Redefining the components of central CO2 chemosensitivity – towards a better understanding of mechanism

Abstract The field of CO2 chemosensitivity has developed considerably in recent years. There has been a mounting number of competing nuclei proposed as chemosensitive along with an ever increasing list of potential chemosensory transducing molecules. Is it really possible that all of these areas and candidate molecules are involved in the detection of chemosensory stimuli? How do we discriminate rigorously between molecules that are chemosensory transducers at the head of a physiological reflexversusthose that just happen to display sensitivity to a chemosensory stimulus? Equally, how do we differentiate between nuclei that have a primary chemosensory function, versusthose that are relays in the pathway? We have approached these questions by proposing rigorous definitions for the different components of the chemosensory reflex, going from the salient molecules and ions, through the components of transduction to the identity of chemosensitive cells and chemosensitive nuclei. Our definitions include practical and rigorous experimental tests that can be used to establish the identity of these components. We begin by describing the need for central CO2 chemosensitivity and the problems that the field has faced. By comparing chemosensory mechanisms to those in the visual system we suggest stricter definitions for the components of the chemosensory pathway. We then, considering these definitions, re-evaluate current knowledge of chemosensory transduction, and propose the ‘multiple salient signal hypothesis’ as a framework for understanding the multiplicity of transduction mechanisms and brain areas seemingly involved in chemosensitivity.


The importance of chemosensitivity for breathing
To maintain homeostasis in the face of ever changing circumstances that arise from activity and environmental conditions, physiological systems are subject to numerous chemosensory feedback mechanisms. Respiratory chemosensitivity is the ability of the brain to detect changes in CO 2 and alter physiological systems to regulate its levels within tightly controlled parameters. CO 2 sensitive cells are predominantly within the brain (Feldman et al. 2003), though a smaller contribution comes from the carotid bodies (Peers & Buckler, 1995). Mean arterial P CO 2 (P a CO 2 ) does not change with exercise (Forster et al. 1986) and remains constant from well below to above sea level, until prolonged periods are spent there or extreme altitudes (>2500 ft (762 m)) are reached (Haldane & Priestley, 1905;Catron et al. 2006). P a CO 2 is even regulated at the expense of P O 2 (Haldane & Priestley, 1905), and an increase in P a CO 2 by just 1 mmHg causes a 20-30% increase in ventilation (Feldman et al. 2003) and increasing inspired CO 2 to 4% increases ventilation by 177% (Haldane & Priestley, 1905). The primacy of CO 2 in controlling respiration is evident from the earliest land dwelling animals, amphibians (Tayor et al. 2003), suggesting a high degree of evolutionary conservation in this vital physiological reflex. Recently, however, the idea that basal levels of P CO 2 drive normal breathing (originating from Haldane & Priestley, 1905) has been called into question, as genetically altered mice apparently lacking CO 2 chemosensitivity can still breathe and survive to adulthood (Ramanantsoa et al. 2011).

Other physiological chemosensitive reflexes
Hypercapnia can become a life threatening physiological condition through dangerous alterations in blood pH and the increased chance of asphyxiation. Thus, several protective physiological reflexes and behaviours exist. The cardiovascular response, hypertension and bradycardia (Oikawa et al. 2005), may allow better alveolar gas exchange or counteract the effects of CO 2 on vasodilatation and autoregulated organs (i.e. the brain and kidney). Renal output alters when blood CO 2 is high, which conserves bicarbonate ions that buffer pH changes caused by hypercapnia. Arousal from sleep and increased vigilance are well known responses to elevated CO 2 (Haxhiu et al. 2001;Johnson et al. 2005;Williams et al. 2007;Buchanan & Richerson, 2010) warning us that either ventilation is inadequate or a dangerous build up of atmospheric CO 2 has occurred. A lack of this response is associated with dysfunctions in chemosensation that lead to death (e.g. sudden infant death syndrome). The association of fear/anxiety to hypercapnia (Papp et al. 1993) causes us to leave, or not enter, areas where CO 2 levels exceed survivable limits.

The problems with chemoreception
Comparison to the visual system. To provide a more encompassing framework to consider CO 2 chemosensory processes, we reflect on a better understood sensory system. In the visual system, there is an easily identifiable sensory organ (the eye), which contains an obvious light-detecting neural nucleus, the retina. The retinal photoreceptors contain the entire transduction process that converts photons of light into the release of a primary transmitter that communicates with second order neurons. The photoreceptors, which are directly sensitive to different aspects of the incident light stimulus, are at the head of parallel pathways of visual information, the processing and abstraction of which begins within the retina itself. These pathways are kept separate and recombined in complex ways through convergence, to abstract further features of the visual input. Even in a system with a recognisable sensory organ and primary neural nucleus, identification of the primary photosensitive cells has not been trivial. Within the retinal ganglion cell population, light-sensitive cells are mixed in with (non-light sensitive) follower cells (Foster et al. 1991;Foster & Bellingham, 2004;Hankins et al. 2008). The discovery of light-sensitive ganglion cells and an understanding of their function, was only achieved recently through identification of the opsin-based transduction process within them. Interestingly, the removal of the opsins that govern vision did not alter the circadian clock; instead circadian input depends on transduction mediated by melanopsin. Thus, in the eye there are parallel sensory systems that are initiated by the same stimulus, but with distinct transducers, which are juxtaposed but subserve different behaviours.
Consideration of the visual system thus provides a useful analogy when thinking about CO 2 chemosensitivity. In this system the central chemosensing organ is the brain itself and easily identifiable nuclei and chemosensory pathways do not exist. Several central chemosensing 'retinas' have been proposed, each using a different signal transduction pathway and releasing distinct neurotransmitters. Like light sensitive ganglion cells of the retina, primary chemosensitive cells in the CNS are mixed in with other cells and are not readily recognizable.
The key message from the retina is that identification of the transducing molecules is an integral part of the definition of primary sensory cells. These primary chemosensitive cells may be organised into chemosensory nuclei and multiple such nuclei may subserve different functional systems. Furthermore within the chemosensitive pathways the second or higher order neurons will respond to chemosensitive signals but may not themselves be intrinsically chemosensitive -just as neurons within the visual pathway respond to light impinging on the retina, and are thus light sensitive, without being intrinsically so. Different aspects of the chemosensitive signal could be handled by parallel pathways that ultimately converge and the chemosensitive nuclei and pathways are likely to be organised into integrated systems (Fig. 1).

Maturation of chemosensory responses -practical
problems for investigators. Central chemosensory responses begin on embryonic day (E14.5) before the onset of fetal breathing movements (Eugenin et al. 2006;Thoby-Brisson et al. 2009). In neonatal preparations hypercapnia stimulates ventilation without eliciting metabolic responses (changes in O 2 consumption, thermoregulation or metabolic rate) that modify demand for ventilation (Mortola & Lanthier, 1996). They increase ventilation via changes in tidal volume prior to postnatal day P3 (Stunden et al. 2001), and frequency thereafter (Wickström et al. 2002). The level of respiratory chemo-sensitivity varies with age , being high between P1 and P5, waning to a nadir at P8, and rising again until the adult response is reached by P21 (Stunden et al. 2001;Putnam et al. 2005). The decrease in the hypercapnic ventilatory response corresponds to a reduction in hypercapnia-induced c-fos staining, occurring after P6 (Wickström et al. 1999). This may arise because chemoreceptors are closer to the surface, where they are more able to detect fluctuations in blood chemistry, in neonates than adults (Forster et al. 1997). Compensation after P8 may be due to the migration of dendrites to the surface to re-establish their association with blood vessels. It appears more likely though that areas important in neonatal chemosensitivity decrease in importance with age between P1 and P8, and areas important in adult chemosensitivity develop around P12. Hence there is a hiatus between when the neonatal areas switch off and the adult regions switch on. Chemosensitivity increases after this point as the adult mechanisms mature. Once adult chemosensitivity is established it appears to remain constant throughout life, at least in humans (Browne et al. 2003).
This has very important consequences for both experimental design and interpretation of data. Several Figure 1. Conceptual organization of chemosensory systems A, a primary chemosensory cell is defined by possession of a molecular transducer for one of the salient chemosensory signals (pH i , pH e , CO 2 , HCO 3 − ). The link between the transducer molecule and the behavioural reflex must have been established by genetic or pharmacological manipulation. B, a primary chemosensory nucleus contains primary chemosensory cells (blue rectangles). These may be intermingled and synaptically connected to non-chemosensitive follower cells (red rectangles). C, a chemosensory system may comprise more than one primary chemosensory nucleus possibly detecting different salient chemosensory signals and abstracting different dynamic features of the stimulus (coloured lines are schematic of different types of post-stimulus time firing histograms for the chemosensory cells). These inputs may pass either directly to the effector nucleus or through secondary relays where integration of the signals may occur. In addition, there may be convergent modulatory nuclei that can alter the responsiveness and sensitivity of the system. J Physiol 589.23 different preparations are used to study chemoreception, each confined to a specific developmental period. The brainstem spinal cord (Suzue, 1984) and acute rhythmic transverse slice (Smith et al. 1991) preparations are limited to the neonates. The in situ preparation (Paton, 1996) can be used for study of mechanisms in juvenile to adult rats. Whole animal experiments are usually limited to the adult. Thus when performing any experiment it is important to use the correct preparation for the appropriate nucleus; it is counterproductive to use neonatal preparations to study nuclei that are important in adulthood and vice versa. The use of inappropriate preparations has led to much conflicting data in the field of chemoreception. It is important therefore to consider the nucleus and the developmental period of the tissue at the time of experimentation when interpreting data.
Molecules and terminology. Once in the extracellular fluid, CO 2 combines with water (facilitated by carbonic anhydrase (CA)) to produce bicarbonate ions and protons ( Fig. 2), resulting in three possible species that could act as chemosensory signals, CO 2 , HCO 3 − and H + . For many years the popular consensus was CO 2 is detected entirely through its proxy, pH (Loeschcke, 1982). At first the effects of pH were deemed to be extracellular, though now changes in intracellular pH are known to be important. This concentration on pH has led to the frequent use of imprecise language to describe chemosensitivity -the term 'CO 2 chemosensitivity' commonly refers to detection of pH. Although the evidence does not rigorously exclude other potential signalling molecules, the involvement of direct detection of CO 2 or HCO 3 − has only recently gained attention. Careful simultaneous adjustment of P CO 2 , [HCO 3 − ] and pH, has shown that different chemosensitive cells respond to different stimuli (Filosa et al. 2002;Wang et al. 2002;Mulkey et al. 2004;Hartzler et al. 2008;Huckstepp et al. 2010a,b). Given that molecular transducers for detection of CO 2 and HCO 3 − have emerged, raising the idea of different forms of chemosensitivity, it is important to adjust the language of the field accordingly, i.e. pH chemosensitivity, bicarbonate chemosensitivity and CO 2 chemosensitivity. Even more importantly, the analysis of mechanisms should identify the relevant signalling species at each of the sites involved in chemosensitive reflexes.
As CO 2 , HCO 3 − and H + are intimately linked via CA it is very difficult to isolate which of them acts as the salient signalling molecule in any physiological reflex. CA in effect plays a key role in creating multiple physiological signalling molecules, and is active in many chemosensory nuclei (Coates et al. 1993). There are several isoforms of CA with differing properties: CA IV requires high levels of CO 2 ∼20-30% (Chandrashekar et al. 2009) whereas CA II only requires ∼1% (Hu et al. 2007); some can be modulated by phosphorylation (Narumi & Miyamoto, 1974); and at least one subtype binds to and modulates acid transporters (Li et al. , 2006b, thus altering intracellular pH by a second and independent mechanism. Therefore, the expression of different CA subtypes may play an important role in the chemosensitivity of different nuclei. A further practical difficulty arises from studies that substitute the physiological CO 2 /HCO 3 − buffering system with Hepes buffered salines (Mulkey et al. 2004;Huang et al. 2006;Dergacheva et al. 2011). This substitution has several potentially serious drawbacks: it alters both extracellular and intracellular buffering capacity (as CO 2 readily crosses membranes); it removes the influence of carbonic anhydrase; the lowered intracellular concentration of HCO 3 − will affect the activation of soluble adenylyl cyclase and a number of transport systems for HCO 3 − ; and Hepes and other organic aminosulphonated compounds can inhibit gap junctions and hemichannels (Bevans & Harris, 1999;Tao & Harris, 2004). At least one gap junction hemichannel protein, connexin26 (Cx26), is important in chemosensory transduction (Huckstepp et al. 2010a,b). Therefore considerable caution when interpreting these data is required, as it cannot necessarily be applied to the in vivo system. An example of this has been reported in a recent study where the presence of the CO 2 /HCO 3 − was essential to see the involvement of ATP in the responses of chemosensory neurons (Wenker et al. 2010).

Anaesthesia, conscious animals and chemosensitivity.
Anaesthetics can depress the firing of neurons and interfere with components of chemosensitivity (Knill et al. 1983;Koh & Severinghaus, 1990;Nattie, 2001), potentially altering the balance of mechanisms in whole animal responsiveness to CO 2 . It is very important to also establish and test mechanisms of chemosensitivity, initially posited from reduced in vitro preparations and anaesthetised animals, in awake freely behaving animals.

Towards a solution -definitions
Chemosensory transducer. As the mechanism of transduction of a chemosensory signal is crucial in classifying intrinsic chemosensory cells and primary chemosensory transmitters ( Fig. 1), we need to define rigorous criteria for the identification of chemosensory transduction molecules. This is not straightforward as it is likely that many molecules have incidental sensitivity to various aspects of chemosensory stimuli, i.e. while a molecule may be sensitive to pH, this may not play a role in the intrinsic chemosensitivity of a given cell or nucleus.
We define a chemosensory transducer as a molecule that responds to fluctuations of a salient chemosensory stimulus (pH (either internal or external), P CO 2 , HCO 3 − ) in a manner that results in significant changes in cellular excitability or cell-cell signalling. Critically, genetic deletion and/or selective blockade of this molecule must reduce or ablate the behavioural chemosensory reflex in conscious animals, thus showing that the molecule is necessary to endow a cell/nucleus with intrinsic chemosensitivity. Expression of the relevant molecule should be sufficient to endow a (previously unresponsive) cell with responsiveness to the salient chemosensory signal. Demonstrating the presence of a pH-CO 2 -or bicarbonate-sensitive molecule in a cell is insufficient; it is essential to show the presence of this molecule is necessary for the chemosensory reflex.
Intrinsic chemosensitivity. An intrinsically sensitive chemosensory cell (or primary chemosensor) must by definition respond to at least one salient stimulus (CO 2 , HCO 3 − or pH (either internal or external)), directly, without dependence on any other cell type in the CNS (Fig. 1). This cell should then communicate, directly or indirectly, with the physiological centres of breathing, arousal, etc. to affect the relevant reflex.
The practical test for intrinsic chemosensitivity has involved demonstration of responses to the relevant chemosensory stimulus following lowering of extracellular Ca 2+ or in the presence of TTX or antagonist cocktails (Kawai et al. 1996;Nattie, 2000;Nattie & Li, 2002;Mulkey et al. 2004;Richerson, 2004). As these traditional experimental criteria do not take into account modern knowledge of the multiplicity of mechanisms for cell to cell communication now evident in the CNS (e.g. transmitter release from glial cells does not require extracellular Ca 2+ and is TTX independent), they are not definitive and leave open the possibility that chemosensitivity is secondary to the release of a transmitter from an (unknown) primary chemosensory cell.
We therefore propose the following definition: an intrinsically chemosensitive cell must contain a molecular transducer that enables it to respond directly to the relevant chemosensory stimulus. The necessity of this molecular transducer for chemosensory responses must be demonstrated by either genetic deletion or selective antagonism. Using this definition, a cell containing a pH/CO 2 /bicarbonate sensitive molecule is not a primary chemosensor unless that molecule has been causally linked to at least one of the relevant behavioural chemosensitive reflexes.
Primary chemosensory transmitters. It follows from the above that a primary chemosensory transmitter is that which is released from an intrinsic chemosensory cell ( Fig. 1). In a complex pathway involving a sequence of several cells with interposed synapses, it is likely that several transmitters will be released in response to a chemosensory stimulus. Many of these, however, will be as a secondary or downstream consequence. Identification of the transmitter released by the intrinsic chemosensors is non-trivial and will depend upon a series of experimental approaches: directly measuring transmitter release, determining the presence of release mechanisms and genetic and/or pharmacological manipulation of the transduction mechanism/signalling pathway.
Following in the footsteps of Werman (1966), we propose the following criteria necessary to demonstrate the involvement of a neurotransmitter in the chemosensory cascade. The candidate transmitter must be: present in the primary chemosensitive cells; released in response to a chemosensory stimulus prior to activation of the network response; and released onto (or able to reach by diffusion) the effector nuclei that produce the physiological reflex. Finally, selective antagonism of the transmitter must reduce or ablate a physiological response during hypercapnia and application of its agonist must evoke the reflex. Primary chemosensory nucleus. A primary sensory nucleus must contain intrinsically chemosensitive cells established by the definition given above (i.e. contain a proven molecular transducer). In the absence of these cells, a chemosensory nucleus must be considered as a relay in the chemosensory reflex ( Fig. 1).

A reappraisal of the current literature
We now re-examine the evidence for the current molecules, transducers, intrinsic chemosensors and pathways in the light of the definitions outlined above.

Modalities of chemosensing: salient stimuli
A recent comprehensive review of salient stimuli and their targets, in all multicellular organisms regardless of function, has been written (Tresguerres et al. 2010). Here we focus specifically on molecules proposed to be involved in the physiological reflexes associated with hypercapnia in mammals.
pH. The primacy of pH in chemoreception is reflected in other fields of physiology, i.e. CO 2 cerebrovascular reactivity and the control of vasodilatation (Brian, 1998;Ainslie & Duffin, 2009). As pH has been largely regarded as the chemosensory signal, much research effort has hinged on studying pH-sensitive ion channels as potential molecular transducers. So far genetic manipulation of these channels has failed to provide evidence for their involvement in mechanisms of chemoreception. In addition, much prior work show strong indications that pH is not the only chemosensitive signal (Pappenheimer et al. 1965;Fencl et al. 1966;Loeschcke, 1982;Eldridge et al. 1985;Shams, 1985). For example, comparing respiratory output after applying acidic saline to the medulla oblongata or inhalation of additional CO 2 to induce the same pH change showed that the combination of an increase in P CO 2 and a decrease in pH was a more powerful stimulant than a decrease in pH alone.  (Eldridge et al. 1985). Furthermore combined CO 2 inhalation with intravenous infusion of balancing HCO 3 − to prevent changes of pH measured at the ventral surface of the medulla showed that an increase of PCO 2 at the medullary surface by itself was able to increase respiratory rate (Shams, 1985). This provides strong support for independent effects of pH and PCO 2 on breathing.
CO 2 . Recognition of CO 2 as a salient chemosensory signal has been slow to emerge as interactions of CO 2 with potential transducer proteins have been largely overlooked by physiologists. Insects such as mosquitoes are sensitive to CO 2 , mediated through G-protein coupled receptors (Jones et al. 2007). That CO 2 binds with haemoglobin and how this may apply to direct detection of CO 2 in other contexts has gone unnoticed; here CO 2 reacts to form a carbamate moiety (via a labile covalent bond) on each of the terminal amines of the two α and two β globin chains (Kilmartin & Rossi-Bernard, 1971). In microbial enzymology (Maveyraud et al. 2000) and in the fixation of CO 2 (Lundqvist & Schneider, 1991), enzymes are activated by formation of a carbamate moiety on specific lysine side chains. Evidence now suggests specific molecular mechanisms by which CO 2 can interact directly with a number of gap junction hemichannels (Huckstepp et al. 2010a,b), affect inwardly rectifying potassium channels (Huckstepp & Dale, 2011) and bind directly to soluble adenylyl cyclase (sAC) (Townsend et al. 2009). There is no longer an absence of a plausible molecular mechanism for the detection of CO 2 , providing an impetus for the recognition of CO 2 itself as a salient chemosensory stimulus.

HCO 3
− . A potential molecular detector for HCO 3 − is sAC (Zippin et al. 2001;Townsend et al. 2009;Tresguerres et al. 2010), giving a possible signalling mechanism through the production of cyclic AMP (cAMP); this has been implicated in the chemosensitive response of both the retrotrapezoid nucleus (RTN) (Ritucci et al. 2005) and glomus cells of the carotid body (Summers et al. 2002;Nunes et al. 2009). However, can the fluctuations of HCO 3 − during hypercapnia be sufficient to change the activation of sAC in a functionally significant manner? sAC is most sensitive to bicarbonate ions in the low millimolar range (Chen et al. 2000), whereas levels in CSF are usually ∼19 mM and rise by only 3 mM during extended periods of hypercapnia (Kazemi et al. 1976), translating into a modest increase in sAC activity (Chen et al. 2000). Whether this increase is functionally significant depends on sAC expression levels, cellular cAMP concentration and the extent to which this varies as a result of modulation of sAC by changing HCO 3 − concentrations. That activation of sAC can cause functional changes in cilia of the airway (Schmid et al. 2007) implies that it may be. Therefore, we think it now useful to evaluate more systematically sAC as a potential chemosensory transducer.

Transducing molecules
Candidates with genetic evidence. Acid sensing ion channel (ASIC). ASICs are expressed in regions of the brain known to be linked to fear associated with hypercapnia, such as the amygdala (Ziemann et al. 2009). Cultured neurons that lack the ASIC1a channel no longer respond to alterations in CO 2 , whereas wild-type neurons show a strong and graded response to it (Ziemann et al. 2009); ASIC1a is therefore sufficient to endow neurons with chemosensitivity. ASIC1a knockout mice have a severely blunted fear response associated with hypercapnia (Ziemann et al. 2009), demonstrating that this molecule is necessary for the full hypercapnic response. This provides the first (and only) genetic evidence for a primary chemosensory transducer (Table 1).
Candidates with pharmacological evidence. Connexin 26 (Cx26). Cx26 is expressed on the leptomeninges and sub-pial astrocytes, including those at the ventral surface of the medulla that respond to hypercapnia (Huckstepp et al. 2010b). It responds to changes in P CO 2 by opening to release ATP, which subsequently causes adaptive changes in ventilation (Huckstepp et al. 2010a,b). Cx26 seems directly sensitive to CO 2 as it remained chemosensitive in isolated patches of both the inside-out and outside-out configuration (Huckstepp et al. 2010a). Expression of Cx26 by itself in HeLa cells was sufficient to make them sensitive to CO 2 and to recapitulate CO 2 -sensitive ATP release, similar to that seen from the surface of the medulla (Huckstepp et al. 2010b). Crucially, pharmacological blockade of this channel reduced hypercapnia elicited ATP release in vitro and in vivo, and reduced the hypercapnia induced adaptive changes in breathing in vivo under anaesthesia (Huckstepp et al. 2010b). Current evidence thus suggests that Cx26 may be both necessary and sufficient for CO 2 chemosensory transduction (Table 1). Nevertheless targeted and conditional genetic manipulation of Cx26 expression (complicated by the lethality of unconditional deletion of this gene) will be required to elucidate the contribution of Cx26 to the full physiological responses to hypercapnia including those in awake animals.
Candidates present with contradictory evidence for involvement. Tandem-pore acid sensing potassium (TASK) channels. TASK-1 channels contribute to the pH sensitivity of raphé neurons (Mulkey et al. 2007b) and maybe the locus coeruleus (LC) (Filosa & Putnam, 2003) and could confer intrinsic chemosensitivity to the preBötzinger Complex (preBötC) (Koizumi et al. 2010) and orexinergic neurons of the lateral hypothalamus (Williams et al. 2007). Since these acid sensing channels are found in the respiratory centre and in cell groups important for chemosensory function, it is possible their activation may lead to adaptive breathing associated with hypercapnia. Nevertheless, constitutive deletion of the TASK-1 and TASK-3 genes does not alter the hypercapnic ventilatory responses (Mulkey et al. 2007b). Thus, these channels are either not essential for the central chemosensitive reflex or compensatory mechanisms obscure their role (Mulkey et al. 2007b;Trapp et al. 2008). Only targeted inducible genetic manipulations will discriminate between these possibilities. Nevertheless, TASK-1 is required for correct function of peripheral chemosensors in the carotid body, and contributes to the whole animal response to lower levels of hypercapnia (Trapp et al. 2008). Overall there is considerable doubt about the role of TASK channels as central chemosensory transducers (Table 1).
pH sensitive inwardly rectifying potassium (KIR) channels. pH-dependent closure of inwardly rectifying potassium channels could underlie the pH sensitivity of some neurons and glia i.e. KIR 4.1/5.1 heteromer (Xu et al. 2000) on sub-pial astrocytes in the RTN (Wenker et al. 2010). Here channel closure leads to depolarization of the astrocytes evoking vesicular ATP release (Wenker et al. 2010), and pH-dependent ATP release from astrocytes in the RTN can enhance breathing . However, the link between KIR 4.1/5.1 and the behavioural reflex remains to be definitively established. This possibility was weakened by recent genetic evidence showing KIR5.1 −/− mice have an intact central chemosensitive response (Trapp et al. 2011), though these experiments may be confounded as these mice have chronic metabolic acidosis (Trapp et al. 2011). Additionally, LC neurons derived from these mice show a significant decrease in their pH sensitivity (D'Adamo et al. 2011), but this effect on isolated neurons has not been extended in vivo or to the behavioural response. Since manipulation of the LC appears to be important in central chemoreception, it is possible that these mice have compensated for the loss of this protein either by upregulation of another subunit, or by plastic changes that place greater emphasis on other nuclei (Trapp et al. 2011).
Overall the evidence for involvement of KIR 4.1/5.1 as a chemosensory transducer is inconclusive and will remain so until targeted conditional deletion of these genes has been attempted (Table 1).
pH i sensitive channel in raphé. The raphé complex probably contains multiple pH-sensitive channels. Whilst TASK-1 may endow neurons with pH e sensitivity (but see above), raphé neurons have also been shown to be sensitive to pH i in vitro Ribas-Salgueiro et al. 2005) and to some extent in vivo (Veasey et al. 1995;Ribas-Salgueiro et al. 2003;Mulkey et al. 2004) suggesting the presence of a further channel sensitive to pH i . These molecules may not necessarily co-exist in raphé neurons at the same time. Instead, it may be that as chemosensitivity, which changes developmentally , switches from its neonatal to adult form, raphé neurons change their expression of pH sensitive channels to alter their sensitivity to hypercapnia. Nevertheless the status of these unknown channels as chemosensory transducers will remain uncertain until they have been identified and clear pharmacological or genetic evidence links them to an adaptive reflex (Table 1).
Outwardly rectifying potassium channel. Although the chemosensitivity of RTN neurons is preserved in the presence of blockers of excitatory transmission and in TTX (Mulkey et al. 2004), these manipulations are not rigorous tests of intrinsic chemosensitivity. Nevertheless RTN neurons may contain a chemosensory transducer of some form and their activation by acidification occurs via reduction of an outwardly rectifying potassium current (Wellner-Kienitz et al. 1998;Mulkey et al. 2004;Guyenet, 2008), making the channel underlying this current a putative candidate transducer. However until the gene encoding this outward rectifier channel is identified a rigorous test of its role in controlling excitability and adaptive ventilatory behaviour remains out of reach.

Chemosensitive anatomical structures
Putative chemosensitive areas are found close to the surface of the brain with a denser capillary network than the surrounding tissue, usually with their neuronal axon terminals and somata abutted to them (Okada et al. 2001;Okada et al. 2002); allowing them to sample J Physiol 589.23 blood before it is affected by the intrinsic metabolism of surrounding tissue. These blood vessels are surrounded by a perivascular space (Okada et al. 2001), with elongated and labryrinthic invaginations of the sub-arachnoid space which form cisternae (Okada et al. 2001), creating a large surface area, with a higher rate of blood delivery, which allows for greater blood gas sampling (Göbel et al. 1990). A large number of putative chemosensory areas have been described in the literature that contribute to several different CO 2 -sensitive reflexes. However, a primary chemosensory nucleus must contain primary chemosensitive cells that are directly activated by hypercapnia, defined through their possession of a chemosensory transducer causally linked to the chemosensitive reflex by genetic deletion or pharmacological experimentation. A primary chemosensory area must show direct or indirect connections to a region that is ultimately responsible for altering the output of the reflex (i.e. in adaptive breathing it must ultimately connect to the preBötC). Chemosensitive areas that lack primary chemosensory cells are most likely relay nuclei. It is not necessary that all chemosensitive structures should be arranged in a serial pathway; some regions may be multifunctional and subserve both the chemosensitive and physiological responses (Arendt, 2008).
Though often studied individually, interactions between the leptomeninges, vasculature, astrocytes and neurons within a chemosensory nuclei may mean that it is better to view these components as a coordinated functional unit. In the RTN/pFRG, meningeal cells control the migration of neurons, angiogenesis and glial function, ensuring the precise development and regulation of this chemosensory unit from before birth (Schwarz et al. 2004;Thoby-Brisson et al. 2009;Grant & Moens, 2010;Rosenstein et al. 2010). A similar chemosensory functional unit may also occur in medullary raphé, as the raphé neurons are in close proximity to the marginal glial layer and leptomeninges (Ribas-Salgueiro et al. 2005) and are closely associated with arteries Severson et al. 2003).
Here we review the putative central chemosensitive areas of the brain that have been described and the extent to which they can be considered as primary central chemosensory nuclei. We will also assess at what stage of development they are most important and which reflex they contribute to.

Raphé nuclei
The medullary raphé are found within close proximity to the VLM surface (Ribas-Salgueiro et al. 2005), whereas the dorsal raphé are found at the surface of the midbrain (Severson et al. 2003). Raphé neurons are highly sensitive to hypercapnia in vitro Severson et al. 2003;Ribas-Salgueiro et al. 2005) and are connected to other chemosensory sites (Stocker et al. 1997;Ribas-Salgueiro et al. 2005;Mulkey et al. 2007b) and the preBötC (Li et al. 2006a;Pace et al. 2007;Hodges et al. 2008;Li & Nattie, 2008b;Ptak et al. 2009). That they remain chemosensitive isolated in cultures and in an antagonist cocktail supports their intrinsic chemosensitivity (Wang & Richerson, 1999). Nevertheless these are soft criteria.
Their involvement in intact animals has been questioned, as only a small number of raphé neurons exhibit chemosensitivity in vivo (Veasey et al. 1995;Ribas-Salgueiro et al. 2003;Mulkey et al. 2004;DePuy et al. 2011). Current evidence suggests that these nuclei only play a facilitatory role (Li et al. 2006a;Mulkey et al. 2007a;DePuy et al. 2011). Medullary raphé neurons are serotonergic and blockade of 5-HT receptors by methysergide (a 5-HT 1/2 receptor antagonist) in neonates reduces both respiratory rate and c-fos staining in respiratory related areas in vivo and in vitro (Bodineau et al. 2004). Mice with deletion of serotonin transporters have increased levels of 5-HT in their extracellular fluid (ECF), leading to desensitization of their receptors, and a decrease in respiratory responses to CO 2 (Penatti et al. 2006;Li & Nattie, 2008b). However, until a chemosensory transducer in medullary raphé neurons can be identified and tested, their status as primary chemosensitive neurons will remain open to doubt. Regardless of whether the contribution of the medullary raphé is mainly facilitatory or is indeed as an intrinsic chemosensitive site, inhibition of the medullary raphé significantly reduces the hypercapnic ventilatory response (Taylor et al. 2005) and the absence of serotonergic neurons, or silencing of them in genetically modified mice, severely blunts the chemosensitive response (Hodges et al. 2008;Ray et al. 2011).
Interestingly, loss of serotonergic neurons also causes the loss of the waking response to hypercapnia -a function thought to be mediated by the midbrain raphé (Buchanan & Richerson, 2010). The dorsal raphé of the midbrain are connected to cortical and hypothalamic regions associated with arousal (Vertes, 1991). However, recent evidence questioning the chemosensitivity of the medullary raphé nuclei raises the possibility that the midbrain raphé may also lack intrinsic chemosensitivity. Once again, targeted genetic manipulation of the key chemosensory transducing molecules within the dorsal raphé is required to settle the issue. Until this is achieved we consider the dorsal raphé to be a provisional, rather than an established, primary chemosensory nucleus.
The chemosensitivity of the medullary raphé varies with age: during the first two postnatal weeks there is an increase in neurons responsive to hypercapnia both in vivo and in culture (Wang & Richerson, 1999). After the second postnatal week the response remains unchanged (Veasey et al. 1995;Okada et al. 2002). It would be of interest to see whether their facilitatory role is also age dependent. The raphé complexes may be more crucial to chemosensitivity during adult life than in the neonate; or during the switch from neonatal to adult mechanisms, which leads to a vulnerable period that may explain why defects in this system often leads to sudden infant death syndrome (SIDS) (Duncan et al. 2010).

The tuberomammillary nucleus (TMN)
The TMN is located near the ventral surface of the posterior hypothalamus and is the sole source of histamine in CNS. Deletion of histidine decarboxylase (HDC) has been used to study the role of the TMN (Parmentier et al. 2002;Anaclet et al. 2009). Mice that lack the HDC gene spend more time in paradoxical sleep than their litter mates (Parmentier et al. 2002;Anaclet et al. 2009) as do mice injected with antagonists of this enzyme (Parmentier et al. 2002). Thus, this nucleus is associated with arousal and vigilance (Parmentier et al. 2002;Anaclet et al. 2009). TMN neurons are excited during hypercapnia (Dillon & Waldrop, 1992) and show increased c-fos immunoreactivity after exposure to elevated CO 2 (Berqin et al. 2000;Haxhiu et al. 2001;Johnson et al. 2005). There is some evidence that activation of neurons in the TMN may increase breathing during hypercapnia (Haxhiu et al. 2001); however, a causal link between activity in the TMN and CO 2 -dependent arousal has not yet been achieved and there are two other nuclei, the midbrain dorsal raphé (see above) and the lateral hypothalamus, that have also been proposed for this role.

The lateral hypothalamus
Orexinergic neurons of the lateral hypothalamus display increased c-fos immunoreactivity after exposure to hypercapnia (Berqin et al. 2000) and increase firing in response to alterations in pH e (Williams et al. 2007). As the lateral hypothalamus is the sole source of orexin, deletion of the prepro-orexin gene (Chemelli et al. 1999) is a genetic functional lesion of this area. Although this manipulation clearly demonstrates the importance of orexinergic signalling in maintaining wakefulness (Anaclet et al., 2009;Carter et al. 2009), the effect of this lesion on CO 2 -dependent arousal has not been examined. Furthermore the link between the putative chemosensory transducer (TASK-1) and CO 2 -dependent arousal has not been demonstrated. Thus, the status of this area as a primary pH chemosensor for CO 2 -dependent arousal is at best only putative.
Genetic deletion of the prepro-orexin gene blunts respiratory responses to hypercapnia (Deng et al. 2007) as do pharmacological antagonists of orexin applied to the medulla (Corcoran et al. 2010), although this is primarily during wakefulness (Dias et al. 2009(Dias et al. , 2010. Orexinergic terminals have been found in close apposition to RTN neurons, which can be excited by exogenous orexin (Lazarenko et al. 2011), though it is unknown whether orexin acts to facilitate or initiate J Physiol 589.23 chemosensitive responses in these neurons. In addition, orexin appears to facilitate GABAergic currents on cardiac vagal neurons, which act to induce the bradycardia associated with hypercapnia (Dergacheva et al. 2011).

The locus coeruleus (LC)
The LC is located at the ponto-medullary border directly underneath the fourth ventricle in the pons and supplies tonic chemosensory input (Infante et al. 2003). Stimulation of the LC causes dopamine and noradrenaline release in the cortex (Devoto et al. 2005), which has been linked to arousal and increased vigilance (Devoto et al. 2005). It also receives projections from the medial habenula (Li & Ku, 2002), and sends efferent projections to the amygdala, and has been linked to fear and anxiety (Gorman et al. 2000;Li & Ku, 2002;Devoto et al. 2005). In addition, the LC has been linked to the respiratory response to hypercapnia. Thus, it is poised to affect many of the physiological behaviours associated with hypercapnia.
Hypercapnia causes an elevation of c-fos immunoreactivity in LC neurons (Berqin et al. 2000) and focal acidification of the LC increases breathing (Coates et al. 1993). Neurons in the LC respond to hypercapnia (Stunden et al. 2001;Filosa et al. 2002) and to pH changes associated with acetazolamide (AZ) microinjections in the LC (Coates et al. 1993). Cultured neurons from the LC remain chemosensitive during pharmacological synaptic blockade and blockade of electrical coupling and in the presence of TTX (Filosa et al. 2002;Filosa & Putnam, 2003;Nichols et al. 2008). Nevertheless these are not definitive criteria and the status of the LC as a primary chemosensory area can only be considered provisional in the absence of a chemosensory transducer.
LC neurons are catecholaminergic (Yao et al. 2000). At the surface of the VLM, noradrenaline (NA) levels increase during hypercapnia (Rentero et al. 1997) and metabolic acidosis (Rentero et al. 1998). Ablation of noradrenergic neurons, by injections of dopamine β-hydroxylase tagged saporin, leads to a reduction in both the hypercapnic ventilatory response and time spent in the awake state  thereby strengthening the argument that NA, and thus the LC, plays a role in the waking response to hypercapnia in addition to its role in the hypercapnic ventilatory response. Further evidence utilizing 6-hydroxydopamine to lesion LC neurons supports a role for the LC in modulating the ventilatory response to hypercapnia (Biancardi et al. 2008).
The LC undergoes maturational changes and there is a dramatic reduction in the number of chemosensitive neurons and in the magnitude of chemosensory responses of single neurons around P10 (Nichols et al. 2008). The physiological role of this nucleus in chemosensory transduction may therefore change with development.

Amygdala
The amygdala is the fear control centre of the brain. Importantly neurons in the amygdala contain ASIC1a channels, which have been causally linked (through genetic deletion) to the CO 2 -evoked fear reflex, so by our definition these neurons are primary chemosensors (Ziemann et al. 2009). Direct stimulation of the amygdala by changes in local pH elicit the fear response and is therefore likely to be one of the areas responsible for fear associated with hypercapnia (Ziemann et al. 2009). As the deletion of ASIC1a was global, the possibility of other chemosensitive nuclei contributing to the fear reflex remains open.

The nucleus tractus solitarii (NTS)
The nucleus tractus solitarii is situated superficially in the dorsal medulla (Okada et al. 2001) and surrounds the fourth ventricle. The NTS is immunoreactive for c-fos after exposure to hypercapnia (Sato et al. 1992;Belegu et al. 1999;Berqin et al. 2000). Unilateral lesions of this region in humans causes reduced responses to hypercapnia (Morrel et al. 1999(Morrel et al. , 2001. Local perfusion of this region with AZ, causing pH changes, leads to increases in phrenic nerve discharge and increased blood pressure (Coates et al. 1993). The chemosensitivity of NTS neurons was only slightly reduced when hypercapnia was applied in the presence of a solution to block synaptic transmission (11.4 mM Mg 2+ /0.2 mM Ca 2+ ) or electrical coupling (carbenoxolone 100 μM) (Nichols et al. 2008). However, as discussed earlier, these are weak criteria for establishing whether the NTS is a site of primary chemoreception.
The properties of, and coupling between, NTS neurons undergo maturation in early development (Vincent & Tell, 1997;Nichols et al. 2008). Thus, caution must be exercised when extrapolating functional roles of the NTS in chemosensitive reflexes at different stages of development. Nevertheless the NTS is a relay for peripheral chemosensory inputs (Nattie & Li, 2002) and is connected to almost all chemosensory and respiratory related nucleus. It is therefore positioned to integrate a number of different inputs and control both respiration and blood pressure during hypercapnia.

Paraventricular nucleus (PVN)
The hypothalamic paraventricular nucleus is located next to the third ventricle (Cham et al. 2005) and is the master controller of blood pressure regulation. Stimulation of the PVN excites the pressor areas of the rostral ventrolateral medulla (RVLM) (Coote et al. 1998), an important nucleus in blood pressure regulation during hypercapnia (Moreira et al. 2006). The PVN also projects to the preBötC, phrenic nucleus, and the spinal cord (Yeh et al. 1997;Mack et al. 2007), and acts upon these nuclei through the release of oxytocin (Mack et al. 2007). Neurons projecting to the phrenic nucleus express c-fos during hypercapnia (Berqin et al. 2000;Kc et al. 2002). PVN stimulation causes increased breathing frequency and amplitude and blood pressure (Yeh et al. 1997). Thus the PVN is poised to influence blood pressure and respiration through intermediate nuclei and motor neurons. At present, this evidence is at best suggestive and further evidence is required if it is to be considered a primary or even secondary site of chemoreception.

The preBötzinger Complex (preBötC)
The preBötC is the site of inspiratory rhythm generation (Smith et al. 1991) and may be intrinsically chemosensitive. In neonatal in vitro preparations all neuronal subtypes of the preBötC were affected by alterations in CO 2 and pH and though no discharge could be seen, they retained their membrane potential alterations in the presence of TTX, cadmium or calcium free solutions to stop synaptic activity (Kawai et al. 1996(Kawai et al. , 2006Koizumi et al. 2010). In addition, focal acidosis of this region in vivo in adult animals also causes alterations in respiratory output (Solomon et al. 2000(Solomon et al. , 2003Krause et al. 2009). However, as discussed above, this is only a weak criterion for establishing whether the preBötC is a primary chemosensitive area.

The retrotrapezoid nucleus (RTN)
RTN neurons are located at 130-230 μm from the ventral surface of the medulla and in the marginal glial layer (MGL) (Mulkey et al. 2004), with extensive secondary dendrites located within the MGL (Mulkey et al. 2004). They show c-fos immunoreactivity in response to elevated CO 2 (Berqin et al. 2000) and are exquisitely sensitive to CO 2 above 4% in vivo (Hewitt et al. 2004;Mulkey et al. 2004) and to hypercapnic acidosis in vitro (Mulkey et al. 2004). They are also activated by injections of AZ directly into this nucleus (to cause acidification) (Hewitt et al. 2004). The RTN shows connections to the preBötC (Mulkey et al. 2004) and either it or the closely localized pFRG acts as an expiratory oscillator (Janczewski & Feldman, 2006a,b;Pagliardini et al. 2011), and thus it is uniquely poised to alter respiration.
While RTN neurons remain responsive to CO 2 in the presence of a cocktail of antagonists (Mulkey et al. 2004), this is not conclusive evidence for their being intrinsically chemosensitive. CO 2 -dependent release of ATP occurs in the vicinity of the RTN neurons. Unlike early reports that used Hepes buffered saline (Mulkey et al. 2006), recent evidence suggest that ATP plays at least a partial role in the chemosensitive responses of these neurons at ages P7-12 (Wenker et al., 2010) and, in the adult, responses of RTN neurons to acidosis appear almost entirely dependent on the prior release of ATP . Thus their intrinsic chemosensitivity appears to alter with age. Though the primacy of the RTN is in question, it remains important as an integrator or relay, as loss of this nucleus severely blunts the response to inspired CO 2 . Whether the RTN is a primary chemosensory nucleus will remain unclear until identification and genetic manipulation of the chemosensory transducer has been achieved.
Introduction of human mutations associated with congenital central hypoventilation syndrome (CCHS) into the mouse Phox2b gene also suggests that the RTN has an important role in chemosensitivity (Dubreuil et al. 2009) as neonatal mice hetero-or homozygous for these mutations completely lack CO 2 chemosensitivity (Dauger et al. 2003;Ramanantsoa et al. 2011). However, the importance of the RTN for adult chemosensitivity is less clear as adult Phox2b heterozygotes have an intact response by P10 (Dauger et al. 2003) and homozygotes regain around 40% of the sensitivity to CO 2 (Ramanantsoa et al. 2011). In addition, c-fos studies show that the RTN is significantly more important in neonatal life than in adults (Belegu et al. 1999). Thus while the RTN may have an important primary role in neonatal chemosensitivity, it may switch to be more of an integrator of other chemosensory inputs in the adult. It must be noted that Phox2b mutations have far reaching effects in humans, affecting the entire peripheral chemosensory arc and other chemosensitive nuclei, and so caution is required when relating this mutation to a specific chemosensory nucleus unless it is targetted to a specific neuronal subpopulation (e.g. Ramanantsoa et al. 2011).

The marginal glial layer (MGL)
The primary cell type at the ventrolateral medulla surface is astrocytes. Presumed glial cells within the MGL depolarise in response to hypercapnia (Fukuda et al. 1978;Ritucci et al. 2005) and perfusion of the rostral VLM with fluorocitrate, a glial toxin, reduced the hypercapnic ventilatory response in anaesthetised rats (Erlichman et al. 1998). Injections of fluorocitrate into the RTN of conscious adult rats increased ventilation and heightened the response to CO 2 , due to increased extracellular acidification (Holleran et al. 2001). However, 50% of the animals went into respiratory distress (Holleran et al. 2001) and for obvious reasons these experiments were discontinued. The results of this investigation could therefore have been biased towards animals where gliotransmitter dysfunction was not large enough to have an effect. Intraperitoneal injections of methoxysulphate, another glial toxin, into neonatal rats, which should affect a much larger proportion of glia compared to site directed injections, resulted in reduced levels of basal ventilation and their CO 2 response (Young et al. 2005).

J Physiol 589.23
Evidence is beginning to unveil the crucial role for glia in chemosensation Huckstepp et al. 2010b;Wenker et al. 2010). RTN astrocytes are intrinsically pH sensitive in neonates (Wenker et al. 2010) and in the adult . Sub-pial astrocytes in the adult are CO 2 sensitive via Cx26 (Huckstepp et al. 2010b) and pH sensitive (via an unknown molecule). Blockade of ATP release (including from astrocytes) during hypercapnia has been shown to reduce the ventilatory response (Gourine et al. 2005Huckstepp et al. 2010a,b;Wenker et al. 2010) and applications of exogenous ATP replicate the response to increased CO 2 (Gourine et al. 2005. Genetic evidence via the removal of the chemosensory transducers to demonstrate alterations in chemosensory reflexes in awake behaving animals is still required for full evaluation of these mechanisms. Nonetheless, neurons of all chemosensitive sites project dendrites to the MGL and many chemosensory nuclei express P2X and P2Y or are affected by ATP (Nieber et al. 1997;Sansum et al. 1998;Ralevic et al. 1999;Yao et al. 2000;Thomas et al. 2001;Whitlock et al. 2001;Fong et al. 2002;Lorier et al. 2004;Sergeeva et al. 2006;Cao & Song, 2007;Lorier et al. 2007). Thus, the MGL is poised to alter every aspect of chemosensory function and has recently been hypothesized to be a unifying factor at all chemosensory sites (Huckstepp et al. 2010b).

Leptomeninges
The leptomeninges, and in particular the pia mater, express Cx26 (Filippov et al. 2003;Huckstepp et al. 2010b) and are responsive to CO 2 (Huckstepp et al. 2010b), and thus they could contribute to chemosensory reflexes. The opening of Cx26 by CO 2 could allow sufficient release of ATP to diffuse into the parenchyma and activate receptors on superficial marginal glial cells and the dendrites of neurons. There is no evidence to support this possibility, but specific deletion of Cx26 within the leptomeninges may provide an answer as to whether this outer covering of the brain, rich in blood vessels, could be part of the chemosensory machinery. Table 2 summarizes and evaluates the contribution of the different brain areas to chemosensitive reflexes. It becomes readily apparent that each chemosensory reflex is subserved by at least two putative chemosensitive areas. What is the reason for such complexity and apparent redundancy? Developing the thinking of Nattie and colleagues (Nattie, 1999(Nattie, , 2000, we believe there are three possible explanations, as outlined below.

The complexity of chemosensitivity
Complexity is only apparent owing to incomplete knowledge. Some of the putative and provisional chemosensitive areas may fail to make the grade when stricter definitions, developed in this review for chemosensory transducers and intrinsic chemosensory cells, are rigorously applied. It may therefore be that for some reflexes only a single primary chemosensitive nucleus is involved. Nevertheless, this seems unlikely and especially so for breathing where six different primary sensor nuclei have been proposed and three of these have either likely or provisional status.
Redundancy hypothesis. Nattie (2000) has proposed that the robust nature of the overall response to CO 2 depends on additive or greater interactions of inputs from multiple locations. The sensitivity of neuronal processing of hypercapnia may be enhanced if all of the neurons in the pathway were chemosensitive (Su et al. 2007). In addition, a multiplicity of chemosensory sites would provide greater overall stability (Li & Nattie, 2008a) by allowing for corrections of regional imbalances of blood flow and metabolism.
Multiple salient signals hypothesis -implications for parallel processing. A more attractive possibility is that the apparent redundancy enables more abstraction of information about physiological state. As the visual system abstracts and computes multiple aspects of the visual stimulus in parallel, the chemosensory system may do the same through parallel processing of multiple salient signals (Fig. 1). In addition to processing pH i , pH e , [HCO 3 − ], and P CO 2 separately, different chemosensors exhibit a range of response characteristics in both the temporal and concentration domains. Some chemosensors give sustained responses (Huckstepp et al. 2010b), whereas others respond in a transient fashion . In breathing, for example, Cx26 in glia can report both increases and decreases of P CO 2 (Huckstepp et al. 2010a), whereas the RTN is only sensitive to CO 2 above 4% (Mulkey et al. 2004). Thus, different chemosensory transducers and nuclei have different 'set points' allowing them to respond to different magnitudes and absolute ranges of change in the relevant salient stimulus. This in turn may mean that tonic baseline input is required from at least some chemoreceptors to allow full expression of the systemic response (Nattie, 2000). Hypoxia can cause intracellular acidification of a range of neurons in the NTS and ventrolateral medulla. In the absence of peripheral chemosensors, this intracellular acidification is not sufficient to cause an enhancement of breathing (Putnam et al. 2004). The reasons for this are not fully understood, but this observation is consistent with the concept that a range of salient chemosensory signals can be processed in parallel to allow accurate computation of the physiological state.
We are a long way from having sufficient knowledge to understand at a systems level how the multiplicity of different transducing molecules in different areas would allow the brain to compute the exact condition of the entire system. This is particularly relevant as metabolic and respiratory acidosis reconfigures the respiratory network in neonates in different ways (Okada et al. 2007). During metabolic acidosis, areas sensitive to changes in either pH i or pH e would be stimulated, whereas those areas sensitive only to HCO 3 − and CO 2 fluctuations would not. By contrast during respiratory acidosis all areas would be excited. During exercise, the reduction of pH arising from increases in anaerobic respiration are likely to be accompanied by a reduction in [HCO 3 − ] as it buffers lactate (Beaver et al. 1986). Thus only pH-sensitive sites would stimulate respiration whereas HCO 3 − -sensitive sites might do the opposite. This general concept that chemosensory pathways are organized to process different types of salient information in parallel pathways has the capacity to allow greater adaptation of the system. By providing a system that monitors multiple facets of the same chemical reaction, the system may still adapt to changes when one of the signalling molecules of this reflex is already saturated.

Concluding remarks
In light of recent evidence we believe that chemosensation is at an exciting stage of development. To progress further we believe that the new criteria proposed here for establishing a chemosensory transducer and a primary chemosensory nucleus need to be rigorously applied across the whole field. We accept that this is potentially very difficult, especially in those cases where the transducer has not yet been identified. This rigorous approach will require targeted and inducible genetic deletions of the relevant molecules. Implicit in this approach is a re-evaluation of the salient chemosensory signals with the possibility that hitherto marginalized molecules (CO 2 and HCO 3 − ) may ultimately play a larger role than suspected and that some putative chemosensors and chemosensory areas will be eliminated from the field.
In addition, careful quantitative characterization of the properties of the transducer molecules, along with the dynamic and quantitative outputs from the different nuclei involved, is needed. This level of detailed information is necessary to allow us to understand how different aspects of the chemosensory stimulus are detected and how the different aspects of information are processed. This will lay the groundwork for a future systems level understanding of chemosensation.