Seizure control through genetic and pharmacological manipulation of Pumilio in Drosophila: a key component of neuronal homeostasis

ABSTRACT Epilepsy is a significant disorder for which approximately one-third of patients do not respond to drug treatments. Next-generation drugs, which interact with novel targets, are required to provide a better clinical outcome for these individuals. To identify potential novel targets for antiepileptic drug (AED) design, we used RNA sequencing to identify changes in gene transcription in two seizure models of the fruit fly Drosophila melanogaster. The first model compared gene transcription between wild type (WT) and bangsenseless1 (parabss), a gain-of-function mutant in the sole fly voltage-gated sodium channel (paralytic). The second model compared WT with WT fed the proconvulsant picrotoxin (PTX). We identified 743 genes (FDR≤1%) with significant altered expression levels that are common to both seizure models. Of these, 339 are consistently upregulated and 397 downregulated. We identify pumilio (pum) to be downregulated in both seizure models. Pum is a known homeostatic regulator of action potential firing in both flies and mammals, achieving control of neuronal firing through binding to, and regulating translation of, the mRNA transcripts of voltage-gated sodium channels (Nav). We show that maintaining expression of pum in the CNS of parabss flies is potently anticonvulsive, whereas its reduction through RNAi-mediated knockdown is proconvulsive. Using a cell-based luciferase reporter screen, we screened a repurposed chemical library and identified 12 compounds sufficient to increase activity of pum. Of these compounds, we focus on avobenzone, which significantly rescues seizure behaviour in parabss flies. The mode of action of avobenzone includes potentiation of pum expression and mirrors the ability of this homeostatic regulator to reduce the persistent voltage-gated Na+ current (INaP) in an identified neuron. This study reports a novel approach to suppress seizure and highlights the mechanisms of neuronal homeostasis as potential targets for next-generation AEDs. Summary: Next-generation anticonvulsant compounds potentiate the activity of the neuronal homeostatic regulator Pumilio.


INTRODUCTION
The number of known contributory genetic loci to human seizure exceeds 500, which greatly increases the challenge of providing personalised medicine through tailoring treatments based on individual gene mutation (Noebels, 2015). An alternative is to target treatment to common modifiers to which larger groupings of individual gene mutations contribute. One obvious modifier is neuronal homeostasis, which acts to stabilise neural circuit activity levels through continual adjustment of neuron excitability (Turrigiano, 2012). However, this opportunity remains unexplored.
Seizures in humans and Drosophila exhibit sufficient parallels to implicate that the underlying neuronal abnormalities are highly similar. This includes defined seizure thresholds, common genetic mutations that modify seizure susceptibility, spread of seizures along defined neuronal tracts and suppression of seizures by recognised AEDs . As in humans, certain mutations in Drosophila genes result in a seizure phenotype (collectively termed bang-sensitive). Seizures can also be induced in Drosophila by exposure to proconvulsants, including picrotoxin (PTX), primarily through block of inhibitory GABA A receptors (Lin et al., 2012).
To model seizure, we used para bss , which is a L1699F point mutation that imparts a gain of function in the sole voltage-gated sodium channel (Na V ) of the fly genome (Parker et al., 2011). Mutations in the human ortholog, SCN1A, are associated with seizure and intractable epilepsy (Escayg and Goldin, 2010). In comparison, we also used exposure to PTX. We exploited the molecular tractability of Drosophila to identify changes to gene transcription that occur during seizure to identify possible pathway nodes exploitable for anticonvulsant therapy. Comparison between the two models identifies 743 common transcriptional changes, including pum. Pum is a translational repressor that binds mRNA transcripts that normally (but not exclusively) contain an 8nucleotide binding motif in their 3′-UTR, termed a Nanos response element [NRE, also known as Pumilio response element (PRE)] (Gerber et al., 2006). A particularly relevant Pum target, with respect to seizure, is Na v . We have previously shown that the fly Na v ( paralytic) is translationally regulated by Pum and also that rat Scn8a (Na v 1.6) is similarly regulated by Pum2 (the closest mammalian homologue to pum) (Driscoll et al., 2013;Mee et al., 2004;. This mechanism forms part of a wellcharacterised homeostatic response that tunes action potential firing to match the changing level of synaptic excitation to which neurons are exposed (Baines, 2005;Weston and Baines, 2007). Two recent studies highlight the potential involvement of Pum in epilepsy. First, a Pum2 knockout mouse exhibits spontaneous seizures (Siemen et al., 2011) and second, PUM2 expression is reduced in human patients suffering temporal lobe epilepsy (TLE) (Wu et al., 2015).
We show here that overexpression of pum in para bss flies is markedly anticonvulsant. By contrast, RNAi-mediated knockdown of pum exacerbates seizure. The likely beneficial effect of upregulation of Pum is through reduction of the voltage-gated persistent sodium current (I NaP ) in central neurons. Thus, our results highlight mammalian Pum2 as a potential target for the design of novel, and possibly wide-spectrum, AEDs. To identify potential compounds that influence Pum activity and/or expression, we constructed a luciferase reporter of Pum activity and screened a comprehensive library of approved compounds. From 785 compounds, we identify 12 that potentiate Pum activity. Further analysis of one of these compounds, avobenzone, shows that it increases transcription of pum, reduces I NaP in identified motoneurons and is potently anticonvulsive in Drosophila.

RNA-sequencing identifies pum as downregulated in seizure
In order to determine changes to gene transcription that occur in seizure-prone CNSs, we used RNA sequencing (RNA-seq) to compare gene transcription in the CNS in two models of seizure: a genetic model ( para bss ) and a chemical model (PTX). Using RNA extracted from the CNS of third instar larvae (L3), we identified transcriptional change in 2246 and 1013 genes, respectively, using an FDR≤1% in WT versus para bss and WT versus WT fed PTX (see Tables S1, S2). Comparison between data sets revealed that 743 common genes exhibit significant change to expression (Fig. 1A inset, see Table S3 for gene details). Of these, 736 showed significant and consistent altered expression in both seizure models. A log 2 plot of fold-change (log 2 FC) showed that 339 (46%) are significantly upregulated and 397 (54%) are significantly downregulated (P=0.001, ANOVA). The remaining seven genes did not show consistent direction of change (Fig. 1A). Identified genes generated a total of 130 functional clusters representing a wide array of functions, including predicted genes encoding ion channels and synaptic proteins (detailed below). The top 20 enriched clusters are shown in Fig. S1. The top four clusters are for genes associated with prereplicative complex assembly, eukaryotic translation elongation factor 1 complex, negative regulation of neuroblast proliferation and translation repressor activity. Genes associated with translational repression include minichromosome maintenance (orthologues 2, 3, 5, 7), elongation factor 1α100E, 1α48D and 1β, anachronism, prospero, musashi, embryonic lethal abnormal vision, brain tumor and pum (Table S3). Twenty genes that we identify have been positively associated with human epilepsy (http://www.informatics. jax.org/humanDisease.shtml) (red dots in Fig. 1A, and described in Table 1). Of these genes, five were upregulated and 15 were downregulated in the Drosophila seizure models. These genes include paralytic (Na v ), nicotinic Acetylcholine Receptor α5, I h channel and Shaker (K + channels) in addition to Syntaxin, Synapsin and unc-13 (synaptic proteins). Seven genes were identified that show particularly large increases in transcription (>3 log 2 FC, blue dots in Fig. 1A). These genes are CG18331 (mucin 68Ca), CG34076 (mitochondrial NADH-ubiquione oxidoreducatse chain 3), CG11205 ( photorepair), CR41620/CR40734 (rRNA genes) and CG7606/CG32198 (unknowns).
Our attention was drawn to pum, which was significantly downregulated in both seizure models [mean±s.d., WT: 602±14 Fig. 1. Analysis of altered gene transcription in seizure models. (A) Cross-comparison shows 743 changes are common to both seizure models. Analysis of direction of log 2 fold-change (log 2 FC) in transcription of the 743 common genes (main figure) shows that 339 are significantly (two-way ANOVA) upregulated and 397 downregulated. Seven genes show differential expression in the two models. pumilio ( pum), which is downregulated, is identified by the orange dot. Genes previously linked to human epilepsy are shown by red dots (described in Table 1). Blue dots highlight genes that show particularly large fold-changes (log 2 FC>3) in expression levels in the seizure backgrounds (see Results text for identity). Inset: analysis of the transcriptome by RNA-sequencing shows change to transcription of 2246 genes in the para bss CNS compared with wild type (WT). Comparison of WT with WT fed picrotoxin (PTX) shows 1013 transcriptional changes. (B) Analysis of pum transcript level in isolated CNS from L3 shows a significant reduction in WT+PTX and para bss compared with WT controls. (C) pum is significantly reduced in adult heads in both seizure models compared with WT controls. The WT value has been set to 1 in each experimental condition. Data are mean±s.d. for n=5 independent samples, *P≤0.05. ***P≤0.001 (unpaired t-test).
vs WT fed PTX: 405±2 and para bss : 381±15 cpm (counts per million), P=1.5×10 −5 , n=3]. This is because its homologue, PUM2, has been reported to be downregulated in humans suffering temporal lobe epilepsy (Wu et al., 2015). Pum is a wellcharacterised translational repressor, which we have previously reported regulates translation of Na v s in both Drosophila and rat to achieve homeostatic control of neuron action potential firing (Driscoll et al., 2013;Mee et al., 2004;. We considered that manipulation of a homeostatic regulator might represent a promising approach to control seizure. To validate RNA-seq data, we undertook RT-qPCR; pum was significantly decreased in WT larval CNS after exposure to PTX (0.83±0.03) and in para bss (0.78±0.03) compared with WT control (set as 1, P=0.03, n=5; Fig. 1B). We observed a similar and significant downregulation of pum transcription in adult heads that contain mostly brain tissue (WT fed PTX: 0.41±0.07 and para bss : 0.58± 0.17) relative to WT control (set as 1, P=0.0002, n=5; Fig. 1C). In addition, we used RT-qPCR to validate the identification of the 20 genes that have been positively associated with epilepsy (red dots in Fig. 1A). This validation was undertaken only for the para bss background. We found consistent change for 14 of the genes (representing a validation rate of 70%). Two genes showed significant change by RT-qPCR but in the opposite direction to RNA-seq, whereas four genes showed no significant change (see Table 1).

Increased pum expression decreases I NaP in motoneurons
Our previous work has shown that Pum regulates I Na through translational regulation of para (Mee et al., 2004;. We recorded from para bss /Y L3 where the expression of transgenic pum was selectively manipulated in only the aCC motoneuron (using RRa-Gal4). Our choice to use this motoneuron is guided by the ability to combine genetics and electrophysiology; a selective Gal4 driver exists to express UAS-transgenes in this neuron, which is also accessible to patch electrodes. That I NaP is greater in amplitude in aCC motoneurons in seizure mutants (Marley and Baines, 2011) is indicative that they share properties with central interneurons in human epilepsy, which can also show increased I NaP (Stafstrom, 2007).
Increased expression of pum in L3 para bss aCC resulted in a striking reduction of I NaP (4.4±4.1 pA/pF vs 12.6±4.0 pA/pF, P=4.9×10 −5 ; Fig. 3A,B,D) but no change to I NaT (Fig. 3E). Analysis of the persistent-to-transient current ratio (P:T) recorded in L3 aCC showed a marked reduction (20.0±18.0% vs 51.0±11.9%, P=5.0×10 −5 ; Fig. 3F). A high P:T ratio (>40%) in central motoneurons has been previously shown to be characteristic of Drosophila seizure mutants and its reduction to be anticonvulsant (Lin et al., 2015;Marley and Baines, 2011). Thus, we conclude that upregulation of pum is anticonvulsant, which is due, at least partially, to its ability to reduce I NaP .
RNAi-mediated downregulation of pum in L3 para bss aCC increased I NaT (31.3±3.3 pA/pF vs 24.7±4.5 pA/pF, P=0.005) but did not affect I NaP or the P:T ratio ( Fig. 3C-F). Analysis of the effect on seizure behaviour following this more selective manipulation of pum expression showed no significant differences to controls ( para bss /Y;; RRa-Gal4/+, data not shown). This is entirely expected given the highly selective cell targeting used in these experiments. However, a more widespread manipulation of pum [e.g. using Cha-Gal4(19B)], which is sufficient to alter seizure duration and/or severity, probably acts via an identical mechanism: through alteration of I Na .
Increasing pum expression in aCC in a WT background resulted in essentially the same changes to I Na as seen with manipulation in the para bss background; I NaP was significantly reduced (2.4± 1.7 pA/pF vs 7.4±4.9 pA/pF, P=0.0028; Fig. 3G) but no change to I NaT was observed (18.8±4.8 pA/pF vs 21.9±2.7 pA/pF; Fig. 3H). By contrast, downregulation using pum RNAi produced a different outcome compared with para bss ; I NaP was significantly increased (11.0±2.4 pA/pF vs 7.4±4.9 pA/pF, P=0.032; Fig. 3G) with no effect on I NaT (24.6±4.7 pA/pF vs 21.9±2.7 pA/pF; Fig. 3H). Analysis of the P:T ratio, however, similarly only showed a significant reduction following upregulation of pum expression in WT (14.7±11.9% vs 33.3±20.2%, P=0.016; Fig. 3I).
On occasion, we noted the appearance of multiple resurgent I Na during the I NaP plateau in the para bss background (Fig. 4A, indicated by arrow). Moreover, we observed a significant correlation between the occurrence of resurgent I Na and pum level (P=0.002, Chi-square test; Fig. 4B). Thus, resurgent I Na was most often observed following RNAi-knockdown, and only rarely following expression of pum. The origin of these currents remains uncertain. Analysis of voltage recordings (Fig. 4A) showed no obvious issue of space clamp, which suggests these currents are not occurring in distal unclamped regions of the neuron. The averaged frequency of the resurgent currents was ∼100 Hz, which did not vary with level of pum expression (Fig. 4C). Resurgent currents are particularly evident at holding potentials between −50 to −20 mV and exhibit highest frequency at −30 mV (RRa-Gal4/UAS-pum RNAi : 104.50± 36.78 Hz; RRa-Gal4/+: 120.00±20.16 Hz; RRa-Gal4/UAS-pum: 115.00±40.93 Hz). Increased resurgent I Na probably supports increased action potential firing consistent with our observation that RNAi-mediated knockdown of pum is proconvulsant (Grieco et al., 2005). Resurgent I Na is only rarely observed (<5%) in WT aCC recordings (data not shown). A screen to identify positive regulators of Pum activity Upregulation of Pum activity, either through increased transcription or post-transcriptional modification might provide an effective means to suppress seizures. To identify possible lead compounds with this mode of action, we constructed a luciferase-based reporter of Pum activity for use in an in vitro S2R+ cell line suited to large-scale screens (Lin et al., 2015). Overexpression of pum is sufficient to repress luciferase activity (due to translational repression), whereas incubation with pum double-stranded RNA is sufficient to increase luciferase activity by reducing endogenous Pum activity. PCR analysis shows that pum is endogenously expressed in S2R+ cells (Fig. S2). Thus, activity of the firefly-luciferase-NRE reporter Increasing expression (UAS) is sufficient to reduce I NaP with no change to I NaT , whereas reduction (RNAi) increases I NaP amplitude but has no effect on I NaT .
(I) Analysis of the P:T ratio in individual cells recorded in G,H shows increased pum is sufficient to reduce the ratio. Data are means±s.d. for n independent cells stated in individual bars. *P≤0.05, **P≤0.01, ***P≤0.001 (two-way ANOVA with Bonferroni's post hoc). Fig. 4. Occurrence of resurgent I Na is related to level of pum. (A) Resurgent I Na (I NaR , arrow) is seen superimposed on repolarization of holding potential used to evoke I NaP . Analysis of the voltage trace (lower trace) shows good control during this step. (B) The occurrence of I NaR , in the para bss background, is highest when pum is reduced (RNAi, 82%, 14 from 17 cells) and lowest when increased (UAS, 21%, 3 from 14 cells). Control (CTRL, para bss , 64%, 9 from 14 cells). Transgene expression was limited to aCC cells using RRa-Gal4. (C) Frequency of I NaR oscillations is unaffected by expression level of pum. Data are means±s.d.
(FF-NRE) reflects the absolute level of Pum function in these cells. A second reporter, which lacked an NRE-motif, was also transfected [renilla (Ren)-luciferase] to allow detrimental effects to cell viability to be determined. The final readout of the assay was a FF:Ren luciferase ratio that would be reduced following upregulation of Pum activity.
We screened 785 compounds from a repurposed library (see Materials and Methods; drugs screened are listed in Table S4). We identified 12 compounds that significantly reduced the FF:Ren ratio at 5 μM (Table 2). Based on structure and/or known drug target, the compounds fall into one of four groupings: those containing a methoxybenzaldehyde moiety (aniracteam and avobenzone); anticancer agents (cladribine, gemcitabine, floxuridine, clofarabine, bleomycin and docetaxel), mTOR inhibitors (temsirolimus and rapamycin) and topoisomerase II inhibitors (mitoxantrone and teniposide). Our attention was particularly drawn to avobenzone because, unlike the other compounds, it had no significant effect on transcription of the control Ren-luciferase reporter (all other compounds also reduced expression of this reporter, in addition to decreasing the FF:Ren ratio). Thus, we took avobenzone forward for further testing.
Our predicted mode of action for avobenzone is inconsistent with an immediate effect of this compound, acting instead to potentiate Pum, which, in turn, downregulates Na v channels in the neuronal membrane. To test this, we recorded from non-drug-exposed L3 para bss aCC and used bath application of avobenzone (5 µM). No changes were observed in either component of I Na (data not shown) and the P:T ratio remained unaffected (Fig. 5H). Higher doses (20 µM), or longer exposure times (10 min) similarly produced no detectable effect (data not shown). This lack of acute effect is consistent with our predicted mode of action. Finally, to directly test this prediction, we measured pum transcript abundance in para bss L3 grown in the presence of avobenzone. We observed a modest, but statistically significant, increase in transcript abundance of ∼20% (1.2±0.17, n=5, P=0.04, t-test, vehicle control set as 1; Fig. 5I). Thus, we conclude that avobenzone, acting to increase the transcription and/or transcript stability of pum, is able to suppress seizure duration through downregulation of I NaP . Finally, we observed equally potent anticonvulsive activity of avobenzone in two other bang-sensitive mutants: easily-shocked (avobenzone: 142±82 vs control: 240±120 s, n=40, P=1.0×10 −5 , L3 electroshock) encoding an ethanolamine kinase (Pavlidis et al., 1994) and slamdance (avobenzone: 178±122 vs control: 272± 108 s, n=40, P=6.8×10 −5 , L3 electroshock) encoding an aminopeptidase (Zhang et al., 2002), indicative that increasing Pum activity might be effective against a broad range of epilepsies.

DISCUSSION
The causes of seizure, even in genetic epilepsies, vary greatly and are not confined to genes with obvious contributions to ion flux across neuronal membranes. This increases the challenge to identify individual mutations, to determine the physiological role of both the WT and mutated protein and, ultimately, to design drugs to minimise the unwanted effect of the mutation. In this study, we identify transcriptional changes that occur in the seizure-prone CNS. We identify over 700 common genes that show altered transcription in two different seizure models. It is noteworthy that we observed approximately double the number of genes showing altered transcription in para bss flies compared with those treated with PTX. The reason for this is unclear but might represent accumulated compensatory changes in the mutant line that have occurred in order to lessen the severity of seizure activity in para bss mutants. These additional genes warrant further investigation as potential seizure suppressors.
Many of the common transcriptional changes we identify, and in particular those that are upregulated (and thus open to inhibition by drug exposure), might provide effective drug targets for novel AED design. However, our attention was drawn to Pum, which we have previously shown orchestrates homeostasis of action potential firing in both Drosophila and rat central neurons (Driscoll et al., 2013;Mee et al., 2004). The degree of seizure suppression achieved by upregulating Pum in para bss flies is considerable and is only matched by the no-action-potential (nap ts ) allele of the maleless (mle) locus in Drosophila, which encodes an ATP-dependent double-stranded RNA (dsRNA) helicase (Ganetzky and Wu, 1982). This mutation causes a catastrophic change in splicing of the Drosophila Na v (Reenan et al., 2000). The net effect of both of these manipulations, increased Pum or the presence of nap ts , is to reduce the availability of functional Na v expressed in central neurons. The direction of change of pum in the two seizure models (that show All but avobenzone also reduce expression of the control Ren luciferase reporter that does not contain an NRE motif. Luciferase values shown are normalised such that 1.0 would represent no effect. reduced expression) might not be ideal with respect to drug development given that disruption of a gene or protein is often more achievable. Nevertheless, we show that upregulation of pum in a Drosophila seizure mutant is potently anticonvulsive and, further, we identify a potential lead anticonvulsive compound that seemingly increases the level of expression of this homeostatic regulator. This compound might catalyse the development of a novel class of AED.
Neurons display an array of homeostatic mechanisms to maintain action potential firing within pre-determined and physiologically appropriate limits (Davis, 2013). Pum is a well-characterised RNAbinding protein that binds mRNA, usually through a specific motif termed the NRE. Once bound, Pum recruits additional cofactors including Nanos and Brain tumor (Brat) to form a complex that is sufficient to prevent translation (Wharton et al., 1998). Our results in this study indicate that increased expression of Pum might have therapeutic benefit for seizure suppression. However, a potential issue in this regard is that a genome-wide identification of RNAs bound to Pum in ovaries identifies upwards of 700 genes (FDR<0.1%) (Gerber et al., 2006). This raises the problem of specificity of effect following global potentiation of level or activity of Pum. This potential issue might, however, be overcome through identifying and targeting neuronal-specific regulators of Pum. One such alternative target might be the inhibition of Myocyte enhancer factor 2 (Mef2)-induced expression of miR-134 in neurons that, in turn, inhibits translation of mammalian PUM2 (Fiore et al., 2009). Additional possibilities include targeting of cofactors required for Pum activity. It is interesting in this regard that a loss-of-function mutation in mei-P26, a homologue of Brat, produces strong seizure suppression in Drosophila bang-sensitive seizure mutants (Glasscock et al., 2005).
Mammalian PUM2 binds transcripts encoding SCN1A (Na v 1.1), and SCN8A (Na v 1.6) (Driscoll et al., 2013;Vessey et al., 2010). A reduction in supply of Na v protein to the neuron membrane is The frequency of cells that exhibit resurgent I Na correlates with avobenzone concentration (P=0.005, Chi-square test). (H) P:T ratio measured from para bss aCC before (CTRL) and after a 1 min bath application of 5 µM avobenzone. (I) Analysis of pum transcript level in isolated CNS from para bss L3 raised on food containing avobenzone (0.4 mg/ml) shows a significant increase compared with para bss L3 raised on food containing an equal amount of vehicle (0.8% DMSO). The control value has been set to 1. Data are means±s.d. for n independent cells stated in individual bars. *P≤0.05, **P≤0.01, ***P≤0.001 (A,H-I, unpaired t-test; D-F, two-way ANOVA with Bonferroni's post hoc). consistent with a reduction in action potential firing and a general anticonvulsant effect (Mee et al., 2004). Analysis of I Na in motoneurons indicates that a likely mechanism includes a marked reduction in I NaP . Increased I NaP is associated with mutations in SCN1A that have been identified from individuals with epilepsy (Meisler and Kearney, 2005) and is specifically reduced by AEDs such as phenytoin, valproate and lamotrigine (Stafstrom, 2007). In light of this, the anticonvulsant effect of increased pum expression is understandable. That reducing pum expression through RNAimediated knockdown is proconvulsive is again both predictable and understandable. However, the effect of this manipulation on I Na is not so clear. Rather than increasing I NaP , I NaT is instead significantly increased together with a novel appearance of resurgent I Na during repolarisation. Increased I NaT would be expected to reduce the threshold for action potential firing (i.e. making firing more likely), whereas resurgent I Na is associated with increased action potential firing frequency, partly by reducing the refractory period (Grieco et al., 2005). Although we have observed this current component in recordings from seizure mutants (including para bss ), it is rarely observed in WT or following expression of transgenic pum.
The ability to manipulate Pum in vivo to determine its anticonvulsive properties in rodent models of seizure will be greatly aided by the identification of chemical compounds that directly potentiate either expression or activity state. We report the use of a suitable cell-based screen to identify such compounds and highlight avobenzone as a potential lead compound for future development. The in vivo toxicity of avobenzone has not been well established and although there are few reports of serious side effects associated with its use as an active ingredient of sunscreen, its tendency to form free radicals might be a potential issue. To our knowledge, this compound has not been used to treat neurological disease, and its mode of action in reducing seizure in Drosophila remains to be determined. Our observations that ingestion of avobenzone result in increased expression of pum is indicative that this compound might mimic elements of the pathway that control expression of this homeostatic regulator.
The output of our screen also provides additional support for the use of rapamycin to control seizure (Lasarge and Danzer, 2014;Russo et al., 2013), indicative that this molecule might influence neuronal homeostasis. The identification of topoisomerase II as a potential target to control seizure also validates previous observations reporting that inhibition of this class of nuclear protein is anticonvulsant (Lin et al., 2015;. Finally, that we identify that the increase in Pum activity by aniracetam might hint at an additional mode of action for this class of known anticonvulsants (Shiotani et al., 2000). The related racetams, levetiracetam and brivaracetam, are currently in clinical use as AEDs, exploiting their capability to bind and inhibit synaptic vesicle protein 2A (SV2A) (Klitgaard et al., 2016).
In summary, we present a description of transcriptional change present in seizure-prone CNS. We identify, in particular, that pum expression is downregulated in both genetic and chemically induced seizure models. This mirrors the reported reduction in PUM2 in human TLE and in rats exposed to the proconvulsant pilocarpine (Wu et al., 2015). It also provides a possible understanding for why Pum2 null mice exhibit spontaneous seizures (Siemen et al., 2011). However, it is perplexing that pum levels should decrease during seizures given that the published model predicts an increase (Mee et al., 2004). As reduced Pum levels are predicted to increase neuronal excitability, it seems that epileptic seizures are associated with a pathological dysregulation of pum expression. We speculate that this occurs because Pum can auto-regulate (the pum transcript contains NRE motifs). Thus, although the neuronal hyperactivity induced by seizures will initially increase Pum expression, the accumulating Pum protein might feed back to downregulate its own transcript (Gerber et al., 2006). Sampling at later stages after seizure occurrence might only report reduced Pum compared with nonseizure controls. Indeed, we have shown that upregulation of pum in the Drosophila CNS, through expression of a wild-type transgene (lacking NRE motifs), results in reduction of endogenous pum transcript level (W.-H.L. and R.A.B., unpublished data). Prevention of this feedback, achievable in this study through expression of transgenic pum lacking an NRE, or exposure to avobenzone, holds significant promise for anticonvulsant therapy.

Fly stocks
Wild type (WT, maintained in the Baines lab) was Canton-S. para bss (bss 1 ), which was obtained from Dr Kevin O'Dell (Institute of Molecular, Cell and Systems Biology, University of Glasgow, UK), is detailed in Parker et al. (2011). The para bss stock (and other transgenic lines used) were not backcrossed to the CS stock. Controls consisted of either untreated para bss and/or parental stocks (i.e. Gal4/+, UAS/+) and are stated in respective figure legends. Slamdance iso7.8 was obtained from Dr Mark Tanouye (Department of Environmental Science, Policy and Management and Department of Molecular and Cell Biology, University of California Berkeley, California, USA). Easily-shocked 2F was obtained from Dr Kevin O'Dell. RRa-Gal4 is expressed in only the aCC and RP2 motoneurons (Lin et al., 2012). We are able to discriminate between these neurons during electrophysiological recordings and use only the aCC neuron in this study. We used  to drive UAS-transgene expression in all cholinergic neurons, which include excitatory premotor interneurons (Salvaterra and Kitamoto, 2001). Pan-neuronal expression was achieved by combining elaV-Gal4 (Bloomington stock no. 8760, 3rd chromosome insert) with para bss . UAS-pum RNAi was obtained from the Vienna Drosophila RNAi Center (stock no. 101399) and UAS-pum is detailed in Schweers et al. (2002). UAS-pum lacks NRE motifs that are present in the 3′-UTR of the endogenous pum gene. All genetic crosses were maintained at 25°C with the exception of overexpression of pum (larvae die as 1st or 2nd instars). These experiments were maintained at 20.5°C. Chemical-induced seizure was achieved by raising WT larvae on food containing 0.25 mg/ml PTX (P1675, Sigma, Poole, UK) until wall-climbing third instar, abbreviated to L3 (Lin et al., 2015).

Library construction and RNA sequencing
CNSs were removed from 50 L3 (mixed sexes) and RNA extracted using the RNeasy mini kit (QIAGEN, Hilden, Germany) as described (Lin et al., 2015). RNA integrity and purity were determined using an Agilent 2200 TapeStation system (Agilent Technologies, Santa Clara, CA). The RNAsequencing library was created using an mRNA Seq library preparation kit as per manufacturer's instructions (Illumina Inc., San Diego, CA). The library products were sequenced, in paired-end reads, using an Illumina HiSeqTM 2000. RNA-sequencing data were analysed using edgeR (empirical analysis of digital gene expression in R) (Robinson et al., 2010). This analysis identified genes with altered levels of expression using a threshold false discovery rate (FDR)≤1%. GO terms for Biological Process, Cellular Component, Molecular Function and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway were used for annotations. We classified differentially expressed genes using the Functional Annotation Cluster (FAC) tool available in the Database for Annotation, Visualization and Integrated Discovery (DAVID) (Huang et al., 2009a,b).

Validation of RNA-sequencing analysis by quantitative PCR
Quantitative RT-PCR was performed using a SYBR Green I real-time PCR method (Roche, LightCycler ® 480 SYBR Green I Master, Mannheim, Germany) as described in Lin et al. (2015). RNA was extracted from either 20 adult heads (3 days old) or 20 L3 CNSs (mixed sexes) using the RNeasy micro kit (QIAGEN). Primer sequences (5′ to 3′) used were: actin-5C (CG4027), CTTCTACAATGAGCTGCGT and GAGAGCACAGCCTGG-AT; pum (CG9755), GCAGCAGGGTGCCGAGAATC and CGCGGCGA-CCCGTCAACG (forward and reverse, respectively). Relative gene expression was calculated as the 2 −ΔCt , where ΔCt was determined by subtracting the average actin-5C Ct value from that of pum.

Luciferase reporter construction
A region of the 3′UTR (NM_169233.2, 2390-2650) of hunchback, containing two pum-binding motifs (NRE 1 and NRE 2 ) (Gupta et al., 2009), was subcloned from UAS-firefly-NRE/pUAST (a gift from Dr Kevin Moffat, University of Warwick, UK) by releasing the DNA fragment using EcoRI and XhoI sites and ligating it into pAc5.1 vector (Invitrogen). Renilla luciferase was subcloned from pRL-CMV vector (Promega) by releasing the DNA fragment using NheI (filling the sticky end to blunt end with Klenow) and XbaI sites and ligating it into EcoRV and XbaI sites of pAc5.1 vector (Invitrogen).

Seizure behaviour test
Twenty virgin females of para bss ; Cha-Gal4(19B) were mated with five males of UAS-pum RNAi , UAS-pum or WT. Because para bss is on the X chromosome and heterozygous para bss /+ females show significantly reduced recovery time, we used para bss /Y male F1 progeny for behavioural screening. For adult seizure determination, male flies (3 days old) were tested at least one day after collection to ensure total recovery from CO 2 -anaesthesia. Ten flies were transferred to an empty plastic fly vial and left to recover for 30 min before a mechanical shock induced by vortexing the vial at maximum speed for 10 s. Recovery time (RT) was calculated from the average time taken for all 10 flies to recover from paralysis to standing (to produce a single value). At least three replicates (of 10 flies per vial) were performed for each condition tested and the recovery time averaged across the three vials. Avobenzone was fed to young adult male flies ( para bss /Y), within 8 h of eclosion. Groups of 10 flies were placed in an empty vial and exposed to drugsoaked filter paper. Drug was first mixed with a sucrose solution (5%) to produce a final concentration of 0.4 mg/ml (1.6% DMSO). Filter paper soaked in this solution was added to vials and left for 24 h before testing.
To measure seizure in larvae, an electroshock assay was performed as previously described (Marley and Baines, 2011). Briefly, L3 male larvae ( para bss /Y) were transferred to a plastic dish after washing to remove food residue and gently dried using paper tissue. Once normal crawling behaviour resumed, a conductive probe composed of two tungsten wires (0.1 mm diameter, ∼1-2 mm apart) was positioned over the approximate position of the CNS, on the anterior-dorsal cuticle of the animal. A 30 V DC pulse for 3 s, generated by a Grass S88 stimulator (Grass instruments, RI, USA) was applied. In response to the electric stimulus, we observed a transitory paralysis in which larvae tonically contracted and, occasionally, exhibited spasms. The time to resumption of normal crawling behaviour was measured as RT. For drug-feeding studies, larvae were raised on food containing avobenzone (PHR1073, Sigma), in 0.8% DMSO, until reaching L3.

Electrophysiology
Whole-cell voltage-clamp recordings were performed on aCC motoneurons at L3 as previously described (Marley and Baines, 2011). Leak currents were subtracted on-line (P/4). The same stimulation protocol was applied three times to each neuron and the recordings averaged. Current amplitudes were normalised for cell capacitance, determined by integrating the area (1 ms time range) under the capacity transients elicited by stepping the cell from −60 to −90 mV for 30 ms. Cells exhibiting no measurable I NaP (resulting from excessive resurgent I Na ) were not included in the quantitative analysis.
To evaluate the effect of pum manipulation on I Na , virgin females of para bss ;; RRa-Gal4 were crossed with UAS-pum RNAi , UAS-pum or WT males. Only para bss /Y male F1 progeny was recorded at L3. To investigate avobenzone action, para bss ;; RRa-Gal4 larvae were raised on food containing 0.8% DMSO or avobenzone at different concentrations (0.1, 0.2 and 0.4 mg/ml) until reaching L3. Acute drug treatment was performed by bath-applying avobenzone to the external saline (0.5% DMSO). I Na was recorded from para bss ;; RRa-Gal4 aCC motoneurons before and 1 min after bath application. Controls were exposed to DMSO alone.

Statistics
Statistical significance between group means was assessed using either a Student's t-test (where a single experimental group is compared with a single control group) or ANOVA followed by the Bonferroni's post hoc test (multiple experimental groups). The Chi-square test was used for statistical analysis of categorized data. Data shown is mean±s.d.