Satiation is the process by which eating and drinking reduce appetite. For thirst,
oropharyngeal cues play a critical role in driving satiation by reporting to the brain
the volume of fluid consumed
1–12
. In contrast, mechanisms that relay the osmolarity of ingested fluids remain poorly
understood. Here we show that the water and salt content of the gastrointestinal (GI)
tract are precisely measured and then rapidly communicated to the brain to control
drinking behavior in mice. We demonstrate that this osmosensory signal is necessary
and sufficient for satiation during normal drinking, involves the vagus nerve, and
is transmitted to key forebrain neurons that control thirst and vasopressin secretion.
Using microendoscopic imaging, we show that individual neurons compute homeostatic
need by integrating this GI osmosensory information with oropharyngeal and bloodborne
signals. These findings reveal how the brain’s fluid homeostasis system monitors the
osmolarity of ingested fluids in order to dynamically control drinking behavior.
Drinking influences the volume and composition of the blood
1–4
. Yet ingestion of water and salt have opposing consequences for fluid balance, which
raises the question of how the brain monitors the osmolarity of ingested fluids.
One way the brain controls drinking is by tracking the passage of fluids through the
mouth and throat. Classic experiments demonstrated that drinking temporarily satiates
thirst even if the ingested water is immediately drained from the esophagus
5–8
, and recent work has identified specific populations of forebrain neurons that receive
this rapid oropharyngeal signal during drinking
9–12
. This mechanism allows the brain to track fluid intake in real-time and thereby quench
thirst and inhibit vasopressin secretion in anticipation of water absorption into
the bloodstream, which occurs gradually over tens of minutes.
Nevertheless, this oropharyngeal signal communicates to the brain only the volume
of fluid ingested, not its composition
9,12
, which suggests that a distinct mechanism tracks fluid osmolarity during drinking.
Taste aversion prevents the consumption of highly concentrated salt solutions
13,14
, but there is little evidence that taste fine-tunes fluid consumption to match physiologic
need. Infusion of fluids into the GI tract and hepatic portal circulation has been
reported to influence drinking behavior and vasopressin secretion in some studies
and species but not others
5–8,15–17
, and it remains unclear where pre-absorptive fluids are monitored in the periphery,
what exactly is sensed, which neurons in the brain receive this information, and how
they use it to regulate behavior
1–4
. A fundamental source of ambiguity in these experiments is that traditional behavioral
and physiologic readouts vary on the same timescale as fluid absorption (minutes),
making it difficult to disentangle pre-absorptive signals that are sensed remotely
in the periphery from systemic signals that are sensed directly in the brain.
To gain new insight into these longstanding questions, we set out to monitor directly
the dynamics of thirst-promoting neurons in the brain while simultaneously manipulating
the fluids ingested or infused into peripheral tissues. We first measured how fluid
osmolarity influences drinking behavior and thirst neuron activity. Mice were equipped
for fiber photometry recordings
18
of glutamatergic neurons in the subfornical organ
9,19,20
that promote drinking and directly monitor blood osmolarity (SFONos1 neurons; Fig.
1a). These animals were dehydrated and then given access to either water or hypertonic
saline. As shown previously
9
, ingestion of either fluid resulted in rapid inhibition of SFONos1 neurons that was
time-locked to the act of drinking (Fig. 1b). SFONos1 neurons remained inhibited following
bouts of water consumption, but their activity rebounded to an elevated, pre-ingestion
level after each bout of saline consumption terminated (Fig. 1c). Based on the kinetics
of this neural activity rebound, we hypothesized that SFONos1 neurons may receive
a signal from the GI tract that depends on the osmolarity of the ingested fluid. Consistent
with this, we found that isolated changes in GI osmolarity were sufficient to influence
salt preference and drinking behavior (Extended Data Fig. 1).
To investigate how information about the water and salt content of the GI tract is
communicated to the brain, we prepared mice with intragastric (i.g.) catheters for
fluid infusion into the stomach
21
as well as fiber photometry implants for recording of SFONos1 neuron dynamics (Fig.
1d). Strikingly, i.g. infusion of water rapidly inhibited SFONos1 neurons (latency
105 ± 13 s, mean ± s.e.m.), whereas i.g. infusion of hypertonic saline activated the
same population (Fig. 1e). Infusion of a range of NaCl concentrations revealed a remarkable
linear correlation between the osmolarity of the infused fluid and the modulation
of SFONos1 neuron activity (Fig. 1f). This response was independent of the initial
hydration state of the animal (Fig. 1g) or the identity of the infused osmolyte (Extended
Data Fig. 2), and was complete before any detectable change in blood osmolarity occurred
(Extended Data Fig. 2), indicating that it reflects local sensing within the GI tract
(likely in the proximal small intestine rather than the stomach itself) or hepatic
portal circulation
3,15–17
. Consistent with this, infusion of hypertonic solutions strongly activated SFONos1
neurons regardless of whether the infused osmolyte could be absorbed into the bloodstream
(Extended Data Fig. 2). Together, these data reveal that osmolarity is precisely measured
within the GI tract and then communicated to thirst neurons in the brain.
We next investigated the significance of this gut-brain communication for thirst satiation.
In dehydrated mice, i.g. infusion of water rapidly inhibited SFONos1 neurons (latency
107 ± 15 s, mean ± s.e.m.) and abolished drinking when water was subsequently presented
(Fig. 2a,b and Extended Data Fig. 3). This indicates that changes in GI osmolarity
(but not distension) are sufficient to satiate thirst and, conversely, that oropharyngeal
cues are not required. The behavioral response to GI osmolarity changes (but not systemic
changes) was hydration state-dependent: that is, i.g. infusion of salt did not stimulate
drinking by animals that were hydrated but did cause an increase in water consumption
by dehydrated animals (Fig. 2b,c and Extended Data Fig. 3). This state-dependent effect
is reminiscent of how inhibition of satiation signals for feeding, such as cholecystokinin,
can increase meal size in hungry animals but typically does not trigger the initiation
of feeding in animals that are well-fed
22
.
One mechanism by which GI osmolarity could regulate satiation is by acting as a “stabilization
signal” that determines whether SFONos1 neurons remain inhibited when water drinking
terminates (Fig. 1b,c). If this were true, then delivery of salt to the GI tract should
cause SFONos1 neuron activity to rebound even after mice drink pure water. To test
this, we gave dehydrated mice an i.g. infusion of salt and then analyzed the dynamics
of SFONos1 neurons during early bouts of water drinking (Fig. 2d–f). Indeed, we found
that the activity of SFONos1 neurons rebounded when drinking stopped, such that the
neural dynamics during water ingestion now resembled those observed during hypertonic
saline ingestion (Fig. 1b,c). This response was specific to salt infused into the
stomach, as an identical dose of salt delivered by systemic injection produced no
equivalent effect. This suggests that GI osmosensation specifies whether the rapid
inhibition of thirst neurons by oropharyngeal cues during drinking is transient or
durable.
We next investigated whether the inhibition of SFO neurons by the GI osmosensory signal
is required for thirst satiation. We surgically equipped mice for i.g. infusion combined
with simultaneous optogenetic activation
23
of SFONos1 neurons (Fig. 2g). As expected, brief (3 min) i.g. infusion of water into
dehydrated mice eliminated drinking when water was subsequently presented. However,
optogenetic re-activation of SFONos1 neurons during and after water infusion completely
blocked this satiating effect (Fig. 2h,i). This indicates that the SFO is a necessary
site in the brain where the osmosensory signal from the GI tract is received to control
the termination of drinking.
The peripheral mechanisms that communicate GI osmolarity to the brain are unknown.
To begin to address this question, we investigated the role of the vagus nerve (cranial
nerve X), which densely innervates the stomach and intestines
24
. Subdiaphragmatic vagotomy significantly attenuated the satiating effect of i.g.
water infusion and, furthermore, greatly reduced the ability of i.g. water or salt
to modulate SFONos1 neuron activity (Extended Data Fig. 4a–d). Genetic ablation of
a subset of vagal sensory neurons had similar effects (Extended Data Fig. 4e–i). Together,
these experiments indicate that vagal afferents are an essential part of the pathway
by which GI osmolarity is communicated to the brain, although they do not rule out
an additional role for circulating signals released from the gut following fluid ingestion.
We next investigated how information about the water and salt content of the GI tract
is represented within the broader neural circuit that controls fluid homeostasis (Fig.
4a). We first recorded by fiber photometry the dynamics of vasopressin-secreting neurons
in the supraoptic nucleus (SONAvp neurons; Fig. 3a). We confirmed that, like SFONos1
neurons
9
, these cells are activated by increases in blood osmolarity (Fig. 3b) and rapidly
inhibited during drinking
7,10,25
(Fig. 3c). By combining i.g. infusions with neural recordings, we showed that SONAvp
neurons are also bidirectionally regulated by GI osmolarity with kinetics similar
to SFONos1 neurons (Fig. 3d and Extended Data Fig. 5). This GI signal may function
in coordination with oropharyngeal cues
10
to pre-emptively regulate vasopressin secretion during eating and drinking.
The median preoptic nucleus (MnPO) is a crucial node in the brain’s fluid homeostasis
system
1,2
that is bidirectionally connected to the SFO through glutamatergic neurons that promote
drinking
11,12,26,27
(defined by expression of Adcyap1, Agtr1a, and Nxph4) and GABAergic neurons that inhibit
drinking
12,26
(defined by expression of Glp1r). Given that this classification understates the heterogeneity
of MnPO cell types
28
and that the single-cell dynamics of these neurons during behavior remain unknown,
we used microendoscopic imaging
29
to investigate how these neurons encode aspects of fluid balance.
To gain genetic access to thirst-promoting, glutamatergic neurons in the MnPO, we
generated knockin mice that express Cre recombinase from the Nxph4 locus (Nxph4-2a-Cre;
Fig. 4b and Extended Data Fig. 6). We then targeted GCaMP to these neurons and implanted
a gradient-index lens above the MnPO in order to record their dynamics in awake, behaving
mice (Fig. 4c). We first tested mice in a paradigm in which they were injected with
isotonic saline, then 10 minutes later injected with salt to induce thirst, and then
10 minutes later given access to water (Fig. 4d,e). Analysis of neural dynamics in
this paradigm by k-means clustering revealed three distinct subpopulations (Extended
Data Fig. 7). One subpopulation (cluster 1, 17%) had no response to isotonic saline
injection, but showed dramatic and sustained activation following salt challenge,
suggesting that these neurons encode blood osmolarity. These same neurons were rapidly
and uniformly inhibited during drinking. In contrast, neurons from cluster 2 (34%)
showed only transient responses that we interpret as likely representing stress or
pain, whereas neurons from cluster 3 (49%) were largely unresponsive. We then investigated
how these neurons respond to changes GI osmolarity. Intragastric infusion of water
into dehydrated mice inhibited a subset of MnPONxph4 neurons (24–26%), whereas infusion
of hypertonic saline into hydrated mice activated a similar proportion of cells (34%;
Fig. 4f,g and Extended Data Fig. 7). Registration of neurons across trials revealed
that these two populations were largely overlapping, indicating that a specific subpopulation
of MnPONxph4 neurons is bidirectionally modulated by GI signals. Moreover, the majority
of GI-tuned MnPONxph4 neurons were robustly activated during thirst (Fig. 4h and Extended
Data Fig. 7). This reveals that individual MnPONxph4 neurons receive ingestion signals
from the oropharynx, satiation signals from the GI tract, and homeostatic signals
from the blood, which they integrate to estimate physiologic state. Experiments that
combined chemogenetic silencing
30
with fiber photometry recordings suggest that these MnPO neurons are required for
relaying GI osmolarity information to the SON, but not the SFO (Extended Data Fig.
8).
In contrast to MnPONxph4 neurons, microendoscopic imaging of intermingled GABAergic
MnPO neurons showed only weak and transient responses during salt challenge (Fig.
5a and Extended Data Fig. 9), indicating that these cells do not encode blood osmolarity
in their baseline activity. We therefore recorded the dynamics of these neurons during
water re-access after dehydration, which revealed strong responses during drinking
(Fig. 5b–d and Extended Data Fig. 9). The majority of MnPOGlp1r neurons were either
activated (28%) or inhibited (36%) during water ingestion; these responses were time-locked
to the act of drinking, such that their activity returned to baseline when ingestion
stopped. The cells exhibiting these varied responses were spatially intermingled (Fig.
5c), suggesting that the functional diversity of the MnPO is not anatomically organized
at the scale of our recordings (500 μm). Of note, an earlier study using fiber photometry
detected only activation of MnPOGlp1r neurons during drinking
12
, which suggests that bulk fluorescence measurements masked the existence of an equal
proportion of ingestion-inhibited cells. In addition, we observed smaller subsets
of MnPOGlp1r neurons that were activated or inhibited by water access alone (Fig.
5b) or by i.g. infusion of either water or hypertonic saline (Fig. 5e–g and Extended
Data Fig. 9). Together, these data indicate that the majority of GABAergic MnPO neurons
are strongly modulated during ingestion, with smaller subsets encoding diverse signals
relevant to fluid balance such as water availability, stress, and GI osmolarity.
Understanding how eating and drinking reduce appetite is one of the fundamental challenges
in physiology. Traditionally, this problem has been studied by measuring the effect
of peripheral manipulations on behavior
31,32
. Here, we have taken a complementary approach that uses the dynamics of appetite-promoting
neurons in the brain as a readout to monitor in real-time the functional significance
of manipulations to peripheral tissues. Using this strategy, we have shown that the
water and salt content of the GI tract are precisely tracked during drinking and then
communicated to the key neurons in the brain that control fluid balance. Furthermore,
we have shown that this gut-to-brain signal requires the vagus nerve and is necessary
and sufficient for the satiation of thirst and vasopressin secretion during normal
behavior.
The GI osmosensory signal described here operates alongside oropharyngeal and bloodborne
cues to regulate drinking behavior. We propose that these anatomically and temporally
distinct signals cooperate to promote satiation in three steps (Extended Data Fig.
10). First, detection of liquid in the mouth generates a rapid signal that reports
the volume of fluid ingested
9–12
. This early estimate of volume inhibits forebrain thirst neurons during the act of
drinking, but is transient. Next, detection in the GI tract generates a second signal
that reports the osmolarity of the ingested fluid. This early estimate of osmolarity
stabilizes the inhibition of forebrain thirst neurons if water was consumed or causes
their activity to rebound if the ingested fluid was hypertonic. Finally, absorption
of water into the bloodstream alters fluid balance throughout the body, leading to
sustained changes in well-characterized signals such as blood osmolarity that are
monitored by the brain directly
1–4
. This three-step mechanism enables the brain to dynamically adjust drinking behavior
to match the level of homeostatic need regardless of the composition of ingested fluids.
The concept of a “set point” or “balance point” has played a dominant role in shaping
how we think about homeostasis
33,34
. Inherent in this concept is the idea that there is an anatomic site where measurements
from the body are integrated to estimate the level of a physiologic variable. While
this is commonly assumed to happen in specific neurons in the brain, it has rarely,
if ever, been directly observed in vivo. We have shown here that individual, genetically
defined thirst neurons in the MnPO integrate information about fluid balance arising
from the oropharynx, GI tract, and blood. This reveals that homeostatic need can be
computed at the level of single neurons in a living animal. Further study of this
single-cell integration may provide new insight into the origin of the enigmatic set-points
that characterize physiologic systems.
Methods
Experimental protocols were approved by the University of California, San Francisco
IACUC following the NIH Guide for the Care and Use of Laboratory Animals.
Mouse strains
Adult mice (>6 weeks old) of both sexes were used for experiments. We obtained Nos1-ires-Cre
knockin mice
35
(Nos1tm1(cre)Mgmj
/J, stock no. 017526), Avp-ires2-Cre knockin mice
36
(Avptm1.1(cre)Hze
/J, stock no. 023530), Glp1r-ires-Cre knockin mice
37
(Glp1rtm1.1(cre)Lbrl
/J, stock no. 029283), Rosa26-lsl-Gfp-Rpl10 knockin mice
38
(Gt(ROSA)26Sortm1(CAG-EGFP/Rpl10a,-birA)Wtp
/J, stock no. 022367), and wild type mice (C57BL/6J, stock no. 000664) from the Jackson
Laboratory. We obtained Ai148D GCaMP6f knockin mice
39
from the Allen Institute for Brain Science and Trpv1-Gfp-2a-Dtr BAC transgenic mice
40
from Mark Hoon at the National Institutes of Health.
We generated Nxph4-2a-Cre knockin mice by CRISPR/Cas9-mediated homologous recombination
as previously described
27
based on published protocols
41,42
. Briefly, homology regions were captured into a plasmid from a BAC containing the
Nxph4 locus by recombineering. The T2A-Cre sequence was inserted immediately upstream
of the endogenous stop codon. The final targeting vector contained ~3 kb (5′) and
~1.3 kb (3′) homology arms and was verified by restriction digest and sequencing.
To generate site-specific double stranded breaks using CRISPR, an sgRNA sequence (GAGTGAGACTGCGATCTGGT)
was selected such that the guide sequence would be separated from the PAM site in
the genomic DNA by the T2A-Cre insertion. This ensured that the targeting vector and
recombined Nxph4-2a-Cre allele were protected from Cas9 nuclease activity. Super-ovulated
female FVB/N mice were mated to FVB/N stud males, and fertilized zygotes were collected
from oviducts. Cas9 protein (100 ng/μL), sgRNA (50 ng/μL), and targeting vector DNA
(20 ng/μL) were mixed and injected into the pronucleus of fertilized zygotes. Injected
zygotes were implanted into oviducts of pseudopregnant CD1 female mice. Founder pups
and offspring were genotyped for the presence of the knockin allele by qPCR. Pups
positive for the knockin allele were crossed to reporter mice, and reporter expression
patterns were identical to endogenous Nxph4 expression in the brain (Extended Data
Fig. 6). All Nxph4-2a-Cre mice used here were maintained on a mixed FVB/C57Bl/6J background.
All mice were maintained in temperature- and humidity-controlled facilities with 12-h
light–dark cycle with ad libitum access to water and standard chow (PicoLab 5053).
Avp-ires2-Cre;Ai148D mice were fed doxycycline (80 mg/kg) chow (BioServ F7515) until
undergoing stereotaxic surgery, after which they were fed standard chow. Mice were
allowed at least two weeks for recovery after stereotaxic surgery before testing.
For dehydration experiments, mice were water-deprived overnight (>16 h) before testing.
Viral vectors
We expressed channelrhodopsin-2 variants
43,44
, GCaMP variants
45
, and chemogenetic receptors
30
in genetically specified neurons by stereotaxic delivery of recombinant AAVs encoding
Cre-dependent or Camk2a-promoter transgene cassettes. We obtained AAV1-CAG-FLEX-GCaMP6s,
AAV5-Camk2a-GCaMP6f, and AAV5-hSyn-FLEX-GCaMP6f from the Penn Vector Core. We obtained
AAV5-Camk2a-hChR2(E123T/T159C)-2a-mCherry and AAV5-Ef1a-DIO-hChR2(H134R)-mCherry from
the UNC Vector Core. We obtained AAV5-hSyn-DIO-hM4D(Gi)-mCherry and AAV5-Camk2a-hM4D(Gi)-mCherry
from Addgene.
Stereotaxic surgery
We performed intracranial surgery using stereotaxic coordinates for the SFO, MnPO,
and SON
46
as previously described
9,27
. For delivery of recombinant AAVs, 100–200 nL of virus was injected (100 nL/min)
at the SFO (−0.60 mm antero-posterior (AP), 0 mm medio-lateral (ML), −2.80 mm dorso-ventral
(DV) relative to bregma) or MnPO (+0.35 mm AP, 0 mm ML, −4.20 mm DV). For SFO and
MnPO photostimulation experiments, a Ø200 μm optical fiber (Thorlabs FT200UMT, CFLC230-10)
was placed 0.30 mm above the injection site in the same surgery. For SFO fiber photometry
experiments, a Ø400 μm optical fiber (Thorlabs BFH48-400, CF440-10) was placed 0.10
mm below the injection site in the same surgery. For SON fiber photometry experiments,
a Ø400 μm optical fiber (Thorlabs BFH48-400, CF440-10) was placed unilaterally above
the SON of Avp-ires2-Cre;Ai148D mice (−0.75 mm AP, −1.20 mm ML, −5.50 mm DV). For
MnPO microendoscope imaging experiments, a Ø500 μm gradient-index (GRIN) lens (6.1
mm length; Inscopix 100-000588) was placed 0.10 mm above the injection site in the
same surgery.
Optical fibers and GRIN lenses were then affixed to the skull using dental cement
(A-M Systems 525000) or MetaBond adhesive cement (Parkwell S380). After at least two
weeks recovery from the lens implantation surgery, mice to be used for microendoscope
imaging were again anaesthetized and a baseplate (Inscopix 100-000279) was placed
above the lens and affixed with MetaBond adhesive cement. When these mice were not
being used for imaging, a baseplate cover (Inscopix 100-000241) was attached to prevent
damage to the GRIN lens.
Intragastric surgery and infusion
We prepared mice for i.g. infusion as previously described
47
based on published protocols
21
. Briefly, catheters were constructed from Silastic tubing (Silastic 508-003), Tygon
tubing (Tygon AAD04119), and a curved metal connector (Component Supply Company NE-9019).
Biologically compatible mesh was attached to the Silastic tubing and around the metal
connector using adhesive (Xiameter RTV-3110 base; Dow Corning 4 catalyst), and a luer
adaptor (Instech LS20) was placed onto the Tygon tubing. Assembled catheters were
sterilized using ethylene oxide. Mice with functional photometry, optogenetic, or
microendoscope implants were anesthetized with ketamine/xylazine and the i.g. catheter
was surgically implanted into the stomach as previously described
47
. Mice were allowed at least one week to recover prior to i.g. infusion and testing.
All i.g. infusions were delivered at a rate of 200 μL/min using a syringe pump (Harvard
Apparatus 70-2001). Solutions of NaCl (75, 150, 250, 375, 500 mM), glucose (1 M),
and mannitol (1 M) were prepared using deionized water, and 1.5 M NaCl solution was
prepared using phosphate-buffered saline (PBS). Previous studies have indicated that
mannitol is not absorbed into the bloodstream from the intestines
48
. We measured the latency for infused fluids to pass through the i.g. catheter itself
en route to the stomach to be approximately 13 sec.
Vagotomy surgery
We prepared mice for i.g. infusion and performed bilateral subdiaphragmatic resection
of the vagus nerve during a single procedure. Mice with functional photometry implants
were anesthetized with ketamine/xylazine and the i.g. catheter was surgically implanted
as described above. Briefly, a 5–7 mm incision was made along the medial line beginning
at the distal edge of the sternum. Within a 3-mm radius around the incision site,
the skin layer was separated from the muscular layer using blunt dissecting scissors.
The liver was then retracted with sterile cotton swabs that had been moistened with
saline so that the distal end of the esophagus could be visualized. Using jeweler’s
forceps, both branches of the vagus nerve were isolated from the esophagus and a 1–2
mm section of each branch of the nerve was resected with scissors. Bilateral subdiaphragmatic
vagotomy was accompanied in the same surgery by pyloroplasty, in which some of the
muscle composing the pyloric sphincter was incised. Briefly, a 2-mm incision was made
along the longitudinal axis of the pylorus without penetrating the lumen. Then each
side of the incision was carefully approximated and sutured with two simple interrupted
stitches. The purpose of the pyloroplasty is to maintain gastrointestinal flow through
the pylorus
49
and thereby alleviate the excessive food retention, gastric distension, and morbidity
that accompanies subdiaphragmatic vagotomy. Control mice for vagotomy experiments
underwent a sham surgery including i.g. catheter implantation and internal organ manipulation
but not vagotomy or pyloroplasty.
To validate the subdiaphragmatic vagotomy, animals received an i.p. injection of wheat
germ agglutinin conjugated to Alexa Fluor 555 (WGA-555; 200 μg per mouse) and were
euthanized four days later. WGA-555 is taken up by axon terminals of intact vagal
motor neurons
50
, whose somas are located in the brainstem and can be visualized by histology (Extended
Data Fig. 4a). Labeling in the dorsal motor nucleus of the vagus was greatly reduced,
but not completely eliminated, by subdiaphragmatic vagotomy. Residual vagal fibers
may be due to incomplete resection during the subdiaphragmatic vagotomy or to regeneration
after surgery
51
.
Behavior
We monitored mouse drinking behavior as previously described
9,27
. All experiments were performed in sound-isolated behavioral chambers (Coulbourn
Habitest Modular System) and were performed during the light cycle to control for
circadian factors. Fluid consumption was monitored with an electrical lickometer and
recorded using Graphic State software (v4.2, www.coulbourn.com/category_s/363.html),
or using LJStreamUD software during fiber photometry experiments (v1.17, www.labjack.com/support/software/applications/ud-series/ljstreamuid)
or nVista software during microendoscope imaging experiments (v2.0, www.inscopix.com/nvista).
Mice were acclimated to the behavioral chamber for at least 10 min at the beginning
of each testing session.
For two-bottle drinking experiments (Extended Data Fig. 1d), wild type mice were dehydrated
and then provided access to two randomly placed bottles (1× water, 1× 300 mM NaCl)
for >10 min before being returned to their home cages. This test was repeated four
times, with the location of the bottles randomly re-assigned on each trial. Mice were
divided into two groups before testing. The first group received a gastric pre-load
of hypertonic saline (1.5 M NaCl; 100 μL) by oral gavage one minute before bottle
access; the second group received a gastric pre-load of isotonic saline (150 mM NaCl;
100 μL) by oral gavage one minute before bottle access.
For three-bottle drinking experiments (Extended Data Fig. 1e–g), SFO photometry mice
were dehydrated and then provided access to three randomly placed bottles (1× water,
2× 300 mM NaCl) for >10 min before being returned to their home cages. This test was
repeated four times, with the location of the bottles randomly re-assigned on each
trial.
Fiber photometry
We prepared mice for in vivo fiber photometry recording as previously described
9
based on published protocols
18
. The fiber photometry signal was output to a lock-in amplifier (Stanford Research
Systems SR810) with time constant 30 ms to allow filtering of noise at higher frequencies.
The signal was then digitized (LabJack U6-Pro) and recorded using LJStreamUD software
(v1.17, www.labjack.com/support/software/applications/ud-series/ljstreamuid) at 250
Hz sampling rate. Photometry data were subjected to minimal processing such as within-trial
fluorescence normalization and temporal downsampling. For these experiments, i.g.
infusions were given in a volume of 1 mL except for Fig. 2c–f and Extended Data Fig.
2h (volume of 200 μL, including ~50 μL dead volume); systemic (i.p.) injections were
given in a volume of 150 μL (1.5 M or 3 M NaCl) using PBS as “vehicle” or in a volume
of 1 mL (water); oral gavages (Extended Data Fig. 8) were given in a volume of 150
μL. Hormones (Extended Data Fig. 4i) were delivered by i.p. injection (volume of 150
μL) for serotonin (2, 20 mg/kg), cholecystokinin (2 mg/kg), ghrelin (2 mg/kg), and
leptin (2 mg/kg) and by subcutaneous injection (volume of 400 μL) for amylin (2 mg/kg)
using PBS as “vehicle”.
Microendoscope imaging
We prepared mice for in vivo microendoscope imaging based on published protocols
29,52
. Videos were acquired at 20 Hz (20% LED power, 2.0 gain) using a miniature microscope
(Inscopix) and nVista software (v2.0, www.inscopix.com/nvista). After acquisition,
videos were first pre-processed, spatially (binning factor of 2) and temporally (binning
factor of 5) downsampled, and motion-corrected using Mosaic software (v1.7, support.inscopix.com/mosaic-workflow).
Activity traces for individual neurons were then extracted from these videos using
the constrained nonnegative matrix factorization (CNMF-E) pipeline
53
(www.github.com/zhoupc/cnmf_e) implemented in Matlab. After initial CNMF-E segmentation,
extracted neurons were manually refined to avoid potential confounds from uncorrected
motion artifacts, ROI duplication, and over-segmentation of the same spatial components.
For each experiment, activity traces for individual neurons were extracted from recordings
from 3–4 mice and then pooled for subsequent analysis. For these experiments, i.g.
infusions were given in a volume of 1 mL; systemic (i.p.) injections were given in
a volume of 100 μL (3 M NaCl) using PBS as “vehicle”.
Optogenetics
We prepared mice for in vivo photostimulation as previously described
9,27
based on published protocols
23,54
. A DPSS 473 nm laser (Shanghai Laser and Optics Century BL473-100FC) was controlled
by Graphic State software (v4.2, www.coulbourn.com/category_s/363.html) through a
TTL signal generator (Coulbourn H03-14) and synchronized with behavior experiments.
The laser power was measured to be ~15 mW at the patch cable tip and was delivered
in 10-ms pulses at 20 Hz. For these experiments, i.g. infusions were given in a volume
of 600 μL.
Chemogenetics
We prepared mice for in vivo chemogenetic inhibition based on published protocols
55,56
. Clozapine N-oxide (CNO, 1 mg/kg) was delivered by i.p. injection (volume of 125
μL) with 0.6% DMSO in PBS as “vehicle”. For these experiments (Extended Data Fig.
8), CNO or vehicle was delivered >15 min before water access after dehydration or
oral gavage of hypertonic NaCl.
Diphtheria toxin receptor ablation
We ablated genetically defined (Trpv1
+) sensory neurons using the DTX receptor-based strategy described in published protocols
40,57
. Briefly, Nos1-ires-Cre;Trpv1-Gfp-2a-Dtr mice were first equipped for fiber photometry
recording of SFO neurons and screened for functionality. Mice with functional photometry
implants then underwent i.g. catheterization surgery and, after recovery, were tested
in a series of experiments (“before DTX” in Extended Data Fig. 4h,i). Mice were then
given two intramuscular injections of DTX (each injection was 50 μL of 25 μg/mL DTX)
separated by three days and then allowed at least five additional days to recover,
after which they underwent the same series of experiments (“after DTX”).
We confirmed the ablation of Trpv1
+ sensory neurons in two ways. First, Nos1-ires-Cre;Trpv1-Gfp-2a-Dtr photometry/i.g.
mice were given access to a single bottle of palatable sugar solution (300 mM sucrose)
that also contained either capsaicin (100 μM) or vehicle (0.15% TWEEN-80 + 1.5% DMSO).
This test was conducted once before and once after DTX treatment, and intake was quantified
from 1–2 h after access (Extended Data Fig. 4h). Second, Trpv1-Gfp-2a-Dtr;hSyn-Nanobody-Rpl10
mice were used for histology of the nodose ganglion and thoracic dorsal root ganglion
(DRG) after either DTX treatment or no treatment. This revealed nearly complete ablation
of Trpv1
+ vagal neurons and more limited ablation of Trpv1
+ DRG neurons (Extended Data Fig. 4e,f). We also observed sparse labeling of Trpv1
+ neurons in the brain
58
, and this was unaffected by DTX treatment (Extended Data Fig. 4g).
Plasma osmolality
Mice equipped with i.g. catheters were fully hydrated and received an infusion (sham,
water, 150 mM NaCl, or 500 mM NaCl) at 200 μL/min for 5 min. Beginning 3 min after
the start of infusion, 125 μL of blood was collected from the tail vein using EDTA-coated
capillary tubes (RAM Scientific 07-6011). The blood collection process took approximately
3 min per mouse. Plasma was isolated by centrifugation (1000 g for 10 min), diluted
in deionized water, and frozen until measurement. Osmolality was then measured in
triplicate for each sample using a freezing point osmometer (Fiske Associates 210).
Mice were allowed one week for recovery between sessions.
Histology
Mice were transcardially perfused with PBS followed by 10% formalin. To visualize
forebrain and hindbrain nuclei, whole brains were dissected, post-fixed in 10% formalin
overnight at 4°C, and then cryo-protected in 30% sucrose overnight at 4°C. Free-floating
sections (40 μm) were prepared with a cryostat, blocked (3% BSA, 2% NGS, and 0.1%
Triton-X in PBS for 2 h), then incubated with primary antibody (chicken anti-GFP,
Abcam ab13970, 1:1000; rat anti-RFP, ChromoTek 5f8, 1:1000; goat anti-mCherry, Acris
ab0040-200, 1:1000; rabbit anti-Fos, Santa Cruz Biotech sc52, 1:500) overnight at
4°C. Sections were then washed, incubated with secondary antibody (Alexa Fluor 488
goat anti-chicken, Life Technologies a11039, 1:1000; Alexa Fluor 568 goat anti-rat,
Life Technologies a11077, 1:1000; Alexa Fluor 568 goat anti-rabbit, Life Technologies
a11011, 1:1000; Alexa Fluor 568 donkey anti-goat, Life Technologies a11057, 1:1000)
for 2 h at room temperature, washed again, mounted with DAPI Fluoromount-G (Southern
Biotech), and then imaged with a confocal microscope (Zeiss LSM-700). To visualize
WGA-555 labelling of the dorsal motor nucleus of the vagus, sections were mounted
and imaged without staining.
To visualize sensory ganglia, the ventral aspects of the skull and vertebral columns
were dissected, post-fixed in 10% formalin overnight at 4°C, and then washed at room
temperature. The nodose ganglion and thoracic DRG were further dissected and then
cryo-protected in 30% sucrose overnight at 4°C. Sections (20 μm) were prepared with
a cryostat, mounted on slides, washed, blocked (5% NGS and 0.1% Triton-X in PBS for
30 min), and then incubated with primary antibody (chicken anti-GFP, Abcam ab13970,
1:1000; rabbit anti-NeuN, Milliipore abn78, 1:1000) overnight at 4°C. Sections were
then washed, incubated with secondary antibody (Alexa Fluor 488 goat anti-chicken,
Life Technologies a11039, 1:500; Alexa Fluor 568 goat anti-rabbit, Life Technologies
a11011, 1:500) for 2 h at room temperature, washed again, mounted with DAPI Fluoromount-G
(Southern Biotech), and then imaged with the confocal microscope.
Images underwent minimal processing, such as brightness and contrast adjustments,
performed using the Fiji distribution of ImageJ (v2.0, www.fiji.sc).
Data analysis
We analyzed behavior data, fiber photometry data, and microendoscope imaging data
using custom Matlab (vR2018a, www.mathworks.com/products/matlab) scripts. Throughout
the paper, a drinking bout is defined as any set of ten or more licks in which no
inter-lick interval is greater than one second. For some assays, multiple trials of
the same experiment were performed for an individual mouse and then averaged and treated
as a single replicate. For photometry data, all responses were normalized using the
function: ΔF/F
0 = (F–F
0)/F
0, in which F is the raw photometry signal and F
0 is the median of F during the baseline period (plotted data before injection/infusion/drinking).
For quantification in bar graphs, the median ΔF/F
0 of a 10-s (for long experiments with >5 min of data plotted) or 2-s (for short experiments
with <5 min of data plotted) window is reported. For peri-event time histograms around
the end of drinking bouts, F
0 was defined as the median of F at the time (±1 s) of the last lick in the bout.
For microendoscope imaging data, all responses were normalized using the function
z = (Craw
–μ)/σ, in which Craw
is the output of the CNMF-E pipeline, μ is the mean of Craw
during the baseline period, and σ is the standard deviation of Craw
during the baseline period. k-means clustering was performed using the built-in Matlab
function (www.mathworks.com/help/stats/kmeans.html) as illustrated in Extended Data
Fig. 7a. For quantification in linear regressions, the mean z-score of the final 1
min of plotted data for i.g. infusions (9–10 min after infusion begins) or of the
first 1 min after drinking onset (first lick in first bout) is reported. The mean
z-score of the first 1 min after the first lick in the first drinking bout was used
to classify GABAergic MnPO neuron responses during ingestion (Fig. 5c,d and Extended
Data Fig. 9b,c). For time-courses of water intake, time 0 is the moment of water access
unless otherwise noted. Initial drinking rate (Fig. 2b) was calculated from the first
1 min of drinking after the first lick in the experiment. Latency to inhibition of
SFONos1neurons by i.g. infusion of water (main text) was calculated as the moment
at which the fiber photometry signal crossed a threshold of four standard deviations
(4σ) below the signal level during the baseline period and was adjusted to account
for the dead volume of the i.g. catheter.
Statistics and reproducibility
We performed statistical analyses using Prism (v7.0, www.graphpad.com/scientific-software/prism).
Throughout the paper, values are reported as mean ± s.e.m. (error bars or shaded area).
In figures with linear regressions, the shaded area represents the 95% confidence
interval for the line-of-best-fit. P-values for pair-wise comparisons were performed
using a two-tailed Student’s t-test (with repeated measures when possible). P-values
for comparisons across multiple groups were performed using ANOVA (with repeated measures
when possible) and corrected for multiple comparisons using the Holm-Šídák method
(within-group comparison to the control condition). *P < 0.05, **P < 0.01, ***P <
0.001, ****P < 0.0001. Randomization, blinding, and statistical methods to predetermine
sample size were not used. Representative images were selected from one to five original
biological replicates, and representative recordings were selected from three to five
original biological replicates. Representative mouse brain schematics (coronal and
sagittal sections) were adapted from ref.
46
. More information about statistics and reproducibility is available in the online
Reporting Summary.
Code availability
Custom Matlab scripts that support the findings of this study are available upon reasonable
request the corresponding author.
Data availability
Data that support the findings of this study are available upon reasonable request
the corresponding author.
Extended Data
Extended Data Figure 1.
GI osmolarity influences drinking behavior and biases salt preference.
Panels a,b present additional data related to Fig. 1b,c. a, Cumulative water or 300
mM NaCl intake after dehydration (n = 5 mice). b, Example SFO neuron dynamics during
drinking after dehydration. Panel c shows that ingestion of hypertonic fluids activates
SFO neurons regardless of hydration state. c, Average SFO activity and drinking behavior
of hydrated mice given ad libitum access to isotonic (300 mM sucrose) or hypertonic
(300 mM sucrose + 600 mM mannitol) sugar solutions of similar sweetness (n = 5 mice).
Panel d shows that increases in GI osmolarity bias salt/water preference. d, Preference
in a two-bottle test after i.g. treatment with hypertonic (red; n = 8 mice) or isotonic
(black; n = 9 mice) NaCl (left; two-way ANOVA, Holm-Šídák correction). Cumulative
water (solid lines) and 300 mM NaCl (dashed lines) intakes in the same two-bottle
test (right). Panels e–g show that post-ingestive SFO neuron activity does not reflect
the delayed consequences of taste or sensorimotor experience associated with an individual
drinking bout. e, Mice initially do not distinguish between bottles containing water
or 300 mM NaCl in a three-bottle test after dehydration (n = 4 mice, linear regression,
R
2
= 0.3163, P = 0.0233). f, Example SFO neuron dynamics during drinking from water (black)
and 300 mM NaCl (blue, red) bottles after dehydration. g, SFO neuron dynamics during
individual water (42 bouts) or NaCl (71 bouts) drinking bouts in trials 1 and 2 of
the three-bottle test (left). Average SFO activity after individual drinking bouts
(right; n = 4 mice). In this experiment (panels e–g), GI osmolarity quickly becomes
hypertonic as the dehydrated mice alternate between drinking from water and NaCl bottles
such that SFO neuron activity “rebounds” even after water drinking bouts, which suggests
that the stabilization signal that either quenches or re-activates SFO neurons after
ingestion reflects GI osmolarity. Error bars represent mean ± s.e.m. Shaded areas
in a,c,d,e,g represent mean ± s.e.m.; in b represent individual licks; in the linear
regression (right) in e represent 95% confidence interval of the line-of-best-fit;
in f represent individual drinking bouts. **P < 0.01, ***P < 0.001.
Extended Data Figure 2.
The GI→SFO osmosensory signal depends on fluid tonicity but not osmolyte identity.
Panels a,b show that i.g. infusion does not rapidly alter the state of the blood.
a, Schematic. b, Plasma osmolality of samples collected during approximately 3–6 min
after the start of the 5-min i.g. infusion (n = 9 mice per group, one-way ANOVA, Holm-Šídák
correction). Panels c–e show that the GI→SFO osmosensory signal depends on fluid tonicity
but not osmolyte identity. c, SFO neuron dynamics of individual mice in response to
i.g. infusion of equiosmotic concentrations of NaCl, which is absorbed into the bloodstream
from the GI tract, and mannitol, which is not absorbed (top; n = 4 mice). SFO neuron
dynamics of a separate cohort of individual mice in response to i.g. infusion of equiosmotic
concentrations of NaCl, which does not permeate cell membranes and has high tonicity,
and glucose, which does permeate cell membranes and has low tonicity (bottom; n =
5 mice). d, Average SFO activity during i.g. infusion of NaCl or mannitol (left).
Quantification (right; n = 4 mice, one-way ANOVA, Holm-Šídák correction). e, Average
SFO activity during i.g. infusion of NaCl or glucose (left). Quantification (right;
n = 5 mice, one-way ANOVA, Holm-Šídák correction). Panels f–h show that SFO neurons
encode systemic and GI osmosensory signals additively rather than hierarchically.
f, Schematic. g, Example (left) and average (right; n = 4 mice) SFO neuron dynamics
during 1.5 M NaCl i.p. injection followed by water i.g. infusion. h, Example (left)
and average (right; n = 3 mice) SFO neuron dynamics during 1.5 M NaCl i.g. infusion
followed by water i.p. injection. Error bars represent mean ± s.e.m. Shaded areas
in summary traces (d,e,g,h) represent mean ± s.e.m. and in example traces (g,h) represent
i.g. infusion. *P < 0.05, **P < 0.01; n.s., not significant.
Extended Data Figure 3.
The GI→SFO osmosensory signal completely satiates but only mildly stimulates thirst.
Panels a,b present additional data related to Fig. 1e–g and Fig. 2a,b. a, Average
SFO activity during i.g. infusions and subsequent drinking while hydrated (left).
Cumulative water intake (right; n = 4 mice). b, Average SFO activity during i.g. infusions
and subsequent drinking after dehydration (left). Cumulative water intake (right;
n = 4 mice). c, Correlation between SFO activity change and latency to drinking after
1 mL infusions into hydrated (black; n = 23 experiments from 4 mice, linear regression,
R
2
= 0.0705, P = 0.2208) or dehydrated (red; n = 12 experiments from 4 mice, linear regression,
R
2
= 0.1321, P = 0.2456) mice. Panel d presents additional data related to Fig. 2c. d,
Average SFO activity after systemic (i.p.) or i.g. treatment with 150 μL NaCl while
hydrated. Shaded areas in a,b,d represent mean ± s.e.m. and in c represent 95% confidence
interval of the line-of-best-fit.
Extended Data Figure 4.
The GI→SFO osmosensory signal involves the vagus nerve.
Panels a–d show that the GI→SFO osmosensory signal is disrupted by subdiaphragmatic
vagotomy. a, Vagal motor neuron somas (located in the brainstem and labeled by i.p.
injection of wheat germ agglutinin, WGA-555) were largely absent following subdiaphragmatic
vagotomy (two examples per condition; scale bar, 1 mm). b, Drinking after dehydration
was less suppressed by i.g. infusion of water in vagotomized mice (middle; n = 7 mice)
compared to sham mice (left; n = 6 mice). Quantification (right; two-tailed Student’s
t-test). c, Drinking was similarly suppressed in both groups by systemic (i.p.) delivery
of water (n = 4 sham and 7 vagotomy, two-tailed Student’s t-test). d, SFO modulation
by water and 500 mM NaCl i.g. infusions, but not by 1.5 M NaCl i.p. injection, was
attenuated in vagotomized mice compared to sham mice (n = 8 mice per group, two-tailed
Student’s t-tests). Panels e–i show that the GI→SFO osmosensory signal involves Trpv1
+ sensory neurons. e, To specifically ablate Trpv1
+ sensory neurons, we treated mice containing a BAC transgene expressing GFP and the
diphtheria toxin (DTX) receptor from the Trpv1 gene start codon (Trpv1-Gfp-2a-Dtr
mice) with DTX (scale bar, 100 μm). f, Quantification (n = 3 control and 2 DTX; NG,
nodose ganglion; DRG, dorsal root ganglion). g, DTX treatment did not ablate Trpv1
+ neurons in the brain (scale bar, 1 mm). h, Hydrated mice avoided drinking 300 mM
sucrose that contained 100 μM capsaicin (Cap.) before, but not after, DTX ablation
of Trpv1
+ sensory neurons (n = 5 mice, two-way ANOVA, Holm-Šídák correction; Veh., vehicle).
i, SFO modulation by water i.g. was significantly attenuated after DTX ablation of
Trpv1
+ sensory neurons, and modulation by 500 mM NaCl i.g. was slightly attenuated (n =
7 mice, two-tailed Student’s t-tests). Panel j shows the response of SFO neurons to
serotonin and other visceral hormones. j, SFO neuron dynamics during injection of
two doses of serotonin (left; n = 5 mice) and to a single dose (2 mg kg−1) of amylin,
cholecystokinin (CCK), ghrelin, or leptin (right; n = 6 mice, one-way ANOVA, Holm-Šídák
correction) in hydrated mice. Error bars and shaded areas represent mean ± s.e.m.
*P < 0.05, **P < 0.01; n.s., not significant.
Extended Data Figure 5.
Vasopressin neurons integrate systemic and GI osmosensory signals and are stress-responsive.
Panels a,b present additional data related to Fig. 3a,b. a, Schematic for fiber photometry
recording of vasopressin neurons (scale bar, 1 mm). b, Vasopressin neuron dynamics
(average, left; individual mice, right) during vehicle or NaCl i.p. injection (n =
7 mice). Panel c shows that vasopressin neurons are stress-responsive. c, Vasopressin
neuron activity during tail suspension (n = 7 mice). Panels d–j present additional
data related to Fig. 3d. d, Schematic. e, Vasopressin neuron activity change after
infusion while hydrated or dehydrated (n = 4 mice, two-way ANOVA, Holm-Šídák correction;
Hyd., hydrated; Dehyd., dehydrated). f, Vasopressin neuron activity during i.g. infusions
while hydrated (n = 4 mice). g, Vasopressin neuron dynamics of individual mice (left)
and distribution of ΔF/F
0 values before and after 500 mM NaCl i.g. infusion for those mice (right). h, Vasopressin
neuron dynamics during i.g. infusions after dehydration (n = 4 mice). i, Vasopressin
neuron activity of individual mice (left) and distribution of ΔF/F
0 values before and after water i.g. infusion for those mice (right). j, GI osmolarity
modulates both the median of ΔF/F
0 (left; used here as a proxy for tonic activity) and the standard deviation (σ) of
ΔF/F
0 (right; used here as a proxy for bursting activity) of vasopressin neurons (n =
4 mice, two-tailed Student’s t-tests). Error bars represent mean ± s.e.m. Shaded areas
in b,c,f,h represent mean ± s.e.m and in g,i represent “before” and “after” infusion
periods. *P < 0.05, **P < 0.01, ***P < 0.001.
Extended Data Figure 6.
Nxph4-expressing MnPO neurons are activated by dehydration and drive thirst.
Panel a presents additional data related to Fig. 4b. a, The Nxph4–2a-Cre recombination
pattern (bottom; crossed to a GFP reporter line) recapitulates the endogenous Nxph4
mRNA expression pattern (top; Allen Institute for Brain Science ISH #73521000) in
the organum vasculosum of the lamina terminalis (OVLT), MnPO, SFO, and paraventricular
hypothalamus (PVH). Panel b shows that MnPONxph4 neurons are activated by dehydration.
b,
Nxph4-2a-Cre recombination (green; crossed to a GFP reporter line) and the immediate
early gene Fos (red; induced by 3 M NaCl i.p. injection) co-localize in the MnPO during
dehydration (scale bar, 100 μm). Panels c,d show that MnPONxph4 neurons drive thirst.
c, Schematic for optogenetic activation of MnPONxph4 neurons (scale bar, 1 mm). d,
Water intake in response to photostimulation (left). Quantification (right; n = 4
mice, two-tailed Student’s t-test). Error bars and shaded areas represent mean ± s.e.m.
*P < 0.05.
Extended Data Figure 7.
In vivo imaging of individual glutamatergic MnPO neurons during thirst, drinking,
and GI manipulation.
Panel a presents additional data related to Fig. 4d,e. a, Workflow for k-means clustering
of individual MnPONxph4 neurons based on their activity during vehicle i.p. injection,
3 M NaCl i.p. injection, and water drinking. Panels b–d present additional data related
to Fig. 4f–h. b, Schematic. c, Dynamics of individual neurons during water i.g. infusion
while hydrated. d, Dynamics of individual neurons tracked during water i.g. infusion
after dehydration (left) and 3 M NaCl i.p. injection (right). Neurons inhibited ≥1σ
after water i.g. infusion were classified as “GI-tuned” (red; 26%) and the remaining
neurons were classified as “GI-untuned” (black; 74%) for the time-course plotted in
Fig. 4h.
Extended Data Figure 8.
Glutamatergic MnPO neurons relay the GI osmosensory signal to vasopressin neurons.
Panels a–d show that glutamatergic MnPO neurons are necessary for relaying GI osmosensory
information to SON vasopressin neurons. a, Schematic for simultaneous fiber photometry
recording of vasopressin neurons and chemogenetic inhibition of glutamatergic MnPO
neurons (scale bar, 100 μm). b, Injection of clozapine N-oxide (CNO) inhibited water
intake after dehydration (n = 5 mice). c, Example vasopressin neuron dynamics during
CNO or vehicle i.p. injection (left) and subsequent 1.5 M NaCl i.g. by oral gavage
(right). Inset, water intake after dehydration for this example mouse. d, Quantification
of vasopressin neuron response to i.p. injection (left) and NaCl i.g. (right; n =
5 mice, two-tailed Student’s t-tests). Panels e–h show that glutamatergic MnPO neurons
are not necessary for relaying GI osmosensory information to SFO thirst neurons. e,
Schematic for simultaneous fiber photometry recording of SFO neurons and chemogenetic
inhibition of glutamatergic MnPO neurons (scale bar, 100 μm). f, Injection of CNO
inhibited water intake after dehydration (n = 5 mice). g, Example SFO neuron dynamics
during 1.5 M NaCl i.g. by oral gavage after CNO or vehicle i.p. injection. Inset,
water intake after dehydration for this example mouse. h, Quantification of SFO neuron
response to i.p. injection (left) and NaCl i.g. (right; n = 5 mice, two-tailed Student’s
t-tests). Panel i shows that CNO inhibits drinking in mice expressing hM4D(Gi) in
glutamatergic MnPO neurons but not in control mice lacking hM4D(Gi). i, Injection
of CNO significantly inhibited water intake after dehydration in MnPOCamk2a::hM4D(Gi)
+ SON photometry mice (n = 5 mice; quantified from panel b) and MnPONos1::hM4D(Gi)
+ SFO photometry mice (n = 5 mice; quantified from panel f) but not in control mice
(n = 6 mice, two-way ANOVA, Holm-Šídák correction). Error bars and shaded areas represent
mean ± s.e.m. *P < 0.05, **P < 0.01; n.s., not significant.
Extended Data Figure 9.
In vivo imaging of individual GABAergic MnPO neurons during thirst, drinking, and
GI manipulation.
Panel a shows that individual GABAergic MnPO neurons do not encode systemic osmolarity
in their baseline activity. a, Dynamics of individual neurons during 3 M NaCl i.p.
injection while hydrated (left). Comparison to thirst-activated MnPONxph4 neurons
(right; “cluster 1” from Fig. 4e). Panels b,c show the dynamics of ingestion-tuned
GABAergic MnPO neurons during hypertonic NaCl drinking. b, Dynamics of individual
neurons during 300 mM NaCl drinking after dehydration (left). Proportion of ingestion-activated
(red; modulated ≥1σ during first min of drinking), ingestion-inhibited (blue; modulated
≤–1σ), and untuned (black) neurons during water (top; n = 77 neurons from Fig. 5c,d)
or 300 mM NaCl (bottom; n = 95 neurons) drinking (right). c, Average responses of
ingestion-activated, ingestion-inhibited, and untuned neurons during 300 mM NaCl drinking
(n = 95 neurons). Note that ingestion of NaCl persists for much longer than ingestion
of water after dehydration (see Fig. 5d and Extended Data Fig. 1a), which may explain
differences in the dynamics of ingestion-tuned GABAergic MnPO neurons when mice drink
these fluids. Panels d,e present additional data related to Fig. 5e–g. d, Schematic.
e, Dynamics of individual neurons during 500 mM NaCl i.g. infusion while hydrated.
Extended Data Figure 10.
Schematic for the neural control of thirst and satiation.
Anatomically and temporally distinct peripheral sensory signals encode information
about the body’s current hydration state (blood) as well as the volume (oropharynx)
and osmolarity (GI tract) of recently ingested fluids. These signals converge on the
brain’s thirst circuit to generate an integrated central representation of fluid balance
at the level of individual neurons, which use this information to dynamically control
drinking behavior and vasopressin secretion in real-time. Illustration from iStock/artsholic.
Supplementary Material
1