Sequential appetite suppression by oral and visceral feedback to the brainstem

The termination of a meal is controlled by dedicated neural circuits in the caudal brainstem. A key challenge is to understand how these circuits transform the sensory signals generated during feeding into dynamic control of behaviour. The caudal nucleus of the solitary tract (cNTS) is the first site in the brain where many meal-related signals are sensed and integrated1–4, but how the cNTS processes ingestive feedback during behaviour is unknown. Here we describe how prolactin-releasing hormone (PRLH) and GCG neurons, two principal cNTS cell types that promote non-aversive satiety, are regulated during ingestion. PRLH neurons showed sustained activation by visceral feedback when nutrients were infused into the stomach, but these sustained responses were substantially reduced during oral consumption. Instead, PRLH neurons shifted to a phasic activity pattern that was time-locked to ingestion and linked to the taste of food. Optogenetic manipulations revealed that PRLH neurons control the duration of seconds-timescale feeding bursts, revealing a mechanism by which orosensory signals feed back to restrain the pace of ingestion. By contrast, GCG neurons were activated by mechanical feedback from the gut, tracked the amount of food consumed and promoted satiety that lasted for tens of minutes. These findings reveal that sequential negative feedback signals from the mouth and gut engage distinct circuits in the caudal brainstem, which in turn control elements of feeding behaviour operating on short and long timescales.


suppression of feeding
In this study, we focused on investigating the two principal cNTS cell types that have been implicated in "non-aversive satiety", i.e. the suppression of feeding in the absence of nausea or sickness [1][2][3][4][5][6][7][8][9][10] .One of these cell types is defined by the expression of prolactin-releasing hormone (encoded by the Prlh gene) and the other by the expression of glucagon-like peptide 1 (encoded by the Gcg gene).These two marker genes are unique in that they label transcriptionally homogeneous cNTS cell types 11 that are intermingled but non-overlapping 11,12 , directly innervated by vagal afferents 2,5,13,14 , activated by ingestion as measured by Fos expression 1,4,15 , and functionally validated to inhibit feeding without inducing aversion 2,6,10,16 (Extended Data Fig. 1).These two genes also overlap with several broader, and more heterogeneous, populations of cNTS neurons that have been independently implicated in the control of feeding (described in Extended Data Fig. 1 and refs. 4,6,11,17).

Post-Ingestive learning does not rescue the response of PRLH neurons to sweet taste
To test whether post-ingestive learning can rescue the rapid activation of PRLH neurons by sweet taste, we gave Prlh Cre Trpm5 -/-mice access to a glucose solution (24%) overnight twice, with a day of separation in between (Methods).Long-term exposure to glucose increased glucose consumption in a subset of taste-blind mice, consistent with previous reports that these animals acquire preferences for nutritive solutions via post-ingestive feedback [18][19][20][21] (Extended Data Fig. 5m).This increase in ingestion rate partially restored the rapid response of PRLH neurons to glucose ingestion (Extended Data Fig. 5n) but did not rescue the neural response per lick, indicating that this requires taste (0.03 ± 0.01 z for learned animals compared to 0.02 ± 0.01 z for naïve animals, p = 0.5476; Extended Data Fig. 5o).Thus, the activation of PRLH neurons by sweet substances requires canonical taste signaling.

PRLH neurons receive orosensory feedback
The rapid activation of PRLH neurons during feeding was driven by an interaction between the taste of food and the rate of ingestion.This was unexpected in part because gustatory signals are relayed primarily to the rostral NTS (rNTS) 22 , whereas the cNTS is associated with visceral feedback transmitted by the vagus nerve.However, PRLH neurons receive descending input from several forebrain structures that could contain information about taste and ingestion dynamics, including the central amygdala, paraventricular hypothalamus, and lateral hypothalamus 23 .In addition, PRLH neurons receive a direct projection from the intermediate reticular formation, a structure that contains premotor neurons for licking 6,24 .The function of these centrifugal projections to the cNTS has received limited attention and dissecting these pathways will be an important area for future investigation.

PRLH neurons do not appear to control sensory-specific satiety
The control of feeding behavior by PRLH neurons appears be unrelated to sensory-specific satiety 25 , the process by which repeated exposure to a taste within a meal reduces further consumption of that taste, but does not affect consumption of different tastes.This is because PRLH neuron responses do not increase as the meal progresses (Fig. 1-2) and because blocking PRLH neuron activation does not increase total consumption (Fig. 4).

PRLH neuron activation is gated by behavioral state
PRLH neurons receive abundant feedback from the vagus nerve 5,13 and remain activated for tens of minutes after nutrient infusion into the stomach (Fig. 1b).Thus, the fact that PRLH neurons track orosensory cues during normal ingestion (Fig. 2) implies that their activation by visceral feedback is in some way suppressed during normal feeding.Indeed, we found that the sustained activation of PRLH neurons observed after IG infusion was absent during normal feeding, even when the amount and duration of food consumption/infusion were matched (Fig. 1 and Extended Data Fig. 3h-j).Consistently, the hormone CCK was required for PRLH neuron activation during IG infusion of fat (CCK) but became dispensable during oral ingestion of the same nutrient (Fig. 1g-h).How this hierarchy of sensory responses is established is unknown but could involve the filtering of predicted visceral inputs by cNTS interneurons 26,27 to shift learned neural responses forward in time, as is observed in motivational circuits 28,29 .In this regard, the sensory signals generated during a meal occur in a specific sequence (from mouth to stomach to intestines), and they may be interpreted correctly by the brain only when this natural progression is respected 30 .Alternatively, there may be a signal from mouth-to-gut that suppresses certain GI signals during normal ingestion The ability to monitor cNTS dynamics during behavior will enable investigation of these basic questions about how signals from the mouth and gut interact during a meal.

Microendoscopic imaging of cNTS neurons in awake behaving mice
In this study, we developed a preparation for recording the single-cell activity of PRLH neurons during ingestion.We implanted an angled GRIN lens above the cNTS (Fig. 3e) and first attempted to perform imaging in freely moving animals, which was associated with severe brainstem motion (Fig. 3e; Supplementary Video 4).This motion was reduced by head-fixing, but not to the extent that it was feasible to track the dynamics of single cells (Fig. 3e; Supplementary Video 5).This is consistent with previous reports of unsuccessful attempts to perform cNTS imaging in awake animals 26 .However, we noticed that the largest motion artifacts were correlated with lower body movements of the mouse, which could be eliminated by further restraining the torso of the head-fixed animals (Fig. 3e).This head-fixed, restrained preparation enabled stable calcium recordings while awake mice consumed liquid diets (Supplementary Video 6).

PRLH neurons do not control motor circuits directly
In Figure 4, we investigated the mechanism by which PRLH neuron activity decreases the size of individual lick bouts.It seemed unlikely that PRLH neurons control motor circuits directly, because manipulating PRLH neuron activity did not change the inter-lick interval (ILI) during a bout (Extended Data Fig. 7i-l), a parameter which is determined by central motor pattern generators [31][32][33] and is invariant during normal behavior.In addition, we failed to observe fictive feeding behaviors (e.g.orofacial movements) in response to PRLH neuron stimulation or silencing in the absence of food, as has been observed for circuits that directly control motor outputs 34 .

Disentangling the effects of nutrients from GI distension
In Figure 5, the activation of GCG neurons by post-ingestive feedback could be due to signals of GI stretch, nutrient sensing, or both [35][36][37] .Of note, because calories and food volume are highly correlated 38 , and because calories can increase gastric distension by delaying gastric emptying 39 , the feeding experiments described in Figure 5 cannot distinguish between these two mechanisms.To test the sufficiency of GI stretch, we infused into the stomach the non-nutritive sugar mannitol, which is not absorbed from the intestines, and therefore induces significant intestinal distension 38 .Mannitol infusion strongly activated of GCG neurons (5.0 ± 1.7 z, p = 0.0002 compared to baseline; Extended Data Fig. 9k), but not PRLH neurons (0.6 ± 0.6 z, p = 0.7082 compared to baseline; Extended Data Fig. 9l), and the magnitude of this activation was at least as large as that caused by equi-osmotic and equi-volemic infusions of glucose and Ensure (3.2 ± 0.6 z for Ensure infusion, p = 0.6282 compared to mannitol infusion; Extended Data Fig. 9k).To present a pure mechanosensory signal to the stomach, we also performed IG infusions of air (1.0 mL), which activated GCG neurons (2.0 ± 0.3 z, p <0.0001 compared to baseline) but not PRLH neurons (-0.1 ± 0.2 z, p = 0.8353 compared to baseline; Extended Data Fig. 9q).Together, these data indicate that GI stretch is sufficient to activate GCG neurons but has little effect on PRLH neurons.

GCG neuron stimulation triggers long-lasting satiety
In Figure 6, we found that GCG neuron pre-stimulation in the absence of food caused a striking, long-lasting suppression of the initiation of new feeding bouts (Fig. 6f).The fact that this response was dose-dependent (Fig. 6g and Extended Data Fig. 10g) suggests that GCG neuron activity can be integrated over time in downstream circuits, likely via release of GLP-1 7 , to produce satiety that lasts much longer.Of note, this provides a mechanism to link the amount of food consumed to the time interval before initiation of the next meal.A similar signal integration mechanism has been proposed for hunger-promoting AgRP neurons 40 , which promote long-lasting hunger via changes in downstream circuits caused by neuropeptide release 41 .It is likely that competition between these and other neuropeptides, which are released in proportion to the duration of food deprivation or ingestion, is an important part of the mechanism that controls the long-lasting transitions between hunger and satiety.