Gut-to-brain signaling restricts dietary protein intake during recovery from catabolic states

Summary

Dietary needs are dynamic with optimal ranges for nutrients varying over time and across physiological states. How optimal nutrient set-points are established and why they are adjusted remains largely unknown. In our efforts to understand the physiology of recovery, we made the surprising observation that mice restrict protein intake at the expense of caloric supply. We identified three amino acids (Q, K, T) within dietary protein, which are necessary and sufficient for protein aversion. The anorexigenic effects of QKT are driven by ammoniagenesis in the gut, sensed by enterochromaffin cells in a TRPA1-dependent fashion, and transduced to brainstem neurons via serotonin-signaling, inducing anorexia. We propose that this mechanism serves as a first-line defense against ammonia toxicity. In summary, we identified a set of adaptive food preferences during recovery (‘recovery behavior’), which has significant implications for understanding diseases of pathologic recovery and the development of therapeutic interventions deployed to enhance recovery.

Introduction

Animals constantly face the decision of when, what and how much to eat. They also need to evaluate which foods are safe and which ones are not. Food choices are vital for organismal fitness because the risk of ingesting something toxic (e.g. pathogens, poisons, xenobiotics) has to be balanced against the necessity to assimilate essential nutrients and energy on a meal-to-meal basis (Carlson and Agrawal, 2023; Florsheim et al., 2021). Among nutrients, protein intake appears to be particularly tightly defended (Johnson et al., 2013). Dilution of dietary protein content promotes increased food intake across species, presumably to meet protein demands (‘protein leverage’) (Gosby et al., 2014; Saner et al., 2023; Solon-Biet et al., 2016; Steck et al., 2018). Although the minimum dietary protein requirements for maintenance of health are relatively clear, the plasticity of these requirements across different physiological states and in response to homeostatic perturbations is less well understood. Patients recovering from acute illnesses are considered a population in whom protein demands are elevated. This assumption is based on the idea that restoration of muscle mass, which is lost during sickness due to proteolysis (catabolism), requires a large supply of dietary protein (Wilmore, 1991). However, recent randomized, controlled clinical trials have shown that exposure to excessive protein can be harmful to such patients (Bels et al., 2024; Heyland et al., 2023). Why excessive protein intake can be toxic and whether physiological mechanisms exist that would otherwise prevent such maladaptive ingestion from occurring, remains unknown.

During acute sickness states, loss of appetite and a desire to eat (anorexia) frequently co-occurs with other behavioral adaptations, collectively referred to as ‘sickness behaviors’ (Hart, 1988), which promote survival (Wang et al., 2016). Adaptive feeding responses during homeostasis and in response to insults are generated by neurons residing in different areas of the brain including the hypothalamus, area postrema (AP) and nucleus tractus solitarius (NTS) (Zhang et al., 2021). These neurons receive strong input from the periphery, including the gut, through endocrine communication and neuronal pathways, leading to an integrative ‘report’ of the animal’s internal state (interoception) (Ran et al., 2022). In many cases, the precise nature of bodily signals causing neurons to adjust feeding behavior as well as the physiological benefits conferred by these adjustments, remains unclear. In contrast to anorexia during acute sickness, much less is known about how feeding behavior and dietary preferences are adjusted after illness resolution and whether this serves an adaptive purpose to optimize recovery (‘recovery behavior’). In fact, how recovery is physiologically achieved is not understood. This large knowledge gap is clinically important because therapeutic tools to support the recovery process remain virtually non-existent.

Here, we report the observation that mice recovering from catabolic challenges voluntarily restrict dietary protein intake at the expense of caloric supply, even if the resulting outcome is death. We found that many mechanisms controlling food intake under homeostatic conditions were not required for protein restriction in the recovery state. Instead, we identified three amino acids (AA), which are necessary and sufficient for protein aversion during recovery from catabolism. The anorexigenic effects of these AA are driven by ammoniagenesis in the gut, which is sensed by enterochromaffin cells in a TRPA1-dependent fashion and transduced to a subset of neurons of the brainstem.

Mice recovering from catabolism defend dietary protein intake against conflicting survival drives.

To gain insight into feeding behavior during recovery, we modelled a post-catabolic state common to most acute illnesses by depriving mice of food for 24h. This predictably led to loss of body weight, including both lean and fat mass (Fig. S1 A,B). Next, we introduced one of three customized diets, which were matched in calories and micronutrients and only differed in the main macronutrient source of the calories (protein, carbohydrate, or fat) (Fig. 1A and Table S1). We thus refer to these three diets as high-fat, high-carb, and high-protein diets (HFD, HCD and HPD). We found that mice consumed large amounts of HFD and HCD upon refeeding and on subsequent days, suggesting that both diets are palatable to rodents (Fig. 1B and Fig. S1C). The absolute intake of HCD and HFD after a fast was comparable to conventional chow, ruling out neophobia as a confounding factor in our experimental paradigm (Fig. S1D). In stark contrast, HPD consumption was low, particularly during the first 24h of refeeding, yielding lower caloric intake (Fig. 1B and Fig. S1C–E) and delayed recovery of body weight (Fig. 1C and Fig. S1F).

Fig. 1. Mice recovering from catabolism defend dietary protein intake against conflicting survival drives.

(A) Experimental paradigm. (B) Food and caloric intake during refeeding of mice subjected to one of the three diets (n=10–12/group). (C) Body weight trajectories (n=6–7/group). (D) Protein intake of fasted mice refed with the indicated diets for 24h (n=10–12/group) and of fed controls subjected to a diet switch (n=4–14/group). (E) Cold exposure as an experimental model for an increased physiological pressure to consume food. (F) Food intake and survival of fasted mice refed with HFD or HPD at 30 or 6°C (n=4/group for food intake, n=4–7/group for survival). (G) Experimental design (H) 24h protein intake and survival of fed and fasted mice exposed to HPD at 6°C (n=10/group) (I) Representative cFos stainings of the AP and NTS of fasted mice treated with isocaloric amounts of olive oil, sucrose or casein intragastrically. Borders of the AP are outlined in white. Scale bar: 100µm. (J) Quantification of cFos+ cells (n=3–9/group) (K) Representative cFos stainings of the AP/NTS of fasted mice pair-fed with HPD or HCD. Borders of the AP are outlined in white. Scale bar: 100µm. (L) Quantification of cFos+ cells (n=3–4/group). Data is presented as mean ± s.e.m *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (B, J) one-way ANOVA with Holm-Sidak’s post hoc test (D,F-left panel) two-way ANOVA with Holm Sidak’s post hoc (H,L) two-tailed, unpaired t-test (F-right panel, H-right panel) log-rank test.

Based on the nutritional composition of the diets, we next calculated the amount of each macronutrient consumed. We observed that total protein intake was closely matched after refeeding across groups, regardless of the protein content of the diet, whereas carbohydrate and fat consumption were not (Fig. 1D and Fig. S1G,H). This behavioral response was specific to the recovery state because healthy, fed mice subjected to a diet switch did not limit protein intake as tightly (Fig. 1D). We confirmed this phenotype in both sexes, a second strain of mice (Balb/c) and at thermoneutrality (30°C) (Fig. S1I–K). Using genetic models, we found that the ability to engage in adaptive thermogenesis did not affect protein consumption (Fig. S1L). Despite inducing severe intestinal distension, a hallmark of germ-free conditions (WOSTMANN and BRUCKNER-KARDOSS, 1959), microbiota depletion by antibiotics (ABX) also had no effect on protein intake upon refeeding (Fig. S1M).

To determine whether these observations were more generalizable, we induced inflammatory catabolism by injection of a sublethal dose of bacterial lipopolysaccharide (LPS). We titrated the LPS dose in a manner such that mice developed anorexia for 24h, after which we introduced one of the three isocaloric diets (Fig.S2A,B). Similar to fasting, LPS exposure induced loss of body weight, lean and fat mass (Fig. S2C). We found that mice recovering from LPS also tightly controlled their intake of dietary protein, but not carbohydrate or fat, and this occurred at the expense of caloric supply (Fig.S2D–F).

The narrow range of dietary protein consumed by mice during recovery suggested that eating too much or too little dietary protein might be unfavorable. Using pair-feeding paradigms, we first established that meeting a minimum dietary protein intake was required for recovery of body weight, fat and lean mass following inflammatory and non-inflammatory catabolism (Fig. S3A,B). This made the observation that mice voluntarily cap their protein intake even more surprising. Accordingly, we next wanted to understand the liability of ingesting too much dietary protein and reasoned that this might be achieved by increasing the physiological pressure to consume food. To do so, we first subjected mice to prolonged starvation for up to three days, which led to exaggerated body weight loss (Fig. S3C). However, protein intake remained unchanged across groups, even as the mice died (Fig. S3D,E). As an orthogonal model, we placed mice into a cold environment (6°), where shivering and induction of adaptive thermogenesis are required for survival (Chouchani et al., 2019). Because both physiological programs are energetically costly, food consumption substantially increases (Deem et al., 2020) (Fig. 1E). As expected, mice offered a HFD ate more food in the cold than those housed at thermoneutrality (30°) (Fig. 1F). This increase in food intake was sufficient to promote body weight recovery and ensure survival post-fasting (Fig. 1F and Fig.S3F). In contrast, HPD intake remained low, which left mice with insufficient caloric supply, resulting in sustained body weight loss and death (Fig. 1F and Fig.S3F). In line with our observation that protein restriction was specific to the recovery state, we found that fed mice subjected to a diet switch in the cold consumed more dietary protein than post-catabolic animals, which enabled survival (Fig. 1G,H). These data demonstrate that mice restrict dietary protein intake at all costs during recovery from a catabolic insult, even if the resulting outcome is death.

Dietary protein intake during recovery from catabolic states is not controlled by canonical modes of appetite regulation.

Having documented a strong aversion to excess dietary protein in mice recovering from a catabolic state, we wondered how such potent control might be achieved. Satiety, nausea or aversion are canonically encoded by neurons residing in the AP and NTS of the brainstem (Bai et al., 2019; Florsheim et al., 2023; Huang et al., 2024; Zhang et al., 2021), so we wondered whether these brain areas are activated in response to protein ingestion. Intragastric administration of isocaloric amounts of casein, sucrose or olive oil showed that protein (casein) most robustly increased neuronal activity in the NTS as measured by cFos expression and, to a lesser extent, AP (Fig. 1I,J). When mice could freely choose how much protein-rich food they wished to consume, we observed a strong increase in NTS, but not AP activation, compared to HCD controls, regardless of whether we pair-fed animals or not (Fig. 1K,L and Data S1). These data suggest that recovering rodents voluntarily restrict protein intake to an extent that prevents AP, but not NTS activation, which is consistent with the fact that only AP activity is reported to lead to aversion (Huang et al., 2024; Roman et al., 2016). We then explored many mechanisms controlling appetite and/or mediating aversion, which converge onto AP/NTS neurons, such as taste, stretch, endocrine hormones including glucagon, fibroblast growth factor 21 (FGF21), cholecystokinin (CCK), growth differentiation factor 15 (GDF15) as well as mast cell mediators (Data S1). Despite validating our experimental approaches (Florsheim et al., 2023) (Data S2), we found no evidence for involvement of any of these mechanisms in protein restriction during recovery (Data S1), suggesting an alternative mode of appetite control.

Three amino acids are necessary and sufficient for protein aversion during recovery.

Complex proteins such as casein are built from individual AA. We thus wondered whether AA and/or their metabolites might restrict protein intake independently of taste, stretch and hormones. To explore this, we created a diet in which casein was exchanged for free AA, while caloric, total protein (i.e. AA) and micronutrient content remained equivalent to the HPD (Fig. 2A and Fig. S4A). The amount of each AA added to the diet was based on the published AA profile of casein(Lapierre et al., 2012), which was validated analytically (Fig. S4B). We found that free AA suppressed appetite even more potently than complex protein during refeeding (Fig. 2A), an effect, that was maintained in the cold, resulting in weight loss and death in the majority of animals (Fig. 2B–D). We thus wished to clarify which AA was sufficient to suppress appetite. To do so, we administered one of the 20 proteinogenic AA as a single bolus intragastrically to fasted mice, followed by refeeding with HFD (Fig. 2E). The dose of each AA was based on the natural abundance in food (casein) (Lapierre et al., 2012) as well as the ad libitum consumption of HPD during the first 24h of refeeding (Fig. 1D). We purposely chose to administer AA in unphysiological kinetics to understand the liability of a breakdown of the behavioral restriction of AA intake that we had documented earlier (Fig.1D,F and Fig. 2A–D). While most AA had little or no effect on food intake and behavior, three AA - glutamine (Q), lysine (K) and threonine (T) - induced anorexia and led to signs of toxicity (e.g. lethargy, diarrhea, muscle spasms, tail twitching and seizures), which culminated in death in many of the mice (Fig. 2F and Fig. S4C). We observed rapid-onset anorexia and mortality when Q, K and T were administered in combination (Fig. 2G and Fig. S4D–F). Consistent with our individual AA screen, we found that these effects were specific to QKT as three other AA with comparable biochemical properties (cysteine, asparagine and histidine, CNH) did not elicit these effects, regardless of whether they were applied in matched or food-adjusted doses (Fig. 2G and Fig. S4D–F). The kinetics of exposure, rather than the cumulative dose, determined QKT toxicity because mice subjected to a split-dose regime (60 and 40% of full dose, 6h apart) survived (Fig. S4G), suggesting that the rate at which mice consume AA is a key mechanism of toxicity prevention. Using the same experimental paradigm, we found that Q, K and T suppressed food intake during refeeding compared to isocaloric amounts of sucrose (Fig. S4H,I).

Fig. 2. Three amino acids are necessary and sufficient for protein aversion during recovery.

(A) Food intake of fasted mice refed with the indicated diets (n=5/group). (B) Food intake of fasted mice refed with the AA diet at room temperature (22°C) or in the cold (6°C) (n=5–6/group). (C,D) Body weight trajectories and survival of the same mice. (E,F) Experimental paradigm and 24h HFD intake of fasted mice treated with one of the 20 proteinogenic AA as a single bolus (n=73 total). (G) Survival of fasted mice treated with an intragastric bolus of QKT or a matched amount of cysteine, asparagine and histidine (CNH) (30 mg/AA) (n=4–5/group) (H,I) Diet intake of mice refed at 22 or 6°C with the full AA diet (AA) or an isocaloric AA diet deficient in Q, K, T with fixed total AA content (n=11–16/group). Survival of mice refed in the cold is shown on the right (n=11/group). (J) Preference of fed or fasted mice for the QKT-deficient over the complete AA diet assessed in a 2-choice assay performed at 22 or 6°C. Results are expressed as change from baseline preference (delta) (see Fig. S4K for experimental design) (n=10–13/group). Data is presented as mean ± s.e.m *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (A,B,H-J) unpaired, two-tailed t-test (D,G,I-right panel) log-rank test.

To understand whether Q was required for protein-induced appetite suppression, we removed this AA from the diet, while proportionally increasing the remaining 19 AA, thus leaving caloric and total protein content fixed. Removal of Q alone was not sufficient to increase protein intake (Fig. S4J). Thus, we next removed Q, K, and T from the AA diet, while fixing total AA and caloric content (AA-(QKT)). The absence of Q, K, and T in the diet allowed for increased AA consumption of mice following a fast, regardless of ambient temperature, which enabled survival in the cold (Fig. 2H,I). Aversion to QKT-rich food was specific to the recovery state because fasted, but not fed mice, developed a strong preference for the QKT-deficient over the complete AA diet when given the choice between the two formulations, regardless of ambient temperature (Fig. 2J and Fig.S4K). Collectively, these data identify a physiological aversion to Q, K, and T that is specific to the recovery state.

Ammonia production from aversive AA is anorexigenic and toxic.

We next set out to clarify the mechanism underlying the anorexigenic and toxic properties of Q, K and T. In line with our previous observations, we did not detect a specific endocrine signature in mice consuming a QKT-deficient versus complete AA diet (Fig. S5A). This supported the idea that intrinsic features of the three AA might be mechanistically relevant instead. The first step of Q metabolism, catalysed by the enzyme glutaminase (GLS), yields release of free ammonia (NH₃), which is the common denominator of AA catabolism in general (Bhutia and Ganapathy, 2016; Nitzahn and Lipshutz, 2020). Ammonia is highly toxic, particularly to neurons, and requires detoxification via the hepatic urea cycle (Albrecht and Norenberg, 2006). We reasoned that accumulation of ammonia would definitely signal a “too much protein ingested” state. Consistent with this notion, force feeding of dietary protein elicited a robust increase in circulating ammonia levels, whereas administration of isocaloric amounts of olive oil did not (Fig. 3A). When we sampled blood from ad libitum refed mice at an early time point at which protein intake was still divergent between dietary groups (Fig. S5B), we detected higher levels of ammonia in mice feeding on protein-rich food compared to HCD or HFD controls (Fig. S5C). Because the rate of protein consumption subsequently dropped sharply in these animals (Fig. S5D), our data suggest that consuming large amounts of protein-rich food after a period of starvation yields a spike in ammonia levels, which is followed by appetite suppression.

Fig. 3. Ammoniagenesis from aversive AA is anorexigenic and toxic.

(A) Serum ammonia levels in fasted mice 45 min after intragastric administration of water or isocaloric amounts of olive oil or casein (n=5–10/group) (B) Circulating ammonia levels in mice treated intragastrically with 50 mg of the indicated AA after a fast (n=3–4/group). (C) Survival of fasted mice treated with an intragastric bolus of 100 mg of D- or L-Gln (n=4/group). (D) Food intake of the same mice (E,F) Quantification of cFos+ cells (n=4–5/group) and representative cFos stainings of the AP/NTS of fasted mice treated with 15 mg NaOAc or NH₄OAc intragastrically. Borders of the AP are outlined in white. Scale bar: 100µm. (G) Food intake of mice refed with HFD at 6°C following treatment with a single bolus of NH₄OAc or NaOAc (n=5/group) (left panel) or continuous exposure to water containing either substance (right panel) (n=11/group) (H) 24h HFD intake of fasted mice treated with a glutamine synthetase inhibitor (MSX) or vehicle 45 min prior to refeeding (n=3–5/group) (I) Survival of fed and fasted mice treated with a single, intragastric bolus of NH₄OAc (n=9–10/group). (J) Serum ammonia levels in fed and fasted mice 30 min following intragastric administration of a sublethal dose of NH₄OAc (n=8/group) (K) Diet habituation paradigm (L) Fold change of transcript levels of hepatic urea cycle enzymes at various time points of HPD exposure visualized as a heatmap (mean of n=4–5 is shown). (M) Blood urea nitrogen (BUN) levels in mice fed with the indicated diets for 1 week post fasting (n=3/group). (N) Survival post NH₄OAc challenge (n=8/group). (O) Preference for the QKT-deficient over complete AA diet assessed in a 2-choice assay performed at 6°C after 1 week of HFD or HPD habituation. Results are expressed as change from baseline preference (delta) (n=10–13/group). (P) Intake of the complete AA diet of mice in the cold following 1 week of HFD or HPD habituation (n=18–19/group). (Q) Representative immunofluorescence staining of the AP/NTS of TRAP2-Ai14 mice in which protein-responsive neurons were permanently labelled with tdTomato (‘TRAPed’) and ammonia responsive neurons were visualized using cFos staining. Scale bar: 100µm. The percentage of cFos+ neurons among TRAPed neurons across both regions is shown below. Data is presented as mean ± s.e.m *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (A,B) one-way ANOVA with Holm-Sidak’s post hoc test (C,I,N) log-rank test (D-H, J,M,O,P) unpaired, two-tailed t-test.

Accordingly, we next wondered whether Q,K and T were more ammoniagenic than other AA. To test this, we employed two orthogonal approaches: first, we supplemented mice with Q,K, and T in doses physiologically consumed during refeeding as a single bolus and measured ammonia in the blood 30 min later. We used cysteine (C) and histidine (H) as neutral and basic AA controls, respectively. We found that Q,K and T yielded higher circulating ammonia levels than C and H (Fig. S5E). Second, we matched the dose of the five AA (50 mg/AA), regardless of their natural occurrence in food. Despite receiving the same amount of AA, mice treated with Q, K and T still showed substantially higher ammonia levels in the circulation than those exposed to C and H (Fig. 3B). These data support the idea that Q,K, and T are intrinsically more ammoniagenic than other AA.

Using a glutaminase inhibitor (BPTES) and the D-enantiomer of Q, which cannot undergo metabolic transformation, we confirmed that ammonia release was necessary for toxicity and appetite suppression (Fig. 3C,D and Fig. S5F,G). Treatment of catabolic mice with ammonium acetate (NH₄OAc) revealed that ammonia accumulation was sufficient to activate AP/NTS neurons and induce anorexia, regardless of ambient temperature and the mode of administration (bolus vs. continuous delivery via the drinking water) (Fig. 3E–G and Fig. S5H). These effects were specific to ammonia, rather than pH or osmolality, because matched amounts of sodium acetate (NaOAc) had no such effects (Fig. 3E–G and Fig. S5H). When we blocked ammonia scavenging using the glutamine synthetase inhibitor methionine-DL-sulfoximine (MSX) following a fast, hungry mice refused to eat over the course of a full day (Fig. 3H). Moreover, treatment with ammonia evoked mortality in a dose-dependent fashion and the resultant clinical phenotype resembled AA poisoning (diarrhea, tail twitching, seizures) (Fig. S5I). Finally, to show that ammonia accumulation is sufficient to induce behavioral aversion, we paired a novel flavor with NH₄OAc or NaOAc and measured flavor preference following a wash-out period. We found that NH₄OA, but not NaOAc, was sufficient to condition flavor aversion (Fig. S5J).

Having documented aversion, anorexia and toxicity upon ammonia accumulation, we wondered whether these effects were specific to the catabolic state. To test this, we first treated fed or fasted mice with the same amount of NH₄OAc intragastrically. We found that fasted mice died within tens of minutes, whereas fed mice all survived (Fig. 3I). Mortality of catabolic mice following delivery of NH₄OAc was paralleled by excessively elevated serum ammonia levels in the absence of a meaningful increase in blood urea nitrogen (BUN) concentration (Fig. 3J and Fig. S6A), suggestive of a saturation of detoxification capacity. This saturation was likely the result of an increased ammonia detoxification demand arising from muscle proteolysis during the preceding catabolic insult (Fig. S1B), as demonstrated by elevated BUN levels after a fast (Fig. S8B). When we administered a sublethal dose of NH₄OAc in the cold and offered animals HFD, fasted, but not fed mice, developed lethal anorexia (Fig. S8C). Consistent with these observations, intragastric protein administration elicited higher circulating ammonia levels in the fasted compared to the fed state, resulting in augmented appetite suppression (Fig. S6D,E). These changes were not artefacts of force-feeding protein, because ammonia levels were also elevated in the blood of fasted mice refed in the cold ad libitum with HPD, but not HCD, compared to fed controls (Fig. S6F).

Next, we wished to clarify if ammonia was required for protein aversion during recovery. Because no single genetic approach to increase ammonia detoxification exists, we physiologically induced this program by exposing mice to HPD for 1 week (Fig. 3K). This intervention predictably led to an induction of hepatic urea cycle enzymes, increased urea production and renal clearance of urea, resulting in osmotic diuresis and a compensatory increase of water intake (Fig. 3L,M and Fig. S6G–I). In mice in which ammonia detoxification physiology was induced, previously lethal doses of ammonia were no longer lethal (Fig. 3N). Further, these mice did not develop aversion to QKT-rich food and voluntarily increased their intake of the QKT-containing, but not QKT-deficient AA diet, leading to enhanced survival in the cold (Fig. 3O,P and Fig. S6J,K). Lastly, we reasoned that if protein ingestion and ammonia accumulation were indeed mechanistically connected, then both cues should be sensed by an overlapping set of neurons in the brain. To explore this, we administered casein to TRAP2 mice crossed to a Cre-dependent reporter strain(DeNardo et al., 2019; Koren et al., 2021), resulting in permanent labelling of neurons activated by protein ingestion. We then challenged these mice with ammonia and stained for cFos. Using this strategy, we found that a large population of protein-responsive neurons in the AP and NTS were also activated by ammonia (Fig. 3Q).

Taken together, these data show that ammoniagenesis from aversive AA is anorexigenic and toxic, which is physiologically counteracted by voluntary restriction of AA consumption.

Protein restriction requires the wasabi receptor TRPA1.

Based on the observation that ammonia generation from dietary AA was necessary for voluntary restriction of protein intake and development of aversion to ammoniagenic AA during recovery, we wished to understand how ammonia was being sensed (Fig. 4A). The mechanistic basis of ammonia sensing in mammals is not entirely clear, but members of the transient receptor potential (TRP) channel family incuding TRPV1 - the “capsaicin receptor”- and TRPA1 - the “wasabi receptor”- which sense noxious chemicals and temperature, have been suggested to be ammonia responsive(Dhaka et al., 2009). We first deleted TRPV1+ sensory neurons by resiniferatoxin (RTX) treatment(Baral et al., 2018), which diminished responsiveness to the TRPV1 agonist capsaicin (Fig. S7A). Using RTX, we found that TRPV1+ sensory neurons were not required for titration of dietary protein intake (Fig. 4B). Non-neuronal cells expressing TRPV1 were also dispensable because transgenic mice with broad depletion of TRPV1+ cells (Rosa26-DTA^Trpv1-Cre) consumed similar amounts of protein as wildtype littermate controls (Fig. S7B). Therefore, we next explored TRPA1. We found that global TRPA1-deficient mice voluntarily consumed more dietary AA, regardless of whether they were offered as a complex protein or in free form (Fig. 4C,D). This behavior was independent of ambient temperature and enabled survival in the cold (Fig. 4C,D and Fig. S7C,D). We further observed that TRPA1-deficient mice did not develop aversion to QKT-rich food during recovery, were resistant to the appetite suppressing effects of ammonia and showed reduced AP/NTS neuron activation in response to protein ingestion and ammonia accumulation (Fig. 4E–H). These effects could not be explained by heightened ammonia detoxification capacity, because Trpa1^−/− mice died just as quickly as wildtype controls upon NH₄OAc treatment following a fast (Fig. S7E). We also ruled out that TRPA1 deficiency was associated with hyperphagia generally (Fig. S7F). Finally, we found no evidence for TRPA1-dependent control of protein intake at homeostasis, because wildtype and Trpa1^−/− mice ate similar amounts of HPD at steady state (Fig. 4I).

Fig. 4. Protein restriction requires the wasabi receptor TRPA1.

(A) Hypothetical sensing of ammonia by a membrane protein (B) Protein intake of mice refed with HPD following ablation of TRPV1+ sensory neurons using resiniferatoxin (RTX). Controls were treated with vehicle solution (n=8/group). (C) Protein intake and survival of WT and Trpa1^−/− mice refed with HPD at 6°C (n=5/group). (D) Intake of the complete AA diet and survival over time of animals of the same genotypes refed in the cold (n=4–7/group). (E) Preference of fasted wildtype and Trpa1^−/− animals for the QKT-deficient over complete AA diet assessed in a 2-choice assay performed at 22°. Results are expressed as change from baseline preference (delta) (n=9–12/group). (F) Intake of HFD in WT and Trpa1^−/− mice exposed to 0.28M NH₄OAc drinking water during refeeding at 6°C (n=8–9/group). (G) Representative cFos stainings of the AP/NTS of WT and Trpa1^−/− mice treated with casein intragastrically after a fast. Scale bar: 100µm. The quantification of cFos+ cells is shown on the right (n=8–9/group) (H) Representative cFos stainings of the AP/NTS of fasted mice of the same genotypes following intragastric administration of a sublethal dose of NH₄OAc (scale bar: 100µm). The quantification of cFos+ cells is shown on the right (n=3/group). (I) Intake of HPD of the two genotypes at steady state (n=6–8/group). (J) HFD intake of mice treated with wasabi or an isocaloric amount of sucrose (n=6/group). (K) Intake of HFD of mice subjected to the indicated treatments (n=7–8/group). Data is presented as mean ± s.e.m *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (B-K) unpaired, two-tailed t test (C-right panel, D-right panel) log-rank test.

To understand whether TRPA1 activation was sufficient to inhibit feeding, we treated mice with the TRPA1 agonist wasabi. We found that wasabi strongly suppressed food intake in fasted mice compared to an isocaloric amount of sucrose (Fig. 4J). We made comparable observations with allyl isothiocyanate (AITC), a naturally occurring TRPA1 agonist present in wasabi (Bellono et al., 2017), which also blunted feeding (Fig. 4K) in a TRPA1-dependent manner (Fig. S7G). Together, these data identify a requirement of the wasabi receptor TRPA1 for protein restriction during recovery from catabolism.

Enterochromaffin cells of the gut sense ammoniagenesis and restrict protein intake in a TRPA1-dependent manner.

Having identified an interoceptive, TRPA1-dependent pathway controlling protein intake during recovery, we next wished to understand which TRPA1+ cell type was responsible for mediating these effects. We reasoned that identifying the main site of ammoniagenesis might aid in narrowing down potential candidate populations. We administered radioactively labelled Q into the stomach of fasted mice and collected tissues 30 min later to trace its distribution. We found that the duodenum took up large amounts of Q, whereas most other tissues including skeletal muscle and different parts of the brain did not (Fig. 5A). To show that AA accumulation in the gut generates ammonia locally, we next performed ¹⁵N stable isotope tracing. Following intragastric administration of a low dose of ¹⁵Q, we detected robust accumulation of ¹⁵NH₄ in the duodenum (Fig. 5B). Consistent with our ¹⁴C tracing results, ammoniagenesis in the jejunum was much less pronounced (Fig. S7H). Urea molecules recovered from the liver contained ¹⁵N atoms, demonstrating that dietary Q-derived ammonia is detoxified via the hepatic urea cycle (Fig. 5C). These observations led us to hypothesize that ammoniagenesis from dietary AA was sensed by a TRPA1+ cell type in the duodenum, where most nutrients are physiologically absorbed (Sarna and Otterson, 1989).

Fig.5. Enterochromaffin cells of the gut sense ammoniagenesis and restrict protein intake in a TRPA1-dependent manner.

(A) Tracing of radioactively (¹⁴C) labelled L-glutamine across organs following intragastric administration [Abbr.: hth=hypothalamus, cereb=cerebellum]. Results are expressed as counts per minute (CPM) per mg of tissue (n= 3–6/organ). (B) ¹⁵NH₄ in duodenal lysates after intragastric administration of the indicated substances expressed as corrected abundance (n=5–6/group) (C) Corrected abundance of ¹⁵N-urea in liver lysates of the same mice (D) relative transcriptional expression (arbitrary units) of TRPA1 and TPH1 across murine small intestinal epithelial cells visualized as a heatmap. (E) Serotonin levels in supernatants of small intestinal explants from WT or Trpa1^−/− mice treated with NaOAc or NH₄OAc for 15 min ex vivo (n=7–9/group). (F) Representative trace of Ca²⁺ responses in WT and Trpa1^−/− EC cells following the indicated stimulations (G) Quantification of NH₄OAc-induced Ca²⁺ responses in WT and Trpa1^−/− EC cells. Scale bars: 0.2 Δratio (340/380) and 40 s (n=6/group) (H) Protein intake of mice with tamoxifen-inducible, EC-specific TRPA1 deletion (Trpa1^ChgaCreER) or wildtype littermate controls (Trpa1^fl/fl) post fasting (n=12/group) (I) Representative cFos stainings of the AP/NTS of mice of the same genotypes treated with casein intragastrically following a fast (scale bar: 100µm). The number of cFos+ cells is shown on the right (n=6–7/group) (J) Schematic illustration of TRPA1-dependent serotonin release from gut EC. The pharmacological targets of fenclonine (fenclo) and ondansetron (onda) are highlighted (K) Protein intake of mice refed with HPD at 6°C following treatment with fenclonine (fenclo) or vehicle solution (n=9–12/group) (L) Protein intake of fasted mice treated with ondansetron (onda) or vehicle (saline) (n=9/group). (M) Protein intake of mice with (Tph1^Vil-cre) or without (Tph1^fl/fl) enterocyte-specific TPH1 deletion refed with HPD at 22°C (n=7–8/group) Data is presented as mean ± s.e.m *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (B,C) one-way ANOVA with Holm-Sidak’s post hoc test (E) two-way ANOVA with Holm-Sidak’s post hoc test (G-M) unpaired, two-tailed t-test.

To understand whether the TRPA1-expressing cell population in the intestine was short- (epithelium, most leukocytes) or long-lived (neurons, some leukocytes), we administered diphtheria toxin (DT) to mice expressing the diphtheria toxin receptor in a TRPA1-dependent manner(Jin et al., 2024). We measured protein intake early (3 days) and late (19 days) following DT administration, reasoning that short-lived, but not long-lived cells, should have recovered by the second testing session. These experiments suggested that the TRPA1-expressing cell type controlling protein intake was short-lived (Fig. S7I), pointing towards an epithelial or immune cell. Screening of publicly available single cell RNA sequencing datasets(Haber et al., 2017) showed that enteroendocrine cells were the only epithelial TRPA1+ cell population in the murine gut (Fig. 5D). Further analysis revealed that these cells expressed tryptophan hydroxylase (TPH1) (Fig. 5D), the rate-limiting enzyme for serotonin synthesis(Mawe and Hoffman, 2013). Enterochromaffin cells (EC) are the main source of serotonin in the body and function as chemosensors, which transduce information from the gut to the brain(Bellono et al., 2017). Using intestinal explants, we found that ammonia induced serotonin secretion in a TRPA1-dependent fashion (Fig. 5E). To demonstrate direct EC activation by ammonia, we isolated EC from the intestines of Tph1^CFP and Trpa1^−/−Tph1^CFP mice. Monitoring of stimulus-evoked changes in cytoplasmic Ca²⁺ in CFP⁺ EC showed that NH₄OAc, but not NaOAc, induced a robust Ca²⁺ response in a TRPA1-dependent manner (Fig. 5F,G). In contrast, TRPA1-deficiency did not affect depolarization-induced Ca²⁺ responses evoked by extracellular K⁺ (Fig. 5F). Together, these data indicate that TRPA1 sensing of ammonia promotes serotonin release from EC.

Therefore, we next deleted TRPA1 from EC using a tamoxifen-inducible system (Trpa1^ChgaCreER)(Chen et al., 2021). Using immunofluorescene staining, we confirmed that TRPA1 expression was lost on EC of mice carrying the Cre allele, but preserved across the submucosal layer, where nerve fibers are abundantly present(Borgmann et al., 2021) (Fig. S7J). Transcript levels of TRPA1 were also readily detectable in dorsal root ganglia (DRG) of Trpa1^ChgaCreER mice, further arguing against off-target effects of this Cre-driver on neurons (Fig. S7K). Consistent with our hypothesis, mice with genetic deletion of TRPA1 in EC voluntarily consumed more dietary protein and failed to activate AP/NTS neurons in response to protein ingestion (Fig. 5H,I).

We then explored if EC-derived serotonin was required for the appetite-suppressing effects of dietary protein and ammonia (Fig. 5J). Global pharmacological serotonin depletion using the TPH1 inhibitor fenclonine (fenclo) showed that serotonin was necessary for restriction of protein intake, regardless of ambient temperature (Fig. 5K and Fig. S7L). Increased caloric supply allowed for prolonged survival of serotonin-depleted mice in the cold (Fig. S7M). These effects required 5HT3 receptor (HTR3A) signaling (Fig. 5L). Consistently, appetite suppression by ammonia was also serotonin-dependent (Fig. S7N). Mice with genetically-induced serotonin deficiency in intestinal epithelial cells (Tph1^Vil-cre) voluntarily consumed more dietary protein than wildtype littermate controls (Fig. 5M), confirming that serotonin release in the gut was necessary for protein restriction.

Finally, to further corroborate that behavioral restriction of protein intake arises from the gut, we chemogenetically silenced duodenum-innervating neurons by injection of a retrograde adeno-associated virus (AAV) encoding for Cre recombinase into the duodenal walls of mice allowing for Cre-dependent expression of an inhibitory designer receptor exclusively activated by designer drugs (DREADD). Controls received a virus encoding for a fluorescent protein (mCherry) only. We confirmed successful transfection of neurons by imaging for mCherry in the nodose ganglion, where the cell bodies of most sensory neurons innervating the small intestine reside (Fig. S7O) (Bai et al., 2019; Williams et al., 2016). Following delivery of the designer drug clozapine-n-oxide (CNO), mice injected with the cre-encoding AAV consumed more dietary protein than controls (Fig. S7P).

In summary, these data support a model in which ammoniagenesis from dietary AA engages an anorexigenic TRPA1-dependent gut-to-brain pathway. We propose that this mechanism serves as an anticipatory first-line defense against ammonia toxicity with an adjustable activation threshold (Fig. 6A,B).

Fig.6. Anticipatory behavioral defense against dietary AA-derived ammonia toxicity during recovery from catabolic states relayed by a gut-to-brain pathway.

(A) Schematic illustration of the interorgan circuit limiting dietary protein intake via intestinal sensing of dietary ammoniagenesis. We propose that this mechanism serves as an anticipatory, first-line defense against accumulation of harmful levels of ammonia (B) Simplified dose-response curve highlighting the proposed relationship between dietary AA (QKT) ingestion and activation of the TRPA1-dependent gut-to-brain pathway. During recovery from catabolism, the activation threshold is shifted to the left because the host’s detoxification capacity operates close to maximum capacity due to muscle proteolysis-driven ammoniagenesis during the preceding insult. The liability of ingesting exogenous sources of ammonia (QKT) is thus enhanced and countered by set point adjustment.

Discussion

Due to the large variety of foods omnivores encounter, selecting the right dietary sources to meet bodily requirements for essential micro- and macronutrients is not trivial. Yet, animals are remarkably efficient at this delicate task (Korn et al., 2024). Even if dietary composition is skewed experimentally, homeostasis can typically still be maintained. One of these nutritional behavioral defenses includes a compensatory increase in food intake upon dilution of dietary protein content (Lee et al., 2008; Solon-Biet et al., 2016; Wali et al., 2021). But how do animals know when enough of a given nutrient has been consumed? In many cases, feeding is terminated in response to a physiological demand (calories, protein and so on) being met by a meal (Tschöp and Friedman, 2023), culminating in satiation and satiety. The responses underlying satiation and satiety can be engaged in an anticipatory (e.g. suppression of hunger-generating hypothalamic neurons upon smell of food (Chen et al., 2015)) or feedback manner (Hill et al., 2022a). In contrast, accumulation of harmful molecules or metabolites is believed to primarily arise from ingestion of toxins or pathogens (Korn et al., 2024). Our study identifies gut TRPA1-dependent sensing of AA-derived ammonia, a potentially harmful and toxic metabolite, as the molecular substrate of behavioral restriction of protein intake during recovery states. Thus, food consumption can not only be controlled by the detection of caloric or nutrient supply, but also via sensing of a harmful product generated upon ingestion of nutrients. Consistent with the idea that ammonia accumulation is highly aversive and restrains appetite, we found that ammonia is a potent anorexigen, which activates both the AP and NTS and is sufficient to drive behavioral aversion. The threshold at which animals capped protein consumption scaled with their ammonia detoxification capacity, strongly suggesting that ammonia is the factor limiting dietary protein intake (Fig. 6B). We also noted that post catabolic mice were highly susceptible to ammonia poisoning due to a transient saturation of detoxification capacity and force-feeding animals with ammoniagenic AA in unphysiological kinetics turned out to be lethal. We thus propose that voluntary restriction of protein intake during recovery serves as a behavioral defense against ammonia toxicity (Fig. 6A,B). Because loss of TRPA1-dependent sensing of ammoniagenesis in the gut did not result in lethal protein hyperphagia, this pathway appears to relay anticipatory behavioral adaptations, rather than encoding the ‘last resort’ of protein restriction. Accordingly, additional mechanisms preventing excessive protein intake must exist.

We found that TRPA1-dependent control of protein appetite did not operate at steady state, suggesting that this pathway reflects a mode of backup regulation, which is only engaged if homeostatic mechanisms restricting protein intake are dysregulated or dysfunctional (Morton et al., 2014). In line with this thought, many canonical regulators of protein intake such as taste and smell, intestinal motility and endocrine hormones are perturbed during and after illness (Ladopoulos et al., 2018; Schiffman, 2007; Vanhorebeek et al., 2006; Wilmore, 1991). Under such circumstances, TRPA1-dependent gut-to-brain communication overrides homeostatic systems regulating feeding to limit protein intake. This mode of operation is consistent with other pathways conferring defense against ingested toxins by activating the AP (Breen et al., 2020; Mulderrig et al., 2021; Zhang et al., 2021). While there might be some fitness costs related to restriction of protein intake at a higher setpoint as observed upon TRPA1 deficiency, the benefits of increased caloric supply clearly offset any of those costs in extreme environments with high energetic demands as illustrated by improved survival of TRPA1-deficient mice in the cold.

The strong physiological aversion to excess dietary protein during recovery from catabolism as well as the detrimental consequences of force-feeding AA uncovered by our experiments have important clinical implications. First, protein supplementation has been a mainstay therapy for critically ill patients for decades (Heyland et al., 2023). Our results suggest that this approach is potentially harmful, which is supported by recently published clinical trials (Bels et al., 2024; Heyland et al., 2023). Accordingly, it will be valuable to systematically assess physiological nutrient preferences and feeding behaviors following various forms of sickness in preclinical models and in humans. These studies need to be complemented by targeted dietary interventions, which should allow for defining general rules of favourable vs. unfavourable recovery diets. Second, we provide a possible mechanistic explanation for the phenomenon of human protein poisoning, in which individuals exclusively feeding on very lean meat suffer from severe malaise (referred to as rabbit starvation) (Bilsborough and Mann, 2006; Tushingham et al., 2021). Third, patients affected by urea cycle disorders are currently advised to restrict dietary protein generally. Our data implicate that avoiding Q-, K- and T-rich protein sources might aid in mitigating damage when ammonia detoxification is impaired. For example, hemp protein contains roughly 50% less QKT than casein (Gorissen et al., 2018). Adjusting dietary recommendations based on protein source and its AA profile rather than global protein intake could allow for meeting the minimal demands for essential AA, while reducing accumulation of harmful ammonia in such individuals. While the molecular basis of the high ammoniagenicity of Q is intuitive, the same is not true for K, or T. Accordingly, more comprehensive studies characterizing the ammonia generating properties of individual AA as well as the underlying molecular mechanisms are needed. Finally, proteolysis is mainly engaged during advanced stages of starvation and sickness-associated anorexia for fuel supply. This is usually explained by the idea that skeletal muscle is precious and should not be degraded unless necessary for survival (Finn and Dice, 2006). However, animals typically carry plenty of muscle mass that is not immediately required for sustaining vital functions or fitness. Our observations suggest that the physiological partitioning of fuel supply (glycogen, followed by fat and then muscle) may serve to minimize damage from AA metabolism and resultant ammonia accumulation during and after catabolic insults.

In summary, our study identifies an anorexigenic gut-to-brain pathway that restricts dietary intake of ammoniagenic AA during recovery from catabolic insults, thus uncovering a fundamental mechanism by which animals titrate ingestion of nutrients conferring toxicity potential.

Limitations

The mechanistic basis for why Q, K and T are more ammoniagenic than other AA remains unknown, which is outside the scope of this study. Due to the lack of other available methods, our studies addressing the requirement of ammoniagenesis to protein restriction were indirect. Whether TRPA1 directly or indirectly senses ammonia is unclear. Blocking serotonin signalling did not fully recapitulate TRPA1 deficiency phenotypically, suggesting that additional mediators downstream of TRPA1 are involved in protein restriction. The molecular identity of protein- and ammonia-responsive neurons needs to be identified to enable functional manipulation. Finally, our experiments were performed in a single animal facility and thus require validation in additional housing environments with distinct microbial communities.

Resource availability

Lead contact

Further information and requests for resources or reagents should be directed to and will be fulfilled upon reasonable request by the lead contact, Andrew Wang (andrew.wang@yale.edu).

Materials availability

This study did not generate unique new reagents.

Data and code availability

• This paper analyzed existing, publicly available single RNA sequencing data as listed in the key resources table.

• This paper does not report original code.

• All original data reported in this manuscript will be shared by the lead contact upon reasonable request.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
c-Fos (9F6) Rabbit mAb	Cell Signaling Technology	Cat#2250S
anti-mouse GDF15	NGM Bio	n/a
anti-keyhole limpet hemocyanin (KLH) isotype control	NGM Bio	n/a
Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 647	Invitrogen	Cat# A31573
Alexa Fluor 488 anti-GFP antibody, clone: FM264G	Biolegend	Cat#338007
Anti-TRPA1 antibody	Alomone Labs	Cat# ACC-037
Chemicals, peptides, and recombinant proteins
cholecystokinin octapeptide	Bachem	Cat# 4033010.0001
recombinant human GDF-15 (CHO-expressed) protein	R&D	Cat#957-GD-025/CF
recombinant human GDF-15 protein	Abcam	Cat#AB125769
diphtheria toxin	Sigma-Aldrich	Cat#D0564
4-hydroxy tamoxifen	Sigma-Aldrich	Cat# H-6278
lipopolysaccharide from E. coli strain 055:B5	Sigma-Aldrich	Cat# L2880
ampicillin	Sigma-Aldrich	Cat#A0166
vancomycin	Sigma-Aldrich	Cat# V2002
metronidazole	Sigma-Aldrich	Cat# M1547
D-mannitol	Sigma-Aldrich	Cat#M4125
methionine-DL-sulfoximine	Medchemexpress	Cat#HY-B1692
ondansetron	Sigma-Aldrich	Cat#O3639
clozapine-n-oxide	Fisher Scientific	Cat#A3317–50
GsMTx4	Tocris	Cat# 4912
devazepide	Medchemexpress	Cat#HY-106301
MK0893	Medchemexpress	Cat#HY-50663
resiniferatoxin (RTX)	Alomone Labs	Cat#R-400
fenclonine	Medchemexpress	Cat#HY-B1368
ally isothiocyanate (AITC)	Sigma-aldrich	Cat# 377430
BPTES	Medchemexpress	Cat# HY-12683
Intesticult™ Organoid Growth Medium (Mouse)	Stemcell Technologies	Cat# 06005
Corning™ Matrigel™ Gfr Membrane Matrix	Corning	Cat# 356231
Tryple	Thermo scientific	Cat# 12604021
Fura-2 AM	Thermo scientific	Cat# F1201
L-Asp	Sigma-aldrich	Cat# A8949
L-Cys	Sigma-aldrich	Cat# 30089
L-His	Sigma-aldrich	Cat# H8000
L-Gly	Sigma-aldrich	Cat# G7126
L-Gln	Sigma-aldrich	Cat# G3126
L-Glu	Sigma-aldrich	Cat#G1251
L-Lys	Sigma-aldrich	Cat# 62840
L-Leu	Sigma-aldrich	Cat# L8000
L-Iso	Sigma-aldrich	Cat# 58879
L-Met	Sigma-aldrich	Cat# M9625
L-Phe	Sigma-aldrich	Cat# P5482
L-Pro	Sigma-aldrich	Cat# P0380
L-Ser	Sigma-aldrich	Cat# S4500
L-Tyr	Sigma-aldrich	Cat# 93829
L-Trp	Sigma-aldrich	Cat# T0254
L-Thr	Sigma-aldrich	Cat# T8441
L-Val	Sigma-aldrich	Cat# 94619
D-Gln	Sigma-aldrich	Cat#G9003
L-Glutamine-13C5,15N2	Sigma-aldrich	Cat# 607983
14C L-Gln	Revvity	Cat#NEC451050UC
solvable	Revvity	Cat# 6NE9100
hionic-fluor	Revvity	Cat# 6013311
sucrose	Sigma-aldrich	Cat#84097
ammonium acetate (NH4OAc)	Sigma-aldrich	Cat#32301
sodium acetate (NaOAc)	Sigma-aldrich	Cat#S8750
Tween80	Sigma-aldrich	Cat#P1754
carboxymethylcellulose	Sigma-aldrich	Cat#C4888
Peg300	Sigma-aldrich	Cat# 8.07484
olive oil	Thermo scientific	Cat#416540250
peanut oil	Sigma-aldrich	Cat#P2144
tamoxifen	Sigma-aldrich	Cat#T5648
Critical commercial assays
Serotonin ELISA	Eagle Biosciences	Cat#EA602/96
Mouse/Rat Fgf21 ELISA	R&D	Cat#MF2100
Mouse/Rat Metabolic Hormone Discovery Assay Array	Eve Technologies	n/a
ammonia assay	Abcam	Cat# ab83360
RNA STAT-60	Tel-Test, Inc.	Cat#CS-110
SuperScript IV	invitrogen	Cat#1890050
RNeasy mini kit	Qiagen	Cat#74106
Experimental models: Organisms/strains
C57BL/6J	The Jackson Laboratory	#000664;RRID:IMSR_JAX:000664
Balb/c	The Jackson Laboratory	#000651;RRID:IMSR_JAX:000651
R26-LSL-Gi-DREADD: B6.129-Gt(ROSA)26Sortm1(CAG-CHRM4*,-mCitrine)Ute/J	The Jackson Laboratory	#026219;RRID:IMSR_JAX:026219
TRAP2: Fostm2.1(icre/ERT2)Luo/J	The Jackson Laboratory	#030323;RRID:IMSR_JAX:030323
Ai14: B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J	The Jackson Laboratory	#007914;RRID:IMSR_JAX:007914
R26-iDTR: C57BL/6-Gt(ROSA)26Sortm1(HBEGF)Awai/J	The Jackson Laboratory	#007900;RRID:IMSR_JAX:007900
Trpa1 IRES Cre	Charles Zuker	n/a
Tph1 flox	G. Karsenty	n/a
Vil Cre: B6.Cg-Tg(Vil1-cre)1000Gum/J	The Jackson Laboratory	#021504;RRID:IMSR_JAX:021504
Trpm5−/−: B6;129-Trpm5tm1Csz/J	Ruslan Medzhitov	#013068;RRID:IMSR_JAX:013068
Tph1CFP: Tg(Tph1-CFP)1Able/J	The Jackson Laboratory	#028366;RRID:IMSR_JAX:028366
Fgf21 flox: B6.129S6(SJL)-Fgf21tm1.2Djm/J	David J. Mangelsdorf	#022361;RRID:IMSR_JAX:022361
Alb-Cre: B6.FVB(129)-Tg(Alb1-cre)1Dlr/J	The Jackson Laboratory	#016832;RRID:IMSR_JAX:016832
Fgf21−/−: Fgf21tm1.1Djm	David J. Mangelsdorf	RRID: MGI_4354176
Trpa1−/−: C57BL/6NJ-Trpa1em1(IMPC)J/Mmjax	The Jackson Laboratory	RRID:MMRRC_0460 59-JA
Trpa1 flox	Cheryl L. Stucky	n/a
chromogranin cre-ER: Chgatm1(EGFP/cre/ERT2)Wtsi	EMMA	RRID:IMSR_EM:095 74
Trpv1-Cre: B6.129-Trpv1tm1(cre)Bbm/J	The Jackson Laboratory	#017769;RRID:IMSR_JAX:017769
Rosa26-DTA flox: B6.129P2-Gt(ROSA)26Sortm1(DTA)Lky/J	The Jackson Laboratory	#009669;RRID:IMSR_JAX:009669
Ucp1−/−: B6.129-Ucp1tm1Kz/J	The Jackson Laboratory	#003124;RRID:IMSR_JAX:003124
Adrb3 −/−	Natasa Petrovic	n/a
Alox5−/−: B6.129S2-Alox5tm1Fun/J	The Jackson Laboratory	#004155;RRID:IMSR_JAX:004155
RMB: B6. Ms4a2tm1Mal	Ruslan Medzhitov	n/a
Hdc −/−	Christopher Pittinger	n/a
Gfral −/−	Ruslan Medzhitov	n/a
Ppara−/−: B6;129S4-Pparatm1Gonz/J	The Jackson Laboratory	#008154;RRID:IMSR_JAX:008154
Oligonucleotides
Fgf21 primer F: CCTCTAGGTTTCTTTGCCAACAG R: AAGCTGCAGGCCTCAGGAT	Sigma	n/a
Cps1 primer F:ACATTGGCAGCAGTGTGGAA R:TGGCCGCCAACTGATATGAT	Sigma	n/a
Ass1 primer F: AATGACCAGGTCCGCTTTGA R: AGGTGCTTGATTCTTGGGGT	Sigma	n/a
Asl primer F: GGCTGTGTTTGAAGTGTCTGAC R: TGCCATGAACACAGCTTTCC	Sigma	n/a
Otc primer F: TGAGTGCTGCAAAATTCGGG R: ATACATTGCCTCCACGTGCT	Sigma	n/a
Rpl13a primer F: GAGGTCGGGTGGAAGTACCA R: TGCATCTTGGCCTTTTCCTT	Sigma	n/a
Trpa1 primer F: GCAGGTGGAACTTCATACCAACT R: CACTTTGCGTAAGTACCAGAGTGG	Sigma	n/a
Recombinant DNA
retrograde pAAV-hSyn-Cre-P2A-dTomato	Addgene	Cat#107738-AAVrg
retrograde pAAV-hSyn-mCherry	Addgene	Cat# 114472-AAVrg
Software and algorithms
Prism 10	Graphpad Inc.	www.graphpad.com
ImageJ	NIH	https://imagej.nih.gov/ij/download.html
Single Cell Portal	Broad Institute	https://singlecell.broadinstitute.org/single_cell
ImarisViewer 10.2.0	Oxford Instruments	https://imaris.oxinst.com/imaris-viewer
BioRender	BioRender	www.biorender.com
Primerblast	NIH	https://www.ncbi.nlm.nih.gov/tools/primer-blast/
Primer3	n/a	https://primer3.ut.ee
NIS-elements	Nikon	https://www.microscope.healthcare.nikon.com/products/software/nis-elements
Other
Rodent Diet, High Protein, Amino Acid Defined, Complete Amino Acid	BioServ	Cat#F10611
Rodent Diet, High Protein, Amino Acid Defined, Lysine, Glutamine and Threonine Deficient	BioServ	Cat#F10610
Rodent Diet, High Protein, Amino Acid Defined, Glutamine Deficient	BioServ	Cat#F10628
Rodent Diet, Purified, High Carbohydrate	BioServ	Cat#F10520
Rodent Diet, Purified, High Protein	BioServ	Cat#F10518
Rodent Diet, Purified, High Fat	BioServ	Cat#F10519
100% micellar casein powder	It’s Just!	n/a
100% beef protein isolate powder	Equip	n/a
Wasabi Powder	Soes	n/a

STAR★Methods

Experimental model and study participants details

Mice

All animal experiments were performed in accordance with institutional regulations after protocol review and approval by Yale University’s Institutional Animal Care and Use Committee. Animals were used between the age of 7–14 weeks, except for mice subjected to RTX treatment as described elsewhere in this section. Mice were obtained from commerical sources or donating investigators as outlined in the key resources table and bred at Yale School of Medicine. Both male and female mice were used. Genotyping of genetically modified mice was performed according to standard protocols. Strains with tamoxifen-inducible Cre expression (Cre-ER) received 1mg tamoxifen i.p. dissolved in peanut oil on 5 consecutive days. Littermate controls (Cre-) were treated the same amount of tamoxifen. Mice were kept under a 12h light:dark cycle in groups of 5 (males) to 6 (females) at 22°C with free access to standard chow and water until they were subjected to experimental procedures as outlined below. All mice were on a C57BL/6J background except Gfral^−/− mice, which were on a Balb/c background.

Recovery from catabolic states and experimental diets

Group-housed mice were fasted in wire-bottom cages at room temperature. Food was removed between ZT22 and ZT1, followed by introduction of customized diets 24 to 72h later. During the fast, mice had free access to water. After completion of the fast, mice were single-housed in GM500 cages (Techniplast) with conventional bedding and 10–25 g of food pellets were provided in the hopper. Food intake was tracked at various intervals as indicated in the figures. Isocaloric, macronutrient-biased (Bioserv) as well as AA-defined diets were designed together with nutritionists from the vendor (for composition see Table 1). Macronutrient intake was calculated based on the nutrient composition of the respective diets. Feeding studies at various ambient temperatures were performed in climatic chambers (Memmert, Germany). Here, cages with bedding and water bottles, but without nestlet material were temperature-adapted for 24h before being used. Mice were fasted at 22°C as outlined above, followed by refeeding at 6, 22 or 30°C. Survival, food intake and body weight were monitored once daily in these experiments. Pair-feeding studies were performed following preliminary experiments establishing kinetics of food consumption over time across paradigms (LPS and fasting). In brief, mice were supplied with new food every day, whereby the amount of food provided was based on the daily average consumption of the HPD group, which consistently ate the least amount of diet. Inflammatory catabolism was induced by intraperitoneal (i.p.) injection of LPS. The effects of LPS are heavily lot-dependent. With our LPS batch, we established that 2.5 mg/kg induced transient anorexia lasting for 24h, together with weight loss and tissue wasting, all of which resembled one day of fasting.

Method details

Nutrient and ammonia supplementation

Micellar casein was dissolved in water and administered intragastrically immediately prior to refeeding (i.e. after 24h of fasting) with HFD. A second bolus was applied 4h later, yielding a total of 0.566g of pure protein. Controls received an equivalent amount of water. The appetite suppressing effects of protein supplementation were calculated based on the % reduction of HFD intake compared to water treated controls (fed and fasted, respectively). For ammonia measurement following protein ingestion, fed and fasted mice were treated with 1 ml of diluted casein (5g powder in 30 ml of water) at ZT15. Blood was collected 45 min later and immediately analysed for ammonia concentration as described elsewhere in this section. In an independent set of experiments, fasted mice received an intragastric bolus of casein or an isocaloric bolus of olive oil or water and blood was sampled 45 min later for ammonia analysis. For the in vivo AA screen, one of the 20 proteinogenic AA was administered intragastrically (dissolved in 1 ml of water) as a bolus, followed by ad libitum refeeding with HFD. The amount of each AA administered was based on the published AA profile of casein(Lapierre et al., 2012) as well as the ad libitum consumption of HPD during refeeding established in preceding experiments. This yielded a dose of each AA, which mice consumed physiologically within the first day post fasting (‘food-adjusted dose). Hydrophobic AA were administered in suspension. Animals were monitored every 15 min for the first 3h, followed by subsequent hourly monitoring. Food intake was tracked after 24h. The same protocol (‘food-adjusted dose’) was used to collect blood 30 min after AA administration for measurement of ammonia. For combinatorial treatment of fasted mice with QKT or CNH, AA doses were food-adjusted as described above or matched (30 mg per AA) and administered intragastrically in 1 ml of water. In an independent set of experiments, the dose of selected AA was split into 2 sub-doses (60 and 40% of total), which were administered intragastrically 6h apart (total volume: 1 ml). Controls received an isocaloric amount of sucrose in these experiments. Sodium acetate (NaOAc) and ammonium acetate (NH₄OAc) were dissolved in water and administered intragastrically or supplied via the drinking water (0.28 M) at 22 or 6°C as indicated in the figures. To compare the anorexigenic properties of NH₄OA between the fed and fasted state, animals were treated with a single intragastric bolus of NH₄OAc (15 mg/mouse), followed by refeeding with HFD in precooled cages at 6°C. The same dose was used for ammonia measurements in the blood. For survival read-outs, mice received 20 mg NH₄OAc. Variations of these experimental approaches are specified in the respective figures legends as applicable. For habituation experiments, mice were group-housed and exposed to isocaloric HFD or HPD for 7 days, before being fasted for 24h in wire-bottom cages. Following completion of the fast, mice were either refed with the complete or QKT-deficient AA diet in precooled cages at 6°C or challenged with 100 mg L-Gln or 20 mg NH₄OAc intragastrically (in 1 ml of water). Survival was monitored or blood was collected for ammonia measurement (30 min post gavage). For preference testing, mice were habituated to HPD or HFD for 7 days before baseline preference was assessed. Following a wash out period and a 24h fast, preference for the AA(-QKT) vs AA diets was measured as described elsewhere in this section. Wasabi powder (Soeos) was diluted in water (20% solution) and administered intragastrically immediately prior to ad libitum refeeding with HFD. Controls received an isocaloric amount of sucrose. Mice were kept on 5% wasabi water or an isocaloric sucrose solution throughout the remainder of the experiment before food intake was tracked after 24h.

Chemogenetics

Mice allowing for cre-dependent expression of an inhibitiory DREADD (R26-LSL-Gi-DREADD) were anesthesized with ketamine/xylazine and placed on a heating pad. The peritoneal cavity was accessed using a small incision. The duodenum was exposed and injected intramurally at mutliple sites with a retrograde AAV encoding for Cre and a fluorescent protein (retrograde pAAV-hSyn-Cre-P2A-dTomato) or a fluorescent protein only (retrograde pAAV-hSyn-mCherry) (both diluted 1:10 in 0.9% saline) using 33G stereotactic hamilton syringes. The abdomen was closed using sterile sutures, followed by injection of 500µl sterile saline. Mice were kept on heating pads for 24h and received moisturized chow. Four weeks later, animals were fasted in wire bottom cages and injected with clozapine-n-oxide (CNO, 1 mg/kg) immediately prior to refeeding with HPD. A second dose of CNO was administered 12h later. The nodose ganglion was collected microscopically and analysed for mCherry expression.

Activity-dependent labelling of neurons (‘TRAPing’)

TRAP2 mice express Cre-ER under the Fos promoter, yielding Cre-ER expression following neuronal activity. We crossed these mice to a reporter strain (Ai14). In the resultant offspring, tamoxifen, delivered during a challenge of interest, allows Cre-ER to translocate to the nucleus and permanently label with TdTomato (‘TRAPing’) (DeNardo et al., 2019; Koren et al., 2021). To ‘trap’ protein-responsive neurons, we acclimatized TRAP2-Ai14 mice for 4 days in temperature-, noise- and light-controlled chambers. All mice were single-housed. We restrained animals daily and subjected them to sham injections to minimise stress during experimental procedures. We then fasted mice for 24h and treated them with an intragastric bolus of casein (1 ml/mouse, 5g casein powder in 30 ml water), immediately followed by i.p. injection of 4-hydroxytamoxifen (50 mg/kg). Mice were maintained in the chambers for an additional 8h without food (the ‘traping’ window), before being refed. Two weeks later, the same mice were fasted, treated with 15 mg NH₄OAc intragastrically and brains were collected 90 min later for immunofluorescence staining.

Conditioned flavor aversion paradigm

Mice were acclimatized in Techniplast rat cages with two food hoppers and two water bottles for 3 days. Water intake from both bottles was measured daily to rule out a detectable side preference at baseline. Next, a single bottle of sugar free cherry-flavored Kool-Aid was introduced for 24h at one side of the cage, whereas the second sipper remained empty. Mice had free access to this flavor for 24h (unpaired stimulus). Next, mice were fasted in wire-bottom cages for 24h, followed by pairing sugar free grape-flavored Kool-Aid with either NH₄OAc or NaOAc. Pairing was achieved by intraoral administration of the flavor using a 1 ml syringe (8×1 drop), immediately followed by intragastric administration of NH₄OAc or NaOAc (15 mg in 1 ml of water/mouse) as previously described(Huang et al., 2024). After a wash-out period with free access to unflavored water and food, the two flavors were introduced (one per sipper), and mice were allowed to freely choose between the two bottles for 24h. Fluid intake was measured on the next day (intake = ml baseline - ml at 24h). Results were expressed as absolute intake of each flavor in ml per 24h. Cages with leaking bottles (>10ml in 24h) were excluded from analysis.

2-choice paradigm

Mice were offered the complete or QKT-deficient AA diet simultaneously in a food hopper with a divider. After 24h, food intake was tracked and the preference ratio for the QKT-deficient diet was calculated (preference for AA(-QKT)=intake AA(-QKT):total intake). All animals then received regular chow diet for 1 day, after which some animals were fasted in wire-bottom cages (fasted group), whereas others continued to received chow for a second day (fed controls). On the third day, all mice were again offered both diets simultanously and the preference ratio was calculated 24h later. Results were expressed as change in preference for the QKT deficient AA diet over time (preference day 2 – preference day 1). The experimental design is schematically summarized in Figure S5 L.

Antibiotics

For microbiota depletion, mice received an antibiotic cocktail (ampicillin, vancomycin and metronidazole at 1, 0.5 and 0.5g per liter, respectively) supplied via the drinking water for 2 weeks before being subjected to experimental procedures. The antibiotics were continued throught the experimental paradigm and changed twice weekly. Splenda (4 packs per liter) was added for taste.

Pharmacological interventions

GsmTx4 (Tocris) was freshly prepared in PBS and injected i.p. at 0.54 mg/kg (as previously described (Hill et al., 2022b)) 0.5 h prior and 6h post refeeding. The same administration scheme was chosen for devazepide and MK0893 (both from Medchemexpress), which were injected at 1 mg/kg and 10 mg/kg, respectively. BPTES (Medchemexpress) was administered i.p. (12.5 mg/kg) 18h and 0.5h prior to an intragastric challenge with 100 mg L-Gln. Neutralizing antibodies directed against GDF-15 or an isotype control (anti-KLH) (both kindly provided by NGM bio) antibody were likewise i.p. injected 18h and 0.5h prior to refeeding. Both antibodies were dissolved in PBS and administered at 10 mg/kg(Luan et al., 2019). Because Gfral^−/− mice were on a Balb/c background, the GDF15-neutralization experiments were performed in Balb/c, rather than C57BL/6J mice. We performed the following validation experiments: 1.) fasted mice were injected with GsmTx4 (0.54mg/kg) or PBS, followed by intragastric administration of D-mannitol (0.256 g/ml, 500µl/mouse) or water(Bai et al., 2019). Brains were collected 90 min later for cFos staining of the AP/NTS; 2.) following treatment with GDF15 neutralizing antibodies or an isotype control (see above), fasted mice received an i.v. injection with rGDF15 (5µg in 200 µl PBS) or vehicle and HFD intake was measured after 1h. Wildtype and Gfral^−/− mice were likewise treated with rGDF15 (2.5µg in 250µl PBS) or vehicle solution after a fast; 3.) devazepide- or vehicle-treated mice received an i.v. injection with CCK octapeptide (0.5µg in 200µl) or PBS immediately prior to refeeding and HFD intake was assessed after 1h. RMB, Alox5^−/− and Trpm5^−/− strains were validated by us previously(Florsheim et al., 2023). Allyl isothiocyanate (AITC) was diluted in olive oil to 10 mg/ml and administered via intragastric gavage at a dose of 50 mg/kg immediately prior to refeeding. A second dose was applied 6h later. Controls received a matched volume of pure olive oil. For fenlconine administration, carboxymethylcellulose (CMC) was dissolved in PBS (0.5%) under continuous shaking over night at 4°C. Next, fenclonine was added to 0.5% CMC to yield a final drug concentration of 10 mg/ml. Fenclonine was then administered intragastrically in suspension at 100 mg/kg 18h and 0.75h prior to refeeding. For studies in the cold, the drug was administered a third time 24h after refeeding. Controls received a matched volume of 0.5% CMC. Methionine-DL-sulfoximine (MSX) was injected at 50 mg/kg 45 min prior to refeeding. Ondansetron was injected immediately prior to food introduction (1mg/kg) and 12h later at the onset of the dark cycle. Controls received vehicle solution (0.9% saline). All other drugs were dissolved in a mix of DMSO (5%), water (60%), Tween80 (5%) and PEG300 (30%) if not specified otherwise. Controls received the corresponding vehicle solution (water, DMSO, Tween80, PEG300).

Metabolic cage studies

Mice were single-housed and acclimated for two days in metabolic cages (Promethion, Sable Systems International) with free access to water, conventional chow and a running wheel prior to experimental procedures. Mice were then fasted for 24h in the metabolic cage (without a wire-bottom), followed by introduction of one of the three isocaloric, macronutrient-biased diets. Activity (beam breaks), wheel running, energy expenditure, food intake, water intake, VO₂, VCO₂, and RQ were measured through indirect calorimetry over 6 days.

EC calcium imaging

Tph1^CFP and Trpa1^−/− Tph1^CFP mice aged 6–10 weeks were used to generate intestinal organoids. The small intestine was isolated and washed with cold PBS and crypts were isolated following dissociation in EDTA. Isolated crypts were suspended in Matrigel. Following polymerization, IntestiCult^™ Organoid Growth Medium (Stemcell) was added and refreshed every 3–4 days. For imaging, organoids were dissociated into single cells with TrypLE (Gibco) and cultured in DMEM containing 5% heat-inactivated FBS, 1% pen/step, 1% L-glutamine, and 10 µM Y-27632. Cells were plated on coverslips coated with 5% w/v matrigel. Before imaging, cells were loaded with 4µM Fura-2 AM (Invitrogen) in culture medium at 37°C for 45 min. Cells were then washed with calcium imaging buffer (135mM NaCl, 5mM KCl, 2mM CaCl₂, 0.5mM MgCl₂, 0.5mM MgSO₄, 0.44mM KH₂PO₄, 0.34mM Na₂HPO₄, 10mM HEPES, 10mM glucose, 30mM sucrose, pH 7.45) and analyzed 30 min later. Fluorescence at 340 nm and 380 nm excitation wavelengths (F340, F380) of CFP⁺ cells was recorded using an inverted Nikon Ti-S microscope with NIS-Elements imaging software (Nikon Instruments). The ratio of fluorescence intensities (F340/F380) was used to reflect [Ca²⁺]i values. The threshold of activation was defined as three standard deviations above the average. 100µM NH₄OAC, 100µM NaOAC, and 100µM KCl were added during the recording as indicated.

TRPV1+ neuron and TRPA1+ cell depletion

Resiniferatoxin (Alomone Labs) was dissolved in pure DMSO and stored in 200µg/ml aliquots. For TRPV1+ neuron ablation, RTX was injected subcutaneously into 4-week-old mice at increasing doses on three consecutive days (30/70/100µg/kg). Controls received an equivalent amount of vehicle (PBS, DMSO and Tween80). Successful depletion of Trpv1 neurons was confirmed by evaluation of the response of mice to a drop of capsaicin onto the eye. All mice (100%) treated with RTX failed to show eye wiping upon capsaicin exposure, whereas all control mice showed vigorous wiping. For depletion of TRPA1+ cells, we crossed mice expressing diphtheria toxin receptor in a Cre-dependent manner with Trpa1-Cre animals. Mice were injected with 200 ng of DT dissolved in PBS i.p. on three consecutive days. Wildtype controls received the same amount of DT. Two and eighteen days after the last DT injection, mice were fasted and refed with HPD 24h later.

Urine quantification and blood/urine chemistry analyses

Daily urine excretion was assessed in a customized metabolic cage system of the George M. O’Brien Kidney Center at Yale School of Medicine. In brief, mice were fasted for 24h, followed by refeeding in cylindrical metabolic cages with grates as flooring (1 mouse/cage). A silicone-coated glass bulb was installed below the grate, allowing the urine to be collected in 10 ml tubes sitting below the bulb. Tubes were changed daily and urine excretion was assessed by measuring the amount of collected liquid. Serum und urine samples were subjected to the core facility for measurement of creatinine, BUN, electrolytes and albumine by HPLC and standard colorimetric assays, respectively.

Validation of dietary amino acid composition

Specimens of the complete AA diet (100g) were shipped to EUROFINS (Iowa, USA) and the complete AA profile was assessed analytically using a modified version of the AOAC 982.30 method. Results were correlated with the AA profile indicated by the supplier of the diets (Bioserv).

Body composition measurements

Analysis of body composition was achieved using an Echo MRI-1000 machine. Baseline measurements for each mouse were taken prior to any interventions. Changes in lean and fat mass were tracked longitudinally and expressed as % change from baseline.

Immunofluorescence staining

Mice were fasted in groups of 5–6 in wire-bottom cages overnight before being subjected to experimental procedures (refeeding, gavage) for which they were single-housed. Animals were sacrificed 90–180 min later. The left ventricle of the heart was punctured using a thin needle and the right atrium was cut with surgical scissors, after which animals were perfused with 10 ml of ice-cold PBS, followed by 10 ml of ice-cold 4% formaldehyde. Brains were collected and stored in 4% formaldehyde for 24h at 4°C before being transferred into a 30% sucrose (in PBS) solution. Next, fixed brains were sliced into 60 µm sections using OCT and a cryostat. Sections were kept at 4°C in PBS prior to staining. For staining of cFos, free-floating brain sections were rinsed with PBS (4×10 min), washed for 1h in PBS supplemented with 0.3 % TritonX-100, followed by 1h incubation in blocking buffer (PBS with 0.3% Triton-X 100 and 10% Blocking One). Sections were then incubated with anti-cFos antibody (1:1000, Cell Signaling Technology) at 4°C overnight. On the next day, sections were rinsed in PBS again (4×10 min), and a fluorescently labeled secondary antibody (donkey anti-rabbit Alexa fluor 647, Invitrogen) was added at 1:1000 dilution for 2–3h at room temperature. Following another rinse in PBS (4×10 min), sections were stained with DAPI (1:2000 in PBS), rinsed again, and mounted onto glass slides. Tiled images were captured on a Leica Stellaris 8 DIVE Multiphoton Microscope using a 20X objective in a blinded fashion. The number of cFos-positive cells was quantified in a blinded manner using ImageJ. For staining of gut sections, the duodenum was collected, flushed with ice-cold PBS to remove stool, and placed in 4% formaldehyde. After fixation, the duodenum was sliced longitudinally and rolled to form a “Swiss roll.” The tissue was flash-frozen using VWR freeze spray and embedded in OCT for cryostat sectioning; 35 µM thick sections were directly mounted on slides. For staining, slide-mounted sections were encircled with a PAP pen. They were then rinsed with PBS (3×5 min), permeabilized with 0.1 % TritonX-100 (1×10 min), and blocked for 1 h in blocking buffer (PBS with 0.1% Triton-X 100 and 10% Blocking One). Slides were then incubated with primary antibodies [AF488 anti-GFP clone FN264G (1:1000, Biolegend) and anti-TRPA1 (1:200, Alomone Labs)] at 4°C overnight. The next day, sections were washed with PBST (3×10min), then incubated for two hours with secondary antibodies [donkey anti-rabbit Alexa fluor 647 (1:1000, Invitrogen)]. Slides were then washed again with PBST (3×10min), followed by one wash with PBS for 10 min. Tiled images were captured on the Leica Stellaris as above.

Metabolic hormone screen

Whole blood (100µl) was collected by retroorbital bleeding using a thin capillary into 1.5 ml eppendorf tubes preloaded with DPPIV (merck millipore, 50µM final concentration) and a protease inhibitor cocktail (Sigma-aldrich, 1x final concentration) to prevent degradation of hormones by endogenous proteases and peptidases. Blood was collected from fed and fasted mice as well as mice refed with different diets 3 and 8h post food introduction as indicated in the figures. Samples were allowed to clot for 20 min at room temperature, followed by centrifugation at 5000 RPM for 15 min at 4°C. Serum was collected, diluted 1:1 in PBS and stored at −80°C until being shipped on dry ice for multiplex analysis of metabolic hormones (Eve Technologies). Results for each hormone were expressed as fold change of fasted controls and visualized using heat maps.

Gene expression analysis

Tissues were collected and immediately placed into ceramic beads containing microtubes loaded with guanidinum thiocyanate sitting on ice. Homogenization of tissue was achieved using a bead mill homogenizer (Omni International). Following chloroforme addition and centrifugation, the aquaeous phase was loaded onto columns and RNA was isolated according to the manufacturer’s instructions (RNaeasy kit, Qiagen). Complementary DNA (cDNA) was synthesized using reverse transcriptase (Superscript IV, invitrogen) and oligo-dT6 primers (Sigma-Aldrich). qRT-PCR was performed on a CFX96 Real-Time System (Bio-Rad) using PerfeCTa SYBR Green SuperMix (Quanta Biosciences). Relative expression units were calculated as transcript levels of target genes relative to Rpl13a. Primers used for qRT-PCR are listed in the key resources table

Ammonia measurement

Whole blood was collected into prechilled 1.5 ml eppendorf tubes sitting on ice. Samples were immediately centrifuged at 4°C for 15 min at 5000 RPM, followed by ammonia measurement using a colorimetric assay (Abcam) within 30 min of collection according to the manufacturer’s instructions. The assay was performed under a hood with continuous air flow to avoid contamination of samples with environmental ammonia. Samples were diluted 1:5 to 1:10 in assay buffer and analyzed on a plate reader (Biotek Synergy HT). Ammonia/ammonium concentration was calculated using the following formula: B/V*D (B=ammonia/ammonium concentration calculated based on standard curve; V= sample volume added to wells in µl [=50] and D= dilution factor (= 5 to 10)).

¹⁴C L-glutamine tracing

Mice received a bolus of radioactively labelled L-glutamine (C¹⁴) intragastrically at a dose of 5µCi/mouse following a 24h fast. Animals were sacrificed 30 min later using ketamine/xylazine anesthesia and cervical dislocation; 50–150 mg of tissue was collected into 20 ml scintillation vials preloaded with 1ml of Solvable (Revvity). The brain was dissected into different anatomical regions, which could be distinguished macroscopically. Tissues were weighed, digested at 55°C under continuous shaking for 1–2h, followed by cooling to room temperature. For all tissues other than skeletal muscle, which was immediately subjected to downstream processing, 2×100µl of 30% hydrogen peroxide was added with swirling between additions. Tissues were then re-heated to 55°C until complete decolorization and 5 ml of Hionic Fluor (Revvity) was added. All samples were temperature and light adapted for 1h in the dark before being analysed on a Beckman LS 6000sc liquid scintillation counter (Beckman Coulter, CA). Result were normalised according to tissue weighed and expressed as counts per minute per mg of tissue (CPM/mg).

¹⁵N stable isotope tracing

Fasted mice received an intragastric bolus of 10 mg ¹⁵N L-glutamine diluted in 1 ml of water. Controls received unlabelled glutamine or water. Tissues were collected 30 min later. Gut segments were flushed with ice cold PBS and immediately snap-frozen in liquid nitrogen. Frozen tissues (15–20mg/sample) were homogenized in a Bullet Blender (Next Advance) in 40µL 80% methanol (VWR Chemicals) per mg of tissue. For extraction, homogenates were kept on dry ice for one hour, followed by a 20-minute centrifugation at 4°C. Ammonium was detected as derivatized indophenol using a modified version of the Berthelot method (Spinelli et al., 2017). The Berthelot method is considered the gold standard method to determine ammonium in samples by colorimetry. Here, the derivatized product, indophenol, was determined in two isotopologues [(0) and (+1)] by mass spectrometry from both 15N and 14N labeled ammonium; 20µL of supernatants were used for derivatization. In brief, samples were incubated for 30 min at 37°C with 100µL of aqueous 250mM NaOH (Fisher Chemicals), 10mM disodium hydrogen phosphate (VWR Chemicals), 10% sodium hypochlorite (Thermo Scientific), and 100µL of 1M phenol (Sigma) with 0.005% sodium nitroprusside (Merck) in ethanol (VWR Chemicals). Derivatized samples were stored at 4°C until use for LC-MS analysis. UPLC-MS was performed with a reverse phase chromatography (Acquity BEH C18 column (2.1×100mm, 1.7µm particle size; Waters) with an Agilent Infinity 1290 UHPLC device) coupled to an Agilent 6495C Triple Quadrupole tandem mass spectrometer (both Agilent). Flow rate was run at 450µL/min. A binary buffer system of buffer A (aqueous 15mM glacial acetic acid (Sigma) and buffer B (15mM glacial acetic acid in methanol) was used. The buffers were applied at 99% A for 1 minute, followed by a linear gradient (99% A to 40% A) over 5 minutes, a linear gradient (40% A to 1% A) over 1 minute, and held at 1% A for an additional 3 minutes, followed by a 0.1-minute gradient from 1% A to 99% A. Column temperature was 50°C. The injection volume was 5µL/sample. Samples were ionized with negative polarity mode: nebulizer 40 psi, capillary −1500V, nozzle voltage: 0V, sheath gas temperature 400°C, and sheath gas flow 11L/min. ¹⁴N- and ¹⁵N-ammonium standards (Sigma, Ammonium-15N chloride, ≥98 atom % 15N, ≥99% (CP)) were derivatized and run in parallel both as separate standards and pooled at defined molar ratios. Quantifier transitions were for Indophenol (0) 198.05 -> 169.9, and for Indophenol (+1) 199.05 ->170.9, both with collision energy of 24 and several other qualifiying transitions. Peaks were integrated with Skyline (Agilent). Natural isotope abundance was corrected empirically using the obtained standard curves with linear fit (R=0.99).

Urea and urea isotopologues were detected using a HILIC chromatography. Liver tissue was homogenized as described above. Metabolites were extracted using standard procedures by incubation at −20°C for two hours(Rinschen et al., 2019). Next, extracts were spun down, and the supernatant was dried down using a speedvac, and resuspended in 200ul of 50%ACN/50%H2O for measurement. UPLC-MS was performed with a HILIC chromatography (Acquity BEH amide column (2.1×100mm, 1.7µm particle size; Waters) with an Agilent Infinity 1290 UHPLC device) coupled to an Agilent 6495C Triple Quadrupole tandem mass spectrometer (both Agilent). Gas temperature was 290°C, flow was 12 l/min, Nebulizer 35 psi, sheath gas temperature 350°C, and sheath gas flow 11 l/min. Capillary voltage was 2000 V in positive ion mode. Column temperature was 25°C. Buffer A was 20 mM Ammoniumformate w. 0.1% formic acid, and Buffer B was Acetonitrile + 0.1% formic acid. Flow rate was 0.4ml/min. Gradient of the binary buffer system was as follows: start 95% B, 1.5min 95% B, 12 min 61% B, 14 min 45% B, 16 min 45% B, 17min 95%B, and hold until 21 min. The mass spectrometer was operated in positive ion mode. Urea was detected using specific retention times (2.1min), as well as specific transitions. For urea, transitions were (m/z): 61 -> 44 (collision energy CE = 24), 61 -> 29.2 (CE = 50). For the +1 isotopologue (¹⁵N₁-urea), transitions were 62 -> 44 and 62 -> 29.2 (CE=24) as well as 62 -> 30.2 (CE= 50). For the +2 isotopologue (¹⁵N₂-urea), transitions were 63 -> 45 and 63 -> 30.2. Authentic standards were run in parallel.

Intestinal explants

The small intestine was removed from the stomach, flushed with ice-cold PBS and the duodenum was collected in DMEM F12 supplemented with 10%FCS and 1% pencillin/streptomycin on ice. Following dissection into two pieces, tissue weights were assessed on a precision scale. Next, duodenal explants were incubated with NaOAc or NH₄Oac (10 mM for both) dissolved in DMEM at 37°C and supernatants were collected 15 min later. Debris was removed by centrifugation (5 min, 4°C, 500g) and supernatants were immediately subjected to downstream analysis.

DRG collection

Dorsal root ganglia (DRG) were collected according to previously published protocols(Sleigh et al., 2016). Mice were euthanised using isoflurane anesthesia. Dorsal skin was removed and the vertebral column was dissected and isolated. Excess tissue was removed and the spine was carefully sectioned to expose the spinal cord. Five to eight DRG were harvested and immediately placed into ice-cold RNA STAT-60. DRG from 5–6 mice were pooled into a single tube to ensure sufficient RNA yield.

Enzyme Linked Immunosorbent Assays

Serum samples were diluted appropriately and levels of FGF21 were assessed using an ELISA according to the manufacturer’s instructions (R&D). Levels of serotonin in supernatants of gut explants were measured by a competitive ELISA as described previously(Bayrer et al., 2023). Briefly, 20µl of supernatant were incubated with equalizing reagent in a 96-well plate for 5 min under steady shaking. Next, 20µl of the corresponding samples were loaded onto antibody coated wells and acetylation reagent was added. Following, antiserum addition, several washing steps and enzyme conjugate addition, substrate was loaded into each well and the reaction was stopped 15 min later. Absorbance at 450nm was measured on a plate reader (Biotek Synergy HT). Results were normalized according to tissue weight.

Quantification and statistical analysis

Experiments were performed in an unblinded manner except for counting of of cFos+ cells, for which the responsible individual was blinded. No sample size calculations were performed prior to conducting in vivo experiments. Statistical analyses were performed using Prism 10 (GraphPad Software, Inc.). Student’s two-tailed, unpaired (two independent samples) or paired (two dependent samples) t-test was used for two-group comparison, whereas groups of three or more were compared using one-way analysis of variance (ANOVA) followed by Holm Sidak’s post hoc test adjusted for multiple comparisons. Groups stratified according to two independent variables (e.g. genotype and treatment or time) were compared by two-way ANOVA with Holm-Sidak’s post-hoc test adjusted for multiple comparisons. The log-rank (Mantel-Cox) test was used to compare survival over time. A p-value of less than 0.05 was considered statistically significant. Data are presented as mean ± SEM if not stated otherwise. All experiments were repeated ≥2 independent times.* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.