Arc controls alcohol cue relapse by a central amygdala mechanism

Alcohol use disorder (AUD) is a chronic and fatal disease. The main impediment of the AUD therapy is a high probability of relapse to alcohol abuse even after prolonged abstinence. The molecular mechanisms of cue-induced relapse are not well established, despite the fact that they may offer new targets for the treatment of AUD. Using a comprehensive animal model of AUD, virally-mediated and amygdala-targeted genetic manipulations by CRISPR/Cas9 technology and ex vivo electrophysiology, we identify a mechanism that selectively controls cue-induced alcohol relapse and AUD symptom severity. This mechanism is based on activity-regulated cytoskeleton-associated protein (Arc)/ARG3.1-dependent plasticity of the amygdala synapses. In humans, we identified single nucleotide polymorphisms in the ARC gene and their methylation predicting not only amygdala size, but also frequency of alcohol use, even at the onset of regular consumption. Targeting Arc during alcohol cue exposure may thus be a selective new mechanism for relapse prevention.


INTRODUCTION
Alcohol use disorder (AUD) is a chronic and, progressive psychiatric disease, with huge costs for afflicted individuals and society, and with limited therapeutic options, that are effective only for some patients [1,2]. Progression of AUD is characterized by periods of abstinence that are followed by alcohol relapses and increasingly longer phases when individuals lose control over alcohol consumption [3][4][5]. As supported by human studies and animal models, alcohol use can be precipitated in abstinent individuals by three main factors that provoke alcohol craving: stress, an alcohol priming dose, or environmental cues that are associated with alcohol availability [5][6][7]. Thus, in order to develop a successful prevention and therapeutic control of AUD progression the neuronal basis of alcohol craving and relapse must be understood [8,9].
Applying animal model of AUD [30], Arc knockout (Arc KO ) mice and local manipulations of the Arc gene in the mouse amygdala with Clustered Regularly Interspaced Short Palindrome Repeats (CRISPR)/Cas9 system [31] delivered by lentiviral vectors we identified AUD-related behaviors that are regulated by Arc. The whole-cell patch-clamp electrophysiology ex vivo was used to characterize Arc-regulated alcohol-induced synaptic plasticity. For translation into humans, we explored the IMAGEN cohort database [32] to investigate a possible association of ARC genetic variations with alcohol-related behavioral variables and amygdala volume in young adults at a time when alcohol consumption reaches its peak in life.

MATERIALS AND METHODS Mice
C57BL/6J mice were purchased from the Medical University of Bialystok, Poland. Arc KO mice were obtained from Dr. Hiroyuki Okuno (Kagoshima University, Kagoshima, Japan) and bred as heterozygotes at Nencki Institute. Mice were 10-week old at the beginning of the experiments. Only female mice were used in the experiments in the IntelliCages, as males are too aggressive for group housing. Animals were housed under a 12/12 h light/dark cycle in standard mouse home cages with ad libitum access to water and food. Experiments were approved by the Animal Protection Act of Poland guidelines and the 1st Local Ethical Committee in Warsaw, Poland (no. 117/2016) in compliance with the ARRIVE (Animal Research: Reporting in vivo Experiments) guidelines. All experiments were planned to reduce the number of animals used and to minimize their suffering.

qPCR for Arc mRNA
The amygdala of Arc WT and Arc KO mice was quickly dissected on ice from the fresh brain, homogenized and stored in RNAlater solution (Invitrogen, AM7020) at 4°C for 24 h and then kept at −20°C till further use. Total RNA was extracted using the RNeasy Mini kit (Qiagen, 74104) according to the manufacturer's recommendations. RNA concentration, quality and integrity were determined using a Nanodrop 1000 (Thermo Scientific) and a Bioanalyzer (Agilent). QuantiTect Reverse Transcription Kit (Qiagen, Cat. No. 205314) was used to quickly prepare gDNA for gene expression analysis. Arc was quantified by using the QuantiFast SYBR® Green PCR Kit (Qiagen, Cat No. 204054). The recommended cycling program from this kit was followed for the amplification. Arc primer was bought from Qiagen (QuantiTect Primer Assay, Arc: QT00250684). As a control, Gusb expression was analyzed (QuantiTect Primer Assay, Gusb: QT00176715). The amplification was conducted with the Step One Plus real-time system (Applied Biosystem). To analyze the fold Arc expression the delta-delta Ct method was used [33].

IntelliCage behavioral training
Female mice were trained for long-term alcohol drinking in the IntelliCage System (NewBehavior; https://www.tse-systems.com/) based on a previously established protocol [30] (sample sizes estimated from previous studies). Animals were briefly anesthetized by isoflurane and subcutaneously injected with RFID transponder tags which allow for their identification in the IntelliCage corners. Mice underwent a week of cage adaptation to habituate to the light/dark cycle and cage conditions. During the cage adaptation period (d1-7) mice had unlimited access to the corners, food and water in the corners. The novelty seeking behavior was used as an anxiety-related trait [30]. The number of exploratory visits to the corners was measured during the first hour of the cage adaptation. Next, mice underwent a nosepoke adaptation (d3-7) under a fixed ratio of reinforcement (FR 1), when each nosepoke in the corner was rewarded by 5-second access to water. When all animals learned how to reach the bottles inside the corners, a free alcohol access period started (d7-47) with alcohol in increasing concentrations (4,8,12,16, 20%, 3 days each) and 12% during the rest of the training. Alcohol availability was signaled by the cue light presented each time a mouse entered the corner. Each time when a mouse entered the alcohol corner a cue-light was on, serving as an alcohol-predicting cue. Each nosepoke to the alcohol corner gave access to alcohol for 5 s. Daily alcohol consumption (g/ kg/ day) was calculated with the formula: Number of licks of alcohol per day × lick volume (1.94 μl) × % alcohol × 1 g/ ml)/mouse weight AUD-related behaviors (d47-97). During a motivation test (d47-53) mice were subjected to an increasing fixed-ratio (FR) schedule to have access to the reward corner. Mice had to perform an increasing number of nose pokes during a visit to the reward corner to have access for 5 s to the bottles. The number of required nose pokes increased (2, 4, 8, 12, 16, 20 …) when a mouse repeated the task 10 times. The FR level reached by a mouse during the test was used as an index of motivation. During the withdrawal test (d60-66) the reward corner was inactive. During this period the cue-light associated with the alcohol was off and no alcohol was available. Alcohol seeking was calculated as an average daily number of nose pokes to an alcohol corner. The test was followed by a cue-induced relapse (d66) during which the alcohol-predicting cue light was presented, but no alcohol was available. Relapse was assessed as a number of nose pokes to an alcohol corner during the first 6 hours of the test. During the alcohol relapse (d67) mice had full access to the reward corner: the cuelight was on and alcohol was present. Alcohol relapse was measured as the amount of consumed alcohol (g/kg/day). The persistence test (d68-71) was divided into two phases, changed every 6 h. The active phase (A)-mice had access to alcohol signaled by the cue light. During the non-active phase (nA), mice had no access to alcohol. Persistence was calculated as a change of nose pokes number to alcohol corner during nA as compared to A (nA-A).
To calculate Addiction index mice performance in all tests was assessed. Mice were labeled as positive in a test if they performed in the top 30% of the population. Addicted drinkers were positive in at least two tests. Nonaddict drinkers were positive maximum in one test. To calculate the individual Addiction score each scores for all tests were normalized and summed up according to the formula: AS =Σ(Vi (individual score) − mean(population))/SD(population)

Western blot
The hippocampus and amygdala of Arc WT and Arc KO mice were quickly dissected on ice from the fresh brain. Tissue was collected in the RIPA lysis buffer (Santa Cruz, sc-24948). After centrifugation the supernatant was stored at −80°C until further analysis. Equal amounts of protein lysate were mixed with the Laemmli buffer and left to denature at 70°C for 10 min. The mixture was loaded on precast gel wells (Bio-Rad, #4568083). The membranes were blocked by 5% milk diluted in TBST (Tris-buffered saline with Tween 20), and incubated with anti-Arc antibody at 1:1000 overnight (o/n) (SYSY156 003; RRID: AB_887694). Anti-Gapdh at 1:50000 o/n (Millipore, MAB374; RRID:AB_2107445) was used as internal control for normalization. The membrane was washed in TBST and incubated with the horseradish peroxidase (HRP) secondary antibodies at 1:5000 (anti-rabbit IgG-HRP, Santa cruz, sc-2030, RRID:AB_631747; mouse-IgGκ BP-HRP, Santa cruz, sc-516102, RRID:AB_2687626). The membrane was visualized by G-Box apparatus using chemiluminescent reagent (Advansta, K-12042-D10).

Immunostaining analysis
The staining was analyzed by SP8 confocal microscope (Leica) using the 63 X objective with 1.66 increasing zoom. Single scan was taken into the analyzed brain structures with a resolution of 148.36 µm × 148.36 µm. Six to eight microphotographs were taken per animal, from every sixth section through the dorsal hippocampus and amygdala. The mean gray value of the images (arbitrary units), density and size of Arc+ puncta were analyzed by ImageJ (1.48 v) software while the data analysis was performed using scripts in Python (3.4.4 v). Hippocampal dissociated cell culture immunostaining was analyzed by All-in-One Fluorescence Microscope BZ-X800 (Keyence) using the 20 X objective. For each transfected neuron the mean gray value was analyzed by ImageJ (1.48 v) software.

Arc gene methylation in mice
The amygdala of alcohol-naïve and alcohol trained C57BL/6J mice was quickly dissected on ice. Samples were immediately frozen in liquid nitrogen. Genomic DNA was extracted by Extractme genomic DNA Kit (Blirt, EM13) following the suggested protocol. DNA was analyzed by EpigenDx (96 South Street Hopkinton, MA 01748, www.epigendx.com) for methylation analysis by Targeted NextGen Bisulfite Sequencing (tNGBS).

Arc KO generation using CRISPR/Cas9 system
The 20-bp sequences directly upstream of 5′-NGG sequences (PAM sequences) in the coding region of Arc gene were identified (Supplementary Materials and Methods). Four sequences with the lowest probability of off-targets and high on-target efficiency were chosen. LentiCRISPR v2 plasmids expressing the CRISPR/Cas9 system with Flag-tag [34] or GFP reporter protein were used (Addgene plasmid #52961; RRID: Addgene_52961; plasmid #82416; RRID: Addgene_82416).
To clone the gRNA into the lentiCRISPR vectors, the cloning protocol of Zhang Lab was used (media.addgene.org/data/plasmids/52/52961/52961-attachment_B3xTwla0bkYD.pdf). Efficiency and specificity of mutation was validated in vitro and in vivo. Five gRNAs for Arc coding sequence were designed and tested to check the efficacy in the depletion of Arc protein expression. gRNA1-2 were designed using http://crispr.dbcls.jp/, gRNA3-4 using https://chopchop.cbu.uib.no/, gRNA5 is from GenScript (cat. SC1678). gRNAs were cloned into plentiCRISPRv2_Flag [34]. The same plasmids without gRNA (empty vector, EV) were used as a control for in vivo experiments. The plasmids were used to generate lentiviral vectors (LV), Arc Crispr and EV, at Nencki Institute core facility Animal Models lab. For in vitro studies, as Flag control plasmid (Ctrl), the Flag tagged KASH fragment [35] was subcloned into the pcDNA3 vector, resulting in the pcDNA3-5x Flag-KASH (ThermoFisher). As GFP control (Ctrl), a monomeric version of EGFP was subcloned into a CAG promoter vector to make pCAG-mEGFP.

Validation of CRISPR/Cas9 system in vitro
The plasmids expressing gRNAs 1-5 were tested in the mouse hippocampal dissociated cell culture. Neurons were transfected with lipofectamine 2000 (Invitrogen, cat. 11668-027) on the 10 th day in vitro (DIV) and after 4 days, tetrodotoxin (TTX, 2 μM, Tocris, Cat. No. 1078) was added to the medium to block neuronal activity. After one day, cells were stimulated with bicuculline cocktail (bicuculline 30 µM (Tocris, Cat. No. 0109), 4AP 100 µM (Sigma-Aldrich, Cat. No. 275875), glycine 100 µM (Sigma-Aldrich, Cat. No. 50046), strychnine 1 µM (Sigma-Aldrich, Cat. No. S0532) to disinhibit cells and induce Arc protein expression. After the 3-h stimulation, cells were fixed in 4% PFA and immunostained to detect Arc protein and GFP/Flag reporters. The in vitro experiments were replicated three times.To confirm the mutation of the Arc gene by CRISPR/ Cas9 system with gRNA1, mouse cortical dissociated cell culture was transduced on DIV5 with LVs coding plentiCRISPRv2 with GFP and gRNA1 (LentiCRISPRv2GFP, plasmid #82416, Addgene). After 2 weeks, the cells were lysed and genomic DNA extracted by Extractme genomic DNA Kit (Blirt, EM13). DNA was used for PCR amplification of the Arc gene with a set of primers specific for the region containing the sequence complementary to gRNA1 (primer forward: 5′ctctgggcctctctagcttc3′, primer reverse: 5′-cgtccaagttgttctccagc3′). The same procedure was applied to nontransduced cells to obtain non-mutated genomic DNA. The amplicon was sent for Sanger sequencing by Genomed S.A. (Warsaw, Poland). The chromatograms were analyzed with the ICE (Synthego) program which gives NGS quality results with Sanger data.

Electrophysiology
The experiments were performed on female and male adult C57BL/6J and Arc KO mice and their wild-type (WT) siblings. Mice were 1-2 months old at the beginning of the behavioral paradigm and 3-4 months old at the time of electrophysiological recordings. Mice were housed separately in their home cages, receiving food and water ad libitum. Mice exposed to longterm alcohol protocol received an additional bottle containing alcohol solution in drinking water. Alcohol was prepared as a solution of 96% ethanol and tap water. The concentration of ethanol in water increased from 4% during the first four days of exposure, to 8% for the remaining 30 days. The alcohol consumption (as well as normal water) was measured every 2 days.
Whole-cell patch-clamp technique was used to analyze AMPA/NMDA EPSCs ratio. Brains from decapitated mice were quickly submerged in the ice-cold cutting solution (135 mM NMDG, 1 mM KCl, 1.2 mM KH 2 PO 4 , 1.5 mM MgCl 2 , 0.5 mM CaCl 2 , 20 mM choline bicarbonate, 10 mM Dglucose, bubbled with carbogene-5% CO 2 , 95% O 2 ). Coronal 250 µmthick slices were prepared using Leica VTS1000 vibratome. Slices containing the amygdala were collected into a chamber filled with artificial cerebrospinal fluid (ACSF,119 mM NaCl, 2.5 mM KCl, 1 mM NaH 2 PO 4 , 26 mM NaHCO 3 , 1.3 mM MgCl 2 , 2.5 mM CaCl 2 , 10 mM D-glucose, bubbled with carbogene) and incubated for at least 60 min at room temperature. Slices were then transferred to the recording chamber, perfused with ACSF solution heated up to 31°C. The stimulating electrode was placed in the axons from the basal amygdala. CeM neurons were identified visually and patched with borosilicate glass capillaries (3-6 MΩ resistance) filled with internal solution (130 mM Cs gluconate, 20 mM HEPES, 3 mM TEA-Cl, 0.4 mM EGTA, 4 mM Na 2 ATP, 0.3 mM NaGTP, and 4 mM QX-314Cl, pH = 7.0-7.1, osmolarity: 290-295 mOsm). Series and input resistances were monitored throughout the experiment. Electrical stimulation was elicited by TTL pulse every 5 s. Recorded currents were filtered at 2 kHz (npi amplifiers) and digitized at 10 kHz (ITC-18 InstruTECH/HEKA). All recordings were performed in the presence of 50 μM picrotoxin (Abcam) in ACSF, to pharmacologically block inhibitory neurotransmission and focus on the excitatory pathway specifically. After reaching the whole-cell configuration, baseline currents were recorded for 2-3 min to ensure the stability of recorded amplitudes. AMPA/NMDA EPSCs Ratio. To measure the AMPA/ NMDA ratio, 50-100 stable sweeps were recorded at −60 mV and +45 mV for AMPA and NMDA EPSCs, respectively, and their peak amplitudes were averaged. Since the EPSC recorded at +45 mV is a composite of AMPA-Rand NMDA-R-mediated current, the amplitude of NMDA response was measured 50 ms after the peak to ensure the absence of the AMPA-R component. All electrophysiological recordings were performed by the experimenter blinded to the mouse genotype. For the analysis of recordings, we averaged the data from multiple cells of one animal, thus showing the overall effect of alcohol on an individual mouse. The number of recorded cells from each experimental group are indicated in the figure legend.

Stereotactic Injection of viral vectors
The surgeries were performed at least 7 days before the beginning of the training. Lentiviruses for the CRISPR/Cas9 system with the gRNA1 or the corresponding empty vector were produced at Animal Models Lab at Nencki Institute (Warsaw, PL). Lentivirus (LV) solution (alcohol and sucrose experiment: 200 nl, 10 7 -10 8 GC/µl; electrophysiology experiment: 900 nl, 10 7 -10 8 GC/µl) was injected into the CeA (Bregma AP −0.12, ML + / − 0.28, DV −0.46) (Paxinos and Franklin, 2008) at the rate of 0.1 µl per minute through the beveled, 26-gauge metal needle and 10 µl microsyringe (SGE010RNS, WPI, USA). The needle was removed from the brain ten minutes after the injection to prevent outflow.

IMAGEN study
Samples were drawn from the IMAGEN cohort, a European multi-center imaging-genetics study of adolescents recruited and tested across eight assessment sites (London, Nottingham, Dublin, Mannheim, Berlin, Hamburg, Paris, and Dresden). The informed consent was obtained from all subjects. The recruitment process, exclusion and inclusion criteria and the procedure have been described elsewhere [36]. From the IMAGEN cohort, data were only included if behavioral, genetic, methylation and MRI data were available. After quality control and removal of outliers, a sample of 1315 19 years-old participants (52.8% females) was used for statistical analysis. Local ethics research committees at each IMAGEN site permitted the study. Alcohol-related behaviors were measured using the European School Survey Project on Alcohol and Drugs (ESPAD) questionnaire with the focus on the self-reported quantity of consumed substance, frequency of drunkenness, and binge drinking over the last 12 months.
Structural MRI data were acquired with 3T magnetic resonance (MRI) scanners (Siemens, GE and Philips). The same scanning protocols were applied across all IMAGEN sites. Full details of MRI acquisition protocols and quality checks have been previously described [37]. FreeSurfer software suite (http:// surfer.nmr.mgh.harvard.edu/) was utilized to segment MRI brain images and inspect each individual cortex for inaccuracies. Participants with cerebral cortex malformations were excluded from analyses. For this study, values referring to the total volume of four brain regions: Left Amygdala, Left Hippocampus, Right Amygdala, Right Hippocampus, were extracted. Since IMAGEN data was collected at 8 different sites, the effect of site was added as a nuisance covariate in all statistical analyses in addition to total intracranial volume [32].

Omics data
DNA was extracted from whole-blood samples. DNA purification and genotyping took place at the Center National de Génotypage in Paris. The whole-blood samples (∼10 mL) were collected from participants at the age of 14 and retained in BD Vacutainer EDTA tubes (Becton, Dickinson & Company) utilizing the Gentra Puregene Blood Kit (QIAGEN) in line with the instructions specified by the manufacturer.
Genotype data were obtained for 582,982 markers utilizing DNA extracted from whole-blood and the Illumina (Little Chesterford, UK) HumanHap610 Genotyping BeadChip. SNPs with call values of <98%, minor allele rate <1% or variation from the Hardy-Weinberg equilibrium (P ≤ 1 × 10 −4 ) were rejected from further examinations. Subjects with an unclear sex code, absent genotypes (failure rate >2%), and outlying heterozygosity (with a frequency of 3 SD from the Mean) were also removed. The identity-by-state resemblance was employed to assess cryptic relatedness for individuals with the PLINK program. Individuals who were closely related to identity-by-descent (IBD > 0.1875) were excluded from the subsequent analysis. The population was investigated using principal component analysis (PCA) in EIGENSTRAT software. The four HapMap populations were managed as reference groups in the PCA and participants with atypical ancestry were also eliminated. Genotype data were coded into minor homozygotes, heterozygotes, and major homozygotes.
ARC gene. For this study, the candidate gene approach was employed. The information about the selected gene of interest, ARC, was extracted from an online genetic database-(GRCh37.p13; https:// www.ncbi.nlm.nih.gov/gene/23237). For the IMAGEN sample, there was only one SNP (rs10097505) located within the ARC gene locus. A total sample of 1315 was extracted from the IMAGEN cohort based on the availability of corresponding ESPAD measures.
In the human study, the significance of the associations with ESPAD variables of interest was corrected for multiple testing and established using permutations (n = 10,000). The p value (two-tailed) was computed by taking the absolute number of permutations that resulted in a higher correlation than the observed absolute value and divided by the total number of permutations. For the statistical tests corrected with permutations, the alpha level of 0.05 was applied, unless otherwise specified. Analyses of associations between genotype and methylation were performed using the Statistical Package for the Social Science (IBM SPSS version 25), while analyses that included ESPAD variables of interest were computed in MATLAB (R2018a) using scripts generated for this study.

Statistical analysis
Data acquisition and quantification were performed by an experimenter blind to experimental groups. All statistical analyses were performed using GraphPad Prism 9 Software. The exact sample size of each experiment is provided in figures or figure legends. For data with normal distribution and equal variance, Student's t-test, two-way ANOVA or a mixed-effects model with repeated measures (RM) and post hoc Sidak's tests for multiple comparison were used. Least significant difference (LSD) post hoc tests were used only for planned comparisons. Data that did not follow normal distributions were analyzed with the Mann-Whitney to compare two groups or Kruskal-Wallis test to compare more than two groups. All data with normal distribution are presented as the mean ± standard error of the mean (SEM). For samples that did not follow a normal distribution, median and interquartile range (IQR) values are shown. The difference between the experimental groups was considered significant at p < 0.05. If not mentioned otherwise all tests were two-sided. Data were excluded from the analysis only if they differed more than 3 SEM from mean.

Distinct regulation of single AUD-related behaviors by Arc
We trained Arc knockout (Arc KO ) mice and their wild-type littermates (Arc WT ) to drink alcohol in an IntelliCage social setting [30] (Fig. 1a). The animals went through a training consisting of alcohol introduction (4-20%) and alcohol free access period (FA, 12%). Next, we assessed five behaviors that operationalize DSM-5 criteria for AUD: [1,30] increased motivation to drink alcohol (Motivation) [38]; alcohol craving during withdrawal (Withdrawal); [39] reactivity to alcohol-predicting cues (Cue-induced seeking response) [40]; lack of control over alcohol consumption (Alcohol relapse); and persistence in alcohol seeking, even during signaled alcohol non-availability (Persistence) [41] (Fig. 1b). An addiction score was calculated as a sum of normalized scores from all AUD tests and an addiction index as a sum of positive results (top 35%) in all tests [30].
Arc KO mice sought significantly more for alcohol during withdrawal and cue-presentation. They showed a higher persistence in alcohol seeking as compared to Arc WT mice and slightly higher motivation to drink alcohol (Fig. 1c). With these behaviors dominating, Arc KO mice reached a higher addiction score and addiction index, than the Arc WT animals (Fig. 1d). We found no significant effect of Arc on novelty seeking or general activity. A lack of Arc did not affect free access alcohol consumption (Fig. 1e) or alcohol consumption during relapse (Fig. 1c). The downregulation of Arc mRNA and protein levels in Arc KO mice, as compared to Arc WT animals, was confirmed by qPCR and western blots from the whole brain homogenates, and by the Arc immunofluorescent staining of the brain sections (Fig. 1f). Altogether, these findings demonstrate a selective role of Arc in AUD-related behaviors, which is confined to alcohol seeking during alcohol withdrawal and cue presentation.
Arc protein is recruited in the mouse amygdala during cueinduced alcohol seeking, but not during drinking In order to further test the role of Arc protein in distinct AUDrelated behaviors, C57BL/6J mice were trained in IntelliCages and sacrificed during period of free access to alcohol (FA, day 95), after alcohol withdrawal (W, day 101), or after 90 min of cue (light)-induced alcohol seeking (CR, day 101). Control animals were also trained in the IntelliCages, but never had access to alcohol, and were sacrificed during the FA phase (day 95) (Fig. 2a). We observed that the alcohol-naïve and alcohol drinking mice did not differ in general activity (Fig. 2b). Previously alcohol drinking mice showed alcohol-seeking behavior, when the alcohol was IntelliCage setup (with a magnified cage corner) and the experimental timeline. Arc KO (n = 12) and Arc WT (n = 12) mice were trained to drink alcohol in the IntelliCage, AUD-related behaviors were tested, addiction score (sum of normalized scores from all AUD tests) and index (sum of positive results (top 35%) in all tests) were calculated. We analyzed motivation to drink alcohol (Motivation, M); alcohol craving during withdrawal (Withdrawal, W); reinstatement of alcohol seeking by alcohol-predicting cues (Cue relapse, CR); lack of control over alcohol consumption (Alcohol relapse, AR); and persistence of alcohol seeking (Persistence, P). abandoned. As expected, this behavior decreased during withdrawal (FA vs. W), indicating the beginning of extinction of alcohol seeking behavior when the reward corner was inactive (Fig. 2b). However, they increased alcohol seeking during cue presentation, as compared to the last day of the withdrawal (W vs CR), indicating a cued recall of the alcohol reward memories (Fig. 2b). Brains were sliced and immunostained to detect Arc and PSD-95, as a marker of the glutamatergic synapse [42]. Arc expression was analyzed in the central (CeA) and basolateral nuclei of the amygdala (BLA) (Fig. 2a) and dentate gyrus of the hippocampus (DG) (Supplementary Fig. 1). We focused on these brain regions as previous work linked them with multiple addiction-related behaviors, including craving and cue-induced reinstatement of alcohol seeking [10][11][12][13][14][15][16][17][18]43]. Arc protein was detected as immunopositive puncta (Arc + ) that often colocalized with PSD-95-positive puncta (PSD-95 + ), indicating their location at excitatory synapses (Fig. 2c, d). We found no effect of the alcohol drinking on Arc levels in CeA, but a significant upregulation of the Arc protein levels after cue-induced seeking. Similar changes of Arc expression were observed in BLA, however, they did not reach statistical significance (Fig. 2e). No changes of Arc protein levels were observed in DG and CA1 (Supplementary Fig. 1). Thus, overall our data indicate a dissociation of Arc protein upregulation in the CeA between cue-induced alcohol seeking, withdrawal-associated alcohol seeking and alcohol drinking.

Alcohol consumption decreases Arc gene methylation in the mouse amygdala
To test a causal mechanism of Arc regulation after alcohol drinking, we investigated ARC gene expression and methylation in the amygdala. We trained mice to drink alcohol in the IntelliCages. Mice had free access to 10% alcohol for 45 days and were sacrificed after 6-day alcohol withdrawal. The control alcoholnaïve mice were also trained in the IntelliCages, but never had access to the alcohol (Fig. 3a). The amygdala tissue was used to analyze Arc mRNA expression by quantitative PCR (qPCR) (Fig. 3b) and methylation of 54 CpG sites within Arc gene promoter and exon 1 using Targeted NextGen Bisulfite Sequencing (tNGBS) (Fig. 3c). We found a significant effect of alcohol exposure on Arc mRNA levels (Fig. 3b) and interaction between alcohol and CpG site on the frequency of methylation (Fig. 3c). In the mice drinking alcohol the levels of Arc mRNA was increased as compared to alcohol-naïve mice (Fig. 3d), and the average frequency of CpGs methylation were significantly decreased both within the Arc promoter and exon 1 (Fig. 3d). Accordingly, we found negative correlations between Arc gene expression and methylation  (Fig. 3e). There was no significant correlation between alcohol consumption and later Arc gene methylation (Fig. 3f). However, Arc methylation correlated with alcohol seeking during withdrawal (Fig. 3g). Altogether, these data suggest that previous alcohol consumption and seeking reduced Arc gene methylation in the amygdala, which enhanced Arc mRNA expression.
Arc mRNA expression is increased in the amygdala of the mice with AUD-resistant phenotype To test a link between Arc mRNA expression and AUD-prone phenotype, we investigated transcriptome in the amygdala of the mice trained to drink alcohol in the IntelliCages and diagnosed as AUD-prone (Addiction index ≥2) or resistant (Addiction index <2) [43] (Fig. 3h). Addiction-prone and resistant mice differed in addiction score, activity as well as AUD behaviors (Fig. 3i, j). The amygdala tissue was collected and total RNA extracted. There were higher levels of Arc mRNA in the amygdala of AUD-resistant mice as compared to AUD-prone drinkers (Fig. 3k), suggesting that Arc mRNA expression has protective effects during AUD progression.
Motivation to drink alcohol and relapse induced by alcoholpredicting cues are regulated by CeA Arc As we identified the CeA as a crucial brain region where Arc activity is associated with cue-induced alcohol seeking, we causally probed this relationship. To this end, we developed a CRISPR/Cas9 system for local Arc gene manipulation in vivo using plentiCRISPRv2 plasmids [34]. First, five Arc gene-specific guide RNAs (gRNAs) were designed and tested in vitro ( Supplementary  Fig. 2a). gRNAs, cloned into plentiCRISPRv2, significantly downregulated Arc protein expression in stimulated mouse dissociated neuronal culture (Supplementary Fig. 2b, c). As gRNA1 is located at the beginning of the Arc gene coding sequence, plentiCRISPRv2 with gRNA1, and plentiCRISPRv2 without gRNA, were used to produce lentiviral vectors (Arc Crispr and empty vector, EV). After . Regression lines and 95% confidence intervals are shown. Each dot represents one mouse. f, g Summary of data showing correlation of Arc gene methylation (promoter and exon 1) in the amygdala of alcohol trained mice with (f) alcohol consumption and (g) alcohol seeking during withdrawal (Spearman correlation; r s , Spearman r). Regression lines and 95% confidence intervals are shown. Each dot represents one mouse. h-k Transcriptomic analysis of AUD-prone and resistant animals. h Experimental timeline. C57BL/6J mice were trained in the IntelliCage. i Addiction score was calculated based on five AUDrelated behaviors. AUD-prone (≥2) and resistant (<2) animals were identified (t(8) = 4.98, p = 0.001). j AUD-related behaviors: motivation for alcohol (Unpaired t-test, t(7) = 1.93, p = 0.048), alcohol seeking during withdrawal (t(7) = 2.82, p = 0.013), and cue-induced relapse (t(7) = 2.96, p = 0.011), alcohol relapse (t(7) = 3.81, p = 0.003) and persistence tests (t(7) = 2.28, p = 0.028). k Summary of data showing Arc mRNA expression in the amygdala of AUD-prone and -resistant animals (one-way ANOVA with post hoc LSD tests, F(2, 11) = 18.6, p < 0.001).
transduction of mouse neuronal culture with Arc Crispr sixteen specific indel mutations were identified within the gRNA complementary sequence of Arc gene ( Supplementary Fig. 2d), confirming specificity of the mutation. None of such mutations were found in the EV neurons. Arc Crispr stereotactically delivered to CeA transduced the cells with 60% efficiency and 96% specificity for neurons ( Supplementary Fig. 2). We found that a local downregulation of Arc reduced the AMPA/NMDA EPSCs ratio in CeA neurons ( Supplementary Fig. 3), and prevented Arc protein expression in CA1 after PTZ kindling ( Supplementary Fig. 4) in the adult mouse brain.
Mice were bilaterally injected into the medial part of CeA (CeM) with Arc Crispr or EV. We focused on CeM since it was previously linked with reward and reward-associated cues [11][12][13][14]. The animals underwent long-term alcohol training in the IntelliCages (Fig. 4a-b). During cue-induced alcohol seeking, Arc Crispr mice made significantly more nose pokes to the alcohol corner than the EV animals (Fig. 4c). Arc Crispr mice also reached a higher breakpoint during the motivation test, as compared to EV mice. There were no significant differences between the Arc Crispr and EV mice in alcohol seeking during alcohol withdrawal, alcohol consumption during alcohol relapse and performance during a persistence test (Fig. 4c). The mice from two experimental groups did not differ in general activity and alcohol consumption during the introduction phase and FA periods in the Intellicages (Supplementary Fig. 5a, b). The post-training analysis of the brain sections confirmed that Arc Crispr and EV were expressed mostly in the CeM (Supplementary Fig. 5c) and Arc Crispr mice had significantly reduced levels of total and synaptic Arc in CeM, as compared to the EV mice ( Supplementary Fig. 5d). Thus, the CeM Arc specifically controls motivation to drink alcohol and alcohol seeking during relapse induced by alcohol-predicting cues. These findings suggest a causal role of Arc in the CeM for the control of cue-induced alcohol seeking.
CeA Arc does not regulate sucrose seeking Since our data indicates that Arc expression in the CeM controls alcohol motivation and seeking during cue-induced alcohol seeking (Fig. 4c), in the next step we tested whether this effect was specific for alcohol-predicting cues or affecting cues associated with other rewards as well. A naïve cohort of mice underwent stereotactic surgery. Arc Crispr and EV were bilaterally injected into CeM and the mice underwent long-term training in the IntelliCages. The training resembled the alcohol training, however, this time 5% sucrose was used as a reward (Supplementary Fig. 6a, c). Arc Crispr mice did not differ from the EV group, as far as visits in new corners, general activity and sucrose consumption were concerned ( Supplementary Fig. 6d), indicating no effect of Arc Crispr on novelty and reward seeking traits that are associated with propensity for AUD [30,44]. We also did not find any effect of Arc Crispr on addiction-related behaviors in the mice trained to drink sucrose ( Supplementary Fig. 6e). These findings suggest no role of CeM Arc in cue-induced seeking of natural rewards.

Arc regulates alcohol-induced synaptic plasticity in CeM
After showing a specific role of Arc in cue-induced alcohol seeking behavior, we asked whether Arc affects plasticity in this process. As Arc regulates internalization of AMPA-R [22,23], we hypothesized that Arc regulates the function of CeM synapses during cueinduced alcohol seeking. We trained Arc KO and Arc WT mice to drink alcohol or water. Thereafter, they were sacrificed, either after alcohol withdrawal or after 90 min of cue-induced alcohol seeking. As a control, alcohol-naïve water drinking mice were used (Fig. 5a). We performed whole-cell voltage clamp recordings from the neurons of the CeM while stimulating axons from BLA (Fig. 5b), and measured the ratio of AMPA-R and NMDA-R excitatory postsynaptic currents (EPSCs). We focused on this projection as former data showed that the BLA-CeM pathway can drive appetitive behaviors [45,46]. We found that the AMPA/NMDA ratio was lower in Arc KO as compared to Arc WT mice when only drinking water. Cue-induced alcohol seeking coincided with a decrease in the AMPA/NMDA EPSCs ratio in the Arc WT mice compared to the withdrawal condition suggesting a weakening of synaptic transmission. The opposite changes were observed in Arc KO mice. The AMPA/NMDA ratio was significantly upregulated during cue-induced alcohol seeking, as compared to the withdrawal condition (Fig. 5c). This phenotype was also replicated in the mice with Arc Crispr expressed in CeA ( Supplementary Fig. 7). These findings suggest that Arc is required for the plasticity of the BLA-CeM synapses during presentation of alcohol-predicting cues when the cues are not accompanied by alcohol reward. Thereby, the CeM Arc would appear as a signal that updates the incentive value of an alcohol cue. Every cue-induced retrieval of the alcohol seeking behavior is at the same time an extinction trial. If not eventually followed by the reward (alcohol), it would weaken the cue-alcohol association, and eventually lead to an extinction of seeking behavior.

ARC genetic variations are associated with alcohol consumption in humans
To check the contribution of the human ARC gene to the expression of alcohol addiction-related behaviors at a time point when they get established, we interrogated the IMAGEN sample of 1315 19-year old participants [32], controlling for gender and site of data collection (Fig. 6, Supplementary Table 1). We found a significant effect of the single nucleotide polymorphism (SNP) rs10097505 (variant: G > A) and the methylation of 15 CpG sites on the amount of consumed alcoholic beverages over 12 months. Thereby, the individuals with the AA variant tended to drink more alcohol than those with GG and GA (Fig. 6, Supplementary Table 2).
Since alcohol consumption may induce DNA hypomethylation [47] (Fig. 3), we investigated blood DNA methylation in the IMAGEN cohort using the Illumina 850 K BeadChip array. The CpG sites analysis demonstrated that the frequency of drunkenness positively correlates with methylation of cg05415840 and the amount of consumed alcohol positively correlates with cg08387463 methylation within exon 1. However, binge drinking frequency negatively correlated with methylation levels of cg04321580 and cg19438565 located within the ARC promoter (Fig. 6, Supplementary Table 3). Moreover, to investigate whether methylation was genotype-dependent, correlations between the methylation of four CpG sites (cg08387463, cg04321580, cg19438565, cg05415840) and the SNP rs10097505 variant were computed (Supplementary Table 3). The analysis revealed that the cg19438565 and cg05415840 methylations were significantly correlated with the variable rs10097505, suggesting that those two CpG sites were genotype-dependent (Fig. 6, Supplementary  Table 3). To test whether the SNP variants influence methylation levels and its association with alcohol consumption, Pearson's correlation test was conducted between the ESPAD variables and two CpG sites, controlling for the genotype (rs10097505) in addition to cells, gender, wave, and sites. The test revealed a significant association between cg19438565 and frequency of binge drinking (Supplementary Table 4). The loss of the previously significant correlation between the frequency of drinking and cg05415840 after controlling for the genotype could be due to the methylation mechanisms being driven by the genotype. On the other hand, the fact that the correlation between binge drinking and cg19438565 did not change substantially, indicates that the tested variables are not influenced by the variation in the genotype, suggesting the environmental effect that modulates this CpG site. Altogether, the findings show that natural variation in the Arc gene and its methylation levels influence key AUD-  related behaviors-amount and frequency of alcohol consumption-that are known to be alcohol cue-controlled.
ARC gene methylation predicts amygdala volume in humans Next, we tested the correlation between the methylation at three CpG sites (cg08387463, cg04321580, cg19438565; excluding cg05415840 since the results showed that the methylation at this CpG site is driven by the genotype) with the amygdala and hippocampus volumes at both hemispheres. Pearson's correlation analysis was conducted controlling for gender, site, cells, wave and total intracranial volume ( Table 1). This analysis showed a negative correlation of the left amygdala with the cg08387463 CpG methylation levels ( Table 1). The result of the analysis indicates that the methylation at the ARC cg08387463 CpG site is linked to the left amygdala volume (Supplementary Table 5).

DISCUSSION
AUD is a common psychiatric disorder with no effective pharmaco-treatment available to interfere with maladaptive behaviors. At the core of this disorder, and criterion for AUD diagnosis, is the uncontrollable cued relapse to alcohol taking after established periods of abstinence. Here we propose a new mechanism how cue-induced alcohol seeking behavior is regulated, which dissociates from mechanisms controlling alcohol consumption per se. The amygdala protein Arc is exclusively enhanced during cue-induced alcohol seeking. Its activity in the CeM determines cue-induced alcohol seeking, but shows no role in later alcohol drinking behavior. As such, it mediates the starting point of a cued relapse episode, but not its end. As Arc is essential for synaptic plasticity of the BLA-CeM projection, it is suggested that Arc in the CeM is recruited for an update of the cue incentive properties and thereby an essential regulator of addiction memory. This conclusion is in line with earlier reports showing that CeA activity is regulated by alcohol cues, as well as with the CeA regulating incentive salience of drug-associated cues [11,12] and reinstatement of drug seeking by cues associated with drug reward [15][16][17][18]. Long-term alcohol consumption affects glutamatergic transmission within the mesocorticolimbic system that ultimately controls drug reward and reinforcement, as well as drug craving and seeking [48][49][50]. Significant changes of AMPA-R and NMDA-R currents, and frequency of silent synapses that lack functional AMPA receptors, were observed after long-term alcohol consumption, alcohol withdrawal, relapse and presentation of alcohol-associated cues in multiple brain regions including the amygdala [10,43,[51][52][53]. In human patients, the genetic variation in the NMDA-R dependent AMPA-R trafficking cascade is associated with alcohol dependence [53,54]. However, the molecular processes that drive the alcohol-induced modifications of the glutamatergic synapses and AUD-related behaviors are only beginning to be elucidated [9].
Here, we found that Arc KO mice had an overall higher addiction index, as compared to Arc WT mice (Fig. 1). This phenotype was a consequence of increased alcohol seeking during withdrawal and relapse induced by alcohol-predicting cues, increased persistence in alcohol seeking and a tendency to display higher motivation for alcohol, but without effects on alcohol consumption. A local knockdown of Arc in CeM confirmed this phenotype in cue relapse and motivation test (Fig. 3). Moreover, we observed higher Arc mRNA levels in the amygdala of the mice diagnosed as AUDresistant, as compared to AUD-prone individuals. Thus, Arc expression is a protective factor for several AUD-related behaviors. This conclusion is in concordance with earlier studies demonstrating decreased Arc mRNA levels in the amygdala of early onset alcoholics [55].
Although Arc protein expression is induced in neurons by multiple stimuli [56], only a few studies showed a causal link between Arc expression and animal behavior. In particular, Arc affects memory formation, cognitive flexibility, anxiety, cocaine self-administration and schizophrenia-related behaviors [22,[57][58][59][60][61]. Decreased Arc expression in the amygdala was implicated in increased early life stress-induced alcohol consumption in animal models [55,57,62]. In human studies, a correlation between ARC gene polymorphism was confirmed solely for schizophrenia and Alzheimer's disease [63,64] so far. Here, we show for the first time a link between ARC gene variation and alcohol consumption in the IMAGEN cohort of young adults. ARC gene rs10097505 SNP affects the amount of consumed alcohol (Fig. 6). The individuals with the AA variant drink more alcohol, then people with GA or GG variants (Supplementary Table 2). Since the AA variant confers also a risk factor for Alzheimer's disease, and these patients have lower levels of ARC mRNA in the cortex as compared to healthy controls [64], our findings suggest that individuals with higher alcohol consumption, which can be seen as the natural result of enhanced occurrence of cue-induced relapse episodes, have lower ARC expression. This is in agreement with the study showing that AUD patients have less ARC mRNA in the amygdala [55]. Moreover, methylation of the ARC promoter (cg04321580 and cg19438565) negatively correlates with frequency of binge drinking, while cg08387463 methylation within exon 1 positively correlates with alcohol consumption. In agreement with previous study [65], increased methylation in the Arc gene body is predictive of lower ARC expression associated with higher and more frequent alcohol consumption.
An important limitation of this finding is that the analysis was conducted in blood samples. Therefore, it is uncertain whether a similar pattern of methylation is observed in the brain and whether it affects Arc expression in the same way. Nevertheless, as we observed that high cg08387463 methylation predicts not only high alcohol consumption, but also low left amygdala volume in the IMAGEN cohort (Table 1), methylation of cg08387463 may be a potential blood marker of alcohol use. Importantly, in concordance with previous studies lower amygdala volume is a risk factor for heavy alcohol drinking in young adults [66][67][68].
We therefore propose Arc in the amygdala as a key regulator for cue-induced alcohol seeking which drives relapse to alcohol Table 1. Analysis of the correlation of ARC gene methylation sites and volume of the amygdala and hippocampus. drinking behavior as part of an AUD. Arc appears to be essential for concurrent updating of the cue incentive salience. Its timedependent selective pharmacological targeting at cue exposure may therefore offer a new therapeutic strategy for one of the key AUD associated behaviors.