Control of complex behavior by astrocytes and microglia

Abstract

Evidence that glial cells influence behavior has been gaining a steady foothold in scientific literature. Out of the five main subtypes of glial cells in the brain, astrocytes and microglia have received an outsized share of attention with regard to shaping a wide spectrum of behavioral phenomena and there is growing appreciation that the signals intrinsic to these cells as well as their interactions with surrounding neurons reflect behavioral history in a brain region-specific manner. Considerable regional diversity of glial cell phenotypes is beginning to be recognized and may contribute to behavioral outcomes arising from circuit-specific computations within and across discrete brain nuclei. Here, we summarize current knowledge on the impact of astrocyte and microglia activity on behavioral outcomes, with a specific focus on brain areas relevant to higher cognitive control, reward-seeking, and circadian regulation.

Glial cells: a brief context on cell diversity and behavioral relevance

The term ‘glia’ refers to a heterogeneous population of cell types. These include: oligodendrocytes, that form axon-insulating myelin sheaths, oligodendrocyte progenitors (also called NG2-cells or polydendrocytes), ventricle-lining ependymal glia, astrocytes, and microglial cells¹. Glia is a contraction of the word ‘neuroglia’ (Nervenkitt) or ‘nerve-glue’, first coined by Rudolf Virchow in 1856, who referred to these cells collectively as the substance “which lies between the proper nervous parts…and gives the whole its form in a greater or less degree”^2,3. A few years after this statement, the close relationship between glial cells and blood vessels was acknowledged by anatomists, but it was not until 1891 that the term ‘astrocyte’ emerged as a distinct cell type mediating this relationship. It took another 30 years before ‘oligodendrocytes’ and ‘microglia’ were recognized as additional glial cell subtypes⁴. Within the broad scientific opinion, the brain glial cells have remained relegated to inert, neuron-supporting roles until about the 1990’s when research into active contributions of glial cells to neuronal activity started to gain ground.

According to recent historical perspectives, oligodendrocytes constitute 45–75% of all glial cells in the human brain, followed by astrocytes (19–40%) and microglial cells (5–10%)^5,6. The frequently cited neuron-to-glia ratios of 1:10⁷ appears to be a mistaken notion that is curiously perpetuated by many textbooks. In the 60–80’s, estimates of the number of neurons and glial cells in the human brain indicated that there are 70–85 billion neurons and 40–130 billion glial cells^8,9, suggesting something close to a 1:1 ratio between neurons and glial cell numbers⁵. Some studies in the vestibular nuclei of the brainstem did find there were ten times more glial cells than neurons, but these estimates were not backed up by experimental evidence of brain-wide counts^5,10–12. Modern techniques, such as isotropic fractionation, flow fractionation and stereological optical fractionation, confirm the estimate of ~1:0.7 ratio between neurons and glial cells that is surprisingly close to the one obtained a long time ago using histological/stereological approaches^9,13–15. Overall, the emerging picture is that the glial cell density is relatively uniform across species and brain regions, and that the glia-to-neuron ratio is simply higher whenever the size of neurons and their processes, which do vary substantially, is higher and neuronal density is correspondingly lower^16–18.

In the evolutionary context, number of cells and, perhaps more importantly, number of cell types, have been proposed to vary linearly with organismal complexity. Indeed, astrocytes in the human brain are substantially more varied than rodent astrocytes^19,20 and some have used this evidence to bolster the argument for glial cell role in cognitive processes⁴. The anatomical and functional heterogeneity of glia is now beyond doubt and mounting evidence suggest substantial transcriptome diversity^21–23. The multiple mechanisms by which glial cells may interact with neurons remain a subject of investigation, and it is not currently known whether unique mechanisms map onto specific computations performed by neuronal circuits mediating discrete behaviors. The impact of oligodendrocytes, oligodendrocyte progenitors, and ependymal glia on complex behavior has so far attracted a relatively small share of investigative attention. In the meantime, the role of astrocytes in behavior has been actively pursued and backed by substantial mechanistic insight ^24,25. Microglia contributions to behavior have also been widely recognized as part of a complex interface between neuronal activity and immune system status²⁶. Given such skewed accumulation of evidence in support of behavioral impact, this review focuses on astrocytes and microglia only. Specifically, here we examine contributions of astrocytes and microglia signals to behaviors thought to arise from activity within three relatively well-established circuits: frontocortical circuits underlying executive function, mesocorticolimbic circuits mediating reward-seeking behaviors, and suprachiasmatic nucleus (SCN) circuits driving circadian homeostasis (Fig.1).

Figure 1.

An outline of the discussed mechanisms underlying glial interactions with neuronal circuitry. Astrocytes (left) and microglia (right) display cytoskeletal changes in response to their microenvironment. These changes are linked to molecular signals underlying structural remodeling and a variety of associated functional changes. Note that binary representation of astrocyte and microglia morphologies is overly simplistic and intermediate phenotypes may be possible as discussed in the text for resting and active microglia. Structural and functional adaptations (dashed arrows) influence glial interactions with neuronal networks and impact diverse neuron types (schematized as purple and green cells). Central origin of the dashed arrows indicates that transition between morphologies is not a pre-requisite for functional changes. Discrete brain regions (top) at the focus of this review, may experience distinct changes in neuroglial dynamics arising from unique combinations of underlying adaptations. FC, frontal cortex; HPC, hippocampus; dSTR, dorsal striatum (including dorsomedial and dorsolateral striatum), NAc, nucleus accumbens (ventral striatum); AMG, amygdala; VTA, ventral tegmental area; SCN, suprachiasmatic nucleus.

Regulation of activity in frontocortical circuits underlying executive function

Many studies support glial cell contributions to neuropathology of cognitive decline, brain trauma, and neurodegeneration as well as regulation of cognitive function in the healthy brain. In this section, we review the literature that examines astrocytes and microglia in the context of their impact on executive control of behavior by the frontocortical circuits, including cortico-striatal, cortico-hippocampal, and cortico-amygdalar connections (Table 1). We begin with astrocytes and highlight converging evidence that regulation of frontocortical function by these cells involves glutamate recycling, release of transmitter molecules, secretion of inflammatory mediators and other signaling factors in addition to cytoskeletal adaptations that likely impose spatial constraints on neuroglial interactions. We recognize that these aspects of astrocyte activity do not represent an exhaustive list of mechanisms by which astrocytes modulate neuronal transmission and that other functions attributed to astrocytes (e.g., water balance, K⁺ buffering, glucose metabolism, etc.) are likely to impact frontocortical-dependent behaviors. Astrocytes represent a regionally heterogeneous population of cells^21–23,27 and it is not yet clear whether any combination of factors (e.g., gene expression, cell morphology, excitability profiles, etc.) can be used to define conserved astrocyte categories regardless of their regional placement in the brain. However, current evidence does indicate unique molecular phenotypes of astrocytes between brain regions and deeper understanding of the importance and role of such phenotypes is beginning to emerge^28,29.

Table 1.

The table focuses on those studies where cell-specific manipulations occurred prior to the observed behavioral effects. Studies in which cellular changes were measured after a behavioral manipulation are excluded with a few exceptions. For constitutive gene knock-out models, brain area specificity column indicates brain areas in which cellular changes were measured. Bi-directional effect column lists evidence, if any, that measured behaviors could be either suppressed or enhanced with opposite cellular manipulations.

	Experimental manipulationREF#	Brain area specificity	Behavioral impact	Cellular mechanism	Bi-directional effect
Astrocytes
Cognitive flexibility	L-AAA injection (astrocyte toxicity)30,31	mPFC	30 ↓cognitive flexibility(set-shifting) 31↓texture discrimination ↓reversal learning and working memory	30 ↓gamma oscillations 31 Not investigated	Not investigated
	hM3 DREADD activation30	mPFC	↑cognitive flexibility (set shifting)	↑Gq signaling in astrocytes	Not investigated
	S100b infusion30	mPFC	↑cognitive flexibility (set shifting)	Not investigated	Not investigated
	GFAP & vimentin knockout mice41	Whole brain	↑memory extinction	Not investigated	Not investigated
	Nestin knockout mice43	Whole brain	↑memory extinction	Not investigated	Not investigated
	EAAT1 knockout mice54	Whole brain	↓ visual discrimination learning	↓ astrocyte glutamate reuptake transport	Not investigated
	Blockade of EAAT2 by dihydrokainate55	mPFC	↑ drinking latency & ↓ responding to PFC electrical self-stimulation55	↓ astrocyte glutamate reuptake transport55	Not investigated
	Mice IP injections of sub-chronic ketamine56	Hippocampus	↓ extinction of operant sucrose self-administration	↓ EAAT2 expression	Not investigated
	dnSNARE knockout mice65	PFC, hippocampus	↓ working and spatial memory	↓ theta-cycle synchronization between hippocampus and PFC	Theta coherence and cognitive deficits rescued by d-serine supplementation
	Deletion of astrocyte GABAB receptors in mice74	PFC	Aberrant cortical synchronization and ↓ T-maze alternation performance	astrocytic GABAB signaling activates mGluRs and recruits PV interneurons	Not investigated
	Urokinase plasminogen activator receptor knockout mice81	orbitofrontal cortex, dorsal striatum	↓ reversal learning	↓ PV interneurons in orbitofrontal cortex and dorsal striatum	Reversal learning deficit rescued by up-regulation of hepatocyte growth factor in astrocytes
	Human Chrdl1 mutations83	Retina, PFC	↑ executive function	Corneal abnormalities and myelination deficits	Not investigated
	Overexpression of MHC1 in mice mPFC astrocytes88	mPFC	↓ reward-based visual discrimination	↓ neuronal spine density in dorsal striatum	Not investigated
	Chronic expression of IL-6 in transgenic mice89	PFC, amygdala	↓ avoidance learning	Loss of synapses and calbindin-containing neurons due to chronic neuroinflammation	Not investigated
Reward seeking	Stimulation of hM3D receptors on astrocytes146	PFC	↑ ethanol drinking	Activation of Gq-coupled signaling on astrocytes	Not investigated
	Stimulation of hM3D receptors on astrocytes147,148	NAc	↓ cocaine reinstatement and motivation to self-administer ethanol147,148	Stimulation of presynaptic mGluRs before hM3D stimulation147	Not investigated
	Optogenetic stimulation of VTA astrocytes149	VTA	↑ avoidance behavior, overriding CPP for cocaine	Astrocytes stimulation VTA GABA interneurons, inhibiting dopamine neurons	Glutamate transporter (GLT-1) from VTA astrocytes blocks avoidance behavior and maintains CPP for cocaine
	Optogenetic activation of hippocampal astrocytes151	Hippocampus	↓ consolidation of contextual fear memory	Release of ATP and adenosine	Pharmacological stimulation of A1 receptors elicited similar effects
	hM3D activation of astrocytes167	Dorsal medial striatum	Shifted behavior from habitual to goal-oriented	hM3D activation reduced sEPSC frequency in D1 MSNs, but increased sEPSC amplitude in D2 MSNs	The effect was not observed in adenosine transporter (ENT1) knockout mice
	Training-induced upregulation of EAAT2/GLT-1171	Dorsal lateral striatum	↑ habit behavior in operant task for chocolate reward	Astrocytes reinforce relative contributions of the DMS and DLS directed behaviors	Inhibition of EAAT2/GLT-1 upregulation promoted goal-directed behavior
	Overexpression of the plasma membrane Ca2+ pump in astrocytes173	Dorsal lateral striatum	↑ excessive and compulsive-like self-grooming behavior	Removal of astrocyte Ca2+ signaling	Not investigated
Circadian rhythms and sleep/wake cycle	Ca2+ imaging of astrocytes in naïve animals270,271,272	Frontal cortex270,272, hippocampus, hypothalamus, pons, cerebellum, barrel cortex270,271	Ca2+ levels increase during wakefulness and decrease during REM sleep in several brain areas270,271,272	Not investigated	Not investigated
	Genetic reprogramming of SCN astrocyte clock genes220	SCN	Genetic reprogramming of astrocyte clock genes reshapes circadian behavior	Not investigated	Not investigated
	Genetic reprogramming of SCN neurons to have incompetent clock gene expression221	SCN	Astrocytes reinstate clock gene expression and circadian function of SCN neurons autonomously	Astrocyte glutamatergic signaling in the SCN	Not investigated
	Experimental manipulationREF#	Brain area specificity	Behavioral impact	Cellular mechanism	Bi-directional effect
Microglia
Cognitive flexibility	Chronic unpredictable stressors in mice117,120	mPFC	↓ PFC-dependent temporal object recognition117	↑ CSF1 receptor and somplement component mRNA117,120	Both mRNA and behavior rescued by treatment with RU486 (glucocorticoid antagonist). 117
	7-day restraint stress119	dmPFC	↓ reversal learning in 4-choice odor discrimination	↑ dmPFC dendritic spine elimination	Not investigated
	Blockade of adenosine2A receptor in prenatal dexamethasone model for anxiety123,124	Hippocampus, PFC	↑ cognitive performance in recognition memory task124	Normalized frontocortical123 and hippocampal124 microglia morphology	Not investigated
	Adolescent social stress model in mice124	PFC	↓ cognitive flexibility performance	Activated microglia release TNFα	Cognitive flexibility deficits were rescued by increased TNFα release by ranylcypromine
	Radioligand TSPO distribution PET scan125,126,127,128	PFC	↓ attention125 ↓ memory and executive function126	↑ TSPO radioligand binding due to increased microglia activation125,126,127,128	Not investigated
			Cognitive decline127,128
Reward seeking	Inhibiting microglia with minocycline in alcohol-dependent mice188,189	Not investigated (IP injection)	↓ withdrawal-induced anxiety188,189	Not investigated	Not investigated
	Depletion of microglia with PLX-5622190	Whole brain	Does not change escalation of voluntary alcohol consumption	Not investigated	PLX-5622 does block escalation under conditions of repeated immune activation
	Depletion of microglia with PLX-5622191	Whole brain	↓ anxiety during alcohol withdrawal ↓ escalation of alcohol intake	Normalized excitatory and inhibitory synaptic plasticity in central nucleus of the amygdala	Not investigated
	Depletion of microglia with PLX-5622192	Whole brain	Normalized performance on marble burying and open field tests after nicotine withdrawal	Elevation of NOX2 release from microglia	Not investigated
	Blockade of TLR4 signaling with LPS-RS or TLR4 knockout mice193	Whole brain	↓ CPP and self-administration of cocaine	Inhibiting TLR4 ↓ extracellular dopamine from cocaine in the NAc by preventing downstream release of IL-1β	Not investigated
	Minocycline treatment194,195	Whole brain	↓ both maintenance and reinstatement of methamphetamine-induced CPP	Minocycline ↓ extracellular dopamine from methamphetamine in the NAc	Not investigated
Circadian rhythms and sleep/wake cycle	Chronic sleep fragmentation model in wild-type mice281	PFC, hippocampus	↓ spatial learning and memory ↑ aggression and anxiety	↑ intracellular amyloid-β, dysfunction of endosome-autophagosome-lysosome pathway	Not investigated
	Depleting microglia with Cx3cr1 transgenic mice282	SCN, hippocampus	Abnormal circadian body temperature and diurnal rhythms	Disrupted expression of clock genes Per1, Per2, and Bmal1	Not investigated
	Ablation of microglia following PLX-5622 treatment in mice283	Whole brain, Hippocampus	↑ sleep duration	↓ in light phase-dependent excitatory synaptic transmission in CA1 pyramidal neurons	Not investigated
	Sleep deprivation in rats288	Whole brain, hippocampus	↑ memory impairments ↓ performance in Morris water maze	Sleep deprivation increases activated microglia activity and minocycline blocks activation of resting microglia	Memory and Morris water maze impairments were rescued by minocycline
	Minocycline treatment in human290 and rodents291,292	Whole brain	Both ↑290 and ↓291,292 sleep quality and episodic memory	minocycline blocks activation of resting microglia288,290,291,292	Not investigated

Astrocyte lesions alter cognitive flexibility –

Several studies have investigated the impact of lesioning astrocytes in the prefrontal cortex (PFC) on relevant behavioral outcomes. In recent work, bilateral administration of the astrocyte-specific toxin, L-AAA, into the medial PFC (mPFC) impaired cognitive flexibility in an attentional set-shifting task and reduced the power of gamma oscillations in the mPFC without any effects on neuron morphology³⁰. In contrast, stimulation of astrocyte activity via G_q receptor coupled DREADDs as well as via mPFC infusions of the astrocyte-secreted calcium-binding protein S100β, in the non-L-AAA lesioned groups, improved cognitive performance³⁰. These observations are consistent with previous work that identified deficits in the ability to discriminate between texture-based, but not odor-based, stimuli, as well as impairments in odor-based reversal learning and working memory after infusions of L-AAA into the mPFC³¹. The L-AAA lesions of the PFC have also been reported to impair sucrose preference and novelty-suppressed feeding whereas excitotoxic lesioning of neurons by ibotenate did not impact performance on these tasks^32,33. Consistent with the postulate of a close-knit relationship between local astrocyte function and neuron structure, substantial reductions in dendritic complexity of mPFC neurons have been observed after L-AAA administration³⁰. In the pre-limbic cortex, L-AAA triggered swelling of neuronal cell bodies, but did not result in measurable neuronal cell death³². The ability of astrocytes to facilitate behavioral dysfunction without an outright neuronal loss has particular relevance to pathophysiology of drug seeking behaviors and depression, conditions characterized more by synaptic reorganization and plasticity rather than extensive neuronal degeneration^34,35. However, results of L-AAA lesioning studies should be interpreted with caution given evidence that L-AAA inhibits glutamate uptake and glutamate metabolism leading to elevated extracellular glutamate levels³⁶. An additional interpretational caveat of L-AAA studies has to do with findings that L-AAA increases microglial infiltration into the lesioned area³⁷ which has clear implications for attribution of the behavioral effects of L-AAA exclusively to astrocytes.

The role of astrocyte cytoskeleton –

Several provocative reports have been published to examine the impact of astrocyte cytoskeletal proteins on behavior. GFAP and vimentin are astrocyte intermediate filament proteins often used for immunohistochemical identification of astrocytes, but also for examination of gross changes in astrocyte morphology. These molecules have been extensively used as markers of reactive gliosis, a phenomenon in which up-regulation of GFAP and vimentin is accompanied by morphological and a wide array of molecular changes³⁸. Nestin is another intermediate filament protein that is normally expressed in progenitor cells during development, but is also re-expressed in mature reactive astrocytes, following brain trauma^39,40. With regard to frontocortical function, one investigation found that transgenic mice null for GFAP and vimentin displayed improved extinction of memories linked to location of a food container in a reversal task⁴¹. While acquisition of place memories relies heavily on the hippocampus, spatial reversal learning implies suppression of existing information and requires additional processing in prefrontal areas⁴². GFAP/vimentin null mice maintained surprisingly normal performance on initial acquisition of hippocampus-dependent place memories, highlighting the specific impact on frontocortical processing⁴¹. Interestingly, mice with constitutive deletion of nestin showed a similar pattern of results: normal memory acquisition, but better extinction of acquired place memories⁴³. Although behavioral specificity of GFAP deletion to reversal learning is intriguing, its absence since embryonic stage may have also impacted adult neuron populations or neural circuit interactions since GFAP is expressed in neural progenitor cells. Indeed, there are suggestions that altered cognitive function may be associated with highly local, cell-layer specific regulation of astrocyte number and morphology as found in a post-mortem analysis of GFAP expression in brains of schizophrenia patients⁴⁴. Pre-clinical studies also support the idea that astrocyte interactions with neurons may be targeted to specific neuronal populations⁴⁵. Overall, a vast number of reports outside the scope of this review has relied on intermediate filament expression, particularly GFAP, to evaluate astrocyte numbers and morphology throughout the brain. The most parsimonious conclusion from this literature is that astrocyte number is bidirectionally sensitive to and can be correlated with many behavioral antecedents, sequelae, or disease pathologies. However, since GFAP staining does not label fine astrocytic processes that are most likely to contact synapses, research using other markers to evaluate astrocyte plasticity is sorely needed^e.g..46 as neither astrocyte numbers on their own nor GFAP-based gross morphology reports are likely to provide mechanistic insight into behavioral relevance of these cells.

Astrocytic control of glutamate recycling and GABA synthesis –

Astrocytes recycle over 90% of extracellular glutamate⁴⁷, with an impact not only on neuronal supply of glutamate, but also on availability of glutamate for GABA synthesis⁴⁸. Among the four major subtypes of glutamate reuptake transporters, two – EAAT1 (GLAST in rodent) and EAAT2 (GLT-1 in rodent) – are abundantly expressed in astrocytes of the CNS. Expression of both EAAT1 and EAAT2 is strongly decreased in schizophrenia and substance use^49,50, disorders known to prominently involve frontocortical activity. A decrease in astrocyte number and EAAT2 expression is also observed in Alzheimer’s disease (AD), in humans and in mouse models, where this deficit has been associated with slower clearance of extracellular glutamate^51,52. Interestingly, overexpression of astrocytic glutamate transport may be beneficial for cognitive function, as suggested, for example, by a post-mortem analysis of entorhinal cortex of AD patients, where overexpression of EAAT2 was found to correlate with the absence of dementia or mild cognitive impairment⁵³. A deficit in visual discrimination learning was reported in mice with constitutive deletion of EAAT1⁵⁴, suggesting impairment of executive function, although the learning deficit prevented a more direct evaluation of frontocortical control in a reversal task⁵⁴. Reminiscent of the effects of astrocyte lesioning in L-AAA studies, blockade of EAAT2-dependent glutamate uptake by dihydrokainate increased drinking latency in a sucrose preference task and led to near-complete cessation of responding for electrical stimulation of the PFC in an intracranial self-stimulation paradigm⁵⁵. In another study, reduced expression of EAAT2 in the mouse hippocampus was observed after treatment with sub-chronic ketamine which was associated with impaired extinction of an operant sucrose self-administration task⁵⁶. These findings were speculated to reflect cognitive deficits characteristic of substance use and resemble data from the PFC of schizophrenic patients⁴⁹. However, extinction of self-administration is not a widely used behavioral indicator of cognitive impairment, and future studies will be needed to more closely examine the impact of astrocytic EAAT2 on frontocortical control of behavior. In the meantime, repeated observations that disrupted glutamate uptake strongly impacts activity of NMDA receptors throughout the brain^57–59, is likely to have implications for astrocyte control of neuronal output as discussed in the next section.

Transmitter release by astrocytes –

A series of seminal studies used a mouse model with astrocyte-specific deletion of the SNARE protein (dnSNARE mice) that was argued to limit transmitter release⁶⁰. Although the specificity of GFAP-driven dnSNARE transgene to astrocytes has been challenged⁶¹, this work formed the foundation for the view that astrocytes may promote synchronous neuronal activity and are instrumental for forms of long-term neuronal plasticity⁶². Later studies provided support for this hypothesis by demonstrating that astrocytes regulate sleep patterns and cortical oscillations with implications for cognitive performance^63,64. The dnSNARE mice have also been recently used to demonstrate impaired theta-cycle synchronization between the hippocampus and the PFC accompanied by deficits in working and spatial memory, but intact neuronal morphology in both the PFC and the hippocampus⁶⁵. Both the theta coherence and the cognitive deficits could be rescued by supplementation with d-serine⁶⁵. D-serine acts as a co-agonist of NMDA receptors at the glycine site and d-serine release from astrocytes has been argued to impact synaptic long-term potentiation in the hippocampus⁶⁶. Additionally, astrocyte-released ATP and its metabolic product, adenosine, may also influence NMDA receptor expression and signaling via activation of purinergic receptors⁶⁷. The interaction between ATP/adenosine receptors and neuronal signals in the PFC is supported by experiments in slices indicating that astrocytic P2Y4 receptors are linked to vesicular release of glutamate and activation of neuronal metabotropic glutamate receptors (mGluRs) and NMDA receptors⁶⁸. Altogether, there is substantial support for the view that NMDA receptors represent a critical component of astrocyte interaction with neurons. Particular attention must be paid to a potentially unique contribution of NMDA receptors located at extrasynaptic sites which have been proposed as primary mediators of neuronal response to astrocyte-released glutamate^69–72.

In addition to glutamate-mediated signaling, astrocytes also release GABA with an impact on inhibitory neurotransmission⁷³. The specific effects of astrocyte-released GABA on behaviorally relevant frontocortical activity remain to be determined. However, a recent publication demonstrated that selective deletion of GABA_B receptors in astrocytes of the mouse PFC resulted in aberrant cortical synchronization and deficits in a T-maze alternation task⁷⁴. The study proposed that astrocytic GABA_B signaling facilitates activation of group 1 mGluRs and recruitment of parvalbumin-expressing interneurons, suggesting a link between GABA_B receptor activation and glutamate release from astrocytes. An interaction between astrocytes and parvalbumin positive interneurons has also been reported in the visual cortex, with an effect on orientation to visual stimuli⁷⁵.

We would like to conclude this section by acknowledging an on-going debate over astrocytes ability to release transmitter molecules through vesicular (i.e. SNARE-mediated) or non-vesicular release mechanisms all grouped under the general definition of “gliotransmission”. The two most prominent aspects of this argument are an unresolved question over the role of astrocyte Ca²⁺ as mediator of gliotransmitter release and the degree to which ex vivo experimental conditions demonstrating astrocyte-mediated release reflect in vivo physiology of these cells⁷⁶. The contentious nature of the debate might arise from the fact that mechanisms of transmitter release from astrocytes are potentially more diverse and certainly much less understood than in neurons^76,77 and evaluations of astrocyte Ca²⁺ may require methods distinct from those applied to neuronal Ca²⁺ transients^78–80. As a consequence, trying to ‘squeeze’ gliotransmission into the framework of extensively characterized neuronal release mechanisms may not provide an accurate depiction of transmitter release from astrocytes. Similarly, any answers as to whether astrocyte contributions to neuronal activity align with intact physiology must await further experimentation, given that the overwhelming majority of current knowledge has been generated using astrocyte cultures or slice preparations.

Other astrocyte-secreted factors –

Astrocytes release a range of neuromodulators that interact with the surrounding neurons and impact behavioral measures of cognition. For example, Plaur transgenic mice, that lack urokinase plasminogen activator (uPA) receptors and show a specific loss of parvalbumin positive interneurons in the orbitofrontal cortex and dorsal striatum, display a pronounced deficit in performance on a reversal learning task⁸¹. This deficit, which does not generalize to a broad learning impairment, can be rescued by up-regulation of the hepatocyte growth factor in GFAP-positive astrocytes⁸¹. Loss of astrocyte reactivity could play a role in these behavioral outcomes since up-regulation of astrocyte uPA receptors in response to uPA secretion from neurons has been reported to mediate astrocyte activation, as measured by increased GFAP expression, after ischemic injury⁸². Another interesting secreted protein is chordin-like 1 (Chrdl1). Chrdl1 acts an antagonist of bone morphogenetic protein, which is critical in CNS development. A human study linked mutations in the Chrdl1 gene to development of corneal abnormalities and myelination deficits, but strikingly, superior performance on tests of executive function⁸³. In mice, secretion of Chrdl1 from astrocytes is essential for increased developmental expression of Ca²⁺-impermeable, GluA2 subunit-containing AMPA receptors that impede synaptic plasticity⁸⁴. Regulated expression of Ca²⁺-permeable versus Ca²⁺-impermeable AMPA receptors has been proposed as one of key characteristics of neuroplasticity within the mesolimbic dopamine reward system associated with cocaine use⁸⁵. Related to this, up-regulation of Chrdl1 RNA was demonstrated in cultured astrocytes treated with dopamine⁸⁶.

A vast literature supports the notion that astrocytic release of pro- and anti-inflammatory molecules as well as astrocytic response to these molecules has consequences for frontocortical signaling⁸⁷. For example, astrocytic overexpression in the mPFC of major histocompatibility complex 1 (MHC1), which is involved in adaptive immune responses, led to decreased neuronal spine density in the dorsal striatum and deficient performance in a reward-based visual discrimination task, despite normal performance in a reversal learning task⁸⁸. In another study, mice that constitutively expressed IL-6 in astrocytes showed deficits in avoidance learning that are thought to rely on PFC interactions with the amygdala⁸⁹. The inflammatory role of astrocytes has been a prominent area of research on neurodegenerative disease with pathologies that span the gamut of behavioral impairments from memory deficits to dysregulation of executive function to depression, anxiety, etc. Numerous astrocyte-specific targets have been identified across neurodegenerative conditions including Parkinson’s disease⁹⁰, Huntington’s disease⁹¹, AD⁹², multiple sclerosis⁹³, stroke⁹⁴, traumatic brain injury (TBI)⁹⁵ as well as normal aging^96,97. Since it appears unlikely that any one of the behavioral measures in these conditions can be attributed to a single astrocyte-specific factor, a deeper understanding of how combinations of secreted factors impact astrocyte interactions with local neurons and long-range signals between brain regions would be extremely valuable.

Microglial contributions to executive function –

Microglia-mediated inflammation has been observed to contribute to frontal cortex dependent behaviors across a range of pathologies, including AD, schizophrenia, TBI, aging, PTSD, depression, neurodegeneration as well as diet-induced inflammation^98–100. Similar to astrocytes, microglial morphology is highly sensitive to a wide range of physiological and pathological stimuli and many studies rely on morphology-based classification of activated microglia as evidence of altered microglial signaling. Indeed, physical appearance of the microglia can be indicative of the neuro-environment¹⁰¹. For example, resting microglia have long, thin, highly branched processes with a small cell body, and they are primarily in charge of querying the neuronal environment for perturbations to ensure effective communication between neurons. In line with their ‘surveillance’ role, these cells are regularly tiled throughout the brain to ensure local oversight by at least one microglial cell. However, when microglia detect a potential neuro-insult, these cells are galvanized into an active state, characterized by retraction of their processes and acquisition of a more amoeboid phenotype. Because these cells share a common lineage to macrophages in the peripheral immune system, microglial activation has been similarly described as falling under M1 (pro-inflammatory and expressing markers such as CD86, iNOS, CD16/32, and MHCII) or M2 (anti-inflammatory and expressing markers such as CD206, FIZZ1, and Arginase I) phenotypes. More recently, however application of M1 and M2 descriptors to microglia has been called into question based on growing support for existence of multiple intermediate and combinative phenotypes¹⁰².

There is substantial evidence that administration of minocycline, a tetracycline antibiotic that disrupts protein synthesis and prevents microglial activation elicits mild-to-substantial cognitive improvement across multiple domains including cognitive symptoms of bipolar disorder¹⁰³, HIV¹⁰⁴, addiction¹⁰⁵, depression¹⁰⁶ and psychotic disorders ^107,108. Pro-cognitive effects of microglia depletion have also been observed after cranial irradiation, often used as a cancer therapy¹⁰⁹. Reducing microglial population in mice by treatment with the colony stimulating factor (CSF)-1 receptor inhibitor, PLX-5622, has been reported to rescue irradiation-induced behavioral deficits on an ‘object in place’ and contextual fear-conditioning tasks both of which involve frontocortical signaling¹⁰⁹. Interestingly, a recent study reported no effect of minocycline on cognitive decline associated with mild AD in human subjects¹¹⁰, although cognitive benefits have been reported in animal models¹¹¹. To reconcile such disparate outcomes, some researchers advocate for use of minocycline in combination with other drugs as a strategy to combat cognitive impairment in AD patients^112,113. It must be noted, however, that similarly to astrocyte depletion with L-AAA, microglia depletion with PLX-5622 or other CSF-1 inhibitors may have off-target effects that involve other cell types¹¹⁴.

Consistent with the body of knowledge that stress is both associated with cognitive deficits and promotes systemic inflammation, stress-activated microglia have also been found to facilitate cognitive impairment. Dysregulated stress responsivity within the PFC is thought to be one of core factors leading to emergence of depression and associated cognitive deficits^115,116. For example, a regime of chronic unpredictable stressors led to elevation of CSF1 receptor and complement components C1q and C3 mRNA in the PFC and behavioral deficits in a PFC-dependent temporal object recognition task¹¹⁷. In this study, both the mRNA and the behavioral changes could be rescued by treatment with the glucocorticoid receptor antagonist, RU486. Glucocorticoid receptors are known to be expressed by a variety of cell types in the brain including neurons and RU486 treatment was noted to reduce stress-associated morphological remodeling of the PFC microglia as well as to prevent reduction of dendritic spine density in PFC layer V pyramidal cells¹¹⁷. These results complement other findings that stress-induced cognitive deficits trigger ramification of microglial processes and increase microglia-neuron contacts in the PFC leading to dendritic spine elimination^118–120. One molecular mechanism that links stress-induced PFC dysfunction with microglial activation has been proposed to involve ATP-gated P2X7 receptor signaling and the downstream release of PGE2 and IL-1β¹²¹. Extracellular ATP is generally subject to quick hydrolysis to adenosine, predicting a role for adenosine receptors in microglial response to stress. Indeed, blockade of adenosine A_2A receptors has been shown to normalize frontocortical microglia morphology in the prenatal dexamethasone model of anxiety in male, but not female rats¹²². Using the same model, normalization of microglial morphology by A_2A antagonism was observed in the hippocampus of female rats which was further linked to increased synchronization of hippocampal-PFC network and improved cognitive performance¹²³.

Some evidence suggests that release of TNFα from activated microglia improves cognitive flexibility in an adolescent social stress model and that administration of the monoamine oxidase inhibitor, antidepressant, ranylcypromine, increases the number of microglia and TNFα release in the PFC¹²⁴, a finding that suggests microglial activation may rescue, rather than promote, PFC-dependent cognitive impairment in some models and behavioral assays. Notwithstanding such evidence, human studies support deleterious effects of activated microglia on cognitive performance. This work has been greatly facilitated by the availability of radioligands to translocator protein (TSPO), a marker of activated microglia utilized in MRI and PET studies. For example, elevated TSPO distribution in the frontal cortex has been associated with lower attention scores in non-medicated patients with a major depressive disorder¹²⁵. Similarly, in HIV patients, increased TSPO radioligand binding in the frontal cortex was linked to lower performance on tests of memory and executive function¹²⁶. Increased microglial activation reported by TSPO binding has also been reported to predict cognitive decline in a model of experimentally-induced meningitis in rats¹²⁷ and in a population of pre-manifest Huntington’s disease patients¹²⁸. Although a specific role of PFC microglia was not investigated in these studies, they suggest that microglial activation can serve as a useful predictor of clinically manifesting symptoms of cognitive decline.

Modulation of reward-seeking behaviors

Regional heterogeneity of astrocyte responses to drugs of abuse –

The role of astrocytes may vary within distinct nodes of the reward circuit, including the PFC, caudate/putamen, nucleus accumbens (NAc), hippocampus, amygdala, and ventral tegmental area (VTA) as most prominent components. Indeed, while our understanding of astrocytes is still in its infancy, accumulating evidence indicates variability exists across brain regions with respect to the basal functional properties and engagement in pathology of these cells^28,129. While a direct comparison has not been made between cortical and striatal astrocytes, several lines of evidence indicate differential features and expression profiles between striatal and hippocampal astrocytes, supporting other evidence for regional variability of both structure and function^{21,28,130–133}. Accordingly, it is not unreasonable to speculate that effects of drug exposure may have dissociable effects within the reward circuitry, and that astrocytes within these regions may differentially influence behavioral responses to drugs.

Early reports reflecting effects of drug self-administration on astrocytes indicated that astrocyte-enriched mediators of glutamate homeostasis, in particular system xC- and EAAT2/GLT-1, are chronically downregulated in the NAc^134–137. It was subsequently revealed that self-administration and extinction of cocaine-seeking led to downregulation of GFAP expression, as well as astrocyte surface area, volume, and colocalization of astrocyte membrane with synapses in the NAc^138,139. This decrease in synaptic colocalization has since also been observed following self-administration and extinction from methamphetamine and heroin^140,141. The effect of cocaine on astrocytes in the NAc was, in contrast, not observed to extend to the prelimbic PFC or the basolateral nucleus of the amygdala, suggesting regional susceptibility of astrocytes in this model¹³⁹. However, decreased expression of EAAT1/GLAST and S100β-positive astrocytes has been observed in the SN/VTA following chronic non-contingent delivery of cocaine or methamphetamine¹⁴². Ethanol consumption in rats exhibits a time-dependent morphological effect on astrocytes in the mPFC, characterized by initial upregulation of GFAP followed by a decrease at 3 weeks of abstinence^143,144. Likewise, adolescent intermittent ethanol administration results in decreased synaptic colocalization of dorsal hippocampal astrocytes in adulthood¹⁴⁵. While many reports have found overall increases in GFAP expression and structural features of astrocytes following ethanol exposure, the general theme of reduced gene expression and structural features following post-administration abstinence has been observed across operant drug paradigms ^{for review, see: 144}.

How then do astrocytes affect drug-related behaviors? One prominent line of evidence comes from studies utilizing astrocyte-specific expression of G_q-coupled DREADD (hM3D) constructs. Stimulation of hM3D receptors within the murine PFC promotes ethanol drinking in ethanol-naïve mice in an intermittent access paradigm with free availability of ethanol every other day¹⁴⁶. However, different results are observed in other brain regions. When expressed in the NAc astrocytes, hM3D stimulation opposes cocaine reinstatement and motivation to self-administer ethanol, respectively^147,148. In the case of cocaine reinstatement, stimulation of presynaptic inhibitory mGluRs subsequent to hM3D stimulation was shown to be responsible for the observed behavioral effect¹⁴⁷. It is unclear whether these divergent results of DREADDs in the mPFC and NAc reflect heterogeneity of astrocyte signaling, or are a function of the stimulation in naïve versus withdrawn, abstinent animals. The parallel between distinct effects on astrocyte morphology and behavioral drug-related output, however, is noteworthy. Moreover, results from the VTA indicate that astrocyte regulation of avoidance behavior is dependent on local circuitry and stimulation of GABAergic neurons¹⁴⁹. Specifically, optogenetic stimulation of VTA astrocytes promotes avoidance behavior which overrides conditioned place preference for cocaine¹⁴⁹. Ventral midbrain astrocytes exhibit distinct gene expression and physiological properties which may contribute to the microcircuit-level outcomes²⁷.

Regional heterogeneity of astrocyte modulation of drug-related behaviors –

What is the mechanism by which striatal astrocytes could oppose drug-related, motivated behaviors? As addressed above, one possibility is stimulation of presynaptic, G_α and G_i/o-coupled metabotropic glutamate receptors¹⁴⁷. In addition, considerable evidence indicates that astrocytes can oppose glutamatergic synaptic transmission via release of ATP/adenosine, acting also on presynaptic inhibitory adenosine A1 receptors¹⁵⁰. For example, optogenetic activation of astrocytes in the hippocampus inhibited consolidation of contextual fear memory via release of ATP and adenosine and pharmacological stimulation of A1 receptors elicited similar effects¹⁵¹. In a similar fashion, glutamatergic synapses in the central nucleus of the amygdala can be inhibited by astrocytes via adenosine A1 receptor stimulation, while inhibitory synapses are activated via stimulation of adenosine A2A receptors¹⁵². Perhaps most germane, Corkrum et al.¹⁵³ recently showed that Ca²⁺ responses are generated in NAc astrocytes subsequent to dopamine release and amphetamine exposure, resulting in ATP/adenosine release and reduced AMPA receptor-mediated synaptic activity. This effect may be facilitated by increased glutamate-mediated neuroglial coupling in the NAc observed after cocaine self-administration⁷², although multiple other regulators of AMPA signaling^{e.g.154–156} could certainly be involved. Independent of drug exposure, Down Syndrome patient-derived induced pluripotent stem cells exhibit increased frequency of spontaneous calcium fluctuations, which inhibit co-cultured neurons via an adenosine-dependent mechanism¹⁵⁷. This is in keeping with a rich literature on varied effects of astrocytes on potentiation of inhibition of synaptic transmission dependent on subtype of adenosine or glutamate receptors^60,158–160. Thus, while significant evidence indicates that astrocytes can promote and support synaptic transmission^158,161 there is also considerable support for the hypothesis that astrocytes may oppose excitatory drive within the reward circuitry, and relatedly oppose behavioral output associated with drug craving and seeking¹⁶².

Striatal astrocytes in habit and compulsive behaviors –

Striatal astrocytes influence behaviors engaged by drugs beyond seeking. For example, both habit-directed and compulsive behaviors are associated with protracted drug seeking¹⁶³. The transition from recreational drug use to more regulated use and relapse is associated with a shift in cellular activity from the ventral to dorsal striatum¹⁶⁴. Relatedly, addiction is also associated with a shift from goal-directed, to more habit-based behaviors that are proposed to rely on recruitment of dorsomedial (DMS) and dorsolateral striatum (DLS), respectively^165,166.

Recent studies have indicated supportive roles for astrocytes in these processes. For example, chemogenetic hM3D activation of astrocytes in the DMS reduced the frequency of spontaneous neuronal excitatory post-synaptic currents (EPSCs) in D1-receptor expressing medium spiny neurons (MSNs), but increased EPSC amplitude in D2-receptor positive MSNs and shifted behavior from habitual to goal-directed¹⁶⁷. The effect of astrocyte G_q stimulation on promoting goal-directed behaviors was dependent on adenosine signaling, as the effect was not observed in adenosine transporter (ENT1) deficient mice. Considerable research has indicated both structural and functional interactions between A1-D1 and A2A-D2 receptors^168–170, underscoring the critical nature of adenosine in the regulation of synaptic and neuronal activity by astrocytes, and providing a mechanism whereby astrocytes may differentially affect different neuronal subtypes within the striatum. Precedent for specific interactions between astrocytes and neuronal subpopulations in the striatum has previously been reported⁴⁵. In contrast to these findings subsequent to astrocyte hM3D stimulation in the DMS, training-induced upregulation of EAAT2/GLT-1 in the DLS resulted in strengthened habit behavior in an operant task for chocolate reinforcers, and inhibition of EAAT2/GLT-1 upregulation promoted goal-directed behavior¹⁷¹. Collectively, these studies in the dorsal striatum suggest that astrocytes may reinforce the relative contributions of the DMS and DLS to habit and goal-directed behaviors, respectively.

Substance use disorders are characterized by three stages: 1) binge and intoxication, 2) withdrawal and negative affect, and 3) preoccupation and anticipation¹⁷². The third stage is associated with impaired PFC-dependent regulation of behavior and compulsive drug seeking. Accordingly, it is of merit to understand how astrocytes may contribute to compulsive-like behaviors. Removal of astrocyte Ca²⁺ signaling in the DLS via expression of the plasma membrane Ca²⁺ pump resulted in excessive and compulsive-like self-grooming behavior associated with increased GABA transporter, GAT-3, expression in astrocytes and impaired tonic inhibition of medium spiny neurons¹⁷³. These findings suggest that astrocytes may suppress dopamine release from axon terminals since compulsions manifesting as motor stereotypies are often observed in conditions characterized by excessive dopamine. Importantly, these results also highlight the ability of striatal astrocytes to modulate inhibitory as well as excitatory neuronal signaling echoing the findings in the frontal cortex.

Microglia and reward-seeking –

Microglia interact extensively with the dopamine system¹⁷⁴ and show sensitivity to drugs across a spectrum of substance use disorder models. In alcohol use disorder, ethanol has been found to increase the number of microglia in the hippocampus and the PFC in the absence of astrocyte activation¹⁷⁵. However, another study has found that although ethanol does induce microglial activation, microglial numbers in the hippocampus decrease after a “binge” pattern of exposure, finding that was supported by evidence of microglial dystrophy¹⁷⁶. Microglial activation has also been shown after exposure to psychostimulant drugs, including amphetamine¹⁷⁷, methamphetamine¹⁷⁸ and cocaine^179–183. Interestingly, methamphetamine-induced activation of microglia could be prevented by maintaining animals at reduced temperature (10–12°C), whereas elevated ambient temperature alone (38–40°C) did not result in microglial activation¹⁷⁸. Additionally, a PET study of human cocaine users using TSPO radioligands found no evidence of activated microglia¹⁸⁴. These somewhat discrepant findings could be reconciled by acknowledging that there is not a binary distinction between activated and resting microglia. For example, while there is converging support for the idea of microglial activation by ethanol using morphological analyses¹⁸⁵, binge ethanol exposure may result in increase of some, but not other markers of immunoreactive microglia¹⁸⁶. Such observations of partial microglial activation call for a more nuanced classification scheme. Indeed, the need for discriminating between multiple microglia phenotypes has been increasingly recognized and particularly aided by the emergence of rapid and cost-effective gene sequencing techniques¹⁸⁷.

In addition to morphological changes and expression of inflammatory markers following drug use, disrupting microglia has also been shown to have an effect on subsequent behavioral read-outs of drug exposure. Alcohol consumption and withdrawal-induced anxiety are both attenuated by inhibiting microglial activation with minocycline^188,189. Depletion of microglia with PLX-5622 does not change escalation of voluntary alcohol consumption, but does block escalation of intake under conditions of repeated immune activation¹⁹⁰. Another study has found that PLX-5622 normalized both excitatory and inhibitory synapse plasticity as well as prevented anxiety during alcohol withdrawal and reduced escalation of alcohol intake¹⁹¹. Consistent with this, microglial depletion normalized behavioral performance on marble burying and open field tests (measures of anxiety) after withdrawal from nicotine, an effect associated with elevation of NOX2, a source of reactive oxygen species expressed predominantly in the microglia¹⁹². Inhibition of microglia has also been generally reported to attenuate behavioral effects of psychostimulants. Thus, preventing microglia activation by blockade of toll like receptor 4 (TLR4) signaling and the downstream release of IL-1β attenuated conditioned place preference and self-administration of cocaine¹⁹³. Minocycline treatment has also been reported to attenuate both maintenance and reinstatement of methamphetamine-induced conditioned place preference^194,195 as well as subjective rewarding effects of dextroamphetamine in human subjects¹⁹⁶. Clinical studies generally support the idea that microglial inhibition suppresses behavioral effects of commonly abused drugs¹⁹⁷, although some results argue against this¹⁹⁸. In preclinical work, locomotor sensitization to cocaine appears to break the pattern with findings that sensitization is suppressed by activated rather than inhibited microglia via release of TNFα¹⁸³ and findings that microglial depletion has no effect on cocaine sensitization¹⁹⁹. It is possible that microglial control of substance-use related behaviors is proportional to the level of stress response that accompanies experimental manipulations. This has been suggested in a recent review²⁰⁰ and is supported by findings that immune activation is required for microglial control of drug-related behavior¹⁹⁰.

Astrocyte control of circadian rhythmicity

In mammals, a specific nucleus of the hypothalamus, the SCN, acts as the master circadian pacemaker. Its function is required to drive daily rhythms in behavior, and lesions of the SCN abolish rhythms in mammals. Accordingly, in hamsters, transplanting an explant of the SCN into an SCN-lesioned animal restores circadian rhythms²⁰¹. Of note, these rhythms have the same period as in the donor, indicating that they arise from cell-autonomous mechanisms present within the SCN²⁰¹. This form of cell-autonomous rhythmicity is generated by two interlocking transcription/translation feedback loops driven by four clock proteins: the activators CLOCK and BMAL1 and the repressors PER and CRY. CLOCK and BMAL1 activate the transcription of the Per and Cry genes. The proteins PER and CRY inhibit the transcriptional activation of CLOCK and BMAL1. As the proteins PER and CRY are degraded through ubiquitination, the repression of CLOCK and BMAL1 is relieved and the cycle begins again^202,203. The period and phase with which these reactions take place can be modified and reset or entrained by cycles of light and darkness in the environment. Light contributes to reset circadian clocks through the activation of retino-hypothalamic (RHT) glutamatergic inputs to the SCN that, according to studies of changes in the expression of the immediate early gene c-Fos, can activate both neurons and astrocytes^204,205.

The components of the molecular machinery regulating clock gene expression in mammals were first characterized in hamster neurons. Multiple groups hypothesized that astrocytes may also contribute and perhaps drive circadian rhythmicity in the SCN and experimental evidence showing that astrocytes express clock genes and function as competent circadian oscillators arrived with the use of reporter gene constructs^206,207. The identification of clock genes in astrocytes puts these cells under the spotlight and suggests that these cells are not mere modulators or targets of neuronal pacemakers, but they can act in conjunction with neurons to establish circadian rhythms²⁰⁶. One should note, however, that the level of expression of clock genes in astrocytes is substantially lower or more diffuse than in neurons, making it perhaps more difficult to detect²⁰⁶. This might have delayed the due recognition of astrocytes in the circadian pacemaker process, despite the early work demonstrating an essential role of astrocytes for the proper function of the circadian clock^208–211. There is now a growing awareness that just like the number of a given cell type does not speak to the relevance of its function, the lower expression of clock genes in astrocytes is not a measure of their physiological relevance in the generation and maintenance of circadian rhythms.

How do astrocytes contribute to the pacemaker activity of the SCN? In 1993, Lavialle and Serviere first reported that, in hamsters, there are light-independent (circadian) changes in the spatial distribution of the astrocyte protein GFAP and in astrocyte morphology, suggesting that these cells undergo circadian structural remodeling²¹². Since astrocyte remodeling is associated with changes in glucose consumption^212,213, and astrocytes are main sources of glycogen, these findings have been thought to reflect the participation of astrocytes in the regulation of energy metabolism^212,214. However, due to their abundant expression of glutamate transporters²¹⁵, a retraction of astrocyte processes could also imply a longer time for glutamate clearance from the extracellular space²¹⁶. This, in turn, could alter the time course of glutamatergic transmission at RHT synapses onto the SCN without the need for altering glutamate transporter expression in these cells²¹⁷. The structural remodeling of astrocytes has been shown to account for circadian changes in the glial coverage of the soma and dendrites of VIP and AVP neurons in the SCN implicated with clock entrainment and pacemaker resetting, respectively^218,219. Despite these findings, there are still uncertainties about the implications that the remodeling of astrocytes in the SCN has for glutamate clearance, synaptic plasticity, and metabolic exchange²⁰⁸. What has emerged in recent years, however, is that SCN astrocytes are not mere followers but rather partners in crime with the neuronal circadian pacemaker machinery²²⁰. Neurons are metabolically active during circadian daytime. Astrocytes, instead, are active during circadian nighttime and suppress neuronal activity by regulating extracellular glutamate concentration and activation of GluN2C-containing NMDA receptors. More importantly, astrocytes in the SCN can instruct neurons without a competent molecular clock to initiate and sustain circadian patterns of activity²²¹.

Astrocytes and circadian clocks outside the SCN –

The first demonstration that astrocytes are competent circadian oscillators came from luciferase assay studies in rat cultures²⁰⁶. This work was important for several reasons that we would like to reiterate. First, it showed that the circadian period of astrocytes is genetically determined and differs between mice and rats, in parallel with period differences in the locomotor behavior of these species, potentially reconciling experimental differences detected when working with experimental preparations from different rodent species. Second, it showed that the expression of clock genes like Per1 is substantially lower or more diffuse in astrocytes than in neurons, potentially explaining initial difficulties in detectability of clock genes in astrocytes. Third, it showed that, in vitro, the circadian rhythms of astrocytes are not detected before the first week of culture. This is in stark contrast with the in vitro developmental timeline of cultured neurons, which display circadian rhythmicity earlier. This finding may provide insights into the different developmental profiles of circadian clocks in neurons and astrocytes in vivo²²². Last, and arguably most importantly, the cultured astrocytes used in this work did not derive from the SCN but from the cortex, suggesting that regions of the brain outside the SCN can act as auxiliary, ancillary or accessory circadian oscillators.

Accordingly, the core circadian molecular machinery of astrocytes has been characterized in several regions of the nervous system (e.g., retina, olfactory bulb and hippocampus) and even outside the brain (e.g., cultured fibroblasts, hepatocytes, leukocytes, adipocytes, and muscle cells)^223–233. Although cells in these tissues can display intrinsic circadian rhythms, each of them would run at its own pace, desynchronize, and dampen population rhythms without SCN activity. This contrasts with what happens in the SCN, where neurons and astrocytes synchronize to each other^234–241. Several studies have attempted to reveal the identity of paracrine signals capable of synchronizing neuronal and astrocytes rhythms in auxiliary circadian oscillators (i.e., regions that express clock genes, but follow the timekeeping system set by the SCN). One of the first ones to be identified was the peptide hormone VIP, which coordinates circadian rhythms among astrocytes where VIP is released and where its receptor VPAC2R is expressed, like the olfactory bulb, retina, and the neocortex^242–245. Although the VIP/VPAC2R signaling pathway dominates circadian rhythmicity, other signaling pathways like the AVP/V1_a/b and the GRP/ BB2 can also sustain daily cycling²⁴⁶.

Astrocytes and circadian clocks in the hippocampus –

Clock genes are expressed in the hippocampus, a brain region involved in encoding spatial and episodic memories²⁴⁷. Recent work has shown that not only clock genes, but also a larger set of hippocampal genes (10%) and proteins (11%) are expressed with circadian rhythmicity (cf. 19% in the SCN)²⁴⁸. These oscillations in the molecular landscape of the hippocampus are altered in experimental temporal lobe epilepsy²⁴⁸, which is of particular interest given that seizure onsets within individuals display strong circadian rhythmicity²⁴⁹.

In vivo behavioral studies indicate that lesioning the SCN eliminates circadian rhythmicity in a passive avoidance task²⁵⁰, and desynchronization of the circadian system impairs recall of a spatial task²⁵¹. More specific tests on hippocampal-dependent learning and memory show that these vary between the circadian daytime and nighttime^252–257. The work of Chaudhury et al. is particularly important in this context as it shows that there is a circadian regulation of memory acquisition, recall and extinction in mice tested using a fear conditioning protocol²⁵⁸.

These periodical variations in hippocampal function can also be detected using in vivo electrophysiology recordings, although there are some differences in the results reported in the literature. Early studies in the dentate gyrus showed that the field EPSP and population spike amplitude are larger during the active phase (nighttime for rodents)²⁵⁹. Potentiation of the field EPSP amplitude, but not of the population spike amplitude, was confirmed in a later study²⁶⁰. Others also showed that the dentate gyrus is more excitable during the inactive phase (daytime for rodents)²⁶¹. When describing these findings, it is important to keep in mind that the results of these works are sometimes described by comparing two clock times which can be misleading, since circadian rhythmicity is a phenomenon with a sinusoidal profile. The use of two points forces linearization which makes the comparison across datasets collected from different laboratories difficult, if not inaccurate, especially when the two time points are not twelve hours apart from each other (i.e., half the period of a circadian cycle) or when different labs use different reference times. This is important because it can lead to opposite conclusions about whether the described phenomenon increases or decreases during daytime/nighttime.

Perhaps surprisingly, there are circadian changes in hippocampal function that are retained in reduced slice preparations. These show that the incidence and magnitude of long-term potentiation (LTP) varies with opposite phase in the dentate gyrus versus hippocampal area CA1, and may therefore shape in different ways the activity of synapses in different domains of the same brain region²⁶². Until recently, the synaptic mechanisms accounting for this effect remained unknown. What has now emerged is that these circadian changes in synaptic plasticity are differentially regulated by neurons and astrocytes²⁶³. During the nighttime, when LTP is reduced in hippocampal area CA1, pyramidal cells reduce their surface pool of functional NMDA receptors²⁶³. At the same time, glutamate clearance from astrocytes becomes slower, due to a retraction of astrocytic processes enriched with glutamate transporters from synapses, similar to what has also been observed in the SCN²¹². Glutamate inactivates AMPA receptors, and the longer glutamate remains in the extracellular space, the longer it takes for these receptors to recover from inactivation²⁶³. This has implications for synaptic integration of EPSPs, which is impaired during nighttime in CA1 pyramidal cells. The diffusible agent mediating these effects is not d-serine, because there are no detectable circadian changes in the occupancy of the NMDA receptor glycine binding site^263,264. It is likely corticosterone, whose peak production time is during the active phase (i.e., nighttime for mice)^263,265. Accordingly, one can rescue the loss of NMDA receptors and synaptic integration in slices prepared during the nighttime by treating them with antagonists of NR3C1/2 mineralocorticoid and glucocorticoid receptors²⁶³. What is also notable from this work is the fact that the magnitude of the circadian changes in synaptic integration are frequency dependent and are most pronounced for stimulation frequencies that can be detected in vivo during exploratory behaviors^266,267. In other words, circadian rhythms do not alter hippocampal activity as a whole: rather, they modulate hippocampal-dependent learning while preserving memory recall.

Together, these findings suggest that in a subordinate oscillator like the hippocampus, the circadian changes in gene and protein expression may change the rules of synaptic plasticity, but the direction and extent of these changes vary not only across but also within brain regions, in cell-specific ways. On the other hand, different sets of cells can be active to mediate the same behavior at different times²⁶⁸. The ultimate effect of circadian clocks on more complex phenomena and behaviors may be activity-dependent, adding additional levels of complexity. Perhaps, our attempts to draw general rules on the functional properties of the brain, has led us to underestimate how dynamic this organ is and how profoundly our circadian clock can shape what we think we can always do well.

Astrocytes and sleep/wake cycles –

The earliest evidence that astrocyte activity impacts sleep homeostasis came from a 2009 publication showing that mice with astrocytic dnSNARE expression show decreased extracellular accumulation of adenosine, a molecule that in wild type mice promotes sleep and increases sleep pressure, also observed with sleep deprivation⁶³. These findings implicated astrocyte-derived adenosine and A1 adenosine receptor activation as the factors underlying these behavioral effects, because pharmacological, in vivo blockade of A1 receptors attenuated sleep pressure accumulation in these mice⁶³. The majority of extracellular adenosine is derived from metabolism of ATP and astrocytes have been recognized as a major source of extracellular ATP²⁶⁹. In addition to these findings, Papouin et al (2017) showed that d-serine levels oscillate in the hippocampus as a function of wakefulness, not of circadian rhythms²⁶⁴. This causes saturation of the NMDA receptor glycine binding site during wakefulness, that wanes down during sleep. Activation of Ca²⁺-permeable α7-nicotinic acetylcholine receptors via septal cholinergic afferents to the hippocampus is thought to be the trigger for d-serine release from astrocytes during wakefulness²⁶⁴. Consistent with these findings, recent Ca²⁺ fiber photometry data show that Ca²⁺ levels in astrocytes increase during wakefulness and decrease during REM sleep in different brain regions including the cortex, hippocampus, hypothalamus, pons and cerebellum^270–272. However, the magnitude of these oscillations and the detailed Ca²⁺ dynamics change between astrocytes in the: (i) cortex and hippocampus; (ii) hypothalamus and pons; (iii) cerebellum²⁷⁰. These findings suggest that astrocytes may contribute differently to the regulation of sleep and wakefulness, depending on the brain region²⁷⁰. The emerging evidence in support of brain region-specific patterns of astrocyte gene expression^{21,132,273,274} may help in identification of molecular drivers underlying such divergent functional and behavioral outcomes.

Microglia involvement in circadian rhythms –

Isolated microglia display intrinsic daily fluctuations of circadian clock genes, Bmal1, Rev-erb, Per1, and Per2²⁷⁵ with inflammatory factors, TNFα, IL-1β, and IL6 oscillating in phase with Per1 and Per2 and in anti-phase with Rev-erb²⁷⁶. Dysregulated Per1 and Per2 expression in aged animals has been associated with elevated TNFα and IL-1β mRNA expression²⁷⁵ linking circadian rhythms with evidence of microglial contribution to age-related neurodegeneration²⁷⁷. Moreover, daily microglial cytokine fluctuations may underlie increased sensitivity of neuroinflammatory response to stimulation by lipopolysaccharide during the resting (sleep) phase²⁷⁶ with the implication that therapeutic efficacy of inflammation-focused treatments may be improved by alignment to daily sleep/wake cycles. The link between microglial immunoreactivity and circadian rhythms is further borne out in studies that reduce or eliminate clock gene expression, although the direction of gene regulation differs between reports. For example, elevated levels of IL-1β and TNFα have been reported in striatal microglia of Bmal1 knock-out mice with further elevation after challenge with LPS or treatment with MPTH, a model of Parkinsonian neurodegeneration²⁷⁸. Elevated levels of IL-1β, IL-6 and TNFα have also been reported in the Rev-erba knock-out model²⁷⁹. However, in homogenized brain tissue, Bmal1 deletion was seen to decrease expression of IL-1β and inflammation-related gene, Nox2²⁸⁰, highlighting potential influence of region-specific differences in microglial response to Bmal1 deletion.

Microglia involvement in sleep/wake cycles –

It is well-documented and, indeed, accepted as common knowledge that disrupted sleep leads to a range of adverse psychological and physiological consequences. Evidence of microglial activation caused by sleep fragmentation²⁸¹ is consistent with this line of reasoning although it is possible that fragmented sleep may arise secondary to other conditions (e.g., depression, drug use) associated with microglial pathologies. Supporting the former contention, depleting microglia by activating the diphtheria toxin receptor under the Cx3cr1 promoter in a transgenic rat model, resulted in disrupted expression of clock genes Per1, Per2, and Bmal1 in the SCN and the hippocampus, abnormal circadian body temperature, and diurnal rhythm disruptions²⁸². Systemic ablation of microglia following PLX-5622 treatment was reported to increase sleep duration and eliminate light phase-dependent difference in synaptic transmission in mouse hippocampal CA1 pyramidal neurons²⁸³. In the healthy brain, microglial morphology fluctuates in rhythm with circadian cycles with longer processes and branching points during wakefulness, phenomena dependent on microglial expression of lysosomal cysteine protease cathepsin S and the adenosine diphosphate, P2Y12 receptors^284,285. Microglial morphology may also be generally sensitive to arousal states with increased arborization of processes and increased process motility observed following anesthesia relative to wakeful conditions²⁸⁶. Partial activation of microglia during sleep may contribute to synapse elimination, suggesting an intriguing mechanism for the role of sleep in memory consolidation²⁸⁷. Indeed, sleep deprivation-induced memory impairments on a Morris water maze could be rescued in rats treated with minocycline²⁸⁸. Minocycline treatment has also been reported to increase slow-wave activity during sleep and improve episodic memory in human subjects²⁸⁹, although others have found sleep disruption by minocycline in both human²⁹⁰ and rodent^291,292 studies. Notably, minocycline effects on sleep are unlikely due to its antibiotic activity as ampicillin treatment was not seen to impact sleep homeostasis²⁹⁰.

Concluding remarks

The emerging picture of structural and functional diversity of glial cells across distinct brain regions suggests co-existence of multiple mechanisms by which glial cells regulate neuronal activity. The diversity of mechanisms likely reflects unique physiological and computational demands placed upon circuits underlying specific behavioral outputs (Table 1). Within brain regions, dynamic glial cell responses to external perturbations, circadian rhythms or other internal changes are echoed by the flexible, adaptive nature of many behaviors. Understanding the functional roles of a molecularly diverse populations of glial cells will advance not only our knowledge of the mechanisms by which these cells control executive function, reward seeking, and circadian function, but will also shed light on phenomenology of a variety of neurodegenerative, substance use, and other disorders. The unique molecular signatures of glial cells make them attractive candidates for development of new therapeutics. Future studies utilizing single cell transcriptomics may shed light on region- and cell-type selective targets for use in both psychiatric and neurological disorders.

Highlights.

• Astrocytes and microglia display diverse properties across brain regions

• Synaptic and neuronal circuit activity are highly sensitive to astrocyte and microglial signals

• Higher cognitive function, reward-seeking, and circadian behaviors are associated with structurally and functionally dynamic responses in glial cells