Adar RNA editing-dependent and -independent effects are required for brain and innate immune functions in Drosophila

ADAR RNA editing enzymes are high-affinity dsRNA-binding proteins that deaminate adenosines to inosines in pre-mRNA hairpins and also exert editing-independent effects. We generated a Drosophila AdarE374A mutant strain encoding a catalytically inactive Adar with CRISPR/Cas9. We demonstrate that Adar adenosine deamination activity is necessary for normal locomotion and prevents age-dependent neurodegeneration. The catalytically inactive protein, when expressed at a higher than physiological level, can rescue neurodegeneration in Adar mutants, suggesting also editing-independent effects. Furthermore, loss of Adar RNA editing activity leads to innate immune induction, indicating that Drosophila Adar, despite being the homolog of mammalian ADAR2, also has functions similar to mammalian ADAR1. The innate immune induction in fly Adar mutants is suppressed by silencing of Dicer-2, which has a RNA helicase domain similar to MDA5 that senses unedited dsRNAs in mammalian Adar1 mutants. Our work demonstrates that the single Adar enzyme in Drosophila unexpectedly has dual functions.

A denosine-to-inosine (A-to-I) RNA editing is an important co-transcriptional RNA modification (for reviews 1,2 ). In A-to-I RNA editing, the ADAR (adenosine deaminase acting on RNA) enzyme binds to double-stranded RNA (dsRNA) and deaminates specific adenosines to inosines. Inosines are generally read by the cellular machinery as guanosines and RNA editing can change encoded protein sequences, as well as altering RNA structure and RNA splicing. ADAR RNA editing occurs throughout the transcriptome at about fifty specific coding sites in humans but site-specific editing is greatly expanded in some invertebrates such as Drosophila, where it occurs at over a thousand specific coding sites in transcripts. Additional promiscuous RNA editing occurs in long dsRNAs formed by inverted repetitive elements in both mammals and Drosophila; in humans this additional ADAR RNA editing occurs at millions of sites, mostly non-specifically and at low efficiency in Alu dsRNA hairpins embedded in protein-coding transcripts [3][4][5] . Unlike DNA mutations, RNA editing usually changes only some proportion of the transcripts of particular genes, and editing levels at individual sites can be different in different tissues and developmental stages [5][6][7][8] , due to different ADAR expression levels and other tissue-specific factors. Thus, RNA editing is a powerful tool that can over-write and enrich the genomically encoded information during the process of gene expression.
ADAR proteins also have editing-independent aspects to their functions 9 ; ADAR binding to dsRNA and editing of adenosines within the dsRNA are separable events, and ADARs bind to some dsRNAs that are not efficiently edited 10 . Studies of ADAR1 effects on pri-miRNA processing show that ADAR binding to dsRNA can interfere with binding of other proteins such as Drosha and Dicer 9,11 . ADARs strongly affect processing of many pri-miRNAs even though very few mature miRNAs are themselves edited 12,13 . Editing-independent effects are likely to involve ADAR1 binding to RNAs or interactions with other proteins, such as Dicer 14 , and it is important to understand such actions of ADARs.
In vertebrates, the major defects in Adar mutants appear to depend almost entirely on Adar RNA editing activity in vivo although the editing-independent effects are also detected. Humans and mice have two catalytically active ADAR proteins: ADAR1 and ADAR2, in addition to ADAR3, which is catalytically inactive. ADAR1 is essential for innate immune function (for review 15,16 ). Mice lacking Adar1 die by embryonic day E12.5 due to the aberrant activation of cytoplasmic, antiviral, innate immune dsRNA sensors, which causes an upregulation of interferon expression and innate immune responses [17][18][19] . Elimination of MAVS, a mitochondria-associated adaptor protein required for innate immune activation by RIG-I-like receptors (RLRs), antiviral cytoplasmic dsRNA sensors, allows Adar1, Mavs double mutant mice to survive to birth, although they die soon after birth 18,19 . The immune regulatory role of ADAR1 is largely dependent on its RNA editing activity; Adar1 E861A mice expressing catalytically inactive Adar1 E861A protein also show upregulation of interferon-stimulated genes (ISG) and embryonic lethality 17 . The Adar1 E861A mutant mice die two days later than Adar1 null mutants, by E14.5. Elimination of MDA5 (encoded by Ifih1), one of three antiviral dsRNA sensor helicases, allows the Adar1 E861A , Ifih1 double mutant to survive to a full life span with no obvious defects 17 . The differences between inactive Adar1 E861A mutant and Adar1 null mutant mice suggest that editing-independent effects are also significant, although restoring Adar1 E861A by transfection in Adar1 mutant cells prevents aberrant innate immune induction only very partially 18 .
In contrast to the innate immune role of Adar1, vertebrate Adar2 is critical for neurological function and shows an even clearer dependence on RNA editing activity. Mice lacking Adar2 die within 3 weeks of birth with increasingly severe seizures associated with effects of increased influx of calcium ions through glutamate receptors on synaptic plasticity across the brain 20,21 . These severe neurological defects are caused by loss of editing at a single glutamine (Q) to arginine (R) editing site in the Gria2 transcript encoding the key GluA2 subunit of AMPA class glutamate gated ion channels 20 .
Drosophila offers an opportunity to study Adar RNA editingdependent and independent functions in more detail. Unlike mammals, Drosophila has only one Adar gene encoding a protein similar to mammalian ADAR2 1,22,23 . Drosophila Adar null mutants such as Adar 5G1 have severe locomotion defects, seldom fly, have aberrant neurotransmitter synaptic vesicle accumulation and sleep defects and develop age-dependent neurodegeneration 24,25 . All Adar 5G1 null mutant phenotypes examined are rescued by expressing Drosophila Adar or human ADAR2 but not human ADAR1 22 , showing the strong conservation of function between these ADAR2type enzymes.
To determine whether Adar mutant phenotypes in Drosophila are fully dependent on RNA editing, we generated a fly strain (Adar E374A ) that has a point mutation in the chromosomal Adar gene resulting in production of catalytically inactive Adar E374A enzyme. We find that loss of Adar RNA editing activity in Drosophila leads to severe locomotion defects and age-dependent neurodegeneration similar to Adar 5G1 null mutant flies 24 . Furthermore, we show that the loss of RNA editing by Adar induces aberrant innate immune gene expression in Drosophila, through Dicer-2 acting as a dsRNA innate immune sensor. This is analogous to the findings in mice where loss of Adar1 editing leads to immune induction through the activation of MDA5 17 ; Drosophila Dicer-2 has a helicase domain of the same type as MDA5 26 . These results show that Drosophila Adar has dual functions of both Adar1 and Adar2 proteins in vertebrates.

Results
An Adar E374A mutant encoding a catalytically inactive Adar. Drosophila deletion mutants with low to completely depleted levels of Adar have been used for characterizing Adar functions 24,27 . However, it is unclear whether Drosophila Adar functions are dependent on RNA editing activity. To uncouple the editingdependent from the editing-independent functions of fly Adar, we used CRISPR/Cas9-mediated homologous recombination to generate a fly strain (Adar E374A ) with a point mutation (Fig. 1a, b), in the catalytic Adar adenosine deaminase domain at the active site glutamate that is critical for transferring a proton during the adenosine deamination reaction. In addition, a synonymous point mutation was introduced at the sgRNA target site to prevent repeated cleavage by Cas9. This Drosophila Adar E374A mutant encodes a stable protein functionally similar to a catalytically inactive Adar1 E861A mouse mutant, which has been critical for showing that ADAR1 has both editing-dependent and editingindependent effects 17 .
The Adar gene is on the first (or X) chromosome and we confirmed that RNA editing was eliminated by sequencing of total, ribosome-depleted head RNA of male Adar E374A and wild type w 1118 flies. Wild-type flies have site-specific editing events with a wide range of editing levels in hundreds of transcripts (Fig. 1c), as previously described 28 , but Adar E374A flies undergo no detectable A-to-I editing (Fig. 1d). The expression level of the Adar E374A transcript is slightly increased compared to wild type Adar ( Supplementary Fig. 1).
RNA editing is required for CNS function. For a simple assessment of Adar E374A catalytically inactive and Adar 5G1 null mutant CNS function, we examined locomotion in male flies. First, as in the original study that reported the Adar 5G1 null mutant 24 , we conducted open field locomotion assays by counting the number of lines crossed by individual flies in a gridded petri plate (Fig. 2a). Each fly was observed for three periods of two minutes each; each period began with tapping the petri plate on the bench but the locomotion is otherwise unstimulated and spontaneous. Neither the Adar E374A nor the Adar 5G1 mutants moved far compared to controls, which indicates their severe locomotion defects (Fig. 2a).
Next, we conducted negative geotaxis assays, commonly used measures of behavioral health, coordination, and neurodegeneration 29 . These assays take advantage of the instinctive climbing behavior of Drosophila, which stimulates them to climb up the walls of vials after they are tapped down. In this assay, flies were placed into glass cylinders, which were tapped on a table, and the number of flies above the 10.5 cm goal height after 30 s was recorded (Fig. 2b). Both Adar E374A and Adar 5G1 flies could barely climb and almost never reached the goal height in the 30 s time periods, whereas wild type flies usually did (Fig. 2b). In addition, similar to the Adar 5G1 null mutants 24 , Adar E374A mutants undergo heatsensitive paralysis and they also do not produce progeny. Therefore, our results demonstrate that fly Adar function in the central nervous system is editing-dependent.
Catalytically inactive Adar impedes rescue by active Adars. Open field locomotion defects and neurodegeneration in Adar 5G1 null mutant flies are fully rescued by expression of a UAS-Adar construct expressing the main adult wild type Adar cDNA isoform under the control of ubiquitously expressed GAL4 drivers 30 . One of the RNA editing events lost in the Adar E374A mutant is at the Adar S/G editing site where editing leads to replacement of a serine residue by glycine in the RNA-binding loop on the Adar deaminase domain 23,30-32 . The genome encoded, unedited Adar S isoform is eightfold more active than the Adar G isoform in assays on RNA editing substrates in vitro 30 and is also less active in comparisons between two fly strains genome-engineered to express the Adar S and Adar G proteins individually 33 . Because the Adar E374A mutant strain produces only the Adar E374A, S isoform, we wished to test the rescue of locomotion in the Adar E374A mutant when expression of catalytically active Adar S is restored by GAL4/UAS-controlled expression from cDNA (Fig. 2a). Expressing the more active, genome-encoded Adar S unedited isoform under actin-5C-Gal4 driver control is lethal with increased editing levels in embryonic and larval transcripts so we instead used the cholinergic neuronspecific ChAT-Gal4 driver (choline acetyltransferase promoter-GAL4 previously named Cha-Gal4) which drives expression in the numerous brain cholinergic neurons that express the acetylcholine synthesizing enzyme choline acetyltransferase 30 . We note that Adar 5G1 is close to being fully rescued by ChAT-Gal4, UAS-Adar 3/4S (i.e. ChAT > Adar 3/4S) expression while Adar E374A is partially rescued (Fig. 2a, b).
The partial rescue of Adar E374A could be due to a dominant negative, interfering effect of having an inactive AdarE374A protein present that may reduce access to editing sites in target RNAs 34,35 . We wished to test the hypothesis that endogenous Adar E374A protein acts as a competitor to the GAL4/UASexpressed catalytically active Adar protein by using RNAi to prevent expression of the catalytically inactive protein (Fig. 2a). This experiment is possible because Adar 5G1 null mutant locomotion defects are also rescued by expressing human ADAR2 22 , which has a different sequence from Drosophila Adar. Therefore, we expressed human ADAR2 under ChAT-GAL4 driver control in the Adar E374A strain, with or without an additional UAS-Adar RNAi transgene driving knockdown of the Drosophila Adar transcript and measured open field locomotion (Fig. 2a). It is clear that human ADAR2 rescues Adar E374A mutant locomotion defects better when Adar E374A transcripts are reduced by RNAi. The locomotion rescue is still not as strong as in Adar 5G1 ; ChAT > hADAR2; however, RT-qPCR analysis of Adar transcripts shows that the Adar E374A RNAi knockdown is only fifty percent complete (Fig. 2c). In addition, the presence of a second UAS construct to express dsRNA against Adar transcript for the knockdown lowers the effectiveness of the GAL4 driver and gives lower ADAR2 protein expression (Fig. 2d). Nevertheless, these results provide clear evidence that inactive ADAR exerts a negative effect on the editing-dependent locomotion function.

Loss of Adar activity causes age-dependent neurodegeneration.
We aged male Adar E374A mutant and wild type w 1118 flies for up to 30 days and prepared haematoxylin and eosin stained, five microns thick paraffin wax head sections from ten flies for each group to visualize neurodegeneration. All the Adar E374A mutant head sections show large vacuoles in various regions of the brain and deterioration in the retina that are not observed in aged wild type flies (Fig. 3a-f). In Adar 5G1 the neurodegeneration develops first in the retina and in the mushroom body calyces and later in the rest of the brain 22 ; in Adar E374A the degeneration may be   b Left: Depiction of the negative geotaxis assay, in which flies are placed in a cylinder, the cylinder is tapped on the table and the number of flies that are above the 10.5 cm mark after 30 s is recorded. Right: The climbing success rate, or percentage of flies that were above the 10.5 cm mark 30 s after the start of the negative geotaxis assay, for the Adar 5G1 and Adar E374A mutants, and for these mutants expressing Adar 3/4S under the control of the ChAT-Gal4 driver. p-values were calculated by a one-way ANOVA followed by Tukey's test (c) Adar transcript expression in wild type w 1118 , Adar E374A , and Adar RNAi knockdown flies Adar E374A ; ChAT > Adar RNAi and Adar E374A ; ChAT > hADAR2; Adar RNAi flies that express both Adar RNAi and hADAR2 under ChAT-GAL4 control. p-value was calculated by Student's t test. d Immunoblot detection in head protein extracts of ADAR2 protein Adar E374A , ChAT > hADAR2 and when an additional UAS transgene construct is present in Adar E374A ; ChAT > hADAR2, Adar RNAi. Quantitation of the immunoblot data is shown below; levels of ADAR2 compared to tubulin in each of the different head protein extracts. p-value was calculated by Student's t test. For all the above panels *p-value < 0.05. **p-value < 0.01. ***p-value < 0.005: n.s-not significant. Error bars: SEM (Standard Error of Mean for biological replicates).
slightly less severe and less focused in the mushroom body calyces. Therefore, neurodegeneration results largely from loss of Adar RNA editing activity when Adar expression levels are near normal, although slight changes in the neurodegeneration in Adar E374A may be due to still having the inactive Adar protein present. We had previously observed suppression of neurodegeneration in Adar 5G1 ; ChAT > UAS-Adar EA expressing catalytically inactive ADAR from a UAS-Adar EA cDNA transgenic construct under ChAT-GAL4 driver control (Fig. 3i, j). The UAS-Adar EA transgenic construct expresses a catalytically inactive protein identical to the predominant adult isoform expected to be expressed from the Adar E374A chromosomal mutant. Even though UAS-Adar EA expression is driven specifically in cholinergic neurons, when transcripts are quantitated in whole fly heads, the transgenic Adar EA transcript is 3-4 times as abundant as the Adar E374A transcript in the Adar E374A chromosomal mutant or the Adar transcript in wild-type w 1118 , indicating that Adar E367A transcript expression in cholinergic neurons is above normal levels (Fig. 3k). Evidently, expressing more of the inactive AdarE367A protein in cholinergic neurons increases the protection against neurodegeneration.
Aberrant innate immune induction in Adar RNA editing mutants. Next, we hypothesized that, because there are such severe CNS defects in the Adar mutants, there might be changes in the expression of neurological genes. As altering a single amino acid in Adar is a less severe change than removing Adar entirely, the CRISPR Adar E374A mutant allows us to focus on editingdependent changes in gene expression. We performed RNA sequencing of fly heads with a minimum of three biological replicates to examine gene expression differences in the Adar E374A and Adar 5G1 mutants compared to the wild type w 1118 . We observed 228 and 751 differentially expressed genes (DEGs) in Adar E374A and Adar 5G1 respectively (10% FDR, abs(log 2 -FoldChange) > = 0.6, Fig. 4a, b, Supplementary Fig. 2). Approximately half (51%) of the 228 genes affected in Adar E374A are also affected in Adar 5G1 , suggesting that they are affected by loss of RNA editing. Gene expression changes in Adar 5G1 only are likely a response to the loss of Adar protein, but not solely due to loss of RNA editing activity. We used DAVID Gene Ontology (GO) term analysis to examine the functions and categories of the most differentially expressed genes in the Adar E374A mutant (Benjamini adjusted p-val < 0.01, abs(log 2 FoldChange) > = 0.6). Although Adar mutants display CNS defects and RNA editing sites are enriched in neuronal genes in Drosophila 8 , we did not observe an enrichment of neurological terms in the genes with increased or decreased expression in the Adar E374A mutant (Fig. 4c, d, Supplementary Fig. 2b, c). However, we did find an enrichment of neurological terms in genes with differential exon usage, indicating that many neurological genes are differentially spliced in ways that depend on Adar RNA editing activity ( Supplementary Fig. 3). Such affects could be due to inosine pairing like guanosine, leading to gain or loss of splice sites or to editing effects on pre-mRNA structure.
Surprisingly, the most enriched term associated with genes with increased expression in Adar E374A was "Immune-induced protein" (DIM, or Drosophila Immune-induced Molecule) (Fig. 4c). Similarly, the DIMs are most enriched among genes with increased expression in Adar 5G1 (Fig. 4d). The 9 DIMs that are significantly overexpressed in Adar E374A are also overexpressed in Adar 5G1 , further suggesting that their upregulation is dependent on the loss of RNA editing. In fact, 5 of the top 10 most enriched terms for genes overexpressed in Adar E374A , along with 7 of the top 10 terms for Adar 5G1 , are related to immune functions (Fig. 4c, d), strongly suggesting that loss of Adar RNA editing induces the expression of immune-related genes in Drosophila. For the genes with increased expression in both Adar E374A and Adar 5G1 compared to the wild type, the enrichment in immune terms is even more apparent (Supplementary Fig. 2a). For genes with decreased expression in both Adar E374A and Adar 5G1 compared to the wild type, we did not observe any enrichment of GO terms with significance. Our RNA sequencing experiments were performed on head samples to identify gene expression changes particularly in neuronal genes. However, the aberrant DIM expression in heads could still be outside of neurons in other cells such as fat cells or digestive system; if this is the case, there may also be strong aberrant DIM expression in the rest of the body. Therefore, we first conducted qRT-PCR to examine the immune induction in whole flies, examining expression levels of the nine DIMs that had significantly increased expression levels in the Adar E374A head RNA-seq data (Fig. 4a, left). All of the 9 DIMs tested had substantially increased expression levels in the Adar E374A mutants compared to the wild type. Despite the high variation in the degree of immune gene induction, we observed a consistent trend of upregulation of all 9 genes, including five that are statistically significant. In addition, we surveyed the expression of six other innate immune genes that were not significantly differentially expressed in the head RNA-seq data, and found that four of them were significantly upregulated in the whole animal of the Adar E374A mutant (Fig. 5a, right). Next, we examined the expression changes in the Adar 5G1 mutant for the nine DIMs and six other innate immune genes (Fig. 5b). Overall, we observed the same trend as we did in the Adar E374A mutant, although the level of gene expression changes may differ for some genes.
DIMs are a broad category of immune genes that includes many uncharacterized or little-studied transcripts. The most studied DIMs are the inducible Anti-Microbial Peptides (AMPs) that kill bacteria. Different AMP transcripts are transcriptionally induced by different pathogens through different defined innate immune signaling pathways. We compared the induction of Mtk (Metchnikowin, normally induced by the Toll-Dif pathway in response to infection with Gram positive bacteria or fungi) in RNA from heads versus whole bodies of Adar mutant flies. We observed that the aberrant induction is much stronger in heads (Fig. 5c), indicating that the aberrant innate immune induction may, after all, be mainly in the brain. Aberrant Mtk induction is much stronger in Adar 5G1 than in Adar E374A , indicating that inactive Adar protein may reduce the aberrant innate immune induction. Similar head-enrichment and weaker induction in the presence of inactive Adar protein is observed for Att-C (Attacin-C, normally induced through IMD-Rel in response to infection with Gram negative bacteria or through Jak-Stat signaling) (Fig. 5d). The Vago transcript which is induced specifically by the distinct antiviral signaling pathway activated downstream of the Dicer-2 innate immune sensor 36,37 , and not by Toll or Imd signaling, is also induced in Adar mutant head RNA (Fig. 5e). We conclude that a set of AMP innate immune transcripts are aberrantly upregulated in fly heads due to the loss of RNA editing; these include AMPs usually activated by either the IMD-Rel or Toll-Dif pathways as well as Vago which indicates aberrant activation through Dicer-2 acting as an antiviral dsRNA-activated innate immune sensor.
With regard to aberrant AMP transcript induction, we had also previously observed that expression of UAS-Adar EA ubiquitously under armadillo-GAL4 driver control suppresses aberrant expression of AMPs as effectively as UAS-Adar wild type expression (Fig. 5f), although TotC (Turandot-C), which is considered more of a stress-induced transcript is not well suppressed. We conclude that a partial inhibition of aberrant Dcr-2 innate immune signaling by AdarE374A protein when expressed at normal levels, becomes much stronger when the catalytically inactive protein is expressed ubiquitously from UAS constructs under arm-GAL4 driver control. This nearly total suppression of innate immune induction by increased inactive Adar (Fig. 5f) is very striking.
Dicer-2 acts as an innate immune sensor in Adar mutant flies. Drosophila does not have a homolog of the vertebrate antiviral cytoplasmic dsRNA sensor Mda5 helicase; the most closely related helicase in Drosophila is Dicer-2, which has been proposed to act as an innate immune sensor for RNAs of some specific viruses 36,38 . To test whether Dicer-2 is involved in the aberrant innate immune induction in Adar mutants we crossed Adar E374A and Adar 5G1 to the Dcr-2 R416X strain. Dcr-2 R416X is a point mutation truncating the Dicer-2 protein within the third Hel2 subdomain of the tripartite N-terminal RNA helicase. We found that Adar mutant locomotion defects are not rescued in the Adar; Dcr-2 R416X double mutants (Fig. 6a). In fact, AMP transcript expression is increased rather than decreased in these double mutants (Fig. 6b) compared to the corresponding Adar mutant, especially in the Adar 5G1 null mutant background.
We considered the possibility that the Dicer2 R416X mutant produces a truncated protein having the Dcr-2 Hel1 and Hel2i dsRNA-activated helicase subdomains 26,39 that might still function as a dsRNA sensor. This truncated protein would lack most of Dicer-2, including the RNAseIII domains, and would be defective for siRNA processing so that long dsRNA would increase in this mutant. If Dcr-2 (1-416) still has dsRNA sensor activity then the increased long dsRNA could cause the increased innate immune induction in the Adar 5G1 ; Dcr-2 R416X double mutant. Intriguingly this increased immune induction is much weaker in the Adar E374A ; Dcr-2 R416X double mutant, suggesting that it is prevented by catalytically inactive Adar protein. To test this idea, we knocked down Dcr-2 by RNAi in cholinergic neurons under ChAT-GAL4 driver control; The knockdown of Dicer-2 transcripts in head RNA is 40-60% complete (Supplementary Fig. 4a), but cholinergic neurons are only a proportion of the cells in a fly head so the Dicer-2 knockdown in cholinergic neurons must be much more complete. AMP transcripts are significantly reduced in Adar E374A ; ChAT > Dicer2 RNAi and Adar 5G1 ; ChAT > Dicer2 RNAi heads (Fig. 6c). We conclude that aberrant innate immune induction in Adar mutants is likely to involve innate immune signaling from Dcr-2.
Signaling downstream of activated Dicer-2 is known to activate expression of Vago which acts as an interferon-like secreted antiviral signaling protein activating the Jak-STAT pathway 37,40 . Vago expression is significantly reduced in Adar E374A ; ChAT > Dicer2 RNAi and Adar 5G1 ; ChAT > Dicer2 RNAi heads (Supplementary Fig. 4b). The AMPs aberrantly induced in Adar mutants, like those induced during virus infections, include a small subset of those normally induced by bacterial or fungal infections. These include Att-C that is typically activated by JAK-Stat signaling, Dpt (Diptericin) that is usually activated by the IMD-Rel pathway, and Drs (Drosomycin) that is usually activated through the Toll-Dif pathway (Fig. 6c). Aberrant AMP induction in Adar mutants involves the Jak-STAT pathway as Adar E374A ; ChAT > Stat93E RNAi flies having RNAi knockdown of Stat93E showed reduced aberrant expression of Attacin-C, Metchnikowin and Drosomycin transcripts; parallel RNAi knockdowns of Rel or Dif did not reduce expression of these same AMP transcripts ( Supplementary  Fig. 5).

Discussion
We generated and characterized a Drosophila chromosomal Adar E374A mutant line expressing a catalytically inactive Adar enzyme in order to separate the editing-dependent functions of Adar from the editing-independent functions. Drosophila has only one Adar gene and no Adar RNA editing remains in the Adar E374A CRISPR mutant. We show that all aspects of the Adar E374A mutant phenotype are similar to those described in the Adar 5G1 null mutant. Furthermore, Adar E374A and Adar 5G1 flies have defects not only in CNS but also in innate immunity, the two systems most affected by Adar mutations in vertebrates. In particular, Adar mutant flies show Dicer-2-dependent aberrant innate immune induction analogous to that caused by unedited dsRNA activating the innate immune sensor MDA5 in Adar1 mutant mice. Therefore, the single fly Adar enzyme has dual functions achieved by Adar1 and Adar2 in mammals, and these functions are dependent on RNA editing activity when Adar is expressed normally from the chromosomal Adar locus. Since expressing human ADAR2 in Adar 5G1 mutant flies rescues all fly Adar mutant phenotypes tested it seems likely that the involvement of Drosophila Adar and not mouse Adar2 in innate immune responses is probably not mainly due to differences between these homologous proteins in vertebrates and flies but due to the particular set of RNAs edited by Adar in Drosophila. Association of ADAR1 type proteins with innate immunity and ADAR2 type proteins with neurological   functions as in vertebrates may have been the original pattern in invertebrates. However, ADAR1 was lost during evolution to insects so ADAR2 proteins may have increased activity on a class of innate immunity-related targets like those edited by ADAR1 in vertebrates and other invertebrates, such as long dsRNAs formed by repetitive sequences.
We observed that although, in general, immune gene induction is seen in both Adar E374A and Adar 5G1 mutants, the degree of induction in whole flies differs between the mutants, suggesting editing-independent Adar effects. For example, the Anti-Microbial Peptide (AMP) transcript inductions are clearly stronger in Adar 5G1 than in Adar E374A , suggesting that the presence of catalytically inactive Adar protein may impede the immune induction. The Adar editing-independent effects are most clearly seen when the catalytically inactive Adar EA is overexpressed and is able to rescue some Adar mutant phenotypes. For example, neurodegeneration (Fig. 3i, j) and aberrant innate immune induction (Fig. 5f) in the Adar 5G1 mutant strain are rescued by overexpressing UAS-Adar EA cDNA constructs using GAL4 drivers. However, this type of rescue experiment involves Adar transgene expression at greater than physiological levels. Nevertheless, expressing UAS-Adar EA cDNA constructs using brain GAL4 drivers did not rescue locomotion defects, consistent with the idea that edited isoforms of CNS proteins are absolutely required for Adar-mediated rescue of locomotion defects. Future work is needed to identify the key editing events that are required for normal locomotion 25,28,41 .
Our gene expression analysis reveals an unexpected function of fly Adar, which is to suppress innate immune induction. Although the only fly Adar is most similar to mammalian ADAR2 in terms of both sequence identity and neurological phenotypes, it appears to have an innate immune role in flies similar to that of ADAR1 in vertebrates. Adar 5G1 aberrant innate immune induction is likely to result from accumulated unedited intracellular dsRNA in Adar 5G1 , paralleling the mouse Adar1 mutant interferon induction through antiviral dsRNA sensors 18 , and the human ADAR1 mutant virus infection mimic syndrome Aicardi Goutières Syndrome 42 and ADAR1 mutant bilateral striatal neurodegeneration 43 . In our initial attempt to test the role of Dicer-2 in Adar mutant innate immune induction the Dcr-2 R416X mutation did not suppress AMP induction in the Adar; Dcr-2 R416X double mutants (Fig. 6b). However, our finding that AMP expression was higher in Adar; Dcr-2 R416X double mutants than in the corresponding single Adar mutant led us to test the effects of Dicer-2 RNAi knockdown instead. The Dicer-2 RNAi knockdown experiments showed that the immune induction in the fly Adar mutants is mediated by Dicer-2 (Fig. 6c), which has a helicase domain closely related to that of MDA5; these helicases may act as ATP-activated translocases on dsRNA 26 (Fig. 6c). Similar to the vertebrate antiviral dsRNA sensors which are inhibited by dsRNA containing inosine 18,44 , ADAR editing of dsRNA inhibits Dicer-2 cleavage in vitro 45 , and may also inhibit innate immune signaling from Dicer-2. A-to-I RNA editing can introduce inosine-uracil wobble base pairs, which weaken dsRNA strand pairing; MDA5-like sensors dissociate more rapidly from imperfectly-paired dsRNAs [46][47][48] . It is surprising that the Dicer-2 (1-416) protein still mediates innate immune induction but it retains the Hel1 and Hel2i subdomains of the helicase domain and their potential dsRNA contacts; and it may also have interactions with other proteins that help it recognize dsRNA. The Hel2i subdomain of helicase DICER proteins is a hotbed of interactions with dsRBD proteins in, for example, human DICER interacting with specialized dsRBDs of TRBP and PACT and with dsRBD3 of human ADAR1, which affects microRNA and siRNA processing 14 . When expressed in Drosophila, human ADAR1 strongly inhibits Dicer-2-mediated RNA silencing through siRNA production, even when catalytically inactive 9 and it may also inhibit innate immune signaling by Dcr-2. We show here that Drosophila Adar EA also strongly blocks Dcr-2 innate immune signaling (Fig. 5f). Failure to rescue Adar mutant aberrant innate immune induction in the Adar; Dicer2 416× double mutant is surprising because this Dcr-2 mutant has been reported to suppress aberrant innate immune induction by long dsRNA overexpressed in Drosophila 49 . We do not know the reason for this difference; it may be due to different dsRNAs driving the immune induction or to different levels of dsRNA expression.
An additional way whereby innate immune induction could contribute to all Adar 5G1 mutant phenotypes is described in a related manuscript 50 . That manuscript shows that reduced autophagy in Adar 5G1 leads to synaptic defects with aberrant accumulation of neurotransmitter synaptic vesicles and associated proteins. In mammalian cells innate immune induction impedes canonical autophagy by diverting the p62 receptor for ubiquitinated proteins to instead form a cytoplasmic innate immune signaling platform 51 ; this cross-regulation of p62 by innate immune signaling helps to redirect autophagy to innate immune defense when required. Dicer-2-mediated aberrant innate immune induction in Drosophila may also impair autophagy in flies. Determining to what extent all the Adar mutant phenotypes arise from aberrant AMP expression leading to neuronal damage or from antiviral innate immune impedance of autophagy, leading to synaptic defects in neurons, requires Adar 5G1 ; Dicer-2 null double mutants to block all immune induction; however, the work described here shows that no such Dicer-2 null allele has been characterized yet.
Given the evolutionary expansion of site-specific RNA editing events in Drosophila, it one would presume that Adar mutant phenotypes would be entirely due to loss of RNA editing in CNS transcripts and to loss of the edited isoforms of encoded proteins. Therefore, it comes as a surprise to find that an important aspect of the Adar mutant phenotype is still an aberrant innate immune induction by unedited dsRNA. This surprise also adds renewed interest to the question of how much of the Adar null mutant phenotype is solely due to loss of edited isoforms of so many CNS proteins. It was already surprising that Adar null mutants flies RT-qPCR on RNA from 2-3 days old whole flies. a Normalized expression levels for the Adar E374A mutant (red) relative to wild type (gray) for DIMs that are significantly upregulated in the RNAseq data (left) and other immune genes (right). b Normalized expression levels for Adar 5G1 (orange) relative to wild type (red) for the same genes as in (a). The y-axis is discontinuous; the break in the y-axis represents the jump in the y-axis scale. Below the break, the y-axis scale is the same as in (a). p value was calculated by Student's t test *p-value < 0.05. **p-value < 0.01. ***p-value < 0.005. Error bars: SEM (Standard Error of Mean for biological replicates). c Metchnikowin transcript expression in Adar mutant head or whole body RNA versus in the corresponding wild type RNA sample. d Attacin-C transcript expression in Adar mutant head or whole body RNA versus in the corresponding wild type RNA sample. e Aberrant Vago transcript expression in Adar mutant heads. (c-e) levels of each AMP transcript in each of the different whole fly RNA samples is compared to that in w 1118 . f Expression of either wild type Adar protein or catalytically inactive Adar E374A protein ubiquitously under armadillo-GAL4 driver expression in Adar 5G1 mutant flies strongly suppresses aberrant AMP transcript expression in whole fly RNA. p-value was calculated by Student's t test *p-value < 0.05. **p-value < 0.01. ***p-value < 0.005: n.s-not significant. Error bars: SEM (Standard Error of Mean for biological replicates).  are viable, suggesting that the RNA editing events have been evolutionarily selected so that basic neuronal functions do not depend on the edited protein isoforms. The new findings now suggest that edited protein isoforms fulfill even more specific roles 25,52,53 that will need to be defined within specific neuronal circuits.

Methods
Fly stocks and sample collection. All stocks were grown on standard fly media. Adar 5G1 ("Adar null") flies are from 31  For the locomotion assays (negative geotaxis and open field locomotion), flies were kept in 12 h/12 h light/dark cycles at 25°C. Replicate experiments were performed at the same time of day. Flies were not exposed to carbon dioxide for at least 24 h before the assays.
Adar E374A mutant generation. The Adar E374A fly line 28 was generated with plasmid and nos-Cas9 fly line CRISPR reagents as described in 54 . sgRNA target sequence CCTGGAGACTATTTCAGCAT was inserted into the U6b-sgRNA-short plasmid and injected into the nos-Cas9 flies (attP2 insertion line) by BestGene. The sgRNA plasmid was injected at 75 ng/ul, and the donor oligo (a ssODN with 60 base homology arms on each side of the mutations C1725T and A1733C, depicted in Fig. 1a) was injected at 750 ng/ul in water. The flies were then crossed multiple times to an FM7a balancer stock (Bloomington #785) to isolate the mutants and then backcrossed for 7 generations to white Canton S (Shohat-Ophir lab stock).
Negative geotaxis assay. Three to four-day-old male flies were placed in two glass cylinders with a fly vial plug at the bottom. The following was repeated five times for each set of flies: The cylinders were banged onto a mouse pad and after 30 s, the number of flies above the 10.5 cm mark was counted. Then there was a 30 s rest period. The average of the percentage of flies above the 10.5 cm mark over the five trials was the "climbing success rate". 14-38 flies were examined for each genotype.
Open field locomotion assay. Two-day-old individual male flies were placed in 35 mm petri dishes divided into eight sections by radii, with an additional center circle. The number of lines crossed in 2 min was recorded three times for each fly, and these values were averaged per fly. 15 flies were examined for each genotype.
Fly ageing and Drosophila head sections and staining. Flies were aged and then fixed in Carnoy's fixation solution (30% chloroform; 10% acetic acid; 60% absolute ethanol) for 3.5 h and flies were then placed in 95% ethanol for 30 min twice and then they were placed in 100% ethanol for 45 min. Flies were kept in methyl benzoate overnight. On the next day, flies were kept in solution of methyl benzoate and paraffin both in equal volume at 60°C for 1 h, then transferred to fresh liquid paraffin for 1 h twice at 60°C and then kept in liquid paraffin overnight at 60°C. Flies were transferred to a plastic petri dish with fresh liquid paraffin at 60°C and then allowed to cool down. Paraffin blocks were made from this. Sections of 5-μm thickness were taken using a LEICA RM 2235 rotary microtome and then dried overnight on the slide warmer at 37°C and then processed for hematoxylin and eosin staining.
Gene expression and splicing analyses. Reads and mapped reads are as in 28 (GEO GSE86056). DESeq2 55 were used to obtain gene expression levels. DEXSeq 56,57 was used to examine differential exon usage. To determine the editing event, we required the editing level to be ≥2% in at least one of the datasets. 89% of the editing sites have significantly higher editing levels with this cut off than the typical A-to-G sequencing error rate. Editing levels were only used if there were at least 20 reads at the editing site.
GO term analysis. DAVID was used for the GO term analysis. The lists of genes tested by DESeq2 (the genes for which the adjusted p-value was not NA) were used as the background. The GO terms used are from these lists (DAVID default): COG_ONTOLOGY, UP_KEYWORDS, UP_SEQ_FEATURE, GOTERM_BP_ DIRECT, GOTERM_CC_DIRECT, GOTERM_MF_DIRECT, KEGG_PATHWAY, INTERPRO, PIR_SUPERFAMILY, SMART. P-values were corrected for multiple hypothesis testing using the Benjamini method, and only the top statistically significant GO terms (adjusted p < 0.01) are shown.
qPCR. cDNA was generated with iScript Advanced (Bio-Rad) or RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific). qPCR reactions using the KAPA-SYBR FAST qPCR kit (Kapa Biosystems) or LightCycler® 480 SYBR Green I Master mix and the primers listed in Supplementary Table 1 were used to measure expression levels on a Bio-Rad CFX96 Touch Real-Time PCR machine or The LightCycler® 480. All primers were tested to ensure the efficiency levels. Expression levels were normalized to those of RP49 and t-tests were used for statistical analysis.
Immunoblotting. Heads of male flies (minimum 20 flies) of the desired genotype were collected and aged for 2 days and then homogenized in NB Buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.5, 2 mM EDTA, 0.1% NP-40). Protein concentration was determined with Pierce BCA Protein Assay Kit. An equal amount of protein was loaded in each lane of a Tris-Glycine SDS Gel and transferred to a nitrocellulose membrane. The membrane was blocked with 5% Milk, incubated with primary antibody overnight. The next day it the membrane was incubated with secondary antibody and developed with Clarity™ Western Blotting Substrate from Bio-Rad. Anti-ADAR2 (HPA018277) (1:250) anti-Tubulin (Developmental Studies Hybridoma Bank) (1:5000). Imaging was performed with ChemiDoc™ XRS + System and signal intensity was quantified with Image J software and Statistical analyses were done using the t-test.
Statistical analysis. qPCR data and immunoblot data were analyzed by Student's t test. A p-value of <0.05 was considered statistically significant. Locomotion and Climbing data p-values were calculated by a one-way ANOVA followed by Tukey's test. The significance of differences between variables was described based on p-values *p-value < 0.05. **p-value < 0.005. ***p-value < 0.001 and n.s (not significant). Error bars: SEM (Standard Error of Mean for biological replicates).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that supports the findings of this study are available from the authors upon request. The source data underlying Figs