Sodium appetite and thirst do not require angiotensinogen production in astrocytes or hepatocytes

In addition to its renal and cardiovascular functions, angiotensin signalling is thought to be responsible for the increases in salt and water intake caused by hypovolaemia. However, it remains unclear whether these behaviours require angiotensin production in the brain or liver. Here, we use in situ hybridization to identify tissue‐specific expression of the genes required for producing angiotensin peptides, and then use conditional genetic deletion of the angiotensinogen gene (Agt) to test whether production in the brain or liver is necessary for sodium appetite and thirst. In the mouse brain, we identified expression of Agt (the precursor for all angiotensin peptides) in a large subset of astrocytes. We also identified Ren1 and Ace (encoding enzymes required to produce angiotensin II) expression in the choroid plexus, and Ren1 expression in neurons within the nucleus ambiguus compact formation. In the liver, we confirmed that Agt is widely expressed in hepatocytes. We next tested whether thirst and sodium appetite require angiotensinogen production in astrocytes or hepatocytes. Despite virtually eliminating expression in the brain, deleting astrocytic Agt did not reduce thirst or sodium appetite. Despite markedly reducing angiotensinogen in the blood, eliminating Agt from hepatocytes did not reduce thirst or sodium appetite, and in fact, these mice consumed the largest amounts of salt and water after sodium deprivation. Deleting Agt from both astrocytes and hepatocytes also did not prevent thirst or sodium appetite. Our findings suggest that angiotensin signalling is not required for sodium appetite or thirst and highlight the need to identify alternative signalling mechanisms.


Introduction
Thirst and sodium appetite are essential for survival when fluid losses cause hypovolaemia. Hypovolaemia initially 0 Lila Peltekian is a medical student who worked as a Research Assistant in Dr Geerling's lab at the University of Iowa. She played a key part in this project and investigated the mechanisms that mediate sodium appetite. She helped Dr Geerling establish his lab and initiate this and other projects. She is currently in her third year of the MD program, preparing for a neurology residency. Silvia Gasparini got her PhD from São Paulo State University in Brazil, where she focused on the central mechanisms involved in the regulation of sodium appetite. After completing a postdoctoral training programme, she had the opportunity to come to the USA and continue her work on sodium appetite at the University of Iowa. Since arriving in Iowa City in 2017, she has been working as a Research Scientist in Dr Geerling's lab, where she is investigating the circuit and molecular mechanisms underlying sodium appetite. (Lowell, 2019;Zimmerman et al., 2017), the precise signalling mechanisms that detect hypovolaemia and activate these neurons remain uncertain. Past work on this topic has focused on endocrine signals, particularly angiotensin II (AngII). Angiotensin signalling is required for normal kidney development and blood pressure (Esther et al., 1996;Kim et al., 1995;Oliverio et al., 1998;Takahashi et al., 2005;Tsuchida et al., 1998). Also, it is widely believed that AngII is necessary for stimulating thirst and sodium appetite during hypovolaemia (Augustine et al., 2020;Crews & Rowland, 2005;Daniels & Fluharty, 2004;Fitzsimons, 1998;Hurley & Johnson, 2015;Lowell, 2019;Matsuda et al., 2017;Sakai et al., 1986;Stocker et al., 2000;Weisinger et al., 1996;Zimmerman et al., 2017). Several lines of evidence support this idea. Hypovolaemia increases circulating AngII (Resch et al., 2017;Russell et al., 1975), and AngII stimulates drinking (Epstein et al., 1970;Fitzsimons, 1998). Neurons that promote thirst and sodium appetite express the angiotensin receptor gene Agtr1a (Allen et al., 2017;Gasparini, Resch et al., 2019;Resch et al., 2017), and eliminating this gene prevents sodium appetite (Matsuda et al., 2017). Losartan, an angiotensin receptor antagonist, prevents exogenous AngII from stimulating water intake (Fregly & Rowland, 1991) and, in larger doses or if injected directly into the brain, it reduces thirst and sodium appetite (Crews & Rowland, 2005;Fitzsimons, 1998;Resch et al., 2017).
Thus, it remains unclear whether AngII is necessary for hypovolaemic thirst or sodium appetite. One way to address this knowledge gap would be to eliminate the gene for angiotensinogen (Agt), which is the precursor for all angiotensin peptides (Alexiou et al., 2005;. Surprisingly, angiotensinogen-knockout (Agt −/− ) mice are polydipsic (McKinley et al., 2008), possibly to compensate for their impaired urinary concentrating ability and low blood pressure (Alexiou et al., 2005;Kihara et al., 1998;Kim et al., 1995;Lu, Wu et al., 2016;Taniguchi et al., 1998). Unfortunately, Agt −/− mice are also refractory to furosemide (Michael McKinley, personal communication), which is the loop diuretic commonly used to test sodium appetite (Jarvie & Palmiter, 2017;Lee et al., 2019;Matsuda et al., 2017;Resch et al., 2017;Rowland & Fregly, 1988), and producing hypovolaemia via an alternative method caused adipsia and anorexia with roughly 50% mortality in this knockout strain (McKinley et al., 2008). Both the tissue origin and necessity of angiotensin peptides for hypovolaemic thirst and sodium appetite remain uncertain.
In the present study, we began by examining the expression of renin-angiotensin system genes in the mouse brain, liver and kidney. We then used Cre-conditional, cell-type-specific deletion of Agt to test whether endogenous production of angiotensin peptides is required for thirst or sodium appetite. We first deleted Agt in astrocytes, which are the main source of angiotensinogen in the brain (La Manno et al., 2021;Milsted et al., 1990;Stornetta et al., 1988;Zeisel et al., 2018). We then deleted Agt in hepatocytes, which produce the majority of circulating angiotensinogen (Campbell & Habener, 1986;Campbell et al., 1984;Matsusaka et al., 2012;Yiannikouris et al., 2015). Our tissue-specific strategy bypassed the renal pathology, furosemide resistance and mortality that complicated earlier experiments after whole-body deletion of Agt (Alexiou et al., 2005;Kim et al., 1995;McKinley et al., 2008;Okubo et al., 1997;Taniguchi et al., 1998;Tanimoto et al., 1994). We designed these experiments to address the importance of central versus peripheral angiotensin production (Fitzsimons, 1998;McKinley et al., 2008;Moe et al., 1984;Sakai et al., 1990;Weisinger et al., 1996;Yang & Epstein, 1991), with a working hypothesis that deleting Agt in the brain would prevent sodium appetite. The Jackson Laboratory 003574; B6.Cg-Speer6-ps1 Tg(Alb-cre)21Mgn /J Postic et al. (1999)
was taken to minimize both the number and any pain and distress of animals used.

Mice
All mice were group-housed in a temperature-and humidity-controlled room with a 12/12 h light/dark cycle. All had ad libitum access to water and standard rodent chow (Envigo 7013; Indianapolis, IN, USA) prior to the experiments described below. We bred 'floxed' angiotensinogen (Agt flox/flox ) mice with two different Cre strains to delete Agt expression in either the brain or the liver. Genotyping (performed by Transnetyx, Cordova, TN, USA using real-time PCR) revealed expected Mendelian ratios for the floxed Agt allele and for both Cre transgenes. Both strains underwent multiple generations of breeding to produce experimental mice (GFAP-Cre;Agt flox/flox or Alb-Cre;Agt flox/flox ) and littermate control mice. The primary controls for each strain were Cre-negative littermates (two separate strains of Agt flox/flox mice, each compared to experimental littermates from their respective strain). We also tested small numbers of 'single-floxed' littermates (GFAP-Cre;Agt flox/+ or Alb-Cre;Agt flox/+ ). The drinking behaviours of these 'single-floxed' mice were highly similar to the primary control mice (Agt flox/flox ), so in each colony we combined Cre-negative and 'single-floxed' control mice as a single group for analysis. The underlying strains we used have been characterized in previous studies to delete angiotensinogen or other genes in the liver, kidney and brain Matsusaka et al., 2012;Sherrod et al., 2005;Yiannikouris et al., 2015), and further information for each strain can be found in Table 1. Due to the reported impact of sex on sodium appetite, which is reduced and more variable in female rats (Stricker et al., 1991), we used male mice exclusively. All mice began habituating to experimental cages between 6 and 12 weeks of age. Body weights ranged 23−32 g at the end of the experiment. In contrast to whole-body Agt-knockout mice (Taniguchi et al., 1998), there was no perinatal mortality in any of our conditionally deleted strains. None of these mice required saline injections or any other special treatments before or after weaning, and there was zero mortality after experimental hypovolaemia. In contrast to the low body weights of Agt −/− mice Massiera et al., 2001;Tanimoto et al., 1994), we did not find any differences in the body weights of mice after conditionally deleting Agt using GFAP-Cre (n = 13 GFAP-Cre;Agt flox/flox 24.9 ± 3.6 g, vs. n = 8 Agt flox/flox controls 26.1 ± 2.1 g; P = 0.370 by Student's two-tailed t test) or Alb-Cre (n = 8 Alb-Cre;Agt flox/flox 24.1 ± 1.4 g, vs. n = 8 Agt flox/flox littermates 25.0 ± 1.5 g; P = 0.243 by two-tailed t test). ANOVA comparing body weight across genotypes from both breeding colonies was non-significant (F = 0.86, P = 0.47). After finding enhanced sodium appetite in Alb-Cre;Agt flox/flox mice, we used an additional cohort of these mice (n = 5) to assess their diuretic responses relative to Cre-negative littermate controls (n = 5) by injecting furosemide (50 mg/kg intraperitoneal (i.p.)) and collecting urine in individual metabolic cages. We measured urine volume, and urine sodium concentration using a flame photometer.
The unexpected results of our initial sets of experiments (deleting Agt in brain or liver, in separate breeding colonies, with separate littermate controls) led us to run an additional experiment testing whether combined deletion of Agt in both astrocytes and hepatocytes prevents sodium appetite. To test this, we deleted Agt in both the brain and the liver by breeding an additional strain of 'double-Cre, double-flox' experimental mice (GFAP-Cre;Alb-Cre;Agt flox/flox ). We tested these mice, along with littermate controls (Agt flox/flox ), in the same protocol of thirst and sodium appetite tests, across four additional cohorts as breeding proceeded. These mice did not require saline injections or any other special treatment. Their perinatal mortality was zero, there was zero mortality after experimental hypovolaemia, and their body weights were not significantly different from littermate control mice (n = 9 'double-Cre, double-flox' GFAP-Cre;Alb-Cre;Agt flox/flox 20.7 ± 3.0 g vs. n = 10 littermate control Agt flox/flox 23.8 ± 3.1 g; P = 0.0512 by two-tailed t test). These mice had grossly normal renal anatomy and histology, including a normal, tubular pattern of Agt mRNA expression in the kidney. We used additional C57BL/6J mice (male, 8 weeks of age) and mice from each of the aforementioned strains for in situ hybridization, immunofluorescence labelling, and control experiments, as described below.

Fluid intake measurement
Up to 12 mice per cohort were housed individually in BioDAQ intake monitoring cages (Research Diets, New Brunswick, NJ, USA). We randomized the proportions of experimental mice and littermate control mice in each cohort, with more than one genotype represented in every cohort. All cages were housed on a stable shelf rack (MetroMax, Wilkes-Barre, PA, USA) to minimize vibration. Each BioDAQ cage had two highly precise scales (±10 mg) continuously weighing a bottle of either water or saline. Mouse cage placement was randomized by genotype in each cohort. We did not vary placement of the water bottle (right) and 3% NaCl bottle (left) because appetite, not gustatory discrimination, was the subject of this study. We used 3% (0.5 M) NaCl, which provides optimal signal-to-noise for assessing sodium appetite; mice and rats show a baseline aversion to this concentration, typically drinking zero to less than 0.1 ml per 24-h period unless sodium appetite is provoked. At 14-day intervals, we cleaned all cages including stainless steel cage mounts, liquid hoppers, blockers and couplings. We re-calibrated all scales at every cage-change, as recommended by the manufacturer. All bottles were cleaned, extensively rinsed with distilled water and refilled between cohorts. Bottles were numbered and used in the same cage position, atop the same scale, for every cohort.
Fluid intake was recorded continuously throughout every experiment, and we specified set intervals a priori for analysis, as detailed below. Screening continuous intake-monitoring data across a total of 10 cohorts including 1472 mouse-days of two-bottle recording data (64 mice × 23-day protocol) revealed brief epochs of technical malfunctions with individual scales or bottles in four mice in our early cohorts. In these cases, we excluded fluid intake data for an individual experiment if the scale during that experiment registered one of the following error conditions: instantaneous, large fluctuations >1 g (>1 ml instantaneous volume change), 0 ml water intake for more than 24 h, or >10 ml intake in less than 24 h (more than double the highest baseline intake volumes we have recorded in mice). These error criteria resulted in the exclusion of data from individual experiments in five mice: the hyperosmotic thirst test from one Alb-Cre;Agt flox/flox mouse and one GFAP-Cre;Agt flox/+ ; the furosemide and dietary sodium deprivation tests from two Agt flox/flox mice (littermates of GFAP-Cre;Agt flox/flox ); and the water deprivation, hyperosmotic thirst and furosemide tests from one GFAP-Cre;Agt flox/flox mouse.

Experimental protocol for thirst and sodium appetite
To test thirst and sodium appetite in the BioDAQ system and to develop the experimental protocol for this study, we first ran two pilot cohorts of C57BL/6J male mice (10−12 weeks of age, 16−32 g). These pilot mice exhibited changes in water and 3% NaCl intake highly similar to previously published rodent studies and led us to design the experimental protocol shown in Fig. 1. This protocol, described below, included a series of four test paradigms evaluating hypovolaemic thirst and sodium appetite, as well as hyperosmotic thirst.
Each mouse was placed in a clean BioDAQ cage, with fresh ALPHA-dri + Plus bedding (Shepherd Specialty Papers, Framingham, MA, USA). Initially, standard rodent chow (Envigo 7013) was provided in an overhead food hopper, with ad libitum access to both dH 2 O and 3% NaCl.
After 2−3 days of habituation and daily gentle handling, all mice were water-deprived for 24 h (beginning at lights-off). This experimental paradigm provokes thirst by a combination of mild hypovolaemia and intracellular dehydration. Water deprivation was accomplished by closing the metal gates that block access to all liquid bottles. The next day, immediately before lights-off, all gates were re-opened, and we compared fluid intakes for the first 6 h of fluid access to the same 6-h time period on day two of habituation.
After 2 days of continued access to both bottles, 2 h before lights-off, all mice received an i.p. 0.5 ml injection of either hypertonic (1 M) or isotonic (0.15 M) NaCl (randomized per mouse in each cohort). After another 2 days, we injected each mouse with the opposite solution (0.15 M or 1 M NaCl i.p.). We compared fluid intakes in the 1 h immediately after 0.15 M injection (control) versus 1 M NaCl (hyperosmotic/hypernatraemic thirst). In contrast to 24-h water deprivation, hypertonic NaCl injection provokes thirst by causing pure intracellular dehydration without hypovolaemia.
Two days later, with continued access to both bottles, we switched the chow in every cage to a low-sodium diet (Teklad 170237; <0.01% Na) and provided clean cages with fresh bedding. Mice were allowed to habituate to the clean cage and low-sodium chow for 3 days. Then, at lights-off, we injected every mouse with a diuretic, furosemide (50 mg/kg i.p.; F4381, lot no. MKCB7376, Sigma-Aldrich, St Louis, MO, USA) and deprived them of sodium by closing the metal gate for just the 3% NaCl J Physiol 601.16 bottle. This experimental paradigm provokes sodium appetite and thirst by causing rapid hypovolaemia without intracellular dehydration. Twenty-four hours later, again at lights-off, we re-opened the 3% NaCl gate. For this test (furosemide sodium depletion), we compared liquid intakes during the 6-h period after opening the 3% NaCl gate to the same 6-h time period on day 2 of habituation to low-sodium chow (1 day prior to furosemide injection).
After two more days with continued access to both bottles, we closed the 3% NaCl gate at lights-off and left it closed for 6 days of dietary sodium deprivation, with continued access to water and low-sodium chow. In contrast to furosemide diuresis, this experimental paradigm gradually provokes sodium appetite by producing mild, prolonged hypovolaemia. After 6 days, we re-opened the 3% NaCl gate at lights-off. For this last test (sodium deprivation), we compared liquid intakes during the 6-h period after re-opening the 3% NaCl gate to the same 6-h period during habituation with low-sodium chow and continuous access to water and 3% NaCl. At the end of the study, all mice were euthanized by transcardial perfusion within 2−3 days, as described below.

Mineralocorticoid and angiotensin receptor blockade
Due to the unexpected increases in 3% NaCl intake in mice with Agt peripheral deletion, we ran additional cohorts of these experimental mice (Alb-Cre;Agt flox/flox , n = 8) with littermate controls (Agt flox/flox , n = 9) to evaluate the effects of two medications previously reported to reduce sodium appetite in mice and rats by blocking the mineralocorticoid receptor (spironolactone) or angiotensin 1a receptor (losartan).
First, mice were habituated to BioDAQ cages for 3 days, with ad libitum access to low-sodium chow, water and 3% NaCl. On the first experimental day, every mouse received a subcutaneous (s.c.) vehicle injection (corn oil, 0.1 ml/10 g mouse) 15−30 min before lights-off. Two days later, to control for any effects of receptor blockade alone on baseline water or saline intake, we made injections of spironolactone (10, 20 or 50 mg/kg in corn oil; product number 53 378, lot no. MKCD7812, Sigma-Aldrich) at the same time of day, 15−30 min before lights-off. Each spironolactone dose was injected on a different day, beginning with the lowest dose, with a 1-day interval between each dose. Next, to test the effect of spironolactone on sodium depletion-induced sodium appetite, 2 days later, we injected every mouse with furosemide (50 mg/kg i.p., as above) and closed all 3% NaCl gates at lights-off. On the next day, corn oil was injected 30 min before lights-off, when the 3% NaCl gates were opened. Two days later, we repeated the same sequence, but injected spironolactone (50 mg/kg s.c.). After an additional 2 days, as an additional control for furosemide depletion, we injected isotonic saline i.p. (instead of furosemide), closed the sodium gates and then reopened the 3% NaCl gates 24 h later. We excluded intake data from two experimental mice (Alb-Cre;Agt flox/flox ) and one littermate control mouse (Agt flox/flox ) with a water bottle scale recording 0 ml intake for >24 h during the baseline period.
To test the effects of losartan (an AT1R antagonist) in Alb-Cre;Agt flox/flox mice (n = 9) relative to Agt flox/flox littermate controls (n = 7), we used the large doses (20 and 100 mg/kg) that had been used to reduce saline intake in previous experiments (Crews & Rowland, 2005;Resch et al., 2017). From a 20 mg/ml stock solution of losartan potassium (61188, lot no. 048M4018V, Sigma-Aldrich) dissolved in dimethyl sulfoxide (DMSO) (D128-500, lot no. 173122, Sigma-Aldrich), we prepared separate 2 and 10 mg/ml solutions, diluted in sterile saline. First, all animals were injected with furosemide (as above) and all 3% NaCl gates were closed. On the next day, mice received a control injection (saline i.p.) and sodium gates were opened 15 min later, at lights-off. Two days later, we injected losartan 20 or 100 mg/kg, without furosemide. Each dose was injected on a different day, with a 1-day interval between them. After testing the effects of losartan alone, at each dose, we again injected every mouse with furosemide and closed all 3% NaCl gates, then injected losartan 24 h later (20 or 100 mg/kg), 15 min before opening the 3% NaCl gate, at lights-off. We excluded intake data from two experimental mice (Alb-Cre;Agt flox/flox ) and one littermate control mouse (Agt flox/flox ) with a water bottle scale recording 0 ml intake for >24 h during the baseline period (before any experimental manipulation).

Perfusions and tissue sectioning
Under deep anaesthesia with ketamine-xylazine (i.p. 150-15 mg/kg), all mice were perfused transcardially with phosphate-buffered saline (PBS) followed by 10% formalin-PBS (SF100; Fisher Scientific, Waltham, MA, USA). After perfusion, we removed each brain and fixed it overnight in 10% formalin-PBS, then submerged it overnight in 30% sucrose-PBS. Using a freezing microtome, we sectioned all brains in the axial plane at 40 µm. We collected separate, one-in-three tissue series in cryoprotectant solution and stored them at −20°C until further processing. After establishing a working 3,3 -diaminobenzidine (DAB)-peroxidase (brightfield) in situ hybridization protocol for hepatic and renal tissue, we began removing, post-fixing, slicing and cryoprotecting livers and kidneys from later cohorts of mice to label Agt.

Fluorescence in situ hybridization
To label mRNA for angiotensinogen (Agt), apolipoprotein E (Apoe), glutamine synthetase (Glul), the neuronal transcription factor Phox2b, the neuropeptide Agrp, the enzymes Ren1 and Ace, and the synaptic vesicular transporters Vmat2 (Slc18a2), VAChT (Scl18a3) and Vglut2 (Slc17a6), we used RNAscope Fluorescent Multiplex Detection Reagents (ref. no. 320851, lot no. 2002259; Advanced Cell Diagnostics, Newark, CA, USA) with probes listed in Table 2. The loxP sites that allow Cre-mediated excision in Agt flox/flox mice flank exon 2 of the Agt gene (Yiannikouris et al., 2012), so to distinguish expression of recombined versus full-length Agt transcripts, we used a custom probe-set designed to specifically target mRNA sequences from exon 2. For background cytoarchitecture, we also labelled ubiquitin C (Ubc, a highly abundant transcript) in every case. For every mouse in every experimental cohort, we labelled Agt and Ubc in 40-µm brain tissue sections through the preoptic area (POA) and through the caudal nucleus of the solitary tract (NTS). We selected the NTS and POA because they contain the highest concentrations of cells that express Agt (Stornetta et al., 1988;Zeisel et al., 2018), as well as neurons essential for thirst and sodium appetite (Allen et al., 2017;Resch et al., 2017).
The afternoon before hybridization, we removed tissue sections from cryoprotectant, rinsed them in PBS, and mounted them on glass slides to dry overnight at 4°C. In the morning, we used a Super-HT PAP pen (Research Products International, Mount Prospect, IL, USA) and vacuum grease (Dow Corning, Midland, MI, USA) to form a hydrophobic barrier around the sections and washed sections for 2 min in PBS twice at room temperature. We then covered sections with Protease IV and placed slides on a covered glass Petri dish floating in a 40°C water bath for 30 min. After two 2 min washes with PBS, we incubated sections in two or more of the probes listed in Table 2 for 2 h at 40°C. After that, we added amplification reagents Amp1-4 in series for 15−30 min each at 40°C, washing 2 × 2 min in RNAscope wash buffer (ref. 320058, lot no. 2001175; Advanced Cell Diagnostics), diluted 1:50 in ddH 2 O, between each step. After Amp4, we rinsed overnight with PBS at 4°C to reduce wrinkles in the tissue. After washing with PBS, we coverslipped all slides using Vectashield with DAPI (Vector Laboratories, Burlingame, CA, USA).

Brightfield in situ hybridization
For brightfield labelling of single transcripts with DAB, we used a C1 RNAscope probe for Ren1 (see Table 2) and the HD Reagent Kit (ref no. 322 310; Advanced Cell Diagnostics). We removed tissue sections from cryoprotectant, rinsed them in PBS, and then incubated them for 30 min in 0.3% hydrogen peroxide (no. H325-100, Fisher Scientific) to quench endogenous peroxidase activity. They were again washed in PBS and then mounted onto glass slides to dry, overnight, at room temperature. The following day, sections were dehydrated for 5 min in 100% ethanol, then dried at room temperature for 20 min. Next, slides were placed in a steamer (Oster) for 60 min followed by 5 min in 1× Target Retrieval solution (10 ml in 90 ml ddH 2 O, warmed for 5 min prior to slide incubation; Advanced Cell Diagnostics), then rinsed twice with ddH 2 O. Slides were next submerged in 100% ethanol for 1 min, then removed and dried at room temperature for 20 min. Next, we used After two final washes in 1× Wash Buffer, we mixed equal volumes of 'DAB-A' and 'DAB-B' (Advanced Cell Diagnostics), then added this combined solution to each slide for 10 min. Finally, we dipped slides in ddH 2 O, then dehydrated the sections in a series of ethanol solutions (50%, 75%, 100%) for 2 min each followed by two xylene solutions for 5 min each. After removal from xylenes, the slides were coverslipped immediately with Cytoseal (no. 8310-16 Thermo Fisher Scientific) and stored at room temperature.

Immunofluorescence
We removed brain tissue sections from cryoprotectant, rinsed them in PBS, and loaded them into a solution containing rabbit anti-Cre (1:1000; Millipore 69050-3) and chicken anti-GFAP (1:2000; Millipore AB5541). These antisera were added serially to a PBS solution containing 0.25% Triton X-100 (BP151-500, Fisher Scientific), 2% normal donkey serum (NDS, 017-000-121, Jackson ImmunoResearch) and 0.05% sodium azide (14 314, Alfa Aesar, Ward Hill, MA, USA) as a preservative (PBT-NDS-azide). We dilution-tested all primary antisera first (1:100-1:16 000) to minimize background labelling and optimize signal-to-noise for established patterns of antigen labelling. We incubated free-floating tissue sections overnight in this primary antibody solution at room temperature, on a tissue shaker. The next morning, after washing three times in PBS, we incubated sections for 2 h at room temperature in PBT-NDS-azide with a combination of species-specific, affinity-purified, donkey secondary antisera conjugated to Cy3 and Alexa Fluor 488 (Jackson ImmunoResearch; diluted 1:1000 and 1:500, respectively). Sections were then washed three times in PBS and mounted on glass slides (no. 2575-plus; Brain Research Laboratories, Waban, MA, USA), then coverslipped using Vectashield with DAPI. Slides were imaged the same day or stored in slide folders at 4°C for imaging the following week.

Immunohistochemistry
For brightfield DAB immunolabelling (immunohistochemistry), we removed tissue sections from cryoprotectant, rinsed them in PBS, then incubated them in 0.3% hydrogen peroxide (no. H325-100, Fisher Scientific) in PBT for 30 min to quench endogenous peroxidase activity. After three washes in PBS, we loaded sections into PBT-NDS-azide with rabbit anti-Cre (1:1000; Millipore 69050-3) overnight, at room temperature, on a tissue shaker. The next morning, after three washes in PBS, we incubated sections for 2 h in a PBT-NDS-azide solution containing biotinylated donkey anti-rabbit (1:500; Jackson ImmunoResearch no. 711-065-152). After three washes in PBS, sections were placed for 1 h in avidin-biotin complex (Vectastain ABC kit PK-6100; Vector Laboratories), washed three times in PBS, then incubated in DAB solution for 10 min. Our stock DAB solution was prepared by adding 100 tablets (no. D-4418, Sigma-Aldrich) into 200 ml ddH 2 O and then filtering. For a working solution, we add this DAB stock solution to PBS in a 1:6.5 ratio. After 10 min, we added 30% hydrogen peroxide (Sigma-Aldrich; 0.8 µl per 1 ml of PBS-DAB solution) and swirled sections for 1−3 min until observing dark brown (DAB) colour change. Finally, after washing sections 3× in PBS, we mounted them on glass slides. Slides were air-dried, then dehydrated in an ascending series of alcohols and xylenes and coverslipped using Cytoseal.

Tissue imaging and analysis
All slides were imaged using a 10× objective on an Olympus (Tokyo, Japan) VS120 slide-scanning microscope. In select hindbrain tissue labelled for Ren1 and other markers, we also collected confocal image stacks using a separate microscope (Leica TCS SP5 II. Deerfield, IL, USA). After collecting digital colour images of brightfield tissue labelling and digital greyscale images of all fluorescence channels, we exported images and used Adobe Photoshop for RGB pseudocolour combinations and to adjust brightness and contrast. Illustrations and figures incorporating tissue images and graphs were arranged in Adobe Illustrator.
After reviewing whole-slide images in OlyVIA or VS-ASW software (Olympus), we cropped the NTS and dorsal vagal complex (at the level of the area postrema) or the POA (beneath the midline decussation of the anterior commissure, using the anterior commissure and 3rd ventricle as guides). We set a standardized intensity for all images by using 'fixed scaling' in OlyVIA and setting the histogram at 2000-10,000 grey values for Agt, 2500-7000 for Ubc, and 200−2000 for DAPI. Cropped images were then opened in ImageJ, converted to 8-bit, and adjusted to have a black background. We used the 'analyse particles' tool in ImageJ, with object size (pixelˆ2) adjusted to 0−500 (Agt) or 0−1000 (Ubc, DAPI) for automated counts of objects (cells or nuclei). We analysed the resulting counts in Prism 7 (GraphPad Software Inc., San Diego, CA, USA), assessing the differences in cell counts between experimental mice (Cre-positive Agt flox/flox ) and littermate control mice (Cre-negative Agt flox/flox ).

Data analysis and statistics
All fluid intake was monitored continuously in the BioDAQ DataViewer software (Research Diets), which allows second-to-second data recording and retrieval. For every mouse cohort, we sorted each data file based on the experimental manipulation (furosemide, water deprivation, sodium deprivation, etc.), with a bout filter minimum set at −0.05 g and the reset period set to bin data hourly. We exported and analysed the first 6 h of liquid intake data (beginning at lights-off) for initial habituation, water deprivation, habituation to low-sodium diet, furosemide-induced sodium depletion, and dietary sodium deprivation. We also exported and analysed the first hour of data immediately after 0.15 M and 1 M saline injections. These data, exported from the BioDAQ software, were opened in Microsoft Excel and organized to calculate total intake volumes of water and 3% NaCl. We then used Prism to plot intake data for each mouse according to genotype.
To assess the statistical significance of fluid-intake data (water and 3% NaCl intake in thirst and sodium appetite experiments) and urinary volume after furosemide diuresis, we used two-way repeated measures ANOVA followed by Bonferroni post hoc correction for multiple comparisons. We used unpaired, two-tailed t tests with a significance threshold of P < 0.05 to compare Agt mRNA expression in the brain between genotypes; angiotensinogen or aldosterone in blood serum between genotypes; or urine volume after furosemide diuresis.

Identifying expression of endogenous renin-angiotensin system genes in the mouse brain
To better understand where renin-angiotensin system (RAS) genes are expressed in the mouse brain, we examined the distribution and identity of cells that express Agt, the precursor gene for all angiotensin peptides. We also examined the distribution of cells that express the gene for the enzyme renin (Ren1), which cleaves angiotensinogen to produce angiotensin I, and the gene for angiotensin converting enzyme (Ace), which cleaves this decapeptide to produce the octapeptide AngII. To identify endogenous mRNA, we performed in situ hybridization in C57BL/6J mice, rather than labelling transgenes or fluorescent reporters.
Agt mRNA was abundant in the hypothalamus and brainstem but absent from the cerebral cortex and basal ganglia. Agt mRNA was prominent in mesial structures, including a region of the POA implicated in thirst ( Fig. 2A) and a region of the nucleus of the solitary tract (NTS) implicated in sodium appetite (Fig. 2J).
Cells expressing Agt were medium-sized, with astrocytic morphology and distribution. To characterize these cells, we compared the appearance and distribution of cells expressing Agt to cells expressing the more broadly expressed gene Ubc. Cells that express Agt had light Ubc labelling and were interposed between many, larger cells that contained abundant Ubc mRNA (Fig. 2B-E and K-N). Next, we compared Agt and Apoe, which is highly expressed in astrocytes (Boyles et al., 1985;Poirier et al., 1991). All Agt-expressing cells co-expressed Apoe (Fig. 2B-E and K-N), which identified additional astrocytes lacking Agt in the cerebral cortex. We also found that all Agt-expressing cells express the enzyme glutamine synthetase (Glul, Fig. 2F-I), which is found in all astrocytes (Martinez-Hernandez et al., 1977;Norenberg & Martinez-Hernandez, 1979). In the NTS, Agt labelling was mutually exclusive with Phox2b, a transcription factor expressed in neurons ( Fig. 2O-R). These results confirm that Agt expression in the mouse brain is astrocyte-specific.
During the preparation and revision of this manuscript, Agt mRNA labelling was reported in Agrp neurons within the arcuate hypothalamic nucleus (Sapouckey et al., 2020), so we analysed both Agt and Agrp mRNA in this region, along with Ubc for cytoarchitectural background. As in the POA, we found strong Agt expression in medium-sized cells with non-neuronal morphology in hypothalamic regions outside the arcuate nucleus. Within the arcuate hypothalamic nucleus, most cells lacked Agt (Fig. 3), but on closer inspection, a minority of Agrp-expressing neurons did contain one or more fluorescent puncta, suggesting that they contain a small amount of Agt mRNA. At our initial, conservative criterion of >10 fluorescent puncta per 'positive' cell, only one Agrp-expressing neuron from one case expressed Agt. However, by reducing our threshold to three puncta per cell, 25% of Agrp-expressing neurons became 'positive' (n = 113 of the 460 Agrp-expressing neurons counted across N = 3 cases). The average number of Agt fluorescent puncta per Agrp neuron was low (1.5 putative transcripts per cell; range 0−19; Fig. 3G), particularly when compared to astrocytes bordering the arcuate nucleus (16 putative Agt transcripts per cell; range 9−33; n = 120 cells analysed, 40 each from N = 3 cases). Consistent with the low Agt transcript counts found in a minority of Agrp neurons using single-cell RNA-Seq (Sapouckey et al., 2017), these results confirm that even in this brain region, Agt is expressed predominantly in astrocytes.
Next, we labelled Ace and confirmed that the choroid plexus contained abundant labelling. Ace mRNA labelling highlighted all choroid plexus epithelial cells in the lateral ventricles, dorsal third ventricle, fourth ventricle and foramina of Luschka (Fig. 4). Circumventricular organs, including the subfornical organ, were devoid of labelling for Ace and Agt mRNA (Fig. 4B). Consistent with publicly available in situ hybridization results in the Allen Mouse Brain Atlas (Lein et al., 2007), we found labelling in other brain regions (including the supramammillary nucleus, cranial motor neurons and dorsolateral striatum) and dense Ace mRNA only in the choroid plexus.
Finally, we examined Ren1. On our initial screens, we found no evidence of Ren1 expression in the mouse brain, challenging the plausibility of angiotensin production within the brain. To increase sensitivity and eliminate the confound of background fluorescence that is inherent to FISH, we switched to a DAB-peroxidase method. After first confirming intense Ren1 mRNA labelling in the juxtaglomerular apparatus of the kidney (not shown), we used this method to label brain sections containing the POA and NTS. Once again, we did not find any Ren1 mRNA labelling in the POA, NTS, subfornical organ or any other part of the brain parenchyma at these levels (n = 3 mice; Fig. 5D and F). However, DAB-peroxidase labelling did identify sparse puncta in some epithelial cells within the choroid plexus ( Fig. 5A and C). To broaden our search, we labelled additional series of tissue sections covering every region of the forebrain, diencephalon, midbrain, cerebellum and hindbrain (n = 5 mice). In this more extensive screen, we identified a small cluster of cells in the brainstem with Ren1 mRNA labelling in the compact formation of the nucleus ambiguus (Fig. 5E1). Other than these, we did not find any evidence of Ren1 expression in the brain parenchyma. To better characterize the Ren1-expressing cells in the nucleus ambiguus, we switched back to FISH labelling for Ren1, along with Slc18a3 (vesicular acetylcholine transporter), Slc18a2 (vesicular monoamine transporter) and Slc17a6 (vesicular glutamate transporter). Ren1-expressing cells did not express the glutamatergic or monoaminergic markers, but did express the cholinergic marker gene Slc18a3, identifying them as vagal motor neurons (n = 3; Fig. 5G-K).
In summary, the three genes required to produce AngII are expressed in the mouse brain, in different patterns. Astrocytes in the hypothalamus and brainstem express Agt, epithelial cells in the choroid plexus abundantly express Ace (and somsparsely express Ren1), and a small number of vagal motor neurons in the nucleus ambiguus compact formation express Ren1.

Assessing tissue-specific deletion of Agt from astrocytes
Because astrocytes are the predominant source of brain angiotensinogen, we used a mouse strain with Cre expression driven by the promoter of a gene expressed in astrocytes, human glial fibrillary acidic protein (GFAP). In wild-type mice, in situ hybridization followed by immunofluorescence labelling revealed fibrous GFAP immunoreactivity surrounding Agt mRNA in many astrocytes ( Fig. 6A-C). And in GFAP-Cre mice, Cre immunolabelling revealed a glial pattern of nuclear immunoreactivity throughout the brain (Fig. 6D), surrounded by fibrous, extra-nuclear GFAP in many astrocytes ( Fig. 6F-N).
Breeding GFAP-Cre mice with 'floxed' Agt mice (Agt flox/flox ) eliminated virtually all Agt mRNA labelling in the brain (Fig. 7A and B), without altering expression in the liver (Fig. 8C) and without reducing the amount of angiotensinogen in the blood plasma ( Fig. 7C; P = 0.465 by two-tailed t test between n = 8 GFAP-Cre;Agt flox/flox experimental mice versus n = 10 littermate control mice). Every Cre-negative littermate (Agt flox/flox , n = 8) and every heterozygous littermate ('single-floxed' GFAP-Cre;Agt flox/+ , n = 2) had a normal number and distribution of Agt-expressing cells in the brain. In contrast, every GFAP-Cre;Agt flox/flox mouse (n = 13) had total or near-total elimination of Agt-expressing cells in brain regions implicated in thirst and sodium appetite (POA and NTS; Fig. 7A and B). The few, remaining cells with Agt mRNA had a distribution, appearance and per-cell hybridization signal identical to astrocytes in control mice, suggesting that Cre-mediated excision had not occurred in at least one floxed allele in these cells.
Thus, GFAP-Cre;Agt flox/flox mice have a near-total deletion of Agt in the brain, without a reduction in the number of cells that normally express it. Liver Agt expression is preserved and circulating levels of angiotensinogen are unchanged. We used these mice to test whether central angiotensin production is required for thirst and sodium appetite.

Assessing tissue-specific Agt deletion from hepatocytes
To reduce circulating angiotensinogen, we used a mouse strain with Cre expression driven by the promotor for rat albumin, which is expressed in hepatocytes. There were no Cre-immunoreactive cells in the brains of Alb-Cre  (Fig. 6E). Breeding these mice with Agt flox/flox mice did not alter the appearance of Agt mRNA labelling in the brain. Every Alb-Cre;Agt flox/flox mouse (n = 8) and every littermate control mouse (Agt flox/flox , n = 8 and Alb-Cre;Agt flox/+ , n = 3) had a normal number and distribution of Agt-expressing cells in both the POA and NTS (Fig. 7D and E). These mice also had normal-appearing kidneys, a normal pattern of renal Agt expression (Fig. 9A), and robust diuretic responses to furosemide (Fig. 9C).
In summary, Alb-Cre;Agt flox/flox mice have near-total deletion of Agt in the liver, with a correspondingly large deficit in circulating angiotensinogen, despite preserved Agt expression in the brain. We used these mice to test Slc18a3 ( whether hepatic angiotensin production is required for thirst and sodium appetite.

Thirst after tissue-specific deletion of Agt
Deleting Agt from astrocytes did not reduce thirst. First, after 24-h water deprivation, GFAP-Cre;Agt flox/flox mice (n = 13) and littermate control mice (n = 12) exhibited similar increases in water intake ( Fig. 10A; P < 0.0001 for both by two-way, repeated measures ANOVA followed by Bonferroni correction for multiple comparisons). There were no differences between strains at baseline (P = 0.999) or after water deprivation (P = 0.958). Control mice consumed slightly more concentrated (3%) NaCl after water deprivation ( Fig. 10B; P = 0.002 relative to their baseline 3% NaCl intake). Between strains, there were no differences in 3% NaCl intake at baseline (P = 0.929) or after water deprivation (P = 0.262), and GFAP-Cre;Agt flox/flox exhibited no difference between baseline and water deprivation (P = 0.958). Second, after hypertonic saline injection, GFAP-Cre;Agt flox/flox mice (n = 12) and littermate control mice (n = 12) again exhibited similar increases in water intake ( Fig. 10C; both P < 0.0001 relative to isotonic saline injection). There were no differences in water intake between strains in the control condition with isotonic saline injection (P = 0.999) or in the experimental condition with hypertonic saline injection (P = 0.138). Some mice in each group exhibited a slight increase in 3% NaCl intake, which was statistically significant only in mice with astrocytic Agt deletion ( Fig. 10D; P = 0.024 GFAP-Cre;Agt flox/flox ; P = 0.600 in littermate controls). Between strains, there were no differences in 3% NaCl intake in the control condition with isotonic saline injection (P = 0.999) or in the experimental condition with hypertonic saline injection (P = 0.072). Deleting Agt from the liver did not reduce thirst. First, both Alb-Cre;Agt flox/flox mice (n = 9) and littermate control mice (n = 11) drank significantly more water after 24-h water deprivation (both P < 0.0001; Fig. 11A). In fact, Alb-Cre;Agt flox/flox mice drank slightly more water than controls, both at baseline (P = 0.023) and after water deprivation (P = 0.008). Alb-Cre;Agt flox/flox mice also drank slightly more 3% NaCl at baseline ( Fig. 11B; P < 0.0001) and after water deprivation (P < 0.0001), with a small but significant increase in this strain (P = 0.013) but not in littermate control mice (P = 0.999). Second, after hypertonic saline injection, Alb-Cre;Agt flox/flox mice (n = 7) and littermate control mice (n = 11) exhibited similar increases in water intake ( Fig. 11C; both P < 0.0001 relative to isotonic saline injection), with no differences between strains in the  Counts and ratios are shown as means ± standard deviation. The proportion of Agt-expressing to total (DAPI-labelled) cells was compared between conditional knockout and littermate control mice using a two-tailed t test. Abbreviations: NTS, nucleus of the solitary tract; POA, preoptic area.

Table 3. Agt-expressing cells and total cells counted in brain regions that are implicated in thirst and sodium appetite (preoptic area and nucleus of the solitary tract)
isotonic control condition (P = 0.999) or hypertonic experimental condition (P = 0.640). Neither strain drank more 3% NaCl after hypertonic saline injection ( Fig. 11D; P = 0.374 Alb-Cre;Agt flox/flox ; 0.999 littermate controls), and there were no differences between strains in the isotonic control condition (P = 0.376) or in the hypertonic experimental condition (P = 0.999).

Effects of mineralocorticoid and angiotensin receptor blockade on sodium appetite in angiotensin-deficient mice
Enhanced sodium appetite in mice with liver Agt deletion was unexpected, so we considered the possibility that aldosterone was responsible for the augmented sodium appetite of liver-Agt-deleted mice. In these mice (n = 8), plasma aldosterone was not significantly different relative to littermate controls (25.5 ± 11.4 ng/dl Alb-Cre;Agt flox/flox , n = 8 versus 25.8 ± 16.7 ng/dl in Agt flox/flox , n = 8; P = 0.960 by two-tailed t test). The natriorexigenic effect of aldosterone can be reduced by blocking the mineralocorticoid receptor (Koneru et al., 2014;Qiao et al., 2016;Sakai et al., 1986), so we first tested the possibility that enhanced sodium appetite in Alb-Cre;Agt flox/flox mice requires mineralocorticoid receptor activation. We treated additional cohorts of Alb-Cre;Agt flox/flox mice (n = 7) and littermate controls (Agt flox/flox , n = 7) with spironolactone, a potent mineralocorticoid receptor antagonist. Sodium depletion with furosemide again increased 3% NaCl and water intake in both genotypes ( Fig. 14A and B; P < 0.0001 in Alb-Cre;Agt flox/flox mice; P = 0.0175 in littermate control mice; two-way repeated measures ANOVA followed by Bonferroni correction for multiple comparisons). At baseline, relative to littermate controls, Alb-Cre;Agt flox/flox mice drank more 3% NaCl (P = 0.0052) but not water (P = 0.193). After sodium depletion, Alb-Cre;Agt flox/flox mice again drank substantially more 3% NaCl and water than littermate controls (P < 0.0001). In littermate control mice, spironolactone (50 mg/kg i.p.) had no effect relative to vehicle pretreatment (3% NaCl intake P = 0.999; water intake P = 0.999), and intakes of both 3% NaCl (P < 0.0001) and water (P = 0.0177) remained different from baseline. In Alb-Cre;Agt flox/flox mice, spironolactone reduced both 3% NaCl (P = 0.0002) and water intake (P = 0.0004) relative to vehicle, without eliminating their elevated intakes above baseline (P < 0.0001 3% NaCl; P < 0.0001 water), and 3% NaCl intake reduced to the baseline range in just one mouse. Spironolactone-treated Alb-Cre;Agt flox/flox mice again drank more 3% NaCl (P = 0.0015) and water (P < 0.0001) than spironolactone-treated littermate control mice. These results are consistent with a role for mineralocorticoid signalling, while also indicating that ongoing mineralocorticoid receptor activity is not required for the enhanced sodium appetite of Alb-Cre;Agt flox/flox mice.
Next, because Alb-Cre;Agt flox/flox mice are deficient in circulating angiotensin peptides (Lu, Wu et al., 2016;Yiannikouris et al., 2015), we used this strain to test the specificity of losartan, an angiotensin receptor blocker, at the large doses used in previous studies reporting that Figure 10. Thirst after brain Agt deletion A and B, water deprivation for 24 h increased water intake similarly in GFAP-Cre;Agt flox/flox mice (n = 13, P < 0.0001) and littermate control mice (n = 12, P < 0.0001). Saline intake was low in both groups, with a small increase only in the control group (P = 0.002), and there were no between-group differences in water or saline intake volumes at baseline or after water-deprivation. C and D, injecting hypertonic saline (intraperitoneal 1 M NaCl) caused an acute increase in water intake in both strains (each P < 0.0001 vs. isotonic saline injection). Saline intake was again low in both groups, with a small increase only in the experimental group (P = 0.024). All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. [Colour figure can be viewed at wileyonlinelibrary.com] J Physiol 601.16 angiotensin signalling is necessary for sodium appetite. To test whether these large doses reduce sodium appetite by blocking angiotensin signalling or via off-target behavioural effects, we used mice with a 90% deficit in circulating angiotensinogen (Alb-Cre;Agt flox/flox ; Fig. 7F), meaning that any drug effect related to angiotensin binding to angiotensin receptors should be less pronounced in these mice than in littermate controls that have normal circulating angiotensin levels. Following sodium depletion with furosemide, we tested the effects of 20 mg/kg and 100 mg/kg losartan (i.p.) pretreatment in Alb-Cre;Agt flox/flox mice (n = 6) and littermate control mice (Agt flox/flox , n = 7). In this experiment, baseline 3% NaCl (P = 0.344) and water (P = 0.999) intakes were not significantly different between genotypes. Without losartan pretreatment, sodium depletion again increased 3% NaCl intake (Fig. 14C,D; P = 0.0003 Alb-Cre;Agt flox/flox ; P = 0.0014 littermate controls), with Alb-Cre;Agt flox/flox mice again drinking more than littermate control mice (P < 0.0001). Sodium depletion increased water intake in Alb-Cre;Agt flox/flox mice (P = 0.0100) but not littermate controls (P = 0.999), and Alb-Cre;Agt flox/flox mice drank more water than littermate controls (P = 0.0001). In sodium-deplete littermate control mice, losartan 20 mg/kg did not reduce 3% NaCl intake (P = 0.999) or water intake (P = 0.999). In sodium-deplete Alb-Cre;Agt flox/flox mice, losartan 20 mg/kg caused adipsia in one mouse without consistently reducing water intake in this group (P = 0.126), and variably reduced 3% NaCl intake (P = 0.0318). Losartan 100 mg/kg had a more pronounced effect, causing adipsia (zero fluid intake) in all but one mouse from each group. Mice normally drink 1−2 ml of water during this test period (6 h after lights-off), so adipsia is inconsistent with the possibility that at this dose losartan simply prevented an increase in thirst or sodium appetite above baseline. In summary, rather than a reduced effect in angiotensin-deficient mice (relative to controls with normal circulating angiotensinogen), both doses produced larger effects in angiotensin-deficient mice. This result argues against the likelihood that effects of losartan at these large doses involve angiotensin-receptor binding.

Combined Agt deletion from brain and liver
We also considered the possibility that angiotensinogen produced in the brain compensates for the peripheral angiotensin deficit of Alb-Cre;Agt flox/flox mice. There is A C D B Figure 11. Thirst after liver Agt deletion A and B, water deprivation for 24 h increased water intake in both Alb-Cre;Agt flox/flox mice (n = 8, P < 0.0001) and littermate control mice (n = 11, P < 0.0001). Alb-Cre;Agt flox/flox mice drank more water than controls at baseline (P = 0.023) and after water deprivation (P = 0.008). Alb-Cre;Agt flox/flox mice also drank more saline at baseline (P < 0.0001) and after water deprivation (P < 0.0001), and they drank slightly more saline after water deprivation relative to baseline (P = 0.013). C and D, injecting hypertonic saline (intraperitoneal 1 M NaCl) increased water intake similarly, in both strains (P < 0.0001, n = 7 Alb-Cre;Agt flox/flox ; P < 0.0001, n = 11 littermate controls), with no difference in saline intake. All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. [Colour figure can be viewed at wileyonlinelibrary.com] no Agt expression in mouse brain regions that are accessible to circulating renin, such as the choroid plexus or circumventricular organs, but it is possible that astrocytes secrete angiotensinogen through distal processes into one of these blood-brain-barrier-deficient sites. Having excluded the necessity of angiotensinogen production by astrocytes in mice with intact peripheral angiotensin (GFAP-Cre;Agt flox/flox ), we next bred and tested 'double-deletion' mice with Agt deletion in both astrocytes and hepatocytes (GFAP-Cre;Alb-Cre;Agt flox/flox , n = 9). We tested thirst and sodium appetite in these mice and compared their responses with littermate control mice (Agt flox/flox , n = 10). First, we tested whether double-deletion of Agt from astrocytes and hepatocytes prevents thirst. Baseline intakes of water ( Fig. 15A; P = 0.999) and 3% NaCl ( Fig. 15B; P = 0.999) were not different between GFAP-Cre;Alb-Cre;Agt flox/flox mice and Agt flox/flox littermate control mice. Water deprivation for 24 h increased water intake (P < 0.0001 in both groups) without increasing 3% NaCl intake (P = 0.401 GFAP-Cre;Alb-Cre;Agt flox/flox ; P = 0.0788 littermate controls). Between genotypes, there was no difference in water (P = 0.999) or 3% NaCl intake (P = 0.476) after water deprivation. Next, we tested hyperosmotic thirst ( Fig. 15C and D). In the isotonic control condition, water and 3% NaCl intakes were not different between genotypes (P = 0.999 water; P = 0.999 3% NaCl). In both groups, hypertonic saline injection increased water (both P < 0.0001) but not 3% NaCl intake (P = 0.648 GFAP-Cre;Alb-Cre;Agt flox/flox ; P = 0.999 littermate controls). Beyond a small difference in water intake in the hypertonic condition (P = 0.0004), these results across two different thirst assays are inconsistent with a necessary role in thirst for angiotensinogen production in astrocytes and hepatocytes.
Finally, we tested whether double-deletion of Agt from astrocytes and hepatocytes prevents sodium appetite. Sodium depletion with furosemide increased both water and 3% NaCl intake in GFAP-Cre;Alb-Cre;Agt flox/flox mice ( Fig. 16A and B; P < 0.0001 water; P = 0.0117 3% NaCl) and in littermate control mice (P = 0.0290 water; P = 0.0172 3% NaCl). Also, GFAP-Cre;Alb-Cre;Agt flox/flox mice drank more water (P < 0.0001), but not 3% NaCl (P = 0.999), than littermate control mice. Water and 3% NaCl intakes were not different between genotypes A C D B Figure 12. Sodium appetite after brain Agt deletion A and B, sodium depletion (furosemide diuresis followed by 24-h low-sodium diet) increased the intake of both 3% saline and water similarly in GFAP-Cre;Agt flox/flox mice (n = 12, P < 0.0001) and in littermate control mice (n = 11, P < 0.0001). C and D, dietary sodium deprivation for 6 days produced a variable increase in 3% NaCl intake, without a change water intake. The increase in 3% NaCl after dietary sodium deprivation was statistically significant only in GFAP-Cre;Agt flox/flox mice (P = 0.0003), not in littermate control mice (P = 0.485). There were no between-group differences in water or saline intake. All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. [Colour figure can be viewed at wileyonlinelibrary.com] J Physiol 601.16 at baseline (P = 0.274 water; P = 0.925 3% NaCl). Additionally, dietary sodium deprivation for 6 days increased 3% NaCl intake (P = 0.0002) but not water intake (P = 0.0515) in GFAP-Cre;Alb-Cre;Agt flox/flox mice ( Fig. 16C and D). Littermate control mice did not drink significantly more water (P = 0.369) or 3% NaCl (P = 0.102) after dietary sodium deprivation. Between genotypes, there was a small difference in water intake after sodium deprivation (P = 0.0369) but not at baseline (P = 0.328), and there were no differences in 3% NaCl intake (baseline P = 0.547; after sodium deprivation P = 0.6745). In summary, without consuming the greatly elevated volumes of 3% NaCl and water that we found in Alb-Cre;Agt flox/flox (hepatocyte Agt deletion) mice, GFAP-Cre;Alb-Cre;Agt flox/flox (Agt double-deletion) mice exhibited robust thirst and sodium appetite responses.

Discussion
This study improves our understanding of hypovolaemic thirst and sodium appetite. Based on these new findings, we rejected our hypothesis that angiotensinogen production in astrocytes is necessary for sodium appetite because deleting Agt in astrocytes and deleting Agt in both astrocytes and hepatocytes did not reduce thirst or sodium appetite relative to littermate control mice with intact angiotensinogen production. Further, deleting Agt in hepatocytes enhanced sodium appetite, and the enhanced salt and water intakes of these mice with circulating angiotensinogen deficiency did not depend on mineralocorticoid receptor activation, suggesting that sodium appetite does not require angiotensin-aldosterone synergy. After discussing limitations of our approach, we highlight additional implications for our understanding of sodium appetite and the brain renin-angiotensin system (bRAS).

Limitations
We used transgenic mice with a partial copy of a Cre-driver gene (human GFAP or rat Alb) inserted outside its normal chromosomal locus to promote Cre transcription. Like knockin-Cre mice, transgenic Cre-drivers can produce tissue-specific recombination, Figure 13. Sodium appetite after liver Agt deletion A and B, sodium depletion (furosemide diuresis followed by 24-h low-sodium diet) greatly increased sodium appetite in Alb-Cre;Agt flox/flox mice (n = 8; 3% NaCl and water intake both P < 0.0001 vs. baseline). Littermate control mice (n = 11) also increased their saline (P = 0.0003) and water intake (P < 0.0001), but Alb-Cre;Agt flox/flox mice drank more of both at baseline (saline P = 0.0006; water P = 0.048) and after sodium depletion (saline and water both P < 0.0001). C and D, in Alb-Cre;Agt flox/flox mice, dietary sodium deprivation for 6 days greatly increased 3% saline intake (P < 0.0001) and slightly increased water intake (P = 0.0019). Littermate control mice had a small increase in water intake (P = 0.012) without a significant increase in saline intake (P = 0.199). Alb-Cre;Agt flox/flox mice again drank more saline and water than littermate control mice at baseline (saline P = 0.0015; water P = 0.0022) and after dietary sodium deprivation (saline and water both P < 0.0001). All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. [Colour figure can be viewed at wileyonlinelibrary.com] Figure 14. Sodium appetite after liver Agt deletion plus spironolactone or losartan A and B, sodium depletion with furosemide followed by 24-h low-sodium diet again increased water and 3% NaCl intake to a greater extent in Alb-Cre;Agt flox/flox mice than in littermate control mice (water and 3% NaCl both P < 0.0001 in n = 7 Alb-Cre;Agt flox/flox versus n = 7 Agt flox/flox ). Pre-treatment with spironolactone (50 mg/kg) lessened the elevated intakes of 3% NaCl (P = 0.0002) and water (P = 0.0004) in Alb-Cre;Agt flox/flox experimental mice, without any effect in littermate controls (P = 0.999 water and P = 0.999 3% NaCl). At baseline, in a euvolaemic control condition (saline I.P. injection, rather than furosemide), Alb-Cre;Agt flox/flox mice drank more 3% NaCl, but not more water, than littermate control mice (P = 0.0052 3% NaCl and P = 0.193 water versus n = 7 Agt flox/flox littermate control mice). C and D, in an additional set of experiments, sodium depletion again increased 3% NaCl (P < 0.0001) and water intake (P = 0.0001) to a greater extent in Alb-Cre;Agt flox/flox experimental mice (n = 6) than in littermate control mice (Agt flox/flox , n = 7). Pre-treating Agt flox/flox control mice with 20 mg/kg losartan did not reduce the water (P = 0.999) or 3% NaCl intake (P = 0.999) caused by sodium depletion. Pre-treating Alb-Cre;Agt flox/flox experimental mice with 20 mg/kg losartan variably reduced 3% NaCl intake (P = 0.0318) and did not have a consistent effect on water intake (P = 0.258). At 100 mg/kg, losartan caused adipsia (eliminated all fluid intake) in all but one mouse from each group. Relative to sodium depletion plus saline pre-treatment, 100 mg/kg losartan had a significant effect on water (P = 0.0004) and 3% NaCl intake (P = 0.0060) in Alb-Cre;Agt flox/flox experimental mice, while in littermate controls its effects on water (P = 0.0527) and 3% NaCl intake (P = 0.0915) were non-significant despite adipsia in all but one mouse in this group. All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. For each condition, each grey line represents data from but caution is warranted because some transgenic strains express Cre ectopically, or fail to express Cre in locations with endogenous expression of the Cre-driver gene (Harno et al., 2013;Schmidt-Supprian & Rajewsky, 2007;Wicksteed et al., 2010;Zhang et al., 2013). To verify the specificity of recombination for our primary variables of interest -brain Agt expression and circulating angiotensinogen protein -we labelled mRNA in the brain and measured protein in the blood and found clear evidence of specific and complementary changes in each. Additionally, we confirmed a loss of Agt mRNA labelling in the liver in Alb-Cre;Agt flox/flox mice, plus normal kidney morphology and intact renal expression of Agt across strains. While we cannot exclude the possibility of Cre-mediated Agt deletion in tissues we did not examine, we are not the first to use Alb-Cre;Agt flox/flox mice, and previous investigators who surveyed additional tissues did not find evidence of Agt deletion outside the liver (Lu, Wu et al., 2016). Even if Cre-mediated deletion did occur outside the brain or liver, it did not affect the two primary variables relevant to this study -brain Agt mRNA and circulating angiotensinogen protein -nor did it alter the normal health, body weight, behaviour or fertility of conditional deletion mice in any of our breeding colonies, relative to their Cre-negative littermates. This point is important because it sharply contrasts the limitations of studying Agt −− mice, which have high perinatal mortality, severe renal pathology and 50% mortality after volume depletion (Alexiou et al., 2005;Kim et al., 1995;McKinley et al., 2008;Okubo et al., 1997;Taniguchi et al., 1998;Tanimoto et al., 1994).
We are not aware of any previous reports using GFAP-Cre;Agt flox/flox mice. It is important to note that GFAP-Cre can cause germline recombination (Zhang et al., 2013), and it is also possible that, in addition to eliminating Agt expression from astrocytes, ectopic Cre expression may have deleted Agt in additional cells, outside the brain. We also do not know whether eliminating Agt expression from astrocytes impacts blood pressure or the function of the RAS outside the brain. However, these mice had intact levels of circulating angiotensinogen protein, as well as intact renal and Figure 15. Thirst after Agt double-deletion from brain and liver A and B, water deprivation for 24 h increased water intake in mice with combined Agt deletion in astrocytes and hepatocytes (Alb-Cre;GFAP-Cre;Agt flox/flox , n = 9) and in littermate control mice (Agt flox/flox , n = 10). Between these two groups, there were no differences in baseline water intake (P = 0.381) or in water intake after water deprivation (P = 0.999), and water deprivation did not increase 3% NaCl intake in either group (P = 0.401 Agt flox/flox ; P = 0.0788 Alb-Cre;GFAP-Cre;Agt flox/flox ). C and D, injecting hypertonic saline (intraperitoneal 1 M NaCl) increased water intake in both strains (both P < 0.0001) and did not increase 3% NaCl intake in either strain (P = 0.999 Agt flox/flox ; P = 0.648 Alb-Cre;GFAP-Cre;Agt flox/flox ). Between these two groups, water intake was slightly higher in the control group (1.2 ± 0.3 ml Agt flox/flox ; 0.9 ± 0.3 ml Alb-Cre;GFAP-Cre;Agt flox/flox P = 0.0004). All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. [Colour figure can be viewed at wileyonlinelibrary.com] hepatic Agt expression and normal ingestive behaviours, so even if Agt deletion did occur in tissues outside the brain, that would strengthen, rather than weaken our conclusion that this gene is not necessary for thirst or sodium appetite.
In the brain, we cannot exclude the possibility that Agt expression rises dynamically under conditions that were not captured in our experiments. Previous claims that neurons produce angiotensinogen relied on transgenic mice that overexpressed a copy of the human AGT gene (Agassandian et al., 2017;Sherrod et al., 2005;Yang et al., 1994) rather than the endogenous mouse Agt gene, or focused on sparse Agt expression in a minority of neurons in the arcuate hypothalamic nucleus, as detailed above ('Identifying expression of endogenous renin-angiotensin system genes in the mouse brain'). Our findings match the results of conventional in situ hybridization in rats (Stornetta et al., 1988) and single-cell RNA-Seq in mice (La Manno et al., 2021;Zeisel et al., 2018), confirming that the vast majority of endogenous Agt mRNA in the mouse brain localizes to a subset of astrocytes. Complementing and supporting this observation, we found that an astrocytic Cre-driver eliminated virtually all Agt mRNA from the brain (GFAP-Cre;Agt flox/flox mice). The sparse, remaining Agt expression in these mice localized to cells with the same astrocytic morphology as in control mice. We cannot exclude the possibility that low-level expression of Agt in some other, unidentified cell type plays a role in thirst or sodium appetite, but our findings indicate that eliminating the vast majority of Agt expression from the brain has no effect on these behaviours.
Despite successfully deleting both copies from nearly every astrocyte in GFAP-Cre;Agt flox/flox mice, 1−3% of cells with this morphology continued expressing Agt. Continued expression in these few astrocytes may reflect the less-than-perfect efficiency of Cre-mediated deletion (Schmidt-Supprian & Rajewsky, 2007), where failure of Cre to recombine one or both alleles in a cell would leave that cell producing a normal or half-normal amount of Agt mRNA. Also, GFAP expression varies among astrocytes and increases with age (Nichols et al., 1993), so the few remaining cells with Agt expression simply may not have expressed the Cre-driver yet.
Likewise, Alb-Cre-mediated deletion in the liver occurs progressively, across postnatal weeks 3−6 (Postic & Figure 16. Sodium appetite after Agt double-deletion from brain and liver A and B, sodium depletion (furosemide followed by 24-h low-sodium diet) increased water intake of mice with combined Agt deletion in astrocytes and hepatocytes (P < 0.0001 Alb-Cre;GFAP-Cre;Agt flox/flox , n = 9) and in littermate control mice (P = 0.0290 Agt flox/flox , n = 10). The experimental strain consumed more water than littermate controls (P < 0.0001). Sodium depletion also increased 3% NaCl intake in each strain (P = 0.0172 Agt flox/flox ; P = 0.0117 Alb-Cre;GFAP-Cre;Agt flox/flox ). Saline intake volumes were more variable in this experiment, with no significant difference between strains after sodium depletion (P = 0.999). C and D, separately, dietary sodium deprivation for 6 days increased 3% NaCl intake (P = 0.0002) but not water intake (P = 0.0515) in Alb-Cre;GFAP-Cre;Agt flox/flox mice. Between genotypes, there was a small difference in water intake after sodium deprivation (P = 0.0369). All statistical comparisons above were made using two-way repeated measures ANOVA with Bonferroni correction for multiple comparisons. [Colour figure can be viewed at wileyonlinelibrary.com] J Physiol 601.16 Magnuson, 2000), and less than 100% recombination could explain the small amount of angiotensinogen protein remaining in circulation in Alb-Cre;Agt flox/flox mice. Alternatively, this may reflect a contribution of other tissues to the circulating pool of angiotensinogen (Yiannikouris et al., 2012). Expression of Agt in the liver is an order of magnitude higher than in the brain and other organs combined (Campbell & Habener, 1986), and investigators using a different floxed-Agt strain reported that Alb-Cre fully eliminated circulating angiotensinogen (Matsusaka et al., 2012). Our 90% reduction falls between this and two other reports of sub-total reductions in circulating angiotensinogen after deleting Agt from the liver (Lu, Wu et al., 2016;Yiannikouris et al., 2015). The differences among studies may reflect variable Agt expression in adipocytes or other extrahepatic tissues (Frederich et al., 1992;Yiannikouris et al., 2012).

Implications for hypovolaemic thirst and sodium appetite
In designing experiments to test whether centrally or peripherally generated angiotensin is necessary, we had not anticipated increases in salt or water intake. In retrospect, however, knowing that the physiological mechanisms driving hypovolaemic thirst and sodium appetite may function without angiotensin, it is not surprising that water and saline intake are augmented in angiotensin-deficient mice because the ethologically relevant stimulus for sodium appetite is low blood volume (Geerling & Loewy, 2008). Angiotensin deficiency lowers blood pressure (Lu, Wu et al., 2016;Olearczyk et al., 2014), so one possible explanation for our findings is that the low blood pressure of angiotensin-deficient mice augmented non-angiotensinergic mechanisms that increase sodium appetite. We favour a similarly straightforward interpretation of earlier evidence that ACE inhibitors increased salt and water intake (Evered & Robinson, 1983;Fregly, 1980;Lehr et al., 1973;Moe et al., 1984;Stricker, 1983). That is, thirst and sodium appetite are appropriate compensatory responses to low blood pressure, which occurs when angiotensin production is blocked.
Indeed, mice with a circulating angiotensin deficiency (Alb-Cre;Agt flox/flox ) consumed more 3% NaCl than littermate control mice. A slight difference was apparent in some mice even at baseline, when we did not provoke sodium appetite. This suggests that blood volume or blood pressure in this strain was low enough to provoke a slight elevation in sodium appetite, even without a hypovolaemic provocation. However, double-deletion mice did not exhibit this baseline increase. The reason for this difference remains unclear, particularly given the lack of any change in sodium appetite after deleting Agt from astrocytes (GFAP-Cre;Agt flox/flox ), but this discrepancy leaves open the possibility that astrocytic Agt production could play a role in the enhanced thirst and sodium appetite of mice with an isolated deficiency of circulating angiotensin (Alb-Cre;Agt flox/flox ). Comparing blood volume and arterial pressure between mice with brain and liver Agt deletion may provide useful information in the future. Overall, however, the lack of a significant decrease in the saline intake of double-deletion mice, relative to littermate control mice, is inconsistent with the hypothesis that angiotensinogen production in either astrocytes or hepatocytes is required for sodium appetite.
The angiotensin-aldosterone 'synergy hypothesis' posits that sodium appetite results from the combined effects of AngII and aldosterone (Fluharty & Epstein, 1983;Sakai et al., 1986;Zhang et al., 1984), and mice with conditional deletion of Agt from the liver or even total body knockout (Agt −/− ) continue producing aldosterone Okubo et al., 1997;Umemura et al., 1998;Yiannikouris et al., 2015). Much like angiotensin-deficient mice, however, aldosterone-deficient rodents and human patients compensate by increasing sodium consumption (Kochli et al., 2005;Richter, 1936;Wilkins & Richter, 1940), and the sodium appetite of aldosterone-deficient rats is further enhanced by switching to a low-sodium diet (Jalowiec & Stricker, 1973). None of this contradicts evidence that AngII and aldosterone can help promote thirst and sodium appetite (Fitzsimons, 1998;Gasparini et al., 2018;Rice & Richter, 1943;Watson, 1986), but these behavioural roles may be less critical than their roles in adjusting renal and cardiovascular function (Cardoso et al., 2022). If neither AngII nor aldosterone is individually necessary for thirst or sodium appetite, some other signalling mechanism must activate sodium appetite to help compensate for their absence.
While it was clear nearly a century ago that adrenal steroids are not necessary for sodium appetite (Richter, 1936), a previous lack of clarity regarding the necessity of angiotensin peptides reflected the technical difficulty of eliminating them. Early methods of eliminating angiotensin production had significant limitations. Whereas experimental animals could survive for weeks or months following adrenalectomy (to eliminate aldosterone), removing the liver (to reduce circulating angiotensinogen) or kidneys (to eliminate renin) did not allow survival experiments, and each procedure altered physiology and behaviour in ways unrelated to the RAS. Similarly, a major challenge with experiments involving Agt-knockout mice was their developmental renal pathology McKinley et al., 2008;Okubo et al., 1997). Separately, pharmacological tools that reduce production of AngII (captopril and other ACE inhibitors) or reduce angiotensin receptor binding (losartan and other angiotensin receptor blockers) produced conflicting results regarding the importance of AngII. The resulting controversy generated new ideas, including a hypothesis that ACE inhibitors increase thirst and sodium appetite by increasing AngI, which then crosses the blood-brain barrier for conversion to AngII in the cerebrospinal fluid (CSF) or brain parenchyma (Elfont et al., 1984;Epstein, 1982;Fitzsimons, 1998;Johnson & Thunhorst, 1997;McKinley et al., 2003;Moe et al., 1984;Sakai et al., 1986;Sakai et al., 1990;Schiffrin & Genest, 1982;Thunhorst et al., 1989;Weisinger et al., 1996;Weiss et al., 1986;Zhang et al., 1984). Also in rodents, doses of 1−10 mg/kg losartan blocked the pressor and dipsogenic effects of AngII, with an EC 50 of roughly 5 mg/kg (Fregly & Rowland, 1991;Wong et al., 1990), but reducing sodium appetite required larger doses that exert off-target effects, including the adipsia we observed in mice receiving 100 mg/kg, as well as increased immobility noted previously at the same dose (Vijayapandi & Nagappa, 2005). These technical challenges left the necessity of angiotensin peptides uncertain.
Our experiments used a different approach to assess possible roles of angiotensin. To our knowledge, the present experiments are the first to test thirst or sodium appetite after eliminating endogenous Agt in specific cell types. The results challenge the idea that angiotensin signalling is necessary for hypovolaemic thirst or sodium appetite. These findings should re-invigorate the search for a hypovolaemic signalling mechanism that is required for activating these vital adaptive behaviours. Identifying this mechanism will require rigorously vetting signals until a candidate mechanism is identified that is not only sufficient but also necessary for stimulating hypovolaemic thirst and/or sodium appetite.

Implications for the brain renin-angiotensin system
Our mice lacking Agt in the brain did not exhibit any apparent abnormalities. Beyond thirst, sodium appetite and body weight, we did not seek alternative roles for brain-derived angiotensin signalling, but the absence of any overt abnormality caused us to reconsider the evidentiary basis of the bRAS concept and the role of angiotensinogen in the brain.
The protein translated from Agt mRNA in the brain is identical to the angiotensinogen protein produced by the liver (Campbell et al., 1984). Angiotensinogen is detectable in the CSF (Ruiz et al., 1983), but it remains unclear how much is retained or secreted by the astrocytes producing it. If exogenous renin is added, brain angiotensinogen can be used as a substrate to produce AngI (Campbell et al., 1984), and ACE derived from the choroid plexus can convert AngI into AngII (Arregui & Iversen, 1978). However, expression of Agt and Ace in the brain parenchyma and choroid plexus did not complete the bRAS puzzle. Renin was the missing piece, and we are not aware of any previous in situ evidence for endogenous renin expression by a specific cell type in the brain.
The bRAS concept originated in the 1970s, when the enzymatic activity of renin was demonstrated in brain tissue from nephrectomized dogs (Ganten et al., 1971). This concept gained momentum in the 1980s, following a report of AngII-like immunoreactivity in axons (Lind et al., 1985). That report gave rise to claims that AngII is synthesized in or incorporated by neurons, which release it as a neurotransmitter. However, subsequent work revealed that the polyclonal antibody responsible for this labelling pattern produced the same labelling in brain tissue from Agt −/− mice (Allen et al., 2009), which have no AngII (Alexiou et al., 2005). Separate claims that the brain produces renin were based on transgenic mice overexpressing a copy of the human renin gene or a rodent transgene (Ren-1c/eGFP), not a reporter knocked into the endogenous Ren1 locus (Lavoie et al., 2004a, b). The neuroanatomical patterns of expression produced by these transgenes differed, and neither matched a pattern of brain regions that reportedly contained mRNA transcribed from the endogenous locus (Lee-Kirsch et al., 1999). Other investigators failed to find evidence for renin enzymatic activity, beyond what circulates into the brain through the cerebral vasculature (Reid, 1979;Reid & Moffat, 1978;van Thiel et al., 2017). The well-characterized transcript expressed in the kidney ('renin-a') was not detected in the brain, and conditional deletion of an exon that is required for this secreted form did not alter fluid intake (Xu et al., 2011). An alternatively spliced transcript ('renin-b') was detected in brain tissue (Lee-Kirsch et al., 1999), but fluid intake values were not reported after deleting a separate portion of the renin gene required for this transcript (Shinohara et al., 2016). Another group of investigators used radioactive in situ hybridization in rats and reported transient expression of renin mRNA after a knife-cut injury to the brain (Lippoldt et al., 2001). Given its centrality to the bRAS concept, the scarcity of evidence for endogenous renin complicated the hypothesis that the brain independently generates AngII. J Physiol 601.16 Here, we identified endogenous Ren1 expression in a subset of neurons in the compact formation of the nucleus ambiguus. Neurons in this cluster extend axons outside the brain into thoracic branches of the vagus nerve that control oesophageal peristalsis (Bieger & Hopkins, 1987;Cunningham & Sawchenko, 1989;Sang & Young, 1998). In these neurons, it is unclear whether Ren1 mRNA is translated into functional renin and, if so, whether the enzyme is secreted within the brainstem or peripherally, in the thorax. It is also unclear what function renin would serve in either location. If these neurons secrete renin locally, it would be well-positioned to cleave angiotensinogen produced and secreted by astrocytes in the ventrolateral medulla. If so, it is possible that AngI produced in this region is converted to AngII by small amounts of ACE produced within the brainstem parenchyma, or by the abundant ACE produced in the choroid plexus of the fourth ventricle or foramen of Luschka. Relative to other tissues, brain concentrations of AngI and AngII are extremely low, such that detecting even femtomolar amounts of either peptide required pooling several brains (Alexiou et al., 2005;Campbell et al., 2004;van Thiel et al., 2017). Our findings open an opportunity to explore possible renin release from Ren1-expressing neurons in the nucleus ambiguus or AngII production within the hindbrain parenchyma.
We also observed sparse Ren1 labelling in choroid plexus epithelial cells, suggesting that angiotensinogen could be processed serially, by renin then ACE, directly within the choroid plexus. Ren1 expression in the choroid plexus was near the limit of detectability in some cases and absent in others, and we did not test whether the renin enzyme is produced (or secreted), but Ren1 expression in the choroid plexus offers a possible way to generate AngII directly within the cerebral ventricles.
The scarcity of renin and AngII in the brain does not discount several lines of evidence that some neurons express the angiotensin receptor AT1R (encoded by Agtr1a) and respond to AngII. The presence of AT1R in the brain is well-established, with convergent evidence from receptor autoradiography (McKinley et al., 1987;Song et al., 1992), single-cell RNA-Seq (Allen et al., 2017;Resch et al., 2017), and in situ hybridization for Agtr1a (Gasparini, Resch et al., 2019), plus Agtr1a-GFP mice (Claflin et al., 2017;Gasparini, Resch et al., 2019;Gonzalez et al., 2012) and physiological effects of exogenous AngII on neuronal activity Resch et al., 2017). Physiological levels of circulating AngII can activate circumventricular neurons and modify ingestive behaviour McKinley et al., 1998;Wong et al., 1990), but it remains unclear whether and how AngII reaches other neurons that are protected by the blood-brain barrier (Ramsay & Reid, 1975). In regions protected by the blood-brain barrier and distant from the cerebral ventricles, AT1R may simply add to the long list of transmitter-receptor 'mismatches' in the mammalian brain (Herkenham, 1987). Alternatively, AT1R-expressing neurons may detect centrally generated AngII, or detect circulating AngII exclusively under pathological conditions that disrupt the blood-brain barrier.
This leaves us with the question, what is the role of angiotensinogen produced abundantly in astrocytes? Many astrocytes express Agt (though less overall than cells in the liver or kidney; van Thiel et al., 2017;Yiannikouris et al., 2012), and angiotensinogen protein appears in the CSF (albeit in a low concentration relative to blood plasma; Ruiz et al., 1983;van Thiel et al., 2017). Despite years of study, what happens to this protein in the brain remains a mystery -including whether and how it is converted into AngII. It is tempting to assume that its only role is to serve as the substrate for generating AngII, but angiotensinogen has other possible functions. Separate studies found that its 442-amino acid, non-substrate portion, des-AngI-angiotensinogen, inhibits angiogenesis (Celerier et al., 2002) and increases body weight gain on a high-fat diet (Lu, Wu et al., 2016). Thus, angiotensinogen produced in astrocytes may play a role that is unrelated to AngII.
Our study is the first to test the necessity of endogenous Agt by deleting it from the mouse brain. Previous investigators cultured astrocytes from whole-animal Agt-knockout mice  or used GFAP-Cre to delete a floxed transgenic copy of human AGT, rather than endogenous Agt (Sherrod et al., 2005). Astrocytes from Agt −/− mice responded more slowly to injury  and overexpressing human AGT in astrocytes was necessary for a separate transgene, overexpressing human renin, to produce hypertension (Sherrod et al., 2005). Others used antisense RNA to reduce Agt expression in rats and found mild diabetes insipidus, slightly reduced blood pressure, and less fluid intake after injecting renin into the brain (Schinke et al., 1999). Using the conditional deletion approach described above (GFAP-Cre;Agt flox/flox mice) will allow future investigators to more directly assess the role of endogenous angiotensinogen in these and other brain functions.

In conclusion
Thirst and sodium appetite do not require Agt expression in astrocytes or hepatocytes. Without contradicting the evidence that angiotensin signalling promotes thirst, our findings indicate that some other signalling mechanism compensates for angiotensin deficiency by increasing salt and water intake during hypovolaemia.