Populations in both developed and developing countries have experienced catastrophic increases in the consumption of processed foods high in sugar and fat5. These changes in dietary intake have been implicated in increased malnutrition, including over-nutrition linked to a wide range of metabolic disorders and related comorbidities1,6,7.

Sugar and fat are essential nutrients and, consequently, animals have evolved taste-signalling pathways that detect and respond to sweet and fat stimuli, leading to appetitive and consummatory behaviour1,8. Remarkably, mice that lack sweet taste receptors8 can still develop a strong behavioural preference for sugar9. This suggested the existence of a taste-independent signalling pathway driving sugar preference. Indeed, it was recently demonstrated that the development of sugar preference is mediated by the gut–brain axis, independently of the taste system4. Furthermore, artificial sweeteners, although capable of activating the same taste receptors as sugar on the tongue8,10, do not activate the gut–brain sugar circuit, and consequently do not create a preference4. Together, these findings revealed a gut-to-brain, post-ingestive intestinal sugar-sensing pathway driving craving and attraction to sugar4,11,12,13,14.

Here we focus our attention on the neural basis of fat preference. We demonstrate that fat, like sugar, uses the gut–brain axis to drive consumption. Then, we dissect the nature of the receptors and neuronal elements mediating the development of fat preference.

The discovery of post-ingestive mechanisms activated by foods rich in sugar and fat can provide valuable strategies to modulate our sugar- and fat-craving eating habits and help combat obesity and associated disorders, including diabetes and cardiovascular disease.

The development of fat preference

To behaviourally monitor the development of post-ingestive fat preference, we presented mice with a choice between an artificial sweetener (3 mM acesulfame K (AceK)) and fat (1.5% Intralipid) (Fig. 1a). Both stimuli are innately attractive to a naive animal8,15 (Extended Data Fig. 1a,b), but artificial sweeteners do not trigger post-oral preference4,16. Therefore, this fat-versus-sweetener test enables us to monitor the emergence of fat preference from an initial state of no preference to a switch into a strongly appetitive stimuli. Indeed, our results showed that although mice initially preferred the artificial sweetener (Fig. 1a,b, pre), their preference is markedly altered within 24 h of exposure to both choices, such that by 48 h, the mice drink almost exclusively from the bottle containing fat (Fig. 1a,b, post). This behavioural switch illustrates the ability of fat stimuli to post-ingestively induce strong consummatory responses and appetitive behaviour1. This switch is also observed when comparing fat to an equicaloric sugar (Extended Data Fig. 1e,f), showing that calories are not driving the development of fat preference.

Fig. 1: The development of fat preference.
figure 1

a, Left, cartoon illustrating the behavioural arena; mice were allowed to choose between a fat emulsion (1.5% Intralipid (IL)) and an artificial sweetener (3 mM AceK). Preference was tracked by electronic lick counters in each port. Right, cumulative licks for each bottle over the 48 h session. The colour bars at the top show lick rasters for fat (red) and sweet (blue) from the first and last 2,000 licks of the behavioural test. Note that by 24 h the mice begin to drink almost exclusively from the fat bottle (red trace). b, Preference plots for fat versus sweet. In these experiments, mice began the preference test preferring sweet (preference index < 0.5), but in all cases they switched their preference to fat (n = 7 mice, two-tailed paired t-test, P = 1.9 × 10−5) The dashed line indicates the equal preference level (50%). c, Schematic showing stimulation of Fos induction by fat ingestion. Strong Fos labelling is observed in the cNST (highlighted yellow) upon ingestion of 20% IL but not by the control stimulus (0.3% xanthan gum (XG)). Scale bars, 100 µm. d, Quantification of Fos-positive neurons. The equivalent area of the cNST (200 µm × 200 µm; bregma −7.5 mm) was processed, and positive neurons were counted for the different stimuli. Two-sided Mann–Whitney U-test between XG and IL (n = 5 mice), P = 7.9 × 10−3. Data are mean ± s.e.m.

It was shown recently that the immediate attraction to fat is dependent on the TRPM5 channel expressed in taste receptor cells3 (Extended Data Fig. 1c). We hypothesized that if the development of fat preference is mediated via post-ingestive, rather than taste-evoked signalling, it should be independent of TRPM5 function, and consequently TRPM5-knockout mice should still be capable of developing behavioural preference for fat. As predicted, TRPM5-mutant mice, although blind to the taste of fat, remain fully capable of developing strong post-ingestive preference for fat3 (Extended Data Fig. 1d).

Fat preference via the gut–brain axis

For an animal to develop a preference for fat over sweetener, it must distinguish between two innately attractive stimuli. We reasoned that if we could identify a population of brain neurons that respond selectively to the consumption of fat, it may provide an entry to reveal the neural control of fat preference and the basis for the insatiable appetite for fat.

We exposed separate cohorts of mice to three different lipid stimuli (Intralipid, linoleic acid or oleic acid) and to fat-free textural controls (xanthan gum or mineral oil). Using Fos as a proxy for neural activity4,17, we found that fat, but not control stimuli, elicited strong bilateral activation of neurons in the caudal nucleus of the solitary tract (cNST) in the brainstem (Fig. 1c,d and Extended Data Fig. 2a–e). The cNST is a nexus of interoceptive signals conveying information from the body to the brain via the gut–brain axis18,19. If the fat-activated brain cNST neurons are receiving signals originating in the gut, then direct delivery of fat stimuli into the gut should also induce activation of the cNST. We implanted an intragastric catheter in the stomach4 and infused either a fat solution or a vehicle control. As predicted, intragastric infusion of fat, but not of a vehicle, was sufficient to activate the cNST (Extended Data Fig. 2j–l).

Next, we reasoned that if the fat-activated cNST neurons are essential for creating fat preference, then blocking their function should prevent the development of fat preference. We used the targeted recombination in active populations (TRAP) system20 to target Cre recombinase to fat-activated cNST neurons, and bilaterally injected an adeno-associated virus (AAV) carrying a Cre-dependent tetanus toxin light chain21 (TetTox) construct to genetically silence synaptic transmission in the cNST neurons responding to fat (Fig. 2a and Extended Data Fig. 3).

Fig. 2: Fat preference is mediated by the gut–brain axis.
figure 2

a, Left, schematic for silencing fat-stimulated cNST neurons. A TetTox virus was targeted bilaterally to the cNST of TRAP2 mice for silencing. Right, the fraction of AceK versus IL consumption after the 48 h preference test, in control (n = 10) versus TetTox mice (n = 9). Two-sided Mann–Whitney U-test for fat, P = 1.4 × 10−3 (total volume consumed: control, 9.9 ± 2.3 ml; TetTox, 8.3 ± 2.1 ml). Control mice developed a strong preference for IL versus sweetener. By contrast, mice in which fat-activated cNST neurons have been silenced do not show a preference for fat over sweetener. Data are mean ± s.e.m. b, Fibre photometry was used to monitor activity in cNST neurons in response to intestinal delivery of fat. c, Neural responses following 10 s intestinal delivery of fat (10% linoleic acid (LA)) or control sugar (500 mM glucose (Glu)). The solid trace is the mean and the shaded area represents s.e.m. Responses after bilateral vagotomy are shown in green. Note total loss of responses following bilateral vagotomy4. n = 6 mice. NR, normalized response. d, Quantification of neural responses pre- and post-vagotomy. Two-tailed paired t-test, P = 4.6 × 10−8 (sugar), P = 4.9 × 10−8 (fat). Data are mean ± s.e.m. e, Imaging of calcium responses in vagal neurons as stimuli are delivered to the intestines. Heat maps depict z-score-normalized fluorescence traces from vagal neurons identified as fat responders (n = 84 out of 515 cells from 8 ganglia). Each row represents the average activity of a single cell to four trials. Stimulus window (10 s) is indicated by dotted white lines. Note the strong responses to intestinal delivery of fat (10% LA) but not to control stimuli (0.1% XG plus 0.05% Tween 80). Shown below are sample traces of responses to alternating 10 s pulses of control (XG) and fat stimuli (LA). f, Heat maps depict z-score-normalized responses to interleaved 10 s stimuli of fat (10% LA) and sugar (500 mM Glu). Each row represents the average activity of a different neuron during three exposures to the stimulus. Top, 151 neurons that responded to intestinal application of both fat and glucose. Bottom, a separate pool of 153 neurons that responded only to fat. n = 22 vagal ganglia; 1,813 neurons were imaged.

To ensure that the genetic silencing did not affect the immediate attraction to fat (that is, the taste-dependent innate attraction), they were first tested in a standard fat-versus-water two-bottle discrimination assay. Our results showed that the silenced mice still exhibited normal immediate attraction to fat, and were indistinguishable from controls (Extended Data Fig. 2f–i). By contrast, they were unable to develop post-ingestive preference for fat, even after prolonged testing sessions (Fig. 2a).

Fat and sugar activated vagal neurons

To investigate how fat signals are transferred from the gut to the brain, we infused fat stimuli into the gut, and used fibre photometry to simultaneously record neural activity in cNST neurons4 (Fig. 2b). Our results showed that cNST neurons are robustly activated by direct intestinal infusion of fat, with responses tracking the delivery of the stimulus (Fig. 2c,d and Extended Data Fig. 3f–i).

The vagus nerve serves as a key conduit for conveying information from the gut to the brain4,12,13,19,22. If the vagus nerve is required for the transmission of fat signals from the gut to the cNST, then transection of the vagus nerve should prevent the signals from reaching the brain. To test this, we infused the gut with fat (or sugar as a control4) and recorded stimulus-evoked responses in the cNST. Indeed, fat-activated neural responses in the cNST were effectively abolished after bilateral vagotomy (Fig. 2c,d and Extended Data Fig. 3f–i), thus establishing the vagus nerve as the conduit for transmitting the fat signal from the gut to the brain.

To directly examine and monitor the fat responses of vagal sensory neurons, we carried out functional imaging of the nodose ganglion (which contains the cell bodies of vagal neurons). We targeted the genetically encoded calcium indicator GCaMP6s23 to vagal sensory neurons using Vglut2-cre mice4,24,25 (Vglut2 is also known as Slc17a6), and used a one-photon calcium imaging setup coupled to synchronous intestinal delivery of fat to record neuronal responses in vivo4 with real-time kinetics (Fig. 2e and Extended Data Fig. 4a–c). To administer the stimuli, a catheter was placed into the duodenal bulb, and an exit port was created by transecting the intestine 10 cm distally. During each imaging session, the intestine was exposed to a pre-stimulus application of PBS, a 10 s (33 μl) exposure to the fat or sugar stimuli (limited to 10 s to prevent activation of non-selective osmolarity responses4,24), and a 180 s post-stimulus wash (see Methods for details); this regime was repeated at least 3 times for each stimulus. Using this preparation, we showed that intestinal infusion of fat (for example, linoleic acid), but not vehicle control, evoked robust responses in a unique subset of vagal neurons (Fig. 2e and Extended Data Fig. 4c,d); the responses were reproducible and time-locked to stimulus delivery (Fig. 2e and Extended Data Fig. 4c). These neurons responded to a variety of dietary fatty acids (Extended Data Fig. 4d–i), thus defining a distinct class of vagal neurons that are reliably activated by intestinal fat stimuli.

We showed previously that intestinal application of glucose also activates a subset of vagal neurons4, and demonstrated that these, in turn, are part of the essential gut–brain axis driving the development of sugar preference. Next, we sought to examine how vagal neurons respond to these fat and sugar nutrient signals in the gut.

We recorded the activity of vagal neurons to alternating gut stimulation with fat and sugar (10 s of 10% linoleic acid and 10 s of 500 mM glucose). Out of more than 1,800 vagal sensory neurons examined from 22 nodoses, we identified two distinct groups of vagal neurons. One group (around 8% of the total imaged neurons) responded to both sugar and fat. The other, a non-overlapping group (also around 8% of the neurons), responded only to fat, but not to sugar (Fig. 2f and Extended Data Fig. 5a). Notably, the subset responding to sugar and fat was also activated by amino acids (Fig. 3a and Extended Data Fig. 5b,c). These results defined two distinct populations of vagal neurons: one, hereafter referred to as sugar/fat responders, function as sensors for all three essential macronutrients in the gut: sugar, proteins and fat. The other population, hereafter referred to as fat-only responders, responds selectively to intestinal delivery of fat. We note that less than one neuron per nodose was found to respond to intestinal delivery of sugar or amino acids but not fat (Extended Data Fig. 5d); however, given such small numbers, these were not considered further (it is likely that they represent sugar or nutrient responders with very small responses to fat).

Fig. 3: Nutrients engage gut-to-vagal CCK-mediated signalling.
figure 3

a, Imaging of calcium responses in vagal sensory neurons4 while delivering fat (10% LA), sugar (500 mM Glu) or amino acid (250 mM amino acid mixture) (AA) stimuli to the intestines (Methods). Heat maps depict z-score-normalized fluorescence traces of sugar/nutrient responders (top) and fat-only responders (bottom) from 641 neurons of 8 mice, before application of CCKAR blocker (pre). The stimulus window (10 s for fat or sugar, 60 s for amino acids) is indicated by dashed white lines. b, To inhibit CCK signalling, we applied devazepide11 (4 mg kg−1, 200 μl), a CCKAR antagonist28 (post) (Methods). Top, note that blocking CCKAR receptor activation abolishes sugar-, fat- and amino acid-evoked activity in nearly all the nutrient responders (compare with a, top). Bottom, by contrast, the CCKAR blocker had no effect on the fat-evoked activity in the fat-only responders (compare with a, bottom). See Extended Data Fig. 6 for results using glutamate receptor blockers. c, Cartoon of the gut-to-brain sugar/nutrient-sensing vagal axis. Bottom right, an expanded view of CCK-expressing EECs in the intestines. Top right, two-dimensional t-distributed stochastic neighbour embedding (t-SNE) plot of the transcriptome of mouse vagal nodose neurons37. Clusters expressing Cckar are shown in red and clusters expressing Vip are shown in green (Methods). d, Calcium responses in vagal ganglia of mice expressing GCaMP6s in VIP neurons during infusion of fat, sugar or amino acids stimuli into the intestines. Heat maps show z-score-normalized fluorescence traces. Approximately 30% of VIP vagal neurons responded to nutrient stimuli (n = 60 out of 203 neurons from 9 ganglia), but only a small fraction (~4%) responded to fat. Stimuli: 10% LA, 500 mM Glu or 250 mM amino acid mixture.

Fat and sugar signalling in the gut

We next explored how sugar or nutrient signals are transmitted from the gut to vagal neurons. Cholecystokinin (CCK)-expressing enteroendocrine cells (EECs) in the intestine have been proposed to function as the sugar-preference gut sensing cells11,26. We hypothesized that CCK may be the signal between the gut and their partner vagal neurons. We thus examined responses of vagal neurons to intestinal application of sugar, fat and amino acids, before and after pharmacologically inhibiting CCK signalling with devazepide27, a CCK-A receptor28 (CCKAR) antagonist (Fig. 3a). Indeed, blocking CCK signalling abolished all the responses of the vagal sugar/fat neurons (that is, to intestinal stimulation with sugar, fat and amino acids). By contrast, the fat-only responses remained robust and reliable (Fig. 3a,b and Extended Data Fig. 6). Given these results, we anticipated that the application of CCK should strongly activate the nutrient responding vagal neurons, but not the fat-only neurons. Our results showed both predictions to be correct (Extended Data Fig. 6f). Finally, we also examined the potential role of glutamate signalling11 by imaging responses of vagal neurons to intestinal sugar stimuli before and after addition of a mixture of l-(+)-2-amino-3-phosphonopropionic acid (AP3) and kynurenic acid, two glutamate receptor antagonists29,30. Our results demonstrated that pharmacological inhibition of glutamate-based signalling has no effect on the gut-to-vagal sugar/nutrient-sensing circuit (Extended Data Fig. 6a–d). Together, these results substantiate CCK as the transmitter mediating sugar/nutrient-sensing in the gut–brain axis, and further distinguishes the CCK-dependent from the CCK-independent fat-sensing gut-to-brain pathways.

Nutrient responders in the nodose

Given that gut sugar, fat and amino acid responders rely on CCK signalling, we expected that vagal neurons receiving this gut-to-brain signal would be defined by the expression of CCK receptors (such as CCKAR) (Fig. 3c). CCK is principally known as a satiety hormone, whose role is to modulate food intake by suppressing appetite31,32. By contrast, the function of nutrient preference circuits is to promote nutrient consumption1,4. Thus, we explored how CCK can function both as a satiety hormone and as a nutrient preference signal in the gut. We reasoned that this conundrum could be easily resolved if a genetically distinct33 subset of CCKAR-expressing vagal neurons mediates nutrient preference.

We engineered Cckar-cre mice by targeting Cre recombinase to the Cckar gene34 (Methods), and used them to functionally validate the nutrient-evoked activation of CCKAR vagal neurons (Extended Data Fig. 7a,b). Next, we used single-cell RNA-sequencing (RNA-seq) data from the nodose ganglion35,36,37 to further characterize subsets of CCKAR-expressing neurons, and generated Cre driver lines expressing GCaMP6s in subsets of candidate clusters. Our results showed that a unique pool of CCKAR-expressing vagal neurons marked by expression of the vasoactive intestinal peptide (VIP) labelled the nutrient responders (with only a small fraction of the fat-only neurons) (Fig. 3c, d, Extended Data Fig. 7e–g). We then further refined this cluster by removing the small number of fat-only responding neurons (Extended Data Fig. 7c,d,g). These results validate the segregation of the nutrient versus the fat-only circuit, and substantiate CCK in the gut as the transmitter mediating sugar/nutrient signals.

An important prediction is that inhibiting signalling from the nutrient-sensing vagal neurons should prevent the activation of the gut–brain axis, and consequently block the development of nutrient preference. Our strategy was to genetically silence the nutrient-sensing vagal neurons by bilaterally injecting the nodose of Vip-cre38 mice with an AAV-Flex-TetTox4 construct (Fig. 4a and Extended Data Fig. 3d). As hypothesized, blocking activity from these neurons markedly impaired the development of nutrient preference (Fig. 4b,c). Importantly, the immediate, innate attraction to sugar and fat in these mice was not affected (Extended Data Fig. 8).

Fig. 4: VIP vagal neurons convey sugar/nutrient preference.
figure 4

a, Silencing VIP neurons in the vagal ganglia by bilateral injection of AAV-DIO-TetTox into the nodose in Vip-cre mice. b,c, Fat and sugar preference tests for control mice and mice with silenced VIP-expressing vagal neurons (Vip-Tx). b, Control mice develop strong preference for fat during a standard 48 h fat-versus-sweetener test (n = 7). By contrast, silencing of VIP vagal neurons abolishes the development of fat preference (n = 8, Vip-Tx mice). Two-sided Mann–Whitney U-test, control versus Vip-Tx fat consumption, P = 3 × 10−4. c, Silencing of VIP vagal neurons also abolishes the development of sugar preference. Control (n = 7) versus silenced mice (n = 8). Two-sided Mann–Whitney U-test, control versus Vip-Tx sugar consumption, P = 6 × 10−4. Data are mean ± s.e.m. d, Strategy for chemogenetic activation of VIP vagal neurons. An excitatory DREADD receptor (via AAV-DIO-hM3Dq) was targeted bilaterally to the nodose of Vip-cre mice. The mice were then tested for their basal preference to cherry or grape flavour (pre). The mice were conditioned and retested using the less-preferred flavour plus the DREADD agonist clozapine (post) (Methods). e, Left, control mice (not expressing DREADD) presented with clozapine (5 mg l−1) in the less-preferred flavour do not switch their preference and maintain their basal, original flavour choice (n = 8 mice; two-tailed paired t-test, P = 0.061). Right, after associating clozapine-mediated activation of VIP vagal neurons with the less-preferred flavour, all the mice expressing DREADD switched their preference (n = 6 mice; two-tailed paired t-test, P = 9.6 × 10−4). Preference index values are mean ± s.e.m.

Finally, we anticipated that artificial activation of this gut-to-brain nutrient preference circuit should afford the development of new preferences, in essence driving appetitive responses to previously unpreferred stimuli. To test this proposal, we bilaterally injected the nodose of Vip-cre mice with a Cre-dependent AAV virus encoding the excitatory designer receptor hM3Dq39, so that nutrient responding neurons could be experimentally activated by the DREADD agonist clozapine40. After allowing expression of DREADD (Extended Data Fig. 3e), mice were exposed to a preference assay using cherry- and grape-flavoured solutions (Fig. 4d), and to enhance attraction of these novel flavours, both solutions were spiked with an artificial sweetener (Methods). Next, we established a baseline preference for each mice (that is, grape vs cherry), introduced clozapine into the less-preferred flavour, and investigated whether clozapine-mediated activation of the nutrient-sensing neurons could create a new preference. Indeed, after 48 h of exposure to both solutions all of the mice markedly switched their preference to the clozapine containing flavour. By contrast, mice without the designer receptor did not develop a new preference, and if anything, were slightly averse to the DREADD activator (Fig. 4e). These results illustrate how non-natural activation of this gut–brain sugar/nutrient-sensing circuit can drive the development of a novel preference.

Fat-only responders in the nodose

We next investigated the identity of vagal neurons mediating the fat-only signals. Using the single-cell RNA-seq atlas from the nodose ganglion35,36,37, we searched for vagal neurons that did not express VIP (as the sugar-, fat- and amino acid-sensing marker), and identified five minimally overlapping candidate clusters (Fig. 5a): Trpa1, Gpr65, Piezo2, Calca and Oxtr. We engineered Trpa1-cre mice using the CRISPR–Cas9 system (Extended Data Fig. 9a and Methods), and obtained Cre driver lines for the other four candidates. Our results (Fig. 5b) demonstrated that the TRPA1-expressing vagal cluster responds selectively to intestinal delivery of fat, but not sugar or amino acid stimuli, thus defining the fat-only responders. Vagal neurons expressing GCaMP6s in Gpr65-cre, Piezo2-cre, Calca-cre or Oxtr-cre mice were unresponsive to intestinal delivery of sugar or fat stimuli (Extended Data Fig. 9b–e).

Fig. 5: TRPA1 vagal neurons mediate fat-specific preference.
figure 5

a, Single-cell RNA-seq atlas of nodose ganglia37, showing vagal clusters for VIP (blue), Trpa1 (red), Gpr65 (orange), Calca (green), Oxtr (brown) and Piezo2 (purple). b, The vagal cluster expressing TRPA1 (Trpa1-GCaMP6s) responded selectively to intestinal delivery of fat (10% LA), but not sugar (500 mM Glu) or amino acid (250 mM amino acids mixture) stimuli. The heat maps show z-score-normalized fluorescence traces. Of 163 imaged neurons from 5 ganglia, approximately 24% responded to fat. See Extended Data Fig. 9 for imaging results for the other vagal clusters. c, Left, strategy for silencing of TRPA1 neurons in the vagal ganglia by bilateral injection of AAV-DIO-TetTox into the nodose of Trpa1-cre mice. Fat and sugar preference tests on control mice (middle) and mice with silenced TRPA1-expressing vagal neurons (Trpa1-Tx) (right). Control mice develop strong preference for fat and sugar after 48 h (n = 7). By contrast, silencing of TRPA1 vagal neurons abolishes the development of fat but not sugar preference (n = 6, right). Two-sided Mann–Whitney U-test, control versus Trpa1-Tx for sugar, P = 0.23; control versus Trpa1-Tx for fat, P = 1.1 × 10−3. Data are mean ± s.e.m.

Next, we reasoned that genetic silencing of the fat-only circuit (that is, TRPA1-expressing vagal neurons) may abolish the development of fat preference but should have no effect on the development of sugar preference. Thus, we bilaterally injected the nodose of Trpa1-cre mice with an AAV-Flex-TetTox construct to silence the fat-only vagal neurons and tested the mice for sugar-versus-fat preference. Indeed, after genetic silencing, these mice no longer developed post-ingestive preference for fat stimuli, but retain their capacity to develop post-ingestive preference for sugar (Fig. 5c). Of note, their immediate attraction to fat was unaffected (Extended Data Fig. 8). Together, these results reveal the identity of the neurons mediating fat-only signals, and uncover their essential role in the gut-to-brain circuit mediating fat preference.

Sugar and fat sensors in the gut

Pharmacological experiments have previously demonstrated that the sodium–glucose-linked transporter 1 (SGLT1) functions as the gut receptor that recognizes glucose and transmits the post-ingestive41, gut-to-brain sugar signals4. Here, we extend the specificity of these findings by generating SGLT1-knockout mice and examining their responses to intestinal stimulation with sugar and fat (Fig. 6a). The data shown in Fig. 6 demonstrate that all vagal responses to intestinal delivery of sugar are abolished in these mice. By contrast, the responses to fat stimuli remain unaffected.

Fig. 6: Intestinal GPR40 and GPR120 fat receptors activate the gut–brain axis.
figure 6

a, We engineered knockout mice for three candidate fat receptors in the gut, and generated mice with every combination of these knockouts. We then recorded vagal responses to intestinal delivery of fat (10% LA) and sugar (500 mM Glu), and tested them for the development of fat and sugar preference. b, Heat maps depict z-score-normalized fluorescence traces from vagal neurons of SGLT1-knockout mice in response to intestinal delivery of fat (10% LA) and sugar (500 mM Glu). As previously shown, SGLT1 functions as the gut-to-brain sugar receptor4, and no vagal neurons responded to sugar in the knockout mice. However, responses to fat were unaffected (n = 174 out of 903 imaged neurons from 10 ganglia). Sglt1 is also known as Slc5a1. c, Heat maps illustrating the selective loss of fat responses in GPR40/GPR120 double-knockout (n = 51 out of 428 imaged neurons from 6 ganglia) and CD36/GPR40/GPR120 triple knockout (n = 44 out of 326 imaged neurons from 6 ganglia) mice. Note the normal responses to intestinal delivery of sugar in these knockout mice. See Extended Data Fig. 10 for imaging results for the other knockout lines. d, Bar graphs comparing vagal neurons responding to intestinal delivery of fat (10% LA) in control mice versus the various receptor knockouts (see Methods). Vagal responses were substantially affected only in the GPR40/GPR120 double-knockout (GPR40/GPR120, n = 7, P = 5 × 10−6) and in the triple knockout (TKO) (n = 6, P = 4 × 10−6) mice. Data are mean ± s.e.m.; statistics are shown in Methods. e, Knockout mice were tested for the development of fat preference. GPR40/GPR120 double knockouts (n = 7 mice, P = 0.81) and CD36/GPR40/GPR120 triple knockouts (n = 9 mice, P = 0.46) did not develop a preference for fat. White bars show initial preference and red bars show preference at the end of the 48 h test. All other combinations of knockouts developed a behavioural preference for fat, similar to control wild-type (WT) mice. Statistics are shown in Methods. Data are mean ± s.e.m. f, As expected, GPR40/GPR120 knockouts still develop preference for sugar. Wild type: n = 10 mice, P = 2.9 × 10−5; GPR40/GPR120: n = 9 mice, P = 8.0 × 10−5; TKO: n = 7 mice, P = 1.9 × 10−3. Data are mean ± s.e.m. Statistics are shown in Methods.

We expected that the development of fat preference would depend on specific fat receptors expressed on the surface of intestinal EECs42. Dietary fat, once ingested and digested, is thought to be sensed by a number of putative gut receptors, including the fatty acid translocase CD36 (refs. 43,44) and the G protein-coupled receptors GPR40 (ref. 45) and GPR120 (refs. 46,47). We anticipated that one or more of these receptors would be used to transmit fat preference46 via the gut–brain axis. Therefore, we used CRISPR–Cas9 to generate mice deficient in all combinations of CD36, GPR40 and GPR120 (single, double and triple mutants) (Extended Data Fig. 9f; see Methods for details).

A key prediction would be that the loss of the essential receptor(s) would abolish vagal responses to intestinal stimulation with fat, thus defining the intestinal sensors for the gut-to-brain fat signals.

Because of the intricacies of breeding such a wide range of knockout combinations, and the need to introduce the GCaMP6s reporter for functional imaging into the various genetic backgrounds, we chose to use a direct fusion of GCaMP6s to Snap25 regulatory sequences48 rather than crossing-in a Cre driver construct and a Cre-dependent GCaMP reporter. Our results showed that the Snap25-GCaMP6s construct is well expressed in vagal neurons, and compares favourably with our studies using other driver lines (Extended Data Fig. 9g,h).

After testing all the fat receptor-deletion combinations (Fig. 6c,d and Extended Data Fig. 10a–f), we found that GPR40 and GPR120 were the essential mediators of intestinal fat signals to the vagal neurons. As expected, vagal neurons responding to sugar were unaffected in all of the mutants (Extended Data Fig. 10a–g). Notably, all fat responses—from both the fat-only and from the sugar-, fat- and amino acid-sensing vagal neurons—were abolished in the GPR40/GPR120 double-knockout mice, demonstrating that the same fat receptors are used in both gut-to-brain signalling pathways (that is, CCK-independent and CCK-dependent, respectively).

An expectation from these imaging results is that the GPR40/GPR120 double-knockout mice (as well as the triple-knockout mice) should not develop preference for fat46, whereas the various single mutants and the other double mutants should be unaffected. We note, however, that these are global knockouts, rather than conditional knockouts. Notably, GPR40, GPR120 and CD36 single mutants, as well as GPR40/CD36 and GPR120/CD36 double mutants were indistinguishable from control wild-type mice (Fig. 6e, right). By contrast, the GPR40/GPR120 double-knockout (and the triple-knockout) mice were no longer capable of developing a behavioural preference for fat (Fig. 6e, left). Importantly, the innate responses to fat stimuli were unaffected in the GPR40/GPR120 double and triple mutants, with the mice exhibiting a strong immediate attraction to fat, illustrating the fundamental difference between the taste and the gut–brain pathways (Extended Data Fig. 10h). As in control mice, fat receptor-knockout mice develop the normal preference for sugar46 (Fig. 6f). Together, these results demonstrate the function of GPR40 and GPR120 as the essential receptors signalling the presence of intestinal fat via the gut–brain axis.


Sugar and fat are indispensable nutrients, and it would be expected that dedicated circuits drive their consumption1,4,13. We have shown that in addition to the taste system, these nutrients rely on a dedicated gut-to-brain system to detect and report the presence of intestinal sugar and fat to the brain.

Here we demonstrate the fundamental role of these nutrient-sensing circuits by showing that genetic or pharmacological blockade of sugar and fat gut-to-brain signals, at any of the four stages following ingestion, abolished the development of nutrient preference: (1) by preventing sugar or fat binding to their corresponding intestinal receptors, (2) by blocking the activated gut cells from signalling to the vagal neurons, (3) by silencing the sugar- or fat-activated vagal neurons and preventing the transfer of their signals to the brain, and (4) by preventing the cNST neurons that receive the gut–brain signals from broadcasting the presence of intestinal sugar or fat to the rest of the brain.

An unexpected finding from these studies was the discovery of a single gut-to-brain pathway, based on CCK signalling, that functions as a generalist detector informing the brain of the intestinal presence of any of the three essential nutrients: sugar, fat and amino acids. Although each nutrient uses its own dedicated receptors in the gut, the convergence of the signal into a unique class of vagal neurons (VIP–UTS2b) highlights the simple and elegant logic of this circuit: after the gut cells are activated, the circuit does not need to preserve the identity of the specific nutrient stimulus, and needs only to ensure that the emerging gut–brain signal triggers behavioural preference4. Given that CCK functions as the signalling molecule in the gut for the sugar and nutrient-sensing pathway, we anticipate that there is a unique subset of intestinal CCK-positive EECs that co-express the sugar (SGLT1) and fat (GPR40 and GPR120) preference receptors (the nature of the amino acid receptor is not yet known). Notably, examination of single-cell RNA atlases from both rodent and human gut tissue suggests that this is probably the case33,49. Future studies should help to define this subtype of CCK-expressing EEC that uses CCK as a transmitter (rather than as a gut neuromodulator or hormone) to activate the gut–brain axis and report the presence of intestinal sugar, fat and amino acid nutrients.

Our results also uncovered two separate gut–brain circuits for intestinal fat sensing (that is, the fat-only and the sugar, fat and amino acid vagal pathways), yet both utilize the same receptors—GPR40 and GPR120—to drive the development of fat preference. Notably, silencing either circuit is sufficient to abolish the preference for fat, demonstrating that both are indispensable for the development of fat preference. Thus, activating the fat intestinal receptors only in the CCK-dependent pathway, or only in the CCK-independent (fat-only) pathway, is not sufficient on its own to trigger fat preference. Indeed, we measured cNST signals activated solely by the fat-only pathway, and they exhibited about 50% of the signal detected when both fat preference pathways were active (Extended Data Fig. 6g–i).

Given the essential role of sugar and fat in a healthy diet, and the importance of these gut–brain pathways in sugar and fat consumption (and most probably in over-consumption), it will be of great interest to determine the brain targets for each, and compare and contrast their function.

Finally, the identification of these gut receptors and gut–brain communication lines could help provide novel strategies to moderate the insatiable appetite for fat and sugar. Additionally, they clarify the fundamental difference between ‘liking’ and ‘wanting’50. Liking sweet and liking fat (that is, the innate attraction to these appetitive stimuli) is the result of activation of the taste system. Wanting sugar and fat, by contrast, is the gut–brain axis.



All procedures were carried out in accordance with the US National Institutes of Health (NIH) guidelines for the care and use of laboratory animals, and were approved by the Institutional Animal Care and Use Committee at Columbia University. Adult mice older than 6 weeks of age and from both sexes were used in all experiments. C57BL/6J (JAX 000664), TRAP2 (JAX 030323), TRPM5 KO (JAX 013068), Ai96 (JAX 028866), Ai162 (JAX 031562), Vglut2-IRES-cre (JAX 028863), Gpr65-IRES-cre (JAX 029282), Vip-IRES-cre (JAX 010908); Uts2b-cre (JAX 035452); Piezo2-cre (JAX 027719); Oxtr-cre (JAX 031303); Calca-cre (JAX 033168); Snap25-2A-GCaMP6s (JAX 025111) and Penk-IRES2-cre (JAX 025112).

Generation of genetically modified mice

To engineer Trpa1-IRES-cre knock-in mice51, a single guide RNA (sgRNA) (targeting CACAGAACTAAAAGTCCGGG) was selected to introduce an IRES-cre construct immediately downstream of the endogenous Trpa1 stop codon. A single-stranded DNA donor containing gene-specific homology arms (150 bp each) and the IRES-cre fragment (Addgene #61574) was generated using the Guide-it Long ssDNA Production System (Takara Bio). Cas9 protein (100 ng μl−1), sgRNA (20 ng μl1) and ssDNA donor (10 ng μl−1) were co-injected into the pronuclei of fertilized zygotes from B6CBAF1/J parents. Founder pups were screened for the presence of the knock-in allele using PCR, and candidates were validated by Sanger sequencing.

SGLT1-knockout mice were generated by co-injecting Cas9 mRNA (100 ng µl−1) with sgRNA (50 ng µl−1) targeting CGCATTGCGAATGCGCTCGT, resulting in a frameshift after the 20th residue and early termination after the 27th residue (wild-type SGLT1 is a 665-amino-acid protein). Homozygous SGLT1-knockout mice were bred and maintained on fructose-based rodent diet with no sucrose or cornstarch (Research Diets #D08040105). The mutant allele was validated by DNA sequencing.

To generate knockout mice for fat receptors (CD36, GPR40 and GPR120), Cas9 protein (50 ng µl−1) was co-injected with a total of 6 sgRNAs (7 ng µl−1 each: CD36: AAATATAACTCAGGACCCCG and TAGGATATGGAACCAAACTG; GPR40: AGTGAGTCGCAGTTTAGCGT and GAAGTTAGGACTCATCACAG; GPR120: CGACGCTCAACACCAACCGG and ACGCGGAACAAGATGCAGAG). The founder mice were validated by DNA sequencing and used to generate various homozygous knockout mice (that is, single, double and triple knockouts). All mutations in the individual homozygous lines were validated by DNA sequencing.

To engineer transgenic mice expressing Cre recombinase from the Cckar gene (Cckar-cre mice), a cre cassette was introduced at the ATG start codon of the Cckar gene using a 151 kb bacterial artificial chromosome (BAC) (RP23-50P5) carrying the Cckar gene, as described previously52.

Fos stimulation and histology

Stimuli consisted of 20% Intralipid (sc215182, Santa Cruz Biotechnology), 10% linoleic acid, 10% oleic acid, 0.3% xanthan gum or 10% mineral oil. Stimuli were emulsified by dilution into milliQ water containing 0.1% xanthan gum and 0.05% Tween 80, and vortexed for a minimum of 10 min. Note that we used high concentration of Intralipid for Fos and TRAP2-labelling experiments to ensure enough Intralipid is consumed and digested during the 90 min stimulation window. By contrast, when performing 48 h behavioural tests examining the development of fat preference, a lower concentration of 1.5% Intralipid was used, particularly to ensure that the fat and the AceK (3 mM) are similarly attractive.

To motivate drinking behaviour during the 90 min Fos induction experiments, C57BL/6J mice were water-restricted for 23 h, given access to 1 ml of water for 1 h, and then water-restricted again for another 23 h. Previously, we showed that such water restriction prior to the 90 min drinking test did not affect the selectivity of cNST labelling4 (for example, no labelling in response to water or AceK; see also Extended Data Fig. 2). Mice had the full complement of food during water restriction (this is essential during Fos labelling experiments as food restriction would activate a wide range of additional circuits, including food-reward circuits upon presentation of sugar or fat stimuli). All Fos experiments consisted of 90 min of exposure to the stimuli; mice were housed individually and all the nesting material and food was removed from their cages. After 90 min, mice were perfused transcardially with PBS followed by 4% paraformaldehyde. Brains were dissected and fixed overnight in paraformaldehyde at 4 °C. The brains were sectioned coronally at 100 μm and labelled with anti-c-Fos (SYSY, no. 226004 guinea pig, 1: 5,000) diluted in 1× PBS with 5% normal donkey serum (EMD Millipore, Jackson ImmunoResearch) and 0.3% Triton X-100 for 48 h at 4 °C, and then Alexa Fluor 647-conjugated donkey anti-guinea pig (Jackson Immuno-Research) for 24 h at 4 °C. Images were acquired using an Olympus FluoView 1000 confocal microscope. Quantification of Fos labelling was carried out by recording the number of positive neurons in an equivalent 200 × 200 μm area of the cNST (bregma −7.5 mm) and area postrema.

For intragastric stimulation, the catheter was placed as previously described4,53. Mice were individually housed and allowed to recover for at least five days before stimulus delivery. A syringe pump micro-controller (Harvard Apparatus) was used to deliver 1.5 ml of the control PBS or 20% Intralipid solution4 at 0.050 ml min−1.

Two-bottle preference assays

No behavioural experiments, including the short-term assays for taste responses, or the 48 h tests examining the development of sugar or fat preference used water-restricted or food-deprived mice. Mice were given ad libitum access to food and water for several days prior to the behavioural tests; any food or water restriction would severely affect the mice’s behaviour in preference or taste responses.

Development of fat preference: mice were first tested for their initial preference between 1.5% Intralipid and 3 mM AceK (pre testing) by completing 100 drinking trials. Each trial was initiated by the first lick and lasted for 5 s; the drinking ports then re-opened after 30 s of inter-trial interval. Next, mice were exposed to 500 licks to both 1.5% Intralipid and 3 mM AceK; this was repeated twice. Mice were then tested for the development of fat preference over 36 h using the 5 s trials. The pre- and post-preference indexes were calculated by dividing the number of licks to fat by the total lick count during the first 2–4 h (100 trials) of baseline measurements (pre) and during the last 2–4 h (100 trials) of the behavioural session (post), respectively.

In order to perform the two-bottle preference assay using large numbers of mice (for example, Figs. 4b, 5c and 6e), the setup was modified by using an LCD-based lick counter. The ‘pre’ preference index was calculated as the number of licks to fat divided by the total lick count during the first 4 h; the ‘post’ preference index was calculated as the number of licks to fat divided by the total lick count during the last 4 h of the session. Mice had ad libitum access to food throughout. The mice with a pre index >0.75 were not used owing to their high initial preference for fat (less than 20% of total tested mice had to be eliminated due to this strong bias).

Fat, sugar and amino acid intestinal stimulation

Stimuli for nodose imaging experiments were as follows. Sugar: 500 mM glucose. Amino acids: a mix consisting of 50 mM methionine, 50 mM serine, 50 mM alanine, 50 mM glutamine and 50 mM cysteine dissolved in PBS. Fat: 10% linoleic acid, 10% linolenic acid, 10% hexanoic acid, 10% DHA, 10% oleic acid, diluted in PBS containing 0.1% xanthan gum and 0.05% Tween 80, and vortexed for a minimum of 10 min. Vehicle control: 0.1% xanthan gum and 0.05% Tween 80. For sugar and fat intestinal stimulation in imaging experiments, we used a 10 s window of stimulation; for amino acids, we used a 60 s stimulus, as we used lower concentrations of each in the mix of several amino acids (see above).

Note that if using Intralipid mix12,13 (a 20% soybean oil emulsion, Santa Cruz) for nodose imaging experiments (rather than consumption where it would be naturally digested and broken down into short, medium and long chain fatty acids), the material needed to be pre-digested with lipases (mimicking its natural course of action upon ingestion). Using undigested complex oils for intestinal stimulation in imaging experiments yielded inaccurate and unreliable responses (data not shown). Intralipid was incubated with 4 mg ml−1 lipase (sigma) in PBS plus 10 mM CaCl2 for a minimum of 5 h at 37 °C.

Stereotaxic surgery

Mice were anaesthetized with ketamine and xylazine (100 mg kg−1 and 10 mg kg−1, intraperitoneal), and placed into a stereotaxic frame with a closed-loop heating system to maintain body temperature. The coordinates (Paxinos stereotaxic coordinates) used to inject and place recording fibres in the cNST were: caudal 7.5 mm, lateral ±0.3 mm, ventral 3.7–4 mm, all relative to Bregma. The fibre photometry experiments used a 400 μm core, 0.48 NA optical fibre (Doric Lenses) implanted 50–100 μm over the left cNST. TRAP2 mice were stereotaxically injected bilaterally in the cNST with AAV9-Syn-DIO-mCherry (300 nl per mouse), AAV9 DIO eGFP-RPL10a (300 nl per mouse) or AAV9 CBA.FLEX-TetTox54 (300 nl per mouse).

Genetic access to fat preference neurons in the brain

The TRAP strategy was used in TRAP220,55 mice to gain genetic access to fat-activated neurons in the cNST. A minimum of 5 days after injection, the AAV-injected TRAP2 mice or TRAP2; Ai9 mice were water-restricted for 23 h, given access to 1 ml of water for 1 h, water-restricted again for another 23 h (with ad libitum food), and then presented with 20% Intralipid ad libitum in the absence of food and nesting material. After 1 h, mice were injected intraperitoneally with 12.5 mg kg−1 4-hydroxytamoxifen (Sigma H6278) and placed back in the same cage for an additional 3 h. Following 4 h of Intralipid exposure, mice were returned to regular home-cage conditions (group-caged, with nesting material, ad libitum food and water). Mice were used for experiments a minimum of 10 days after this TRAP protocol. C57BL/6J and TRAP2 mice expressing TetTox in the cNST were tested in the two-bottle Intralipid versus sweetener preference assay for 48 h, as described previously4. Note that mice were never food-deprived prior to TRAPping, so as to prevent unrelated labelling and confounds from the activation of feeding and food-reward responding neurons.

Fibre photometry

Vglut2-cre;Ai96 mice were placed in a stereotaxic frame and implanted with a 400 μm core, 0.48 NA optical fibre (Doric Lenses) 50–100 μm over the left cNST. Photometry experiments were conducted as described previously4,56. To quantify the effects of vagotomy, we calculated the ratio of stimulus-related peak amplitude of the normalized trace (within 120 s of stimulus onset) prior vagotomy versus after vagotomy.

The duodenal catheterization surgery was carried out as described previously4. Stimulus delivery was performed via a series of peristaltic pumps (BioChem Fluidics) operated via custom Matlab software and Arduino microcontroller. Stimuli and washes were delivered through separate lines that converged on a common perfusion manifold (Warner Instruments) connected to the duodenal catheter. Trials consisted of a 60-s baseline (PBS 200 μl min−1), a 30 s stimulus (200 μl min−1), and a 3-min washout period (150 s at 600 μl min−1, and 30 s at 150 μl min−1). Stimuli were each presented three times in an interleaved fashion. The vagotomy procedure was carried out after the first round of stimulus as described previously4,57.

Nodose ganglion injection experiments

Genetic vagal silencing experiments

Cre-expressing mice (Vip-cre and Trpa1-cre) were anaesthetized with ketamine and xylazine (100 mg kg−1 and 10 mg kg−1, intraperitoneal). The skin under the neck was shaved and betadine and alcohol were used to scrub the skin three times. A midline incision (~1.5 cm) was made and the trachea and surrounding muscles were gently retracted to expose the nodose ganglia. AAV9 CBA.FLEX-TetTox (600 nl per ganglion) containing Fast Green (Sigma, F7252-5G) was injected in both left and right ganglia using a 30° bevelled glass pipette (custom-bevelled Clunbury Scientific). At the end of surgery, the skin incision was closed using 5-0 absorbable sutures (CP medical, 421A). Mice were allowed to recover for a minimum of 26 days before 2-bottle preference tests for sugar and fat. We note that almost all of the Vip-cre mice survived the surgical procedure and bilateral injections, whereas only 50% of the Trpa1-cre mice survived.

The Trpa1-cre knock-in line was validated by in situ hybridization experiments (Extended Data Fig. 9a). Fixed frozen nodose ganglia were sectioned at 16 μm thickness and processed for mRNA detection using the RNAscope Fluorescent Multiplex Kit (Advanced Cell Diagnostics) following the manufacturer’s instructions. The following RNAscope probes were used: Trpa1 (catalogue no. 400211-C3) and Cre-O4 (catalogue no. 546951).

Chemogenetic activation experiments

For gain-of-preference experiments, Vip-cre mice were injected bilaterally with 600 nl per ganglion of an AAV carrying the Cre-dependent activator DREADD (AAV9-Syn-DIO-hM3Dq-mCherry)37,39 and were allowed to recover for a minimum of three weeks before behavioural tests. Control and Vip-cre mice were tested in a two-bottle grape versus cherry flavour-preference assay (grape: 0.39 g l−1 Kool-Aid Unsweetened Grape, cherry: 0.36 g l−1 Kool-Aid Unsweetened Cherry, both containing 1 mM AceK). Flavour-preference tests were carried out as previously described4.

Vagal calcium imaging

Each mouse was anaesthetized with ketamine (100 mg kg−1) and xylazine (10 mg kg−1). The mice were tracheotomized, and the nodose ganglion was exposed for imaging exactly as previously described4.

For CCKAR blocker experiments, devazepide (Sigma) was dissolved in DMSO and diluted to a final dose of 4 mg kg−1 in saline11. For glutamate receptor blocker experiments, a mixture of metabotropic glutamate receptor antagonist AP3 (2 mg kg−1) and ionotropic glutamate receptor antagonist kynurenic acid (300 μg kg−1) was used. CCKAR and glutamate receptor blockers were delivered both into the intestines and abdominal cavity11; after a 5 min incubation period, the imaging session was started. For CCK application, the intestines, still attached to the anaesthesized mouse, were partly placed on a 25 mm petri dish to allow delivery (60 s) and washout (>180 s) of the stimuli (1 μg ml−1 CCK peptide; Bachem 4033101).

Note that for nodose imaging experiments using sugar, glucose stimuli consisted of 10 s pulses since stimulating with high concentration (>250 mM) for long pulses (60 s or more) strongly activates nutrient-independent vagal responses4,22,58, severely masking sugar/nutrient-evoked responses.

Calcium imaging data collection and analysis

Imaging data was obtained using an Evolve 512 EMCCD camera (Photo-metrics). Data was acquired at 5 Hz. A single field of view was chosen for recording and analysis from each ganglion. Calcium imaging data collected at 5 Hz was downsampled by a factor of 2, and the images were stabilized using the NoRMCorre algorithm59. Motion-corrected movies were then manually segmented in ImageJ using the Cell Magic Wand plugin. Neuropil fluorescence was subtracted from each region of interest with the FISSA toolbox60, and neural activity was denoised using the OASIS deconvolution algorithm61.

Neuronal activity was analysed for significant stimulus-evoked responses as described previously4,62. Note that for the fat receptor-knockout imaging studies, the minimal peak amplitude for defining responders was set to 1% ΔF/F. To quantify responses in fat receptors knockouts (Fig. 6d and Extended Data Fig. 10g), the number of responding neurons over the total number of imaged neurons per ganglia was normalized to the number of responders in wild-type control mice (set to 100%).

For experiments using blockers, two repeat trials per stimuli were used to accommodate the expanded time scale of the session (that is, before and after), and a neuron was considered a responder if it responded in both trials. The two-trial average area under curve for each stimulus was used to quantify the before and after responses (Extended Data Fig. 6d,e).

Imaging data is presented as heat maps of z-score-normalized responses (see also ref. 4). Equivalent results are obtained when using absolute ΔF/F (data not shown)


No statistical methods were used to predetermine sample size, and investigators were not blinded to group allocation. No method of randomization was used to determine how mice were allocated to experimental groups. Statistical methods used include one-way ANOVA followed by Tukey’s honest significant difference post hoc test, two-tailed t-test, two-way ANOVA or the two-sided Mann–Whitney U-test, and are indicated for all figures. Analyses were performed in MATLAB and GraphPad Prism 8. Data are presented as mean ± s.e.m.

Figure 6d: ANOVA with Tukey’s test compared to Snap25-GCaMP6s control. CD36 KO (n = 6 mice) vs control, P = 0.99; GPR40 KO (n = 7 mice) vs control, P = 0.89; GPR120 KO (n = 6 mice) vs control, P = 0.53; CD36/GPR40 double KO (n = 6 mice) vs control, P = 0.96; CD36/GPR120 double KO, (n = 8 mice) vs control, P = 0.99; GPR40/GPR120 double KO (n = 7 mice) vs control, P = 5 × 10−6; CD36/GPR40/GPR120 triple KO (n = 6 mice) vs control, P = 4 × 10−6.

Figure 6e: Two-tailed paired t-tests evaluating pre versus post fat preference. Wild-type mice (n = 11 mice) pre vs post, P = 2 × 10−6; CD36 KO (n = 8 mice) pre vs post, P = 4.8 × 10−3; GPR40 KO (n = 12 mice) pre vs post, P = 1 × 10−4; GPR120 KO (n = 14 mice) pre vs post, P = 1.03 × 10−4; CD36/GPR40 KO (n = 5 mice) pre vs post, P = 2 × 10−2; CD36/GPR120 KO (n = 6 mice) pre vs post, P = 1.7 × 10−3; GPR40/GPR120 double KO (n = 7 mice) pre vs post, P = 0.81; CD36/GPR40/GPR120 triple KO (n = 9 mice) pre vs post, P = 0.46.

Figure 6f: Two-tailed paired t-tests evaluating pre versus post sugar preference. Wild-type mice (n = 10 mice) pre vs post, P= 2.9 × 10−5; Gpr40−/−Gpr120−/− (n= 9 mice), pre vs post, P= 8.0 × 10−5; Cd36−/−Gpr40−/−Gpr120/− (n= 7 mice), pre vs post, P= 1.9 × 10−3.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.