Nutrient sensing in the nucleus of the solitary tract mediates non-aversive suppression of feeding via inhibition of AgRP neurons

The nucleus of the solitary tract (NTS) is emerging as a major site of action for the appetite-suppressive effects of leading pharmacotherapies currently investigated to treat obesity. However, our understanding of how NTS neurons regulate appetite remains incomplete. Objectives In this study, we used NTS nutrient sensing as an entry point to characterize stimulus-defined neuronal ensembles engaged by the NTS to produce physiological satiety. Methods We combined histological analysis, neuroanatomical assessment using inducible viral tracing tools, and functional tests to characterize hindbrain-forebrain circuits engaged by NTS leucine sensing to suppress hunger. Results We found that NTS detection of leucine engages NTS prolactin-releasing peptide (PrRP) neurons to inhibit AgRP neurons via a population of leptin receptor-expressing neurons in the dorsomedial hypothalamus. This circuit is necessary for the anorectic response to NTS leucine, the appetite-suppressive effect of high-protein diets, and the long-term control of energy balance. Conclusions These results extend the integrative capability of AgRP neurons to include brainstem nutrient sensing inputs.


INTRODUCTION
The nucleus of the solitary tract (NTS) is established as a major brain site for the sensing and integration of signals relevant to the control of feeding behavior. It is a neuroanatomical hub for ascending vagal afferents activated by ingested foods, corticolimbic-descending inputs encoding homeostatic, cognitive, and motivational controls of feeding, and blood-borne signals diffusing from the adjacent area postrema (AP) that lacks a blood-brain barrier [1,2]. Molecularly, it is enriched in specialized interoceptive neuronal populations equipped to monitor circulating levels of nutrients, gut hormones, and adiposity signals [3]. NTS processing of these diverse inputs is classically described as the main mediator of the short-term negative feedback control of ingestion (or satiation) via recruitment of medullary motor output circuits [1]. The NTS also relays processed information to the lateral parabrachial nucleus (lPBN), established as a common target for NTS efferents in the central representation of aversive and avoidance feeding-related cues [4,5]. In both cases, the NTS outputs interrupt food ingestion and until recently had not been implicated in the regulation of hunger, the longterm control of satiety, or hedonic feeding. Studies applying molecular genetics or modern circuit analysis tools to the functional characterization of NTS neurons revealed that the NTS can in fact modulate a much larger range of behavioral effectors of energy balance including meal initiation and satiety [3,6e9]. However, little is known about the neural mechanisms through which the NTS regulates forebrain hunger and satiety circuits and the physiological contexts in which these NTS feeding-regulatory forebrain-projecting outputs are engaged. Conceptually, a key question is whether different behavioral effectors of ingestion (that is, satiation, avoidance/aversion, satiety, and foodseeking) are engaged by distinct and functionally specialized NTS neuronal subsets. Evidence that segregated subsets of CCK NTS or TH NTS neurons project to the lPBN or hypothalamus to produce either avoidance/aversive anorexia vs satiety or glucoprivic feeding support this view [10,11]. Alternatively, or in addition to this possibility, recruitment of the same neurons could simultaneously or gradually induce many of these behavioral outputs. The fact that NTS catecholaminergic neurons send collaterals to midbrain and forebrain targets provides a neuroanatomical basis for the latter [12], which could explain the ability of high doses of satiation hormones such as CCK to recruit aversive circuits [13] and/or provide a mechanism for the synergistic feeding suppressive effects produced by the combination of anorectic signals [3,14,15]. Addressing this question with molecularly defined circuit analysis tools is difficult because most identified NTS neurochemical subsets are functionally heterogeneous, respond to multiple cues, and project widely throughout the neuraxis [7,16]. Instead, it may be possible to better understand the functional organization of NTS feeding-regulatory circuits using functionally defined circuit mapping, which could be particularly insightful if subsets of NTS neurons are specialized in the transmission of highly specific sensory cues and organized in a similar fashion as gustatory and vagal sensory neurons [17,18]. Applying such a strategy to signals able to produce satiation or satiety without negative consequences may lead to an important new understanding of how to pharmacologically suppress appetite without undesirable side effects. We previously showed that NTS sensing of the branched-chain amino acid leucine not only modulates the control of meal size, but also rapidly suppresses hunger in fasted animals and increases satiety without the production of conditioned taste aversion [19,20]. In this study, we used NTS leucine sensing as a functional entry point to investigate ascending neural circuits engaged by NTS neurons to modulate hunger and satiety. In these experiments, leucine is injected into the NTS at physiologically relevant doses to model the postprandial increase in brain leucine levels seen in response to the consumption of high-protein meals, a dietary paradigm that potently suppresses food intake. We employed an activity-dependent labeling and circuit mapping strategy that allowed the expression of circuit analysis tools specifically in leucine-sensing neurons and downstream circuits.

Experimental models
All the experiments were conducted on male mice in accordance with the Animals (Scientific Procedures) Act 1986 and approved by the local animal ethics committees. The mice were obtained from Charles River UK (8-week-old C57/bl6J) or the Jackson Laboratories (Agrp-ires-cre, Th-cre, and NPY-hrGFP), housed in individually ventilated cages with standard bedding and enrichment, and maintained in a humiditycontrolled room at 22e24 C on a 12 h light/dark cycle with ad libitum access to water and standard laboratory chow diet unless otherwise stated. Isocaloric modified diets with varying protein amounts were custom made by Research Diets as per the formulations in Suppl. Table 1. For all the experiments using Cre reporter lines, we conducted the work in hemizygous males or wild-type littermates randomly assigned to the experimental groups. For studies on wildtype mice, weight-matched groups were compared. Before each dietary change, the mice were briefly exposed to the new diets to avoid neophobia or other novelty-related responses in subsequent experiments.

Stereotaxic surgical procedures
Surgical procedures were conducted on 9-to 11-week-old mice under isoflurane anesthesia. All the animals received Metacam prior to surgery and 24 h after surgery and were allowed a 1-week recovery period during which they were acclimatized to injection procedures. The mice were stereotactically implanted with bilateral steel guide cannulas (Plastics One) positioned 1 mm above the ARH (A/P: 1.1 mm, D/V: 4.9 mm, and lateral: þ/À0.4 mm from the bregma) or the DMH (A/P: 1.5 mm, D/V: 4 mm, lateral: and þ/À0.4 mm from the bregma) or 2 mm above the caudomedial NTS (cannula-holding bar in a 10 rostro-caudal angle with coordinates relative to the occipital suture: A/ P: þ0.5 mm, D/V: 3 mm, and lateral: þ/À0.4 to the midline). Beveled stainless steel injectors (33 gauge) extending 1 mm (for ARH and DMH) and 2 mm (for NTS) from the tip of the guide were used for injections. For chronic cannula implantation, the cannula guide was secured in place with Loctite glue and dental cement (Fujicem2). Correct targeting was confirmed histologically postmortem. The mice were allowed 1 week of recovery during which they were handled daily and acclimatized to the relevant experimental settings.

NTS leucine injection and acute food intake assessments
Studies were conducted in a home cage environment. For NTS leucine injection, the mice were food deprived for 6 h during the day before receiving a bilateral parenchymal injection of L-leucine (Sigma, 2.1 mM, 50 nl/side, and 50 nl/min) or aCSF (R and D) and either immediately returned to their home cage for food intake analysis or perfused 80/90 min later for histological assessments. For food intake studies, the injection occurred 1 h before dark onset. The mice were refed after the injection and their food intake was monitored over Original Article various time points after refeeding. For the meal initiation experiment, digital cameras were used to record the first 30 min of feeding response after the mice received the brain injection and were provided a food pellet. All the studies were conducted in a cross-over randomized manner on age-and weight-matched groups, and at least 4 days elapsed between each brain injection.
2.5. Activity-dependent induction of cre expression The mice that received AAV-fos-CreER into the NTS or DMH and were chronically equipped with a cannula guide targeting the NTS and underwent a series of induction sessions as follows. The mice received an injection of leucine into the NTS in their home cage as previously described and 80 min later were dosed with tamoxifen metabolite 4hydroxytamoxifen (4-PHT, Sigma, 40 mg/kg i.p.) prepared using a formulation previously described [24].  [30]. The sections were then mounted on slides and coverslipped with Prolong Diamond (Thermo Fisher Scientific). The sections were imaged using a Zeiss Axio slide scanner with a 20x objective or a Leica SP8 confocal microscope with 40x or 63x objectives. The imaging settings remained the same between the experimental and control conditions. Images of tissue sections were digitized, and areas of interest were outlined based on cellular morphology and using Paxinos and Franklin's brain atlas [31]. The images were analyzed using the ImageJ manual cell counter or Zeiss ZEN 2.3 software. To assess projection coverage to the ARH, Imaris software (Oxford Instruments plc) was used to 3D reconstruct the ARH image stacks acquired by the SP8 microscope and analyze the contact areas.

Multiplexed FISH with RNAscope
The mice were perfused as previously described. Their brains were post-fixed in 4% PFA solution overnight and then cryoprotected in 30% sucrose solution in PBS for up to 24 h. The tissue was covered with optimal cutting temperature (OCT) media and then sliced to 16 mm thick using a Leica CM1950 cryostat directly onto Superfrost Plus slides (Thermo Fisher Scientific) in an RNase-free environment. The slides were then stored at À80 C.
Multiplexed fluorescence in situ RNA hybridization (FISH) was performed using RNAscope technology. After epitope retrieval and dehydration, the sections on the slides were processed for multiplexed FISH using an RNAscope LS Multiplex Assay (Advanced Cell Diagnostics). The samples were first permeabilized with heat in Bond Epitope Retrieval solution 2 (pH 9.0, Leica AR9640) at 95 C for 2 min, incubated in protease reagent (Advanced Cell Diagnostics) at 42 C for 10 min, and then treated with hydrogen peroxide for 10 min to inactivate endogenous peroxidases and the protease reagent. The samples were then incubated in z probe mixtures for 2 h at 42 C and washed 3 times. DNA amplification trees were built through incubations in AMP1 (preamplifier), AMP2 (background reducer), and then AMP3 (amplifier) reagents (Leica) for 15e30 min each at 42 C. Between incubations, the slides were washed with LS Rinse buffer (Leica). The samples were then incubated in channel-specific horseradish peroxidase (HRP) reagents for 15 min at 42 C, TSA fluorophores for 30 min, and HRP blocking reagent for 15 min at 42 C. The following TSA labels were used to visualize the z probes: Cy3 (1:500), FITC (1:500), and Cy5 (1:500) fluorophores (PerkinElmer). The brain sections were imaged using a spinning disk Operetta CLS (PerkinElmer) in confocal mode using an sCMOS camera and a 40x automated water-dispensing objective. The sections were imaged with z stacks at intervals of 1 mm. ROIs included the PVH, DMH, NTS, AP, and DMX. Gain and laser power settings remained the same between the experimental and control conditions during each experiment. Harmony software (PerkinElmer) was used to automatically quantify the number of labeled RNA molecules (spots) per cell and the number of labeled cells among other metrics.

Statistical analysis
All the data presented as means AE SEM were analyzed using GraphPad Prism 8. For all of the statistical tests, an a risk of 5% was used to define statistical significance. Dietary and aCSF/leucine treatments where allocated randomly in weight-matched groups. When possible, we performed within-mice comparisons and treatments were delivered in a cross-over manner in weight-matched groups. All the kinetics were analyzed using repeated-measures two-way ANOVAs and adjusted with Bonferroni's post hoc tests. Multiple comparisons were tested with one-way ANOVAs and adjusted with Tukey's post hoc tests. Single comparisons were conducted using two-tailed student's t tests. We used blinding (to mouse genotype, viral treatment, or drug delivered) for in vivo experiments and to conduct image analyses. Additional statistical details on each experiment can be found in the figures or figure legends.

RESULTS
3.1. NTS amino acid sensing inhibits AgRP neurons via a polysynaptic circuit NTS leucine sensing rapidly reduces hunger in fasted rodents [3,32], but the underlying neural circuit mediating this response in unknown.
To characterize the ascending neural circuits engaged by NTS leucine sensing to rapidly inhibit appetite, we first assessed neuronal activation throughout the neuraxis in response to local bilateral NTS leucine administration. Mice were fasted for 6 h and received a site-specific injection of 50 nl of leucine per side into the caudomedial NTS as previously described [32] ( Figure 1A). Neuronal activity was assessed using c-Fos immunohistochemistry 80 min later. NTS leucine induced robust c-Fos expression in the caudomedial NTS as well as in the adjacent area postrema (AP) ( Figure 1B,C). Outside this region, only a few brain sites were significantly activated by local NTS leucine administration compared to aCSF vehicle: the locus coeruleus (LC) and the paraventricular, ventromedial, and dorsomedial nuclei of the hypothalamus ( Figure 1B,C). In contrast, NTS leucine produced a w50% decrease in c-Fos immunolabeling in the ARH ( Figure 1B,C). Of note, NTS leucine did not produce neuronal activation in the parabrachial nucleus (PBN, Figure. 1B), consistent with the lack of conditioned avoidance in response to parenchymal NTS leucine administration in the mice [20]. The ARH contains intermingled orexigenic and anorexigenic neurons including AgRP neurons, which are critical for the development of foodseeking behavior and meal initiation in hungry mice [33,34]. We previously found that NTS leucine sensing robustly increases first-meal latency in fasted mice, hence reducing the drive to approach and consume food [3,32]. This, together with the reduced c-Fos expression in the ARH following NTS leucine administration, prompted us to hypothesize that hindbrain leucine sensing may rapidly inhibit AgRP neurons. To test this, we repeated the same experiment ( Figure. 1A) in the Npy-hrGFP transgenic mice, where the hrGFP signal in the mediobasal hypothalamus selectively labels all AgRP neurons [35]. A majority of ARH NPY/AgRP neurons were activated under control conditions ( Figure 1D,E). As predicted, NTS leucine produced a w50% decrease in NPY/AgRP neuronal activation throughout the rostrocaudal extent of the ARH ( Figure 1D,E). Thus, NTS leucine sensing rapidly inhibits ARH NPY/AgRP neurons. Previous research indicated that NTS inputs can modulate the activity of AgRP neurons [11,20], but the neuroanatomical organization of these inputs and the physiological conditions in which they are engaged to modulate feeding remain unclear. To establish that AgRP neurons are synaptically connected to NTS neurons, we conducted a series of retrograde viral tracing studies. Pseudorabies virus Bartha strain (PRV-Bartha) is a neuroanatomical tracer that is transmitted retrogradely across synapses and can be used to define polysynaptic inputs to infected neurons [36]. PRV-Introvert is a newly developed version of PRV-Bartha in which retrograde viral propagation and reporter expression are activated only after exposure to Cre recombinase with high specificity [21]. We used PRV-Introvert in Agrp-ires-cre mice to serially label chains of presynaptic neurons projecting to ARH AgRP neurons. The mice were sacrificed 0, 24, 48, 72, or 96 h after local  Original Article ARH inoculation with PRV-Introvert and their brains were examined for HA reporter expression ( Figure. 1F). As expected, we did not detect any HA immunolabeling in the brains of wild-type mice injected with the virus (Suppl. Fig. 1a), confirming Cre dependency. At 24 h after injection, PRV-Introvert was detectable in the ARH, indicating that Cremediated recombination occurred locally within 24 h of PRV injection. After 48 h, spread of the Cre-activated PRV virus was observed in multiple hypothalamic sites including the arcuate, ventromedial, dorsomedial, lateral, and paraventricular nuclei (Suppl. Fig. 1b and 1c).
After 72 h, the medial amygdala was labeled (Suppl. Fig. 1d). After 96 h, we detected PRV in several pontine, midbrain, and hindbrain structures. These included the medial parabrachial nucleus (mPBN) (Suppl. Fig. 1e), the ventrocaudal part of the spinal trigeminal nucleus (vcSPVC) (Suppl. Fig. 1f), the rostral ventrolateral medulla (RVLM) (Suppl. Fig. 1e), the AP and NTS, both in its rostral portion and in the lateral portion of the caudomedial NTS ( Figure 1G and Suppl. Fig. 1fe  1g). Thus, ARH AgRP neurons receive inputs from multiple midbrain and hindbrain sites, including the caudomedial NTS. The long survival time necessary to detect the presence of PRV in the NTS suggests that the NTS / AgRP circuit contains more than 1 synapse. Alternatively, the long distance that the virus needs to travel to label hindbrain sites may also explain the lack of signals at 48 and 72 h. To clarify this, we performed Cre-dependent monosynaptic retrograde viral tracing in the Agrp-ires-cre mice using an envelope protein (EnvA) pseudotyped glycoprotein (g)-deleted rabies virus modified to express mCherry (SADDG-mCherry) (EnvA) [22,23]. AgRP Cre-expressing neurons were first genetically modified to co-express TVA (a receptor for the avian sarcoma leucosis virus glycoprotein EnvA) and oG (optimized rabies envelope glycoprotein) via targeted unilateral injections of rAAV8-hSyn-FLEX-TVA-P2A-eGFP-2A-oG into the ARH (Suppl. Fig. 1h). AgRP neurons infected with this construct became selectively competent for transduction by SADDG-mCherry(EnvA) and expressed eGFP. Three weeks later, the mice received a unilateral injection of SADDG-mCherry(EnvA) in the same injection site and were sacrificed at various survival times (up to 14 days). The brains were processed to examine mCherry expression. Two weeks after the SADDG-mCherry(EnvA) injection, we observed dense eGFP expression in the ARH (Suppl. Fig. 1i) together with dense mCherry immunolabeling in the ARH, DMH, and PVH (Suppl. Fig. 1i and 1j).
Overall, 42% of AgRP neurons expressing eGFP co-expressed mCherry. We carefully examined the NTS of 8 successfully infected animals throughout the rostro-caudal extent of the NTS but did not detect rabies-infected cell bodies (Suppl. Fig. 1k). These data support the conclusion that the NTS does not send monosynaptic inputs to AgRP neurons.

NTS PrRP neurons are leucine-sensing and project to AgRP neurons
Previous research showed that a majority of NTS leucine-sensing neurons express TH [3,32], but TH labels a molecularly and functionally diverse group of neurons, prompting us to further analyze the neurochemical identity of NTS neurons responsive to leucine. The NTS contains several neuronal subpopulations responsive to aversive gastrointestinal stimuli or nutritional stress, leading to the formation of visceral malaise, taste aversion, or avoidance [22,23,37e39]. In contrast, some NTS neuronal subtypes are recruited preferentially in response to physiological satiation cues and do not produce aversive anorexia even in the context of pharmacological activation. These include subsets of TH NTS neurons expressing prolactin-releasing peptide (PrRP) [40,41] and recently characterized Calcr NTS neurons [20]. We previously showed that NTS leucine does not produce conditioned avoidance [20], leading us to hypothesize that leucine specifically engages either PrRP NTS or Calcr NTS neurons to suppress feeding. To examine this possibility, we used RNAscope multiplex in situ hybridization (ISH) against Fos, Prlh (transcript for PrRP), and Calcr in caudomedial hindbrain sections of mice that received local NTS leucine injections as previously described (Figure. 1A). We found a significant overlap between PrRP NTS and Calcr NTS neurons (Figure 2A): 75.1 AE 1.3% PrRP NTS neurons expressed Calcr, while 68.3 AE 0.9% Calcr NTS neurons expressed Prlh. Most of the NTS Calcr þ /Prlh À neurons where concentrated in the dorsal NTS, and Calcr was also expressed in a dense Prlh À neuronal population in the AP ( Figure. 2A). Consistent with c-Fos immunolabeling, we found that Fos expression rapidly increased in the NTS and AP in response to local leucine delivery (Figure 2AeB). Leucine activated on average 80% of the PrRP NTS neurons (Figure 2A,C), which represented 34.7 AE 8.6% of the total population activated by leucine in the caudomedial NTS. Analysis of high-resolution ISH images revealed that the number of PrRP NTS neurons was similar between conditions ( Figure. 2D), but leucine increased the expression of Prlh ( Figure. 2E), introducing a role for PrRP neurotransmission in leucine-sensing neurocircuits. We then determined whether PrRP NTS neurons project to ARH AgRP neurons using brain sections from the Agrp-ires-cre mice infected with PRV-Introvert and killed 96 h after infection. We found that a majority of the HA þ neurons labeled 96 h after ARH PRV-Introvert delivery colocalized with TH (61 AE 8%) and PrRP (35 AE 2%) ( Figure 2F,G), confirming that PrRP NTS neurons project to ARH AgRP neurons. Of note, NTS Ha þ neurons did not express GDF15 receptor GFRAL (Suppl. Fig. 2a), indicating that GDF15 does not engage the NTS / AgRP ARH circuit to suppress feeding.
3.3. NTS leucine sensing activates DMH LepR þ /GPR10 neurons projecting to AgRP neurons We next investigated the neuronal populations relaying NTS leucinesensing inputs to AgRP neurons. Given that PrRP NTS neurons represent only one-third of NTS leucine-sensing neurons and one-third of NTS neuron projection to AgRP ARH , and in the absence of a known specific molecular marker for NTS leucine-sensing neurons, we developed a strategy of activity-dependent circuit mapping following NTS leucine administration. We used the AAV8-Fos-ERT2-Cre-ERT2-PEST (AAV-Fos-CreERT2) virus to translate temporally delimited neuronal activity into sustained reporter expression [24]. Neurons expressing AAV-Fos-CreERT2 do not express Cre unless acutely exposed to an activating stimulus together with tamoxifen. A low dose of the tamoxifen metabolite 4-hydroxytamoxifen (4-OHT) allows genetic labeling of transiently activated neurons with high temporal specificity and low background [24]. The Npy-hrGFP mice received a co-injection of AAV-Fos-CreERT2 and AAV8-EF1a-DIO-hChR2(H134R)-mCherry viruses in the caudomedial NTS. Three weeks later, the mice were exposed to 3 experimental inductions, each separated by 96 h ( Figure 3A). During each of these, the mice were fasted for 4 h during the light phase and received a bilateral NTS injection of aCSF or leucine followed 80 min later by an i.p. Injection of 40 mg/kg 4-OHT (hereafter designated as aCSF induced and Leu induced mice, respectively). Access to food was restored 4 h later to avoid food-induced Cre recombination. Mice were sacrificed 14 days after the last NTS injection. Using mCherry immunodetection in brain tissues, we characterized the neuroanatomical distribution of axonal projections and synaptic terminals of NTS leucine-sensing neurons. mCherry expression was dense in the caudomedial NTS of mice induced with NTS leucine injections, confirming the success of the approach (Figure. 3B). We did not detect mCherry þ signals in the ARH of Leu induced mice, indicating that NTS leucine-sensing neurons do not project directly to the ARH ( Figure. 3C). In contrast, we found mCherry þ fibers and terminals in the PVH and ventral DMH of the Leu induced mice compared to controls ( Figure 3D-E). Thus, NTS leucine-sensing neurons project to the PVH and the DMH. The PVH and DMH are both good candidates to relay NTS leucinesensing inputs from PrRP NTS neurons to AgRP neurons. Both the PVH and DMH receive dense projections from TH NTS neurons (Suppl. Fig. 3ae3c), are innervated by PrRP þ fibers, and express GPR10, the receptor for PrRP [42]. Previous monosynaptic retrograde tracing studies identified the PVH and DMH as the main sources of presynaptic inputs to AgRP neurons [43], and channelrhodopsin-assisted circuit mapping studies showed that all PACAP PVH and LepR DMH neurons project to and directly regulate the activity of AgRP neurons [43,44]. However, there is limited understanding of how these neuronal inputs to AgRP neurons may be engaged under physiological conditions to modulate appetite. To examine the role of PACAP PVH and LepR DMH in relaying leucine-sensing information from the NTS to AgRP neurons, we first used RNAscope to colocalize Fos, Adcyap1 (transcript for PACAP), and Prlhr (transcript for PrRP receptor) or Fos, LepR, and Gpr10 in the PVH and DMH respectively of mice that received NTS aCSF or leucine as previously described (Figure. 1A). These experiments confirmed that NTS leucine produces a significant increase in the number of Fos-expressing neurons in the PVH compared to vehicle injection (Suppl. Fig. 3d), but NTS leucine did not increase the number of Prlhr þ , Adcyap1 þ , or Prlhr þ /Adcyap1 þ PVH neurons expressing Fos (Suppl. Fig. 3de3e). In the DMH, NTS leucine increased Fos expression in a group of neurons concentrated in the caudal DMH ( Figure. 3F).
Overall, 30% of the DMH neurons co-expressed LepR and Prlhr (Suppl. Fig. 3g), and NTS leucine significantly increased the number of DMH Prlhr þ , Lepr þ , and Lepr þ /Prlhr þ neurons expressing Fos ( Figure 3F-H). Thus, NTS leucine activates neurons in the DMH that are well positioned to receive inputs from NTS PrRP neurons and project to AgRP neurons. To confirm that the DMH relays NTS leucine-sensing inputs to AgRP neurons, we conducted activity-dependent circuit mapping from DMH neurons activated by NTS leucine. We delivered AAV-Fos-CreERT2 and AAV8-EF1a-DIO-hChR2(H134R)-mCherry viruses into the DMH of the NPY-hrGFP mice and exposed them to the similar induction paradigm previously used ( Figure. 3I) to label axons and synaptic terminals of DMH neurons activated by NTS leucine. In the presence of 4-OHT, NTS leucine induced a significant increase in the number of neuronal cell bodies labeled with mCherry in the DMH ( Figure 3J,K), confirming the success of the approach to label DMH neurons responsive to NTS leucine. We observed mCherry-labeled axons in the ARH ( Figure. 3l) but failed to identify additional mCherry labeling in other brain regions (not shown). These data indicate that DMH neurons activated by NTS leucine provide axo-somatic innervation of the ARH. The projection field labeled with mCherry overlapped with the hrGFP immunofluorescent labeling of NPY/AgRP neurons ( Figure. 3L). Analysis of mCherry þ puncta contacting ARH NPY-GFP neurons suggested that DMH leucine-sensing neurons were synaptically connected to AgRP neurons ( Figure 3L-N). This claim was further supported by 3D immunofluorescent imaging analysis indicating that the contact surfaces between mCherry þ punta/fibers and NPY-GFP cells also colocalized with a presynaptic protein marker synaptophysin (SYP), suggesting the presence of synaptic contacts (Suppl. Fig. 3he3i). Thus, NTS leucine activates DMH neurons that project to AgRP neurons.
3.4. DMH neurons responsive to NTS leucine sensing are necessary for the anorectic effect of NTS leucine and high-protein diets and the long-term control of energy balance To directly test the role of DMH neurons engaged downstream from NTS leucine sensing in the appetite-suppressing effect of NTS leucine, we selectively silenced DMH neurons activated by NTS leucine using the cell-specific expression of tetanus toxin (TT) to prevent synaptic neurotransmitter release. To achieve this, we co-injected AAV-Fos-CreERT2 and AAV-DJ-CMV-DIO-eGFP-2A-TeNT into the DMH of the wild-type mice and exposed to the same induction paradigm as previously described via NTS injections of aCSF or leucine (DMH aCSFÀTT and DMH LeuÀTT , respectively) ( Figure 4A). We then compared the anorectic response to NTS leucine in the DMH LeuÀTT and DMH aCSFÀTT Figure 4: DMH neurons responsive to NTS leucine sensing are necessary for the anorectic effect of NTS leucine and high-protein diets. Diagram of the experimental paradigm to selectively silence DMH neurons receiving inputs from NTS leucine-sensing neurons (a, the DMH aCSFÀTT and DMH LeuÀTT mice). Diagram of the experimental paradigm used to test the feeding effect of NTS leucine in the DMH aCSFÀTT and DMH LeuÀTT mice (b). First meal latency (c), first meal size (d) 24 h body weight change (e), and 24 h food intake (f) in the DMH aCSFÀTT and DMH LeuÀTT mice following an acute injection of aCSF or leucine into the NTS. Average food intake (g) and weight change (h) of the DMH aCSFÀTT and DMH LeuÀTT mice during the 4 weeks following DMH injection with the Tet-Tox virus. Average food intake (i) and body weight (j) in the DMH aCSFÀTT and DMH LeuÀTT mice during transitions from diets containing 20% or 45% of energy as proteins. A schematic summary of the PrRP NTS / LepR/GPR10 DMH / AgRP ARH circuit mediating NTS leucine anorexic effects (k). All the results are shown as means AE SEM. *p < 0.05, **p < 0.01, ***p < 0.001, and ***p < 0.0001 vs aCSF or control group. mice ( Figure. 4B). In DMH aCSFÀTT controls, NTS leucine produced the expected behavioral response, including an increase in meal latency and a decrease in food intake following the injection ( Figure 4C,D, Suppl. Movie 1, and Suppl. Fig. 4a and b). In contrast, NTS leucine failed to increase meal latency and decrease food intake in DMH LeuÀTT mice ( Figure 4C,D, Suppl. Movie 2, and Suppl. Fig. 4a and 4b). Thus, DMH neurons engaged by NTS leucine sensing are required for the acute effects of NTS leucine on meal initiation and satiation. In addition, NTS leucine-induced reductions in 24 h food intake and 24 h weight change were blunted in the DMH LeuÀTT mice ( Figure 4E,F), supporting a role for the NTS / DMH leucine-sensing circuit in the long-term feeding and metabolic consequences of NTS leucine sensing. Of note, over time, the DMH LeuÀTT mice developed a slight but significant hyperphagia ( Figure. 4G) and gained significantly more weight than the DMH aCSFÀTT controls ( Figure. 4H), supporting a role for DMH neurons receiving NTS leucine-sensing inputs in the chronic maintenance of energy balance.
Supplementary video related to this article can be found at https://doi. org/10.1016/j.molmet.2020.101070 Next, we assessed whether this newly characterized circuit was relevant not only for the feeding-suppressive effect of NTS leucine, but also for the anorectic response to high-protein feeding. In mice, acute exposure to a high-protein diet reduces appetite and weight gain [45]. While the central mechanisms mediating these responses are poorly characterized, the NTS is established as a neuroanatomical site responding to high-protein diets [46]. Furthermore, a single highprotein meal is sufficient to increase brain leucine concentration [46], supporting the possibility that NTS leucine-sensing neurons could mediate appetite suppression in response to dietary proteins. To address this, we exposed the DMH LeuÀTT and DMH aCSFÀTT mice to a high-protein diet containing 45% of energy as protein (P45) and isocaloric with a control maintenance diet containing 20% energy as protein (P20, control maintenance diet). To avoid a neophobic response to the P45 diet, the mice were first briefly exposed to P45 pellets (3 times for 30 min on 3 consecutive days). A week later, the mice were switched to the P45 diet for 6 days. In the control DMH aCSFÀTT mice, the P45 diet produced a rapid 30% decrease in energy intake, followed by a sustained 10e15% reduction in daily energy intake in the following days ( Figure 4I and Suppl. Fig. 4c). The anorectic response to the high-protein diet was associated with sustained weight loss ( Figure 4J and Suppl. Fig. 4d), confirming the feeding and metabolic effects of the P45 diet under these conditions. In contrast, the anorectic response to P45 was blunted in the DMH LeuÀTT mice ( Figure 4I and Suppl. Fig. 4c) and remarkably, the P45 diet did not produce a weight response in these mice ( Figure 4J and Suppl. Fig. 4d). Thus, DMH neurons activated by NTS leucine are required for the acute anorectic response to high-protein diets, while other pathways likely mediate the sustained anorectic effect of dietary proteins. Unexpectedly, these results indicate that the NTS / DMH leucine-sensing circuit contributes to the metabolic effect of high-protein diets. These results provide a central mechanism for the behavioral and metabolic responses to dietary proteins. We then examined the neuronal activation patterns of various AgRP output areas in the DMH aCSFÀTT and DMH LeuÀTT mice after an acute terminal NTS leucine injection with the paradigm outlined in Figure 1A.
We observed approximately twice as many eGFP-TeNT-expressing cells in the DMH LeuÀTT over DMH aCSFÀTT mice while the NTS leucine-induced DMH activation patterns were comparable between the two groups, indicating the success of our targeting strategy, which did not appear to affect the responsiveness of DMH neurons to NTS leucine sensing (Suppl. Fig. 4ee4f). Of note, approximately one-half of all of the eGFP-TeNT-labeled DMH cells were activated by the terminal NTS leucine injection in the DMH LeuÀTT mice while only approximately less than 10% of the labeled cells were activated by the aCSF injection (Suppl. Fig. 4g), validating the strategy's specificity. As expected, the inhibitory effects of the NTS leucine injection on AgRP ARH neurons was blunted in the DMH LeuÀTT mice (Suppl. Fig. 4h). We then analyzed neuronal activation in a range of previously identified AgRP output areas [47]. Consistent with ARH activation patterns, the inhibitory effects of NTS leucine on the paraventricular nucleus of the thalamus (PVT) and lateral hypothalamus (LH) were blunted in the DMH LeuÀTT mice (Suppl. Fig. 4h). However, we did not observe significant modifications of NTS leucine-induced PVH activation, demonstrating the complexity of the relative contributions to those brain areas receiving influences from both the NTS and AgRP ARH (Suppl. Fig. 4h). We did not observe robust c-Fos expression in the anterior bed nucleus of the stria terminalis (aBNST) and central amygdala (CeA) across all of our conditions in our mild 6 h fasting paradigm (Suppl. Fig. 4i). Together, these results revealed the efficacy of NTS leucine sensing in regulating the downstream targets of AgRP ARH neurons and corroborating the role of DMH as an inhibitory relay between NTS leucine sensing neurons and AgRP neurons.

DISCUSSION
Our findings reveal a mechanism through which nutrient sensing in the NTS regulates food-seeking behavior, satiety, and long-term energy balance through polysynaptic inhibition of AgRP neurons. We demonstrate that PrRP NTS neurons engage this circuit in response to the detection of the branched-chain amino acid leucine, a signal of dietary protein availability. Silencing of DMH neurons responsive to NTS leucine sensing blunts leucine's appetite-suppressive effects and dampens the anorexic and weight loss responses to a high-protein diet, extending the role of this circuit in the behavioral and metabolic responses to dietary proteins. Our data expand the characterization of the functional diversity of NTS TH neurons to include a subset of neurons expressing PrRP and CTR, projecting to the DMH and modulating feeding initiation and satiety via downstream projections to AgRP neurons. While other neuronal populations (including NTS and PVH neurons) activated by NTS leucine are likely involved in other behavioral, metabolic, and neuroendocrine outputs of NTS leucine sensing, the circuit described in this report is sufficient to entirely explain the appetitive consequences of NTS leucine detection. These results support a model in which specialized neuronal populations regulate specific behavioral outputs. Intriguingly, although PrRP and CTR neurons of the NTS have been shown to project to the PBN [42,48], this site is not activated by NTS leucine, suggesting further functional diversity among these neurons. This highlights the relevance of the activity-dependent labeling and circuit-mapping strategy for the characterization of functionally relevant brain relays connecting NTS leucine-sensing neurons and AgRP neurons. Recent research indicated that AgRP neurons integrate various sensory inputs including environmental food-related cues and visceral mechanosensory and nutritional inputs [18,49e51]. Our research extends the integrative capability of AgRP neurons to include brainstem nutrient-sensing inputs. Given that visceral vagal afferents terminate in the caudomedial NTS where PrRP/CTR neurons are concentrated, it is likely that intestinal sensory inputs engage the same circuit as NTS leucine to inhibit AgRP neurons. Thus, the PrRP NTS / LepR DMH circuit may be specialized in integrating nutritional cues arising from multiple central and peripheral interoceptors. These sites may be only partially functionally redundant given the indication that caloric density is the primary information carried by vagal afferents activated by nutrients in the gut [17], whereas NTS nutrient sensing surveys nutritional status and postabsorptive nutrient availability. With the ability to monitor the availability of specific nutrients and relay this information to forebrain centers mediating the long-term control of energy balance, PrRP NTS neurons and their projections to LepR DMH neurons are well positioned to contribute to the production of nutrient-specific satiety, a wellestablished behavior lacking mechanistic characterization [52]. DMH LepR neurons also process environmental food-related cues in the regulation of AgRP activity, which may be integrated with former signals as well [44]. To the best of our knowledge, these are the only inhibitory inputs to AgRP neurons identified to date. However, polysynaptic retrograde tracing from AgRP neurons in this study revealed additional medullary inputs to AgRP neurons that may extend the inhibitory control of this population. While the novel circuit we identified in this study provides new insights into hindbrain ascending pathways for the control of feeding, additional research is needed to clarify how the NTS integrates distinct food-related signals to produce a variety of behavioral responses. In particular, whether NTS leucine-responsive cells are also responsive to other nutrients such as glucose and free fatty acid remains to be determined. Collectively, this study resolves the mechanisms through which NTS nutrient-sensing modulates food-seeking behavior and provides insights into the functional organization feeding regulatory circuits, creating new opportunities for the treatment of hyperphagic obesity and related metabolic disorders.

AUTHOR CONTRIBUTIONS
AHT conducted the experiments, analyzed the data, and drafted the manuscript. DN conducted the experiments and data analysis. TD and HS conducted the experiments. CB designed and conducted the experiments, analyzed the data, and prepared the manuscript.

MATERIALS AND CORRESPONDENCE
Correspondence and material requests should be addressed to Clemence Blouet.