• Open Access

Efficient Generation of Astrocytes from Human Pluripotent Stem Cells in Defined Conditions§


  • Author contributions: A.S.: collection and/or assembly of data, data analysis and interpretation, and manuscript writing; J.P. and Q.L.: collection and/or assembly of data; M.S.R.: conception and design and manuscript writing; X.Z.: conception and design, data analysis and interpretation, manuscript writing, and final approval of manuscript.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLSEXPRESS January 22, 2013.


Astrocytes can be generated from various tissue sources including human pluripotent stem cells (PSC). In this manuscript, we describe a chemically defined xeno-free medium culture system for rapidly generating astrocytes from neural stem cells derived from PSC. We show that astrocyte development in vitro, mimics normal development in vivo, and also passes through a CD44+ astrocyte precursor stage. Astrocytes generated by our method display similar gene expression patterns, morphological characteristics and functional properties to primary astrocytes, and they survive and integrate after xenotransplantation. Whole genome expression profiling of astrocyte differentiation was performed at several time points of differentiation, and the results indicate the importance of known regulators and identify potential novel regulators and stage-specific lineage markers. STEM CELLS 2013;31:941–952


Astrocytes comprise a large portion of the cells in the central nervous system (CNS) and are both functionally and morphologically heterogeneous [1]. Glial fibrillary acidic protein (GFAP)-positive cells develop at various time points during normal development [2]. Early neuroepithelial cells mature to form radial glia which express GFAP and may still retain stem cell-like properties including the ability to make neurons [3–7]. GFAP-positive cells that are restricted to the astrocyte lineage arise later in perinatal development and rapidly proliferate and migrate throughout the brain to comprise more than 50% of the cells in the adult brain [8].

It is thought that these astrocytes arise via two separate lineage pathways: a prominent pathway, whereby astrocytes are differentiated from neural stem cells (NSC) via an intermediate precursor that expresses CD44 [9–11]; and an indirect pathway that involves the differentiation of a glial precursor that is restricted to make astrocytes and oligodendrocytes. The glial precursors generate CD44 immunoreactive astrocyte precursor cells (APC) [12–15] that have downregulated A2B5 and NG2 immunoreactivity [16–18]. The second pathway is more likely active in the adult brain, and in response to injury, where the glial-restricted progenitor cells (GRP) are the predominant dividing cells present [16, 19–21]. These intermediate stages of cells, such as the NSC, APC, and GRP, have shown to propagate in vitro and can be frozen and transplanted [22–24]. These cells can be harvested from fetal tissue, adult tissue [25, 26], and more recently, from pluripotent stem cells (PSC) including embryonic stem cells (ESC) [12, 27] and induced pluripotent stem cells (iPSC) [28]. Much of this work has been demonstrated in rodents and key results have been validated with human fetal cells.

In the past few years, accumulating evidence suggests that astrocytes are heterogeneous with respect to their morphology [29], developmental origin, and functional properties both in normal and diseased brain [30, 31]. Astrocytes from different brain regions as well as locally restricted astrocytes display diversity in both functional and morphological characteristics [31]. This heterogeneity of astrocytes reflects restriction on cell identity imposed by regional transcriptional regulation in response to external cues [32]. Similarly, in vitro, restriction of developmental potential in response to morphogens has been identified during astrocyte differentiation in ESC [33]. Recent emerging evidence suggests that astrocyte diversity is guided developmentally by intrinsic differences among progenitors from which they arise [32, 34–36]. Whether the regional identity of ESC-derived astrocytes is determined by inherent difference in astrocyte populations remains to be investigated.

Generation of functional astrocytes from human PSC has been reported previously [33]. However, the protocol is time consuming (>180 days in culture) and is difficult to scale-up. In this manuscript, we have generated astrocytes (up to 80% of astrocytes over a short period of 5–6 weeks) from PSC-derived NSC, including ESC and iPSC-derived NSC, using a defined medium system, and determined the properties of these cells using immunocytochemistry and gene expression profiling as well as in vivo transplantation study. We show that NSC derived from PSC can generate astrocytes, and they do so by first generating a CD44+ intermediate stage cell lineage that is restricted to maturing into astrocytes. We show that bone morphogenetic protein (BMP) and ciliary neurotrophic factor (CNTF) can direct differentiation of this precursor. In addition, neuregulin, like BMP and CNTF, can also promote astrocyte differentiation. Our gene expression profiling studies suggest that notch signaling, nuclear factor 1A (NFIA), and hairy and enhancer of split (HES) pathways are potential regulators of cell fate. In summary, these astrocytes in culture display similar gene expression patterns, morphological and functional characteristics to primary astrocytes, and survive in rodent brains after transplantation.


Cell Culture

ESC- and iPSC-derived NSC line H9, H14, and BC1 were derived as previously described [28] and maintained on culture dishes coated with Geltrex (Invitrogen, Carlsbad, CA). Cells were maintained in a humidified incubator with 5% CO2 at 37°C. Control or treated cells received fresh medium and growth factors every other day. GRP isolated from fetal brain were purchased from Q Therapeutics Inc., Salt Lake City, UT; www.qthera.com.

Astrocyte Differentiation

NSCs were plated at 50%–60% confluence on poly-ornithine-laminin or Geltrex (Invitrogen)-coated dishes in NSC medium which consist of neurobasal medium containing B27 supplement (1×), nonessential amino acids (1×), GLutaMAX-(1×), Anti-Anti(1×) (or 50 μg/ml Penn-Strep) (all from Invitrogen), and 20 ng/ml of basic fibroblast growth factor (bFGF; Sigma, St. Louis, MO). The next day, cells were switched to fresh NSC medium containing 5 ng/ml of CNTF, 10 ng/ml BMP (PEPROTECH, Rocky Hill, NJ), and 8 ng/ml bFGF in presence or absence of 1% fetal bovine serum (FBS) (Invitrogen), or to StemPro medium with supplement containing 10 ng/ml of Activin A, 10 ng/ml of Heregulin 1β, and 200 ng/ml of IGFI analog (all from Invitrogen) [37]. Medium was changed every other day and cells were passaged at least five to six times during differentiation process.

Cell counts were expressed as a percentage of total cells in a field. Total number of cells was represented by the number of Hoechst-labeled nuclei on each image. GFAP-positive cells were counted by analyzing fluorescent images using Photoshop. Four different randomly chosen fields from two independent experiments were counted by three different individuals. Values were obtained by evaluating at least 800 GFAP-positive cells per experiment. Statistical analysis was performed using the Student's t test with two-tailed distribution and assuming equal variance.


Immunocytochemistry and staining were performed as described previously. Briefly, cells were fixed with 2% paraformaldehyde for 15 minutes at room temperature. Fixed cells were washed with phosphate-buffered saline (PBS) twice and blocked using a buffer containing 10% goat serum, 4% Bovine Serum Albumin (BSA), and 0.1% Triton X-100 for 1 hour at room temperature. Primary antibody was diluted in antibody buffer containing 8% goat serum, 4% BSA, and 0.1% Triton X-100 and incubation was performed at 4°C overnight. Appropriate secondary antibodies were used in single and double labeling at room temperature for 1 hour. All secondary antibodies were tested for cross reactivity and nonspecific reactivity. The primary antibodies used here as follows: anti-GFAP antibody (1:2,000; DAKO, Glostrup, Denmark), anti-β-tubulin III clone SDL 3D10 (1:1,000; Sigma-Aldrich), anti-Nestin (1:500; BD Transduction Laboratories, Sparks, MD), anti-SOX1 and SOX2 (1:500; Chemicon, Billerica, MA), orthodenticle homeobox 2 (OTX2) (1:500; Abcam, Cambridge, MA), NKX2-2 (1;100; DSHB, Iowa City, IA), and secondary antibodies: Alexa Flour 488 Goat Anti-Rabbit, Alexa Flour 594 Goat Anti-Rabbit, and Alexa Flour 594 Goat Anti-Mouse (Invitrogen). Hoechst (Molecular Probe, Grand Island, NY; H3570) was used at 1:1,000 dilutions for nuclei staining. Images were captured using Nikon fluorescence microscope.

Gene Expression and PCR Analysis

Total RNA was prepared using the RNeasy Mini kit according to the manufacturer's instructions (Qiagen, Valencia, CA). RNAs were isolated from H19 and BC1. NSC were hybridized to Illumina Human HT-12 BeadChip (Illumina, Inc., San Diego, CA, performed by Microarray core facility at the Burnham Institute for Medical Research). All the data processing and analysis were performed using the algorithms included with the Illumina BeadStudio software. The background method was used for normalization. The maximum expression value of gene for probe set was used as the expression value of the gene. For the processed data, the dendrogram was represented by global array clustering of genes across all the experimental samples, using the complete linkage method and measuring the Euclidian distance. A differentially expressed gene was defined if the gene showed twofold expression change between any two samples. A heat map was generated by unsupervised two-way hierarchical clustering of differentially expressed genes using log 2 signal values for each gene across all samples analyzed using “The Institute for Genomic Research (TIGR) Multiexperiments Viewer v4.5.1 [38].” High expression relative to mean was colored red, whereas low expressions were colored green. Black represented no significant change in expression level between mean and sample. Expression and all cell line correlations were a measure of Pearson's coefficient, implemented in R System. cDNA was prepared from RNA using SuperScriptReverse Transcriptase kit (Invitrogen) according to manufacturer's instruction. Real-Time PCR analysis was performed using SYBR Green (Invitrogen) and gene-specific primers.

Coculture Experiment and Synapse Quantification

ESC-derived neurons were differentiated from NSC using our previously published protocol [28] for 28 days. Twenty-eight-day differentiated astrocytes were plated on a layer of neurons (4 × 105 astrocytes per 1 × 106 neurons). Cells were kept in astrocyte differentiation medium for the next 7 days and medium was changed every other day. On the seventh day, the medium was removed, and cells were washed with PBS, fixed, and stained with SYNAPSIN-1 (1:500; Abcam Cambridge, MA) and Tuj1 (1:1,000; Sigma-Aldrich) antibodies according to immunostaining procedure described previously. To quantify the synapse formation, individual neurons were randomly selected. In general, a neuron with obvious cell body was selected, which was separated and distant from other neurons by three cell bodies. On the selected neurons, yellow puncta were identified along a single axon and counted by analyzing fluorescence images using photoshop. Seventy randomly selected neurons from three independent experiments were counted blindly by three different individuals. Fold difference in number of yellow puncta by ESC-derived neurons in presence or absence of astrocytes was calculated by the mean number of puncta counted. Statistical analyses were performed using the Student's t test with two-tailed distribution and assuming equal variance.

Glutamate Uptake

The uptake protocol was a modified protocol adapted from that used by Drejer and later by Szymocha et al. [39, 40]. Cells were rinsed with 1 ml of the phenol-free Dulbecco's modified Eagle's medium (DMEM) without glutamate and glutamine (Life Technologies, Grand Island, NY) and then starved in phenol-free DMEM without glutamate and glutamine for 30 minutes at 37°C. After starvation, the cells were equilibrated in uptake buffer containing phosphate-buffered saline (with 135 mM NaC1, 3.0 mM KC1, 1.0 mM CaCl2, 0.6 mM MgSO4, 1.7 mM KH2PO4, 8.0 mM Na2HPO4, and 6 mM glucose, pH 7.4) for 5 minutes at 37°C. The assay buffer was prepared by adding L-3,4-[3H]glutamic acid (PerkinElmer, MA) to the uptake buffer (1 μCi/ml, i.e., 50 nmol), and the final glutamate concentrations were obtained with increasing concentration of unlabeled L-glutamate (5, 10, 25, 50, and 100 μM). Cells were incubated with assay buffer (2 ml) and glutamate uptake was performed at 37°C. The kinetics of glutamate uptake is linear within the first 10 minutes [41]. The uptake medium was removed after time indicated (15 minutes), cell plates were kept on ice, and the wells were rinsed with 2 ml of ice-cold PBS buffer containing excess (1 mM) glutamate to avoid reverse transport. The cells were immediately lysed with radioimmunoprecipitation assay buffer (150 mM NaCl, 1% Triton X-100, 0.1% sodium dodecyl sulfate, 10 mM Tris-HCl [pH 7.2], 1 mM EDTA, 1% sodium deoxycholate, 1% aprotinin; 500 μl/well) and 100-μl aliquots were analyzed for incorporated radioactivity by scintillation spectrometry (5 ml of UltimaGold; PerkinElmer) at an efficiency of approximately 50% (Beckman LS 6500 analyzer). Glutamate uptake was normalized to the total protein in each well. The protein content was determined by a bicinchoninin acid (BCA) protein assay (Thermo Fisher Scientific Inc, Pitsburgh, PA). For the time course assay, the unlabeled glutamate concentration was kept at 50 μM. The specificity of glutamate transport was assessed using 1 mM competitive inhibitor, DL-threo-hydroxyaspartic acid (β-THA) [42], and/or TNF-a (10 μg/ml) [40]. HEK293 cells, which do not significantly uptake glutamate compared to primary astrocytes, were used as controls. The uptake of glutamate by cells was reported as nmol of glutamate per mg of protein after being normalized. Statistical analysis was determined by Student's t tests, and significance was calculated as p < .05.

Transplantation Procedure

Eight-week-old male C57BL/6 mice were obtained from the Jackson Laboratory (Bar Harbor, ME). Mice were injected with ESC-derived astrocytes (35 days differentiated) (approximately 1 × 106 cells in 3 μl of cell preparation medium) or vehicle with a Hamilton syringe into the striatum (0.74 mm anterior to bregma, 1.7 mm lateral to midline, and 3.5 mm beneath dura). Mice were intraperitoneally injected with cyclosporine A (10 mg/kg; Novartis Pharmaceuticals, Annandale, NJ) 24 hours before transplant, and every day afterward, until the mice were sacrificed. Histological analysis of brain sections was performed 2 weeks post-transplantation. Experimental protocols were in accordance with the National Institutes of Health Guidelines for Use of Live Animals and were approved by the Animal Care and Use Committee at the Buck Institute.

For histological analysis, brains were removed and immersion-fixed overnight at room temperature. Brains were then dehydrated in graded ethanol, cleared in xylene, and paraffin-embedded [43]. Seven micrometer-thick serial coronal sections were cut and mounted on glass slides, which were dried overnight at 42°C. Sections were deparaffinized, then rehydrated through a graded series of ethanol, and washed in water. For immunostaining, sections were incubated with blocking solution using a buffer containing (2% horse serum, 1% bovine serum albumin, and 0.1% Triton X-100 in phosphate-buffered saline, pH 7.5) for 1 hour at room temperature, followed by primary antibodies incubation overnight at 4°C. Appropriate secondary antibodies were used in blocking solution for single and double labeling at room temperature for 2 hours. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole using proLong Gold antifade reagent (Invitrogen). Mouse polyclonal anti-human nuclear antigen (anti-HNA) (1:300: Chemicon), mouse monoclonal anti-human GFAP (1:500, Covance, Emeryville, CA), anti-GFAP antibody (1:400; DAKO, Glostrup, Denmark), and rabbit monoclonal anti-Thy-1 antibody (1:200, Abgent, San Diego, CA) were used in this experiment.

Statistical Analysis

Statistical analyses were performed using the Student's t test with two-tailed distribution and assuming equal variance. Newman-Keuls post hoc analysis was used when differences were observed by analysis of variance testing (p < .05).


Astrocytes Can Be Generated from ESC- and iPSC-Derived NSC

We have previously shown that NSC derived from multiple hESC and iPSC lines can be maintained in culture for prolonged periods, without losing their NSC identity and ability to differentiate into neurons and glia [44]. NSCs derived from hESC (H9 and H14) and iPSC (BC1) lines were used for this study. supporting information Figure S1 shows that >95% of H14-derived NSC was positive for SOX1 (supporting information Fig. S1A, S1B), SOX2 (supporting information Fig. S1C, S1D), PAX6 (supporting information Fig. S1E, S1F), and NESTIN, markers well-known for NSC identity. Less than 5% of the cells in our NSC culture expressed β-III-tubulin (neuron), GFAP (astrocyte), and GalC markers (oligodendrocyte) (data not shown). Similar results were obtained for H9- and BC1-derived NSC (data not shown).

To generate astrocytes, NSC were cultured and differentiated for 35 days in DMEM/F1-based medium supplemented with heregulin (a neuregulin splice variant), IGFI, activin, and FGF2, or NB medium containing CNTF and BMP2 in presence or absence of 1% FBS as described in “Materials and Methods.” Within 3 weeks of differentiation, a progressive increase in the number of cells immunostained for GFAP was observed in the culture, and the increase in GFAP staining was accompanied by morphological changes (flat and distinct astrocytic morphology) as well as a decrease in growth rate. Cells differentiated in the presence of CNTF and BMP2, regardless of the presence of serum in the culture media and showed significantly higher levels of GFAP+ cells compared to control cells in NSC medium (Fig. 1A–1F). The increase in GFAP+ cells coincided with a decrease in the number of Tuj1+ neurons. Inclusion of 1% FBS enhanced astrocyte differentiation slightly but significantly. The most dramatic effect with respect to astrocyte differentiation was observed in the presence of neuregulin (Fig. 1G, 1H), consistent with previous report on the importance of epithelial growth factor receptor family member ErbB signaling in astrocyte development [45].

Figure 1.

Differentiation of neuroepithelial cells into astrocytes. (A, B): Human NSC (H14) cultured in neurobasal medium supplemented with B-27 and FGF2. (B): Higher magnification of (A). (C, D): Human NSC (H14) differentiated for 5 weeks in neurobasal medium supplemented with B-27 (10 ng/ml), FGF2 (5 ng/ml), CNTF (10 ng/ml), and BMP2 and in presence of 1% FBS or (E, F) absence of 1% FBS. (D, F): Higher magnification of (C) and (E), respectively. (G, H): Human NSC (H14) in defined medium displayed characteristics of astrocytes as determined by GFAP immunostaining. (H): Higher magnification of (G). (I): The percentage of cells displaying GFAP-positive staining. Cells in four different fields per three independent experiments were counted blindly. The percent of GFAP-positive cells was calculated as the ratio of cells showing GFAP+ staining compared to the total number of cells with Hoechst-stained nuclei (**p < 0.001 [t test]). (J, K): Human NSC (BC1) and (L, M) Human NSC (H9) differentiated for 5 weeks in defined medium display high percentage of GFAP immunostaining. Scale bars in A, C, E, and G = 100 μM. Scale bars in B, D, F, H, and J–M = 100 μM. Abbreviations: CNTF, ciliary neurotrophic factor; DM, defined medium; FBS, fetal bovine serum; GFAP, glial fibrillary acidic protein; NSC, neural stem cells.

Quantification of GFAP+ cells after 5 weeks of differentiation showed that 24%, 34%, and 62% of total cells were expressing GFAP in medium with (a) CNTF and BMP, (b) CNTF, BMP, and FBS, and (c) neuregulin, respectively (Fig. 1I). Similar results were obtained from NSCs derived from BC1 and H9 (Fig. 1J–1M and supporting information Fig. S2A). Astrocyte identity of these cells was further validated by S100β immunostaining, which showed colocalization with GFAP in majority of cells (supporting information Fig. S2B, S2C). In addition to quantifying GFAP+ cells by immunostaining, we used flow cytometry to independently verify the quantification of GFAP+ cells. supporting information Figure S3A, S3B shows that cells, after 5 weeks of differentiation, in the presence of neuregulin displayed a high percentage of GFAP positivity (69% for H14 and 80% for BC1) in cells of live population when compared with isotype-treated control.

Development Passes Through a CD44+ Intermediate Stage and Cells at the Intermediate Stages can be Cryopreserved

We performed a time course study of the expression of GFAP and other markers in medium supplemented with neuregulin and showed that astrocyte differentiation went through a CD44+ cell stage. CD44 was expressed nearly uniformly in day 12 culture prior to a robust increase in GFAP staining, which marks differentiation into mature astrocytes (Fig. 2A, 2B). At 14 days, a small percentage of cells stained positive for GFAP, while the majority of cells were stained positive for Tuj1 (Fig. 2C, 2D). The percentage of GFAP-positive cells increased gradually during the following week (21 days) (Fig. 2E, 2F). By 28 days, a significant, robust increase in GFAP staining could be observed, with the majority of cells displaying positive immunoreactivity to GFAP antibodies (Fig. 2G, 2H). By 35 days, approximately 60%–80% of the cells was immunopositive for GFAP (Fig. 2I, 2J). Similar results were obtained by NSCs derived from iPSC line BC1 (supporting information Fig. S4).

Figure 2.

Time course study of neural stem cells (NSC) (H14) differentiation using defined medium into astrocytes as imaged by GFAP staining. (A, B): CD44 immunoreactivity precedes GFAP staining upon differentiation of NSC in defined medium. (A): Live staining of NSC (H14) in defined medium at day 12, showing majority of cells immunopositive for CD44 staining. (B): Phase contrast of NSC (H14) in defined medium for 12 days merged with CD44. (C, D): GFAP immunostaining of cells cultured in defined medium for 2 weeks. Scale bar in C = 100 μM. (D): Higher magnification of (C). Scale bar in D = 100 μM. (E, F): At 3-week time point, the number of cells immunopositive for GFAP staining began to increase and continued. Scale bar in E = 100 μM. (F): Higher magnification of (E). Scale bar in F = 100 μM. (G, H): By 4 weeks, a robust increase in GFAP staining was observed, and by 5 weeks majority of cells stained positive for GFAP and had differentiated into astrocytes (I, J). Scale bars in G and I = 100 μM. (H, J): Higher magnification of (G) and (I), respectively. Scale bars in H and J = 100 μM. Abbreviation: GFAP, glial fibrillary acidic protein.

We then assessed whether cells at intermediate stages of differentiation can be cryopreserved with sufficient viability while maintaining the ability to differentiate into astrocytes. The viability upon thawing, using trypan exclusion assay, for H14 was 78% and 70% at day 10 and day 14, respectively (supporting information Fig. S5A–S5C). Furthermore, after thawing, these cells retained their ability to differentiate into astrocytes upon culture to a comparable level observed previously (supporting information Fig. S5D, S5E). The ability to store cells at intermediate stages is of obvious relevance to scaling-up and obtaining sufficient numbers of cells for screening or for cell replacement therapy.

Transcriptome Profiling of Astrocytes Derived from PSC

We next used microarray platforms to generate whole genome expression profiles of astrocytes generated from ESC and iPSC. To identify gene expression changes during astrocyte differentiation, we performed a side by side whole genome microarray comparison of differentiation at days 14, 21, 28, and 35. The overall correlation coefficient was calculated between each population by Genome Studio software and presented as a dendrogram (Fig. 3A). Clustering of cells at various time points of differentiation revealed a distinct gene expression pattern for astrocytes compared to NSCs from which they were derived. The dendrogram showed that NSC, regardless of ESC (H14) or iPSC (BC1) origin, clustered closer during early differentiation process (NSC and day 14 stages) than late stages (days 21, 28, and 35 stages). Unsupervised hierarchical gene clustering of all the genes expressed showed clustering between the NSC samples, as well as their differentiated progenies, and that the gene cluster of astrocytes was clearly distinct from those of NSC (Fig. 3B).

Figure 3.

Expression profiling demonstrated differences between NSC and NSC-derived astrocytes. (A): Dendrogram and sample clustering of data from various time points following incubation of NSC with defined medium. Within each branch, vertical distance represents similarities of gene expression among samples. Both H14- and BC1-derived NSC displayed similarities in gene expression represented by short vertical distance, at early step of differentiation. Gene expression became less correlated as H14 and BC1 continued differentiation pass 21 days. (B): Microarray heat map presented using unsupervised hierarchical clustering of whole genome expression data from undifferentiated NSC in presence or absence of defined medium revealed two distinct cell populations with differential gene expression patterns. Samples with higher than average expression are shown in red, and samples with lower than average expression are shown in green. Each time point is represented by a single column. Each gene is represented by single row of colored boxes. (C): Validation of candidate astrocyte-associated genes from the array data. RNA was isolated from NSC or their differentiated progenies at different time points using the RNeasy Mini kit according to the manufacturer's instructions (Qiagen). cDNA was made from RNA using SuperScriptReverse Transcriptase kit (Invitrogen) according to manufacturer's instruction. Real-time PCR analysis was performed using SYBR Green (Invitrogen) and gene-specific primers. Results are representative of two independent experiments performed in triplicates. ESC-derived NSC differentiated for 35 days in defined medium expressed higher levels of CHD5, NFIX, NFIA, TMOD2, OMG, LHX2, HOPX, and TLE2 in addition to GFAP when compared with NSC from which they were derived. Error bars represent ± SEM. Abbreviations: GFAP, glial fibrillary acidic protein; NSC, neural stem cells; OMG, oligodendrocyte-myelin glycoprotein; TMOD, tropomodulin.

Quantitative comparison of gene expression of cells at various days of differentiation allowed for identification of differentially expressed genes in each stage. As expected, several known NSC markers such as SOX1, SOX2, NESTIN, and PAX6 were highly expressed in NSC samples, and were absent or minimally expressed in the differentiated population. In concordance with our time point study, the array data showed a gradual increase in the expression of GFAP during differentiation, with maximum expression at 28 and 35 days of differentiation. Neurogenesis in the early stages of differentiation (e.g., days 14 and 21) was evident by the high expression of neuron-associated genes such as neural cell adhesion molecule 1, β-III tubulin, neurogenin 2, and neurogenic differentiation 1, which were downregulated in day 35 when astrocyte became the predominant cell type in the culture (please refer to the complete array dataset).

The complete gene expression data and differentially expressed genes at different time points of differentiation are available in Table A (available upon request). We focused our analysis on the clustered genes with significant changes during differentiation as potential key regulators of this process. The list of genes expressed 30-fold higher at 35 days differentiation time point compared to those of NSC (supporting information Tables S1, S2). In addition to the well-known astrocyte markers, such as GFAP and S100β [46], several genes that are highly enriched in our day 35 culture have been described as markers for astrocytes. These include transcription factors, such as NFIA [47], NFIX [48], as well as aquaporin 4 [49], in addition to s100β [46] and GFAP (Table S3 and Fig. S6 available as supporting information).

A closer examination of genes highly expressed in astrocytes along with the published data from rodent [50–52] studies suggested that several signaling pathways may play roles in astrocyte differentiation. These included Notch signaling, NF1A, NFIX, Hairy/enhancer-of-split related with YRPW motif protein 2 (HEY2), and HES6. Not surprisingly, SMAD, BMPs, and CNTF were also expressed as well as CNTF via Jak-Stat signaling.

We then used quantitative RT-PCR analysis to confirm some of the new stage-specific genes in astrocytes. These include top astrocyte enriched genes, such as tropomodulin (TMOD) [53], which has been implicated in negative regulation of neurite survival and formation, chromodomain helicase DNA-binding protein 5 (CHD5) [54, 55], NFIA [47], NFIX [48], TMOD2 [53], oligodendrocyte-myelin glycoprotein (OMG) [56], LIM-homeodomain (LIM-HD) (ILH2) [57], homeodomain only protein (HOPX) [58], and transducin-like enhancer of split (TLE2) [59] (Fig. 3C).

Since comprehensive gene expression signature of human astrocytes is currently lacking, we compared our data with a previously published murine astrocyte profile [51]. Clustering of the top 40 candidate astrocyte-associated genes from murine astrocyte profile with our array data revealed similarities in a number of genes enriched in both mouse and human astrocytes (supporting information Fig. S7A). Of the 40 most highly expressed transcripts in murine primary astrocytes, 26 genes were present in our array dataset. Many of these shared genes were either astrocyte-specific, such as ALdhL1, or were implicated to play a role in astrocyte physiology, such as SLC25A18 (solute carrier family 25, member 18), GLAST (GLutamate ASpartate Transporter), SLC4A4 (solute carrier family 4, member 4), ATP1A2 (ATPase, Na+/K+ transporting, alpha 2 (+) polypeptide), F3 (factor III), ALDOC (aldolase c), and FZD2 (frizzles family receptor 2). Genes that were absent in our dataset included: CHRDL1, TTPA, DIO2, and TLR3 (supporting information Table S4 for the complete list). These were also absent in the dataset of primary APC we have published earlier [52], suggesting that murine and human astrocytes, despite their similarities, may be subject to different signaling pathways, perhaps due to their differences in developmental complexity (supporting information Fig. S7A, S7B).

Human-Derived Astrocytes Are Heterogeneous in Cell Fate Identity

To determine the regional identity of the generated astrocytes, we used our gene expression data and examined the expression patterns of several well-known homeodomain markers important for anterior posterior axis formation. We found that at the onset of astrogenesis (day 28), concomitant with an increase in GFAP expression, the levels of engrailed 1 (EN1), homeobox gene B4 (HOXB4), OTX2, and paired box 2 (PAX2), markers well known for the midhind brain fate, increased (supporting information Fig. S8A). supporting information Figure S8B shows the expression of NKX2-2, a hind brain and ventral spinal cord marker, and OTX2, a mid-brain marker, in our astrocyte culture. Although anterior hind brain markers such as growth response 2 (EGR2; Krox20) and gastrulation brain homeobox 2 were expressed, their levels were decreased in day 35 astrocytes (data not shown). More caudal homeodomain markers such as HOXA1, HOXB1, HOXB6, and HOXC6 were also present (data not shown). Markers representative of forebrain identity such as forkhead box G1 (FoxG1) and LIM homeobox 2 (LHX2) were both present and upregulated, respectively. Although these data suggest that these PSC-derived astrocytes exhibit heterogeneity in their cell fate identity, it remains to be examined directly.

Astrocytes can be Efficiently Generated from Fetal-Derived Cells Using the Same Medium

Given the availability of fetal derived cells, we wish to test whether our differentiation protocol for PSC-derived NSCs can be used for generating astrocyte from their fetal counterparts. We show that fetal-derived GRP displayed a progressive increase in GFAP immunostaining and a decrease in Tuj1 immunostaining when cultured in our defined medium during a 3-week time course of differentiation (supporting information Fig. S9A–S9F). Greater than 80% of the cells expressed GFAP after 3 weeks of differentiation. This result suggests that our protocol allows generation of astrocytes, not only from NSC but also from their restricted progeny, GRP [60].

ESC-Derived Astrocytes Exhibit Functional Characteristics Similar to Those of Primary Astrocytes

To test functionality of the NSC-derived astrocytes in our defined medium, we examined their abilities to enhance synapse formation by ESC-derived neurons. We used our previously established protocol for derivation of neurons from ESC-derived NSC using a xeno-free defined medium [28] and cultured 28 days differentiated astrocytes on a feeding layer of predifferentiated neurons (day 28 of differentiation). After 7 days of coculture, we quantified synapse formation by coimmunostaining with SYNAPSIN-1 and Tuj1 antibodies. Figure 4A–4D shows colocalization of SYNAPSIN-1 and Tuj1 as determined by the appearance of yellow puncta, which are few when ESC-derived neurons are cultured in the absence of astrocytes. The number of synapse formation (yellow puncta) increased significantly (approximately sevenfold) when astrocytes were cocultured with neurons for 7 days (Fig. 4B, 4D, 4E). Interestingly, the number of synaptic puncta remained significantly higher in the presence of astrocytes even when we used ESC-derived neurons that were only differentiated for 14 days (data not shown).

Figure 4.

Neural stem cells (NSC)-derived astrocytes enhance synapse formation by embryonic stem cell (ESC)-derived neurons. (A–D): Synapse formation is determined by colocalization of SYNAPSIN-1 (red) and Tuj1 (green) as yellow puncta on DA neurons. (A, C): ESC-derived neurons in the absence of astrocytes have a few yellow puncta. (B, D): The number of yellow puncta increased when ESC-derived neurons were cocultured with exogenous astrocytes for 7 days. (E): Quantification of synapse formation by measuring the number of yellow puncta per unit of neurons (total of 70 neurons per experimental condition) showed a sevenfold increase of synaptic puncta in the coculture system (arrows: yellow puncta hence colocalization). Error bars represent the SE. Results are representative of three independent experiments. Scale bars in A and B = 50 μM and in C and D = 75 μM. (F): Kinetic of glutamate uptake by human ESC-derived astrocytes. Velocity of glutamate uptake in astrocyte culture as a function of exogenous glutamate concentration. Astrocytes were starved for glutamate for 30 minutes prior to measurement of glutamate uptake. The media were replaced with uptake buffer in the presence of increasing amounts of unlabeled L-glutamate (5, 10, 25, 50, and 100 μM) and using 1 μCi/ml [3H]glutamate. Glutamate uptake was then determined after a 15-minute incubation. Data were converted to Lineweaver-Burk plots to obtain Vmax of 77.27 and Km of 46.61. Each time point is the mean ± SE, and n = 3. (G): Kinetic of cellular uptake of [3H]-L-glutamate in cultures of ESC-derived astrocytes was measured in the presence of 50 μM concentrations of cold L-glutamate [G°] (black bars), with glutamate transport inhibitors, TNF-a (10 ng/ml) (dotted bar), and/or β-THA (100 μM) (horizontal bars) (n = 3 for each group) or combination of the two (vertical bar). Results are means SEM of triplicates from two independent experiments. Statistically significant difference, relative to glutamate uptake by 293T cells; **p < 0.001 (Student's t test).

Another functional characteristic of astrocytes is their ability to regulate the glutamate homeostasis. To determine whether ESC-derived astrocytes had retained their ability to transport glutamate, we measured the kinetics of the cellular uptake of L-glutamate, a substrate for glutamate transporter. As summarized in Figure 4F, The Michaelis-Menten kinetic parameters, Km and Vmax, of glutamate uptake were measured following culture of astrocytes in the presence of labeled [3H]-glutamate and increasing concentration of unlabeled glutamate (5, 10, 25, 50, and 100 μM). Figure 4F shows the uptake (in nmol/mg per minute) as a function of glutamate concentration. The curve is fitted to the experimental data points by computer software (GraphPad Prism 6). The half-maximal transport constant (Km) of 46.61 and Vmax of 77.27 was determined by conversion of data to Lineweaver-Burk plots. Figure 4G (white bars) shows that ESC-derived astrocytes exhibited a time-depended increase in the mean values of glutamate uptake which remain linear up to 60 minutes. The rate of uptake of [3H]-L-glutamate was significantly higher in ESC-derived astrocytes at 60 minutes when compared with that of HEK293T cells. Inclusion of TNF-a (10 ng/ml) (dotted bar) and/or glutamate transport inhibitors β-THA (100 μM) (bars with horizontal or vertical lines) abolished the uptake of functional glutamate receptors and transporters.

Astrocytes Derived from PSC Can Survive and Integrate In Vivo

To examine the functionality of astrocytes generated from PSC by our protocol, we transplanted day 35 astrocytes into the mouse striatum. Histological analysis of brain section 2 weeks post-transplantation revealed the existence of donor cells of human origin, as determined by positive staining of HNA at the site of transplantation (supporting information Fig. S10A). Using a human-specific antibody against GFAP, we showed the survival of human astrocytes at the site of transplantation (Fig. 5A–5C), whereas no human GFAP+ cells were observed in mouse brain sections of a corresponding region on the contralateral side (Fig. 5D–5F). This data were further confirmed by colocalization of HNA and GFAP staining in same donor cells (Fig. 5G–5J), suggesting that human astrocytes can indeed survive the transplantation process. We did not observe any Tuj1+/HNA+ cells in the graft (data not shown), nor detected a localization of HNA and Thy1, a marker for mature neurons [61–63], suggesting that donor cells do not differentiate into neurons (supporting information Fig. S10A–S10D) nor function as stem cells.

Figure 5.

Human donor cells expressed GFAP at the site of transplantation. (A): Confocal imaging of immunostained cells at the site of transplantation with a human-specific anti-GFAP antibody. (B): Hoechst nuclear staining of the panel in (A). (C): Majority of nuclei at the site of transplantation belonged to human astrocytes. Scale bars in A, B, and C = 20 μM. (D–F): Mouse brain cells did not show staining with human GFAP antibody. (D): Mouse brain section was prepared from a site that corresponds to the site of transplantation in experimental animal. Confocal microscopy of mouse brain cells failed to detect immunoreactivity for human GFAP antibody. (E): Hoechst staining of the panel in (D). (F), (D), and (E) merged. (G–J): Confocal microscopy of brain section double stained with anti-HNA and GFAP showed that some human cells also expressed GFAP. Scale bars in D, E and F is 100 μM. (G): Nuclei staining of the cells at the site of transplantation using HNA antibodies. (H): Hoechst staining. (I) Confocal imaging of the transplantation site showed the presence of GFAP immunopositive cells. (J): GFAP-positive cells at the site of transplantation also stained positive for HNA. Scale bars in G–J = 20 μM. Abbreviations: GFAP, glial fibrillary acidic protein; HNA, human nuclear antigen.


In the past, efforts have been made to isolate astrocytes from the developing brain, but the recent advancement in PSC generation and differentiation offers a renewable source of neural cells which hold a great promise for regenerative medicine [64–66]. Recently, astrocytes have been generated from human PSC [33, 67] with different protocols, and in general, it takes a prolonged period of culture (e.g., over 180 days) to generate astrocytes from human PSC [68]. In this study, we described a xeno-free defined medium system that allows for an efficient and rapid generation of functional astrocytes from NSC derived from human PSC (∼70% GFAP+ cells in 5 weeks), as assessed by in vitro gene expression analyses and in vivo transplantation study. Our protocol is suitable for scale-up, as cells at various time points can be cryopreserved without impacting their ability to differentiate into astrocytes.

In developing brain, proliferation and differentiation of NSC, as well as those of glial progenitors, are controlled by various signaling pathways and factors. Cytokines such as CNTF, a member of IL-6 family, have been reported to induce differentiation of cerebral cortical precursor cells into astrocytes and inhibit their neuronal differentiation [69]. CNTF exerts its effect through downstream Janus tyrosine kinase/signal transducers and activators of transcription (JAK/STAT) pathway, leading to phosphorylation and nuclear localization of STAT3. BMP2, a member of transforming growth factor-β superfamily, has also been implicated in astrocyte specification. Treatment of multipotent NSC with BMPs (2, 4, 5 and 7) induces their differentiation into astrocytes [70]. In the nucleus, activated SMAD cooperates with STAT3 through a complex formation that involves p300 leading to synergistic induction of the GFAP gene [70, 71]. In concert with previously published data, we showed that treatment of NSC, with both CNTF and BMP2, enhanced the number of GFAP-positive cells in our differentiation culture. However, of the factors we tested, the highest efficiency of astrocyte differentiation was achieved in culture differentiated in the presence of heregulin [37], an alternative spliced form of neuregulin, and a member of epidermal growth factor (EGF)-like ligands that promote astrocyte maturation and survivability in primary rat culture [72]. Heregulin is synthesized by neuron and non-neuronal tissues such as astrocytes [72], oligodendrocytes [73], and Schwann cells [74], and acts as paracrine/autocrine ligands for the HER family receptors, activating downstream signaling, Rat sarcoma (RAS)/mitogen-activated protein kinase, and/or phosphatidylinositol 3-kinase/AKT pathways. A more recent report has demonstrated the activation of STAT3 by EGF and heregulin, that signal through receptor tyrosine kinases, as well as nonreceptor tyrosine kinases such as Src kinase [75–77].

Our data suggest that neuregulin plays a role in enhancing astrocytes generation, as has been shown previously for oligodendrocyte development [78]. Examination of our array dataset shows a high level expression of epidermal growth factor receptor ERBB2, a more modest level of ERBB3, with no detectable expression of ERBB4, suggesting that the predominant receptor is a complex containing ERBB2. The ERBB receptor complex is known to activate the Akt/phosphatase and tension homolog (PTEN) pathway. Examination of our array data shows that all components of this pathway are present, which is consistent with the hypothesis that heregulin regulates glial maturation and acts via ErbB receptor signaling to activate the Akt/PTEN pathway.

Activin, another growth factor in the medium supplement, is a member of TGF-β superfamily which was implicated in enhancement of the leukemia inhibitory factor (LIF)-mediated astrocyte differentiation in neural precursor cells [79, 80]. This enhancement was postulated to be as a result of cooperation of SMAD, a downstream signal transducer of TGF-β-signaling, with Stat transcription factor. However, in our experiments, activin did not promote astrocyte generation (unpublished results).

In addition to astrocytes that constitute the vast majority of GFAP immunoreactive cells in the brain, several specialized glia are also present. The GFAP expressing cells we detected in culture appeared to be astrocytes: they did not express low affinity nerve growth factor receptor, estrogen receptor alpha (hallmarks of olfactory ensheathing glia) [81] or the characteristic radial morphology of Bergman and Muller cells, or radial glial [82]. Our culture conditions did not prove regional differentiation into cerebellar or retinal phenotypes, further suggesting that these specialized glia were not present in our cultures [83, 84]. Unlike radial glia which can generate neurons [85], our cells did not differentiate into neurons either in vitro or in vivo once they began expressing CD44 and GFAP, suggesting that they are GFAP immunoreactive cells.

The in vivo data regarding S100β and its relevance as a late marker of astrocyte development are not conclusive. Many reports suggest that the onset expression of S100β characterizes a terminal maturation stage as it is the case for cortical astrocytes, while others have reported a high level of S100β in undifferentiated astrocytes [86, 87]. In our astrocyte cultures, the expression pattern of S100β coincides with the onset of astrogenesis. However, the induction of S100β transcript (day 21) occurs prior to upregulation of GFAP expression (day 28) (supporting information Fig. S6). Validation for this observation is supported by a time course study for S100β and GFAP expression analyzed by immunostaining (data not shown). Although GFAP expression proceeds that of S100β, and colocalizes with nuclear or cytoplasmic S100β protein in majority of cells, few cells remain GFAP+/S100β−. Whether the presence of these GFAP+/S100β− recapitulates the heterogeneity nature of astrocytes remains to be determined, but we cannot exclude the possibility of the loss of S100β in a fraction of astrocytes as they mature.

Comparing the properties of the astrocytic cells we have generated with fetal tissue derived cells, we find that they appear quite similar. Both cells respond to the same media components and growth factors in a similar fashion. Both cells can be readily propagated in culture, in serum-free medium, and the media condition we identified for optimal propagation of PSC derived astrocytes works equally well with fetal astrocytes.

Transcription profiling of NSC and astrocytes during four time points of differentiation (days 14, 21, 28, and 35) potentially reveals several biological pathways that may participate in the differentiation of astrocytes. Here, we identify CHD protein 5, NFIA, X, and OMG among most strongly expressed genes, in our defined medium, as potential astrocyte markers. Interestingly, CHD5 was previously reported to negatively regulate neural cell fate decisions in murine CNS as a part of nucleosome remodeling and histone deacetylation complex [55, 88]. NFIA, a transcription factor with CCAAT box element-DNA binding domains [89, 90], was recently implicated in onset of gliogenesis in mouse embryonic spinal cord progenitors, and was shown to be necessary for the maintenance of HES1, an effector of Notch signaling important in inhibition of neurogenesis [47]. Both NFIA and NFIX have overlapping expression pattern during embryogenesis in murine model [90], and their knockout phenotype was shown to reduce expression of late astrocytic markers, such as GFAP [91–93], through a mechanism that involves demethylation of astrocyte-specific gene and alteration in chromatin structure [94]. In keeping with previous reports from rodent and human primary astrocytes culture, our data suggest a similar potential instructive role for NFI proteins in promoting human ESC gliogenesis [48, 95]. Our gene profiling also identifies HOPX and LHX2 as possible putative astrocyte markers. Although increased expression of LHX2 was previously observed in gene profiling of human primary astrocyte culture [57, 96], potential involvement of HOP in mediating differentiation of astrocytes from PSCs is novel and requires further investigation. We anticipate that HOP may participate in the process by nucleosome remodeling as it does in cardiac muscle differentiation [97].

It has been postulated that astrocyte heterogeneity, in early development, is determined by position-dependent extrinsic signals as well as regional restrictions imposed by combinatorial interactions of homeodomain and bHLH transcription factors [32, 35]. Interestingly, this diversity of astrocyte phenotypes is also reflected in their gene expression profile as astrocytes from distinct regions of CNS have shown to exhibit molecular and functional differences [96, 98]. In our analysis of gene expression data, we identify a temporal pattern of homeodomain code for ESC/iPSC-derived astrocytes. The homeodomain code of these astrocytes indicates cell fate of forebrain, hind brain, and spinal cord identity. Our work supports the previously published data on PSC-specified astrocytes in retaining a regional identity during in vitro differentiation process [33]. The relevance of these multiple regional markers, detected from our array data, to the biological and functional properties of astrocytes remains to be addressed. Challenges remain to investigate if their regional identity is more restricted in vivo under the guidance of the regional signaling milieu, whether they differentiate into region-specific cells following transplantation; or if a more heterogeneous population of astrocytes will be more plastic in their positional identity.


Overall, our results suggest that astrocytes derived by this methodology will be useful for both cell therapy and screening applications. These astrocytes express major components of the glutamate transport and have appropriate glucose and iron metabolism and the appropriate transporters. Astrocytes can be obtained in sufficient numbers and can be frozen, thawed, and aliquoted into 96-well plates for screening. These cells can be readily transfected with reporters, and we can generate astrocytes from engineered lines as well as disease-specific iPSC lines.


This work was supported in part by California Institute for Regenerative Medicine Grants TR-01856 and CL1-00501. We are thankful to Dr. Andrzej Swistowski (XCell Science Inc.) for providing dopaminergic neurons for coculture experiment. We thank Steven S. Spusta for technical assistance with the flow cytometry and Justine Montoya-Sack for her assistance with the cell counting.


The authors indicate no potential conflicts of interest.