydrophobic part of the active site. In gad67 however, a spatially more favorable interaction is observed between the carboxyl moiety of ligands and ARG457.B, explaining the slightly improved Ki for ligands in the gad67 isozyme compared to gad65. and EKP (n = 58). Statistical analysis was performed using Student's unpaired t-test. Values that were significantly different compared to VHC were indicated with ****p ≤ 0.0001. Error bars on all graphs represent s.d. (C) EKP-induced seizure upregulated c-fos expression in 7 dpf zebrafish larvae as assessed by qPCR (PTZ was used as positive control). c-fos expression is normalized to the housekeeping gene ef1α and β-actin 1. The data were analyzed using the ΔΔCq method and reported as mean ± s.e.m. of three experiments performed in triplicate.

Discussion
In this study we provide evidence that EKP acts as a strong inducer of epileptiform activities in zebrafish larvae. Compared with AG 13 , EKP induced hyperactivity at a 100-fold lower concentration. In addition, hyperactivity occurred more synchronously and with a fast onset. We also found a clear increase of epileptiform discharges in EKP-treated larvae that matches well with the outcome of the locomotor assay and the c-fos expression study.
Moreover, also a neuroluminescent approach was used to investigate the proconvulsant effect of EKP. The genetically modified larvae used to this end specifically express a Ca 2+ sensor that generates bioluminescent signals upon strong neural activation 38 . We found that EKP exerted a dramatic effect on neuronal brain activity, as seen from the emission of large waves of neuroluminescence that could easily be measured and quantified.
From the experimental work it also became clear that the neuroluminescence-dependent assay offers several advantages over LFP recordings: (i) the larvae can freely move and hence there is no need to embed or paralyze them, (ii) recordings do not depend on the correct positioning of a needle that causes inherently interexperimental inconsistencies, (iii) brain activity is assessed as a whole as opposed to electrode recordings that monitor only a very small part of the brain, and (iv) long-term recordings are possible, which might be essential for zebrafish seizure models with infrequent and conditional seizure activities. Furthermore, the method is also compatible with future simultaneous live and video-based recordings of movement and epileptiform activities as an equivalent to clinical video-EEG monitoring.
Protein docking revealed that both KPA and EKP exhibited similar GAD binding affinity as glutamate. The interaction of the carboxyl (in KPA and glutamate) or ester group (in EKP) of these ligands with an arginine residue in the active site contributed strongly to protein binding. The hypothesis that dimers are formed and that a dynamic loop is present correlated with the observed difference in inhibition constants between the monomeric and dimeric models. Two of the mutations in the active site (Phe184 vs Tyr184 and Gly325 vs Ser325) could explain the small difference in affinity of the inhibitors towards gad65 and gad67, however altogether it appears that the proconvulsant activity observed in the different assays can be attributed to direct inhibition of both enzymes.
Further, we evaluated the efficiency of commonly used ASDs representing multiple mechanism of actions (MOA) to reduce the epileptiform zebrafish phenotype induced by EKP as assessed by the three different assay tools. To make a straightforward comparison, a heat map was generated that provides a visual summary of the EKP-related responses upon treatment with different ASDs. Total number of movements, epileptic events and photons emitted after ASDs pretreatment were normalized against the EKP-treated group (Fig. 5D). Overall, EKP-treated larvae responded poorly to most of the tested ASDs suggesting that EKP-induced seizures in zebrafish are treatment-resistant. Despite the fact that EKP is a potent inhibitor of GAD leading to GABA depletion For each experiment a group of three larvae was used. All results (A, B, C) were analyzed using one-way ANOVA. Significant differences compared to EKP group are marked with *, **, ***, **** (p ≤ 0.05, p ≤ 0.01, p ≤ 0.001 and p ≤ 0.0001 respectively). (D) Heat map comparison of larval responses against EKP after different ASD treatment as examined by locomotor tracking, electrographic activity recording and neuroluminescence measurements. Data from ASD-treated larvae were normalized against results of the EKP-treated groups.
SCIENTIfIC REPoRtS | 7: 7195 | DOI:10.1038/s41598-017-06294-w no correlation could be found between the anti-seizure response and a specific MOA of different ASDs. This contrasts with a number of widely used seizure models utilizing GABA A inhibiting compounds (e.g. PTZ, picrotoxin or bicuculline) which are most sensitive to GABAergic ASDs (e.g. barbiturates or benzodiazepines). Consequently, our observations could indicate that the EKP model does not show bias for any particular MOA of existing ASDs and may potentially lead to the discovery of mechanistically novel anti-seizure compounds.
Nevertheless, we observed that three ASDs, namely PER, ZSM and TPM, exhibited consistent reduction in epileptiform events. Of note, TPM as well as PHT were only slightly active in the locomotor activity assay. Additionally, PHT and RGB were found to be active in the neuroluminescence assay, but did not show significant effects in LFP recordings. In fact, certain ASDs may have preferential effects against secondary generalization and spread of seizures, which had been observed in rodent models like the amygdala kindling 39 . The observed difference may also stem from the fact that neuroluminescence monitor whole brain activity, whereas LFP measurements detect signals generated from a small population of optic tectal neurons. As a result, LFP-based assays tend to be less sensitive in detecting signals occurring in brain locations that are distant from the recording spot.
Thus, neuroluminescence offers a window of opportunity to identify compounds that would have passed unnoticed by standard electrophysiological methods. Furthermore, neuroluminescence assay may be able to detect seizure-like activity that is not associated with clear-cut convulsive behaviors in zebrafish, e.g. non-convulsive seizures. Since our study represents the first comprehensive evaluation of ASDs efficacy in a convulsant-induced seizure zebrafish model using a neuroluminescence readout, the model will be further tested with the criteria of reliability and usefulness using other convulsant compounds. This will provide more information about its general applicability and detection rates of false positive and negative compounds. In addition, further developments are possible as it feasible to combine the synchronous recording of bioluminescent signals and locomotor behavior of the freely moving larvae 38 . Moreover, zebrafish expressing genetically encoded calcium indicators (GCaMPs) have successfully been used to monitor in real-time whole brain 3D neuronal activity  in immobilized zebrafish larvae 40 . When compared to the bioluminescent approach that does not provide any spatial information on the emitted light signal, this fluorescence-based method would possibly allow to localize the EKP-induced seizures and discern whether they are focal or not, and investigate further into detail the anti-seizure activity of a few selected compounds of interest. Epilepsy is a very heterogeneous disease and acute preclinical seizure models typically do not mimic any particular epilepsy syndrome, but rather reflect certain seizure types (e.g. MES, PTZ or 6 Hz). Yet, it can be argued that animal models of seizures have shown generally high predictive validity for a therapeutic drug response in patients. Thus, any seizure model does not need to be a perfect replication of the clinical condition, but it is important that the validation provided for a given model is "fit for purpose" 41 .
We have demonstrated for the first time a robust convulsive effect of EKP, a potent GAD inhibitor, in 7 dpf zebrafish larvae. EKP increased locomotor activity, induced epileptiform events and a corresponding increase in neuroluminescence, and increased synaptic activity-regulated c-fos expression.
EKP-induced seizures were similar to those caused by AG, but with much shortened seizure latency and a more synchronized seizure onset. However, it remains to be established if this model, like many other acute seizure models in both zebrafish and rodents, represent any particular epilepsy syndrome despite strong evidence that GAD has been implicated in pathophysiology of epilepsy. Our findings support the suggestion that the EKP zebrafish model is characterized by poor response to several existing ASDs and may become a useful addition to the armamentarium of animal models of drug resistant seizures. In fact, no single model has been validated for use to identify potential compounds effective against drug resistant seizures. Consequently, it is suggested that a battery of such models should be employed, thus enhancing the sensitivity to discover novel, highly effective ASDs 42 .
In conclusion, the zebrafish EKP model can therefore be used for screening of drug-like hits in an early step of the discovery of mechanistically novel ASD candidates. Furthermore, it is anticipated that by providing quantifiable whole brain-related Ca 2+ signals in freely moving transgenic animals, the neuroluminescence technique might turn out to be of particular interest for future drug discovery strategies in the field of epilepsy and epileptogenesis.

Materials and Methods
Animals and maintenance. Adult zebrafish (Danio rerio) were maintained in a UV-sterilized rack recirculating system equipped with a mechanical and biological filtration unit and kept under a 14/10 hour light/dark cycle at the temperature of 27-28 °C and pH of 6.8-7.5. Water quality was monitored for pH, temperature, conductivity, ammonia, nitrite (SL1000 Portable Parallel Analyzer, Hach Instruments, USA) and nitrate levels (Tetra, Melle, Germany). Zebrafish were fed three times per day with flake food (TetraMin, Tetra, Germany) and Artemia (brine shrimp). Embryos were obtained via natural spawning, then sorted and kept in petri dishes (92 × 16 mm, Sarstedt, Nümbrecht, Germany) at 28  GlaxoSmithKline) and coelenterazine-h (CLZN-h) (NanoLight ® Technologies). All compounds were dissolved in dimethyl sulfoxide (DMSO) and kept as a stock at −20 °C or −80 °C. Stock solutions were further diluted in embryo medium for locomotor activity assay and local field potential recordings, or E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl 2 and 0.33 mM MgSO 4 ) for neuroluminescence recording to achieve a final DMSO concentration of 1% w/v. As vehicle control (VHC) 1% w/v DMSO in embryo or E3 medium was used.

Generation of Tg(elavl3:eGFP-apoAequorin) line. A transgenic fish line Tg(elavl3:eGFP-apoAequorin)
was generated with the pan-neuronal elavl3 promoter that drives the expression of a fusion of eGFP and apoAequorin. Briefly, the elavl3 promoter (a gift from Dr. Florian Engert, Harvard, USA) and the coding sequence for the eGFP-apoAequorin fusion protein 43 (a gift from Dr. Ludovic Tricoire, Université Pierre/Marie Curie, France) were cloned into a gateway expression vector flanked with tol2 recognition sites. To generate stable transgenic zebrafish, 20 pg of the elavl3:eGFP-apoAequorin plasmid DNA was co-injected with 100 pg tol2 transposase mRNA into the cytoplasm of single cell stage fertilized nacre zebrafish embryos. Injected embryos were screened for eGFP expression at 2-5 dpf. eGFP positive individuals (F0) were grown to adulthood and out-crossed with nacre zebrafish. All the experiments with were performed with progeny (F3) from intercrossing stable F2 transgenics.
Analysis of gad1b and gad2 expression using qPCR. A. RNA extraction and cDNA synthesis: Total RNA was extracted from 12 embryos on 1 to 7 dpf using TRIzol (Life Technologies) according to manufacturer's SCIENTIfIC REPoRtS | 7: 7195 | DOI:10.1038/s41598-017-06294-w instructions. Following DNase (Roche) treatment, RNA was quantified by NanoPhotometer P330 (Implen). cDNA was synthesized from 1 μg of total RNA using random primers and SuperScript III reverse transcriptase (Invitrogen) according to the manufacturer's instructions and further diluted 1:20.
B. qPCR analysis: For analysis of gad1b and gad2 mRNA expression, 4 μl of each cDNA was amplified with Bio-Rad 20x PrimePCR assays (gad1b assay ID: qDreCID0019786; gad2 assay ID: qDreCID0014650) and 2x SsoAdvanced Universal SYBR Green Supermix (Bio-Rad). H 2 O was used as none template control (NTC). qPCR reactions were performed with HardShell ® Low-Profile Thin-Wall 96-Well Skirted PCR Plates (Bio-Rad) on CFX96 Touch Real-Time PCR Detection System (Bio-Rad) under cycling conditions according to the manufacturer's protocol. Data generated by real-time PCR were compiled using CFX Manager Software (Bio-Rad). The gad1b and gad2 transcripts were normalized against ef1α and 18 s (primers: Table S1). The relative expression levels were quantified using the comparative Cq method (ΔΔCq). Amplification specificity was monitored by examining the final melting curve.
Validation of larval zebrafish EKP seizure model. Locomotion activity. Larvae (6 dpf) in 100 µl VHC were arrayed individually in a 96-well plate (tissue culture plate, flat bottom, Falcon, USA) and kept for 18 h in the dark at 28 °C. Prior to tracking the following day (7 dpf) 100 μl of VHC or EKP stock solution was added to each well to obtain a range of EKP concentrations (200 μM-800 μM). The plates were placed in an automated video tracking device (ZebraBox TM apparatus; Viewpoint, Lyon, France) and 5 min later the locomotor behavior of the larvae was monitored for 2 h in the dark at 28 °C. Locomotor activity was quantified using ZebraLab ™ software (Viewpoint, Lyon, France) and expressed in "actinteg" units per 5-min interval. The actinteg value is defined as the sum of all image pixel changes detected during the time window.
Video recordings of individual wells were used between 0-30 min to assess the cumulative number of larvae exposed to the different EKP concentrations that had exhibited phase 2 seizures. These seizures are defined as rapid corkscrew-like accompanied with entire body jerking and twitching, which are easily distinguished from other behavioral abnormalities (for description of phase 1 and 3 seizures, see results).
Survival rate. Larvae (7 dpf) were arrayed individually in a 96-well plate, exposed to EKP (200 μM-800 μM) or VHC (control) and immediately scored under a stereomicroscope for lethality every 5 min for 1 h.
Local field potential (LFP) recordings. Larvae (6 dpf) in 100 µl VHC were arrayed individually in a 96-well plate and kept for 18 h in the dark at 28 °C. An equal volume of 800 μM EKP (2x solution) or VHC (control) was added to each well and hence larvae were incubated with 400 μM EKP for 15 min. Next, the treated larva was embedded ventral side down in 2% low melting point agarose bathed in artificial cerebrospinal fluid (ACSF, 124 mM NaCl, 2 mM KCl, 2 mM MgSO 4 , 2 mM CaCl 2 , 1.25 mM KH 2 PO 4 , 26 mM NaHCO 3 , and 10 mM glucose). A blunt glass electrode (soda lime glass, Hilgenberg, Germany) was pulled with DMZ Universal Puller (Zeitz, Germany) to an opening of 15-20 microns, filled with ACSF and placed on the skin above the optic tectum. The recordings were performed using WinEDR (John Dempster, University of Strathclyde, UK). Differential signal was band pass filtered at 0.3-300 Hz and digitized at 2 kHz via a PCI-6251 interface (National Instruments, UK). LFP recordings started each time exactly 2 min after the removal of the larva from the EKP solution (or VHC) and were continued for 10 min at room temperature (24 °C).
EKP-induced events were considered as epileptiform activity when their signal exceeded three times the baseline and lasted for minimum 100 msec, as described previously 44 . Electrophysiological recordings were analyzed in a blinded way using Clampfit 10.2 software (Molecular Devices, USA).
Analysis of c-fos expression in larval heads using qPCR. To measure the seizure-related changes in c-fos mRNA expression in the CNS, 12 larvae (7 dpf) were treated with either 20 mM PTZ, 400 μM EKP or VHC for 15 min and then rapidly decapitated. The heads were immediately processed for RNA isolation followed by cDNA synthesis as described for gad1b and gad2. qPCR analysis of c-fos mRNA expression using Bio-Rad 20x PrimePCR assays (c-fos assay ID: qDreCED0015103) was performed as described for gad1b and gad2, and normalized to the housekeeping gene ef1α and β-actin 1 (primers , Table S1).
Neuroluminescence recording. Tg(elavl3:eGFP-apoAequorin) zebrafish larvae were visually screened for even GFP expression at 3 dpf. Selected larvae were incubated with VHC in 40 µM CLZN-h for 18-24 h in the dark at 28 °C 38 . Excess of CLZN-h was removed by washing larvae repeatedly in E3 medium afterwards. A group of 3 larvae was then transferred into a light-tight thermostated perfusion chamber (27-28 °C) containing either VHC or 400 µM EKP. The recordings started each time exactly 5 min after EKP exposure and were continued for 30 min. Photons emitted were detected by a photon-counting tube (Type H3460-04, Hamamatsu Photonics, Japan) that was positioned about 2 cm above the larvae. Light impulses were discriminated, prescaled and counted with a PC-based 32-bit counter/timer board (PCI-6601, National Instruments Corporation, Austin, TX, USA). The number of impulses occurring during a 5-sec time interval was monitored with custom-built software. Analysis of the bioluminescence recordings was done in a blinded way.
Pharmacology of larval zebrafish EKP seizure model. A schematic comparison of timelines of the different experiments is depicted in Fig. 7.
Maximum tolerated concentration (MTC). Before performing pharmacological experiments, MTC of each ASD was determined as described previously 11 .
SCIENTIfIC REPoRtS | 7: 7195 | DOI:10.1038/s41598-017-06294-w Effect of ASDs on EKP-induced locomotion activity. Larvae (6 dpf) were exposed to VHC or ASDs at their respective MTC for 18 h in the dark at 28 °C. Then, the larvae were treated with 400 µM EKP (or VHC) and the locomotion activity was monitored for 30 min (after 5 min of habituation) as described earlier. The average amount of movement per 5-min intervals was measured, quantified and normalized to the EKP control group for the entire 30-min tracking session or per 5-min interval.
Effect of ASDs on EKP-induced epileptiform discharges. Larvae (6 dpf) were exposed to VHC or ASDs at their respective MTC for 18 h in the dark at 28 °C. Then, the larvae were treated with 400 µM EKP (or VHC), embedded in agarose, and the LFP recordings were performed as described earlier.
Effect of ASDs on EKP-induced neuroluminescent events. eGFP-positive Tg(elavl3:GA) zebrafish larvae (6 dpf) were exposed to VHC or ASDs at their respective MTC (as determined in E3-CLZN solution) and 40 µM-h CLZN for 18 h in the dark at 28 °C. Larvae were then washed repeatedly in E3 medium and groups of three larvae were transferred into a light-tight thermostated perfusion chamber (27-28 °C) containing either VHC or 400 µM EKP. Recordings were performed as described earlier.
Protein ligand docking. Using i-Tasser 45 , homology models for gad65 and gad67 were created for Uniprot sequences F1R9E8 and Q7ZUS3 and pdb structures 2OKK and 2OKJ as templates respectively 46 . Cα traces in dimer models are constructed with COTH 47 starting from 2OKJ using 484 residues in each of the subunits (first 97 residues in gad65 and 101 in gad67 are omitted). For the ease of comparison, residue numbering in both models starts at 1. The SABBAC server 48 and AMBER software 49 were used for backbone reconstruction and final energy minimization respectively. Loop remodeling was performed with the loop refinement tool in USCF Chimera 50 . Finally, a short molecular dynamics simulation of 20 nanoseconds (nsec) was performed. Ramachandran plots were hereafter generated to verify the quality of the homology model 51 (Fig. S1).
Flexible docking was performed using Autodock 52 in one of the active sites of the dimeric homology models. A grid unit of 0.375 A was used with a 3-dimensional box having sides of 40 units. The grid box center for gad67 and gad65 was chosen to be (11.693, 17. Within each gad, the conformations were superimposed to visualize differences in binding mode. Analysis of hydrogen bonds and hydrophobic interactions was performed by USCF Chimera and LigPlot+ 53 . N6-(pyridoxal phosphate) lysine was parametrized based on empirical data for all docking and molecular dynamics simulations. Data analysis. Statistical analysis was performed using one-way or two-way ANOVA followed by Dunnett's multiple comparison test with GraphPad Prism Version 7.0c (San Diego, CA). Data availability. All data supporting this work are included in this manuscript.