Gs-coupled GPCR signalling in AgRP neurons triggers sustained increase in food intake

Agouti-related peptide (AgRP) neurons of the hypothalamus play a key role in regulating food intake and body weight, by releasing three different orexigenic molecules: AgRP; GABA; and neuropeptide Y. AgRP neurons express various G protein-coupled receptors (GPCRs) with different coupling properties, including Gs-linked GPCRs. At present, the potential role of Gs-coupled GPCRs in regulating the activity of AgRP neurons remains unknown. Here we show that the activation of Gs-coupled receptors expressed by AgRP neurons leads to a robust and sustained increase in food intake. We also provide detailed mechanistic data linking the stimulation of this class of receptors to the observed feeding phenotype. Moreover, we show that this pathway is clearly distinct from other GPCR signalling cascades that are operative in AgRP neurons. Our data suggest that drugs able to inhibit this signalling pathway may become useful for the treatment of obesity.

D uring the past decades, obesity has emerged as a major health crisis in most parts of the world 1 . Obesity represents a major risk factor for many severe diseases including type 2 diabetes, cancer and cardiovascular disease, thus posing an enormous socio-economic burden 2 .
Given the limited success of behavioural approaches towards achieving long-lasting weight loss, there is a clear need for effective pharmacological strategies to stem the current obesity epidemic 3 . Unfortunately, the number and efficacy of appetite-suppressing drugs that have been approved for clinical use is severely limited 3 .
To guide the design of novel appetite-suppressing drugs, it is essential to map the neuronal circuits that regulate food intake under physiological and pathophysiological conditions. In the present study, we focused on a small subpopulation of hypothalamic neurons located in the arcuate nucleus (ARC) of the hypothalamus, which synthesize and release agouti-related peptide (AgRP), a neuropeptide endowed with potent, long-lasting orexigenic activity 4 . These neurons, which are generally referred to as AgRP neurons, release two additional agents that promote acute feeding, neuropeptide Y (NPY) and GABA, a biogenic amine neurotransmitter 4,5 . At present, the relative roles of these three orexigenic agents in stimulating appetite in response to different hormones or neurotransmitters are not well understood.
Numerous studies have shown that AgRP neurons play a key role in regulating food intake and energy homeostasis [6][7][8][9][10] . For example, acute ablation of AgRP neurons in adult animals results in the cessation of feeding and, ultimately, starvation 6,7 .
As is the case with essentially all other cell types, the activity of AgRP neurons is predicted to be regulated by cell surface receptors belonging to the superfamily of G protein-coupled receptors (GPCRs), which are linked to distinct functional classes of G proteins 11,12 . So far, the functional consequences of activating the various GPCR/G protein signalling pathways in AgRP neurons have not been studied systematically. To address this issue, we started to employ a new set of pharmacological tools referred to as DREADDs (designer receptors exclusively activated by designer drugs). DREADDs represent mutant muscarinic acetylcholine receptors that are unable to bind acetylcholine or any other endogenous ligands 13 . However, DREADDs can be selectively activated by clozapine-N-oxide (CNO), a compound that is otherwise pharmacologically inert [13][14][15][16][17] . Importantly, we and others recently developed DREADDs endowed with distinct coupling G protein properties [15][16][17] , making it possible to monitor the in vivo consequences of activating distinct GPCR signalling pathways in a drug (CNO)-dependent fashion in specific cell types. Such studies are not possible to perform with native GPCRs, which, as a general rule, are expressed in multiple tissues and cell types 18 .
We recently generated mice that selectively express a G q -DREADD (hM3Dq; ref. 13) or a Gi-DREADD (hM4Di; ref. 13) in AgRP neurons of the ARC 19 . We found that CNO treatment of these animals resulted in greatly enhanced or reduced food intake, respectively 19 . In the present study, we selectively expressed a G s -linked DREADD (GsD; ref. 16) in mouse AgRP neurons. We demonstrate that drug-mediated stimulation of this receptor in AgRP neurons leads to a sustained increase in food intake that almost exclusively depends on the release of AgRP. In addition, we delineate a novel cellular pathway predicted to underlie this activity and show that this pathway is clearly distinct from the one involved in triggering acute food intake via activation of G q -coupled GPCRs in AgRP neurons. These new findings suggest that drugs capable of reducing G s signalling in AgRP neurons might become clinically useful as appetite-suppressing drugs.

Results
Expression of a G s -coupled DREADD in AgRP neurons. In the present study, we selectively expressed a G s -coupled designer GPCR (G s DREADD ¼ GsD) in AgRP neurons of the mouse ARC. The generation and pharmacological properties of this construct have been described previously 14 . The GsD designer receptor, like other members of the DREADD family 13,[15][16][17] , is no longer recognized by endogenous ligands, but can be selectively activated by CNO, an otherwise pharmacologically inert compound. To facilitate the detection of GsD expression in the mouse brain, we fused an enhanced green fluorescent protein (eGFP) tag to the C terminus of GsD (GsD-eGFP). Studies with transfected COS-7 cells demonstrated that the eGFP tag did not interfere with proper receptor expression and the ability of CNO to induce GsD-dependent G s activation, as measured in cAMP accumulation assays (Supplementary Table 1).
To selectively express GsD (GsD-eGFP) in mouse AgRP neurons, we used a Cre-recombinase-dependent adeno-associated virus (AAV) to target GsD to AgRP neurons. Stable transgene inversion was achieved by using FLEX Switch technology 20 . We stereotaxically injected an AAV that contained the GsD-eGFP coding sequence in the reverse orientation (AAV-hSyn-DIO-GsD-eGFP) into the ARC of AgRP-ires-Cre mice (bilaterally), which selectively express Cre recombinase in AgRP neurons. To demonstrate the selective expression of GsD-eGFP by AgRP neurons, we injected the same AAV into the ARC of AgRP-ires-Cre; R26-loxSTOPlox-L10-GFP mice (L10-GFP mice) 21 . In this mouse line, eGFP is selectively expressed, in a Cre-dependent fashion, in AgRP neurons 21 . GsD-eGFP expression was detected by an antibody directed against the eGFP tag (note that no native GFP fluorescence was observed with this construct, most likely due to low GsD-eGFP expression levels). This approach revealed the selective expression of the GsD-eGFP designer receptor in AgRP neurons of the ARC (Fig. 1a,b). For brevity, we refer to the AAV-hSyn-DIO-GsD-eGFP-injected AgRP-ires-Cre mice simply as 'GsD-AgRP mice'.
For control purposes, we used the same approach to inject the AAV-hSyn-DIO-hM3Dq-mCherry virus into the ARC of AgRP-ires-Cre mice 19 (note that hM3Dq represents a G q -coupled DREADD). We refer to these mice, which express the hM3Dq designer receptor selectively in AgRP neurons 19 , as 'hM3Dq-AgRP mice' throughout the manuscript. Whole-cell patch clamp electrophysiological recordings revealed clear differences in CNO responses between GsD-AgRP and hM3Dq-AgRP neurons. We previously reported that CNO rapidly depolarizes the resting membrane potential of hM3Dq-AgRP neurons (see Fig. 1c, bottom panel, for a representative trace; see ref. 19). In contrast, bath application of CNO (10 mM) had no significant effect on the membrane potential of GsD-AgRP neurons (baseline, À 44.3 ± 5.5 mV; after CNO, À 44.8±5.6 mV; n ¼ 7; Fig. 1c, top panel, d).
To trigger GsD signalling selectively in mouse AgRP neurons, we injected GsD-AgRP mice with a single dose of CNO (1 mg kg À 1 intraperitoneal (i.p.)) or saline (as a negative control). Gene expression studies were performed 2 h after CNO injections. CNO treatment of GsD-AgRP mice had no detectable effect on c-fos expression in AgRP neurons (Fig. 2b,g and Supplementary  Fig. 1b). GsD activation is predicted to increase cellular cAMP levels and to stimulate the activity of protein kinase A (PKA) 14,22 , resulting in the phosphorylation of many cellular proteins including the transcription factor CREB. Consistent with this concept, we found that CNO treatment of GsD-AgRP mice promoted CREB phosphorylation selectively in AgRP neurons (Fig. 2a,e and Supplementary Fig. 1a). In contrast, saline-injected GsD-AgRP mice showed little or no pCREB staining in AgRP neurons (Fig. 2a,e and Supplementary Fig. 1a). These data clearly indicate that the GsD designer receptor is functional in AgRP neurons.
Interestingly, hM3Dq-AgRP mice displayed a functional profile that was opposite to that observed with the GsD-AgRP mice. In hM3Dq-AgRP mice, CNO treatment triggered strong c-fos staining in hM3Dq-expressing AgRP neurons, while little or no c-fos expression was observed with saline-treated hM3Dq-AgRP mice (Fig. 2d,h and Supplementary Fig. 1d; also see ref. 19). In contrast to GsD-AgRP mice, CNO treatment of hM3Dq-AgRP mice did not result in a detectable increase in CREB phosphorylation in AgRP neurons (Fig. 2c,f and Supplementary  Fig. 1c). A quantitative summary of the pCREB and c-fos expression data is given in Fig. 2e-h.
GsD activation promotes sustained feeding. We next examined whether GsD signalling in AgRP neurons was able to modulate food intake. Specifically, we injected GsD-AgRP mice with a single dose of CNO (1 mg intracerebroventricular (i.c.v.)) or saline early in the morning (between 9:00 and 10:00) when mice are satiated and consume little food. In most feeding experiments, we delivered CNO i.c.v. since this route of application yielded more consistent data, as compared with i.p. administration of CNO. Subsequently, we monitored acute food intake over a 4-h period as well as long-lasting effects on food intake over the following 3 days. In wild-type (WT) control mice, CNO administration (1 mg i.c.v.) had no significant effect on food intake ( Supplementary Fig. 2). In contrast, CNO treatment of GsD-AgRP mice triggered a robust increase in food intake during the first 4 h after CNO injection (Fig. 3a). Surprisingly, this CNO-induced orexigenic effect was very long-lasting (for 3 days; Fig. 3b). The orexigenic effects of CNO were dose dependent, as shown in Supplementary Fig. 3. Consistent with the food intake data, CNO-injected GsD-AgRP mice displayed a significant increase in body weight on days 2 and 3 after CNO administration (Fig. 3c).
For control purposes, we carried out analogous experiments with hM3Dq-AgRP mice. As reported previously 19 , CNO treatment (1 mg i.c.v.) of hM3Dq-AgRP mice triggered a striking increase in food intake during the first 24 h after CNO administration (Fig. 3d,e). This effect was even more robust than that observed with GsD-AgRP mice (please note that the y axes have different scales in Fig. 3a,d). However, in contrast to GsD-AgRP mice in which CNO triggered long-lasting orexigenic effects (Fig. 3b), CNO-treated hM3Dq-AgRP mice showed normal food intake on days 2 and 3 after CNO injection (Fig. 3e). In agreement with this observation, CNO-treated hM3Dq-AgRP mice did not show any changes in body weight on days 2 and 3 after CNO injection, as compared with saline-treated hM3Dq-AgRP mice (Fig. 3f).
Previous studies have demonstrated that AgRP treatment of rodents has long-lasting effects on food intake 23,24 . We therefore speculated that the sustained feeding phenotype displayed by the CNO-injected GsD-AgRP mice was due to GsD-dependent AgRP release. To address this issue, we injected WT control mice with a single dose (3 mg i.c.v.) of an active AgRP fragment, AgRP 83-132 , using the same experimental set-up as described above. Strikingly, the AgRP 83-132 -injected WT mice showed increases in food intake and body weight that were characterized by a pattern very similar to that observed with CNO-treated GsD-AgRP mice ( Fig. 3g-i).
GsD-induced feeding requires AgRP release. To further test the hypothesis that the orexigenic effects observed after activation of the GsD designer receptor require AgRP, we used two different experimental strategies.
First, we generated GsD-AgRP mice that carried a single copy of the agouti 'lethal yellow' mutant allele (A y ) (GsD-AgRP, A y mice). For comparison, we also introduced the A y mutation into hM3Dq-AgRP mice (hM3Dq-AgRP, A y mice). The A y mutation leads to the ectopic expression of agouti protein in the brain (as well as other tissues) where it blocks central melanocortin receptors, in a fashion similar to AgRP 25 . As a result, mice carrying the A y mutation are predicted to show reduced sensitivity to the orexigenic actions of AgRP 26 . We found that The upper trace shows a representative recording from a GsD-AgRP neuron in response to CNO bath application. For comparison, the lower trace depicts a representative recording derived from an hM3Dq-AgRP neuron, indicating that CNO depolarizes hM3Dq-AgRP neurons (also see ref. 19). In this latter experiment, AgRP-ires-Cre mice were injected with the AAV-hSyn-DIO-hM3Dq-mCherry virus, which codes for a G q -coupled DREADD (hM3Dq). Data are given as means±s.e.m. (n ¼ 7 neurons from 4-5 mice).
CNO failed to stimulate food intake in GsD-AgRP mice carrying the A y mutation (Fig. 4a). In contrast, the presence of the A y mutation did not interfere with the ability of CNO to promote robust food intake in hM3Dq-AgRP mice (Fig. 4b), in agreement with the outcome of a previous study employing optogenetic interrogation of AgRP neurons 26 . Second, we co-injected GsD-AgRP mice with CNO (0.5 mg i.c.v.) and an anti-AgRP antibody (1 mg i.c.v.) 27 . Treatment of GsD-AgRP mice with the anti-AgRP antibody alone had no significant effect on food intake ( Supplementary Fig. 4). Strikingly, the anti-AgRP antibody completely blocked the ability of CNO to promote food intake in GsD-AgRP mice (Fig. 4c). In contrast, treatment of hM3Dq-AgRP mice with the anti-AgRP antibody had no significant effect on CNO-mediated orexigenic effects (Fig. 4d).
These data clearly indicate that the orexigenic effects mediated by G s -and G q -coupled receptors selectively expressed in AgRP neurons involve different cellular/neuronal mechanisms.
GsD-induced feeding does not require NPY/GABA release. A recent study 5 demonstrated that the acute orexigenic effect observed with CNO-treated hM3Dq-AgRP mice requires the release of NPY and/or GABA. We therefore explored the possibility that NPY or GABA might also contribute to the CNO-mediated increase in food intake in GsD-AgRP mice. Initially, we demonstrated that treatment of GsD-AgRP mice with NPY (0.85 mg i.c.v.) caused a marked increase in food intake ( Supplementary Fig. 5a), consistent with the wellknown orexigenic activity of NPY 28 . Co-injection of NPY with BIBO3304 (3 mg i.c.v.), a selective NPY Y 1 receptor antagonist 29 , completely abolished NPY-induced feeding in GsD-AgRP mice ( Supplementary Fig. 5a), indicating that the orexigenic effects of NPY are mediated by NPY Y 1 receptors. Treatment of control mice (AgRP-ires-Cre mice that had been injected with AAV-hSyn-DIO-mCherry) with BIBO3304 alone had no effect on basal food intake ( Supplementary Fig. 5b). We next injected GsD-AgRP mice with either CNO alone (0.5 mg i.c.v.) or with CNO plus BIBO3304 (0.5 and 3 mg i.c.v., respectively). BIBO3304 treatment had no significant effect on CNO-induced food intake at the 0.5, 1 and 2 h time points but caused a partial reduction in food intake 4 h after CNO injection (Fig. 5a). This observation suggests that NPY does not make a major contribution to the orexigenic effect caused by activation of the GsD designer receptor in AgRP neurons.
To test the possibility that GABA release from AgRP neurons might contribute to the ability of CNO to increase food intake in GsD-AgRP mice, we generated GsD-AgRP mice that selectively lacked the vesicular GABA transporter (Vgat) in AgRP neurons (GsD-AgRP-ires-Cre, Vgat flox/flox mice). In these mice, GABA cannot be released from AgRP neurons since it cannot be packaged in synaptic vesicles. We found that selective deletion of the Vgat gene in AgRP neurons had no significant effect on the ability of CNO (1 mg kg À 1 i.p.) to stimulate food intake in GsD-AgRP mice (Fig. 5b). This observation indicates that the orexigenic effect triggered by activation of the GsD designer receptor in AgRP neurons does not require the release of GABA.
GsD activation promotes hypothalamic AgRP release. To further confirm that activation of GsD expressed by AgRP neurons promotes hypothalamic AgRP release, we carried out ex vivo AgRP secretion experiments. Specifically, we incubated hypothalamic slices prepared from GsD-AgRP mice, with artificial cerebrospinal fluid (aCSF) for 1 h at 37°C (to determine basal AgRP release), followed by a second 1-h incubation in the presence of CNO (15 mM). In parallel, we carried out analogous experiments with hypothalamic slices prepared from mCherry-AgRP control mice (AgRP-ires-Cre mice that had been injected bilaterally with the AAV-hSyn-DIO-mCherry virus). Strikingly, CNO triggered a significant increase in AgRP release from hypothalamic preparations of GsD-AgRP mice (Fig. 6a). This effect was absent in hypothalamic slices prepared from mCherry-AgRP control mice (Fig. 6b). Taken together, these data clearly indicate that activation of GsD, a G s -coupled designer receptor, in AgRP neurons stimulates hypothalamic AgRP release.
GsD activation increases AgRP and KLF4 expression. After having demonstrated that AgRP is essential for the GsD-mediated increases in food intake, we next examined whether hypothalamic AgRP expression levels were altered in CNO-treated GsD-AgRP mice. Specifically, we injected GsD-AgRP mice with CNO (1 mg kg À 1 i.p) and prepared hypothalamic RNA 2 h later.  For control purposes, we carried out similar experiments with hM3Dq-AgRP mice. We found that CNO treatment of GsD-AgRP mice led to a robust (B2.5-fold) increase in hypothalamic AgRP expression, as compared with saline-injected GsD-AgRP control mice (Fig. 7a). In contrast, under the same experimental conditions, hypothalamic AgRP expression levels remained unchanged in hM3Dq-AgRP mice (Fig. 7a). Hypothalamic NPY expression was not significantly affected by CNO treatment of either GsD-AgRP or hM3Dq-AgRP mice (Fig. 7a), indicating that activation of GsD in AgRP neurons selectively upregulates AgRP transcript levels.
Krüppel-like factor 4 (KLF4) is a zinc finger transcription factor, which is expressed in various tissues and modulates many developmental processes and biological functions 30,31 .  Interestingly, recent studies have shown that drugs that increase intracellular cAMP levels promote KLF4 expression in different cell types 32,33 , indicative of a possible link between enhanced G s signalling and increased KLF4 expression. Moreover, Ilnytska et al. 34 recently demonstrated that KLF4 is a potent activator of the AgRP promoter. We therefore speculated that activation of the GsD designer receptor in AgRP neurons might lead to an increase in hypothalamic KLF4 expression. In agreement with this hypothesis, we found that CNO treatment (1 mg kg À 1 i.p) of GsD-AgRP mice resulted in a significant increase in hypothalamic KLF4 expression levels (by B50%, as compared with saline-injected control mice; hypothalamic RNA was prepared 2 h after CNO injections; Fig. 7a). In contrast, CNO treatment of hM3Dq-AgRP mice did not lead to an increase but rather a reduction in hypothalamic KLF4 expression levels (Fig. 7a).
To investigate whether the GsD-mediated stimulation of KLF4 mRNA expression also led to increased levels of KLF4 protein, we microinjected the AAV-hSyn-DIO-GsD-eGFP virus unilaterally into the ARC of AgRP-ires-Cre mice (right side), leading to the expression of the GsD designer receptor in AgRP neurons of the AAV-injected side only (Fig. 7b). Treatment of the AAV-injected mice with CNO (1 mg kg À 1 i.p.) resulted in a robust increase in KLF4 protein expression in AgRP neurons expressing the GsD receptor, measured 1 h after CNO administration (Fig. 7b, right side). Only weak KLF4 staining was detected on the side that did not express the GsD receptor (Fig. 7b, left side).
GsD-mediated increase in AgRP expression requires KLF4. We next studied whether enhanced KLF4 expression is required for To test this hypothesis, we took advantage of the availability of a compound called kenpaullone (Ken), a suppressor of KLF4 transcription 35 . Consistent with our hypothesis, we found that pretreatment of GsD-AgRP mice with Ken (2 mg i.c.v.) led to a pronounced reduction in CNO-mediated increases in both KLF4 and AgRP hypothalamic expression levels (Fig. 7c). In this figure, gene expression levels are displayed normalized relative to those obtained with CNO-treated GsD-AgRP mice that had been pretreated with i.c.v. saline rather than Ken. Ken pretreatment had no significant effect on hypothalamic NPY expression levels in CNO-injected GsD-AgRP mice (Fig. 7c). These data support the concept that GsD-mediated stimulation of AgRP expression requires an increase in KLF4 activity.
The KLF4 promoter is a direct target of activated CREB. To examine whether the KLF4 promoter is directly activated by CREB, we carried out a series of KLF4 promoter-luciferase reporter assays (Supplementary Fig. 6a). The mouse KLF4 promoter contains two half CREB-responsive element (CRE) sites (CGTCA) that form a cluster within 300 nucleotides upstream of the transcription start ( Supplementary Fig. 6c) 36 . We found that mutational modification of the first half CRE site ( À 269 site) had no significant effect on forskolin-mediated stimulation of KLF4 promoter activity. In contrast, mutation of the second half CRE site ( À 232 site) or of both half CRE sites virtually abolished this response ( Supplementary Fig. 6b), indicating that only the second half CRE site is transcriptionally relevant. Taken together, these data strongly suggest that the KLF4 promoter is a direct target of activated CREB. PAC 1 receptors promote KLF4-dependent AgRP expression. We next examined whether stimulation of endogenously expressed G s -coupled receptors resulted in cellular actions similar to those observed with the GsD designer receptor. Initially, we carried out studies with hypothalamic GT1-7 cells 37 , which are known to release AgRP depending on their energy status 38 . We found that incubation of GT1-7 cells with forskolin (10 mM) led to a time-dependent increase in the expression of both AgRP and KLF4 ( Supplementary Fig. 7). These forskolin-induced effects could be completely blocked by pretreatment of cells with the PKA inhibitor, H89 ( Supplementary Fig. 7), strongly suggesting that cAMP-mediated activation of PKA promotes AgRP and KLF4 expression. GT1-7 cells express various G s -linked GPCRs, including the PAC 1 receptor, a pituitary adenylate cyclase-activating peptide (PACAP) receptor subtype 39,40 . We therefore treated GT1-7 cells with PACAP 1-38 (1 mM) to stimulate endogenously expressed PAC 1 receptors. PACAP 1-38 , similar to forskolin, triggered pronounced increases in the expression of both AgRP and KLF4 (Fig. 8a,b). These effects were not observed in the presence of H89, indicating that the stimulatory effects of PACAP 1-38 on AgRP and KLF4 expression were due to receptor-mediated stimulation of the G s -cAMP-PKA cascade. Interestingly, treatment of GT1-7 cells with Ken, an inhibitor of KLF4 transcription (Fig. 8d), completely abolished the PACAP 1-38 -induced increase in AgRP expression (Fig. 8c). These results support the concept that PACAP 1-38 induces AgRP transcription in a KLF4-dependent manner.

PAC 1 receptors stimulate AgRP expression and food intake.
Recent work has shown that AgRP neurons, like GT1-7 cells, express PAC 1 receptors 21,41 . The potential role of these receptors in regulating food intake remains unclear. On the basis of our findings obtained with CNO-treated GsD-AgRP mice, we tested the hypothesis that selective activation of PAC 1 receptors endogenously expressed by AgRP neurons might cause a sustained increase in food intake.
To stimulate these receptors in vivo, we took advantage of the recent finding that PACAP-containing glutamatergic neurons in the paraventricular nucleus of the hypothalamus (PVN) are monosynaptically connected to AgRP neurons 21 . To selectively activate this neuronal pathway, we first injected the AAV-hM3Dq virus (AAV-hSyn-DIO-hM3Dq-mCherry) into the PVN of PACAP-ires-Cre mice (bilateral injections; use of 'Flex Switch' technology). This strategy resulted in the selective expression of the hM3Dq designer receptor in PACAP neurons of the PVN 21 . For the sake of brevity, we refer to these mice as hM3Dq-PACAP mice. To activate the hM3Dq-expressing PACAP neurons in vivo, we injected hM3Dq-PACAP mice with a single dose of CNO (1 mg kg À 1 i.p.). Before CNO injections, hM3Dq-PACAP mice received an i.c.v. injection of either saline or PACAP 6-38 (1.2 mg), a potent PAC 1 receptor antagonist. CNO treatment of hM3Dq-PACAP mice pre-injected with saline resulted in a robust, time-dependent increase in acute food take (Fig. 9a). Strikingly, this response was greatly reduced in CNO-injected hM3Dq-PACAP mice that had been pretreated with PACAP 6-38 (Fig. 9a). PACAP  had no significant effect on food intake in WT mice (Supplementary Fig. 8).
We next examined whether CNO treatment of hM3Dq-PACAP mice resulted in changes in the hypothalamic transcript levels of AgRP, KLF4 and NPY. Specifically, hM3Dq-PACAP mice that had been pretreated i.c.v. with either saline or PACAP 6-38 (1.2 mg) were injected with CNO (1 mg kg À 1 i.p.), and hypothalami were isolated and processed for quantitative reverse transcription-PCR studies 2 h later. In the absence of PACAP  , CNO treatment led to significant increases in hypothalamic AgRP and KLF4 expression levels, which were similar in magnitude to those observed with CNO-treated GsD-AgRP mice (Fig. 9b). In contrast, in hM3Dq-PACAP mice that had been pretreated with PACAP 6-38 , CNO-induced AgRP and KLF4 expression (but not NPY expression) was significantly suppressed (Fig. 9b). These results suggest that activation of PAC 1 receptors endogenously expressed by AgRP neurons promotes food intake by stimulating the KLF4-AgRP axis. Finally, we investigated whether CNO treatment of hM3Dq-PACAP mice had long-lasting effects on food intake, as observed with CNO-injected GsD-AgRP mice. We found that a single injection of CNO (1 mg kg À 1 i.p.) stimulated food intake for 3 days, similar to the results obtained with GsD-AgRP mice (Figs 3b and 9c). The long-lasting stimulatory effect of CNO on food intake was accompanied by an increase in body weight (note that this effect was highly significant on day 3 after CNO injection; Fig. 9d). Taken together, these data strongly suggest that activation of PAC 1 receptors endogenously expressed by AgRP neurons mimics the sustained orexigenic effect triggered by GsD signalling in these neurons.
Fasting promotes pCREB expression in AgRP neurons. To explore whether G s -pCREB signalling in AgRP neurons is sensitive to altered metabolic conditions such as fasting when AgRP neurons show increased activity, we studied pCREB expression in AgRP neurons of Npy-hrGFP mice (Npy-GFP mice) 42 . This mouse line expresses GFP selectively in NPY/AgRP neurons 42 . Specifically, we prepared hypothalamic sections from Npy-GFP mice that had either free access to food or had been fasted for 20 h. AgRP neurons were identified by GFP fluorescence (green) while pCREB expression was probed with an anti-pCREB antibody (red fluorescence; Fig. 10 and Supplementary Fig. 9).We found that fasted mice showed a marked increase in the number of AgRP neurons staining positive for pCREB (Fig. 10c). This observation suggests that pCREB expression in AgRP neurons is likely to be of physiological relevance.

Discussion
Food intake studies showed that treatment of GsD-AgRP mice with a single dose of CNO resulted in a long-lasting (for several days) orexigenic effect that was associated with a significant increase in body weight (Fig. 3a-c). In contrast, the stimulatory effect observed with CNO-treated hM3Dq-AgRP mice lasted only for a single day (Fig. 3d-f), most likely reflecting differences in cellular activity caused by the activation of G s -versus G q -dependent signalling pathways (Figs 1c and 2).
AgRP neurons store and release the neuropeptides AgRP and NPY, as well as the biogenic amine, GABA, all of which promote food intake 4 . Using a combined genetic/pharmacological approach, we demonstrated that NPY and GABA release do not contribute to a significant extent to the orexigenic effects mediated by the GsD designer receptor expressed by AgRP neurons (Fig. 5). In contrast, the GsD-mediated increase in food intake was completely abolished by i.c.v. treatment of GsD-AgRP mice with an anti-AgRP antibody (Fig. 4c). These data clearly indicate that the ability of the G s -coupled designer receptor to promote food intake is mediated by the release of AgRP.
In striking contrast, the anti-AgRP antibody had no significant effect on the CNO-induced increase in food intake observed with hM3Dq-AgRP mice (Fig. 4d). In agreement with this observation, Krashes et al. 5 recently reported that the release of either NPY or GABA alone is sufficient to drive acute feeding in CNO-treated hM3Dq-AgRP mice and that AgRP does not contribute to this response.
Given the key role of AgRP in mediating the orexigenic effects displayed by CNO-treated GsD-AgRP mice, we speculated that CNO stimulation of GsD might affect AgRP expression. Consistent with this hypothesis, CNO treatment of GsD-AgRP mice resulted in a robust increase in hypothalamic AgRP transcript levels (Fig. 7a). In contrast, hypothalamic AgRP expression remained unaffected in CNO-injected hM3Dq-AgRP mice (Fig. 7a). It is likely that the increase in GsD-mediated AgRP expression represents a key factor contributing to the long-lasting orexigenic effects observed with CNO-treated GsD-AgRP mice.
Since drugs that enhance the production of cAMP can stimulate KLF4 expression in different cell types 32,33 , we tested the hypothesis that GsD-mediated stimulation of KLF4 expression is required for the GsD-dependent increase in AgRP levels observed with GsD-AgRP mice. We first showed that CNO treatment of GsD-AgRP mice led to a significant increase in hypothalamic KLF4 mRNA levels and a pronounced increase in KLF4 protein levels in AgRP neurons expressing the GsD designer receptor (Fig. 7a,b). Studies with a chemical inhibitor of KLF4 transcription 35 indicated that the GsD-mediated increase in hypothalamic AgRP expression requires KLF4 (Fig. 7c). Importantly, overexpression of KLF4 in AgRP neurons led to a significant increase in food intake (Fig. 7d,e). Taken together, these data strongly support the concept that activation of the G s -linked designer receptor (GsD) promotes food intake by stimulating KLF4-dependent expression of AgRP.
Studies with cultured hypothalamic cells demonstrated that ligand-induced activation of an endogenously expressed G s -linked receptor, the PAC 1 receptor subtype, also promoted KLF4 and AgRP expression, similar to the findings obtained with CNO-treated GsD-AgRP mice. Additional studies indicated that these responses were PKA dependent and that increases in KLF4 expression triggered the upregulation of AgRP expression (Fig. 8). Collectively, these data strongly support the concept that activation of G s -coupled receptors endogenously expressed by AgRP neurons (for example, PAC 1 receptors) stimulates a PKA-dependent signalling cascade that promotes the expression of KLF4 and AgRP in a sequential fashion (Fig. 9).
A recent study 43 reported that overexpression of KLF4 in the ARC of rats abolished the anorectic actions of leptin. Taken together, these findings indicate that KLF4 represents an important central regulator of the activity of hypothalamic AgRP neurons.
In agreement with the in vitro studies, in vivo experiments demonstrated that the activation of G s -linked PAC 1 receptors endogenously expressed by AgRP neurons closely mimicked the stimulatory effects on food intake and body weight observed with CNO-treated GsD-AgRP mice (Figs 3 and 9). In this set of experiments, we used a chemogenetic strategy to selectively activate the PACAP-containing neurons of the PVN that are monosynaptically connected to AgRP neurons 21 . Importantly, the increase in food intake observed after activation of this pathway could be greatly reduced by pretreatment of mice with PACAP 6-38 , a PAC 1 receptor antagonist (Fig. 9a). These data strongly support the concept that stimulation of PAC 1 receptors or other G s -coupled receptors endogenously expressed by AgRP neurons leads to a sustained increase in food intake (Fig. 9e). Interestingly, microarray data have shown that AgRP neurons express additional G s -coupled GPCRs, besides the PAC 1 receptor, including calcitonin and MC 3 melanocortin receptors (Geo dataset GSE45858; also see ref. 44). It is likely that signalling through all of these receptors contributes to G s -mediated AgRP release.
As pointed out at the beginning of the Discussion section, CNO treatment did not depolarize GsD-AgRP neurons in ex vivo whole-cell patch clamp experiments (Fig. 1c,d). However, this finding does not exclude the possibility that CNO-mediated stimulation of GsD affects the activity of downstream signalling molecules and/or ion channels that facilitate the activation of AgRP neurons in vivo where these neurons are exposed to a multitude of hormones and neurotransmitters. Consistent with this concept, activation of the G s -cAMP signalling pathway by G s -coupled receptors has been shown to promote neuronal excitability [45][46][47] . It is likely that the sustained orexigenic effect observed with CNO-treated GsD-AgRP mice is due to cAMP-PKA-mediated changes at the transcriptional level (Figs 7-9) and/or G s -dependent changes in synaptic plasticity 48 , which remain to be explored.
Our findings raise the possibility that drugs that block the activity of G s -coupled receptors endogenously expressed by AgRP Generation of AAV-hSyn-DIO vectors. The generation of the AAV-hSyn-DIO-hM3Dq-mCherry and AAV-hSyn-DIO-mCherry viral vectors has been described previously 19 . To generate the AAV-hSyn-DIO-GsD-eGFP vector, we replaced the hM3Dq-mCherry coding sequence in the pAAV-hSyn-DIO hM3Dq-mCherry plasmid (a kind gift by Dr Bryan Roth, UNC, Chapel Hill, NC) with the HA-GsD-eGFP coding sequence 14 . The AAV-hSyn-DIO-rM3D(Gs)-mCherry virus was provided by the UNC Vector Core (Chapel Hill, NC). This latter virus is identical with the AAV-hSyn-DIO-GsD-eGFP virus except for the different fluorescence tags (mCherry versus eGFP, respectively). To generate the AAV-hSyn-DIO-KLF4 vector, we first fused the human influenza hemagglutinin (HA) tag sequence to the KLF4 coding sequence (Addgene plasmid #15920) by using a PCR strategy. Then, the KLF4-HA coding sequence was cloned into the pAAV-hSyn-DIO vector. All AAVs were packaged in serotype 8 by the UNC Vector Core.
Stereotaxic injections. Male mice that were at least 6 weeks old were anaesthetized with isoflurane, and placed into a stereotaxic apparatus (David Kopf Instruments, model 940A with 923B mouse gas aesthesia head holder). The skull was exposed via a small incision and a small hole was drilled (0.45-mm drill bit) into the skull for the injection of AAVs. A Hamilton 10-ml syringe with a 30-or 33-gauge blunt-end needle was inserted into the brain for virus delivery. Postoperative analgesia was provided (ketoprofen, 5 mg kg À 1 subcutaneous). Bilateral injections (200 nl of AAV) were made in the ARC. The coordinates were (from bregma) as follows: anterior-posterior, À 1.46 mm; lateral (from midline), ± 0.3 mm; dorsal-ventral, À 5.80 mm). In a similar fashion, bilateral injections were made in the PVN. The coordinates were (from bregma) as follows: anterior-posterior, À 0.82 mm; lateral (from midline), ± 0.25 mm; dorsal-ventral, À 4.80 mm (from the surface of the skull). Mice were allowed to recover for 10 days before the start of feeding experiments.
Immunohistochemistry and imaging studies. Mice were perfused with 4% paraformaldehyde in 0.1 M phosphate buffer fixative (pH 7.4). Tissue was post-fixed in this solution overnight and transferred to 20% sucrose in 1 Â PBS. Hypothalamic slices (20-30-mm thick) were incubated overnight at 4°C with primary antibodies diluted in PBS, supplemented with 1% BSA and 0.1% Triton X-100. Slices were then washed three times and incubated with fluorophore-conjugated secondary antibodies for 2 h at room temperature. Slices were rinsed three times in PBS containing 0.1% Triton X-100 and then mounted for imaging using Antifade Reagent (SlowFade Gold with DAPI, S36939; Life Technologies, Grand Island, NY) or Vectashield technology (Vector Labs, Burlingame, CA). Fluorescence images were taken with a Zeiss Imager D1 fluorescent microscope and a Zeiss LSM-700 confocal microscope.
Quantification of c-fos and pCREB expression. To quantitate CNO-induced changes in c-fos and pCREB expression in DREADD-expressing AgRP neurons or changes in pCREB expression in AgRP neurons of fed and fasted Npy-GFP mice 42 , we carried out a series of immunohistochemical studies. CNO (1 mg kg À 1 )injected GsD-AgRP and hM3Dq-AgRP mice were killed 2 h after drug treatment.
For fasting experiments, Npy-GFP mice were killed at 9:00 in either the ad lib fed state or after a 20-h fast (food was removed at 12:00 on the previous day). Hypothalamic sections were prepared and processed for immunohistochemical staining as described above. AgRP neurons/DREADD-expressing AgRP neurons staining positive for pCREB or c-fos were visualized and counted using confocal images taken under Â 20 and Â 40 power of magnification. Fluorescent images were captured using a Zeiss LSM-700 confocal microscope. Cells were counted by an observer that was blinded to mouse treatment or genotype of the mice. Data are expressed as the percentage of all AgRP neurons staining positive for pCREB (experiments with Npy-GFP mice) or as the percentage of DREADD-expressing AgRP neurons staining positive for pCREB or c-fos, respectively. The antibodies used are listed in Supplementary Table 2.
I.c.v. injections. For i.c.v. injections of drugs or antibodies, mice were anaesthetized with 2-3% isoflurane and placed into a stereotaxic apparatus (see 'Stereotaxic injections'). The surface of the skull was exposed with sterile surgical instruments, and a small hole was drilled through the skull and a guide cannula was placed into the lateral ventricle via the stereotaxic manipulator. For i.c.v. injections, the following coordinates were used (from bregma): anterior-posterior, À 0.34 mm; lateral (from midline), þ 1 mm; dorsal-ventral, À 2.3 mm (from the surface of the skull). The guide cannula (Plastic One; C315GS-2/SP, 2.6 mm with 2 mm pedestal; Roanoke, VA) was placed and secured using Loctite 454 Glue. After a 2-week recovery period, a small injection cannula (Plastic One; C315IS-2/SPC, fit 2.6 mm C315GS-2 with 1 mm projection) attached to a catheter and a Hamilton syringe was inserted into the guide cannula. The injection cannula protruded 1 mm beyond the end of the guide cannula into the ventricle. Drugs or antibodies were delivered slowly in a 2-ml volume via the injection cannula. After an additional 7 min, the injection cannula was removed and capped with a dummy cannula (C315FDS-2; Plastic One). The following drugs or antibodies were delivered via this route: Assessment of neuropeptide and KLF4 mRNA levels. AgRP, KLF4 and NPY mRNA levels were measured using RNA prepared from cultured GT1-7 cells or from mouse hypothalamic tissue. One day before the assay, GT1-7 cells were seeded into six-well plates at a density of 1 Â 10 6 cell per dish and cultured in DMEM containing 10% fetal bovine serum and penicillin (100 units per ml)/ streptomycin (100 mg ml À 1 ) in a humidified atmosphere of 5% CO 2 /95% O 2 at 37°C. On the day of the assay, the medium was replaced with fresh DMEM.  table below). PCR cycling conditions were as follows: 50°C for 2 min, 95°C for 10 min; and 40 cycles at 95°C for 15 s and 60°C for 1 min, respectively. The expression of b-actin RNA served as an internal control. The results were expressed as fold change in expression of a particular RNA transcript relative to b-actin RNA expression between untreated and drug-treated cells.
In some experiments, AAV-injected mice (hM3Dq-AgRP, GsD-AgRP or hM3Dq-PACAP mice) were injected i.p. with saline or CNO (1 mg kg À 1 ). In a subset of experiments, i.c.v. injections of kenpaullone or PACAP 6-38 were performed 30-45 min before i.p. saline or CNO injections. Two hours after the administration of CNO (or saline), RNA was extracted from mouse hypothalami using the QIAzol Lysis reagent (Qiagen). cDNA synthesis and qPCR studies were performed as described in the previous paragraph. Experimental data were expressed as fold change in expression of a particular RNA transcript relative to b-actin RNA expression between a specific test group and the corresponding control group (for primer sequences, see Supplementary Table 3).
For whole-cell patch clamp electrophysiological recordings, brain slices containing the ARC were incubated at B30°C in a heated, oxygenated holding chamber containing aCSF (composition in mM): 124 NaCl, 4.4 KCl, 2 CaCl 2 , 1.2 MgSO 4 , 1 NaH 2 PO 4 , 10.0 glucose and 26.0 NaHCO 3 , and then transferred to a submerged recording chamber maintained at B30°C (Warner Instruments, Hamden, CT). Recording electrodes (3)(4)(5) were pulled with a Flaming-Brown Micropipette Puller (Sutter Instruments, Novato, CA) using thin-walled borosilicate glass capillaries. These recording electrodes were filled with a solution of the following composition (in mM): 135 K þ -gluconate, 5 NaCl, 2 MgCl 2 , 10 HEPES, 0.6 EGTA, 4 Na 2 APT and 0.4 Na 2 GPT. To block neural firing, tetrodotoxin (0.5 mM) was added to the perfusing aCSF solutions described above. Signals were acquired via a Multiclamp 700B amplifier (Molecular Devices, Sunnyvale, CA), digitized at 20 kHz, filtered at 3 kHz and analysed using Clampfit 10.4 software (Molecular Devices). Input resistance and access resistance were monitored before and after each experiment. Experiments in which changes in access resistance were 420% were not included in the data analysis. A maximum of two neurons were analysed per mouse. Membrane potential shift was compared between a 4-min bin average during baseline and a 4-min bin average at the end of drug application. NPY-hrGFP mice were used to visualize AgRP neurons. In the case of GsD-AgRP and hM3Dq-AgRP mice, AgRP neurons were identified by mCherry fluorescence (both GsD and hM3Dq were fused to mCherry at their C termini).
Measurement of AgRP release from hypothalamic slices. The release of AgRP from mouse hypothalamic slices was measured ex vivo, in a fashion similar to that described by Enriori et al. 50 The 2-mm-thick hypothalamic slices containing the ARC and PVH were prepared from GsD-AgRP and mCherry-AgRP control mice (AgRP-ires-Cre mice that had been injected bilaterally with the AAV-hSyn-DIO-mCherry virus). The slices were then incubated in aCSF plus 1.7 ml ml À 1 of protease inhibitor cocktail (Sigma) and equilibrated with 95% O 2 and 5% CO 2 at 37°C for 1 h. Hypothalamic slices were then incubated for 1 h in aCSF to determine basal AgRP release. Subsequently, slices were incubated for another 1 h period in the presence of CNO (15 mM). Tissue viability was verified by exposure to 56 mM KCl for 1 h. Supernatants were collected, and AgRP concentrations were determined using an AgRP fluorescence EIA kit (Phoenix Pharmaceuticals) according to the manufacturer's protocol. The amount of AgRP secreted into the medium in the presence of CNO was normalized to basal AgRP release determined during the initial aCSF incubation period.
Generation of KLF4 promoter-luciferase reporter constructs. The pGL3-pKLF4 luciferase reporter construct 51 containing B1.5 kb of the mouse KLF4 promoter was kindly provided by Dr John Christman (The Ohio State University Wexner Medical Center, Columbus, OH). The two half CRE sites (CGTCA) present in the KLF4 promoter ( À 269 to À 265 and À 232 to À 228, respectively) were mutated to CATGG (ref. 52) using the QuikChange Site-Directed Mutagenesis Kit (Agilent). The correctness of all constructs was verified by sequencing. The numbering of KLF4 promoter sequences is based on the NM_010637 NCBI reference sequence.
Luciferase reporter assays. HEK293T cells were transfected with the pGL3-pKLF4 reporter construct using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Where indicated, cells were transfected with the pCMV-FLAG-ACREB construct (kindly provided by Dr Rebecca Berdeaux) or pcDNA3.1 (empty vector) to adjust for the amount of transfected DNA. All cells were co-transfected with the pACLA-RSV-b-Gal plasmid 52 , which served as internal control to normalize the luminescence signal for transfection efficiency. About 24 h after transfection, cells were incubated in 24-well plates for 5 h at 37°C with forskolin (10 mM) or vehicle (DMSO), and luciferase assays were performed as described previously 53 .
Statistical analysis. Data are expressed as means ± s.e.m. for the indicated number of observations. The statistical tests used are indicated in the figure legends.