Abstract
The anorexigenic peptide glucagon-like peptide-1 (GLP-1) is secreted from gut enteroendocrine cells and brain preproglucagon (PPG) neurons, which respectively define the peripheral and central GLP-1 systems. As peripheral satiation signals are integrated in the nucleus tractus solitarius (NTS), PPGNTS neurons are assumed to link the peripheral and central GLP-1 systems, forming a unified GLP-1 gut-brain satiation circuit. This hypothesis, however, remains unsubstantiated. We report that PPGNTS neurons encode satiation in mice, consistent with vagal gastrointestinal distension signalling. However, PPGNTS neurons predominantly receive vagal input from oxytocin receptor-expressing vagal neurons, rather than those expressing GLP-1 receptors. Furthermore, PPGNTS neurons are not necessary for eating suppression induced by the GLP-1 receptor agonists liraglutide or semaglutide, and semaglutide and PPGNTS neuron activation additively suppress eating. Central and peripheral GLP-1 systems thus suppress eating via independent gut-brain circuits, hence PPGNTS neurons represent a rational pharmacological target for anti-obesity combination therapy with GLP-1 receptor agonists.
Glucagon-like peptide 1 (GLP-1) acts as an incretin hormone and anorexigenic neuropeptide, prompting the successful and ongoing development of GLP-1-based therapies for type 2 diabetes and obesity1,2. Endogenous GLP-1 is produced both by enteroendocrine cells in the gut, and preproglucagon (PPG) neurons in the brain, which are the defining populations of the peripheral and central GLP-1 systems, respectively3,4. PPG neurons in the nucleus tractus solitarius (PPGNTS neurons) suppress eating when chemogenetically or optogenetically activated5–7, consistent with substantial pharmacological evidence for anorexigenic GLP-1 signalling in the brain (reviewed by Muller et al4). Physiologically, PPGNTS neurons are the major source of GLP-1 in the brain, are necessary for stress-induced hypophagia, and their inhibition or ablation elicits transient hyperphagia during large intakes7. Although glutamate is a co-transmitter in these neurons8,9, selective Ppg knockdown has confirmed the necessity of proglucagon-derived peptides for their anorexigenic role10. PPGNTS neurons are thus the crucial component of the central GLP-1 system, which they comprise along with numerous populations of GLP-1 receptor (GLP-1R)-expressing neurons found throughout the brain11,12. Whilst it is widely assumed that endogenous peripheral GLP-1 interacts with the central GLP-1 system (via vagal and/or endocrine activation of PPGNTS neurons) to control eating under physiological conditions, this link remains subject to debate and has not been demonstrated empirically4,13–15. The difficulty in substantiating this link partly arises from the inherent complexity of interrogating these widely-distributed systems using pharmacological approaches, compounded by observations that native GLP-1 and pharmacokinetically-optimized therapeutic GLP-1 receptor agonists (GLP-1RAs) suppress eating via divergent signalling pathways16–22.
Numerous studies demonstrate the value of selective transgenic manipulations to determine the neuroanatomical organization and physiological functions of specific cell populations involved in eating control, including those comprising parts of the peripheral and central GLP-1 systems7,23–26. Here we utilized similar transgenic and viral approaches to address whether PPGNTS neurons have a role in physiological satiation, and determined their anatomical and functional connectivity to molecularly defined neuronal populations mediating gut-brain satiation signalling. Specifically, we tested the prevalent but unsubstantiated hypothesis that peripheral GLP-1 signals to the brain to suppress eating via vagal and/or endocrine activation of central GLP-1-producing preproglucagon (PPG) neurons, i.e. that peripheral and central GLP-1 systems comprise a unified, directly connected gut-brain satiation circuit. We furthermore tested the role of PPGNTS neurons in eating suppression induced by the anti-obesity GLP-1RAs liraglutide and semaglutide, to establish whether this neuronal population has translational importance as a distinct therapeutic target for obesity treatment.
Results
PPGNTS neurons selectively encode large meal satiation
PPGNTS neurons are not necessary for control of daily or long term food intake or bodyweight in ad libitum eating mice7. However, it is unknown whether they regulate within- or between-meal parameters, or whether the absence of an ablation-induced bodyweight phenotype masks more subtle alterations in energy expenditure or physical activity. We addressed these questions by metabolic phenotyping of ad libitum eating mice following viral ablation of PPGNTS neurons (Fig. S1a). Food intake over the circadian cycle was unaffected by neuronal ablation in either sex (Fig. S1b-c,k-l), and there was no effect on meal size, frequency or duration (Fig. S1e-g). Similarly, ablation did not affect locomotion, energy expenditure, bodyweight or water intake (Fig. S1d, i,j,m). As a positive control in this model we did, however, successfully replicate a previous report7 that ablation induces hyperphagia both during post-fast refeeding, and after a short liquid diet preload (Fig. S1n-o).
The replicable observation that ablation of PPGNTS neurons elicits transient hyperphagia only under conditions manipulated to induce large intakes7 is consistent with results in rats indicating that these neurons are activated during ingestion of unusually large meals27. We thus tested the hypothesis that hyperphagic responses observed after PPGNTS neuronal ablation are specifically due to a delay in meal termination, to establish a bona fide role for PPGNTS neurons in the process of satiation. We conducted high resolution meal pattern analysis using home cage FED pellet dispensers (Feeding Experimentation Device28), and observational analysis of liquid diet intake, to test the effects of acute chemogenetic inhibition of PPGNTS neurons on termination of large solid and liquid meals (Fig. 1a,g). PPGNTS inhibition increased fasting-induced pellet refeeding during hour 1 in a sex-independent manner, and this effect was driven by increased meal size, rather than frequency (Fig. 1b-f). The hyperphagic effect in this model was also confirmed to be specific to large meals, as inhibition had no effect under ad libitum eating conditions (Fig. S1p-r). We then modified a previously used Ensure liquid diet preload paradigm7 to test whether PPGNTS neurons are necessary for satiation during consumption of large liquid meals (Fig. 1g), as suggested by a previous cFos expression study in rats29. Under these conditions, Ensure intake was increased by PPGNTS neuron inhibition, and, consistent with effects on pellet intake in the FED system, this increase was driven by increased duration of Ensure eating (Fig. 1h-k). These data demonstrate that both under conditions of normal and negative energy balance, PPGNTS neurons are recruited to encode physiological satiation specifically by ingestion of large meals.
PPGNTS neurons suppress eating without behavioural disruption
The observation that PPGNTS neurons selectively encode satiation during large meals, but apparently do not control intake under ad libitum eating conditions, suggests they have capacity to suppress eating when stimulated. Evidence for such capacity has been reported previously5–7, and could indicate translational potential for PPGNTS neurons as a target for therapeutic suppression of eating, provided that the eating suppression is robust, is not compensated for, and is not associated with nausea/malaise. We therefore extended these studies by testing whether hypophagia induced by chemogenetic activation of PPGNTS neurons was followed by compensatory rebound hyperphagia, or elicited significant disruption to the behavioural satiety sequence (Fig. 2a,e). In ad libitum eating mice, PPGNTS activation reduced intake by ~40% in the first 24 hours (Fig. 2b) in a sex-independent manner (Fig. S2b), predominantly driven by reductions in the first 5 hours of the dark phase (Fig. S2a). No compensatory hyperphagia occurred, hence cumulative intake and bodyweight were still reduced 48 hours after acute CNO administration (Fig. 2c-d). This robust suppression of eating with sustained reduction in intake was also observed when PPGNTS neurons were activated immediately prior to dark onset refeeding after a prolonged (18hr) fast (Fig. 2e, S2f). We thus combined the FED system with infrared video in this paradigm to investigate changes to the behavioural satiety sequence under relatively naturalistic home cage conditions. Following a period of eating (Fig. S2g), PPGNTS neuron activation advanced the point of satiation by ~15 minutes (shift from time bin 4 to 1; Fig. 2f-g), and the stochastic sequence of satiety behaviours (eating → grooming → inactive) was not disrupted. Quantitative analyses revealed that PPGNTS activation did not significantly alter eating rate (Fig. S2h), but reduced eating duration during the first 15 minutes (Fig. 2h), as expected from the left-shifted satiation point. Time inactive was modestly increased, and grooming and active behaviours appeared to be correspondingly decreased, however the temporal patterns of these behaviours were maintained (Fig. 2i-k). That the robust suppression of intake elicited by their activation is not associated with significant alteration to the behavioural satiety sequence further supports the idea that PPGNTS neurons have translational potential as a pharmacological target for eating suppression. These observations also contrast with previously reported effects in this assay of the emesis/nausea-inducing agent lithium chloride and the GLP-1RA Exendin-4, which reduce eating rate and almost completely suppress grooming and other active behaviours30,31. Instead, and consistent with prior evidence that PPGNTS activation does not condition flavour avoidance5, the absence of behavioural satiety sequence disruption reported here supports the view that PPGNTS neuronal activation suppresses eating without inducing nausea/malaise, in contrast to the effects of peripherally administered GLP-1RAs.
Glp1r-expressing vagal afferent neurons suppress eating and condition flavour avoidance
Direct synaptic input from undefined population(s) of vagal afferent neurons (VANs) to PPGNTS neurons26,32 presumably underlies the ability of gastrointestinal distension and large liquid intakes to induce cFos expression in this NTS population29,33, and to drive their role in large meal satiation. However, the molecular identities of VAN inputs to PPGNTS neurons remain to be characterized. VANs defined by their expression of the GLP-1 receptor gene (Glp1r) innervate the gut and have been identified as a predominantly mechanosensory population that encodes gastrointestinal distension, as well as likely mediating paracrine satiation signalling by peripheral GLP-114,24,25. However, it is unknown whether PPGNTS neurons are a major synaptic target of Glp1r VANs, and thus to what extent direct vagal communication between peripheral and central GLP-1 systems is neuroanatomically plausible. To address this question, we developed approaches for viral targeting and activation of Glp1r VANs with chemogenetic and optogenetic effectors (Fig. 3a,f), to determine whether these manipulations produced effects on eating consistent with Glp1r VANs being part of a unified gut-brain satiation circuit with PPGNTS neurons. Chemogenetic activation of Glp1r VANs in ad libitum eating mice suppressed eating during the dark but not subsequent light phase (Fig. 3b, S3a). Notably, however, the magnitude of this hypophagic effect appeared less robust than that following chemogenetic activation of PPGNTS neurons. Bodyweight was also transiently decreased, driven by suppressed eating rather than increased energy expenditure (Fig. 3c-e, S3b-c). The modest anorexigenic effect of chemogenetically activating Glp1r VANs in ad libitum eating mice precluded use of the 18hr fasted BSS paradigm in this model. We therefore instead tested whether optogenetic activation of the central axon terminals of ChR2-transduced Glp1r VANs was able to condition avoidance of, or a preference for, a paired novel flavour of Kool-Aid (Fig. 3f-g). Optogenetic activation of Glp1r terminals within the NTS conditioned avoidance of the paired flavour (Fig. 3h), and also modestly suppressed eating in a subsequently conducted acute eating test (Fig. 3i). This finding contrasts with the lack of disruption to the BSS we observed following activation of PPGNTS neurons, and a previous report that chemogenetic activation of PPGNTS neurons does not condition flavour avoidance5, but is consistent with several reports that GLP-1RAs condition flavour avoidance and reduce reward-related behaviours34–36. These results support prior evidence that Glp1r VANs at least partly mediate endogenous or exogenous peripheral GLP-1 satiation signalling20,22. However, the modest anorexigenic effect and conditioning of avoidance produced by chemogenetic and optogenetic activation of Glp1r VANs argues against the hypothesis that this population are the primary driver of eating suppression by PPGNTS neurons.
Oxtr rather than Glp1r VANs are the major vagal input to PPGNTS neurons
We next tested the neuroanatomical connectivity between PPGNTS neurons and Glp1r VANs, using two complementary circuit mapping approaches. Utilizing a cross of GLP-1R-Cre and PPG-YFP mouse strains37 combined with unilateral viral targeting of nodose ganglia, we selectively labelled the NTS terminal fields of Glp1r VANs with tdTomato, allowing simultaneous visualization with YFP-expressing PPGNTS somata and dendrites (Fig. 4a). While extensive innervation by both right and left branch Glp1r VANs was observed along the rostro-caudal extent of the NTS, there was little regional overlap with PPGNTS neurons. In the caudal NTS, Glp1r vagal afferents predominantly terminated dorsomedial to PPGNTS somata, and their terminal fields extended considerably beyond the rostral extent of the PPGNTS population (Fig. 4b-c).
The absence of overlap between Glp1r VAN terminals and PPGNTS somata does not preclude some vagal input via their distal dendrites in the dorsomedial NTS. We therefore quantified direct synaptic connectivity between Glp1r VANs and PPGNTS neurons by Cre-dependent monosynaptic retrograde rabies virus tracing in combination with RNAscope fluorescence in situ hybridization for Glp1r and oxytocin receptor gene (Oxtr) expression in nodose ganglia (Fig. 4d). Expression of Oxtr was investigated based on reports of target-based scRNAseq analysis of VANs (target-scSeq), which suggest that mechanosensation of gastric and intestinal distension are predominantly encoded by VANs defined by Glp1r and Oxtr expression, respectively (with some overlap), and that an additional subpopulation of chemosensitive VANs expressing Glp1r (but not Oxtr) innervate the intestine and mediate paracrine GLP-1 satiation signalling14,25. As previously reported26, we observed that rabies virus-GFP was expressed extensively in nodose ganglia, confirming substantial monosynaptic vagal innervation of PPGNTS neurons. Surprisingly, however, we found that <5% of PPGNTS neuron-innervating VANs express Glp1r alone, and similarly <5% of VANs expressing Glp1r alone synapse onto PPGNTS neurons (Fig. 4e-f). Conversely, 33% of all PPGNTS neuron-innervating VANs express Oxtr alone, and 21% of VANs which express Oxtr alone synapse onto PPGNTS neurons (Fig. 4g-h). VANs expressing both Oxtr and Glp1r (which are presumably mechanosensory) comprise 20-25% of these populations (Fig. S4c), and 26% of these Oxtr / Glp1r VANs synapse onto PPGNTS neurons, comprising an additional 9% of vagal input to this population (Fig. S4d-h). We thus identified PPGNTS neurons as an important synaptic target of Oxtr VANs, in addition to the catecholaminergic target population previously identified25. Our findings strongly suggest that Oxtr-rather than Glp1r-expressing VANs are the primary source of gastrointestinal distension signals driving PPGNTS neuron-mediated satiation. Furthermore, as the overwhelming majority of VANs expressing Glp1r but not Oxtr (which presumably includes the chemosensory population) do not synapse onto PPGNTS neurons, they are highly unlikely to be a functionally relevant target of vagal-dependent paracrine signalling from the peripheral GLP-1 system.
PPGNTS neurons are necessary for oxytocin-induced eating suppression
Having identified Oxtr-expressing VAN input to PPGNTS neurons, we next characterized the effects of oxytocin itself on this population. We first performed ex vivo calcium imaging using coronal brainstem slices from transgenic mice expressing GCaMP3 in PPGNTS neurons38,39. Slices were taken at a rostro-caudal level containing the majority of PPGNTS neurons, and which is reported to contain substantial Oxtr VAN terminal fields25. Superfusion of oxytocin activated 84% of glutamate-responsive PPGNTS neurons, as determined by increased calcium-dependent fluorescence (Fig. 5a-e). Notably, superfusion of GLP-1 has no effect on PPGNTS neurons in the same preparation37.
Peripherally administered oxytocin is reported to suppress eating in a vagal-dependent manner40,41, so we subsequently tested whether PPGNTS neurons were necessary for this effect, given their direct synaptic inputs from Oxtr VANs. While oxytocin acutely suppressed eating in ad libitum eating control mice, this effect was completely abolished in PPGNTS neuron-ablated mice (Fig. 5f-i, S5a-b), confirming these neurons as a necessary component of the gut-brain circuit recruited by peripheral oxytocin to suppress eating.
PPGNTS neurons are not a major synaptic target of area postrema Glp1r neurons
Having determined the neuroanatomical implausibility of vagal transmission of peripheral GLP-1 signals to PPGNTS neurons, we next investigated a potential route for endocrine GLP-1 signalling to these NTS neurons via Glp1r neurons in the area postrema (AP), which lacks a blood-brain barrier and has been implicated as a site where circulating GLP-1 and GLP-1RAs may act to suppress eating17,20,42,43. Monosynaptic retrograde tracing from PPGNTS neurons combined with in situ hybridization for Glp1r was again utilized (Fig. 6a). In contrast to robust synaptic input from VANs, synaptic inputs to PPGNTS neurons from the AP were relatively sparse (Fig. 6b). Of these sparse inputs from the AP, 25% expressed Glp1r, however these represented <3% of all Glp1r AP neurons (Fig. 6c). As catecholaminergic AP neurons express GLP-1R and have been proposed to link peripheral GLP-1 signalling to central nuclei (including the NTS) involved in eating control43, we further characterized PPGNTS neuronal input from tyrosine hydroxylase immunoreactive AP neurons (Fig. 6d-e, S6a-c). ~20% of the sparse AP inputs to PPGNTS neurons are catecholaminergic, consistent with a report that PPGNTS neurons are indirectly activated by noradrenaline44. However, these presynaptic AP neurons comprised only 3% of all catecholaminergic AP neurons (Fig. 6e). Therefore, PPGNTS neurons are unlikely to be a functionally relevant target for peripheral GLP-1 and/or GLP-1RAs acting via the AP to suppress eating.
Liraglutide and semaglutide suppress eating independently of PPGNTS neurons
Limitations to rabies virus propagation efficiency inevitably result in an underestimate of the total number of neurons (including Glp1r VANs and AP neurons) providing direct synaptic input to PPGNTS neurons. Nevertheless, it is apparent that the majority of Glp1r VANs and AP neurons are not directly presynaptic to the central GLP-1 system. However, one or both of these Glp1r populations may be polysynaptically connected to PPGNTS neurons, in which case vagal and/or endocrine peripheral GLP-1 could still provide substantial input to the central GLP-1 system.
PPGNTS neurons may alternatively (or additionally) receive input from other Glp1r-expressing neuronal populations which are accessible to peripheral GLP-1/GLP-1RAs and are reportedly necessary for their hypophagic effects, such as glutamatergic neurons45 or GABAergic NTS neurons46. We therefore investigated whether PPGNTS neurons are a necessary component of any neurocircuits recruited by peripheral GLP-1RAs to suppress eating, by testing whether PPGNTS neuronal ablation attenuates the anorexigenic effects of two long-acting anti-obesity GLP-1RAs, liraglutide and semaglutide (Fig. 7a). Liraglutide robustly suppressed intake and bodyweight over 24 hours in ad libitum eating eGFP-transduced control mice, however ablation of PPGNTS neurons had no effect on the magnitude of eating suppression at any timepoint, or on 24h bodyweight loss (Fig. 7b-d, S7c-g). Semaglutide suppressed eating to an even greater extent than liraglutide, and similarly PPGNTS ablation had no impact on acute or delayed eating suppression, or on bodyweight loss (Fig. 7e-g, S7h-l).
These findings demonstrate that PPGNTS neurons are not necessary for GLP-1RA-induced suppression of eating. While access to the brain by GLP-1RAs is limited, they are able to access several circumventricular Glp1r-expressing nuclei (in addition to the AP), which may be upstream of PPGNTS neurons, or part of a subset of the downstream targets of these neurons12,17,21,47. Relevant GLP-1RA-accessible downstream Glp1r populations likely include neurons in the hypothalamic arcuate nucleus, which are at least partly necessary for liraglutide-induced suppression16. Administration of GLP-1RAs to PPGNTS neuron-ablated mice cannot differentiate between whether there are any Glp1r populations upstream of PPGNTS neurons that are functionally dispensable for eating suppression, or if GLP-1RAs only recruit circuits downstream or entirely independent of PPGNTS neurons. We therefore investigated whether the same highly anorexigenic dose of semaglutide used in the ablation experiment was able to induce neuronal activation of PPGNTS neurons, by quantifying cFos expression in the PPG-YFP mouse line47,48. Semaglutide induced robust cFos expression within the AP and throughout the rostro-caudal extent of the NTS (Fig. 7h-j), and additionally in hypothalamic and parabrachial nuclei (Fig. S7m-v). However, consistent with a report that the GLP-1RA Exendin-4 does not increase cFos expression in PPGNTS neurons49, and that these neurons themselves do not express Glp1r32,37, we found that <3% of PPGNTS neurons were activated by semaglutide (Figure 7k). This finding demonstrates that systemically-administered GLP-1RAs act centrally via ascending circuits parallel to, but independent of, PPGNTS neurons, and/or by partially bypassing them to activate a subset of their downstream targets.
Semaglutide and PPGNTS neurons additively suppress eating
The convergent lines of neuroanatomical and functional evidence reported here suggest that, rather than comprising part of a unified GLP-1 gut-brain circuit, PPGNTS neurons suppress eating via circuits which are anatomically and functionally distinct from those recruited by peripheral endogenous GLP-1 and peripherally administered GLP-1RAs. To support the hypothesis that the circuits mediating eating suppression by peripheral GLP-1RAs and PPGNTS neurons are indeed entirely independent, or at least only converge at limited peripherally-accessible downstream population(s), it is a necessary to demonstrate that their concurrent activation is capable of suppressing eating in an additive manner. We therefore tested this hypothesis by administering the same dose of semaglutide that elicited robust eating suppression and neuronal activation in earlier experiments, in combination with chemogenetic activation of PPGNTS neurons, and assessed intake and bodyweight over 72 hours (Fig. 8a). As expected, either manipulation alone suppressed eating over the first 24 hours, with semaglutide eliciting the stronger effect. Crucially, their combined effect was significantly additive to semaglutide alone, throughout the duration of acute chemogenetic activation (Fig. 8b-e). Consistent with our observation that PPGNTS neuron activation suppresses eating without compensatory rebound hyperphagia, both cumulative intake and bodyweight were reduced at 24 and 48 hours in both of the semaglutide-treated groups. The apparent floor effect on eating suppression at 24 hours confirmed that an appropriately high dose of semaglutide was used, but this likely precluded the ability to elicit significantly additive weight loss at these later timepoints (Fig. 8f-i). Nevertheless, the additive effect of semaglutide and PPGNTS activation on cumulative intake was significant even 72 hours after a single CNO dose (Fig. 8j), supporting the translational potential of this combined approach. The observed additivity could theoretically be explained by incomplete GLP-1R saturation by semaglutide within a peripherally-accessible subset of Glp1r-expressing nuclei downstream of PPGNTS neurons. However, as we deliberately used a high dose of semaglutide to overcome this possibility, and chemogenetic activation is itself a strong and robust stimulus, this explanation is unlikely. Rather, as GLP-1RAs do not suppress eating via PPGNTS neurons, and since these neurons project to numerous central nuclei involved in eating control which are not accessible to GLP-1RAs, the most parsimonious explanation is that the observed additivity must derive from concurrent activation of distinct anorexigenic neurocircuits.
Discussion
Here we report that PPGNTS neurons encode satiation specifically during large meals, and have capacity for pharmacological activation to suppress eating without compensatory rebound hyperphagia or behavioural disruption. Activation of Glp1r VANs similarly suppressed intake, but did condition flavour avoidance, and complementary circuit mapping approaches demonstrated that PPGNTS neurons are not a major synaptic target of this vagal population. We report that PPGNTS neurons instead predominantly receive vagal input from Oxtr VANs, and are required for peripheral oxytocin-induced eating suppression. Similarly, PPGNTS neurons are at most a minor synaptic target of Glp1r neurons in the area postrema, suggesting that endocrine GLP-1 signalling from the periphery by this route does not require PPGNTS neurons. Consistent with this observation, PPGNTS neurons are not recruited by peripherally administered semaglutide, or required for the anorexigenic effects of liraglutide or semaglutide, and concurrent administration of semaglutide and activation of PPGNTS neurons suppresses eating in an additive manner. We therefore conclude that, contrary to the prevalent hypothesis, the peripheral and central GLP-1 systems are components of anatomically and functionally independent eating control circuits. Furthermore, while pharmacokinetically-optimized GLP-1RAs may access a limited subset of Glp1r neuron populations downstream of PPGNTS neurons, such partial convergence of recruited circuits does not preclude additive suppression of eating. PPGNTS neurons are thus a rational pharmacological target for obesity treatment both in their own right and for combination therapy with GLP-1RAs.
Author Contributions
Competing interests statement
The FR + FMG laboratory receives funding from AstraZeneca, Eli Lilly and LGM for unrelated research and FMG consults for Kallyope (New York). All other authors have nothing to declare.
Methods
Animals
We used 116 mice of both sexes (≥10 weeks) from five previously reported strains, all maintained on C57BL/6 backgrounds and bred in house. For selective Cre-dependent viral targeting and ex vivo Ca2+ imaging of PPG neurons, we used mGlu-Cre/tdRFP 50 and mGlu-Cre/GCaMP3 strains 38, referred to herein as PPG-Cre:tdRFP and PPG-Cre:GCaMP3, respectively. For visualization of PPG neuron somata, axons and dendrites, we used the mGlu-YFP strain (Reimann et al., 2008; referred to as PPG-YFP). For selective Cre-dependent viral targeting of GLP-1 receptor-expressing neurons we used the Glp1r-Cre/tdRFP strain 51, or a cross with the PPG-YFP strain (Card et al., 2018; referred to as GLP-1R-Cre x PPG-YFP). All mice were kept on a 12h light/dark cycle with chow and water available ad libitum and group housed until surgery and/or behavioural experiments. Within-subjects design experiments were conducted using sex-balanced cohorts of appropriate genotype littermates as far as possible. Similarly, for between-subjects and mixed model design experiments, littermates were semi-randomly allocated to virus groups to ensure groups were balanced for sex and age as far as possible. Power calculations were not performed, appropriate group sizes (detailed in figure legends) were determined from pilot experiments and our previously published studies using these models and behavioural paradigms 7,49,52.
Experiments conducted in the UK were performed in accordance with the U.K. Animals (Scientific Procedures) Act 1986 and with appropriate institutional approval. Experiments conducted in the U.S. were performed in accordance with the U.S. Public Health Service’s Policy on the Humane Care and Use of Laboratory Animals and experimental protocols were approved by the Institutional Animal Care and Use Committees of Florida State University and the University of Florida. Experiments conducted in Switzerland were approved by the Canton of Zurich’s Veterinary Office (ETH Zurich).
Stereotaxic Surgery
NTS Virus Injections
Injection of viral vectors targeted to the caudal NTS were performed as previously described 7,26,49. Mice were anaesthetized with intramuscular medetomidine (1 mg/ kg) + ketamine hydrochloride (50 mg/kg) or 1.5-2.5% isoflurane, and given carprofen analgesia (5mg/kg, s.c.). Core temperature was maintained using a homeothermic monitoring system, and appropriate depth of surgical anaesthesia was determined by absence of pedal reflex. The skull was restrained in a stereotaxic frame, and the head flexed downwards such that the nose and neck were at a right angle. The scalp was incised from the occipital crest to first vertebrae, and muscle layers parted to expose the atlanto-occipital membrane. The membrane was bisected horizontally with a 30G needle to expose the brainstem surface with obex as an anatomical landmark. Viral vectors (as detailed in figures and Methods Table 1) were injected via pulled glass micropipettes at the following coordinates from obex: +0.1mm rostral, ±0.5mm lateral and −0.35mm ventral. Viruses encoding chemogenetic effectors, diptheria toxin subunit A (DTA) and control reporters were all bilaterally injected in volumes of 200-250nl. Mice were allowed to recover for a minimum of 3 weeks before behavioural experiments began. For monosynaptic retrograde tracing from PPG neurons in the caudal NTS, 300nl of a 1:1 mix of AAV5-EF1a-FLEX-TVA:mCherry and AAV8/733-CAG-FLEX-RabiesG were bilaterally injected +0.1mm rostral, ±0.4mm lateral and 0.35mm ventral to obex. 21 days later, 400nl in total of (EnvA)-RABV-∆G-GFP was unilaterally injected at two injection sites: +0.1mm rostral, +0.25mm lateral and 0.35-0.45mm ventral to obex; and +0.1mm rostral, +0.4mm lateral and 0.35-0.45mm ventral to obex. Mice were transcardially perfused for histological processing and in situ hybridisation 7 days later. We have previously optimised, validated and published all viral targeting strategies used in the present study 7,26,49,52. Post-mortem tissue sections were processed from all mice used in behavioural studies and verified for appropriate expression of fluorescent reporters. Mice in which viral injection targeting was inaccurate, or transduction efficiency below expected levels, were omitted from analyses.
Nodose Ganglia Virus Injections
Injection of viral vectors targeted to the somata of vagal afferent neurons in the nodose ganglia were performed as previously described 52. Mice were anaesthetized with 1.5-2.5% isoflurane and given carprofen analgesia (5mg/kg, s.c.), then the ventral surface of the neck was incised, and muscles parted to expose the trachea. The vagus nerve was separated from the carotid artery to allow access to the nodose ganglia. In each nodose ganglia, a total volume of 500nl of viral vector (AAV5-hSyn1-DIO-hM3Dq:mCherry, AAV5-EF1a-DIO-hChR2(H134R):mCherry or AAV-PHP.S-CAG-DIO-tdTomato, as detailed in figures) was injected into sites rostral and caudal to the laryngeal nerve branch, using a bevelled tip pulled glass micropipette and pneumatic microinjector. Viruses encoding chemogenetic and optogenetic effectors were injected bilaterally, and the virus for tdTomato-visualized projection tracing was injected unilaterally into left or right nodose ganglia. Mice were allowed to recover for a minimum of 2 weeks before behavioural experiments or transcardial perfusion.
Optical Fibre Implantation
Two weeks after bilateral injection of AAV5-EF1a-DIO-hChR2(H134R):mCherry in nodose ganglia, optical fibres were prepared and implanted as previously described 53. Optical fibres (CFLC230–10 ceramic ferrules with FT200UMT fibre, Thorlabs) were unilaterally implanted above the right caudal NTS, 7.5mm caudal, 0.25mm lateral and 4.0mm ventral to Bregma. Skull screws, superglue and dental cement were used to secure the fibre, and mice were allowed to recover for an additional 2 weeks prior to behavioural testing.
Behavioural Studies
Drug Administration
Clozapine N-oxide (CNO; Hello Bio / Enzo) was administered intraperitoneally at 2 mg/kg in 2 ml/kg dose volume for all experiments, typically 15 minutes prior to dark onset. We have previous determined that in our hands CNO at this dose does not affect eating behaviours in control virus transduced mice, and similarly that the chemogenetic effectors used do not have any constitutive activity which affects eating behaviours 7,52. To minimize animal use and maximize statistical power, all chemogenetic experiments in the present study were therefore conducted in hM3Dq-expressing mice using a within-subjects design. All mice received both CNO and saline vehicle in a counterbalanced manner, and hence acted as their own controls. Similarly, when assessing the effect of PPGNTS ablation on the anorexigenic actions of oxytocin, liraglutide and semaglutide, we used a mixed model design, whereby mice in DTA-ablated and control cohorts all received drug and vehicle in a counterbalanced manner. Oxytocin (Tocris) was administered intraperitoneally 15 minutes prior to dark onset at 0.4 mg/kg, based on reports that this dose and route of administration elicits vagal-dependent eating suppression in mice 40,41. Liraglutide and semaglutide (gift from Lotte Bjerre Knudsen, Novo Nordisk) were administered subcutaneously 30 minutes prior to dark onset at 0.2 mg/kg and 0.06 mg/kg, respectively, in 5 ml/kg dose volume, based on recommendations from LBJ and previous reports of the anorexigenic effects of these drugs in mice 16,17,54.
Eating Behaviour Paradigms
Drug- or neuronal manipulation-induced changes to eating behaviour were assessed from dark onset in ad libitum eating or fasted mice. In the ad libitum eating paradigm, mice were habituated (≥5 sessions) to being fasted for the final 3 hours of the light phase and receiving saline/vehicle injections 5-30 mins prior to return of food at dark onset. This protocol minimized hypophagia from handling and injection stress, and entrained mice to eat consistently from dark onset, without needing to induce negative energy balance. All experiments using either manual or automated measurement of food intake used this protocol for assessment of ad libitum eating, except for metabolic phenotyping of PPGNTS ablated mice experiment, in which mice were not handled or injected and were already habituated to test cages. To assess the effect of chemogenetic manipulations on the behavioural satiety sequence, and large meals driven by refeeding after a prolonged fast, mice were fasted for 18 hours prior to dark onset. To assess the effect of optogenetic activation of Glp1r-expressing vagal afferent neurons on acute feeding, mice were fasted for the entire light phase and intake measured during the first 30 minutes of the dark phase. The effect of optogenetic activation was also assessed using a within-subjects design, with all mice tested for 30 minute intake under ‘laser on’ (20ms blue light pulse every 3 seconds, ~5mW intensity) and ‘laser off’ (tethered but no light pulses) conditions in a counterbalanced manner.
Food Intake Measurement
Food intake was measured manually, using open source FED pellet dispensers (Feeding Experimentation Device; Nguyen et al., 2016), or using commercially-available Phenomaster (TSE Systems) or Promethion (Sable Systems) systems. For manual measurement of intake in the ad libitum eating paradigm, mice were weighed at the start of the 3 hour fast, then a pre-weighed amount of food was returned at dark onset. Food was again weighed at 1, 2, 4, 6 (GLP-1RA experiments only), and 21 hours, at which point 24 hour bodyweight change was also determined. FED dispensers were used for all experiments involving chemogenetic manipulations of PPGNTS neurons, the Phenomaster system for chemogenetic activation of Glp1r-expressing vagal afferent neurons, and the Promethion system for metabolic phenotyping of PPGNTS ablated mice. For all food intake measurement systems and eating behaviour paradigms, mice were habituated to the test equipment and all aspects of the paradigm (including vehicle dosing where appropriate) prior to the start of testing. Mice were considered habituated after their intakes during ≥3 consecutive habituation sessions were not significantly different, and no directional trend was apparent. Meal pattern analysis was conducted on automated food intake data from experiments testing the effect of inhibition or ablation of PPGNTS neurons. A meal was defined as the sum of all bouts ≥0.02g with intra-meal intervals <10 minutes, based on the standard operating procedure of the UC Davis Mouse Metabolic Phenotyping Centre 55. Consumption of Ensure liquid diet was measured manually, and the temporal pattern of Ensure drinking was measured by offline video coding of licking duration (blinded to drug treatment), using the BORIS open source video coding software package 56.
Behavioural Satiety Sequence Analysis
Alterations to the behavioural satiety sequence (BSS) following chemogenetic activation of PPGNTS neurons were determined using the continuous monitoring BSS protocol 57,58, adapted for use with mice and FED dispensers. Mice were fasted for 18 hours, FED dispensers were returned to home cages at dark onset, and behaviour recorded for 40 minutes using infrared video cameras. Behaviours were subsequently coded offline using BORIS software by trained observers (Cohen’s κ for inter-rater reliability >0.9) blinded to treatment group. Behaviours were coded as mutually-exclusive categories: eating, drinking, grooming (including scratching), inactive (resting and sleeping) and active (locomotion and rearing). In pilot experiments, the duration of water drinking was found to be extremely low and unaffected by our manipulations, hence was omitted from further analyses. The total duration mice spent exhibiting behaviours in the remaining 4 categories were calculated for 8 × 5 minute time bins. For qualitative and semi-quantitative evaluation of the stochastic sequence of satiety behaviours (eating → grooming → inactive), the mean durations of these 3 categories were plotted across all time bins separately for saline control and CNO activated conditions. To aid visualization of neuronal activation-induced acceleration of the sequence, horizontal dotted lines were added denoting the time bin during which the (probabilistic) transition from feeding to resting occurs, which is typically considered the satiation point / onset of satiety. Data were also presented and analysed to quantitatively test the effect of chemogenetic activation over time in each behaviour category.
Indirect Calorimetry
We measured respiratory exchange ratio and energy expenditure concurrently with food intake from PPGNTS ablated and Glp1rNodose-hM3Dq mice, using the indirect calorimetry functionality integrated into the Phenomaster and Promethion systems. Mice were habituated to test cages for ≥3 days before testing, and metabolic data were collected for 24 hours for between-subjects analysis (PPGNTS-DTA ablated vs PPGNTS-mCherry controls), or during two 24 hour test sessions (Glp1rNodose-hM3Dq, counterbalanced for CNO administration), separated by ≥48 hour washout periods during which mice remained in test cages. Respiratory exchange ratio (RER) was obtained from measurement of mice’s O2 consumption (ml/kg/hr) and CO2 production (ml/kg/hr), using the equation: RER = VCO2 / VO2. Energy expenditure (EE) was calculated using the Weir equation: EE = 3.941 × VO2 + 1.106 × VCO2. Raw data from both systems were used to generate standardized output files, which were imported into the CalR analysis tool 59 for production and analysis of intake and metabolic data over light and dark phases and total circadian cycle. As now recommended for analysis of calorimetry data from these systems, energy expenditure was not normalized to bodyweight.
Conditioned Flavour Preference
Whether optogenetic activation of Glp1r vagal afferent neurons conditioned a preference for (or avoidance of) a flavour was assessed as previously described 52. Experiments were conducted within sound-attenuated cubicles, using behavioural chambers equipped with two sipper tubes connected to contact-based licking detection devices, allowing high resolution measurement of licking responses (Med Associates Inc.). Following recovery from surgeries for nodose ganglia virus injection and optical fibre implantation, individually-housed mice were placed on a food and water restriction regime, under which they were maintained at 90% of starting bodyweight and were limited to 6 hours of water access per day. Mice were habituated to behavioural chambers (including being tethered to fibre cables) and trained to lick for a 0.025% saccharin solution during daily 1 hour habituation sessions, conducted during the light phase. Mice were considered trained to saccharin licking once they showed <10% between-session variability in the number of licks, a criterion all mice reached within 10 sessions.
Once trained, a ‘pre’-test was conducted in which mice were given access to two novel Kool-Aid flavours (cherry or grape, both 0.05% in 0.025% saccharin solution) for 10 minutes, with sipper bottle positions switched after 5 minutes to avoid position bias. Mice then underwent 3 × 1 hour training sessions for each flavour (alternately over 6 days), in which both bottles contained the same flavour. One flavour was paired with laser stimulation (CS+), such that licking triggered blue light laser stimulation via a TTL output signal. Specifically, 10 licks triggered a 20ms light pulse of ~5mW intensity, with additional licks during the following 10 seconds having no programmed consequences. Further bouts of ≥10 licks triggered additional pulses in the same manner throughout the 1 hour session. During training sessions with the unpaired (CS-) flavour, mice were tethered but licking did not elicit laser stimulation. Upon completion of these training sessions, mice underwent a ‘post’-test identical to the ‘pre’-test, i.e. both flavours were available, and licking did not elicit laser stimulation. The number of licks for the laser-paired flavour during ‘pre’ and ‘post’ tests was used to calculate preference ratios (CS+ licks / total licks) for the flavour before and after training, to determine if optogenetic stimulation of Glp1r vagal afferent neurons increased or decreased preference for the paired flavour.
Immunohistochemistry & In Situ Hybridization
Tissue Preparation
Mice were deeply anaesthetized then transcardially perfused with ice-cold PB/PBS (0.1M, pH 7.2) then 4% formaldehyde in PB/PBS. Brains and NG (when required) were extracted and post-fixed in 4% formaldehyde at 4°C overnight (≤2 hours for NG), before being cryoprotected in 20-30% sucrose solution for ≥24 hours at 4°C. Brains were sectioned into 30-35μm coronal sections, collected free-floating and stored at 4°C until processing for immunofluorescent labelling as detailed below. NG were sectioned into 10μm sections, collected on Superfrost Plus microscope slides and stored at −20°C until processing for in situ hybridization as detailed below.
Immunofluorescent labelling
Brain sections were processed for amplification of fluorescent reporter signals by immunofluorescent labelling of tdRFP, mCherry, eYFP, eGFP and/or GCaMP3 as previously described 7. Briefly, sections were incubated free-floating with primary antibodies (see Methods Table 1) overnight at 4°C in PBS with 2% normal goat/donkey serum and 1% BSA, followed by 2 hours at room temperature with secondary antibodies conjugated to fluorophores appropriate for the native fluorescent reporter being amplified (i.e. Alexa Fluor 488 for eYFP/eGFP/GCaMP3 and Alexa Fluor 568 for tdRFP/mCherry).
RNAscope In situ Hybridization (Nodose Ganglia)
Sections from nodose ganglia of mice previously injected with viruses for monosynaptic retrograde rabies tracing were processed for in situ hybridization of Glp1r and Oxtr mRNA using the RNAscope assay as previously reported 60. Sections were cut at 10μm on a cryostat and collected on Superfrost Plus slides, then allowed air-dry at room temperature for one hour.
Slides were then dipped in molecular grade ethanol and further air-dried overnight at room-temperature. RNAscope in situ hybridization was performed on these sections using the RNAscope Multiplex Fluorescent Kit v2 (Advanced Cell Diagnostics) as per the manufacturer’s instructions, with a modification to the pre-treatment procedure (Protease IV incubation conducted for 20 min at room temperature) that allows for preservation of the fluorescent reporter signal while also providing optimal signal from the target mRNAs. Probes for Glp1r, Oxtr and appropriate positive (Ubc) and negative (DapB) controls (detailed in Methods Table 1) were hybridized and after completion of the procedure slides were immediately cover slipped using Prolong Antifade medium.
RNAscope In situ Hybridization (Brainstem)
Brainstem sections containing the area postrema were pre-treated with hydrogen peroxide for 30 minutes at room temperature, slide-mounted in dH2O and air dried overnight. Sections were subsequently processed for in situ hybridization of Glp1r mRNA using the same reagents and protocol as nodose ganglia sections, followed by additional processing for immunofluorescent labelling of GFP and tyrosine hydroxylase (TH). Incubation with primary antibodies against GFP and TH was performed concurrently overnight at room temperature, with the remaining protocol conducted as described above for labelling of fluorescent reporters. Sections were then dehydrated in increasing concentrations of ethanol, cleared in xylene and cover slipped using Cytoseal 60.
cFos Quantification
For immunohistochemical validation of chemogenetic PPGNTS neuron activation, brains from PPGNTS-hM3Dq mice were processed for immunofluorescent labelling of cFos as previously reported 7. For quantification of semaglutide-induced cFos expression, mice expressing eYFP in PPG neurons were habituated to handling and the standard ad libitum eating behaviour paradigm, including vehicle injection 30 minutes prior to dark onset. Food intake manually quantified after 3 hours during habituation sessions and on the day semaglutide was administered, allowing a within-subjects quality control for the effect of semaglutide in this experiment. On the test day, mice were injected with vehicle or semaglutide (0.06mg/kg as per behavioural studies) and transcardially perfused 4 hours later. Coronal brain sections were prepared as above, then processed for immunoperoxidase labelling of cFos with DAB-Ni followed by immunofluorescent amplification of eYFP as previously described 26.
Imaging
Brain sections labelled for fluorescent reporters and/or cFos expression were imaged using an upright epifluorescence and brightfield microscope (Leica) with a Retiga 3000 CCD camera (QImaging). For co-localization of DAB-Ni labelled cFos and PPG-eYFP neurons, brightfield and fluorescence images were sequentially captured in the same focal plane. Quantification of cFos expression and co-localization was conducted using merged native brightfield (DAB) and fluorescent (PPG-eYFP) images. For clarity of presentation, brightfield DAB images were inverted and pseudocolored prior to merging with fluorescent channels. Nodose ganglia and brainstem sections processed for in situ hybridization and/or fluorescent reporters were imaged with a Keyence BZ-x700 at 20x or 40x in 0.6μm optical sections, or a Leica TCS SP8 confocal microscope at 20x. For imaging of sections processed for in situ hybridization, sections hybridized with positive and negative control probes were used to determine exposure time and image processing parameters necessary for optimal visualization of mRNA signals and control for possible degradation. Generation of montages from individual images, brightness and contrast adjustment, and quantification of cFos expression using the Cell Counter plugin were all performed using Fiji open source biological image analysis software 61.
Brain Slice Ca2+ Imaging
Imaging Data Capture
Coronal brainstem slices (200μm) were obtained from PPG-Cre:GCaMP3 mice and used to assess the effects of bath-applied oxytocin on PPGNTS neuron calcium dynamics using methods previously described in detail 39. Oxytocin was dissolved in aCSF (3mM KCl, 118mM NaCl, 25mM NaHCO3, 5mM glucose, 1mM MgCl2, 2mM CaCl2; pH 7.4) to give a bath concentration of 100nM, based on reports that this concentration elicits robust activation of vagal afferent neurons under ex vivo conditions 40. Slices were superfused with aCSF for ≥10 minutes, with the final 5 minute period prior to oxytocin application used to determine baseline fluorescence intensity. Slices were then superfused with oxytocin solution for 3-5 minutes, washed with aCSF for ≥10 minutes, then finally superfused with 100μM glutamate for 1 minute as a positive control to confirm imaged neurons were healthy and responsive to glutamatergic input. GCaMP3 fluorescence was excited at 460 ± 25 nM using an LED light source, for 250ms every 5 seconds. Imaging was conducted using a widefield microscope (Zeiss) with 40x water immersion lens and captured at 12-bit on a CCD camera (QClick, QImaging). Data were obtained from 8 experiments (i.e. recordings from single slices) from 3 mice.
Imaging Data Analysis
Time-lapse image recordings were imported into FIJI software, with the StackReg plugin used to correct for XY drift. Regions of interest (ROIs) were manually drawn around all PPGNTS somata in the field of view, with additional ROIs used to determine background intensity for each experiment. Background intensity was subtracted from ROIs and a cubic polynomial function was used to adjust for bleaching. Data are presented as ΔF/F0, where F0 is the mean fluorescence intensity over the 5 minute baseline period, and ΔF is the intensity at each timepoint with F0 subtracted. Response magnitudes were determined using the area under the curve (AUC) over 4 minutes from first application of oxytocin. To ensure artifactual fluctuations were not included in analyses, only fluorescence changes for which the magnitude was greater (or less) than 3 standard deviations of the baseline period AUC were considered to be ‘responders’. As the noise level (i.e. variability in baseline AUC) differs between slice recordings, this threshold is not absolute, hence there is some degree of overlap between the AUC of ‘non-responsive’ ROIs from more noisy recordings and ‘responsive’ ROIs from less noisy recordings. As a further quality control, oxytocin-responsive ROIs were only included for analysis if they subsequently were responsive to glutamate.
Quantification and statistical methods
Data are presented as mean ± SEM, and were analysed for statistical significance as detailed in figure legends using Student’s t-test, one-way within-subjects or two-way within-subjects/mixed-model ANOVA (with the Greenhouse-Geisser correction applied where appropriate). Where data were not normally distributed, non-parametric equivalents were used as detailed.
Significant one-way ANOVA tests were followed by pairwise comparisons with Tukey’s correction for multiple comparisons. For two-way ANOVA, either simple main effects were reported, or significant interactions were reported and followed by pairwise comparisons with Sidak’s correction for multiple comparisons. The threshold for statistical significance was considered <0.05, and significant comparisons are reported in all figures as: * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001. For transparency, all comparisons in which p<0.1 (but ≥0.05) are additionally reported with exact p values shown. Statistical analyses were conducted using GraphPad Prism 7 or IBM SPSS Statistics 26.
Data availability
The datasets supporting the current study are available from the lead contact on request. All mouse lines, plasmids and reagents used in this study have been previously published and/or are commercially available. Further information and requests for resources and reagents should be directed to and will be fulfilled by Prof. Stefan Trapp (s.trapp{at}ucl.ac.uk).
Acknowledgements
We would like to thank Lotte Bjerre Knudsen at Novo Nordisk for valuable discussions and provision of liraglutide and semaglutide. We would also like to thank Myrtha Arnold at ETH Zurich and Yalun Tan at the University of Florida for their expert technical assistance, and Kevin Beier at UC Irvine for providing rabies virus for retrograde tracing. This study was supported by MRC project grant MR/N02589X/1 to ST. The initial collaboration between the Trapp, de Lartigue and Langhans labs which generated this work was made possible thanks to a UCL Global Engagement Fund award to DB, and a UCL Neuroscience ZNZ Collaboration award to ST and WL. Research in the de Lartigue lab was funded by the NIH (NIDDK grant R01 DK116004) and with institutional support from the University of Florida College of Pharmacy. Research in the Reimann/Gribble laboratories was funded by the Wellcome Trust (106262/Z/14/Z and 106263/Z/14/Z) and the MRC (MRC_MC_UU_12012/3). Research in the Rinaman laboratory was funded by the U.S. National Institutes of Health (MH059911 and DK100685).