Individual paraventricular hypothalamic GLP-1R neurons track ingestion across energy states

Summary

Energy homeostasis is tightly regulated to ensure metabolic health in uncertain environments. Neurons expressing the glucagon-like peptide-1 receptor (GLP-1R) contribute to maintaining body weight, but how their activity is regulated according to energy state remains unclear. Here, we investigated the function and cellular dynamics of GLP-1R-expressing neurons of the paraventricular hypothalamus (PVH^GLP-1R) during food consumption, providing strong evidence of their ability to bidirectionally control feeding. Elevating intracellular signaling through 3′,5′-cyclic adenosine monophosphate (cAMP), the major pathway downstream of GLP-1R, suppressed feeding. Using two-photon calcium imaging, we observed heterogeneous single-cell responses to ingestion that occurred regardless of energy state or tastant identity. Longitudinal tracking of individual PVH^GLP-1R neurons revealed dynamic shifts in ingestion-responsive activity between satiety states while the overall population activity remained stable. These findings reveal stable yet dynamic changes in network activity patterns between energy states, highlighting avenues for further investigation of the effects of obesity and GLP-1R agonists.

Graphical abstract

Highlights

• •Increasing cAMP in PVH^GLP-1R neurons drives burst firing and suppresses feeding
• •Energy state dictates individual PVH^GLP-1R neuron responsivity to ingestion
• •Average PVH^GLP-1R neuron population activity appears stable across energy states
• •The distribution of PVH^GLP-1R neuron responses appears stable across energy states

Physiology; behavioral neuroscience; cellular neuroscience

Introduction

Animals adapt their behavior in response to fluctuating food availability to meet energy needs. Changes in energy states or circulating hormonal signals are accompanied by altered neuronal function across the brain.^1,2 The ability of neural circuits to respond to changes in energy state allows for adjusting feeding behavior and energy homeostasis. However, where and when such adaptations occur within the brain are not yet fully established.

The paraventricular nucleus of the hypothalamus (PVH) regulates satiety and contributes to autonomic and neuroendocrine functions.^3,4,5 The PVH contains several molecularly distinct neuron populations that respond differentially during feeding and encode relevant behavioral states.^6,7,8 Among these, PVH neurons expressing the glucagon-like peptide-1 receptor (GLP-1R; PVH^GLP-1R neurons) are vital to energy homeostasis.^7,9,10 Although these neurons appear dispensable for the appetite-suppressing effects of GLP-1R agonists,^7,9,11 PVH^GLP-1R neurons, and GLP-1 signaling within the PVH are important for influencing feeding behavior and body weight control. Nevertheless, how PVH^GLP-1R neurons exert their effects over food intake behavior remains an active area of investigation. In vivo fiber photometry recordings have demonstrated that food-related sensory cues rapidly activate PVH^GLP-1R neurons in food-deprived mice.^7,9 However, population dynamics recorded with calcium signal fiber photometry do not have the cellular resolution to provide information about response heterogeneity.

Here, we investigated the relationship between PVH^GLP-1R neuron activity and feeding behavior. We demonstrated that these neurons bidirectionally control feeding. Transient suppression of neuronal firing enhanced feeding in ad libitum fed mice. PVH^GLP-1R neurons respond to transient accumulation of cAMP, the main mediator of GLP-1R signaling,¹² with high-frequency firing and suppression of feeding. We then used multiphoton microscopy to longitudinally track the calcium dynamics of individual PVH^GLP-1R neurons during ingestion across energy states. We observed phasic calcium responses of individual PVH^GLP-1R neurons time-locked to ingestion that did not vary across tastant identity. Finally, we show that although average population activity was similar across energy states, responses of individual PVH^GLP-1R neurons varied such that distinct ensembles responded to specific energy states.

Results

Photoinhibition of PVH^GLP-1R neurons induces feeding

Based on observations that food-related cues rapidly activate PVH^GLP-1R neurons,^7,9 we tested whether optogenetic inactivation is sufficient to induce food intake. We injected AAV-SIO-stGtACR2-FusionRed, expressing the light-activated chloride conducting channel, into the PVH of Glp1r-ires-Cre mice (Figures 1A and 1B). As expected, blue light stimulation suppressed firing in stGtACR2-positive PVH^GLP-1R neurons induced by current injection (Figures 1C and 1D). We then expressed stGtACR2-FusionRed or tdTomato in PVH^GLP-1R neurons and implanted an optic fiber above the PVH to allow for photoinhibition during a free-feeding assay (Figures 1E and S1). In ad libitum fed mice, silencing PVH^GLP-1R neurons induced food intake within minutes (Figure 1F). Photoinhibition of PVH^GLP-1R neurons had no effect on real-time place preference (RTPP), suggesting that these neurons are unlikely to be involved in appetitive reinforcement (Figures 1G and 1H). Taken together, these results confirm that PVH^GLP-1R neurons rapidly regulate feeding.^7,10

Figure 1.

Inhibition of PVH^GLP-1R neurons acutely evokes feeding

(A) Experimental strategy for expression of the GtACR-FusionRed construct and implantation of optic fiber above PVH in Glp1r-ires-Cre male mice.

(B) Representative image of GtACR expression and optic fiber implant in the PVH. Scale bars, 100 μm.

(C) Representative trace of action potential (AP) firing in GtACR-expressing PVH^GLP-1R neurons at varying levels of current injection with and without blue light delivery.

(D) Quantification of AP firing in (C)—n = 4 cells, two-way ANOVA; effect of current step: F(1.52, 18.25) = 3.282, p = 0.072; effect of light: F(3, 12) = 7.55, p = 0.004; effect of subject: F(12, 192) = 28.60, p < 0.0001; interaction: F(48, 192) = 3.282, p < 0.0001.

(E) Schematic of strategy to inhibit PVH^GLP-1R neurons during feeding.

(F) Food intake in ad libitum fed mice before, during, and after 5 min of photoinhibition—n = 6 control/5 GtACR, two-way ANOVA with Šídák’s multiple comparisons; effect of time: F(1.699, 15.29) = 5.598, p = 0.0183; effect of group: F(1, 9) = 8.440, p = 0.0183; interaction: F(2, 18) = 3.014, p = 0.0743; subject: F(9, 18) = 0.4, p = 0.919.

(G) Representative heatmaps of mouse position frequency in the RTPP assay. Blue bar indicates side of arena paired with light delivery.

(H) Preference for light-paired side (n = 6 control/5 GtACR, unpaired two-tailed t test, t(9) = 0.2714, p = 0.7922). ∗p < 0.05, ∗∗p < 0.01. Data represent mean ± SEM.

Elevating cAMP signaling in PVH^GLP-1R neurons suppresses appetite

While acute activation of PVH^GLP-1R neurons or GLP-1 release in the PVH has been shown to suppress feeding,^7,9,10 it is unclear whether intracellular signaling pathways downstream of GLP-1R can similarly induce rapid anorectic effects. As cAMP is the major second messenger downstream of GLP-1R signaling,^12,13 we tested whether elevating cAMP in PVH^GLP-1R neurons could acutely affect feeding. We therefore expressed the photoactivatable adenylyl cyclase biPAC¹⁴ in PVH^GLP-1R neurons (Figures 2A and 2B). We first validated photoactivation of AAV-DIO-mKate2-biPAC in PVH^GLP-1R neurons ex vivo (Figure 2C). A 2-s-long pulse of blue light increased neuronal firing rate during a ramping current injection (Figure 2D). Additionally, spontaneous firing increased significantly for several seconds following biPAC photoactivation (Figure 2E). Thus, we predicted that biPAC stimulation would suppress feeding in vivo. We expressed mKate2-biPAC or tdTomato in PVH^GLP-1R neurons and implanted an optic fiber above the PVH to deliver pulsed light stimulation during a post-fast refeeding paradigm (Figures 2F and S1). Food intake was reduced during the stimulation period and returned to control levels after cessation (Figure 2G). Moreover, photoactivation of biPAC in PVH^GLP-1R neurons did not influence RTPP (Figures 2H and 2I). Collectively, these results indicate that cAMP signaling in PVH^GLP-1R neurons is sufficient to acutely suppress feeding within minutes, potentially mediating the anorectic effects of GLP-1 signaling in the PVH.¹⁵

Figure 2.

Intracellular cAMP signaling excites PVH^GLP-1R neurons and suppresses feeding

(A) Experimental strategy for expression of the biPAC-mKate2 construct and implantation of optic fiber above PVH in Glp1r-ires-Cre male mice.

(B) Representative image of biPAC expression and optic fiber implant in the PVH. Scale bars, 100 μm.

(C) Representative trace of AP firing in biPAC-expressing PVH^GLP-1R neuron. Ramping current injection was given before and after blue light delivery.

(D) Quantification of AP firing during each current injection with and without light—n = 6 cells, two-way ANOVA with Šídák’s multiple comparisons; effect of ramp: F(1, 10) = 5.084, p = 0.0203; effect of light: F(1, 10) = 1.881, p = 0.2003; interaction: F(1, 10) = 7.594, p = 0.0203; effect of subject: F(10, 10) = 4.543, p = 0.0126.

(E) Quantification of spontaneous AP firing before or at certain times after light delivery—n = 5 cells, repeated measures one-way ANOVA, treatment F(1, 4) = 8.062, p = 0.0469; individual F(4, 8) = 1; p = 0.4609.

(F) Schematic of strategy to elevate intracellular cAMP in PVH^GLP-1R neurons during feeding.

(G) Food intake in food-deprived mice before, during, and after 5 min of biPAC stimulation—n = 6 control/5 biPAC, two-way ANOVA with Šídák’s multiple comparisons, effect of time: F(1.803, 16.23) = 2.121, p = 0.1549; effect of group: F(1, 9) = 0.1198, p = 0.7372; interaction: F(2, 18) = 8.292, p = 0.0028; effect of subject: F(9, 18) = 5.024, p = 0.0018.

(H) Representative heatmaps of mouse position frequency in the RTPP assay. Blue bar indicates side of arena paired with light delivery.

(I) Preference for light-paired side—n = 6 control/5 biPAC, unpaired two-tailed t test, t(9) = 0.7781, p = 0.4565. ∗p < 0.05. Data represent mean ± SEM.

Distinct subsets of PVH^GLP-1R neurons track ingestion across energy states

Given that PVH^GLP-1R neurons can bidirectionally control feeding (Figures 1 and 2), we next investigated their in vivo calcium dynamics during feeding across energy states. To do so, we recorded neural activity with two-photon calcium imaging in head-fixed mice while randomly delivering small volumes (3 μL) of 10% sucrose solution. This approach distinguished responses to ingestion from food-related cues to which PVH^GLP-1R neurons also respond^7,9 by minimizing the influence of olfactory, visual, or auditory cues. We expressed the calcium indicator GCaMP8m in PVH^GLP-1R neurons and implanted a gradient refraction index (GRIN) lens above the PVH (Figures 3A–3C). Mice readily learned to lick for sucrose within days, and individual PVH^GLP-1R neurons showed transient activation or inhibition time-locked to consummatory licking (Figures 3D–3F).

Figure 3.

Distinct subsets of PVH^GLP-1R neurons transiently respond to ingestion in fed and fasted mice

(A) Experimental strategy for expression of the GCaMP8m construct and implantation of GRIN lens above PVH in Glp1r-ires-Cre male mice.

(B) Top: Representative histology of GCaMP expression and lens implant in the PVH. Scale bars, 250 μm. Bottom: mean intensity projection of imaged PVH^GLP-1R neurons. Right: neurons segmented from corresponding calcium imaging experiment.

(C) Representative examples of dF/F from neurons labeled in (B) (multiple trials stitched together).

(D) Schematic for trial-by-trial imaging of PVH^GLP-1R neuronal activity during sucrose consumption in head-fixed male mice.

(E) Representative lick raster during experiment shown in (B).

(F) Representative heatmap of trial-by-trial activity aligned to lick initiation (dotted line) in neurons labeled in (B), with histogram below corresponding to average lick rate of raster shown in (E).

(G) Representative mean images of PVH^GLP-1R neurons in fasted (top) and fed (bottom). Arrow points to example neuron identified in both experiments.

(H) Trial-by-trial activity (left) and trial-averaged activity (right) of neuron in (A).

(I) Representative mean images as in (A) (different mouse).

(J) Trial-by-trial activity (left) and trial-averaged activity (right) of neuron in (C).

(K) Heatmap of trial-averaged dF/F aligned to lick initiation in fasted (left) and fed (mice), with rows in each corresponding to the same cell (n = 100 cells).

(L) Average change in activity for each mouse (neuron-averaged) when cells are sorted independently (“sorted” or control) or tracked across experiments—n = 5 mice, two-tailed paired t test, t(4) = 3.730, p = 0.0203.

(M) Accuracy of decoding energy state based on the trial-by-trial activity of individual neurons identified in both experiments (n = 100 cells, Mann-Whitney test, U = 1724, p < 0.0001).

(N) Heatmap of trial-averaged dF/F of activated PVH^GLP-1R neurons in fed mice (left) and their corresponding activity when fasted (same neuron on same row).

(O) Neuron-averaged activity of data shown in (H).

(P) Quantification of response shown in (I) (n = 22 cells, Wilcoxon test, W = 227, p < 0.0001).

(Q) Heatmap of trial-averaged dF/F of inhibited PVH^GLP-1R neurons in fed mice and their corresponding activity when fasted (same neuron on same row).

(R) Neuron-averaged activity of data shown in (K).

(S) Quantification of response shown in (L)—n = 6 cells, two-tailed paired t test, t(5) = 4.915, p = 0.0044. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗∗p < 0.0001. Data represent mean ± SEM.

We next examined how varying energy state affects these responses at the single-cell level. Individual PVH^GLP-1R neurons were tracked longitudinally to determine whether representations of ingestion are stable across energy states. We identified neurons that displayed changes in response magnitude or sign when satiety state changed (Figures 3G–3J). This observation was consistent across the entire population (Figure 3K). To quantify this change in cellular representation across mice, we defined a metric (change in responsivity; see STAR Methods) to compare the response following ingestion, regardless of its sign, across the two states. To statistically compare the observed changes when neurons are tracked, we created a null model reflecting a situation in which the same neurons respond in each state: neurons were sorted by their response amplitude independently in each state and paired together. Hence, the most active or inhibited neurons were paired with each other to yield a change in responsivity near zero. With this approach, a greater shift was found in the observed data than would be expected by the null model across all mice (Figure 3L). Consistently, classifiers trained on trial-by-trial data from single neurons could successfully predict the energy state of the mouse (Figure 3M). However, classifiers failed to perform better than chance when trained on a random subset of the population in each state (Figure S2A). These analyses suggest that the set of neurons responding to ingestion differs in each energy state.

To further visualize this change in representation within the population, we evaluated the subsets of PVH^GLP-1R neurons statistically classified as responsive to ingestion in a specific state. Neurons showing activation in fed mice shifted toward diminished activity under food deprivation (Figures 3N–3P). Similarly, neurons inhibited in fed mice showed little to no responses in fasted mice, though few neurons of this type were observed (Figures 3Q–3S). Similar results were obtained by conversely analyzing neurons responsive under food deprivation in fed mice (Figures S2B–S2G). Thus, these results suggest that PVH^GLP-1R neurons shift their activity dynamics in response to ingestion as energy state changes.

Population-averaged responses of PVH^GLP-1R neurons to ingestion do not depend on satiety state or tastant identity

To test whether these shifts in responses among individual neurons reflect changes in overall population-averaged responses, we examined the distribution and mean responses to ingestion among all neurons. We included all neurons regardless of whether they were tracked across sessions in this analysis (Figure 4A, ∼160 neurons in each state). Averaging the responses of the population revealed an increase in activity following sucrose consumption that was similar between energy states (Figures 4B and 4C). Moreover, the distribution of neurons classified as activated, inhibited, or unresponsive was similar. In total, neurons showing significant responses to ingestion comprised 25% (36/144 cells) and 27% (45/164 cells) of the observed fed and fasted populations, respectively (Figure 4D). Within these responsive populations, response magnitude did not differ by energy state (Figures 4E–4H). We also compared responses to sucrose ingestion with responses to other taste solutions by interleaving trials of sucrose delivery with water or the non-caloric sweetener sucralose (Figures 5A–5D). Response amplitude and distribution were similar between sucrose and water or sucralose (Figures 5E–5M). Similar results were obtained with the bitter tastant quinine, although activated neurons showed a slightly greater transient amplitude than in response to sucrose (Figures 5E and 5N–5Q). Thus, a subset of PVH^GLP-1R neurons responds to ingestion independent of caloric value or taste. While the individual neurons displaying this response are dynamic, population-averaged activity appears static across energy states.

Figure 4.

Average population dynamics of PVH^GLP-1R neurons are time-locked to ingestion independent of energy state

(A) Heatmap of trial-averaged dF/F aligned to lick initiation (dotted line) in food-deprived (top) and fed (bottom) mice (n = 5 mice, 164 fasted cells, 144 fed cells).

(B) Population-averaged response of PVH^GLP-1R neurons to ingestion.

(C) Quantification of response (area under curve 2 s after lick initiation) to ingestion in fasted and fed mice (Mann-Whitney test, U = 11494, p = 0.6877).

(D) Distribution of neurons classified as activated, inhibited, or unresponsive in each state (Fisher’s exact test, p = 0.7931).

(E) Average response of activated PVH^GLP-1R neurons in each state.

(F) Quantification of response shown in (K) (n = 33 fasted cells, 28 fed cells, Mann-Whitney test, U = 403, p = 0.3999).

(G) Average response of inhibited PVH^GLP-1R neurons in each state.

(H) Quantification of response shown in (M)—n = 11 fasted cells, 10 fed cells, two-tailed t test, t(19) = 1.251, p = 0.2261. Data represent mean ± SEM or mean +SEM (bar graphs).

Figure 5.

PVH^GLP-1R neurons respond similarly to ingestion of stimuli of varying caloric content, taste, and valence

(A) Schematic for trial-by-trial imaging of PVH^GLP-1R neuronal activity during interleaved trials of sucrose and sucralose, water, or quinine delivery in food-deprived mice.

(B) Average lick rate to sucrose or water within the same imaging session (n = 5 mice).

(C) Average lick rate to sucrose or sucralose within the same imaging session (n = 5 mice).

(D) Average lick rate to sucrose or quinine within the same imaging session (n = 5 mice).

(E) Heatmap of trial-averaged dF/F of PVH^GLP-1R neuron responses to licking sucrose (left) or an alternative tastant (right). Activity from the same neuron is shown on across rows (n cells = 163 [water], 164 [sucralose], 173 [quinine]).

(F) Population-averaged activity in response to sucrose or water ingestion.

(G) Quantification of response shown in (F) (Mann Whitney test, U = 11968, p = 0.1220).

(H) Neuron-averaged activity of activated (top) or inhibited (bottom) PVH^GLP-1R neurons in response to sucrose or water ingestion.

(I) Quantification of corresponding responses shown in (H)—n = 28 sucrose/36 water activated cells, Mann-Whitney test, U = 493, p = 0.8878; n = 17 fasted/6 fed water inhibited cells, two-tailed t test, t(21) = 0.8447, p = 0.4078.

(J) Population-averaged activity in response to sucrose or sucralose ingestion.

(K) Quantification of response shown in (J)—unpaired t test, t(326) = 0.7747, p = 0.4391.

(L) Neuron-averaged activity of activated (top) or inhibited (bottom) PVH^GLP-1R neurons in response to sucrose or sucralose ingestion.

(M) Quantification of corresponding responses shown in (L)—n = 43 sucrose/45 sucralose activated cells, Mann-Whitney test, U = 808, p = 0.1854; n = 5 sucrose/8 sucralose inhibited cells, two-tailed t test, t(11) = 0.0805, p = 0.9373.

(N) Population-averaged activity in response to sucrose or quinine ingestion.

(O) Quantification of response shown in (N) (Mann-Whitney test, U = 13877, p = 0.2426).

(P) Neuron-averaged activity of activated (top) or inhibited (bottom) PVH^GLP-1R neurons in response to sucrose or quinine ingestion.

(Q) Quantification of corresponding responses shown in (P)—n = 53 sucrose/32 quinine activated cells, Mann-Whitney test, U = 469, p = 0.0005; n = 17 sucrose/12 quinine inhibited cells, t(27) = 1.115, p = 0.2748. ∗∗∗p < 0.001. Data represent mean ± SEM or mean +SEM (bar graphs).

Discussion

We found that PVH^GLP-1R neurons can bidirectionally regulate feeding: inhibition of PVH^GLP-1R neurons induces feeding and elevating intracellular cAMP levels suppresses food intake. Using in vivo calcium imaging to track individual PVH^GLP-1R neurons, we found they transiently respond to ingestion, regardless of the tastant being consumed. These responses are heterogeneous, with a large fraction of neurons exhibiting no response to ingestion in either ad libitum fed or fasted states. By longitudinally tracking neuron activity, we found that individual PVH^GLP-1R neurons tune their response profiles according to energy needs. These findings demonstrate one way in which hypothalamic neuronal circuits may remodel in response to changing energy demands to maintain energy homeostasis.

PVH^GLP-1R neurons are necessary for normal body weight regulation.¹⁰ Chemogenetic silencing elicits food intake, and activation suppresses feeding.^7,9 Here, we extend these findings by demonstrating that PVH^GLP-1R neurons rapidly regulate feeding on the order of minutes. Increasing intracellular cAMP levels potently elevated PVH^GLP-1R neuronal firing in brain slices. Interestingly, PVH neurons expressing the melanocortin-4 receptor do not increase spiking in vivo after cAMP elevation.¹⁶ Whether this coupling between cAMP and spiking across PVH cell types is heterogeneous, as shown for other brain regions,¹⁷ requires further study. This may occur due to differential expression of hyperpolarization-activated cyclic nucleotide-gated ion channels or phosphodiesterases.^17,18 Moreover, the contributions of GLP-1 to cAMP signaling and spiking under physiological circumstances, given its binding to a G_s-coupled receptor that increases cAMP,^12,19 remain to be investigated but may depend upon its cellular location of action. For example, GLP-1 can enhance synaptic strength by modulating presynaptic release probability or excitatory postsynaptic currents.^9,10 Whether spontaneous firing increases in vivo needs further study. Moreover, there are neuropeptide receptors co-expressed on PVH neurons.^7,8 Whether GLP-1 is the key regulator of cAMP signaling among all other neuromodulators affecting PVH^GLP-1R neurons is unclear.

Here, we showed that individual PVH^GLP-1R neurons respond to fluid ingestion during both fasted and ad libitum fed states. We speculate that the transient response to ingestion limits food intake, as these neurons can reduce ongoing consumption.^9,20 Because PVH^GLP-1R neurons have previously been shown to respond to food-related cues prior to the initiation of feeding only in food-deprived mice,^7,9 we speculate that food cues and ingestion elicit distinct responses. This suggests roles for PVH^GLP-1R neurons in both the appetitive (i.e., sensory detection) and consummatory (i.e., ingestion) phases of feeding, though the taste of the small volume of sucrose may also act as a sensory cue.²¹ How sensory modulation of PVH^GLP-1R neuronal activity relates to satiety is unclear. One possibility is that the appetitive response of PVH^GLP-1R neurons drives anticipatory satiety or halts exploratory behavior, as has been proposed for the melanocortin system.^{7,22,23,24,25,26,27} Further studies are needed to determine PVH^GLP-1R neuron responsivity to food cues at single-cell resolution. We recently reported that PVH^GLP-1R neurons send descending projections to the brainstem where synapse strength varies with energy state.⁹ Such presynaptic plasticity may allow distinct neuronal ensembles to convey information about energy state to downstream targets. However, it is unclear whether the set of highly responsive neurons we imaged corresponds to those undergoing state-dependent synaptic plasticity. Future work should examine whether projection-defined PVH^GLP-1R neurons exhibit similar activity dynamics.

We observed no differences in the magnitude or distribution of PVH^GLP-1R neuron population dynamics during ingestion across different energy states regardless of taste or caloric content. This suggests that some PVH^GLP-1R neurons may signal mechanosensory aspects of ingestion, which are likely driven by synaptic input. Along with dense input from arcuate AgRP and POMC neurons, the PVH receives input from neurons sensitive to caloric intake and interoception. For example, several classes of neurons of the nucleus tractus solitarius (NTS) respond to aspects of feeding, including mechanosensation and nutrient sensing.^28,29,30,31 Many NTS neurons receive ascending mechanosensory information related to ingestion,^32,33 but whether they project to the PVH is uncharacterized. While NTS GLP-1 neurons compose the primary source of GLP-1 to the PVH,^34,35 it is unlikely that the rapid changes in PVH^GLP-1R neuron activity during ingestion are driven directly by GLP-1 signaling. NTS GLP-1 neurons respond specifically to visceral feedback over minutes.^30,31 Thus, the rapid PVH^GLP-1R neuron dynamics during ingestion are likely to be driven by synaptic input from fast neurotransmitters rather than slower neuropeptide signaling. Clarifying the connectivity patterns of PVH^GLP-1R neurons with known neuronal populations involved in sensory regulation of food intake may shed light on this issue. PVH^GLP-1R activity may also follow the slow changes in activity observed in GLP-1 neurons.^10,30 However, the current trial-by-trial imaging strategy is designed to investigate only short-term changes during ingestion.

We observed shifts in cellular encoding of ingestion among PVH^GLP-1R neurons across energy states. This parallels findings of representational drift in sensorimotor cortices and hippocampus.^36,37 It is possible that this drift simply reflects stochastic changes in the strength of afferents to PVH^GLP-1R neurons, as has been shown to occur for other hypothalamic cell types.^{38,39,40,41,42} Intrinsic excitability differences could also change the sensitivity of PVH^GLP-1R neurons to synaptic input.^10,43 Alternatively, PVH^GLP-1R neurons may reflect a heterogeneous population of multiple functionally distinct cell types that are defined by other neurochemical markers.^7,8 Nevertheless, recent reports indicate that hypothalamic neurons can undergo state-dependent shifts in the cellular encoding of innate behaviors such as sleep and food consumption.^44,45 These changes may enhance flexibility in regulating energy homeostasis in the face of external challenges like obesity that may cause damage or dysfunction in individual neurons.^41,46 The effects of conditions like diet-induced obesity on the stability of representations of feeding require further investigation.^42,47,48 Approaches that leverage optical methods at single-cell resolution hold great potential in advancing these studies.

Limitations of the study

Several methodological and interpretive limitations should be considered when evaluating these findings. Many PVH cell types are tuned to multiple behavioral states, including acute stress. Although mice were habituated prior to testing, the neural activity we recorded may be influenced by stress associated with head fixation. It is unclear whether subsets of PVH^GLP-1R neurons alter their ingestion response profiles during times of elevated stress. Similarly, the present experiments cannot determine whether similar responses would be observed in freely moving mice. Future work integrating molecular identity and single cell imaging in unrestrained mice may yield insights into these issues.

Resource availability

Lead contact

Additional information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Mark A. Rossi (mark.rossi@rutgers.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Original data and analysis code are available at Zenodo: https://doi.org/10.5281/zenodo.17353213.
• •Code used to analyze imaging and behavioral data is deposited in a GitHub repository (https://github.com/RohanSavani/PVN_GLP1R_2P).
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Animal models
GLP-1R-ires-Cre mice	Williams et al.32	RRID: IMSR_JAX:029283
Bacterial and virus strains
AAV1-SIO-hSyn1-stGtACR2-FusionRed	Addgene	Cat#105667
AAV2/DJ-DIO-EF1α-mKate2-biPAC	Boston Children’s Hospital Viral Core	Cat#169127
AAV1-FLEX-tdTomato	Addgene	Cat#28306
AAV5-Syn-FLEX-jGCaMP8m	Addgene	Cat#162378
Software and algorithms
Suite2p	Pachitariu et al.49	RRID:SCR_016434
GraphPad Prism 10	https://www.graphpad.com	RRID:SCR_002798
Adobe Illustrator 2022	https://www.adobe.com	RRID:SCR_010279
BioRender	https://biorender.com/	RRID:SCR_018361
Prairie View	https://bruker.com	RRID:SCR_017142
Deposited data
Imaging data and analyses	This paper	https://doi.org/10.5281/zenodo.17353213
Other
Optical fiber	RWD Life Sciences	Cat# R-FOC-BL200C-50NA
GRIN lens	Inscopix	Cat# 1050-004597

Experimental model and study participant details

Animals

Experimental procedures involving mice were approved by the Rutgers University Institutional Animal Care and Use Committee (Protocol #PROTO201702609) and conformed to guidelines for the Care and Use of Laboratory Animals set by the National Institutes of Health. Mice were housed in the Child Health Institute of New Jersey vivarium facilities (21°C–23°C, 30–70% humidity, 12-h light-dark cycle) with ad libitum access to water and standard chow (TestDiet, 5058-PicoLab), with all experiments conducted during the light phase. Homozygous Glp1r-ires-Cre hybrid C57BL/6J mice were used.³² Male mice were used for all experiments. For experiments involving multiple groups, littermate co-housed mice were randomized to each experimental group on the basis of body weight. Investigators were blinded to group assignment during experiments. Sample sizes required (n = 5–7) for each group were estimated from previous studies.^9,42 Mice that underwent two-photon imaging were single-housed following GRIN lens implantation.

Method details

Surgeries

5-7-week-old mice were anesthetized with isoflurane (induction at 3%, 1–1.5% for maintenance) and placed in a stereotactic frame with a heating pad. Following application of ophthalmic ointment, local subcutaneous injection of bupivacaine (0.025%), subcutaneous injection of carprofen (5 mg/kg), and scalp shaving and scrubbing (betadine, ethanol), a midline incision was made to expose the skull. A dental drill was then used for craniotomy (0.5–1 mm) above the injection and implantation site. Viruses were injected at a rate of 1 nL/s (150 nL total volume) using a glass pipette controlled by Nanoject III (Drummond Scientific). The needle was slowly withdrawn after 5 min. Following injection and implantation, a stainless-steel ring was also cemented atop the head for head fixation. Mice were monitored daily (administered carprofen for 3 days) and recovered for at least 4 weeks prior to experimentation.

For optogenetics experiments, AAV1-SIO-hSyn1-stGtACR2-FusionRed (2.4 × 10¹³ gc/mL, Addgene Cat#105667), AAV2/DJ-DIO-EF1α-mKate2-biPAC (prepared by Boston Children’s Hospital Viral Core, 1.66 × 10¹³ gc/mL, Addgene Cat# 169127), AAV1-FLEX-tdTomato (control for both groups, 2.0 × 10¹³ gc/mL, Addgene Cat# 28306) were injected bilaterally into the PVH (relative to bregma, −0.8 mm AP, ±0.18 mm ML, −4.75 mm DV). A 200 μm diameter optical fiber (RWD Life Sciences) was unilaterally implanted (0 mm ML) 100–150 μm above the injection site.

For two-photon imaging experiments, AAV5-Syn-FLEX-jGCaMP8m (1.2 × 10¹³ gc/mL, Addgene Cat# 162378) was injected unilaterally into the PVH (relative to bregma, −0.8 mm AP, +0.18 mm ML, −4.75 mm DV). A GRIN lens (0.6 mm diameter, 7.3 mm length, Inscopix) was then slowly implanted 200 μm above the injection site (+0.20 mm ML) and affixed with dental cement.

Optogenetics

Behavior experimental design

Behavioral paradigms were adapted from prior work.⁹ Light was delivered by a 473 nm DPSS laser at different frequencies, coordinated by custom Arduino programs. 4.5 mW (measured at tip of fiber) constant stimulation was used for photoinhibition with stGtACR. 7.5 mW pulsed stimulation was used for optogenetic elevation of cAMP with biPAC. For measurements of food intake in freely moving mice, food pellets were manually weighed (including crumbs left in the cage) after each phase of the experiments. For measuring real-time place preference to light activation of biPAC and GtACR, mice with attached optic fiber patch cables were habituated to an acrylic arena (20 × 20 × 42 cm). A camera tracked positions of each mouse within the arena for 10 min. Whenever a mouse entered the stimulation-paired side of the arena, light was delivered (same parameters as above). Custom MATLAB scripts determined mouse position based on fur coat contrast with the arena to determine entry into each zone. Data were processed in Python. For analysis of real-time place preference experiments, a preference index was calculated as the time spent in the light-paired side relative to total time of experiment.

Validation with electrophysiology

Brain slice whole-cell patch-clamp electrophysiology was performed as previously described.⁹ Mice were anesthetized and decapitated. Brains were removed and immersed in cold (4°C) oxygenated cutting solution containing (in mM) 50 sucrose, 2.5 KCl, 0.625 CaCl₂, 1.2 MgCl₂, 1.25 NaH₂PO₄, 25 NaHCO₃, and 2.5 glucose. 300 μm coronal sections were cut using a vibratome (Leica, VT 1200S) and collected in artificial cerebrospinal fluid (ACSF) bubbled with 5% CO₂ and 95% O₂ and containing 125 NaCl, 2.5 KCl, 2.5 CaCl₂, 1.2 MgCl₂, 1.25 NaH₂PO₄, 26 NaHCO₃, and 2.5 glucose. After 1 h recovery, slices were transferred to a recording chamber and constantly perfused with bath solution (33°C) at a flow rate of 2 mL/min. Patch pipettes (5–8 MΩ) were made from borosilicate glass (World Precision Instruments) with a pipette puller (Narashige, PC-10) and filled with 126 K-Gluconate, 4 KCl, 10 HEPES, 4 Mg-ATP, 0.3 Na₂-GTP, 10 phosphocreatine (pH to 7.2 with KOH). For GtACR validation, after successful whole-cell patch clamp, current clamp recordings of action potentials at stepped current injections were performed in the same cell with and without 473 nm constant light stimulation of the brain slice. For biPAC validation, each cell underwent the same protocol twice, first without light stimulation and then with light. A ramping pulse of current (1 s duration, 30 pA) was injected twice, once before and once after the light stimulation would occur. MultiClamp 700B, Digitizer 1440, and pClamp v.10.5.0.9 (Molecular Devices) were used for data acquisition. Spontaneous and evoked spikes were manually counted offline using ClampFit 10.2.

Two-photon calcium imaging

Behavioral design

Following 3 to 4 weeks of recovery, mice were habituated to head fixation for at least 3 days, during which randomly delivered drops of 10% sucrose (in water, 2–3 μL, every 20-30s) were delivered for at least 15 min. Liquids were delivered via gravity-driven, solenoid-controlled tubing coordinated with lick sensing from a custom-built lickometer via Arduino and MATLAB programs. Mice were initially trained while water-deprived and were then food-deprived to begin experiments, once they licked rapidly in response to sucrose delivery. To assess state-dependent neural responses to ingestion, mice received 10% sucrose drops every 25-30s for 30 trials while ad libitum fed or food-deprived (overnight, ∼16h). When fasted across successive days, mice received 2-3g of normal chow to maintain ∼90% of their body weight. For experiments in which multiple liquids were delivered, two solenoids separately delivered liquids to the spout in randomly interleaved trials. Fasted mice received 50 trials for sucrose/sucralose comparisons (25 trials of 10% sucrose, 25 trials of 6.25 mM sucralose) and sucrose/water comparisons (30 10% sucrose, 20 tap water trials). Water-deprived mice received 60 trials for sucrose/quinine comparisons (45 10% sucrose, 15 0.5 mM quinine trials). Mice were water-deprived for the latter experiments to increase their motivation to lick on trials, as fasted mice often ceased licking altogether.

Data collection and analysis

2-photon imaging was performed using an Ultima Investigator laser scanning microscope (Bruker, 10× air objective, 8 mm working distance, 0.5 NA, Thorlabs TL10X-2P), as previously described.⁴² An ALCOR 920 nm 2W fixed-wavelength laser (100 fs pulses, 80 MHz, Spark Lasers) acted as the light source. Imaging was performed trial-by-trial (25s total, from 10s pre-delivery to 15s post-delivery) and was coordinated by Arduino-driven TTL inputs to Prairie View software (Bruker). Images were collected using resonant scanning (30 Hz, 2-frame-averaged online to 15 Hz). Fields of view were kept consistent across multiple sessions. Laser power and PMT gain were kept consistent for experiments where data were directly compared.

After movie acquisition, data were analyzed using Suite2P⁴⁹ for motion correction, automated cell segmentation, and fluorescence intensity extraction. A custom Cellpose⁵⁰ model (cyto architecture) trained on 22 manually segmented enhanced mean projection images (each with 15–60 ROIs) for 7500 iterations was used for automated segmentation. Further analysis was performed using custom Python scripts. Motion-corrected time series data were normalized trial-by-trial to the first 9 s of each trial to compute dF/F values. Trials in which mice failed to lick at least thrice (with an interlick interval of at most 500 ms) within 5 s of liquid delivery were discarded. All mice had at least 10 successful trials (average of 15 trials for ad libitum fed mice and 22 trials for fasted). Data were then aligned to the time of first lick following reward and averaged across trials. Data as presented in figures are smoothed/downsampled for presentation (not statistics). Area under the curve (AUC) of the dF/F time series was computed using the 2 s following lick bout onset.

To classify cells as activated or inhibited in response to ingestion, we applied a “circular shift” method previously described.⁵¹ Briefly, data under the null hypothesis that pre- and post-lick data are drawn from the same distribution were simulated by shifting calcium traces from randomly selected neurons (pooled from all recordings for each satiety state or trial type, 1000 simulations per mouse). Then, Wilcoxon rank-sum test statistics comparing baseline (−5 s to −3 s before reward delivery) and post-lick activity (0 s–2 s after lick bout onset) on each trial were summed and normalized by the number of trials used in summing. A two-tailed p-value was computed comparing this statistic for the observed data to those in the null distribution. Statistically significant (p < 0.05) responses for a given cell indicated responsiveness.

Identical cells across different experiments were tracked manually and aligned/compared using custom Python scripts. To calculate the change in each neuron’s response between caloric state, the AUC of mean fluorescence 2 s following lick bout onset in ad libitum fed mice was computed and subtracted from that in fasted mice. The absolute value of this change was then compared to data under the hypothesis that each neuron responded similarly in both states. Neurons were sorted independently in each state such that the most active/inactive neurons in each state were paired. The same absolute value in the change in AUC was calculated using this sorted pair. This metric was then averaged for all neurons for each mouse and used for statistics.

To determine whether the activity of individual neurons could predict satiety state, we applied classification methods as previously described.^45,47 Support vector machines were trained with cross-validation (StratifiedShuffleSplit, 5 splits, 20% test set size, C parameter range of [1e-3, 1e-2, 1e-1, 1, 10, 100, 1000], gamma parameter range for rbf of [1e-3, 1e-2, 1e-1, 1, 10, 100, 1000]) on neurons tracked across both satiety states. A classifier was fit to the trial-by-trial calcium imaging data for each neuron using mean activity 2 s following lick bout onset on each trial. To test whether this model performed better than chance, the same procedure was applied to data with shuffled satiety state labels. To determine if the trial-averaged activity of all neurons could predict satiety state, support vector machines were trained with cross-validation similarly. The mean and max activity of all neurons in the training set (90%) following lick bout onset was used to train classifiers. This was repeated for 1000 iterations, with the highest accuracy score for each run across all parameters of the grid search being used. The same procedure was performed using shuffled state labels to compute a p-value for decoding accuracy.

Histology

Mice were deeply anesthetized under isoflurane and transcardially perfused with PBS followed by 4% paraformaldehyde (PFA). Brains were post-fixed in 4% PFA overnight and switched to 30% sucrose the next day for >24 h. Brains were embedded in OCT (Tissue-Tek), cryosectioned (50 μm coronal slices), and stored in PBS. Following this or immunostaining, sections were mounted onto glass slides with Fluoroshield (with DAPI). Images were acquired via confocal microscopy (Zeiss, LSM 700). Injection and implantation sites were confirmed in all mice. 2 mice were excluded from the GtACR and biPAC experimental groups due to off-target injection and/or implants. 1 mouse was excluded for off-target GRIN lens placement for calcium imaging. 6 excluded implanted mice had insufficient GCaMP expression during calcium imaging.

Quantification and statistical analysis

Statistical analyses were performed using GraphPad Prism (10.0) and Python (3.9). All data, unless otherwise specified, are presented as mean ± standard error of the mean. Data were tested for normality and equal variance prior to application of tests (details of statistical tests used in figure legends). All tests were two-sided and corrected for multiple comparisons, normality, and unequal variance where needed. Asterisks in figures denote statistical significance: ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

Acknowledgments

This work was supported by grants from the Robert Wood Johnson Foundation to the Child Health Institute of New Jersey (RWJF grant no. 74260), NIH grant no. NIDDK R01DK131452 (Z.P.P.), NIH grant no. NIDDK R01DK136641 (M.A.R.), and the Whitehall Foundation grant (no. 2022-12-051, M.A.R.). We thank Dr. Jorge Luis-Islas for technical assistance and the members of the Pang and Rossi labs for helpful discussions and feedback on this work.

Author contributions

R.H.S. and L.W. conceived the project and designed the experiments with input from M.A.R. and Z.P.P.; L.W. performed all surgeries and electrophysiological validation; R.H.S. collected and analyzed two-photon imaging data with assistance from L.W. and Y.L.; R.H.S. collected and analyzed data from optogenetics experiments with assistance from T.L.; R.H.S., E.P., and M.Y. performed histology and confocal imaging; and R.H.S. and L.W. wrote the manuscript with M.A.R. and Z.P.P., with input from all authors.

Declaration of interests

The authors declare no competing interests.

Published: November 15, 2025