Introduction

The hypothalamus is a critical hub for the integration of interoceptive signals and the initiation of homeostatic behaviors regulating need states such as hunger and thirst. In particular, molecularly defined subpopulations of neurons in the arcuate nucleus, paraventricular nucleus, and lateral hypothalamic area control food intake [1, 2], while median preoptic nucleus excitatory and inhibitory neurons promote and suppress, respectively, water drinking [3, 4]. Beyond physiological needs, hypothalamic nuclei also control food and fluid consumption driven by hedonic motives (i.e., pleasurable sensory perception or affective state), as they interface with corticolimbic and midbrain monoaminergic systems that encode reward and decision making [5, 6].

In recent years, the parasubthalamic nucleus (PSTN), a differentiation of the premammillary lateral hypothalamic area that is highly interconnected with brain regions regulating interoception and appetite, has emerged as a new player in the regulation of ingestive behaviors (see [7] for review). Significant induction of c-Fos expression, a marker of neuronal activation, has been observed in the rodent PSTN following predatory hunting of cockroaches, early active phase surge of food intake, refeeding after food deprivation, anorexia-inducing amino-acid deficient diet refeeding, and sucrose drinking following water deprivation [8,9,10,11,12,13,14]. While all PSTN neurons are glutamatergic (VGluT2-positive), they comprise subpopulations expressing high levels of either Tac1 (encoding preprotackykinin-A, a precursor of substance P and neurokinin A) or Crh (encoding corticotropin-releasing factor, CRF) with minimal overlap [14,15,16]. Both PSTNTac1 and PSTNCrh neurons show strong activation in response to refeeding in food-deprived mice [14].

Manipulations of PSTN neurons via optogenetic, chemogenetic, and targeted cell ablation approaches have further demonstrated their functional implication in consummatory behaviors (see [7] for review). Specifically, stimulation of PSTNVGluT2 neurons projecting to the paraventricular nucleus of the thalamus (PVT), PSTNTac1 neurons as a whole, as well as PSTNTac1 neurons projecting to the central nucleus of the amygdala (CeA), PVT, parabrachial nucleus, or nucleus of the tractus solitarius, reduces food intake in ad libitum fed mice [14, 17]. This effect is not observed when stimulating PSTNCrh neurons or PSTNTac1 neurons projecting to the bed nucleus of the stria terminalis (BNST), highlighting the cell-type and pathway specificity of the PSTN’s influence on food intake [14]. Activation of PSTNAdcyap1 neurons, which overlap with both the Tac1 and Crh populations, also reduces the time spent feeding in hungry mice regaining access to food [18]. On the other hand, inhibiting PSTNTac1 somas counters the anorectic effect of a malaise-inducing agent (lipopolysaccharide), neophobia (first-time exposure to sucrose), or appetite-suppressing hormones (amylin, cholecystokinin, peptide YY), while ablating PSTNVGluT2 neurons abolishes anorexia induced by glucagon-like peptide-1 [12, 14, 19] and silencing PSTN neurons as a whole (CaMKIIα promoter) attenuates cholecystokinin-induced and fear-induced feeding suppression [18, 20]. While these studies elegantly demonstrated the ability of several PSTN subpopulations to suppress feeding, none of them directly addressed the role of PSTN neurons activated upon food access resumption in hungry animals in the control of ensuing food consumption. To address this gap of knowledge, we used chemogenetics in “Targeted Recombination in Active Populations” mice (TRAP2 mice, in which the sequence encoding iCre-ERT2 is inserted in the Fos locus without disrupting endogenous Fos expression [21]), to selectively re-activate the ensemble of refeeding-activated PSTN neurons and determine their influence on consummatory behaviors. Moreover, in addition to measuring the amount of food and fluid consumed (as was done in previous studies), we recorded the latency to first bite of food or first lick of fluid as a behavioral indicator of the mouse’s motivation to eat or drink, which emerged as highly sensitive to PSTN manipulations.

We found that, in hungry mice, the PSTN refeeding ensemble drastically increases the latency to initiate refeeding with both familiar chow and a novel, palatable food (Froot Loops), but does not control the amount of food consumed. In thirsty mice, this ensemble also delays sucrose consumption but accelerates water consumption, with again no influence on the amount of fluid consumed. Given that previous studies only reported measures of intake (i.e., amount consumed), we next sought to examine which subpopulations of PSTN neurons might be driving these latency effects, using cell-type and pathway-specific chemogenetic manipulations.

Materials and methods

Animals

TRAP2 (Fostm2.1(icre/ERT2)Luo/J, stock #030323 [21]), Crh-Cre (Crh-IRES-Cre, B6(Cg)-Crhtm1(cre)Zjh/J, stock #012704 [22]), and Tac1-Cre (B6;129S-Tac1tm1.1(cre)Hze/J, stock #021877 [23]), breeders were obtained from The Jackson Laboratory. C57BL/6 J mice were obtained from Scripps Research rodent breeding colony. All mice used for experimentation were heterozygous for the Cre allele. All experimental subgroups contained a mix of age-matched males and females. Mice from the same litter were distributed across experimental subgroups.

Mice were maintained on a 12/12 h light/dark cycle. Food (Teklad LM-485, Envigo) and reverse osmosis purified water were available ad libitum, except for a 24-h period of food deprivation or 4-h water deprivation in relevant experiments. All mice were single-housed in static caging with Sani-Chips (Envigo) bedding one week prior to behavioral assays and remained in these housing conditions for the duration of experimentation. All mice were at least 10 weeks old at the time of surgery. All procedures adhered to the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of The Scripps Research Institute.

Drugs

For iCre-ERT2 activation in TRAP2 mice [21, 24], the estrogen receptor ligand 4-hydroxytamoxifen (4-OHT) was obtained from Hello Bio (HB6040). 4-OHT was dissolved in 100% ethanol and then mixed 1:1 with a mixture of 1 part castor oil (Sigma-Aldrich 259853) and 4 parts sunflower oil (Sigma-Aldrich S5007). Ethanol was then removed via vacuum centrifugation and the remaining oil was again diluted in the same 1:4 oil mixture to achieve a concentration of 10 mg/mL before intraperitoneal (i.p.) injection at a dose of 50 mg/kg (5 mL/kg injection volume, 23-gauge needle). Clozapine-N-oxide (CNO) was used as the ligand for hM3Dq and hM4Di designer receptors [25, 26] and was obtained from Enzo Life Sciences Inc. (freebase, BML-NS105-0025) or Hello Bio (dihydrochloride, HB6149). CNO freebase was dissolved in dimethyl sulfoxide (DMSO) and diluted in 0.9% saline (0.5% final DMSO concentration) and CNO dihydrochloride was dissolved directly in saline for i.p. injection (3 mg/kg body weight, 10 mL/kg injection volume, 27-gauge needle). Salvinorin B (SalB) was used as the ligand for the κ-opioid receptor-based designer receptor (KORD [27]) and was obtained from Hello Bio (HB4887). SalB was dissolved in 100% DMSO for subcutaneous (s.c.) injection (10 mg/kg body weight, 1 mL/kg injection volume using a Hamilton 250 μL syringe #81108, 27 gauge, point style 4 needle). Prior to SalB testing, mice were habituated for two days to DMSO s.c. injections.

Viral vectors

Adeno-associated viral serotype 2 (AAV2) vectors encoding the hM3Dq excitatory or hM4Di inhibitory designer receptor fused to the red fluorescent protein mCherry, or mCherry alone, under the control of the human synapsin promoter (hSyn) and in a Cre-dependent manner (Double-floxed Inverted Open reading frame, DIO), were obtained from the Vector Core at the University of North Carolina at Chapel Hill (AAV2-hSyn-DIO-hM3Dq-mCherry, Addgene plasmid # 44361, lot 8269, titer 1.5 × 1013 vg/mL; AAV2-hSyn-DIO-hM4Di-mCherry, Addgene plasmid # 44362, lot 8268, titer 1.8 × 1013 vg/mL; AAV2-hSyn-DIO-mCherry, Addgene plasmid # 50459, lot 8267, titer 1.3 × 1013 vg/mL) [26]. An AAV8 encoding the inhibitory designer receptor KORD fused to the fluorescent protein mCitrine under the control of the human synapsin promoter in a Cre-dependent manner was obtained from Addgene (AAV8-hSyn-dF-HA-KORD-IRES-mCitrine, viral prep # 65417-AAV8, lot v43122, titer 2.1 × 1013 gc/mL) [27]. A retrograde AAV [28] encoding the Cre enzyme fused to the green fluorescent protein (GFP) under the control of the human synapsin promoter was obtained from Addgene (AAVrg.hSyn.HI.eGFP-Cre.WPRE.SV40, viral prep # 105540-AAVrg, lot V102961, titer 2.5 × 1013 gc/mL). An AAV8 vector expressing hM3Dq-mCherry under a short EF1α promoter in a Cre- and Flp- dependent manner was generated, packaged, and purified by the laboratory of Karl Deisseroth (AAV8-nEF-Con/Fon-hM3Dq-mCherry, lot 7280, titer 1.23 × 1012 gc/mL) and used in conjunction with a retrograde AAV encoding the Flpo enzyme under an EF1α promoter (AAVrg-EF1a-Flpo, Addgene viral prep # 55637-AAVrg, lot v56725, titer 1.02 × 1012 gc/mL) [29].

Experimental cohorts

The data were collected from seven separate cohorts of mice.

A cohort of 26 TRAP2 mice (13 males + 13 females) featured three experimental subgroups: food-deprived mice injected with 4-OHT upon refeeding (n = 9), food-deprived mice injected with vehicle upon refeeding (to control for leaky Cre activity, n = 9), and food-deprived mice injected with 4-OHT without refeeding (to control for PSTN activity during fasting, n = 8). These mice were refed 8 h after 4-OHT administration. In this experiment, all mice were injected with the hM3Dq vector and the effect of CNO was tested according to a between-subject design. For the induction of Cre activity, all mice were transferred to a clean, new cage (without food), and 24 h later, 4-OHT (or vehicle) was injected immediately prior to the placement of chow pellets in the wire lid of food-deprived cages. Deprivation-induced body weight loss was confirmed and vigorous interactions of refed mice with the food hopper were noted, but consumption measures were not collected to avoid interfering with the cages during the time window of Cre activation. An additional cohort of 16 TRAP2 mice (7 males + 9 females) featured three experimental subgroups: food-deprived mice injected with 4-OHT (n = 6), food-deprived mice injected with vehicle (to control for leaky Cre activity, n = 5), and ad libitum fed mice injected with 4-OHT (to control for baseline PSTN activity, n = 5).

A cohort of 46 Tac1-Cre mice (22 males + 24 females) featured three experimental subgroups injected with either the hM3Dq (n = 16), hM4Di (n = 13), or mCherry (n = 15) vectors. The effect of CNO was tested according to a between-subjects design.

A cohort of 19 Crh-Cre mice (10 males + 9 females) were all injected with the hM3Dq vector and were treated with either CNO or vehicle for a between-subject analysis. The treatment assigned to each mouse remained the same across all assays.

A cohort of 19 C57BL/6 J mice (12 males + 7 females) featured three experimental subgroups that were co-injected with the hM3Dq and KORD vectors (1:1 premixed cocktail) in the PSTN and the retrograde Cre vector in the CeA (n = 6), the BNST (n = 7) or the PVT (n = 6) for pathway-specific manipulations. In this experiment, all mice were injected with chemogenetic ligands (CNO or SalB) and their respective vehicles for within-subject analysis (i.e., each mouse was tested four times). CNO was tested first, then SalB. In each case, the order of ligand and vehicle administration was counterbalanced between mice.

A cohort of 6 Tac1-Cre mice (1 male + 5 females) and 7 Crh-Cre mice (3 males + 4 females) were injected with the Con/Fon hM3Dq vector into the PSTN and retrograde Flpo vector in the CeA. All mice were injected with CNO and vehicle for within-subject analysis.

A cohort of 10 Tac1-Cre mice (6 males + 4 females) and 10 Crh-Cre mice (5 males + 5 females) were injected with the Con/Fon hM3Dq vector into the PSTN and retrograde Flpo vector in the BNST. All mice were injected with CNO and vehicle for within-subject analysis.

Stereotaxic surgery, histology, and behavioral testing

Methodological details are provided as Supplementary information.

Statistics

Raw data was processed in Microsoft Excel and statistical analysis was performed using GraphPad Prism software. Experiments evaluating between-subject effects of CNO in TRAP2 (including mCherry counts) and Tac1-Cre cohorts were analyzed using ordinary one-way ANOVAs, with an alpha of 0.05. Multiple comparisons were conducted with the Tukey’s test for TRAP2 mice and with the Dunnett’s test for Tac1-Cre mice (using the mCherry group as control condition). Experiments comparing the effect of CNO versus vehicle were analyzed using either unpaired (Crh-Cre mice) or paired (pathway-targeted and Con/Fon cohorts) two-tailed t-tests, with an alpha of 0.05. In each graph, individual values are plotted, bars show group averages, and error bars represent standard error of the mean.

Results

The PSTN refeeding ensemble delays the latency to initiate feeding

Subpopulations of PSTN neurons become highly active in response to refeeding after food deprivation, as indexed by c-Fos induction [10, 13, 14]. To determine the functional significance of this neuronal ensemble, TRAP2 mice were injected with a Cre-dependent hM3Dq-encoding virus in the PSTN and administered 4-OHT upon refeeding following 24 h of food deprivation. In addition to this experimental group, a first control group was injected with 4-OHT in a fasted state to control for PSTN activity related to hunger, and another control group was injected with vehicle upon refeeding to control for leaky (i.e., 4-OHT-independent) Cre recombination (Fig. 1A). Native fluorescence of the mCherry reporter was used to estimate the size of the PSTN neuronal ensemble targeted in each of the three experimental groups. As expected, there were very few mCherry labeled cells in vehicle-injected mice (Fig. 1B). Accordingly, there was a significant main effect of group (F2,23 = 22.52, p < 0.0001), whereby mice injected with 4-OHT at the time of refeeding (p = 0.0001) and those injected with 4-OHT in a fasted state (p < 0.0001) had significantly more mCherry positive cells than vehicle-injected controls. The ensemble captured in a fasted state was not significantly different than that captured upon refeeding (p = 0.3691), providing a sizeable population of PSTN cells to inform on the functional specificity of the PSTN refeeding ensemble upon chemogenetic manipulation.

Fig. 1: The ensemble of PSTN neurons activated by refeeding delays consumption of familiar and palatable foods but hastens water drinking.
figure 1

A Experimental design and representative image of mCherry fluorescence showing targeted recombination in the PSTN of a mouse injected with 4-OHT immediately prior to refeeding (scale bars: gray, 500 μm; white, 200 μm). B The number of PSTN mCherry-positive cells illustrates the extent of recombination across experimental conditions. Behavioral testing was performed 30 min after injection of the chemogenetic actuator CNO, and 24 h after food deprivation for chow (C) and Froot Loops (D), or 4 h after water deprivation for water (E) and sucrose (F). In each panel, the latency to initiate feeding/drinking is shown on the left and the amount of food/fluid consumed is shown on the right. Bars represent mean ± s.e.m. and individual values are overlaid. Data were analyzed using one-way ANOVA followed by Tukey’s posthoc comparisons when appropriate, *p < 0.05; **p < 0.01; ***p < 0.001, ****p < 0.0001.

Full size image

Mice were again deprived of food for 24 h and injected with CNO 30 min prior to regaining access to chow. There was a significant main effect of group on latency (F2,22 = 21.61, p < 0.0001) but not consumption (F2,22 = 0.75, p = 0.4859) (Fig. 1C). Re-activating the refeeding ensemble resulted in significantly longer latencies compared to both fasted controls (p = 0.0001) and vehicle controls (p < 0.0001).

When provided with novel Froot Loops after 24 h of food deprivation, there was again a significant main effect of group on latency (F2,21 = 4.82, p = 0.0189) but not consumption (F2,21 = 0.94, p = 0.4080) (Fig. 1D). Activation of the refeeding ensemble resulted in significantly longer latencies compared to both fasted controls (p = 0.0443) and vehicle controls (p = 0.0336).

We next sought to determine the impact of the PSTN refeeding ensemble on fluid consumption. Mice were deprived of water for 4 h at the onset of the dark phase and injected with CNO 30 min prior to regaining access to water. There was a significant main effect of group on latency (F2,21 = 24.86, p < 0.0001) and 2-h consumption (F2,21 = 3.48, p = 0.0497) (Fig. 1E). Interestingly, this is the only situation in which activation of a PSTN subpopulation resulted in a significant decrease in latency compared to vehicle controls. Activation of the refeeding ensemble resulted in significantly shorter latencies compared to vehicle controls (p < 0.0001) and to fasted controls (p = 0.0023), which also had significantly shorter latencies compared to vehicle controls (p = 0.0124). Activation of the refeeding ensemble ultimately led to significantly greater consumption of water compared to vehicle controls (p = 0.0393). This unique effect may reflect prandial thirst, which is normally triggered during food ingestion to anticipate changes in extracellular fluid osmolality following the absorption of solutes into the bloodstream [4].

When provided with a novel 5% sucrose solution after 4 h of water deprivation, there was a significant main effect of group on latency (F2,21 = 10.49, p = 0.0007) but not overall consumption (F2,21 = 2.61, p = 0.0970) (Fig. 1F). Activation of the refeeding ensemble resulted in significantly longer latencies compared to fasted controls (p = 0.0026) and vehicle controls (p = 0.0019).

We conducted a complementary experiment to corroborate these observations and include mice injected with 4-OHT in a sated state as additional control condition (Supplementary Fig. 1, details of statistical analysis are provided in Supplementary Information). In this cohort, mice injected with 4-OHT at the time of refeeding and those injected with 4-OHT in a sated state had significantly more mCherry positive cells than vehicle-injected controls, but the sated ensemble was significantly smaller than the refeeding ensemble (Supplementary Fig. 1B). Chemogenetic activation of the PSTN refeeding ensemble had similar effects as in the former cohort, with significantly longer latencies to eat familiar chow (Supplementary Fig. 1C) and novel Froot Loops (Supplementary Fig. 1D), shorter latency to drink water (Supplementary Fig. 1E), longer latency to drink novel sucrose (Supplementary Fig. 1F), and no difference in the amount of food or fluid consumed (Supplementary Fig. 1C–F) compared to vehicle controls. Furthermore, there was no significant difference between the latencies and intakes of sated controls vs. vehicle controls.

Taken together, these data demonstrate that the PSTN refeeding ensemble is functionally different from both the fasted and sated PSTN ensembles and drastically delays feeding initiation in hungry mice but does not control the amount of food consumed. This conclusion also applies to sucrose ingestion in thirsty mice. Furthermore, while both the PSTN refeeding and fasting ensembles accelerate water drinking, the refeeding ensemble exerts a far more drastic influence. It is possible that the refeeding, fasted, and sated ensembles partially overlap (i.e., a given PSTN cell may participate in two or three of these ensembles), but the refeeding ensemble stands out in its unique capacity to influence behavior. This functional specificity is not related to a threshold number of PSTN cells that need to be activated, as the fasted and refeeding ensembles had similar sizes. We next sought to examine which subpopulation(s) of PSTN neurons might be driving these striking latency effects.

PSTNTac1 neurons suppress both the initiation and the execution of consummatory behaviors

The PSTN contains two non-overlapping populations of neurons expressing either Tac1 or Crh, and both populations are activated by refeeding [14]. PSTNTac1 neurons were previously demonstrated to lower the amount of sucrose consumed by thirsty mice under conditions of novelty (neophobia) or sickness [12]. They also contribute to the reduction in meal frequency induced by anorexigenic hormones and reduce food intake upon chemogenetic or optogenetic stimulation [14]. In contrast, PSTNCrh neurons do not exert such influence [14]. These studies, however, did not examine how manipulating the two populations might alter the delay to engage in consummatory behavior. We used Cre-dependent expression of chemogenetic actuators in the PSTN of Tac1-Cre and Crh-Cre mice to address this question.

Cre-dependent constructs encoding hM4Di, hM3Dq, or mCherry alone were virally transferred into the PSTN of Tac1-Cre mice (Fig. 2A). Mice were first assessed for refeeding with familiar chow following 24 h of food deprivation. There was a significant main effect of vector on both measures (latency: F2,42 = 24.85, p < 0.0001; consumption: F2,42 = 13.39, p < 0.0001) (Fig. 2B). Activation of PSTNTac1 neurons significantly increased the latency to first bite of chow (p < 0.0001), while their inhibition had no significant effect (p = 0.66). PSTNTac1 activation significantly reduced the amount of chow consumed (p = 0.0434). In contrast, PSTNTac1 inhibition significantly increased chow consumption (p = 0.0138). These data indicate that chemogenetic stimulation of PSTNTac1 neurons in food-deprived mice is sufficient to delay and reduce chow consumption despite the motivational state produced by hunger. They also show that the endogenous activity of PSTNTac1 neurons during refeeding suppresses the consumption of familiar chow, thereby opposing the homeostatic drive to feed.

Fig. 2: PSTNTac1 neurons delay and reduce food and fluid consumption independently of caloric content, metabolic state, and novelty.
figure 2

A Experimental design and representative images of mCherry fluorescence in the PSTN illustrating expression of hM3Dq in PSTNTac1 neurons (scale bars: gray, 500 μm; white, 200 μm). Behavioral testing was performed 30 min after injection of the chemogenetic actuator CNO, and 24 h after food deprivation for chow (B) and Froot Loops (C, D, no food deprivation but same circadian timepoint in E), or 4 h after water deprivation for water (F) and sucrose (G, H). Testing was performed upon first-time access to Froot Loops and sucrose (C, G), and again following habituation (D, H). Froot Loops consumption was also tested after habituation to scheduled access with concurrent ad libitum access to chow (E). In each panel, the latency to initiate feeding/drinking is shown on the left and the amount of food/fluid consumed is shown on the right. Bars represent mean ± s.e.m. and individual values are overlaid. Data were analyzed using one-way ANOVA followed by Dunnett’s posthoc comparisons to mCherry controls, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Full size image

In a similar experimental design, mice were provided with a novel and presumably palatable high-sugar food, Froot Loops, after 24 h of food deprivation to assess the sensitivity of PSTNTac1 neurons to other dimensions of consumption behaviors. There was a significant main effect of vector on both measures (latency: F2,42 = 6.46, p = 0.0036; consumption: F2,42 = 9.85, p = 0.0003) (Fig. 2C). Activation did not affect the latency to first bite of Froot Loops (p = 0.9670), nor the amount of Froot Loops consumed over 30 min (p = 0.3267). In contrast, inhibition resulted in significantly shorter latencies (p = 0.0086) and led mice to consume significantly more Froot Loops than controls (p = 0.0082). Interestingly, the results here demonstrate that novel palatable foods likely increase the endogenous activity of PSTNTac1 neurons to its maximal extent given that no difference was observed in response to chemogenetic activation. This idea is bolstered by the ability of inhibition to reduce latencies to first bite to levels similar to familiar chow and to produce overconsumption compared to controls.

To disentangle the features of Froot Loops being simultaneously a novel and palatable food, mice were given both chow and Froot Loops ad libitum for 5 days to dampen novelty. Preference measures for chow versus Froot Loops (no CNO) were collected over a 24-h period during this habituation period and all mice demonstrated highly significant preference for Froot Loops compared to chow (p < 0.0001), thereby confirming the strong palatability of Froot Loops for these mice (Supplementary Fig. 2A). After habituation, mice were again tested with Froot Loops after 24 h of food deprivation. There was a significant main effect of vector on both measures (latency: F2,42 = 16.18, p < 0.0001; consumption: F2,42 = 14.17, p < 0.0001) (Fig. 2D). Activation of PSTNTac1 neurons significantly increased latencies (p = 0.0003) and reduced intake (p = 0.0192) compared to controls. Inhibition did not significantly affect latency (p = 0.3238) but led mice to consume significantly more Froot Loops than controls (p = 0.0219). Strikingly, manipulations of PSTNTac1 neurons while refeeding with habituated Froot Loops yielded effects similar to those observed for familiar chow. In both cases, activation caused a major hesitancy to begin eating and ultimately led to significantly less overall consumption. Inhibition, in contrast, did not increase the speed at which mice took their first bite but did produce overconsumption after 30 min. These data also strengthen the idea that PSTNTac1 neurons may promote hyponeophagia via increased endogenous activity. Altogether, in hungry mice given access to food, the endogenous activity of PSTNTac1 neurons delays feeding initiation selectively under conditions of novelty while it lowers the amount of food consumed regardless of novelty.

In the above-described experiments, food deprivation was used to motivate the mice to readily engage in food consumption. To control for the hunger state as a potential factor underlying our results, consummatory behaviors for habituated Froot Loops were also measured in a sated state, i.e., with ad libitum access to chow. Mice were habituated to receiving scheduled access to Froot Loops for 7–8 days to encourage consumption on test day. Upon CNO administration, there was a significant main effect of vector on both latency (F2,42 = 7.16, p = 0.0022) and consumption (F2,42 = 4.47, p = 0.0174) (Fig. 2E). Similar to hungry mice given access to familiar food, activation led sated mice to increase their latencies (p = 0.0122), while inhibition had no significant effect (p = 0.7091). Furthermore, activation significantly decreased consumption (p = 0.0128), while no difference was observed in response to inhibition (p = 0.7441). This outcome is compelling given the drastic difference in metabolic state of mice deprived of food for 24 h versus those in a sated state. These parallel results highlight that a state of hunger is not necessary for PSTNTac1 neurons to drive feeding suppression, in line with data published by Kim et al. [14].

We next sought to determine the impact of chemogenetic manipulations of PSTNTac1 neurons on fluid consumption. For these assays, mice were deprived of water for 4 h starting at the onset of the dark phase. When access to water was resumed after CNO administration, there was a significant main effect of vector on both measures (latency: F2,41 = 4.65, p = 0.0151; consumption: F2,41 = 6.621, p = 0.0032) (Fig. 2F). Activation did not affect latency to first lick (p = 0.1791) nor the amount of water consumed (p = 0.9990). Inhibition, however, resulted in significantly shorter latencies (p = 0.0078) and significantly higher consumption compared to controls (p = 0.0058). These observations reveal that PSTNTac1 neurons exert potent inhibitory control over water consumption despite water restriction. Strikingly, the effect of inhibiting PSTNTac1 neurons on the latency to rehydrate is comparable to the effect of activating the PSTN refeeding ensemble, which highlights a functional disconnect between these two PSTN subpopulations. This mismatch suggests that only a subset of PSTNTac1 neurons may participate in the PSTN refeeding ensemble, in line with the Fos/Tac1 colocalization analysis conducted by Kim et al. [14], and that PSTNTac1 neurons do not belong to the ensemble driving prandial thirst.

To gain insight into whether PSTNTac1 neurons are similarly sensitive to a novel and palatable fluid, mice were provided with a 5% sucrose solution after 4 h of water deprivation. There was a significant main effect of vector on both measures (latency: F2,40 = 34.58, p < 0.0001; consumption: F2,40 = 20.04, p < 0.0001) (Fig. 2G). Activation led to significantly longer latencies when compared to controls, similar to the effect seen with chow and habituated Froot Loops (p < 0.0001). In contrast, inhibition resulted in significantly shorter latencies, similar to the effect seen with novel Froot Loops (p = 0.0058). After 30 min, activation did not alter consumption (p = 0.9219), while inhibition led mice to consume significantly more sucrose solution compared to controls (p < 0.0001). The results here demonstrate that PSTNTac1 neurons similarly modulate the consumption of both palatable solid and liquid foods.

As was done for Froot Loops, mice were habituated to the sucrose solution ad libitum for 5 days alongside a separate water bottle to disentangle novelty from the palatable nature of sucrose. Preference measures for water versus 5% sucrose (no CNO) were collected over a 72-h period and as expected, all mice demonstrated highly significant preference for sucrose compared to water (p < 0.0001) (Supplementary Fig. 2B). Upon CNO administration, there was a significant main effect of vector on both measures (latency: F2,41 = 6.36, p = 0.0039; consumption: F2,41 = 4.787, p = 0.0135) (Fig. 2H). Parallel to observations with habituated Froot Loops, activation significantly increased latencies (p = 0.0292), while no difference was observed with inhibition (p = 0.5328). Similarly, activation led mice to consume significantly less habituated sucrose (p = 0.0088), while inhibition had no significant effect (p = 0.6415). Building on the body of evidence collected in previous appetitive assays, inhibition of PSTNTac1 neurons has a stronger overconsumption effect on novel palatable substances than when these items are habituated over time, likely indicating higher endogenous activity of PSTNTac1 neurons during hyponeophagia.

PSTNCrh neurons promote the consumption of novel palatable substances

In the PSTN, Tac1- and Crh-expressing neurons represent non-overlapping populations of the PSTN [14]. Accordingly, we tested whether PSTNCrh neurons might exert a differential influence on consummatory behaviors compared to the PSTNTac1 population. A Cre-dependent construct encoding the excitatory designer receptor hM3Dq was virally transferred into the PSTN of Crh-Cre mice (Fig. 3A).

Fig. 3: PSTNCrh neurons delay refeeding of familiar foods and increase consumption of novel palatable substances.
figure 3

A Experimental design and representative images of mCherry fluorescence illustrating expression of hM3Dq in PSTNCrh neurons (scale bars: gray, 500 μm; white, 200 μm). Behavioral testing was performed 30 min after injection of the chemogenetic actuator CNO or vehicle, and 24 h after food deprivation for chow (B) and Froot Loops (C, D), or 4 h after water deprivation for water (E) and sucrose (F, G). Testing was performed upon first-time access to Froot Loops and sucrose (C, F) and again following habituation (D, G). In each panel, the latency to initiate feeding/drinking is shown on the left and the amount of food/fluid consumed is shown on the right. Bars represent mean ± s.e.m. and individual values are overlaid. Data were analyzed using unpaired t tests, *p < 0.05; **p < 0.01.

Full size image

Mice were deprived of food for 24 h and injected with CNO or vehicle 30 min prior to the beginning of experimental sessions. Mice were first assessed for refeeding with familiar chow (Fig. 3B). Activation of PSTNCrh neurons by CNO resulted in significantly longer latencies compared to vehicle (t18 = 3.15, p = 0.0055). The amount of chow consumed in 30 min, however, was indistinguishable between treatments (t18 = 0.12, p = 0.9101). The results demonstrate that PSTNCrh neurons can drive hesitancy to first bite of chow, but do not influence overall consumption.

Mice were provided with novel Froot Loops after 24 h of food deprivation to assess the influence of PSTNCrh neurons on other dimensions of appetitive behaviors. Activation of PSTNCrh neurons did not affect the latency to first bite (t16 = 0.05, p = 0.9627) but significantly increased Froot Loops consumption (t16 = 3.39, p = 0.0037) (Fig. 3C). These results demonstrate that activation PSTNCrh neurons leads to overconsumption of a novel, palatable food, i.e., similar to the effect of inhibiting PSTNTac1 neurons. Altogether, these observations suggest that the two populations oppose each other’s function in situations of hyponeophagia.

To disentangle the novelty of Froot Loops from their palatability, mice were given both chow and Froot Loops ad libitum for 5 days. After habituation, mice were again tested for Froot Loops after 24 h of food deprivation. As was observed with chow, activation resulted in significantly longer latencies (t15 = 3.37, p = 0.0042), with no significant effect on consumption (t16 = 0.95, p = 0.3574) (Fig. 3D).

We next sought to determine the impact of chemogenetic excitation of PSTNCrh neurons on liquid consumption after 4 h of water deprivation. When access to water was resumed after CNO or vehicle administration, there was no significant effect of treatment on latency (t16 = 0.72, p = 0.4850) or consumption (t16 = 0.65, p = 0.5231) (Fig. 3E).

When provided with a novel, palatable 5% sucrose solution, there was no significant effect of treatment on latency (t16 = 1.54, p = 0.1432) but CNO increased sucrose consumption compared to vehicle (t16 = 3.15, p = 0.0062) (Fig. 3F). These results are congruent with the first-time Froot Loops assay, such that excitation of PSTNCrh neurons significantly increase the consumption of novel, palatable solid and liquid substances without affecting the motivation to begin consuming.

Mice were habituated to the sucrose solution ad libitum for 5 days to disentangle novelty from the palatable nature of sucrose. There was no significant main effect of treatment on latency (t16 = 0.15, p = 0.8853) or consumption (t16 = 1.26, p = 0.2276) (Fig. 3G). Altogether, activation of PSTNCrh neurons did not affect the consumption of familiar chow, habituated Froot Loops, or habituated sucrose and selectively increased first-time consumption of Froot Loops and sucrose. These results underscore the sufficiency of PSTNCrh neurons to promote the consumption of novel palatable substances without altering the latency to initiate their ingestion.

PSTN neurons projecting to the CeA suppress feeding initiation

A dual-vector, pathway-specific chemogenetic approach was employed to determine the contribution of PSTN projection neurons. Previous neuroanatomic tracing demonstrated prominent projections from the PSTN to the CeA, BNST, and PVT [7, 14, 15], three structures that are involved in feeding regulation [30,31,32,33]. While PSTNTac1 neurons send inputs to all three areas, food consumption is reduced by photoactivation of CeA-projecting and PVT-projecting, but not BNST-projecting, PSTNTac1 neurons, highlighting the differential role of these pathways [14]. Furthermore, stimulation of PVT-projecting PSTNVGluT2 neurons reduces food intake [17]. We therefore sought to assess the influence of CeA-, BNST-, and PVT-projecting PSTN neurons in the control of feeding initiation. In C57BL/6 J mice, a retrograde Cre-expressing virus was injected into the CeA, BNST or PVT permitting the Cre-dependent expression of hM3Dq and KORD viral constructs co-injected into the PSTN to excite or inhibit those projections neurons upon administration of CNO or SalB, respectively (Fig. 4A).

Fig. 4: CeA- but not BNST- nor PVT-projecting PSTN neurons mediate refeeding delay.
figure 4

A Experimental design and representative images of GFP fluorescence illustrating Cre expression and mCherry fluorescence illustrating hM3Dq expression in the PSTN (scale bars: gray, 500 μm; white, 200 μm). Cre was expressed from a retrograde vector injected either in the CeA (B, C), BNST (D, E), or PVT (F, G). Behavioral testing was performed 30 min after injection of the chemogenetic actuators CNO (B, D, F) or SalB (C, E, G), or their respective vehicles, 24 h after food deprivation. Chow was presented either in the home cage (B, D, F) or in a novel arena (C, E, G). In each panel, the latency to initiate feeding/drinking is shown on the left and the amount of food/fluid consumed is shown on the right. Bars represent mean ± s.e.m. and individual values are overlaid. Data were analyzed using paired t tests, *p < 0.05; **p < 0.01; ***p < 0.001.

Full size image

Mice were deprived of food for 24 h and injected with chemogenetic ligands (CNO, SalB, or their respective vehicles) 30 min prior to refeeding with familiar chow. SalB testing was conducted in an open arena to increase feeding latency and facilitate the detection of downward shifts. PSTNCeA activation significantly increased latency (t5 = 4.33, p = 0.0075) but did not alter consumption (t5 = 0.19, p = 0.8530) (Fig. 4B). Conversely, PSTNCeA inhibition significantly shortened latency (t5 = 3.97, p = 0.0106) with no effect on consumption (t5 = 0.90, p = 0.4118) (Fig. 4C). PSTNBNST activation also increased latency (t6 = 8.22, p = 0.0002) without altering consumption (t6 = 0.50, p = 0.6313) (Fig. 4D). PSTNBNST inhibition had no significant effect on either latency (t6 = 1.71, p = 0.1365) or consumption (t6 = 1.27, p = 0.2496) (Fig. 4E). PSTNPVT activation similarly increased latency (t5 = 3.35, p = 0.0202) but did not alter consumption (t5 = 0.23, p = 0.8241) (Fig. 4F). PSTNPVT inhibition had no significant effect on either latency (t5 = 1.2, p = 0.2838) or consumption (t5 = 1.27, p = 0.2597) (Fig. 4G). Altogether, these data point to a prominent role of the endogenous activity of PSTNCeA neurons in delaying food consumption initiation in hungry mice.

CeA-projecting PSTNTac1 neurons delay refeeding

Given that both PSTNTac1 and PSTNCrh neurons can delay feeding initiation in food-deprived mice given access to chow (Figs. 2B and 3B) and that they both project to the CeA [12, 14], either cell type may contribute to the influence of the PSTNCeA projection on refeeding latency. To explore this possibility, we used a dual-vector, intersectional strategy to drive the expression of hM3Dq in PSTNTac1 and PSTNCrh neurons projecting to the CeA. This selective targeting was driven by the expression of Cre in Tac1-Cre or Crh-Cre mice respectively, the injection of a retrograde Flp-encoding vector in the CeA, and the injection of an INTRSECT (INTronic Recombinase Sites Enabling Combinatorial Targeting [29]) vector encoding hM3Dq-mCherry upon Cre AND Flp recombination in the PSTN (Fig. 5A).

Fig. 5: CeA-projecting PSTNTac1 neurons delay refeeding while CeA-projecting PSTNCrh neurons hasten refeeding.
figure 5

A Experimental design and representative images of mCherry fluorescence illustrating hM3Dq expression in CeA-projecting PSTNTac1 (B, C) and PSTNCrh (D, E) neurons (scale bars: gray, 500 μm; white, 200 μm). Behavioral testing was performed 30 min after injection of CNO or vehicle, and 24 h after food deprivation (B, D) or 4 h after water deprivation (C, E). In each panel, the latency to initiate feeding/drinking is shown on the left and the amount of food/fluid consumed is shown on the right. Bars represent mean ± s.e.m. and individual values are overlaid. Data were analyzed using paired t tests, *p < 0.05; ***p < 0.001.

Full size image

Following 24-h food deprivation, activating CeA-projecting PSTNTac1 neurons significantly increased the latency to initiate familiar chow consumption (t5 = 9.32, p = 0.0002) but did not affect the amount consumed (t5 = 0.434, p = 0.6822) (Fig. 5B). These results further underscore the importance of this molecularly defined projection given that activation of PSTNTac1 neurons (Fig. 2B) and CeA-projecting neurons (Fig. 4B), in isolation or combined, significantly increases latency to first bite of familiar chow. Total consumption of chow was not affected by the activation of CeA-projecting PSTNTac1 neurons, in contrast to the significant decrease observed with general PSTNTac1 activation (Fig. 2B), but consistent with the outcome of activating CeA projectors as a whole (Fig. 4B). Most relevant to our goal of characterizing the neuronal subpopulations activated by food access in hungry mice, this pattern is identical to the effect of re-activating the PSTN refeeding ensemble (Fig. 1C and Supplementary Fig. 1C).

We then determined the effect of CeA-projecting PSTNTac1 activation on water drinking. When access to water was resumed after CNO or vehicle administration, there was no significant effect of treatment on latency (t5 = 1.82, p = 0.1270) or consumption of water (t5 = 1.32, p = 0.2431) (Fig. 5C), indicating that these neurons do not contribute to the PSTN subpopulation triggering prandial thirst.

Activating CeA-projecting PSTNCrh neurons after 24-h food deprivation significantly reduced the latency to initiate familiar chow consumption (t6 = 2.94, p = 0.0258) without affecting the amount consumed (t6 = 1.308, p = 0.2388) (Fig. 5D). CeA-projecting PSTNCrh thereby stands out as the only population in our study sufficient to accelerate refeeding, which underscores not only an opposing role of Tac1 and Crh subsets within the PSTNCeA projection, but also the unique influence of CeA-projecting Crh cells among all PSTNCrh neurons (Fig. 3B).

Activating CeA-projecting PSTNCrh neurons after 4 h of water deprivation had no significant effect on the latency to initiate drinking (t6 = 1.94, p = 0.1011) or the amount of water consumed (t6 = 0.914, p = 0.3959) (Fig. 5E), similar to the results observed with general PSTNCrh activation (Fig. 3E).

BNST-projecting PSTNCrh neurons hasten water drinking

Given that both PSTNTac1 and PSTNCrh neurons project to the BNST [14] and that general activation of the PSTNBNST pathway robustly delays feeding initiation in food-deprived mice given access to chow (Fig. 4D), we used the same dual-vector, intersectional strategy to drive the expression of hM3Dq in PSTNTac1 and PSTNCrh neurons projecting to the BNST and tease apart their respective roles on feeding and drinking behavior (Fig. 6A).

Fig. 6: BNST-projecting PSTNCrh neurons hasten water drinking.
figure 6

A Experimental design and representative images of mCherry fluorescence illustrating hM3Dq expression in BNST-projecting PSTNTac1 (B, C) and PSTNCrh (D, E) neurons (scale bars: gray, 500 μm; white, 200 μm). Behavioral testing was performed 30 min after injection of CNO or vehicle, and 24 h after food deprivation (B, D) or 4 h after water deprivation (C, E). In each panel, the latency to initiate feeding/drinking is shown on the left and the amount of food/fluid consumed is shown on the right. Bars represent mean ± s.e.m. and individual values are overlaid. Data were analyzed using paired t tests ***p < 0.001.

Full size image

Following 24-h food deprivation, activating BNST-projecting PSTNTac1 did not impact the latency to initiate familiar chow consumption (t9 = 0.66, p = 0.5247) or the amount consumed (t9 = 1.46, p = 0.1775) (Fig. 6B). These results demonstrate that BNST-projecting PSTNTac1 neurons do not contribute to the robust latency increase triggered by general PSTNTac1 activation (Fig. 2B), unlike their CeA-projecting counterparts (Fig. 5B).

We then determined the effect of BNST-projecting PSTNTac1 activation on water drinking. When access to water was resumed after CNO or vehicle administration, there was no significant effect of treatment on latency (t9 = 0.02, p = 0.9868) or consumption of water (t9 = 0.20, p = 0.8496) (Fig. 6C).

Likewise, activating BNST-projecting PSTNCrh neurons after 24-h food deprivation did not alter the latency to initiate familiar chow consumption (t9 = 1.65, p = 0.1325) or the amount consumed (t9 = 0.51, p = 0.6218) (Fig. 6D). These results show that neither BNST-projecting PSTNTac1 nor PSTNCrh are sufficient to drive the increase in latency observed following general PSTNBNST activation (Fig. 4D), suggesting a role for yet another subpopulation of BNST-projecting PSTN neurons.

In contrast, activating BNST-projecting PSTNCrh neurons after 4 h of water deprivation significantly shortened the latency to initiate drinking (t9 = 5.17, p = 0.0006) without altering the amount of water consumed (t9 = 0.25, p = 0.8056) (Fig. 6E), similar to the pattern evoked by the PSTN refeeding ensemble (Fig. 1E and Supplementary Fig. 1E). These results support the contribution of BNST-projecting PSTNCrh neurons to the PSTN subpopulation triggering prandial thirst, and further emphasize functional heterogeneity with PSTN cell types as this effect was not captured by activating PSTNCrh neurons as a whole.

Discussion

The results of this study demonstrate that the PSTN exerts a profound effect on the motivation to initiate feeding and drinking. The PSTN is known to become active in response to diverse anorexigenic signals including binge-like refeeding following food deprivation, sickness, nutrient-deficient diets, novelty, and homeostatic hormonal signaling [8,9,10,11,12,13,14]. Our study highlights that the functional significance of this activation is to suppress the initiation of food consumption, an influence that extends to palatable solids and calorie-containing liquids, with no impact on the amount of food consumed (Fig. 1). In parallel, this ensemble promotes the initiation of water intake. We further find that these refeeding-sensitive cells are phenotypically similar to subpopulations of PSTNTac1 neurons (Fig. 2) but distinct from PSTNCrh neurons activated as a whole (Fig. 3). Furthermore, we show that PSTNCeA neurons activated by food access in a hungry state play a prominent role in suppressing feeding initiation (Fig. 4). Specifically activating CeA-projecting PSTNTac1 neurons was sufficient to delay refeeding, while activating CeA-projecting PSTNCrh neurons hastened refeeding, highlighting the functional heterogeneity of PSTN subpopulations within a given anatomical pathway (Fig. 5). In contrast, activating BNST-projecting PSTNCrh neurons exclusively hastened water drinking without impacting the latency to feed (Fig. 6).

Compared to previous studies that have investigated the influence of PSTN cell types and projections on consummatory behaviors [12, 14, 17,18,19,20], these results are the first to demonstrate a role of PSTN neurons in controlling the latency to initiate feeding or drinking, and the unique effects we observed on water drinking initiation are entirely novel to the field. Our study also disentangled the roles of palatability and novelty in the influence of PSTNTac1 neurons on reinforcer consumption, with similar effects seen with familiar chow vs. Froot Loops or sucrose following habituation to these novel, palatable substances.

Strikingly, several of our experiments detected significant changes in latencies to first bite or first lick in the absence of changes in the amount of food or fluid consumed. This disconnect may reflect diverging biological underpinnings for appetitive motivation vs. real-time calorie or osmolarity homeostasis, or it could be related to the limited temporal resolution of our assays, which may have failed to detect differences in front loading behaviors. The latter limitation may explain why stimulating the PSTNPVT projection did not alter food intake in our study, while it did in a previous study that used real-time measurement of food consumption in Tac1-Cre mice [14]. Another possible explanation for this discrepancy is that the influence of the PSTNPVT pathway on food intake may be limited to sweet foods [14, 17].

Altogether, our data suggests that CeA-projecting PSTNTac1 neurons represent a major component of the PSTN ensemble that drives hesitancy in hungry animals given access to food. By enhancing our understanding of the brain mechanisms controlling feeding onset, independently of consumption per se, this finding may have therapeutic relevance for the treatment of eating disorders. Activating CeA-projecting PSTNTac1 neurons could help alleviate the craving component of bulimia nervosa and binge-eating disorders, while inhibiting this circuit may loosen pathological self-control in restrictive anorexia nervosa – and neither of these manipulations would be expected to negatively affect homeostatic food consumption. Future studies will aim to elucidate the molecular signaling events underlying the suppression of feeding initiation by CeA-projecting PSTNTac1 neurons to afford pharmacological access to this circuit.

While PSTN neurons activated upon refeeding of hungry animals with familiar chow did not impact the amount of food consumed, our results from chemogenetic inhibition show that the endogenous activity of PSTNTac1 neurons exerts a strong influence on consumption when mice are given access to a novel palatable food or fluid, consistent with the effect reported by Barbier et al. in mice given access to sucrose for the first time [12]. We found that this influence is reduced by habituation, which is consistent with reduced c-Fos induction in the PSTN of rats habituated to sucrose consumption [12], and further blunted by removing the state of metabolic need triggered by food deprivation. As expected, reducing the endogenous tone of PSTNTac1 activity by using a familiar food or fluid was necessary for chemogenetic stimulation to significantly reduce consumption, in line with the effect of optogenetic stimulation of PSTNTac1 neurons in ad libitum fed mice [14]. Interestingly, the endogenous activity of PSTNTac1 neurons also suppressed water drinking, indicating that the influence of this subpopulation on consummatory behavior is independent of caloric content. This observation contrasts starkly with the influence of the PSTN refeeding ensemble on water intake (discussed in more detail below), indicating that these two PSTN subpopulations only have a partial overlap.

Another important outcome of our study was the opposing influence of PSTNTac1 vs. PSTNCrh neurons on palatable food consumption, which was suppressed by the former but increased by the latter. While previous studies had already identified an important role of PSTNTac1 neurons in feeding suppression [12, 14], less is known regarding the physiological role of PSTNCrh neurons [7]. PSTNCrh neurons are activated by refeeding [14] and Crh expression in the PSTN is upregulated by an anorexia-inducing valine-deficient diet [9], but they do not influence food consumption ([14] and results reported here). Aside from consummatory behaviors, PSTNCrh neurons have been shown to mediate defensive responses to acute predator threats and to regulate REM sleep [34]. Our observation that their activation promotes the consumption of sweet food and liquids suggests an important role in the control of hedonic feeding. Future studies will be needed to determine if this influence may extend to the consumption of other rewarding substances (e.g., addictive drugs) or experiences (e.g., social interaction).

Our results also uncover an unexpected implication of the PSTN in water drinking initiation. Re-activating the PSTN refeeding ensemble accelerated the onset of water consumption, and this phenotype was replicated when specifically activating BNST-projecting PSTNCrh neurons. These data raise the possibility that BNST-projecting PSTNCrh neurons may be connected to circuits controlling prandial thirst, such as subfornical organ Nos1 neurons or posterior pituitary-projecting, vasopressin-secreting neurons [4, 35,36,37]. Our finding aligns with the recent discovery of a pathway-specific influence of insular cortex layer 5 projection neurons on water intake, whereby the subpopulation projecting to the PSTN promotes water licking in thirsty mice [38]. Interestingly, the shortening of drinking latency driven by the PSTN refeeding ensemble was selective for water, as an opposite effect was observed when mice were given access to a sucrose solution. Since latency modulation occurs before the mice get to sample the fluid’s taste, the differential effects of the PSTN refeeding ensemble on water vs. sucrose drinking initiation probably involve olfactory cues that enable the mice to predict the composition of the available solution [39].

While our study sought to both highlight and disentangle the heterogenous make-up of the PSTN, it is important to note that other cell types beyond Tac1 and Crh neurons are found within the PSTN and are likely to influence the feeding and drinking behaviors reported here. A previous study determined that ~70% of PSTNTac1 neurons and ~80% of PSTNCrh neurons express Fos following refeeding [14], which justified focusing on these two populations in the context of the present study, but it also identified Fos in PSTN cells that express neither Tac1 nor Crh (their proportion was not quantified). For instance, PSTNAdcyap1 neurons only partially overlap with the Tac1 (~40%) and Crh (~20%) populations and may contribute to the PSTN refeeding ensemble given their known influence on feeding behavior [18]. Likewise, we chose to probe projections to the CeA, BNST, and PVT based on the established role of these brain areas in feeding control [30,31,32,33], but future studies will be needed to explore the role of PSTN projections to cortical, midbrain, and brainstem structures [14, 15].

The intersectional strategy we used for pathway-specific manipulations does not preclude that behavioral effects might be driven by collateral projections to other brain targets. Consistent with this possibility, there was a partial overlap in the phenotypes induced by the global manipulation of PSTN efferents to the CeA, BNST, and PVT (Fig. 4). However, there was a stark dichotomy in the outcome of stimulating cell-type and pathway-specific subpopulations (Fig. 5 vs. Fig. 6), which demonstrates the functional independence of these different circuits and invalidates a potential role of collaterals. Furthermore, while we were able to visualize mCherry-immunolabeled fibers in the targeted projection area (i.e., where the retrograde vector was injected), such labeling was not detected in other brain regions.

Another element that should be considered for the interpretation of our data relates to the cell-type specificity of Cre activity in the different mouse lines. We confirmed that there is minimal Cre-driven recombination in TRAP2 mice in the absence of 4-OHT (Fig. 1 and Supplementary Fig. 1). However, the size of the ensembles captured upon 4-OHT injection in sated mice and fasted, non-refed mice was larger than anticipated based on published cFos+ cell counting studies [10, 13, 14]. This discrepancy may result from the time window of Cre activity being longer than that of cFos protein expression [21] and/or from higher baseline PSTN activity during the dark phase [8]. Importantly, despite their sizes, the PSTN ensembles captured in sated and fasted mice did not trigger the same behavioral phenotypes as in refed mice, further demonstrating the functional specificity of the latter ensemble.

The fidelity and penetrance of Cre activity in Tac1-Cre and Crh-Cre lines can be visualized by double in situ hybridization in the offspring of crosses with the Cre reporter line Ai14 [40], as reported by the Allen Mouse Brain Connectivity Atlas (https://connectivity.brain-map.org/transgenic/, [41]). Overall, tdTomato mRNA extensively co-localizes with Tac1 and Crh in adult PSTN cells of Tac1-Cre;Ai14 mice (experiment 180301447, images 32–34, and experiment 267222532, images 29–30) and Crh-Cre;Ai14 mice (experiment 182530933, image 12), respectively. We would like to note that experiment 156348221 has high and uneven background in the tdTomato channel, which had previously led us to conclude about poor fidelity of Cre activity in Crh-Cre mice [7]. In Tac1-Cre;Ai14 brains, there are tdTomato+ cells in the subthalamic nucleus, which is generally considered devoid of Tac1 expression [16], but the tdTomato signal does co-localize with faint Tac1 signal, consistent with very low levels of Cre being sufficient to induce reporter expression. In Crh-Cre;Ai14 PSTN, the tdTomato signal is strikingly bimodal, as it is intense and filling in some cells vs. faint and punctate in other cells. Overall, when including the cells with low, punctate tdTomato signal, there is excellent overlap between Crh and tdTomato within the PSTN for both fidelity and penetrance. However, there are Crh-negative cells with strong tdTomato labeling outside of the PSTN cluster, which emphasizes the importance of viral vector transduction anatomical accuracy. As expected, the regional selectivity of reporter expression (PSTN vs. adjacent brain areas) was higher in intersectional strategies (Figs. 4–6) than in mice injected with a single vector in the PSTN (Figs. 1–3).

In conclusion, our study identifies a novel circuit that suppresses feeding initiation despite hunger, which emphasizes the relevance of the PSTN as a key brain site in the control of appetite and food rejection [19] and has important implications for our understanding of the neural mechanisms that may be disrupted in eating disorders. It also identifies a novel circuit that promotes water drinking initiation, implicating for the first time the PSTN in the brain network driving prandial thirst.