Large-Scale Phenotype-Based Antiepileptic Drug Screening in a Zebrafish Model of Dravet Syndrome

1Dravet syndrome is a catastrophic childhood epilepsy that is resistant to available medications. Current animal models for this disease are not amenable to high-throughput drug screening. We used a zebrafish model for Dravet syndrome and screened >1000 compounds. We report the identification of compounds with the ability to suppress seizure behavior and electrographic seizure activity. This approach provides an example of precision medicine directed to pediatric epilepsy.

2Dravet syndrome (DS) is a devastating genetic epileptic encephalopathy that has been linked to more than >300 de novo mutations in a neuronal voltage-gated sodium channel (SCN). Children with DS are at a higher risk for sudden unexplained death in epilepsy and episodes of uncontrolled status epilepticus. Delayed language development, disruption of autonomic function, and motor and cognitive impairment are also associated with this disease. Seizure management includes treatment with benzodiazepines, valproate, and/or stiripentol. Some reduction in seizure activity has been reported with the use of bromides and topiramate, or a ketogenic diet. Despite these options, available antiepileptic drugs (AEDs) do not achieve adequate seizure control in most DS patients, making the identification of new drugs a critical unmet need. High-throughput screening offers a powerful tool to identify new drug candidates for these patients. However, commonly available screening approaches rely on in vitro cell-based assays and do not recapitulate the complicated neural networks that generate seizures in vivo. Given the need for new treatments for children with DS, and the growing number of genetic epileptic encephalopathies that are medically intractable, we developed an alternative phenotype-based in vivo drug-screening strategy. While cell-based in vitro screening assays can efficiently identify compounds that bind specific targets, whole-organism-based screens are more likely to reliably predict therapeutic outcomes as they maintain the complex neural circuitry involved in the underlying disease process. Whole-organism screens do not require well validated targets to identify compounds that yield a desirable phenotypic outcome, but can be prohibitively costly and time consuming in mammals. As a simple vertebrate with significant genetic similarity to human, zebrafish are now recognized as an ideal cost-effective alternative to achieve rapid in vivo phenotype-based screening.

3Using scn1a mutant zebrafish larvae with a gene homologous to human and spontaneously occurring seizures, we screened, in a blinded manner, a repurposed library of ∼1000 compounds for drugs that inhibit unprovoked seizure events. We also screened two compounds (huperzine A and fenfluramine) that were discovered in rodent-based assays using acquired seizure protocols and that were recently suggested as potential treatments for DS. Only 20 compounds in the repurposed drug library reduced swim behavior to control levels. However, many of these compounds were toxic or were not confirmed on retesting, and only four compounds advanced to a second-stage in vivo electrophysiology assay. Of these compounds (cytarabine, dimethadione, theobromine, and norfloxacin) only dimethadione, a T-type calcium channel antagonist previously reported to have anticonvulsant activity, reduced ictal-like electrographic discharges seen in scn1Lab mutant larvae. This two-stage phenotype-based screening approach, using a genetic DS model with >75% genomic similarity to human, is a sensitive, rapid means to successfully identify compounds with antiepileptic activity.

4Zebrafish were maintained in a light- and temperature-controlled aquaculture facility under a standard 14:10 h light/dark photoperiod. Adult zebrafish were housed in 1.5 L tanks at a density of 5-12 fish per tank and fed twice per day (dry flake and/or flake supplemented with live brine shrimp). Water quality was continuously monitored: temperature, 28-30º C; pH 7.4-8.0; conductivity, 690-710 mS/cm. Zebrafish embryos were maintained in round Petri dishes (catalog #FB0875712, Fisher Scientific) in “embryo medium” consisting of 0.03% Instant Ocean (Aquarium Systems, Inc.) and 000002% methylene blue in reverse osmosis-distilled water. Larval zebrafish clutches were bred from wild-type (WT; TL strain) or scn1Lab (didys552) heterozygous animals that had been backcrossed to TL wild-type for at least 10 generations. Homozygous mutants (n = 6544), which have widely dispersed melanosomes and appear visibly darker as early as 3 d postfertilization (dpf; Fig. 1b ), or WT larvae (n = 71) were used in all experiments at 5 or 6 dpf. Embryos and larvae were raised in plastic petri dishes (90 mm diameter, 20 mm depth) and density was limited to ∼60 per dish. Larvae between 3 and 7 dpf lack discernible sex chromosomes. The care and maintenance protocols comply with requirements outlined in the Guide for the Care and Use of Animals (ebrary Inc., 2011) and were approved by the Institutional Animal Care and Use Committee (protocol #AN108659-01D).

5Zebrafish larvae were placed individually into 1 well of a clear flat-bottomed 96-well microplate (catalog #260836, Fisher Scientific) containing embryo media. Microplates were placed inside an enclosed motion-tracking device and acclimated to the dark condition for 10-15 min at room temperature. Locomotion plots were obtained for one fish per well at a recording epoch of 10 min using a DanioVision system running EthoVision XT software (DanioVision, Noldus Information Technology); threshold detection settings to identify objects darker than the background were optimized for each experiment. Seizure scoring was performed using the following three-stage scale: Stage 0, no or very little swim activity; Stage I, increased, brief bouts of swim activity; Stage II, rapid “whirlpool-like” circling swim behavior; and Stage III, paroxysmal whole-body clonus-like convulsions, and a brief loss of posture. WT fish are normally scored at Stage 0 or I. Plots were analyzed for distance traveled (in millimeters) and mean velocity (in millimeters per second). As reported previously, velocity changes were a more sensitive assay of seizure behavior. For electrophysiology studies, zebrafish larvae were briefly paralyzed with α-bungarotoxin (1 mg/ml) and immobilized in 1.2% agarose; field recordings were obtained from forebrain structures. Epileptiform events were identified post hoc in Clampfit (Molecular Devices) and were defined as multispike or polyspike upward or downward membrane deflections greater than three times the baseline noise level and >500 ms in duration. During electrophysiology experiments zebrafish larvae were continuously monitored for the presence (or absence) of blood flow and heart beat by direct visualization on an Olympus BX51WI upright microscope equipped with a CCD camera and monitor.

6Compounds for drug screening were purchased from MicroSource Discovery Systems, Inc. (PHARMAKON 1600) and were provided as 10 mm DMSO solutions (Table 1). Test compounds for locomotion or electrophysiology studies were dissolved in embryo media and were tested at an initial concentration of 100 µm, with a final DMSO concentration of <2%. In all drug library screen studies, compounds were coded and experiments were performed by investigators who were blind to the nature of the compound. Baseline recordings of seizure behavior were obtained from mutants bathed in embryo media, as described above; a second locomotion plot was then obtained following a solution change to a test compound and an equilibration period of 15–30 min. Criteria for a positive hit designation were as follows: (1) a decrease in mean velocity of ≥44% (e.g., a value based on the trial-to-trial variability measured in control tracking studies; Fig. 1c ); and (2) a reduction to Stage 0 or Stage I seizure behavior in the locomotion plot for at least 50% of the test fish. Each test compound classified as a “positive hit” in the locomotion assay was confirmed, under direct visualization on a stereomicroscope, as the fish being alive based on movement in response to external stimulation and a visible heartbeat following a 60 min drug exposure. Toxicity (or mortality) was defined as no visible heartbeat or movement in response to external stimulation in at least 50% of the test fish. Hyperexcitability was defined as a compound causing a ≥44% increase in swim velocity and/or Stage III seizure activity in at least 50% of the test fish. Hits identified in the primary locomotion screen were selected from the PHARMAKON 1600 library and rescreened using the method described above. Select compound stocks that were successful in two primary locomotion assays, and were not classified as toxic in two independent clutches of zebrafish, were then purchased separately from Sigma-Aldrich for further testing. Drug concentrations between 0.5 and 1 mm were used for electrophysiology assays to account for more limited diffusion in agar-embedded larvae.

7Data are presented as the mean and SEM, unless stated otherwise. Pairwise statistical significance was determined with a Student’s two-tailed unpaired t test, ANOVA, or Mann–Whitney rank sum test, as appropriate, unless stated otherwise. Results were considered significant at p < 0.05, unless otherwise indicated.

8Locomotion tracking is a reliable and rapid strategy with which to monitor behavioral seizures in freely swimming larval zebrafish. In these locomotion plots, high-velocity movements of ≥20 mm/s correspond to paroxysmal whole-body convulsions, referred to as Stage III, and are consistently observed in unprovoked scn1Lab mutant larvae but not in age-matched wild-type siblings. Using automated locomotion tracking, we performed a phenotype-based screen to identify compounds that significantly reduce mutant swim behavior to levels associated with Stage 0 or Stage I (e.g., activity equivalent to that seen in normal untreated WT zebrafish). In a 96-well format, we tracked mutant swim activity at baseline, and then again after addition of a test compound (100 µm); each compound was tested on six individual mutant larvae (Fig. 1a ), and larvae were sorted based on pigmentation differences (Fig. 1b ). Mutant swim activity between two consecutive recording epochs in embryo media is tracked on every plate as an internal control. A box plot showing the change in swim velocity in untreated mutants is shown in Figure 1c (n = 112) and defined as the control. Based on an SD of 21.8 for these data, we set the detection threshold as any compound that inhibits movement (measured as a change in mean velocity) by >2 SDs (or ≥44%). This approach was previously validated using standard antiepileptic drugs in this model. Next, we screened a repurposed library in which all compounds have reached the clinical evaluation stage (PHARMAKON 1600 Collection; http://www.msdiscovery.com/pharma.html). Among the 1012 compounds screened (Fig. 1d ) only 20 (or 1.97%) were found to significantly inhibit spontaneous seizure behavior in scn1Lab mutants. All 20 compounds were subsequently retested in a separate clutch of scn1Lab mutants at a concentration of 100 µm (Fig. 2a, trial 2; N = 6 fish/compound). A total of 154 compounds were classified as “toxic” (Table 2); 55 compounds were classified as “hyperexcitable” (Table 3). Representative locomotion tracking raw data plots for gemfibrozil, a toxic nonsteroid nuclear receptor ligand, and mepivacaine, a hyperexcitable proconvulsant anesthetic, are shown in Figure 2b.

9Extracellular recording electrodes are a reliable, reproducible, and sensitive approach to monitor electroencephalographic activity in agar-immobilized larval zebrafish. Field electrodes offer high a signal-to-noise ratio and can be placed, using direct visualization in transparent larvae, into specific CNS structures (i.e., telencephalon or optic tectum). Using a local field electrode, we can efficiently monitor the occurrence of electrographic seizure events in the same zebrafish that were previously tested in the locomotion assay. Based on a positive nontoxic result in two independent locomotion assays, four drugs moved on to electrophysiology testing at concentrations between 500 µm and 1 mm (Fig. 3). Consistent with a “false-positive” classification, spontaneous epileptiform discharge activity was observed for three of these drugs: norfloxacin, theobromine, and cytarabine. Dimethadione, previously shown to inhibit spontaneous epileptiform discharges in thalamocortical slices at concentrations between 1 and 10 mm, suppressed burst discharge activity in scn1Lab mutant larvae (Fig. 3a,b ). To identify whether any of these four compounds exert nonspecific effects on behavior, they were also tested on freely swimming WT zebrafish larvae (5 dpf) at a concentration of 500 µm. Comparing the total distance moved during a 10 min recording epoch before, and after, the application of a test compound failed to reveal any significant changes in locomotor activity (Fig. 3c ).

10Next, we tested two additional compounds that were not in our drug library, but have recently been described as potential antiepileptic treatments for DS. Huperzine A, a small-molecule alkaloid isolated from Chinese club moss with NMDA-type receptor blocking and anticholinesterase activity, has purported antiepileptic actions against NMDA- or soman-induced seizures. In the locomotion assay, huperzine A failed to significantly alter scn1Lab seizure behavior at any concentration tested (Fig. 4a,b ). In contrast, huperzine A was effective at 1 mm in the acute pentylenetetrazole (PTZ) assay (Fig. 4b ). Fenfluramine is an amphetamine-like compound that has been reported to successfully reduce seizure occurrence in children with DS as a low-dose add-on therapy. In the locomotion assay, fenfluramine significantly reduced mutant mean swim velocity at concentrations between 100 and 500 μm (Fig. 4c,d ); 1 mm fenfluramine was toxic in the scn1Lab and PTZ assays (Fig. 4d ). The fenfluramine-treated scn1Lab mutant exhibited a suppression of spontaneous electrographic seizure discharge to levels similar to controls at 500 μm, but only a partial reduction in electrographic activity at 250 μm (Fig. 4e ).

11Zebrafish and humans share extensive genomic similarity. With regard to disease, 84% of genes known to be associated with disease states in humans have a zebrafish homolog. This genetic similarity and the characteristic of zebrafish larvae to exhibit quantifiable seizure behaviors or electrographic seizure discharge that is fundamentally similar to that recorded in humans make this an ideal system for drug discovery. Behavioral assays customized for automated evaluation of locomotion make moderate-to-high-throughput phenotype-based drug screening in zebrafish possible. Using this approach and a zebrafish scn1 mutant, we successfully identified antiepileptic compounds. Here we report results from screening ∼1000 compounds from a repurposed drug library and present data that will be periodically updated on-line using this open-access publishing mechanism.

12As a model system, the scn1Lab mutant zebrafish has many advantages. First, in contrast to transient and variable knockdown of gene expression using antisense morpholino oligonucleotides, scn1Lab mutants carry a stable and heritable amino acid substitution at position 1208 in the third domain of SCN1A that shares 76% homology with humans. Mutations in this channel are one of the primary genetic causes underlying DS. As zebrafish possess two scn1 genes, homozygous mutants for scn1Lab are comparable to the haploinsufficient clinical condition, and there is no variability from larvae to larvae, or clutch to clutch, with respect to gene inactivation, as is commonly observed with morpholino injections. Although crosses of heterozygotes produce only one-quarter homozygous scn1Lab mutants per mating, there are virtually no limitations on maintaining a large colony of healthy, adult breeders for these types of large-scale screens. Second, it is possible to observe and monitor seizure-like behavior consisting of rapid movements and whole-body convulsions in freely swimming scn1Lab mutants as early as 4 dpf that persist for the life of the larvae (∼12 dpf). These behaviors are comparable to those observed with exposure to a common convulsant agent (PTZ) and classified as Stage III. In addition, clear evidence for epileptiform discharge generated in the CNS of immobilized scn1Lab mutant larvae has been obtained at ages between 4 and 8 dpf. Both zebrafish measures of seizure activity are sensitive to inhibition by AEDs commonly prescribed to children with DS (e.g., valproate, benzodiazepines, and stiripentol), but are resistant to many antiepileptic compounds (e.g., phenytoin, carbamazepine, ethosuximide, decimemide, tiletamine, primidone, phenacemide, and vigabatrin). Pharmacoresistance is defined as the inability to control seizure activity with at least two different AEDs, and, with demonstrated resistance to eight AEDs, our model clearly fits this definition. This level of model validation has not been possible with morpholinos probably owing to the high degree of variability, or off-target effects, associated with this technique.

13Our screening results highlight the stringency of our approach with a positive hit rate of only 1.97% on the first-stage locomotion assay, and successful identification of 1 compound (of 1012 compounds) with known antiepileptic activity (i.e., dimethadione, a T-type channel antagonist). In additional testing, we confirmed an antiepileptic action for fenfluramine (serotonin uptake inhibitor). Similar to ethosuximide, a reduction in regenerative burst discharges associated with neuronal T-type calcium currents could be the underlying mechanism for dimethadione in DS mutants; however, it is worth noting that T-type channel blockers ethosuximide and flunarizine were not similarly effective (; this article), and that dimethadione can cause arrhythmia owing to blockade of cardiac human ether-a-go-go-related gene potassium channels. Modulation of serotonin [5-hydroxytryptamine (5-HT)] signaling by blocking uptake or increasing release from neurons by acting as substrates for 5-HT transporter (sertraline) proteins may be the mechanism of action for fenfluramine in patients with DS, though a detailed analysis of precisely how fenfluramine modulates excitability via this signaling pathway has not been performed. Nonetheless, both drugs probably exert a direct effect on network excitability (at neuronal or synaptic levels, respectively) to suppress electrographic discharge and the associated high-velocity seizure behavior seen in scn1Lab mutants, and may be potential targets for clinical use. In contrast, three other drugs identified in the primary locomotion assay were not effective in suppressing electrical events and were designated as false positives. This is not altogether surprising given that the xanthine alkaloid (theobromine), chemotherapeutic (cytarabine), and antibiotic (norfloxacin) mechanisms for these compounds would not be consistent with seizure inhibition. Moreover, the variability inherent in behavioral experiments performed on different zebrafish larvae, in different microplates, and on different days may contribute to these false-positive designations in locomotion assays, and is evident in the range of mean velocity values seen during tracking episodes from control studies (Fig. 1c ) or in the failure of many of the initial 20 lead compounds to be confirmed on subsequent retesting (see Fig. 2a ). This is a limitation of locomotion-based screening assays and is another reason why a secondary electrophysiology assay on the same zebrafish is a critical advantage of our approach.

14An additional advantage of in vivo screening with zebrafish larvae is the simultaneous identification of compounds resulting in toxicity. Zebrafish-based anticonvulsant drug-screening assays based primarily on in situ hybridization detection of early gene expression at 2 dpf do not routinely monitor spontaneous swim behavior, heart rate, or response to external stimuli. Lacking these real-time measures of toxicity, compounds observed to induce fatality in a freely swimming scn1Lab-based behavioral assay (e.g., gemfibrozil, suloctidil, pimozide, or dioxybenzone) were mistakenly classified as seizure-suppressing compounds in the PTZ-based c-Fos in situ hybridization assay. Indeed, 41% of the “anticonvulsant” compounds positively identified at 2 dpf in were toxic, proconvulsant, or simply not effective in scn1Lab mutant assays at 5-6 dpf. Similarly, it is critical to monitor blood flow and heart activity even in the agar-immobilized electrophysiology assay as compounds effective in suppressing electrical activity can also be toxic. These discrepancies highlight the potential problems associated with zebrafish drug-screening strategies that do not encompass multiple readouts and suggest the need for a note of caution when comparing screening results from different laboratory groups. While any lead compound identified in a zebrafish-based screening assay will, ultimately, need to be independently replicated and/or validated in additional mammalian model systems, the ability to rapidly identify such compounds, while simultaneously identifying potential negative side effects, makes genetically modified zebrafish a unique resource for drug discovery in an age of personalized medicine.