The integrated brain network that controls respiration

1From the first cry to the last gasp, the respiratory system should never fail to supply sufficient oxygen to meet metabolic demands during every possible event throughout life. The respiratory pattern is not only determined by physical activity, but also reflects the emotional state, and volitional control of respiration can be used to alter affection and reduce stress. Indeed, multiple behaviors, such as swimming, playing musical instruments, parturition, or meditation depend on precise respiratory control, and for many sports and arts, it is often the control of respiration that separates mediocre from top performance. Thus, indeed, respiratory control affects all aspects of life.

2Although control over respiration can be voluntary, most of it is subconscious, even during voluntary respiration. In this review, we discuss the integrated network of brain regions most involved in the control of respiration, their connections, and possible clinical consequences of their pathology. Throughout, we discuss how respiratory control and other motor and non-motor systems interact.

3Despite their vast differences in body size and ecological niches, mammals possess similar basic mechanics of ventilation, with some variations between species or sexes. Both lung and tidal volumes scale linearly with body weight, and the higher metabolic rate of smaller mammals is accounted for by a faster respiratory rate. The force required for inspiration is delivered by so-called pump muscles that expand the rib cage. The diaphragm and external intercostal muscles are the most prominent inspiratory pump muscles, but also parasternal intercostal, sternocleidomastoid and scalene muscles can act as such. During active expiration, expiratory pump muscles, in particular the internal intercostal and abdominal muscles, are active. The diaphragm is under control of motor neurons in the phrenic nucleus located in the anterior ramus of the third to sixth cervical vertebrae. The intercostal and abdominal muscles are innervated from the thoracic spinal cord (Figure 1A–C).

4Valve muscles regulate the air flow by adjusting the resistance of the upper airways. Activity of the facial nucleus can lead to opening of the nasal valve, via the dilator naris anterior muscle and the alar part of the nasal muscle. Contractions of these muscles do not only facilitate inspiration, but can also relate to sniffing. Just before the start of inspiration, motor neurons of the hypoglossal nucleus activate tongue muscles, reducing collapsibility of the pharynx. Indeed, tongue deformation, for example as a consequence of excessive fat depositions in obesity, can be related to obstructive sleep apnea. Activation of the trigeminal motor nucleus can contribute to jaw movements. Finally, contractions of the larynx muscles show a bimodal pattern; initially, they are dilated to allow airflow into the lungs, while at the end of inspiration these muscles contract to reduce the outflow of air, prolonging the period of gas exchange. Laryngeal constriction is controlled by the nucleus ambiguus via the vagus nerve. The nucleus ambiguus houses also motor neurons controlling swallowing, the control of which is strongly coupled to that of respiration.

5Respiratory muscles typically serve multiple functions. Rib cage muscles, for instance, control both respiration and arm movements, so that locomotion and respiration are tightly coupled during quadrupedal locomotion. Indeed, one could argue that the change from quadrupedal to bipedal locomotion during hominid evolution paved the way for the intricate breathing control required for human speech.

6Below, we describe the brain areas most involved in subconscious control of respiration, and their main connections. Most of these pathways are bilateral, but ipsi- and contralateral projections can differ in strength or in their ratio between excitatory and inhibitory fibers, potentially introducing asymmetries in motor activity. As detailed studies on monosynaptic projections are sparse in humans, we base our summary on animal studies, with earlier descriptions mostly concerning cats, and more recent ones often performed in rats or mice (Figure 2, Supplementary file 1). Connections labeled as sparse in the original papers are not included in this overview.

7Many brain regions lack clear borders. In particular when different species are compared, this may lead to some variations in the interpretation of anatomical projections. On top of this, one should also take into account that anatomy and physiology do not always match. For instance, the inspiratory neurons originally considered to be located in the pre-Bötzinger complex are actually distributed around the region of the pre-Bötzinger complex and are partially intermingled with expiratory neurons originally considered to be located in more caudal nuclei. When reading this review, please note that the use of anatomical names is to help orient oneself, but in reality, borders are often fuzzy. Genetic markers may help to define more homogeneous populations of neurons, and when this information was available, we mention that in the text and figures.

8Given that the neuronal mechanisms of respiratory rhythm generation are evolutionary well conserved, inter-species differences are expected to be relatively minor. Important exceptions, however, are the elongation of the pharyngeal region, and the development of complex muscle control of pharynx and mouth related to human speech. Since neural control of speech is outside the scope of this review, this will not be further discussed.

9Quiet breathing, or eupnea, is a rhythmic alternation between inspiration and passive expiration or post-inspiration (; Figure 1G). During periods with higher metabolic demand, hyperpnea occurs, which entails also active expiration (;; Figure 1D–E). It must be noted that the different respiratory muscles cover different aspects of the respiratory cycle. As a result, a clear distinction between passive and active expiration cannot be made. For example, during passive expiration, some of the expiratory muscles can be active, and active expiration encompasses passive elastic contractions as well. In humans, post-inspiration can contribute to longer periods of relatively constant air flow as required for speech or singing (;;; Figure 1F). Here, additional muscles become active that are not active during solely post-inspiration. Thus, to what extent the respiratory cycle within the brainstem can indeed be divided into rhythmogenic phases or whether these phases are just seen at the motor output remains topic of further research. As stated, we discuss in this review multiple mechanisms that modulate the rhythmicity of respiration when confronted with respiratory challenges, such as a change in air pressure or increase in CO2 concentration (hypercapnia).

10During regular breathing, the respiratory cycle is determined by brainstem central pattern generators. Inspiration is triggered by activity of neurons in the pre-Bötzinger complex, most of which fire in phase with inspiration and indirectly drive the inspiratory pump muscles. Inspiration may be terminated by activation of the Kölliker-Fuse nucleus, and possibly also the postinspiratory complex can contribute to this. Finally, the lateral parafacial nucleus that is silent during eupnea becomes active during active expiration (;;; Figure 3). Without input from the brainstem, the spinal circuitry cannot organize respiration. It does contribute to sequential contraction of thoracic muscles to optimize the inflow of air by adjusting muscle control to body biomechanics.

11The pre-Bötzinger complex contains a network of neurons generating rhythmic activity that is both essential and sufficient to drive inspiration (;;;;;;; Figure 3B). In rodents, the pre-Bötzinger complex consists of around 3,000 neurons on each side of the brain, with approximately equal numbers of excitatory and inhibitory neurons. Inspiratory neurons with comparable dynamics may also be found in surrounding regions of the ventral respiratory column, pointing toward a more diffuse spatiotemporal network than originally described. The rhythmogenic kernel is formed by somatostatin-negative (SST-) excitatory interneurons. Excitatory neurons in the pre-Bötzinger complex have a refractory period that prevents them from generating bursts at a high frequency, and this refractory period can be shortened by inhibitory input. Hence, although inhibitory interneurons are not essential for generating rhythmicity, they may modulate the breathing frequency and contribute to the termination of inspiration. The output of the pre-Bötzinger complex is composed of both inhibitory and SST+ excitatory neurons.

12Although there are direct projections from the pre-Bötzinger complex to multiple respiratory motor nuclei, indirect projections appear to be more common. As such, the phrenic nucleus is predominantly targeted via the rostral ventral respiratory group (rVRG), and the thoracic motor neurons that control abdominal muscles via the caudal ventral respiratory group (cVRG). In addition, the hypoglossal nucleus is principally reached via the parahypoglossal region of the reticular formation, and the facial nucleus via the intermediate reticular formation.

13In addition, there are substantial projections to other respiratory control areas: the Bötzinger complex, Kölliker-Fuse nucleus, postinspiratory complex (PiCo), and lateral parafacial nucleus. Furthermore, also the retrotrapezoid nucleus, nucleus tractus solitarii (NTS), lateral and dorsomedial hypothalamus, lateral and medial parabrachial nuclei, and periaqueductal gray are targeted. A specific subset of Cdh9-neurons projects to noradrenergic neurons in the locus coeruleus. Finally, there are strong projections to the contralateral pre-Bötzinger complex to promote left/right synchrony during respiration.

14The adjacent Bötzinger complex houses mostly inhibitory neurons that either show decrementing activity during post-inspiration or incrementing activity during expiration, and both groups contribute to the inhibition of inspiratory activity in the pre-Bötzinger complex during expiration. These activity patterns partially reflect sensory feedback: while augmenting neurons are inhibited by lung inflation, decrementing neurons are excited by it. Next to the adjacent pre-Bötzinger complex, also other respiratory control centers are innervated by the Bötzinger complex: the Kölliker-Fuse and lateral parafacial nuclei.

15The Bötzinger complex also inhibits premotor areas: rVRG and to a lesser extent also cVRG, and directly inhibits motor neurons in the phrenic nucleus. Other targets are the NTS, lateral parabrachial nucleus and periaqueductal gray.

16As the respiratory pattern has to be coordinated with ongoing behavior, the pre-Bötzinger complex receives input from many brain regions, like the Kölliker-Fuse nucleus, PiCo, cVRG, NTS, retrotrapezoid nucleus, locus coeruleus, caudal raphe, lateral and paraventricular hypothalamus, central amygdala, lateral and medial parabrachial nuclei, periaqueductal gray, spinal trigeminal nuclei, and reticular formation. The pre-Bötzinger complex receives also direct input from the forebrain, including several regions of the neocortex presumably involved in voluntary control of respiration, but these connections are relatively sparse. The inputs of the Bötzinger complex are similar to those of the pre-Bötzinger complex, although less widespread (; Supplementary file 1).

17The pre-Bötzinger complex is not the only area essential for rhythmic respiration. Selective damage to the brainstem at the level of the pons leads to impaired transition from inspiration to expiration, resulting in prolonged periods of inspiration, a condition called apneusis. This effect was later localized in parts of the parabrachial complex, initially referred to as pneumotaxic center, and later as the pontine respiratory group. The parabrachial complex is composed of the lateral and medial parabrachial nuclei, that have predominantly ascending projections carrying sensory information, and the Kölliker-Fuse nucleus that primarily targets subcortical structures. Accordingly, of the parabrachial complex it is mainly the Kölliker-Fuse nucleus that influences the switch from inspiration to expiration (;; Figure 3B). Neurons of the Kölliker-Fuse nucleus show activity related to specific phases of respiration, with most neurons being active during inspiration; these latter neurons abruptly stop firing at the end of inspiration, marking the inspiration-expiration transition. A bilateral block of activity in the Kölliker-Fuse nucleus prolonged the inspiratory activity of the phrenic nerve in an in situ preparation, but did not completely block the termination of inspiratory activity, which would be in line with a prominent, but not exclusive role of the Kölliker-Fuse nucleus for the termination of inspiration.

18The Kölliker-Fuse nucleus receives input from the pre-Bötzinger and Bötzinger complexes, and from several central chemoreceptor areas: the NTS, retrotrapezoid nucleus, and cerebellar fastigial nucleus. In addition, the Kölliker-Fuse nucleus also receives input from the rVRG, cVRG, periaqueductal gray, spinal trigeminal nucleus, paratrigeminal nucleus, pedunculopontine tegmental nucleus (PPTg), and vestibular nuclei. Finally, there are descending inputs from the lateral, dorsomedial and paraventricular hypothalamus.

19Glutamatergic projections directly and indirectly (via the rVRG) target the phrenic nucleus, as well as the ambiguus, hypoglossal and facial nuclei. The latter connections allow premotor neurons in the Kölliker-Fuse nucleus to constrict valve muscles, reducing outflow during post-inspiration.

20Further excitatory projections target, in addition to the other nuclei of the parabrachial complex, the pre-Bötzinger complex, PiCo, and lateral parafacial nucleus. Also the cVRG, reticular formation, retrotrapezoid nucleus, NTS, and periaqueductal gray receive glutamatergic input. The caudal part of the Kölliker-Fuse nucleus sends inhibitory projections mainly to the sensory trigeminal nucleus, but also to the dorsomedial hypothalamus. Finally, there are projections to the raphe nuclei, vestibular nuclei, and cerebellar cortex.

21Recently, a second brain region involved in controlling post-inspiration has been identified in the ventral part of the intermediate reticular formation: the postinspiratory complex, or PiCo, consisting of cholinergic neurons. Isolated in a tissue slice, the PiCo can generate rhythmic activity that peaks during post-inspiration, and thus potentially contributes to a biphasic (pre-Bötzinger complex – PiCo) or triphasic (pre-Bötzinger complex – PiCo – parafacial nucleus) oscillator controlling respiration. Optogenetic stimulation in vivo confirmed that PiCo activity can prolong post-inspiration.

22The role of the PiCo in controlling post-inspiration has not yet been fully investigated. Although this is not a proof against a coordinating role of the PiCo for post-inspiration, systematic recordings in a whole-brainstem preparation found widespread activity during post-inspiration, but more in the pontine region than focused in the area of the PiCo. Furthermore, inhibition of the PiCo did not affect the duration of inspiration, while in the same study the PiCo was shown to be involved in gating swallowing motor patterns to the respiratory system. In conclusion, it seems likely that the PiCo is involved in the neural control of post-inspiration, but probably more as part on an integrated network than as primary pattern generator.

23The PiCo receives strong input from the Kölliker-Fuse nucleus and periaqueductal gray, but there are also substantial connections from the caudal and intermediate NTS and the hypothalamic paraventricular nucleus. Also the pre-Bötzinger complex projects to the PiCo. As far as we are aware, there are no systematic studies on PiCo efferents, but projections to the pre-Bötzinger complex and retrotrapezoid nucleus have been demonstrated.

24At the end of the respiratory cycle, the lateral parafacial nucleus can trigger active expiration by recruiting expiratory abdominal muscles via the cVRG. In the literature, the term parafacial respiratory group is sometimes used as synonym for the lateral parafacial nucleus, or for the combination of the lateral and the ventral parafacial nucleus, or even for the latter together with the retrotrapezoid nucleus.

25In embryonic and newborn rodents, the lateral parafacial nucleus is rhythmically active at rest, but this activity wanes during early development, and in adults the lateral parafacial nucleus is generally silent during eupnea, but rhythmically active during hyperpnea (;;; Figure 3C–D). The inactivity at rest could be due to tonic inhibition from the medial NTS. Excitatory input from the commissural NTS, conveying chemosensitive input from the carotid bodies, can activate neurons in the lateral parafacial nucleus during periods with elevated blood CO2 levels. In addition, the parafacial nucleus receives direct input from chemoreceptors in the adjacent retrotrapezoid nucleus. Furthermore, input comes also from the pre-Bötzinger and Bötzinger complexes, rVRG, Kölliker-Fuse nucleus, reticular formation, caudal raphe, lateral and medial parabrachial nuclei, periaqueductal gray, PPTg and vestibular nuclei.

26To adapt the depth of ventilation to the metabolic state, the blood gas balance is continuously monitored. At the same time, sensory systems of the airways and lungs, consisting of thermo-, mechano-, and chemoreceptors, survey ongoing respiration and detect environmental irritants and inflammatory mediators. Muscle spindles give feedback on posture and respiratory muscle performance. Altogether, sensory feedback affects ventilatory control, and can trigger respiratory reflexes aimed at maintaining homeostasis upon adverse events. In addition, also hormones can modulate respiratory behavior.

27Chemoreceptors monitor the partial pressures of O2 (pO2) and CO2 (pCO2) in blood and cerebrospinal fluid. Since CO2 reacts with water to form HCO3- and H+, increased pCO2 leads to acidosis that can cause adverse effects on tissue structures, and may result in headaches, delirium and eventually coma. Regulation of pCO2 is therefore, next to that of pO2, of great importance, and a major drive for the level of ventilation. The concentrations of both gasses are continuously measured by peripheral chemoreceptors in the carotid bodies. The carotid bodies project mainly to the NTS, but also to the cVRG.

28Acidosis can also be caused by inflammation, ischemia or defective acid containment. Consequently, acid sensing is not restricted to respiratory control and has evolved as an important property of neurons with unmyelinated and thinly myelinated fibers. Only those areas that sense pH changes and directly affect ventilation are considered to be central chemoreceptor areas. In this respect, most attention has been given to the retrotrapezoid nucleus, but also the NTS, locus coeruleus, raphe nuclei, lateral hypothalamus, and cerebellar fastigial nucleus are central chemoreceptor areas (;;;;;;;; Figure 4A).

29Several molecular mechanisms for central chemoreception have been proposed. First, in astrocytes, increased levels of HCO3- can activate the electroneutral Na+/HCO3- co-transporter NBCn1 (SLC4A7), causing Na+ influx that in turn activates the astrocytic Na+/Ca2+ exchanger Ncx103 (SLC8A1-3; ). The resulting increase in intracellular Ca2+ triggers ATP release. Purinergic receptors on nearby neurons are activated by ATP, leading to neuronal excitation (; Figure 4B–C).

30Second, inward-rectifier K+ (Kir) channels can be inhibited by a decrease in pH, and trigger depolarization and consequently ATP release by astrocytes (; Figure 4C).

31A third mechanism, one that is independent of pH changes, involves the ability of CO2 to induce conformational changes in the connexin-hemichannel Cx26 located in gap junctions between glial cells, rendering Cx26 permeable to ATP (;;;; Figure 4E). Later studies revealed the interaction between CO2 and Cx26 to be more complicated, and probably context-dependent.

32In addition to glial-mediated chemoreception, also neurons express acid-sensitive ion channels (ASICs) and receptors. Among these are some members of the family of inwardly rectifying K2P channels, such as TASK1 and TASK3 that are co-expressed with ASIC1 in the ventrolateral medulla, and contribute to central chemoreception in rats. Another family member is the TASK-2 K+ leak channel, whose absence leads to impaired ventilatory responses to hypercapnia (;; Figure 4D).

33The KCNA1 gene encodes the α subunit of Kv1.1 voltage-gated potassium channels that show particularly strong expression in hippocampus, cerebellum, neocortex and peripheral nerves. Mutations in the KCNA1 gene can lead to the development of episodic ataxia type 1 (EA1), an autosomal dominant disorder with multiple symptoms, most prominently episodes of cerebellar ataxia and myokymia. These episodes are often triggered by physical and emotional stress, which could be related to a defect in respiratory chemoreception. Furthermore, mutations in Kv1.1 channels have been associated with epilepsy, and Kcna1-deficient mice are considered to be a model of sudden unexpected death in epilepsy (SUDEP), while also showing progressive respiratory dysfunction. As respiratory dysfunction is hypothesized to be a primary risk factor for susceptibility to the cardiorespiratory dysfunction in epilepsy, this could reveal a new role for KCNA1 channelopathies in the regulation of basal respiratory physiology.

34The lungs and airways, in particular the larynx, are lined with sensory receptors associated with the nervus vagus. Flow receptors sense air temperature, which is typically colder for inhaled than for exhaled air, and drive receptors measure the laryngeal wall pressure that correlates with its conductivity. Flow and drive receptors are active during each breath, while other mechano- and chemoreceptors are only activated during adverse conditions, such as an obstruction, the presence of irritants, or increased CO2 concentrations. Sensory information from the distal airways is transferred via the nodose ganglion to the NTS, while that of the proximal airways goes predominantly via the jugular ganglion to the paratrigeminal nucleus.

35Vagal stretch receptors, associated with Aβ fibers, can be subdivided into slowly and rapidly adapting receptors: SARs and RARs, respectively. SARs sense inflation and can stay activated for sustained periods, while RARs are more sensitive to acute changes in pressure. Although all RARs rapidly adapt to a persistent mechanical stimulus, a subset of them, the so-called irritant receptors, can remain activated by specific chemicals for a prolonged period. Environmental irritants and inflammatory mediators can trigger activity of unmyelinated C fibers, while localized mechanical stimuli and acid can activate thinly myelinated Aδ fibers that are also known as cough receptors, and that are located in the airway epithelium and mucosa.

36Coughing is a vital action to clear the airways. Coughing involves first inspiration, then obstruction of airways to build up pressure, and subsequent explosive expulsion of air. Ineffective cough reflexes, for example in patients with dementia, can lead to lethal aspiration pneumonia. Inversely, the relatively common condition of chronic cough, affecting more than 5% of the adult population, is thought to be caused by hypersensitivity to airway stimulation.

37The cough reflex is typically triggered by stimulation of the cough receptors and possibly also by activation of C fibers in the airways. Mechanical stimulation of the vocal cords can trigger the expiration reflex, which resembles cough without the initial inspiration.

38Stimulation of irritant receptors in the upper airways can evoke a cough reflex via activation of the paratrigeminal nucleus, and this reflex can be suppressed by the submedial thalamic nucleus and the upstream ventrolateral orbital cortex. In the periaqueductal gray, the excitatory drive from the paratrigeminal nucleus and the inhibitory input from the ventrolateral orbital cortex come together. The periaqueductal gray is hence an important intermediate between the forebrain and the cVRG. The cVRG itself is essential for the cough reflex, as it can activate both expiratory motor neurons in the thoracic spinal cord, controlling abdominal muscles, and in the ambiguus nucleus, controlling laryngeal muscles (Figure 5F).

39Stimulation of the lower airways or the lungs activates RAR relay neurons in the NTS. These neurons receive also input from various other sources that can contribute to suppression or facilitation of coughing. Thalamocortical loops can further modulate coughing and exert voluntary control. Like the paratrigeminal nucleus, the NTS innervates the cVRG. In addition, the NTS affects also the phrenic nucleus, directly as well as indirectly via the rVRG. These connections are complemented with direct output to the ambiguus nucleus from both the paratrigeminal nucleus and the NTS (;;; Figure 5F).

40In the nasal mucosa, irritant receptors expressing Trpv1, a capsaicin-sensitive cation ion channel, that can be activated by histamine H1R receptors, are present on thin sensory fibers of the ethmoidal nerve that terminates in the sneeze-evoking region. The sneeze-evoking region is a dedicated part of the spinal trigeminal nucleus that projects to the cVRG.

41The external and internal intercostal muscles assist with, respectively, expansion and contraction of the rib-cage during breathing (Figure 1A–E). In addition, the intercostal muscles exert postural control, and they combine their respiratory and postural activity in a superposed manner. Single intercostal 1a afferents project to the region of Clarke’s column, to the intercostal motor nucleus, and to the intermediate regions, conveying sensory feedback originating from muscle spindles in the intercostal muscles. From Clarke’s column, information is projected to the cerebellar cortex via the spinocerebellar tract, originating from two types of respiration-related neurons in the lower thoracic segments (T9-T12). This spinocerebellar tract contains uncrossed as well as crossed ascending axons that play different roles in transmitting signals between the spinal cord and the cerebellum. Rhythmic activity in uncrossed spinocerebellar tract neurons, located in and around Clarke’s column, reflects afferent activity from the chest wall, whereas that of crossed neurons, located in laminae VII and VIII, reflect descending influence from the respiratory centers with or without peripheral influences. It has been suggested that the cerebellum uses the posture-related information of the thoracic spinocerebellar tract neurons to adjust posture or coordination of whole body movements.

42Several hormones can affect development and metabolism, and thus have an indirect effect on respiration. Thyroid hormones, for instance, are critical for the development of the respiratory system, and their dysfunction can lead to respiratory failure, including the respiratory distress syndrome. Hormones with a specific role in respiratory control are discussed below.

43Progesterone, estradiol and testosterone can all affect respiratory parameters, and these effects could help explain differences in respiratory behavior between males and females, as well as during different life stages. Indeed, sex hormone receptors are widely distributed in the brain, including central chemoreceptor areas, certain respiratory motor neurons, such as the hypoglossal and phrenic nuclei, and cerebellar Purkinje cells. Progesterone levels correlate with the muscle tone of the genioglossus muscle, and thus with upper airway rigidity, which could play a role in the reduced occurrence of obstructive sleep apnea in females when compared to males.

44Leptin is primarily secreted by adipocytes. An increase in adipose tissue therefore leads to more leptin secretion, which results in increased breathing activity. Leptin can activate a specific subset of excitatory neurons in the NTS projecting to respiratory premotor neurons in the rVRG, as well as to the dorsomedial hypothalamus. Next to the direct projection from the NTS to the rVRG, an indirect pathway via the lateral parabrachial nucleus to the pre-Bötzinger complex is also likely to contribute to the impact of leptin on respiratory frequency. The dorsomedial hypothalamus, which itself also contains leptin receptors, can contribute to upper airway control during respiration. The impact of the leptin-mediated pathways may be compromised in obesity, as leptin resistance can contribute to the development of obesity hypoventilation syndrome and central sleep apnea.

45Full-length leptin receptors (LEPRB) are not restricted to the NTS and dorsomedial hypothalamus, but can also be found in the carotid bodies, neocortex, substantia nigra and cerebellum. Although it is not clear whether these all affect respiratory control, it has been shown that activation of leptin receptors in the carotid bodies can affect breathing and induce ventilatory response to hypoxia.

46Not only increased metabolic activity, but also emotional arousal can cause an increase in the level of ventilation. While such a stress-induced reaction makes sense as preparation for a fight or flight reaction, it can derail during a panic attack. In the absence of increased metabolic demands, hyperventilation induces a decrease in pCO2, which increases blood pH. Hyperventilation can be associated with several symptoms of panic, including shortness of breath, heart racing, dizziness, and fear of dying. Although hyperventilation is typically triggered by stress, anxiety or panic, in rare cases it can also have a neurological cause, and central neurogenic hyperventilation is often associated with pontine damage.

47The retrotrapezoid nucleus is considered as the main central chemoreceptor area, and inhibition of its activity reduces the ventilatory response to hypercapnia, while optogenetic stimulation of the retrotrapezoid nucleus can increase the breathing rate by reducing the duration of expiration. Cholinergic input, probably from the PPTg and the PiCo, can increase the activity of chemoreceptors, while serotonergic input from the caudal and dorsal raphe can enhance the chemosensitive response in the retrotrapezoid nucleus.

48The retrotrapezoid nucleus consists in mice of around 700 Phox2b-positive cells located ventrolateral to the facial nucleus. The locations of these neurons partially overlap with those of the lateral parafacial nucleus. The retrotrapezoid nucleus houses also around 200 biochemically and morphologically different neurons that lack CO2-sensitivity, and that could be related to sighing via their direct projection to the pre-Bötzinger complex.

49Next to pCO2, also sensory input relayed via the caudal and commissural parts of the NTS affects the activity of the retrotrapezoid nucleus. In addition, direct inputs come from the pre-Bötzinger complex, Kölliker-Fuse nucleus, and lateral and medial parabrachial nuclei, rVRG and cVRG, lateral and paraventricular hypothalamus, and central amygdala and periaqueductal gray.

50The output of the retrotrapezoid is glutamatergic and affects respiration directly via projections to the pre-Bötzinger complex, rVRG and cVRG, as well as to the cervical and thoracic spinal cord, and indirectly via the Bötzinger complex, Kölliker-Fuse nucleus, lateral parafacial nucleus, NTS, and lateral and medial parabrachial nucleus.

51The NTS houses respiratory chemoreceptors, and is the prime recipient of visceral input. Especially the caudal part of the NTS receives direct input from pulmonary and cardiovascular baro-, chemo- and stretch receptors, as well as from chemoreceptors in the carotid bodies (;;;;;;;;;; Figure 5A–B). These inputs allow the NTS to contribute to metabolic homeostasis, affecting not only cardiorespiratory function, but also food intake and digestion. The NTS is not essential for inspiration, but NTS dysfunction impairs the response to hypercapnia, which could underlie congenital central hypoventilation syndrome. The ventrolateral NTS is also known as the dorsal respiratory group.

52The NTS contains multiple types of respiratory neurons, some relate to inspiration, others to expiration or are phase-independent (;;;;; Figure 5D). Among these are neurons, located mainly in the commissural and medial NTS, that respond to pulmonary irritant receptors, including RARs and C-fibers mediating cough reflexes, and that are suppressed by the pump neurons (P-cells;;;;; ).

53P-cells of the ventrolateral and other parts of the caudal NTS respond to SARs reporting airway stretch. The firing rates of the P-cells relate to lung volume, possibly modulated by the respiratory rhythm. Activation of these P-cells results in an inhibition of the rVRG and phrenic motor nucleus.

54In addition, the NTS is involved in several other mechanisms controlling respiration. These entail a strong and direct projection to the phrenic motor nucleus. Furthermore, P-cells of the intermediate NTS mediate the Hering-Breuer inspiratory reflex that protects the lungs against overinflation.

55In addition to the extensive visceral inputs, the caudal NTS also receives central input, in particular from other regions of the NTS, lateral and paraventricular hypothalamus and central amygdala, but also from the pre-Bötzinger complex, Bötzinger complex, rVRG, Kölliker-Fuse nucleus, retrotrapezoid nucleus, caudal raphe, bed nucleus of the stria terminalis, lateral and medial parabrachial nuclei, periaqueductal gray, spinal trigeminal nucleus, paratrigeminal nucleus, as well as from insular and infralimbic areas of the cerebral cortex (; Figure 5C). Evidence for significant projections from the cerebellum to the NTS is currently lacking.

56As mentioned above, strong and direct projections from the NTS to the phrenic nucleus have been reported, while the NTS targets also the upper airway motor nuclei. Other target areas are the pre-Bötzinger complex, Bötzinger complex, PiCo, lateral parafacial nucleus, cVRG, retrotrapezoid nucleus, locus coeruleus, dorsal raphe, lateral, paraventricular and dorsomedial hypothalamus, bed nucleus of the stria terminalis and central amygdala, lateral and medial parabrachial nuclei, periaqueductal gray, spinal trigeminal nucleus, pedunculopontine tegmental nucleus (PPTg), cerebellum, and inferior olive (;; Figure 5C).

57The locus coeruleus has broad impact on brain activity, affecting among others attention, motivation, memory, and the level of arousal through its widespread network of noradrenergic fibers. The locus coeruleus can also mediate sensory-evoked awakenings from sleep and has been proposed to be a key center in coupling brain activity with the respiratory cycle.

58Hypercapnia leads to increased activity in the locus coeruleus, which is an evolutionary conserved phenomenon observed in amphibians and mammals. This increase in neural activity can lead to stronger basal ventilation, although this effect was not observed in all studies. The impact of the locus coeruleus may therefore depend on behavioral or experimental conditions.

59Stimulation of the locus coeruleus can modulate activity in the pre-Bötzinger complex via a direct projection. The locus coeruleus also targets many other brain regions with three diverging pathways, with individual locus coeruleus neurons projecting to functionally related areas. The ascending pathway targets the thalamus and cerebral cortex, but also the paraventricular and lateral hypothalamus, periaqueductal gray, dorsal and caudal raphe nuclei, bed nucleus of the stria terminalis, and central amygdala. Other targets include the nuclei of the mesodiencephalic junction (MDJ). The cerebellar pathway targets both the cerebellar nuclei and cortex. The third, descending pathway targets the brainstem and spinal cord, including the ambiguus, hypoglossal and facial nuclei, dorsal raphe, and the pedunculopontine tegmental nucleus. The spinal projections are, as typical for locus coeruleus projections, widespread and involve also the phrenic nucleus, without showing a specific concentration of terminals in the latter.

60In turn, the locus coeruleus receives widespread but relatively sparse projections from the cerebral cortex, and denser projections from the hypothalamus, central amygdala, bed nucleus of the stria terminalis, raphe nuclei, Kölliker-Fuse nucleus and adjacent parabrachial nuclei, and periaqueductal gray. There is also input from the pre-Bötzinger complex, NTS, PPTg, and cerebellar nuclei, in addition to numerous direct projections from cerebellar Purkinje cells.

61The raphe nuclei consist of clusters of neurons along the midline of the midbrain, pons and medulla. Of these, the dorsal and caudal raphe both house CO2 chemoreceptors, and together form the main source of serotonin in the respiratory system (;;; Figure 6A). Disturbances in serotonin release have been linked to respiratory dysfunction in Prader-Willi syndrome as well as to SIDS. The risk of the latter may be increased by prenatal exposure to nicotine that can later cause impairments in serotonin release during hypercapnia.

62The dorsal raphe nucleus mediates the CO2 arousal reflex via its projection to the lateral parabrachial nucleus. Other projections, sometimes involving collateral fibers, target predominantly forebrain regions, including the lateral hypothalamic nucleus, bed nucleus of the stria terminalis, central amygdala, and vestibular nuclei. Descending projections target the hypoglossal, trigeminal and facial motor nuclei, retrotrapezoid nucleus, locus coeruleus, periaqueductal gray, and PPTg (;; Figure 6B).

63The dorsal raphe nucleus receives input from many areas, with relatively dense projections originating in the cerebral cortex, lateral, paraventricular and dorsomedial hypothalamus, bed nucleus of the stria terminalis, central amygdala, periaqueductal gray, lateral and medial parabrachial nuclei, Kölliker-Fuse nucleus, NTS, locus coeruleus, and vestibular nuclei. In addition, also the cerebellar nuclei project to the dorsal raphe nucleus, but this projection seems to be relatively sparse.

64The caudal, or medullary, raphe nuclei can also directly affect respiration. The caudal raphe nuclei exist of the raphe magnus, raphe obscurus, and raphe pallidus, and contain serotonergic, non-serotonergic and mixed neurons (; Figure 6A). Accordingly, neural responses of neurons in the caudal raphe are heterogeneous and hypercapnic acidosis can activate some, and inhibit other neurons of the caudal raphe, the former category comprising serotonergic neurons, the latter not.

65Egr2-Pet1-expressing serotonergic neurons of the raphe magnus mediate the respiratory CO2 chemoreflex. These chemoreceptor neurons target predominantly other chemoreceptor areas in the brainstem: the retrotrapezoid nucleus, NTS, locus coeruleus, pre-Bötzinger complex and medial parabrachial nucleus. Tac1-Pet1-expressing serotonergic neurons of the raphe obscurus do not express chemoreceptor properties themselves, but are most likely downstream of Egr2-Pet1 neurons and preferentially innervate motor nuclei, including the phrenic, facial, trigeminal, hypoglossal, and ambiguus nucleus, but also the pre-Bötzinger complex and NTS.

66Orexinergic neurons of the lateral hypothalamus are the only neurons of the diencephalon with central chemoreceptor properties. Orexins, neuropeptides exclusively produced by the lateral and posterior hypothalamus, are distributed widely throughout the brain and promote wakefulness. The activity of orexinergic neurons is largely restricted to the awake state, and dysfunction of the orexinergic system can cause narcolepsy.

67Orexins can also mediate respiratory chemoreflex responses and increase the tidal volume. To this end, orexinergic fibers innervate the phrenic and hypoglossal nuclei. Other targets include the pre-Bötzinger complex, Kölliker-Fuse nucleus, retrotrapezoid nucleus, NTS, locus coeruleus, dorsal and caudal raphe nuclei, bed nucleus of the stria terminalis, central amygdala, periaqueductal gray, and PPTg. There is also widespread innervation of the cerebellar cortex and nuclei, in particular of the fastigial nucleus.

68Orexinergic cells of the lateral hypothalamus receive input from other areas of the hypothalamus, including the paraventricular and dorsomedial nuclei, and from the bed nucleus of the stria terminalis, central amygdala, periaqueductal gray, dorsal raphe nucleus, and lateral parabrachial nucleus. Other studies revealed input from the pre-Bötzinger complex, NTS, locus coeruleus, medial parabrachial nucleus, PPTg, cerebellar interposed nucleus, and cerebellar dentate nucleus.

69Electrical or chemical stimulation of the rostral part of the cerebellar fastigial nucleus can affect the respiratory pattern. Furthermore, bilateral lesions of the fastigial nucleus suppressed spontaneous breathing in anesthetized cats. This effect was likely to be specific, as bilateral lesions of the cerebellar dentate nuclei had no impact of spontaneous breathing in that same study. An impact of anesthesia cannot be excluded, though, as this result could not be reproduced in awake goats. In contrast, the latter study describes a reduction in hypercapnia-induced increases in ventilation following bilateral lesioning of the fastigial nucleus, suggesting that the impact of the fastigial nucleus becomes especially apparent during periods with increased pCO2. Neural recordings demonstrated activity patterns in phase with respiration at rest, but in particular during respiratory challenges such as tracheal occlusion, bilateral carotid occlusion, or injection of sodium cyanide. Thus, the evidence converges on a role for the fastigial nucleus in mediating ventilatory responses to hypercapnia. The other two cerebellar nuclei lack the chemoreceptor abilities of the fastigial nucleus, and putative other roles of the interposed and dentate nuclei in respiratory control have not been extensively studied. However, electrical stimulation of these nuclei seemed to promote expiration, indicating a more widespread involvement of the cerebellum in respiratory control. The connections to and from the fastigial nucleus are described in the section on the cerebellum.

70The NTS and the paratrigeminal nucleus together form the main entrances of vagal sensory input, and the paratrigeminal nucleus receives mainly input from the proximal airways via the jugular ganglion. The paratrigeminal nucleus also receives primary sensory input from the trigeminal, glossopharyngeal and lingual nerves, and from the upper cervical cord.

71The strongest projections from the paratrigeminal nucleus target the NTS and the lateral and medial parabrachial nucleus, often involving collateral projections. Other targets are the Kölliker-Fuse nucleus, ambiguus nucleus, periaqueductal gray, spinal trigeminal nucleus, and inferior olive. A projection to the cerebellum has been described, but this has later been questioned.

72The trigeminal nerve conveys sensory input from the face and terminates in the sensory trigeminal nuclei that consist of a primary and a spinal nucleus. The spinal nucleus appears particularly relevant for respiratory control, receiving not only tactile information, for example, from the mystacial vibrissae, but also olfactory input from the nasal mucosa, bypassing the forebrain olfactory system. Stimulation of the trigeminal olfactory system can lead to sniffs and respiratory depression without involvement of the forebrain. Activation of the trigeminal olfactory system occurs mainly by noxious or irritable gasses, and the resultant respiratory depression is supposed to protect the lungs, and may contribute to the diving reflex.

73The anterior ethmoidal nerve, the branchlet of the trigeminal nerve that innervates the nasal mucosa, terminates in the interpolar and caudal subnuclei of the spinal trigeminal nucleus. From these regions, projections target the pre-Bötzinger complex, Kölliker-Fuse nucleus, NTS and lateral parabrachial nucleus, as well as the facial nucleus, lateral parafacial nucleus, cVRG, medial parabrachial nucleus, and periaqueductal gray. In addition, there is direct output to the cerebellum and inferior olive. The latter is also indirectly targeted via the MDJ.

74Although direct projections from central pattern generators and respiratory sensory areas to motor neurons do exist, it is likely that indirect pathways via premotor areas are more prominent.

75The Bötzinger and pre-Bötzinger complexes form a cell column in the ventrolateral medulla that continues caudally until the first cervical spinal segment. Caudal to the pre-Bötzinger complex is the retroambiguus nucleus that houses inspiratory premotor neurons in its rostral part and expiratory premotor neurons in its caudal part. The respiratory neurons of the retroambiguus nucleus have later been termed the ventral respiratory group. Somewhat confusingly, some authors use the terms retroambiguus nucleus and ventral respiratory group as synonyms, while others refer to the retroambiguus nucleus specifically as the cVRG, or consider the retroambiguus nucleus and cVRG as overlapping areas within the caudal medullary reticular formation. For the purpose of this review, we use the terms rVRG for the inspiratory and cVRG for the expiratory part of the ventral respiratory column.

76The rVRG provides monosynaptic input to the phrenic motor nucleus, and is the most prominent intermediate between the pre-Bötzinger complex and the diaphragmatic motor neurons. The direct pathway is complemented by a presumably weaker disynaptic pathway via premotor interneurons in the upper cervical (C1 and C2) segments. In addition, the rVRG projects also to the ambiguus, hypoglossal and facial motor nuclei. The rVRG likely contributes to shaping the pattern of respiratory motor output, processing and transmitting sensory afferent information, coordinating ventilation with motor activity, and regulating accessory and respiratory muscle activity.

77The output of the rVRG is not limited to primary motor areas, but targets also the cVRG and Kölliker-Fuse nucleus, and to a lesser extent also the lateral parafacial nucleus, reticular formation, retrotrapezoid nucleus, NTS and lateral parabrachial nucleus, spinal trigeminal nucleus, cerebellum, and inferior olive.

78Input to the rVRG does not originate solely in the pre-Bötzinger complex, but comes also from the Bötzinger complex, Kölliker-Fuse nucleus, cVRG, NTS, retrotrapezoid nucleus, and medial parabrachial nucleus.

79The cVRG is home to expiratory pre-motor neurons, and therefore crucial for the control of expiration-related behavior, like vocalization and expulsive reflexes such as vomiting, coughing and sneezing. While vocalization depends on the projections to the nucleus ambiguus and spinal cord, expulsive reflexes use predominantly the latter only. In particular during coughing, the cVRG may trigger also the inspiratory phase.

80Contrary to the rVRG, the cVRG does not project to the phrenic nucleus, but instead innervates motor neurons of abdominal muscles in the thoracic spinal cord, and those of the upper airway muscles in the ambiguus, hypoglossal, trigeminal, and facial nuclei. Other outputs target pre-Bötzinger and Bötzinger complexes, Kölliker-Fuse nucleus, rVRG, retrotrapezoid nucleus, lateral parabrachial nucleus and periaqueductal gray. Vocalization-related input to the cVRG comes largely from the periaqueductal gray, and this connection is also associated with sexual behavior.

81The cVRG receives both excitatory and inhibitory input from the pre-Bötzinger complex. Also the Bötzinger complex, Kölliker-Fuse nucleus, lateral parafacial nucleus and rVRG project to the cVRG. Other direct inputs come from the NTS, retrotrapezoid nucleus, and lateral and medial parabrachial nuclei. A direct input from the carotid bodies to the area of the cVRG has been described, but this connection did not receive as much attention as that to the NTS.

82The hindbrain reticular formation is a highly heterogenous region spanning from the pons to the caudal medulla, counting many subdivisions with often unclear borders. While many respiratory regions are connected with parts of the reticular formation, this connectivity is difficult to compare between different studies, especially considering that the nomenclature of the reticular subdivisions has been quite fluid. For this reason, we focus only on those connections with a clear role in respiratory control; other connections are summarized in Supplementary file 1. In particular, we mention three connections from the pre-Bötzinger complex in which the reticular formation serves as pre-motor area. First, there is the connection via the perihypoglossal area to the hypoglossal nucleus that is relevant for upper airway control. Second, two regions serve as premotor areas for the part of the facial nucleus that controls among others nose movements: the retrofacial area just caudal to the facial nucleus, and a part of the intermediate reticular formation, and both areas probably receive direct input from the pre-Bötzinger complex. Third, next to the nose region of the intermediate reticular formation is the vibrissal region that links the pre-Bötzinger complex to the vibrissal part of the facial nucleus. The latter two pathways can contribute to the coupling of nose movements, whisking and respiration. The reticular formation could serve a similar pre-motor function for the diaphragm as well, as in particular the gigantocellular part projects directly to the phrenic nucleus.

83Of the limbic system, the amygdala and hypothalamus are involved in subconscious control of respiration. The former is particularly important for handling fear responses, for example, as a consequence of high levels of CO2, while the latter is especially important for coupling with the endocrine system, as well as for peptidergic modulation of respiration. The lateral hypothalamus is a central chemoreceptor area, and – although part of the limbic system – discussed in the section on central chemoreception.

84The central amygdala plays a role in CO2-induced fear behavior, and electrical stimulation of the central amygdala can indeed affect breathing. In humans, amygdalar stimulation leads to hypopnea or even apnea during nasal breathing, but not during mouth breathing, indicating stronger impact on the upper airways than on respiratory rhythmogenesis itself.

85Renewed attention for a respiratory role of the central amygdala comes from findings in sudden unexpected death in epilepsy (SUDEP). Typically, this sequence leads to SUDEP: rapid breathing after a seizure, followed by apnea, bradycardia, and finally cardiac arrest. Furthermore, EEG recordings in patients with epilepsy correlate amygdalar activity with seizure-induced apnea.

86It has been proposed that the impact of the amygdala on respiratory control is mainly exerted via its direct projection to the bed nucleus of the stria terminalis. However, as the central amygdala gives rise to widespread GABAergic projections, also other connections may be relevant for respiratory control. Target areas include the pre-Bötzinger complex, hypoglossal nucleus, retrotrapezoid nucleus, NTS, locus coeruleus, dorsal raphe, caudal raphe, lateral hypothalamus, lateral and medial parabrachial nuclei, periaqueductal gray, and PPTg.

87The central amygdala receives input from the NTS, locus coeruleus, dorsal raphe, lateral and paraventricular hypothalamus, bed nucleus of the stria terminalis, lateral and medial parabrachial nuclei, periaqueductal gray, and PPTg.

88The bed nucleus of the stria terminalis is part of the extended amygdala and a main output station of the central amygdala. It is central to fear, aggression and stress responses. In relation to respiration, the bed nucleus of the stria terminalis may be particularly relevant for anxious responses, or even panic caused by increased levels of CO2. To this end, the bed nucleus of the stria terminalis expresses relatively high levels of acid-sensing ion channel 1A (ASIC1A). ASIC1A in the bed nucleus of the stria terminalis is also essential for the freezing reaction of mice, which implicates a strong reduction in breathing, upon exposure to a predator odor.

89Apart from the central amygdala, other areas targeting the bed nucleus of the stria terminalis are the lateral and dorsomedial hypothalamus, NTS, locus coeruleus, dorsal raphe nucleus, lateral parabrachial nucleus, and periaqueductal gray. Relating to their functional similarities, the central amygdala and bed nucleus of the stria terminalis receive input from the same brain regions, often even from collaterals, except from the substantia nigra pars compacta, that does not project to the bed nucleus of the stria terminalis. This does not imply that all projections are equally strong. In particular, the parabrachial nuclei project stronger to the central amygdala, and the NTS stronger to the bed nucleus of the stria terminalis.

90Output reaches the central amygdala, lateral, dorsomedial and paraventricular hypothalamus, NTS, locus coeruleus, dorsal raphe nuclei, caudal raphe, lateral and medial parabrachial nuclei, and PPTg.

91The paraventricular nucleus of the hypothalamus is vital for the hypoxia reflex: decreased heart rate and increased blood pressure and phrenic nerve activity. The paraventricular nucleus is reciprocally connected with the NTS, and the NTS serves both as intermediate for the ascending input coming from the carotid bodies, and for the descending output affecting the phrenic nucleus. The impact of these NTS-projecting paraventricular fibers may be complemented by vasopressinergic connections to the phrenic nucleus, oxytocinergic projections to the pre-Bötzinger complex, or by projections to the Kölliker-Fuse nucleus, PiCo, ambiguus, hypoglossal and facial motor nuclei, reticular formation, retrotrapezoid nucleus, locus coeruleus, dorsal and caudal raphe, lateral hypothalamus, lateral and medial parabrachial nuclei, periaqueductal gray, bed nucleus of the stria terminalis, central amygdala, spinal trigeminal and paratrigeminal nuclei, PPTg, and cerebellum.

92Apart from the NTS, the lateral and dorsomedial nucleus of the hypothalamus, locus coeruleus, bed nucleus of the stria terminalis, and lateral parabrachial nucleus project to the paraventricular nucleus.

93The dorsomedial hypothalamus is essential for mediating respiratory effects of stressful stimuli. To this end, it receives input from other hypothalamic regions and the bed nucleus of the stria terminalis, as well as from the periaqueductal gray, lateral parabrachial nucleus, NTS, pre-Bötzinger complex, and Kölliker-Fuse nucleus.

94The dorsomedial hypothalamus projects predominantly to other hypothalamic nuclei, in particular to the paraventricular nucleus, but also to the bed nucleus of the stria terminalis, periaqueductal gray, NTS, Kölliker-Fuse nucleus, dorsal and caudal raphe, and cerebellar cortex.

95On top of the central pattern generators, (pre)motor areas, sensory areas and limbic system, also other brain regions can affect specific aspects of respiration, often to adapt respiration to other behaviors. In fact, few brain regions are completely unrelated to respiration. In the following section, we restrict ourselves to those brain regions of which a non-voluntary respiratory function has been clearly described.

96The parabrachial nuclei produce a tonic excitatory drive that can, in conjunction with output from the Kölliker-Fuse nucleus, contribute to setting the duration of inspiration and expiration. In addition, the lateral parabrachial nucleus is also crucial in the hypercapnic arousal reflex that can trigger waking up during sleep. It has been suggested that frequent hypercapnic arousals contribute to day-time fatigue and cardiovascular problems in patients with obstructive sleep apnea. The lateral parabrachial nucleus has a tonic descending drive and an event-driven ascending one. The latter is consistent with other functions of the lateral parabrachial nucleus, as it responds to various aversive signals, varying from food poisoning to itch and hypercapnia, and can send a general alarm signal to the forebrain.

97The main outputs target the lateral hypothalamus, bed nucleus of the stria terminalis, central amygdala, and thalamus. Other targets are the pre-Bötzinger and Bötzinger complexes, lateral parafacial nucleus, cVRG, ambiguus nucleus, hypoglossal nucleus, retrotrapezoid nucleus, NTS, locus coeruleus, dorsal and caudal raphe, paraventricular nucleus, periaqueductal gray, and PPTg.

98The main ascending inputs to the lateral parabrachial nucleus come from the spinal cord and NTS. Other inputs come from the pre-Bötzinger and Bötzinger complexes, Kölliker-Fuse nucleus, rVRG, NTS, retrotrapezoid nucleus, locus coeruleus, dorsal raphe, bed nucleus of stria terminalis, central amygdala, paraventricular nucleus, spinal trigeminal nucleus, paratrigeminal nucleus, PPTg, and fastigial nucleus.

99Activity of the medial parabrachial nucleus affects, like that of the lateral parabrachial nucleus, the breathing frequency by modulating the duration of expiration. The medial parabrachial nucleus contains more expiratory neurons than the neighboring Kölliker-Fuse nucleus that has more inspiratory and phase-spanning neurons. The medial parabrachial nucleus has a dense expression of µ-opioid receptors and plays an important role in the mediation of respiratory depression by opioids. Unlike the lateral parabrachial nucleus, the medial parabrachial nucleus is not involved in the hypercapnic arousal reflex.

100The medial parabrachial nucleus has both excitatory and inhibitory projections to pre-inspiratory neurons in the pre-Bötzinger complex and to expiratory neurons in the Bötzinger complex. Activity of the medial parabrachial nucleus can be affected by input from pulmonary stretch receptors coming via the NTS, from which it receives direct input. In addition, it receives input from chemosensitive neurons of the caudal raphe nucleus.

101The medial parabrachial nucleus receives also input from the pre-Bötzinger complex, Kölliker-Fuse nucleus, rVRG, NTS, retrotrapezoid nucleus, locus coeruleus, caudal raphe, bed nucleus of stria terminalis, central amygdala, paraventricular nucleus, spinal trigeminal nucleus, paratrigeminal nucleus, Purkinje cells in the cerebellar cortex, and fastigial nucleus.

102The medial parabrachial nucleus projects to the pre-Bötzinger and Bötzinger complexes, lateral parafacial nucleus, rVRG and cVRG, ambiguus and hypoglossal nuclei, retrotrapezoid nucleus, NTS, locus coeruleus, dorsal and caudal raphe, lateral hypothalamus, bed nucleus of the stria terminalis, central amygdala, periaqueductal gray, and cerebellum.

103The periaqueductal gray, also known as the central gray, contributes to maintaining the homeostatic balance of an individual. To this end it participates in controlling defensive, emotional, social, sexual and autonomic behaviors, but also in modulating unpleasant sensations, such as pain, itch or the urge to cough. In addition, the periaqueductal gray is central to the control of vocalizations via its strong projection to the cVRG that is the only brain area directly targeting all motor areas required for vocalization. This pathway is especially relevant for emotional vocalizations, like crying and laughing, that do not rely on Broca’s area, as human speech does. For this reason, the periaqueductal gray can be seen as the laughing center, receiving excitatory input from emotion-related pathways originating from the basal temporal and frontal lobes, limbic system, and basal ganglia, and inhibition from voluntary systems, in particular (pre)motor cortices. As to the suppression of involuntary laughter, the periaqueductal gray can also contribute to the suppression of the urge to cough. The periaqueductal gray sends predominantly GABAergic fibers to the NTS (; Figure 5F). This can have clinical relevance, as the common condition of chronic cough is thought to be caused by hypersensitivity to airway stimulation.

104Although the periaqueductal gray lacks clearly discernible cell groups, specific functions can be roughly attributed to four longitudinal columns that differ in their connectivity patterns. The lateral (lPAG) and dorsolateral (dlPAG) columns are mostly involved in active coping strategies, such as fight or flight, while the ventrolateral (vlPAG) column is associated with passive coping strategies, like freezing.

105The periaqueductal gray subserves several respiratory functions to support stereotypical behavioral responses. These functions are distributed over the four cell columns, largely in line with their general function. Chemical stimulation of the dlPAG triggers active breathing, as occurring during fight and flight situations, while stimulation of the dorsomedial and the medial part of the lPAG resulted in slow deep breathing and dyspnea, and inspiratory apneusis, respectively. Finally, stimulation of the lateral parts of lPAG and vlPAG induced respiratory patterns associated with vocalizations.

106The periaqueductal gray is an important hub between the fore- and hindbrain, receiving input from multiple cortical areas as well as from the lateral, dorsomedial and paraventricular hypothalamus. Ascending input to the periaqueductal gray comes directly from the spinal cord, and in particular from the upper cervical cord, to the lPAG and vlPAG. Also the NTS, the prime recipient of vagal sensory input, projects to the ventrolateral and medial PAG. Other respiratory centers in the brainstem also project to the lPAG and vlPAG: the pre-Bötzinger and Bötzinger complexes, Kölliker-Fuse nucleus, locus coeruleus, dorsal and caudal raphe, lateral and medial parabrachial nuclei, paratrigeminal nucleus, and fastigial and dentate cerebellar nuclei.

107Descending output from the periaqueductal gray targets multiple respiratory centers. All four columns, but in particular the lPAG and vlPAG, target the pre-Bötzinger complex, Kölliker-Fuse nucleus and lateral parafacial nucleus. Ipsilateral connections to the parafacial nucleus are predominantly glutamatergic, while contralateral projections have a mixture of glutamatergic and GABAergic fibers. The periaqueductal gray innervates also the PiCo, retrotrapezoid nucleus, NTS, locus coeruleus, dorsal and caudal raphe, PPTg, and inferior olive. Forebrain projections include the lateral hypothalamus, bed nucleus of the stria terminalis, and central amygdala.

108The pedunculopontine tegmental nucleus (PPTg) and the adjacent laterodorsal tegmental nucleus provide important sources of cholinergic fibers, the activation of which is sufficient to induce REM sleep during non-REM sleep. To this end, the pedunculopontine and laterodorsal tegmental nuclei provide widespread innervation of the thalamus and basal ganglia, but also of the retrotrapezoid nucleus. As the chemoreceptor activity in the retrotrapezoid nucleus can be modulated by acetylcholine, this could imply a different sensitivity to pCO2 during REM sleep. Indeed, partial lesioning of the PPTg resulted in abnormal sleep patterns during intermittent hypoxia, while leaving basic respiratory parameters intact.

109Pharmacological activation of muscarinic acetylcholine receptors in the PPTg resulted in decreased breathing rates in awake rats, and also glutamate injections could suppress breathing in anesthetized rats. Inhibition of the PPTg resulted in increased inspiratory activity. Overall, it seems that the PPTg can inhibit the function of the retrotrapezoid nucleus, and thereby modulate the respiratory drive. This may be particularly relevant during REM sleep, given the interaction between PPTg, hypoxia and disrupted sleep patterns.

110The target areas of the PPTg are not restricted to those mentioned above, but also include the Kölliker-Fuse nucleus, lateral parafacial nucleus, hypoglossal, trigeminal and facial motor nuclei, locus coeruleus, dorsal and caudal raphe, lateral hypothalamus, bed nucleus of the stria terminalis, central amygdala, lateral parabrachial nucleus, spinal trigeminal nucleus, and all three cerebellar nuclei.

111Input to the PPTg comes from the NTS, locus coeruleus, dorsal and caudal raphe, lateral and paraventricular hypothalamus, bed nucleus of the stria terminalis, central amygdala, lateral parabrachial nucleus, and cerebellar nuclei.

112Despite their widespread connectivity and importance for motor control, the basal ganglia are not typically considered to be part of the respiratory control network. Possibly the best-known consequence of basal gangliar dysfunction is Parkinson’s disease, caused by progressive degeneration of dopaminergic neurons in the substantia nigra. Patients with Parkinson’s disease can be confronted with respiratory problems, but these may be attributed to degeneration of respiratory muscles, and therefore not directly to respiratory control. Patients with Parkinson’s disease may also experience sleep-disturbed breathing or dyspnea. With a prevalence of 12%, dyspnea is a relatively rare symptom of Parkinson’s disease, and its cause is unclear. Overall, the basal ganglia, and the substantia nigra in particular, therefore, do not seem to play a major role in respiratory control.

113Nevertheless, there are anatomical connections between the substantia nigra and the respiratory control centers, including a direct pathway to the pre-Bötzinger complex. In a rat model for Parkinson’s disease, modulation of the retrotrapezoid nucleus by the substantia nigra indirectly, via the periaqueductal gray, could contribute to respiratory control. Given the uncertainty of the impact of the basal ganglia on respiratory control, these outputs of the substantia nigra have been hypothesized to provide respiratory control centers with information on other ongoing movements, to facilitate respiratory control into the behavioral context of the animal. For this reason, we did not further consider the basal ganglia as control center for the subconscious regulation of respiration.

114Breathing is rather unique among the autonomic functions as it can be subjected to volitional control, modulated by the emotional state, adapted to other ongoing behaviors, and vital for vocalizations and human speech. It is quite likely that the cerebral cortex plays an important role in these functions. Unlike lesions of the subcortical, and in particular of the medullar respiratory centers, lesions of cortical respiratory areas do not abolish respiration. The anatomical aspects of cortical modulation of respiration have received relatively little attention, and we consider the integration of the thalamocortical system into the anatomical scheme of Figure 7 an important future task.

115There is early evidence for a direct projection from the cat motor cortex to the phrenic nucleus, bypassing the medullary respiratory circuitry. In addition, a wide spectrum of projections from different cortical areas targets subcortical respiratory control areas, with the most numerous projections targeting the periaqueductal gray and the Kölliker-Fuse nucleus, and less abundant projections to the caudal raphe nucleus, Bötzinger and pre-Bötzinger complexes. In this study, the Kölliker-Fuse nucleus was predominantly targeted from the (facial) somatosensory, endopiriform and rhinal cortices, and the periaqueductal gray from the insular, cingulate, motor, pre- and infralimbic cortices. In addition, the NTS receives a direct projection form the insular cortex.

116In the context of the present review, especially the function of the insular cortex is relevant. It appears to be the prime cortical target area of visceral input, and it has a viscerotopic organization. EEG recordings of the insular cortex revealed coherence with breathing. Depending on the location within the insular cortex, microstimulation could either increase or decrease the respiratory rate, suggesting that the insular cortex can affect respiration in response to vagal sensory input.

117The cerebellum is composed of a cortex surrounding the fastigial, interposed and dentate nuclei. The output of the cerebellar cortex is formed by Purkinje cells that predominantly target the cerebellar nuclei, although direct connections to the vestibular nuclei, locus coeruleus, and lateral and medial parabrachial nuclei exist as well. From the cerebellar nuclei, there are projections back to the cerebellar cortex, inhibitory projections to the inferior olive, and excitatory projections to the MDJ and many brain regions relevant for respiratory control.

118Strikingly, there are no direct connections to the phrenic nucleus, nor to the pre-Bötzinger complex. Instead, diaphragmatic activity can be affected by a direct projection to the premotor rVRG. Although there are direct projections from the cerebellum to the trigeminal motor nucleus, affecting upper airway muscles, facial nucleus, and hypoglossal nucleus, these direct projections to motor nuclei are typically relatively sparse so that indirect projections, for example, via the reticular formation, are probably much more abundant.

119Other cerebellar projections that can be relevant for respiratory control target the Kölliker-Fuse nucleus, locus coeruleus, dorsal raphe nuclei, periaqueductal gray, lateral and paraventricular hypothalamus. Many of these areas, in particular the Kölliker-Fuse nucleus, have profound projections to medullary respiratory circuit, so that these projections can explain how the cerebellum can affect respiratory control.

120Despite the absence of strong pathways from the cerebellum to the respiratory pattern generators and motor nuclei, cerebellar activity can profoundly modulate respiration. Functional brain imaging of healthy human subjects revealed that the cerebellum is particularly active during coping with respiratory challenges: broad activation during hypoxia and slow breathing, and specific activation in lobules V and VI during hypercapnia. Also the cerebellar nuclei are active during hypercapnia and hypoxia, as well as during volitional expiration. Changes in environmental air pressure can trigger cerebellar activity as well (;; Figure 6G–J).

121The notion that the cerebellum is particularly relevant for deviating from rhythmic breathing is corroborated by the finding that mice suffering from a complete lack of output from the cerebellar cortex or from a partial loss of cerebellar output neurons display overly regular breathing at rest (;; Figure 6F). Stimulation of the fastigial nucleus in cats can affect phrenic nerve activity, and in particular terminate expiratory activity when stimulated during expiration, analogous to optogenetic stimulation experiments of Purkinje cells (; Figure 6E). And indeed, Lurcher mice that suffer from a complete loss of Purkinje cells during their development, demonstrated impaired responses to hypercapnic and hypoxic challenges. Neuronal activity of the cerebellar cortex and the fastigial nucleus is linked to respiration (;;;;; Figure 6C–D). At rest, modulation of the activity of Purkinje cell occurs without phase lead or lag relative to the respiratory cycle, and one could speculate that the cerebellum does not only act to adapt respiration to ongoing behavior, but also facilitates adaptation of other behavior to respiration, in line with observations in cerebellar patients.

122The cerebellum may, in addition, have a particular role in the regulation of breathing during sleep, which includes different stages with different respiratory and oculomotor dynamics. Indeed, when shifting from awake to subconscious behavior and vice-versa, different brain structures such as cerebellum, hippocampus and amygdala, are dynamically activated and de-activated in line with changes in breathing. Accordingly, the impact of sleep on breathing and its role in the development of diurnal respiratory failure in patients suffering from hypoventilation and/or cerebellar disorders can be overlooked. It remains to be elucidated which sets of mechanisms are involved in this process, but it should be noted that sleep by itself does not only reduce respiratory drive, but also diminishes responsiveness to hypoxia and hypercapnia. For example, acute increases in CO2 during rapid eye movement sleep can initiate the process of bicarbonate retention, which further depresses ventilatory responsiveness and which in turn can affect activity in the cerebellar nuclei.

123Mossy fibers constitute one of the two main glutamatergic input pathways to the cerebellum. They carry ascending as well as descending information to granule cells in the cerebellar cortex, while often forming collaterals to the cerebellar nuclei. Cerebellar granule cells make up about halve of all neurons in the brain, and their bifurcating axons, the parallel fibers, target Purkinje cells directly, and indirectly via inhibitory interneurons.

124The pontine nuclei, consisting of the basal pons and the nucleus reticularis tegmenti pontis, are the prime source of mossy fibers; not only are they the main intermediate for descending pathways from the cerebral cortex, they also receive subcortical input from among others the lateral hypothalamus, central amygdala, periaqueductal gray, spinal trigeminal nucleus, PPTg, MDJ, and cerebellar nuclei.

125Other important mossy fiber sources are the two vagal sensory nuclei relating respiratory visceral input to the brain, the NTS and paratrigeminal nucleus. Strong mossy fiber connections bind also the spinal trigeminal nucleus, medial parabrachial nucleus, and several parts of the medullary reticular formation with the cerebellum. This is also true for multiple spinal regions. Weaker projections come from the lateral parabrachial nucleus, Kölliker-Fuse nucleus, and rVRG. The pre-Bötzinger complex provides only sparse projections to the cerebellum.

126The other main input to the cerebellum is formed by the climbing fibers that originate exclusively from the contralateral inferior olive and that form extraordinarily strong glutamatergic synapses on Purkinje cells, next to relatively weak collateral projections to the cerebellar nuclei. Where each adult Purkinje cell can receive input from over 100,000 parallel fibers, it is typically innervated by only a single climbing fiber. Climbing fiber spikes invariably trigger complex spike firing by the postsynaptic Purkinje cells, and these consist of normal action potentials in conjunction with dendritic spikelets caused by profound influx of calcium. Because of their impact on intracellular calcium levels, complex spikes affect the synaptic strength of parallel fiber inputs, and thereby control cerebellar learning. Accordingly, climbing fiber activity is strictly regulated, occurs at a sustained, but low rate, and can thereby serve a homeostatic function at rest, while reporting salient events when they occur. In general, the complex spikes therefore modulate the activity pattern of simple spikes, that organize motor output.

127Apart from the cerebellum, the most important input areas to the inferior olive are the spinal cord, periaqueductal gray, spinal trigeminal nucleus, and MDJ. The MDJ is the main intermediate between the cerebral cortex and the inferior olive: it relays input from the rostromedial and caudal parts of the cerebral cortex to the principal olive, and from the rostrolateral parts of the cerebral cortex to the medial accessory olive. In addition, the MDJ receives input from the spinal trigeminal nucleus and the cerebellar nuclei. Input to the inferior olive comes also from the two vagal sensory recipient nuclei: NTS and paratrigeminal nucleus, as well as from rVRG.

128Other cerebellar afferents could potentially contribute to respiratory control, although their precise impact has not yet been studied. Hypothalamocerebellar connections are evolutionary preserved, but more prominent in primates than in rodents. They originate from several hypothalamic nuclei. In the context of this review, orexinergic fibers from the lateral hypothalamus to the cerebellar nuclei seem to be the most relevant, possibly in conjunction with a projection from the paraventricular nucleus.

129The locus coeruleus provides widespread noradrenergic innervation of the cerebellar cortex and nuclei, which could contribute to a general modulation of cerebellar output. Also both the dorsal and caudal raphe nuclei project to the cerebellar cortex and nuclei, providing serotonergic input, which can have profound impact on cerebellar function. In mild forms of cerebellar ataxia, application of a serotonin receptor agonist could alleviate various motor symptoms. These, and the other anatomical projections, are summarized in Figure 7.

130Although the role of the cerebellum in respiratory control is not yet settled in clinical settings, several lines of evidence indicate a specific role in respiratory control. For instance, in some developmental disorders, cerebellar abnormalities and respiratory dysfunction co-occur. Arguably, the most explicit example is sudden infant death syndrome (SIDS), the unexplained abrupt death of infants, typically during sleep, and most likely related to respiratory problems. Several studies link SIDS to cerebellar malformations. For instance, post-mortem analyses relate SIDS to the presence of a wide external granular layer, which is a sign of developmental delay of the cerebellum. Also malformed and displaced Purkinje cells were observed in several cases. Additionally, the dentate-olivary system can be affected in SIDS. Note that malformations of the cerebellum and related structures have been demonstrated only in a subset of cases, as SIDS can have multiple causes (; Figure 8B).

131Nevertheless, respiratory deficiencies and prolonged dependency on assisted or even mechanical ventilation are frequently observed after cerebellar damage. Relatively mild aberrations in respiratory functions were also noted in different types of spinocerebellar ataxia, while in a mouse model for spinocerebellar ataxia type 7 (SCA7) abnormal respiratory patterns were recorded (Figure 8B–C). Distinct types of spinocerebellar ataxia can affect various brain regions relevant for subconscious control of respiration (Figure 8B, Supplementary file 1). Finally, chronic cough may be one of the earliest symptoms of cerebellar ataxia with neuropathy and vestibular areflexia syndrome (CANVAS), and manifests in the absence of any perceived lung abnormalities.

132Lesions, e.g., as a consequence of hemorrhage, at different levels of the brain can result in different patterns of disordered breathing (Figure 8A).

133A bilateral hemispheric or diencephalic lesion can lead to Cheyne-Stokes respiration. This is characterized by an undulating pattern of breathing with increasing depth and frequency alternated with waves of shallow, slower breathing.

134Apneustic breathing, characterized by the absence or extreme delay of the inspiration to expiration switch, has originally been described as the consequence of a lesion at the level of the Kölliker-Fuse nucleus or the surrounding area.

135Biot’s respiration is characterized by irregular periods of apnea alternated by several breaths of identical depth. Although, the irregularity of this breathing pattern has led to the more common term ‘ataxic breathing’, the cerebellum is not primarily involved in this type of respiration. However, Biot’s respiration pattern can result after a lesion in the medulla. Biot’s respiration can also be a complication of long-term opioid use.

136Finally, a respiratory arrest can result after damage to the pre-Bötzinger complex or the upper spinal cord.

137Animal experiments have demonstrated that a lack of cerebellar output can lead to hyperregular respiration. Cerebellar patients, however, can also display the opposite and present with irregular respiration. Both, hyperregular as well as irregular breathing, can be the consequence of impaired coordination of respiration with ongoing behavior, as described for instance by.

138While each of these brain areas result in a specific type of respiratory pattern. It must be noted that a specific respiratory pattern can have multiple causes such as metabolic or cardiopulmonary disorders. In addition, also peripheral diseases, like COPD, can cause disordered breathing. COPD is a chronic obstructive pulmonary condition characterized by abnormalities, especially narrowing of the small airways, of the lung which leads to limitation of the airflow.

139Throughout this review, we have discussed several behaviors that are tightly coupled to respiration, such as those involved in airway clearance, feeding, and vocalization, as well as adjusting the rate of ventilation to metabolic demands. In addition, also postural control and cardiovascular function are linked to respiration. All pump muscles have dual functions; they do not only enable ventilation but control also posture and movements. The coordination between these different tasks can be affected in cerebellar patients, displaying regular breathing at rest, but arhythmic breathing during arm movements.

140Global and local rhythms abound in the brain, and they have been proposed to arrange long-range synchrony and functional coupling, and thus play a vital role in motor control, sensory perception, and conscious processing. There is accumulating evidence that the respiratory rhythm can affect these brain rhythms.

141In the olfactory system, a coupling between respiration and rhythmic activity was noticed already a long time ago. As odorants are carried by air, each sniff brings in new information, and can be considered as the temporal unit of olfaction. Sniffing can, however, be more widely seen as patterning sensory processing, as it, for instance, can dynamically synchronize with whisking. A hint that respiration is even more related to brain function comes from the observation that humans tend to inhale at the start of a cognitive test, even if that does not involve olfaction. This goes so far that a change in performance on a visuospatial task varies between inspiration and expiration was observed in line with alteration in the EEG spectrum. In fact, respiration-induced patterns in neural oscillations are ubiquitously observed throughout the brain.

142Changes in brain oscillations are by no means restricted to the frequency domain of respiration or sniffing. Gamma oscillations (30–100 Hz), which could reflect sensorimotor integration, can show changes in power that are phase-locked to the respiratory rhythm. Respiration can, therefore, serve as a scaffold for sensory, sensorimotor and cognitive functions.

143Ventilation is only one aspect of the ultimate goal of respiration: getting oxygen to mitochondria throughout the body, while maintaining balanced concentrations of O2 and CO2 in the blood. Hence, respiration has to be coordinated with cardiac and vascular regulation, a process collectively known as cardiorespiratory function. Ineffective cardiorespiratory function is an important predictor for development of averse cardiovascular events and mortality, and has a higher predictive value than for example, smoking or hypertension.

144An example of the integrated control of cardiovascular and respiratory function is the fastigial pressor response that entails simultaneous increases in heart rate and arterial pressure with changes in ventilation. The specificity of the fastigial nucleus for triggering the fastigial pressor response has been called into question, arguing that it was actually caused by activation of the passing afferents to the lateral parabrachial nucleus. Yet, lesioning of the fastigial nuclei does lead to impairment of the fastigial pressor response (; Figure 8D).

145Cardiac and respiratory regulation are coupled on a cycle-by-cycle basis, as the heart rate increases during inspiration, and decreases again during post-inspiration and active expiration. Respiratory sinus arrhythmia is caused by respiration-related fluctuations in activity of inhibitory preganglionic parasympathetic cardiac vagal neurons that are primarily located in the nucleus ambiguus. The Kölliker-Fuse nucleus provides the phasic excitatory drive to these cardiac premotor neurons that is required for respiratory sinus arrhythmia.

146Stimulation of pro-opiomelanocortin-producing neurons in the NTS can augment respiratory sinus arrhythmia, along with triggering suppressed breathing (bradypnea) and cardiac function (bradycardia). Pro-opiomelanocortin is a precursor of the opioid ß-endorphin, and its impact of cardiovascular function can be mimicked by administered opioids. The pro-opiomelanocortin neurons of the NTS directly target important respiratory centers as the pre-Bötzinger complex, giving rise to bradypnea, the ambiguus nucleus, giving rise to bradycardia, and the hypoglossal nucleus, rVRG, hypoglossal nucleus, and raphe obscurus nucleus (; Figure 5E).

147It is an attractive idea to have one or a few central pattern generator areas that control the respiratory rhythm. In particular, the pre-Bötzinger complex is essential in this respect, as inspiration depends on it. In this view, projections organized in a hierarchical fashion pass the respiratory control from the pattern generator(s) to the respiratory motor neurons. Throughout this review we summarize, however, that many other brain regions, including areas like the cerebellum and the limbic system that are not classically considered to be respiratory areas, can affect respiratory control (Figure 8). As all these areas are interconnected (Figure 7), the picture of an integrated network emerges. Thus, depending on specific behavioral needs, these other brain regions may modulate or even overrule the central pattern generators. This does not contradict the idea of central pattern generators being the main players that control rhythmic respiration, but postulates that these pattern generators together with their downstream areas are themselves embedded in a larger network that allows for flexibility within the respiratory control system. We therefore suggest to view the respiratory control system primarily as an integrated network, rather than a hierarchical system. Consequently, we believe that integration of behavioral conditions will add to the pathogenesis of respiratory disease.