2.1 Norepinephrine

Various lines of evidence report that systemic injections of NE enhance long-term retention of memory when administered shortly after training (Colucci et al., 2019; Costa-Miserachs et al., 1994; Gold & van Buskirk, 1975, 1978; Liang et al., 1986; McGaugh & Cahill, 1997; Roozendaal et al., 2009). Similar effects were found in object recognition using yohimbine, a selective presynaptic α2-adrenoceptor antagonist, that enhances NE outflow (Roozendaal et al., 2006). Consistently, pharmacological depletion of NE impairs memory consolidation in the two-way shuttle box avoidance task (Archer et al., 1984), spatial water maze (Cahill et al., 1994), and inhibitory avoidance discrimination test (Atucha & Roozendaal, 2015; Cahill et al., 1994).

Mechanistically, NE enhances long-term potentiation (LTP)—one of the major cellular mechanisms that may underlie learning and memory processes (Bliss & Collingridge, 1993; Cooke & Bliss, 2006; Kessels & Malinow, 2009; Nabavi et al., 2014)—in both hippocampus (Huang & Kandel, 1996; Jȩdrzejewska-Szmek et al., 2017) and amygdala (Tully et al., 2007). Hu et al. (2007) reported that NE enhances phosphorylation of GluA1 and lowers the threshold for LTP and memory formation. Moreover, NE primes synapses for future changes in long-lasting plasticity (Maity et al., 2015).

2.2 Glucocorticoid hormones

GCs, via activation of MRs, are involved in appraisal and response selection in novel contexts (Oitzl & de Kloet, 1992; Sandi & Rose, 1994). Moreover, activation of MRs mediates effects of stress on memory systems, promotes habitual processes both in rodents and humans (Arp et al., 2014; Schwabe et al., 2010, 2013; Vogel et al., 2015, 2017), and is involved in spatial and fear-related memory processes (Berger et al., 2006; Joëls et al., 2018; Oitzl & De Kloet, 1992; Zhou et al., 2010). Activation of GRs is critical for an enhanced memory consolidation that is observed with elevated GC levels (Chen et al., 2012; Donley et al., 2005; Hui et al., 2004; Krugers et al., 2011; Oitzl et al., 2001; Pugh et al., 1997; Revest et al., 2005, 2010, 2014; Roozendaal & McGaugh, 1997; Sandi & Rose, 1994; Xiong & Krugers, 2015; Zhou et al., 2010). At the cellular level, activation of MRs, within minutes, enhances the frequency of hippocampal mEPSCs, enhances the mobility of GluA1/2 AMPARs, and retention of GluN2B containing N-methyl-D-aspartate receptors (NMDARs) (Groc et al., 2008; Karst & Joëls, 2005; Mikasova et al., 2017). At a longer time-scale, GCs, acting through GRs, enhance the lateral diffusion of AMPARS, hippocampal synaptic retention of AMPARs (Groc et al., 2008; Martin et al., 2009; Xiong & Krugers, 2015), and enhance the amplitude of miniature excitatory postsynaptic currents (mEPSCs) (Karst & Joëls, 2005; Xiong & Krugers, 2015; Xiong et al., 2016). Prolonged exposure to stress and GCs hampers learning and memory (Conrad et al., 1996; Krugers et al., 1997; Wei et al., 2014), synaptic plasticity (e.g. Krugers et al., 2006; Wei et al., 2014), and reduces prefrontal synaptic transmission by reducing neuronal glutamate receptor expression (Yuen et al., 2012). Together, the effects of glucocorticoids on synapses - at least in part - mediate the regulation of stress on learning and memory, although the exact molecular trajectories between receptor activation and functional outcome remain largely unknown (Liu et al., 2010; Wei et al., 2014; Yuen et al., 2009, 2011).

2.3 (Nor)epinephrine–glucocorticoid interactions

NE and GCs also in synergy, modulate neuronal function and behavior (Hermans et al., 2014; Joëls et al., 2011). For example, metyrapone (a GC synthesis inhibitor) attenuates the memory-enhancing effects of NE (Roozendaall et al., 1996), while β-adrenergic receptor activation is required for GCs to modulate memory storage processes (McReynolds et al., 2010). Similarly, corticosterone enhances object recognition memory, an effect that is also impeded by β-adrenoceptor antagonist treatment (Okuda et al., 2004). These findings strongly suggest that adrenergic activation is essential for enabling GC enhancement of memory consolidation.

At the cellular level, NE and GCs interact to modify AMPAR function (Zhou et al., 2012). Corticosterone, at a dose that activates both MRs and GRs, rapidly increases AMPAR surface expression - a postsynaptic index of glutamatergic transmission –particularly, when combined with moderate doses of the β-adrenergic receptor agonist isoproterenol. Moreover, the combination of both hormones enhances mEPSC frequency, which reflects enhanced glutamatergic transmission. Accordingly, co-application of GCs with isoproterenol led to a significant enhancement of the early, but not the later LTP phase (Pu et al., 2009). On the contrary, brief administration of GCs several hours before application of isoproterenol markedly suppressed the efficacy of isoproterenol to potentiate the later phase of LTP (Pu et al., 2009). While the studies mentioned here only reflect a summary of the extensive literature, there is substantial evidence that activation of the autonomic nervous system and HPA axis, alone as well as together, modify neuronal function and behavioral adaptation to stressful experiences: NEs and GCs promote behavioral adaptation to stressors by enhancing learning and memory processes, while prolonged (chronic) exposure to stress and stress hormones hampers learning and memory, presumably by altering synaptic function.

3.1 Microglia express stress-hormones receptors

Tanaka et al. (1997) demonstrated the expression of GRs and MRs in microglial cells by immunoblotting. Sierra et al. (2008) confirmed the presence of these receptors by real-time PCR in isolated hippocampal microglia using Fluorescence Activated Cell Sorting (FACS). These studies suggest that GRs are more abundant than MRs in microglia at both mRNA and protein level. In functional terms, the deletion of GR in microglia reduces microglial motility, increases amoeboid morphology, and elicits greater neuronal damage and demyelination after intraparenchymal injection of lipopolysaccharide (LPS) (Carrillo-de Sauvage et al., 2013). In contrast, activation of GRs in a model of Alzheimer's disease increased the pro-inflammatory marker (cluster of differentiation) CD68, microglial density per se, as well as activation in the hippocampal CA1 area (Pedrazzoli et al., 2019). Interestingly, other conditions with increased GCs levels, such as aging, have been associated with a decrease in microglial complexity (van Olst et al., 2018). Specifically, enhancing GCs levels in young mice enhanced microglial ramifications, while the specific GR knockdown in old mice aggravated age-associated microglial amoebification in the hippocampus. This suggests that microglial GR regulates essential microglial properties during inflammation, and have a key role in neuronal survival and function.

Receptors for NE are also present in microglial cells. Mori et al. (2002) demonstrated, via real time-PCR, that microglial cells express mRNAs Fα1A, α2A, β1, and β2 adrenergic receptors. However, its expression depends on its activity state. Analysis of adrenergic receptor expression with quantitative PCR indicated that resting microglia primarily express β2 receptors but switch expression to α2A receptors under pro-inflammatory conditions modeled by LPS (Gyoneva & Traynelis, 2013). In both conditions, treatment with NE caused a retraction of microglial processes by blocking the ATP-induced process extension and migration in vitro (Gyoneva & Traynelis, 2013). In addition, selective stimulation of adrenergic receptors reduced the release of pro-inflammatory mediators such as nitric oxide, IL-6, and TNF- α (Färber et al., 2005). Similarly, β2-adrenergic stimulation after stroke caused an enlarged morphology of microglia, impaired microglial proliferation, and reduced expression of TNFα (Lechtenberg et al., 2019). Hence, microglia may be kept quiescent and in a non-inflammatory state via NE release. This is key as deficiency of NE under pathological conditions may then facilitate an inflammatory state that can impair normal learning and memory processing and even prevent microglial phagocytosis of amyloid plaques (Heneka et al., 2010).

3.2 Microglia and memory

A role for microglia in learning and memory processes is suggested by different lines of evidence. Torres et al. (2016) depleted microglia (a) locally by injecting clodronate in the hippocampus or (b) systemically using a Csf1R inhibitor (thus, blocking the cytokine Csf—which involved in microglial growth and differentiation). The depletion resulted in an impaired performance in the Barnes maze that was rescued after microglial repopulation 20 days after clodronate treatment. Chaaya et al. (2019) reported an increase in microglial numbers and a change in microglial morphology after fear conditioning. Examination of microglial morphology within the DG revealed altered branching, with processes with significantly shorter and less complex extensions. Microglia also play a role in the maintenance of dendritic spine densities in the adult brain, which is relevant for learning and memory (Hayashi-Takagi et al., 2015). Depletion of microglia via diphtheria toxin showed a decrease in learning-induced synaptic remodeling (formation and elimination of hippocampal dendritic spines) related to deficits in multiple learning tasks (Parkhurst et al., 2013). Recent evidence suggests that microglia are also involved in forgetting. Wang et al. (2020) used CD11b-DTR mice, in which diphtheria toxin receptor (DTR) is specifically expressed in CD11b-expressing myeloid cells, including microglia (Prinz et al., 2011). By administering diphtheria toxin (DT), they depleted the microglia in the hippocampus and showed that these mice showed significantly higher freezing levels at the retrieval phase (35 days after training) and confirmed this by targeting microglia with a Csf1R inhibitor. These results show that microglia are required to mediate the forgetting of remote memories. Finally, several lines of evidence in humans (using, e.g., PET imaging) suggest that activation of microglia is related to cognitive alterations (Cagnin et al., 2001; Eckenweber et al., 2020; Fan et al., 2017; Hamelin et al., 2016).

3.3 Microglia and synapses

During development and learning experiences, morphological maturation of microglial cells takes place in close proximity to the period of spine formation, which may suggest that microglia are actively involved in the synaptogenesis of neuronal circuits (Dalmau et al., 1998; Matcovitch-Natan et al., 2016; Reemst et al., 2016). Also, microglia are canonically associated with synaptic pruning in physiological (Schafer et al., 2012) and pathological (Sellgren et al., 2017; Stephan et al., 2012) conditions. Moreover, microglia have been shown to mediate synaptic loss in Alzheimer models (Hong et al., 2016) and to show distinct patterns of activation in response to amyloid and tau in Alzheimer's disease (Gerrits et al., 2021). Moreover, microglia may participate in functional synapse maturation through the modulation of excitatory synaptic transmission. In vitro studies have demonstrated that conditioned media from cultured microglia potentiated the amplitude of NMDA-induced currents and could enhance LTP in synapses of the Schaffer collateral pathway (Hayashi et al., 2006). Consistent with this, LTP was reduced in organotypic hippocampal slices prepared from Cx3cr1KO mice—which have impaired proper neuron–microglial communication (Rogers et al., 2011). Using the same animal model (Parkhurst et al., 2013), further demonstrated a decline in the frequency of NMDAR- and AMPA-mediated miniature excitatory postsynaptic currents (mEPSCs). Moreover, administering minocycline (a specific inhibitor of the pro-inflammatory response in microglia—Garrido-Mesa et al., 2013) could partly restore the decreased induction of LTP in hippocampal pyramidal cells after stress, which indicates a role of pro-inflammatory microglia in the impairment of spatial learning and memory in the Morris water maze (Liu et al., 2015). Finally, targeting microglia with a mutation of KARAP/DAP12—selectively expressed in microglia in the brain—leads to an altered synaptic phenotype in hippocampus that involves reduced synaptic recruitment of GluR2 that triggers an NMDAR-independent form of LTP (Roumier et al., 2004).

4.1 Number and morphology of microglia

Tables 1 and 2 show the effects of acute (Table 1) and chronic stressors (Table 2) on microglia number and morphology. As shown in Table 1, acute stress, in general, does not alter microglia number or morphology in the hippocampus, although Sugama et al. (2007) report an increase in the microglial surface area after acute stress exposure. Chronic stress exposure, on the other hand, increases the number and complexity of microglia. This has consistently been shown using different microglial markers such as ionized calcium-binding adapter molecule 1 (Iba-1), also known as allograft inflammatory factor 1 (AIF-1) (Ito et al., 1998), CD68 (Walker & Lue, 2015), CD11b, and CD45 Greter et al., 2015), ED-1 Graeber & Streit, 2010), and translocator protein (TSPO) (Beckers et al., 2018). It is important to mention though that some of these markers are also expressed by peripheral immune cells, and results have to be carefully interpreted in conditions when the blood–brain barrier is disrupted (Grassivaro et al., 2020), which may occur after chronic stress (Lehmann et al., 2018). In addition, chronic stress alters the morphology of microglia. In general, chronic stress induces microglia to enter a reactive phase characterized by enlarged somas, and retraction and thickening of processes, with an excess of short and thick hyper-ramified processes which give them a bushy appearance. These hypertrophic microglia are functional cells that can exert increased response to pro-inflammatory stimuli (Perry & Holmes, 2014), by increasing their pro-inflammatory cytokines expression. Moreover, the increased number of distal processes is often associated with exacerbated interactions with neurons and synapses (Šišková & Tremblay, 2013).

Mechanistically, there is evidence that these changes in microglial number might be mediated via stress-induced elevation in stress hormones. Nair and Bonneau (2006) demonstrated that blocking GC synthesis using metyrapone reversed the increase in hippocampal microglia number found on the fourth day of a chronic restrain stress paradigm. In the same way, animals treated with the GR antagonist RU486 prior to the stressor did not exhibit the same increase in microglia number when restrained. Accordingly, NE could be also involved, as Sugama et al. (2019) probed that the increase in microglial number and change in morphology (enlarged with shorter processes and enlarged soma) after restraint stress were significantly inhibited by pre-treatment with the β-blocker propranolol. This was confirmed in a double knockout for β-1 and β-2 adrenergic receptors. However, the specific mechanisms underlying these effects remain unknown. Chronic stress not only increases the microglial population in the hippocampus and promotes a shift toward a more inflammatory phenotype, but it also promotes peripheral recruitment and migration of peripheral macrophages and/or bone marrow–derived monocytes into the brain. Chronic stress exposure, e.g., potentiates the influx of bone marrow–derived cells by extravasation into the brain parenchyma of the ventral hippocampus and differentiation into microglia (Brevet et al., 2010). Finally, in contrast to chronic stress in adulthood, stress during the early postnatal period affects microglia in a different manner when measured in early stages. Réus et al. (2019) demonstrated that maternal deprivation decreases the level of Iba-1 immunolabeling in the hippocampus at PND10, which then increases at PND20 and remains elevated from PND 30 onwards. In agreement, Hoeijmakers et al. (2017) showed reduced coverage of Iba1+ cells in the dentate gyrus at P9 in a model of fragmented maternal care, with specifically reduced complexity of the most distal branching of microglia of the stressed exposed offspring. At 4 months of age, increased Iba1 density and CD68 immunoreactivity were found (Catale et al., 2020; Hoeijmakers et al., 2017; Reemst et al., 2016). Together, these studies show that microglial sensitivity to chronic stress differs in its responses at different ages of exposure.

4.2 Gene expression

Table 3 (acute stress) and Table 4 (chronic stress) summarize the changes in microglia-related gene expression after stress. It is important to note that in some cases, such changes in, e.g., cytokine expression, might be mediated by other cell types, and are likely not always exclusively related to microglia. Several studies have further used minocycline (Han et al., 2019; Wadhwa et al., 2017; Wang et al., 2017; Zhao et al., 2015), and reported a reversal in specific gene expression changes when microglia could not be activated, thereby identifying microglia-specific effects.

4.3 Microglial–neuronal crosstalk

The crosstalk between microglia and neurons is important for neuronal function as discussed before (Boxes 1 and 2). Evidence now suggests that this interaction, and in particular Cx3CL1-Cx3Cr1 signaling, may be affected by stress. Firstly, NE output via reboxetine (a norepinephrine reuptake inhibitor) stimulates the production of CX3CR1 (González-Prieto et al., 2020). Secondly, deletion of Cx3cr1 prevented the stress-effects on microglial morphology and phagocytosis of cellular elements (Milior et al., 2016), In addition, Cx3cr1 deficiency has profound defects in synapse elimination following sensory lesioning (Gunner et al., 2019). This leads to deficits in glutamatergic transmission, as well as short- and long-term neuronal plasticity, highlighting the importance of this crosstalk for memory processing. In particular, CA1 synapses of Cx3cr1 KO show a decrease in AMPA/NMDA ratio (pointing to immature synapses) and a defective AMPA receptor EPSCs (Basilico et al., 2019). Cx3cr1-deficient mice also exhibit increased fear acquisition as well as retrieval (Schubert et al., 2018) and increased freezing during the reinstatement, suggesting that the retrieval of fear was potentiated in these mice. This may be particularly relevant considering recent findings on the role of microglia in the process of forgetting (Wang et al., 2020). In summary, neuron–microglia signaling through CX3CR1 is necessary for synaptic function and prides a potential connection among microglia, stress, and memory.

4.4 Synaptic transmission

As discussed before, in particular, chronic stress alters microglial function and this may mediate stress-effects on learning and memory. Microglial cells may also mediate stress-effects on synapses. Liu et al. (2015) demonstrated that chronic unpredictable stress in adult rats leads to a pro-inflammatory phenotype in hippocampal microglia in relation to the attenuated phosphorylation of glutamate receptor GluA1, an effect that is inhibited by minocycline treatment. In addition, minocycline prevented stress-effects on LTP induction (Liu et al., 2015). Similarly, rats that underwent social isolation during adolescence showed a downregulation of the expression of hippocampal NMDAR subunit NR1 and AMPAR subunits GluA1 and GluA2, which were restored after minocycline treatment (Wang et al., 2017). These data suggest a role of microglia in synaptic plasticity through regulation of AMPA/NMDA signaling after stress exposure. Interestingly, the relation between microglia and NMDA receptors is bidirectional. Nair and Bonneau (2006) demonstrated that chronic stress induces microglial proliferation which was mediated by GC regulation of NMDARs. Recently, studies have shown that NE is involved in the role of microglia in modifying synaptic contacts, by increasing the microglia–dendrite contact area and contact duration (Liu et al., 2019; Stowell et al., 2019). Taken together, there is evidence that prolonged stress might alter synaptic mechanisms via changes in microglia.

4.5 Kynurenine pathway

In the brain, tryptophan is converted into several bioactive molecules, the best known of which is serotonin. However, only a small percentage of tryptophan is metabolized into serotonin. More than 95% of tryptophan is converted into kynurenine in a process known as kynurenine pathway. This pathway is present in many different tissues, notably in the liver and brain. In the brain, microglia play a critical role in producing the enzyme indoleamine 2,3-dioxygenase (IDO), an enzyme that catalyzes the conversion of tryptophan into kynurenine (Savitz, 2020). Kynurenine is further metabolized to, among others, quinolinic acid (Schwarcz et al., 2012). The role of kynurenine metabolites in memory is mainly (either direct or indirect) via the regulation of neurotransmitter systems. For example, kynurenine and quinolinic acid have opposing effects on NMDAR function (Perkins & Stone, 1985). Quinolinic acid contributes to the activation of NMDAR itself and by inhibiting glutamate re-uptake, resulting in excessive extracellular glutamate, inducing further NMDAR agonism (Maddison & Giorgini, 2015). Also, quinolinic acid has been shown to stimulate the release of glutamate from neurons, inhibiting its re-uptake by astrocytes (Jo et al., 2015). Overall, Rahman et al. (2018) have demonstrated that intraventricular quinolinic acid infusion results in significant impairment of learning in adolescent animals.

There is growing evidence for kynurenine pathway hyperfunction in stress-related disorders (Savitz et al., 2015). Brooks et al. (2016) showed that GCs induce amplified the ability of interferon-γ to increase Ido1 expression—and, in consequence, shifting the route toward more kynurenine production—in organotypic hippocampal slice cultures. Also, both GR (using the agonist dexamethasone) and MR (using the agonist aldosterone) are involved in this increase. Vecchiarelli et al. (2016) have demonstrated that acute restraint stress increases Ido1 expression in the amygdala, promoting the production of excitotoxic NMDAR agonist quinolinic acid, and thus potentially increase excitability. Similar findings were found after chronic stress exposure. Fuertig et al. (2016) demonstrated that chronic social defeat results in higher kynurenine levels in both blood and brain (specifically in memory-related structures such as amygdala and hippocampus), related to an increase in freezing behavior during fear conditioning. On the other hand, IDO inhibition did reduce chronic social defeat-induced fear memory such that the extent of fear expression got close to that of non-stressed animals. In concordance, Réus et al. (2019) demonstrated that maternal deprivation causes a decrease in mRNA Ido expression in the hippocampus at PND10. In conclusion, kynurenine metabolites—for whose production microglia is key—are overexpressed after stress exposure and might affect memory processing via altered glutamate transmission.

5.1 Do stress hormones alter learning and memory via microglia?

The information collected in this review points toward the possibility that microglia might be playing a determinant role in stressful memories, although the underlying mechanisms are still unknown. As microglia are known to express receptors of both main stress hormones - GCs and NE (Mori et al., 2002; Sierra et al., 2008; Tanaka et al., 1997) - animal models lacking these receptors specifically in microglia (Carrillo-de Sauvage et al., 2013) will be critical to understanding how these hormones/transmitters affect synaptic structure and function, and memory processes such as consolidation, retrieval, and extinction, through modulation of microglial function. As stress-effects on both microglia and memory seem to depend - among others - on the duration of exposure, it will be important to understand whether/how microglia are involved in effects of acute versus chronic stress exposure on memory formation.

5.2 Are microglia involved in sex differences in stress-effects on learning and memory?

Effects of (acute and chronic) stressors on memory are often found to be sex dependent (Bowman et al., 2009; Foy et al., 1999; Vouimba et al., 2000; Warren et al., 1995; Woolley & McEwen, 1992, 1993). A potential role for microglia will be particularly relevant given that (a) several microglial processes, such as maturation, have been reported to be sex dependent (Thion et al., 2018), (b) microglia are affected by stress in a sex-specific manner (Bollinger et al., 2017), specifically in corticolimbic regions, and (c) microglia express receptors not only for stress hormones—which in the case of GR is more abundant in females (Sierra et al., 2008)—but also for estrogen and progesterone (Sierra et al., 2008),. Therefore, sex steroids may exert a modulatory effect on stress-induced microglial cell activation and memory. In fact, some of the molecular properties of microglia in the hippocampus are also differentially modified by sex including gene expression of several neuronal crosstalk markers and cytokines that are also crucial for memory (Bollinger et al., 2017). This might be crucial to understand the sex differential vulnerability to some memory-related stress conditions, such as post-traumatic stress disorder (PTSD; Shansky, 2015).

5.3 Role of microglia in effects of early life stress on learning and memory

Stress early in life is widely known to affect memory and synapses (e.g. Brunson et al., 2005; Lesuis et al., 2019; Liu et al., 2000; Naninck et al., 2015; Oomen et al., 2010). Evidence suggests that these effects in this specific time window may as well be mediated by microglia. Shortly after early life adversity, coverage of microglia appears to be reduced, while later in life the coverage is enhanced (Hoeijmakers et al., 2017; Réus et al., 2019, but see also Delpech et al., 2016 and Roque et al., 2016). As microglia are involved in synapse pruning (Paolicelli et al., 2011) and developmental apoptosis (Wakselman et al., 2008) in the hippocampus during the neonatal period, while also contributing to neurogenesis and neuroprotection in the developing brain (Cowan & Petri, 2018; Rodríguez-Iglesias et al., 2019; Sierra et al., 2010), alterations in microglia function may have long-lasting consequences for brain structure, neuronal function and memory over the lifespan. In line with this, it has been reported that transient depletion of microglia from PND0 to PND14 is accompanied by later working memory deficits and an altered expression of fear, anxiety-like and risk-assessment behaviors (VanRyzin et al., 2016). It will therefore be important to investigate in detail the relationship among early life experiences, microglia, and the sensitivity to stress and memory later in life (Desplats et al., 2019; Weber et al., 2019).