The ADAR RNA editing enzyme controls neuronal excitability in Drosophila melanogaster

RNA editing by deamination of specific adenosine bases to inosines during pre-mRNA processing generates edited isoforms of proteins. Recoding RNA editing is more widespread in Drosophila than in vertebrates. Editing levels rise strongly at metamorphosis, and Adar5G1 null mutant flies lack editing events in hundreds of CNS transcripts; mutant flies have reduced viability, severely defective locomotion and age-dependent neurodegeneration. On the other hand, overexpressing an adult dADAR isoform with high enzymatic activity ubiquitously during larval and pupal stages is lethal. Advantage was taken of this to screen for genetic modifiers; Adar overexpression lethality is rescued by reduced dosage of the Rdl (Resistant to dieldrin), gene encoding a subunit of inhibitory GABA receptors. Reduced dosage of the Gad1 gene encoding the GABA synthetase also rescues Adar overexpression lethality. Drosophila Adar5G1 mutant phenotypes are ameliorated by feeding GABA modulators. We demonstrate that neuronal excitability is linked to dADAR expression levels in individual neurons; Adar-overexpressing larval motor neurons show reduced excitability whereas Adar5G1 null mutant or targeted Adar knockdown motor neurons exhibit increased excitability. GABA inhibitory signalling is impaired in human epileptic and autistic conditions, and vertebrate ADARs may have a relevant evolutionarily conserved control over neuronal excitability.


INTRODUCTION
RNA editing by ADARs (adenosine deaminases acting on RNA) has been proposed to diversify transcripts, particularly in the brain, to meet the physiological needs of the organism (1,2). The ADAR enzymes convert specific adenosines to inosines within duplex regions of transcripts. Many of the transcripts edited by ADARs are expressed in the central nervous system (CNS). Ribosomes decode inosine as guanosine, and editing events within open reading frames lead to production of edited isoforms of ion channel subunits and other proteins. Production of edited subunit isoforms affects the pharmacological properties of channels and receptors such as AMPA receptors, potassium Kv1.1 channels and serotonin 5HT 2c receptors (3)(4)(5)(6). Various neurological disorders including Motor Neuron Disease, depression and epilepsy have been linked to defects in RNA editing, particularly to ADAR2 and to the transcripts that it edits (7)(8)(9)(10)(11). Developmental and tissue-specific ADAR regulation is also important, and both edited and unedited isoforms of channels have function; unedited GABA A (g-aminobutyric acid) receptor, for instance, is crucial for synapse formation in the developing vertebrate brain (12,13).
Homozygous Adar2 null mice die at or before weaning owing to seizures principally attributed to loss of editing at the Q/R site in the Gria2 (previously GluR-2), transcript encoding subunit 2 of the glutamate receptor, resulting in the production of abnormal calcium-permeable AMPA receptors (14,15). The Gria2 Q/R site is constitutively edited to 100% efficiency (16)(17)(18). Editing at this site is crucial for viability, and mutating the chromosomal Gria2 gene to express only the GRIA2 R edited isoform rescues the Adar2 mutant lethality (14). Prevention of excessive Ca 2+ influx through GRIA 2-containing AMPA receptors appears to be the primary function of ADAR2 in mice (15,16,19).
Drosophila Adar is a homologue of mammalian Adar2 (20), although Drosophila transcripts encoding subunits of excitatory glutamate receptors are not edited (21). Despite the loss of editing at specific sites in a large set of transcripts, Drosophila Adar null mutant flies are viable and morphologically normal. However Adar mutant flies do show uncoordinated locomotion, loss of the male courtship display, leg tremors, temperature-sensitive paralysis and age-dependent progressive neural degeneration (21,22). The defects of the Adar mutant are likely to derive from malfunctioning of multiple membrane channels and trafficking proteins that comprise the largest functional grouping among the edited transcripts (21,23,24). Although effects of editing on several different Drosophila ion channels have been characterized after Xenopus oocyte expression, the overall physiological role of ADAR-mediated RNA editing is incompletely understood in either insects or mammals (15,(25)(26)(27). Recently, Drosophila ADAR was shown to limit synaptic release at larval neuromuscular junctions (NMJs) (28), raising the possibility that the major physiological role of ADAR is in modulating neuronal activity and/or maintaining neural homeostasis.
Expression of Adar transcripts from the chromosomal gene is low at the earlier stages of development and increases at metamorphosis. We also previously described a potential autoregulatory circuit in adult flies in which ADAR proteins edit the Adar transcript to produce an ADAR isoform with lower RNA editing activity (29). In adult flies, which express primarily the more active Adar 3/4 splice form, dADAR protein edits Serine (S) codon 391 to Glycine (G), changing a residue on the RNAbinding face of the deaminase domain. The 'genomeencoded' dADAR 3/4S protein has an 8-fold higher editing efficiency in vitro than the edited dADAR 3/4 G isoform after purification from overexpressing yeast cultures (29).
An 'ineditable' UAS-Adar 3/4 S cDNA construct in which the self-editing (S/G) site is mutated to an alternate serine codon 'TCT', which cannot be edited, produces only the dADAR 3/4S isoform (29). Overexpression of Adar 3/4S in Drosophila under the control of the Actin 5c-GAL4 driver, which expresses ubiquitously and strongly from embryonic stages, is lethal in larvae and pupae. This is probably because the overexpressed Adar 3/4S bypasses the normal self-limitation of dADAR RNA editing activity. Lethality depends on the editing activity of dADAR 3/4S. Taking advantage of the power of Drosophila genetics, we performed a screen for genetic suppressors of this Adar overexpression lethality. We find that genetic manipulation to reduce GABA signalling rescues the lethality caused by dADAR 3/4S overexpression. We present evidence that Drosophila ADAR acts cell-autonomously to fine-tune neural activity. In vivo extracellular recordings show that dADAR 3/4S overexpression dampens larval motor neuron excitability. Also, Adar 5G1 null flies exhibit increased neuronal excitability and viability, and locomotion defects are rescued by chemical enhancement of GABA signalling. Our data indicate that dADAR is required to protect neurons from hyper-excitability and that dADAR suppresses neuronal excitability in concert with GABA signalling.

Drosophila strains
Fly stocks were maintained at 18 C on a 12 h light/dark cycle and were raised on standard corn meal agar media.

Viability screen
w 1118 ; Actin 5c-GAL4 /SM5 CyO; UAS-Adar 3/4S, Tub-GAL80 ts , virgin female flies were collected at 18 C, crossed with deficiency-bearing male flies at 27 C, and the number of progeny of each genotype was counted. Twelve to sixteen virgin female flies were crossed with 5-8 male flies in each standard vial. The parent flies were transferred three times at 2-day intervals, and their progeny were collected only up to 12 days after mating to avoid counting any second-generation progeny. To estimate the viability, the number of viable Adar 3/4S overexpressing progeny flies obtained in the heterozygous deficiency background was divided by the number of sibling heterozygous deficiency flies that had the SM5 Cy second chromosome rather than the Actin 5c-GAL4 driver second chromosome and did not overexpress Adar 3/4S. Statistical significance of rescue was calculated with the binomial test to compare the viability of the surviving adult flies with the expected un-rescued viability of 0%. Statistical significance was further corrected using False Discovery Rate correction (FDR, Bonferroni, otherwise stated).
qRT PCR cDNA made from 500 ng of RNA with oligo-dT primers was used for quantification, with minus RT control and water-only negative controls. qRT PCR was performed with SYBER GREEN master mix, with either a BioRad (C1000 TM Thermal Cycler) instrument, or a Light Cycler Õ 480 (Roche). All primers were tested for correlation factor and efficiency. All qRT PCR results were normalized to Gapdh level and to additional standards. P-values were calculated with the unpaired two-tailed Student's t-test. For each comparison, the cDNAs used were made at the same time with the same amount of RNA. The PCR primers used for PCR are Adar forward: TGGACCTTC AGTGCAATCA, reverse: CCTCACCGGACTCGATTT .
Editing assay cDNA made from total RNA was amplified with specific primers for the regions of interest in the transcript. The PCR product was either directly sequenced or cloned into pGEM-T Easy and individual clones sequenced. Two methods of quantitation for editing efficiency were used. The first method involved sequencing of total RT-PCR product pools and measuring relative peak heights at the edited site. Editing levels of PCR products were measured by comparing the heights of the Adenosine and Guanosine peaks at the same position (Editing percentage = the height of Guanosine peak/the height of both Guanosine+Adenosine peaks). Three PCR sequences were used to estimate an average editing level. For the second method of editing quantification, 60-100 colonies were picked and each clone was sequenced with both T7 and SP6 primers. P-values were calculated using the Mann-Whitney U test.

Drug administration
Clonazepam (SIGMA-ALDRICH, C1277) stock solution (63.2 mM in acetone) was diluted in distilled water to obtain a series of concentrations from 1.58 to 15.8 mM. Valproic acid (SIGMA-ALDRICH, P4543) 2-mM stock solution was made in water prior to use, and further diluted to 0.1-1 mM in water. 150 ml of the final drug dilution was added on the food surface in a standard food vial, and the food was subsequently mashed. Mated flies were placed in the drug-containing vial once most of the liquid had evaporated. Flies were transferred to fresh drug-containing food every 2 days and the number of eclosed male progeny was counted over 15 days.

Climbing analysis
A column of 1.5 cm diameter and 20 cm height, cut from a 25-ml plastic pipette, was used for the climbing assay. The height of the column was divided into 120 equally spaced lines. For each test, one 2-day-old fly was placed into the column. The highest line the fly reached in 1 min was recorded. For each genotype, 6-10 individual flies were tested three times to acquire an average score. A twotailed Welch t-test was performed to calculate the P-value compared with the Adar 5G1 null fly group. The climbing assay was performed at room temperature, at the same time each day.

In vivo extracellular recordings from larval motor neurons
Intact CNS from wandering third-instar larvae was dissected under external saline, leaving imaginal discs and peripheral nerves attached (30). An isolated CNS was positioned on top of a cured Sylgard-coated coverslip (SYLGARD Õ 184 Silicone elastomer kit). The CNS was placed with its dorsal surface uppermost and immobilized with tissue glue (Vet Bond, WPI) applied to stretched peripheral nerves.
To access motor neuron somata, a patch pipette ($10 mm opening) filled with 1 mg/100 ml protease (type XIV from Streptomyces griseus, 1 mg/ml, Sigma), diluted in external saline, was used to make a small hole in the glial sheath on or near the midline of the CNS (31). Once neuron somata were free from attached glia, the enzyme pipette was removed, and a patch pipette filled with external saline was used for extracellular current recording. Gentle suction was applied to achieve a loose-seal on the aCC soma (identified by GFP). Recordings were made for 5 min with an Axopatch 200B amplifier in voltage clamp mode (V hold 0 mV, Molecular Devices). Recordings were digitised at 20 KHz with an Axodata 1332A A/D board and collected with pClamp software.
For each genotype, four to eight neuronal extracellular current records were made. The analysis used 3-min recording periods taken shortly after the second minute of recording data. Counting the number of events (action potentials) was performed automatically in Clampfit. A burst was defined as a minimum of four events within an interval of 25 ms. P-values between different groups were calculated with the Poisson exact test.

Overexpression of adult Adar 3/4 S isoform is lethal in Drosophila
The lethality caused by overexpression of the adult Adar 3/4 S391 isoform in larvae and pupae provides an opportunity to perform a genetic screen to identify interacting genes in which mutations prevent lethality. Lethality depends on the RNA editing activity of the expressed dADAR since expressing an Adar 3/4 E329A construct, which encodes a catalytically inactive protein, or expressing the 'edited' dADAR 3/4 S391G isoform do not cause lethality (29). Genes suppressing the lethality might be key targets of RNA editing, or they may encode proteins that increase dADAR editing activity. Suppressor genes may also encode proteins that alleviate the critical toxic physiological effect of dADAR overexpression by some other means.
To perform this screen for genetic modifiers, the expression of the lethal dADAR 3/4 S protein was controlled by generating a strain that stably encodes both the Actin 5C-GAL4 driver and the UAS-Adar3/4 S construct by addition of a Tub-GAL80 ts construct (this strain will be referred to as Adar 3/4S OE). The GAL80 ts repressor protein is expressed constitutively in all cells from the tubulin promoter and binds to GAL4 to prevent transcription activation from the UAS (upstream activating sequence). At 25 C or lower, the temperature-sensitive GAL80 ts protein prevents expression of Adar 3/4S by binding to GAL4, thereby allowing adult flies to be obtained (32). At 27 C or at higher temperatures the GAL80 ts protein is no longer repressive, Adar 3/4S is expressed and no larvae or pupae survive to adulthood. Raising the temperature to 27 C has little effect on the viability of w 1118 wild-type flies ( Figure 1A).
The lethality of Adar 3/4S OE occurs throughout larval and pupal stages ( Figure 1A). RT-PCR analyses show that during embryonic stages at 27 C, the level of total Adar transcript in Adar 3/4S OE is approximately 1.5Â higher on average than at 25 C whereas total Adar transcript is 7Â higher at 27 C than at 25 C during the L3 larval stage ( Figure 1B). RT-PCR primers are complementary to the Adar coding region and detect both endogenous Adar transcript and transcript expressed from the Adar 3/4S cDNA construct. Adar transcripts are normalized to Gapdh. Interestingly, Adar 3/4S OE flies raised at lower temperatures can survive at higher temperatures after they have eclosed from the pupae, which implies that the lethal effect of ADAR overexpression is limited to the earlier developmental stages when RNA editing levels are normally lower (data not shown). The viability of the adult flies facilitates the screen for genetic modifiers because it permits crosses with them to be performed at any temperature.
A wild-type Adar gene copy is present on the X Chromosome in the Adar 3/4S OE strain but crossing in the Adar 5G1 deletion mutant that deletes the Adar gene entirely does not rescue lethality at 27 C of male Adar 5G1 deletion progeny overexpressing Adar 3/4S ubiquitously ( Figure 2A). Therefore, lethality is not mediated through the endogenous Adar gene and does not depend on additional expression of the endogenous transcript nor on any feedback effect on expression of Adar isoforms from the endogenous locus. Overexpressing an Adar RNA hairpin (Adar siRNA) fully rescues the Adar 3/4S OE lethality, confirming that lethality is owing to excess Adar (Figure 2A).
A genetic screen identifies chromosomal deficiencies that rescue lethality caused by Adar 3/4S overexpression To identify genetic modifiers of Drosophila Adar 3/4 S OE lethality, we performed a forward genetic screen by crossing Adar 3/4 S OE adult virgin flies from stock maintained at 25 C with males of heterozygous deficiency lines and raising the progeny at non-permissive temperatures >25 C. When the Adar3/4S OE flies were crossed at 29 C to flies of 128 stocks bearing chromosomal deficiencies that cover approximately 70% of Chromosome 3 in large blocks (33)  crosses were repeated at 27 C, three heterozygous deficiencies on Chromosome 3L rescued the lethality (Supplementary Figure S1, Supplementary Table S1). We found three other weakly rescuing deficiencies on Chromosome 3R (Supplementary Table S1). A similar screen of deficiencies on Chromosomes 1 and 2 did not identify any further rescuing deficiencies. We focused our research on the Df(3L)RDL-2 deficiency because it rescued the lethality of Adar 3/4S OE most efficiently, to 90% of the level obtained with Adar RNAi (FDR P < 1.0E-312, Figure 2A). One lethality suppressor we identified on Chromsome 3R was the ubiquitin ligase slmb (supernumerary limbs); however, we did not study it further. Heterozygous mutation in slmb prevented Adar transcript overexpression. Reduced dosage of this ubiquitin ligase may affect the GAL4/UAS expression system, possibly by contributing to turnover of the GAL80 ts protein at nonpermissive temperatures (data not shown).
We were unable to map the effects of some other rescuing deficiencies down to single genes, as testing rescue by overlapping deficiencies gave uninterpretable outcomes. This is a significant difficulty that frequently arises with deficiency screens and the reasons for such inconsistencies are not understood. The Df(3L)RDL-2 deficiency is a gamma ray-induced chromosomal deletion at the 66F5 region that deletes Rdl (Resistant to dieldrin), nwk (nervous wreck), Tequilla and S(CypE JP )3.4 (34) (flybase: http://flybase.org/reports/FBab0024018.html).
The most interesting suppressors are those that allow continued overexpression of Adar while preventing the lethal effect. To determine whether the Df(3L)RDL-2 deficiency rescues without preventing Adar overexpression, we compared Adar transcript levels in rescued Adar3/4S OE flies heterozygous for the Df(3L)RDL-2 deficiency with Adar transcript levels in wild-type and in Adar3/4S OE flies rescued by RNAi against the overexpressed Adar. The Df(3L)RDL-2 deficiency-rescued flies had highly elevated expression of Adar transcript, with about 10fold more Adar transcript in whole flies compared with w 1118 wild-type flies or Adar3/4S OE flies rescued by RNAi against Adar ( Figure 2B). Therefore, it appears that rescue by Df(3L)RDL-2 is not caused by reduced expression of the Adar transcript, but is likely to involve mitigation of the effects of Adar overexpression. Because of Adar3/4S OE lethality, we could not compare the expression level of Adar transcript in the rescued adult flies with the level in Adar3/4S OE adult flies obtained through the same rearing protocol at 27 C.
Increased editing at sites in larval transcripts is a possible cause of the lethality observed on overexpression of highly enzymatically active adult ADAR 3/4 S isoform during embryonic and larval stages. This inappropriate editing could lead to expression of adult protein isoforms, which would mediate the lethal effect of Adar overexpression. We previously reported increased editing of the Ca-1D transcript encoding a voltage-gated calcium channel in muscle and CNS in Adar 3/4S-overexpressing larvae (35).
Increased editing may still occur in the Df(3L)RDL-2rescued adult flies, as they express elevated levels of Adar.
To determine whether Adar 3/4S OE flies rescued by the Df(3L)RDL-2 deficiency have above-normal RNA editing activity in some transcripts, we examined editing in the CG33205 transcript from rescued adult flies. We chose this transcript because it has 10 edited sites with low or moderate editing levels close together in the 3'UTR. Its editing profile gives us an advantage in evaluating the editing level using a single pair of primers only, as most individual RNA editing sites require a separate set of RT PCR products for each site. The CG33205 transcript is of no other particular interest except that it is widely expressed and might be edited even outside the nervous system in Adar 3/4S OE. One site (Position 8) showed a significant increase (1.65 fold, P 0.034) in the editing level in Adar 3/4S OE; Df(3L)RDL-2 flies ( Figure 2C), indicating that the overexpressed Adar 3/4S OE does lead to increased editing activity. We do not expect to see above-normal RNA editing at all editing sites, as previously shown (21,36). Preventing Adar overexpression by RNAi against Adar prevents increased editing in the CG33205 transcript. Indeed Adar 3/4S OE; Adar RNAi shows a trend for reduced editing at some sites below that seen in w 1118 wild-type flies, which may reflect reduction in isoforms produced from the endogenous Adar locus. These findings show that heterozygous Df(3L)RDL-2 rescues the Adar 3/4S OE lethality even though Adar is still highly overexpressed and still maintains above-normal editing activity.
Reduced Rdl (Resistant to dieldrin), gene dosage or Rdl mutants rescue Adar 3/4S overexpression lethality Among the candidate genes in the Df(3L)RDL-2 deficiency, only mutations in Rdl were able to rescue the lethality of Adar 3/4S OE when tested. Heterozygosity for the Rdl null allele Rdl 1 , a large intragenic inversion covering multiple exons in Rdl (37), rescues Adar 3/4S OE viability to 51.9% (n = 94; Figure 2A), supporting our conclusion that the effect of the Df(3L)RDL-2 deficiency is due to reduced expression of the Rdl gene. Rdl 1 is a gamma ray-induced inversion with both breakpoints within the Rdl transcript. Rdl 1 homozygotes do not produce any Rdl transcript, which results in late embryonic stage lethality (37). Two Rdl RNAi lines were also tested at 27 C, but we could not determine whether they rescue the lethality of Adar 3/4S OE. Rdl is an essential gene and knocking down Rdl ubiquitously from embryonic stages by combining the Actin 5C-GAL4 driver with these Rdl RNAi constructs in an Adar wild-type background is also lethal. The encoded RDL protein (also known as GABA A receptor subunit beta) is the pore-forming subunit of the ligand-gated inhibitory GABA A receptor which is a major class of inhibitory receptor in the fly. RDL mediates fast synaptic inhibition through GABA A receptors and shares 30-38% identity with vertebrate GABA A receptor beta subunits (38). RDL can form functional GABA A receptors as a homomer when expressed in Xenopus laevis oocytes (38,39). Although the physiological compositions of the insect GABA A receptors are not clear, the homooligomeric GABA A receptor formed by RDL subunits expressed in Xenopus closely resembles native insect neuronal GABA A receptor in pharmacological sensitivities (40,41). The Rdl transcript is also alternatively spliced and alternate splice forms in combination with RNA editing profoundly influence GABA-mediated inhibition (25).
To confirm that mutations affecting RDL channel activity rescued Adar overexpression lethality, the Adar 3/4 S OE flies were crossed with three additional Rdl mutant lines-Rdl CB-2L , Rdl MD-RR and Rdl CB . All three are ethyl methanesulfonate (EMS)-induced Rdl mutations causing amino acid replacements; the Rdl CB-2L M267I mutation is on the extracellular part of the channel whereas Rdl CB-2 and Rdl MD-RR change single amino acids at several sites in the intracellular part of the channel (http://flybase.org/reports/FBgn0004244.html) (Supplementary Figure S2). Rdl CB-2L /+ rescued lethality by 57.9% (n = 55, FDR P < 1.0E-312; Figure 2A) but neither of the other alleles did. A limited number of Rdl mutations are available, but the data are consistent with the hypothesis that RDL channel function is the basis for rescue.

No significant increase in editing is observed in the Rdl transcript in Adar 3/4S overexpressing or Df(3L)RDL-2-rescued animals
We investigated editing levels at sites in Rdl transcripts in Adar3/4S OE L3 larvae at 27 C. Among the five known edited sites, only the 735RG site showed a trend of increased editing (6.5-10.5%; Figure 3A). The 1218IV site, which is 100% edited in wild-type, is edited about 100% in Adar3/4S OE, and the other sites that are not edited in wild-type are also not edited in Adar3/4S OE ( Figure 3A). In rescued adult Adar 3/4S OE; Df(3L)RDL-2 and Adar 3/4S OE; Adar RNAi flies, editing levels at sites in the Rdl transcript were not significantly different from w 1118 ( Figure 3B). Thus, we do not have evidence showing that altered editing of the Rdl transcript underlies Adar 3/4S OE lethality, although editing may influence expression of RDL channels.
Increasing or decreasing GABA signalling causes reciprocal effects on phenotypes of Adar 5G1 null mutant and Adar3/4S OE flies To test whether the rescue of Adar 3/4S OE lethality by Rdl mutants was due to a reduction in GABA signalling, we examined the effect of decreasing GABAergic input on Adar 3/4S OE lethality. Gad1 L352F is a strong hypomorphic mutant allele of Gad1 (Glutamic acid decarboxylase 1), which encodes the enzyme that synthesizes GABA. The heterozygous Gad1 L325F mutant has approximately 50% of the enzymatic activity of wild-type Gad1, and this is predicted to lead to reduced GABA inhibitory signalling (42). Crossing Adar 3/4S OE to Gad1 L325F rescued lethality to 59.5% (n = 107, FDR P < 1.0E-312; Figure 4A), demonstrating that the decrease in GABA inhibitory signalling can be the mechanism underlying rescue by reducing Rdl expression.
Because reduction in Rdl expression or GAD1 activity rescued the lethality caused by Adar 3/4S overexpression, we sought a reciprocal rescue of Adar 5G1 null mutant phenotypes by chemically enhancing GABA signalling. Clonazepam (CLZ), is a benzodiazepine drug that binds to the benzodiazepine site of the GABA receptors, increasing the effect of GABA binding on neurons (43). CLZ has also been shown to weakly activate GABA A receptors in insects (44). We also fed Adar 5G1 with a histone deacetylase inhibitor, Valproic Acid (VPA). VPA has a wide spectrum of activities including enhancing GABA neurotransmission by inhibiting GABA transaminase, and this function is conserved in lower animals including Drosophila (45,46). Adar 5G1 null male flies have significantly reduced viability compared with FM7 male balancer siblings among progeny of Adar 5G1 /FM7 females ( Figure 4B). The FM7 male siblings have a wildtype Adar gene on a balancer first chromosome with multiple chromosomal inversions marked with a dominant Bar eye marker. Adar 5G1 /FM7 flies were reared on either 1.58-15.8 mM clonazepam (CLZ) or 0.1-20 mM VPA-containing food to ensure that parent flies were exposed to the drug. These flies were then allowed to lay eggs on drug-containing food. Both CLZ (15.8 mM Â 150 ml) and VPA (0.1 mM Â 150 ml) increased the viability of Adar 5G1 null flies relative to FM7 siblings among the progeny in the next generation ( Figure 4B).
Adar 5G1 adult flies display locomotion defects and have reduced ability to climb up in a narrow column ( Figure 4C). A fly in a narrow column climbs up to the top naturally, a response known as negative geotaxis (47). Most wild-type flies climb to the top in 20 s after being tapped to the bottom of the column but Adar 5G1 flies climb up to approximately 20% of the height of the column in 1 min ( Figure 4C). Adar 5G1 null flies grown on the food containing high concentration (15.8 mM) CLZ or low dosage (0.1 mM) VPA showed statistically significant improvement in climbing ability by day 2 (FDR P < 0.02), although they were still defective compared with wild-type flies (FDR P < 1.0E-4; Figure 4C). Therefore, enhanced GABA signalling does ameliorate mutant phenotypes in Adar 5G1 flies, consistent with the hypothesis that these flies have increased excitability due to loss of RNA editing in neuronal transcripts.
Adar 5G1 null mutant and dADAR-overexpressing larvae display reciprocal abnormalities in aCC motor neuron excitability A possible explanation for the rescue of the Adar 3/4 S OE lethality by Rdl mutants is that lethality arises from suppressed neuronal excitability. We hypothesized that lethality is rescued by increased excitability due to an overall reduction in fast inhibitory GABA signalling through the RDL channel. If this hypothesis is correct, the Adar 3/4S OE flies would have lower neuronal activity while Adar null flies would have higher neuronal firing activity.
If phenotypes of Adar mutants or Adar overexpressors are due to aberrant excitability that can be corrected by manipulating GABA receptors then aberrant excitability is likely to occur in many different types of neurons that express Rdl. The larval motor neuron preparation gives access to a well-studied neuron in which we can measure the effects of increased or decreased Adar expression on neuronal excitability. The aCC motor neurons in the larval ventral ganglion project to dorsal muscles, receive GABA input, express Rdl (37) and can be identified when UAS-GFP is expressed specifically in these neurons under the control of an RRa-GAL4 driver. We recorded endogenous action potential firing in aCC motor neurons of L3 larvae using an in vivo extra-cellular recording procedure. When additional Adar 3/4S is expressed in the Adar wild-type in the aCC motor neurons specifically using the RRa-GAL4 driver, the aCC motor neurons show significantly reduced firing activity ( Figure 5A). Reciprocally, aCC motor neurons of Adar 5G1 null mutants were hyper-active ( Figure 5A). aCC motor neurons with RRa-GAL4-driven Adar siRNA knockdown in an Adar wildtype background were also hyper-active ( Figure 5A).
The firing frequency, represented by the number of firing events (Figure 5B), and the number of bursts in 3 min ( Figure 5C) were significantly decreased in Adar 3/4S OE aCC motor neurons (FDR P < 1.0E-6), but increased in Adar 5G1 null or Adar siRNA aCC motor neurons (FDR P < 1.0E-7). All these phenotypes are more severe in the Adar siRNA aCC motor neurons than in the Adar 5G1 null larvae. Therefore, these are likely to be cell-autonomous effects of dADAR level. We reason that Adar 5G1 larval aCC motor neurons show less severe phenotypes than Adar siRNA aCC motor neurons because non-cell-autonomous effects differ depending on whether other projecting neurons or niche cells also have Adar expression. The mean number of the events in each burst, however, was significantly increased only in the Adar 5G1 null larval aCC motor neurons (FDR P 2.27E-5, Figure 5D), indicating that, unlike the other effects, the effect on burst length is not cell-autonomous but is a systemic physiological response of the neurons. These in vivo extracellular recordings of Adar mutant larval aCC motor neurons strongly support the genetic data, indicating that a major physiological role of dADAR is to control neuronal excitability.

DISCUSSION
Ubiquitous expression of a highly active adult ADAR isoform at larval and pupal stages is lethal in Drosophila. Taking advantage of the GAL4/UAS system, we placed the lethal Adar overexpression under temperature-dependent control, constructing a Drosophila strain that can be reared at 25 C for the purpose of performing a genetic screen. When progeny from this line (termed Adar 3/4S OE) or from crosses of this line to wild-type are reared at 27 C, all Adar 3/4S OE progeny die during pre-adult stages. Adar 3/4S OE females were crossed to males carrying defined chromosomal deletions, and we identified heterozygous deletions that prevent Adar 3/4S OE-induced lethality. The Rdl gene is responsible for the effect of one strongly rescuing deletion. Rdl is an essential gene encoding a pore-forming subunit of the important inhibitory GABA A receptors in Drosophila.
Among     flies by feeding flies with CLZ or VPA improved survival and early-stage fitness. Compared with CLZ, which is specific for GABA receptors, VPA has a much wider spectrum of effects, including increased autophagy. Nevertheless, both drugs improved Adar 5G1 fly defects to similar degrees, indicating that additional effects of VPA do not further improve mutant fly viability and locomotion.
The firing frequency of single motor neurons is controlled cell autonomously by dADAR. Adar 5G1 larval aCC motor neurons have increased firing frequency and knocking down Adar in aCC motor neurons further increases firing frequency. Overexpressing dADAR 3/4S in aCC motor neurons reduces spontaneous firing activity. These cell autonomous effects are more likely to be mediated mainly through the editing status of ion Tonic burst firing pattern Neuron acitivity threshold Figure 6. Drosophila CNS edited transcripts encode a range of proteins likely to underlie aberrant neuronal excitability in Adar-overexpressing or Adar null mutants. Solid lines indicate direct control and dashed lines show indirect effects. Each box represents a category of edited transcripts encoding proteins that either directly or indirectly influences neuronal excitability by effects on the action potential firing threshold or tonic or burst firing pattern control. Transcripts edited by dADAR encode ligand-gated channels, potassium channels and sodium channels that set the neuronal activity threshold. Rdl, encoding a GABA-gated GABA A receptor subunit, is highlighted in the box at the right lower corner. Various transcripts encoding modulators, including ion transporters, neurotransmitters and G-protein coupled signalling proteins are also edited by dADAR. Functional changes in these proteins will also indirectly modulate the ion channels that determine the neuronal activity threshold. dADAR is also expected to modulate the neuron firing pattern. An edited slo transcript that encodes the Calcium-dependent large potassium (BK) channel is known to directly regulate the tonic or burst firing pattern of the neuron. Ca 2+ directly regulates BK channel activity; therefore, all genes listed in the 'Ca 2+ signal' box both influence BK channel activity and express transcripts edited by dADAR. In addition, some edited transcripts encode cAMP signalling and PKA family proteins that may indirectly control BK channel activities.
channel receptors on motor neuron membranes than by changes in upstream inputs or in presynaptic interneurons. However, the defect in cell-autonomous firing frequency control is more significant in the neurons with targeted knockdown of Adar than in the neurons from the Adar 5G1 null mutant larva. The physiological feedback on motor neurons from other cells with which they interact is likely to be affected by whether those cells also have Adar. We demonstrate that dADAR dampens neuronal spontaneous firing activity cell autonomously in concert with the GABA A fast inhibitory signal and suggest that this is the likely basis for lethality rescue by mutations in Rdl.
Tonic and burst firing pattern control is likely to be modified by dADAR non-cell autonomously. Adar 5G1 larval motor neurons show prolonged bursts whereas knockdown of Adar in motor neurons did not change the length of each burst or the pattern of tonic and burst firing. We recorded aberrant excitability in Adar 5G1 null and Adar 3/4S overexpressing larval motor neurons, but it is likely that many other types of neurons are similarly affected. We do not know where in the body the lethal effect of Adar 3/4S overexpression is caused although it is likely to be in GABA-responsive neurons that express Rdl.
We do not fully understand how changes in ADAR dosage cause changes in excitability. The lethal effect of Adar 3/4 S overexpression is dependent on the high RNA editing activity of this isoform, and the increased editing at sites in larval transcripts that we reported previously has also been observed at further sites in this study. However, the number of identified sites of increased editing remains surprisingly limited. Editing levels at many sites appear relatively unresponsive to changes in dADAR levels or isoforms. A recent report from Savva et al. (36) examined the editing site specificity of dADAR S and dADAR G proteins expressed from the chromosomal Adar locus and found that most sites are edited similarly by the two ADAR isoforms, with only a few sites preferring dADAR S. Purified ADAR proteins edit individual RNA substrates specifically in vitro, but other proteins interacting with RNA or with dADAR may have strong effects on editing levels at individual sites in vivo. The rescue by altered Rdl dosage raises the question of whether altered editing of Rdl drives the changes in neuronal excitability seen in Adar mutants or overexpressors. Rdl transcripts are edited at one site in larvae and at additional sites in adult flies, but alterations in the pattern of editing in the Rdl transcript itself do not appear to be central to Adar 3/4S OE lethality. Indeed, Rdl editing at most sites increases the EC50 values for homomeric RDL channel responses to GABA agonists. If increased editing also makes native RDL channels similarly less responsive to GABA then effects of Rdl editing seem unlikely to mediate the neuronal excitability changes observed (25). Therefore, aberrant editing of transcripts other than Rdl may underlie the effects of increased or decreased dADAR on neuronal excitability.
The reciprocity of the excitability effects of increased and decreased dADAR, however, raises the intriguing possibility that these effects arise from either too much or too little editing in the same set of transcripts. The effects of editing alterations on the function of each individual encoded protein may be mild (36), though cumulative effects may be significant ( Figure 6). The frequency of neuron firing events is known to be controlled directly by large-conductance calcium and voltage-gated potassium channels (BK channels), and indirectly by calcium channels (48). The slowpoke (slo) transcript encoding the BK Ca 2+ -channel ( Figure 6) has four editing sites, including two that result in amino acid changes in the extracellular vestibule and conducting pore of the channel (21,23,49). Many transcripts encoding synaptic release machinery are also edited by dADAR, including the synaptotagmin transcript encoding the Ca 2+ sensor for synaptic vesicle fusion (23,50). The increase in the length of burst may be caused mainly by altered presynaptic Ca 2+ release in the Adar 5G1 larvae, due to loss of editing in transcripts encoding neurotransmitter receptors and to altered abundance of synaptic release machinery as is observed in the NMJ of Adar 5G1 larvae (28). A full survey of editing events in Adar 3/4S OE will be required to search for key targets. The effects on excitability could also arise in more indirect ways through editing of microRNAs or repetitive RNA sequences leading to stress or immune responses (51)(52)(53).
We demonstrate for the first time that RNA editing by dADAR can fine-tune neuronal activities in concert with GABA fast inhibitory signalling. Many, though not all, of the effects of editing on ion channels in flies and vertebrates can be accommodated by the hypothesis that A-to-I RNA editing reduces neuronal excitability (15,27,54). For instance, editing reduces calcium permeability of glutamate-gated AMPA channels, decreases serotonergic potency of 5HT 2C receptors and reduces the inactivation rate of the Kv1.1 channel (4,55,56). Therefore, RNA editing can create protein diversity though why it has evolved to specifically modulate neuronal excitability is a question that remains to be answered.