Detection of amino acid and peptide transmitters in physiologically identified brainstem cardiorespiratory neurons

Most of the CNS neurons that regulate circulation and respiration reside in regions of the brain characterized by extreme cellular heterogeneity (nucleus of the solitary tract, reticular formation, parabrachial nuclei, periaqueductal gray matter, hypothalamus, etc.). The chemical neuroanatomy of these regions is correspondingly complex and teasing out specific circuits in their midst remains a problem that is usually very difficult if not impossible to solve by conventional tract-tracing methods, Fos methodology or electrophysiology in slices. In addition, identifying the type of amino acid or peptide transmitter used by electrophysiologically recorded neurons has been until recently an especially difficult task either for lack of a specific marker or because such markers (many peptides for example) are exported to synaptic terminals and thus undetectable in neuronal cell bodies. In this review, we describe a general purpose method that solves many of these problems. The approach combines juxtacellular labeling in vivo with the histological identification of mRNAs that provide definitive neurochemical phenotypic identification (e.g. vesicular glutamate transporter 1 or 2, glutamic acid decarboxylase). The results obtained with this method are discussed in the general context of amino acid transmission in brainstem cardiorespiratory pathways. The presence of markers of amino acid transmission in specific aminergic pre-sympathetic neurons is especially emphasized as is the extensive co-localization of markers of GABAergic and glycinergic transmission in the brainstem reticular formation.

Introduction

Most of the CNS neurons that regulate circulation and respiration reside in regions of the brain that serve quite heterogeneous functions (nucleus of solitary tract, reticular formation, parabrachial nuclei, periaqueductal gray matter, hypothalamus, etc.). The chemical neuroanatomy of these regions is correspondingly complex and teasing out specific circuits in their midst presents a major challenge.

In vivo, reticular formation neurons can be classified into numerous functional categories based on anatomical location, discharge pattern, differential response to activation of selected inputs and axonal projections. Except in very specific cases, the vast majority of these cells can no longer be identified in vitro because the connections that produce their characteristic discharge pattern in vivo are severed and intrinsic properties rarely provide unambiguous characterization in a very complex mix of neurons. Furthermore, very practical considerations such as tissue survival or cell visibility generally restrict the study of brainstem neurons in vitro to neonates.

This review describes an approach to identify the transmitters of neurons that contribute to brainstem cardiorespiratory control circuits as judged by their electrophysiological properties in vivo. In theory, most of these methods should be applicable to the identification of neurons recorded in the arterially perfused working heart brain preparation (WHBP) (Boscan et al., 2002).

Identifying the biochemical phenotype of neurons recorded in vivo is by no means a novel idea (Pilowsky et al., 1991) but the traditional method was laborious because of the need to perform intracellular recording to label cells. Furthermore, until very recently, the neurochemical information derived from cells labeled following recording was limited by the shortage of reliable histological tools to identify ionotropic neurotransmitters (e.g. glutamate, GABA and glycine) or peptides. The present review emphasizes several recent developments that have greatly increased the breadth of information that can be derived from labeling neurons characterized electrophysiologically in vivo. The first development is the introduction of juxtacellular labeling, a relatively high yield method of labeling extracellularly recorded neurons in vivo (Pinault, 1996). The second is the discovery of new biological markers whose presence in neurons seems necessary and sufficient to identify which amino acids and/or peptides they use as transmitters. Henceforth, we will call this type of marker a “diagnostic marker”.

The main advantage of the approach emphasized in the present review is that it can be used to identify the neurochemical phenotype of any physiologically identified neuron, no matter how complex or heterogeneous the surrounding neuropil. Its primary disadvantage is the small number of markers that can be tested per labeled cell. The information derived from labeling single physiologically identified neurons has been generally congruent with the conclusions derived from other methods such as Fos expression or purely anatomical methods (retrograde and anterograde tracings, pseudorabies virus, etc.; Loewy, 1998, Dampney et al., 1995). However, the single cell labeling of physiologically recorded neurons provides results that are more definitive and it has a more general applicability than most other methods.

Methods for labeling neurons intracellularly with sharp electrodes are well described and need no elaboration (for some recent examples pertinent to brainstem cardiorespiratory circuitry, see Lipski et al., 1995, Otake et al., 1987, Pilowsky et al., 1990). The juxtacellular labeling method introduced by Pinault (1996) will be reviewed because it is relatively new, highly reliable and very productive for identifying the neurochemical phenotype of neurons recorded in vivo. The method fills cell bodies and dendrites with biotinamide without spillage of the marker into the extracellular space (Schreihofer and Guyenet, 1997, Schreihofer et al., 1999). Juxtacellular labeling probably relies on single cell electroporesis, a technique used by molecular biologists to introduce plasmids and DNA oligonucleotides into single cells (Haas et al., 2001, Rathenberg et al., 2003). Single cell electroporesis consists of delivering pulses of current through a glass pipette of conventional size and resistance (tip size around 1 μm, 20–40 MΩ) with intensity sufficient to overcome the dielectric properties of the cell membrane. The current is assumed to create transient pores through which the chemical present in the pipette gains access to the interior of the cell (Teruel and Meyer, 1997). Negative current is typically used for single cell electroporesis in vitro because the point of most experiments is to expel negatively charged oligonucleotides or plasmids (Haas et al., 2001, Rathenberg et al., 2003). The in vivo juxtacellular method relies on the use of positive current pulses (400 ms, 1.5–9 nA). These pulses produce bursts of action potentials that should be continuously monitored and serve as a measure of the magnitude of the membrane disruption (Schreihofer and Guyenet, 1997, Schreihofer et al., 1999)(Fig. 1D). Such monitoring is the key to success of the procedure because it serves to insure that the recorded cell is the one that is labeled and it allows adjusting the current intensity to the lowest level needed to disrupt the membrane. The requirement for positive current stimulation means that only positively charged molecules can be introduced into neurons by the in vivo juxtacellular method. So far only biotinamide or biocytin have been used for this purpose but any other positively charged molecules are likely to work.

The advantages of juxtacellular labeling over the traditional intracellular method are its simplicity, reliability and efficiency. Extracellular single-unit recording allows one to study neurons for extensive periods of time prior to labeling, which is often essential to obtain an accurate characterization of neuronal types in vivo. It also produces a much higher cell yield than sharp electrode recording and probably minimizes the sampling bias associated with somatic size. Labeling efficiency is high. In our experience, over 50% of attempts at labeling neurons succeed in introducing enough biotinamide to locate the soma and main dendrites after histological processing. Cells with apparently completely filled soma and dendrites are obtained less frequently however (around 20% of labeling attempts). The juxtacellular method achieves neuronal labeling without dialysis or mechanical disturbance of the cytoplasm. When used in conventional anesthetized preparations, the method provides high quality subsequent light microscopic and ultrastructural analysis of the labeled cells and their immediate environment (Aicher et al., 2001, Llewellyn-Smith et al., 2001). The juxtacellular method also leaves the mRNA content undisturbed as judged by the fact that labeled cells exhibit as high a hybridization signal as non-recorded cells of the same type (Fig. 2). To our knowledge, mRNA detection by in situ hybridization (ISH) has not been performed in cells labeled in vivo with sharp electrodes but the procedure is likely to be successful in minimally damaged neurons. A priori, whole cell recording would seem detrimental to subsequent mRNA detection since this mode of recording dialyses the cytoplasm but evidence might prove otherwise.

The juxtacellular method has several disadvantages. First, the method relies on extracellular recording; therefore silent cells typically escape detection. Second, when biotinamide is used, the method only produces partial filling of the main axon and, in our hands at least, rarely fills small axonal collaterals and synaptic boutons. In this respect, the juxtacellular method is clearly inferior to intracellular labeling with sharp or patch electrodes (Deuchars et al., 2000, Pilowsky et al., 1990). Third, on occasion, more than one cell is labeled. In the reticular formation where cell density is low, this inconvenience is negligible since multiple cell labeling occurs at most 5% of the time. In areas of the brain where neuronal density is much higher, the incidence of multiple labeling may have to be reassessed. Finally, juxtacellular labeling in vivo seems to require positive current pulses. Unless this requirement is overcome, it may exclude the possibility of injecting oligonucleotides or plasmids into identified neurons.

The steps necessary to successfully find and label specific populations of autonomic neurons in the reticular formation are illustrated in Fig. 1. The first requirement (Fig. 1A) is to limit the search for units within the limited region of the reticular formation where the population of interest is most concentrated. In the rat, this process usually requires focusing on a brain region that is typically less than 600 μm long, 300 μm wide and 200 μm deep. This degree of precision is required due to the generally narrow distribution of neuronal populations within the reticular formation (Fig. 3). For example, finding pre-sympathetic neurons in significant numbers requires sampling within a 4–500-μm long region located in close proximity to the caudal end of the facial motor nucleus (Fig. 1, Fig. 3). The second requirement is to identify a set of physiological criteria that are characteristic of the cell type of interest. In the particular case of the pre-sympathetic neurons, a pulse-synchronous discharge, extreme sensitivity to changes in blood pressure and the lack of effect of arterial CO₂ under hyperoxic conditions are indicative, providing that the recordings are done in the proper anatomical region (Fig. 1B). Identifying the cells by their axonal projections is also useful. Antidromic activation from the thoracic cord (Fig. 1C) distinguishes pre-sympathetic cells from reticular formation neurons with virtually identical discharge patterns but that innervate the hypothalamus, not the spinal cord. The last step is juxtacellular labeling (Fig. 1D), which must result in a clean entrainment of the cell for at least 30 s in order to label the cell body and its major dendrites.

As in other brain areas, virtually all spontaneous, miniature or evoked PSCs recordable in brainstem slices yield to blockade of neuronal glutamatergic, GABAergic or glycinergic receptors (Lin et al., 1998, Hayar and Guyenet, 1998, Hayar and Guyenet, 1999, Doyle and Andresen, 2001). The most probable reason is that the rest of the ionotropic receptors (e.g. P2X and nicotinic) are largely presynaptic and modulate the exocytosis of the big three (glutamate, GABA or glycine) (McGehee et al., 1995, Shigetomi and Kato, 2004). At the EM level, virtually all terminals in brainstem or spinal cord contain one of these three amino acids (Llewellyn-Smith et al., 1995, Llewellyn-Smith et al., 2001). Finally, in vivo, brainstem neuronal activity evoked by brain stimulation or peripheral afferent stimulation is usually severely attenuated if not blocked by administration of antagonists of ionotropic glutamate, GABA or glycine receptors (for review, see Guyenet and Stornetta, 2004, Nakamura et al., 2004). In the brainstem and spinal cord, GABA and glycine can be released either separately or together by brainstem neurons (O'Brien and Berger, 1999, Jonas et al., 1998). Co-release of GABA and glutamate by brainstem neurons is not documented yet. Thus, as a first approximation, the brainstem may largely consist of excitatory glutamatergic neurons and inhibitory GABAergic and/or glycinergic neurons. Under these conditions, it would seem that the most basic information needed to understand the role of a given neuronal population in any circuit is whether the neuron is excitatory (releases glutamate) or is inhibitory (releases GABA and/or glycine).

It is now finally possible to determine without ambiguity whether a given neuron is glutamatergic, GABAergic and/or glycinergic because diagnostic markers are available. GABAergic neurons are identifiable by the presence of glutamic acid decarboxylase (GAD) (Chan and Sawchenko, 1998, Weston et al., 2003, Schreihofer and Guyenet, 2003) of which two isoforms exist, GAD₆₅ and GAD₆₇ (Erlander et al., 1991, Esclapez et al., 1993). In the brainstem, GAD₆₇ is expressed at higher levels than GAD₆₅ but both isoforms seem to coexist in most GABAergic neurons (Stornetta and Guyenet, 1999). GABAergic neurons can also be identified by the presence of GABA vesicular transporter (VGAT) but this marker does not provide definitive identification since the transporter is also expressed by glycinergic neurons that do not make GABA (Chaudhry et al., 1998). In the brainstem as elsewhere, most glutamatergic neurons can be identified by the presence of one of two glutamate vesicular transporters (VGLUT1 and VGLUT2) whose expression is largely but not exclusively complementary (Takamori et al., 2000, Takamori et al., 2001). In the brainstem, VGLUT1 or 2 are absent from neurons that express GAD and/or GLYT2 (GABAergic or glycinergic neurons) (Stornetta et al., 2002a, Stornetta et al., 2003a, Weston et al., 2003). Furthermore VGLUT1 or 2 are exclusively present in terminals that make asymmetric synapses, widely thought to be excitatory (Takamori et al., 2000, Takamori et al., 2001, Fremeau et al., 2001). A third vesicular transporter also exists (VGLUT3) but its role in the exocytotic release of glutamate has not yet been proven. Consequently, its usefulness to identify neurons that use glutamate as their classic ionotropic transmitter remains uncertain (Gras et al., 2002, Schafer et al., 2002, Fremeau et al., 2004). In the brainstem and spinal cord, VGLUT1 is almost entirely expressed by neurons that contribute mossy fiber inputs to the cerebellum (Hisano et al., 2002) and unpublished results of Stornetta and Guyenet). The remainder of the lower brainstem glutamatergic neurons, especially neurons located in the reticular core and the nucleus of the solitary tract, seem to express VGLUT2 (Guyenet et al., 2002, Stornetta et al., 2002a, Stornetta et al., 2002b, Stornetta et al., 2003a, Stornetta et al., 2003b, Weston et al., 2003). There are other markers of glutamatergic neurons (e.g. plasmalemmal glutamate transporters, (Amara and Fontana, 2002) but their expression is not limited to glutamatergic neurons. Other markers have not been demonstrated to identify glutamatergic cell bodies consistently and reliably (glutamate itself; phosphate-activated glutaminase, PAG). The presence of PAG in identified brainstem respiratory motor neurons and Bötzinger neurons is a case in point (Pilowsky et al., 1997). Motor neurons are cholinergic, Bötzinger neurons contain GlyT2 and clearly release an inhibitory transmitter that triggers a chloride current postsynaptically (Schreihofer et al., 1999, Ezure et al., 2003). In our hands, brainstem somatic motor neurons in rats express neither VGLUT1 nor VGLUT2 mRNA consistent with the lack of immunoreactive VGLUT1 protein in cranial somatic motor neurons in the mouse (Kraus et al., 2004). However, the matter may not be fully settled since others have detected low levels of both VGLUT1 and VGLUT2 mRNA in lumbar motor neurons using digoxigenin-labeled cRNAs (Landry et al., 2004). Conceivably, spinal cord motor neurons could have biochemical characteristics distinct from those located in the brainstem but we lean towards the possibility that Landry's observation could be an ISH artifact to which motor neurons are prone when the procedure is performed with digoxigenin-labeled cRNAs. In addition, we also cannot detect VGLUT1 mRNA in the compact division of nucleus ambiguus, which contains esophageal motor neurons in rats though the cognate protein has been detected by immunohistochemistry in the end plates of esophageal motor neurons in the mouse (Kraus et al., 2004). As mentioned above, glycinergic neurons can be identified by the presence of glycine transporter 2 (GlyT2), a neuron specific plasmalemmal transporter that is somehow essential to provide neurons with glycine destined for synaptic vesicle loading (Jursky et al., 1994, Jursky and Nelson, 1995, Zafra et al., 1997, Gomeza et al., 2003). GlyT2 mRNA is frequently co-localized with GAD₆₇ mRNA in the brainstem (Fig. 4; from unpublished work by Stornetta and Guyenet). GlyT2 mRNA is rarely present in the nucleus of the solitary tract (Fig. 4B1–B2) and absent from the area postrema (not illustrated), both regions that contain large numbers of GAD67-expressing neurons. GlyT2 mRNA appears to be excluded from cells that contain VGLUT1 or 2 (unpublished results from Stornetta and Guyenet).

A practical experimental problem arises from the fact that each of the diagnostic markers mentioned above (GAD₆₅ or GAD₆₇, VGLUT1 or 2, GlyT2) is a molecule designed for export to synaptic terminals. These markers are highly concentrated in terminals but their level in the soma and dendrites of neurons is generally too low to be of practical use for routine immunohistochemistry. VGLUT3 is an exception to this rule since its presence can be detected in somata by immunohistochemistry (Nakamura et al., 2004). The low and usually undetectable level of these markers in neuronal somata leaves only two options to the investigator who wishes to identify whether a recorded neuron is glutamatergic, GABAergic or glycinergic. The first option is to label the recorded neuron to such an extent that some of the marker will migrate to its terminals and will be detectable at this level. If this can be achieved, then conventional immunohistochemistry can be used to test whether the labeled terminals contain the diagnostic protein of interest (GAD₆₇, GlyT2 or VGLUT2 immunoreactivity) since excellent antibodies are commercially available for these proteins. This approach is very difficult however because the migration of a marker from the cell body to its terminals may take many hours. Also the process probably requires injecting large amounts of the marker into the soma, a requirement that essentially precludes the use of the juxtacellular labeling method. The alternative approach that we have chosen is to identify the cognate mRNAs rather than the proteins. The obvious advantage of this approach is that mRNAs are confined to the cell bodies therefore even the lightest juxtacellular labeling of a recorded neuron (~20–30 s of entrainment) is sufficient to establish whether it contains the mRNA of interest (VGLUT2, GlyT2 or GAD67 mRNA). The productive use of this approach requires a robust and reliable ISH method because the search for neurons labeled in vivo requires processing large numbers of sections in a systematic manner. Non-radioactive cRNA probes are therefore ideal for this purpose. In our experience, the level of expression of the mRNAs that encode GAD₆₇, GlyT2 or VGLUT (1 or 2) is sufficiently high for efficient use of probes labeled with digoxigenin-UTP (Schreihofer et al., 1999, Stornetta et al., 1999, Stornetta et al., 2002a, Guyenet et al., 2002). This ISH method is highly reliable, it is performed using conventional free-floating sections from paraformaldehyde-fixed brains and it labels neuronal somata with high intensity (Figs. 2, Fig. 4, Fig. 5). Probes labeled with FITC-UTP (e.g. Fig. 4A1, B2, C2 and D2) also work well but require a more complex histological procedure based on tyramide amplification (Stornetta et al., 2002a). Radioactive probes are typically more sensitive but their use is cumbersome, which may account for the fact that, to our knowledge, such probes have not been used to identify the neurochemical phenotype of neurons recorded in vivo.

Intra- or juxtacellular cell labeling with biotinamide can be combined with the immunohistochemical detection of one or more proteins including membrane receptors (Schreihofer and Guyenet, 1997, Sartor and Verberne, 2002, Jones et al., 2002, Gourine et al., 2003). When antibodies are not available, ISH is again useful to identify the probable presence of the cognate proteins within recorded neurons. The mRNAs that encode the protein precursors to classic neuropeptides such as enkephalin, neuropeptide Y and somatostatin are typically expressed by brainstem neurons at levels that are high enough to allow detection by non-radioactive ISH (Stornetta et al., 1999, Stornetta et al., 2001, Stornetta et al., 2003a). Methods based on mRNA identification rather than immunohistochemistry for neuropeptides or their precursor proteins have three essential advantages. First, neuropeptides, like the markers of excitatory or inhibitory transmission discussed above, are exported to terminals and thus frequently undetectable in cell bodies by immunohistochemistry without blocking axonal transport with colchicine. Second, the manufacture of cRNA probes is now a routine and rapid laboratory procedure that only requires knowledge of the gene sequence, information widely available in silico. Finally, whereas the selectivity of an antibody for immunohistochemistry is often difficult to prove unless a viable knock-out animal is available, cRNA probes can be designed to hybridize with highly specific mRNA sequences and thus can usually be designed to distinguish clearly between mRNAs that encode closely related proteins. One drawback of using mRNAs for phenotypic identification is that, in some rare cases, the cognate protein may not be translated. A second limitation of the approach is that only a few mRNAs per cell can be labeled with currently available techniques. Given that the juxtacellular labeling uses the biotin–avidin method, a biotin-labeled riboprobe would not be compatible. The only other available options for non-radioactive labeled nucleotides are FITC, rhodamine and digoxigenin, limiting detection in theory to three mRNAs per biotin-labeled neuron. The final limitation of the juxtacellular/ISH combination method is that the detection of low abundance mRNAs in recorded cells remains a challenge. Most G-protein-coupled receptors fit this description and these receptors are typically not detectable with existing non-radioactive ISH methods.

Recent evidence based in large part on mRNA detection by ISH suggests that a very large proportion of the brainstem neurons that regulate circulation and probably also respiration are indeed either glutamatergic (VGLUT2 positive), GABAergic (GAD₆₇ positive) or glycinergic (GlyT2-positive)(Schreihofer et al., 1999, Stornetta et al., 2002b, Stornetta et al., 2003b, Schreihofer and Guyenet, 2003, Ezure et al., 2003, Rosin et al., 2003). The list of likely glutamatergic cells includes the C1 “adrenergic” neurons with pre-sympathetic function (for recent reviews, see (Dampney et al., 2003, Guyenet and Stornetta, 2004), which, despite their catecholaminergic phenotype, contain VGLUT2 mRNA in their cell bodies and VGLUT2 immunoreactivity in their terminals (Stornetta et al., 2002b) (Fig. 5). This list also includes the orexinergic neurons (Rosin et al., 2003) and at least some of the bulbospinal “pre-autonomic” neurons of the hypothalamic paraventricular nucleus (Stornetta and Guyenet, unpublished). In our hands, neither motor neurons, nor the noradrenergic A5 neurons of the pons contain any of the classic mRNA markers of glutamatergic, GABAergic or glycinergic transmission (VGLUT1-3, GAD₆₇ and GlyT2).

Lower brainstem serotonergic neurons lack VGLUT1, VGLUT2 and GlyT2 but subsets of these neurons contain VGLUT3 and/or contain GAD₆₇ (Stornetta and Guyenet, 1999, Weston et al., 2004, Nakamura et al., 2004). The GAD₆₇ mRNA-containing serotonergic cells of the medullary raphe are bulbospinal, presumably pre-sympathetic and reside mostly within the marginal layer of raphe pallidus and the parapyramidal region (Fig. 6) (Stornetta and Guyenet, 1999, Weston et al., 2004). Serotonergic neurons located in similar regions of the raphe also express VGLUT3 suggesting that a subset of serotonergic neurons with pre-sympathetic function may use glutamate as a transmitter (Oliveira et al., 2003, Herzog et al., 2004, Nakamura et al., 2004). These observations support the conclusions of much older studies which had suggested the presence of partly serotonergic sympathoexcitatory neurons in the rostral midline raphe (Minson et al., 1987, Pilowsky et al., 1995). VGLUT3-expressing neurons that do not contain detectable levels of serotoninergic markers are also present in raphe pallidus in a region that regulates sympathetic tone to brown adipose tissue (BAT) and skin blood flow (Blessing and Nalivaiko, 2001, Nakamura et al., 2004, Morrison, 2004). One possible interpretation of this data is that the medullary raphe, especially raphe pallidus, contains a population of glutamatergic (VGLUT3-containing) pre-sympathetic neurons that express the serotonergic phenotype to varying degree. The parallel with the blood pressure-regulating pre-sympathetic neurons of RVLM is striking if one accepts the view that the latter cells are primarily glutamatergic neurons that also express a monoaminergic phenotype (adrenergic in this case) to a varying degree (Stornetta et al., 2002a, Stornetta et al., 2002b, Guyenet and Stornetta, 2004). Finally, the medullary raphe also contains serotonergic neurons that, like the A5 cells of the pons, express none of the three VGLUTs nor GAD₆₇ nor GlyT2 (Stornetta and Guyenet, unpublished results). These particular aminergic neurons may therefore lack an amino acid transmitter, contrary to the C1 neurons of the RVLM and the VGLUT3-expressing pre-sympathetic neurons of the medullary raphe.

Judging from electrophysiological results, glutamate is certainly the primary ionotropic transmitter of vagal afferents and, most probably, that of arterial baroreceptor afferents although the presence of a vesicular transporter in these cells has not been described (Andresen et al., 2001, Andresen et al., 2004). Based on Fos expression studies, some of the putative second-order neurons in the baroreflex pathway that project to the caudal ventrolateral medulla (CVLM) express VGLUT2 and are thus also glutamatergic (Weston et al., 2003). This conclusion verifies the interpretations of prior pharmacological experiments (reviewed by Sved and Gordon, 1994).

The main transmitter of several types of inhibitory neurons with a defined cardiovascular or respiratory role has been definitively identified in the brainstem reticular formation by Fos and/or by single cell labeling in vivo (Chan and Sawchenko, 1998, Schreihofer et al., 1999, Weston et al., 2003, Schreihofer and Guyenet, 2003, Ezure et al., 2003). Each neuronal type identified definitively so far contained either GAD₆₇ or GlyT2 mRNA but not both. In particular, the ventrolateral medullary interneurons that mediate the sympathetic baroreflex are GABAergic but do not express GlyT2 mRNA (Fig. 6) (Chan and Sawchenko, 1998, Weston et al., 2003) and the reverse applies to the expiratory augmenting (Bötzinger) and decrementing expiratory neurons of the ventral respiratory column (Schreihofer et al., 1999, Ezure et al., 2003). However, the number of cell types analyzed so far is very small and, in fact, there is a considerable degree of overlap between GAD₆₇ and GlyT2 mRNAs within the reticular formation of the medulla oblongata (Stornetta and Guyenet, unpublished; Fig. 4). At least one type of premotor neuron (premotor to hypoglossal motor neurons) has been physiologically identified as having a dual GABA/glycine phenotype based on the pharmacology of the mPSCs that it generates (O'Brien and Berger, 1999). This result agrees with the presence of many dual labeled GAD₆₇/GlyT2 neurons (GABA/glycine phenotype) in the immediate proximity of hypoglossal motor neurons (Fig. 4C1–C2). Whether these premotor neurons process respiration-related information is unknown. In any event, GABA/glycine neurons almost certainly contribute to autonomic regulation since neurons in the rostral ventromedial medulla have this mixed phenotype (Fig. 4, Fig. 6) and many of them are rapidly infected after introduction of pseudorabies virus into the adrenal medulla (unpublished data of Stornetta and Guyenet).

The recent discovery of markers that are diagnostic for specific types of neurotransmission (notably VGLUT1, VGLUT2, GlyT2 and GAD67) and the introduction of more efficient ways to label neurons recorded in vivo have greatly facilitated the biochemical identification of electrophysiologically recorded neurons. These methods provide a powerful general purpose approach to study neuronal connectivity in brain regions characterized by extreme neuronal heterogeneity. They are thus especially suited to the study of brainstem autonomic circuits that typically consist of very small populations of functionally related neurons (few hundred per brain) buried within a very heterogeneous neuronal mix.