Agouti-related peptide neural circuits mediate adaptive behaviors in the starved state.

In the face of starvation, animals will engage in high-risk behaviors that would normally be considered maladaptive. Starving rodents, for example, will forage in areas that are more susceptible to predators and will also modulate aggressive behavior within a territory of limited or depleted nutrients. The neural basis of these adaptive behaviors likely involves circuits that link innate feeding, aggression and fear. Hypothalamic agouti-related peptide (AgRP)-expressing neurons are critically important for driving feeding and project axons to brain regions implicated in aggression and fear. Using circuit-mapping techniques in mice, we define a disynaptic network originating from a subset of AgRP neurons that project to the medial nucleus of the amygdala and then to the principal bed nucleus of the stria terminalis, which suppresses territorial aggression and reduces contextual fear. We propose that AgRP neurons serve as a master switch capable of coordinating behavioral decisions relative to internal state and environmental cues.

Ecological studies reveal that prey species display cost-benefit decision making when foraging for food. The costs of foraging include food-seeking energy demands along with environmental threats such as predation and thermal challenges. Many prey species forage within a familiar territory, in zones that are protected from predators and have moderate temperatures [1-3]. However, when challenged with starvation, behavioral priorities adapt and prey species display higher-risk behavior to find food [4-7].

Orexigenic AgRP neurons are active in a starved state [8, 9] and elicit signals that are paramount to the sensation of hunger [10-12]. Coined for the expression of agouti-related peptide (AgRP), AgRP neurons are inhibitory projections neurons; they are GABAergic and express two inhibitory neuropeptides, neuropeptide Y (NPY) and AgRP [8, 13-15]. Somewhat paradoxically, AgRP neurons appear to stimulate hunger by inhibiting downstream brain regions involved in satiety. AgRP neurons are derived from at least two progenitors [16] and project (with minimal collaterals) to approximately 15 unique downstream brain regions [14, 17]. Activation of distinct AgRP projections revealed a “parallel and redundant” signaling network, but interestingly, some AgRP target regions do not evoke a feeding response [17]. We propose that the heterogeneous AgRP population functions to coordinate numerous behavioral and physiological adaptations that prioritize food seeking and energy conservation under conditions of starvation.

AgRP neurons may influence behavioral decisions by signaling to brain regions that are involved in sensory processing. For example, a subset of AgRP neurons project to the medial amygdala (MeA) [14, 18, 19], a brain region implicated in innate social behaviors including aggression [20]. Chemosensory cues of conspecific mice activate cells in the MeA, as indicated by the expression of Fos [21, 22], and acute activation of GABAergic cells in the posterior dorsal MeA can induce attack behavior [23]. Under conditions of starvation, AgRP signaling to the MeA may alter an animal's normal response to chemosensory cues, shifting behavior away from protecting an energy-depleted territory and toward exploratory, food-seeking behavior. To test this idea, we used a combination of viral and genetic tools to activate AgRP neurons, and compared the behavior of these mice to those in the fasted state. We describe a specific starved-state neural circuit that influences innate and learned behavioral responses ().

RESULTS

AgRP circuits promote risk taking and reduce territorialism

Starvation promotes higher-risk foraging behavior such that prey species become willing to search for food in exposed areas outside of their territorial safe zone [4]. To model this behavioral shift in a laboratory setting, we designed an experiment that challenges mice to search for food in a chamber that they are conditioned to associate with a mild foot shock (). We observed that, under normal conditions, mice avoided the shock-associated area, spending only 24.5 ± 2.5% of the trial in this chamber. Fasted animals, however, overcame the conditioned threat and spent more than 40% of the time in the shock-associated side. During habituation and training, food was present below the floor grid in the shock-associated chamber. On test day, the food either remained under the floor grid (food-blocked group), or was presented in the chamber and available for consumption (food-access group). The food-access group spent 46.9 ± 4.6% of the trial in the shock chamber; similarly, the food-blocked group spent 43.2 ± 1.6% of the trial in the shock chamber (, black outlined bars; Supplemental STATISTCS provides details of all tests performed). We questioned whether the fasted state or the food cues biased the animals’ behavior. However, when the experiment was performed in the absence of food entirely, fasted mice behaved similar to fed controls, spending only 25.7 ± 4.7 % of the trial in the shock chamber ().

We reasoned that AgRP neurons may promote high-risk exploration in the starved state. To test this idea, we made AgRP neurons excitable by transducing Agrp mice with a conditional virus containing the stimulatory DREADD (designer receptor exclusively activated by designer drugs), hM3Dq ()[24]. The designer receptor ligand, clozapine-n-oxide (CNO), induces Gαq-mediated signal transduction and can be used to activate AgRP neurons via intraparietal delivery [11]. We activated AgRP neurons in fed mice and asked whether this isolated circuit could recapitulate the foraging behavior of fasted mice in the food-challenge assay described above. Similar to fasted animals, we found that AgRP neuron-stimulated mice spent more time in the shock-associated chamber relative to controls. Both food-available and food-blocked groups spent 40.0 ± 5.6% and 45.6 ± 5.6% of the trial in the shock chamber, respectively; while control animals displayed an aversion to the shock-associated side and spent only 22.7 ± 3.2% of the trial in this chamber (, red bars). These data support that, beyond promoting food intake, AgRP neurons can influence the behavioral response to environmental threats.

In a second test, we evaluated innate anxiety-like behavior by assessing the willingness of animals to enter an exposed platform on an elevated maze. In support of previous literature [25], both fasted and AgRP neuron-stimulated animals spent significantly more time in the exposed platforms compared to controls (). AgRP neuron activation has been demonstrated to promote locomotor activity [11] and, consistent with this, stimulated mice moved an average total distance of 3148 ± 148.3 cm, while controls moved an average of 2024 ± 186.5 cm on the maze. The stimulated mice not only spent more time but also moved a greater distance in the open arms than the controls (stimulated, 48.2 ± 3.1%; controls, 30.7 ± 4.5%).

During starvation, the costs associated with foraging are not limited to environmental threats, but also include the threat of dwindling energy reserves. Consequently, organisms forage in a way that minimizes the energy costs associated with food seeking [1-3, 26]. For example, territorial-defense behavior is not an efficient use of energy if a territory is depleted of resources [27]. Experimentally, we evaluated territorial behavior using the resident-intruder assay and define territory as a defended area. Resident males were sexually experienced and territorialized to an isolated home cage. The intruder test evaluates aggressive territorial behaviors including: holding, fighting (boxing/attacking/mounting), high-speed chasing, and nudging when an intruder (younger, sexually naïve, group-housed littermates that do not display aggression toward residents) is placed in the cage. (). Compared to the fed state, fasted animals displayed less home-cage aggression toward an intruder (, black bars). We observed that fasted residents spent significantly more time investigating the snout of the intruder—perhaps smelling food odorants on the intruder's snout—and displayed escape behaviors including rearing and jumping (). We used a 48 hr fast to maximally activate the feeding circuits; however, 24-hr fasted residents also displayed decreased home-cage aggression ().

To test the role of AgRP neurons in fasting-related territorial behavior, we evaluated activated AgRP neurons in fed resident mice. Similar to fasting, AgRP neuron-stimulated mice displayed less home-cage aggression toward an intruder (, red bars). If food was presented during the trial, AgRP neuron-stimulated residents spent the majority of the 10-min trial eating, consuming 0.34 ± 0.03 g ().

AgRP → MeA signaling influences territorial behavior

Because MeA neurons are involved in innate social behavior, including territorialism, we reasoned that the inhibitory AgRP → MeA circuit may be responsible for starved-state decreases in territorial aggression. To test whether AgRP fibers can directly inhibit cells in the MeA, we transduced AgrpCre mice with a conditional channelrhodopsin-2 (ChR2)-expressing virus ()[28]. We photostimulated AgRP fibers in the MeA and performed whole-cell recordings in slice preparations. MeA soma in close proximity to fluorescent AgRP fibers were patched; 4 of 11 cells from 2 mice displayed a light-evoked inhibitory post-synaptic current (IPSC) that was blocked by the GABAA receptor antagonist picrotoxin (PTX) but not by glutamate receptor antagonists (). We used retrograde tracing to quantify the subset of AgRPMeA-projecting neurons. Fluorescent RetroBeads injected into the MeA () were retained in 7.1 ± 0.6 % of AgRP-expressing cells (167 ± 15 of 2356.3 ± 146 total hemisphere, n = 3; ). These data establish that a subset of AgRP neurons make direct inhibitory connections onto MeA neurons.

We used optogenetic fiber stimulation to probe the behavioral impact of the AgRP → MeA circuit (). Similar to fasting, stimulation of AgRP axons in the MeA elicited less territorial aggression relative to a non-stimulated state (). To gauge the specificity of the AgRP → MeA circuit on territorial behavior, we stimulated AgRP neurons that project to the PVH. Prior studies demonstrated that the AgRP → PVH circuit can induce food intake equivalent to that observed following a fast [29]. If fasting-induced territorial behavior is a consequence of hunger, then this circuit should also modulate home-cage aggression. We did not observe a significant change in aggression upon AgRP → PVH stimulation (). Together, these data indicate that a distinct population AgRP neurons can mediate unique behaviors.

Along with social behavior, evidence suggests that the MeA is also involved in feeding behavior and body-weight regulation [30-32]. We found that AgRP → MeA fiber-stimulated mice ate significantly more than non-stimulated, fiber-attached controls (0.87 ± 0.12 g vs 0.10 ± 0.02 g; ). To gauge the magnitude of this effect, we measured light-evoked food intake from AgRP → PVH fiber stimulation. The hyperphagia induced by AgRP → PVH fiber stimulation was 2.5-fold higher than AgRP → MeA fiber stimulation (). These data add to evidence that there are redundant AgRP circuits that promote feeding [17].

Manipulating cells downstream of AgRP in the MeA

AgRP neurons co-express NPY[8, 14]; therefore, post-synaptic targets of AgRP neurons likely express NPY 1 or 5 receptors (Npy1R or Npy5R) [33] and the melanocortin 4 receptor (MC4R). To evaluate the function of MeA cells that receive information from AgRP neurons, we generated an Npy1r knock-in mouse line ().

We validated the correct targeting of this knock-in using multiple approaches. Prior to injection, neomycin-resistant ES cell colonies were screened for the proper insertion of Cre in the targeted Npy1r allele by Southern blot analysis. We also evaluated transcripts expressed in Npy1r cells by crossing this line to a Cre-dependent RiboTag mouse that expresses an epitope-tagged ribosomal protein (RPL22:HA)[34]. The conditional expression of the hemagglutinin (HA)-tagged ribosomes in Cre-positive cells allows us to isolate mRNA transcripts from these cells. Npy1r-RiboTagged cells in the MeA () were enriched in both Npy1r and Cre transcripts relative to transcripts expressed in all cells within the same region ().

To further profile the Npy1RMeA cells, we mined the RiboTag-isolated transcriptome, probing for genes characteristic of excitatory, inhibitory and glial cells. We found that Npy1R-RiboTagged cells were enriched for both Mc4r and Gad2 transcripts. The glutamate transporter, Slc17a6 (Vglut2) was not enriched and the glial cell marker, Cnp was de-enriched (). These data suggest that Npy1R cells in the MeA may be inhibitory neurons.

Npy1RMeA cells are anatomically distributed throughout the MeA, with a slightly biased distribution in the anteroventral subdivision ( and ), a pattern that resembles Mc4r expression [18, 19]. To investigate whether Npy1RMeA cells are involved in feeding behavior or aggression, we bilaterally transduced the MeA of Npy1r mice with AAV1-DIO-hM3Dq:YFP (). Because inhibitory AgRP → MeA fiber stimulation decreased territorial aggression, we predicted that activation of target Npy1R neurons would have the opposite effect. We found that CNO-induced Npy1RMeA neuron activation significantly increased territorial aggression (). Consistent with previous work using DREADDs to manipulate social behavior [23], we observed a scale of aggressive phenotypes upon CNO-induced activation. Four of nine stimulated animals displayed overt attack behavior, while the rest engaged in other aggressive behaviors including, nudging, aggressive grooming, and holding. To determine the degree of aggression evoked, we evaluated the overtly aggressive males in the presence of an anesthetized intruder; all four mice attacked the anesthetized conspecific within 59.0 ± 5.9 s (), a behavior never observed in non-stimulated mice. Along with changes in aggression, activation of the Npy1RMeA neurons significantly decreased food consumption within the first 4 h of the dark cycle ().

To determine the necessity of the Npy1RMeA population for satiety, territorial aggression, and high-risk exploration, we used a viral approach to chronically inhibit Npy1r-expressing cells in the MeA. Mice were transduced with a conditional virus containing the light-chain of the tetanus toxin (AAV1-DIOGFP:TetTox) which can silence neurons by preventing synaptic transmission [35, 36] (). In the food-challenge assay, TetTox-silenced Npy1RMeA mice spent more time in the shock-associated chamber postconditioning (). They also gained significantly more weight after viral transduction relative to controls (), but there was no statistical difference between TetTox-silenced mice relative to controls in territorial aggression (). Because acute inhibition from AgRP → MeA fibers was sufficient to suppress territorial aggression, this finding was surprising but may be the result of compensatory phenomena. We also evaluated anxiety using the elevated plus maze. There was no difference in open-arm exploration between TetTox-silenced mice relative to YFP controls (17.2 4.2% for TetTox versus 21.2 3.1% for YFP controls, P = 0.5; ).

A disynaptic network: AgRP → Npy1RMeA → pBNST

To identify candidate secondary targets of the AgRP → MeA circuit, we mapped the projection field of Npy1RMeA cells using a virus expressing a Cre-dependent, synapse-specific reporter, synaptophysin:YFP(). We observed dense reporter expression in the posterior principle region of the bed nucleus of the stria terminalis (pBNST), along with several other brain regions including the lateral hypothalamic area, periaqueductal grey, parabrachial nucleus, ventromedial hypothalamus and anterior olfactory bulb (). To determine whether these were secondary targets of AgRP neurons, we injected a Cre-dependent and trans-synaptic anterograde tracing virus, H129Δ-fs-TK-TT [37], into the ARH of Agrp animals (). TdTomato fluorescence was observed in many sites throughout the brain, notably the MeA and the pBNST ().

The pBNST is an established target of the MeA [38, 39] and while AgRP fibers have been found in the anterior BNST [14, 17], they have not been observed in the pBNST. Likewise, we observed few, if any, fluorescent cell bodies in the ARH following injection of fluorescent RetroBeads into the pBNST (). To test the idea that the pBNST is a secondary target of AgRP neurons via the MeA, we co-injected RetroBeads into the pBNST and H129Δ-fs-TK-TT into the ARH of Agrp mice (). We observed expression of both reporters in the MeA (), consistent with the idea that the pBNST is a secondary target of the AgRP → MeA circuit (). Due to the nature of H129 infection, the cells were not healthy enough to quantify the overlap of these reporters in the MeA. Instead, we used an alternative approach to characterize the cells in the MeA that receive input from AgRP neurons and project to the pBNST.

Numerous cell types in the posterior MeA have been defined based on electrophysiological and morphological properties [40-42]. To investigate the properties of pBNST-projecting MeA neurons we injected RetroBeads into the pBNST and performed whole-cell recordings on bead-labeled cells in the MeA (). When subjected to current-step injections, we observed a prominent hyperpolarization-activated voltage sag (h-current, denoted by the arrow, in 10 of 13 cells recorded), which has been described in type 1 GABAergic projections neurons in the posterior MeA [42]. In addition, we discovered that that a subset of these neurons expressed a T-type calcium current (denoted by the arrow head, in 7 of 13 cells; ).

We determined that AgRP neurons can evoke GABA-mediated IPSCs in MeA neurons. Using the RetroBead labeling described here, we sought to determine whether ChR2-expressing AgRP neurons could evoke light-induced responses in pBNST-projecting MeA cells (). Bead-positive MeA soma in proximity to YFP fibers demonstrated light-evoked IPSPs in 3 of 11 recorded cells (). Similar to the recordings in , the light-evoked inhibitory response occurred with a short latency to the photostimulation, suggesting a direct connection. To support this idea, we identified a shifted (smaller and longer latency) light-evoked IPSP in the presence of the action-potential blocker, TTX along with a potassium channel blocker, 4-aminopyridine (4-AP)—a property indicative of monosynaptic connections in ChR2-assisted circuit mapping () [43].

The light-responsive, bead-positive MeA cells also respond to NPY. Bath application of NPY in the presence of TTX resulted in an outward current when held at a membrane potential of −60 mV in 4 out of 4 cells tested (). To investigate the NPY-induced current, voltage ramps were performed in the presence and absence of NPY (). Consistent with the idea that NPY-induced current is mediated by G protein-coupled, inwardly-rectifying potassium channel activation [44, 45], we found that the reversal potential for the outward current was at -85 mV, close to the Nernst equilibrium potential for potassium.

Based on the firing properties recorded in bead-positive cells () along with enrichment of Gad2 in Npy1R-RiboTagged cells (), we predicted that these neurons were GABAergic[42]. Following recording, we harvested the cytosol of the cells [46] and performed single-cell RT-PCR to test for Slc32a1 (Vgat) expression. The majority of cells were Vgat-positive—out of 11 successfully harvested Actb-positive cells, 7 were positive for Vgat expression, while the remaining 4 did not amplify either Vgat or Vglut templates (). These data are consistent with the idea that a subset of AgRP neurons synapse on a population of inhibitory, NPY-responsive cells in the MeA that project to the pBNST.

Npy1RMeA → pBNST signaling influences territorial behavior

The Npy1RMeA population projects to numerous efferent targets, some of which have been implicated in aggressive behavior, including the VMH and PAG [22, 47]. Because AgRP neurons make a disynaptic connection to the pBNST via the MeA, we hypothesized that this circuit may be involved in aggressive/territorial behaviors. We virally transduced Npy1r mice unilaterally with ChR2:YFP virus and placed a fiber optic cannula above either the ipsilateral VMH or pBNST (). Optogenetic stimulation of Nyp1RMeA fibers in the pBNST evoked significantly more territorial defensive behavior (). However, rather than overt attack behavior, the pBNST-fiber stimulation increased nudging activity; the resident mouse followed the intruder for the majority of the assay, constantly nudging the intruder into the wall of the cage (). As opposed to violent aggression, a nudging threat display may be adequate for territorial defense from most competitors (). We activated Npy1RMeA soma using metabotropic hM3Dq DREADD receptors while we stimulated the Npy1RMeA fibers in the pBNST using ionotropic photostimulation; hence, we cannot exclude the possibility that this difference accounts for the behavioral difference.

Optogenetic stimulation of Npy1RMeA fibers in the VMH did not result in significant differences in territorial aggression (). We also measured the effect of photostimulating Npy1RMeA axons in the pBNST or VMH on food intake, but did not observe a significant effect of stimulating either of these projections (). Because Npy1R neurons project to numerous downstream targets, it is likely that the overt/violent aggression observed following Npy1RMeA neuron stimulation (), and also feeding behavior () are orchestrated by projections to targets other than the pBNST or VMH. The modulatory effect of AgRPMeA fiber stimulation on territorial behavior likely involves Npy1R neurons that project to the pBNST, a circuit that can be modulated under physiological conditions of negative energy balance.

DISCUSSION

Risk assessment and territorialism require sensory processing of environmental cues. Rodents select a territorial domain for nesting and foraging with respect to the risk of predation and will defend the limited resources of this area from conspecific intruders. However, under conditions of starvation, mice forage in more exposed or threatening areas and are less willing to defend a territory that is depleted of resources, exemplifying a behavioral adaptation that is associated with an internal state change [48].

Hungry animals will aggressively defend limited food resources from competitors. However, if food is depleted, starving mice try to escape from their territorialized home cage (), display less aggression toward an intruding conspecific, are less anxious, and engage in risky exploration to seek food. This shift in behavior is accompanied by a coincident change in AgRP neuron activity. In the absence of food, AgRP neurons are activated by interoceptive cues of negative energy balance, but when food or food-related cues are present, AgRP neuron activity is rapidly silenced [49, 50]. We demonstrate that a subset of AgRP neurons can evoke GABA-mediated inhibition of the MeA and argue that this circuit is responsible for modulating aggressive territorial behavior when food is limited. If food is discovered during foraging, this cue should rapidly relieve GABA-mediated AgRP inhibition of the MeA, providing a switch to adjust behavior for food acquisition. This behavioral switch is difficult to model in the confines of a small arena and isolated housing conditions. When food was presented to hungry residents (artificially induced by AgRP neuron-stimulation) during an intruder trial, they choose to eat rather than interact with the intruder (). Fluctuations in territorial aggression levels with respect to a limited or depleted food source could be evaluated with the development of techniques to study and track individual animals in a large, group-housed arena.

We propose that the AgRP → MeA circuit is involved in territorial adaptations during starvation, but questioned whether hunger itself could influence territorialism. In this experiment we targeted the AgRP → PVH circuit which has been demonstrated to evoke robust feeding behavior, equivalent to that following a fast. Unlike activating AgRP → MeA fibers, stimulation of the AgRP → PVH circuit did not reduce territorial aggression (). These data support the idea that hunger itself does not change territorial behavior. We define an MeANpy1R → pBNST circuit that is downstream of AgRP and can alter territorial behavior. There are, however, many other AgRP targets throughout the brain, and similar to feeding, they may play redundant roles in territorial adaptations.

The behavioral adaptations that occur during starvation facilitate food acquisition and minimize unnecessary energy expenditure, a complex state-change likely attributed to the broad projection profile AgRP neurons throughout the brain. Not all anatomically distinct AgRP subsets contribute equally to food-intake behavior. Optogenetic stimulation of AgRP fibers in the PVH evokes maximal food consumption, while stimulation of AgRP fibers in the paraventricular thalamus and parabrachial nucleus evoke lesser, if any, influence on food consumption [17, 29]. Similarly, we find that AgRP fibers in the MeA can evoke food intake, but the magnitude of this effect is much less compared to equivalent fiber stimulation in the PVH. The long-term consequence of inhibiting cells in the MeA is a significant increase in body weight (our data and ref [31]). One idea to resolve the differing degrees of food consumption observed by activating distinct AgRP neuron target regions is that their contributions are additive; however, because PVH stimulation is equivalent to stimulating AgRP cell bodies, that explanation is unlikely. Instead, brain regions where AgRP-axon stimulation promotes less food consumption may help coordinate non-feeding behaviors with hunger. For example, AgRP-mediated inhibition of the MeA induces feeding and suppresses territorialism, providing a circuit that can function independently to coordinate two behaviors. It is also possible that under some conditions, select populations of AgRP neurons become activated. The potential for AgRP neurons to orchestrate a complex behavioral response is broad. Future studies detailing the behavioral and physiological contribution of other AgRP targets will provide a complete profile of the starved-state behavioral response.

Online Methods

Animals

Adult male mice (2-6 months old) were bred onto a C57BL/6 background and housed on a 12-hr light cycle (5:00 – 17:00). All experiments were approved by The Animal Care and Use Committee at the University of Washington, and in accordance with NIH guidelines. Agrp and RiboTag (Rpl22) knock-in mice have been characterized [34, 51]. The Cre-dependent reporter strain, Rosa26Sor , was acquired from Jackson Laboratories (stock# 007914).

Npy1r knock-in mice were generated by gene targeting in ES cells. A Cre:GFP fusion gene was inserted just 5’ of the normal initiation codon for Npy1r. The targeting construct had 3.5 kb 5’ and 3’ arms that were prepared by PCR (Npy1r gene) and inserted into a targeting vector with SvNeo for positive selection and HSV-TK and PGK-DTa for negative selection (Npy1r targeting construct, ). After targeting, Cre:GFP expression is under control of Npy1r regulatory elements. The Sv40-Neo selectable gene can be removed by action of FLP recombinase.

Stereotaxic injections and tissue preparation

Stereotaxic surgery and injection coordinates

Mice were anesthetized and positioned on a stereotaxic alignment device (David Kopf instruments). During the procedure, body temperature was maintained with a heating pad and a nose cone delivered isoflurane (1.5 – 2%). Either a Hamilton syringe (88000) or pulled glass capillary was used to inject the target brain regions; ARH / bregma −1.25, lateral ±0.25, ventral −5.8; MeA (viral injections) / bregma −1.0, lateral ±2.25, ventral −5.55 and pBNST / bregma 0.4, lateral ±0.9, ventral −4.25. Cannulas were implanted 0.5 mm above the position of the injection with the exception of the MeA in which the cannula was implanted caudal to the injection at bregma −1.5.

Fiber placement

MeA and PVH tracks were evaluated at bregma −1.5 mm and −0.8 mm, respectively. Terminal placement is represented in . It is likely that some of the MeA-targeted fibers terminated in the lateral ventricle, as we did not find terminals for every implant.

Viruses and neuronal tracers

AAV serotype 1 viruses (AAV1-Ef1α-DIO-hM3Dq:YFP, AAV1-Ef1α-DIO-hM3Dq-mCh, AAV1-CAG-DIO-GFP:TeTx, AAV1-Ef1α-DIO-ChR2:YFP, AAV1-Ef1α-DIO-Synaptophysin:YFP, AAV1-Ef1α-DIO-YFP) were generated at the University of Washington as described [52]. We did not observe side effects in animals injected with any of the AAV1 viruses. AAV1 virus (500 nL) was injected at a titer of ~109 viral particles per μl. Behavioral tests began following a minimum 2-wk, post-operative incubation. For viral tracing using Syn:GFP, the tissue was harvested after a minimum of 5 days after viral injection. H129Δfs-TkTT was generously provided by David Anderson and Liching Lo [37]. H129 virus (300 nL) was diluted and injected at ~106 viral particles per μl. Following a 5-day incubation, H129-injected animals were euthanized and the tissue collected for analysis. To label afferents, we injected 300 nL of green RetroBeads (Lumafluor, RetroBeads™ IX) into the desired target location.

Histology

Mice were anesthetized and transcardially perfused with saline, followed by 4% (w/v) paraformaldehyde in 0.1 M phosphate buffer (PB). Brains were post-fixed (6 hr), washed, cryoprotected in 0.1 M PB with 30% sucrose (w/v), embedded in OCT and frozen at −80 C. For immunohistochemistry, 30-μm floating sections were stained for YFP or tdTomato with the primary antibodies: rabbit anti-GFP (Life Technology A11122; diluted 1:2000) or rabbit anti-DsRed (Clonetech 632496; diluted 1:1000). Antibodies were diluted in 0.1M PB with 0.1% Triton and 2% donkey serum and developed overnight at 4 C.

CNO administration for Fos analysis

CNO was injected at a dose of 1.0 mg/kg body weight 2 hr prior to euthanizing the animals. The tissue was stained with goat anti-Fos (Santa Cruz 48869, diluted 1:300).

Electrophysiology

Figure 2b

Whole-cell recordings were made using an Axopatch 700B amplifier (Molecular Devices) with filtering at 1 KHz using 4-6 MΩ electrodes. Coronal brain slices (250 μm) were prepared in an ice slush solution containing (in mM): 250 sucrose, 3 KCl, 2 MgSO4, 1.2 NaH2PO4, 10 D-glucose, 25 NaHCO3, 0.1 CaCl2. Slices recovered for 1 h at 34°C in artificial cerebral spinal fluid (ACSF) continually bubbled with O2/CO2 and containing (in mM): 126 NaCl, 2.5 KCl, 1.2 NaH2PO4, 1.2 MgCl2 11 D-glucose, 18 NaHCO3, 2.4 CaCl2. Patch electrodes were filled with an internal solution containing (in mM): 130 CsCl, 2 MgCl2, 0.5 EGTA, 10 HEPES, 0.25 Na-GTP, and 2.5 Mg-ATP, pH 7.2-7.4, 280 mOsm. ACSF at 32°C was continually perfused over slices at a rate of ~2 ml/min during recording. Drugs were purchased from Abcam and were applied to the bath where indicated at the following concentrations: CNQX: 10 μM; APV: 100 μM; picrotoxin: 100 μM. For light-evoked responses the fiber optic cable was lowered into the bath and 10-ms light pulses (10 mW) were delivered at a rate of 0.1 Hz while cells were held in voltage-clamp mode at −70 mV. Example traces are averages of 15 sweeps.

Figure 5

Coronal brain slices were prepared as described [53]. Whole-cell patch recordings were performed in voltage clamp and current clamp using an Olympus BX51W1 upright microscope equipped with video-enhanced, infrared-differential interference contrast (IR-DIC) and an Exfo X-Cite 120 Series fluorescence light source. Electrodes were fabricated from borosilicate glass (1.5 mm outer diameter; World Precision Instruments, Sarasota, FL) and filled with a normal internal solution (in mM): 128 potassium gluconate, 10 NaCl, 1 MgCl2, 11 EGTA, 10 HEPES, 2 ATP, and 0.25 GTP (pH was adjusted to 7.3-7.4 with 1N KOH, 290-300 mOsm); for measurement of IPSCs, patch pipettes were filled with a high chloride solution (in mM): 140 mM KCl, 10 mM Hepes, 0.1 mM EGTA, 5 mM MgCl2, 0.3 mM Na-GTP, and 5 mM K2-ATP, pH was adjusted to 7.3-7.4 with 1N KOH, 290-300 mOsm. Pipette resistances ranged from 3–5 MΩ. In whole cell configuration, access resistance was less than 20 MΩ; access resistance was 80% compensated. For optogenetic stimulation, a light-induced response was evoked using a light-emitting diode (LED) 470 blue light source controlled by a variable 2A driver (ThorLabs, Newton, NJ) with the light path directly delivered through an Olympus 40× water-immersion lens. Example traces are averages of 10 sweeps. Electrophysiological signals were digitized with Digidata 1322A (Molecular Devices, Foster City, CA), and the data were analyzed using p-Clamp software (version 9.2, Molecular Devices). The liquid junction potential was corrected for all data analysis.

Behavior

To minimize the hierarchical dominance behavior observed in adult group-housed male mice, our behavior studies were performed on singly-housed mice. The behavior results were scored blindly.

AgRP neuron-stimulated cohort: prescreening

Numerous studies using viral targeting as a means of activating AgRP neurons have found that the efficiency (number of cells transduced) correlates with the magnitude of the feeding response when activated [11, 12]. Based on this, we used a prescreening criteria to select animals that consumed at least 1 g of food 4 hr after administration of CNO (1mg/kg).

Food-challenge assay

Rooms of a two-chamber arena were distinguished by visual (vertical striped or wood grain wall paper), olfactory (Nestlet below the floor containing 3 drops of either almond or orange extract) and tactile (wire mesh versus metal bar flooring) cues (). Animals were exposed to this context on 3 consecutive days during the light cycle (5:00 – 17:00). On habituation Day 1, mice were placed in the almond side of the cage and allowed to explore both chambers of the arena for a 30-min session. Mice were excluded from the test if they displayed ≥70% preference for either room. On training Day 2, mice were again placed in the almond side; upon movement into the orange chamber, the door dividing the chambers was closed and a shock paradigm (2 s, 0.3-mA shock, every 3 min over the course of 30 min) was initiated. Immediately following the last shock interval, mice were removed and placed in their home cage. Following a 2 hr inter-trial interval, mice were returned to the almond chamber, the dividing door was shut and the mice did not receive shocks during this 30-min session. Food was present below the floor grid, on the orange side on Days 1 and 2. On testing Day 3, mice were placed on the almond side with the chamber door open and their movement was tracked for a 30-min session. Depending on the experimental conditions, food was either available above the floor grid (food-available cohort) or was again placed below the floor grid (food-blocked cohort) in the orange/shock chamber. Food-available mice were excluded from the test if they moved the food pellet into the almond side during the test session. Video recordings on Days 1 and 3 were evaluated manually in a blinded manner. Group 1, fed versus fasted: all animals were food deprived between Day 1 and Day 2 training. Food was returned to “fed” animals following training, while “fasted” animals were continuously deprived. Group 2, AgRP stimulated (hM3Dq + CNO) versus AgRP controls (YFP + CNO): on test day, food was removed from the home-cage hopper 1 h prior to testing and animals were injected with CNO (1 mg/kg body weight). Group 3, Npy1R silenced (TetTox) versus Npy1R controls (YFP): food was provided ad libitum throughout the trial. Note, we did not repeat this test on individual subjects, each animal was evaluated once.

Elevated Plus Maze

Movement within the arms of a plus maze was recorded and scored for time spent in the open arms. The trial was conducted during the light cycle (5:00 – 17:00). The open-arm score is the cumulative time the animal spent in either open arm during the 10-min session. The center of the maze was excluded from scoring. The animal's movement was video recorded and scored using Ethovision XT (Noldus) video-tracking software. Mice were excluded from the study if they failed to stay on the maze for the duration of the trial. Group 1, fed versus fasted: fasted animals were food deprived for 48 h prior to testing. A separate cohort of age-matched mice, given ad libitum access to food was used as controls. Group 2, AgRP-stimulated (hM3Dq + CNO) versus AgRP controls (YFP + CNO): on test day, food was removed from the home-cage hopper 1 h prior to testing and animals were injected with CNO (1mg/kg). Group 3, Npy1R silenced (TetTox) versus Npy1R controls (YFP): food was provided ad libitum throughout the trial. We found that a second exposure to this maze will bias the animals’ movement. Thus, we evaluated only one trial for each animal.

Resident-intruder assay

Resident males were sexually experienced and territorialized to an isolated home cage for a period of two weeks prior to testing. Intruder males were group-housed littermates between 7 to 10 weeks of age and weighed less than resident animals (). The test consisted of exposing a resident animal to an unfamiliar intruder for a 10-min session. In order to expose the cage arena for video recording, the wire rack (containing food and water) was removed from the cage 1 h prior to testing. CNO versus saline control trials: CNO (Tocris ®, diluted in saline and administered intraperitoneally at 1 mg/kg body weight); saline (administered intraperitoneally at 10 uL/g body weight). Injections were performed 30 min prior to testing. Optogenetic trials: light stimulation (pulse paradigm ) versus tethered controls (attached to dummy patch cords with no light stimulation). Stimulated animals were exposed to light 30 min prior to, and during testing. Anesthetized intruders were injected with ketamine (0.75 mg/kg; Ketaset® Pfizer). All trials were conducted within the first 2 h of the dark cycle (17:00 – 19:00) and video recorded with infrared lighting. The videos were scored in a blinded manner using the manual settings of Ethovision XI. Aggressive behavior included: holding, fighting (boxing/attacking/mounting), high-speed chasing, and nudging (). We did not observe aggressive behavior initiated by intruders. In some cases aggression was tested twice on the same animal (paired tests). Because aggression levels may change with repeated testing, we designed a randomized crossover study with respect to the experimental and control conditions. The tests were separated by a week and unique intruders were used in the second trail.

Optogenetics

Fiber-optic cannulas were assembled as described [54]. A 0.22 NA, 220-μm coated multimode fiber (Thorlabs, FG200LEA) was fit into a 2.5-mm ferrule (custom bore hole of 230 μm; Precision Fiber). The fiber lengths for targeted sites were: MeA, 5.0; pBNST, 3.4; VMH, 5.0 mm. Custom duel-fiber-optic cannulas (fibers 1 mm apart and 4.5 mm long, Doric Lenses) and patch cords were used for PVH fiber stimulation. Before implantation, each cannula was evaluated for power loss at the fiber ending. Fiber-optic cannulas were permanently fixed to the skull using a layer of C&B Metabond (Parkell), followed by a top coat of Jet Acrylic cement (Lang Dental). Cannulas were connected to a blue (473 nm) laser (LaserGlow, LSR-0473) via a patch cord (0.22 NA, Doric Lenses), and set to deliver light at 10 mW from the end of the implanted fiber. Pulses (5 ms) were delivered at 10 Hz for 5 s, followed by a 2 s no-light recovery. The laser waveform was generated by Master-8 pulse stimulator (AMPI).

RiboTag

The MeA was isolated from a 2-mm-thick coronal section. Tissue punches from 6 male animals (Npy1r) were pooled and homogenized in buffer, as described previously [34]. An antibody against HA (mouse-α-HA; Covance MMS-101R) was used to precipitate the epitope-tagged ribosomes and associated transcripts while the remaining lysate was used as the input sample as described [34]. Transcript expression levels were quantified by qRT-PCR.

PCR analysis

RiboTag qRT-PCR analysis

Transcript expression levels were quantified using either SYBR Green or TaqMan PCR assays from Agilent Technologies as described [51]. The following primer sets were used for SYBR Green primer-mediated amplification: (Npy1r) 5′-TGATCTCCACCTGCGTCAAC/ 5′-ATGGCTATGGTCTCGTAGTCAT; (Slc17a6) 5’- AGAGAGCGCAAATCTGCTAGGT/ 5’-GCGTAGACGGGCATGGAT; (Cre) 5’-CTGCCACCAGCCAGCTAT/ 5’- GGGCACTGTGTCCAGACC; (Actb) 5’-AGTGTGACGTTGACATCCGTA/ 5’- GCCAGAGCAGTAATCTCCTTCT. The following probes were used for TaqMan amplification: Mc4r, ID Mm00457483_s1; Gad2, ID Mm00484623_m1; Cnp, ID Mm01306641_m1

Post-recording single-cell PCR

Following whole-cell recordings, the content of the patched cell was harvested by applying negative pressure to gently aspirate the cell content into the tip of the recording pipette. The cell content was expelled into a 500 μl harvesting tube containing 5 μl of RT solution and stored at −80 C until further processing. RNA transcripts in the harvested lysate were reverse transcribed and PCR was performed as previously described [46] using the following primer sets: Vglut2 (Slc17a6) 5’-ATCTGCTAGGTGCAATGG/ 5’- TAAGCTGGCTGACTGATG; (Slc32a1) 5’- GTCACGACAAACCCAAGATCAC/ 5’- GGCGAAGATGATGAGGAACAAC; (Actb) 5’- AAGGCCAACCGTGAAAAGAT; 5’-GTGGTACGACCAGAGGCATAC

STATISTICS

Statistical analysis was performed with GraphPad's Prism software. A P value of < 0.05 was the threshold for significance (*P < 0.05, **P < 0.01, ***P < 0.001), and all error bars represent standard errors of the mean. Unless otherwise indicated, all n values represent individual mice. No statistical methods were used to pre-determine sample sizes because many of the outcomes were unknown. When possible, sample sizes were based on previous studies [11, 17, 23], but all data was gathered from at least three independent experiments. The data distribution was assumed to be normal, but this was not formally tested. Please note, a supplementary methods checklist is also available through Nature Neuroscience.

Sample size: ad lib, n = 15; fasted food-available, n = 7; fasted food-blocked, n = 7; YFP + CNO, n = 13; hM3Dq + CNO food-available, n = 6; hM3Dq + CNO food-blocked, n = 7. 1-Way ANOVA: F (5, 49) = 10.35, ***P < 0.0001. Bonferroni's multiple comparison tests: ad lib versus fasted food available, 95% CI (−10.67 to −2.78) ***P; ad lib versus fasted food blocked, 95% CI (−9.58 to −1.69) **P; YFP + CNO versus hM3Dq + CNO food available, 95% CI (−9.44 to −9.32) **P; YFP + CNO versus hM3Dq + CNO food blocked, 95% CI (-10.93 to -2.85) ***P.

Sample size: ad lib, n = 8; fasted, n = 7; YFP + CNO, n = 12; hM3Dq + CNO, n = 8. 1-Way ANOVA: F (3, 31) = 8.58, ***P = 0.0003. Bonferroni's multiple comparison tests: ad lib versus fasted, 95% CI (−3.58 to −0.05) *P; YFP + CNO versus hM3Dq + CNO, 95% CI (−3.79 to −0.68) **P < 0.01

Sample size: ad lib vs fasted, n = 12; hM3Dq + saline vs hM3Dq + CNO, n = 10. Paired two-tailed Student's t-test: ad lib (M = 52.42, SD = 21.10) versus fasted (M = 33.58, SD = 13.60): t (11) = 2.87, *P = 0.015; hM3Dq + saline (M = 57.61, SD = 28.16) versus hM3Dq + CNO (M = 11.73, SD = 9.76): t (9) = 5.16, ***P = 0.0006

Sample size: ChR2 off vs ChR2 stimMeA, n = 8; ChR2 off vs ChR2 stimPVH, n = 6. Paired two-tailed Student's t-test: ChR2 off (M = 58.62, SD = 17.97) versus ChR2 stimMeA (M = 27.93, SD = 11.27): t (7) = 4.67, **P = 0.0023; ChR2 off (M = 58.73, SD = 12.54) versus ChR2 stimPVH (M = 76.37, SD = 22.82): t (5) = 2.02, not significant P = 0.0999

Sample size: ChR2 off n = 18; ChR2 stimMeA, n = 8; ChR2 stimPVH, n = 10. 1-Way ANOVA: F (2, 33) = 58.68, ***P < 0.0001. Bonferroni's multiple comparison tests: ChR2 off versus ChR2 stimMeA, 95% CI (−1.29 to −0.25) **P; ChR2 off versus ChR2 stimPVH, 95% CI (−2.56 to −1.59) ***P.

Sample size: YFP + CNO, n = 8 vs hM3Dq + CNO, n = 10. Unpaired two-tailed Student's t-test: YFP + CNO (M = 57.88, SD = 16.42) versus hM3Dq +CNO (M = 140.0, SD = 54.23): t (16) = 4.11, ***P = 0.0008

Sample size: YFP + CNO vs n = 8; hM3Dq + CNO, n = 8. Unpaired two-tailed Student's t-test: YFP + CNO (M = 1.25, SD = 0.19) versus hM3Dq + CNO (M = 0.49, SD = 0.18): t (14) = 8.15, ***P < 0.0001

Sample size: YFP, n = 6 vs TetTox, n = 6. Unpaired two-tailed Student's t-test: YFP (M = 7.16, SD = 2.26) versus TetTox (M = 12.71, SD = 2.31): t (10) = 4.21, **P = 0.0018

Sample size: TetTox, n = 6; YFP, n = 6. 2-Way repeated measures ANOVA. Bonferroni's multiple comparisons: week 8, **P < 0.01; week 9 *P < 0.05; week 10 **P < 0.01; week 11 ***P < 0.001

Sample sizeYFP, n = 6 vs TetTox, n = 6. Unpaired two-tailed Student's t-test: YFP (M = 46.83, SD = 28.32) versus TetTox (M = 31.05, SD = 20.66): t (10) = 1.10, not significant P = 0.296

Sample size: ChR2-off vs ChR2 stimpBNST, n = 5; ChR2-off vs ChR2 stimVMH n = 3. Paired two-tailed Student's t-test: ChR2-off (M = 45.40, SD = 20.45) versus ChR2 stimpBNST (M = 211.0, SD = 82.79): t (4) = 5.86, **P = 0.004; ChR2-off (M = 38.33, SD = 8.02) versus ChR2 stimVMH not significant (M = 86.67, SD = 16.17): t (2) = 3.46, not significant P = 0.074

Sample size: ChR2-off vs ChR2 stimpBNST, n = 5; ChR2-off vs ChR2 stimVMH n = 3. Paired two-tailed Student's t-test: ChR2-off (M = 1.34, SD = 0.12) versus ChR2 stimpBNST (M = 1.47, SD = 0.04): t (2) = 1.60, not significant P = 0.251; ChR2-off (M = 1.00, SD = 0.58) versus ChR2 stimVMH (M = 1.16, SD = 0.56): t (2) = 0.73, not significant P = 0.540

Sample size: ad lib (n = 6) vs 48-hr fasted (n = 6). Unpaired two-tailed Student's t-test: ad lib (M = 7.65, SD = 3.09) versus 48-hr fasted (M = 7.71, SD = 3.45): t (10) = 0.03, not significant P = 0.976

Sample size: intruders (n = 23); residents, ad lib (n = 12); residents, fasted (n = 12). 1-Way ANOVA: F (2, 44) = 46.64, ***P < 0.0001. Bonferroni's multiple comparison tests: intruder versus resident, ad lib, 95% CI (−9.17 to −5.61) ***P, intruder versus resident, fasted, 95% CI (−4.85 to −1.29) ***P.

Sample size: ad lib vs 24-hr fasted (n = 6). Paired two-tailed Student's t-test: ad lib (M = 59.27, SD = 27.92) versus 24-hr fasted (M = 27.62, SD = 18.09): t (5) = 3.41, *P = 0.019

Sample size: ad lib vs 48-hr fasted (n = 9). Paired two-tailed Student's t-test: ad lib (M = 17.54, SD = 4.15) versus fasted (M = 57.82, SD = 13.53): t (8) = 8.90, *** P = 0.0001.

Sample size: ad lib vs 48-hr fasted (n = 9). Paired two-tailed Student's t-test: ad lib (M = 7.29, SD = 4.10) versus fasted, (M = 3.75, SD = 22.17): t (8) = 3.63, **P = 0.007

Sample size: hM3Dq + saline or hM3Dq + CNO (no food), n = 10; hM3Dq + CNO (with food), n = 7. 1-Way ANOVA: F (2,24) = 20.90, ***P < 0.0001. Bonferroni's multiple comparison tests: saline versus CNO (no food), 95% CI (25.91 to 65.85) ***P; saline versus CNO (with food), 95% CI (28.89 to 72.90) ***P.

Sample size:YFP vs n = 4; TetTox, n = 4. Unpaired two-tailed Student's t-test: YFP (M = 2.11, SD = 0.62) versus TetTox (M = 1.72, SD = 0.84): t (6) = 0.75, P = 0.479.