Agrp neuron activity is required for alcohol-induced overeating

Alcohol intake associates with overeating in humans. This overeating is a clinical concern, but its causes are puzzling, because alcohol (ethanol) is a calorie-dense nutrient, and calorie intake usually suppresses brain appetite signals. The biological factors necessary for ethanol-induced overeating remain unclear, and societal causes have been proposed. Here we show that core elements of the brain's feeding circuits—the hypothalamic Agrp neurons that are normally activated by starvation and evoke intense hunger—display electrical and biochemical hyperactivity on exposure to dietary doses of ethanol in brain slices. Furthermore, by circuit-specific chemogenetic interference in vivo, we find that the Agrp cell activity is essential for ethanol-induced overeating in the absence of societal factors, in single-housed mice. These data reveal how a widely consumed nutrient can paradoxically sustain brain starvation signals, and identify a biological factor required for appetite evoked by alcohol.

D rinking an apéritif to stimulate appetite has been mentioned as a human practice since at least the 5th century AD 1 . Modern studies confirm that alcohol intake acutely stimulates eating, and correlates with obesity 2-6 . Due to the rising incidence of obesity and its co-morbidities, the link between alcohol intake and overeating is becoming a recognized clinical concern 2-4 . Popular explanations for alcohol-induced overeating include an alcohol-induced loss of self-control, leading to a disregard for societal constraints on eating 4,6,7 . Biological factors required for alcohol-induced overeating remain elusive [2][3][4] . In homoeostatic models of eating control, the brain's appetite-stimulating signals are negatively regulated by nutrients [8][9][10][11] . An essential eating-stimulating signal is provided by Agrp neurons, a starvation-activated, molecularly distinct hypothalamic cell type 12,13 . The Agrp cells are found in mice and humans, and their chemogenetic or optogenetic stimulation is sufficient to cause rapid overeating, even in the absence of energy shortage [14][15][16][17] . Such eating despite energy sufficiency resembles the nutritionally paradoxical nature of alcohol-induced overeating. Although ethanol (EtOH, the active ingredient in alcoholic beverages) is the second most calorie-dense nutrient after fat, it stimulates rather than inhibits eating [2][3][4] . This suggests that EtOH creates a hunger state, but does not resolve whether EtOH highjacks normal homoeostatic signals of starvation, or evokes a different (for example, hedonic) hunger. Causal links between EtOH-induced overeating and brain hunger signals remain elusive. Establishing such a link would require a demonstration that EtOH-induced overeating is prevented by a selective interference with a molecularly defined hunger pathway.
Most molecular pathways associated with human obesity reside in the brain 18 , and are conserved in lower animals. Therefore, we hypothesized that EtOH may stimulate eating across species by distorting brain hunger signals. Here we probed this by: (1) determining EtOH impact on natural eating of animals isolated from social pressures, (2) examining EtOH action on genetically defined hunger signals and (3) establishing causality between specific hunger signals and EtOH-induced overeating. We present results suggesting that, in mice, EtOH sustains the activity of eating-stimulating hypothalamic Agrp cells, and Agrp cell activity is required for eating stimulation by EtOH.

Results
Effect of EtOH on food intake in mice. First, to test if EtOH-induced overeating occurs in non-human mammals, we examined the effects of EtOH on food intake of mice. We used single-housed mice to minimize social confounders, male and female mice to examine gender differences, and intraperitoneal (i.p.) administration of EtOH for precise dosage and to remove taste confounders. In an 'alcoholic weekend' experiment, each animal was given EtOH for three days, and saline for 3 days before and after (see Methods for consideration of EtOH dosage). The food intake was significantly and reversibly increased on the EtOH days, and the magnitude of this increase was similar in males and females (Fig. 1a,b; statistics are given in the figure legends). The effect of EtOH on food intake was also apparent at finer (hourly) time points ( Fig. 1c; Supplementary Fig. 2). These data suggest that alcohol-induced overeating is an evolutionarily conserved biological phenomenon occurring across mammals, irrespective of aesthetic beliefs and social conditioning.
Effect of EtOH on hypothalamic Agrp cells. Second, we sought to establish whether EtOH rapidly modulates endogenous hunger signals. To test if Agrp cell activity is modulated by EtOH, we injected a Cre-inducible ultrasensitive calcium activity reporter GCaMP6s (ref. 19) into the hypothalamic arcuate nucleus of Agrp-Cre mice ( Fig. 2a; Methods). Subsequent imaging of the Agrp-GCaMP6s network in brains slices revealed rapid, dose-dependent and reversible EtOH-induced activation of Agrp neurons (Fig. 2b,c). After i.p. injection of EtOH concentrations similar to those used in our in vivo experiments, deep brain EtOH concentrations have been measured to reach 30-80 mM in rodents 20-23 (see 'Ethanol dosage' in Methods for further discussion). We found that such EtOH concentrations produced similar Agrp cell activation (E40% cytosolic calcium elevation, Fig. 2c) to that caused by fasting or physiological hunger hormones 24 .
The calcium liberation implies activation of biochemical cascades in neurons 25 , but does not directly prove that the electrical spiking of Agrp neurons, the signal driving eating 16 , is increased. Thus, we also examined the electrical activity and membrane currents of Agrp cells in brain slices. In cell-attached electrical recordings (Methods), EtOH acutely and reversibly accelerated Agrp cell spiking (Fig. 3a). Similar effects were seen in whole-cell membrane potential recordings under pharmacological synaptic isolation (Fig. 3b,c), consistent with a direct postsynaptic action of EtOH. In contrast, non-Agrp arcuate  neurons (that is, putative satiety cells 16 ) were not excited by EtOH ( Supplementary Fig. 1A). While this suggests a degree of cell-type selectivity of EtOH actions in the arcuate, we note that the unidirectional inhibition of satiety cells by hunger cells 26,27 implies that such selectivity may not be critical for increasing eating (Discussion). We next addressed mechanisms of the EtOH-induced excitation. Membrane potential recordings revealed that EtOH-induced excitation involves membrane depolarization and increased conductance (Fig. 3b,c). Whole-cell current-voltage relations further showed that this conductance increase is due to an EtOH-activated non-selective ionic current (Fig. 3d, equilibrium potential ¼ À 32±3.1 mV; n ¼ 7 cells). Buffering cytosolic Ca 2 þ increases by 10 mM BAPTA in the pipette solution abolished EtOH-induced excitatory actions (Fig. 3e), suggesting that a rise in cytosolic calcium is required for EtOH action. Cytosolic calcium elevations can activate plasmalemmal 3Na þ /Ca 2 þ exchanger (NCX), which can trigger membrane depolarization and excitation due to NCX electrogenicity [28][29][30][31][32] . Thus, we tested whether EtOHinduced excitation of Agrp neurons requires this transporter. Bath application of the selective NCX inhibitor KB-R7943 (refs 31,33) prevented EtOH-induced excitation (Fig. 3f). Together, these data suggest that EtOH-induced excitation of Agrp cell membrane is mediated by cytosolic calcium signals and plasmalemmal NCX activity. Overall, these data demonstrate that EtOH evokes a functional remodelling in the brain's biophysical generators of hunger drive, thereby sustaining false 'starvation alarms' despite extracellular nutrient sufficiency.
Effect of Agrp cell silencing on EtOH-induced eating. Third, to investigate whether Agrp cell activity is required for EtOH-induced overeating, we targeted Cre-dependent neuroinhibitory receptor hM4Di, or control protein ChR2-mCherry, to Agrp neurons in Agrp-Cre mice ( Fig. 4a; Supplementary Fig. 1B). This made it possible to block Agrp-hM4Di neuron spiking by the physiologically inert hM4Di receptor ligand CNO (Fig. 4b,c). The CNO concentrations we used did not affect spiking of Agrp-ChR2 neurons ( Supplementary Fig. 1B), nor eating in mice not expressing hM4Di ( Supplementary Fig. 1C,D), confirming that CNO did not modulate the parameters under study without hM4Di. In the absence of CNO, EtOH increased food intake in mice with Agrp-hM4Di or Agrp-ChR2 neurons, (Fig. 4d), confirming that transgene expression per se does not prevent ethanol-associated eating. We then gave all mice CNO, to suppress Agrp neurons of Agrp-hM4Di mice. Under this condition, eating of Agrp-hM4Di mice is reduced (Fig. 4d), as previously reported 17 , but-importantly-despite this lower baseline, eating stimulation remains possible by eating-control neurons downstream of Agrp 34 . However, despite this existence of neurons that can increase eating in the absence of Agrp 34,35 , we found that in the presence of CNO, EtOH no longer stimulated food intake of Agrp-hM4Di mice (but stimulated it normally in control Agrp-ChR2 mice, Fig. 4d, left plot). While these findings do not preclude EtOH actions on targets other than Agrp, they do imply that such actions are insufficient to cause eating without Agrp cell activity.
Finally, we analysed whether EtOH controls the contribution of the natural activity of Agrp cells to eating in vivo. We hypothesized that if EtOH-activated Agrp neurons in vivo (as predicted by in vitro data, Figs 2 and 3), then EtOH should increase Agrp cell-dependent eating. Because CNO silences Agrp-hM4Di cells (Fig. 4b,c), the magnitude of the CNO effect on eating in Agrp-hM4Di mice reflects Agrp cell-mediated eating, that is, in vivo Agrp cell activity (if Agrp cell activity is low, its silencing will have a small effect on eating, and if Agrp cell activity is high, its silencing would have a large effect). We found that the effects of CNO on eating in Agrp-hM4Di mice were significantly greater in the presence of EtOH (Fig. 4d, right plot), consistent with the idea that EtOH increases Agrp cell contribution to eating in vivo (Discussion).
Together, these data identify Agrp cell activity as a critical process required for the EtOH action on food intake.

Discussion
Agrp cell activity potently increases motivation for feeding, and has been called a neural correlate of hunger 17 . Thus our data suggests that alcohol sustains fundamental appetite signals, not just disinhibits their behavioural manifestation. Apart from the Agrp cells, other powerful controllers of food intake exist in the brain, for example, MC4R neurons of the paraventricular hypothalamus 36 , Vgat neurons of the lateral hypothalamus 37 and neurons of the parabrachial nucleus 34 . Crucially, the control of feeding by some of these modules does not require Agrp neurons 35 . Therefore, alcohol actions on the latter non-Agrp eating controllers would be expected to alter eating even when Agrp cells are silenced. However, we found that when Agrp cells were chemogenetically silenced, alcohol no loger modulated food intake. Thus, we propose that the alcohol-associated activity of Agrp neurons-mediated either by direct alcohol actions described here, or via a still-unknown upstream element-is the critical step in alcohol-induced overeating.
There are previous reports of ethanol actions on diverse cellular targets, including eating-regulating systems such as midbrain dopamine neurons, and on diverse and ubiquitously  expressed molecular targets, such as neuronal GABA and NMDA receptors [38][39][40][41][42] . We believe that these previous reports are compatible with our findings and conclusions. First, our finding that the Agrp cells are an essential locus of alcohol-induced eating does rule out that there are other sites in the brain, where alcohol can plausibly influence eating. Second, similar to previous studies of other neurons, we found that ethanol acts on Agrp neurons by modulating fairly ubiquitous molecular operators (calcium and NCX), rather than a process that is unique to Agrp cells. We propose that this lack of specificity at the cellular and molecular levels is compatible with a specific action at the whole-system level, due to the non-random, directed interactions between different neuronal controllers of behaviour 43 . Nonrandom neuronal connectivity has been reported throughout the brain 43 , including appetite systems (for example, Agrp cells inhibit POMC cells but not vice versa 27 ). Such a unidirectional network connectivity would 'filter' non-selective actions at the cellular/molecular levels (for example, direct ethanol-induced excitation of both hunger and satiety cells) into a behaviourselective action (that is, eating versus non-eating) at the system output level. Testing this proposal by developing tools for disrupting directed connectivity, in conjunction with cell-type specific in vivo recordings, is an interesting direction for future work, especially considering that Agrp neurons can be modulated differently in vitro and in vivo. Overall, our findings provide an explanation for how a commonly consumed nutrient may generate a positive feedback on energy intake, offering new mechanistic and conceptual insights into pathological overeating behaviour linked to morbid disorders.

Agrp-Cre
Cre-dependent hM4Di-mCherry (c) Left, group data of membrane potential effect in experiment shown in b (n ¼ 5 cells). P value is from a paired t-test, t ¼ 6.689, d.f. ¼ 4. Right, Spike firing responses in experiment shown in b for all cells. (d) Left, effects of EtOH (2 g kg À 1 i.p.) and CNO (5 mg kg À 1 i.p.) on food intake of Agrp-Cre mice expressing hM4Di or ChR2 in Agrp cells. Food intake under each drug condition represents a mean of 3 days of drug treatment (same design as in Fig. 1a). Four male mice per group (age/gender-matched littermate pairs). Two-way repeated measures analysis of variance: treatment F(3, 18) ¼ 129.9, Po0.0001; genotype F(1, 6) ¼ 45.44, P ¼ 0.0005; interaction F(3, 18) ¼ 53.28, Po0.0001; numbers between bars are P values from Tuckey's multiple comparison corrections. Right, Agrp cell-driven food intake (defined as reduction in food intake evoked by CNO in Agrp-hM4Di mice) in the presence and absence of EtOH (n ¼ 8 mice). Number above bars is P value from a two-tailed paired t-test, t ¼ 5.055, d.f. ¼ 7. Values shown are individual points and/or means ± s.e.m.

Methods
Animals and feeding studies. All animal procedures followed United Kingdom regulations (Animals, Scientific Procedures Act 1986) and were approved by local welfare committees and veterinarians. Adult (10-12-week old) male and female mice on the C57/BL6 background were used for all experiments.
For experiments involving transgene expression in Agrp neurons, we used, where indicated, two previously described and characterized transgenic mouse lines: Agrp-IRES-Cre (Jax no. 012899 (ref. 44)) and NPY-hrGFP (Jax no. 006417 (ref. 45)). Mice were kept on a standard 12 h light-dark cycle (lights on at 7:00), and on standard mouse chow (LabDiet 5021) and water ad libitum. Room temperature was kept at 22°C. During food intake measurements, the chow was provided in 'anti-spill' cylindrical hoppers (TSE Systems). In Fig. 1a,b,d, measurements were performed in home cages by lowering the hoppers onto cage floor through a modified cage lid, and food intake was measured daily by manual weighing of the hoppers. In Fig. 1c, food was weighed automatically in suspendedhopper cages (TSE PhenoMaster) equipped with custom-elevated floors (with openings to maintain hopper suspension) that brought food closer to the floor to facilitate access despite potential ethanol-induced motor incoordination 46 . When placed in unfamiliar cages, mice were allowed to acclimatize for at least one week before measurements.
Drugs (EtOH 2 g kg À 1 , CNO 5 mg kg À 1 ) were sterile-filtered and injected i.p. at 12.65 ml g À 1 total volume (0.9% NaCl saline was used as vehicle). Control i.p. injections contained saline only (12.65 ml g À 1 ). Mice were injected at 19:00 (the beginning of the dark phase) and subsequent eating was measured by weighing the food consumed at different times after injection. In an alternative experimental design (which produced similar data to the main findings), mice received a single 1 g kg À 1 EtOH i.p. injection at 16:00 and food intake was monitored using standard TSE Phenomaster cages ( Supplementary Fig. 2).
Ethanol dosage. For in vivo experiments, we used EtOH doses of 2 g kg À 1 in Figs 1 and 4, and 1 g kg À 1 in Supplementary Fig. 2; for in vitro experiments, we used 10-100 mM EtOH. These doses were based on the following considerations. First, the i.p. EtOH doses we used are compatible with studies of mouse behaviour 47 , and have been reported to elevate deep brain EtOH concentrations to 20-80 mM in rodents in vivo [20][21][22][23] , which we here mimicked in brain slices. We confirmed these reports of deep brain ethanol elevations using a custom EtOH sensor in the arcuate nucleus ( Supplementary Fig. 3).
Second, these doses resemble those relevant for humans (after correcting for body weight). Specifically, these doses would approximately represent the amounts reported to be consumed by 27% of young UK people at least once per week (418 units alcohol, according to a UK government survey for 2012: http://www.ons.gov.uk/ons/rel/ghs/opinions-and-lifestyle-survey/drinking-habitsamongst-adults--2012/sty-alcohol-consumption.html). Assuming that 1 alcohol unit contains 8 g EtOH, the 18 units consumed by a 70 kg human equates to an ethanol dose of 2.06 g kg À 1 . The blood EtOH concentration of a 70 kg human male was estimated (Widmark formula) to be B50 mM after consuming the 18 units of alcohol. Since EtOH readily crosses membranes and rapidly equilibrates across the blood-brain barrier, similar EtOH concentrations occur in the blood and the brain, and in different brain regions 20,23 .
Network activity imaging. For calcium imaging of Agrp cell activity, AAVs carrying Cre-dependent GCamp6s (rAAV9.CAG.Flex.GCaMP6s.WPRE.SV40; titre 2.74 Â 10 13 gc ml À 1 , Penn Vector Core) were stereotaxically injected into the arcuate nucleus of the Agrp-Cre mice. Two 100 nl injections of the GCaMP6s virus were made at: À 1.4 mm from Bregma; ± 0.3 mm from midline; and À 5.80 and À 5.70 mm from skull surface. Cre dependency of viral expression was confirmed by injections of the Cre-dependent AAV into the brains of Cre-negative mice (n ¼ 3 mice). Acute coronal brain slices were prepared 10-14 days after AAV injection, and confocal imaging and analysis of GCaMP6s signals was performed as previously described 48 .
Single-neuron electrophysiology. Acute coronal brain slices were prepared and cell-attached and whole-cell recordings performed as previously described 49 . Briefly, fluorescent neurons were visualized an upright Olympus BX51WI microscope equipped with an oblique condenser, a xenon lamp and appropriate filter sets. Somatic recordings were carried out at 37°C using a HEKA EPC 10 USB patch-clamp amplifier controlled by Patchmaster software (HEKA Elektronik, Germany), after slices were allowed to equilibrate in the recording chamber for 10-15 min. Arcuate Agrp neurons were identified for slice electrical recordings using either Cre-dependent fluorescent transgenes in Agrp-Cre mice (hD4Di-mCherry, ChR2-mCherry); or NPY-hrGFP mice since all NPY expression in arcuate neurons co-localizes with Agrp 50 .
Immunohistochemistry. To confirm hM4Di-mCherry transgene targeting to the arcuate nucleus, mice were perfused with 4% paraformaldehyde, and dissected brains were cryoprotected in 30% sucrose solution for a minimum of 48 h. Free floating cryosections (50 mM) of the arcuate nucleus and cell nuclei were counterstained with Hoescht.
Statistical analyses. Statistical tests and descriptive statistics were performed as described in the figure legends. In general, we found that 20-30% of Agrp neurons did not respond to EtOH application in our in vitro preparation (for example, see Fig. 2b); these non-responding cells were included in group analysis of calcium responses to assess net effect of EtOH on all Agrp neurons (Fig. 2c), but excluded from group quantification of electrical responses to measure EtOH response size at the level of individual cells (Fig. 3a-d). In parametric tests, normality was assessed with Shapiro-Wilk, Lilliefors, Anderson-Darling and Kolmogorov-Smirnov (KS) tests, and variances were assessed for homogeneity using Levene's, Brown-Forsyth and F-tests. The analysis was performed using Origin 2015 (OriginLab) or GraphPad Prism 6.
Data availability. The data that support the findings of this study are available from the authors on reasonable request.