An airway-to-brain sensory pathway mediates influenza-induced sickness

Pathogen infection causes a stereotyped state of sickness that involves neuronally orchestrated behavioural and physiological changes1,2. On infection, immune cells release a ‘storm’ of cytokines and other mediators, many of which are detected by neurons3,4; yet, the responding neural circuits and neuro–immune interaction mechanisms that evoke sickness behaviour during naturalistic infections remain unclear. Over-the-counter medications such as aspirin and ibuprofen are widely used to alleviate sickness and act by blocking prostaglandin E2 (PGE2) synthesis5. A leading model is that PGE2 crosses the blood–brain barrier and directly engages hypothalamic neurons2. Here, using genetic tools that broadly cover a peripheral sensory neuron atlas, we instead identified a small population of PGE2-detecting glossopharyngeal sensory neurons (petrosal GABRA1 neurons) that are essential for influenza-induced sickness behaviour in mice. Ablating petrosal GABRA1 neurons or targeted knockout of PGE2 receptor 3 (EP3) in these neurons eliminates influenza-induced decreases in food intake, water intake and mobility during early-stage infection and improves survival. Genetically guided anatomical mapping revealed that petrosal GABRA1 neurons project to mucosal regions of the nasopharynx with increased expression of cyclooxygenase-2 after infection, and also display a specific axonal targeting pattern in the brainstem. Together, these findings reveal a primary airway-to-brain sensory pathway that detects locally produced prostaglandins and mediates systemic sickness responses to respiratory virus infection.

Pathogen infection causes a stereotyped state of sickness that involves neuronally orchestrated behavioural and physiological changes 1,2 . On infection, immune cells release a 'storm' of cytokines and other mediators, many of which are detected by neurons 3,4 ; yet, the responding neural circuits and neuro-immune interaction mechanisms that evoke sickness behaviour during naturalistic infections remain unclear. Over-the-counter medications such as aspirin and ibuprofen are widely used to alleviate sickness and act by blocking prostaglandin E2 (PGE2) synthesis 5 . A leading model is that PGE2 crosses the blood-brain barrier and directly engages hypothalamic neurons 2 . Here, using genetic tools that broadly cover a peripheral sensory neuron atlas, we instead identified a small population of PGE2-detecting glossopharyngeal sensory neurons (petrosal GABRA1 neurons) that are essential for influenza-induced sickness behaviour in mice. Ablating petrosal GABRA1 neurons or targeted knockout of PGE2 receptor 3 (EP3) in these neurons eliminates influenza-induced decreases in food intake, water intake and mobility during early-stage infection and improves survival. Genetically guided anatomical mapping revealed that petrosal GABRA1 neurons project to mucosal regions of the nasopharynx with increased expression of cyclooxygenase-2 after infection, and also display a specific axonal targeting pattern in the brainstem. Together, these findings reveal a primary airway-to-brain sensory pathway that detects locally produced prostaglandins and mediates systemic sickness responses to respiratory virus infection.
Respiratory infections caused by influenza and other pathogens are leading causes of death and hospitalization worldwide 6 , and the recent COVID-19 pandemic has broadly disrupted human society. Feeling sick can be profoundly debilitating, and most people are sick several times a year. The sensation of sickness represents a neural response to infection that may provide a highly coordinated and adaptive strategy to promote recovery 1,2 . Animals infected with various pathogens display common behavioural and physiological responses that can include fever, lethargy, loss of appetite, headache and pain, mood changes and decreased socialization, suggesting a common sickness state involving shared neural circuits 2,7 . In addition to common symptoms, other sickness responses are tailored to the site of infection; for example, some respiratory infections induce cough, congestion and bronchoconstriction, whereas some gut infections induce nausea, diarrhoea and vomiting. Infection-specific behavioural responses suggest multiple body-brain communication pathways for pathogen detection.
Several cytokines and immune mediators can induce sickness behaviour when administered in isolation, including interleukins, interferons, tumour necrosis factor (TNF) and eicosanoids 1,2,4,8 . These and other cytokines, as well as pathogen-derived factors such as lipopolysaccharide, bacterial toxins and formyl peptides, can activate an assortment of central and/or peripheral sensory neurons 4,9 . Together, these observations raise the possibility that there are many pathways for neuro-immune crosstalk, although it is unclear when or whether each of these pathways is engaged during naturalistic infections.
Chemicals such as salicylic acid from willow bark, aspirin and ibuprofen block biosynthesis of key infection-induced lipid mediators through inhibition of cyclooxygenase enzymes, and have provided a historically effective approach to manage sickness symptoms 5,10 . Prostaglandin E2 (PGE2) is a key cyclooxygenase-dependent metabolite that evokes sickness behaviour, and knockout of other enzymes downstream of cyclooxygenase in the PGE2 biosynthesis pathway also ameliorates sickness responses 11 . PGE2 is detected by a small subfamily of 4 G-protein-coupled receptors (EP1-EP4) 12 , and some but not all pyrogen-induced sickness responses are thought to be mediated by the EP3 receptor 2 . The EP3 receptor is expressed in various brain regions, including the hypothalamus and circumventricular organs as well as peripheral neurons, immune cells and many other cell types 12,13 . Region-specific knockout of the EP3 receptor in the median preoptic nucleus (MnPO) of the hypothalamus diminished lipopolysaccharide-induced fever responses 14 , with PGE2 receptors in other areas reportedly being relevant for effects on arousal and feeding 2 . These and other findings have led to several possible models in which (1) PGE2 can directly cross the blood-brain barrier owing to its hydrophobicity, (2) PGE2 can be detected or enter the brain at circumventricular organs, and/or (3) PGE2 can be synthesized in the

Glossopharyngeal neurons and sickness
The EP3 receptor is expressed in several classes of central and peripheral neurons. To determine the key site of EP3 receptor action in influenza-induced sickness, we obtained mice with an allele (Ptger3 flox ) for Cre-dependent knockout of the Ptger3 gene 14 , which encodes the EP3 receptor. We then crossed Ptger3 flox mice with either Nestin-cre or Advillin-cre ER mice, which target Cre recombinase to most central or peripheral neurons 20,21 , respectively (Extended Data Fig. 4a), and measured influenza-induced sickness behaviours. Advillin-cre ER ; Ptger3 flox mice were treated with tamoxifen to induce Cre-mediated recombination at least one week before virus administration; prior tamoxifen treatment of control mice did not affect the subsequent behavioural responses to influenza infection (Extended Data Fig. 4b). Influenza-induced decreases in feeding, drinking, movement, body weight and survival were attenuated in Advillin-cre ER ; Ptger3 flox mice (Fig. 2b), with effect magnitudes similar to ibuprofen treatment, but persisted in Nestin-cre; Ptger3 flox mice (Fig. 2a). Similar effects of Ptger3 gene deletion on sickness behaviour were observed when lower, less lethal doses of influenza virus were used (Extended Data Fig. 5). Mice with attenuated sickness behaviour displayed a similar recovery time, as normal feeding, drinking and motility were observed by about two weeks after infection in all experimental groups. These observations indicated that influenza induces sickness responses through PGE2 action on peripheral sensory neurons.
The EP3 receptor is expressed in several classes of peripheral neurons marked in Advillin-cre ER mice, including vagal sensory neurons, glossopharyngeal sensory neurons, spinal sensory neurons of the dorsal root ganglia, and potentially in other neuron types 20,22 . Glossopharyngeal and vagal sensory neurons were considered primary candidates for the detection of respiratory pathogens as they account for most of the innervation in the upper and lower airways 23 . In mice, the soma of vagal (nodose and jugular) and glossopharyngeal (petrosal) sensory neurons are fused into a large superganglion on each side of the body 23 . We injected an adeno-associated virus (AAV) with a constitutive cre allele (AAV-cre) bilaterally into both nodose-jugular-petrosal (NJP) ganglia of Ptger3 flox mice. Effective knockout of Ptger3 in NJP ganglia was confirmed two weeks after AAV injection by RNA in situ hybridization (Extended Data Fig. 6). NJP ganglion-targeted Ptger3 knockout caused a marked attenuation of influenza-induced sickness behaviour (Fig. 2c). These findings indicate that influenza induces behavioural changes through EP3 receptor expressed on vagal and/or glossopharyngeal sensory afferents.
Next, we examined how manipulations of sensory neurons affected viral transcript levels and cytokine production. Influenza infection induced PGE2 to similar levels in plasma and BALF of Gabra1-IRES-cre;

Article
Ptger3 flox , Phox2b-cre; Ptger3 flox , Advillin-cre ER ; Ptger3 flox and Ptger3 flox mice (Extended Data Fig. 9a), consistent with changes in PGE2 detection rather than PGE2 synthesis underlying the observed behavioural differences. In control mice, viral transcript levels peaked three days after infection in the upper airways and five days after infection in the lungs. In Gabra1-IRES-cre; Ptger3 flox mice, viral transcript levels were partially reduced in the upper airways and were decreased, delayed and persistent in the lungs (Extended Data Fig. 9b). Levels of interferon-gamma (IFNγ), TNF and interleukin 6 (IL-6) similarly peaked in BALF of control mice 5 days after infection, and were likewise decreased and delayed in Gabra1-IRES-cre; Ptger3 flox mice (Extended Data Fig. 9c). These findings indicate that targeted EP3 receptor knockout in sensory neurons not only affects sickness behaviour, but also the immune response and the transition from upper to lower respiratory tract infection.
vagal-glossopharyngeal sensory neurons in influenza-induced sickness, and rule out a role for other Gabra1 expression sites. Mouse cells are normally resistant to diphtheria toxin, but can be rendered sensitive by the expression of the human diphtheria toxin receptor (DTR). NJP ganglia of Gabra1-IRES-cre; lsl-DTR mice were bilaterally injected with diphtheria toxin (we term these Gabra1-ABLATE mice), resulting in highly efficient ablation of vagal and glossopharyngeal GABRA1 neurons (Fig. 4c). We previously demonstrated that this approach does not affect Cre-negative neurons in NJP ganglia, or remotely located Cre-positive neurons 27 . Gabra1-ABLATE mice displayed attenuated sickness responses to influenza infection that were similar to those in Gabra1-IRES-cre; Ptger3 flox mice or ibuprofen-treated mice (Fig. 4a).

Article
calcium transients in GABRA1 neurons acutely cultured from NJP sensory ganglia of Gabra1-IRES-cre; lsl-tdTomato mice (Fig. 4d, 16 out of 26 or 61.5% of tdTomato-positive, KCl-responsive neurons). Thus, rare glossopharyngeal NP9 neurons detect the presence of a respiratory viral infection indirectly through the EP3 receptor and, in response, evoke a multi-pronged programme of sickness behaviour.

A sensory arc from nasopharynx to brain
Gabra1-IRES-cre mice provide a valuable tool for visualizing scarce glossopharyngeal sensory neurons involved in neuro-immune crosstalk, so we used genetic mapping approaches to mark the peripheral and central projections of GABRA1 neurons (Fig. 5a). NJP ganglia of Piezo2 Gabra1-IRES-cre mice were injected with either an AAV containing a Cre-dependent reporter gene encoding tdTomato (AAV-flex-tdTomato) or alkaline phosphatase (AAV-flex-AP), as well as an AAV containing a Cre-independent reporter gene encoding GFP (AAV-GFP) to visualize the global NJP projection field. Fibres were then visualized in the airways and the brain.

Time after infection (days) Time after infection (days) Time after infection (days) Time after infection (days)
Centrally, sensory neurons of the nodose and petrosal ganglia cross the skull and innervate brainstem regions that include the nucleus of the solitary tract (NTS) and area postrema 32 , whereas jugular sensory neurons innervate the paratrigeminal nucleus 33 . Various Cre-defined vagal afferents target spatially restricted subregions of the NTS, with some but not all afferent types also accessing the area postrema 24,34,35 . The axons of NJP GABRA1 neurons were highly restricted in the lateral NTS (Fig. 5a), not observed in the area postrema, and remote from NTS locations targeted by axons of some other Cre-defined NJP neurons, such as gut-innervating neurons marked by expression of Gpr65 or Glp1r 35 . These findings further support a model for topographic organization in the brainstem 36 , with neurons relaying the presence of an airway infection spatially restricted from at least some other interoceptive inputs.
In the periphery, labelled GABRA1 axons were observed in only a few internal organs known to receive vagal and glossopharyngeal innervation. We observed densest innervation in the nasopharynx and oral cavity including taste papillae, some innervation of tracheal and pharyngeal muscles, and sparse innervation of stomach muscle (Extended Data Fig. 10b). Innervation was not observed in many other internal organs, including the heart, oesophagus, aorta and carotid sinus. The nasopharynx is a rich site for immune surveillance, as it Data are mean ± s.e.m.; n = 8 mice per group. Food intake: P < 0.0001; water intake: P < 0.0001; motility: P < 0.0001; body weight: P = 0.0004; survival: P = 0.049. b, Wild-type mice with bilateral glossopharyngeal nerve transection surgery or sham surgery were infected with influenza A virus and monitored as indicated. Data are mean ± s.e.m.; n = 8 mice per group. Food intake: P < 0.0001; water intake: P < 0.0001; motility: P < 0.0001; body weight: P < 0.0001; P = 0.036.
Two-tailed unpaired t-test as detailed in Fig. 1 for behaviour or physiology analysis; log-rank (Mantel-Cox) test for survival analysis. c, Immunostaining for DTR in whole-mount preparations of NJP ganglia four weeks after injection. Scale bars, 200 μm. d, Calcium transients evoked by PGE2 (1 μM) or KCl (150 mM) were imaged using Calbryte 520 AM in tdTomato-positive neurons acutely collected from NJP ganglia of Gabra1-IRES-cre; lsl-tdTomato mice. Scale bar, 10 μm. Images are representative of three technical replicates. AU, arbitrary units.

Article
connects the nasal cavity to the rest of the respiratory system and provides an early line of mucosal immune defence 37 . Influenza infection increases PGE2 levels locally within the nasopharynx 38 , and in humans leads to inflammation of nasopharyngeal tonsils 39 . Mice have an orthologous system known as nasopharynx-associated lymphoid tissue 40 , so we tested whether GABRA1 axons were located near relevant immune mediators. GABRA1 axons were enriched dorsally in epithelial and subepithelial layers of the nasopharynx (Fig. 5b) and notably, cyclooxygenase-2 expression was likewise detected in dorsal epithelium of the nasopharynx (Fig. 5c). Moreover, cyclooxygenase-2 expression was markedly upregulated in dorsal nasopharynx after influenza infection in mice (Fig. 5c). Thus, GABRA1 neurons occupy a strategic and privileged position for detection of the increased PGE2 levels that occur during viral infection.

Discussion
Despite seasonal vaccination and antiviral therapeutics, influenza A viral infection remains one of the most severe human threats, affecting millions of people worldwide each year. Cyclooxygenase-2 inhibitors are among the most prevalent anti-sickness and anti-pain medications, with around 30 billion doses of nonsteroidal anti-inflammatory drugs (NSAIDs) consumed annually in the USA alone 41 . Blockade of PGE2 production alleviates sickness, and several models have been proposed for how the brain receives input about a peripheral infection through PGE2 (ref. 2 ). Here, we observe that influenza-induced sickness behaviours were similarly attenuated by 1) ibuprofen, aspirin and EP3 receptor antagonism, 2) Ptger3-targeted gene knockout using Advillin-cre ER , Phox2b-cre and Gabra1-IRES-cre mice, and direct AAV-cre injection in NJP ganglia, 3) ablation of GABRA1 NJP neurons, and 4) and glossopharyngeal nerve transection. From these combined results, we conclude that a small cluster of Gabra1-expressing glossopharyngeal sensory neurons detects PGE2 and induces a neuronally orchestrated state of sickness associated with a suite of behavioural responses to respiratory viral infection.
Prostaglandins and many other cytokines can activate vagal and other peripheral afferents in vitro 8,29,31 , but their physiological roles in naturalistic infections have been difficult to parse. Vagotomy blocks sickness responses to cytokines and lipopolysaccharide in some studies but not others 42,43 , and this variability may be owing to the location or dose of pyrogen administration. Our findings indicate a clear role for sparse glossopharyngeal sensory neurons that are anatomically poised to detect influenza-induced PGE2. PGE2 has limited stability in vivo 44 , so peripheral sensory neurons that directly detect PGE2 at the infection site would seemingly provide a fast and robust conduit for information transfer to the brain.
Sickness behaviours have been proposed to help conserve energy to fight off infection 1 . However, eliminating the influenza detection pathway of the glossopharyngeal nerve not only attenuates sickness behaviour, but also promotes survival. Knockout of EP3 from a small group of glossopharyngeal neurons is sufficient to mimic the profound survival phenotype observed in full-body knockouts. One model is that pathogen-induced anorexia is beneficial in some infection models but harmful in others. For example, blocking PGE2 production is harmful during mycobacterium 3 infection 45 , but promotes survival during influenza infection 19,46 . Moreover, glucose gavage is detrimental during bacterial sepsis, where blood glucose may provide direct fuel for the pathogen, but protective during influenza infection 47 . Thus, disrupting neuronal responses to infection may decrease mortality by attenuating changes in feeding behaviour. Moreover, an infection-activated, anorexia-promoting neural pathway may provide a net survival benefit during certain bacterial infections, and thus is maintained over evolution, but is harmful during viral infection, as observed here. In addition, sickness behaviour may provide a separate population-level benefit as sick animals seek isolation and thereby limit pathogen transmission to kin 48 .
These findings indicate that targeted EP3 receptor knockout in sensory neurons not only affects sickness behaviour, but also the immune response and viral transition from the upper to lower respiratory tract. A parsimonious interpretation of these findings is that petrosal GABRA1 neurons, on detection of PGE2, engage neural circuits that evoke coordinated responses that include sickness behaviours as well as motor reflexes that affect immune function. Additionally, changes in feeding behaviour may secondarily affect immune function 47,48 and activation of petrosal GABRA1 neurons may affect levels of cytokines other than PGE2 (Extended Data Fig. 9c), which can potentially elicit further neuronal feedback and enhance sickness behaviour. All such responses to infection would critically depend on the EP3 receptor in petrosal GABRA1 neurons, which serves an essential role in first relaying the presence of an upper respiratory infection to the brain, ultimately evoking sickness behaviour.
Influenza infection-induced sickness behaviour was attenuated, but not eliminated, following NSAID treatment, targeted EP3 receptor knockout, targeted neuronal ablation and glossopharyngeal nerve transection, suggesting other routes to sickness. Of note, the kinetics of residual sickness behaviour following manipulation of petrosal GABRA1 neurons matches the time course of virus accumulation in the lungs, indicating that there are probably two phases of influenza-induced sickness behaviour. The first phase occurs when the virus is most prevalent in the upper respiratory tract, and sickness is primarily mediated by PGE2-detecting glossopharyngeal sensory neurons marked by GABRA1, which project to the nasopharynx. The second phase of sickness occurs when the virus is most prevalent in the lower respiratory tract, and the lung is primarily innervated by vagal rather than glossopharyngeal sensory neurons. Our data raise the possibility that this second phase of sickness involves another neuronal pathway independent of PGE2, EP3 and glossopharyngeal sensory neurons. Inhibitors of residual neuro-immune interaction pathways, once identified, could potentially provide improved relief of influenza-induced malaise when used in combination with NSAIDs. We also note that PGE2 receptors are expressed in multiple NJP sensory neuron types, as well as in the brain and dorsal root ganglia. It is likely that the nervous system uses multiple sensory pathways to detect peripheral infections, with different sensors perhaps specialized for particular pathogen types or infection locations in the body. In this way, gut and respiratory sensory neurons could potentially engage different neural circuits that lead, for example, to either cough or nausea 31,49 . Consistent with this notion, various routes to sickness are differentially sensitive to pharmacological inhibition; for example, gut malaise is treated clinically using antagonists of the serotonin receptor HTR3A 50 . Understanding the diversity of sensory pathways to sickness and when they are engaged by different pathogen infections will provide an essential framework for deciphering this complex and poorly understood physiological state, and may enable improved therapeutic interventions.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-05796-0.

Influenza A virus inoculation
Influenza A/PR/8/34 (H1N1) was purchased from Charles River Avian Vaccine Services (10100374) with a EID 50 of 10 12 ml −1 . Virus was diluted in sterile saline for the administered dosage, which was EID 50 10 6 ml −1 unless otherwise indicated. Viral administration was adapted from previous studies 52 . Awake mice were manually restrained and influenza A virus was delivered through each nostril (25 μl total volume) slowly using a P200 pipette to prevent spillage or delivery to the mouth. After administration, the mice were held with nostrils upright for 10-15 s, and returned to their cage.

Behavioural monitoring For analysis of influenza A virus infection-induced behavioural changes,
six-to eight-week-old mice without sex bias were singly housed with free access to food and water. Baseline values were obtained by measuring daily food intake, daily water intake, body weight, and motility for three days before inoculation. After viral infection, all measurements were expressed as a percentage change from baseline values. Food and water intake were calculated daily by measuring the change in weight of remaining food and water in the cage. Animal movement was recorded (C922x Pro Stream Webcam, Logitech) in the home cage for 30 min daily and motility expressed as the total distance moved, as determined by the MTrack2 plug-in available in ImageJ (1.53q, NIH). For measurement of acute food intake, mice were fasted for 24 h, and at dark onset, were administered with PGE2 or sulprostone (EP3 agonist, 14765, Cayman Chemical Company) via the exposure route and doses indicated in Figure legends. Animals were then singly housed in clean cages with ad libitum access to food and water, and the amount of food consumed for 1 h determined by measuring the food in the cage before and after the test period.

Body temperature measurements
Core body temperature was monitored by a rectal probe or radio telemetry. Rectal probe measurements were made hourly each day from noon to midnight using a mouse rectal temperature probe (Kent Scientific, RET-3, 0.16 mm tip diameter) connected to a thermometer (Kent Scientific WD-20250-9). Probes were sterilized (70% ethanol), lubricated (Aquagel lubricating gel, ParkerLabs, 57-05), and inserted ~5-7 mm into the rectum of physically restrained mice, and temperature was recorded for 30 s. For radio telemetry, a radio telemetry device that reports body temperature (HD-X11, DSI) was surgically implanted in the abdominal cavity. Surgery was performed under anaesthesia with avertin (200 mg kg −1 , intraperitoneal injection) and the abdominal area was disinfected and shaved. A ~1 cm incision was made, and the abdominal peritoneum at the linea alba was gently exposed to insert the sterile radio telemetry device. The surgery site was sealed with VICRYL Sutures ( J392H, Ethicon). Animals received Buprenorphine SR (BupSR-LAB, ZooPharm, 20 mg kg −1 subcutaneously at the back of the neck), and were allowed to recover for at least a week. Body temperature data were collected over 15 min intervals with a receiver (easyTEL, Emka Technologies) connected to a computer running iox software v.2.10.8. All data points from one mouse on a given day were averaged.

Ganglion injection
NJP ganglia were injected with AAVs or diphtheria toxin as previously described 34 , with minor modifications. Mice were anaesthetized (200 mg kg −1 avertin, intraperitoneal injection), and depth of anaesthesia ensured throughout the procedure by lack of a toe pinch response. NJP ganglia were exposed and serially injected (10 × 13.8 nl) with saline solution containing either AAVs (titre > 6.7 × 10 12 vg ml −1 ) or diphtheria toxin ( Animals recovered for at least two weeks for behavioural analysis or four weeks for histological analysis.

Nerve transection
Mice were anaesthetized (200 mg kg −1 avertin, intraperitoneal injection), and depth of anaesthesia ensured throughout the procedure by lack of a toe pinch response. NJP ganglia were surgically exposed, and the glossopharyngeal nerve was identified and severed immediately adjacent to the ganglia using microscissors. Sham surgeries involved NJP ganglia exposure without nerve transection. Surgical wounds were closed with coated VICRYL Sutures ( J392H, Ethicon) and animals received 20 mg kg −1 Buprenorphine SR (BupSR-LAB, ZooPharm, subcutaneously at the back of the neck) as an analgesic. Animals recovered for at least two weeks before behavioural analysis.

Single-cell transcriptomics
All UMAP plots in this manuscript were made from published single-cell transcriptome data (GEO Accession ID: GSE145216) 22 using Seurat (4.1.0) and R Studio (4.1.2).

Quantitative PCR analysis
Levels of mouse Ptger3, Gapdh transcript and the influenza virus nucleoprotein (NP) were measured in cDNA obtained from hypothalamus, NJP ganglia, nasal lavage (for upper respiratory tract analysis), and/or BALF (for lower respiratory tract analysis). BALF was collected by surgically opening the neck and cannulating a 21G catheter into the trachea. In total, 3 × 1 ml of PBS was slowly injected into the lungs then solution was gently aspirated back to the syringe and collected. Collected BALF was centrifuged (7 min, 400g, 4 °C), and the supernatants were stored (−80 °C) until analysis. Nasal lavage was performed by inserting a needle (26G) into the upper trachea angled toward the nose. Sterile PBS (1 ml) was administered through the needle via a cannulated syringe (PE20 polyethylene cannula, inner diameter 0.38 mm), and collected after expulsion from the nose. RNA was collected and cDNA synthesized from BALF, nasal lavage fluid, and homogenized tissue using Trizol (15596026, ThermoFisher) and the High-Capacity cDNA Reverse Transcription Kit (4368813, ThermoFisher) according to the manufacturer's protocols. qPCR analysis was performed on cDNA using gene-specific primers, PowerTrack SYBR Green Master Mix (A46012, ThermoFisher) on QuantStudio 7 qPCR instrument (ThermoFisher). Gapdh primers were GGGTGTGAACCACGAGAAATATG and TGTGAGGGAGATGCTCAGTG TTG; Ptger3 primers were CAACCTTTTCTTCGCCTCTG and TTTCTG CTTCTCCGTGTGTG; and influenza virus nucleoprotein primers were GACGATGCAACGGCTGGTCTG and ACCATTGTTCCAACTCCTTT. Expression levels of Ptger3 and nucleoprotein were normalized to Gapdh levels. Normalization was by the 2 -ΔCT method for Extended Data Fig. 4a, and the 2 -ΔΔCT method 53 for Extended Data Fig. 9b, with added normalization using values from uninfected controls.

Fibre photometry
Fibre photometry of AGRP neurons was performed as described 54 with PGE2 (intraperitoneal injection, 0.5 mg kg −1 ) or PBS acutely injected during a brief manual restraint using Synapses software (Build: 94-42329P, Tucker-Davis Technology). Data were analysed using Matlab (R2020a). Sample sizes were chosen based on prior expertise and publications in our field 26,55 and are disclosed in the Figure legends. Animal groups were randomly assigned and control animals were age-matched to experimental animals. The same investigator performed genotyping and analysis of sickness responses, so data were not generated blind to genotype or experimental group.

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

Data availability
All data used to generate figures, including behavioural data points from each individual mouse are provided as source data. All reagents that may not be commercially available including certain transgenic mice and AAVs will be made freely available upon reasonable request. Source data are provided with this paper. Statistical values are identical to those in Fig. 1a. b, PGE2 levels in plasma (left, middle) and BALF (right) were measured by ELISA at time points indicated after exposure to influenza A virus (red, blue) or vehicle control (black). Some virus-infected mice (blue) were additionally given ad libitum access to ibuprofen (1 mg/ml) in drinking water from 3 days prior to infection (left) or received daily aspirin administration (IP, 20 mg/kg), mean ± sem, n: 3 mice per group, ***p < 0.0005 by two-way ANOVA followed by Bonferroni's multiple comparison test with comparisons made between red and black curves (red stars) or red and blue curves (blue stars). p values in b are <0.0001 for all indicated stars.   a, Core body temperature in flox-Ptger3 and Gabra1-ires-Cre; flox-Ptger3 mice was measured daily after administration of influenza virus (red, blue) or saline (black) using a rectal probe (left) or radiotelemetry (right), mean ± sem, n: 6 (left) and 3 (right) mice per group, ***p < 0.0005 by one-way ANOVA Dunnett's multiple comparison test (left); ***p < 0.0005 by two-tailed unpaired t-test for right as detailed in Fig. 1 for behavior/physiology analysis (right), comparisons between red and black curves (red stars) or between red and blue curves (blue stars). For data on the right, two of three control flox-Ptger3 mice died on day 5, so subsequent data is represented as a dashed line and statistical analysis was performed only on data from days 1-4. b, Mice were given daily injections of aspirin (IP, 20 mg/kg), red or vehicle alone (black) throughout the paradigm. After three days, mice were infected with influenza A virus and subsequently monitored as indicated, mean ± sem, n: 6 mice per group, ***p < 0.0005 by two-tailed unpaired t-test as detailed in Fig. 1   Days Post Infection EID 50 10 4 /ml EID 50 10 4.5 /ml EID 50 10 5 /ml EID 50 10 5.5 /ml EID 50 10 6 /ml EID 50 10 6.5 /ml EID 50  previously injected with tamoxifen) or flox-Ptger3 mice (black) indicated were infected with sublethal doses of influenza A virus (top: 10 5 EID 50 or LD 21 , bottom: 10 5.5 EID 50 or LD 39 ) and monitored daily as indicated, mean ± sem, n: 6 mice per group, ***p < 0.0005 by two-tailed unpaired t-test as detailed in Fig. 1 for behavior/physiology analysis, *p < 0.05, ns: not significant by a log-rank  39 ) and monitored daily as indicated, mean ± sem, n: 6 mice per group, **p < 0.005, ***p < 0.0005 by two-tailed unpaired t-test as detailed in Fig. 1