Astrocyte heterogeneity and interactions with local neural circuits

Astrocytes comprise the largest class of glial cells in the mammalian brain and are essential for central nervous system development and function [1]. Differences between astrocytes have been reported. Classically, astrocytes have been classified based on gross morphology into protoplasmic (gray matter) astrocytes and fibrous (white matter) astrocytes. In addition, there is known variability in the expression of marker proteins, such as glial fibrially acidic protein (GFAP) and excitatory amino acid transporter 1 (EAAT1/GLAST), as well as in astrocyte numbers across brain regions. Finally, various types of specialist astrocytes have been reported in retina (Müller glia) and cerebellum (Bergmann glia). Despite these obvious differences, however, astrocytes have traditionally been viewed as a largely homogeneous cell type [2].

Recent advances in tools to specifically label and manipulate astrocytes are beginning to change this view [3]. It is now widely acknowledged that astrocytes interact with a wide variety of neuronal types [4,5] and are key players at all stages of the synaptic life cycle, including synapse formation and elimination, synapse maturation, maintenance of synaptic homeostasis and modulation of synaptic transmission [6]. This diversity of functions raises the rather obvious question of whether they are performed by all astrocytes, as presupposed, or whether functional subsets of astrocytes exist.

Evidence is now accumulating that astrocytes show profound differences in developmental origin, molecular profile, physiology and functional outputs, which combine to produce a previously unsuspected level of heterogeneity both between and within brain regions. The goal of this mini-review is 3-fold. First, I aim to discuss recent advances in our understanding of astrocyte heterogeneity: to do this, I will concentrate the discussion on protoplasmic astrocytes, principally in mouse cortex, where we arguably have the most complete picture of the astrocyte life-cycle from birth through to adulthood – although other examples will be used when instructive. Readers interested in other astrocyte types are, therefore, directed to recent review articles, which cover aspects of their development and function [7,8]. Second, I aim to discuss the evidence for a central role of neurons in determining local (protoplasmic) astrocyte identity as a means to ensure optimal local circuit formation and/or function. Finally, as astrocyte processes are in direct with synapses, I advance the idea that subcellular compartmentalization is likely a principal driver of (functional) heterogeneity (Figure 1).

Given the short-format of this article, readers interested in a comprehensive review of the literature relating to aspects of protoplasmic astrocyte development, form or function are pointed towards several recent articles, each of which covers a specific topic in depth [2,9–12].

The production of neurons and astrocytes during development is intimately connected and (arguably) best understood in the mouse cortex. Its basic anatomy is generally defined based on the anatomical arrangement of excitatory neurons into six horizontal layers and numerous vertical columns, which form the basis of cortical microcircuits [13]. The generation of cortical neurons begins at approximately embryonic (E) day 10–11. Regional patterning events generate progenitor cells (radial glia) in the ventricular zone. During mid-to-late embryonic development, radial glia generate the vast majority of cortical excitatory neurons. As radial glia extend long processes which span the cortex, these cells also acts as guides for migrating neurons, explaining how the cortex is built in an ‘inside-out’ fashion (in contrast, cortical interneurons are generated outside of the developing cortex in the subpallium and migrate tangentially over long distances into the cortex) [13].

In cortex, radial glia switch to the production of astrocytes (the ‘gliogenic’ switch) around E18. The events underlying the gliogenic switch are incompletely understood. However, it appears that neuronal feedback may play a key role in promoting local astrocyte formation. Both neuron-committed intermediate progenitors and young neurons express the Notch ligands Jagged 1 and Delta-like 1, which activate Notch signaling in radial glia promoting astrogenesis through promoter demethylation, allowing cell type specific gene expression [14]. Cytokine release by neurons (particularly members of the interleukin 6 family) has also been proposed to stimulate gliogenesis, through stimulation of the JAK-STAT signaling pathway [15]. However, it appears that radial glia have a limited capacity for astrocyte generation, having generally disappeared by birth, when only a few astrocytes can be found in the cortex [16]. Expansion of the astrocyte population to populate the entire cortex is thought to arise from extensive local proliferation in the first few weeks of post-natal life [17] (Figure 2). Although neurons appear during embryonic development, assembly and functional maturation of synapses is concomitant to this astrocyte proliferation and maturation [13] (Figure 3).

It appears increasingly likely that this close ontological relationship between radial glia, neurons and astrocytes has important consequences for the formation of cortical microcircuits. It is now recognized that diversity in the progenitor pool has an important influence on the final identity of neurons [18,19]. Furthermore, it appears that clones of related neurons not only form a precise columnar architecture within the cortex but preferentially develop synapses among themselves [20]. The coincident generation of astrocytes from the same progenitors and their close association with sibling neurons to form a stable clonal unit [21–23], suggests a process whereby neurons act as an ‘instructional’ scaffold which ‘sculpts’ astrocytes to support the formation, maintenance and function of specific synapses within distinct local microcircuits, which form the basis of information processing in the CNS.

Protoplasmic astrocytes have an instantly recognizable morphology, consisting of a central soma, and four to ten primary branches, which in turn branch off creating many higher order branchlets and terminal leaflets (Figure 4A’) [24]. (The specialized end-foot structure, which is responsible for astrocyte interactions with blood vessels, will not be considered in this review). This branching is paralleled by a progressive decrease in size from the µm (soma) to nm (leaflets) size domain. Astrocyte morphology appears to develop within the first three post-natal weeks [25], concomitant to synaptogenesis (Figure 3), and appears heavily dependent on a complex interplay between physical interactions with neurons [26] and the effects of soluble neurotransmitters [27].

Mature astrocytes display a variety of morphologies. A systematic study of mouse hippocampal and striatal astrocytes revealed that while these cells have equal somatic volumes, numbers of processes and cell volumes, striatal astrocytes generally occupy a larger tissue area than hippocampal astrocytes [28]. While striatal astrocyte territories typically contain many neuronal cells bodies compared with hippocampal astrocytes, the latter contain more excitatory synapses, implying distinct regional differences in how astrocytes interact with local neuronal circuits [28]. Morphological heterogeneity even extends to astrocytes occupying the same brain region. For example, cortical protoplasmic astrocytes were found to display four prominent morphological subtypes, which distribute in differing proportions through the various cortical layers [29], which may reflect differences in local neuronal activity [30]. Likewise, astrocytes in the dentate gyrus (DG) have also been shown to display morphological heterogeneity linked to DG layering [31].

Astrocyte peripheral processes have received considerable attention as their terminal ends are thought to interact with neurons – creating a structure widely referred to as the ‘tripartite synapse’. Although the degree of this interaction between astrocytes and neurons varies between brain regions (77% of synaptic interfaces in striatum, 86% in hippocampus, 90% in cortex [28,32]), this structure is obviously remarkably common. An elegant electron microscopy study has recently reconstructed astrocytes in mouse somatosensory cortex with unprecedented resolution, revealing an elaborate heterogeneity in process nanostructure [33]. In this study, processes appear as branches of undulating thickness (comprising both expanded and constricted segments), often ending with a constricted tip that is physically associated with a synapse. However, a number of connections were also apparent within the process structure, creating en-passant synapses: it is for this reason that the term ‘peripheral process' is likely to be more accurate than ’perisynaptic process’, which strongly implies a tripartite structure formed by a terminal process ensheathing a synapse [34,35]. Furthermore, synapses around astrocytes tend to organize into clusters of between 4 and 55 structures, with mitochondria positioned closer to larger synaptic clusters, presumably to match the high energetic demands of synaptic transmission (Figure 4A’’). Interestingly, in this study, astrocyte processes were rarely seen looping back on themselves [33]. Hence, astrocytes were proposed to possess a structure largely reminiscent of a ‘thorny bush’, rather than a ‘sponge’. This is in direct contradiction with two other recent studies of astrocyte morphology, one of which also used electron microscopy [36], while the other used super-resolution stimulated emission depletion (STED) microscopy (albeit on live rather than fixed tissue) [37]. Both these studies reported ‘frequent’ loop type structures – although, crucially, neither study quantified the frequency at which these structures appear. At present, the most parsimonious explanation for the discrepancy between these studies is the relatively small number of somatosensory astrocytes reconstructed by Salmon and colleagues [33], combined with potential morphological variability both within and between brain regions (see above) [28,29,31] – although methodological differences cannot be completely discounted at this stage. Nevertheless, the general consensus appears to be that astrocyte processes are much more structurally heterogeneous than previous thought, and this is likely to support specialized interactions with local neuronal circuits.

At present, the degree to which tripartite synapses can be remodeled (if at all) remains unclear. Historically, there is extremely strong evidence for extensive structural remodeling of synapses in brain regions exhibiting powerful biological rhythmicity, such as circadian time setting in the suprachiasmatic nucleus [38], or those involved in modulating responses during extreme physiological conditions, such as the hypothalamus in pregnancy and lactation [39]. However, the situation in other brain regions, under normal physiological conditions, remains unclear. Experiments using a Förster resonance energy transfer (FRET)-based system to measure the distance between astrocyte peripheral processes and neurons in acute slices of striatal tissue suggest that interactions between the two cell types are stable [40]. In contrast, imaging of fluorescently labeled astrocytes and neurons, both in slices and in vivo, suggests that remodeling of peripheral processes can occur rapidly in response to local neuronal simulation in mouse hippocampus and somatosensory cortex [41,42], with process withdrawal impacting directly on local synaptic transmission [43]. The reasons for this discrepancy are unclear. It may reflect intrinsic differences in the biology of the striatum compared to hippocampus and cortex, or it may relate to differences in the methodologies used (or a combination of both factors).

However, what is clear is that dynamic process remodeling will involve extensive modifications, presumably driven by the actin cytoskeleton [44], and will result in the redistribution of essential proteins, protein signaling complexes and mitochondria. Processes with distinct morphologies and signaling machinery may respond differently to synaptic activity. Hence, many questions remain as to how key machinery is positioned within individual processes during various states of cell activity – yet this will undoubtedly be a factor in determining functional output and influence on local circuits.

Our understanding of astrocyte heterogeneity has been massively accelerated by sequencing-based approaches. Droplet-based single cell techniques (allied to relatively shallow sequencing) [45,46] and translating ribosome affinity purification (TRAP)-based experiments [28,47,48] have been performed to compare astrocyte transcriptomes across brain regions. The results clearly demonstrate that this cell type is much more molecularly diverse than previously thought and appears to change with age [48,49]. Although the degree to which transcriptome accurately reflects the cellular proteome is unclear [28,50], a systematic comparison of hippocampal and striatal astrocytes suggests that transcriptomic diversity may well be reflected at the protein level [28].

These groundbreaking studies have since been augmented by work clearly indicating regional substructure in cortex. This can have important consequences for local circuit formation and function, as demonstrated by the work of Miller and colleagues [51]. Using a 8.3 kb fragment of the excitatory amino acid transporter 2 (EAAT2/GLT1) promoter to drive TdTomato expression resulted in a subset of astrocytes in the region of cortical layer 5 being labeled. Crucially, this subset of astrocytes was found to express and secrete the protein Norrin, which regulates local dendrite growth and spine formation [51]. Mutations in the Ndp gene encoding Norrin lead to Norrie disease, which is characterized by retinal abnormalities, intellectual disabilities and behavioral abnormalities, consistent with a central role for the protein in synaptogenesis [52]. This was soon followed by a study which used large-scale single molecule fluorescence in situ hybridization (smFISH) [53] to map single cell transcriptome data obtained using a SMART-seq2-based protocol optimized for use with astrocytes [54]. This approach revealed cortical layering of astrocytes distinct from traditional neuronal laminae. These astrocytic laminae were defined by specific patterns of gene expression, including genes involved in synapse formation and maturation, such as Chrdl1, among others [53]. A similar layering of molecularly distinct astrocytes in DG [31], and the differential expression of µ-crystallin along the dorsal–ventral axis in striatal astrocytes, suggests regional substructure may be a more widespread phenomenon and not restricted to cortex [28].

Hence, current work seems to point strongly to subregional specialization of astrocytes to support local circuits. In this respect, a crucial question is how astrocyte diversity is achieved in the adult brain. Recent work suggests that the pool of developmental progenitors is limited, with astrocytes passing through various immature developmental stages after exiting the cell cycle, before becoming fully mature in the adult brain [55]. There is strong evidence that maturation and diversification of adult astrocytes is strongly dependent on local neuronal activity (Figure 2). Gross manipulations which invert neuronal layering in the cortex result in aberrant astrocyte layering [29,53]. More subtle manipulations which influence local signaling have also been shown to have profound effects. Neuronal activity is synonymous with neurotransmitter release, and work from multiple groups has shown that cortical glutamate release regulates astrocyte maturation, including the acquisition of molecular identity [27,56,57]. In addition to glutamate, other signaling molecules appear capable of regulating astrocyte gene expression. Most prominent among these is sonic hedgehog (SHH). In a seminal study, Farmer and colleagues demonstrated that neurons control Bergmann glia diversity through SHH release in the cerebellum, as well as the acquisition of astrocyte identity in hippocampus [58]. Interestingly, SHH is also produced in adult mouse cortex and SHH receptors are expressed on a proportion of cortical astrocytes [59], an arrangement which appears necessary for the expression of layer-specific astrocyte genes essential for correct synapse development [60,61]. Of course, this does not preclude a contribution of other signaling pathways, such as fibroblast growth factor (FGF) signaling [62]. This idea that neurons provide an ‘instructional scaffold’ for astrocytes is made more attractive by the recent finding that in visual cortex, neuronal identity and function is refined post-natally, in an activity dependent manner [30]. Thus, it appears that neurons and astrocytes may mature into a function circuit together.

To date, the issue of molecular identity has generally been discussed at the whole cell level. However, recent work, combining use of astrocyte-specific TRAP with subcellular fractionation, has allowed the purification of ribosomes from astrocyte peripheral processes associated with synapses (synaptosomes), strongly suggesting the possibility of local translation in astrocytes [63]. Furthermore, this local translatome appears to be dynamic and capable of modification by neuronal activity [64,65]. In mice, translation of new proteins appears to drive, at least in part, the proteomic changes that occur in peripheral astrocyte processes following exposure to a fear conditioning paradigm [65], with the resultant behavioral modifications presumably arising due to changes in astrocyte–neuron interactions which impact circuit activity. In addition, local protein trafficking appears to play a role in defining the composition of the astrocyte process, in response to neuronal activity. In mature cells, the glutamate transporter GLT1 is trafficked to the cell surface and is concentrated at synaptic sites [66], in order to terminate synaptic transmission. As the transport activity of GLT1 is slow and the number of GLT1 molecules present at the synapse sterically limited, it is thought that active glutamate uptake needs to be maintained either by the rapid resupply of ‘fresh’ transporters to the plasma membrane via local recycling [67], or by the lateral exchange of GLT1 transporters from a membrane-resident pool [66]. Interestingly, astrocytes in primates and humans display increased morphological complexity compared with rodent astrocytes, with interlaminar astrocytes extending long peripheral processes which span cortical layers [68], which would further support the concept of local signaling mechanisms as a means of ensuring efficient local function (Figure 4B).

The underlying message appears to be that astrocytes are plastic and respond to their local environment. Such changes can even be deleterious, as in human astrocytes associated with tumor margins, which downregulate genes involved in synaptic function, suggesting the involvement of peritumor astrocytes in tumor-associated neural circuit dysfunction [69]. At the extreme, it appears that striatal astrocytes may be so plastic that they are able to engage a latent neurogenic program following injury [70]. Such plasticity does not appear limited to mammalian glia. Model organisms such as Caenorhabditis elegans, Drosophila and Zebrafish all appear to possess astrocyte-like cells [12]. In the case of Drosophila, astrocytes also appear extremely plastic [71]. The power of the genetic toolbox available in Drosophila, allowing simultaneous labeling and manipulation of astrocytes and their local neurons, suggests that such model organisms may provide highly useful insights into the mechanisms driving astrocyte diversity.

As astrocytes were generally considered to be electrically non-excitable, the early discovery that they elevate their intracellular free Ca²⁺ levels in response to stimuli has attracted a lot of attention, as it suggests a means of cellular communication. Exploration of Ca²⁺ signaling has benefitted from the development of genetically encoded calcium indicators (GECIs), which can be targeted to various cellular locations (although live imaging is generally limited by the diffraction limit of fluorescence microscopy, meaning terminal leaflets are not visible). Use of these indicators has revealed a rich diversity of Ca²⁺ signals in cells, ranging from the generation of signaling microdomains [72,73], through to global waves that encompass entire astrocytes, including their cell bodies [74,75]. Ca²⁺ responses can occur spontaneously [73,76], or be evoked by neuronal activity [77,78].

The most frequent events are the formation of so-called ‘Ca²⁺ microdomains’ in astrocyte peripheral processes, which most likely occur due to the close physical proximity of processes to neurons and their exposure to neurotransmitters and/or neuromodulators [79]. There are likely to be multiple sources of Ca²⁺ within astrocyte processes. Ca²⁺ may enter directly in response to neuronal activity through the opening of ionotropic glutamate receptors, purinergic P₂X receptors or nicotinic receptors. Alternatively, uptake of glutamate and γ-aminobutyric acid (GABA) leads to Na⁺ influx, which activates Na⁺/Ca²⁺ exchangers triggering Ca²⁺ transients. Activation of Gq-coupled metabotropic receptors leads to phospholipase C (PLC) activation and hydrolysis of phosphatidylinositol 4,5-bisphosphate (PIP₂), resulting in the production of inositol 1,4,5-trisphosphate (IP₃) (as well as diacylglycerol). High concentrations of cytoplasmic IP₃ trigger Ca²⁺ release through type 2 IP₃ receptors (IP₃R₂) in the endoplasmic reticulum: crucially, a local ER-derived Ca²⁺ signal may then propagate into a self-sustaining Ca²⁺ wave, due to Ca²⁺-induced Ca²⁺ release (CICR). Ca²⁺ release from mitochondria resident in astrocyte processes (opposite synapses) has also been reported (summarized in [11] and references therein). Such a complex and interlinked system likely explains why astrocytes are capable of responding to single action potentials with ‘microdomain’ events [78] and encoding stimuli of increasing intensity as a gradual accumulation and spread of Ca²⁺ along the process [77], with strong activity leading to global changes, which appear capable of inducing modifications in gene expression through cAMP response element binding protein (CREB) activation [27] (Figure 4C).

In hippocampal astrocytes, oscillations in intracellular Ca²⁺ appear superimposed on a heterogeneous resting Ca²⁺ landscape within the cell, with a gradient extending from the cell soma to the processes. This gradient is age-dependent and astrocytes with high and low Ca²⁺ occupy contiguous space [80]. The basal Ca²⁺ level is set by many factors, including Ca²⁺ leak from internal stores or through plasma membrane channels, the levels and distribution of fixed and mobile Ca²⁺ buffers and uptake/extrusion mechanisms [11]. Hence, heterogeneity in the basal Ca²⁺ reflects the Ca²⁺ homeostasis machinery at work in astrocytes (including presumably the peripheral processes), which will ‘sculpt’ the dynamics of evoked Ca²⁺ increases [81]. Variations in the cellular Ca²⁺ handling machinery presumably underlie the differences in spontaneous and evoked Ca²⁺ activities in hippocampal and striatal astrocytes [28] – and may go some way to explaining the differences in activation state evoked by Gi-coupled DREADDs (Designer Receptors Exclusively Activated by Designer Drugs) in these cells [28]. An additional level of complexity arises from the fact that astrocyte structure is known to be plastic and changes according to activity (see ‘Morphology’), meaning it is possible that Ca²⁺ signaling in a given cell may even change according to its previous activity.

Despite (these) major advances in our understanding of Ca²⁺ signal types, and the mechanisms underlying their generation, their physiological significance and information content are not well understood – although recent work suggests that reducing the amplitude and duration of Ca²⁺ signals in striatal astrocytes leads to excessive self-grooming in mice [82], while impairing Gi-coupled Ca²⁺ signaling brain wide leads to impaired spatial memory [83]. The introduction of next generation signal processing tools [84–86], capable of accurately describing the dynamics of (subcellular) Ca²⁺ transients, will undoubtedly help link the appearance of specific astrocytic Ca²⁺ signals to specific functional outputs. This will have to be allied to improvements in imaging to capture the true spatial dynamics of the signal (3D imaging) [87], over extreme time domains extending from milliseconds to seconds [88]. Major hurdles will still remain, however, to understand the molecular mechanisms linking a specific Ca²⁺ signal to its function output, such as control of local translation, local protein trafficking and protein complex formation, triggering of gliotransmitter release etc.

Crucially, a focus on intracellular Ca²⁺ elevations as activity signals may be unnecessarily constraining our view of astrocyte physiology, as dopamine appears to both elevate and lower astrocyte Ca²⁺ in the stratum radiatum [89], and it will be interesting to assess the functional consequences of such decreases. Beyond Ca²⁺, the use of genetically encoded indicators for cAMP [90] and lactate [91] has revealed transient changes in the intracellular concentrations of these molecules under defined conditions - and these signaling pathways have been reported to have crucial roles in astrocyte maturation and function [27]. In this respect, it is not surprising that neuronal activity appears to engage multiple second messenger pathways in a transmitter-specific, context dependent manner, providing yet another layer of complexity to astrocyte signaling and information processing [92].

Finally, although astrocytes were long assumed to be electrically passive cells, recent experiments with genetically encoded voltage sensors suggest that neuronal activity induces large, rapid, focal and pathway-specific depolarizations in astrocyte peripheral processes. These are driven primarily by action-potential mediated K⁺ efflux and the activity of electrogenic glutamate transporters [93]. These focal depolarizations inhibit astrocyte-mediated glutamate clearance during neuronal activity, enhancing glutamate receptor-mediated transmission in cortical pyramidal neurons. These findings further reinforce the view that astrocyte processes represent functionally independent units capable of influencing local neuronal activity. It is to be expected that local changes in membrane potential will exert powerful effects on astrocyte functions other than neurotransmitter clearance.

Perhaps the single most important advantage of astrocyte diversity is that it enables the creation of specialized neuron–glia units, which can drive complex behaviors. Strong evidence exists that specialist populations of spinal cord astrocytes influence local neural circuits. Astrocytes in the mouse ventral spinal cord express Sema3a, which is essential for axon integrity and survival of α-motor neurons during development [94]. Similarly, ventral horn astrocytes are also enriched in transcripts for Kcnj10, which encodes the astrocyte K⁺ channel Kir4.1. Interestingly, expression of Kir4.1 at sufficient levels was dependent on receiving glutamatergic input from local motor neurons, with loss of astrocyte Kir4.1 selectively impacting on the size of fast α-motor neurons and their function, implying a complicated functional feedback loop [95]. Evidence for astrocyte participation across a broad range of key biological functions has been obtained, including generation of circadian rhythms [96], respiration [97], food intake [98] and processing [99], and memory acquisition [100]. Recent work suggests that hippocampal astrocytes may even act as spatial processors, which encode reward location [101], although the encoding mechanism(s) used by the astrocyte and the consequences for local circuit activity are unknown at the time of writing.

Functional heterogeneity may not only be represented in a region-specific manner. Recent work supports the idea that astrocytes control cortical information processing: endocannabinoid release from cortical neurons leads to astrocyte activation and ATP/adenosine release, which acts to transiently depress synaptic transmission in both a column and layer-specific manner, consistent with a view of self-contained neuron-astrocyte signaling units in cortex (see also ‘Development’) [102]. Evidence for a similar degree of discrimination, based on endocannabinoid signaling, has also been found in the dorsal striatum, which contains intermingled subtypes of medium spiny neurons, that express either dopamine D1 or dopamine D2 receptors. When stimulated, endocannabinoid release from specific striatal neurons leads to increases in intracellular Ca²⁺ in different subsets of striatal astrocytes: crucially, mimicking this activation, using Ca²⁺ uncaging, evokes a synaptic potential only in homotypic pairs of synaptically connected neurons [103].

As astrocytes occupy unique, non-overlapping spatial positions in the parenchyma, it is likely that an individual astrocyte will contact 10,000s if not 100,000s of synapses [28,104,105]. Hence, from a pure spatial perspective, it is likely that astrocytes will contact both excitatory and inhibitory synapses. An interesting idea is that astrocyte peripheral process heterogeneity underlies differential synapse formation. For example, the expression of proteins such as neuroligin 2 and thrombospondin 1 may orchestrate excitatory synapse formation in one process [106], with inhibitory synapse formation relying on the specific expression and trafficking of proteins such as neuronal cell adhesion molecule 1 (NRCAM1) in a different process [107]. Alternatively, both sets of proteins may be expressed in discrete ‘signaling microdomains’ within the same process (Figure 4D). Furthermore, it is likely that the protein composition of astrocyte processes will dynamically change with functional consequences. For example, developmental upregulation of connexin 30 is essential for closing the critical period in mouse visual cortex [108], which it achieves by inhibiting expression of the matrix degrading enzyme matrix metallopeptidase 9 (MMP9), leading to stabilization of perineuronal nets, increased interneuron activity and circuit maturation.

Taken together, these examples provide clear proof that specialized astrocyte subsets impact the function of specific neural circuits and are likely capable of synapse-specific regulation. Unfortunately, at present our ability to identify unique astrocyte subtypes, usually by sequencing, has largely outstripped our ability to manipulate them functionally: given the complex transcriptional ‘fingerprints’ which increasingly define astrocyte subtypes, it is likely that intersectional genetic strategies will be needed for labeling [109]. Although new tools, such as adeno-associated virus (AAV)-based vectors, are simplifying genetic manipulations in mice [3], experiments in classical genetic models appears a promising parallel strategy (see also ’Transcriptome/Proteome’). For example, the power of genetics to help dissect the role(s) of glia in the control of circuit function was recently demonstrated in Zebrafish [110]. In response to failed swim attempts, sustained noradrenergic signaling led to the accumulation of Ca²⁺ in radial astrocytes, ultimately driving activity in GABAergic neurons to inhibit swimming – which presumably reflects an evolutionary constrained behavior to prevent unnecessary energy expenditure on futile tasks. In Drosophila, neurexin-neuroligin interactions were recently shown to close a critical period of motor circuit plasticity [111].

Astrocyte heterogeneity is now an established concept. Aspects of development, molecular profile, intrinsic physiology and function all combine to produce a unique astrocyte identity in the adult (mouse) brain. Emerging evidence suggests that astrocytes show a high degree of plasticity, which allows them to adapt to the needs of local neurons and circuits. This heterogeneity may even exist at the level of individual peripheral processes, and may help fine tune synaptic transmission at individual synapses. However, a number of key questions remain unanswered. These include, but are not limited to, how is heterogeneity established and maintained at the level of individual processes, how is it modified over the complete range of synaptic/circuit activity and what is the potential influence of ageing and/or disease and injury?

• Astrocytes show heterogeneity in developmental origin, mature morphology, molecular profile, intrinsic physiology and functional output.

• Heterogeneity occurs not only between brain regions but also within brain regions.

• Heterogeneity likely underpins the formation of specific astrocyte-neuron interactions essential for the formation of the specialized neural circuits that drive complex behaviors.

• Mounting evidence suggests that heterogeneity extends to the subcellular level, with astrocyte peripheral processes acting as local signaling domains capable of responding to, and influencing, local neuronal activity.

• New tools and experimental strategies – such as live super-resolution imaging combined with genetic methodologies allowing real-time monitoring of mRNA and local protein trafficking, or fluorescent probes to monitor astrocyte and neuron structure and function – are needed to understand the exact range of astrocyte heterogeneity and its functional consequences. In this respect, the more extensive use of genetic model organisms (C. elegans, Drosophila, Zebrafish) offers great potential to drive advances in the field.

M.G.H. has acted as a scientific advisor to Resolve Biosciences on the development of their spatial transcriptomics (Molecular Cartography™) platform.

M.G.H. acknowledges previous support from the European Research Council (ERC) [grant number 281961]; KU Leuven (C1) [grant number C14/20/071]; Fonds Wetenschappelijk Onderzoek (FWO) [grant numbers G066715N and 1523014N] and The Foundation for Alzheimer Research (SAO-FRA: S#16005) for support on work relating to astrocyte heterogeneity. M.G.H. is currently the ERA Chair (NCBio) at i3S Porto funded by the European Commission (H2020-WIDESPREAD-2018-2020-6; NCBio) [grant number 951923].

M.G.H. is solely responsible for the content of this article.

M.G.H. would like to thank Dr Sergey Kasparov, Dr Jérôme Wahis, Dr Lynettte Lim and Dr Marika Marino for discussions, and Dr. Olga Sin for administrative support. M.G.H. apologizes if any relevant articles were omitted due to space constraints. Figures were prepared by David Pennington (@PenningtonArt).

CNS

Central nervous system

GFAP

Glial fibrially acidic protein

EAAT1

Excitatory amino acid transporter 1/GLAST

GFP

Green fluorescent protein

GPCR

G-protein coupled receptor

IP₃R₂

Type 2 inositol 1,4,5-trisphosphate receptor

MCU

Mitochondrial Ca²⁺ uniporter

(m)NCX

(Mitochondrial) Na⁺/Ca²⁺ exchanger

mPTP

Mitochondrial permeability transition pore

PMCA

Plasma membrane Ca²⁺ ATPase

SERCA

Sarcoplasmic/endoplasmic reticulum Ca²⁺ ATPase

α2δ-1

Voltage-dependent calcium channel subunit α2δ-1

NL

Neuroligin

NRCAM

Neuronal cell adhesion molecule

NRX

Neurexin

TGF-β1

Transforming growth factor-β1

TSP

Thrombospondin

DG

Dentate gyrus

STED

Stimulated emission depletion

FRET

Förster resonance energy transfer

TRAP

Translating ribosome affinity purification

EAAT2

Excitatory amino acid transporter 2/GLT1

smFISH

Single molecule fluorescence in situ hybridization

SHH

Sonic hedgehog

FGF

Fibroblast growth factor

GECI

Genetically encoded Ca²⁺ indicator

GABA

γ-aminobutyric acid

PLC

Phospholipase C

PIP₂

Phosphatidylinositol 4,5-bisphosphate

IP₃

Inositol 1,4,5-trisphosphate

CICR

Ca²⁺-induced Ca²⁺ release

CREB

cAMP response element binding protein

DREADD

Designer receptor exclusively activated by designer drugs

MMP9

Matrix metallopeptidase 9

AAV

Adeno-associated virus