Mammalian brainstem chemosensitive neurones: linking them to respiration in vitro

Abstract

Neurones which are excited by CO₂ or H⁺ are present in a number of brainstem structures in addition to the ventrolateral region of the medulla, the site at which the respiratory response to hypercapnia is traditionally believed to originate. In this review we examine recent work concerned with establishing the relationship between these chemosensitive neurones and respiration, the emphasis being placed on the use for this purpose of in vitro preparations of the mammalian brainstem.

Under normoxic conditions CO₂ is the most powerful natural stimulus to respiration in man and other air breathing vertebrates (Dejours, 1988). In the steady state much of the resulting ventilatory response (up to 80 % during hyperoxia; Heeringa et al. 1979) depends on the action of CO₂ (or H⁺) at sites located within the lower brainstem. These sites, usually held to be neurones, are thus sometimes referred to as central (respiratory) chemoreceptors, though the term itself is little more than a useful fictional device for structures which have proved to be very difficult to identify with any confidence. The purpose of this article is to review recent work concerned with one specific aspect of this problem, namely with the question whether brainstem neurones which belong to the respiratory network are themselves CO₂ and/or pH chemosensitive.

A commonly held view is that the ventilatory response to hypercapnia or acidosis is mediated by structures (chemoreceptors) located at, or close to, the ventrolateral surface of the medulla (reviewed by Loeschke, 1982). Nothing which has emerged from more recent experimental work has diminished what appears to be the particular importance of this region in stimulating ventilation, and indeed has tended to reinforce it by adding previously unavailable detail. First, by directly demonstrating the presence within the ventrolateral medulla of chemosensitive neurones (Jarolimek et al. 1990; Kawai et al. 1996), some of which are located immediately adjacent to, and others within a few hundred micrometres of, the surface (Richerson, 1995). Second, by showing that even where chemosensitive neurones in this region are located at some distance from the surface they may nevertheless be morphologically specialized to detect pH changes occurring in the vicinity of the surface (see Pilowsky et al. 1993; Kawai et al. 1996). And third, by showing that, at least in the non-awake animal, the functional integrity of a specific, superficially located group of (putative) chemosensitive cells within the rostral ventrolateral medulla (retrotrapezoid nucleus) is required for the generation of a respiratory response to systemic hypercapnia (Nattie et al. 1991; Nattie & Li, 1994; Erlichman et al. 1998).

That said, what has changed in the last few years is the belief that the ventrolateral medulla is uniquely chemosensitive. The picture which is now beginning to emerge is one in which CO₂/pH chemosensitivity is a property shared by a significantly large number of neurones belonging to several quite distinct brainstem networks. The experimental background to this revised view is largely the result of two developments. First, improvements to technique have made it possible in in vivo preparations to produce localized areas of tissue acidosis (300-500 μm in extent) within the brainstem, including deep-lying structures. So it is possible to demonstrate that focal acidification at a number of individual, widely distributed sites stimulates the respiratory motor (phrenic) output (Coates et al. 1993; Nattie & Li, 1996; Bernard et al. 1996; Li & Nattie, 1997; Erlichman et al. 1998). Second, the introduction of a variety of in vitro preparations – some of which show a rudimentary, but still recognizably respiratory, motor output – has made it possible to extend the search for chemosensitive neurones beyond the traditionally chemosensitive region of the ventrolateral medulla and to include identified respiratory neurones in this search (for references see below). The encouraging feature of these developments is that, at least on a regional basis, there is a large measure of agreement between in vitro and in vivo work as to the distribution of chemosensitive structures, but in all this, the problem remains of linking chemosensitive neurones directly to the control of respiration, and it is with this issue that we are concerned here. In the remainder of this article we will use the term ‘CO₂’ in referring to the stimulus because its concentration is the most commonly employed experimental variable; the nature and site of action of the adequate stimulus are still very much subjects of debate (e.g. Voipio & Ballanyi, 1997; Ritucci et al. 1997).

Chemosensitive neurones are present in several anatomically distinct brainstem structures

From work on in vitro preparations it is clear that significantly large numbers of brainstem neurones, present in several anatomically distinct structures and exhibiting very different patterns of activity, can be classified as chemosensitive. This is conventionally understood to mean that they exhibit an excitatory response to an increase in CO₂ or H⁺ concentration which they retain under conditions in which chemical synaptic transmission is minimal or absent. Within the lower brainstem the currently known distribution of such neurones falls into four main groups: the pontine locus coeruleus (LC; Pineda & Aghajanian, 1997; Oyamada et al. 1998); the nucleus tractus solitarii (Dean et al. 1989, 1990; Huang et al. 1997); the midline (raphe) nuclei of the ventral medulla (Richerson, 1995; Wang et al. 1998) and the ventrolateral quadrant of the medulla (VLM; Jarolimek et al. 1990; Richerson, 1995) where they are distributed over a fairly extensive column along the rostro-caudal axis (Kawai et al. 1996).

The actual incidence of such neurones in these various structures is difficult to assess accurately because of differences in recording technique and in the preparations which have been employed, but it ranges from in excess of 90 % in the LC, which is a homogeneous structure in the rat (Pineda & Aghajanian, 1997; Oyamada et al. 1998) to about 14 % in the ventral raphe and adjacent rostral VLM where CO₂-excited neurones constitute about half of the responsive population (Richerson, 1995).

We will consider later two classes of chemosensitive neurone for which it is possible to infer a link with respiratory control because they form part of a synaptically connected network of brainstem neurones showing respiratory rhythm. However, beyond this specific relationship the evidence provided by trans-synaptic (pseudorabies virus) tracing of connections back from the phrenic nucleus reveals that each of those structures (see above) now known to contain chemosensitive neurones forms part of a substantially larger network involved in controlling the input to this nucleus, and therefore the contraction of the diaphragm (Dobbins & Feldman, 1994). For example, the ventral respiratory group of neurones, located within the VLM, represents the main source of rhythmic respiratory input to the phrenic nucleus (in the rat), but it is also a point of convergence for numerous additional pathways, including those originating in nuclei outside the VLM and which contain chemosensitive neurones. Thus, there are reasonable grounds for suspecting that these various chemosensitive neurones are involved in some way in controlling the diaphragm, whether for its primary purpose in ventilating the lungs or as part of a wider range of homoeostatic adjustments.

One approach to looking for connections between chemosensitive neurones and the neurones controlling ventilation is to use the respiratory rhythm itself as an indication of connectivity, and to ask whether neurones showing this rhythm are also chemosensitive. A start has been made in this direction by using in vitro preparations of the brainstem and spinal cord, and this is described next.

Respiratory chemosensitivity in vitro

A number of in vitro preparations of the brainstem show the property that they spontaneously generate a rhythmic output on respiratory nerve roots which is sustained for more or less long periods (hours) after isolation. Only a few have been specifically examined from the point of view of chemosensitivity (Suzue, 1984; Morin-Surun et al. 1995; Eugenin & Nicholls, 1997) and the most commonly used in work on single neurones is the isolated and superfused brainstem- spinal cord of the neonatal rat. In this preparation the rhythm is invariably present on the phrenic (inspiratory) nerve roots, but when stimulated by lowering the bath pH it also appears with the appropriate phasing in expiratory intercostal nerves (Iizuka, 1999). The two important experimental features of this preparation are as follows.

First, it exhibits a reproducible response to increasing the CO₂ concentration of the superfusate (hypercapnic acidosis). The most prominent feature of this response is an increase in phrenic burst (= respiratory) frequency with (typically) little change in burst structure and just a slight shortening in burst duration (see phrenic trace in Figs 2A–C and 4A; Suzue, 1984; Okada et al. 1993; Brockhaus et al. 1993; Kawai et al. 1996; Voipio & Ballanyi, 1997; Oyamada et al. 1998; Iizuka, 1999; but see also Harada et al. 1985). Neither the pons nor the dorsal half of the medulla is required for the response, while microinjection of a small volume of CO₂-tonometered saline at numerous individual sites just below the ventrolateral surface of the medulla elicits at short latency a transient increase in respiratory frequency (Issa & Remmers, 1992), i.e. the VLM contains all that is necessary to generate a CO₂-sensitive rhythm.

Figure 2. Chemosensitivity in respiration-modulated LC neurones in vitro.

A, the acid shift in bath pH which follows switching from 2 to 8 % CO₂-equilibrated superfusate, the subsequent increase in phrenic burst frequency (∞Phr), the depolarizing shift in membrane potential (V_m) and increase in discharge frequency in an LC neurone. B and C, short episodes on a fast time base taken from the record in A. D (different LC neurone), the depolarizing response and increase in discharge frequency are retained after blocking phrenic discharge and respiratory modulation of membrane potential with TTX (0.2 μM). (Modified after Oyamada et al. (1998) and reproduced with permission.)

Figure 4. Chemosensitivity in an inspiratory VLM neurone in vitro.

A and B, the same neurone recorded in perforated patch configuration. A, switching from 2 to 8 % CO₂ resulted in an acid shift in bath pH, an increase in phrenic burst (∞Phr) frequency and membrane depolarization. B, TTX suppressed phrenic discharge and correspondingly phased synaptic input to the neurone, but left the depolarizing response to CO₂ intact (A. Kawai, D. Ballantyne, K. Mückenhoff & P. Scheid, unpublished data shown with permission).

Second, with extracellular or whole-cell recording techniques it has been possible to identify respiratory neurones within the VLM, to map their distribution and to classify individual neurones according to the timing of their activity in relation to phrenic discharge (Smith et al. 1990, 1992; Onimaru & Homma, 1992; Brockhaus et al. 1993). For present purposes the details of this classification are less important than the fact that the existence of various sub-types of respiratory neurone is evidence both of a substantial synaptic circuitry and of differences, experimentally verified in a few cases (Onimaru et al. 1996; Rekling et al. 1996; Koshiya & Smith, 1999), in their intrinsic membrane properties.

Together these two features – a reproducible response to CO₂ and identifiable respiratory neurones – would seem to make the brainstem preparation well suited to asking whether the latter function as chemoreceptors. However, there are also ‘problems’ with the in vitro rhythm: it is slow, the inspiratory burst is brief and decrementing in pattern rather than incrementing as in eupnoeic respiration. Most importantly, the way it changes in response to CO₂ is almost exactly the reverse of the response to hypercapnia seen in the vagotomized (adult) animal in vivo where respiratory frequency remains largely unchanged at increased tidal volume. Some of these differences have been variously attributed to the effects of (low) temperature, de-afferentation, tissue acidosis and hypoxia in the in vitro preparation (see St John, 1998, and Rekling & Feldman, 1998, for opposing views as to their origin). However they arise, it is not unreasonable to ask whether it is worthwhile studying the chemosensitivity of what would certainly be an abnormal respiratory rhythm in a brain with intact blood supply. We believe that it is because it allows questions to be asked about chemosensitivity which cannot at present be readily answered either in less reduced preparations showing a normal eupnoeic rhythm or in conventional slice preparations lacking an identifiable respiratory rhythm.

Chemosensitivity within the rhythmically active brainstem respiratory network

In the neonatal rat brainstem in vitro both LC neurones and some VLM neurones have been shown both to be responsive to CO₂ and to exhibit a rhythmic synaptic input which is synchronized to the burst discharge of phrenic motoneurones (Fig. 1A and B; Kawai et al. 1996; Oyamada et al. 1998). Since there is no peripheral afferent feedback in this preparation this input must ultimately be attributed to the way these neurones are connected to the respiratory rhythm generator which, in this preparation, is located within the ventral medulla. Most current evidence places this location immediately rostral to the ventral respiratory group of neurones (defined using the classification of Dobbins & Feldman, 1994) within a functionally identified region referred to as the ‘pre-Bötzinger complex’ (PBC) (Smith et al. 1991; Johnson et al. 1994; Rekling & Feldman, 1998; Koshiya & Smith, 1999). However, there is also evidence for endogenously rhythmic (pre-inspiratory) neurones outside the PBC at a more rostral location within the medulla (RVL; Onimaru et al. 1988). Since these may be important in generating respiratory rhythm in vitro (reviewed by Ballanyi et al. 1999), we have included both the PBC and RVL as possibilities in the schematic rhythm generator of Fig. 1C. The issue is of little consequence for the present discussion since both LC neurones and those medullary respiratory neurones which have been examined in any detail from the point of view of their chemosensitivity are located ‘downstream’ of the rhythm generator, i.e. their respiratory-phased activity is dependent upon intact chemical synaptic transmission. Since the most prominent feature of the response to CO₂in vitro is an increase in respiratory frequency, the obvious question to ask is what purpose is served by the chemosensitivity of ‘downstream’ neurones. To answer this it is first necessary to examine the form which their chemosensitivity takes.

Figure 1. Respiratory-phased innervation of neurones in the locus coeruleus (LC) and ventrolateral medulla in the neonatal rat brainstem in vitro.

A and B, spontaneous activity (membrane potential, V_m) recorded with a whole-cell pipette from a respiration-modulated LC neurone (A) and an inspiratory neurone of the ventral respiratory group (VRG) (B). Inspiratory discharge was recorded with a suction pipette from a phrenic nerve root and is shown in integrated form (∞Phr). The large difference in respiratory frequency between A and B is typical of preparations with (A) and without (B) the pons. C shows schematically the relationship of these synaptically driven respiratory neurones to the respiratory-rhythm generator. Since there is still uncertainty as to whether the essential rhythm generating neurones are located within the pre-Bötzinger complex (PBC) or the rostroventrolateral medulla (RVL), both possibilities are shown. VRG neurones are shown with dendritic projections to the ventrolateral surface. The existence of gap junction coupling of LC neurones within the pericoerulear dendritic network (PCD) is based on inference (see text). The + and – symbols refer to excitatory and inhibitory synaptic input, respectively.

The LC as a chemosensory nucleus

The respiratory input to LC neurones generates a biphasic pattern of excitation and inhibition, the former occurring approximately synchronously with the phrenic burst, and the latter occupying the early stage of the expiratory interval between bursts (Fig. 1A). The effect of this input is to modulate at respiratory frequency the on-going discharge of LC neurones which may take the form of a regular rhythm of single spikes, but more commonly in the neonate (in vitro) is patterned into spike doublets or triplets by an underlying subthreshold oscillation of the membrane potential. Since this non-respiratory component of spike discharge continues in the absence of chemical synaptic input (Williams et al. 1984) it can be regarded as being endogenously generated.

Bath application of glutamate antagonists suppresses both components of the respiratory input, while the inhibitory component alone is suppressed by the specific α₂-adrenoceptor antagonist, idazoxan. In Fig. 1C we have indicated two possibilities, for both of which there is evidence, which may determine the timing of the inhibitory input. The first is to suppose that LC neurones are connected via collateral axonal or dendritic noradrenaline-releasing inhibitory pathways (Aghajanian et al. 1977; Egan et al. 1983; Ennis & Aston-Jones, 1986; Harris et al. 1992) which are activated by the preceding phase of excitation, the latter deriving from the medulla. Such an intracoerulear inhibitory mechanism would be consistent with the fact that in the brainstem preparation exposure to α₂-adrenoceptor antagonists not only suppresses respiratory-phased inhibition with no effect on the preceding excitation, but also results in an increase in frequency of on-going spike discharge. A second possibility is that the sequence reflects the coordinated activation of excitatory and inhibitory afferents to the LC. If this is the case these afferents are likely to originate in the nucleus paragigantocellularis (located within the RVL) since this is the main medullary source of both excitatory amino acid-mediated excitation and adrenergically mediated inhibition of the LC, and is itself a point of convergence for a variety of afferent inputs involved in the control of the LC (reviewed by Aston-Jones et al. 1991). One reason why it will become important in future to clarify the source of respiratory-phased inhibition is that its suppression contributes significantly to the excitatory response of the LC to increased CO₂ (see below).

What makes the respiratory innervation of the LC of interest is the very extensive nature of the noradrenergic output projections from this nucleus (Foote et al. 1983). Since, on the evidence of trans-synaptic tracing, these targets appear to include (functionally unidentified) elements of the medullary network of phrenic premotoneurones (Dobbins & Feldman, 1994) the LC may feed back directly to the respiratory network. We will consider the possible significance of this later. At this stage we note that if the respiratory-phased burst is regarded as an output ‘signal’ transmitted by the LC, and the on-going spike activity as ‘noise’, then the signal-to-noise ratio becomes smaller as the CO₂ concentration is increased. Two factors contribute to this. The first is that, while the excitatory component of the respiratory input is essentially unaffected by an increase in CO₂ (from 2 to 8 %), the inhibitory component becomes weaker so that the corresponding period becomes occupied by action potentials (Fig. 2A–C). The second is that increasing the CO₂ concentration depolarizes the neurones and results in an increase in the frequency of the non-respiratory component of their spike discharge (Pineda & Aghajanian, 1997; Oyamada et al. 1998). Thus, the modulating effect of the respiratory input, prominent at a low CO₂ concentration, is diminished at a higher concentration. These effects are reversible, they follow the same time course as the CO₂-induced increase in respiratory frequency recorded on the phrenic roots, and so, after a short delay, they also follow the time course of the change in bath pH. That the overall response is physiologically relevant is suggested by the fact that under in vivo conditions systemic hypercapnia results in an increase in LC discharge frequency which parallels the increase in sympathetic outflow (Elam et al. 1981) and phrenic activity.

What is the significance of the respiratory-phased burst discharge of LC neurones? A clue is suggested by the fact that the intensity of this burst (measured as spike frequency) is virtually unchanged over a range of CO₂ concentrations which result in a fairly large (about 50 %) change in the frequency of spontaneous spike discharge, i.e. the response to CO₂ shows selectivity. There is nothing unique about CO₂ in this regard. For example, bath application of noradrenaline suppresses the on-going spike activity and subthreshold oscillation of LC neurones, but leaves their respiratory-phased (presumed glutamatergic) excitation intact (Oyamada et al. 1998) and their burst response to iontophoretically pulsed glutamate almost unaffected (Aston-Jones et al. 1991). This suggests that transmission of the respiratory-phased burst may continue under conditions in which background activity in the LC is negligible. This in turn may reflect a more general requirement for the LC, perhaps connected with its arousal function, to remain responsive to afferent inputs under such conditions. If the LC does innervate the medullary respiratory network then the continued phasic release of noradrenaline may be an important factor in controlling the average level of excitability of the network or certain elements of it (see e.g. Arata et al. 1998) even if the phasic character of this input is subsequently filtered out postsynaptically. In any event, it is clear that the relatively selective nature of the effect of CO₂ on LC activity raises questions as to the mechanism and site of action of this stimulus, and we examine these aspects next.

Are LC neurones intrinsically chemosensitive?

In their study of adult LC neurones (pontine slice) Pineda & Aghajanian (1997) showed that the CO₂-induced increase in tonic discharge frequency was dependent on intracellular acidification and that this was associated with inhibition of the outward current component through a potassium inward rectifier. In addition they reported that the response was retained in the presence of idazoxan, a result which makes it very unlikely that the response is mediated indirectly via depression of noradrenaline release, in effect disinhibiting the synaptically coupled network. Other evidence favouring a mechanism intrinsic to LC neurones is the fact that in the in vitro neonatal brainstem exposure to TTX, at a concentration sufficient to suppress phrenic activity and the corresponding respiratory synaptic input to the LC, leaves both the depolarizing component of the response to CO₂ and the increase in the frequency of the endogenously generated rhythm of (Cd²⁺-blockable) spikes intact (Fig. 2D). The point is important because LC neurones are equipped with a variety of receptors which act to modulate discharge frequency via inwardly rectifying potassium channels, though it must be said there is no direct evidence that TTX-sensitive afferent inputs to the LC are completely blocked under these conditions.

One possibility which should be further investigated follows the finding that LC neurones are equipped with P2 purinoceptors which, when stimulated by (enzymatically stable) ATP, generate a depolarizing response and an increase in spike frequency (Harms et al. 1992). The co-release of ATP from noradrenergic LC neurones into an acidic extracellular space – which may be expected to potentiate the purinergic (P2 X₂) response (King et al. 1996) – may contribute to the TTX-resistant CO₂ response (see Thomas et al. 1999 for a discussion of this point in relation to ventral medullary chemosensitivity). How significant purinergic transmission may be in chemosensitivity is difficult to assess at present because LC neurones – as indeed VRG inspiratory neurones – remain responsive to CO₂ in a low Ca²⁺-high Mg²⁺ medium (Kawai et al. 1996; Oyamada et al. 1999), which certainly depresses Ca²⁺-dependent synaptic transmission, and which in the LC also suppresses the excitatory response to exogenously applied methylene-ATP (Harms et al. 1992).

Electrical coupling and synchronization of LC neurones is retained during their chemosensory response

Since dendritic mechanisms have been implicated in the timing of spontaneous discharge in LC neurones (Williams et al. 1984; Williams & Marshall, 1987) the dendritic tree may also represent a site at which the CO₂ sensitivity of the membrane potential acts to control the output from the nucleus. An important aspect of this control in the neonate arises from the fact that LC neurones form a dye- and electrically coupled network with the coupling apparently mediated by dendro-dendritic interactions (Christie et al. 1989; Christie & Jelinek, 1993; Travagli et al. 1995; Oyamada et al. 1999). Electrical coupling may also be present under some circumstances in the adult (Travagli et al. 1995), for in the adult slice, the capacity of the nucleus to synchronize its output depends on the physical presence of a dense pericoerulear network of dendrites (PCD, Fig. 1C) (Ishimatsu & Williams, 1996). This network, extending for several hundred micrometres outside the compact cell soma region of the nucleus proper (Shipley et al. 1996), may be presumed therefore to represent a major site at which electrically mediated interactions occur.

The most prominent manifestation of such electrotonic coupling in the neonatal LC is the synchronized occurrence of spontaneous subthreshold oscillations of membrane potential in paired recordings. These persist in the absence of chemical synaptic transmission and after pharmacological blockade of a variety of potential transmitters (Christie et al. 1989; Travagli et al. 1995). The frequency of oscillation is sensitive to CO₂, approximately doubling when the concentration is raised from 2 to 8 % (Oyamada et al. 1998). Since the effect of this presumed dendritically located oscillation (Williams & Marshall, 1987) is to trigger spike discharge (Fig. 3A) the mechanism of oscillation becomes crucial to understanding the action of CO₂. Although little is known about the mechanism, some of the conditions influencing its occurrence are beginning to emerge.

Figure 3. Synchronization of the electrically coupled rhythm within the LC network remains effective during the chemosensory response.

A–C, perforated patch recording (V_m) from the same LC neurone under control conditions (A) and during exposure to a medium of reduced Ca²⁺ (0.2 mM) and increased Mg²⁺ (5 mM) concentration (B and C). The frequency of this rhythm was increased and intraburst discharge frequency reduced on switching from 2 to 8 % CO₂. Note suppression of phrenic discharge (∞Phr) and respiratory synaptic input in low Ca²⁺-high Mg²⁺. D shows rhythmic activity in low Ca²⁺-high Mg²⁺ as a control. E (same neurone), loss of membrane potential oscillation in low Ca²⁺-high Mg²⁺ plus carbenoxolone (300 μm, 25 min), but retention of the capacity to discharge spikes in response to current injection. F, an extracellular recording showing multi-unit synchronization in low Ca²⁺-high Mg²⁺ as a control. G (same recording as F), desynchronization after 15 min exposure to carbenoxolone. (Modified after Oyamada et al. (1999) and reproduced with permission from Respiration Physiology, Elsevier Science.)

In the brainstem preparation exposed to a low Ca²⁺-high Mg²⁺ medium all LC neurones exhibit a regular rhythm of large amplitude, slow oscillations of membrane potential (Fig. 3B). The effect of each cycle of depolarization is to trigger a rapid train of smaller oscillations – resembling those seen in normal medium and which we will refer to as dendritic spikes – and these, in turn, trigger full-sized spikes. Extracellular recordings show that this rhythm is synchronized throughout the nucleus. The large oscillation is blocked by Cd²⁺. The entire rhythm is suppressed by TTX, but the large oscillation is restored by addition of Ba²⁺ and TEA. Thus, in low Ca²⁺-high Mg²⁺ LC neurones exhibit a synchronized pattern of burst discharge which is dependent on an external source of Ca²⁺ or a source of depolarizing current via Cd²⁺-sensitive channels. If voltage-sensitive Na⁺ channels are normally involved in sustaining the large oscillation, they are resistant or inaccessible to somatic injection of QX314. The emergence of such synchronization is somewhat similar to the situation shown by hippocampus CA1 neurones in a zero Ca²⁺ medium where the effect appears to be due to enhanced gap junction coupling (Perez-Velazquez et al. 1994). There is, however, an important difference between these neurones. In the LC the effect of increasing the CO₂ concentration (from 2 to 8 %), and therefore probably producing a significant intracellular acidification (Ritucci et al. 1997; Huang et al. 1997), is not to suppress the rhythm as in the hippocampus, but to increase its frequency (on average by 35 %), i.e. the rhythm remains synchronized and the only obvious effect which might be due to pH-induced gap junction uncoupling is a (reversible) reduction in intraburst discharge frequency (Fig. 3B and C). Exposure to the gap junction uncoupler carbenoxolone (or to very high concentrations (> 20 %) of CO₂) suppresses the oscillation of the membrane potential, but has no effect on the capacity of LC neurones to discharge full-sized spikes (Fig. 3D and E; see also Travagli et al. 1995). Under these conditions the discharge becomes desynchronized (Fig. 3F and G) and individual neurones discharge at a frequency determined by the prevailing membrane potential (Fig. 3E; see also Travagli et al. 1995). This, in turn, depends on the CO₂ concentration (Oyamada et al. 1999).

There are several reasons why this low Ca²⁺-high Mg²⁺-induced rhythm is of interest. First, it reinforces the notion, previously proposed for the LC (Ishimatsu & Williams, 1996) and some other networks (e.g. Draguhn et al. 1998), that electrical coupling may be a requirement for the generation of some forms of synchronized oscillatory activity. Second, it reveals evidence of a CO₂-sensitive oscillation of potential which appears to trigger dendritic spikes. Third, it suggests that electrical coupling within the LC network is retained at CO₂ levels which are unlikely to be exceeded in vivo (see also Huang et al. 1997). Finally, the behaviour of LC neurones in low Ca²⁺-high Mg²⁺ reveals a remarkable similarity to the behaviour of certain classes of VLM respiratory neurone which have been implicated in respiratory rhythmogenesis in vitro (see below).

Chemosensitivity in respiratory rhythmogenic and non-rhythmogenic regions of the VLM

There is direct evidence for the existence within the PBC of identifiable inspiratory neurones in which the rhythm shows voltage-sensitive properties, i.e. the occurrence of rhythm is conditional on depolarization to an appropriate range of membrane potentials (Koshiya & Smith, 1999). However, one very striking feature of some PBC respiratory neurones (Johnson et al. 1994) and also of pre-inspiratory neurones of the RVL (Onimaru et al. 1989, 1995) is that when they are exposed to a low Ca²⁺-high Mg²⁺ medium they lose their type-specific pattern of discharge and exhibit instead a rhythmic burst discharge which resembles that shown by LC neurones in this medium. This similarity is reinforced by the finding that in low Ca²⁺-high Mg²⁺ the burst frequency of pre-inspiratory neurones is increased when the CO₂ concentration is raised (Onimaru et al. 1989). Although dye (biocytin) and electrical coupling have been demonstrated between inspiratory neurones of the nucleus ambiguus (Rekling & Feldman, 1997) and coupling is a prominent feature of dorsal medullary chemosensitive neurones (Huang et al. 1997), there is, as yet, no evidence to our knowledge for its existence within the pre-inspiratory network or among the variety of neurone types within the PBC. The possibility nevertheless remains that, at least in low Ca²⁺-high Mg²⁺, these neurones form one or more gap-junction coupled network(s) and that the rhythm is the product of the interaction of this coupling with the intrinsic conductances of these neurones (see also Butera et al. 1999). If this putative coupling occurs at the dendro-dendritic level, then it may provide a means of synchronizing slow CO₂-induced changes in membrane potential within the network.

There are at present no reports of the membrane potential response of rhythmogenic respiratory neurones to a change in CO₂ concentration. However, this has been examined in non-rhythmogenic inspiratory neurones of the VRG, and here the membrane potential is clearly sensitive to CO₂ (Kawai et al. 1996). In neurones of this kind the phasic character of their synaptically driven inspiratory discharge is retained at increased CO₂ concentrations when the neurone depolarizes (Fig. 4A), and eliminating voltage-sensitive Na⁺ currents (with TTX) results in a flat membrane potential with a level dependent on the prevailing CO₂ concentration (Fig. 4B). These neurones do not show rhythmic activity in low Ca²⁺-high Mg²⁺, though they do retain their depolarizing response to CO₂.

The behaviour shown by inspiratory neurones of this kind might be explained if the CO₂-induced depolarization, which is accompanied by a modest increase in input resistance, is of remote dendritic origin, perhaps even originating in other cells electrically coupled to the recorded cell. In this case, when viewed from the cell soma, remotely located membrane chemosensors may function as a current source which passively displaces the membrane potential. This might account for the fact that in these neurones the only obvious consequence of such depolarization is to reduce the amplitude of their respiratory-phased excitatory synaptic potential, typically leaving intraburst discharge frequency unaffected. There is no direct evidence for such a remotely located chemosensor, which remains therefore no more than conjecture, but there is one piece of morphological evidence which would be at least consistent with this general idea. This consists in the finding that chemosensitive neurones in this region, which are not confined to respiratory neurones, exhibit dendritic projections to within a short distance (< 50 μm) of the ventrolateral surface, and that this pattern of projections is independent of the depth of the parent cell soma which may vary over several hundred micrometres (Fig. 1C; Kawai et al. 1996), i.e. they project into the ‘traditional’ site of respiratory chemosensitivity. If it is the function of these dendrites to detect pH changes at the surface and to transmit the response to the deeper lying somata (but see Voipio & Ballanyi, 1997), then this function is presumably dependent on the typically low frequency character of physiologically occurring changes in pH and P_CO2. The chemosensor itself could be located within the dendrite, within some associated cell connected by gap junctions, or perhaps in one releasing transmitter with low or no Ca²⁺ sensitivity and not dependent on TTX-sensitive Na⁺ conductances.

Finding a respiratory function for chemosensitive neurones

As pointed out earlier the structure of the inspiratory burst in VRG neurones is almost unchanged by CO₂-induced depolarization. In essence this resembles the situation in LC neurones, but with the important difference that the latter are equipped with conductance mechanisms which convert the depolarizing response to an increased frequency of tonic spiking, i.e. it is their endogenously generated rhythm which is sensitive to CO₂. In other words in both classes of respiratory-innervated neurone the action of CO₂ appears to be relatively selective. Are there any grounds for believing that this kind of chemosensitivity is at all relevant to the control of respiration? In the case of the LC a plausible case can be made for supposing that the increased activity which accompanies hypercapnia may feed back at some stage to the medullary respiratory network, and may exert a stimulating action on some endogenous bursters (Arata et al. 1998). It is very much more difficult to suggest a function for chemosensitive VRG inspiratory neurones in relation to the in vitro respiratory rhythm, but this may simply reflect the way this rhythm is generated. Chemosensitivity at the VRG level may have (or acquire) a function when the rhythm is of the normal eupnoeic kind, and therefore requiring a different preparation for its study.