3.1. Epilepsy-associated astrogliosis

Reactive astrogliosis is present in almost all forms of epilepsy, but it is most notable in presence of hippocampal sclerosis (HS), which is often associated with MTLE and other epilepsy syndromes (Thom, 2014). Indeed, besides a severe loss of principal neurons observed in CA1 and CA3 and granule cell dispersion, HS is characterized by a chronic and fibrillary gliosis in CA1 and radial gliosis in the dentate gyrus, where the length of GFAP⁺ fibres is directly correlated with the degree of cell dispersion in the dentate gyrus (Fahrner et al., 2007). Furthermore, in HS, together with increased conventional GFAP expression, a novel GFAP isoform has been identified in small multinucleate CA1 and CA4 astrocytes, GFAP-γ, which is speculated to regulate astrocyte size and motility (Martinian et al., 2009). Whether HS is a primary cause of epilepsy or the result of repeated epileptic seizures is still controversial. Even if the prevailing view tends to consider HS as a secondary consequence of epilepsy, experimental data on surgical samples and autoptic tissues suggest that HS aetiology is multifactorial. Febrile seizures, genetic susceptibility, alterations of hippocampal development, head injuries, infections and inflammatory and neurodevelopmental factors have indeed been identified as predisposing elements to HS development (Sendrowski and Sobaniec, 2013, Thom, 2014, Walker, 2015).

3.2. Kir channels and K⁺ homeostasis

It is well known that astrocytes are key players in the regulation of extracellular K⁺ ([K⁺]_o), which can transiently accumulate during prolonged neuronal activity and cause neuronal depolarization and hyper-excitability if uncompensated (Heinemann and Lux, 1977). [K⁺]_o homeostatic control is performed by K⁺ uptake and by spatial K⁺ buffering: while the former is mediated by glial and neuronal Na,K-ATPase or Na-K-Cl cotransporters, the latter is driven by the difference between the glial syncytium negative membrane potential and the local K⁺ equilibrium potential. This results in redistribution of K⁺ from sites of high neuronal activity to sites of lower [K⁺]_o through gap junction (GJ)-connected astrocytic networks (Orkand et al., 1966, Walz, 2000). This peculiar astrocytic property is mainly mediated by Kir4.1 K⁺ channels, which are weakly-rectifying channels possessing a high open probability at rest and showing conductance increase at high [K⁺]_o (Butt and Kalsi, 2006). Considering their role in K⁺ homeostasis and since increased levels of [K⁺]_o have been associated to the pathophysiology of epilepsy, Kir channels have been investigated in experimental (see (Chever et al., 2010, Djukic et al., 2007)) and human epilepsy. By using K⁺-sensitive microelectrodes, measurements of stimulus-induced [K⁺]_o increases have been performed in human sclerotic and non-sclerotic hippocampal tissues (HS and non-HS) in presence of Ba²⁺, a blocker of Kir channels (Kivi et al., 2000). These recordings highlighted an impaired K⁺ buffering in HS slices: indeed, they showed that Ba²⁺ was able to induce [K⁺]_o accumulation in hippocampal slices from lesion-associated TLE patients with no histopathological hippocampal alterations; on the contrary, it failed to do so in hippocampi from drug-resistant TLE patients showing strong neuronal loss and gliosis (Gabriel et al., 1998, Heinemann et al., 2000, Jauch et al., 2002, Kivi et al., 2000). Furthermore, patch-clamp recordings and single cell RT-PCR performed on human sclerotic hippocampus revealed a significant decrease (Hinterkeuser et al., 2000, Schröder et al., 2000) or a complete loss (Bordey and Sontheimer, 1998) of Kir4.1 currents; this has also been recently supported by Western blot and immunohistochemistry analysis showing a decrease of Kir4.1 protein levels in HS tissues compared to non-HS TLE patients or sudden-death controls (Das et al., 2012, Heuser et al., 2012). Thus, in HS TLE, the reduced expression and functionality of Kir4.1 channels determine an impaired K⁺ buffering and enhance seizure susceptibility, even if it is still controversial whether this reduction is a cause or a consequence of TLE.

Interestingly, several variations in human Kir4.1 gene (KCNJ10) have been identified and associated to different seizure phenotypes. Indeed, Buono and colleagues found that patients with refractory MTLE, childhood absence and juvenile myoclonic and idiopathic generalized epilepsy carried a missense mutation in the C-terminal domain of KCNJ10. Such domain is involved in ionic conductance, channel subunit dimerization and anchoring to the plasma membrane (Buono et al., 2004). Furthermore, other single nucleotide variations in non-coding regulatory KCNJ10 sequences have been found in TLE patients with febrile seizures (Heuser et al., 2010). Moreover, patients with EAST/SeSAME syndrome, an autosomal recessive multiorgan disorder characterized by seizures, ataxia, sensorineural deafness, mental retardation and electrolyte imbalance, presented loss-of-function mutations in highly conserved amino acid sequences of Kir4.1 gene, which causes impairment in channel conductance and Ba²⁺-sensitivity (Bockenhauer et al., 2009, Reichold et al., 2010, Scholl et al., 2009, Williams et al., 2010). Another cohort of patients displayed instead a gain-of-function mutation in KCNJ10 affecting the N-terminus and the first transmembrane region of the channel: this variation caused an increase in Kir4.1 surface expression or conductance and patients developed seizures in association with autism spectrum disorders and impaired cognitive abilities (Sicca et al., 2011).

3.3. Aquaporin-4 (AQP4)

Ion homeostasis in the brain depends not only on proper functioning of ion channels, but also on water transport. The influx of water between the blood and the brain parenchyma is tightly regulated by blood vessels. Glial ensheathment of blood vessels also contributes to water influx. The movement of water across cellular compartments is mediated by aquaporins (AQPs), transmembrane proteins which act as water channels in different cells and tissues. In the brain, the predominant form is AQP4, which is localized, as Kir4.1 channels, in astrocytic perivascular endfeet and perisynaptic processes. AQP4 participates in the control of extracellular fluid osmolarity and extracellular space volume by regulating water flow and K⁺ buffering (Nagelhus et al., 2004). It appears that water movement is altered in the hippocampus of HS TLE patients, as it has been shown that it presents an increased T2 signal density on magnetic resonance imaging and higher apparent diffusion coefficients, indicating water accumulation (Bronen et al., 1991, Hugg et al., 1999, Wieshmann et al., 1999). A potential impairment of AQP4 function has been proposed in TLE patient hippocampus: RT-PCR, immunohistochemistry and gene chip analysis have revealed an increased expression of AQP4 in HS tissues; this was accompanied by reduced levels of dystrophin, a protein involved in anchoring AQP4 to the membrane in perivascular endfeet (Lee et al., 2004). These results thus suggested an altered trafficking and distribution of AQP4 on the membranes. Quantitative electron microscopy later confirmed this hypothesis: the levels of AQP4 are indeed increased in MTLE compared to non-MTLE hippocampi. However, AQP4 density along the perivascular membrane domain of astrocytes was reduced by half in MTLE CA1 region, while no difference was found in AQP4 density on astrocyte membranes facing neuropil (Eid et al., 2005). These changes were secondary to altered perivascular dystrophin expression in sclerotic areas (Eid et al., 2005, Lee et al., 2004), which caused the loss of perivascular AQP4 and resulted in an impaired water flux through astrocytes. Given that in presence of high neuronal activity K⁺ and water are taken up by the astrocyte membrane facing the neuropil, transported through astrocytic syncytium and siphoned into blood or CSF through perivascular endfeet membrane (Paulson and Newman, 1987), this altered flow of water affected extracellular K⁺ buffering and contributed to epileptogenicity. Similar changes in AQP4 distribution have also been found in cortical samples of epileptic focal cortical dysplasia type IIb patients. AQP4 was indeed found to be enriched in the neuropil and around dysplastic neurons and reduced at the perivascular membranes due to disrupted dystrophin complex (Medici et al., 2011). The involvement of AQP4 in human epileptogenesis has been further confirmed by a genetic study which identified several polymorphisms of AQP4 gene associated to MTLE in combination with KCNJ10 single nucleotide polymorphisms (Heuser et al., 2010). These data support that alterations in water and K⁺ transport contribute to the aetiology of epilepsy.

3.4. Connexin43 and gap junctions

The role of GJ-connected astrocytic networks in the pathophysiology of epilepsy is still controversial, since they can act both as antiepileptic, by clearing and redistributing extracellular K⁺, glutamate and GABA (Orkand et al., 1966, Walz, 2000), and proepileptic, by fueling neurons with glucose and its metabolites in an activity-dependent manner (Rouach et al., 2008). Several studies have analysed changes in connexin (Cx) expression and coupling in human epileptic tissues: Naus and colleagues first reported increased Cx43 mRNA levels in the temporal cortex of epileptic patients (Naus et al., 1991). This change was confirmed by other groups both at the mRNA and protein levels (Aronica et al., 2001, Collignon et al., 2006, Fonseca et al., 2002). However, unchanged levels of Cx43 have also been reported in hippocampal tissues from patients presenting a complex partial seizure disorder in the medial temporal cortex and hippocampus (Elisevich et al., 1997). Whether expression and function of Cx30, the other astrocytic gap junction forming subunit, are altered in epilepsy remains unclear. Studies in rodents indeed report different results, which may be related to the diversity of the epilepsy models (Akbarpour et al., 2012, Condorelli et al., 2002, Söhl et al., 2000), and no data on human epileptic brain are yet available. Cx expression does not necessarily reflect active and functional coupling between astrocytes, since post-translational modifications can alter GJ properties, such as conductance, open probability and trafficking. Few functional data on GJ are available in human tissue. An early study performed in vitro on astrocytic cultures from tissues of MTLE patients reported an enhanced cell coupling by using the fluorescence recovery after photobleaching technique (Lee et al., 1995). However, recent work showed ex vivo a complete lack of glial GJ coupling in sclerotic hippocampal tissues from MTLE-HS patients, and a reduced coupling during the epileptogenic phase in the kainate mouse model of TLE (Bedner et al., 2015). It is unclear whether cultured astrocytes maintain the functional properties they display in the diseased tissue. Alternatively, gap junction coupling may be differentially regulated over the course of epilepsy, as increased astroglial coupling was found following status epilepticus in the kainate mouse (Takahashi et al., 2010).

3.5. Glutamate and GABA homeostasis

Neuronal activity leads to the release of excitatory and inhibitory neurotransmitters (mainly glutamate and GABA) in the extracellular space. To ensure appropriate synaptic responses and the maintenance of excitation-inhibition balance and to avoid neuronal loss and seizures, proper glutamate and GABA removal from the synaptic cleft becomes essential. Astrocytes play a crucial role in this phenomenon through the glutamine-glutamate-GABA cycle (Eid et al., 2012): extracellular glutamate and GABA released vesicularly are taken up by astrocytes via GLAST and GLT-1 glutamate transporters and GAT3 GABA transporter, respectively. Glutamine synthetase (GS) then directly converts glutamate into glutamine, while GABA enters the TCA cycle as succinate and is subsequently converted into alpha-ketoglutarate, glutamate, and glutamine. Specialized N-transporter proteins then extrude glutamine from astrocytes into the extracellular space, while glutamatergic and GABAergic neurons import glutamine through system A transporters, and convert it to glutamate and GABA, respectively. Using in vivo microdialysis, it has been shown that MTLE patients present five-fold higher extracellular glutamate levels in the epileptic sclerotic hippocampus compared to non-HS and non-epileptic hippocampal formation, despite neuronal loss and a two-fold increase in glial density (Cavus et al., 2005, Cavus et al., 2008). This suggested that an impaired neurotransmitter homeostasis could participate in the pathophysiology of epilepsy (Coulter and Eid, 2012). Indeed, alterations of glutamate and GABA transporters as well as GS have been identified in epileptic patients.

Studies on glutamate transporters in MTLE have reported contradictory results. An initial study found unchanged GLAST or GLT-1 expression through in situ hybridization and Western blot analysis in TLE hippocampus and cortex compared to post-mortem controls (Tessler et al., 1999). Similar results have also been reported in two other more recent studies (Bjørnsen et al., 2007, Eid et al., 2004). Conversely, a general decrease in GLT-1 and GLAST expression levels was found in HS TLE patients compared to non-HS TLE samples as assessed by immunocytochemical analysis (Mathern et al., 1999, Proper et al., 2002). This suggests that astrocytic glutamate uptake may play a crucial role in preventing epileptogenesis, even if the mechanism involved remains unknown.

Glutamate excess in TLE patients may be explained not only by a decreased expression in glutamate transporters, but also by impaired GS function. Indeed, it has been demonstrated that there is a reduction in GS expression and functionality in astrocytes of human sclerotic hippocampi (Eid et al., 2004, van der Hel et al., 2005): this results in slower rates of glutamate-glutamine cycling, accumulation of glutamate in astrocytic cytoplasm and decreased synaptic glutamate clearance (Petroff et al., 2002). On the contrary, subiculum astrocytes still express GS, but the protein is almost absent in most of the distal astrocytic processes, compared to non-epileptic controls (Eid et al., 2004). Furthermore, genetic mutations in GS gene (GLUL) have been found in epileptic patients: two GLUL congenital homozygous mutations occurring at GS active sites have been identified in two unrelated newborns, one of them displaying severe brain malformations, almost no EEG activity except short theta bursts and generalized seizures (Haber et al., 2006, Häberle et al., 2005). Noteworthy, a third mutation has recently been found in a child affected by epileptic encephalopathy and psychomotor retardation (Häberle et al., 2011).

A deficiency in GABA and GABA-mediated inhibition is thought to contribute to neuronal hyperexcitability in TLE. An in vivo microdialysis study has shown that the epileptogenic hippocampus of TLE patients presented lower extracellular GABA levels and higher glutamate concentrations just before the onset of seizures compared to the non-epileptogenic hippocampus of the same patient (During and Spencer, 1993). Interestingly, it has been demonstrated that this was due to an increased expression of the astrocytic GABA transporter GAT3, which is normally only weakly expressed in human hippocampal astrocytes (Lee et al., 2006): indeed, GAT3 was more prominently expressed in cells resembling protoplasmic astrocytes, located in dentate gyrus and hilus of sclerotic hippocampal formations. This increased GAT3 expression may thus explain GABA reduced extracellular levels during seizures, via an increased uptake by astrocytes (During and Spencer, 1993). Differently from TLE, absence epilepsy is characterized by an increased tonic GABA_A receptor-mediated inhibition in thalamocortical neurons, which is necessary and sufficient for the generation of non-convulsive seizures, typical of this form of epilepsy. This alteration of GABA levels is caused by a dysfunction of GABA transporter GAT-1, which is exclusively located in astrocytes in the thalamus of both humans and rodents (Crunelli et al., 2011, Pirttimaki et al., 2013).

Altogether these results indicate that an impairment of glutamate and GABA extracellular levels, due to altered expression of their transporters and reduced expression and functionality of GS, can play a major role in the pathophysiology of epilepsy.

4.1. Structural and morphological changes of astrocytes

Glioma cells share many characteristics with non-tumoural astrocytes, but also present several structural and morphological differences. In tissue of glioma patients, an overexpression of GFAP has been identified as being positively correlated with glioma size, but not with the degree of malignancy (Herpers et al., 1986, Lee et al., 2011a). For this reason, GFAP is used as a reliable immunohistochemical marker to stain surgically resected brain tumours in order to verify their astrocytic origin (Vinters et al., 1998). A strong GFAP immunoreactivity is present also in reactive astrocytes within or surrounding non-glial tumours (Raore et al., 2011). In grade I to IV astrocytomas, GFAP and vimentin are co-expressed but in different cellular compartments: indeed, while vimentin is located closer to cell nucleus, GFAP is preferentially detected in cellular processes (Bordey and Sontheimer, 1998, Cosgrove et al., 1989, Herpers et al., 1986). Besides GFAP, changes in other types of intermediate filament structural proteins have been observed: astrocytoma cells express the vimentin binding, 300-kDa intermediate filament associated protein (IFAP–300 kDa), which is normally present only in radial glia and immature astrocytes and absent in normal adult brain (Yang et al., 1993, Yang et al., 1994); furthermore, in astrocytomas, the proportion of keratin-containing cells, which are normally detected in the neuroectoderm, is directly linked to the degree of tumour malignancy, thus supporting a dedifferentiation of tumoural cells (Yang et al., 1994).

The modified expression of cytoskeletal elements may be responsible of the altered structure of astrocytic glioma cells: indeed, compared to non-tumoural astrocytes, which present a stellate shape with 2–3 major and many smaller processes, low-grade pilocytic astrocytoma cells display only 2–3 thick processes (Bordey and Sontheimer, 1998). Furthermore, astrocytic cells in low-grade gliomas display minimal to moderate nuclear atypia, scant cytoplasm and a high nucleus-to-cytoplasm ratio (Burel-Vandenbos et al., 2011).

4.2. Extracellular glutamate

Extracellular glutamate concentration is elevated in tumoural and peritumoural regions, especially close to tumours containing necrotic areas in high-grade astrocytoma patients (Roslin et al., 2003). Similar results have also been observed in oligodendrogliomas, which present high levels of glutamate and glutamine in the peritumoural area, as assessed by magnetic resonance spectroscopy (Rijpkema et al., 2003). This altered glutamate homeostasis explains why during the time course of the disease, 60–80% of glioma patients experience seizures (Kurzwelly et al., 2010, Lynam et al., 2007), which originate close to the tumour mass (Pallud et al., 2013, Patt et al., 2000). Various studies aiming at identifying the source of peritumoural glutamate reported an impaired expression of glutamate transporters on glioma cells: brain tissues from glioblastoma patients indeed display a strong reduction in GLT-1 levels, while GLAST is normally expressed but is thought to be mislocalized in cell nuclei rather than at the plasma membrane (Lynam et al., 2007, Savaskan et al., 2008, Ye et al., 1999). Decreased GLT-1 levels have also been observed in high-grade compared to low-grade astrocytomas and normal brains (de Groot et al., 2005). Furthermore these changes are accompanied by an altered expression of the cysteine-glutamate system (xc system), a Na⁺-independent exchanger that controls the intracellular glutathione levels by importing one molecule of extracellular cysteine (required in glutathione synthesis) per released glutamate. Glioma cell lines originated from tumours and brain specimens from glioblastoma patients express the xc system at significantly higher levels compared to human tissue samples without malignant transformation (Savaskan et al., 2008, Ye et al., 1999). Furthermore, a recent study has shown that 50% of patient-derived gliomas have elevated expression of SLC7A11, the catalytic subunit of the xc system responsible for xc-mediated glutamate release (Robert et al., 2015). Interestingly, when these glioma cells implanted intracranially in mice propagated in vivo as flank tumour xenolines, they caused seizures, tumour-associated excitotoxicity and shortened survival. Altogether these results thus suggest that high levels of this system contribute to the release of cytotoxic glutamate levels, which promote seizures and act as an autocrine/paracrine signal sustaining tumour growth and invasion (Lyons et al., 2007).

5.1. Astrogliosis in AD

Reactive astrogliosis is a well-known hallmark of AD, even if its role has not been clearly understood yet. It is identified by an increased expression of GFAP and hypertrophy of astrocytes in the vicinity of amyloid plaques. Post-mortem tissues from AD patients indeed display increased GFAP levels in temporal (Griffin et al., 1989, Simpson et al., 2010), occipital, parietal and frontal lobes (Kashon et al., 2004). Moreover, in the cerebrospinal fluid of AD patients, higher levels of GFAP concentrations have been measured compared to age-matched controls (Jesse et al., 2009). Interestingly, some degree of correlation has been found between GFAP expression and AD progression, with higher GFAP levels at increasing Braak groups (Simpson et al., 2010) or duration of clinical illness (Serrano-Pozo et al., 2013). However, this correlation remains uncertain, since another work showed no difference in GFAP expression between demented and non-demented brains within the same Braak stage (Wharton et al., 2009). During AD progression, 8 of the 10 different GFAP isoforms described (Hol and Pekny, 2015) are upregulated. For instance, reactive astrocytes in dentate gyrus subgranular zone, hilus and CA4 area of AD patients display a prominent expression of GFAPδ, but only CA1, CA3 and subiculum astrocytes surrounding plaques showed GFAPδ upregulation with increasing AD stage (Kamphuis et al., 2014). Additionally, the number of human-specific astrocyte subtypes expressing the frame-shifted GFAP variant, GFAP⁺¹, is increased with AD progression, but only few of these GFAP⁺¹-expressing cells has been identified as associated to plaques, with processes protruding through them (Kamphuis et al., 2014, Middeldorp et al., 2009). Astrogliosis and GFAP upregulation are also accompanied by dysregulation in the expression of other astrocytic cytoskeleton proteins. For instance, in the lateral temporal cortex of advanced AD stages, there is a significant decrease in transcripts encoding members of the myosin and kinesin family and other cytoskeletal proteins, such as actin β, dynein and integrin α. Moreover, transcripts encoding tight junction proteins and adherens junctions are also reduced during AD progression (Simpson et al., 2011). The effect of the altered expression of these genes in astrocytes still remains unclear, but it may affect various intracellular signalling pathways.

5.2. Interactions between astrocytes and Aβ

In AD brains at early stages of the pathology, activated reactive astrocytes are predominant in the molecular layer of the cerebral cortex and close to amyloid plaques in pyramidal cell layers (Wisniewski and Wegiel, 1991). In several brain regions such as cortex, hippocampus and cerebellum, proliferating processes of hypertrophic astrocytes nearest to amyloid deposits contact and surround the plaques; they penetrate more into non-cored primitive plaques, compared to classic compact cored amyloid deposits, thus merging with them and contributing to their fragmentation, dispersion and the observed variety of plaque morphology (Kato et al., 1998, Wisniewski and Wegiel, 1991). Interestingly, in the visual cortex of AD brains with severe pathology, GFAP-immunoreactive astrocytes and plaques are arranged in a specific laminar distribution: indeed gliosis is preferentially localized in laminae II, III, IVa and IVc, the latter presenting a discrete plaque-associated glyotic horizontal band at the lower edge (Beach and McGeer, 1988). Furthermore, a more recent study has shown that reactive astrocytes together with microglial cells form specific 3D reactive glial nets around plaques in a plaque-specific way: at Aβ dense-core plaques, astrocytic processes are intermingled with microglial cell bodies which envelop the core Aβ structure; while at fibrillary plaques, a higher number of glial cells are recruited to reactive glial net formation and both microglial and astrocytic processes invade the plaque area and interact with Aβ protein (Bouvier et al., 2016).

The stimulus capable of inducing astrocyte reactivity in AD brains is still under debate; however, studies using aggregated Aβ protein and the intact core of Aβ plaques isolated from AD brain tissue have shown that Aβ can trigger activation of astrocytes in vitro, causing GFAP up-regulation and morphological changes (DeWitt et al., 1998). Furthermore, amyloid plaques colocalize with reactive astrocytes in the absence of dystrophic neurites in the hippocampus of mild AD brains, while decreasing over the course of the pathology (Pike et al., 1995). This suggests that plaque-associated astrocytosis can act as a contributory event in AD pathology and that once activated, astrocytes can participate in Aβ metabolism. Aβ42 indeed accumulates in the cytoplasm and processes of reactive astrocytes in the molecular layer completely devoid of amyloid plaques in AD brains. Aβ42 also builds up in astrocytes associated with plaques in pyramidal cell layers of the entorhinal, parietal, occipital and temporal cortex (Akiyama et al., 1999, Funato et al., 1998, Kurt et al., 1999, Nagele et al., 2003, Thal et al., 1999, Thal et al., 2000). In these cells, non-fibrillar Aβ42 localizes to small lysosomal lipofuscin-like granules, organized in clusters and mainly located in the perinuclear cytoplasm (Funato et al., 1998, Nagele et al., 2003, Thal et al., 1999, Yamaguchi et al., 1998). Moreover, it has recently been reported that more than 90% of activated astrocytes in AD brain frontal cortex can accumulate a newly described class of amyloid structures formed by Aβ, the annular protofibrils. These fibrils are absent in amyloid plaques and in the brain of age-matched controls, and can induce reactive oxygen species generation and inactivation of GS in the cell (Lasagna-Reeves and Kayed, 2011).

The accumulation of Aβ42 in astrocytes is directly linked to the severity of local AD pathology. Aβ42⁺ material in astrocytes is indeed proportional to the amount of Aβ42 contained in surrounding neurons and to the local density of amyloid plaques within the pyramidal cell layer of the entorhinal cortex, while the amount of Aβ42 in astrocytes of the molecular layer devoid of plaques well correlates with the severity of the pathology in the sub-adjacent cortical laminae (Nagele et al., 2003). Interestingly, astrocyte-accumulated Aβ42 is not produced by astrocytes, but has a neuronal origin, deriving from the internalization of degenerating synapses and dendrites by phagocytosis: astrocytic Aβ42 colocalizes with choline acetyltransferase and α7 nicotinic acetylcholine receptors, which are neuron-specific proteins accumulating in astrocytes as a consequence of their debris-clearing activity (Nagele et al., 2004). Furthermore, with the progression of the disease, Aβ42-burdened astrocytes can undergo lysis and form small spherical GFAP⁺ amyloid plaques, first appearing in the subpial portion of the cortical molecular layer close to astrocytes containing large Aβ42 deposits (Nagele et al., 2003).

These results indicate that not only neurons, but also astrocytes are capable of giving rise to amyloid plaques and causing morphological modifications within these plaques; this ability can thus account, at least in part, for the variety of plaque morphology identified in AD brains. Blocking the initial accumulation of Aβ42 in neurons represents the main early target to control AD pathology. However, in view of the ability of astrocytes to interact with plaques and participate in Aβ metabolism, limiting their recruitment could also contribute to limiting or delaying AD progression.

5.3. Glutamate and GABA homeostasis

There are several indications that astrocytic glutamatergic function is impaired in AD. GLT-1 immunoreactivity is reduced in the frontal cortex of AD patients (Li et al., 1997, Tian et al., 2010) and this reduction is inversely correlated with amyloid precursor protein mRNA levels, while no change has been observed for GLAST expression. GLT-1 reduced expression is also accompanied by decreased glutamatergic transport activity and increased mRNA levels, thus indicating an impairment at post-transcriptional level (Li et al., 1997). Similar results have also been obtained in the hippocampus and gyrus frontalis of AD patient brains. A marked impairment in GLT-1 and GLAST expression at both gene and protein levels occur already at early stages of the disease, particularly in the vicinity of amyloid plaques (Jacob et al., 2007). Furthermore, in AD lateral temporal cortex, GLT-1 expression tends to decrease with higher Braak group. This correlation is however still under debate, since contradictory results have been obtained due to high variability of glutamate transporter expression between AD individuals (Beckstrøm et al., 1999). Together with reduced expression, protein splice variants can also account for changes in the functionality of glutamate transporters. Interestingly, disease- and pathology-specific changes in GLT-1 splice variant expression occur in autoptic AD brains, which can account for the reduced astrocytic glutamate uptake efficiency in AD. In particular, a reduction in the functional splice variants b, which is able to uptake glutamate, has been identified in several brain regions, together with a significant increase of exon-skipping variants, characterized by reduced transport capacity (Scott et al., 2011). Altogether, these results suggest an impairment of astrocytic glutamate clearance capability in AD, which may lead to neurotoxic effect and contribute to the increased prevalence of seizures in AD patients (Scarmeas et al., 2009).

Furthermore, the levels of GS are also altered in AD brain: temporal neocortex astrocytes from AD brain samples display reduced GS expression due to impairment of post-translational modifications, which renders the protein highly sensitive to oxidative lesioning (Le Prince et al., 1995). In contrast, total GS levels in the CSF of AD patients remain unchanged (Timmer et al., 2015).

GABAergic dysfunction also plays a role in AD, and several recent studies point to abnormalities of GABA homeostasis in reactive AD astrocytes. An accumulation of GABA together with an increase in GAD67 and GAT3 have been found in GFAP⁺ astrocytes of dentate gyrus molecular layer and inferior temporal cortex from post-mortem AD patients, but not in age-related healthy subjects (Mitew et al., 2013, Wu et al., 2014). This suggests an increased GABA production in astrocytes. Consistent with this hypothesis, astrocytic monoamine oxidase-B, the enzyme responsible for GABA production, is up-regulated in post-mortem brains of AD individuals. This has also been shown in AD mouse models, where accumulating GABA is abnormally released from reactive astrocytes through bestrophin 1 channels, which impairs learning and memory (Jo et al., 2014) The high astrocytic GABA level identified in human brains with high Aβ load may thus be used as a novel biomarker and diagnostic tool for AD. Furthermore, since it has been shown that GABA transporters can reverse their transport direction in presence of excessive intracellular GABA (Lee et al., 2011b, Richerson and Wu, 2003), GAT3 may serve as a new drug target.

6.1. Astrocyte density and morphology

Many histopathological studies performed on post-mortem brain samples have unveiled prominent decreases in astrocyte number and packing density in MDD subjects compared to age-matched non-psychiatric controls (Cotter et al., 2002, Cotter et al., 2001, Gittins and Harrison, 2011, Ongür et al., 1998, Rajkowska et al., 1999). Several brain regions display a reduced astrocytic population, such as dorsolateral prefrontal (Cotter et al., 2002, Rajkowska et al., 1999), orbitofrontal (Rajkowska et al., 1999), subgenual (Ongür et al., 1998) and anterior cingulate cortex (Cotter et al., 2001) and amygdala (Altshuler et al., 2010, Bowley et al., 2002). However, an increase in glial cell density has also been reported in hippocampal regions and dentate gyrus of MDD patients (Stockmeier et al., 2004), while no change has been observed in the orbitofrontal cortex and in the supragenual region of the anterior cingulate cortex in late-life depressed patients (Khundakar et al., 2011a, Khundakar et al., 2011b) and in the hippocampus (Cobb et al., 2013). Changes in glial density may thus differentially affect specific brain regions. Interestingly, it has been shown that this alteration is age-dependent: in grey matter of dorsolateral prefrontal cortex of younger depressed patients (<50 years old), the density of GFAP⁺ astrocytes is significantly reduced compared to controls of similar age; in contrast, older subjects with late-onset depression presented increased astrocytic population in the same area (Miguel-Hidalgo et al., 2000), probably reflecting a compensation to neuronal loss observed in older MDD patients (Rajkowska et al., 2005).

In parallel to alterations of astrocyte packing density, the size of glial nuclei seems to be affected in MDD. Astrocytes with larger nuclei have been observed in the dorsolateral prefrontal cortex (Rajkowska et al., 1999) and in the grey and white matter of the anterior cingulate cortex: in these regions fibrous astrocytes had bigger cell bodies and more ramified processes in depressed subjects committing suicide compared to matched sudden-death controls (Chana et al., 2003, Torres-Platas et al., 2011). On the contrary, three other studies observed unaltered glial size in the prefrontal and orbitofrontal cortex and in the hippocampus (Cotter et al., 2002, Cotter et al., 2005, Stockmeier et al., 2004).

Studies on GFAP expression additionally revealed marked alterations in MDD subjects. Immunohistochemical analysis aimed at quantifying the area covered by GFAP⁺ cell bodies and processes has shown a predominant decrease of GFAP in grey matter of the prefrontal cortex in young depressed subjects compared to controls (Miguel-Hidalgo et al., 2000), in white matter of anterior cingulate cortex (Gittins and Harrison, 2011), in orbitofrontal cortex (Miguel-Hidalgo et al., 2010), CA1 and CA2 hippocampal regions (Müller et al., 2001), locus coeruleus (Chandley et al., 2013) and cerebellum (Fatemi et al., 2004). Moreover, GFAP decrease has also been confirmed at mRNA and protein levels (Chandley et al., 2013, Fatemi et al., 2004, Johnston-Wilson et al., 2000, Miguel-Hidalgo et al., 2000, Webster et al., 2005) and it is correlated with age and onset of depression: the levels of GFAP protein are indeed significantly lower in less than 60 years-old depressed patients compared to age-matched controls, with no change observed between older MDD subjects and their controls (Si et al., 2004). In contrast, an increase of GFAP occurred in dorsolateral prefrontal cortex in late-onset MDD patients (Davis et al., 2002, Miguel-Hidalgo et al., 2000).

6.2. Glutamate and GABA homeostasis

Recent neuroimaging and post-mortem studies on depressed subjects have revealed a dysfunction of astrocytic-mediated regulation of glutamate homeostasis. MDD subjects presented lower levels of glutamate, glutamine or combined glutamate/glutamine (Glx), as assessed by magnetic resonance spectroscopy, in several brain regions such as prefrontal areas (Hasler et al., 2007), frontal lobe (Yildiz-Yesiloglu and Ankerst, 2006), anterior cingulate cortex (Auer et al., 2000, Mirza et al., 2004, Pfleiderer et al., 2003) and amygdala (Michael et al., 2003), and in plasma (Altamura et al., 1995). This suggests an impaired glutamate/glutamine metabolism. These changes may however not appear homogenously in all brain areas and in all depressed patients, since two other studies reported glutamate increases in occipital and frontal cortex (Hashimoto et al., 2007, Sanacora et al., 2004), probably reflecting region-specific alterations which can also depend on patient age and chronicity of depression.

The impaired glutamate homeostasis observed in MDD patients may be related to the reduced astrocytic packing density detected from post-mortem studies on MDD brain samples. Furthermore, reduced expression of GLAST and GLT-1 occur in anterior cingulate (Choudary et al., 2005), dorsolateral prefrontal (Choudary et al., 2005, Klempan et al., 2009) and orbitofrontal cortex (Miguel-Hidalgo et al., 2010), and in locus coeruleus (Bernard et al., 2011, Chandley et al., 2013) and hippocampus (Medina et al., 2016) of subjects diagnosed with MDD. Finally, different cortical and subcortical regions of depressed suicide victims, such as prefrontal and premotor cortex and the amygdala, also display reduced levels of GS (Miguel-Hidalgo et al., 2010, Sequeira et al., 2009). Interestingly, it has been shown that the impairment in glutamate-related gene expression is specific to astrocyte in MDD, since it does not occur in neurons of MDD brains (Bernard et al., 2011), thus underlying the astrocytic basis of MDD pathology.

In MDD not only glutamate regulation is impaired, but also GABA homeostasis. Indeed, reduced GABA levels have been found in the dorsomedial/dorsal anterolateral prefrontal cortex (Hasler et al., 2007), and in the occipital cortex (Sanacora et al., 2004) of depressed but not remitted MDD patients (Schür et al., 2016). Furthermore, in the dorsolateral prefrontal cortex, GABA_A receptor subunits are up-regulated (Choudary et al., 2005, Sequeira et al., 2009). Interestingly, it has been shown that α1 and β3 subunit expression is selectively increased in suicide completers (Choudary et al., 2005).

7.1. Astrocytic over-population and structure

Despite the reduced number of neurons, which is due to severely impaired proliferation and increased apoptosis (Contestabile et al., 2007), DS brains present an altered number of glial cells, especially astrocytes. In particular, DS foetuses between 17 and 21 weeks of gestation have a higher percentage of cells with astrocytic phenotype and a smaller percentage of cells with neuronal phenotype in the hippocampus (Guidi et al., 2008), thus indicating a shift from neurogenesis to gliogenesis. Similar results were also found in the DS frontal lobe, where astrocytes were more numerous and morphologically more mature than those of age-matched controls (Zdaniuk et al., 2011). Furthermore, a two-fold developmental bias toward astrocyte differentiation in DS neural cultures (Briggs et al., 2013), which occurs through the release of astrocytic S100β (Chen et al., 2014), was reported using DS-induced pluripotent stem cells obtained by episomal reprogramming.

Astrocytes in DS are not only more proliferative and abundant, but they also display altered processes. DS brains display an impaired development of interlaminar astrocytes, which starts normally around 20–40 days after birth, but manifests a breakdown by the first year of age. This results in decreased number of interlaminar processes and immature astroglial layout, which can occur in several brain regions with different gravity (Colombo et al., 2005). Moreover, decreased GFAP mRNA levels have been found in the superior prefrontal cortex of DS subjects and in temporal lobe white matter of DS foetuses during middle pregnancy period (Goodison et al., 1993, Kanaumi et al., 2013). Similar results have also been obtained in the frontal lobe white matter of older patients. Interestingly, the pattern was reversed in the cortex, where an increase in GFAP⁺ cell number was observed, probably reflecting premature dementia-associated development of plaques and neurofibrillary tangles (Jørgensen et al., 1990, Mito and Becker, 1993).

This work was supported by grants from INSERM, CNRS, College de France and ERC to N.R., from Neuropôle de Recherche Francilien (NeRF) and Ligue Francaise contre l’Epilepsie (LFCE) to E.D., and from Observatoire B2V des memoires to F.V.