All procedures followed the Guide for Care and Use of Laboratory Animals provided by the National Institutes of Health, and with approval of the Institutional Animal Care and Use Committee of UCSF. Male C57BL/6 mice, 6–8 week of age, were purchased from Jackson Laboratories. Animals were single-housed under a reverse 12:12 light:dark cycle, with lights off at 10:00 a.m. Food and water were available, ad libitum, for all subjects.

We used a modified drinking in the dark paradigm as previously described (Lei et al., 2016a,b,c). After ∼2 weeks acclimation, mice were first given 24 h two-bottle choice access to 15% alcohol (v/v) and water. Thereafter, mice drank under a limited daily access paradigm, where they were given two-bottle choice of 15% alcohol and water for 2 h/day. 5 day/week starting at ∼3 h after lights off. To test for specificity of drugs on alcohol drinking, a subset of mice drank a 0.05% saccharin solution an identical schedule to that used for alcohol; this concentration was previously determined to yield a similar volume of intake as alcohol (Lei et al., 2016a,b).

After ∼2 weeks acclimation, mice drank under an intermittent access schedule where they received overnight (∼24 h) two-bottle choice access to 20% alcohol (v/v) and water starting on Monday, Wednesday, and Friday at ∼3 h after lights off (Hwa et al., 2011; Morisot et al., 2018).

After ∼2-week of alcohol access under 2bdc-DID or 2bc-IA, surgery was performed to bilaterally implant guide cannulae (Plastics One) aimed at either the Shell (AP +1.5, ML ± 0.5, and DV -4.8 mm), NAc Core (AP +1.5, ML ± 1.0, and DV -4.0 mm) or vmPFC (AP +1.7, ML ± 0.4, and DV -2.7 mm). All given coordinates are relative to Bregma from skull surface. After surgery, mice were allowed to recover for at least 3 days before alcohol drinking was resumed. At the end of experiments, brains were harvested for histology and cannula placement verification.

Agents, doses, and relevant references for the doses used are given in Table 1. The selective OX1R antagonist, SB-334867 (Tocris) and the PKC inhibitor, chelerythrine (Abcam), were dissolved in 100% DMSO. OrexinA peptide (Sigma), the selective OX2R antagonist, TCS-OX2-29 (TCS, Tocris), and the NMDAR antagonist, AP5 (Tocris) were dissolved in 0.9% saline, as was the D1R selective antagonist, SCH-23390 (SCH, Sigma). A cocktail of the GABAA agonist muscimol (Sigma) and GABAB agonist baclofen (RBI) (M/B) were dissolved in DMSO, and 50 ng/side each was injected intracranially. Some mice were injected with 1 or 5 mg/kg baclofen i.p., dissolved in saline. The compounds we tested have been widely used and thus we tested a single dose taken from literature. Also, although the SCH23390 dose is high, it has been previously shown to be selective in that it reduces acquisition of alcohol CPP without altering LiCl CPA (Pina and Cunningham, 2014). While 100% DMSO is a high dose for intracranial, we (Lei et al., 2016b) and other groups have used this intracranial vehicle (Pierce et al., 1999; James et al., 2011; Simms et al., 2011; Vendruscolo et al., 2015). Importantly, our studies were performed with a randomized, Latin-squares design, with alcohol drinking on days in between intracranial test sessions, and any possible lingering toxicity of DMSO should impact drinking on subsequent days, which was not observed. We also note that other studies have used concentrations of SB that are much lower from what we utilize, e.g., where Thorpe and Kotz (2005) used 6 ng in Shell to block orexin enhancement of feeding. However, this paper uses aCSF as the vehicle (see also Zheng et al., 2007). In recent times, many studies, including our own, use a DMSO-based strategy (e.g., see James et al., 2011), and we and others do not get SB solubility in aCSF. We also note that comparison of doses with older studies may be challenging given observations that there can be batch-related differences in SB solubility and color (Jeff Simms, personal communication, a co-author from Richards et al., 2008, and other orexin-related publications).

All animals were habituated to handling and injection prior to drug treatment sessions. After 2–3 days of simple handling, animals have 2–3 days of handling where cannula plug is removed and returned; finally, animals have one saline injection prior to drug test sessions. Drug deliveries occurred 30 min prior to drinking. Drugs and their vehicles were injected in a counter-balanced manner in different cohorts of mice. For many cohorts, vehicle and drug were each tested twice in each animal in a counter-balanced, randomized fashion, and averaged to give a single vehicle intake value and single drug intake value for each animal. SCH, orexinA ± TCS, vmPFC AP5, and some SB and M/B in 2bc-DID were tested with only a single injection of vehicle and drug. There was at least one drinking day between drug treatments and not more than two injections per week. For i.p. delivery, drugs were injected at a volume of 10 mL/kg. For intracranial injections, drugs were bilaterally injected at a volume of 200 nL at a rate of 200 nL/min, with the exception of OXA, which was injected at a volume of 300 nL. The infusion needles (Plastics One) were left in place for an extra 60 s before retraction. For Shell and Core, needles projected 0.3 mm past end of guide cannulae, and for vmPFC needles projected 0.5 mm.

After each drinking session, alcohol (g/kg of body weight) or saccharin intake (mL/kg of body weight), water intake, and the preference ratio (volume of alcohol intake/total volume of fluid intake) were measured. Data were statistically analyzed using Prism (GraphPad), SPSS (IBM), and R v3.4.4 (R Foundation for Statistical Computing).

Alcohol drinking was analyzed with paired t-tests comparing vehicle vs. drug treatment within each animal. Non-normal data, including concurrent water intake, were tested with Wilcoxon match-paired signed rank test (for paired data) or Mann–Whitney test (for unpaired data). To compare the basal intake vs. change in drinking slope relationships across different groups, a one-way ANCOVA was performed with a dependent variable of Change in Drinking, an independent variable of Treatment Group, and a covariate of Basal Drinking. The presence of a statistically significant interaction between Treatment Group and Basal Drinking indicates a dissociable effect of Treatment Group on the slope of a regression line which examines Change in Drinking as a function of Basal Drinking. All bar data are shown as mean ± SEM.

We were particularly interested in understanding the relationship between basal drinking levels and the requirement for Shell Ox1Rs/activity for driving alcohol intake. Thus, we took advantage of the large data sets we had acquired to study a Shell inhibition group, which consisted of data combined from Ox1R or global GABA-mediated inhibition within the Shell. Since there was variability in mouse intake, we also compared Shell inhibition results with data combined from a series of control groups where no change was observed with drug treatment. This gave us a robust comparison to help account for intrinsic variability in mice drinking and allowed us to study whether a greater Shell contribution in higher-drinking individuals represented mathematical effects such as regression to the mean. Basal alcohol intake was normally distributed in both groups, and there was no difference in basal intake level between the Shell inhibition group and control group (t₂₀₂ = 0.8891, p = 0.3750).

We first compared basal drinking levels with the actual change in alcohol intake with drug treatment. In addition, because lower basal drinkers have a smaller maximal change in drinking levels, we next normalized the drug-related drinking change to account for different basal intake levels in each animal. In particular, we calculated the log[100^∗(intake during drug treatment)/(intake during vehicle)]. This allows us to express the impact of Shell inhibition or other drug treatment on drinking relative to baseline consumption levels, but also partially correct for the high percent increases that can be observed when lower basal intake levels go up. Using this measure, a log value of 2 (log[100]) indicates no intake change with treatment. For subjects where treatment reduced drinking to zero, the log was set to 1. This method thus reflected a useful compromise to examine the percent drop in drinking so that it could be compared with basal drinking levels across individuals. In addition, in order to examine the distribution of treatment effects across individuals (shown in Figure 2G,H), we determined the number of animals in each of the following bins of log[100^∗(intake during drug treatment)/(intake during vehicle)] values: 1–1.3, >1.3–1.5, >1.5–1.7, >1.7–1.9, >1.9–2.1, >2.1–2.3, >2.3–2.5, >2.5.

Previous studies using systemic Ox1R inhibition have suggested that Ox1Rs promote intake in higher-drinking individuals (Moorman and Aston-Jones, 2009; Alcaraz-Iborra et al., 2017; Moorman et al., 2017). We previously demonstrated that Shell Ox1Rs promote 2bc-DID drinking (Lei et al., 2016b) (see section “Materials and Methods”), and that animals drinking under this model reach binge-level blood alcohol concentrations (Lei et al., 2016a). To help understand whether Shell Ox1Rs underlie excessive consumption in higher-drinking individuals, we examined intake under two different high alcohol drinking models, 2bc-DID and 2bc-IA (see section “Materials and Methods”). Drug versus vehicle were tested within each animal using a randomized, counterbalanced design.

Ox1R inhibition within the Shell, using SB-334867 (SB, 3-μg/side) (Hollander et al., 2008; Espana et al., 2010; James et al., 2011; Plaza-Zabala et al., 2012; Qi et al., 2013; Brown et al., 2015; Lei et al., 2016b), significantly reduced 2bc-DID alcohol intake (Figure 1A; t₁₃ = 2.256, p = 0.0420) as previously observed (Lei et al., 2016b). Shell Ox1R inhibition, with the same SB dose, also significantly decreased intake during 2bc-IA intake (Figure 1B; t₂₉ = 3.004, p = 0.0054; measured during the first 2 h of intake). SB at a 1-μg/side dose in Shell did not reduce 2bc-DID intake (n = 11; veh: 2.00 ± 0.31 g/kg intake; 1-μg SB: 1.78 ± 0.14 g/kg intake; t₁₀ = 0.699, p = 0.5008), suggesting a dose-dependent effect. Thus, Shell Ox1Rs critically contributed to driving binge alcohol intake under both 2bc-DID and 2bc-IA models. To further understand how the Shell contributes to alcohol intake, we examined whether inhibiting the Shell with muscimol/baclofen (M/B, 50 ng of each/side), which produces more global inhibition of activity (Chaudhri et al., 2010; Millan et al., 2010; Wilden et al., 2014), would reduce 2bc-DID intake. Like Ox1R inhibition, GABA-mediated Shell inhibition also significantly reduced 2bc-DID alcohol intake (Figure 1C; t₁₀ = 2.710, p = 0.0219). Because administration of Ox1R blocker into Shell could result in drug diffusion to the adjacent nucleus accumbens core (Core) and act there to inhibit intake, we also infused SB directly into the Core, which did not reduce 2bc-DID alcohol consumption (Figure 1D; t₉ = 1.427, p = 0.1874). These results suggest that NAc Ox1Rs contributed to binge intake in a subregion-specific manner, and that Shell inhibition significantly reduced multiple forms of excessive alcohol consumption.

Reduced alcohol drinking after Shell inhibition could indicate more general motor and motivational effects. However, we previously showed that Shell Ox1R inhibition does not reduce saccharin intake (Lei et al., 2016b), and inhibition here of Shell Ox1Rs or general activity had no effect on concurrent water intake (Figure 1E–G; 2bc-DID SB: p = 0.6698; 2bc-IA SB: p = 0.2920; 2bc-DID M/B: p = 0.1563), similar to previous studies (see section “Discussion”). In addition, we observed no changes in preference (Table 2), perhaps due to the low level of concurrent water drinking across the 2-h session (Dhaher et al., 2009; Seif et al., 2015; Hartog et al., 2016; Lei et al., 2016b), although concurrent water intake was greater during 2bc-IA versus 2bc-DID sessions (p = 0.0032 Mann–Whitney) and 2bc-IA preference change was nearly significant (not shown). Thus, Shell Ox1Rs and activity promoted excessive alcohol drinking, rather than regulating intake more generally.

Since we had similar effects of Ox1R block in different alcohol drinking models, and congruent effects of more global Shell inhibition, we were in a unique position to aggregate these findings to examine whether the Shell is a critical region that promotes excessive intake in higher-drinking individuals. Thus, we generated a large, combined data set from data in Figure 1A–C and previous Shell Ox1R 2bc-DID results (Lei et al., 2016b). Importantly, we examined the relationship between basal intake levels (alcohol drinking on vehicle test days) and the consumption difference between drug and vehicle sessions. In this way, we could determine whether Shell inhibition had a greater impact on alcohol consumption in higher drinkers relative to more moderate drinkers. In addition, since mouse drinking exhibits variability, Shell inhibition experiments were compared with studies combined from ineffective drug treatments (which we call the control, no-change group, described in detail in Figure 4). This no-change group allows better understanding of the basal-intake/drug-effect relationship, including possible mathematical effects (in particular, where higher basal intake might be more likely to show a decrease, and vice versa).

We first examined basal alcohol intake relative to the actual intake difference between drug and vehicle sessions; negative values indicate reduced consumption during drug relative to vehicle sessions. Figure 2A demonstrates that Shell Ox1R or GABAergic inhibition reduced alcohol drinking in individuals with higher basal drinking, with no overall effect in more moderate drinkers. Thus, basal drinking was significantly and negatively correlated with the change in drinking by Shell inhibition (F_1,70 = 60.44, p < 0.0001). In the no-change group, basal drinking was also correlated with effect of drug treatment (Figure 2B; F_1,130 = 7.077, p = 0.0088), an indicator that higher basal drinking is more likely to drop and lower intake likely to rise, separate from drug treatment (a mathematical artifact likely reflecting regression to the mean). Importantly, however, the slope of the Shell inhibition group (slope = -0.8445, R² = 0.4633) was significantly larger than the slope of the no-change group (slope = -0.2428, R² = 0.0516) (ANCOVA: Group: F_1,200 = 37.044, p < 0.001; Basal intake: F_1,200 = 44.656, p < 0.001; Group × Basal Interaction: F_1,200 = 17.237, p < 0.001). These results suggest that Shell inhibition caused a greater disruption of alcohol intake in higher drinkers, indicating a more important role for Shell Ox1Rs in promoting excessive intake in higher-drinking individuals.

Since lower basal drinking limits the actual change in drinking levels, we next normalized the treatment-related intake change (or lack thereof) in relation to the basal consumption levels in each animal. In particular, we determined the percent change in drinking with treatment, and took the logarithm of this, determined as log[100^∗(intake during drug treatment)/(intake during vehicle)] (see section “Materials and Methods”). Using this analysis, the Shell inhibition group still showed an overall similar distribution as in Figure 2A, with a bigger decrease in alcohol consumption with Shell inhibition in higher-drinking individuals (Figure 2C; F_1,70 = 6.742, p = 0.0115, slope = -0.1155, R² = 0.08786). In contrast, control animals showed much less difference between animals with moderate and higher intake (Figure 2D; F_1,130 = 5.476, p = 0.0208, slope = -0.0499, R² = 0.04042), although the slope was still significant, likely for mathematical reasons described above. We note that Shell inhibition did reduce drinking in some moderate-drinking subjects, and thus we performed several analyses to better understand Shell inhibition’s impact in subjects with different basal consumption. First, we performed a median split to divide animals into moderate versus higher drinkers (Moorman and Aston-Jones, 2009; Alcaraz-Iborra et al., 2017; Moorman et al., 2017). Figure 2E,F shows alcohol intake pre- and post-treatment in moderate versus higher drinkers, and demonstrate that Shell inhibition strongly reduced alcohol intake levels only in higher drinkers (Figure 2E), while treatment had no effect in the control group (Figure 2F). In addition, Shell inhibition caused a significantly greater decrease in alcohol intake in higher drinkers relative to moderate drinkers (Figure 2G; p = 0.0170 Mann–Whitney). In contrast, there was no difference in treatment effect between moderate and higher drinkers in the control (no-change) group (Figure 2H; p = 0.2102 Mann–Whitney). Thus, these results suggest that Shell Ox1Rs and activity played a stronger role in driving alcohol consumption in higher-drinking individuals.

We also generated histograms to examine the distribution of treatment effect across moderate and higher basal drinkers. In mice with higher basal intake (Figure 2I), control animals showed a normal distribution centered on log value of 2, indicating no change with treatment (100% of baseline). In contrast, higher-drinking Shell inhibition animals showed a clear shift to the left, indicating a significant decrease in intake compared with controls (p < 0.0001 Mann–Whitney). Mice with moderate basal intake (Figure 2J) had more subjects with increased intake, likely reflecting that drinking levels can increase more when starting from lower basal levels. In addition, there was no difference in the distribution of Shell inhibition and control moderate drinkers (p = 0.4925 Mann–Whitney). Together, these results confirm that Shell Ox1Rs and activity were essential for driving the excessive alcohol binging in higher drinkers, with limited contribution across moderate drinkers.

The overall lack of effect of Shell inhibition in moderate drinkers could reflect a floor effect, where drinking could not be reduced further. This possibility was unlikely for several reasons. We examined the impact of the GABAB receptor agonist baclofen, which reduces alcohol intake in humans and animals (Mirijello et al., 2015; Bell et al., 2017). Baclofen (5 mg/kg, i.p.) significantly decreased 2bc-DID drinking (Figure 3A; t₁₄ = 3.816, p = 0.0019), and, importantly, this was observed in both moderate and higher drinkers (Figure 3B), since the slope of basal intake versus treatment effect was nearly zero (slope = 0.0186, F_1,13 = 0.0311, p = 0.8628). The baclofen reduction in intake was specific to alcohol, since 5 mg/kg baclofen did not reduce concurrent water intake during alcohol sessions (Figure 3C; p = 0.3594), and did not reduce saccharin consumption in separate mice (Figure 3D; t₈ = 0.392, p = 0.7051). 1 mg/kg baclofen also slightly but significantly reduced alcohol drinking (Figure 3E; t₃₁ = 2.257, p = 0.0312), without an effect on concurrent water intake (Figure 3F; p = 0.8987). Thus, baclofen produced a specific reduction in alcohol drinking that was similar across moderate and higher drinkers. In addition, to further rule out simple mathematical effects, we examined concurrent water intake, which was not reduced overall by Shell Ox1R inhibition (Figure 1D,E). If anything, Shell SB had the opposite on water intake relative to basal alcohol drinking levels, with a bigger drop in water intake in moderate basal drinkers; this was significant for 2bc-DID (Figure 3G; F_1,29 = 4.456, p = 0.0435) although not 2bc-IA (Figure 3H; F_1,28 = 1.822, p = 0.1879). Taken together, these results suggest that the minimal impact of Shell inhibition across moderate drinkers was highly unlikely to be due to a floor effect or other artifact (such as higher basal intake tending to go down, and lower to go up). Also, the median alcohol drinking level in the moderate drinkers was ∼1.55–1.65 mg/kg, and we previously showed that ∼1.6 g/kg/2 h leads to ∼80 mg% BACs (Lei et al., 2016a), suggesting that even moderate drinkers on average consumed binge alcohol levels. Thus, these findings together strongly support our hypothesis that Shell Ox1Rs and activity were specifically critical for driving the excessive binging in higher-drinking individuals.

In addition to examining regulation of binge alcohol intake by Shell Ox1R- and muscimol-baclofen-sensitive activity, we also examined the importance of other signaling pathways which turned out to not alter intake; these became part of an aggregated control (no-change) group (as shown, e.g., in Figure 2B). First, evidence suggests that nicotine seeking is regulated by NAc Ox1Rs, PKC, and NMDARs (Plaza-Zabala et al., 2013). However, alcohol drinking was not impacted by inhibiting Shell PKC (Figure 4A; 0.03–0.4 μg/side chelerythrine; t₁₀ = 0.325, p = 0.7521) or NMDARs (Figure 4B; 0.3 μg/side AP5; t₁₂ = 1.036, p = 0.3205). Second, dopamine receptors are necessary for some alcohol-related behaviors (Hodge et al., 1997; Chaudhri et al., 2009; Pina and Cunningham, 2014; Hauser et al., 2015) but not others (Dickinson et al., 2003; Doherty and Gonzales, 2015), and systemic administration of the D1R blocker SCH23390 (0.3 mg/kg; Pina and Cunningham, 2014) did not alter intake (Figure 4C; t₁₁ = 0.724, p = 0.4842). Third, since Shell Ox1R blockers decrease intake, we tested whether orexinA peptide (100 pmol/side) might increase drinking, but found that Shell orexinA actually decreased intake (Figure 4E; t₇ = 4.188, p = 0.0041). In this regard, co-infusion of the Ox2r blocker TCS-OX2-29 (TCS, 3 ug/side, see Lei et al., 2016b) prevented the orexinA reduction in intake (Figure 4F; t₆ = 0.793, p = 0.4580). Shell Ox2rs do not regulate alcohol drinking (Lei et al., 2016b), but these results are consistent with Shell orexinA enhancing locomotion through Ox2rs but not Ox1Rs (Thorpe and Kotz, 2005; Kotani et al., 2008) (and we speculate that increased locomotion could disrupt focus on alcohol drinking). Finally, we previously showed that vmPFC Ox1Rs promote binge alcohol drinking in addition to Shell Ox1Rs (Lei et al., 2016b). However, NMDAR block with AP5 within vmPFC did not reduce alcohol drinking (Figure 4D; t₇ = 0.143, p = 0.8901). Together, these findings suggest that Shell PKC and NMDARs, vmPFC NMDARs, and D1Rs (tested systemically), were not critical for driving alcohol consumption. Thus, in order to construct the large no-change group to compare with our aggregated Shell inhibition group (Figure 2), results in Figure 4A–D,F were combined with data from no-change groups whose average values were previously reported (from Lei et al., 2016b: intra-Shell TCS, intra-Insula SB; from Lei et al., 2016a: systemic vehicle treatment during alcohol-only or quinine-alcohol drinking, 0.3 mg/kg SB for alcohol-only drinking, and systemic TCS for alcohol-only drinking).

Since Shell Ox1R and activity were critical for driving alcohol drinking in animals with higher basal intake, it would be particularly interesting to determine whether higher basal intake reflected a trait within particular individuals. Again, taking advantage of the large data set that we possess (n = 281), we examined whether alcohol drinking on the first day of access (the initial 24 h intake session in 2bc-DID drinkers) predicted 2 h intake levels averaged within each week of 2bc-DID intake. In fact, the initial-day drinking significantly predicted subsequent 2bc-DID intake levels on week 1 (Figure 5A; F_1,279 = 30.65; p < 0.0001; R² = 0.099), week 2 (Figure 5B; F_1,279 = 24.52; p < 0.0001; R² = 0.081), and week 3 (Figure 5C; F_1,279 = 35.11; p < 0.0001; R² = 0.112). Also, alcohol intake slightly but significantly increased across the first 3 weeks of 2bc-DID consumption (week 1: 2.535 ± 0.054 g/kg/2 h; week 2: 2.976 ± 0.052 g/kg/2 h; week 1: 3.139 ± 0.052 g/kg/2 h; F_2,280 = 57.78, p < 0.0001 one-way RM ANOVA; p < 0.01 difference between intake on any pair of weeks). Our results support the possibility that inter-individual variation in drinking levels is a trait, although there is some variability in the data that make it possible that some individuals increase or decrease intake across weeks of drinking. Thus, basal intake for all other studies was determined later in intake, nearer to the actual test sessions.