A hypothalamic circuit for anticipating future changes in energy balance

Summary

AgRP neurons cause hunger, the drive to seek and consume food. Their activation by fasting is key for survival and is thought to be triggered by feedback when energy stores are low. However, we know that environmental cues can also regulate AgRP neurons, since cues that predict future food intake rapidly inhibit AgRP neurons. But is the converse true: can the prediction of future fasting rapidly activate AgRP neurons? Here we show in mice that such rapid fasting activation of AgRP neurons does occur. This rapid activation is driven by excitatory input from paraventricular hypothalamic neurons expressing Sim2, which are bidirectionally sensitive to predictions of future energy state. Thus, cognitively-processed contextual information conveyed by PVH^Sim2 neurons strongly activates AgRP neurons. Lastly, chronic silencing of PVH^Sim2 neurons causes persistent hypophagia. This PVH^Sim2 to AgRP neuron circuit, by anticipating and preventing negative energy balance, provides an important new dimension of hunger regulation.

Graphical Abstract

Introduction

Tight regulation of food intake is critical to fitness and survival, and is achieved by a variety of neuron types in the brain. How these neurons integrate long-timescale signals of systemic energy balance with shorter-timescale sensory and cognitive signals remains unclear.

Neurons in the arcuate nucleus of the hypothalamus expressing the neuropeptide AgRP (ARC^AgRP neurons) are key regulators of energy balance¹. ARC^AgRP neurons have been termed “hunger neurons” since they show increased activity following fasting², and experimentally increasing their activity recapitulates many behavioral, physiological, and neural consequences of fasting, including increased food intake and food-seeking^1,3–6.

During eating-induced satiation, ARC^AgRP neurons integrate rapid feedforward sensory information with post-ingestive gut signals and slower, hormonal signals of fat stores, producing a rapid and sustained drop in activity in response to food^7–12. In contrast, how fasting increases ARC^AgRP neuron activity, and over what timescale, remains poorly understood. ARC^AgRP neurons are sensitive to circulating feedback signals of energy balance (e.g. leptin, insulin, ghrelin)^12–16, and their activity is thought to gradually increase as energy stores are depleted^1,2. Recent technical advances have enabled continuous long-term ARC^AgRP neuron recordings^17,18, and such recordings throughout fasting have challenged this view¹⁷. They show ARC^AgRP neuron activity increasing very early in fasting, likely before energy stores are depleted¹⁷. The origin of this rapid early activation is unknown.

Another key nucleus in food intake control is the paraventricular nucleus of the hypothalamus (PVH)¹⁹. The PVH contains neurons that regulate diverse autonomic, neuroendocrine, and behavioral processes, many related to energy balance^20,21. It has long been considered a satiety center: Complete PVH ablation or inhibition increases food intake^22,23, whereas activation inhibits feeding²⁴. These effects are attributable to PVH “satiety neurons” that are inhibited by ARC^AgRP hunger neurons^2,19,25,26.

Embedded within the PVH “satiety center”, however, is a subset of neurons with an opposing effect. Activating PVH neurons expressing Trh and Adcyap1 increases food intake in sated mice²⁷ via excitatory output onto ARC^AgRP neurons. These PVH→ARC^AgRP connections undergo profound energy state-dependent plasticity^28,29, and preventing this plasticity by deleting NMDA receptors from ARC^AgRP neurons reduces body weight and food intake²⁹. Nevertheless, what information PVH^Trh/Adcyap1 neurons transmit to ARC^AgRP neurons, and how they contribute to physiological food intake regulation, remains unclear¹. This is largely because both Trh and Adcyap1 are expressed across multiple PVH cell types with diverse functions^20,30–32, precluding selective access to PVH→ARC^AgRP afferents.

In this study, we mined our transcriptomic atlas of the mouse PVH³² to identify a specific PVH neuron type (PVH^Sim2 neurons) that provides excitatory input to ARC^AgRP neurons and drives feeding. We demonstrate that these PVH^Sim2 neurons transmit bidirectional predictions of future energy balance to ARC^AgRP neurons. Finally, we show that this excitatory input to ARC^AgRP neurons is critical for long-term food intake regulation, since chronic inhibition of PVH^Sim2 causes a chronic reduction in food intake.

Results

Sim2 expression selectively labels Trh/Adcyap1 neurons in the PVH

PVH neurons expressing Trh and Adcyap1 synapse onto ARC^AgRP neurons and regulate food intake^27,28. However, these genes are expressed in multiple, functionally distinct PVH neuron types (for example, Trh is also expressed in neuroendocrine neurons that regulate thyroid hormone release³³). To gain selective genetic access to ARC^AgRP neuron afferents, we aimed to pinpoint their precise transcriptional identity (Figure 1A). We specifically looked for neurons co-expressing both Trh and Adcyap1 in our PVH neuron atlas³² as likely AgRP neuron afferents, since Trh+/Adcyap1+ co-expressing neurons are present in the PVH²⁷, show increased expression of Fos following overnight fasting²⁸, and increase food intake when activated (Figure S1).

Figure 1. PVH^Sim2 neurons excite ARC^AgRP neurons and drive feeding.

A) PVH→ARC^AgRP schematic. B) Representative Trh (green), Adcyap1 (yellow) and Sim2 (magenta) expression in anterior PVH, detected by smFISH (RNAscope). C, D) Percentage of PVH neurons coexpressing the indicated genes. (N=3 mice, mean±SEM). E) Sim2-expressing neurons per section at different anterior-posterior levels of PVH. A: anterior; M: mid, P: posterior. (N=3 mice, mean±SEM; ordinary one-way ANOVA). F) Left: injection of AAV-DIO-ChR2-mCherry into PVH of a Sim2–2a-Cre::Npy-hrGFP mouse. Middle: mCherry expression in PVH, and not surrounding areas. Right: mCherry-labelled PVH^Sim2 neuron projections (magenta), and GFP-expressing ARC^AgRP neuron cell bodies (cyan), in ARC. G) Fraction of Npy-hrGFP-positive (top) and -negative (bottom) ARC neurons that show leEPSCs with monosynaptic latency following optogenetic activation of PVH^Sim2 neuron projections by CRACM. H) Example recording from Npy-hrGFP-positive neuron with optogenetic stimulation of PVH^Sim2 neurons in artificial cerebrospinal fluid (aCSF) alone (top), or with tetrodotoxin (TTX, middle), or with TTX and 4-aminopyridine (4-AP, bottom). I) Left: in vivo optogenetic stimulation of PVH^Sim2 terminals with simultaneous fiber photometry recording of ARC^AgRP neuron activity. Right: mean±SEM ARC^AgRP neuron activity before, during, and after 10s stimulation of PVH^Sim2 axons in ARC. (N=3 mice, mean over 15 stimulations per mouse). J) Left: AAV-DIO-hM3Dq-mCherry injection into PVH of Sim2–2a-Cre mice. Right: mCherry expression (magenta) in PVH. K) Daytime (ZT3–6) food intake following injection of vehicle or CNO. Mean±SEM; thin lines represent individual mice. (N=11 mice; two-way repeated measures ANOVA; post-hoc comparisons at each time point with Sidak correction). L) PVH^Sim2→ARC^AgRP circuit and its role in feeding. *p<0.05; **p<0.01; ***p<0.001.

The majority of Trh+/Adcyap1+ neurons belong to a transcriptional cluster uniquely expressing the transcription factor Sim2 (not to be confused with Sim1, which is broadly-expressed in most PVH neurons³⁴). We confirmed these results using three-color single molecule fluorescent in situ hybridization (smFISH) against Trh, Adcyap1 and Sim2 (Figure 1B). Throughout the PVH, 76.9 ± 3.7% (mean±SEM) of Trh+/Adcyap1+ neurons also expressed Sim2 (Figure 1C). Similarly, 77.2 ± 1.8% of Sim2-expressing neurons expressed both Trh and Adcyap1, and most of the remaining Sim2+ neurons either expressed Adcyap1 only (13.3 ± 1.2%) or Trh only (3.9 ± 1.1%) (Figure 1D). Interestingly, adult Sim2-expressing neurons were primarily located in the anterior and mid-PVH, with very few cells in posterior PVH (Figure 1E), consistent with a previous study of newborn mice³⁵ and our recent PVH MERFISH study³². This anterior localization is also consistent with monosynaptic rabies mapping of ARC^AgRP neuron afferents²⁷; and contrasts with PVH satiety neurons expressing Mc4r and Pdyn^25,26, which are located primarily from mid- to posterior PVH.

To gain specific genetic access to PVH^Sim2 neurons, we generated recombinase driver mice expressing Cre and FLPo under the control of the endogenous Sim2 gene (Sim2–2a-Cre, Figures S2A and S2B; and Sim2–2a-FLPo, Figures S2D and S2E). More tdTomato+ neurons were recombined using Sim2–2a-Cre than Sim2–2a-FLPo, likely due to Cre’s higher efficiency in marking neurons with transient developmental or low-level Sim2 expression (Figure S3), as previously described^36–38. Importantly, regions with cells recombined by Sim2–2a-FLPo were also seen with Sim2–2a-Cre. Using recombinase-dependent tdTomato reporter mice, both lines drove expression in several brain regions, including the PVH and mammillary bodies, where Sim2 expression has been previously described (Figure S3)^39,40. smFISH confirmed high overlap between tdTomato and Sim2 expression in the PVH (Figures S2C and S2F). Consistent with most PVH neurons being glutamatergic, almost all PVH^Sim2 neurons expressed VGlut2 (Slc17a6) (Figure S2C). Importantly, PVH^Sim2 neurons are almost exclusively centrally-projecting, and not neuroendocrine, since the vast majority (88.6 ± 3.3%) were not labelled by peripheral injection of fluorogold (Figure S4A; mean±SEM, N=4).

PVH^Sim2 neurons excite ARC^AgRP neurons and drive feeding

To test whether PVH^Sim2 neurons send axonal projections to the mediobasal ARC, where ARC^AgRP neurons are located, we labelled PVH^Sim2 neuron projections and ARC^AgRP neurons in the same mouse, using Sim2–2a-Cre::Npy-hrGFP mice (since ARC^AgRP neurons coexpress Npy⁴¹) injected with AAV-DIO-ChR2-mCherry in the PVH (Figure 1F). We then imaged a 700μm-thick section of ARC cleared using CUBIC⁴². We observed dense synaptic projections in the ARC, specifically in the region containing GFP+ ARC^AgRP neurons. Moreover, many GFP+ ARC neurons were densely wrapped by mCherry+ fibers, suggesting that PVH^Sim2 neurons may provide direct excitatory input to ARC^AgRP neurons.

To visualize PVH^Sim2 neuron projections more comprehensively, we injected AAV-DIO-Syp-mCherry into the PVH of Sim2–2a-Cre mice and examined labelled axonal projections throughout the brain (Figure S4B). This confirmed that PVH^Sim2 neurons are concentrated in the anterior PVH, and densely and specifically target the mediobasal ARC. We also observed dense PVH^Sim2 projections in the bed nucleus of the stria terminalis (BNST), and sparse projections in a few other forebrain regions (Figure S4B).

We next asked whether individual PVH^Sim2 neurons project to single targets or collateralize to multiple target sites. Retrograde viral collateral mapping using AAVretro-fDIO-Cre revealed that at least a subset send collaterals to both the ARC and BNST (Figures S4C and S4D), confirming that at least some individual PVH^Sim2 neurons project to multiple sites.

To test whether PVH^Sim2 neurons provide direct monosynaptic input to ARC^AgRP neurons, we used channelrhodopsin (ChR2)-assisted circuit mapping (CRACM)⁴³, using Npy-hrGFP to label ARC^AgRP neurons. In ex vivo brain slices, activating ChR2-expressing PVH^Sim2 axons resulted in light-evoked excitatory post-synaptic currents (leEPSCs) with monosynaptic latency (~6ms) in ~40% of Npy+ neurons (Figure 1G). These leEPSCs were abolished by tetrodotoxin (TTX), but reinstated by co-application of 4-aminopyridine (4-AP), confirming monosynaptic connectivity (Figure 1H). No leEPSCs were observed in Npy-negative ARC neurons (Figure 1G). The PVH^Sim2→ARC^AgRP connectivity rate is lower than the ~75% reported for PVH^Trh neurons²⁸, potentially reflecting lower ChR2 expression driven by Sim2–2a-Cre (Sim2 is expressed at lower levels than Trh, see expression levels in Figure 1B); or alternatively because PVH afferents to some ARC^AgRP neurons express Trh but not Sim2. Regardless, these results establish that PVH^Sim2 neurons provide selective monosynaptic glutamatergic input to ARC^AgRP neurons. Notably, given that the majority of PVH^Sim2 neurons express Adcyap1 (Figure 1D), and that its neuropeptide product PACAP increases AgRP neuron firing rates ex vivo²⁷, PVH^Sim2 neurons likely also influence AgRP neuron activity via neuropeptide release alongside fast glutamatergic transmission.

To test whether PVH^Sim2 neurons influence ARC^AgRP neuron activity in vivo, we optogenetically activated their projections to ARC while monitoring ARC^AgRP neuron activity using fiber photometry (Figure 1I). We expressed the red-light-gated cation channel ChrimsonR in the PVH under Sim2–2a-Cre, and the calcium-dependent fluorophore GCaMP6s in the ARC under Npy-IRES-FLPo, and placed an optical fiber for light delivery and collection above the ARC. Brief (10s) stimulations of PVH^Sim2 neurons increased ARC^AgRP neuron activity, which returned to baseline after stimulation offset (Figure 1I). Thus, PVH^Sim2 neurons can increase ARC^AgRP neuron activity in vivo. Of note, ARC^AgRP neurons project reciprocally to the PVH, where they directly inhibit satiety neurons and indirectly increase the activity of PVH^Crh neurons controlling the hypothalamic-pituitary-adrenal axis^5,23,25,26. PVH^Sim2 neurons, however, are notably unaffected by ARC^AgRP neuron activation (Figures S5A and S5B), underlining the unidirectional nature of the PVH^Sim2→ARC^AgRP neuron circuit.

Activation of ARC^AgRP neurons, or their PVH afferents labelled by Trh-IRES-Cre or Pacap-IRES-Cre, drives food intake^3,4,27. Thus, we hypothesized that activating PVH^Sim2 neurons would similarly increase food intake through activation of ARC^AgRP neurons. To test this, we expressed the excitatory chemogenetic receptor hM3Dq in PVH^Sim2 neurons, and measured food intake in response to injection of vehicle or the hM3Dq ligand clozapine N-oxide (CNO) in the early light phase, when mice usually eat little (Figures 1J and 1K). Activation of PVH^Sim2 neurons by CNO injection strongly increased food intake over 3 hours (Figure 1K). Next, to isolate the influence of ARC-projecting PVH^Sim2 neurons, we specifically activated the PVH^Sim2 neurons that project to the ARC. To achieve this, we injected AAVretro-fDIO-Cre into the ARC of Sim2–2a-FLPo mice to express Cre recombinase in ARC-projecting Sim2+ neurons, and then injected AAV expressing Cre-dependent, mCherry-tagged hM3Dq into the PVH (Figures S5C and S5D). Activating just the ARC-projecting PVH^Sim2 neurons by CNO injection increased food intake (Figure S5E). This indicates that PVH^Sim2 neurons can drive food intake through the activation of ARC^AgRP neurons (Figure 1L). In support of this, inhibition of ARC^AgRP neurons has been shown to prevent the ability of PVH afferents (PVH^Trh neurons) to drive food intake²⁷.

As noted above (Figure 1C), a small proportion of Trh+/Adcyap1+ neurons do not express Sim2. Our sc/snRNA-seq atlas suggested that a small subset of Trh/Adcyap1-coexpressing PVH neurons belong to a cluster uniquely expressing Ucn3, consistent with reports of Trh and Adcyap1 in Ucn3+ neurons^31,32,44,45. However, PVH^Ucn3 neurons project weakly to the ARC, do not synapse onto ARC^AgRP neurons, and do not increase food intake when activated in vivo (Figure S6). Ucn3 therefore does not label the excitatory PVH input to ARC^AgRP neurons.

Pre-ingestive inhibition of PVH^Sim2 neurons by food

Having identified a specific marker for the PVH neurons that synapse onto ARC^AgRP neurons (Sim2) and generated recombinase driver mouse lines to provide highly selective access to them (Sim2–2a-Cre and Sim2–2a-FLPo), we can now study the information these inputs convey by recording their activity in vivo. We started by characterizing the response of PVH^Sim2 neurons to food, given that ARC^AgRP neurons show a strong reduction in activity in response to food in the fasted state^7–9. We expressed GCaMP6s in PVH^Sim2 neurons and monitored population calcium activity using fiber photometry (Figures 2A and 2B). Presentation of a standard chow pellet (~2 g) to a fasted mouse led to a rapid drop in PVH^Sim2 activity (Figure 2C-F), which was complete within seconds (τ = 7.5 ± 6.2 s from pellet presentation, where τ represents the time at which 63.8% of the total activity drop has occurred). This rapid drop in activity is reminiscent of the ARC^AgRP neuron response to food^7–9. As with ARC^AgRP neurons⁷, the food response of PVH^Sim2 neurons was state-specific, since these neurons showed little response to food in the fed state (Figures 2G and 2H). Further, this response was specific to food, and not other motivationally salient stimuli, since following overnight water deprivation, presentation of water did not inhibit PVH^Sim2 neuron activity (Figures 2I and 2J). In fact, water presentation slightly increased PVH^Sim2 neuron activity, which may reflect increased food-seeking drive following the relief of dehydration-induced anorexia⁴⁶.

Figure 2. Pre-ingestive inhibition of PVH^Sim2 neurons by food.

A) Fiber photometry recordings from PVH^Sim2 neurons. B) GCaMP6s (green) in PVH^Sim2 neurons; dashed line indicates optic fiber. C) Experimental setup. D) Heatmap of PVH^Sim2 neuron activity before and after chow pellet presentation. Each row is one trial; black line indicates pellet delivery. (N=20 trials from 7 mice). E) Mean (±SEM) response of PVH^Sim2 neurons to pellet. F) Mean activity in 3 minutes before and after pellet delivery. Each line indicates one trial; box and whiskers represent upper/lower quartiles and minimum/maximum, respectively. (N=20 trials from 7 mice; two-tailed paired t-test). G, I) Response of PVH^Sim2 neurons to pellet delivery in the fed or fasted state (G), or to water bowl delivery in the water-deprived state and pellet delivery in the fasted state (I). (N=7 mice; mean±SEM over mice). H, J) Mean activity in 5 minutes following pellet delivery in fed and fasted states (H), or following water bowl delivery in the water-deprived state and pellet delivery in the fasted state (J). Each point represents mean over trials for one mouse; lines connect data points for the same mouse. (N=7 mice; two-tailed paired t-test). K) Mean±SEM activity of PVH^Sim2 neurons as in (E), but aligned to the first bite. L) Mean activity in different bins relative to pellet delivery, contact, and consumption (see Methods); each line indicates one trial. (N=20 trials from 7 mice; two-way repeated measures ANOVA; post-hoc comparisons between adjacent time points with Sidak correction). M) Schematic of fiber photometry recordings from PVH^Sim2 neurons during intragastric infusion. N) Mean (±SEM) response of PVH^Sim2 neurons to intragastric infusion of water (blue trace) and Ensure (orange trace), and to pellet drop (black trace). Ensure and pellet responses recorded after overnight fasting; water responses in the ad libitum fed state. Shading indicates 10-minute gastric infusion. (Ensure: N=8 trials from 4 mice; water: N=6 trials from 4 mice; pellet drop: N=6 trials from 3 mice). O) Mean±SEM activity in the early (first 5 minutes after infusion start/pellet drop); late (last 5 minutes of infusion); and post-infusion periods (5 minutes after end of infusion). Each point represents mean over trials for one mouse. (Early: ordinary one-way ANOVA; late/post: two-tailed paired t-test). **p<0.01; ***p<0.001; ns = not significant.

The rapid response of PVH^Sim2 neurons suggests that they may be sensitive to sensory stimuli that are predictive of food. To further investigate this hypothesis, in a subset of mice, we recorded video together with fiber photometry, and manually annotated the interactions of mice with food. PVH^Sim2 neuron activity dropped prior to the first bite of food, and even before the mouse made contact with the food pellet (Figures 2K and 2L). Thus, similar to ARC^AgRP neurons^7–9, PVH^Sim2 neurons are sensitive to pre-ingestive sensory cues that predict upcoming food intake.

Notably, the activity of PVH^Sim2 neurons gradually recovered following food presentation. This contrasts with the drop in ARC^AgRP neuron activity in response to food, which is sustained over at least tens of minutes after food consumption⁷. Thus, the initial fall in ARC^AgRP neuron activity is likely due, in part, to a reduction in excitatory input from PVH^Sim2 neurons, in addition to an increase in inhibitory input from DMH^Lepr/Glp1r and ARC^Bnc2 neurons^47–50. The sustained reduction in ARC^AgRP neuron activity, however, is likely mediated by other inputs, such as post-ingestive signals originating in the gastrointestinal tract^10,11,51,52. We therefore hypothesised that, unlike ARC^AgRP neurons, which are strongly inhibited by intragastric nutrient infusion^10,11, PVH^Sim2 neurons would be insensitive to gastrointestinal nutrient signals. To test this, we used fiber photometry to record PVH^Sim2 neuron activity during infusion of a complete liquid diet (Ensure) through a gastric catheter in the fasted state (Figure 2M). In contrast to the rapid response to food presentation, PVH^Sim2 neuron activity was unaffected in the first 5 minutes of Ensure infusion, and remained unchanged in the late and post-infusion phases (Figures 2N and 2O). Gastric distension can also inhibit food intake⁵³; but like ARC^AgRP neurons¹¹, stomach distension through intragastric infusion of water did not affect PVH^Sim2 neuron activity (Figures 2N and 2O). Likewise, we observed no PVH^Sim2 neuron response to intragastric infusion of saline or to a sham infusion (Figures S7A and S7B). PVH^Sim2 neurons were also insensitive to hormonal signals of energy state. Intraperitoneal injection of leptin in fasted mice did not affect PVH^Sim2 neuron activity (Figure S7C). Likewise, PVH^Sim2 neuron activity was unaffected by injection of ghrelin in the ad libitum fed state, which strongly activates ARC^AgRP neurons⁷ (Figure S7D). Thus, in stark contrast to ARC^AgRP neurons, PVH^Sim2 neurons are not sensitive to signals of food after ingestion, but rather detect external, pre-ingestive sensory cues that are predictive of food.

Removal of food rapidly increases ARC^AgRP and PVH^Sim2 neuron activity

ARC^AgRP neurons are bidirectionally sensitive to energy status: while food presentation inhibits their activity, overnight fasting strongly increases it^8,54. Using immediate early gene (c-Fos) expression as a readout of neuronal activity, we found that PVH^Sim2 neurons similarly showed increased activity following overnight fasting (Figure S7E). Thus, PVH^Sim2 neurons are likely to contribute to the fasting activation of ARC^AgRP neurons.

We next sought to determine the timescale over which fasting activates ARC^AgRP neurons. Traditionally, the activation of ARC^AgRP neurons is thought to be driven largely by feedback from the body as energy stores are gradually depleted^1,55,56. However, the PVH, where the Sim2+ inputs to ARC^AgRP neurons are located, is not thought to directly sense bodily feedback signals related to the fasted state, such as leptin⁵⁷. Moreover, the timescale over which fasting activates ARC^AgRP neurons is unclear. We recently developed a novel long-term fiber photometry approach to continuously measure population activity over days¹⁷. Using this methodology, we found that ARC^AgRP neurons show a remarkably rapid increase in activity when fasting is initiated by the removal of food¹⁷. This suggests that rapid feedforward neuronal inputs play a role in the response of ARC^AgRP neurons to fasting.

To confirm these fasting responses, we performed long-term fiber photometry recordings from ARC^AgRP neurons before, during, and after overnight fasting, and under ad libitum conditions (Figures 3A-D). We recorded activity from ZT6 (Zeitgeber Time 6; 6 hours after lights-on) until ZT5 the following day; food was removed at ZT12 (lights-out) and returned at ZT2 (2 hours after lights-on). As previously described^17,18, with ad libitum food, ARC^AgRP neuron activity gradually decreased over the dark period (Figure 3E). In the fasting condition, however, food removal caused a large, rapid, and sustained rise in ARC^AgRP neuron activity, within 1 hour of fasting onset (Figures 3E and 3F). This increase was complete within ~2 hours (time constant (τ) = 1.78 ± 0.40 h, mean±SEM). ARC^AgRP neurons respond before the body reaches negative energy balance: while 1 hour of fasting from lights-out significantly increased ARC^AgRP neuron activity, it did not significantly affect circulating ghrelin, insulin, or leptin (Figures S7G-I). Thus, ARC^AgRP neuron activity anticipates the metabolic consequences of food unavailability early in fasting.

Figure 3. PVH^Sim2 neurons mediate the rapid activation of ARC^AgRP neurons upon food removal.

A) Experimental design: activity is recorded from ZT6 to ZT5 the following day (23 hours), and in the fast condition, food is removed between ZT12 and ZT2 (14 hour fast).

B) Home-cage recording setup. C) Fiber photometry recording from ARC^AgRP neurons.

D) GCaMP6s expression (green) in ARC^AgRP neurons; dashed line indicates optic fiber.

E) ARC^AgRP neuron activity in the fast and ad libitum conditions (mean±SEM over mice; 30-minute moving mean). Vertical lines indicate start and end of fasting. (N=6 mice). F) Mean activity in 1-hour bins between ZT11 and ZT16. Two-way repeated measures ANOVA; post-hoc comparisons at each time point with Holm-Sidak correction. G, H) As (E, F) but for PVH^Sim2 neurons in Sim2–2a-FLPo mice. (N=12 mice). I) Fiber photometry recording from ARC^AgRP neurons (using Npy-IRES-FLPo) with chemogenetic inhibition of PVH^Sim2 neurons. J) Representative mCherry (red) and GCaMP6s (green) expression in PVH (left) and ARC (right). K) ARC^AgRP neuron activity (mean±SEM) in fast condition with injection of vehicle or CNO at ZT11.5. Inset: magnification of activity between ZT11 and ZT15. (N=6 mice). L) Time constant (τ) of increase in ARC^AgRP neuron activity for recordings in (K). Two-tailed paired t-test. M) As (F), but for recordings in (K). Two-way repeated measures ANOVA; post-hoc comparisons at each time point with Holm-Sidak correction. *p<0.05; **p<0.01; ***p<0.001. For boxplots, lines indicate individual mice; box and whiskers indicate upper/lower quartiles and minimum/maximum, respectively. Outliers indicated by symbol.

Given that PVH^Sim2 neurons provide excitatory input to ARC^AgRP neurons, we speculated that they may convey this rapid fasting response to ARC^AgRP neurons. To test whether PVH^Sim2 neurons are similarly sensitive to food removal, we recorded PVH^Sim2 neuron activity under the same conditions (Figures 3G and 3H). In the ad libitum state, PVH^Sim2 activity was dynamic, peaking in the early dark phase. Strikingly, following food removal at ZT12, PVH^Sim2 neuron activity increased rapidly (within 1 hour; τ = 0.87 ± 0.14 h (mean±SEM)). Activity remained elevated above the ad libitum state for at least 3 hours following food removal before dropping. Consistent with Figure 2, returning food at ZT2 led to a rapid drop in PVH^Sim2 neuron activity. The rapid fasting response is food-specific because similarly timed water deprivation did not increase ARC^AgRP or PVH^Sim2 neuron activity relative to the ad libitum condition (Figures S8A-D).

To confirm that removal of food early in the dark phase leads to a rapid increase in PVH^Sim2 neuron activity, we used immediate early gene (c-Fos) expression as an orthogonal readout of neuronal activity. Brief 2-hour fasting at the beginning of the dark cycle (ZT12–14) significantly increased c-Fos expression in PVH^Sim2 neurons relative to mice with ad libitum food access (Figure S7F). Together with the photometry results above, this indicates that food removal transiently increases PVH^Sim2 neuron activity over a period of hours, and suggests that these neurons contribute to the fasting activation of ARC^AgRP neurons.

In addition to excitatory input from the PVH, ARC^AgRP neurons also receive strong inhibitory drive from Lepr/Glp1r-expressing neurons in the ventral dorsomedial hypothalamus (vDMH^LepR)^47,50, and the rapid response of these neurons to food contributes to the inhibition of ARC^AgRP neurons by feedforward food-predictive cues⁴⁸. To test whether vDMH^LepR neurons might also contribute to increased ARC^AgRP neuron activity early in fasting, we monitored their activity under the same conditions (Figures S8E and S8F). As previously reported⁴⁷, vDMH^LepR activity strongly increased when food was presented at the end of fasting (Figure S8G). However, we observed no difference in vDMH^LepR activity between ad libitum and fasted states within the first 6 hours of fasting (Figures S8G and S8H), indicating that vDMH^LepR neurons do not contribute to the activation of ARC^AgRP neurons by food removal.

PVH^Sim2 neurons mediate the rapid ARC^AgRP neuron response to food removal

Given that the activity of both PVH^Sim2 and ARC^AgRP neurons increases in response to food removal, and that PVH^Sim2 neurons project to and activate ARC^AgRP neurons, we hypothesized that PVH^Sim2 neurons drive the ARC^AgRP neuron response to food removal. To test this, we inhibited PVH^Sim2 neuron activity while recording ARC^AgRP neuron activity (Figures 3I and 3J). In Sim2–2a-Cre::Npy-IRES-FLPo mice, we expressed the inhibitory chemogenetic receptor hM4Di in PVH^Sim2 neurons using a Cre-dependent AAV. We then used a FLP-dependent AAV to express GCaMP6s in ARC^AgRP neurons, and implanted an optical fiber above the ARC for fiber photometry recordings. We recorded ARC^AgRP neuron activity before, during, and after an overnight fast as described in Figure 3A, and injected CNO or vehicle at ZT11.5 (30 minutes before food removal). Following vehicle injection, ARC^AgRP neurons responded to food removal with a rapid, large increase in activity (Figures 3K-M). Inhibition of PVH^Sim2 neurons by CNO injection delayed this response, increasing the time constant (τ) by 1.80 ± 0.51 hours (mean±SEM). Likewise, inhibiting PVH inputs to ARC^AgRP neurons using the less-specific but broader Trh-IRES-Cre mouse line similarly delayed the increase in ARC^AgRP neuron activity (Figures S9A-E; Δτ = 2.03 ± 0.34 h (mean±SEM)). Thus, the rapid increase in ARC^AgRP neuron activity early in fasting is driven by excitatory input from PVH^Sim2 neurons.

ARC^AgRP neuron activity is essential for the high level of food intake after overnight fasting⁵⁸. Chemogenetic inhibition of PVH^Sim2 neurons, in contrast, did not affect food intake in the first 3 hours of refeeding (Figures S9F and S9G). This suggests that other factors (e.g. circulating hormones) may sustain ARC^AgRP neuron activity and drive food intake following long-term fasting. Notably, acute inhibition of PVH^Trh neurons does reduce feeding early in the dark cycle²⁷, suggesting that PVH afferents contribute to ad libitum food intake (see also Figure 6).

Figure 6. PVH^Sim2 neurons are necessary for normal food intake.

A) AAV-DIO-GFP-2a-TeNT injection into PVH of Sim2–2a-Cre mice. B) GFP expression (green) in PVH. C) Percentage change in body mass of males relative to pre-injection. (Two-way repeated measures ANOVA; post-hoc comparisons at each time point with Sidak correction). D) Fat mass as a percentage of total body mass in male mice at 13 weeks post-injection. (Two-tailed unpaired t-test). E) Mean weekly food intake of male mice from weeks 4–12 post-injection. (Two-tailed unpaired t-test). C-E: mean±SEM, N=5–6 mice per condition; thin lines/points represent individual mice. F) Mean hourly energy expenditure (over two days) vs body mass of males at 4 weeks post-injection per mouse. Lines represent linear regression fit for each group. (N=7 mice per condition; generalized linear model). G) Mean distance travelled per day (over two days) at 4 weeks post-injection. Box/whiskers represent upper/lower quartiles and minimum/maximum, respectively. (N=7; two-tailed unpaired t-test). H) Mean weekly food intake of female mice from weeks 4–12 post-injection. (Two-tailed unpaired t-test). I) Percentage change in body mass of females relative to pre-injection. (Two-way repeated measures ANOVA: no significant effect of AAV or interaction). J) Fat mass as a percentage of total body mass in female mice at 13 weeks post-injection. (Two-tailed unpaired t-test). H-J: mean±SEM, N=6–7 mice per condition; thin lines/points represent individual mice. K) Fat mass as a percentage of total mass in GFP-injected control male and female mice at 13 weeks post-injection. (Mean±SEM, N=5–7 mice per condition; two-tailed unpaired t-test).

In the experiments presented above, fasting always began at ZT12, when mice usually begin to eat at the highest rate¹⁷. We wondered whether the neuronal response to food removal might change if we removed food at different times relative to the light cycle. To test this, we removed food either 3 or 6 hours before the dark phase (ZT6 or ZT9, respectively), and recorded ARC^AgRP and PVH^Sim2 neuron activity (Figures S10A-C). Food removal at ZT9 did lead to small increases in ARC^AgRP and PVH^Sim2 neuron activity within the subsequent 3 hours, relative to the ad libitum state. However, no matter when food was removed, both ARC^AgRP and PVH^Sim2 neuron activity in the fasted state always peaked early in the dark phase (Figures S10D and S10E), with activity peaking around 2 hours later in ARC^AgRP than PVH^Sim2 neurons. The consistency in the timing of peak ARC^AgRP and PVH^Sim2 neuron activity irrespective of fast start time suggests that it is not the removal of food per se that most potently increases the activity of these neurons. Rather, the early dark phase is when mice usually eat at the highest rate, and have the highest motivational drive to find and consume food. Thus, this suggests that the unavailability of food at the time when mice have a strong drive to eat triggers the rapid increase in PVH^Sim2 and ARC^AgRP neuron activity.

Unsuccessful food-seeking increases PVH^Sim2 and ARC^AgRP neuron activity

The results above suggest that increased PVH^Sim2 and ARC^AgRP neuron activity early in fasting may be triggered by food unavailability at times when mice are actively seeking food. In other words, ARC^AgRP neuron activity may increase as mice realize that food is unavailable, because their search for food is unsuccessful. However, our prior experiments could not precisely identify when food seeking occurred and was unsuccessful. To establish when mice realize that food is unavailable, we measured food-seeking behavior during fasting. We trained mice to receive all their food via an operant task on a fixed-ratio 1 (FR1) schedule, where each poke leads to delivery of one small (20 mg) food pellet, using an operant device (FED3)⁵⁹. Then, we recorded ARC^AgRP neuron activity in mice either with continuous operant food access or throughout an overnight fast, during which mice could poke, but no pellets were delivered (Figures 4A and 4B).

Figure 4. Unsuccessful food seeking increases ARC^AgRP and PVH^Sim2 neuron activity.

A) Operant task using FED3: in ad libitum state, each nose poke releases one pellet. During fasting (ZT12-ZT2, 14 hour fast), mice poke but receive no pellets. B) Cumulative pellets retrieved in fast (green) and ad libitum (purple) conditions. C) Rolling mean poke rate in fast (dotted green) and ad libitum (dotted purple) conditions. D) Overlay of poke rate (dotted blue) and ARC^AgRP neuron activity (solid green, 30-minute moving mean) during the first 5h of fasting. E) Overlay of poke rate (dotted blue) and PVH^Sim2 neuron activity (solid green) during the first 5h of fasting. B-E: mean±SEM over mice. F,G) ARC^AgRP (F) and PVH^Sim2 (G) neuron activity aligned to pokes (mean±SEM over pokes). B,C,D,F: N=7; E,G: N=4 mice.

At fasting onset, mice poked vigorously, then poking gradually declined over hours (Figure 4C). This decline represents extinction – the weakening of operant nose-poking as mice learn it no longer yields food – while the initial high poke rate represents an extinction burst: an initial increase in operant behavior frequently observed early in extinction⁶⁰. This extinction burst indicates that mice have detected the change in contingency, while the decline in poke rate shows that they are learning the new contingency (i.e. that pokes no longer deliver food). As in our previous experiments, ARC^AgRP neuron activity increased early in fasting, and then remained elevated (Figure 4D). Notably, this increase coincided with the decline in food-seeking (nose-poking, Figure 4D). As such, ARC^AgRP neuron activity increases as mice enter a state of awareness that food is not available, and we speculate that this awareness drives the increase in ARC^AgRP neuron activity.

Given that PVH^Sim2 neurons are also active early in fasting, we reasoned that they may similarly respond to unsuccessful food-seeking. Like ARC^AgRP neurons, the activity of PVH^Sim2 neurons was also elevated during the initial burst of poking early in fasting, demonstrating that PVH^Sim2 neurons are also active during unsuccessful food-seeking (Figure 4E). However, ARC^AgRP neurons and PVH^Sim2 neurons showed different temporal dynamics: while ARC^AgRP neuron activity increased gradually as poke rate declined, and then remained elevated, PVH^Sim2 neuron activity increased more rapidly and was high throughout the period of high poke rate, but then declined together with the poke rate. This indicates that PVH^Sim2 neurons respond transiently to food unavailability, and suggests that ARC^AgRP neurons may use these transient signals to obtain a stable prediction of food availability.

To gain a closer understanding of how the awareness of food unavailability impacts ARC^AgRP and PVH^Sim2 neuron activity, we looked at their response to individual pokes. We aligned neural activity to pokes, selecting pokes only during the fasting or equivalent ad libitum period (ZT12-ZT2). We selected only pokes with no pokes during the preceding baseline period (45 s), although this filter did not affect the results (Figures S11A and S11B). Under ad libitum conditions, ARC^AgRP neuron activity dropped precipitously immediately preceding a poke, and then recovered to a level below its initial baseline (Figure 4F), consistent with ARC^AgRP neuron activity around free-feeding bouts¹⁷. During fasting, however, this initial drop was smaller, perhaps because the expectation of a poke leading to food was reduced; and notably, the unrewarded poke was followed by a large increase in ARC^AgRP neuron activity. This response to unrewarded pokes was fairly consistent throughout an overnight fast (Figure S11C). Thus, unsuccessful food-seeking behavior increases ARC^AgRP neuron activity.

Like ARC^AgRP neurons, PVH^Sim2 neuron activity dropped at the time of poking when food was available (Figure 4G). In contrast, during fasting PVH^Sim2 neuron activity increased with unrewarded pokes. This response occurred earlier than that in ARC^AgRP neurons; in fact, increased PVH^Sim2 neuron activity preceded pokes. This increased activity before poking developed gradually over fasting: early in fasting, PVH^Sim2 neuron activity increased only after poking, once food was not delivered; but this response shifted earlier as fasting progressed (Figure S11D). We speculate that later in fasting, PVH^Sim2 neurons show an anticipatory response reflecting the prediction that food will not be delivered. Thus, both PVH^Sim2 and ARC^AgRP neurons show increased activity in response to unsuccessful food seeking, with PVH^Sim2 neurons showing more rapid responses, particularly later in fasting.

PVH^Sim2 neurons receive diverse monosynaptic input from across the brain

Our previous data highlight that PVH^Sim2 neurons: i) are rapidly inhibited by sensory signals predictive of food; ii) rapidly increase in activity following unsuccessful food-seeking; and iii) are required for the rapid fasting activation of ARC^AgRP neurons (Figures 2–4). To understand how these features of PVH^Sim2 neuron activity might come about, we identified the brain regions that provide direct input onto PVH^Sim2 neurons. We performed monosynaptic rabies tracing from PVH^Sim2 neurons using the CVS-N2c rabies strain, which labels more starter cells and is less neurotoxic than the traditional SAD-B19 strain (Figures 5A and 5B)^61,62. We applied the QUINT workflow⁶³ to systematically map all traced cells onto the Allen Mouse Brain Common Coordinate Framework version 3 (CCFv3). We found that although ~33% of all inputs to PVH^Sim2 neurons come from within the hypothalamus, these neurons also receive diverse input from across the brain (Figure 5C). This includes areas like the cortex and hippocampus that, on the one hand, are atypical inputs for a homeostatic control neuron, while on the other hand, are consistent with PVH^Sim2 neurons responding to cognitively-processed contextual cues (Figure 5C; Table S1). Three areas are particularly striking given their association with such contextual cues. First, the anterior cingulate area (ACAv in Figures 5D and 5E, part of the prefrontal cortex) receives input from cortical sensory systems, updates associations as contingencies change, and selects behaviors optimal for changed contingencies⁶⁴. We also found rabies-traced neurons in other regions of the prefrontal cortex, notably layers 5 and 6 of the prelimbic and infralimbic areas (Figure 5E and Table S1). Second, the ventral hippocampus (CA1 and ProS in Figures 5D and 5E) processes contextual cues to regulate motivated behavior and emotion⁶⁵. Third, the lateral septum (LSr and LSv in Figures 5D and 5E) is a major relay for output from the ventral hippocampus⁶⁶. Of note, according to the ventral hippocampal subdivision proposed by Ding and colleagues^67,68, the rabies+ neurons in the ventral hippocampus, denoted as CA1 and ProS in Figure 5D, are most likely almost all within the prosubiculum (see Figure 5E, lower panels), which was previously not well distinguished in mice. Based on their detailed analysis of its inputs and outputs, this area is proposed to regulate motivated behavior and emotion⁶⁸.

Figure 5. PVH^Sim2 neurons receive diverse monosynaptic input from across the brain.

A) Experimental design. B) Representative helper virus expression (TVA-FLAG, cyan) and rabies-infected neurons (tdTomato, magenta). Starter neurons co-expressing TVA-FLAG and tdTomato marked by white arrows. C) Labeled neurons by gross anatomical area as percentage of all labeled neurons, per Allen Mouse Brain CCFv3, using QUINT workflow⁶³. D) As (C), but by fine anatomical area (top 30 regions shown). C,D: N=3 mice, mean±SEM. E) Rabies-infected neurons in selected sites (tdTomato-positive, magenta). F) Top: tdTomato+ rabies-infected neurons (magenta) and Npy-hrGFP (cyan) in ARC. Bottom: tdTomato+ and Npy-hrGFP+ neuron quantification in ARC across 2 mice. G) Left, upper: AAV-hSyn-hChR2(H134R)-eYFP injections into ProS-v and AAV-hSyn-DIO-mCherry into PVH of a Sim2–2a-Cre mouse. Left, lower: resultant mCherry+ cell bodies and ChR2-eYFP+ fibers in PVH; and ChR2-eYFP in ProS-v. Middle: example recording from mCherry+ PVH^Sim2 neuron with optogenetic stimulation of ProS-v fibers in aCSF alone (top), or with TTX (middle), or with TTX and 4-AP (bottom). Blue squares show light pulse. Right: monosynaptic connection probability from ProS-v fibers to mCherry+ (N=8 mice, 31/60 neurons connected) and mCherry- (N=4 mice, 1/15 neurons connected) PVH neurons. (Mann-Whitney U test). *p<0.05. Abbreviations: ACAv: anterior cingulate area, ventral part, layer 5; AH: anterior hypothalamic nucleus; ARC: arcuate hypothalamic nucleus; BNST: bed nuclei of the stria terminalis; CA1: Field CA1; CB: cerebellum; CEAm: central amygdalar nucleus, medial part; CTX: cortex; DMH: dorsomedial nucleus of the hypothalamus; FT: fiber tracts; HB: hindbrain; HC: hippocampus; HY: hypothalamus; LH: lateral hypothalamic area; LPOA: lateral preoptic area; LSr: lateral septal nucleus, rostral part; LSv: lateral septal nucleus, ventral part; MB: midbrain; MEA: medial amygdalar nucleus; MPN: medial preoptic nucleus; MPOA: medial preoptic area; MRN: midbrain reticular nucleus; MY: medulla; NAc: nucleus accumbens; OLF: olfactory areas; PA: posterior amygdalar nucleus; PAG: periaqueductal gray; PAL: pallidum; PBN: parabrachial nucleus; PH: posterior hypothalamic nucleus; PMv: ventral premammillary nucleus; ProS: prosubiculum; PVp: periventricular hypothalamic nucleus, posterior part; SI: substantia innominata; STR: striatum; TH: thalamus; TN: tuberal nucleus; VMH: ventromedial hypothalamic nucleus; VS: ventricular system; ZI: zona incerta. See also Supplementary Table 1. In D) HY (other), MB (other) and STR (other) refer to neurons in those gross anatomical regions but not within finer subregions as determined by QUINT.

Other top regions providing input to PVH^Sim2 neurons have previously been implicated in the control of feeding behavior, including the bed nucleus of the stria terminalis (BNST), medial amygdala (MeA), lateral hypothalamus (LH), parabrachial nucleus (PBN) and ARC (Figure 5D; Table S1)^2,69–76. Two of the mice included in this experiment also carried the Npy-hrGFP allele, which allowed us to examine whether ARC^AgRP neurons provide input to PVH^Sim2 neurons. Co-localization analysis between Npy-hrGFP and td-Tomato+ monosynaptic inputs to PVH^Sim2 neurons in the ARC revealed very little overlap (Figure 5F), confirming our in vivo functional connectivity studies showing that ARC^AgRP neuron activation does not influence PVH^Sim2 neuron activity (Figure S5B). The molecular identity of the ARC neuron subtype that provides input to PVH^Sim2 neurons remains an open question.

Remarkably, one of the top individual input sites for PVH^Sim2 neurons was the ventral hippocampus (likely the ProS, Figure 5D and 5E; Table S1). Given that this site is an unexpected input for a homeostatic drive control neuron, we confirmed this finding using CRACM. We injected Cre-independent ChR2 into the ventral ProS (ProS-v), and labelled PVH^Sim2 neurons by injecting Cre-dependent mCherry into the PVH of Sim2–2a-Cre mice (Figure 5G, left). Activating ChR2-expressing ProS-v axons in ex vivo brain slices resulted in light-evoked excitatory post-synaptic currents in PVH^Sim2 neurons, which were abolished by TTX and reinstated by co-application of 4-AP, confirming their monosynaptic nature (Figure 5G, middle). leEPSCs were observed in 31/60 PVH^Sim2 neurons, and in only 1/15 neighboring Sim2-negative PVH neurons (Figure 5G, right). This indicates that ProS-v sends highly specific projections to PVH^Sim2 neurons, and not other PVH neuron types. Given this strong confirmation of specific ProS-v→PVH^Sim2 connectivity, and the presumed function of ProS, it is likely that contextual information about food availability is routed, at least in part, through the ProS-v to the PVH^Sim2→ARC^AgRP circuit.

PVH^Sim2 neurons are necessary for normal food intake

We next asked how the ongoing activity of PVH^Sim2 neurons contributes to long-term energy balance in mice with ad libitum access to standard chow. To do so, we chronically silenced synaptic output from PVH^Sim2 neurons by injecting AAV-DIO-GFP-2a-TeNT into the PVH of Sim2–2a-Cre mice (Figures 6A and 6B). We then measured body weight and food intake weekly over 12 weeks, relative to control (GFP-only) mice. While control males gained weight throughout the experiment, silencing PVH^Sim2 output drastically reduced body weight gain (Figure 6C). Moreover, at the end of this experiment, PVH^Sim2-TeNT males had reduced body fat relative to controls (Figure 6D). Consistent with the ability of these neurons to drive food intake when activated (Figure 1K), we found that chronic silencing of PVH^Sim2 output persistently reduced food intake (Figures 6E and S12A).

To test whether the reduced body weight of PVH^Sim2-TeNT male mice might be due to this reduction in food intake, or to metabolic alterations, we used indirect calorimetry to measure various metabolic parameters at 4 weeks post-injection (when the weight difference is beginning to develop, Figure S12B). Silencing PVH^Sim2 output greatly reduced energy expenditure (Figures 6F and S12C) and locomotor activity (Figures 6G and S12D), without affecting substrate utilization (measured by respiratory exchange ratio, Figure S12E). All else being equal, this reduced metabolic activity would be expected to increase weight gain – in contrast to the reduced weight gain observed in PVH^Sim2-TeNT males. Thus the reduced weight following PVH^Sim2 silencing is not due to metabolic alterations, but rather to reduced food intake. Consistent with this, we found a significant reduction in feeding rate in the early dark cycle in PVH^Sim2-TeNT males relative to GFP controls in this experiment (Figures S12F and S12G). Of note, reduced energy expenditure is unlikely to be caused by reduced ARC^AgRP neuron activity, since this is expected to increase energy expenditure^4,77. Instead, we propose that this reduction in energy expenditure may be a metabolic adaptation to minimize negative energy balance in the face of reduced energy consumption.

We also measured the effect of PVH^Sim2 synaptic silencing on body weight and food intake in female mice. As in males, chronic silencing of PVH^Sim2 output greatly reduced food intake (Figures 6H and S12H). Unlike in males, however, this reduced food intake did not translate to a statistically significant reduction in body weight or body fat (Figures 6I and 6J). This is likely due to a concurrent reduction in energy expenditure (as seen in males, Figure 6F), which was perhaps more effective in maintaining body weight due to the greater compensatory responses of these female mice to defend their much lower body fat content (Figure 6K: GFP controls measured at 13 weeks post-injection). Taken together, PVH^Sim2 neuron activity is essential for normal levels of food intake in ad libitum-fed mice.

Discussion

ARC^AgRP neurons are thought to be gradually activated by circulating factors^1,2, but they also receive direct neuronal input, including significant input from excitatory PVH neurons²⁷. However, the precise molecular identity of PVH inputs to ARC^AgRP neurons, the information that they convey, and their role in the regulation of ARC^AgRP neuron activity and food intake, has been unknown.

Here, we show that this PVH→ARC^AgRP circuit conveys cognitively-processed contextual information to increase ARC^AgRP neuron activity, and is important both for the rapid response of ARC^AgRP neurons to anticipated energy deficit, and for ad libitum food intake. We first interrogated a single-cell/nucleus RNA-seq atlas to identify the PVH transcriptomic subtype providing this excitatory drive to ARC^AgRP neurons. PVH neurons expressing Sim2 synapse onto and excite ARC^AgRP neurons, and drive food intake. Moreover, PVH^Sim2 neuron activity is bidirectionally regulated by food availability: food presentation to a fasted mouse markedly decreases PVH^Sim2 neuron activity over minutes (Figure 2), whereas removal of food when mice would usually eat sharply increases PVH^Sim2 neuron activity over 30–60 minutes (Figure 3). This latter response is triggered by unsuccessful food seeking (Figure 4), perhaps explaining why it is most prominently seen at times mice are more likely to eat (Figure S10). We demonstrate that PVH^Sim2 neuron activity is required for the rapid, feedforward activation of ARC^AgRP neurons early in fasting. Importantly, PVH^Sim2 neuron activity is not regulated by gut nutrients, gastric distension, leptin, or ghrelin (Figures 2 and S7). Instead, the activation of PVH^Sim2 neurons by cognitively-processed cues likely relates to their input from “higher-order” brain regions atypical of homeostatic control neurons, including prefrontal cortex, hippocampus, and lateral septum (Figure 5). Finally, at longer timescales, PVH^Sim2 neuron activity is required to support ad libitum food intake (Figure 6).

The neuron population identified here is transcriptomically, anatomically, and functionally unique among PVH neurons. While essentially all PVH neurons express the Sim1 transcription factor³⁴, its paralogue Sim2 is highly specific to a subset of anterior PVH neurons co-expressing Trh and Adcyap1 (Figures 1C-E). This anterior localization contrasts with other Trh-expressing PVH populations⁷⁸. Moreover, PVH^Sim2 neurons largely project to hypothalamic and forebrain targets, with few caudal projections (Figures 1 and S4). This contrasts with most previously described brain parenchyma-projecting PVH subpopulations, which project heavily to midbrain, brainstem, and spinal cord^{24–26,79–85}. Interestingly, most of these subpopulations also suppress food intake when activated (for example, those expressing Mc4r, Pdyn, TrkB, Glp1r, Calcr, Irs4, Avp, and Nos1)^{24–26,79–81,84,86}, contrasting with PVH^Sim2 neurons, which increase food intake. Of note, PVH^Sim2 neurons project to multiple targets that may contribute to the behavioral effects of PVH^Sim2 neuron activation/inhibition, beyond ARC^AgRP neurons.

Typical for PVH neurons, PVH^Sim2 neurons are glutamatergic. Additionally, they express multiple neuropeptide genes, including Trh and Adcyap1 (encoding PACAP, Figure 1D), suggesting they may regulate ARC^AgRP neuron activity through peptidergic as well as glutamatergic transmission, particularly since PACAP increases ARC^AgRP neuron firing²⁷. Notably, Sim2 likely plays a developmental role in this subpopulation, since knockout of Sim2 reduces Trh expression in anterior PVH³⁵. The role of Sim2 expression in these neurons during adulthood remains unknown.

Predicting future energy states allows the brain to anticipate and prevent deviations from homeostasis before they occur⁸⁷. Several feedforward signals regulate ARC^AgRP neuron activity in anticipation of future energy deficit or surfeit. First, following fasting, food presentation causes a rapid and sustained drop in ARC^AgRP neuron activity, driven by external sensory cues predictive of food, the taste of food in the mouth, and post-ingestive nutrient detection in the gastrointestinal tract^{7–11,51,52,88}. Inhibitory inputs to ARC^AgRP neurons from both DMH (DMH^Lepr/Glp1r neurons) and ARC (ARC^Bnc2 neurons) show increased activity in response to food-predictive sensory cues^47–50. Our results suggest that a coordinated increase in inhibitory input with a concomitant decrease in excitatory input from PVH^Sim2 neurons likely mediates the rapid feedforward inhibition of ARC^AgRP neurons.

A second component of feedforward ARC^AgRP neuron regulation is their rapid increase in activity over the first hour of fasting (Figure 3 and 17), well before bodily energy stores are reduced. We show that this increased activity is likely driven by unsuccessful food-seeking, which signals food unavailability (Figure 4). This ARC^AgRP neuron responses is driven by PVH^Sim2 neurons, which are also rapidly activated by food unavailability, since PVH^Sim2 inhibition delays ARC^AgRP activation. Notably, PVH^Sim2 neurons respond more rapidly than ARC^AgRP neurons to unsuccessful food-seeking, both at gross (tens of minutes; Figures 4D, 4E and S10), and fine (seconds; Figures 4F and 4G) timescales. We interpret ARC^AgRP neuron activity as a low-pass filter of PVH^Sim2 neuron activity, accumulating PVH^Sim2 input over time. Two potential mechanisms may be co-release of glutamate and PACAP from PVH^Sim2 neurons (Figure 1), with PACAP acting more slowly to gradually increase ARC^AgRP neuron excitability²⁷; and activity-dependent synaptic potentiation, previously demonstrated at the PVH→ARC^AgRP synapse (see below)²⁸.

While PVH^Sim2 neurons are required for the initial ARC^AgRP response to fasting, we cannot definitively rule out a role in sustaining high ARC^AgRP neuron activity later in fasting, since a single CNO injection is unlikely to inhibit PVH^Sim2 neurons throughout a 14-hour fast⁸⁹. However, similarly to the inhibition of ARC^AgRP neurons upon food presentation and consumption, the response to fasting likely involves both fast- and slow-acting signals, with feedforward PVH^Sim2 input being the first, and slower signals subsequently sustaining ARC^AgRP neuron activity. These may include circulating signals of energy deficit²; synaptic potentiation (see below)²⁸; and/or yet-unidentified neuron populations.

What is the functional role of rapid PVH^Sim2/ARC^AgRP neuron activation following food removal? ARC^AgRP neuron activation drives food-seeking and consumption⁴, so this rapid fasting response would be expected to increase efforts to seek food. Food-seeking in anticipation of predicted energy deficit could allow mice to find and consume food before reaching this deficit, facilitating the maintenance of energy homeostasis. Additionally, ARC^AgRP neurons drive numerous physiological fasting responses, including reduced energy expenditure, hypothalamic-pituitary-adrenal axis activation, peripheral insulin sensitivity, and liver autophagy^4,5,90,91. Inducing these responses in anticipation of energy deficit may help to conserve energy stores and maintain blood glucose as energy deficit develops during extended fasting. Notably, such anticipatory responses to predict and prevent deviations from homeostasis are widespread across species⁸⁷.

Increased activity in PVH inputs to ARC^AgRP neurons has been shown to drive synaptic plasticity by increasing the number of synaptic contacts from PVH^Trh to ARC^AgRP neurons²⁸. Likewise, fasting similarly increases the strength of this connection, an effect that depends on NMDA receptor expression in ARC^AgRP neurons^28,29. The substantial increase in PVH^Sim2 neuron activity we observe early in the dark phase during fasting (Figure 3) raises the intriguing possibility that this high PVH^Sim2 activity may be responsible for driving synaptic plasticity. Of note, PVH^Sim2 activity peaks earlier in fasting than ARC^AgRP neuron activity (Figure S10), consistent with the idea that strengthening of PVH^Sim2→ARC^AgRP neuron connections may contribute to the early response of ARC^AgRP neurons to fasting. This strengthened PVH^Sim2→ARC^AgRP neuron connection could contribute to sustaining high ARC^AgRP neuron activity throughout fasting, even as the activity of PVH^Sim2 neurons drops. Notably, our sc/snRNA-seq data indicates that PVH^Sim2 neurons express a number of mediators of excitatory synaptic plasticity at high levels, including Bdnf and Cbln2, a result that we confirmed with smFISH (Figure S13). Indeed, within the PVH, Cbln2 expression is largely selective to PVH^Sim2 neurons. Future studies will determine whether PVH^Sim2 neuron activity early in fasting drives plasticity at PVH^Sim2→ARC^AgRP neuron synapses, the molecular mechanisms responsible for this plasticity, and its role in sustaining ARC^AgRP neuron activity during fasting.

A third feedforward input to ARC^AgRP neurons is the circadian clock, which regulates ARC^AgRP neuron activity by time of day to appropriately time feeding^17,18. We similarly observed variability in PVH^Sim2 neuron activity across the day under ad libitum conditions, peaking in the early dark phase (Figure 3). However, PVH^Sim2 neurons are unlikely responsible for circadian regulation of ARC^AgRP neuron activity, since ablating PVH^Trh neurons did not affect ARC^AgRP neuron rhythmicity under ad libitum conditions¹⁸. Indeed, we recently demonstrated that circadian rhythmicity in ARC^AgRP neuron activity is mediated by excitatory DMH input, which transmits circadian signals originating in the suprachiasmatic nucleus¹⁷. This is consistent with PVH^Sim2 neurons being dispensible for circadian regulation of ARC^AgRP neuron activity under ad libitum conditions, but instead playing a specific role in the rapid response of ARC^AgRP neurons when food is desired but unavailable.

Beyond fasting, our results demonstrate that PVH^Sim2 neurons also regulate long-term ad libitum food intake (Figure 6), since synaptic silencing of PVH^Sim2 neuron output markedly reduced food intake in both sexes. While chronic silencing or ablation of ARC^AgRP neurons in adult mice does not affect body weight gain or ad libitum food intake^58,92, reduced excitatory input to ARC^AgRP neurons by AgRP neuron-specific NMDA receptor knockout substantially reduces body weight and food intake²⁹. This suggests that reduced excitatory input to ARC^AgRP neurons likely accounts for reduced food intake in mice lacking PVH^Sim2 neuron output. Given the signals transmitted by PVH^Sim2 neurons (Figures 2–4), this suggests the intriguing possibility that cognitively-processed, feedforward cues play an important role in driving food intake under ad libitum conditions.

In summary, we identified a highly specific PVH neuron subtype that drives food intake and is essential to long-term energy balance. These PVH^Sim2 neurons transmit bidirectional anticipatory signals to enable ARC^AgRP “hunger” neurons to predict future changes in energy balance, both when food is unexpectedly available, and when desired food is unavailable. Our findings demonstrate the importance of cognitively-processed, feedforward information in controlling hunger and energy balance, and provide an entry point to understand how the brain computes these feedforward signals. These results also identify a possible new neuronal target for cell-type-specific therapies that could potentially sustain reduced body weight after dieting.

Limitations of the study

While our study pinpoints the subset of PVH neurons providing input to ARC^AgRP neurons, we cannot exclude the existence of multiple PVH^Sim2 subpopulations with distinct functions. The viral approaches used activate/silence all PVH^Sim2 neurons, not only those responsive to feedforward/predictive signals. Additionally, we do not address all factors contributing to increased ARC^AgRP neuron activity during fasting. These likely include many factors besides PVH^Sim2 neurons, particularly later in fasting – although we cannot rule out PVH^Sim2 neurons contributing to sustained ARC^AgRP neuron activity later in fasting, due to limitations of our neuronal silencing approach. Lastly, we compared male and female mice in each experiment and observed no clear sex differences besides those described for chronic silencing in Figure 6. However, we cannot conclusively demonstrate an absence of sex differences due to the relatively low number of mice per sex. Given the apparent absence of sex difference in initial cohorts, we combined results between sexes for all experiments except Figure 6.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by Bradford B. Lowell (blowell@bidmc.harvard.edu).

Materials availability

Knock-in mouse lines will be deposited at Jackson Laboratory.

Data and code availability

• All photometry and metabolic data has been deposited on GitHub: https://github.com/sjwalker223/Walker_et_al_2026/ https://doi.org/10.5281/zenodo.20059773. All other data used to produce the graphs is in files Data S1 and Data S2.

• All Python code used to analyze the photometry and metabolic data is available in the same GitHub repository: https://github.com/sjwalker223/Walker_et_al_2026/ https://doi.org/10.5281/zenodo.20059773.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals

All animal care and experimental procedures were approved in advance by the National Institutes of Health and the Beth Israel Deaconess Medical Center Institutional Animal Care and Use Committee. Mouse health was monitored daily, and mice were housed in a humidity- and temperature-controlled environment (22–24°C) with a 12 h light / 12 h dark cycle and ad libitum access to standard chow (LabDiet 5001) and water, unless otherwise specified. All lines were maintained on a mixed genetic background, and all experimental mice were heterozygous for the allele(s) of interest. For experiments involving across-animal comparisons, littermates of the same sex were randomly assigned to experimental or control groups. Both male and female mice were used for all studies. Results for male and female mice were compared for each experiment, and no clear sex differences were observed besides those described for chronic silencing in Figure 6. However, we cannot conclusively demonstrate an absence of sex differences due to the relatively low number of mice per sex. Given the apparent absence of sex difference in initial cohorts, we combined results between sexes for all experiments except Figure 6. The following mouse lines were obtained from Jackson Laboratory: Npy-IRES-FLPo⁹³ (#030211), Lepr-IRES-Cre⁹⁴ (#032457). The following mouse lines have been described previously: Npy-hrGFP⁹⁵, Agrp-IRES-Cre⁷⁷, Trh-IRES-Cre²⁷, Pacap-IRES-Cre²⁷, Trh-2a-Dre²⁸. Ucn3-Cre was obtained from MMRRC (#032078-UCD) through Catherine Dulac.

Generation of mouse lines

Sim2–2a-Cre and Sim2–2a-FLPo mouse lines were generated using Easi-CRISPR⁹⁶. A p2a-Cre or p2a-FLPo cassette was inserted immediately upstream of the stop codon in the Sim2 gene. To achieve this, fertilized eggs from FVB dams were injected with CAS9 protein (PNA Bio, CP01), sgRNA, and single-stranded DNA template (ssDNA) containing the p2a-recombinase cassette in addition to 100nt 5’ and 3’ homology arms. ssDNA sequences can be found in Figure S2. ssDNA was synthesized by Genscript (Sim2–2a-Cre) or IDT (Sim2–2a-FLPo). sgRNA was synthesized by Genscript, with the following sequence:

AGGGCCAGUCCCGCUCGGCA

Samples from the resultant offspring were screened for cassette insertion by PCR, using two pairs of primers targeting short amplicons straddling the 5’ and 3’ ends of the inserted cassette. Dimethyl sulfoxide (DMSO, 5%) was added to PCR reactions due to high G/C content. Positive samples were then amplified using two pairs of primers that each target a ~1kb amplicon straddling the 5’ or 3’ end of the inserted cassette, with the two amplicons overlapping in the middle of the cassette. The resultant cDNA samples were verified for sequence integrity using Sanger sequencing (Eton Biosciences). One founder mouse was selected for each line and bred to wild-type mice of mixed background.

Primer pairs used for screening offspring were as follows:

Sim2–2a-Cre:

Short, left arm: TCGTGCTGCTCAACTACCAC and CATGTCCATCAGGTTCTTGC

Short, right arm: CGCTGGAGTTTCAATACCGG and CCAACCCTCTACACCAGCAT

Long, left arm: TCGTGCTGCTCAACTACCAC and AACCAGCGTTTTCGTTCTGC

Long, right arm: AGCCGAAATTGCCAGGATCA and CCAACCCTCTACACCAGCAT

Sim2–2a-FLPo:

Short, left arm: TCGTGCTGCTCAACTACCAC and TCGAACTGGCTCATCACCTT

Short, right arm: GAGCAGCTACATCAACAGGC and CCAACCCTCTACACCAGCAT

Long, left arm: TCGTGCTGCTCAACTACCAC and ACTGAATGATCACGCCCAGG

Long, right arm: CCACATTCATCAACTGCGGC and CCAACCCTCTACACCAGCAT

METHOD DETAILS

Single molecule fluorescent in situ hybridization (RNAscope)

Mice were terminally anaesthetized using 7% chloral hydrate (500mg kg⁻¹; Sigma Aldrich) diluted in saline (0.9%), and transcardially perfused, first with 0.1 M phosphate-buffered saline (PBS), then with 4% paraformaldehyde (PFA). Brains were extracted and post-fixed overnight at 4°C in 4% PFA, and then cryopreserved in 15% and 30% sucrose for one day each at 4°C. The following day, brains were cut into 40 μm sections on a freezing microtome (Leica Biosystems). The sections were washed in PBS containing 0.5% Triton X-100, mounted on microscope slides (Fisherbrand Superfrost Plus), and dried at room temperature (RT) for one hour. Sections were then fixed for 15 minutes in 4% PFA at 4°C, washed in PBS, dehydrated in an ethanol series, and then air-dried. A barrier was drawn around the sections using an ImmEdge Hydrophobic Barrier Pen. The sections were then incubated in Protease III for 30 minutes at RT, then incubated with target probes: C3-Sim2 (Cat# 1110401-C3), C2-Trh (Cat# 436811-C2), C1-Adcyap1 (Cat# 405911), C1-Slc17a6 (Cat# 319171), C2-Bdnf (Cat# 424821), and C1-Cbln2 (Cat# 428551) for two hours at 40°C. Slides were stored overnight at 4°C in 5x SSC before being treated with Amp 1–3 at 40°C. Probes were visualized using HRP-C1/2/3 and TSA-fluorophores: Cf488a, Cf568, and Cf640r (Biotium). Sections were counterstained with DAPI, coverslipped with mounting medium (SouthernBiotech #0100–01) and imaged using a confocal microscope (Zeiss LSM 880). DAPI-labelled cells containing each of the probes were counted manually using QuPath or FIJI.

Stereotactic surgical procedures

Stereotactic injections were performed as previously described⁵. Mice were anesthetized by injection of xylazine (10 mg kg⁻¹) and ketamine (100 mg kg⁻¹) diluted in sterile saline (0.9%), and their heads fixed into a stereotaxic apparatus (Kopf). A skin incision was used to expose the skull, and a small drill hole was made above the injection site. AAVs were injected using a pulled glass micropipette (20–40 μm tip diameter) at the following coordinates (from Bregma):

PVH – AP: −0.70, ML: ±0.2, DV: −4.85

ARC – AP: −1.65, ML: ±0.3, DV: −5.90 and −5.80

vDMH – AP: −1.80, ML: ±0.3, DV: −5.20

BNST – AP: +0.40, ML: −0.55, DV: −4.15

Virus delivery was controlled by an air pressure system using picoliter air puffs through a solenoid valve (Clippard EV 24VDC) pulsed by a Grass S48 stimulator. The pipette was removed 5 minutes post-injection, and the incision was closed using VetBond (unless an optic fiber was also implanted – see below). Post-operative analgesia was provided by subcutaneous injection of sustained-release Meloxicam (4 mg kg⁻¹). All subjects determined to be “misses” based on little or absent reporter expression at the target site upon histological examination were excluded from analyses.

Viruses

The viruses were injected at the following specified volumes (per hemisphere). The following viruses were obtained from Addgene: AAV9-CAG-Flex-rev-ChR2-tdTomato (#18917 – 35nl, unilateral), AAV5-Syn-Flex-rc[ChrimsonR-tdTomato] (#62723 – 35nl, unilateral), AAV8-EF1α-fDIO-GCaMP6s (#105714 - PVH: 35nl; ARC: 200nl per site, unilateral), AAV8-hSyn-DIO-hM3Dq-mCherry (#44361 – 35nl, bilateral), AAV1-Syn-Flex-GCaMP6s (#100845 – 200nl per site, unilateral), AAV8-hSyn-DIO-hM4Di-mCherry (#44362 – 50nl, bilateral) and AAVretro-EF1α-fDIO-Cre(HA) (#121675 - BNST: 50nl, unilateral; ARC: 75nl at −5.90 DV and 25nl at −5.80 DV). The following viruses were obtained from the Stanford Gene Vector and Virus Core (RRID:SCR_023250): AAVDJ-CMV-DIO-eGFP-2a-TeNT (GVVC-AAV-71 – 50nl, bilateral) and AAVDJ-CMV-DIO-eGFP (GVVC-AAV-12 – 50nl, bilateral). AAV8-EF1α-DIO-Syp-mCherry (35nl, unilateral) was obtained from the MGH Vector Core Facility, Charlestown, MA, USA. Plasmid DNA for pAAV-DIO-EF1a-TVA-2A-N2cG was obtained from Addgene (#172360) and packaged in-house into an AAV9 vector (50nl, bilateral). Rabies virus Rbv-CVS-N2c-dG-tdTomato(EnvA) was obtained from UNC Neurotools (50nl, bilateral).

AAV8-nEF-Cre^ONDre^ON-hM3Dq-mCherry was generated by first inserting FREX from an AAV-CAG-FREX vector (kindly provided by T. Badea) into a pAAV-nEF backbone⁹⁷ at BamHI and EcoRI sites. We designed an hM3Dq-mCherry sequence split into 3 exons with synthetic introns, and where the central exon was double-floxed in reverse orientation, following the approach of ⁹⁷. This split-hM3Dq-mCherry sequence was inserted into the vector in reverse orientation between the double inverse rox12 and roxP sites at AscI and NheI sites. We verified that the resultant AAV drove mCherry expression in the presence of both Cre and Dre, but no expression in the presence of Cre alone, and minimal expression in the presence of Dre alone (Figure S1B).

Optic fiber implants

For fiber photometry experiments, following AAV injection, an optic fiber (400 μm core diameter, 0.5 NA, multimode, Thorlabs) affixed into a metal ferrule (Precision Fiber Products) was implanted 100 μm dorsal to the AAV injection site (ARC – AP: −1.65, ML: −0.3, DV: −5.75; PVH – AP: −0.70, ML: −0.2, DV: −4.75; vDMH: AP: −1.80, ML: −0.3, DV: −5.10). The skull and ferrule were scored to improve adhesion, and the ferrule was affixed to the skull using dental acrylic mixed with glue (Krazy Glue). Fiber tip location was determined upon post-mortem histological examination.

Chemogenetic food intake studies

Experimental mice were injected with AAV driving expression of excitatory (hM3Dq) or inhibitory (hM4Di) chemogenetic receptors in PVH. At least 2 weeks later, animals were subjected to sham intraperitoneal (IP) injections and extensive handling every day for at least 1 week prior to food intake studies. All food intake studies were conducted starting at ZT3 (9am), when ad libitum fed mice have low drive to feed. Each animal was subjected to at least one trial with IP vehicle (1% DMSO in 0.9% saline) injection, and the same number of trials with IP CNO (1 mg kg⁻¹ in 1% DMSO, 0.9% saline) injection. Where more than one trial was performed per condition, the mean over trials was calculated for each mouse.

For experiments involving hM3Dq, mice were moved to a new cage with food the preceding afternoon to minimize the amount of food in bedding. The next morning, 15 minutes prior to the start of the food intake study, vehicle or CNO was injected, and food was removed from the cage. Food was placed in the hopper at ZT3 and weighed every hour for 3 hours. For experiments with AAV8-nEF-Cre^ONDre^ON-hM3Dq-mCherry, food was weighed at 8 hours.

For experiments involving hM4Di, mice were fasted by moving to a new cage without food at ZT12 the preceding day. Vehicle or CNO was injected at ZT2 (i.e., one hour prior to the introduction of food. Food was placed in the hopper at ZT3, and weighed every hour for 3 hours.

Immunofluorescence

Mice were terminally anesthetized using 7% chloral hydrate (500mg kg-1; Sigma Aldrich) diluted in saline (0.9%), and transcardially perfused, first with 0.1 M phosphate-buffered saline (PBS), then with 10% neutral-buffered formalin (NBF; Thermo Fisher Scientific). Brains were extracted and post-fixed overnight at 4°C in 10% NBF, and then cryopreserved overnight at 4°C in 20% sucrose. Brains were then cut into 40–50 μm coronal sections using a freezing microtome (Leica Biosystems) and stored at 4°C in PBS until staining. Subsequent steps were performed at room temperature. Sections were washed (4 × 3 minutes) in PBS, and then incubated for 1 hour in blocking solution (5% normal donkey serum and 0.5% Triton X-100 in PBS). Sections were then incubated overnight with the following primary antibodies diluted in blocking solution: Rat anti-mCherry (Invitrogen #M11217, 1:3000), Rabbit anti-GFP (Invitrogen #A11122, 1:3000), Rabbit anti-fluorogold (Millipore #AB153-I, 1:400) and/or Rabbit anti-c-Fos (Synaptic Systems #226–003 [since discontinued], 1:1000). The next day, sections were washed (4 × 3 mins) in PBS, then incubated for 1–2 hours with the following Alexa Fluor-conjugated Donkey secondary antibodies at 1:1000 concentration in blocking solution: anti-rat 594 (Invitrogen #21209), and anti-rabbit 488 (Invitrogen #A21206). Finally, sections were washed (4 × 3 mins) in PBS, mounted on gelatin-coated slides, air-dried, mounted using mounting medium (SouthernBiotech #0100–20) containing DAPI, and covered with a coverslip. Fluorescent images were captured using an Olympus VS120 slide-scanning microscope.

Thick tissue immunofluorescence

700 μm-thick coronal sections through the hypothalamus were obtained using a microtome (Leica VT1000S). Sections were cleared using CUBIC Trial Kit (Code No. 290–80801)⁴² and processed as in⁹⁸. Briefly, sections were immersed in ScaleCUBIC-1 diluted 1:1 with MQ-H₂O overnight in a 37°C incubator on a shaker. The tissue was then placed in ScaleCUBIC-1 for two days in a 37°C incubator on a shaker. Once the sections were sufficiently cleared, they were washed overnight at RT in PBS + 0.2% Triton X-100. Next, we incubated the sections with the primary antibody (Rat anti-mCherry, Invitrogen #M11217, 1:500) diluted in modified blocking solution (PBS with 10% Triton X-100, 5% normal horse serum and 300mM NaCl) in a 37°C incubator on a shaker for 4 days. Sections were washed overnight at RT in PBS + 0.2% Triton X-100 before being incubated in the secondary antibody (anti-rat 594, Invitrogen #21209, 1:500) and DAPI in the same blocking solution as above in a 37°C incubator on a shaker for 4 days. The sections were then washed overnight at RT in PBS + 0.2% Triton X-100 before being gently post-fixed in 1% PFA in PBS for 1 hour at RT before being washed overnight at RT in PBS + 0.2% Triton X-100. Sections were then immersed in ScaleCUBIC-2 diluted 1:1 with MQ-H₂O overnight in a 37°C incubator on a shaker, before being immersed in ScaleCUBIC-2 overnight in a 37°C incubator on a shaker. Sections were imaged in a 2:8 mix of mounting solution 1 and 2 (CUBIC Trial Kit, Code No. 290–80801) to achieve a refractive index of ~1.49. Sections were imaged using a Leica SP5 confocal microscope at 10x magnification.

Fluorogold labelling

We used peripheral injection of fluorogold to label neuroendocrine neurons that project outside the blood-brain barrier (BBB), since fluorogold cannot cross the BBB, but is taken up by axon terminals outside the BBB and transported retrogradely to label cell bodies⁶. Sim2–2a-Cre::tdTomato mice were injected intraperitoneally with fluorogold (50μl of a 5% solution in 0.9% NaCl, Fluorochrome), and mice were perfused as described above 5 days later. Sections from each mouse were matched at 4 anterior-posterior levels of PVH, visualized using immunofluorescence, and imaged using a confocal microscope (Zeiss LSM 880). Cell bodies labelled with tdTomato and/or fluorogold were counted manually using QuPath.

Collateral mapping

To map collaterals of ARC- and BNST-projecting PVH^Sim2 neurons, Sim2–2a-FLPo mice were injected with AAVretro-EF1α-fDIO-Cre(HA) in the ARC or BNST, respectively, and AAV8-EF1α-DIO-Syp-mCherry in PVH. Animals were perfused 3 weeks after injection. The presence of collaterals in BNST/ARC was determined based on immunofluorescent detection of mCherry expression.

Channelrhodopsin-assisted circuit mapping

Sim2–2a-Cre::Npy-hrGFP or Ucn3-Cre::Npy-hrGFP mice were injected with AAV9-CAG-Flex-rev-ChR2-tdTomato in the PVH. To obtain ex vivo sections containing the ARC, mice were deeply anaesthetized using isoflurane and decapitated, and the brain was removed and immediately submerged in ice-cold cutting solution (in mM: 92 choline chloride, 10 HEPES, 2.5 KCl, 1.25 Na₂HPO₄, 30 NaHCO₃, 25 glucose, 10 MgSO₄, 0.5 CaCl₂, 2 thiourea, 5 sodium ascorbate, 3 sodium pyruvate) saturated with carbogen (95% O₂, 5% CO₂) and with measured osmolarity 310–320 mOsm l⁻¹ and pH 7.4. Coronal sections (275–300 μm thick) were then cut using a vibratome (Campden 7000smz-2) and incubated in carbogen-saturated cutting solution for 10 minutes at 34°C. Sections were then transferred to oxygenated artificial cerebrospinal fluid (aCSF – in mM: 126 NaCl, 21.4 NaHCO₃, 2.5 KCl, 1.2 NaH₂PO₄, 1.2 MgCl₂, 2.4 CaCl₂, 10 glucose) at 34°C for 15 minutes. Sections were then kept in oxygenated aCSF at room temperature until use. For recordings, a single section was placed into the recording chamber, where it was continually perfused with oxygenated aCSF at 3–4 ml min⁻¹. Neurons were visualized using an upright microscope (SliceScope Pro 1000, Scientifica) equipped with infrared differential interference contrast and fluorescence optics. Patch pipettes with open-tip resistances of 3–5 MΩ were backfilled with internal solution (in mM: 135 CsMeSO₃, 10 HEPES, 1 EGTA, 3.3 QX-314 (Cl^- salt), 4 Mg₂-ATP, 0.3 Na₂GTP, and 8 Na₂-phosphocreatine) with measured osmolarity 290 mOsm l⁻¹ and pH 7.35. Optogenetically-evoked excitatory post-synaptic currents (oEPSCs) were isolated in whole-cell voltage clamp at V_h=−70 mV. To photostimulate ChR2-expressing fibers in ARC, blue light from an LED (470 nm, Cool LED pE-100) was focused onto the back aperture of the microscope objective (40x) to produce wide-field illumination around the recorded neuron at a power of 10–15 mW mm⁻² (as measured using an optical power meter, PM100D, Thorlabs). The blue LED was controlled by a programmable pulse stimulator (Master 8, A.M.P.I.) and pClamp 10.5 software (Axon Instruments). The stimulation protocol consisted of four blue light pulses (470 nm wavelength, 5 msec) administered 1 s apart during the first 4 s of an 8-s sweep, repeated for a total of 20 sweeps. EPSCs with a latency of ≤6 ms following light delivery were considered as likely monosynaptic oEPSCs. To confirm the monosynaptic nature of PVH^Sim2→ARC^AgRP connections, for a subset of recorded neurons, oEPSCs were recorded in the presence of tetrodotoxin (TTX, 1 μM) alone and then in combination with 4-aminopyridine (4-AP, 500 μM). Signals were amplified through a Multiclamp 700B amplifier, and data were filtered at 2 kHz and digitized at 20 Hz. Access resistance (<30 Ω) was continuously monitored by a voltage step, and recordings were used for analysis if resistance changed by <15%. All recordings were analyzed using Clampfit 10.5.

Fos studies

Sim2–2a-FLPo::tdTomato mice were single-housed and habituated to handling for at least one week prior to the start of the experiment. Then, cages were changed at ZT10, and no food was introduced into the new cage of the fasting group. The next day at approximately ZT6, mice were anesthetized and perfused as described above. Brains were cut into 40 μm sections, and for each mouse, 4 sections were selected at matched intervals spanning the anterior-posterior extent of the PVH for manual quantification. Counts of tdTomato- and c-Fos-expressing neurons were summed across the 4 sections for each mouse.

Monosynaptic rabies mapping

Tracing of monosynaptic inputs to PVH^Sim2 neurons using rabies was performed as previously described^61,62. We injected AAV9-DIO-Ef1a-TVA-2A-N2cG into the PVH of Sim2–2a-Cre animals, waited two weeks, and then injected Rbv-CVS-N2c-dG-TdTomato(EnvA) rabies virus into the PVH of the same animals. Animals were perfused 1 week after injection of rabies virus.

Whole-brain monosynaptic inputs to PVH^Sim2 neurons were quantified and mapped to the Allen Mouse Brain Common Coordinate Framework version 3 (CCFv3)⁹⁹ using the QUINT workflow⁶³. First, we sectioned the entire mouse brain at 50μm into 3 series, and performed immunofluorescence against tdTomato on one series (encompassing 1/3 of the brain) as described above, and counterstained using DAPI. We registered our imaged sections to the Allen Mouse Brain CCFv3 using QuickNII software, and then segmented the tdTomato+ neurons with ilastik software. Next, we used the Nutil software to overlay the segmented tdTomato+ neurons onto the Allen Brain Atlas registrations, allowing us to quantify the location of the tdTomato+ neurons in relation to brain regions as defined by the Allen Mouse Brain CCFv3.

Hormone measurements

Wild-type mice were habituated to extensive handling for at least 10 days prior to sacrifice. Then, these mice were placed into a fresh cage with food around ZT6. At ZT12, mice in the “ZT12 ad lib” condition were sacrificed; food was removed from mice in the “ZT13 fast” condition, and for mice in the “ZT13 ad lib” condition, the cages were opened briefly but food was not removed. Then at ZT13, the remaining mice were sacrificed. All mice were sacrificed by rapid decapitation, and trunk blood was collected into K2EDTA tubes on ice.

Collected trunk blood was centrifuged, and plasma was collected into aliquots and stored at −80°C. For ghrelin measurements, the K2EDTA collection tubes contained 4 μL of 100 mM p-hydroxymercuribenzoic acid in potassium phosphate buffer (for a final concentration of 1 mM); and plasma was mixed with 1 N hydrochloric acid to reach a final concentration of 0.1 N. Plasma aliquots were then run in duplicate for each test. Enzyme-linked immunosorbent assay (ELISA) kits were used to measure ghrelin (Millipore, EZRGRA), insulin (Crystal Chem, 90080), and leptin (R&D Systems, MOB00B) in 96-well plates, according to the manufacturer’s instructions. Absorbance was measured using a plate reader.

Fiber photometry

Fiber photometry acquisition was controlled using a pyPhotometry microcontroller¹⁰⁰, and excitation LEDs (405 nm and 465 nm) and optical paths were contained within a fluorescence minicube (Doric Lenses ilFMC4). A low-autofluorescence optic fiber cable (1m length, 400 μm diameter, NA 0.48) transmitted excitation and emission light between the sample and minicube, and was affixed to the metal ferrule on the mouse’s skull through a ceramic mating sleeve (Thorlabs) with a small dab of UV-curing glue to prevent movement relative to the sample. Emitted light (500–550 nm) was measured using a photodetector embedded in the minicube, and digitized by the pyPhotometry board. Excitation light power was adjusted such that emission light fell roughly in the middle of the range of the photodetector. The sample was illuminated at 10 Hz using time-division multiplexing. All recordings were performed in the home cage, and mice were allowed to habituate to patch cord attachment for at least 10 minutes prior to the start of the recording.

Prior to short-term recordings of neural responses to food and water, mice were allowed ad libitum access to food, or denied access overnight to either food or water, as indicated. The same mice were subjected to different conditions and stimuli to allow for within-mouse comparisons. Immediately prior to recording, both food and water were removed from the cage. After 3 minutes of baseline recording, a pellet of chow (LabDiet 5001) or bowl of water was placed into the mouse’s cage, and the recording continued for a further 24 minutes. For a subset of these recordings, video recordings of the mouse were simultaneously acquired at 10 Hz using a camera (Allied Vision Mako U-130B), with acquisition controlled by Bonsai¹⁰¹. Video and photometry data were synchronized using a TTL pulse sent by the camera to the pyPhotometry board upon each acquired frame.

For long-term recordings, mice expressing GCaMP6s in PVH^Sim2 or ARC^AgRP neurons were first habituated to the recording setup by being tethered to the patch cord for at least 30 minutes per day over 2–3 days. Then, during long-term recordings, mice had access to food and water in separate bowls in their home cage, and recordings began before ZT6. For experiments involving fasting or water deprivation, food or water was manually removed from the cage at the specified time, and returned the following morning at ZT2. The recording was stopped after ZT5. Following experiments involving fasting or water deprivation, mice were allowed at least 5 days for recovery before being used for another experiment. In all studies, the same animals were subjected to conditions being compared, to allow within-animal comparisons.

For all experiments involving recording from ARC^AgRP neurons, each mouse was first screened for signal quality by measuring the neural response to a food pellet following overnight fasting. As previously demonstrated^7–9, ARC^AgRP activity drops substantially in response to food, so mice that showed no drop in signal over 3 trials on different days were excluded from further experiments.

To record from PVH^Sim2 neurons, Sim2–2a-FLPo mice expressing AAV8-EF1α-fDIO-GCaMP6s in the PVH were used. To record from ARC^AgRP neurons, Agrp-IRES-Cre mice expressing AAV1-hSyn-DIO-GCaMP6s in ARC were used. To record from ARC^AgRP neurons while inhibiting PVH^Sim2 or PVH^Trh neurons, Npy-IRES-FLPo::Sim2–2a-Cre or Trh-IRES-Cre mice, respectively, expressed AAV8-EF1α-fDIO-GCaMP6s in the ARC and AAV8-hSyn-DIO-hM4Di-mCherry in the PVH. In the latter experiments, clozapine N-oxide (CNO, 1 mg kg⁻¹) or vehicle (1% DMSO in 0.9% saline) was injected intraperitoneally at ZT11.5.

For experiments involving concurrent photometry recording and intragastric infusion, mice were either fasted overnight (Ensure infusion), or allowed to feed ad libitum (water, saline, sham infusions) prior to the infusion. In order to attach both the gastric catheter and patch cord animals were briefly anaesthetized with isoflurane and allowed to recover for 45 minutes before the start of the recording. The implanted vascular access button was then attached to a tether and swivel system (Instech- KVAH62T) that allowed free movement. During infusion, mice did not have access to chow, and water was available ad libitum. 1 ml of Liquid diet (Ensure, 0.93 Kcal/ml), water, or 0.9% saline was infused over 10 minutes at a constant rate of 100 μL/min using a syringe pump (New Era NE-4000). Sham infusion was performed by connecting the infusion apparatus, but without actual infusion. The line was flushed daily with 1 mL water (0.1 ml/minute) to prevent clogging and maintain patency.

For experiments involving concurrent photometry recording and optogenetic stimulation, mice used were either Sim2–2a-Cre::Npy-IRES-FLPo or AgRP-IRES-Cre::Sim2–2a-FLPo, expressing AAV5-Syn-Flex-rc[ChrimsonR-tdTomato] in the PVH/ARC and AAV8-EF1α-fDIO-GCaMP6s in the ARC/PVH, respectively. 630 nm excitation light from an LED light source (Plexon) was passed through a fluorescence minicube (Doric Lenses) together with the photometry excitation light as described above, and delivered to the brain through the same optic fiber. For these experiments, photometry data was acquired using time-division multiplexed illumination at 20 Hz. Optogenetic pulses were delivered at 20 Hz, and were interleaved between the 465 nm and 405 nm photometry excitation pulses to prevent bleedthrough using custom code running on an Arduino board synchronized to the pyPhotometry system.

Intragastric catheter placement

Gastric catheter surgeries were performed as previously described¹⁷. Prior to surgery, we assembled gastric catheters by threading a 10 cm of Micro-Renathane 0.037” × 0.023” per ft. Braintree Scientific MRE 037 100FT tubing through the center of a 1 cm × 1 cm square of Bard Mesh Flat Sheet 10”×14” (112660 eSutures), leaving 1 cm length of tubing extended through the mesh. The corners of the mesh were cut to ensure no sharp edges remained, and the tubing was secured via Krazy Glue and allowed to dry overnight. Before gastric catheter implantation, animals were fasted overnight, and allowed to re-feed for 2 hours before surgery. This ensured that their stomachs were full, allowing us to more easily perform the surgery. Mice were anesthetized by injection of xylazine (10 mg kg⁻¹) and ketamine (100 mg kg⁻¹) diluted in sterile saline (0.9%), and received extended-release buprenorphine (1.2 mg kg⁻¹) administered subcutaneously for analgesia. Next, we shaved the surgical sites on the back of the mouse below the skull and abdomen and treated the skin with betadine and alcohol. A 0.5 cm straight line incision was made between the bottom of the skull and the shoulder blades. The animal was placed on its back, and a second, 3 cm long incision was made in the midline extending from below the xyphoid process. A small incision was made in the abdominal cavity from which the stomach was gently removed and placed on sterile gauze moistened with 0.9 % sterile isotonic saline. The exposed stomach was kept moist by periodic application of 0.9 % sterile isotonic saline. A hole was made in the forestomach using an 18 G needle (BD- 305195). The tip of the catheter was inserted into the hole, and between 4–6 sutures (5–0 Medtronic Surgipro Blue 36” CV-23 Taper, Double Armed (eSutures-VP556X)) were used to firmly attach the mesh to the stomach. The stomach was gently placed back in the abdominal cavity, and the free end of the catheter was threaded through a subcutaneous tunnel across the left flank, to the previously made incision between the bottom of the skull and the shoulder blades. The peritoneum (5–0 Polysorb Undyed 18” P-12 Cutting (eSutures- SL5636)) and abdominal skin (5–0 Silk 30” NFS-2 Needle (Med Vet International- MV682)) were then sutured and the mouse was gently placed on its stomach. The end of the catheter was trimmed so that only 1 cm extended from the opening on the back of the mouse between the bottom of the skull and the shoulder blades. The end was then glued to the end of a vascular access button (VAB62BS/22 Mouse (Instech)) using Krazy Glue. The button was sutured with 6 sutures to the intrascapular muscle, and the skin was closed around the button (5–0 Monosof black 18” P-12 Cutting (eSutures SN5666)). Mice were placed in their home cage on a heating pad until they were mobile. Mice were allowed to recover for at least 14 days before gastric infusion experiments.

Operant task

The operant task was administered in the home cage using FED3 devices⁵⁹ (Open Ephys) running custom Arduino code. At all stages of conditioning, mouse body weight was monitored daily, and where necessary, each step was extended until mice maintained at least their starting body weight. Mice were first habituated to 20 mg chow pellets (LabDiet 5TUM) in a bowl overnight. The next day, bedding was changed and the FED3 was introduced in free-feeding mode (i.e. mice could access pellets ad libitum). After one day of free-feeding, the FED3 was switched into fixed-ratio 1 (FR1; i.e. 1 nose-poke leads to pellet delivery). In FR1, both nose ports were active, and no conditioned stimulus was delivered besides the pellet. Mice spent at least 3 nights in FR1 mode, and maintained a stable body weight for at least 2 days, before photometry experiments began. For fasting, pellet delivery was paused between ZT12 and ZT2.

Fiber photometry data analysis

Fiber photometry data were analyzed using custom Python scripts (https://github.com/sjwalker223/Walker_et_al_2026). First, for short experiments (< 1 hour), data from both channels was filtered using a Gaussian filter (window = 7 samples, sigma = 1) to remove noise. Then, for all experiments, correction for motion artifacts was achieved by using linear regression to predict the calcium-dependent (465 nm-evoked) fluorescence based on calcium-independent (405 nm-evoked) fluorescence at each time point, subtracting this predicted signal from the measured 465 nm signal, and adding the mean measured 465 nm signal across the whole recording. Next, this corrected signal was aligned to the time of pellet drop. ΔF/F₀ was calculated as (F-F₀)/F₀ at each time point, where F₀ is the mean signal from 3 minutes to 10 seconds prior to pellet drop. Finally, zΔF/F₀ was calculated by taking the z-score of this signal, using the formula (F-M)/S, where F is the ΔF/F₀ at a given timepoint; and M and S are the mean and standard deviation, respectively, of the ΔF/F₀ signal from 3 minutes to 10 seconds prior to pellet drop.

For comparing activity before and after pellet delivery (Figure 2F), we took the mean zΔF/F₀ in the 3 minutes before (pre) and after (post) pellet drop. For comparing the response between conditions (Figures 2H and 2J), we took the mean ΔF/F₀ in the 5 minutes following pellet drop. For comparing different time windows relative to food contact/consumption, we took the mean ΔF/F₀ for each trial in the following windows: pre-drop=30–20s before pellet delivery (to avoid contamination by cues before drop, such as hand over cage); pre-contact=drop to 1s before contact; pre-bite=contact to 1s before bite; bite=10s following first bite.

We calculated time constants (τ) of the food response as follows: for each mouse, within a window of 30 seconds before to 5 minutes after food presentation, we first determined the minimum signal, and then identified τ as the earliest time point at which 63.8% of this minimum had occurred, relative to the pre-stimulus baseline.

To measure responsiveness to hormone injections (Figures S7C and S7D), we calculated zΔF/F₀ as described above, except that F₀ was calculated from 10 minutes to 10 seconds prior to intraperitoneal hormone injection. Photometry signals were recorded for 115 minutes post-injection, and responses were quantified as the mean zΔF/F₀ value for each mouse from 90 to 115 minutes post-injection. Doses were chosen based on previous studies of ARC^AgRP neuron activity: leptin (2 mg/kg in 10 μL/g saline) or vehicle (10 μL/g saline)¹¹; and ghrelin (60 μg in 200 μL saline) or vehicle (200 μL saline)⁷.

For long-term recordings (≥23 hours), motion correction was first performed as described above. Then, ΔF/F₀ was calculated as (F-F₀)/ F₀ at each time point, where F₀ is the mean signal from ZT6 to ZT9. Given our interest in slow-timescale activity changes, we then filtered out fast-timescale fluctuations by taking the rolling mean of this ΔF/F₀ signal over a 30-minute window preceding each data point. Next, zΔF/F₀ was calculated by taking the z-score of this filtered signal, using the formula (F-M)/S, where F is the signal at a given timepoint; and M and S are the mean and standard deviation, respectively, of the signal from ZT6 to ZT9. For recordings where fasting was initiated at ZT6, the window (ZT3-ZT6) was instead used for calculating F₀, M, and S. Finally, for mice with more than one trial for a given condition, we took the mean over trials for each mouse, so that subsequent analyses were performed on a per-mouse basis.

For quantifying the response to food removal, we took the mean zΔF/F₀ signal in 1-hour bins from ZT11 to ZT16 for each mouse. We calculated time constants (τ) of the response to food removal as follows: for each mouse, within a window from ZT11 to ZT18, we first determined the maximum signal, and then identified τ as the earliest time point at which 63.8% of this maximum had occurred, relative to the mean signal from ZT11 to ZT12.

For experiments with photometry recordings during the operant task, we first aligned photometry and FED3 data. Then, for each trial, we calculated the cumulative numbers of pellets retrieved and pokes (combining both nose ports). Poke rate was calculated as the mean number of pokes per minute over a sliding window of 30 minutes. To calculate activity aligned to pokes, we used only pokes during the fast period (ZT 12-ZT2) that had no poke in the preceding 45 seconds. We first filtered the GCaMP6s signal (after correcting for motion artifacts – see above) using a Gaussian filter as described above for short experiments. Then, for each poke, we took this filtered signal from 45 seconds before to 100 seconds after each poke, and calculated zΔF/F₀ as described above for short experiments, using the window of 45 s to 1 s before the poke to calculate F₀, M, and S. Finally, we took the mean (± SEM) zΔF/F₀ signal over all pokes from all mice. To split pokes by percentile (Figures S11C and S11D), for each trial, we again used only pokes with no poke in the preceding 45 seconds. From these pokes, we calculated the cumulative percentage of pokes during the fast period at each poke time. We then binned pokes into deciles, and took the mean poke-aligned trace in each decile. For ease of visualization, we present four equally-spaced deciles.

Long-term body mass studies

Sim2–2a-Cre mice (8–13 weeks old) were injected with AAVDJ-CMV-DIO-eGFP-2a-TeNT (“TeNT”) or AAVDJ-CMV-DIO-eGFP (“GFP”) bilaterally in the PVH. Starting body weight was measured immediately prior to surgery, and following surgery mice were individually housed, and both food and body mass were weighed weekly for 12 weeks. At 13 weeks post-injection, body composition was measured using an EchoMRI 3-in-1 Analyzer without anesthesia. A single female was removed from the GFP control group for food intake studies due to evidence of food shredding.

Indirect calorimetry

Sim2-TeNT and Sim2-GFP mice were injected and housed as described above. At 4 weeks post-injection, mice were weighed, and their body composition analyzed as described above. They were then introduced to individual indirect calorimetry cages (Sable Systems Promethion) in the BIDMC Energy Balance Core. Mice were housed in these cages overnight (>18 h) before recording the 48 h experiment. During the experiment, the Promethion system measured physical activity (beam breaks), oxygen consumption, carbon dioxide production, body mass, and food/water consumption every 3 minutes. From these measurements, energy expenditure, respiratory exchange ratio, and total distance travelled were calculated. At each time point, the mean value over the two days of the experiment was used for further analysis. For non-cumulative time series (Figures S12C and S12E), the mean value was calculated over a sliding window of 30 minutes for each mouse. For Figure S12F, food consumption was measured in 3-minute bins. For Figure S12G, food consumption was summed in 1-hour bins. Analyses and statistical tests were performed using a custom Python script, except for comparing energy expenditure. Since energy expenditure can be affected by body mass, we employed a generalized linear model to account for possible effects of both treatment and body mass on energy expenditure, using the average energy expenditure over the 48 h experiment as the dependent variable¹⁰². This statistical model was implemented in CalR2 (https://bankslab.shinyapps.io/prod_ver/).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were performed using Python (for photometry and indirect calorimetry data), CalR2 (for energy expenditure measurements), or GraphPad Prism 6 or 10 (all other data), as described above. Sample sizes (N) and statistical tests are reported in figure legends. Briefly: when comparing two groups at a single time point, we used two-tailed t-tests (paired for within-subject comparisons). When comparing more than two groups, we used one-way ANOVA. When comparing groups at multiple time points, we used two-way repeated measures ANOVA. If a significant interaction (treatment*time, p<0.05) was identified, we then performed post-hoc tests at each time point, or between adjacent time points, with Holm-Sidak correction for multiple comparisons. To compare energy expenditure between groups while accounting for body mass differences, we used a generalized linear model, as described above. For all tests, α=0.05 was used as the threshold for significance.

For histological quantifications, we used QuPath or FIJI software to manually count neurons labelled by each of the specified probes/markers within the PVH at the specified anterior-posterior level(s).

Supplementary Material

MMC3

Document S1. Figures S1-S13.

MMC2

Data S1. Gross and fine whole-brain quantification of monosynaptic rabies tracing from PVH^Sim2 neurons, Related to Figure 5.

Percentage of all tdTomato-labelled neurons across the brain that belong to the specified gross (left tab) and fine (right tab) brain regions, as represented in Figures 5C and 5D, respectively.

MMC1

Data S2. Source data, Related to all Figures.

Source data for all graphs in the paper.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rat anti-mCherry (1:3000)	Invitrogen	Cat#M11217
Rabbit anti-GFP (1:3000)	Invitrogen	Cat#A11122
Rabbit anti-fluorogold (1:400)	Millipore	Cat#AB153-I
Rabbit anti-c-Fos (1:1000)	Synaptic Systems	Cat#226–003
Donkey anti-rat igG, secondary antibody, Alexa 594 Conjugate (1:500–1000)	Invitrogen	Cat#21209
Donkey anti-rabbit igG, secondary antibody, Alexa 488 Conjugate (1:1000)	Invitrogen	Cat#A21206
Bacterial and virus strains
AAV9-CAG-Flex-rev-ChR2-tdTomato	Atasoy et al.103	Addgene 18917-AAV9
AAV5-Syn-Flex-rc[ChrimsonR-tdTomato]	Klapoetke et al.104	Addgene 62723-AAV5
AAV8-EF1α-fDIO-GCaMP6s	Gift from Rylan Larsen (unpublished)	Addgene 105714-AAV8
AAV8-hSyn-DIO-hM3Dq-mCherry	Krashes et al.4	Addgene 44361-AAV8
AAV1-Syn-Flex-GCaMP6s	Chen et al.105	Addgene 100845-AAV1
AAV8-hSyn-DIO-hM4Di-mCherry	Krashes et al.4	Addgene 44362-AAV8
AAVretro-EF1α-fDIO-Cre(HA)	Schneeberger et al.106	Addgene 121675-AAVrg
AAVDJ-CMV-DIO-eGFP-2a-TeNT	Stanford Gene Vector and Virus Core	Cat#GVVC-AAV-71
AAVDJ-CMV-DIO-eGFP	Stanford Gene Vector and Virus Core	Cat#GVVC-AAV-12
AAV8-EF1α-DIO-Syp-mCherry	MGH Gene Delivery Technology Core	N/A
pAAV-DIO-EF1a-TVA-2A-N2cG	Sumser et al.62	Addgene 172360
Rbv-CVS-N2c-dG-tdTomato(EnvA)	UNC NeuroTools	N/A
AAV8-nEF-CreONDreON-	This paper	N/A
hM3Dq-mCherry
Chemicals, peptides, and recombinant proteins
Standard rodent chow	LabDiet	Cat#5001
Fluoromount-G Mounting Medium	SouthernBiotech	Cat#0100–01
DAPI Fluoromount-G Mounting Medium	SouthernBiotech	Cat#0100–20
Clozapine N-oxide (CNO)	NIMH Drug Supply	Program Cat#C-929
Dimethyl Sulfoxide (DMSO)	Thermo Fisher Scientific	Cat#036480.K2
Fluoro-Gold	Fluorochrome	N/A
Tetrodotoxin (TTX)	Tocris	Cat#1069
4-aminopyridine (4-AP)	Tocris	Cat#0940
20 mg grain-based rodent tablets	LabDiet	Cat#5TUM
CAS9 Protein	PNA Bio	Cat#CP01
Critical commercial assays
RNAscope® Probe-C3-Sim2	ACD	Cat#1110401-C3
RNAscope® Probe-C2-Trh	ACD	Cat#436811-C2
RNAscope® Probe-C1-Adcyap1	ACD	Cat#405911
RNAscope® Probe-C1-Slc17a6	ACD	Cat#319171
RNAscope® Probe-C2-Bdnf	ACD	Cat#424821
RNAscope® Probe-C1-Cbln2	ACD	Cat#428551
RNAscope® Multiplex Fluorescent Regent Kit V2 Assay	ACD	Cat#323100
CUBIC Trial Kit	TCI	Cat#C3942
Rat/Mouse Ghrelin ELISA Kit	Millipore	Cat#EZRGRA
Mouse Insulin ELISA Kit	Crystal Chem	Cat#90080
Mouse/Rat Leptin Quantikine ELISA Kit	R&D Systems	Cat#MOB00B
Deposited data
Data S1 - Rabies source data	This paper	N/A
Data S2 - Source data	This paper	N/A
Photometry and metabolic source data	This paper.	https://github.com/sjwalker223/Walker_et_al_2026; https://doi.org/10.5281/zenodo.20059773
Experimental models: Organisms/strains
Mouse: Npy-IRES-FLPo: B6.Cg-Npytm1.1(flpo)Hze/J	The Jackson Laboratory; Daigle et al93	Jax: 030211; RRID:IMSR_JAX:030211
Mouse: Lepr-IRES-Cre: B6.129-Leprtm3(cre)Mgmj/J	The Jackson Laboratory; Leshan et al94	Jax: 032457; RRID:IMSR_JAX:032457
Mouse: Npy-hrGFP: B6.FVB-Tg(Npy-hrGFP)1Lowl/J	The Jackson Laboratory; van den Pol et al95	JAX: 006417; RRID:IMSR_JAX:006417
Mouse: Agrp-IRES-Cre: Agrptm1(cre)Lowl/J	The Jackson Laboratory; Tong et al77	JAX: 012899; RRID:IMSR_JAX:012 899
Mouse: Trh-IRES-Cre: B6;129S-Trhtm1.1(cre)Mjkr/LowlJ	The Jackson Laboratory; Krashes et al27	JAX: 032468; RRID:IMSR_JAX: 032468
Mouse: Pacap-IRES-Cre	Krashes et al27	N/A
Mouse: Trh-2a-Dre	Grzelka et al28	N/A
Mouse: Ucn3-Cre: Tg(Ucn3-cre)KF43Gsat/Mmucd	MMRRC	Cat#032078-UCD
Mouse: Sim2–2a-Cre	This paper	N/A
Mouse: Sim2–2a-FLPo	This paper	N/A
Oligonucleotides
Sim2 sgRNA	Genscript	N/A (This paper)
Sim2 primers	Eton Biosciences	N/A (This paper)
Recombinant DNA
Sim2–2a-Cre ssDNA	Genscript	N/A (This paper)
Sim2–2a-FLPo ssDNA	IDT	N/A (This paper)
Sim2–2a-FLPo ssDNA	IDT	N/A (This paper)
Software and algorithms
Python 3.11	Python Software Foundation,	https://www.python.org/ RRID: SCR_008394
Photometry and metabolic analysis scripts	This paper.	https://github.com/sjwalker223/Walker_et_al_2026; https://doi.org/10.5281/zenodo.20059773
Prism 6/10	GraphPad	https://www.graphpad.com/features RRID: SCR_002798
CalR2	Mina et al.102	https://bankslab.shinyapps.io/prod_ver/
QuPath	Bankhead et al.107	https://qupath.github.io/ RRID:SCR_018257
Fiji	Schindelin et al.108	https://imagej.net/software/fiji/ RRID:SCR_002285
QuickNII	Yates et al.63	https://quicknii.readthedocs.io/en/latest/ RRID:SCR_016854
Ilastik	Berg et al.109	https://www.ilastik.org/download.html RRID:SCR_015246
Nutil	Yates et al.63	https://www.nitrc.org/projects/nutil RRID:SCR_017183
Bonsai	Lopes et al101	https://bonsai-rx.org/ RRID:SCR_017218
Other
pyPhotometry	Open Ephys; Akam & Walton100	SKU: OEPS-8010; RRID:SCR_022940
FED3.2	Open Ephys; Matikainen-Ankney et al59	SKU: OEPS-7511