Gliotransmission and adenosinergic modulation: insights from mammalian spinal motor networks

Abstract

Astrocytes are proposed to converse with neurons at tripartite synapses, detecting neurotransmitter release and responding with release of gliotransmitters, which in turn modulate synaptic strength and neuronal excitability. However, a paucity of evidence from behavioral studies calls into question the importance of gliotransmission for the operation of the nervous system in healthy animals. Central pattern generator (CPG) networks in the spinal cord and brain stem coordinate the activation of muscles during stereotyped activities such as locomotion, inspiration, and mastication and may therefore provide tractable models in which to assess the contribution of gliotransmission to behaviorally relevant neural activity. We review evidence for gliotransmission within spinal locomotor networks, including studies indicating that adenosine derived from astrocytes regulates the speed of locomotor activity via metamodulation of dopamine signaling.

INTRODUCTION

Locomotor behaviors such as walking, flying, and swimming permit mammals to interact with their environment, thereby satisfying basic requirements of survival. In vertebrates, locomotion arises from the biomechanical properties of the skeletomuscular system and activity within dedicated neural circuitry in the central nervous system (CNS) (Dickinson et al. 2000; Grillner 2006; Kiehn 2006 2016; Orlovsky et al. 1999). Whereas the planning, initiation, and maintenance of movements involve various brain regions, including the cortex, basal ganglia, midbrain, and hindbrain (Orlovsky et al. 1999; Takakusaki et al. 2004), executive control over the timing and coordination of muscle activity is invested in networks of interneurons in the ventral horn of the spinal cord. These networks, like others in the spinal cord and brain stem that produce a stereotyped motor output, are referred to as central pattern generators (CPGs).

Spinal motor networks, including the motoneurons that provide their final common output, are subject to extensive neuromodulation by substances such as the biogenic amines 5-hydroxytryptamine (5-HT/serotonin), dopamine (DA), and norepinephrine, and amino acids including γ-aminobutyric acid (GABA) and glutamate (Miles and Sillar 2011). A given neuromodulator may act at multiple receptor subtypes and in a cell type-specific fashion. In addition, the effects of an individual neuromodulator may be modified by a second-order neuromodulator, a process termed metamodulation (Katz 1999; Miles and Sillar 2011). The resulting multiplicity of signaling mechanisms extends the repertoire of motor patterns that can be generated by a finite population of cells, contributing to the behavioral plasticity necessary for an animal to successfully negotiate its environment.

Although in many cases neuromodulators derive from neurons, they also may be secreted by astrocytes (Araque et al. 1999, 2014), as well as other glial cells (Jackson et al. 2017). Evidence gathered over the past two decades indicates that astrocytes release neuromodulators in response to activity in neighboring neurons, acting as the third partner, along with pre- and postsynaptic neurons, at tripartite synapses throughout the nervous system. So-called gliotransmission represents a form of information processing that extends the canonical role of astrocytes as cellular housekeepers, responsible for tasks including ion homeostasis, metabolic support, and neurotransmitter clearance (Verkhratsky and Butt 2013). However, the importance of gliotransmission to the operation of the nervous system in healthy animals remains controversial, in part because of a paucity of evidence that it contributes to the production of behavior (Agulhon et al. 2012; Bazargani and Attwell 2016; Hamilton and Attwell 2010; Nedergaard and Verkhratsky 2012; Sloan and Barres 2014).

Because the activity of neurons within the spinal cord determines quantifiable behaviors, spinal cord preparations may provide a tractable model for the study of gliotransmission and its role in the operation of neural networks. The lumbar spinal locomotor CPG controls the rhythmic activation of hindlimb muscles during walking and running, and rhythmic, behaviorally relevant patterns of activity can be generated by isolated CPG networks in the absence of descending and peripheral inputs. This has permitted the study of diverse neuromodulatory systems and their roles in networks controlling defined behaviors (Dickinson 2006; Harris-Warrick 2011; Marder and Bucher 2001; Miles and Sillar 2011; Whelan 2010). In the present article, we review recent studies indicating that astrocytes within the spinal cord are a source of neuromodulatory adenosine. We then discuss the modulation of motor networks by adenosine, including recent evidence that it functions principally as a metamodulator of dopamine during motor behaviors. For the most part, we discuss evidence from mice and rats, the models used in nearly all studies of gliotransmission to date. However, a recent report implicates bidirectional signaling between neurons and glia in the regulation of behavior in Drosophila, suggesting the possibility that a greater range of organisms may be investigated in future (Ma et al. 2016).

SPINAL LOCOMOTOR NETWORKS AND NEUROMODULATION

Networks of premotor interneurons selectively excite and inhibit pools of motoneurons with precise timing to ensure appropriate contraction and relaxation within antagonistic pairs of muscles, resulting in smooth, controlled movement (Orlovsky et al. 1999). During rhythmic locomotor behaviors, such as walking, flying, and swimming, spinal networks activate muscles in a cyclically repeated sequence. Like other CPGs within the brain stem and spinal cord, including the networks that coordinate muscle contractions during chewing, swallowing, and respiration (Feldman and Del Negro 2006; Jean 2001; Lund and Kolta 2006; Marder and Bucher 2001), locomotor networks remain capable of generating a basic motor pattern when inputs from descending and peripheral pathways are removed (Brown 1911; Guertin 2009).

The effects of neuromodulators have been studied extensively within spinal cord preparations, most notably from lampreys, Xenopus tadpoles, cats, rats, and mice (Dickinson 2006; Harris-Warrick 2011; Katz 1999; Katz and Frost 1996; Miles and Sillar 2011). Postnatal rodent spinal cords may be isolated and sustained in artificial cerebrospinal fluid in vitro, and rhythmic locomotor-related activity can be evoked in the hindlimb motor circuitry by application of agonists of glutamate and monoamine receptors or by electrical or optogenetic stimulation (Hägglund et al. 2013; Kiehn and Kjaerulff 1996; Kudo and Yamada 1987; Smith and Feldman 1987; Whelan et al. 2000). Isolated networks generate rhythmic activity in hindlimb muscles, if present, or in transected ventral roots containing the motoneurons that innervate them (Hayes et al. 2009; Kudo and Yamada 1987; Smith and Feldman 1987). In preparations in which the ventral roots are transected, field potential (electroneurogram) recordings reveal a pattern of rhythmic bursting in the lumbar ventral roots corresponding to activity observed during treadmill locomotion in adult animals (Cowley and Schmidt 1994; Kiehn and Kjaerulff 1996; Whelan et al. 2000). Bursting in lumbar ventral roots L₁–L₄ is in phase with hindlimb flexors and out of phase with bursting in roots L₅ and L₆, which corresponds to bursting in extensors.

The basic output of the locomotor CPG is determined by the intrinsic electrical properties of the neurons that constitute the network and by the synaptic connections between them (Grillner 2006). Neuromodulators adjust both, enabling a circumscribed population of neurons to produce diverse outputs and modes of adaptive behavior (Dickinson 2006; Harris-Warrick 2011; Katz 1999; Katz and Frost 1996; Miles and Sillar 2011). Alterations to motor network output allow animals to adjust their behavior according to developmental stage and physiological state and to meet the challenges posed by different environmental conditions (Grillner 2006). For instance, changes in flexor-extensor coordination within locomotor networks underlie transitions from one gait to another, and gait changes allow animals to efficiently vary the speed of their locomotion (Bellardita and Kiehn 2015; Orlovsky et al. 1999). Multifarious neuromodulatory influences acting in concert give rise to a vast repertoire of network and behavioral outputs and are thus a source of considerable behavioral flexibility (Harris-Warrick 2011; Miles and Sillar 2011). The iterative nature of fictive locomotor activity facilitates comparison of control and treatment conditions, and isolated spinal networks have been used to study the effects of diverse neuromodulators at the network level, by application of pharmacological and genetic tools. Although neuromodulation in locomotor networks has received considerable attention, evidence that neuron-glia interactions play a role has only recently emerged (Acton and Miles 2015; Witts et al. 2012).

GLIOTRANSMISSION

Glia are a class of cell within the central and peripheral nervous systems consisting of macroglia, which include astrocytes, oligodendrocytes and Schwann cells, and microglia (Verkhratsky and Butt 2013). Astrocytes, like the other glial cell types, are traditionally considered to have a merely passive, supportive function within neural networks, with well-established roles in ion homeostasis and the synthesis and clearance of neurotransmitters (Verkhratsky and Butt 2013). However, substantial evidence now indicates that astrocytes are active participants at tripartite synapses, dynamically regulating neuronal excitability and synaptic strength by the release of gliotransmitters (Araque et al. 1999, 2014; Bazargani and Attwell 2016; Haydon and Nedergaard 2015). In its original formulation, the tripartite-synapse model entails 1) the binding of a neurotransmitter released during synaptic activity to an astrocytic Gα_q-linked G protein-coupled receptor (GPCR); 2) the production of inositol trisphosphate (IP₃) and activation of IP₃ receptors, triggering release of Ca²⁺ from stores within the astrocyte; 3) Ca²⁺-dependent release of a gliotransmitter such as glutamate (Bezzi et al. 1998; Han et al. 2013; Oh et al. 2012; Parpura et al. 1994; Pasti et al. 1997), adenosine triphosphate (ATP) (Newman 2001; Pryazhnikov and Khiroug 2008), or d-serine (Henneberger et al. 2010; Mothet et al. 2005) by vesicular exocytosis or via ion channels; and 4) the activation of metabotropic or ionotropic receptors on either the pre- (Carlsen and Perrier 2014; Jourdain et al. 2007) or postsynaptic neuron (Angulo et al. 2004; Fellin et al. 2004; Parri et al. 2001) (Fig. 1) of the same (Martín et al. 2015; Navarrete and Araque 2010) or a different (Pascual et al. 2005; Serrano et al. 2006; Zhang et al. 2003) synapse. Although signaling in this manner has been denoted “gliotransmission,” this is arguably a misnomer: modulation of neuronal activity by substances derived from astrocytes occurs on a variable timescale but, because of its dependency on metabotropic receptors, is necessarily slower than acute transmission mediated by ionotropic receptors (Agulhon et al. 2012; Araque et al. 2014). “Gliomodulation” has therefore been proposed as a more appropriate term to distinguish bidirectional signaling between neurons and glia (Agulhon et al. 2012) but is not widely used (“gliotransmission” is used in this review).

Fig. 1.

Established model of bidirectional signaling between neurons and glia at the tripartite synapse. Neurotransmitters activate astrocytic Gα_q-linked G protein-coupled receptors (GPCRs), stimulating the production of inositol trisphosphate (IP₃), activation of astrocytic IP₃ receptors (IP₃R2), and release of Ca²⁺ from internal stores. An increase in intracellular Ca²⁺ triggers release of gliotransmitters including glutamate, ATP, and d-serine by a vesicular or channel-mediated mechanism; these in turn bind to pre- or postsynaptic GPCRs, or in the case of d-serine, N-methyl-d-aspartate (NMDA) receptors, to modulate synaptic strength or neuronal excitability.

The steps in the pathway described above have been deduced largely from Ca²⁺ imaging of astrocytes and electrophysiological recordings from neurons and are supported by numerous studies (Haydon and Nedergaard 2015). However, the extent to which gliotransmission is important for the operation of neural networks and in the production of behavior in healthy animals remains controversial (Agulhon et al. 2012; Bazargani and Attwell 2016; Hamilton and Attwell 2010; Nedergaard and Verkhratsky 2012; Sloan and Barres 2014). The coding of information by astrocytes in the form of Ca²⁺ fluctuations, which vary in kinetics and subcellular localization, is poorly understood, and the physiological relevance of experimental manipulations used to investigate Ca²⁺ signaling is disputed (Nedergaard and Verkhratsky 2012; Rusakov 2015). Crucially, evidence is lacking that gliotransmission dependent on astrocytic Gα_q-linked GPCRs and Ca²⁺ signaling is important in the generation of behavior.

GLIOTRANSMISSION AND BEHAVIOR

Reports have indicated astrocytic involvement in diverse behaviors (reviewed in Oliveira et al. 2015); these include sleep (Foley et al. 2017; Halassa et al. 2009; however, see Fujita et al. 2014), pain aversion (Foley et al. 2011), recognition memory (Florian et al. 2011; Halassa et al. 2009), motor coordination (Watase et al. 1998), whisker-dependent perception (Han et al. 2014), olfactory responses (Martin et al. 2012), and anxiety-like and depressive behavior (Banasr and Duman 2008; Cao et al. 2013; John et al. 2012). However, despite extensive evidence for bidirectional communication between neurons and astrocytes in in vitro preparations, evidence that gliotransmission is important for the production or modulation of behavior is sparse (for further discussion, see Agulhon et al. 2012; Bazargani and Attwell 2016; Hamilton and Attwell 2010; Nedergaard and Verkhratsky 2012; Sloan and Barres 2014). Uncertainty about the contribution of gliotransmission to the generation of behaviors stems from a lack of acute and selective techniques for the manipulation of astrocytes in vivo and uncertainty about the physiological relevance of the techniques that have been applied to the study of gliotransmission to date (reviewed in Bazargani and Attwell 2016 and Nedergaard and Verkhratsky 2012). Importantly, mice lacking IP₃R2, the IP₃ receptor isoform that is preferentially expressed in astrocytes, (IP₃R2KO mice) have been tested for a range of behaviors, including locomotion, learning, memory, and anxiety, but do not differ from wild-type littermates on most measures (Agulhon et al. 2013; Petravicz et al. 2014); however, the possibility that the importance of IP₃R signaling is masked in these studies by compensatory mechanisms during development cannot be excluded. Until further behavioral evidence for gliotransmission in healthy animals is provided, it will remain controversial.

Given the limitations of the techniques currently available for studying gliotransmission in behaving animals, in vitro and in vivo rhythmically active brain stem preparations and in vitro spinal cord preparations may provide insight into the role of bidirectional neuron-glia signaling in behavior. Studies of brain stem networks for respiration and mastication have provided cogent evidence for behaviorally relevant gliotransmission. In the brain stem, astrocytic Ca²⁺ elevations are evoked by reductions in both pH (Gourine et al. 2010; Kasymov et al. 2013) and the partial pressure of oxygen (Angelova et al. 2015), in both cases causing release of ATP, which stimulates breathing. In intact rats, optogenetic stimulation of astrocytes also triggers release of ATP to stimulate breathing (Gourine et al. 2010), whereas expression of tetanus toxin in astrocytes to inhibit vesicular release prevents increases in respiration normally observed in response to reduced oxygen availability (Angelova et al. 2015). The release of modulators from astrocytes described by these studies departs from the tripartite synapse model, however, because it is stimulated by astrocytic chemoception, rather than detection of a neurotransmitter. Glia are also proposed to modulate neuronal responses to ATP within the pre-Bötzinger complex (pre-BötC), an area of the medulla critical for production of the respiratory rhythm, by releasing glutamate in a Ca²⁺-dependent manner (Huxtable et al. 2010). A further example of neuron-glia cross talk in a rhythmic network is provided by the brain stem circuity for mastication, where sensory input is proposed to activate astrocytic N-methyl-d-aspartate (NMDA) receptors, eliciting release of the Ca²⁺-binding protein S100β, with the resulting decrease in extracellular Ca²⁺ concentration conferring rhythmic bursting properties on neighboring neurons (Morquette et al. 2015). Evidence from networks that generate rhythmic, stereotyped motor behaviors (including evidence from the locomotor CPG, described below) therefore may help to illuminate the extent to which astrocytes participate in information processing within the nervous system of intact animals.

EVIDENCE FOR GLIOTRANSMISSION IN THE SPINAL CORD

Glial Ca²⁺ Signaling

Although a substantial literature is dedicated to the role of astrocytic information processing in in vitro brain preparations, research into gliotransmission in the spinal cord is in its infancy. Information about the dynamics of Ca²⁺ signaling in astrocytes has been provided for the most part by studies in the cortex, hippocampus, and cerebellum. Ca²⁺ signaling in astrocytes has been extensively studied by imaging cells expressing a genetically encoded Ca²⁺ indicator or loaded with Ca²⁺-sensitive dyes by means of a patch pipette. Astrocytes are acutely sensitive to neuronal activity, to the extent that they can respond to basal synaptic activity stimulated by a single action potential (Panatier et al. 2011). They display intracellular Ca²⁺ elevations in response to neuronal activity in both neonatal and adult rodents in vitro and in vivo (Araque et al. 2014; Bazargani and Attwell 2016; Rusakov 2015; Shigetomi et al. 2016). Ca²⁺ signaling is observed in awake mice in response to sensory stimulation in the brain and dorsal horn of the spinal cord (Sekiguchi et al. 2016; Srinivasan et al. 2015) and during locomotion in the cortex (Dombeck et al. 2007). Given these observations, astrocytes within the ventral horn of the spinal cord may be expected to display rhythmic oscillations corresponding to those of neuronal components of the locomotor CPG during network activity. Although imaging of ventral horn astrocytes in behaving animals may be challenging because of the difficulty of accessing tissue, it is possible to detect Ca²⁺ signaling in astrocytes in in vitro preparations during network activity. Preliminary studies have reported rhythmic Ca²⁺ fluctuations in putative astrocytes during pharmacologically induced locomotor-related activity in in vitro hemicord preparations (Chub et al. 2012). Furthermore, Ca²⁺ fluctuations are selectively blocked during inhibition of metabotropic glutamate receptor 1 (mGluR1) during ventral root stimulation in disinhibited preparations, implying a role for that receptor in neuron-glia signaling in the spinal cord, as reported in the brain (Chub and O’Donovan 2011). Consistent with these results, a subset of astrocytes in the pre-BötC displays rhythmic Ca²⁺ activity immediately preceding rhythmic respiratory-related neuronal bursts (Okada et al. 2012; Oku et al. 2016). Interestingly, pre-BötC astrocytes also display low-frequency synchronized Ca²⁺ transients during network activity, indicating a complex relationship between activity in neurons and astrocytes in this network. This finding suggests that astrocytes in locomotor networks may also display patterns of Ca²⁺ signaling that do not directly correspond to rhythmic locomotor-related activity in neurons. Apart from preliminary evidence from the Ca²⁺ imaging studies cited above, indirect evidence also supports neuron-to-astrocyte signaling in the spinal cord; this is discussed below.

Stimulation of Spinal Cord Astrocytes

A range of techniques is available for the experimental induction of astrocytic Ca²⁺ elevations (for further information, see Araque et al. 1999; Bazargani and Attwell 2016; Nedergaard and Verkhratsky 2012), including ultraviolet (UV) photolysis of caged Ca²⁺ or IP3 introduced via a patch pipette (Hua et al. 2004; Martín et al. 2015; Wang et al. 2013), depolarization of the astrocyte under whole cell patch-clamp conditions (Jourdain et al. 2007; Kang et al. 1998), mechanical stimulation (Hua et al. 2004; Lee et al. 2015), activation of endogenous or transgenically expressed GPCRs, including DREADDs (designer receptors exclusively activated by designer drugs; Agulhon et al. 2010, 2013; Chen et al. 2016; Rae and Irving 2004; Shigetomi et al. 2008), and activation of transgenically expressed channelrhodopsins (Beppu et al. 2014; Gourine et al. 2010; Li et al. 2012). Recent studies of gliotransmission in the spinal cord have exploited protease-activated receptor-1 (PAR1) to trigger or enhance astrocytic Ca²⁺ signaling (Acton and Miles 2015; Carlsen and Perrier 2014). PAR1 is an endogenous GPCR associated with Gα_q proteins, and selective activation by a synthetic ligand results in Ca²⁺ elevations in astrocytes (Lalo et al. 2014; Lee et al. 2007; Shigetomi et al. 2008). In the spinal cord (Acton and Miles 2015), as in the brain (Junge et al. 2004; Weinstein et al. 1995), PAR-1 is expressed by cells that also express glial fibrillary acidic protein (GFAP), an intermediate filament protein used widely as an astrocyte marker (Wang and Bordey 2008); by contrast, neurons do not appear to express the receptor in the ventral horn (Acton and Miles 2015). Cells expressing PAR1 in the spinal cord are therefore proposed to be astrocytes. However, the possibility that a subset of neural precursor cells of the astrocyte lineage also expresses PAR1 has not been excluded; gliogenesis continues during postnatal stages at which network-level and single-cell recordings are typically made (Tien et al. 2012), and it has been proposed that some neural precursor cells express GFAP in the spinal cord during postnatal development (Chvátal et al. 1995). In addition, the adult spinal cord contains GFAP+ multipotent neural stem cells that are quiescent within the intact spinal cord (Fiorelli et al. 2013). Such populations of nonastrocytic GFAP+ cells must therefore be taken into account when studies using GFAP as a marker of astrocytes in the spinal cord are being interpreted. Whether PAR1 activation results in Ca²⁺ signaling exclusively in terminally differentiated astrocytes remains to be confirmed, as does the proportion of astrocytes that expresses PAR1. Nevertheless, PAR1 activation offers several advantages as a means of stimulating astrocytes in the spinal cord: activation of PAR1 does not affect activity in locomotor networks following pharmacological ablation of astrocytes, indicating that it does not directly modulate the activity of other cell types (Acton and Miles 2015); similarly, it does not alter the excitability of neurons in the brain (Lalo et al. 2014; Lee et al. 2007; Shigetomi et al. 2008); it is expressed from early postnatal stages in cells across the ventral horn, allowing those cells to be stimulated concurrently during activity of the locomotor CPG by bath application of a selective PAR1 agonist; and finally, because it is an endogenous receptor, experimental PAR1 activation may result in Ca²⁺ elevations with spatiotemporal properties close to those generated by GPCR activation during neural activity in vivo, which may not be the case for transgenically expressed receptors.

During pharmacologically induced network activity in in vitro spinal cord preparations from postnatal mice, stimulation of glia by PAR1 activation results in a transient reduction in the frequency but not the amplitude of rhythmic bursting (Fig. 2, A–D) (Acton and Miles 2015). This effect is blocked by theophylline, a nonselective adenosine receptor antagonist, and cyclopentyl dipropylxanthine (DPCPX), a selective antagonist of A₁-adenosine receptors, indicating that it is mediated by adenosine. Extracellular adenosine is largely formed when ATP is released from cells and then hydrolyzed by ectonucleotidases in the extracellular space, although adenosine may also be secreted from cells directly (Cunha 2001; Klyuch et al. 2012; Wall and Dale 2013). Given that the activity of spinal neurons is unaltered when glia are stimulated in the presence of an ectonucleotidase inhibitor, spinal glia appear to release ATP rather than adenosine itself (Acton and Miles 2015; Carlsen and Perrier 2014).

Fig. 2.

Adenosine derived from glial cells acts via A₁ receptors to modulate locomotor-related activity in a murine spinal cord preparation. A: an in vitro spinal cord preparation in which suction electrodes are used to recording compound action potentials from motoneurons in transected ventral roots. B: sample raw (top) and rectified/integrated (bottom) ventral root traces showing rhythmic locomotor-related bursting evoked by bath application of 5-hydroxytryptamine (10 µM), NMDA (5 µM), and dopamine (50 µM) in control conditions, following application of Thr-Phe-Leu-Leu-Arg-NH₂ trifluoroacetate salt (TFLLR; 10 µM) to selectively activate glial protease-activated receptor-1 (PAR1) and during washout. Ci: locomotor-burst frequency over 3 min during a control period, immediately following TFLLR application, and following a 20-min washout period. Individual data points are shown in gray, and means are represented by black lines. Error bars indicate ± SE Cii: time course plot of normalized data aggregated into 1-min bins showing a reduction in burst frequency upon application of TFLLR (n = 10 preparations). Di: locomotor-burst amplitude is unaffected by TFLLR application. Dii: time course plot showing no change in burst amplitude upon application of TFLLR. Ei: sample traces showing the effects of the nonselective adenosine receptor antagonist theophylline (20 µM) on locomotor-related activity. Eii: theophylline increases the frequency of locomotor-related bursting, revealing the modulatory role of endogenous adenosine derived from sources within the ventral horn. Values are means ± SE; n = 6. Eiii: the selective A₁-adenosine receptor antagonist cyclopentyl dipropylxanthine (DPCPX; 50 µM) also reveals modulation of locomotor-related activity by endogenous adenosine and prevents further frequency increases in the presence of theophylline, demonstrating that adenosine acts via A₁ receptors. Values are means ± SE; n = 6. Fi: traces showing the effects of the glial toxin fluoroacetate (FA; 5 mM) when coapplied with glutamine (1.5 mM) and a lack of modulation by endogenous adenosine following glial ablation, as revealed by application of theophylline (20 µM). Fii: theophylline has no effect on the frequency of locomotor activity when applied in the presence of the glial toxins FA and methionine sulfoximine. Values are means ± SE; n = 6. *P < 0.05; **P < 0.01. [B–D are adapted from Acton and Miles (2015). E and F are adapted from Witts et al. (2012).]

It is interesting that inhibition of A₁ receptors during fictive locomotion is sufficient to block the effects of PAR1 stimulation, suggesting that ATP-adenosine is the only gliotransmitter produced in these experiments (Acton and Miles 2015). By contrast, stimulation of astrocytes in the dorsal horn has been shown to elicit release of glutamate (Bardoni et al. 2010; Nie et al. 2010), and astrocytes in the brain release substances including glutamate, d-serine, and GABA (Araque et al. 1999, 2014), all of which modulate spinal motor networks (Acton and Miles 2017; Bertrand and Cazalets 1999; Iwagaki and Miles 2011; Miles and Sillar 2011; Wang and Dun 1990). Astrocytes in the ventral horn may not, therefore, match neurons in the diversity of neuromodulatory substances they secrete. However, some techniques fail to elicit gliotransmitter release in preparations where others are effective (Shigetomi et al. 2008; Wang et al. 2013). Activation of either PAR1 or the purinergic receptor P2Y1, both endogenous astrocytic Gα_q-linked GPCRs, elicits Ca²⁺ elevations of similar amplitude in the somas of astrocytes in the CA1 region of the hippocampus, but only PAR1 activation stimulates gliotransmitter release. The cause of this discrepancy is unknown, but could, for instance, be related to differences in the subcellular localization of the receptors in relation to Ca²⁺ stores and sites of gliotransmitter release (Bazargani and Attwell 2016). For this reason, it is desirable that several techniques for stimulating astrocytes be compared in future studies. In addition, the experiments described above do not fully consider state-dependent differences in gliotransmission. It is conceivable that the influence of astrocyte-derived modulators varies according to the fluctuating neuromodulatory milieu of the spinal cord (Acevedo et al. 2016; Bazargani and Attwell 2016; Gerin et al. 1995; Gerin and Privat 1998). State-dependent modulation of networks by astrocytes may involve changes in the competence of astrocytes to detect neuronal activity or release modulators, or changes in neuronal sensitivity. The possibility that adenosinergic modulation is dependent on the actions of dopamine is discussed below.

In agreement with adenosine release during network activity, PAR1 activation elicits the release of ATP-adenosine from ventral horn glia in slice preparations from postnatal mice, reducing the amplitude of evoked excitatory postsynaptic currents (EPSCs) recorded from interneurons via a presynaptic mechanism mediated by A₁ receptors (Carlsen and Perrier 2014). Also, consistent with findings from rhythmically active preparations, this depends on extracellular ectonucleotidase activity, indicating that adenosine but not ATP modulates ventral horn interneurons. However, despite apparent agreement, it remains unclear whether adenosine acts via a common mechanism in spinal cord slices and rhythmically active isolated spinal cord preparations (see below).

Suppression of Gliotransmission in the Spinal Cord

Endogenous release of substances from astrocytes can be suppressed by inhibition of astrocytic Ca²⁺ signaling by various strategies, including dialysis of astrocytic syncytia by Ca²⁺ buffers and inhibition of IP₃R signaling (for further information, see Bazargani and Attwell 2016; Hamilton and Attwell 2010; Nedergaard and Verkhratsky 2012). In addition, secretion of gliotransmitters can be suppressed by inhibition of G protein signal transduction (Di Castro et al. 2011), disruption of vesicle release mediated by N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) proteins (Araque et al. 2000; Gourine et al. 2010; Jourdain et al. 2007; Lalo et al. 2014; Sultan et al. 2015), or by conditional expression of toxins (Angelova et al. 2015) or dn-SNARE, a dominant-negative cytosolic domain of vesicle-associated membrane protein 2 (VAMP2; a.k.a. synaptobrevin 2) (Fellin et al. 2009; Fujita et al. 2014; Lalo et al. 2014; Pascual et al. 2005; Sultan et al. 2015).

Suppression of astrocytic Ca²⁺ signaling has not been achieved during locomotor-related network activity in spinal cord preparations, and techniques for selective and acute inhibition of astrocytic Ca²⁺ signaling are currently unavailable. In acute spinal cord slices from mice, loading astrocytes with a Ca²⁺ chelator enhances the amplitude of EPSCs in neighboring neurons (Carlsen and Perrier 2014). Although this is proposed to reflect a reduction in Ca²⁺-dependent release of ATP-adenosine from glia, this has not been tested directly. It has, however, been shown that adenosinergic modulation of network activity is prevented in murine preparations in which astrocytes have been treated with gliotoxins, which comprehensively disrupt astrocytic function (Clarke 1991; Fonnum et al. 1997; Huxtable et al. 2010; Li et al. 2013; Wall and Dale 2013; Witts et al. 2012; Zhang et al. 2003). Network activity persists following application of the gliotoxins methionine sulfoximine or fluoroacetate if glutamine, the astrocyte-derived precursor of glutamate and GABA, is supplemented to support rhythmic activity (Witts et al. 2012). Methionine sulfoximine irreversibly inhibits the astrocytic enzyme glutamine synthetase and causes glycogen accumulation and cell damage in astrocytes, but not other glial cells (Aschner and Kimelberg 1996; De Vellis 2002; Phelps 1975; Ronzio et al. 1969). Fluoroacetate, a precursor of the glial aconitase inhibitor fluorocitrate (Fonnum et al. 1997), is proposed to disrupt astrocyte metabolism by reversible inhibition of the Krebs cycle, resulting in decreased production of ATP (Aschner and Kimelberg 1996; Keyser and Pellmar 1994; Paulsen et al. 1987). Following astrocytic ablation by either substance, blockade of adenosine receptors by DPCPX or theophylline has no effect on network output, revealing that endogenous adenosine no longer modulates network activity (Fig. 2, E and F) (Witts et al. 2012). It should be noted that extensive disruption of astrocytic metabolism is an imprecise technique for suppressing gliotransmission, and off-target effects of gliotransmitters have been reported, including changes in extracellular K⁺ levels (Largo et al. 1997) and glycogen deposition in cranial motoneurons (Young et al. 2005). However, spinal cord preparations remain sensitive to adenosine following gliotoxin treatment, consistent with disruption of endogenous release of adenosine from astrocytes (Witts et al. 2012). Thus astrocytes are proposed as the source of modulatory adenosine in the mammalian spinal cord during locomotor-related activity, despite reported release of adenosine from neurons and microglia in other systems (Jackson et al. 2017; Wall and Dale 2013).

Adenosine is also released in the spinal cord during hypoxia and hypocapnia, and depresses reflex potentials recorded from the ventral roots (Lloyd et al. 1988; Otsuguro et al. 2006, 2009, 2011). Release of adenosine during hypoxia, although not hypocapnia, is inhibited by fluoroacetate, implying an astrocytic source (Takahashi et al. 2010). However, in this context it is not clear whether adenosine acts on spinal cord neurons or on peripheral afferents. In addition, it should be noted that in primary cultures of brain cells, oxidative stress was found to elicit release of adenosine from microglia and not from astrocytes or neurons (Jackson et al. 2017).

Together, these experiments demonstrate that activation of Gα_q-coupled GPCRs expressed by putative astrocytes in the spinal cord results in the production of adenosine, which in turn modulates neuronal activity via A₁-adenosine receptors. Production of modulatory adenosine following release of ATP from astrocytes appears to be widespread throughout the CNS, having also been reported in brain preparations (Panatier et al. 2011; Pascual et al. 2005; Serrano et al. 2006).

Indirect Evidence for Neuron-to-Glia Signaling in Motor Networks

The tripartite synapse model implies detection of neuronal activity by astrocytes, which respond with elevations in cytosolic Ca²⁺. Indirect support for neuron-to-astrocyte signaling during locomotor network activity is provided by the observation that modulation of the frequency of locomotor-related activity by endogenous adenosine scales with network activity (Acton and Miles 2015). This presumably represents activity-dependent release of ATP-adenosine by astrocytes and implies that ventral horn astrocytes detect neuronal activity, for instance, by detection of neurotransmitter release, as predicted by the tripartite synapse model. Astrocytes are therefore proposed to mediate a negative feedback loop, whereby activity-dependent release of adenosine depresses and perhaps stabilizes network output, resulting in smooth, controlled movement (Acton and Miles 2015). However, further Ca²⁺ imaging is needed to confirm spinal astrocytic responsiveness to neuronal activity, as has been demonstrated in the brain (Araque et al. 2014; Bazargani and Attwell 2016; see above).

Interestingly, it has been observed that astrocytes express receptors for the neurotransmitters released specifically by the neurons they make contact with, implying that astrocytic receptors function to detect synaptic transmission (Verkhratsky and Butt 2013). Thus spinal cord astrocytes are unusual in expressing receptors for glycine, the dominant inhibitory neurotransmitter in the spinal cord (Bowery and Smart 2006; Pastor et al. 1995). Both glycine and GABA_A receptors expressed by spinal cord astrocytes carry a Cl⁻ current, as they do in neurons (Pastor et al. 1995). Spinal cord astrocytes also express NMDA receptors and are sensitive to glutamate (Ziak et al. 1998); however, the physiological importance of these receptors remains unknown.

ADENOSINERGIC MODULATION OF SPINAL MOTOR CIRCUITRY

The purines ATP and its derivative adenosine are involved in a myriad of biological processes, most notably energy transfer. In the nervous system, they also function as neuromodulators and are involved in diverse processes in health and disease, including sleep homeostasis, memory, and the regulation of mood (Burnstock 2007; Cunha 2001; Fredholm et al. 2005). Experiments indicating release of modulatory adenosine from astrocytes in the ventral horn are complemented by studies into the functions and mechanisms of purinergic signaling during motor activity. However, some data point to unresolved complexity.

There are four adenosine receptor subtypes, designated A₁, A_2A, A_2B, and A₃, which differ in affinity for adenosine, structure, cellular distribution, and G protein coupling (Cunha 2001; Dunwiddie and Masino 2001). A₁ and A_2A receptors have the highest affinity for adenosine and are the best characterized (Cunha 2001). Both receptors are expressed throughout the spinal cord (Deuchars et al. 2001; Paterniti et al. 2011; Reppert et al. 1991). A₁ receptors are tightly coupled to the Gα_i pathway, which mediates inhibition of adenylyl cyclase and reduced production of cAMP (Cunha 2001). However, A₁ receptors are also coupled to other G proteins and act via adenylyl cyclase-independent pathways to inhibit spontaneous and evoked neurotransmitter release (Cunha 2001). A₁ receptors typically mediate presynaptic inhibition via the adenylyl cyclase-independent inhibition of N-type Ca²⁺ channels (Cunha 2001; Ribeiro 1995). A_2A receptors are primarily coupled to Gα_s, but also to Gα_i and Gα₁₂, and have diverse effects mediated by protein kinase A (PKA), protein kinase C (PKC), and N-type and P-type Ca²⁺ channels (Cunha 2001). In many systems, the inhibitory actions of A₁ receptors are countered by the facilitatory actions of A_2A receptors in a concentration-dependent manner (Cunha 2001).

The role of adenosine signaling in the modulation of mammalian locomotor behavior has been addressed by injection of intact animals with receptor antagonists, such as caffeine, which typically have an excitatory effect. This effect has been proposed to be mediated by either A₁ (Goldberg et al. 1985; Snyder et al. 1981), A_2A (El Yacoubi et al. 2000; Ledent et al. 1997; Lindskog et al. 2002; Svenningsson et al. 1995, 1997) or both A₁ and A_2A receptors (Antoniou et al. 2005; Karcz-Kubicha et al. 2003; Kuzmin et al. 2006), perhaps indicating a role for both receptor subtypes. However, interpretation of studies in which antagonists are chronically applied is complicated by the reported acquisition of receptor tolerance, to which A₁ receptors are particularly susceptible (Karcz-Kubicha et al. 2003). It is also difficult to draw conclusions about the locus or loci of adenosinergic modulation: it is likely that adenosine regulates locomotion in multiple regions of the nervous system involved in motor control, including the basal ganglia (Popoli et al. 1996) and ventral spinal cord (Acevedo et al. 2016; Witts et al. 2012), and the effects of adenosine blockade in one network may mask its effects in another.

Experiments in which in vitro rodent spinal cord preparations were exposed to adenosine-receptor agonists and antagonists have addressed the role of adenosine specifically within spinal locomotor networks. During locomotor-related activity induced by bath application of 5-HT, NMDA, and DA in neonatal mouse preparations, blockade of A₁ receptors results in increased burst frequency (Acevedo et al. 2016; Witts et al. 2012), with no effect on amplitude, whereas blockade of A_2A receptors has no effect (Acevedo et al. 2016; Witts et al. 2012). Conversely, bath application of adenosine or an A₁-selective agonist, but not an A_2A agonist, results in reduced burst frequency in a dose-dependent manner, again with no effect on amplitude (Acevedo et al. 2016; Witts et al. 2012). The frequency effects are associated with changes in burst and cycle duration, implying no change in duty cycle (Acevedo et al. 2016). Bath application of ATP has a similar effect to adenosine in reducing burst frequency (Witts et al. 2012); however, both blockade of adenosine receptors and application of ATP have no effect in the presence of an ectonucleotidase inhibitor. This indicates that endogenous adenosine is derived from ATP released into the extracellular space, that ATP itself does not modulate locomotor-related activity, and that both endogenous and exogenous ATP are efficiently degraded to adenosine within the murine spinal cord (Witts et al. 2012). The effects of bath-applied and endogenous adenosine therefore closely resemble those of ATP-adenosine released upon glial stimulation during network activity (Acton and Miles 2015).

Modulation of the frequency but not the amplitude of locomotor-related activity by adenosine suggests that it acts on the premotor circuitry rather than on motoneurons (Miles and Sillar 2011). Consistent with this hypothesis, neither exogenous adenosine nor adenosine released following PAR1 activation modulates disinhibited activity mediated by motoneurons and excitatory components of the locomotor circuitry alone (Acton and Miles 2015; Witts et al. 2012), implying that adenosine exerts its depressive effects via inhibitory interneurons, which may include the V1 population, for example (Fig. 3A). Ablation or inhibition of V1 interneurons reduces the frequency of locomotor-related activity, similar to the effects of adenosine acting at inhibitory A₁ receptors (Gosgnach et al. 2006). However, disinhibited activity, which is generated by spinal cord preparations following application of GABA_A and glycine receptor antagonists, is produced by a modified network, and caution must therefore be applied when comparing the effects of adenosine under disinhibited conditions and during fictive locomotion.

Fig. 3.

Schematics illustrating proposed release of ATP-adenosine from astrocytes and inhibition of D₁-like receptor signaling within locomotor networks. A: outline of the spinal locomotor CPG as a bilateral network comprising flexor and extensor and rhythm- and pattern-generating modules for the coordination of muscle activation during locomotion. Molecularly defined populations of postmitotic ventral horn neurons (V1, V2a, V2b, V0v, V0d, V3) are represented as blue circles, and some of their proposed interactions are indicated by arrows, signifying excitation, or bars, signifying inhibition. Astrocytes are proposed to modulate the rhythm-generating circuity by exerting an inhibitory effect, via secretion of ATP-adenosine, on an inhibitory population of interneurons that regulates locomotor frequency, possibly the V1 population (see text). For details of the roles of the neuronal populations indicated, see Goulding (2009) and Kiehn (2016). B: ATP is released from putative spinal cord astrocytes during network activity in response to an unidentified neuronal signal and upon experimental activation of the Gα_q-linked receptor PAR1 by TFLLR, which is proposed to mimic the endogenous action of neurotransmitters on astrocytic GPCRs. Extracellular ectonucleotidases mediate the hydrolysis of ATP to adenosine, which activates neuronal Gα_i-linked A₁ receptors to inhibit signaling through Gα_s-linked D₁-like receptors at the level of adenylyl cyclase. Reduced cAMP production by adenylyl cyclase results in reduced activation of PKA. PKA acts on unidentified targets, perhaps including ion channels and ionotropic receptors, to increase neuronal excitability; PKA inhibition results in a reduced frequency of locomotor-related activity. IN, interneuron; MN, motoneuron.

The effects of bath-applied adenosine on neurons in acute slices from postnatal mice have also been assessed. Adenosine modulates a broad population of ventral horn interneurons, representing both excitatory and inhibitory populations. In these cells, adenosine acts via presynaptic A₁ receptors to reduce the frequency of miniature postsynaptic currents (mPSCs; Witts et al. 2015) and the amplitude of evoked excitatory postsynaptic currents (eEPSCs) in a paired simulation protocol (Carlsen and Perrier 2014). Thus adenosine acting at A₁ receptors inhibits synaptic transmission onto interneurons. In addition, A₁-receptor activation induces a hyperpolarizing current in ventral horn interneurons (Witts et al. 2015). By contrast, adenosine induces a depolarizing current in motoneurons by an unknown mechanism that is sensitive to TTX, although adenosine does not appear to directly modulate synaptic inputs onto motoneurons (Witts et al. 2015). Interestingly, A₁-receptor blockade alone has no effect in recordings from neurons within spinal cord slices. It is possible that there is insufficient adenosine present in slices for the tonic activation of these receptors or that adenosine at the concentrations at which it exists in the spinal cord requires concurrent activation of D₁-like receptors to exert modulatory effects (Acevedo et al. 2016; see below).

Data suggest differences in purinergic modulation between rats and mice. In isolated rat spinal cord preparations, bath-applied adenosine enhances burst amplitude during pharmacologically induced locomotor-related activity and depresses the frequency of disinhibited bursting in an A₁-dependent manner (Taccola et al. 2012), effects not observed in mouse preparations (Acevedo et al. 2016; Witts et al. 2012). In addition, the duration of bouts of locomotor-related activity induced by dorsal root stimulation in rat spinal cord preparations is reduced by exogenous adenosine (Taccola et al. 2012). A₁ blockade alone has no effect on drug-induced or electrically stimulated locomotor-related activity in rats (Taccola et al. 2012), whereas in mice, A₁ blockade increases the frequency of fictive locomotion (Acevedo et al. 2016; Witts et al. 2012). Differences between mice and rats in purinergic modulation of respiratory networks have been attributed to species-specific ectonucleotidase expression (Huxtable et al. 2009; Zwicker et al. 2011), which results in the mouse network being more strongly modulated by adenosine, and the rat network, by ATP. Further experiments are required to assess whether similar species differences account for the reported differences in the sensitivity of rat and mouse spinal locomotor networks to adenosine. Differential responses to adenosine may also relate to differing roles of coactivation of D₁-like receptors (Acevedo et al. 2016), but the potential role of D₁/A₁ coactivation remains to be investigated in the rat spinal cord.

METAMODULATION OF D₁-LIKE DOPAMINE RECEPTORS BY A₁ RECEPTORS

Recent data suggest that endogenous adenosine exerts its effects on locomotor-related activity via second-order modulation of dopamine signaling within the mouse spinal cord (Acevedo et al. 2016). Dopamine is a potent modulator of spinal motor circuitry, acting via diverse mechanisms to regulate motor behavior (Miles and Sillar 2011; Sharples et al. 2014). All segments of the mammalian spinal cord receive dopaminergic or l-DOPAergic inputs from descending fibers originating in the brain, most notably from the A11 region of the hypothalamus (Björklund and Skagerberg 1979; Commissiong et al. 1979; Fleetwood-Walker et al. 1988; Hökfelt et al. 1979; Koblinger et al. 2014; Skagerberg and Lindvall 1985), and dopamine levels increase within the ventral horn during locomotion (Gerin et al. 1995; Gerin and Privat 1998).

The five dopamine-receptor subtypes fall into two families: the D₁-like receptors are the D₁ and D₅ subtypes, and the D₂-like receptors are the D₂, D₃, and D₄ subtypes (Neve 2010; Pieper et al. 2011). Whereas D₁-like receptors signal through Gα_q to stimulate phospholipase C and through Gα_s to stimulate adenylyl cyclase, D₂-like receptors signal through Gα_i to inhibit adenylyl cyclase (Abdel-Majid et al. 1998; Pieper et al. 2011). Adenylyl cyclase catalyzes the synthesis of the second messenger cyclic adenosine monophosphate (cAMP), which activates various proteins including PKA. PKA regulates a number of proteins, including Na⁺-dependent ion transporters, various ion channels, cAMP-responsive element binding protein 1 (CREB1), and dopamine and cyclic AMP-regulated phosphoprotein of 32 kDa (DARPP-32) (Abdel-Majid et al. 1998; Undieh 2010). Dopaminergic modulation of locomotor-related activity in spinal networks is mediated by both D₁-like and D₂-like receptors (Humphreys and Whelan 2012; Madriaga et al. 2004; Sharples et al. 2015).

A₁-adenosine receptors are reported to interact with D₁-like dopamine receptors in the spinal locomotor circuitry (Acevedo et al. 2016), as is observed in the basal ganglia (Popoli et al. 1996), where they colocalize (Ferré et al. 1992). Dopamine stabilizes locomotor-related activity (Barrière et al. 2004; Humphreys and Whelan 2012; Jiang et al. 1999; Madriaga et al. 2004; Sharples et al. 2015; Whelan et al. 2000) and is coapplied with 5-HT and NMDA to induce locomotor-related activity in many studies. Unless exogenously applied, it is presumed to be absent at functionally relevant concentrations in isolated lumbar spinal cord preparations because descending dopaminergic inputs are severed in in vitro preparations, and dopamine receptor antagonists do not alter locomotor-related activity induced in the absence of a dopamine receptor agonist (Barrière et al. 2004). When dopamine is absent or when D₁-like receptors are selectively blocked, A₁ blockade no longer alters the frequency of locomotor-related bursting (Acevedo et al. 2016). A₁ blockade is similarly ineffective when the Gα_s signaling pathway through which D₁-like receptors signal is inhibited at the level of PKA. However, when forskolin is applied to activate adenylyl cyclase in a receptor-independent manner, the effect of A₁ blockade on the frequency of locomotor-related bursting is restored.

Collectively, data from isolated spinal cord preparations suggest a model in which astrocytes secrete ATP during locomotor-related activity in an activity-dependent manner. Adenosine produced by hydrolysis of this ATP acts through A₁ receptors to regulate the activity of locomotor networks by inhibiting ongoing signaling mediated by D₁-like receptors and PKA (Fig. 3B). This may occur through direct inhibition of adenylyl cyclase mediated by the Gα_i pathway to which A₁ receptors are coupled, and it is possible that A₁ and D₁-like receptors form heterodimers (Franco et al. 2000); however, colocalization of A₁ and D₁-like receptors has not yet been demonstrated in the spinal cord.

It is unclear how these results relate to experiments in which adenosine was found to modulate neuronal activity in the absence of dopamine (Carlsen and Perrier 2014; Lloyd et al. 1988; Otsuguro et al. 2006, 2009, 2011; Taccola et al. 2012; Witts et al. 2015). It is possible that high concentrations of adenosine, such as those used for bath applications or produced during hypoxia and hypocapnia, modulate neural activity independently of dopamine, whereas low levels of endogenous adenosine require coactivation of D₁-like receptors by dopamine. Consistent with this, A₁ blockade has not been shown to modulate neuronal activity in the absence of dopamine (Acevedo et al. 2016; Taccola et al. 2012; Witts et al. 2012), although it does inhibit effects mediated by exogenous adenosine or adenosine released in response to a stimulus (Lloyd et al. 1988; Otsuguro et al. 2006, 2009, 2011; Taccola et al. 2012; Witts et al. 2015); presumably, this indicates low basal levels of adenosine within the spinal cord. Interestingly, all dopamine-dependent and dopamine-independent effects so far described in in vitro spinal cord preparations from healthy animals are mediated by A₁ receptors, despite expression of A_2A receptors in the spinal cord. It is conceivable that A_2A receptors have a functional role only in responses to injury: following spinal cord injury, A_2A receptors are upregulated in the spinal cord, and administration of a selective A_2A antagonist exerts neuroprotective effects, perhaps by reducing excitotoxicity (Paterniti et al. 2011). Thus A₁ receptors may participate in multiple signaling pathways, depending on agonist concentration.

The proposed antagonistic interaction between A₁ and D₁-like receptors implies that D₁-like receptors, which are typically excitatory, enhance the frequency of locomotor-related bursting in neonatal mice. However, this has not been clearly demonstrated. Although D₁-like receptors typically have excitatory effects, they have not been reported to increase the frequency of ongoing locomotor-related activity in mice (Humphreys and Whelan 2012; Sharples et al. 2015), but may instead enhance the stability of rhythmic bursting (Sharples et al. 2015). Bath-applied dopamine or a D₁-like receptor agonist stimulates locomotor-related activity in neonatal rat preparations (Barrière et al. 2004; Kiehn and Kjaerulff 1996) but not in neonatal mouse preparations (Jiang et al. 1999; Sharples et al. 2015; Whelan et al. 2000); however, the selective D₁-like receptor agonist SKF-81927 is able to stimulate locomotor activity in spinalized adult mice (Lapointe et al. 2009). Thus, although D₁-like receptors have excitatory effects within the spinal cord, the enhancement of burst frequency that A₁ receptors are proposed to counteract in spinal cords from neonatal mice has not been detected.

The effects of dopamine on network activity likely reflect multiple effects mediated by both D₁-like and D₂-like signaling pathways within diverse cell populations: dopamine has been shown to modulate synaptic strength and passive electrical properties of motoneurons and interneurons (Barrière et al. 2008; Carp and Anderson 1982; Clemens and Hochman 2004; Han et al. 2007; Han and Whelan 2009; Humphreys and Whelan 2012; Maitra et al. 1993; Seth et al. 1993). The extent of second-order modulation of D₁-like receptors by adenosine is currently unclear: further experiments are required to reveal whether adenosine acts on all cells expressing D₁-like receptors or only on a subset involved in the regulation of speed. Nevertheless, the model by which adenosine is proposed to modulate the effects of D₁-like receptor activation during fictive locomotion is consistent with other examples of metamodulation, which is proposed as an efficient mechanism for refining the effects of a first-order neuromodulator; this may be important when the first-order modulator is widespread and/or is capable of acting promiscuously on various cell types to exert diverse effects within a network (Katz 1999). Metamodulation may entail a second-order neuromodulator regulating the availability (release or uptake) of a first-order neuromodulator or, as in the present example, modulation of its effect on a target neuron. Other examples of metamodulation in spinal locomotor networks are provided by Xenopus tadpoles, in which nitric oxide (NO) modulates the release of norepinephrine (McLean and Sillar 2004), and lampreys, in which NO modulates the activity of endocannabinoids (Song et al. 2012).

THE PHYSIOLOGICAL IMPORTANCE OF GLIAL ADENOSINE IN SPINAL LOCOMOTOR NETWORKS

Adenosinergic neuromodulation is proposed to have a homeostatic role in diverse systems, preventing metabolic exhaustion and excitotoxicity (Cunha 2001; Fredholm et al. 2005; Wall and Dale 2008). Activity-dependent production of adenosine has been detected in the mammalian brain and brain stem and in the spinal cord of Xenopus tadpoles (Wall and Dale 2008). In some cases, this may be coupled to the degradation of ATP. ATP is consumed by neurons as a source of energy during activity, and adenosine produced by its hydrolysis may be released directly via neuronal equilibrative nucleoside transporters. Subsequent autocrine inhibition of activity via A₁ receptors is proposed as a mechanism to prevent metabolic exhaustion (Cunha 2001; Fredholm et al. 2005). However, this mechanism does not appear to operate in systems, including spinal motor networks in tadpoles and mice, in which the source of adenosine is ATP released into the extracellular space from either neurons or astrocytes (Carlsen and Perrier 2014; Panatier et al. 2011; Pascual et al. 2005; Serrano et al. 2006; Wall and Dale 2008; Witts et al. 2012). Nevertheless, it remains possible that adenosine functions to avert metabolic exhaustion, even if its source is not ATP expended as energy.

Xenopus tadpoles provide an example of purinergic signaling during motor behavior in a nonmammalian vertebrate. ATP released within the spinal cord during episodes of swimming excites the locomotor CPG, extending the duration of swimming bouts (Dale 1998; Dale and Gilday 1996). As swimming progresses, ATP is hydrolyzed to adenosine, which activates A₁ receptors to drive down network activity (Brown and Dale 2000). This mechanism may exist to prevent metabolic exhaustion and excitotoxicity (Wall and Dale 2008). The cellular origin of ATP in the tadpole spinal cord is unclear, and given recent evidence from Drosophila larvae that neuron-glia cross talk regulates behavior of nonmammalian species (Ma et al. 2016), glial release of purines should not be ruled out. The temporal dynamics of adenosine release in the mammalian spinal cord have not been investigated; it is possible that gradual increase in adenosinergic tone during motor behaviors could also have a protective function by preventing metabolic exhaustion.

SUMMARY

Recent studies have provided evidence for the release of ATP-adenosine from astrocytes in the spinal cord, as previously demonstrated elsewhere in the CNS. Adenosine operates by multiple pathways within the spinal motor circuitry. During fictive locomotion, it is proposed to function as a second-order modulator of D₁-like receptors (Acevedo et al. 2016). This mechanism is likely to be important in refining the influence of dopamine, which has manifold cellular activities during locomotor behaviors. The consequent reduction in locomotor speed may be important in ensuring fluency of movement or have a role in averting cellular fatigue (Acevedo et al. 2016; Acton and Miles 2015; Witts et al. 2012). Adenosine also acts by a dopamine-independent pathway or pathways (Carlsen and Perrier 2014; Lloyd et al. 1988; Otsuguro et al. 2006, 2009, 2011; Taccola et al. 2012; Witts et al. 2012), which may be activated only by high concentrations of adenosine. It will be interesting to compare tissue concentrations of adenosine during inactivity, locomotion, and injurious conditions in which adenosine is released, such as hypoxia and hypercapnia. The agreement between dopamine-dependent effects mediated by A₁-adenosine receptors during fictive locomotion and the effects of adenosine released upon glial stimulation suggests that the latter acts via D₁-like receptors to modulate rhythmic activity, but this remains to be tested directly.

Although evidence exists for glial release of adenosine (Acton and Miles 2015; Carlsen and Perrier 2014; Witts et al. 2012), detailed Ca²⁺ imaging of astrocytes is required to confirm a role for Ca²⁺-dependent neuron-glia signaling in the ventral horn. However, it should also be noted that several studies have provided evidence for cross talk between neurons and astrocytes in a manner that departs from the established tripartite synapse model (Angelova et al. 2015; Gourine et al. 2010; Gourine and Kasparov 2011; McDougal et al. 2013; Torres et al. 2012). Neurotransmitter uptake is as important as release in shaping patterns of synaptic transmission and may be regulated in an activity-dependent manner independently of Ca²⁺ signaling (Al Awabdh et al. 2016; Attwell et al. 1993; Li et al. 2009; Marcaggi and Attwell 2004; Perego et al. 2000; Roux and Supplisson 2000; Shibasaki et al. 2017). Although PAR1 activation does not appear to elicit release of glutamate or its coagonists at NMDA receptors, glycine and d-serine (Acton and Miles 2015), it has been proposed that coagonist concentrations are regulated by astrocytes in an activity-dependent manner during motor behaviors (Acton and Miles 2017). In addition, activation of astrocytic GPCRs may result in the modulation of neuronal activity by Ca²⁺-independent pathways. Selective activation of a Gα_q-coupled DREADD expressed by astrocytes lacking IP₃R2 receptors results in diverse behavioral modifications on a timescale longer than that predicted for Ca²⁺-dependent responses; these may involve signaling via Gβγ or PKC, for instance (Agulhon et al. 2013). These examples suggest that information processing by astrocytes is underexplored and may be crucial in the generation of behaviors by spinal networks, exceeding their established role as cellular housekeepers.

GRANTS

D. Acton was supported by funds from a Wellcome Trust Institutional Strategic Support Fund grant. G. B. Miles received support from Biotechnology and Biological Science Research Grant BB/M021793/1.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

D.A. prepared figures; D.A. drafted manuscript; D.A. and G.B.M. edited and revised manuscript; D.A. and G.B.M. approved final version of manuscript.