We now understand that the traditional view of astrocytes as merely supportive cells providing only structural and metabolic support to neurons is limited and studies of the last 20 years show that astrocytes exert a series of complex functions that go well beyond the uptake and recycle of neurotransmitters and the buffering of extracellular potassium (Volterra and Steinhauser 2004; Allen and Barres 2005; Volterra and Meldolesi 2005; Fellin et al. 2006b; Haydon and Carmignoto 2006; Iadecola and Nedergaard 2007). Morphologically, astrocytes are characterized by a highly ramified structure of thin processes with which they contact neurons, blood vessels and other astrocytes. Astrocyte–astrocyte contacts mediate gap junction coupling between adjacent glial cells to form a cellular network called the astrocytic ‘syncytium’ (Giaume and McCarthy 1996). Although astrocytes form a highly interconnected network, they structurally occupy separate domains with very limited or no overlap between the region occupied by one astrocyte and the regions occupied by its neighboring cells (Bushong et al. 2002; Halassa et al. 2007b) (Fig. 1b). Within the volume of one astrocytic domain, it has been estimated that a single glial cell can contact up to 105 synapses (Bushong et al. 2002; Halassa et al. 2007b) and hundreds of dendrites (Halassa et al. 2007b) (Fig. 1a). The contact between the astrocyte and the neuron is a highly dynamic structure (Hirrlinger et al. 2004; Haber et al. 2006) and the extent of astrocytic coverage of the neuronal terminals is activity-dependent (Genoud et al. 2006). This complex anatomical organization serves a number of different functional roles. For example, astrocytes display a form of excitability that is based on changes in the intracellular Ca2+ concentration (Cornell-Bell et al. 1990) and are capable of releasing a number of signaling molecules that have a complex effect on neuronal function (Haydon 2001; Newman 2003; Fellin and Carmignoto 2004). Although it has been shown that the release of chemicals from astrocytes influences a variety of brain processes such as the genesis (Christopherson et al. 2005) and stabilization of synapses (Ullian et al. 2001), the control of the vasculature tone (Zonta et al. 2003; Mulligan and MacVicar 2004; Metea and Newman 2006; Takano et al. 2006) and the modulation of neuronal excitability, many aspects of this astrocyte-to-neuron signaling are still incompletely understood (Agulhon et al. 2008). This review first highlights some recent results describing the properties of Ca2+ signaling in astrocytes in vivo and then discusses some examples of signaling between astrocytes and neurons focusing on the importance of astrocyte-to-neuron communication in the modulation of neuronal function at the level of single synapses and neuronal networks.
Neuromodulation is a fundamental process in the brain that regulates synaptic transmission, neuronal network activity and behavior. Emerging evidence demonstrates that astrocytes, a major population of glial cells in the brain, play previously unrecognized functions in neuronal modulation. Astrocytes can detect the level of neuronal activity and release chemical transmitters to influence neuronal function. For example, recent findings show that astrocytes play crucial roles in the control of Hebbian plasticity, the regulation of neuronal excitability and the induction of homeostatic plasticity. This review discusses the importance of astrocyte-to-neuron signaling in different aspects of neuronal function from the activity of single synapses to that of neuronal networks.
miniature excitatory postsynaptic current
metabotropic glutamate receptor
tumor necrosis factor α
Astrocytes: from supportive to signaling cells
For several decades, the inability of astrocytes to fire action potentials has supported the view of these glia as supportive, non-excitable cells. Some of the first evidence that astrocytes are signaling cells that can release chemical messengers to affect neuronal function was provided almost 20 years ago (Muller and Best 1989). In that study, the authors showed that injection of immature astrocytes into the visual cortex of adult cats in vivo reopens the window of ocular dominance (OD) plasticity. Since its discovery more than 40 years ago (Wiesel and Hubel 1963), OD plasticity has been a preferred model to investigate the cellular and molecular mechanism of plasticity in vivo. In higher mammals, the visual cortex is organized in OD columns consisting of neurons responding preferentially to one but not the other eye. If during a precise postnatal time period, called the critical period, one eye is deprived of vision, there is a dramatic change in the neuronal circuits underlying the OD columns because of the imbalance of visual inputs. OD columns responding to the deprived eye shrink while the OD columns responding to the non-deprived eye expand. Although visual deprivation performed after the critical period usually results in little plasticity, Muller and Best show that transplantation of immature astrocytes from newborn kittens into the visual cortex of adults cats is able to induce OD plasticity after the end of the critical period (Muller and Best 1989). Based on these results, the authors hypothesized that immature astrocytes release a permissive factor that allows OD plasticity to occur and suggested that the end of the critical period could coincide with the maturation of cortical astrocytes.
Although the molecule secreted by astrocytes that restores OD plasticity was not identified, this study suggested that astrocytes are capable of releasing neuroactive molecules and thus have the potential to be not only supportive but also signaling cells in the brain. Since then, a number of different investigations, performed mainly in vitro and in situ, have identified several molecules released by astrocytes in a process named ‘gliotransmission’, including, among others, glutamate (Parpura et al. 1994; Bezzi et al. 1998), ATP (Cotrina et al. 1998; Coco et al. 2003), d-serine (Mothet et al. 2000) and tumor necrosis factor α (TNFα) (Stellwagen and Malenka 2006). The study of the intracellular signaling pathways regulating the release of ‘gliotransmitters’ has revealed a central role of cytosolic Ca2+ concentration for some of these molecules. The use of fluorescent Ca2+ indicators demonstrated that astrocytes display spontaneous and neuronal activity-evoked oscillations in their intracellular Ca2+ concentration (Cornell-Bell et al. 1990; Charles et al. 1991) which can then trigger the release of gliotransmitters such as glutamate (Parpura et al. 1994), d-serine (Mothet et al. 2005) and ATP (Coco et al. 2003). These findings led to the proposal that astrocytes are excitable cells, yet their excitability is not based on changes in the voltage of their membrane but rather on changes in the intracellular concentration of the Ca2+ ion (Halassa et al. 2007a).
Ca2+ signaling in astrocytes: lesson from in vivo experiments
The introduction of two-photon microscopy has revolutionized fluorescence imaging permitting in vivo functional imaging with cellular resolution (Denk et al. 1990; Helmchen and Denk 2005) and allowing an appreciation of the complexity of the Ca2+ signaling dynamics of astrocytes in vivo. Using this technique, Hirase et al. first described the occurrence of Ca2+ signals in cortical astrocyte from living rats (Hirase et al. 2004). In anesthetized animals, more than 60% of the imaged astrocytes displayed a complex pattern of changes in the intracellular Ca2+ concentration, including brief (5–50 s) repetitive spikes and long-lasting (>50 s) plateaus. Plateaus and Ca2+ spikes occurred with a relatively low frequency (0.1 event/min) under basal conditions and showed a limited degree of correlation among nearby astrocytes (Hirase et al. 2004; Nimmerjahn et al. 2004). Under basal conditions and in the absence of sensory stimulation, Ca2+ signaling in astrocytes is thus organized in domains that largely coincide with the tiling organization of the structural domains (Fig. 2a and b).
In addition to spontaneous oscillations, in situ studies have demonstrated that neuronal activity can trigger Ca2+ elevations in astrocytes (Porter and McCarthy 1996; Pasti et al. 1997). The first evidence suggesting that this could happen in vivo was the observation that the application of the GABAA receptor antagonist bicuculline (Hirase et al. 2004) or picrotoxin (Gobel et al. 2007), which increases neuronal activity by triggering epileptic-like discharges, resulted in an increase in Ca2+ signaling in astrocytes. Additionally, these Ca2+ signals were correlated between pairs of nearby astrocytes (Hirase et al. 2004), suggesting that neuronal activity-driven oscillations may not be limited to single astrocytes but may occur synchronously in multiple cells. Subsequently, sensory stimulation was shown to increase Ca2+ signaling in astrocytes in anesthetized animals. In the mouse, whisker (Wang et al. 2006), limb (Winship et al. 2007) and odor stimulation (Petzold et al. 2008) causes Ca2+ transients in astrocytes in the barrel, the primary somatosensory cortex and the olfactory bulb respectively, while a recent study in ferrets (Schummers et al. 2008) demonstrates that visual stimulation triggers a large increase in astrocytic Ca2+ in the visual cortex. Typically activity-mediated Ca2+ signals in astrocytes are delayed by a few seconds in relation to neuronal responses (Wang et al. 2006; Schummers et al. 2008), although a subpopulation of glial cells (Winship et al. 2007) can respond as quickly as neurons. When whisker stimulation is used to evoke Ca2+ signals in cortical astrocytes, astrocytic Ca2+ responses peak at a whisker stimulation frequency that induces the maximal neuronal response (Wang et al. 2006), suggesting that the astrocytic response linearly correlates with the strength of the sensory stimulation. Astrocyte activation by sensory stimulation is extremely tuned and stimulus selective and display all the features typical of the neuronal response (Schummers et al. 2008): (i) astrocytes respond to visual stimuli over a comparable region of nearby neurons and the x–y two-dimensional receptive field and maximal response of astrocytes are similar to the neuronal ones; (ii) the Ca2+ response of astrocytes is a function of the orientation and spatial frequency of the visual stimulus and when compared to neuronal responses, astrocytes display sharper orientation and spatial frequency tuning; (iii) single cell-based orientation preference maps across the cortical surface show that small groups of cells (from 2 to >10) respond preferentially to one stimulus orientation and not to others demonstrating that Ca2+ signaling in astrocytes evoked by sensory stimulation is organized in functional domains that are not restricted to single cells but involve multiple astrocytes (Fig. 2c and d). Whether the astrocytes that form a functional domain are coupled with each other to form distinct syncytia or whether they are separately, but simultaneously activated by neuronal activity is still not known.
All the in vivo data discussed so far were obtained in anesthetized animals but confirmation that activity-dependent Ca2+ oscillations in astrocytes occur in behaving mice comes from a recent report (Dombeck et al. 2007). When head restrained mice were free to run on a spherical treadmill and simultaneous two-photon Ca2+ imaging was performed in the cortex, Dombeck et al. could observe behaviorally correlated Ca2+ signals in astrocytes. Large, repetitive Ca2+ signals in astrocytes are associated with the running behavior and are temporally correlated in multiple glial cells over a distance of almost 100 μm (Dombeck et al. 2007).
Taken together these results show that in different preparations from mice (Wang et al. 2006), to rats (Hirase et al. 2004) and ferrets (Schummers et al. 2008), in anesthetized as well as in behaving animals, astrocytes display a complex pattern of spontaneous and neuronal activity-driven Ca2+ signals. Spontaneous Ca2+ signals are mostly restricted to single astrocytes and occur independently of the activity of neighboring astrocytes (Hirase et al. 2004), while activity-mediated Ca2+ oscillations occur in groups of different astrocytes that respond to the same sensory stimulus (Schummers et al. 2008). Astrocytic Ca2+ signaling is thus organized in functional domains that, in the case of spontaneous oscillations, correspond to the non-overlapping anatomical domain (Fig. 2a and b). In contrast, in the case of activity-evoked signals, the functional domains of the Ca2+ response are larger than the structural domains and can involve several glial cells (Fig. 2c and d).
Different mechanisms underline the generation of spontaneous and activity-evoked Ca2+ signals in astrocytes
Astrocytes express a wide range of receptors for different neurotransmitters (Haydon 2001). In brain slice preparations, several lines of evidence demonstrate that the astrocytic Ca2+ response to excitatory neuronal activity is largely mediated by the activation of metabotropic glutamate receptors (mGluRs) (Pasti et al. 1997). In vivo, application of the mGluR agonist 3,5-dihydroxyphenylglycine (DHPG) or t-ACPD evokes robust Ca2+ transients in cortical astrocytes, while co-application of the mGluR1 antagonist LY367385 and the mGluR5 antagonist 2-methyl-6-(phenylethynyl)pyridine (MPEP) reduces the whisker-evoked Ca2+ response in astrocytes by more than 70% (Wang et al. 2006). 2-Methyl-6-(phenylethynyl)pyridine (MPEP) and LY 367385 alone are equally effective in inhibiting the whisker-evoked response (40–50% inhibition), suggesting that both mGluR1 and mGluR5 contribute to astrocyte activation (Wang et al. 2006). Iontophoretic application of the alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)/kainate receptor antagonist 6-Cyano-7-nitroquinoxaline-2,3-dione (CNQX) reduces the post-synaptic response with no effect on whisker-evoked Ca2+ oscillations in astrocytes. These results suggest that, in vivo, astrocytes can sense glutamate release at excitatory synapses which in turn activates metabotropic glutamate receptors on their plasma membrane to generate intracellular Ca2+ increases. Nonetheless, a recent study suggests a role for glutamate transporters in the regulation of astrocytic Ca2+ signaling (Schummers et al. 2008). Application of the glutamate transporter inhibitor threo-b-Benzyloxyaspartic acid (TBOA) in vivo drastically reduces Ca2+ signals in astrocytes evoked by visual stimulation, with a modest increase in the neuronal responses.
The activity-driven response in astrocytes is extremely sensitive to the level of neuronal activity: 1% increase in the isofluorane concentration causes a modest (16%) decrease in the neuronal response to visual stimulation but a dramatic (77%) decrease in the astrocytic response (Schummers et al. 2008). This finding suggests that there is a steep threshold for astrocyte activation and this has been proposed to account for the sharper tuning to orientation and spatial frequency of the astrocytic response compared to the neuronal one (Schummers et al. 2008).
Differently from sensory-evoked Ca2+ signals, spontaneous Ca2+ oscillations seem to be generated independently from neuronal activity (Takata and Hirase 2008). In vivo application of tetrodotoxin (TTX) to block neuronal action potential-dependent release at pre-synaptic terminals significantly reduces neuronal activity with no effect on spontaneous Ca2+ oscillations in astrocytes. At the same time, application of mGluR and purinergic receptor antagonists does not result in a reduction of spontaneous Ca2+ signals in astrocyte (Takata and Hirase 2008). These findings agree with previous studies in brain slice preparations (Nett et al. 2002), suggesting that spontaneous Ca2+ oscillations in astrocytes do not depend on the activation of glutamate and purinergic receptors, occur independently of neuronal activity and are probably generated by intrinsic mechanisms.
Altogether these in vivo data demonstrate that Ca2+ signaling in astrocytes is a complex and sophisticated phenomenon. It consists of a complicated pattern of spikes and plateaus and has two principal modes of operation: under basal conditions, when no sensory stimulation is present, sporadic Ca2+ signals occur mainly in single astrocytes and independently of neuronal activity (Hirase et al. 2004; Takata and Hirase 2008). In contrast, in response to sensory stimulations, astrocytes display increased Ca2+ signaling that involves multiple glial cells (Schummers et al. 2008) and is generated by glutamate released at the synapse that activates receptors (Wang et al. 2006) and transporters (Schummers et al. 2008) on the astrocytic membrane (Fig. 3a). While recent evidence shows that these astrocytic Ca2+ signals are important for the modulation of the arteriole diameter (Takano et al. 2006) and the regulation of the hemodynamic response that generates the intrinsic optical signal (Schummers et al. 2008), the following paragraphs will focus on the importance of Ca2+ signaling in astrocytes in the control of neuronal excitability and plasticity at the level of single synapses and neuronal networks.
Gliotransmission: the release of chemical transmitters from astrocytes
Since the observation that transplantation of immature astrocytes in the visual cortex of adult cats reopens the critical period for OD plasticity (Muller and Best 1989), several studies have shown that astrocytes can release a variety of different molecules that can have complex effects on neuronal function (Parpura et al. 1994; Araque et al. 1998; Kang et al. 1998). These studies led to the introduction of the concept of gliotransmission and of the so-called ‘tripartite synapse’ (Araque et al. 1999) in which astrocytes are considered together with the pre- and post-synaptic terminals a third, signaling element at the synapse. For example, astrocytes release glutamate which acts at the pre- (Fiacco and McCarthy 2004; Jourdain et al. 2007; Perea and Araque 2007) and post-synaptic level (Parri et al. 2001; Angulo et al. 2004; Fellin et al. 2004; Perea and Araque 2005) to modulate synaptic transmission and neuronal excitability at both excitatory (Bezzi et al. 1998; Pasti et al. 2001) and inhibitory (Kang et al. 1998; Liu et al. 2004) synapses, while ATP released from astrocytes modulates neuronal excitability through the activation of purinergic ionotropic receptors (Gordon et al. 2005). At the same time, after degradation to adenosine, ATP released from astrocytes is responsible for activity-dependent heterosynaptic depression at excitatory synapses (Pascual et al. 2005; Serrano et al. 2006).
It is important to note that a link between gliotransmission and Ca2+ signaling in astrocytes has been described only for some of the molecules released by these glia. While Ca2+-dependent release has been proven for glutamate (Parpura et al. 1994; Bezzi et al. 1998), ATP (Coco et al. 2003) and d-serine (Mothet et al. 2005), for many other molecules the Ca2+-dependency has not been proven or not tested yet. Furthermore, different routes of release for the same gliotransmitter can coexist in astrocytes (Fellin and Carmignoto 2004). For example, multiple mechanisms of release that are independent on Ca2+ mobilization have been described for glutamate (Kimelberg et al. 1990; Duan et al. 2003; Ye et al. 2003) and ATP (Cotrina et al. 1998). Despite the different mechanisms that govern the release of gliotransmitters, all these studies support the view that astrocytes by releasing chemical transmitters play a fundamental role in the modulation of synaptic transmission and neuronal function.
Coincidence detection of astrocytic and neuronal signals at single synapses
Using Ca2+ uncaging together with minimal stimulation to activate single synapses, Perea and Araque show that astrocytes control synaptic transmission at single hippocampal synapses (Perea and Araque 2007). UV uncaging of Ca2+ in single astrocytes causes a transient increase in synaptic transmission which is due to the pre-synaptic increase in the probability of release with no change in synaptic potency. Astrocyte-induced change in synaptic transmission is transient, lasts for less than a minute and is generated by glutamate released from astrocytes activating pre-synaptic mGluRs. Most importantly, the transient increase in synaptic transmission becomes long-lasting when photolytic stimulation of astrocytes is associated with a neuronal post-synaptic depolarization. Indeed, mild depolarizations of the post-synaptic neurons to −30 mV, that do not modify synaptic transmission per se, lead to long-term potentiation (LTP) of synaptic transmission when astrocytes are photostimulated (Perea and Araque 2007). This elegant study demonstrates the importance and complexity of astrocytic function at the tripartite synapse. Ca2+-dependent glutamate release from astrocytes causes a transient potentiation in synaptic transmission but when this glial signal is coupled to a post-synaptic depolarization it leads to a previously unrecognized form of LTP. Whether this astrocyte-mediated form of LTP occurs during spontaneous and activity-driven Ca2+ oscillations in astrocytes and whether this phenomenon is important for learning and memory is still unknown. These findings, together with the observation that Ca2+ signaling in astrocytes is able to sense the level of neuronal activity (Pasti et al. 1997) and to integrate signals from different synaptic pathways (Perea and Araque 2005), suggest that astrocyte are not only signaling cells, but may also contribute to integration and processing of information in the brain (Fig. 3b).
Glutamate release from astrocytes influences network activity
Spontaneous Ca2+ oscillations in astrocytes occur in vitro (Nett et al. 2002) and in vivo (Takata and Hirase 2008) independently from neuronal activity and their function is still not clear. The observation that spontaneous Ca2+ spikes in astrocytes correlates in thalamic neurons with the detection of slow inward currents (SICs) mediated by the NMDA receptor (Parri et al. 2001), suggests that these spontaneous Ca2+ signals mediate the release of glutamate from astrocytes to modulate neuronal excitability. Since then, a number of studies have investigated the nature of these slow inward currents and showed that different protocols that stimulate Ca2+ signaling in astrocytes also increase the frequency of observation of SICs in neurons. Pharmacological stimulation (Angulo et al. 2004; Kang et al. 2005; Fellin et al. 2006c; Kozlov et al. 2006; Nestor et al. 2007; Navarrete and Araque 2008; Shigetomi et al. 2008), single astrocyte stimulation (Fellin et al. 2004; Perea and Araque 2005; D’Ascenzo et al. 2007), single neuron depolarization (Navarrete and Araque 2008) and extracellular stimulation of neuronal afferents (Fellin et al. 2004; D’Ascenzo et al. 2007), all resulted in Ca2+ elevations in astrocytes and detection of SICs in nearby neurons. Furthermore, these studies demonstrated that glutamate released from glia generates SICs by activating a particular set of NMDA receptors located at extrasynaptic sites and containing preferentially the NR2B subunit (Fellin et al. 2004). When recorded in current-clamp, SICs can cause suprathreshold depolarization and thus trigger action potential firing in neurons (Fellin et al. 2006a). Most importantly, SICs occur with a high degree of synchronization in different neurons (Angulo et al. 2004; Fellin et al. 2004; Kozlov et al. 2006). When two neurons were simultaneously recorded in voltage clamp and were <100 μm apart, SICs occur with a high temporal correlation in the two cells, while no synchronous SICs were recorded when the two neurons were more than 100 μm apart. These electrophysiological results are confirmed by confocal Ca2+ imaging experiments showing that SICs occur synchronously in small groups (from 2 to more than 10) of contiguous neurons (Fellin et al. 2004). Several studies and different groups described these currents in various brain regions including the thalamus (Parri et al. 2001), the cortex (Ding et al. 2007), the hippocampus (Angulo et al. 2004; Fellin et al. 2004; Cavelier and Attwell 2005; Kang et al. 2005; Perea and Araque 2005; Nestor et al. 2007; Shigetomi et al. 2008), the nucleus accumbens (D’Ascenzo et al. 2007) and the olfactory bulb (Kozlov et al. 2006). While it could be argued that the majority of these studies were performed using artificial means to stimulate Ca2+ signaling into astrocytes (flash photolysis, pharmacological stimulation), it must be noted anyway that SICs were first described to be correlated with endogenous, spontaneous oscillations in thalamic slices (Parri et al. 2001).
A recent study generated transgenic mice in which a Gq-coupled receptor is expressed selectively into astrocytes and is activated specifically by an exogenous ligand (Fiacco et al. 2007). Using this approach, the authors could reliably induce long-lasting (>1 min) Ca2+ plateaus in the majority of hippocampal astrocytes, but failed to detect glutamate release from astrocytes and, as a consequence, SICs in neurons. Nevertheless, when the authors generate faster (∼10–20 s) Ca2+ transient in astrocytes by means of IP3 uncaging they could induce glutamate release from astrocytes and measure its effect on neuronal excitability (Fiacco et al. 2007), suggesting that the duration of Ca2+ oscillations may be crucial to determine or not the release of glutamate from these glia. A very recent study (Shigetomi et al. 2008) confirms and expands this hypothesis. When agonists for P2Y1 and PAR-1 receptors are applied to the slice, they have no significant effect on the membrane properties of neurons but generate Ca2+ signals in astrocytes with significantly different time courses and kinetics. Electrophysiological recordings reveal that only activation of PAR-1 receptors and not of P2Y1 receptors is associated with glutamate release and SICs generation in neurons, demonstrating that Ca2+ signaling in astrocytes is not an all-or-none phenomenon but that the duration and kinetics of the Ca2+ transients in astrocytes are key factors controlling the release of gliotransmitters (Shigetomi et al. 2008). Moreover, a different study confirms that activation of PAR-1 receptors triggers glutamate release from astrocytes which depolarizes neurons by activating NMDA receptors, but shows that PAR-1 receptor-dependent glutamate release from astrocytes causes a tonic depolarization and no SICs in neurons (Lee et al. 2007). These findings raise the potential that there are yet unknown mechanisms that control the quantity of glutamate released or the mechanism of Ca2+-dependent release from astrocytes.
Altogether these studies show that Ca2+ signaling in astrocytes is a complex phenomenon and that Ca2+ increases of specific kinetics and duration result in glutamate release from astrocytes and the generation of SICs in nearby neurons. Moreover, by activating NR2B-containing NMDA receptors, glutamate release from glia can generate synchronous depolarization and action potential firing in small groups of contiguous neurons. Recently, it has been suggested that this form of astrocyte-to-neuron communication could act as a source of runaway excitation in seizure pathologies and a few studies have been performed to address this hypothesis (Kang et al. 2005; Tian et al. 2005; Fellin et al. 2006a). While Ca2+ oscillations in astrocytes and, as a consequence, SICs in neurons are found to be increased in a number of in vitro models of epilepsy (Tian et al. 2005; Fellin et al. 2006a), pharmacological inhibition of SICs does not prevent the generation of epileptiform activity but significantly attenuates the ictal, seizure-like event (Fellin et al. 2006a). These data suggest that glutamate released from astrocytes, although not necessary for the generation of epileptiform activity, may act as a feed-forward mechanism to enhance neuronal excitation during seizure pathologies (Fellin and Haydon 2005; Tian et al. 2005). Besides their role in the modulation of neuronal excitability, it has been suggested that SICs, by activating NR2B-containing NMDA receptors, could also generate signaling messages for cellular death (Carmignoto and Fellin 2006). Indeed, the selective activation of NR2B-containing NMDA receptors is linked to the activation of cellular death pathways (Hardingham et al. 2002), raising the possibility that under specific conditions astrocytes could modulate this signaling pathway. A recent study confirms this hypothesis and demonstrates that pharmacological attenuation of this astrocytes-to-neuron signaling pathway provides significant protection from the neuronal death that follows a brain insult such as status epilepticus (Ding et al. 2007).
TNFα released by astrocytes mediates synaptic scaling
When neurons are subjected to prolonged periods of altered synaptic activity, they scale their output to compensate for the global change in network activity (Davis 2006; Rich and Wenner 2007). All synapses are potentiated or depressed proportionally and all synapses are scaled independently from their initial strength. This form of plasticity tends to stabilize the neuronal network in face of a perturbation and is called homeostatic plasticity to distinguish it from Hebbian, synapse-specific forms of plasticity such as LTP (Turrigiano and Nelson 2004; Turrigiano 2007). A typical protocol consists of blocking neuronal activity by prolonged incubation of neuronal cultures with tetrodotoxin and then measuring the effect of this treatment on miniature excitatory post-synaptic currents (mEPSCs) (Turrigiano et al. 1998). Under these conditions, the amplitude distribution of mEPSCs is scaled up and for this reason this form of homeostatic plasticity is usually named ‘synaptic scaling’. In cortical and hippocampal pyramidal cells, one mechanism by which synapses adapt to the overall decreased activity and become stronger is by increasing the number of alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors in the post-synaptic membrane (Rich and Wenner 2007; Turrigiano 2007). A recent study demonstrates that the release of the cytokine TNFα from astrocytes contributes to this form of plasticity (Stellwagen and Malenka 2006). By combining the use of transgenic mice with a pharmacological approach, the authors show that TNFα mediates the increase in synaptic strength that follows a prolonged incubation of the neuronal culture with TTX. In contrast, TNFα signaling is not required for other forms of plasticity such as LTP and long-term depression. Using a clever approach utilizing mixed cultures from wild-type (WT) and TNFα knock-out mice, the authors demonstrate that astrocytes are the primary source of this pro-inflammatory cytokine (Stellwagen and Malenka 2006). When neurons from WT or TNFα knock-out mice are plated with WT astrocytes the increase in mEPSCs amplitude in response to TTX incubation is normally observed. However, when WT neurons are plated together with astrocytes from TNFα knock-out mice, no homeostatic increase in mEPSCs amplitude is observed. Most importantly, astrocytic TNFα signaling is involved only in the scale-up of synapses following TTX treatment, while it has no role in the scaling down of synapses that follows a prolonged increase in neuronal activity (Stellwagen and Malenka 2006).
These results demonstrate that the gliotransmsitter TNFα is the mediator of the specific form of synaptic scaling that occurs in response to prolonged blockade of neuronal activity. Glia are thus one of the sensors for homeostatic regulation and are crucial modulators of neuronal network activity (Fig. 3b). Nevertheless, a number of exciting questions are still to be addressed. For example, a recent report shows that TNFα-mediated synaptic scaling is involved in OD plasticity (Kaneko et al. 2008). After monocular deprivation, neurons in the binocular region of the visual cortex decrease their response to the closed eye and increase their responsiveness to the open eye. Using transgenic mice, this study suggests that the increase in the open eye response is a homeostatic process mediated by TNFα (Kaneko et al. 2008). Given that astrocytes are a major source of TNFα, that they respond to visual stimulation and that they are able to secrete permissive factors for OD plasticity, these results suggest that astrocytes have the potential to be key elements in the control of neuronal network function and plasticity in vivo. Future experiments will investigate how astrocytes sense the global change in network activity, which are the molecular mechanisms controlling the release of TNFα and whether other gliotransmitters are involved in different forms of homeostatic plasticity.
Gliotransmission: understanding space and time
Given that gliotransmission modulates various aspects of the neuronal function, understanding where it occurs and the position of the neuronal receptors involved in this process is of fundamental importance. It is usually assumed that the release of chemical transmitters from glia occurs at the process of the astrocyte contacting the synapse but where exactly the sites of release are and whether different gliotransmitters have different release locations is not clear. Structural evidences support the view that for some transmitters the release may occur close to the synapse since glutamate-containing vesicles have been identified in the astrocytic process facing synaptic structures (Bezzi et al. 2004). Moreover, functional evidence provided by Oliet and coworkers in the supraoptic nucleus of the hypothalamus suggests that the astrocytic process contains the site of release for d-serine (Panatier et al. 2006). Under normal conditions in this brain region astrocytic processes enwrap neurons, but during lactation the astrocytic morphology is drastically changed and astrocytic processes are completely withdrawn. Using this experimental model, the authors were able to demonstrate that the astrocytic process releases d-serine into the synaptic cleft to modulate synaptic but not extrasynaptic NMDA receptors (Panatier et al. 2006). On the other hand, different studies demonstrate that gliotransmission impact receptors located at extrasynaptic or perisynaptic sites with little or no effect on synaptic receptors (Araque et al. 1998; Fellin et al. 2004), raising the possibility that different transmitters could be released at different locations. Given that we know very little about the spatial locations where gliotransmitters are released, it should be remembered that astrocytes contact neurons not only at synapses, but also at cell bodies (Peters et al. 1991; Volterra et al. 2002) and thus it cannot be excluded that part of the signaling between astrocytic and neuronal cells occurs at these latter locations.
As the coincidence detection of the neuronal and astrocytic input determines the outcome of synaptic plasticity (Perea and Araque 2007), understanding the timing of gliotransmission with respect to neurotransmission is of equal importance. For example, in vivo and in situ we have very limited information about how in the extracellular space the concentration profile of most gliotransmitters varies with time, whether astrocytes are constantly releasing these molecules, or whether the release is strictly dependent on the neuronal activity. Nevertheless, for a few gliotransmitters, as for example, ATP, we do know that astrocytes release it both tonically and in an activity-dependent way. Constant release of ATP that is rapidly degraded to adenosine provides a tonic suppression at hippocampal synapses, while activity-mediated ATP release from astrocytes is responsible for heterosynaptic depression (Pascual et al. 2005).
Is gliotransmission organized in domains?
If we restrict our discussion to Ca2+-dependent gliotransmission, where we can monitor Ca2+ which is the stimulus for transmitter release, the observation that Ca2+ signaling in astrocytes is organized in functional domains (Fig. 2) would support the hypothesis of a domain organization for gliotransmission. Moreover, the existence of at least two different modes of Ca2+ signaling, spontaneous and activity-evoked, with distinct spatial patterns suggest that gliotransmission evoked by spontaneous or activity-driven oscillations may impact different spatial regions and potentially have different functional roles (Fig. 4). Given that spontaneous Ca2+ signaling in astrocytes displays a low level of correlation among different glial cells, one can envision that intrinsic spontaneous oscillations would cause gliotransmission to impact only the neuronal structures within the domain of a single astrocyte (Fig. 4a). In contrast, gliotransmission triggered by activity-mediated Ca2+ signals would be correlated in many more cells (2 to more than 10 different glial cells, Fig. 2). The number of neurons that can be influenced by activity-evoked gliotransmission may thus be up to one order of magnitude higher than that influenced by gliotransmission driven by spontaneous oscillations (Fig. 4b).
Besides the different number of neurons affected, these two modes of signaling could lead to profound differences also in the effect of gliotransmitters on neuronal function. For example, glutamate release evoked by spontaneous oscillations by activating pre-synaptic mGluRs would have a high probability of generating short- rather than long-term plasticity (Perea and Araque 2007). Indeed, spontaneous Ca2+ oscillations occur independently of neuronal activity and thus the probability of having release from astrocytes coincident with a post-synaptic depolarization would be low. In contrast, activity-dependent release from astrocytes would have a higher chance of generating long- rather than short-term plasticity given that neuronal activity would cause both Ca2+ signals in astrocytes and neuronal depolarization. Furthermore, it must be remembered that most activity-evoked astrocytic responses occur with a delay of a few seconds with respect to the neuronal responses (Wang et al. 2006; Schummers et al. 2008). This suggests that gliotransmission would lead to a long- rather than short-term effect only if the neuronal stimulus that determines the astrocytic response lasts for more than a couple of seconds or if the first neuronal stimulus is followed by a second neuronal input that generates post-synaptic depolarization during gliotransmission. Activity-driven release of gliotransmitters could thus contribute to the refinement of neuronal connectivity during activity-dependent plasticity, while gliotransmission evoked by spontaneous oscillations would provide an activity-independent feedback to synapses.
It is now clear that astrocytes are more than the traditional supportive cells providing only structural and metabolic support to neurons. In vivo experiments show that astrocytes are endowed with an intrinsic Ca2+ excitability (Takata and Hirase 2008) and that they can respond to increased neuronal activity with augmented Ca2+ signaling (Wang et al. 2006; Schummers et al. 2008). Astrocytes are involved in key functions of the brain such as the control of cerebral blood flow (Zonta et al. 2003; Mulligan and MacVicar 2004; Takano et al. 2006), synapse formation and stabilization (Christopherson et al. 2005) and are crucial element for the structural plasticity of the synapse (Haber et al. 2006). Additionally, they release chemical transmitters to modulate synaptic transmission and neuronal excitability (Parpura et al. 1994; Bezzi et al. 1998; Kang et al. 1998; Pascual et al. 2005; Panatier et al. 2006). The results discussed in this review suggest that gliotransmission regulates fundamental aspects of neuronal function. By releasing glutamate, astrocytes can influence Hebbian plasticity at single synapses (Perea and Araque 2007) and generate coordinated activity in groups of neurons (Fellin et al. 2004), while, by releasing TNFα, they mediate global synaptic scaling of neuronal networks (Stellwagen and Malenka 2006). Although these recent studies give examples of the potential of astrocytes to control neuronal excitability and synaptic plasticity, the importance of this astrocyte-to-neuron communication in brain physiology and pathology in vivo is still largely unknown. Understanding how all the different effects generated by distinct gliotransmitters integrate with one another, which are the mechanisms controlling the release of different gliotransmitters, which are the neuronal signals activating diverse pathways of gliotransmission, and what is the importance of these different pathways in brain physiology and pathology represents the current challenge in the field. The development of transgenic mouse models bearing specific mutations in different intracellular signaling pathways together with the use of physiological techniques for the study of neuronal function in vivo will represent a valid model for addressing some of these questions.
This work was supported by the Italian Institute of Technology. I am grateful to F. Benfenati, M. D’Ascenzo, R. Fellin, R. R. Gainetdinov, and P. G. Haydon for critical reading of this manuscript and helpful comments.