To test whether the preference to consume ethanol in flies underlies similar pharmacological properties as relapse behavior in humans, we tested whether acamprosate, a clinically approved medication for alcohol relapse prevention (19), is able to reduce ethanol consumption preference after abstinence in flies. To measure voluntary ethanol intake, we used a capillary feeder assay (Figure 1A). Flies preferred to feed on 15% ethanol enriched with 5% sucrose for five consecutive days (Figure 1B). On day 6, one-half of the flies continued to have a choice between different sucrose and sucrose + ethanol, whereas the experimental groups had access only to 5% sucrose for 24 h. After abstinence, and the re-introduction of ethanol, the experimental group significantly preferred to consume ethanol. They did not differ in their preference to controls that did not undergo abstinence (Figure 1B). Since in humans, steady-state concentrations of acamprosate is reached only after 5 to 7, days (20), Acamprosate was added to both solutions from days 1 to 7 (Figure 1C). During the first 5, days, the acamprosate-treated flies had similar levels of fluid intake (controls: 3.59, ±, 0.14, μL per fly and the experimental flies: 3.63, ±, 0.13, μL per fly). After abstinence, the acamprosate-treated flies significantly decreased their ethanol preference compared with that of the control flies and their own baseline ethanol consumption preference on days 1 and 5 predeprivation (Figure 1C). Acamprosate reduced preference to consume ethanol-containing food, similar to what is seen in comparable animal models and humans. Thus, these data provide pharmacological validity to the Drosophila model of alcohol relapse.

To analyze whether we can identify mutants with selective defects in their ethanol consumption preference after abstinence, we analyzed hang mutants for their preference to consume ethanol-enriched food over non-ethanol-containing food after abstinence (Figure 2). The hang gene defines a major cellular stress pathway underlying behavioral responses to repetitive ethanol exposure (10). During the initial 5 days, hang^AE10 flies consistently preferred ethanol-containing food. After deprivation, on day 7, the deprived group significantly reduced their ethanol intake in comparison with hang mutants that did not undergo a period of abstinence. The preference of the deprived hang flies was also significantly lower than the preference at the start of the experiment and before deprivation. Thus, hang mutants show a selective defect in ethanol preference after abstinence.

Next, we wanted to identify novel players underlying the regulation of ethanol preference after abstinence. We thought to identify these genes in a set of preselected gene loci implicated in alcohol use disorder by GWAS. As entry point, we focused our analysis on the GWAS performed by Treutlein et al. (5). First, we selected candidate genes that had been tagged by SNPs identified with AUDs by GWAS and in follow-up studies [see Table 1 in Treutlein et al. (5)]. The list consists of 16 SNPs; 12 of which are localized within introns (one SNP is associated with the introns of two genes and two additional SNPs are inserted in the intron of one of the genes without affecting the other) and four are between genes. In total, the SNPs are associated with 19 different genes (Supplementary Table 3). Two genes that encode a nuclear RNA and a pseudogene were excluded from the analysis. For the remaining 17 human candidate proteins, we performed a BLASTP analysis with the Drosophila melanogaster proteome. We further excluded calpastatin, interferon regulatory factor 5 (IRF5), and centrosomal protein 83 (CEP83) because we did not detect any similar proteins in Drosophila. In our analysis, we also included the Ku80, the Drosophila homolog of human XRCC5, because previous studies implicated Ku80 in the susceptibility to AUDs and because previous studies implicated Ku80 in cellular stress responses (8, 21).

Since four out of the remaining 14 genes are known to be involved in the regulation of cellular stress (Supplementary Table 3) as well as the hang (10) and Ku80 (21), we focused our analysis on the following candidate genes: peroxisomal trans-2-enoyl-CoA reductase (PECR), serine/threonine-protein phosphatase 2A, 55 kDa regulatory subunit B beta isoform (PPP2R2B), estrogen receptor 1 (ESR1), glycerol-3-phosphate dehydrogenase 1-like (GPD1L), and XRCC5. The Drosophila CG10672/dehydrogenase/reductase 4 protein is with 35% amino acid identity the most similar to human PECR; thus, we named this candidate gene from here on dPECR. The Drosophila Twins is the most similar to the human PPP2R2B protein, with 79% amino acid identity. The most similar protein to the human ESR1 protein in Drosophila is the estrogen-related receptor (ERR), with 35% amino acid identity. Human glycerol-3-phosphate dehydrogenase 1-like (GPD1L) shares the highest homology, with approximately 60% identity, to Drosophila glycerol-3-phosphate dehydrogenase 1 (Gpdh1).

First, we screened for viable homozygote mutants. Next we determined the level of transcripts used qRT-PCR with cDNA isolated from adult Drosophila heads (Figure 2). For five mutant lines, homozygous flies were viable and qRT-PCR analysis performed on polyA-selected head RNA revealed a significant and pronounced reduction in transcripts of the respective candidate genes (Figures 2A–E). However, no viable mutants for Ku80 were available. Therefore, we used RNA interference (RNAi) to reduce Ku80 expression under the control of Appl-Gal4 in a broad set of neurons (Figure 2F). Next, we analyzed the mutant flies to determine their baseline ethanol preference and ethanol reconsumption after deprivation. The control for the respective mutants consumed ethanol after a period without ethanol to a similar extend as flies that did not undergo a period without ethanol (Figures 2G–M). In contrast to the controls, the dPECR and ERR mutant flies showed significantly reduced preference for ethanol after deprivation (Figures 2H,I). The reduction in twins, Gpdh1, or neuronal Ku80 did not alter ethanol reconsumption after abstinence (Figures 2K–M). Whether these candidate genes can be completely excluded from playing a role in regulating ethanol preference after abstinence remains to be determined, since twins mutation showed significantly increased ethanol preference on day 1 compared with day 7 after abstinence, and the knockdown of Ku80 and Gpdh1 might not have been sufficient to alter gene function to influence ethanol consumption preference.

In summary the hang, dPECR, and ERR mutant flies were specifically impaired in terms of ethanol consumption after abstinence and together with the observation that acamprosate reduces ethanol consumption reference these results suggest that dPECR, ERR, and hang regulate relapse-like behavior.

Since the here-studied genes are known to be involved in the regulation of cellular stress [Supplementary Table 3; (22, 23)], we next asked whether changes in stress sensitivity in neurons correlates with changes in ethanol reconsumption after abstinence. Ethanol promotes the formation of free radicals that cause cellular stress and in the end cell death that might affect the survival of the organism (24). Therefore, we analyzed whether the reduction in ERR and dPECR function in neurons changes the survival rate in response to the free radical forming agent H₂O₂. We reduced ERR and dPECR using RNAi transgenes in different sets of neurons using different GAL4 drivers and assayed survival in response to oxidative stress (Figure 3). First, to insure that the transgenes are functional, we determined that the expression of the RNAi transgenes UAS-ERR^miRNA and dPECR^HMS00752 under the control of the Appl-Gal4 driver effectively reduced gene expression using qRT-PCR (Supplementary Figure 1). For the survival experiments, the function of ERR and dPECR were reduced in a broad set of neurons using the Appl-Gal4 driver (25). Different types of neurons are differently sensitive to cellular stress. To alter gene function selectively in glutamatergic, taurine-containing neurons, or GABAa receptor-expressing neurons, we used the following Gal4 drivers: the DVGlut-Gal4 driver to target glutamatergic neurons that express the vesicular glutamate transporter (26); the dEAAT2-Gal4 driver to target neurons that express the excitatory amino acid transporter 2 (dEAAT2) and to label taurine-containing neurons (27); and the Rdl-Gal4 driver to target neurons that express the GABA_A receptor resistance to dieldrin (Rdl), a postsynaptic marker for GABAergic signaling (28). Reduction in ERR in a broad set of glutamatergic, dEAAT2-expressing and Rdl-expressing neurons significantly increased resistance to stress caused by oxygen radicals (Figures 3A–D). The reduction in dPECR in a broad set of neurons or glutamatergic neurons did not interfere with survival, but reduction in dPECR in taurine-containing neurons and Rdl-expressing neurons significantly increased resistance to free radicals (Figures 3A–D). Thus, loss of ERR and dPECR protects neurons against oxidative stress and dPECR especially in specific subset of neurons.

Changes in transcript levels in response to ethanol are widely used as markers for genes involved in the regulation of ethanol-induced behaviors. Thus, we determined whether the transcript levels of the selected candidate genes changed upon ethanol exposure in fly heads using qRT-PCR analysis (Figure 4). We focus on difference in the head, since we were interested in detecting difference in genes that might regulate behavior. After the ethanol vapor exposure 4 h later, the transcript level of ERR was significantly increased by approximately 8-fold and the transcript level of twins by approximately 1.7-fold, whereas the transcript level of dPECR was significantly reduced by approximately 68% (Figure 4A).

The hang gene encodes a neuronal nuclear RNA-binding protein that regulates transcript levels (12), and the ERR encodes a nuclear hormone receptor that induces a transcriptional switch (29). To investigate whether hang regulates the transcript of candidate genes, the transcript levels of the candidate genes were determined in the heads of the hang^AE10 mutants (Figure 4B). The transcripts of ERR and twins were not changed, but those of dPECR, Gpdh1, and Ku80 were significantly reduced. To investigate whether these genes regulate hang, the hang transcript level was analyzed in the heads of the respective mutant flies (Figure 4C). In ERR, Gpdh1, and twins mutants, hang was significantly upregulated, whereas in the dPECR mutants, hang was not changed. In summary, ethanol increased the transcript levels of ERR and twins, and both negatively regulated the transcript level of hang; whereas, ethanol suppressed the dPECR transcript level, and hang is a positive regulator for dPECR, Ku80, and Gpdh1 transcript levels. Thus, ERR is ethanol inducible and upstream of the hang transcription or RNA stability or RNA turnover, whereas hang is upstream of dPECR and is a positive regulator of dPECR.

To identify physical interactions of the candidate proteins, we performed a database analysis to identify protein-protein interactions (Figure 4D; Supplementary Table 4). We identified three different groups of interactions at the protein level. The first group contained nuclear proteins such as ERR, Ku80, pannier, and Hang, for which we found indirect protein-protein interactions via one additional shared partner, Sima.

For certain shared partners, genetic and protein interactions have been previously shown (Supplementary Table 4). For example, Ku80 together with Sima/Hif-1α regulates the PDK-1 function required for apoptosis (30), and ERR binds to Sima/Hif-1α to induce the Hif-1α-dependent transcriptional program under hypoxia (23), whereas the interaction of Sima and Hang was identified by coimmunoprecipitation (31). Whether these interactions happen at the same time needs to be further analyzed. Twins belongs to the second group of proteins, which are involved in insulin signaling. The final group contained dPECR and Gpdh1. dPECR interacts with several other proteins before interconnecting with Gpdh1, indicating that dPECR and Gpdh1 have a very indirect interaction. dPECR and ERR were not found in the same interacting protein-protein complex, which is not surprising given the different subcellular locations of these proteins.