Corresponding author James Duffin: Departments of Physiology, University of Toronto, Medical Sciences Building, Room 3326, 1 King's College Circle, Toronto, Ontario, M5S 1A8, Canada. Email: firstname.lastname@example.org
This article presents a short overview of current knowledge about the medullary respiratory neurones and the generation of breathing rhythm. The background respiratory neurophysiology of the medulla and pons is briefly reviewed, with some current ideas about the organization of the pontine–medullary respiratory control system and its development. Questions and speculations about the organization and generation of respiratory rhythm are included, with a view to stimulating experiments to provide answers.
Breathing provides the gas exchange essential to life and is therefore under automatic control, responding to afferent feedback information from many sources such as the respiratory chemoreflexes, and receptors in the airway and lungs. Failure of the respiratory rhythm generator and controller in the medulla and pons has been implicated in trauma involving these locations, sudden infant death, and sleep apnoea.
Despite the importance of this automatic rhythm, and considerable experimental work over many years, its generation is still not fully understood, although significant advances in understanding have been made in recent years. The aim of this article is to provide a short overview of current knowledge in this active research area, with some added speculations of my own; some of them may be considered outlandish and some merely truisms. However, my intention is to place current research endeavours in context, especially for those not directly involved in this field, and to pose questions for future investigations to those that are.
Much of the following basic description derives from experiments on adult rats, although the picture was originally built largely on past experiments on adult cats. Nevertheless, experiments on a broad selection of other species including cats, dogs, mice, frogs and turtles continue to provide useful information about the control of breathing by the neurones of the medulla and pons.
Respiratory neurones are classified or identified by three main attributes: (1) location; (2) pattern of activity; and (3) function.
LocationFigure 1 shows the central respiratory control system. Not shown are the medullary elements sensing hydrogen ion concentration that constitute the central chemoreceptors (Nattie, 2001; Okada et al. 2001) that are located close to the ventral surface and elsewhere, which provide a drive to breathing. Also not shown are the important peripheral chemoreceptor inputs (Paton et al. 1999) and afferent inputs from lung receptors of various types (Ezure et al. 2002) that converge to the nucleus tractus solitarius (nTS).
Respiratory neurones are located in three main, bilaterally symmetric, longitudinally distributed nuclei in the pons, and the dorsal and ventral respiratory groups in the medulla. These neurones are involved in rhythm generation, afferent processing and premotor output shaping.
The afferent processing and relay neurones of the nTS constitute the dorsal respiratory group (DRG) in the rat (Hilaire et al. 1990), although in the cat this group also contains premotor output neurones that project to motoneurones in the spinal cord (Duffin & Lipski, 1987).
The ventral respiratory group (VRG) has been subdivided into a caudal part, where many expiratory premotor neurones are found (Shen & Duffin, 2002), and a rostral part, where many premotor inspiratory neurones are found (Stornetta et al. 2003), based on the location of the obex, the point where the central canal surfaces into the fourth ventricle. An intermediate locus about the obex contains both inspiratory and expiratory neurones. Intermingled with these neurones are the cranial motoneurones of the nucleus ambiguus (nA; Núñez-Abades et al. 1992).
Close to the midline are the hypoglossal motoneurones (nXII) that innervate the tongue (Fregosi & Fuller, 1997), and caudal to the ventral respiratory group are the mysterious upper cervical inspiratory neurones (UCINs; Lipski et al. 1993).
Above the ventral respiratory group, or a rostral extension of it depending on one's viewpoint, lies the Bötzinger complex in the rostro-ventrolateral medulla (RVLM), which contains a variety of neurone types including cranial motoneurones and a major group of inhibitory expiratory neurones (Ezure et al. 2003b). Here also is the para-facial respiratory group (pFRG; Onimaru et al. 1988; Onimaru & Homma, 2003), and just caudal to the Bötzinger complex lies the pre-Bötzinger complex (Smith et al. 1991; Rekling & Feldman, 1998). Neurones in both of these locations are associated with rhythm generation in neonates.
Finally, the pons and its role in respiratory control must be considered. Both anatomical (Gang et al. 1995) and electrophysiological tracing experiments (Ezure et al. 1998) show that pontine neurones receive information from the medulla. Similarly, pontine stimulation (Fung & St-John, 1998) demonstrates that the pons has profound influences on respiratory rhythm (Okazaki et al. 2002), especially if the vagus nerves are not intact (Jodkowski et al. 1994). However, the traditional views of pontine ‘apneustic’ and ‘pneumotaxic’ centres as fundamental to respiratory rhythm generation are giving way to a perhaps vaguer view of the pons as involved in respiratory control for more complex behaviours, a view supported by the finding of third-order neurones from the output network for diaphragm motor control in pontine locations (Dobbins & Feldman, 1994; Travers & Rinaman, 2002). These include the role of the pons in vocalization (Jurgens, 2002) and as a source of state-dependent control of respiratory activity, such as the control of airway muscles during sleep (Haxhiu et al. 2003) and exercise (Plowey et al. 2002). This review therefore concentrates on the medulla as the fundamental source of respiratory rhythm generation.
Patterns of activity As well as location, respiratory neurones are also classified according to their pattern of activity. Recordings from different locations show examples of the patterns observed (Fig. 2A). I have already introduced the inspiratory (I) and expiratory (E) designation, and it can be seen (Fig. 2B) that the frequency of firing can change during the bursts of action potentials, so that the pattern can be designated as augmenting (AUG) or decrementing (DEC), and with little change in firing frequency as constant (CON). Closer examination of the respiratory cycle and the activity of the types of respiratory neurones shows that the expiratory phase of the cycle can be divided into early or E1 or postinspiratory, and late or E2 or pre-inspiratory subdivisions.
Function While the function of the inspiratory phase is obvious as the pump muscle driving phase, the subdivision of the expiratory phase into two is not. The purposes of the two expiratory phases differ, with the postinspiratory activity acting to control airway muscle activity such that an expiratory resistance to flow prolongs the inflation of the lungs to allow better gas mixing (Paton & Dutschmann, 2002) and the late expiratory activity simply assuring that inspiration is not activated inappropriately.
The function of a respiratory neurone is of course intimately related to its axonal projections and connections, and vice versa. Whether it is inhibitory or excitatory, and whether it is a cranial motoneurone, a premotor neurone, an afferent relay neurone, a rhythm-generating neurone or an output-shaping neurone will determine its connections. There are several ways of determining connections, and three are illustrated here.
Figure 3A shows how anatomical methods can be used to find connections. In this elegant study by Lipski et al. (1994), both a VRG inspiratory neurone and a phrenic motoneurone have been stained with different dyes and their morphology reconstructed to show that the medullary neurone probably projects to, and synapses with, a phrenic motoneurone.
Connections can also be inferred from the patterns of neuronal activity. In Fig. 3B it can be seen that phrenic motoneurone activity coincides with that of a VRG inspiratory neurone, suggesting that the VRG neurone is the source of the phrenic drive. Similarly, chloride reversal of phrenic motoneurones shows the existence of a late expiratory inhibition, coinciding with the activity of Bötzinger complex expiratory neurones, suggesting an inhibitory connection.
While anatomical techniques can show likely connections and connections can be inferred from the correlation of patterns of activity, only electrophysiological techniques like cross-correlation and spike-triggered averaging can demonstrate functional connections. In Fig. 3C the cross-correlogram of the activity of a single VRG inspiratory neurone and the phrenic nerve discharge show a peak after a short delay for transmission that indicates a monosynaptic excitatory connection. Since the same neurone projected to and excited both left and right phrenic motoneurones, a cross-correlogram of the left and right phrenic nerve discharges showed a central peak because of the near simultaneous excitation of the left and right phrenic motoneurones (Duffin et al. 2000). A triggered average of the intracellular potential of a phrenic motoneurone (Fig. 3D) reveals an inhibitory postsynaptic potential, demonstrating the inhibitory connection from the triggering Bötzinger complex expiratory neurone.
While Fig. 3 shows how connections between individual neurones may be discovered, by extending this technique, using arrays of multiple electrodes, many neurones can be recorded at once. This ability allows the experimenter to study the behaviour and interconnections of large numbers of neurones, and Morris et al. (2001) have shown that this recording technique can provide insights into the behaviour of groups of neurones as network modules.
Experiments like these, carried out in a number of laboratories throughout the world, demonstrated connections and suggested basic network models of rhythm generation for the adult (Ezure, 1990; Duffin, 1991; Richter et al. 1992) like those shown in Fig. 4 from a previous review (Duffin et al. 2000), with complex models based on these core generators providing realistic simulations of neuronal activities (Ryback et al. 1997). But note that although time has passed since the figure was first presented, the dotted connections still remain to be functionally demonstrated.
In vitro neonatal rat preparations
Thus far I have confined my review of the basics to findings from adult preparations, but in vitro preparations have also been used to study the activity, connections and functions of respiratory neurones, albeit in neonatal rats. As Fig. 5 shows, the superfused brainstem–spinal cord preparation has rhythmic phrenic and cranial nerve activities, and a transverse medullary slice also shows a rhythmic hypoglossal nerve activity. These experiments have produced new ideas about respiratory rhythm generation and recently raised some interesting questions. (1) Are there two rhythm generators? (2) Did they evolve from two breathing functions? (3) What are their developmental changes from neonate to adult? (4) How might they relate to eupnoea and gasping?
Two rhythm generators There are two rhythm generators in the frog, one for buccal breathing and one for lung breathing. Figure 6, from Wilson et al. (2002), shows the two rhythms, with the small fluctuations indicating a continuing buccal rhythm and the large periodic fluctuations a synchronized lung rhythm. These investigators have recently suggested that the rat may have inherited these two rhythm generators, which suggests a positive answer to the second question. They have also proposed the interactions between the two generators shown in Fig. 6, and so these might also be found in the rat. Since buccal breathing is primarily an expiratory act, like the breathing in some of our reptilian ancestors, where expiration is the active phase rather than inspiration (Glass & Wood, 1983), and frog lung breathing is primarily an inspiratory act, one might expect these actions to be reprised in the rat.
Indeed, evidence has existed for some time that there were two sources of rhythm in the neonatal rat in vitro preparation. The description by Onimaru et al. (1988) of neurones generating a respiratory rhythm in the RVLM was made some time ago. Smith et al. 1990) summarized their findings of rhythmic neurones in a location they named the pre-Bötzinger complex.
The inspiratory nature of the pre-Bötzinger rhythm generator has been demonstrated many times and its pacemaker properties determined (Del Negro et al. 2002), so that models of their rhythm generation have been developed (Butera et al. 1999; Rybak et al. 2004). The experimental observations from Johnson et al. (2001) shown in Fig. 7 illustrate these points. Figure 7 shows that recordings made from the hypoglossal nerve and the pre-Bötzinger complex neurones in the thin medullary slice exhibit rhythmic bursting, and these bursts can also be recorded from islets containing the pre-Bötzinger complex rhythm generator neurones alone. That the rhythm is intrinsic is demonstrated by its persistence after blockade of GABA and glycine inhibition with strychnine and bicuculine, respectively. The pre-Bötzinger complex inspiratory neurones may excite each other across the midline; cross-correlation of the activity from left and right pre-Bötzinger complex neurones shows peaks either side of time zero, indicating a mutual excitation (Li et al. 2003), and similar cross-connections occur between neurones in the RVLM (Kashiwagi et al. 1993).
The neurones of the rhythm generator in the RVLM have recently been visualized using an exciting new optical technique that shows their spatiotemporal activity in the neonatal rat brainstem–spinal cord in vitro preparation with a voltage-sensitive dye (Onimaru & Homma, 2003). This rhythm generator introduced a new pattern of activity description; the preinspiratory or pre-I neurone, so called because its activity precedes phrenic inspiration, although the activity of pre-I neurones is not confined to this preinspiratory period but also occurs in the postinspiratory phase (see Fig. 8A), and this pattern of activity might be better termed ‘biphasic expiratory’, as others have done (Smith et al. 1990). As Takeda et al. (2001) showed, the activity of pre-I neurones can continue through to the early part of expiration if not inhibited by inspiratory neurones. That the pre-I neurones are intrinsically rhythmic (Onimaru et al. 1989) is demonstrated by their continued rhythmic activity, albeit with a changed pattern, after blockade of GABA and glycine inhibition with strychnine and bicuculine, respectively, as shown in Fig. 8D, a recording from Brockhaus & Ballanyi (1998).
Figure 8A, B and C, from a combined effort of the laboratories formerly championing the individual rhythm generators (Janczewski et al. 2002), shows how the two rhythm generators in the neonatal rat may interact. As Fig. 8A illustrates, the lumbar motoneurones have a pre-I pattern of activity, and these authors traced the excitatory pathway from pre-I neurones to lumbar expiratory motoneurones via caudal VRG expiratory neurones. μ-Opioids were used to suppress the rhythmic drive to the phrenic nerves from the pre-Bötzinger inspiratory neurones (Fig. 8B) and, when phrenic bursts were absent, the inspiratory inhibition of abdominal expiratory activity was also absent, demonstrating that an inhibitory connection to the RVLM pre-I neurones must exist, probably originating from the pre-Bötzinger inspiratory neurones. These experiments also demonstrate the expiratory nature of the RVLM pre-I rhythm generator, reinforcing the idea that it has evolved from its frog buccal or reptilian past, but notice that the hypothesized interaction (Fig. 8C) between the two rhythm generators in the rat is the opposite of that for the frog buccal and lung rhythm generators.
That the pre-I pFRG rhythm generator has both inhibitory and excitatory connections to the pre-Bötzinger complex rhythm generator has been demonstrated by Mellen et al. (2003). They showed that with μ-opioid depression of the pre-Bötzinger rhythm generator and elimination of its drive to phrenic motoneurones, both excitatory depolarizations and inhibitory hyperpolarizations with the same rhythm as the RVLM pre-I rhythm generator can be identified in pre-Bötzinger complex neurones.
Developmental changes: pacemakers to networks With the existence of two sources for rhythm generation apparently confirmed for the neonatal rat, we are still faced with the differences observed in the adult, where inhibitory network interconnections predominate; respiratory rhythm is disrupted when inhibition is blocked (Hayashi & Lipski, 1992; Paton & Richter, 1995).
Despite this change in the mechanism of respiratory rhythm generation with development, it appears that the two neonatal rhythm generators are still present in the adult. Using μ-opioid agonists to depress the pre-Bötzinger rhythm generator but leave the RVLM pre-I rhythm generator unaffected, Mellen et al. (2003) changed the balance between the two rhythm generators in favour of the RVLM pre-I generator. With increasing doses of μ-opioid the bursting frequency decreased gradually at first but then it slowed in steps (quantal slowing). These authors suggested that the phrenic rhythm became determined by the fixed RVLM pre-I rhythm generator, whose rhythm was unaffected by the μ-opioid agonist, rather than the pre-Bötzinger generator, whose activity was depressed by the μ-opioid agonist. The slowing of rhythm was therefore in steps, the result of the depressed pre-Bötzinger neurones failing to drive phrenic premotor neurones at every cycle so that some cycles were skipped.
Developmental changes will not only occur in the mechanisms of rhythm generation themselves but also with the balance between the two generators in controlling the respiratory rhythm. If the neurones of the RVLM pre-I rhythm generator are descended from frog buccal control then perhaps they are linked to control of the airway. This idea suggests that the RVLM pre-I generator may dominate hypoglossal motoneurone control, whereas the pre-Bötzinger generator assumes control of the phrenic motoneurones.
Some credence to this idea may be given by recent studies from Ezure's laboratory (Saito et al. 2002; Ezure et al. 2003a). As Fig. 9A illustrates, these authors showed that manipulation of lung inflation can produce changes in hypoglossal activity independent of phrenic activity. Their findings suggest an independence of control that might relate to a looser coupling of the two rhythm generators in the adult. Indeed, it has been shown that phrenic premotor neurones do not drive hypoglossal motoneurones (Peever et al. 2001, 2002), which adds support to the idea of divided control functions. Figure 9B shows that hypoglossal activity precedes that of the phrenic nerve in the adult rat at a preinspiratory time in the cycle, and differences in power spectra for preinspiratory and inspiratory portions of the hypoglossal discharge suggest different origins for drive during these phases (Leiter & St-John, 2004). In the neonatal in vitro preparation, however, the coupling appears to be stronger than in the adult, at least judged by the closer time relations. Other airway control motoneurones, such as the postinspiratory motoneurones in the caudal VRG, may also be controlled from the RVLM pre-I rhythm generator; although no such connection has been demonstrated, blockade of glycine receptors in neonatal in situ preparations moves their activity to the inspiratory phase (Paton & Dutschmann, 2002), evoking comparison with the scenario described above for abdominal motor activity and μ-opiod depression. However, as a caveat to these ideas, it should also be borne in mind that since projections from the pons to hypoglossal motoneurones have been demonstrated in cats (Smith et al. 1989; Kuna & Remmers, 1999), alterations in pontine influences might also explain these observational differences.
Eupnoea and gasping The balance between the two generators in controlling the respiratory rhythm may not just change with development, but also with conditions. The recording shown in Fig. 9C was made in a 39-day-old juvenile rat in situ preparation. It shows that the change from an augmenting to a decrementing phrenic bursting pattern appears to result from the appearance of pre-inspiratory activity. In this case the preparation had been used for several hours and its condition was probably deteriorating. Similar effects, produced by hypoxia, have been previously noted by St-John & Paton (2003). So, is gasping the result of the RVLM pre-I rhythm generator becoming more active and dominating the control of phrenic motor output? As the senior rhythm generator in terms of evolution it may be the more robust and the last to succumb to deteriorating conditions.
How do the neurones in the two neonatal rhythm generators, where pacemaker properties predominate, develop into the neurones that are found in the adult rat, where inhibitory connections assume importance?Figure 10A, from Paton & Richter (1995), shows that while blockade of inhibition disrupted respiratory rhythm in mice over 15 days old, it did not in neonatal mice under 8 days old. Ezure (2004) has recently considered this question, considering the types of neurone found in adult and neonatal rats as assessed by their firing patterns, and how the neonatal RVLM pre-I generator neurones and the pre-Bötzinger complex inspiratory neurones might be transformed into mature neurones. Figure 10B shows Ezure's suggestions. Identification of neurones with these changing firing patterns with development will not only show how the activity is transformed by development but also show the transformation in terms of their spatial distribution.
The in situ preparation
How might we answer Ezure's challenge to link the neurones identified in neonatal in vitro preparations with those of mature animals? The insitu or working heart–brainstem preparation (Fig. 11), the brainchild of Paton (1996), which can be used over the course of development, offers one possibility. In this decerebrate preparation, the animal below the diaphragm is discarded, and the descending aorta is perfused with an artificial CSF to which an oncotic agent has been added. Not only can the temperature, perfusion pressure, oxygen and pH be controlled, but various pharmacological agents can also be applied, all while recording from neurones in the medulla, pons and various respiratory nerves.
My laboratory began using this preparation with a simple question of development in mind that turned into a more insightful examination as it proceeded. Previous experiments in decerebrate adult rats, cross-correlating left and right phrenic discharges, revealed central peaks in the cross-correlograms indicating a short time scale synchronous drive to both left and right phrenic motoneurones, probably from bilaterally projecting VRG inspiratory neurones. We did not find such peaks in our in vitro brainstem–spinal cord neonatal preparations, and that result persuaded us that only unilateral spinal projections existed in the neonate for these medullary phrenic premotor neurones.
However, using the in situ preparation we found central peaks in the cross-correlograms at all ages, as Fig. 12A illustrates. Was our previous interpretation incorrect? We did note that the peak widths varied with age, and this observation suggested the possibility that the mechanism producing the peaks might differ between neonates and adults. We hypothesized that the central peaks in cross-correlograms from neonatal preparations resulted from a left–right synchronization of medullary premotor neurones with unilateral spinal projections in the neonate. Gap junction coupling (Solomon et al. 2003) could play a role in this left–right synchronization of medullary phrenic premotor neurones, because as Fig. 13B shows that carbenoxalone (CBX, a gap junction blocker) was able to disrupt the medullary connections that gave rise to the short time scale synchronization of left and right medullary premotor neurones that produce the central peaks in the cross-correlograms. However, these CBX gap junction blockade experiments were inconclusive, because both CBX and its inactive analogue, glycyrrhzic acid (GZA), eliminated the central peaks in such cross-correlograms.
Finally, we speculated that the lack of central peaks in cross-correlograms from in vitro neonatal preparations, but their presence in cross-correlograms from in situ neonatal preparations was due to differences in preparation condition (Duffin, 2003). Observations such as those shown in Fig. 13 confirmed this view and illustrate the necessary cautions that must apply when comparing different preparations.
In this brief review I have offered evidence from a number of investigators that suggest the existence of two respiratory rhythm generators in both neonates and adults. I have presented speculations that they may be assigned to expiratory and inspiratory control and airway and pump muscle control as dictated by their evolutionary roles. Future experiments must examine these rhythm generators in both adults and neonates to determine how their functions change with conditions and with development, the latter not only in terms of their mechanisms of rhythm generation but also in terms of their interaction. I hope that experimenters will continue to discover the functional relations between neurones so that modellers can pursue the goal of a comprehensive understanding of the central control of respiration that will assist in the solution of clinical problems.
I thank my colleagues and students for their enthusiastic participation in my laboratory experiments, and The Canadian Institutes of Health Research for financial support.