One of the oldest known functions of astrocytes is to supply energy in the form of lactate to neurons. Glucose is mainly stored as glycogen in astrocytes, where it is metabolized to pyruvate and lactate and then transported via MCTs across the cell membrane. The transported lactate is then utilized by neighboring neurons and metabolized (Magistretti et al., 1999). Apart from glucose metabolism, astrocytes are also involved in glutamate uptake via two pathways. The first pathway involves the direct conversion of glutamate to α-ketoglutarate through NAD-dependent oxidative deamination catalyzed by GDH, and the second pathway is an ATP-requiring reaction in which ammonium is catalyzed by GS to yield glutamine. This glutamate-glutamine shuttle protects against the toxic effects caused by extracellular glutamate (Sonnewald et al., 1997).

One essential function of astrocytes is to maintain brain homeostasis through multiple dynamic equilibrium adjustments, including water balance, ion distribution, glutamate buffering, and recycling (Wang and Qin, 2010; Coulter and Eid, 2012). High levels of synaptic glutamate can cause over-activation of neurons which may lead to excitotoxicity; thus rapid removal of extracellular glutamate from the synaptic cleft is essential for neuronal survival (Dong et al., 2009). This is accomplished by Na⁺ dependent transporters on astrocytes, EAAT1 and EAAT2, respectively.

Apart from glutamate clearance, astrocytes can control cerebral glutamate levels (Stobart and Anderson, 2013). Glutamate that is taken up by the astrocytes is converted to glutamine by GS, then later passed back to the synaptic terminal where it is converted back to glutamate (Danbolt, 2001; Parpura and Verkhratsky, 2012). There is increasing evidence that the uptake of glutamate also induces glycolysis in astrocytes, resulting in the production and secretion of lactate for the neighboring neurons (Ricci et al., 2009; Bélanger et al., 2011; Stobart and Anderson, 2013). This mechanism, the astrocyte to neuron lactate shuttle, regulates lactate delivery in an activity-dependent manner (Pellerin et al., 1998; Stobart and Anderson, 2013).

Together with endothelial cells and pericytes of the brain microvessels, astrocytes form the BBB, a physical diffusion barrier which restricts the exchange of most molecules between blood and brain (Abbott et al., 2006; Macvicar and Newman, 2015). Astrocytes are also involved in regulating cerebral blood flow by a K⁺ siphoning mechanism, releasing K⁺ onto blood vessels from their end-feet in response to neuronal activity (Paulson and Newman, 1987). It has been suggested that the release of prostaglandins from astrocytes results in increased Ca²⁺ that evokes vessel dilation (Zonta et al., 2003). Likewise, they are also involved in regulating BBB permeability from the bloodstream to brain parenchyma by the activation of tight junction proteins through NF-κB (Brown et al., 2003; Abbott et al., 2006). BBB defects are involved in many neuroinflammatory and neurodegenerative diseases, including multiple sclerosis, where the specialized brain endothelial cells which comprise the BBB are diminished, causing a loss of protective function during the progressive phase of disease (Weiss et al., 2009).

There is now abundant evidence to support the notion that astrocytes are actively involved in the formation and refinement of neural networks (Oberheim et al., 2006; Araque and Navarrete, 2010). During development, billions of neurons connect to form functional networks via synapses, with the control of synapse development by astrocytes highly conserved across species. A distinctive attribute of astrocytes in synapse formation is to increase the number of synaptic structures (dendritic spine) within the neural circuits (Ullian et al., 2001; Slezak and Pfrieger, 2003; Stevens et al., 2007; Stipursky et al., 2011; Clarke and Barres, 2013). The first evidence for astrocytes being involved in synapse formation came from the rodent RGC study, which showed that culture with astrocytes resulted in a 10-fold increase in excitatory synapse and synaptic functionality (Meyer-Franke et al., 1995). Later, in vitro studies confirmed that astrocytes can also instruct synapse formation for human neurons (Diniz et al., 2012).

Astrocytes are also involved in the refinement of the neural network by synaptic pruning - the elimination of extra synapses to increase the precision and efficiency of neural circuits (Clarke and Barres, 2013). The mouse retinogeniculate system, an excellent model system for studying synapse refinement and elimination (Hong and Chen, 2011), has been used to show that signals released from astrocytes in the postnatal brain induced the expression of the complement component C1q that executes synapse elimination by astrocytes via phagocytosis (Stevens et al., 2007). Notably, astrocytes employ this mechanism throughout the nervous system (e.g., in the uninjured brain or in response to glioma or trauma). However, further work is required to investigate the phagocytic pathway of astrocytes in human models.

Studying the direct communication between astrocytes and neurons is a rapidly expanding field of neuroscience. The term “tripartite synapse,” was proposed 20 years ago to describe synaptic physiology involving astrocytes, in addition to pre- and post-synaptic neurons. In this context, astrocytes release neuroactive molecules (such as glutamate, ATP, nitric oxide, prostaglandins, and D-serine) in the presence of elevated Ca²⁺, which in turn influence neuronal excitability (Araque et al., 1999; Perea et al., 2009; Eroglu and Barres, 2010).

The concept of “gliotransmission,” which was first hypothesized in the 1980s, involves the active vesicular release of neurotransmitters and glutamate by astrocytes (Halassa et al., 2007). In other words, the GPCR-mediated Ca²⁺ variations in astrocytes can trigger the release of glutamate, D-serine, and ATP. So far, various mechanisms have been proposed for gliotransmission, including Ca²⁺-regulated vesicular exocytosis (Agulhon et al., 2012) and non-vesicular release. Additionally, Santello et al. (2011) found that cytokines are required for functional glutamatergic gliotransmission. While the importance of gliotransmission is supported by findings that dysfunction of astrocytic proteins involved in transmitter release can cause severe brain disorders (Rossi et al., 2011), there are still several features of gliotransmission that are controversial and require further clarification (Hamilton and Attwell, 2010; Sahlender et al., 2014; Sloan and Barres, 2014).

Radial glial are the earliest cells to be derived from the lateral wall of the neural tube (Anthony et al., 2004; Howard et al., 2006; Mo et al., 2007) (Figure 1). In humans, RG cells are also found in the oSVZ in the developing neocortex, which is absent in rodents (Pollen et al., 2015). The oSVZ contains a large proportion of outer RG (oRG) cells that act as guides for neuronal migration. Particularly in humans, the oRG cells appear to contribute to neocortical expansion by increasing the number of neural precursor cells (NPCs) (LaMonica et al., 2013).

Radial glial cells usually divide at the apical (ventricular) surface within the ventricular zone (VZ) to generate new neurons or single IPC by asymmetric division, or self-amplify progenitors by symmetric division. The majority of these progenitors migrate radially with the rapid increase in the width of the cerebral wall (Hartfuss et al., 2001; Marín and Rubenstein, 2001; Anthony et al., 2004; Götz and Huttner, 2005; Sild and Ruthazer, 2011; Florio and Huttner, 2014; Schitine et al., 2015). Just before birth, the RGs “accelerate the expansion of the neuronal population” and switch to gliogenesis to produce astrocytes. This transition from neurogenesis to astrogenesis is mediated by known soluble factors including IL-6 and BMP-4 protein (Miller and Gauthier, 2007). By the end of the cortical development, most of the RG cells lose their ventricular attachments and migrate toward the cortical plate to form different cortical layers.

In general, there is a remarkable overlap in the expression patterns and cell proliferative processes of human and rodent, but there are species differences in structural organization and complexity. In mice, the majority of cells form a single compact layer, whereas in humans they are dispersed throughout a larger zone. IPCs in human VZ regions can be divided into two subpopulations, apical and basal, that have discrete molecular profiles. Apical IPCs are defined by mitosis occurring at the ventricular surface and the basal IPCs are defined by mitosis occurring at an aventricular location and the absence of ventricular contact (Florio and Huttner, 2014). In general, IPCs can generate neurons (nIPCs) or glial cells, including oligodendrocytes (oIPCs) or astrocytes (aIPCs). Mouse IPC go through only one round of cell division to produce two neurons (LaMonica et al., 2013), whereas human IPC undergo several rounds of cell division before producing neurons (Hansen et al., 2010). The molecular mechanisms that underlie IPC division are still not clear but their transcriptional regulators, such as TBR2, EOMES are likely involved (Kowalczyk et al., 2009; Fietz et al., 2010; Hansen et al., 2010). Many reports have suggested that SVZ in humans can be subdivided into the iSVZ and oSVZ (Hansen et al., 2010; Lewitus et al., 2013; Thomsen et al., 2016). A new type of outer radial glia-like cells, basal RGs (bRGs), was identified in the oSVZ of developing cortices of humans (Betizeau et al., 2013; Pilz et al., 2013). The translocation of RG cells and differentiation to astrocytes has been visualized through time-lapse imaging (Noctor et al., 2008), as well as a similar transformation in the sub-cortical telencephalon (Barry and McDermott, 2005). To date, five different sources of cortical astrocytes have been identified: (i) RG cells within the ventricular zone; (ii) RG cell transformation; (iii) intermediate progenitors; (iv) glial progenitors in marginal zone; and (v) superficial layer progenitors. However, only a limited number of studies on astrogliogenesis have been performed in the human (Kanski et al., 2014; Paşca et al., 2015), in contrast to the numerous reports on rodents (Chi et al., 2006; Ventura and Goldman, 2007; Widestrand et al., 2007; Demars et al., 2010; Ge et al., 2012).

Examination of human astrocytes from post-mortem tissue have led to a better understanding of brain diseases and opened doors toward generating more efficient in vitro based models. The first human astrocytes were cultured from fetal or adult post-mortem tissue (Ennas et al., 1992; Lee et al., 1993), but were often contaminated with microglia and other cell types which were difficult to separate during dissections. Additionally, in many cases, biopsies represent the end stage of the disease and control tissue is obviously inaccessible due to ethical concerns and potential health risks. Given all the practical limitations of human brain tissue research, murine and rat astrocytes have mostly been used to study astrocyte physiology. For this purpose, murine or rat astrocytes are usually purified from the cortex. Alternatively, astrocytes may be generated from murine PSCs (Kleiderman et al., 2016). These generalized approaches do not account for regional heterogeneity of astrocytes, such as the expression of the transporter OCT3, which is high in striatal astrocytes and low in cortical astrocytes (Cui et al., 2009). Especially for disease studies, astrocytes may therefore be purified from the relevant brain regions. For instance, cerebellar, cortical, striatal, and nigral astrocytes show differences in dopamine or angiotensin signaling (Yu et al., 1996; Reuss and Unsicker, 2000). Thus, astrocytes from specific brain regions offer an attractive alternative source to study astrocyte function in vitro. Mice models offer further advantages since methods to manipulate the genome (knock-out or knock-in genes) are well-established. With these approaches it needs, however, to be considered that astrocyte heterogeneity may not be a cell-intrinsic property, but be decided by the surrounding neurons, and may therefore get lost in cell culture (Farmer et al., 2016).

The neural induction protocol involves in dissociating PSCs and plating them on a feeder or feeder-free adherent culture system. The media for neural induction usually consist of the neurobasal medium or DMEM/F12 medium, or combination of both. The cells are rapidly induced with antagonists, either LDN or noggin to inhibit the BMP pathway and SB431542 to inhibit the TGF-ß pathway, along with additional components to enhance neuronal precursors (Chambers et al., 2009) (Figure 2). The primitive neuroectodermal aggregates (3D system) or neuroepithelial sheets (2D system) are then plated on the adherent substrate to promote the definitive neuroectoderm fate. Upon reaching the “end phase,” NPCs are organized into polarized structures called neural rosettes. These neural rosettes are selected and cultured for several passages and then directed toward astroglial progenitors with different combinations of morphogens (CNTF, SHH, FGF, and RA) in defined culture medium. Numerous studies have utilized each of these methods, often with minor variations. However, it is not always clear why a particular method was chosen, and so it is very difficult to evaluate the exact effect of these small changes. In the next section, we have summarized the most efficient ways to generate human astrocytes from PSC-derived NPCs through in vitro culture, and discuss the maturity of the derived astrocytes.

For neural differentiation of hESCs, most of the available methods are based on either the initial formation of embryoid bodies or on co-culture with stromal cells. First, we would like to compare the aggregate differentiation from monolayer differentiation.

Zhang et al. (2001) was the first to develop a novel technique to isolate and culture human astrocytes in serum-free conditions, thus recapitulating the resting, non-reactive state normally observed in healthy astrocytes in vivo, and went on to describe a differentiation technique for human ES cells that can be applied to model and study fetal and mature astrocytes in health and disease. In their first report (Zhang et al., 2001), differentiation was initiated by culturing the cells as 3D spheres in chemically defined medium containing FGF-2 for a short period. Subsequently, the EBs were grown in adherent culture to form neural rosettes. This unique cross-sectional arrangement of epithelial cells is now considered a hallmark of successful neural induction. By day 7, almost all EBs generated neural tube structures to produce neural progenitor cells and neural stem cells, and morphological analysis confirmed the expression of neural markers Nestin, Musashi-1 and neural cell adhesion molecule (NCAM). This group were the first to show the incorporated hESC-derived neural precursors in different brain regions upon transplantation into the neonatal mouse, with no teratoma formation. This remarkable development generated GFAP⁺ astrocytes, oligodendrocytes both in vitro and in vivo.

In the same year, two other EB based differentiation studies were published (Carpenter et al., 2001; Reubinoff et al., 2001). Both studies also used hESCs in a 3D aggregate culture system in the presence of FGF-2 to induce neural differentiation. In one, EBs were differentiated into neuronal progenitors in the presence of RA along with selective morphogens (Carpenter et al., 2001) (detailed in Table 3), while in the other, differentiation occurred in the presence of just EGF and FGF-2 (Reubinoff et al., 2001). Therefore both the studies obtained populations with a wide panel of morphological characteristics, such as positive for Nestin, PS-NCAM, and A2B5. FACS analysis revealed 96% of the cells were positive for A2B5 (a marker for astroglial progenitors) (Carpenter et al., 2001).

Later, Tabar generated a neural progenitor population by inducing hESCs on stromal cells (MS5) in serum-free media (Tabar et al., 2005). To enhance the neural induction noggin was added to the N2 media. After a month neural precursors were isolated and maintained in N2 media containing FGF-2 and EGF. At this stage, the majority of cells (>90%) were immunoreactive for neural precursor markers (Nestin, Musashi-1) and a glial progenitor marker (A2B5).

A better understanding of differentiation protocols has led us to conclude that the majority of hESC studies utilize the classic aggregation method to yield functional astrocytes in combination with different growth factor cocktails. Despite the potential benefits of hESC technology, the direct differentiation of hESCs toward astrocytes is rarely undertaken, for ethical reasons. Therefore we now compare some of the available protocols from human iPSCs technology (detailed in Table 3 and Figures 3A,B). For neural differentiation of hiPSCs, most methods available are either based on the initial formation of embryoid bodies or monolayer base methods.

Differentiation of hiPSC-derived NSC into astrocytes was first reported following the observation that astrocyte progenitor cells were S100β positive after four weeks of differentiation and GFAP positive around the 12th week of differentiation (Hu et al., 2010). The study used a similar aggregated differentiation method to that of Zhang et al. (2001) to obtain neurons. To further enhance the glial differentiation the progenitors were expanded in N2B27 media containing additional factors, i.e., cAMP, T3, platelet-derived growth factor (PDGF), Insulin-like growth factors (IGFs), and neurotrophin-3 (NT-3), with the aim of increasing the formation of OLIG2 positive ventral progenitors and reducing the formation of HB9⁺ postmitotic motoneurons (Hu et al., 2010).

In the subsequent years, additional studies have described the generation of astroglial cells from human PSCs, including a landmark study that generated functional astrocytes from iPSC that were indistinguishable from hESC-derived astrocytes (Emdad et al., 2012). The study involved culturing EBs on low attachment plates in neurobasal media for 2 weeks to develop columnar epithelium within the formed EBs. And also investigates the astrocyte differentiation potential with three different conditions: (1) CT-1 alone, (2) CT-1 in combination with CNTF, and (3) recombinant Jagged-1 (JAG1-Fc) in conjunction with CNTF. The results have showed high number of hiPSC-derived NE cells after sequential treatment with FGF- 2 + CNTF followed by CNTF alone. However, no incremental effect on astrocytic differentiation was observed when CT-1, Notch activator JAG1-Fc was used.

Later, Lafaille described the efficient generation of astroglial cells within 10 weeks (Lafaille et al., 2012), adapting previously described protocols to enhance neural differentiation (Zhang et al., 2001; Elkabetz et al., 2008). The derived neural crest stem cells (NCSc) and NPCs from hPSCs were expanded in defined conditioned medium supplemented with EGF and FGF-2 for 8 weeks, followed by 5% FBS treatment for 20 days, leading to 90% GFAP⁺ cells.

The next study went on to show the efficient generation of astroglial cells within the short time frame of 6 weeks from hPSC-derived NSCs, using a defined medium system (Shaltouki et al., 2013). As might be expected, the differentiation protocol was EB based, with the colonies cultured in suspension for 8 days before surface attachment. Formed neural rosettes were manually isolated and replated to acquire a homogenous population of NSCs, that was then stimulated with neurotrophic factor (CNTF) and proteins (BMP, FGF-2) to significantly increase the proportion of GFAP positive cells. These differentiated astrocytes were then plated on neurons to quantify synapse formation. Interestingly, the number of synaptic puncta remained significantly higher in the presence of astrocytes, indicating that these in vitro astrocytes displayed similar functional characteristics and morphological patterns to primary astrocytes.

Other studies have added further evidence for the generation of astroglial cells. Sareen et al. (2014) generated so-called “EZ spheres” from hiPSCs that could be differentiated into neural progenitor cells with an astroglial predisposition. The “EZ sphere” is a free-floating cell aggregate containing pre-rosette stem cells, that is generated directly from PSCs. The EZ spheres were caudalized using all-trans-RA in neurobasal medium and maintained their proliferative capacity for 30 passages. For astroglial differentiation, they were dissociated and plated as single cells (Sareen et al., 2014). Like Sareen, the study by Serio et al. (2013) opted for EZ sphere formation; neurospheres were mechanically chopped at the beginning of the enrichment phase and cultured in neurobasal medium for 2–4 weeks, before being dissociated to form NPCs. Astrocyte populations were obtained by differentiating NPCs in CNTF (Serio et al., 2013). The resulting population robustly expressed both developmental (Nestin, vimentin, GFAP) and mature astrocyte markers (S100β).

The most striking feature of these studies is the formation of the EZ sphere (∼200 μm) which represents the primitive type of neural stem cells at a stage before early neural tube formation. While various protocols exist for the generation of PSC-derived neural progenitor cells, including the complex four stage systems involving RA-mediated induction, adherent culture, and neural rosette isolation, all these methods can be laborious when compared to EZ sphere formation.

Mormone and co-workers then generated iPSC-derived astrocytes with characteristics similar to those differentiated from hESCs by culturing clusters of cells in low attachment plates in the presence of neurobasal media supplemented with different factors such as FGF-2, EGF, and CNTF, at various time points. Cells were then subjected to immunopurification via magnetic cell sorting for the “positive selection of A2B5 cells,” before being plated in neurobasal medium containing CNTF for an additional 2 weeks. A2B5 is mostly expressed in embryonic neural tissue and is therefore considered to be a marker for immature glial-committed precursors that give rise to glial types such as astrocytes and oligodendrocytes. The authors reported 99% GFAP-positive cells at fourth week of differentiation with the absence of teratoma formation in long-term experiments (Mormone et al., 2014).

Roybon et al. (2013) generated two subpopulations of human astrocytes in vitro: the first subpopulation was generated from stem cells (derived from human PSCs) and was capable of an immunological response similar to human primary astrocytes, while the second subpopulation contained mature, quiescent astrocytes (Roybon et al., 2013). When co-cultured with MNs, these astrocytes also enhanced neuronal survival and neurite outgrowth. Strikingly, they also found that the addition of either FGF1 or FGF2 was sufficient to promote transitions, maturation/quiescence without triggering inflammatory responses (TNFα or IL-1). It should be noted that although the majority of the existing protocols administer FGF2 into the media (Table 3), the exact role of FGF2 in in vitro astrocyte differentiation is uncertain. There is evidence to suggest that FGF-2 has an additional feature in gliogenesis beyond driving the propagation of neural stem cells both in vitro and in vivo. The possible mechanisms of action are: (i) FGF2 in combination with a second factor (e.g., CNTF or BMP2), but not FGF2 alone, may trigger astrocytogenesis; (ii) FGF2 may trigger early commitment of NSC to astrocytes, but is not sufficient to transform the cells to astrocytes; or (iii) astrocytogenesis may be a stressful process for NSC, and the presence of FGF2 keeps the cells alive and proliferative, while other factors could decide the way of the differentiation (Song and Ghosh, 2004; Kang and Song, 2010).

The strategy to recapitulate neurogenesis and astrogenesis during early development has gained increased attention in modern science. The generation of human RG-type cells from PSCs in a chemically defined medium without adding morphogens validates that cultured hRGs maintain a cell-intrinsic clock that regulates the progressive generation of stage-specific neuronal and glial subtypes (Duan et al., 2015). One step further, a 3D culture approach for generating human cortical spheroids (hCSs), a laminated cerebral cortex-like structure, from PSCs initially showed only a few GFAP positive cells after 35 days of differentiation. This later increased to 8% on the 76th day and almost 20% after 180th day (Paşca et al., 2015). Finally, Caiazzo et al. (2014) demonstrated that the direct conversion of human fibroblast by overexpressing a defined group of TFs (NFIA, NFIB, and SOX9) led to the rapid single-step generation of human induced astrocytes. The study makes use of inducible TFs which are selectively expressed in astrocytes. Unlike any other studies, this research indicated a direct reprogramming process with no intermediate states resembling either ESCs or NSCs. To our knowledge, this is the first report that shows astrocyte-like cells derived from fibroblasts by entirely skipping the iPSC generation. The expression of astrocyte marker proteins (S100β and GFAP) was detected in induced astrocytes. However, further experiments are required to prove the functional similarity to human astrocytes (Caiazzo et al., 2014). Therefore, regardless of approach, it is a matter of generating authentic RG-like NSC/NPCs which can be further differentiated into astroglial-like cells. To obtain a more gliogenic NSCs/NPC type, it is, therefore, necessary to induce the differentiation with RA and subsequently immune-isolate for CD133-positive cells. This principle was shown for human cells (Gorris et al., 2015) and murine cells (Kleiderman et al., 2016).

Astrocyte maturation occurs through a series of complex events, which remain poorly understood. Some reports have suggested that IL-6 family cytokines (such as LIF, CNTF) (Bonni et al., 1997) or BMPs (BMP2, BMP4) (Gross et al., 1996) have a comparable potential to induce differentiation of astrocytes from neural progenitor cells. Some published protocols have achieved astroglial maturation by culturing cells in the presence of different combinations of FGF-2, EGF, and CNTF for changing time periods (Emdad et al., 2012). Others have achieved astrocyte maturation by any of the following combinations: (i) maintenance media along with 10% fetal bovine serum (FBS) (Palm et al., 2015); (ii) maintenance media in the presence of BMP-4 (preferably for rodents); (iii) maintenance media with CNTF; or (iv) maintenance medium with bFGF, NT3 and/or PDGF-AA (Juopperi et al., 2012; Haas and Fischer, 2013; Tanga et al., 2013). These methods can generate robust populations of functionally diversified astrocytes with high efficiency, based on staining for β-III-tubulin, GFAP, and AQP4, but their functional maturation and efficacy varies considerably among different studies and cell lines.

Astrocytes can also emerge spontaneously in neuronal cultures by the “spontaneous emergence approach” (Figure 3C). For example, human iPSC-derived NPCs have been differentiated toward various neuronal lineages such as cortical neurons (Li et al., 2009), midbrain dopaminergic neurons (Yan et al., 2005) and spinal MNs, where astrocytes and oligodendrocytes frequently appear as side products in these preparations, however, the efficiency is low and not reliable. A numbers of protocols have been derived from this approach. A recent study showed the emergence of endogenous astrocytes from neuronal cultures via increased expression of the glial fibrillary acidic protein (GFAP), a marker of astrocytes, at day 100, with confirmation through the NanoString platform (Muratore et al., 2014). Other groups have also confirmed the emergence of astrocytes with long-term differentiation of hiPSCs-derived neural progenitors (Shi et al., 2012; Espuny-Camacho et al., 2013). A new 3D based protocol published by our group was able to generate neuronal cultures enriched in GFAP⁺ and Aquaporin-4⁺ astrocytes (Zhou et al., 2016). However, all of these protocols are complex, time-consuming, difficult to expand and have fewer functional properties than the in vivo models or directed approach.

The generation of astrocytes from human PSC has reached a state where the cells can be considered for clinical applications and various research questions. It is now important to take stock, to identify remaining research needs and to work toward protocols that provide standardized sets of cells. Concerning the latter point, it would be important to develop consensus criteria of how a ‘high quality astrocyte population’ should be defined. Such criteria will need to account for different applications of the astrocytes. For instance, the requirements for cells intended for transplantation will be different from those required for pre-clinical target discovery, or disease modeling, or construction of in vitro 3D tissues. However, in all instances, the cell population will need to be characterized for: (i) purity according to positive markers and also for negative markers; (ii) for maturity/region-selectivity, e.g., by using markers, secretion of growth factors and/or gene expression profiling; (iii) function, e.g., by functional assays like looking at glutamine synthesis or inflammatory activation or by quantifying the support of neuronal growth and synapse formation; as well as (iv) cell cycle state - most of the cells should be non-proliferative. To our knowledge, none of the available protocols has been fully characterized for all this information, and each available protocol still has drawbacks: (i) some of the protocols do not provide sufficient detail for reproduction; (ii) astrocyte purity and maturity is not comparable to that obtained with primary isolated astrocytes; (iii) there are technical issues that prohibit widespread use, even though the protocol may be suitable (e.g., long-term induction process that might result in >120 days protocol to generate astrocytes); and (iv) the emerging astrocytes are often poorly characterized (e.g., are only stained for classic markers such as GFAP and S100β), but hardly for maturity markers, regional identity, negative markers and additional astrocyte functional characterization. Some astrocytes have been generated from PSC in the presence of patterning molecules known to shape the nervous system. In this way either spinal cord phenotypes (Krencik et al., 2011; Roybon et al., 2013; Serio et al., 2013; Sareen et al., 2014) or astrocytes with forebrain identity have been generated (Shaltouki et al., 2013). However, most studies have not explored the region-specific phenotype of the astrocytes generated in vitro.

For this reason, there is no protocol at present that has become a gold standard, and there is still a need both for further protocol optimization and for better characterization and standardization of the resultant cell types.

Alexander disease is a progressive astrogliopathy caused by a dominant gain-of-function mutation in the GFAP gene that maps to chromosome 17q21 (Brenner et al., 2009). It is a primary disease of astrocytes that affects neural development and causes mental retardation, seizures, and megaencephaly in early childhood. There are three age-dependent clinical subtypes of AxD: infantile, juvenile and adult-onset. Infantile AxD is very aggressive and is characterized by seizures, and bulbar dysfunction with low life expectancy. Juvenile AxD is characterized by hyperreflexia and bulbar symptoms affecting children between 2 and 14 years of age, and is milder than the infantile form. Adult AxD which affects the late adolescence and beyond, has similar characteristics to the juvenile form (Prust et al., 2011).

In this disease, there is no known metabolic abnormality. Pathophysiology reveals abundant astrocytic accumulations of eosinophilic cytoplasmic inclusions, known as RF, one of the hallmarks of AxD (Molofsk et al., 2012). RF are protein aggregates composed of Vimentin, GFAP and small heat shock proteins that are present in the cytoplasm of astrocytes. The clinical signs of AxD are likely due to astrocyte dysfunction rather than astrocyte loss (Messing and Goldman, 2003). In vitro studies have provided additional clues that astrocyte-mediated effects in AxD are due to oxidative stress and reduction in the GLT-1. Therefore these observations suggest a possible pathogenic mechanism underlying neuronal loss in AxD. Other possible astrocytic failures that contribute to myelin degeneration in AxD are K⁺ buffering, and Na⁺/K⁺ATPase activity (Lanciotti et al., 2013). To address the clinical importance of astrocyte dysfunction in AxD, there is a critical need for the new model system, as so far only a Drosophila model exists (Wang L. et al., 2011).

Amyotrophic lateral sclerosis is an adult-onset neurodegenerative disease manifested by degeneration of MNs in the motor cortex, brain stem, and spinal cord, resulting in muscle paralysis and ultimately motor neuronal death (Rowland and Shneider, 2001; Kiernan et al., 2011). Most of the cases of ALS are sporadic with unknown genetic mutation; only about 5–10% of ALS cases are caused by known genetic mutations, among which 20% are caused by mutations in the copper/zinc SOD-1 gene. Mutations of SOD1 were identified through an autosomal-dormant inheritance pattern (Rosen et al., 1993). Gurney et al. (1994) were the first to recapitulate the hallmarks of ALS using a transgenic mouse model (SOD1 G93A mutant mouse) that expresses human copies of SOD1 enzyme with glycine-alanine substitution at the 93rd codon of the coding sequence (Gurney et al., 1994; Rothstein et al., 1995). Over the years, a variation of SOD1 mutants have been subsequently developed as G37R, G85R, and G86R transgenic; although these models differ in protein content, they develop the same MN degeneration that is characteristic of ALS (Turner and Talbot, 2008). In the past two decades exhaustive efforts have been made to significantly improve the understanding of ALS disease, and at least eight variations of the disease have now been identified (ALS1–ALS8) indicating the complexity of MN degeneration. At present, the G37R, G85R, and G86R transgenic models are considered to be the most reliable and accurate animal models of ALS, where they have been extensively used in understanding the mutation of SOD1 that causes cell death.

Despite the improvement in understanding ALS pathogenesis from the use of transgenic models, there is still a lack of understanding of, or effective treatment for, ALS disease. However, modeling ALS using rodents with an ALS-carrying mutation only represents a subset of the disease, and since ALS is a slowly progressive neurodegenerative disease, modeling ALS using transgenic animal models requires months of study which results in increased cost. To overcome these limitations, the research community has shown keen interest in using iPSC models for ALS disorders. Recent studies have emphasized the involvement of astrocyte dysfunction in the pathogenesis of ALS. One of the standard features of both familial and sporadic ALS is the loss of astrocyte GLT-1, (EAAT2) from the synaptic cleft; regulating synaptic transmission and preventing glutamate excitotoxicity. Supporting the hypothesis of MN degeneration caused by excitotoxicity (Barbeito et al., 2004). Multiple groups have performed ALS iPSC disease modeling from patients with familial and sporadic ALS (Chestkov et al., 2014; Liu et al., 2014). Work by Krencik et al. (2011) in was significant in broadening the tools for ALS research by demonstrating that astrocytes could be generated from human ESCs. When astrocytes generated from hPSCs were transplanted into the lateral ventricles of neonatal mice, they observed that the regional identity specified in vitro was retained. Strikingly, the transplanted cells also developed end-feet suggesting that these cells were able to contribute in the formation of the BBB (Krencik and Ullian, 2013). Recently, astrocytes directly reprogrammed from ALS patients carrying TARDBP mutations showed abnormalities typical of a TDP-43 proteinopathy, including its cytoplasmic mislocalisation. Imaging of mutant astrocytes revealed that TDP-43 mislocalisation decreased cell survival, suggesting that mutant TDP-43 is responsible for astrocyte pathology. Apparently, mutant astrocytes were not toxic when co-cultured with either control or mutant TARDBP MNs (Serio et al., 2013). In two other reports hiPSCs were utilized to study SOD1 mutation in MNs. Kiskinis et al. (2014) used high-resolution RNA sequencing technologies to identify the transcriptional and functional changes induced by the SOD1A4V dominant mutation in human MNs and found that the expression of mitochondrial-related (Jiang et al., 2013) and ER-stress-mediated genes were significantly reduced. While the other study by Chen et al. (2014) discovered an autonomous behavior of mutated SOD1 in MNs caused by destabilization of neurofilament subunits.

Hepatic Encephalopathy is another example of how astrocyte dysfunction can cause a neurological disease. HE accompanies both acute and chronic liver failure and is characterized by the accumulation of ammonia in the brain that is detoxified by astrocytic GS, resulting in increased osmotic forces, which then leads to cytotoxic oedema and astrocyte swelling. It seems that altered neurotransmission is responsible for the excessive ammonia levels that lead to the cognitive and motor impairments seen in patients with HE. In the brain of HE patients, neurons appear to be morphologically normal, but astrocytes show signs of Alzheimer type II degeneration, i.e., nuclear enlargement, prominent nuclei, chromatin changes, and neurotransmitter receptor alteration (Felipo and Butterworth, 2002; Butterworth, 2010). There have been some animal studies evaluating BBB integrity in acute liver failure, but there has been less research on this subject conducted on humans. To our knowledge, Chen et al. (2012) evaluated the hepatoprotective property of 3-genes iPSC transplantation in a carbon tetrachloride (CCl4)-induced AHF model in mice. Their findings revealed that 3-genes iPSC-based therapy not only improved hepatic functions and animal survival, but also improved CCl4-induced HE. After Chen’s group (Chen et al., 2012), there have been only limited reports addressing the functionality of hepatic cells derived from hiPSCs/hPSCs (Takayama, 2015).

Alzheimer’s disease is characterized by the progressive deterioration of cognitive functions such as memory and mental processing. Most cases of AD are sporadic, but about 1–2% are genetically linked with the early onset (EOAD) of dementia. There are two major histopathological hallmarks in the brain of AD patients, the deposition of extracellular SPs composed of the Aβ peptide, and neurofibrillary tangles, which are intracellular inclusions of hyperphosphorylated tau protein in selective regions of the brain (Sidoryk-Wegrzynowicz et al., 2011). SPs are deposits of extracellular Aβ protein derived from Aβ42, a peptide fragment of 42 amino acid residues derived from the sequential step of proteolytic processing of amyloid precursor protein by β and γ secretase. As the disease progresses, synaptic loss and neuronal death become prominent, which consequently lead to the shrinkage of the brain.

The best-studied risk factor for sporadic, LOAD is the presence of ApoE (APOE ε4), a protein involved in lipid metabolism. The presence of one APOE ε4 allele increases the risk of LOAD by threefold while two alleles of APOE ε4 can increase the risk for late-onset AD by 12-fold. Multiple studies from various populations have confirmed the increase in LOAD is caused by the presence of an APOE-ε4 allele. In contrast, APOE-ε2, another variation of this allele, confers protection against developing AD. A recent finding suggests that astrocytes play a significant role in the clearance of the Aβ peptide thus preventing the formation of plaques in the healthy brain (Wyss-Coray et al., 2003; Li et al., 2011). The precise mechanism by which astrocytes recognize and degrade Aβ is not known, but ApoE which is primarily expressed in astrocytes and to a lesser extent in microglia has been proposed to be responsible for this cellular action (Koistinaho et al., 2004; Castellano et al., 2011). The importance of APOE ε4 in AD has been well established, although further studies are needed to understand the exact molecular mechanism that leads to an increased susceptibility for AD.

Another significant aspect of AD pathogenesis is the interaction of microglia and astrocytes through the production of neurotoxic molecules. The role of astrocytes in inflammatory processes is complicated to address, though activated astrocytes are capable of phagocytosis. Activated astrocytes are characterized by hypertrophic somata in neurodegenerative diseases. The process of astrocyte activation results in reactive gliosis, which appears in late stage AD. Reactive astrogliosis in AD can be triggered by various elements, including damaged neuronal signaling and extracellular deposition of the β-amyloid peptide. Once substantial accumulation occurs, astrocytes themselves undergo apoptosis resulting in the formation of amyloid plaques positive for GFAP and S100 β. Recently, iPSC-derived neurons from different patients were found to show different accumulation of Aβ oligomers in AD models (Israel et al., 2012; Kondo et al., 2013) Therefore the use of iPSC-derived AD model can act as a testing platform for the optimal pharmacological regimen.

Neuromyelitis Optica is a primary astrocytopathy disease affecting the CNS. NMO was first described in the 19th century and was long considered to be a variant of multiple sclerosis. Pathological descriptions such as the frequent occurrence of necrosis with cavitation are used to distinguish NMO from multiple sclerosis (Marignier et al., 2010). The disease is commonly associated with diffuse cerebral white matter lesions that resemble acute disseminated encephalomyelitis, and severe demyelination affecting the optic nerve and spinal cord. The evidence of this disease reveals the loss of neurons and astrocytic damage (Wingerchuk et al., 2007). AQP-4 is one of the most valuable tools for the diagnosis of NMO (Lennon et al., 2004), and a novel potential therapeutic approach using a recombinant monoclonal anti-AQP4 antibody that selectively inhibits NMO-IgG binding to AQP-4 has been proposed (Tradtrantip et al., 2012). This study demonstrates a direct way in understanding the pathogenesis of autoimmune disease of NMO in mouse models.

Indeed, several studies have highlighted the relevance of stem cells derived astrocytes in disease modeling. Some examples are mentioned below (Table 5).