The ins and outs of the caudal nucleus of the solitary tract: An overview of cellular populations and anatomical connections

Abstract The body and brain are in constant two‐way communication. Driving this communication is a region in the lower brainstem: the dorsal vagal complex. Within the dorsal vagal complex, the caudal nucleus of the solitary tract (cNTS) is a major first stop for incoming information from the body to the brain carried by the vagus nerve. The anatomy of this region makes it ideally positioned to respond to signals of change in both emotional and bodily states. In turn, the cNTS controls the activity of regions throughout the brain that are involved in the control of both behaviour and physiology. This review is intended to help anyone with an interest in the cNTS. First, I provide an overview of the architecture of the cNTS and outline the wide range of neurotransmitters expressed in subsets of neurons in the cNTS. Next, in detail, I discuss the known inputs and outputs of the cNTS and briefly highlight what is known regarding the neurochemical makeup and function of those connections. Then, I discuss one group of cNTS neurons: glucagon‐like peptide‐1 (GLP‐1)‐expressing neurons. GLP‐1 neurons serve as a good example of a group of cNTS neurons, which receive input from varied sources and have the ability to modulate both behaviour and physiology. Finally, I consider what we might learn about other cNTS neurons from our study of GLP‐1 neurons and why it is important to remember that the manipulation of molecularly defined subsets of cNTS neurons is likely to affect physiology and behaviours beyond those monitored in individual experiments.

from autonomic outflow to motivated behaviour. 1,4 Unfortunately, much of the anatomical organization uncovered in the last century is often forgotten in contemporary neuroscience reports, perhaps as a result of the lack of easily accessible, recent overviews. This review is intended to provide exactly that: an overview of the efferent and afferent connections of the NTS, as well as the current state of knowledge regarding the neuropeptidergic cell types residing within the cNTS.
First, I briefly describe the architecture of the cNTS and its resident cell types. Then, I review the anatomical configuration of the inputs and outputs of the cNTS. Finally, as perhaps the best studied peptidergic population of cNTS neurons, glucagon-like peptide-1 (GLP-1)-expressing neurons will be used as an example of secondorder neurons, which receive substantial vagal sensory input, as well as input from both forebrain and hindbrain regions. This review will not cover the function of the cNTS in detail, and readers are referred to excellent available reviews on the subject. 1,[3][4][5][6] An important note on species is warranted: this review covers only preclinical data, most of which was collected in rodents. Indeed, most of the anatomical data that will be discussed were collected in rat. A small number of tracing and cytoarchitecture studies have been conducted in rabbit, 7 hamster 8 and cat, 9 and, more recently, presumably as a result of the increased popularity of the mouse as an experimental model, reports of the anatomy of the mouse NTS have been added to the literature. [10][11][12] For clarity and to emphasize the idea that not all species can be reasonably assumed to be anatomically and functionally identical, I will indicate the species that the data were collected from.

| NTS ARCHITECTURE
In rodents, the NTS is traditionally, if somewhat arbitrarily, divided into two parts, sometimes three, 13 based on their relative rostrocaudal location 10 Figure 1C). In coronal sections, the NTS is oval in appearance at more rostral levels ( Figure 1D) but takes a triangular shape at the level of the AP, with the DMV situated at its ventral border. At its rostral extreme, the NTS is at its most lateral and gradually moves more medial until it finally surrounds the midline at the very caudal end of the nucleus ( Figure 1D).
In rodents, the NTS can be subdivided into a number of subnuclei based on the location, size, shape, density, and staining intensity of neuronal cell bodies following Nissl, silver or Golgi staining. Following this approach, Ganchrow et al. 10 thoroughly mapped the cytoarchitecture of the mouse NTS and compared this to previous reports in hamster and rat, 10 which were largely similar. Because Ganchrow et al. 10 provides such an excellent and exhaustive description of the subnuclear organization of the rodent NTS, this particular aspect of NTS anatomy will not be described in detail here. However, for convenience Figure 1E presents an overview of the subnuclear division of the cNTS.

| CELL TYPES OF THE CNTS
The cNTS is cellularly heterogeneous with a multitude of neuropeptides ( Figure 2), small-molecule neurotransmitters and receptors expressed in distinct or overlapping neuronal populations (Table 1).
Not all have been investigated in detail beyond demonstrating their expression in the cNTS and only few have been selectively targeted to study their physiological roles ( Figure 2, Table 1). In the last decade, advances in chemo-and optogenetic manipulation have made it possible to selectively activate or inhibit cells in an anatomically and genetically defined manner. These advances will not be discussed in detail here, but are highlighted with appropriate references in Figure 2 and

| AFFERENT CONNECTIONS OF THE CNTS
Input to the cNTS arises from widespread regions in the brain, as well as peripheral sites (Figure 3), comprising an anatomical organization that is reminiscent of the significant variety of physiological and psychogenic stimuli, which modulate the activity of the NTS. 92 These inputs have been reported predominantly in rats, although studies using mice, rab-  Table 1. Highlighted in green are cell types that have been manipulated chemo-or optogenetically to investigate their function as indicated in Table 1. There is conflicting evidence on the location of cocaine-and amphetamine-regulated transcript (CART) neurons, possibly as a result of species differences. For details, see Table 1. Abbreviations are indicated in Table 2 T A B L E 1 Expression of neuropeptides and selected small-molecular neurotransmitters, intracellular proteins, and receptors in the caudal nucleus of the solitary tract (cNTS)

| Sensory inputs to the cNTS
The cNTS is directly sensitive to blood-borne signals, including changes in glucose, 94  inputs. These monosynaptic inputs from peripheral organs makes the NTS the primary brain region to receive and process rapid, neurochemical information regarding the internal environment of the body.
Indeed, the cNTS is robustly activated in response to interoceptive stimuli. 4,6,92 Interestingly, however, the cNTS also receives widespread central inputs and is engaged following psychogenic stressors, suggesting that the cNTS is also sensitive to information regarding emotional states. 3,12,102

| Central inputs to the cNTS
Central inputs to the NTS, which appear to be similar in rats and mouse, 11 are depicted in Figure 3 alongside efferent outputs. Below, I   Table 2 describe, in some detail, our current state of knowledge of the central inputs to the NTS. Of note, few studies have been able to limit the injection of retrograde tracers to the cNTS without significant leakage to more rostral areas or to the AP and DMV. This limitation makes it difficult to conclude with certainty which regions provide input to the cNTS specifically. In addition, many retrograde tracers, including the widely used CTb, are taken up by fibres of passage. In the case of the NTS, this could mean any descending projections to the spinal cord not terminating in the NTS may take up and transport CTb. 103 For a few brain regions, those limitations have been addressed by combining anterograde and retrograde tracing.

| Telencephalic inputs
The insular, prelimbic and infralimbic cortices all provide significant bilateral input to the NTS in mice 11,12 and rats. 11,104,105 In rats, infralimbic neurons directly synapse onto catecholaminergic neurons in the NTS. 106 These descending inputs appear to mediate cortical modulation of sympathetic and parasympathetic activity 107,108 and, as such, may represent a functional link between emotional processing and autonomic outflow.
Subcortically, several regions of the extended amygdala innervate the NTS, including the central amygdala and bed nucleus of the stria terminalis of rat, 11,14,105 mouse 11,12 and rabbit. 109 Interestingly, all of these extended amygdala inputs appear to be exclusively ipsilateral. 11 Most input from the central amygdala arises from the medial subdivision in both rats 11,110 and mice, 12 although the lateral subnucleus also provides some synaptic input. 12 In the rat cNTS, central amygdala inputs terminate mostly in the medial and dorsomedial subnuclei and not only are predominantly GABAergic, 14 but also may release a range

| Diencephalic inputs
Arguably the densest central input to the NTS arises from the paraventricular nucleus of the hypothalamus (PVN), evidenced in rats 11,105,113 and mice. 11,12 This input is bilateral, 11 primarily originates in the more caudal parts of the PVN 114 (rat) and appears to represent a distinct parvocellular population, which does not overlap with neuroendocrine magnocellular PVN neurons in mice 11 and rats. 115 In rats, 60% of NTS-projecting PVN neurons express the stress neuropeptide corticotropin-releasing hormone (CRH), 114 and PVN CRH neurons project directly to the NTS in mice. 116 Evidence from rats suggests that a much smaller population (6%-10%) of NTS-projecting PVN neurons express oxytocin, 114,117 PVN axons in the NTS express oxytocin, 118 and electrical stimulation of the PVN leads to release of oxytocin into the dorsal vagal complex. 119 In mice, PVN oxytocin cells do not provide significant direct synaptic input to the cNTS, although oxytocinergic fibres are clearly visible in the NTS of mice. 120 Finally, a subset of NTS-projecting PVN neurons express the melanocortin 4 receptor. 121 Removal of this descending input from the PVN leads to the development of obesity, 122 although data from mice suggests this pathway has no effect on ad libitum feeding. 123 One possibility is that stress, a powerful stimulus to suppress eating in rodents, activates NTS-projecting PVN neurons in mice, 12 which in turn mediate stress-induced activation of cNTS neurons, including those that express catecholamines in rats. 124 Other hypothalamic inputs include the arcuate nucleus, the dorsomedial hypothalamus, and the lateral hypothalamus in both mice 11,12 and rats. 11,105 In addition, neurons in the ventromedial hypothalamus may innervate the NTS in mice, 125 although not every comprehensive study reported input from the dorsomedial and ventromedial hypothalamus in mouse. 11,12 Interestingly, descending input from the arcuate nucleus does not appear to arise from agouti-related peptide or pro-opiomelanocortin (POMC) neurons in mouse, 93 while in rat, evidence suggest a small population of POMC neurons do project to the dorsal vagal complex. 126 Finally, the parasubthalamic nucleus provides heavy, unilateral input to the NTS in mice 11,12 and rats, 11 a pathway that may mediate fear-induced changes in autonomic outflow, 127 although studies of this particular nucleus are scarce. Indeed, the phenotype of these NTS-projecting parasubthalamic neurons remains unknown, but may include tachykinin-expressing, 128 CRH-expressing 129 and/or glutamatergic neurons. 127

| Mesencephalic and hindbrain inputs
The periaqueductal grey, Edinger-Westphal nucleus, parabrachial nucleus, Kölliker-Fuse nucleus and Barrington's nucleus all provide direct input to the NTS in rats and mice. 11,12,130,131 Parabrachial input appears to mainly arise from glutamatergic, non-calcitonin-gene related peptide neurons in mice, 132 whereas tachykinin-expressing neurons in the periaqueductal grey may be the source of input to the NTS in rats. 133,134 We recently found that NTS-projecting Barrington's nucleus neurons are activated in response to acute restraint stress in mice and express the stress neuropeptide CRH, 12 supporting the idea that the NTS is engaged following psychogenic stimuli.
Finally, multiple lower brainstem regions provide input to the NTS in the mouse and rat, including the raphe obscurus, the raphe magnus, the reticular nucleus, the parapyramidal regions, the gigantocellular nucleus 11,12

| Circumventricular organs
Neurons in the cNTS send projections to a number of sensory circumventricular organs: the AP, 141 the subfornical organ 142 and the vascular organ of laminar terminalis. 143 NTS input to the subfornical organ is inhibitory and may relay signals from peripheral baroreceptors in the rat. 144

| Telencephalic projections
Notably, there is no evidence that the cortex or any of the hippocampal regions receive monosynaptic input from NTS neurons. Subcortically, the entire extended amygdala receives input from NTS neurons: The bed nucleus of the stria terminalis, the nucleus accumbens, the medial septum, the substantia innominata and the central amygdala are all synaptic targets of cNTS neurons in the rat. 31,143,145,146 At least a subset of NTS inputs to the bed nucleus of the stria terminalis in mice are GABAergic, 140 suggesting that this pathway is partly inhibitory, although other NTS cell types are known to project to these regions as well, including GLP-1 neurons in mouse 35 and rat 143 and catecholaminergic neurons in rat, 146 but not NTS POMC neurons in mouse. 93

| Diencephalic projections
Diencephalic targets include multiple regions in the hypothalamus.
The PVN is a particularly densely innervated region in rats 139,143,147 and mice, 140 and the input is at least partly made up of GLP-1 35,143 (mouse and rat), catecholaminergic 143 (rat), GABAergic 140 (mouse) and POMC fibres (mouse). 93 Other hypothalamic targets include the dorsomedial hypothalamus, the lateral hypothalamus and the arcuate nucleus. 143,147 These inputs are at least partly made up of GLP-1 and catecholaminergic projections in the rat 143

| Mesencephalic and pontine projections
In the midbrain, the ventral tegmental area, the dorsal raphe and the periaqueductal grey all receive input from the NTS in the rat. 143 Further caudal, the Kölliker-Fuse nucleus, parabrachial nucleus, locus coeruleus and Barrington's nucleus are targets of NTS efferents in rats. 143 Efferents to the parabrachial nucleus are assumed to drive suppression in appetite, 148 and, in the mouse, include input from POMC, 93 GLP-1, 35 CCK 21 and noradrenergic neurons. 22 In the very caudal pons, the rostroventrolateral medulla 9 (cat) and DMV 9,149 (cat and rat) make up a subset of the brain-wide autonomic control centres, which receive dense projections from the NTS. Finally, the neighbouring AP receives light input from the cNTS in the cat. 141

| Spinal connections
In cats, the NTS projects to the thoracic ventral horn, the encoding GFP or mCherry represented a significant step forward. 155 In combination with Cre-expressing transgenic mice or rats, this genetically modified rabies virus is efficient, exclusively retrograde, strictly monosynaptic and cell-type specific. 155 Using such a EnvA-ΔG-RABV we recently mapped the monosynaptic inputs to GLP-1 neurons in the mouse cNTS 12 and found that GLP-1 neurons receive dense monosynaptic input from many of the same regions that provide input to the cNTS as a whole. Notable exceptions included cortical regions, the arcuate nucleus, the ventral tegmental area and the linear raphe nucleus. 12 We also identified polysynaptic inputs, including the hippocampal formation, the arcuate nucleus and the paraventricular thalamus. 12 6.3 | Does anatomy predict function? Mapping these circuits is one step, although understanding their role in the modulation of behaviour and physiology is hampered by our difficulty in specifically manipulating subsets of neurons that provide direct synaptic inputs to molecularly defined cell types. The limiting factor has been the toxicity of rabies virus, leading to cell death within weeks of infection. 157 Further improvements in the toxicity of rabies viruses have been reported and may allow specific populations of input neurons to be manipulated for behavioural testing. 157 Regarding efferent connections, cNTS GLP-1 neuron projections appear to be significantly more widespread than those of NTS POMC neurons in the mouse. 93 Although GLP-1 neurons innervate multiple regions in the extended amygdala and hypothalamus, 35,143 the only forebrain regions to receive input from cNTS POMC neurons are the PVN, the PSTh, and the medial subnucleus of the central amygdala. 93 It will be interesting to determine whether these differences in inputs and outputs are matched by differences in function. Although both populations decrease food intake, an important functional difference appears to be their effect on heart rate: optogenetic activation of cNTS POMC neurons leads to a decrease in heart rate, 54 but chemogenetic activation of GLP-1 neurons increases heart rate. 37 Interestingly, both GLP-1 and POMC neurons are activated by solitary tract stimulation and CCK, [158][159][160] suggesting at least some overlap in the stimuli that engage them.
6.4 | The importance of remembering the bigger picture in the study of single subpopulations of cNTS neurons Peptidergic cNTS neurons, and perhaps GLP-1 neurons in particular, 92,102 are exquisitely well-positioned to integrate interoceptive or psychogenic signals. It is their anatomical configuration that enables this integration of multimodal signals of physical and mental well-being.
In turn, GLP-1 neurons have the ability to impact truly varied processes both autonomic and behavioural (Table 1) (Table 1) and, when investigating the function of these other populations, we should keep in mind that they are not unlikely to modulate multiple downstream targets and, as a result, multiple physiological and behavioural processes simultaneously, as discussed for GLP-1 neurons above. Examples of this ability to modulate multiple processes are provided in Table 1. Subpopulations of neurons do not work in isolation in the living organism and their artificial activation through chemo-or optogenetics is likely to have impact beyond the single output measured in most experiments. The cNTS is perhaps particularly sensitive to this as a result of its position as a link between sensory and emotional inputs, as well as its ability to modulate both behaviour and physiology.

| CONCLUSIONS AND FUTURE DIRECTIONS
The afferent inputs to the NTS make it ideally suited to respond to both psychogenic and interoceptive stimuli, whereas its efferent connections facilitate widespread modulation of autonomic function and motivated behaviour. However, the specific circuits and cell types contributing to the functions of the NTS are still not fully understood. Future studies should take advantage of recently developed retrograde AAVs and the targeting of ChR2-expressing terminals to selectively manipulate neurons, based on not only their neurochemical phenotype, but also their projection targets.
These circuit-and cell type-specific studies will, in combination with previously published classic knife-cut and toxin studies, provide new insights into the functions of subpopulations of cNTS neurons.

CONFLICTS OF INTEREST
The author declares that he has no conflicts of interests.

DATA AVAILABILITY
Data sharing is not applicable to this review because no new data were created or analyzed.