Transformation of Our Understanding of Breathing Control by Molecular Tools

Abstract

The rhythmicity of breath is vital for normal physiology. Even so, breathing is enriched with multifunctionality. External signals constantly change breathing, stopping it when under water or deepening it during exertion. Internal cues utilize breath to express emotions such as sighs of frustration and yawns of boredom. Breathing harmonizes with other actions that use our mouth and throat, including speech, chewing, and swallowing. In addition, our perception of breathing intensity can dictate how we feel, such as during the slow breathing of calming meditation and anxiety-inducing hyperventilation. Heartbeat originates from a peripheral pacemaker in the heart, but the automation of breathing arises from neural clusters within the brain-stem, enabling interaction with other brain areas and thus multifunctionality. Here, we document how the recent transformation of cellular and molecular tools has contributed to our appreciation of the diversity of neuronal types in the breathing control circuit and how they confer the multifunctionality of breathing.

INTRODUCTION

Breathing is the flow of oxygenated air into the lungs during contraction of the diaphragm muscle (inspiration), followed by expulsion of carbon dioxide–enriched air, driven mostly by the stretched lungs’ elastic recoil (expiration). Layered upon this is a brief constriction of the upper airway larynx immediately after inspiration (postinspiration), which momentarily traps oxygenated air in the lungs to maximize exchange of gases between the environment and the body (1) (Figure 1).

Figure 1.

Breath is composed of three sequential phases that are encoded by the brainstem breathing control circuit. First, oxygenated air is drawn into the lung by contraction of the diaphragm muscle (inspiration; blue, left). This phase is generated by the autonomous, rhythmic activity of the breathing pacemaker, the preBötC (blue, right), which transmits its inspiratory signal to the rVRG diaphragm premotor neurons. Second, air is momentarily trapped in the lung by closure of the larynx upper airway (postinspiration; gray, left). This signal originates in the PBN/KF and the PiCo and terminates inspiration (gray, right). The third phase is when deoxygenated air is pushed out of the body by the lung’s elastic recoil (expiration). Expiratory duration and strength are controlled by the BötC and pFRG, which delay the onset of the preBötC burst and activation of muscles that drive expiration, such as the abdominal muscles, via the cVRG. Breathing rate and breath phases are constantly modulated by central and peripheral sensory signals, including those from the RTN and lung, which are transmitted to each component of the brainstem breathing control circuit. Abbreviations: BötC, Bötzinger complex; cVRG, caudal ventral respiratory group; PBN/KF, parabrachial nucleus/Kölliker-Fuse complex; pFRG, parafacial respiratory group; PiCo, postinspiratory complex; preBötC, preBötzinger complex; RTN, retrotrapezoid nucleus; rVRG, rostral ventral respiratory group.

For a single breath to become rhythmical breathing, the diaphragm must cycle between activity and inactivity. This rhythmic cycle is controlled by a small collection of neurons in the ventrolateral medullary brainstem called the pre-Bötzinger complex (preBötC), named after a German wine (2). The rhythmic preBötC inspiratory signal is transmitted to the diaphragm via adjacent premotor neurons in the rostral ventral respiratory group (rVRG). Postinspiration results from reciprocal connections between the preBötC and the pontine parabrachial nucleus/Kölliker-Fuse complex (PBN/KF) (3) and likely also the medullary postinspiratory complex (PiCo) (4) (Figure 1).

Although expiration during calm breathing is thought to be a passive process that does not require muscle activity, it is typically controlled by signals from the Bötzinger complex (BötC) and the parafacial respiratory group (pFRG). These regions retime the onset of preBötC-dependent inspiration to prolong or shorten expiration and also increase the strength of expiratory muscle contraction via premotor neurons in the caudal VRG (cVRG) (1, 5) (Figure 1).

Because this circuit gives rise to an autonomous and reproducible rhythmic motor pattern, it fulfills the criteria to be a central pattern generator (CPG) (6, 7) (Supplemental Figure 1). However, the speed and pattern produced by this CPG are constantly modulated by a number of sensory inputs, which are powerful enough to stop breathing or transform a basic breath into a sigh. Fundamental examples of these modulatory signals include lung volume and oxygen-sensing inputs from the lungs and carotid body, which traverse to the breathing CPG via the vagus nerve and glossopharyngeal, respectively, via the nucleus tractus solitarius (NTS) (8) and carbon dioxide–sensing signals from the medullary retrotrapezoid nucleus (RTN) and caudal serotonergic raphé nucleus (9–11). Other interesting inputs that are currently unmapped are likely to include those that connect breathing with emotional states or allow volitional control.

In this review, we describe how cellular and molecular tools have enabled granular descriptions of the different neurons that comprise the CPG and its modulatory signals from the raphé and vagus nerve and the way in which they give rise to the multifunctional nature of breathing. We have not included a detailed discussion of the current models for preBötC rhythm generation, which have been adequately described elsewhere (1, 12–16).

MOLECULAR TOOLS AND THEIR ACCURATE INTERPRETATION ARE ESSENTIAL FOR DISSECTING CENTRAL PATTERN GENERATORS

Decades of studies on invertebrate CPGs have shown that a key way to understand them is to identify the specific purpose(s) of each neuron within the context of its overall function (e.g., pacemaking). Could this be achieved for the breathing CPG? One possible approach would be to identify cell types using molecular signatures, which could then be leveraged to carry out genetic characterizations of the intrinsic biophysical properties of these cells and selectively perturb their activity or delete them to define their purpose.

However, for a research community to effectively use molecular cell type definitions, a consistent language must be chosen. The cells identified by genetic lineage tracing (Cre lines crossed to a reporter) cannot be equated to mature cells that express the same gene. For example, the μ-opioid receptor gene Oprm1 traces half of the preBötC neurons (17), even though only 10% of mature neurons within the preBötC express Oprm1 (18–20). In this review, to provide a model for moving forward, we distinguish these two descriptions by using ^L for lineage and ⁺ for transcript expression. Because reporter transgenes are designed to reflect transcripts, we also annotate transgene expression using ⁺. Note that protein expression is also marked with ⁺, but consideration should be taken when comparing protein versus transcript expression. Each experimental approach has tremendous value, but thoughtful and precise definitions must be used to avoid confusion between studies.

MOLECULAR MARKERS CAN DELINEATE CENTRAL COMPONENTS OF THE BREATHING CONTROL CIRCUIT

Most of the clusters within the breathing CPG are found within the lateral paragigantocellular reticular and intermediate reticular nuclei and are distinguished by fuzzy anatomical borders that appear to expand and contract (14, 21). This anatomy, which contrasts with the discrete architecture of brain regions such as the hippocampus, cerebellum, and cortex, poses a challenge to selectively and reproducibly study the components of the breathing CPG. One possibility is that distinct molecular markers will be able to define the well-described properties of each cluster, for example, inspiratory versus expiratory or rhythmogenic versus premotor. If so, these markers could be leveraged to identify and then individually perturb each cluster.

A blueprint for this molecular code approach is the RTN. The RTN is developed from lineages including atonal 1 (Atoh1) and early growth response 2 (Egr2), and mature RTN neurons express genes such as paired-like homeobox 2b (Phox2b), neuromedin B (Nmb), and tachykinin 1 receptor [hereby referred to as neurokinin 1 receptor (Nk1R) in accordance with the literature]. When combined, these markers form a molecular code that can define the RTN: Phox2b⁺Nmb⁺Nk1R⁺Atoh1^LEgr2^L. In fact, this code could even be further developed (22–30). Indeed, unique or restricted markers like Phox2b and Nmb have proven their worth in numerous studies elucidating the role of the RTN in breathing behavior.

Although molecular markers for components of the breathing CPG have been described (Supplemental Figure 1), they are far from ideal. The most specific markers are for the RTN (see above) and choline acetyl transferase (ChAT⁺) and vesicular glutamate transporter 2 (Vglut2⁺) neurons in the PiCo (4, 31). Molecular markers for the remaining groups are rarely apparent in the majority of relevant neurons and do not demarcate the boundaries between nearby groups. For example, although most glutamatergic neurons within the preBötC are derived from the developing brain homeobox 1 (Dbx1^L) lineage, so are many neurons in the adjacent premotor rVRG (32–34). In addition, inhibitory neurons in the BötC and rVRG are both derived from ladybird homeobox 1 and paired homeobox 2 (Lbx1^LPax2^L) developmental lineages (34, 35). Furthermore, as well as defining core respiratory groups, it will be important to identify molecular markers for nearby or intermingled nodes that pattern breathing-dependent behaviors, including vocalization (36), whisking (37), and licking (38).

COMBINATIONS OF MARKERS REVEAL FUNCTIONALLY DISTINCT NEURONS WITHIN THE preBötC

The molecular definition of the RTN was facilitated by its cellular composition, comprising few glutamatergic neurons whose biophysical properties and purpose in breathing control (pH chemosensation) are relatively homogeneous (25). Unlike the RTN, however, most other breathing CPG clusters contain multiple neuronal types, such as neurons that are active at early or late expiration within the BötC (39–47). It is therefore imperative to define the molecular cell types within these heterogeneous clusters as well as markers for each cluster. This has motivated studies over the last decade, and indeed, the molecular genetic definitions for cell types within the preBötC have made it possible to discover novel roles for breathing control neurons in respiratory function and other physiological processes. Below are four examples of molecularly defined murine preBötC cell types (Supplemental Figure 1) and their physiological roles. Importantly, some of these molecular markers have enabled identification of the preBötC in the human medulla, confirming the conservation of neuronal types across evolutionarily distant mammalian species (48).

Sst⁺Nk1R⁺Dbx1^LRobo3⁺Penk⁻ Neurons Bilaterally Synchronize Breath Shape

Nearly two decades ago, the neuropeptide somatostatin (Sst) was found to have restricted expression in ∼100–200 glutamatergic neurons in the preBötC (49). Sst⁺ neurons are Dbx1^L derived (32, 33), express Nk1R⁺ (49), and are mostly commissural (49). Conditional expression of the light-gated ion channel channelrhodopsin 2 (ChR2) to briefly or continuously excite Sst⁺ neurons (Sst-Cre with preBötC AAV-DIO-ChR2) induces ectopic or augmented breaths, respectively, and slows breathing (50, 51). Slower breathing is likely secondary to an increase in breath volume, which prolongs the duration of inspiration and expiration. This phenotype contrasts with the rapid increase in breathing rate after excitation of all Dbx1^L neurons (50, 52), demonstrating that Sst⁺Nk1R⁺Dbx1^L neurons participate in shaping, but not pacing, breath and that pacing neurons must be Dbx1^LSst⁻. About 30% of all Dbx1^L-derived neurons express the commissural axon out-growth cue, roundabout guidance receptor 3 (Robo3⁺), and its conditional genetic deletion from Dbx1L prevents essentially all axons from crossing the midline, resulting in desynchronization of the bilaterally located preBötCs and breathing muscles (32). One-quarter of Robo3⁺Dbx1^L neurons are Sst⁺ and nearly three-quarters of Sst⁺ neurons are commissural, but Sst⁺ neurons do not express the premotor neuron marker preproenkephalin peptide (Penk) (49, 53). This indicates that the molecular code of nearly all Sst⁺ neurons is Sst⁺Nk1R⁺Dbx1LRobo3⁺Penk⁻ and that they function to synchronize breathing pattern bilaterally (Figure 2). However, these neurons will have additional roles in breathing control as they project to multiple brain sites, including the PBN, periaqueductal gray, thalamus and hypothalamus (54).

Figure 2.

Example to illustrate the molecular identification of a candidate preBötC rhythm-generating neuronal type. Hierarchical tree of the molecular subtypes that constitute Nk1R⁺ neurons (a) and the phenotypes (b) supporting the hypothesis that a subset that express Sst⁺ are involved in patterning breath, while the remaining neurons are breathing rhythm generators. Percentages (noted in gray) represent the proportion of neurons in each subdivision. Question marks indicate an anticipated cell type without known molecular markers. Abbreviations: ChR2, channel rhodopsin 2; Dbx1, developing brain homeobox 1; Lbx1, ladybird homeobox 1; Lhx9, lim homeobox 9; MafB, maf bzip transcription factor B; Nk1R, neurokinin 1 receptor; NTS, nucleus tractus solitarius; Oprm1, μ-opioid receptor 1; Pax2, paired homeobox 2; Penk, preproenkephalin; Pons, pontine brainstem; preBötC, preBötzinger complex; Robo3, roundabout guidance receptor 3; Sst, somatostatin; Sst2aR, somatostatin 2a receptor; Vglut2, vesicular glutamate transporter 2. Symbols: ⁺, gene expressed; ⁻, gene not expressed; ^L, developmental lineage.

Nmbr⁺Grpr⁺ Neurons Produce Sigh Breaths

The neuromedin B receptor (Nmbr) is a G protein–coupled receptor that responds to the RTN neuropeptide Nmb and is a member of the same family as the gastrin-releasing peptide receptor (Grpr). These two receptors are selectively expressed within ∼200 preBötC neurons. Microinjection of Nmb or GRP into the preBötC induces sighing, ectopic hyperpolarization of the RTN inhibits sighing, and ablation of Nmbr⁺Grpr⁺ preBötC neurons eliminates sighing (24, 55). These phenotypes appear to be selective, as basal and hypoxic breathing were unchanged after ablation of these neurons. Two important findings that resulted from these studies are that sighs appear to be critical for sustaining life, as breathing becomes severely ataxic several days after Nmbr⁺Grpr⁺ ablation (24), and that stress-induced emotive sighing is initiated by a neuropeptide signal from the hypothalamus (55).

Cdh9⁺Dbx1^L Neurons Convey Breathing Information to the Entire Brain

Approximately 175 preBötC neurons express the adhesion molecule cadherin 9 (Cdh9⁺) and are also Dbx1^L (Cdh9⁺Dbx1^L). These neurons display inspiratory-related activity and project to, synapse onto, and activate the locus coeruleus, the key arousal center of the brain. Ablation of Cdh9⁺Dbx1^L neurons does not impact the pace or pattern of breathing but instead modulates arousal state and results in calmer mice (56). Thus, the molecular definition and genetic deletion of these neurons enabled the discovery of a corollary (efferent) signal that directly links fast breathing to aroused states like anxiety and slow breathing to calmness.

Oprm1⁺Dbx1^LFoxp2^L Neurons Mediate Opioid-Induced Respiratory Depression

Exogenous opioids significantly slow breathing, an effect that can be mitigated by deletion of Oprm1 from the preBötC (18, 57). Fifty percent of Oprm1⁺ preBötC neurons are from the Dbx1^L lineage, therefore glutamatergic, and number between 150 and 250 depending on the method of labeling them (transgenic versus in situ) (18, 58). Furthermore, ∼50 of these Oprm1⁺Dbxl^L neurons are forkhead box protein P2 (Foxp2^L) derived. Because the preBötC is estimated to contain several thousand neurons, the Oprm1⁺Dbxl^L population comprises ∼5–10% and the Oprm1⁺Dbx1^LFoxp2^L population just ∼1–3%. Deletion of the Oprm1 gene from Dbxl^L neurons completely prevents suppression of the preBötC rhythm by Oprm1 agonists in vitro, and genetic deletion from Foxp2^L neurons largely prevents opioid-induced slowing (18, 58). These 200 or so Orpm1⁺ preBötC neurons have perhaps a privileged role in rhythm generation, as breathing dramatically slows when opioids silence these neurons. The remaining 50% of Oprm1⁺ neurons express vesicular GABA transporter (Vgat⁺) and are thus inhibitory.

These success stories should motivate similar approaches to dissect the diverse cell types that comprise other breathing control clusters in the brainstem and their varied roles in breathing behavior. This includes the PBN/KF, involved in postinspiration patterning, hyperventilation, apnea, arousal, and opioid-induced respiratory depression (3, 18, 57, 59, 60). In fact, these molecular concepts have already been used to begin to tease apart PBN/KF neurons, including calcitonin gene-related peptide (CGRP⁺) neurons that mediate hypercapnic arousal (59) and Oprm1⁺ neurons that pattern postinspiratory activity and basal respiratory rate as well as contribute to opioid-induced respiratory depression (57, 61–63).

THE QUEST TO IDENTIFY BREATHING RHYTHM-GENERATING NEURONS IN THE preBötC

The preBötC is essential for inspiratory rhythm generation and breathing (2, 64), and ablation of the several hundred glutamatergic Nk1R⁺ neurons within it (53, 65) is sufficient to halt breathing (49, 66). Most Nk1R⁺ neurons are commissural neurons or local preBötC interneurons (53, 65), and some have activity just prior to the onset of inspiration (preinspiratory neurons) (67). Furthermore, activation of NK1R with its cognate peptide speeds up breathing in vitro and in vivo (19, 68). These characteristics make preBötC Nk1R⁺ neurons prime candidates for key neurons to pace breathing. However, because Nk1R⁺ neurons are divisible, it is possible that just a subset represent the elusive breathing rhythm-generating neurons.

Germline mutants of Dbx1 and maf bzip transcription factor B (MafB) develop few, if any, Nk1R⁺ neurons, fail to breathe at birth, and do not produce a robust preBötC rhythm in vitro (32, 33, 69). Therefore, the Nk1R⁺ rhythm generators are likely to be Nk1R⁺Dbx1^LMafB^L. As noted above, one-third of Nk1R⁺ neurons are Sst⁺, thereby disqualifying them. Thus, the relevant 100–200 Nk1R⁺ neurons must be Nk1R⁺Dbx1^LMafB^LSst⁻. Additionally, because 60% of Nk1R⁺ neurons express Oprm1 (19) and profoundly slow breathing and the preBötC rhythm in vitro upon silencing by opiates (18, 57, 58), the candidate breathing pacers are likely Nk1R⁺Dbx1^LMafB^LSst⁻Oprm1⁺ (Figure 2). When coupled with modern intersectional mouse genetics, the molecular code approach will enable this important hypothesis to be rigorously tested.

ENDOGENOUS NEUROMODULATORY PATHWAYS CONTROL

THE preBötC RHYTHM

Although the preBötC can function autonomously, in reality it does not. The preBötC constantly receives signals that alter its speed, including from endogenous and exogenous neuromodulators. In particular, the action of exogenous opiates to slow, and ultimately stop, breathing has been the cause of nearly 900,000 deaths since the onset of the opioid epidemic in 1999. Opioid-induced respiratory depression mostly occurs by opioid-dependent silencing of preBötC Oprm1⁺ neurons, suggesting that these neurons are key to the generation of rhythmic breathing (18, 57).

Further examples include endogenous serotonin (5-HT) and substance P (SP), which are essential to stabilize breathing in postnatal life and prevent the devastating pathology of sudden infant death syndrome (SIDS). Ectopic application of 5-HT and SP can stimulate a fictive pre-BötC breathing rhythm in medullary slices, and their microinjection into the preBötC in vivo is sufficient to increase breathing rate (Figure 3a) (70–75). Some serotonergic neurons in the caudal medulla (raphé obscurus) also express SP, and transgenic expression of synaptically targeted green fluorescent protein into these 5-HT⁺SP⁺ neurons reveals monosynaptic connections with core components of the breathing CPG, including the preBötC and NTS as well as multiple cranial nerve nuclei (76, 77). Stimulation of the raphé obscurus in preBötC medullary slices and whole animal preparations consistently elicits 5-HT- and SP-dependent depolarization of some preBötC neurons (77–79) as well as an increase in breathing speed (19, 74, 77, 80, 81). Likewise, optogenetic stimulation of the raphé obscurus in awake animals (ePet-Cre neurons with ChR2) increases respiratory rate by 20%, an effect that is blunted by a 5-HT receptor antagonist (82) (Figure 3a). Thus, the raphé is sufficient for the stimulation of breathing by the action of neuromodulators on a population of preBötC neurons.

Figure 3.

Idealized data demonstrating endogenous modulation of breathing rate by 5-HT- and SP-expressing raphé neurons. (a) Ectopic application of 5-HT and SP to the preBötC in vitro increases the rate of the inspiratory rhythm. Likewise, stimulation of the raphé in vitro and in vivo increases the neuromodulatory effects of 5-HT and SP. (b) 5-HT and SP antagonists slow the preBötC in vitro rhythm. In combination, the rhythm is completely silenced. (c) Neonatal mice harboring mutations that prevent the development of serotonergic neurons display many more apneas (~ 30% of the time) than do control mice (~ 8% of the time; dashed black line). Systemic injection of a 5-HT_2A agonist reduces apneas to near wild-type rates. (d) Model of the endogenous neuromodulation of breathing rate by serotonergic neurons during the transition from fetal to postnatal stages. Neurons within the preBötC that express both 5-HT_2A and Nk1R receptors represent a subset of Nk1R⁺ neurons that are inspiratory rhythm generators. Abbreviations: 5-HT, serotonin; 5-HT_2A, serotonin receptor 2a; Nk1R, neurokinin 1 receptor; preBötC, preBötzinger complex; SIDS, sudden infant death syndrome; SP, substance P.

Because 5-HT and SP synergistically increase the speed of the preBötC rhythm in vitro and breathing in vivo, it is instructive to consider the effect of their antagonism. Antagonists to the serotonin receptor 2A (5-HT_2A) partially slow the preBötC, and subsequent Nk1R antagonism completely eliminates the inspiratory rhythm (77, 80, 83) (Figure 3b). This effect can be reversed by the ectopic application of either neuromodulator to the preBötC. Furthermore, the slow breathing observed in genetic neonatal mouse models that lack serotonergic neurons can be rescued by activation of 5-HT_2A and Nk1R receptors (84). These pharmacological and genetic dissections demonstrate how the endogenous release of 5-HT and SP, presumably from the same raphé neurons, is necessary to drive the preBötC rhythm and thus stabilize postnatal breathing by tonic depolarization of 5-HT_2A- and Nk1R-expressing cells. The candidate rhythm-generating neurons described above (Nk1R⁺Dbx1^LMafB^LSst⁻Oprm1⁺) express Nk1R receptors (Figure 3d); thus, it will be important to determine whether they also respond to 5-HT via 5-HT_2A.

Consistent with this hypothesis, genetic mutations that prevent the development of serotonergic neurons cause breathing to slow in the neonatal period and reduce animal viability. Germline deletion of the FEV transcription factor (Pet-1) causes a 70% decrease in the number of serotonergic neurons, accompanied by a 30–50% reduction in postnatal breathing rate and preBötC rhythm in vitro compared to control littermates. These changes are believed to underlie the 30% mortality in Pet-1^−/− neonates (85). Similarly, deletion of Lim homeodomain transcription factor 1β (Lmx1b) from the Pet-1 lineage eliminates all serotonergic neurons and results in neonates that display multiple apneas (pauses in breathing), sometimes lasting up to 35 s, together with a 50% decrease in breathing rate and reduced preBötC rhythm in vitro (84). These effects are reversed by a 5-HT_2A agonist (Figure 3c). Importantly, a null mutation for a key enzyme in 5-HT synthesis, tryptophan hydroxylase-2 (Tph2), increases neonatal mortality, decreases breathing rate, and causes irregular respiration with long pauses or apneas (86–88). These breathing abnormalities are reversed with systemic delivery of a 5-HT precursor (87). Combined, these studies validate the concept that serotonin, and probably other neuromodulators released by raphé neurons such as SP, augment breathing in the early postnatal period.

Like manipulations that deplete serotonergic neurons, genetic deletion of the ion channel that mediates SP-induced membrane depolarization, sodium leak channel nonselective (Nalcn), is neonatal lethal, and its conditional deletion from the Dbx1^L lineage is ∼30% lethal (78, 79). Neonates with a Dbx1^L–Nalcn deletion display more apneas, reduced breathing frequency, and slower in vitro preBötC rhythm, mirroring deletion of raphé neurons or disrupted 5-HT synthesis. This suggests that SP has a role similar to 5-HT in supporting neonatal respiration, consistent with a model where the caudal raphé neurons supply both 5-HT and SP to the preBötC to promote breathing in neonates. However, an important experiment that has yet to be carried out is the selective deletion of SP from 5-HT neurons.

SIDS, the unexpected death of an infant within the first year of life that usually occurs during sleep, appears to be due, at least in part, to the inability of the respiratory system to spontaneously respond to challenges like hypoxia (autoresuscitation). The brainstem serotonergic system has been genetically and molecularly linked to SIDS. Moreover, murine neonatal studies have shown that inhibition of serotonergic neurons, or genetic mutations that fail to produce serotonin, prevent autoresuscitation following anoxic challenges (85, 88–91). The critical role of 5-HT and SP in stimulating the preBötC and breathing in neonates suggests a plausible mechanism to explain the pathophysiology of SIDS. Importantly, this understanding paves the way for the development of pharmacological stimulants for breathing that might be given to infants at risk, such as those born prematurely.

THE SENSORY COMPONENTS OF THE BREATHING CONTROL CIRCUIT CONTAIN MULTIPLE NEURONAL TYPES

An important component of the breathing control circuit is the sensory feedback from the lungs that continually modulates breathing. These peripheral signals are critical to prevent the lung from being overstretched or to change breathing in response to chemical irritants. A simple model is that the key breathing rhythm-generating neurons (likely Nk1R⁺Dbx1^LMafB^LSst⁻Oprm1⁺) ultimately receive this sensory input and modify breathing accordingly. Recently, a comprehensive molecular characterization of pulmonary sensory neurons has been completed (described below), providing a framework to understand how regulatory reflexes control the breathing CPG in the brainstem.

Historically, pulmonary sensory neurons were broken into multiple cell types: fast-conducting A-type fibers, further differentiated by lung innervation patterns and slow or rapid adaptation to lung stretch (slowly adapting receptors, SARs, in smooth muscle and rapidly adapting receptors, RARs, in subepithelial tissue) (92–95), and at least two slow-conducting unmyelinated C-type fibers (largely capsaicin sensitive) innervating either the larger airways (the trachea and bronchi) or the alveoli (pulmonary parenchyma) (96, 97). C-fibers, which represent ∼75% of sensory neurons in the pulmonary vagus nerve (98), and also some RARs respond to chemical irritants and endogenous inflammatory molecules, either directly or indirectly (94, 95). The responses generated are protective in nature, for example, causing breathing to become rapid and shallow during lung inflammation and edema, triggering deep sighs to stretch a noncompliant lung, or eliciting coughing to expel an irritant (95, 99). In contrast, the lung stretch/inflation signal from SARs serves as regulatory feedback, for example, to terminate ongoing inspiration and/or prolong expiration, depending on lung volume (100–102). However, this sensory neuron classification is a simplification, as anatomical, physiological, and molecular diversities have been described within each category (97, 102, 103). The physiology of pulmonary sensory neurons has been reviewed in detail elsewhere (95, 99, 104–108).

Pulmonary vagal sensory neurons reside in the jugular (superior) and nodose (inferior) vagal ganglia. Neurons in these two locations are physiologically distinct and regulate different pulmonary reflexes (109). Jugular neurons preferentially innervate extrapulmonary sites (such as the trachea and larynx), signal to the paratrigeminal brainstem nucleus (110, 111), are molecularly defined by genes like PR/SET domain 12 (Prdm12⁺) (112, 113), and derive from the neural crest developmental lineage (114). In comparison, nodose neurons monitor both the extra- and intra-pulmonary status, relay sensory signals to the NTS (110, 111), express markers like purinergic receptor P2X 2 (P2rx2⁺) (115), and derive from the Phox2b^L visceral nervous system lineage (116) (Figure 4). Beyond the lung, each ganglion also innervates other structures or organs such as the ear (jugular), gastrointestinal system (nodose), and heart (both) (117).

Figure 4.

Molecular subtypes of vagal sensory neurons that modulate breathing. (a) Receptive fields for nodose (red) and jugular (dark gray) vagal neurons that originate in the trachea (external to the lung) and also within the lung. Also represented are the sensory neuronal types within each ganglion and their differential inputs to the medullary brainstem. The nodose signals are integrated within the NTS, which broadly projects to the breathing central pattern generator. The jugular sensory neurons signal to the trigeminal sensory system via the para5. (b) Molecular identification of vagal neurons groups clustered by sensory neuronal type and breathing physiology phenotype from experimental studies. Within each cluster, the neuronal subsets are annotated by the cluster number used in the scRNA-seq studies in References 112 (black) and 113 (gray, italics). Dashed red outline represents the hypothesized primary lung stretch neurons that mediate the Hering-Breuer reflex (Lamp5⁺). Abbreviations: A, pulmonary sensory A-fiber; BötC, Bötzinger complex; C, pulmonary sensory C-fiber; ChR2, channel rhodopsin 2; Lamp5, lysosomal-associated membrane protein family member 5; Npy2r, neuropeptide Y receptor Y2; NTS, nucleus tractus solitarius; P2ry1, P2Y purinoceptor 1; para5, paratrigeminal nucleus; PBN/KF, parabrachial nucleus/Kölliker-Fuse complex; Phox2b, paired-like homeobox 2b; Piezo2, piezo-type mechanosensitive ion channel component 2; Prdm12, pr/set domain 12; preBötC, preBötzinger complex; RAR, rapidly adapting pulmonary stretch receptor; RTN, retrotrapezoid nucleus; rVRG, rostral ventral respiratory group; SAR, slowly adapting pulmonary stretch receptor; scRNA-seq, single-cell RNA sequencing; SP, substance P; Trpv1, transient receptor potential cation channel subfamily V member 1. Symbols: ⁺, gene expressed; ⁻, gene not expressed; ^L, developmental lineage.

MOLECULAR CLASSIFICATION OF PULMONARY VAGAL SUBTYPES ENABLES FUNCTIONAL CHARACTERIZATION

Physiological studies have shown that sensory fibers within each classically defined group respond to distinct stimuli, leading to the hypothesis that each fiber type can be further subdivided (97, 111). Only recently has this idea been comprehensively explored by unbiased classification of vagal sensory neurons using transcriptional profiles. Clustering algorithms of single-cell transcripts obtained from single-cell RNA sequencing (scRNA-seq) are able to distinguish neurons from the jugular and nodose ganglia and also many subgroups within each (112–118). The complementary studies identified six groups of jugular neurons, three of which could be further divided, suggesting a total of ten jugular cell types (Supplemental Figure 2). The nodose ganglion is composed of at least 12 to 18 neuronal groups, which upon further subdivision reveal a total of 27 to 52 possible cell types (Supplemental Figure 2). Building upon these global characterizations of vagal sensory neurons, scRNA-seq studies specific to sensory neurons in the lung have identified up to six cell groups (115, 118), which upon alignment with the unbiased studies described above, suggest a total of ∼15 to 30 types of pulmonary sensory neurons (5–10 jugular and 10–20 nodose) (Figure 4). One clear outcome of these studies is that C-fibers can be molecularly identified by markers like the transient receptor potential cation channel subfamily V member 1 (Trpv1⁺) and low expression of neurofilament light chain (Nefl^Low). In addition, most RARs and SARs can be identified by piezo-type mechanosensitive ion channel component 2 (Piezo2⁺, which bestows stretch sensitivity), P2Y purinoceptor 1 (P2ry1⁺), and Nefl^High. Combined with genetic tools, these molecular identities have enabled the identification of specialized roles for several sensory cell types in murine breathing control. Two examples are described below.

P2ry1⁺ Neurons Stop Breathing

P2ry1⁺ is expressed in three pulmonary sensory neuron types that are also Nefl^High and Piezo2⁺, suggesting they may be mechanosensitive. However, these neurons are not responsive to lung stretch, as necessitated for SARs, so their mechanosensitivity appears to serve a separate, unknown purpose. Physiological characterization of P2ry1⁺ fibers reveals they have fast transduction speeds and lack responses to capsaicin, demonstrating their Trvp1⁻ (A-fiber) identity (119). These neurons course along or below the smooth muscle of the lung’s airways and innervate clusters of sensory cells termed neuroepithelial bodies (118, 119). They project to medial and lateral subregions of the NTS. Despite this, a role for these P2ry1⁺ pulmonary sensory neurons in regulation of breathing remains undefined. Lastly, P2ry1⁺ neuronal types also innervate the larynx and function as sensors that can trigger swallowing and pause breathing (termed apnea) (113, 119).

Npy2r⁺ Neurons Make Breathing Rapid and Shallow

Neuropeptide Y receptor Y2 (Npy2r⁺) neurons are also Nefl^Low, Trvp1⁺, and slow conducting (119), confirming their identity as C-fibers. Like pulmonary Trvp1⁺ neurons, they course along or beneath the smooth muscle of the airways to the alveoli and project to the medial NTS compartment (118, 119, 120). Optogenetic excitation of these neurons elicits rapid and shallow breathing, as anticipated from classical studies of C-fibers (109). Future loss-of-function manipulations and physiological characterization will be required to reveal the protective role that they play in breathing.

THE NUCLEUS TRACTUS SOLITARIUS RELAYS PERIPHERAL SENSORY SIGNALS TO THE BREATHING CENTRAL PATTERN GENERATOR

Pulmonary jugular vagal sensory afferents primarily enter the central nervous system via the NTS, distributing their signals to the various parts of the brainstem breathing center. The NTS forms a column that spans the dorsal medulla, and lung afferents innervate the intermediate and caudal portions. SARs, RARs, and C-fibers have distinct innervation patterns in the NTS, reflecting their different effects on breathing. In turn, the various compartments of the NTS have subtly different patterns of projection to the breathing control groups (121). For example, the lateral NTS at the level of area postrema projects to the entire breathing control circuit and regulates reflexes like the Hering-Breuer apnea reflex initiated by lung stretch (122). In contrast, the caudal NTS projects to select areas of the breathing circuit, including the RTN, rVRG, and cVRG, but not the preBötC rhythm generator (41, 121). Thus, any impact of the caudal NTS on breathing speed must be indirect. Respiratory responses mediated by the caudal NTS include the lung chemoreflex stimulated by irritants like capsaicin (apnea followed by rapid, shallow breathing) (8, 95, 123) and slowing of breathing following 5-HT activation of C-fibers (124–126). Future studies of the second-order NTS neurons targets throughout the breathing control circuit may reveal unique projection patterns for neurons receiving specific pulmonary signals. Ultimately, defining neurons that target the CPG will reveal the cell types that have important functions in breathing control, such as the control of speed or depth of breath. Examples of NTS cell types that transform sensory signals from the lung are described below and also reviewed elsewhere (8, 105). None of these neurons are molecularly defined and thus largely uncharacterized.

1. SAR afferents activate pump cells (P-cells) and inspiratory-β (Iβ) cells by releasing glutamate and adenosine in the NTS at the level of the area postrema (127–129). P-cells are active during lung inflation, mirroring their excitatory SAR input, and send excitatory or inhibitory signals to the caudal commissural and contralateral regions of the NTS, the PBN/KF, the noradrenergic A5 region, and medullary breathing control centers including the RTN, BötC, preBötC, and rVRG (105, 125, 130–132).

2. Iβ cells are excited by SARs and also by a corollary (efferent) signal from the medullary breathing control circuit and project to the spinal cord where they monosynaptically excite motor neurons for the diaphragm (133, 134). What role they play in breathing needs investigation, particularly as Iβ cells have largely been characterized in cats and dogs but not rodents.

3. Unlike the limited innervation pattern of the NTS by SARs, individual RARs project to multiple NTS regions, reflecting their multifunctionality in breathing control (94, 135–137). The densest projections are to the caudal and medial NTS, which is not involved in the Hering-Breuer effect, the classic stretch receptor response (122, 138, 139). NTS neurons that receive RAR input exhibit spontaneous activity that is modulated by lung volume (inflation and deflation) and a central breathing signal (such as Iβ cells). However, unlike SARs, RARs are only sensitive to large or rapid lung inflation. NTS RARs project to the PBN/KF and the medullary premotor breathing control centers, rVRG and cVRG (138, 140). Therefore, any impact on breathing rhythm is via an indirect signal to the preBötC. Finally, it is presumed that NTS RARs excite the largely uncharacterized Iγ cells, which transmit lung deflation signals to the spinal cord (141).

4. C-fibers terminate within the medial ipsilateral and contralateral NTS. This bilateral innervation is significant, as silencing the caudal contralateral NTS pathway impairs the reflexive breathing changes that are associated with C-fiber excitation (126, 142). The second-order NTS neurons for C-fibers remain undefined.

THE HERING-BREUER REFLEX IS A MODEL OF THE INTERACTION BETWEEN PERIPHERAL AND CENTRAL NEURONS

The Hering-Breuer reflex (HBR) prevents overstretching of the lung by terminating an ongoing inspiration (102). If inflation is sustained into expiration, such as during airway obstruction, the HBR lengthens expiration. At its extreme, the HBR halts breathing during maintained inflation of the lung by completely suppressing the rhythmogenic capacity of the central breathing control circuit. Early work by Knowlton & Larrabee in 1946 (92) revealed that lung stretch is sensed by SARs. More recent work that we describe below has elucidated the mechanism of mechanosensation, a molecular definition of the sensory neurons, and the NTS targets that appear to silence breathing.

The mechanosensitive ion channel Piezo2 is expressed by nodose and jugular pulmonary sensory neurons, mostly fast conducting A-fibers (143). Optogenetic excitation of Piezo2⁺ neurons slows, and sometimes stops, breathing, mirroring the HBR. Conditional genetic deletion of Piezo2 from the Phox2b lineage eliminates both the HBR and vagal nerve responses to lung stretch. Combined, these data show that the HBR is mediated by Piezo2⁺ SARs in the nodose ganglia and that lung stretch is directly sensed by the Piezo2 ion channel. Furthermore, the volume of air inspired during basal breathing increases after conditional deletion of Piezo2 from adult sensory neurons, demonstrating that lung stretch and SARs play an ongoing role in the regulation of basal breathing.

It has been postulated that a single class of SAR afferent mediates the HBR. Can a candidate be identified from scRNA-seq studies? Many fast-conducting nodose sensory neurons from the lung express P2ry1⁺, yet these neurons appear to be insensitive to ectopic lung stretch (113). In addition, their optogenetic excitation induces swallowing, and they do not innervate the NTS region implicated in the HBR (119, 122). These results indicate that the P2ry1⁺ subset in the nodose ganglion does not mediate the HBR. Remarkably, most of the Piezo2⁺ nodose sensory neurons are P2ry1⁺, narrowing candidate sensory neurons for the HBR to perhaps one cell type, namely the cell type denoted as NP18 (113), NG 9 (112), or C3 (118) in the scRNA-seq studies to date (defined as P2ry1⁻Piezo2⁺Lamp5⁺) (Figure 4). And consistently, P2ry1⁻Piezo2⁺Lamp5⁺ compose most of the stretch-responsive sensory neurons (113, 118). This hypothesis could be tested by independently deleting Piezo2 from P2ry1⁺ and Lamp5⁺ neurons, although it is possible that the HBR requires stretch signaling from multiple sensory neuron types in the lung.

The second stage of the HBR is transmission of the sensory stretch signal to second-order P-cells within the NTS, whose activity patterns mimic those of SARs. Although P-cells can be found in multiple regions of the NTS, excitation of the region immediately medial to the solitary tract at the level of the area postrema causes apnea that resembles the HBR. In addition, when excitatory synaptic signaling is blocked in this region, breathing becomes slow and deep like that following vagotomy (122). This portion of the NTS displays rhythmic release of glutamate and adenosine (signaling molecules that control the rhythmic activity of P-cells), which is ultimately due to lung inflation and dependent on the vagal sensory neuron input (127–129). Furthermore, lesions of other NTS regions, including lateral and caudal compartments, do not eliminate the HBR (138, 139). Thus, the medial NTS at the level of the area postrema appears to be the region that mediates the HBR. However, it will be critical to discover molecular markers for the P-cells in this area to selectively manipulate them and determine whether they are necessary and sufficient for the HBR.

The P-cell population appears to be a mixture of excitatory and inhibitory neurons (144) that project to multiple components of the brainstem breathing control circuit (105, 130–132). Indeed, electrophysiological recordings have revealed that these compartments receive excitatory and inhibitory synaptic inputs coincident with lung stretch, SAR activity, and P-cell activity (145). Pharmacological blockade of GABAergic and glycinergic signaling within the BötC (expiratory center) and preBötC (inspiratory center) eliminates the HBR, as does optogenetic inhibition of BötC and preBötC glycinergic neurons (146, 147). This suggests that excitation of the NTS ultimately drives inhibition within the preBötC to suppress the inspiratory rhythm. Other brainstem breathing centers that are modulated by P-cells likely play a modulatory role during the HBR, including sensitivity tuning of the BötC and preBötC by neurons within the PBN (148), or regulate other parts of the upper airway such as the trachea (149).

Although the HBR was discovered in the late 1860s, it has remained a model breathing reflex to this day. Originally, it was hypothesized that lung stretch activates NTS P-cells, which excite glycinergic BötC neurons that are active in the early phase of expiration, thus suppressing the preBötC and inspiration to cause HBR apnea. Although the critical molecular mechanism for sensing lung stretch has been identified, the entire pathway of cells involved in the reflex arc is only partially known (Figure 5). Molecular identification of NTS P-cells and the brainstem breathing control neurons they directly modulate is the next important step. It is also important to understand why neuronal activity in the majority of the brainstem breathing control circuit is modulated by lung stretch, and what purpose this sensory signal serves beyond the HBR. Perhaps some of these breathing control centers relay sensory information from the lung to the higher brain to arouse the feeling of shortness of breath. The study of reflexes that have profound effects on breathing rhythm, such as the HBR and apnea, will yield the identity of key neurons that generate this rhythm. It is possible, for example, that Nk1R⁺Dbx1^LMafB^LSst⁻Oprm1⁺ preBötC neurons are the cellular targets of the HBR that are silenced to induce apnea. It seems logical that the preBötC neurons targeted by the reflex to stop breathing would also be the cells that are used to generate it. It will also be interesting to determine whether the same cells are targeted by distinct reflexes that stop breathing, such as during apnea when we dive under water or inhale ammonia.

Figure 5.

Transmission of the lung stretch sensory signal that initiates the Hering-Breuer reflex and halts breathing. The cell types and respective activity during the breath (a) and the anatomical location and signaling hierarchy (b) of the neurons that transmit lung stretch to the preBötC and lead to the Hering-Breuer reflex during sustained lung inflation. (a) Activity of the NTS P-cells (red circle), BötC expiratory neurons (green), and preBötC inspiratory neurons (blue) with respect to the breathing airflow (black). (b) the presumed Hering-Breuer reflex arc. The sustained lung inflation is sensed by Piezo2⁺ nodose neuron ①, which excites the P-cells via glutamate and ATP ②, which maintain the BötC neural activity via glutamate ③ that is presumed to suppress the preBötC with glycine and thereby cause apnea ④. This reflex arc is presumed because the molecular identities of the P-cells, BötC, and preBötC neurons involved in the reflex remain undefined. Abbreviations: BötC, Bötzinger complex; Glyt2, glycine transporter 2; NTS-L, nucleus tractus solitarius lateral; NTS-M, nucleus tractus solitarius medial; P2ry1, P2Y purinoceptor 1; P-cell, pump cell; Phox2b, paired-like homeobox 2b; Piezo2, piezo-type mechanosensitive ion channel component 2; preBötC, preBötzinger complex; SAR, slowly adapting pulmonary stretch receptor. Symbols: ⁺, gene expressed;⁻, gene not expressed; ^L, developmental lineage.

CONCLUSIONS

Breathing is mostly passive and therefore something that we typically ignore, but when our physiology is challenged, it becomes hard to focus on anything else. Breathing has the remarkable ability to direct and amplify emotions like anxiety and calmness, demonstrating its powerful influence on brain-wide activity. The breathing CPG is a valuable model system to study the fundamental principles of systems neuroscience. Its limited size and autonomous rhythmic activity make it an ideal mammalian node to study how cellular biophysical properties and circuit connectivity interact to produce a behavior, reminiscent of the important contributions that studies of the stomatogastric ganglion have made to neuroscience. Furthermore, breathing is a perfect system to study the temporal dynamics and impact of neuromodulation. An actively growing field that was only briefly touched upon in this review is the exploration of how the breathing CPG integrates with other patterning systems that utilize the airway system, including chewing, swallowing, and vocalization. Future studies of breathing should apply the understanding gained from decades of basic research to identify the cellular and molecular mechanisms of major societal issues such as the abuse of opioids, common afflictions like obstructive sleep apnea, and devastating events like SIDS. Ultimately, a cellular and molecular understanding of breathing will enable the development of novel pharmacological approaches that could revolutionize medical fields such as anesthesia, critical care, and neonatology.