All mice were housed in temperature-controlled and humidity-controlled facilities, with a 12-h light-to-dark cycle and free access to water and standard chow, unless otherwise noted. The ambient temperature was kept at about 23-26 °C and humidity was about 50-60%. Adult mice aged at least 8 weeks were used for data collection. Both males and females were used for virus-tracing studies and plasma osmolarity or Na tests. Only male mice were used for free-moving behavioral studies. The mice used in this study were vGAT-Cre (Jackson, cat. no. 016962), CaMKIIa-Cre (Jackson, cat. no. 005359), wild-type C57BL/6J mice (Slac Laboratory animal). All mice used for experiments were heterozygotes, maintained on the C57BL/6J background. For whole-brain Fos labeling, the TRAP2 (ref. ) (Jackson, cat. no. 030323) mice were crossed with Ai47 reporter mice (with three tandemly linked GFP molecules, EmGFP-T2A-TagGFP2-P2A-hrGFP, a gift from Z. Qiu, Chinese Academy of Sciences, Shanghai, China) to visualize active neurons. Trap2 mice were given 20 mg kg of aqueous 4-hydroxytamoxifen (MedChemExpress. cat. no. HY-16950) by intraperitoneal injection at the timepoint indicated in Fig. 1b to induce the expression of Fos and CreERT2. All experiments were performed in accordance with protocols approved by the ethical committee of Zhejiang Chinese Medical University and in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

AAV2/2Retro-hSyn-EGFP-WPREs (viral titers, 1.31 × 10 particles ml), AAV2/2Retro-hEF1a-DIO-mCherry-WPRE-pA (viral titers, 3.38 × 10 particles ml), AAV2/9-hEF1a-DIO-oChIEF-tDtomato-WPREs (viral titers, 1.70 × 10 particles ml), AAV2/9-mCaMKIIa-hM3D(Gq)-EGFP-WPREs (viral titers, 1.32 × 10 particles ml), AAV2/8-mCaMKIIa-GCaMP6s-WPREs (viral titers, 2.02 × 10 particles ml), AAV2/9-mDlx-oChlEF-mCherry-ER2-WPRE-PA (viral titers, 2.39 × 10 particles ml), AAV2/2Retro Plus-hEF1a-DIO-EGFP-WPRE-pA (viral titers, 1.07 × 10 particles ml) and AAV2/2Retro Plus-hSyn-FLEX-GCaMP6s-WPRE-pA (viral titers, 1.83 × 10 particles ml) were purchased from Taitool Bioscience Co., Ltd; rAAV-EF1a-DIO-axon-GCaMP6s-WPREs (viral titers, 5.09 × 10 particles ml), AAV-hSyn-DIO-ChrimsonR- mCherry (viral titers, 2.57 × 10 particles ml), rAAV-EF1a-DIO-tettoxlc-P2A-mCherry-WPREs (viral titers, 5.38 × 10 particles ml) and rAAV- EF1a-DIO-mCherry-WPREs (viral titers, 5.87 × 10 particles ml) were purchased from BrainVTA Co., Ltd; pAAV-CAG-FLEX-ArchT-GFP (viral titers, 1.75 × 10 particles ml) and pAAV-EF1a-DIO-GCaMP6m-WPREs (viral titers, 2.70 × 10 particles ml) were purchased from OBio Technology Co., Ltd; rAAV-CAG-DIO-mWGA-mCherry (viral titers, 5.10 × 10 particles ml), rAAV-hSyn-mWGA-Cre (viral titers, 3.22 × 10 particles ml), rAAV1-hSyn-Flpo (viral titers, 1.00 × 10 particles ml), rAAV-hSyn-Con Fon-taCasp3-T2A-TEVp-P2A-mCherry (viral titers, 2.00 × 10 particles ml), rAAV-hSyn-Con Fon-mCherry (viral titers, 5.00 × 10 particles ml), rAAV-hSyn-hM4D(Gi)-mCherry (viral titers, 2.08 × 10 particles ml), rAAV-EF1a-FDIO-taCasp3-T2A-TEVp-P2A-EGFP (viral titers, 5.01 × 10 particles ml), rAAV-EF1a-FDIO-taCasp3-EGFP (viral titers, 5.03 × 10 particles ml) and rAAV-mDlx-DIO-Axon-jGCaMP7f (viral titers, 3.00 × 10 particles ml) were purchased from Brain Case Co., Ltd. For rabies tracing experiments, rAAV-EF1a-DIO-mCherry-F2A-TVA-WPREs (viral titers, 5.33 × 10 particles ml), rAAV-EF1α-DIO-oRVG-WPREs (viral titers, 5.27 × 10 particles ml) and RV-ENVA-ΔG-EGFP (viral titers, 2.0 × 10 particles ml) were purchased from BrainVTA Co., Ltd; rAAV-EF1α-DIO-N2cG (viral titers, 5.16 × 10 particles ml), rAAV-EF1α-DIO-EGFP-T2A-TVA (viral titers, 2.50 × 10 particles ml) and CVS-EnvA-ΔG-tdTomato (viral titers, 2.00 × 10 particles ml) were purchased from Brain Case Co., Ltd. All viral vectors were aliquoted and stored at -80 °C until use.

Animals were anaesthetized with sodium pentobarbital, injected intraperitoneally (i.p.) at a dose of 50 mg kg. Each mouse was then placed in a stereotaxic apparatus (RWD Life Science) on a heating pad at 37 °C. After shaving the hair and disinfecting the scalp with iodine solution, an incision was made to expose the skull surface. A dental drill (RWD Life Science) was used to perform the craniotomy. Viral infusions were administered using a 1-µl syringe (Gaoge Industrial and Trading Co., Ltd) driven by a micropump (Micro 4, World Precision Instruments) at a rate of 40 nl min. The total infusion volume was 80 nl for SFO injection, 70-200 nl for MS injection (70 nl for experiments in Fig. 3, 200 nl for other experiments with strict post-hoc exclusion of animals showing notable MnPO labeling) and 100 nl for PBN injection. After viral infusion, the needle was maintained at the injection site for at least 5 min to ensure proper viral distribution and minimize backflow along the injection tract. The viruses were injected at the following coordinates: MS (anteroposterior (AP), +1.0 mm; mediolateral (ML), 0.0 mm; dorsoventral (DV), -4.0 mm); SFO (AP, -0.45 mm; ML, 0.0 mm; DV, -2.80 mm); PBN (AP, -5.2 mm; ML, ±1.25 mm; DV, -2.8 mm). Optic fibers (Inper Ltd) were implanted into the SFO (AP, -0.45 mm; ML, 0.0 mm; DV, -2.70 mm). Animals were recovered from anesthesia on a heating pad and then returned to their home cages for 3-4 weeks for viral expression and recovery from surgery before behavioral testing.

For retrograde monosynaptic tracing, a cocktail of rAAV-EF1a-DIO-mCherry-F2A-TVA-WPREs and rAAV-EF1α-DIO-oRVG-WPREs mixture (1:2) was injected into the SFO of vGAT-Cre or CaMKIIa-Cre mice. After 2 weeks, RV-ENVA-ΔG-EGFP was injected into the same site. The animals were sacrificed 7 d later to identify and examine the presynaptic neurons in the MS by histology. For additional trans-synaptic spread from the monosynaptic inputs to the SFO neurons, a cocktail of rAAV-EF1α-DIO-N2cG and rAAV-EF1α-DIO-EGFP-T2A-TVA was injected into the MS of vGAT-Cre mice with CVS-EnvA-ΔG-tdTomato injected into the SFO at the same time. The animals were sacrificed after at least 2 weeks for histology.

The procedures for oral and gastrointestinal catheter implantation were performed as previously described. Specifically, silicone oral catheters (4 cm in length, 0.5 mm inner diameter (i.d.) and 1.5 mm outer diameter (o.d.)) and gastrointestinal catheters (6 cm in length, 0.5 mm i.d. and 1.5 mm o.d.) were implanted. For oral catheter implantation, a 5-mm incision was made in the cheek skin and a 10-mm midline dorsal head incision was made from the eyes to the ears. A subcutaneous tunnel was created between the incisions and the sterilized catheter was inserted into the oral cavity and secured with 6-0 nylon sutures. The cheek muscle and skin were sutured and the catheter was anchored to the skull with a light-cured composite.

For gastrointestinal catheter implantation, a 10- to 15-mm incision was made under the costal margin and a 10-mm dorsal head incision. A subcutaneous tunnel was established to connect the incisions and the sterilized catheter was guided through the tunnel. For gastric implantation, the stomach was externalized, a small puncture was made to insert the catheter, which was secured with 9-0 nylon sutures, and then the stomach returned to the abdominal cavity. All animals were allowed to recover for 1 week after the implantation surgery before performing in vivo calcium imaging.

For two-photon imaging experiments, intraoral infusions were delivered via catheter at three different flow rates (5, 10 and 20 µl s) for 3.8 s. These parameters were carefully selected to prevent potential aspiration, because anesthetized animals may have impaired swallowing reflexes that could lead to nasal regurgitation. Intragastric infusions were administered at a constant rate of 100 µl min for 10 min. After each experiment, catheters were flushed with saline and filled with heparinized glycerol.

For intragastric infusion experiments in Extended Data Fig. 5, mice were fasted overnight before the procedure. Either 0.9% saline or ddHO (1-ml total volume) was delivered over 3 s using a gavage needle. For the sham group, the gavage needle was gently inserted into the esophagus and maintained in position for 3 s without fluid delivery.

For the drinking or feeding behavioral test, animals were tested in a behavioral cage and habituated to the behavioral cage for 3 d before the first experiment. All water-intake and food-intake experiments were performed between 8am and 3pm to avoid the circadian rhythm influence on drinking or feeding behavior. Fluid consumption was monitored with an electrical lickometer (QAXK-WLD, Thinkertech) and recorded using InperStudio software (Alpha 8.2, Inper) or Multichannel Fiber Photometry software (v.2.0.0.33169, RWD Life Science) during fiber photometry or optogenetic experiments.

The water-restrained experiments were conducted as previously described; mice were water deprived for 48 h and provided with 1 ml of water on the second habituation day before water-restrained experiments. Before food-restrained experiments, mice were fasted for 24 h and the total food intake was measured in a 30-min session experiment. For 24-h recording of food intake, water intake and locomotion experiments, mice were individually housed in a behavioral phenotyping system (TSE PhenoMaster Systems). All measurements were collected and processed automatically using PhenoMaster software from TSE Systems.

After the completion of behavioral assays, animals were perfused with 4% paraformaldehyde (PFA). The target sites were then examined to verify viral expression and confirm the positions of the optical fibers. Animals that failed to exhibit detectable viral expression or had misplaced optical fibers in the target sites were excluded from further analysis.

The fluorescence signal was recorded using an Inper Multi-Channel Fiber Photometry Device (Inper, cat. no. FPS-410/470). To calculate the resultant ΔF/F (where F is fluorescence and F baseline fluorescence), the following formula was used: ΔF/F = (F - F)/F. Signals were motion corrected using the InperDataProcess software (v.0.7.2, Inper) to minimize motion-related artifacts. The 410-nm signal was scaled using least-squares regression and then subtracted from the 470-nm signal to generate the fitted 470-nm signal.

For long-term tests, the F was determined as the mean fitted 470-nm signal within a 2-s time window before water access. The ΔF/F was calculated for a 10-min window before water access (pre-session) and a 10-min window after water access (post-session). For brief access tests, the F was determined as the mean fitted 470-nm signal within a 2-s time window before the lick onset of a drinking bout. The AUC (ΣΔF) was calculated for a window spanning 2 s before and 5 s after the onset of each drinking bout. A lick bout was defined as any set of licks that lasted >5 s and is separated from the previous lick bout by at least 5 s. Experiments that combine photometry with optogenetics were performed as previously described. Briefly, 589-nm light was delivered to activate the ChrimsonR-effected neurons (10 ms, 20 Hz, 5 mW). Photometry recordings were removed if there was a technical issue with the lickometer (for example, did not track licks).

GLM models were established using GLM2 (v.1.2.1) in R (v.4.3.3). The model incorporated three continuous predictors (cumulative licks in the preceding 5 s, instantaneous lick rate and bout size (total licks per drinking bout)) and one categorical predictor (osmolality: water, dry lick, saline and 2% NaCl). The dependent variable was the percentage change in neuronal activity (ΔF/F, %), measured during 7-s drinking bouts relative to a 2-s pre-lick baseline period. Initial modeling with all four osmolality conditions revealed significant differences only in water (Supplementary Table 1), justifying reduction to a binary predictor (water versus non-water) in subsequent analyses. Model performance metrics included Akaike information criterion and m.s.e. The significance of model coefficients from the Gaussian GLM was assessed using Student's t-tests. For each predictor variable, the test statistic (t-ratio) was calculated by dividing the estimated coefficient by its s.e.m. and the corresponding two-tailed P values were derived from the Student's t-test reference distribution. P < 0.05 was considered statistically significant. Effect sizes and partial R² for each covariate were quantified using rsq (v.2.7) and lm.beta (v.1.7.2) packages, respectively.

For in vivo two-photon experiments, 80 nl of virus of AAV2/2Retro Plus-hSyn-FLEX-GCaMP6s-WPRE-pA was injected into the SFO of vGAT-Cre mice. After viral injection, a gradient index lens with 0.5-mm diameter and 6-mm length (Go!Foton) was slowly implanted above the MS (AP, +1.10 mm; ML, 0.0 mm; DV, -3.75 mm).

Two-photon calcium imaging was performed using a miniature two-photon microscope FHIRM-TPM V4 (Transcend Vivoscope Biotech Co., Ltd) equipped with an FHIRM-U headpiece (field of view, 491.88 × 419.74 μm; resolution, ~900 nm; frame rate, 4.56 Hz; working distance, ~1,000 μm). A 920-nm light (ALCOR 920-1, Spark lasers) was used for excitation in all experiments. Images were processed using the open-source Suite2p package (https://github.com/MouseLand/suite2p), implementing nonrigid motion correction for optimal image registration. Neurons detected by the pipeline were subsequently curated manually in the Suite2p graphical user interface. To account for background contamination, fluorescence traces were neuropil corrected using the established equation: F = F - 0.7 × F. For experiments in Extended Data Fig. 6a-g, we also utilized deconvolved calcium events with superior temporal resolution, which more accurately capture neuronal firing patterns and facilitate the analysis of rapid sequential neural activities.

For oChIEF-mediated drinking suppression, animals were subjected to water restriction for 48 h before behavioral experiments. Laser pulses (473 nm, 10 ms) at 20 Hz were delivered through an optic fiber (Inper) using a laser pulse generator (Inper Remote). The laser output was maintained at 10 mW as measured at the tip of the fiber. Each trial consisted of an OFF-ON-OFF cycle lasting 30 s, with animals receiving photostimulation for up to 10 s during the ON phase (10- to 20-s window). Photostimulation was triggered manually in each trial. Animals underwent testing for a minimum of three trials per condition and the number of licks was averaged across trials. A trial was considered positive if the animal exhibited continuous drinking for >5 s before the stimulation. The total number of licks was measured before (0- to 10-s window), during (10- to 20-s window) and after (20- to 30-s window) the stimulation session.

For ArchT-mediated stimulation of drinking, sated animals were used in behavioral experiments. In each 30 min of the ON-OFF-ON trial, constant 589-nm light (10 mW at fiber tip) was started 10 min after water presentation and maintained until the end of the trial. The number of licks in a 5-s window after the first lick was analyzed. Animals were tested for three to ten trials each and the number of licks was averaged across trials.

For real-time place preference tests, mice were placed in a two-chambered Plexiglas box (50 × 50 × 50 cm each) and habituated to optic patch cables for 5 min over 3 d before the experiments start. Mice were randomly placed in either chamber, with active entry triggering 473-nm (20 Hz, 10 ms, 10 mW) or constant 589-nm light until the mouse crossed back. Sessions lasted for 20 min and the trace and time spent in each compartment were recorded using video tracking software (ANY maze Video Tracking System v.4.98, Stoelting).

For open field tests, mice were placed in a 45 × 45 × 45 cm open field arena under low illumination conditions. The center zone was defined as the area 11.25 cm from the edge of the arena. A video camera mounted above the arena recorded the movement of the mice throughout the 10-min sessions. The collected data were analyzed using the ANY maze Video Tracking System (v.4.98) to quantify various parameters of exploratory behavior and anxiety-related measures.

For the chemogenetics gain-of-function experiments, a fresh stock solution of CNO (Abcam, cat. no. ab141704) was prepared by dissolving it in dimethyl sulfoxide and then diluting it in saline. To activate the Gq cohort, mice were injected i.p. with 3 mg kg of CNO 20 min before the experiments. The mCherry control mice received the same dose of CNO as the experimental cohort.

To measure plasma osmolality and plasma Na concentration, both mCherry and Casp3 groups underwent water restriction for 24 h before being provided access to water at time t = 0. About 100 μl of blood samples was collected at a single timepoint per session: sated (baseline), dehydrated (-1 min) or after rehydration (10 min, 30 min, 60 min and 90 min). Blood was collected from the facial vein using heparin lithium-coated capillary tubes (JAENAVRL). Plasma was isolated by centrifugation at 4 °C and 1,000g for 10 min. Plasma osmolality was measured using a freezing point osmometer (Fiske Model 210) and plasma Na concentration was determined using a sodium assay kit (Elabscience, cat. no. E-BC-K207-S). Mice were allowed 1 week of recovery between sessions.

Mice were anesthetized by injection i.p. of sodium pentobarbital, perfused with saline, followed by ice-cold PFA. Brains were then postfixed for 24 h in 4% PFA and equilibrated in 30% sucrose at 4 °C for at least 24 h. Virus expression and fiber positions were confirmed post-hoc for all mice.

For immunohistochemistry staining, brains were sliced on a freezing microtome (Thermo Fisher Scientific, cat. no. NX70) into 30-µm coronal sections. Brain slices were blocked for 1-2 h with 0.3% Triton X-100, 5% normal donkey serum in phosphate-buffered saline (PBS) solution at room temperature and incubated in primary antibodies overnight at 4 °C. Next, brain slices were washed 3× with PBS for 15 min each to wash away unbound primary antibody and then incubated with secondary antibodies for 1-2 h at room temperature. Finally, brain slices were washed again with PBS 3× for 15 min each to wash away unbound secondary antibody before imaging. Anti-CCK (Abcam, cat. no. ab37274, 1:500), anti-CaMKII (Abcam, cat. no. ab92332, 1:500), anti-AT1R (Abcam, cat. no. ab124505, 1:50), anti-Sox6 (Abcam, cat. no. ab314209, 1:50), donkey anti-mouse immunoglobulin (Ig)G H&L (Alexa Fluor-488) (Abcam, cat. no. ab150105, 1:500) and donkey anti-rabbit IgG H&L (Alexa Fluor-647) (Abcam, cat. no. ab150075, 1:500) antibodies were used in this study. Brain slices were imaged using a laser confocal microscope (Olympus, cat. no. FV3000).

For combined FISH, brains were sliced on a freezing microtome (Thermo Fisher Scientific, cat. no. NX70) into 14-µm coronal sections. The mouse Gad1 (cat. no. 400951), Slc32a1 (cat. no. 319191-C2), Fos (cat. no. 316921), Slc17a6 (cat. no. 319171-C3) and GLP1r (cat. no. 418851-C3) probes, purchased from Advanced Cell Diagnostics, were used in this study. Brain slices were imaged using a laser confocal microscope.

The in vivo two-photon imaging data were processed for clustering and visualization using a customized Python 3 code. First, the calcium signal data for each neuron were preprocessed, including filtering based on frame ranges specified by the user. All data were truncated to ensure uniform length across neurons and normalized to facilitate clustering. For unsupervised clustering, K-means clustering was performed on the processed dataset. Using the elbow method and silhouette coefficient, K = 3 was determined as the optimal number of clusters. For visualization, principal component analysis was applied to reduce the dimensionality of the dataset to two components. The clustering results were visualized as scatter plots in principal component analysis space, with confidence ellipses (95% confidence interval) drawn for each cluster. The average activity patterns of each cluster were visualized using a PSTH with s.e.m. shading.

To identify gene markers of SFO-projecting MS neurons, we utilized LCM combined with Smart-seq2 (Spatial-seq) technology. Specifically, fluorescence-enhanced AAV2/2Retro Plus-hEF1a-DIO-EGFP-WPRE-pA was injected into the SFO of vGAT-Cre mice to selectively label MnPO and MS neurons projecting to the SFO. Target neurons were identified based on their fluorescent signals and manually dissected using a laser microdissection system (equipped with a DM68 microscope, a Leica LMD6 laser cutter and a single-cell capture collector LMT350, Leica Microsystems). This LCM system uses ultraviolet light for precise, contact-free and contamination-free isolation of target neurons from tissue sections. For each animal, SFO-projecting MS neurons and MnPO neurons were collected separately, with cells from each region pooled together. Library preparation and sequencing were performed by Annoroad Gene (Beijing, China) following the Smart-seq protocol.

Two-way analysis of variance (ANOVA) was used to assess how behavior was affected by two independent factors. Unpaired Student's t-testing was used for comparisons between two groups. Two-sided tests were used throughout, with α = 0.05. To facilitate the log(transformation) of data containing zero values in Fig. 5e, a small constant (0.1) was added to each bout number before applying the logarithm. This approach ensures that the logarithm is defined for all data points while minimally affecting the overall data distribution. To quantify neuronal responses in Extended Data Fig. 6c,g, the response index (Ig) was calculated as: lg = (Average normalized activity rate + 0.01)/(Average normalized activity rate + 0.01). Neurons with a response index between -0.1 and 0.1 were classified as unresponsive, those with a response index >0.1 as activated and those with a response index <-0.1 as inhibited. This classification was used to group neurons into different response categories for further analysis. Volcano plots showing differential gene expression in Extended Data Fig. 3f were processed and analyzed using the limma package in R (v.4.4.1), incorporating the voom transformation to account for the mean-variance relationship in count data. The voom function was applied to the normalized counts to transform the data into log(counts per million) while estimating precision weights based on the observed mean-variance trend. Genes with a P < 0.05 and an absolute log(fold-change) were considered statistically significant. Statistical analyses were conducted using GraphPad Prism 10 Software (GraphPad Software) and Microsoft Excel 2019. Recording experiments used nonrandomized trials with within-participant controls when possible. For behavioral experiments, mice were pseudorandomly assigned to groups before surgery or experimental manipulations to minimize confounds and ensure balanced allocation. For the RTPP test, mice were randomly placed in either chamber with active entry triggering light stimulation. For other experiments, standardized randomization was not performed. All experiments in this study were repeated at least 3×. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications. Individual data points have been shown where feasible and nonparametric tests were used where data were not expected to be normally distributed. Otherwise, data were assumed to be normal, although this was not formally tested. Multiple comparison adjustments and significance definitions are included in Supplementary Table 1. Behavioral and quantification experiments were done in a blinded manner.

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