A vagal reflex evoked by airway closure

Airway integrity must be continuously maintained throughout life. Sensory neurons guard against airway obstruction and, on a moment-by-moment basis, enact vital reflexes to maintain respiratory function1,2. Decreased lung capacity is common and life-threatening across many respiratory diseases, and lung collapse can be acutely evoked by chest wall trauma, pneumothorax or airway compression. Here we characterize a neuronal reflex of the vagus nerve evoked by airway closure that leads to gasping. In vivo vagal ganglion imaging revealed dedicated sensory neurons that detect airway compression but not airway stretch. Vagal neurons expressing PVALB mediate airway closure responses and innervate clusters of lung epithelial cells called neuroepithelial bodies (NEBs). Stimulating NEBs or vagal PVALB neurons evoked gasping in the absence of airway threats, whereas ablating NEBs or vagal PVALB neurons eliminated gasping in response to airway closure. Single-cell RNA sequencing revealed that NEBs uniformly express the mechanoreceptor PIEZO2, and targeted knockout of Piezo2 in NEBs eliminated responses to airway closure. NEBs were dispensable for the Hering–Breuer inspiratory reflex, which indicated that discrete terminal structures detect airway closure and inflation. Similar to the involvement of Merkel cells in touch sensation3,4, NEBs are PIEZO2-expressing epithelial cells and, moreover, are crucial for an aspect of lung mechanosensation. These findings expand our understanding of neuronal diversity in the airways and reveal a dedicated vagal pathway that detects airway closure to help preserve respiratory function.

methods to decrease functional airway volume in the mouse: (1) airway compression by inflating a cuff surrounding the thorax; (2) airway suction by applying negative pressure within the trachea; and (3) bronchoconstriction by tracheal delivery of nebulized methacholine (Fig. 1a).Physiological changes were recorded by measuring tracheal pressure (which reports the frequency and magnitude of each inspiration and exhalation), oesophageal pressure (as a proxy for intrathoracic pressure) and heart rate.Each method of airway closure triggered reflexive gasps (Fig. 1b), with a stereotypical pattern characterized by a powerful, deep inspiration and subsequent rapid exhalation (defined by a >50% increase in expiration compared with the previous and subsequent breath).Gasps were also associated with increased activity of intercostal muscles and the diaphragm (Extended Data Fig. 1a,b), as measured by electromyography.The frequency of gasping depended on the depth of anaesthesia, and under urethane anaesthesia, gasping was rarely observed at baseline but was routinely observed following airway compression (Extended Data Fig. 1c).Gasping was also triggered by hypoxia (10-12% O 2 ), to a lesser extent by nebulized citric acid and not by other stimuli, including inhaled particulates (microbeads), hypercapnia or hyperoxia (Fig. 1c).Mechanical stimuli evoked a higher gasp frequency than chemical stimuli tested, and this is probably due, at least partially, to a shorter time delay to first gasp after introduction of the stimulus (Extended Data Fig. 1d).We observed that thoracic compression, under these conditions, promoted additional changes in lung mechanics, including decreased lung compliance through chest wall restriction (Extended Data Fig. 1e).Onset of airway compression reduced airway pressure and respiratory rate (Extended Data Fig. 1f), presumably because increased intrathoracic pressure provides dominant suppression over any compensatory inspiratory drive, but had no effect on blood oxygen saturation (Extended Data Fig. 1f-h).
Gasping responses to thoracic compression persisted following transection of several vagus nerve branches, including the superior laryngeal nerve (SLN), the recurrent laryngeal nerve (RLN) and the glossopharyngeal nerve.Responses were instead abolished after transection of the vagus nerve trunk below the SLN departure point (Fig. 1d), a procedure that eliminates vagal fibres below the larynx, including those to the lung.By comparison, gasps and increases in tidal volume induced by hypoxia persisted after transection of the vagus nerve trunk below the SLN, but instead were lost after transection of the glossopharyngeal nerve, consistent with a role for carotid body chemosensation 29 .Together, these findings indicate that airway closure, induced by airway compression, suction or bronchoconstriction, induces a gasping reflex in the mouse through the vagus nerve.

Article
airway stretch 19,22,23,30 .Here we used in vivo calcium imaging within vagal ganglia 6,20 to investigate vagal responses to airway closure.Vagal ganglion imaging enables a parallel analysis of real-time responses in >100 individual neurons per experiment.We used the progeny of lsl-SALSA mice mice, which express the calcium reporter GCaMP6f-tdTomato (also called SALSA 31 ) from a Cre-dependent allele, crossed with Vglut2-ires-cre mice, in which Cre recombinase is expressed in all vagal sensory neurons.We imaged vagal ganglia while connections to the lungs were preserved (Fig. 2).As we previously observed 20 , airway stretch evoked calcium transients in a small group of vagal sensory neurons (7.1%, 93 out of 1,303 neurons, 7 mice) that express PIEZO2 (refs.6,20,21).Compressing the airways by inflating a thoracic cuff triggered acute calcium responses in 9.2% of vagal sensory neurons (120 out of 1,303 neurons, 7 mice).Most airway closure receptors (75.8%, 91 out of 120) did not respond to airway stretch, but some (24.2%, 29 out of 120) responded weakly.Airway suction activated a smaller group of vagal sensory neurons (5.1%, 25 out of 495), most of which also responded to airway compression.Nebulized methacholine also stimulated some vagal neurons (Extended Data Fig. 2c,d), which largely overlapped with compression-sensing neurons (60.7%, 54 out of 89) but not neurons selective for lung inflation (33 out of 89 responded to compression but not inflation, 21 out of 89 responded to both compression and inflation and 3 out of 89 responded to inflation but not compression) (Extended Data Fig. 2d and Supplementary Video 1).The percentages of neurons that responded to different stimuli across trials were generally conserved across mice (Extended Data Fig. 2a,b).Together, these findings indicate that airway compression acutely activates a subset of vagal sensory neurons, with the major cohort unresponsive to airway stretch, which suggests that there is a dedicated vagal pathway for detecting airway closure.

Vagal PVALB neurons mediate gasping
Vagal ganglia contain dozens of molecularly distinct sensory neuron types 6,15,17,32,33 , so we asked which sensory neurons mediate airway-closure-induced gasping.First, we used optogenetics to activate various vagal sensory neurons and measured reflexive gasping behaviour.We expressed the light-activated ion channel channelrhodopsin-2 from a Cre-dependent allele (lsl-ChR2) in different vagal sensory neuron types using Cre driver mice, including P2ry1-ires-cre, Pvalb-t2a-cre, Crhr2-ires-cre, Npy2r-ires-cre, Gpr65-ires-cre and Vglut2-ires-cre mice.As previously reported 6,14,20,34 , we then stimulated vagal sensory neurons by shining light on the vagus nerve trunk, ganglion or particular nerve branches (Fig. 3a).Activating all vagal sensory neurons in Vglut2-ires-cre;lsl-ChR2 mice evoked reflexive gasps, with a frequency of 6.0 gasps per min of illumination, whereas activating CRHR2, NPY2R, GPR65 or other neuron types did not (Fig. 3b).Optogenetic stimulation of PVALB or P2RY1 neurons similarly caused fictive gasping, with PVALB predominantly expressed in subsets of vagal P2RY1 neurons.Activation of P2RY1 but not PVALB neurons also caused swallowing, as previously reported 6 (Extended Data Fig. 3a,b,c).In Pvalb-t2a-cre;lsl-ChR2 mice, gasping was evoked by illumination of vagal ganglia or the vagal trunk distal to the SLN departure point, consistent with a role for lung afferents (Extended Data Fig. 3d).Gasps were evoked by ganglion illumination after transection of the trunk, which indicated a role for sensory neurons that transmit the information to the brain (Extended Data Fig. 3d).Pvalb is enriched in three clusters of vagal sensory neurons, only one of which also expresses Olfr78.We generated Olfr78-p2a-cre mice, and optogenetic stimulation of vagal OLFR78 sensory neurons also evoked gasping behaviour (Extended Data Fig. 3e).These experiments pinpoint transcriptome-defined vagal neuron subtypes that mediate gasping (Fig. 3c), and these neurons are distinct from P2RY1 neurons that mediate swallowing 6 .
We then asked whether vagal PVALB neurons would respond to airway closure.We crossed mice to contain a constitutive neuronal GCaMP reporter (Snap25-GCaMP6s), Pvalb-t2a-cre and a Cre-dependent fluorescent reporter (lsl-tdTomato) to distinguish Cre-positive and Cre-negative neurons.We then performed calcium imaging in vagal ganglia as described above.We observed that PVALB neurons accounted for most neurons that responded to both airway suction and compression (13 out of 14, 92.9%; Extended Data Fig. 3f).Moreover, airway closure activated 24.8% of vagal PVALB neurons, a result consistent with transcriptomics analysis indicating that gasp-promoting vagal sensory neurons represent 23.1% (161 out of 698) of all PVALB neurons (Extended Data Fig. 3e).
Pvalb-t2a-cre mice are an effective tool to mark some neurons that evoke gasps and respond to airway closure, so we asked whether ablating vagal PVALB neurons eliminated gasping to airway closure.Vagal PVALB neurons were engineered to express the human receptor (DTR) for diphtheria toxin (DT) using a Cre-dependent allele 35 , and DT was injected directly into vagal ganglia (resulting in PVALB-Vagal ABLATE mice; Fig. 3d).This DT-based approach has been effectively used for targeted cell ablation in the vagus nerve, and results in efficient ablation of Cre-positive cells in vagal ganglia but not Cre-negative cells

Article
or Cre-positive cells in other locations 6,34,36 .We verified similarly efficient removal of vagal PVALB neurons using this approach compared with control mice (Fig. 3e and Extended Data Fig. 4a), which were age-matched Cre-negative littermates injected with DT.Control mice displayed normal gasping responses to airway compression, but airway compression-evoked gasps were lost in PVALB-Vagal ABLATE mice (Fig. 3f) as well as in P2RY1-Vagal ABLATE mice (Extended Data Fig. 4b,c).PVALB-Vagal ABLATE mice also exhibited decreased gasping to airway suction and methacholine, but normal rates of hypoxia-evoked gasping compared with littermates (Fig. 3f,g and Extended Data Fig. 4e).No significant changes in tidal volume or breathing rate during relaxed breathing were observed (Extended Data Fig. 4d), but suction-induced increases in inferred tidal volume were reduced in PVALB-Vagal ABLATE mice (Extended Data Fig. 4f).Finally, the Hering-Breuer inspiratory reflex was normal in PVALB-Vagal ABLATE mice (Extended Data Fig. 4g), which indicated that reflexes evoked by airway stretch and airway closure are mediated by distinct vagal neuron types.Taken together, activating vagal PVALB neurons evokes gasps, whereas ablating them eliminates gasping to airway closure.

Airway closure neurons appose NEBs
Various terminal types have been proposed in the airways 1,2,6,14,17 , but the location and structure of neuronal terminals sensitive to airway closure are unknown.We previously developed a genetic approach to trace the terminals of Cre-defined vagal sensory neurons 14 , and here performed similar experiments in Pvalb-t2a-cre mice to genetically mark airway-closure-sensing terminals in the lungs.We injected the vagal ganglia of Pvalb-t2a-cre mice with an adeno-associated virus (AAV) harbouring a Cre-dependent reporter allele encoding tdTomato (AAV-flex-tdTomato).Alternatively, we visualized labelled neurons directly in Pvalb-t2a-cre;Vglut2-Flp mice containing a Cre-dependent and Flp-dependent tdTomato reporter gene.We observed dense innervation of the lungs in Pvalb-t2a-cre mice (Fig. 3h), with all visualized lung axons displaying characteristic candelabra terminals in airway epithelium.PVALB candelabra terminals uniformly apposed clusters of specialized lung secretory cells termed NEBs, which were visualized by immunostaining for neural cell adhesion molecule 1 (NCAM1) (Fig. 3i).We observed innervation of 62% of lung NEBs by vagal sensory neurons in general using a Cre-independent AAV-based tracer (130 out of 209 NEBs, 18 mice), and innervation of a similar percentage of NEBs (60%) by vagal PVALB neurons (Extended Data Fig. 4h,i).PVALB neurons innervated NEBs as densely as vagal P2RY1 neurons 14 .Moreover, NEBs were not innervated by other airway vagal sensory neurons marked in many other Cre lines, including Crhr2-ires-cre, Gabra1-ires-cre, Npy2r-ires-cre, Mc4r-ires-cre and Gpr65-ires-cre.By comparison, vagal CRHR2 neurons promoted apnoea and instead innervated near terminal bronchioles and alveoli, but did not innervate NEBs.Thus, vagal PVALB neurons mediate airway closure responses and densely innervate NEBs in the lung.

NEBs mediate gasping to airway closure
NEBs are clusters of rare epithelial cells (<1% of all epithelial cells) located predominantly near branch points of the major conducting airways in the lung 37 .The dense innervation of NEBs suggests that NEBs are primary sensory cells analogous to taste cells or Merkel cells, other excitable epithelial cells that communicate with neurons.It has been proposed that NEBs detect a variety of stimuli, including hypoxia, acid and mechanical stretch [38][39][40][41][42] , and they have been linked to neuroimmune signalling and airway pathophysiology [43][44][45][46] .Our observations that NEBs are densely and specifically innervated by vagal PVALB neurons raised the possibility that they may be involved in detecting airway closure.
We developed intersectional genetic approaches to gain selective access to NEBs in vivo.We combined a Flp allele (Nkx2.1-Flpo)that broadly marks pulmonary epithelial cells 47 (Fig. 4a) and a Cre allele that marks NEBs but not other pulmonary epithelial cells (Piezo2-ires-cre; see single-cell analysis of NEBs below).We also generated Ascl1-creER;Nkx2.1-Flpomice (followed by tamoxifen administration in adults), as Ascl1 is a classical marker for NEBs 48 .In both Piezo2-ires-cre;Nkx2.1-Flpo mice and in Ascl1-creER;Nkx2.1-Flpomice, expression of Cre-dependent and Flp-dependent reporter genes were driven with high selectivity and efficiency in NEBs (Fig. 4b).Selective labelling of NEBs within the lung was verified by co-staining for NCAM1, with NCAM1 also labelling nearby neuronal fibres that were not labelled by intersectional genetic tools.In both allele combinations, reporter genes were not observed in any other cell types of the lung, neurons of vagal ganglia, spinal cord or dorsal root ganglia, or in cells of the carotid body or aortic arch (Extended Data Fig. 5).Solitary neuroendocrine cells in the larynx and trachea were effectively labelled in Ascl1-creER;Nkx2.1-Flpomice but generally not labelled (<1%) in Piezo2-ires-cre;Nkx2.1-Flpo mice (Extended Data Fig. 5b).Additionally, we noted reporter expression in the thyroid and the hippocampus of Ascl1-creER;Nkx2.1-Flpomice but not in the thyroid or brain of Piezo2-ires-cre;Nkx2.1-Flpo mice.Based on these data, we conclude that Piezo2-ires-Cre;Nkx2.1-Flpo mice (hereafter termed NEB INTER mice) are an effective tool for NEB-selective genetic manipulations.
Next, we used chemogenetic approaches to activate NEBs in freely behaving mice, and measured the effects on respiratory physiology and behaviour by whole-body plethysmography (Fig. 4c).We used an intersectional reporter (inter-Gα q -DREADD) that drives expression of a designer Gα q -coupled receptor for the synthetic agonist clozapine-N-oxide (CNO) 49 .CNO elicited calcium transients in NEBs in live lung slice preparations from Gα q -DREADD-expressing mice (tamoxifen-treated Ascl1-creER;lsl-SALSA;lsl-Gα q -DREADD mice; Extended Data Fig. 6a), which confirmed that chemogenetic tools can directly stimulate NEBs.In freely behaving mice, activating NEBs with CNO caused reflexive gasping in NEB INTER ;inter-Gα q -DREADD (termed NEB Gαq-DREADD ) mice (Fig. 4d).Breathing rate and tidal volume were not significantly altered (Extended Data Fig. 6b,c), but NEB activation changed the respiratory pattern, with accelerated inspiration and increased peak inspiratory flow (Fig. 4e).Similar results were observed whether NEBs were intersectionally targeted with Piezo2-ires-cre or Ascl1-creER alleles (Extended Data Fig. 6d,e).Gasps induced by NEB activation displayed a similar characteristic pattern to gasps induced by airway closure or to spontaneous sighs (Extended Data Fig. 6f,g), which suggests a common motor command circuit.CNO-induced gasps in NEB Gαq-DREADD mice persisted under urethane anaesthesia and after transecting the SLN, RLN or glossopharyngeal nerves, but were abolished by transection of the vagus nerve trunk after the SLN departure point (Extended Data Fig. 7a), a result consistent with a role for lung afferents.In addition to respiratory changes, chemogenetic activation of NEBs in freely behaving mice caused apparent respiratory distress, with a characteristic hunching posture and immobility (Supplementary Video 2 and Extended Data Fig. 7b,c), as might be expected for a mouse experiencing air hunger.
We then used intersectional genetic approaches to ablate NEBs.We crossed NEB INTER mice to contain an intersectional allele driving expression of DTR (inter-DTR) 50 .DT administration in NEB INTER ;inter-DTR mice effectively ablated 87% of NEBs (termed NEB ABLATE mice) compared with DT-injected control littermates lacking Cre recombinase (Fig. 4f).We first asked how NEB ablation affected vagal responses to airway closure, as measured by whole nerve electrophysiology.Vagal responses to airway suction were reproducibly observed across multiple trials in control mice, but were absent in NEB ABLATE mice.By comparison, vagal responses to airway inflation were observed in both control and NEB ABLATE mice, a result consistent with a model in which distinct sensory pathways mediate airway closure and inflation (Fig. 5g).Furthermore, NEB ABLATE mice failed to gasp in response to airway closure caused by airway compression, suction or methacholine, whereas normal gasping responses were preserved in control littermates (Fig. 4g-i).Breathing rate and tidal volume were unchanged in NEB ABLATE mice during relaxed breathing, but we noted that NEB ABLATE mice had decreased compensatory changes in tidal volume during and after airway compression (Extended Data Fig. 8).NEB ABLATE mice also had decreased lung compliance and inspiratory capacity, which indicated that NEBs

Article
have an active role in maintaining normal mechanical properties of the lung.Spontaneous sighs were still displayed, but with a 36% reduction in frequency, a result consistent with the presence of NEB-independent pathways for sigh generation 13,29 .The Hering-Breuer inspiratory reflex was normal in NEB ABLATE mice across a range of airway distension volumes (Fig. 4j), so NEBs are dispensable for PIEZO2-mediated detection of airway inflation 21 .Taken together, NEBs are not required for normal breathing but are essential to maintain normal lung mechanics and to induce gasps when the airways are challenged by closure.

NEB-localized PIEZO2 and airway closure
Physiological responses to airway closure require NEBs, which suggests that at least some NEB cells express an airway-closure-activated mechanosensor.NEBs have been proposed to sense a diversity of airway cues 38,45,51 , but the underlying sensory receptors that mediate airway closure are unclear.Here we analysed the transcriptomes of individual NEB cells to search for mechanosensory proteins that detect airway closure in the lungs.NEBs are sparse cells (constituting around 0.4% of lung epithelial cells), so we used genetic tools to collect them for single-cell RNA sequencing (Fig. 5a).Immune-cell-depleted suspensions of lung cells were made from mice in which NEBs were fluorescently labelled (Calca-eGFP or Ascl1-creER;lsl-tdTomato), and NEBs were enriched by fluorescence-activated cell sorting (FACS), with only a minority of non-NEB cells also captured and included for comparative analysis.Individual cells were then encapsulated in nanolitre-sized droplets using the 10x Genomics platform, and cell-barcoded cDNA was synthesized and sequenced.Transcriptomes were obtained from 4,954 cells, including 2,975 NEB cells (defined based on known markers), and an atlas of cell types was created using unsupervised bioinformatics analysis (Fig. 5b and Extended Data Fig. 9a-c).Pathway analysis revealed that NEBs had enriched expression of neurotransmitters and genes associated with synapse formation (Extended Data Fig. 9d).Depending on the stringency of cell-type assignment criteria, a few NEB cell types could be subclustered 44,52 .Immunochemistry previously revealed that the mechanosensor PIEZO2 is expressed in at least some NEB cells 21 .Notably, single-cell RNA sequencing data here revealed uniform Piezo2 expression by NEB cells (Fig. 5c), a result similarly observed by Piezo2 RNA in situ hybridization (Extended Data Fig. 10a) and by genetic marking in Piezo2-ires-cre mice (Fig. 4b).PIEZO2 acts directly in other vagal sensory neurons to detect airway stretch and mediate the Hering-Breuer inspiratory reflex 21 and, together with PIEZO1 in other vagal sensory neurons, to detect changes in blood pressure underlying the baroreceptor reflex 34,53 .Notably, PIEZO2 is expressed in NEBs and some PVALB neurons, which raised the questions of whether PIEZO2 acts in the sensation of airway closure and, if so, where it acts.Global Piezo2 knockout is lethal soon after birth owing to respiratory distress 21 , so we used approaches for cell-selective elimination of Piezo2.We previously described Phox2b-cre;flox-Piezo2 mice and Phox2b-cre;flox-Piezo1;flox-Piezo2 mice 21,53 , in which PIEZO2 or both PIEZO1 and PIEZO2 are deleted from nodose/petrosal ganglia but not NEBs (herein termed Vagal-ΔPIEZO2 mice and Vagal-ΔPIEZO1/2 mice, respectively).We additionally generated Ascl1-creER;flox-Piezo2 mice, in which PIEZO2 expression is removed from NEBs but not vagal or glossopharyngeal sensory neurons after tamoxifen administration (NEB-ΔPIEZO2 mice).Selective and effective deletion was verified by Piezo2 RNA in situ hybridization (Extended Data Fig. 10a,b).
Next, we measured responses to airway closure following Piezo2 deletion in either NEBs or vagal afferents (Fig. 5d-f and Extended Data Fig. 10c).Knockout of Piezo2 in NEBs eliminated gasping responses to airway closure evoked by airway compression, airway suction or nebulized methacholine.Gasping evoked by hypoxia persisted in NEB-ΔPIEZO2 mice (Extended Data Fig. 10d), which suggested that there are distinct pathways for detecting airway closure and hypoxia, with hypoxic gasps presumably due to carotid body chemosensation 29 .Furthermore, whole nerve electrophysiology revealed that vagal responses to airway closure were absent in NEB-ΔPIEZO2 mice (Fig. 5g).By contrast, gasping to airway compression and suction persisted after knockout of PIEZO channels in vagal sensory neurons (in both Vagal-ΔPIEZO2 and Vagal-ΔPIEZO1/2 mice), even though these mice had a dysfunctional Hering-Breuer inspiratory reflex 21 (Figs.4j and 5g and Extended Data Fig. 4g).Similar to NEB ABLATE animals, NEB-ΔPIEZO2 mice displayed normal eupnoeic breathing, spontaneous sighs with a slightly decreased frequency, decreased inspiratory capacity and decreased lung compliance (Extended Data Fig. 10e,f).Together, these findings indicate that NEBs first sense airway closure through PIEZO2 and then communicate to vagal PVALB neurons, which relay the information to the brain to ultimately evoke reflexive gasping behaviour.

Discussion
Sensory neurons densely innervate the airways to ensure the constancy of breathing throughout life.Recent studies have revealed a diversity of transcriptome-defined vagal sensory neurons in the airways 6,15-17 , many of which are 'orphan neurons' without known functions or sensory properties.Here we ascribe a function to an orphan vagal neuron type, finding that it is required for protective gasping responses to airway closure.
Airway closure threatens pulmonary function, alters breathing patterns, evokes sensations of dyspnoea or breathlessness and promotes gasping.The forceful exchange of air that occurs during a gasp helps open the airways and restore lung capacity 9 .Here we used intersectional genetics, live cell imaging, anatomical mapping, targeted neuronal manipulations and physiology to reveal the following findings: (1) a vagal pathway that senses airway closure and promotes gasping in response; (2) the localization of gasp-promoting vagal terminals at NEBs; (3) a required role for NEBs in airway-closure-induced gasping; (4) a cell atlas of NEB expression; and (5) a required role for NEB-localized PIEZO2 in airway closure responses.Together, these results define a vagal reflex and underlying sensory mechanism that involves PIEZO2 and epithelial cell-to-neuron communication for sensing and responding to airway closure.
Gasping is also evoked by hypoxia through the carotid body 29,54 , which suggests the convergence of different sensory stimuli onto gasping control circuits in the brainstem, which involve around 200 neurons in the preBötzinger complex that express bombesin-like neuropeptide receptors 13 .The preBötzinger complex receives input from key respiratory control nuclei, including the nucleus of the solitary tract (NTS) 55 , which is the predominant target of vagal and glossopharyngeal afferents in the brain.The NTS displays spatial organization 56 , with pulmonary P2RY1 neurons terminating in lateral NTS regions associated with breathing control 14 .In future studies, it will be interesting to understand how NTS neurons responsive to airway stretch and airway closure differentially communicate with downstream breathing control circuits 55,[57][58][59] .
NEBs were previously proposed to communicate with neurons 37 , but their sensory functions remained unclear.Here we used intersectional genetics to selectively eliminate NEBs from the mouse and observed that NEB loss does not change normal breathing but instead eliminates gasping evoked by airway closure.Single-cell RNA sequencing revealed uniform expression of PIEZO2 in NEBs, and selective knockout of Piezo2 in NEBs eliminated airway-closure-induced gasps.Although some heterogeneity was observed across NEB single-cell transcriptomes, the uniform expression of PIEZO2 raises the possibility that they serve a homogeneous mechanosensory function in detecting airway closure.We propose that NEBs are similar to Merkel cells in the skin involved in touch sensation in that they are PIEZO2-utilizing epithelial cells that communicate with peripheral sensory neurons 3,4 .However, there are residual touch responses following Piezo2 knockout in Merkel cells, whereas airway closure responses reported here are absent following Piezo2 knockout in NEBs.In addition to the key mechanosensory role defined here, it is possible that NEBs are polymodal sensory cells that detect other gasp-inducing stimuli; transcriptome data should provide a framework for understanding the comprehensive functions of this poorly understood cell type.NEB frequency is dynamic 51 , with the number of NEBs reported to increase in various pulmonary diseases, including neuroendocrine cell hyperplasia of infancy, chronic obstructive pulmonary disorder and asthma, each of which present with symptoms of dyspnoea 8,60 .It will be interesting to investigate whether NEBs underlie physiological responses to airway closure in humans and induce the sensation of air hunger.
Over 150 years ago, classical studies revealed the first mechanosensory reflex of the vagus nerve in which increases in airway volume cause apnoea or a reduction in breathing.Here we described the workings of another mechanosensory reflex of the vagus nerve within the lungs and showed that this reflex is triggered by airway closure and leads to gasping behaviour.One model to explain these findings is that airway closure prevents dissipation of inspiratory pressure across the entire pulmonary tree, which leads to local pressure increases in conducting airways where NEBs are enriched.A model involving mechanosensation at airway bottlenecks would also explain why some classically defined deflation receptors are also activated by large lung inflations, with transient activity (apparent rapid adaptation) potentially explained by a short-lived pressure increase in large conducting airways that soon disperses throughout the lungs.PIEZO2 acts in NEBs to detect airway closure in the lung, whereas PIEZO2 instead acts directly in vagal sensory neurons to detect lung inflation without utilizing NEBs, indicating a heterogeneity of mechanosensory pathways within the lungs.This dedicated vagal pathway for detecting airway closure is therefore distinct from classical lung stretch receptors in that airway closure responses involve different vagal neurons, a specialized Article mechanosensory structure involving terminals at NEBs and a distinct physiological response.The existence of distinct classes of airway mechanoreceptors presumably enables precise, bidirectional control of breathing.

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/s41586-024-07144-2.

Physiological measurements
Mice were anaesthetized with urethane (1.6-1.8 mg g -1 intraperitoneal injection at least 30 min before surgery) or by isoflurane inhalation (1.5-2%) and warmed on a heated platform.Urethane was used for all experiments involving anaesthesia, except those in Extended Data Fig. 1, which explicitly describes the use of isoflurane in the figure legend.A tracheostomy was performed by inserting a cannula (18 or 20 gauge) to the carina and attaching the cannula to multipronged tubing with three openings: one to the atmosphere, one to a pressure transducer and one to an in-line gas and nebulized aerosol delivery port through which the animals are exposed to constant low-level flow rate (40 ml min -1 , which creates a tracheal pressure of 2-4 mmH 2 O, controlled by a SAR-1000 ventilator, CWE, room air in Fig. 1 and 100% oxygen in subsequent figures to minimize hypoxic sighs).The following parameters were measured: respiration was measured using an in-line pressure transducer; heart rhythm by electrocardiogram recorded with three needle electrodes placed subcutaneously in paws; oesophageal, pharyngeal and/or thoracic pressure by a fluid-filled pressure transducer; and respiratory muscle contraction by electromyographic recording with a concentric bipolar needle electrode coupled to an amplifier (1-2 kHz sampling, MP150 amplifier system, Biopac Acq-Knowledge v.4.2, v.4.5 or v.5.0).Where indicated, electromyography signals were digitally integrated (τ = 0.02 s).Pulse oximetry monitoring was performed using a MouseSTAT Jr with a mouse paw sensor (Kent Scientific).
Thoracic compression was applied by affixing a cuff around the rib cage spanning from the forelimbs to the xiphoid process and inflating the cuff slowly over 5 s to achieve a 40-60% reduction in peak tracheal pressure per breath for 10 s, unless otherwise indicated.The cuff pressure varied by animal based on size and cuff fit, and the maximal pressure was typically 5-30 cmH 2 O. Airway suction was applied (5 s) by switching the in-line gas and nebulized aerosol delivery port to a digitally controlled vacuum reservoir (SCIREQ).The final applied suction pressure was determined by a pressure transducer in the trachea (low, −5 cm H 2 O; high, −10 cm H 2 O).Inhaled gases (Airgas, as in figures and legends, remaining percentage is N 2 ) were delivered in-line through the intake port on the ventilator (40 ml min -1 , 5 min trials).For measurements of the Hering-Breuer inspiratory reflex, lung inflation was achieved by increasing air flow through the ventilator (10-25 ml min -1 g -1 body weight, 10 s).Aerosols were administered in saline (PBS) and delivered (5 s, room temperature) by a nebulizer (ANP-1100 from SCIREQ with a 50% duty cycle).Reflexes were monitored for the subsequent 5 min.Nebulized aerosols were methacholine (10 mg ml -1 , PBS, Cayman, 23092), citric acid (30% w/v, Sigma, C1909), KCl (Sigma, 12636) and microbeads (Thermo Scientific, 0.2 mm F8811, 2.0 mm F8827).Gasps were defined as single-breath expirations with >50% amplitude increase compared with the previous and subsequent breath, as inferred by electromyography and tracheal or oesophageal pressure measurements.For stimulus-evoked changes in breathing, data were normalized by comparison to values from a 10 s baseline immediately before stimulus introduction.
Respiratory mechanics (Extended Data Figs.1e, 8c and 10f) were measured using a flexiVent computer-controlled piston ventilator (SCIREQ).Animals were anaesthetized, tracheostomized (18 or 20 g cannula inserted to the carina) and attached to the ventilator.In Extended Data Fig. 1e, closed-chest animals were then paralyzed (1 mg kg -1 pancuronium, intraperitoneally, Sigma-Aldrich, P1918); measurements shown in Extended Data Figs.8c and 10f were performed using open-chest animals.Mice were ventilated at 150 breaths per min, a tidal volume of 10 ml kg -1 and 3 cmH 2 O positive end expiratory pressure with room air, unless otherwise indicated.Respiratory mechanics were assessed using the forced oscillation technique.Forced-expiratory volumes and pressure-volume loop manoeuvres were controlled by flexiVent software (flexiWare v.8.2).

Calcium imaging in vagal ganglia
In vivo imaging of vagal ganglia was performed as previously described 20,56 with minor modifications.In brief, mice were anaesthetized with urethane as described above and given PBS (300 μl, intraperitoneally) early in the surgery for homeostatic support.The left vagal ganglia was surgically exposed with branches superior to the ganglion transected and immobilized on a glass imaging platform attached to a manipulator.Calcium imaging was performed in most experiments (4 out of 7 mice) by two-photon microscopy (Olympus FVMPE resonant-scanning two-photon microscope with a piezoelectric Z-stepper (P-915, Physik Instrumente) and ×25, NA1.0 water-immersion objective) using a Ti:sapphire laser with dispersion compensation (MaiTai eHP DeepSee, SpectraPhysics), with excitation tuned to 940 to 975 nm, and fluorescence emission filtered with a 570 nm long-pass dichroic and 495-540 nm bandpass filter for GCaMP6 and a 575-645 nm bandpass filter for TdTomato signals.Volumetric images were typically collected at 1.5-3 Hz with focal planes 40-60 μm apart (Olympus FluoView software vFV31S-SW).For some experiments (3 out of 7 mice), calcium imaging was performed by confocal microscopy (Leica SP5 II with ×20, NA1.0 water-immersion objective) as previously described 20 .
Two-channel images were motion-corrected using the 'Image Stabilizer' plugin in Fiji ImageJ (v.1.52p).Red fluorescence channel images were averaged to delineate individual cells and to demarcate regions of interest (ROIs).Unhealthy cells typically exhibited distinctively strong and unvarying GCaMP fluorescence relative to baseline and were excluded.Baseline fluorescence (F 0 ) was calculated from a 20 s period before stimulus onset, and ratiometric ΔF/F 0 intensity was calculated and normalized to tdTomato fluorescence intensity at each ROI to control for photobleaching, motion and GCaMP6 expression.Cells were coded as responsive if stimulus-evoked increases in ΔF/F 0 were at least 3 s.d.above the average fluorescence across the entire imaging session.For each responsive cell, the ratio (R c /R i ) of response (ΔF/F 0 ) to compression and inflation was calculated; cells were classified as compression-selective if R c /R i > 2, as inflation-selective if R c /R i < 0.5 or as polymodal if 0.5 < R c /R i < 2. In Extended Data Fig. 2c, cells that did not respond to either airway inflation or airway closure were subsequently separated based on responsiveness to methacholine.Only some non-responsive neurons were selected for inclusion in indicated heatmaps based on computer randomization.

Cell ablations
Vagal sensory neurons were ablated as previously described 6,34 with DT (Sigma, D0564) solution (2-5 ng DT, PBS with 0.05% Fast Green FCF dye) injected (10 × 10 nl, serially) into surgically exposed vagal ganglia using a Nanoject III injector (Drummond).NEB ablation was achieved by intranasal administration (daily for 4 days) of solution containing 10 ng DT in 30 μl PBS.Cell ablation controls involved DT-administered Cre-negative littermates.Animals were allowed to recover for at least 2 weeks before subsequent experiments.

Whole-body plethysmography
Whole-body plethysmography was performed in freely behaving animals using a VivoFlow chamber system (SCIREQ).Chamber airflow was measured by a pneumotach at constant temperature and humidity with 0.5-0.6 l min -1 bias flow, and respiratory measurements were amplified, digitized and recorded using the VivoFlow-usbAMP and lox2 software (v.2.10.5.28,SCIREQ).Gas challenges involved hypoxia (12% O 2 ), hypercapnia (5% CO 2 , 21% O 2 ) and normoxia (21% O 2 ) balanced with nitrogen (Airgas).Animals were acclimated in the plethysmography chamber for 40-60 min, and then baseline respiratory data were recorded for 30 min.CNO injections involved brief removal of the animal from the chamber for administration of CNO (3 mg kg -1 , intraperitoneally, 100 μl PBS), and animals were immediately returned to the chamber for further recordings (30 min).Breaths were assigned and respiratory parameters (tidal volume, breaths per minute (BPM), minute volume) were calculated using Iox2 software (v.2.10.5.28 SCIREQ).Gasp-like breaths were manually identified from pneumotachographs and defined as a 50% increase in both inspiration and expiration compared with preceding and subsequent breaths.For quantification of eupneic breathing parameters, data were filtered to exclude respiratory events outside typical adult mouse breathing (tidal volume >2 ml or <0.05 ml; BPM > 400), averaged over the recording period (with a 7 min delay after CNO introduction), and breathing measures dependent on airway volume were normalized to the body weight of the animal.

Single-cell transcriptomics
Whole lungs below the trachea were collected from 10 Calca-eGFP and 10 Ascl1-creER;lsl-tdTomato mice (5-7 weeks old, equal male and female, 10 days after tamoxifen injection), pooled by strain, minced and incubated (60 min, 37 °C) in oxygenated papain dissociation buffer (Worthington Biochemical, LK003150).Residual tissue was mechanically dissociated through a 100 μm cell strainer, pelleted by centrifugation (400g, 7 min, 4 °C), washed, resuspended in red blood cell lysis buffer (150 mM NH 4 Cl, 10 mM NaCHO 3 and 0.1 mM EDTA) for 5 min, pelleted and resuspended in FACS buffer (0.5% BSA, 2 mM EDTA, PBS, 4 °C).Immune cells were depleted using anti-CD45 magnetic beads according to the manufacturer's instructions (BioLegend, 480027), and the remaining cells were resuspended in viability buffer (TO-PRO-3 and CellTrace Violet, both 1:10,000, in RPMI 1640; Thermo Fisher, T3605, 65-0854-39 and 11835030, respectively).Cells were collected by FACS using a FACS Aria II (BD Bioscience) with gates to select for fluorescent reporter expression and viability (CellTrace Violet positive, TO-PRO-3 negative).Collected cells were individually encapsulated in nanodroplets using a 10x Genomics platform (v.3 chemistry).Single-cell cDNA was prepared according to the manufacturer's protocol and sequenced at the Harvard Medical School Biopolymers Facility on a NextSeq 500 platform.For analysis, sequence reads were aligned to the mm10 reference transcriptome, and feature barcode matrices were generated using Cell Ranger (10x Genomics; pipeline v.3.1.0),and analysed in R (v.4.1.3)using Seurat (v.4.1.1)for quality control, pre-processing, normalization, clustering and differential expression analysis.Transformed matrices from both strains were integrated (nFeature = 3,000) before cluster identification and UMAP representation.Analysis used a standard process excluding cells with >15% mitochondrial reads or <500 unique features.Neuroendocrine cell clusters were identified for enriched expression of Epcam, Calca and Ascl1; genes to define additional lung cell types are depicted in Extended Data Fig. 9a.After differential expression analysis, gene ontology enrichment analysis used the top 50 most enriched genes ranked by significance (P value) using Enrichr 61 (https://maayanlab.cloud/Enrichr/).

Whole-nerve electrophysiology
Whole vagus nerve electrophysiology recording was performed as previously described 14,21,53 with minor modifications.In brief, urethane-anaesthetized animals (1.6 mg g -1 ) were surgically prepared to administer airway suction, as described above for airway physiology measurements.The left vagus nerve was then transected, and the lung-connected nerve end was desheathed and placed on a pair of platinum-iridium electrodes.The nerve and electrode were immersed in halocarbon oil, and a ground electrode was placed on nearby muscle.Multiunit nerve activity was amplified (CP511, Grass Technologies), digitized (MP150, Biopac), recorded (AcqKnowledge software, v.4.5, Biopac) and integrated (Elenco, RS-400).Stimulus-induced responses were calculated as a percentage change from baseline activity and normalized to the response to serotonin (intraperitoneally, 10 mM, 400 μl PBS) over 100 s after administration.

Behaviour coding
Animals were video recorded (Logitech C920 HD PRO camera) during whole-body plethysmography.After acclimation (1 h), behaviours were manually scored using BORIS software (v.8.20.4) 62 by a genotype-blinded investigator who measured time exploring, rearing, grooming, sniffing or hunching for 10 min periods before and at minutes 7-17 after CNO administration (3 mg kg -1 , intraperitoneally).Hunching was defined based on a characteristic recumbent posture and was typically associated with laboured breathing and ruffled fur.

Fig. 2 |
Fig. 2 | Imaging neuronal responses to airway closure.a, Cartoon depicting vagal ganglion imaging (top) and a two-photon image (bottom) of SALSA fluorescence in vagal ganglia of Vglut2-ires-cre;lsl-SALSA mice.b, Heatmap depicting vagal sensory neuron calcium responses (ΔF/F colour coded, 468 imaged neurons, 2 representative mice) to airway compression (red bar) and airway inflation (black bar).All 55 responsive and some randomly selected non-responsive neurons are shown.c, Quantification of neurons responsive to only compression (red, 91 out of 1,303 neurons), only inflation (black, 64 out of 1,303), both (purple, 29 out of 1,303) or neither (grey, 1,119 out of 1,303) across 1,303 neurons, 7 mice.d, Representative traces of ratiometric GCaMP6f fluorescence from individual vagal neurons responding to the indicated stimuli.e, Representative images of maximal GCaMP6f fluorescence in vagal ganglion neurons during airway compression (blue) and inflation (red).Scale bars, 50 μm (a,e).

Fig. 1 |
Respiratory measurements of airway closure.a, Representative physiological measurements before, during and after suction (grey shading) with magnifications (right) depicting data from magenta boxes, red: gasps.b, Quantification of physiological changes induced by airway suction (low), expressed as a percentage increase per breath from baseline (n = 12 trials from 3 animals), mean ± sem, dots: individual trials, one-way ANOVA with Bonferroni post hoc test: Treatment x Measure F (1.57,17.32)= 2.753, ns.c, Quantification of gasp frequency with or without thoracic compression, mean ± sem, dots: average per animal across at least three trials, n = 4 animals per group, two-way ANOVA with Bonferroni post hoc test: Anesthesia x Condition F (1,8) = 37.19, p = 0.0003; multiple comparisons (Isoflurane versus Urethane): (−), ns; (+), p < 0.0001.d, Raster plots indicating gasps (red stripe) before, during (beige shading), and after stimuli indicated, 1-4 trials per each of 5 mice.e, Respiratory measurements before, during and after thoracic compression (purple shading) measured in ventilated mice, black: mean, grey: individual animals, (n = 5 mice, one-way ANOVA with Bonferroni post hoc test:

Fig. 2 | 3 |
Imaging vagal responses to airway stimuli.a, Quantification of neurons responsive to compression only, inflation only, or both, data from individual mice (dots, n = 7) are connected (dashed lines), bar: mean.b, Heat map depicting calcium responses in vagal sensory neurons (ΔF/F color coded) to airway compression (red bar) and airway inflation (black bar) across multiple trials; 152 neurons were imaged, all responsive and only some randomly selected non-responsive neurons are shown.c, Heat map (left) depicting vagal sensory neuron calcium responses (ΔF/F color coded) to airway compression (red bar), airway inflation (black bar), and nebulized methacholine (orange bar); 286 neurons were imaged, all responsive and only some randomly selected non-responsive neurons are shown.Average neuron response (right) of indicated type to stimuli depicted; dark line: mean, light line: SEM, dotted line: time of methacholine administration.d, Proportion of neurons responsive to nebulized methacholine (orange slice, 89/468 neurons, 2 mice) and distribution of methacholine-responsive neurons activated by compression and/or inflation.Extended Data Fig.See next page for caption.