Human stem cell-derived spinal cord astrocytes with defined mature or reactive phenotypes

SUMMARY

Differentiation of astrocytes from human stem cells has significant potential for analyzing their role in normal brain function and disease, but existing protocols generate only immature astrocytes. Using early neuralization, we generated spinal cord astrocytes from mouse or human embryonic (ESCs) or induced pluripotent (hiPSCs) stem cells with high efficiency. Remarkably, short exposure to FGF1 or FGF2 was sufficient to direct these astrocytes selectively toward a mature quiescent phenotype, as judged both by marker expression and functional analysis. In contrast, TNFα and IL-1β but not FGFs, induced multiple elements of a reactive phenotype but did not affect maturation. These phenotypically defined, scalable populations of spinal cord astrocytes will be important both for studying normal astrocyte function and for modeling human pathological processes in vitro.

INTRODUCTION

Astrocytes play multiple roles in the central nervous system (CNS). Many of these are critical to normal function in the healthy adult, where astrocytes act as support cells for neurons, regulating cerebral blood flow, energy reserves and neurogenesis (Abbott, 1988; Allaman et al., 2011; Lie et al., 2005; Okamoto et al., 2011; Takano et al., 2006). Moreover, their involvement in synaptic transmission and plasticity has been the object of intensive study (Bernardinelli et al., 2011; Panatier et al., 2011). Cell culture models used to study normal astrocyte function should therefore reflect this mature, quiescent phenotype.

However, astrocyte phenotypes are extremely dynamic. Reactive phenotypes can be induced by proinflammatory factors such as TNFα, IL-1β and IFNγ in a wide variety of traumatic and pathological contexts (Gwak et al., 2012). Indeed, multiple classes of reactive astrocytes can be identified depending on the nature of activating stimulus and the time elapsed post-activation (Zamanian et al., 2012). Astrocyte activation leads to production of a wide array of mediators, including chemokines, inflammatory cytokines and growth factors (Allaman et al., 2011), as well as transcriptional changes of genes associated with cell-to-cell communication (e.g. connexins), cytoskeletal structure (e.g. GFAP) and many others (for review, see Sofroniew, 2009). It is therefore essential to precisely model the reactive state for in vitro studies of human disease.

Human astrocytes have been cultured from fetal or adult post-mortem CNS using expansion of neural precursors (Caldwell et al., 2001; Haidet-Phillips et al., 2011; Lee et al., 1993; Verwer et al., 2007) but such preparations are rare and intrinsically variable. One of the first protocols to report differentiation of hESCs (human embryonic stem cells) into astrocytes was that of Krencik et al. (2011). However, one practical drawback of the method is that it necessitates 6 months of culture to generate a sufficiently pure population (Krencik et al., 2011; Krencik and Zhang, 2011). Since then, other protocols starting from neural precursor cells have reported generation of astrocytes within 35–80 days (Emdad et al., 2012; Juopperi et al., 2012; Lafaille et al., 2012; Serio et al., 2013; Shaltouki et al., 2013). However, by the criteria discussed below, the astrocytes generated in each case are immature and do not fully model normal astrocyte function.

Astrocyte maturation occurs through a complex series of events that remain incompletely understood. There is considerable overlap between expression of different markers and it is likely that the precise sequence of their appearance varies from one region of the CNS to another. Nevertheless, we have constructed a tentative synthesis of the appearance of known markers during maturation, based on spinal cord data where available (Figure S1). Overall, astrocyte development and maturation encompasses two phases (see Supplementary Text for full review, abbreviations and citations). During the first - mainly embryonic – phase, subsets of astrocytes are generated from radial glia and progressively take on their principal functional phenotypes. Subsequently, over the first postnatal weeks in rodents, astrocytes adopt a mature, quiescent morphology and phenotype. Although all potential marker genes have not been studied in parallel in a single brain region, the sequence of appearance of markers during the first, embryonic phase is likely NF1A>GLAST>ALDH1L1>Cx43>S100β>CD44>AldolaseC>GFAP. The NF1A⁺/GFAP⁺ cells generated by extant stem cell differentiation protocols (see above) likely correspond to this first, immature stage. In contrast, the second, maturational phase is associated with down-regulation of GFAP, GLAST and ALDH1L1, while GFAP expression persists in white matter astrocytes. In parallel, there is continued accumulation of Cx43, followed by Aqp4 and the mature astrocyte glutamate transporter GLT1. Therefore, mature gray matter astrocytes should not be expected to express high levels of GFAP, so other markers are needed to monitor their maturation and purity.

To summarize, astrocytes may adopt either a quiescent state with protoplasmic morphology, characterized by low GFAP and high GLT1, or a fibrous, reactive phenotype characterized by high GFAP and low GLT1. Standard preparations of cultured GFAP⁺ astrocytes (McCarthy and de Vellis, 1980) reflect only the latter (Zamanian et al., 2012). Therefore, a robust model of mature, quiescent astrocytes would be a significant step forward for in vitro studies of human neural function as well as disease. This is especially significant given the different effects of immature and mature astrocytes on axonal regeneration (Goldshmit et al., 2012; Tom et al., 2004). Here we report that using identified signaling factors, mouse or human spinal cord astrocytes generated from either ESCs or hiPSCs (human induced pluripotent stem cells) can be induced to adopt phenotypes that correspond to those of either mature or reactive astrocytes in vivo. Such defined populations of mature human astrocytes will considerably increase the resolution of studies of normal astrocyte function, while fully activated preparations open the door to more precise disease modeling based on patient-derived human iPS cells.

RESULTS

Astrocytes derived from mouse embryonic stem cells are immature

Since the development and maturation of astrocytes are better understood in rodents than in human, we first generated astrocytes from mouse ES cells (mESCs). Our overall aim was to generate astrocytes of spinal cord identity as a tool for modeling motor neuron diseases such as amyotrophic lateral sclerosis (ALS). We therefore used an induction protocol that specifies ventral spinal progenitors (Wichterle et al., 2002) followed by neurosphere expansion, which amplifies neural progenitors and generates largely glial cells (Kuegler et al., 2012; see Figure S1 and Experimental Procedures). After 1 week, cultures of dissociated neurospheres contained cells that stained for canonical markers of developing astrocytes: GFAP, Aqp4, S100β, vimentin and NF1A (Figure 1A, B) and their spinal cord identity was confirmed by co-expression of HoxB4 (Figure 1A). The pluripotency marker Oct4 was undetectable (Figure S2C), and neurons (βIII-tubulin⁺), microglia (Iba1⁺) and oligodendrocytes (CNPase⁺ or Olig2⁺) were rare or absent (Figures 1B and S1D). The mESC-derived cultures therefore contained essentially only astrocytes and glial precursors. However, they were not homogeneous since many cells were not double-stained for any given pair of antigens (Figure 1A) and less than one-third of the cells positive for vimentin and NF1A expressed the markers Aqp4, S100β or GFAP (Figure 1B). This is comparable to the heterogeneity in cultures of neonatal astrocytes (Imura et al., 2006). Our data suggested that the cultures contained developing astrocytes at different stages of maturity.

Figure 1. Mouse ESC-derived spinal cord astrocytes show an immature phenotype.

A. Differentiated mESC cultures contain glial cells expressing canonical astrocyte markers together with the spinal cord marker HOXB4. Insets show higher magnification of regions indicated with dotted lines. Scale bars: 75 μm.

B. Percentages of mESC-derived glial cells expressing markers for astroglia (Aqp4, S100β, vimentin, NF1A and GFAP), oligodendroglia (Olig2 and CNPase), neurons (β-III-tubulin), microglia (Iba1) and spinal cord identity (HOXB4). Mean ± s.e.m.; n=3–4 independent differentiations.

C. Western blot of mESC-derived astrocytes as compared to astrocytes derived from E12.5 mouse embryo spinal cord. GFAP protein is present in both, but ALDH1L1, GLAST and GLT1 are only detectable in the mE12.5-derived astrocytes. α-tubulin was used as loading control.

See also Figure S2.

To assess their state of maturation, we used Western blots to quantify levels of three markers known to be upregulated in astrocytes as they mature: GLAST and ALDH1L1, which appear at mid-embryogenesis, and GLT1, a postnatal marker (Figure 1C; see Supplementary Text). Although 4 weeks had elapsed since the mESCs were put into culture, and although GFAP was clearly present, none of these markers could be detected. This was in contrast to astrocytes generated from primary neural precursors of embryonic day 12.5 (E12.5) spinal cords using a neurosphere-based approach (Figure S2A), which expressed robust levels of GLAST, ALDH1L1 and GLT1 (Figure 1C). We therefore sought treatments through which mESC-derived spinal astrocytes could be brought to a similar degree of maturity.

FGF1 promotes maturation of mESC-derived astrocytes

In vivo, FGFs have been proposed to act as differentiation signals for astrocytes (Irmady et al., 2011; Morrow et al., 2001) and FGF1 is strongly expressed in postnatal spinal neurons (Elde et al., 1991). Addition of FGF1 to mESC-derived astrocytes from days 28–35 triggered a >4-fold increase in the number of cells expressing GLT1, with an EC₅₀ of 5 ng/ml (Figure 2A, B). Similar results were obtained using FGF2 in place of FGF1 (Figure S3). Western blotting confirmed that FGF increased overall levels of both GLT1 and GLAST (Figure 2C, D and Figure S3C) as well as other maturation markers such as aldolase C, Cx43 and ALDH1L1 (Figure 2D and Figure S3C). In parallel, GFAP levels were strongly downregulated (Figure 2B, D and Figure S3A, C). Thus, mESC-derived cultures treated with FGFs express high levels of glutamate transporters and low GFAP, a phenotype reminiscent of mature quiescent astrocytes.

Figure 2. FGF1 promotes maturation of mESC-derived astrocytes.

A. Effects of increasing dose of FGF1 on the relative abundance of GLT1-expressing astrocytes in mESC-derived cultures (mean ± s.e.m. for 2 independent experiments, 3 replicates per condition). One-way ANOVA reveals an effect of treatment above 5 ng/ml. P=0.013, F_(8;9)=5.01. *P < 0.05.

B. Compared to an FBS control (left), 50 ng/ml FGF1 (right) triggers a strong increase in GLT1 staining and a nearly complete loss of GFAP immunoreactivity. Scale bar: 75 μm.

C. Increased GLT1 expression following FGF1 treatment of astrocytes derived from two independent mESC lines (Hb9::GFP and wildtype) revealed by Western blot analysis. Results are representative of 3 independent experiments.

D. FGF1 is sufficient to induce appearance of GLAST, CX43 and ALDH1L1 but strongly decreases GFAP expression. Results are representative of 3 independent experiments.

E. Na⁺-dependent L-(³H)-glutamate transport using two mESC lines differentiated into astrocytes shows an average 2-fold increase in uptake following treatment with FGF1 (bars show mean ± s.e.m.; n=4; P=0.0095, F_(3;3)=1.286. **P < 0.01).

See also Figure S3.

To determine whether this biochemical maturation resulted in functional changes, we measured glutamate uptake, a key role of mature astrocytes (Huang and Bergles, 2004). In mESC astrocytes from two different cell lines grown in FBS alone we detected significant levels of glutamate uptake that were Na⁺-dependent as expected (Figure 2E; Mitani and Tanaka, 2003; Rothstein et al., 1996). Exposure to FGF1 for 7 days led to a 2-fold increase in glutamate transport (Figure 2E), reflecting upregulation of the transporters responsible (Figures 2B–2D). Thus, functional astrocytes can be efficiently generated from mESCs and FGF1 promotes their biochemical and functional maturation to a quiescent state.

Human spinal cord astrocytes derived from hES and hiPS cells following early neuralization

We then used the findings with mouse ESCs to generate mature human astrocytes. To accelerate astrocyte production, rather than passing through a stage of cycling neural precursors (Shaltouki et al., 2013), we turned to early neuralization by dual inhibition of SMAD signaling, which had been shown by Chambers et al. (2009) to enhance production of CNS precursors and neurons but whose effect on glial generation had not been studied. To inhibit SMAD signaling we employed the ALK4/5/7 inhibitor SB431542 together with LDN193189, a potent derivative of dorsomorphin that inhibits TGFβ1/activin receptor-like kinase (Boulting et al., 2011; Kriks et al., 2011; Yu et al., 2008).

We used one hESC line (R1; James et al., 2006) and one hiPSC line (18c; Bock et al., 2011; Boulting et al., 2011) derived from a healthy volunteer, which we have shown to have a strong propensity for neural differentiation. When SB431542 (10 μM) and LDN193189 (0.2 μM) were added to cultures from 1 to 5 DIV (Figure 3A), 70–80% of cells in the culture at 10 DIV were PAX6⁺ Oct4^– neural progenitors (not shown). To direct the cells toward a caudal ventral identity, RA and SHH-C were added as indicated and neurotrophic factors were provided to support neuronal survival (Figure 3A). By 40 DIV, as with earlier protocols (Hu et al., 2010; Karumbayaram et al., 2009; Li et al., 2005; Shaltouki et al., 2013), the cultures were enriched for markers of ventral neuronal populations (OLIG2, HB9, ISL1/2, NKX2.2 and NKX6.1; not shown). Critically for our studies, they showed spinal cord (HOXB4; >90%) but not midbrain (OTX2; <2%) identity (Figure 3B). Early neuralization did not therefore perturb the patterning of these neural precursors.

Figure 3. Rapid generation of spinal cord astrocytes from hESCs and hiPSCs following early neuralization.

A. Treatments used for efficient generation of hESC- and hiPSC-derived astrocytes. P: passage; LDN: LDN193189; SB: SB431542; RA: retinoic acid; ASAC: ascorbic acid; N2: N2 supplement; SHH-C: sonic hedgehog cysteine mutated; B27: B27 supplement, NTFs: neurotrophic factors (CNTF, BDNF, IGF-1, GDNF).

B. Efficient conversion of human pluripotent stem cells - here shown for hESC (R1) line - into HOXB4⁺ OTX2⁻ spinal cord cultures. Representative of 3 independent experiments. Scale bar: 50 μm.

C. At 80 DIV, no neurofilament (NF-H)-expressing neurons are found in hESC- and hiPSC-derived cultures (here shown for line 18c). At 40 DIV, the fraction of NF-H-expressing cells was ~40% (not shown). Scale bar: 75 μm.

D. At 80 DIV, very few CNPase-expressing oligodendrocytes are detected. Mean abundance ± s.e.m. (n=2–3 independent differentiations). Scale bar: 50 μm.

E. At 45 DIV, GFAP-expressing cells display bipolar morphology and resemble radial glia.

F. At 90 DIV, cultures are enriched in GFAP-expressing astrocytes, which retain their spinal cord identity as revealed by HOXB4 expression (mean ± s.e.m.; n=3–4).

G. At 80 DIV, hESC- and hiPSC-derived cultures (here shown for R1) contain abundant astrocytes that express the canonical marker CD44 (mean ± s.e.m.; n=2). Scale bar: 50 μm.

H. The number of cells expressing the canonical astrocyte marker S100β increases over time to nearly 100% (mean ± s.e.m.; n=2–3). Scale bar: 50 μm.

I. At 80 DIV, hESC- and hiPSC-derived cultures (R1 and 18c) contain astrocytes that express canonical markers together with the EAAT2 transporter. Mean abundance ± s.e.m. (n=3–5 independent differentiations). Scale bar: 75 μm.

J. Western blotting confirms the expression of the astroglial markers in all lines tested. Human adult spinal cord (HASC) tissue was used as positive control. Blot representative of 2 independent experiments.

We monitored the appearance of different cell classes over the course of culture in FBS (1%) of these and two further hESC (H13) and hiPSC (11a) lines. Neurons were initially abundant at early stages but disappeared over the period 30–60 DIV until by 90 DIV there were no neurons remaining (Figure 3C). Moreover, few CNPase-positive oligodendrocytes could be detected at either 40 or 90 DIV (<3% of total cells; Figure 3D) and whereas many cells at early stages expressed the neural crest marker p75 these too were spontaneously eliminated (not shown). GFAP staining was first observed from 20–45 DIV, in the form of scattered GFAP⁺ and A2B5⁺ cells (Figure 3E and not shown) whose long processes suggested they might be radial glia (Hirano and Goldman, 1988). Over the period 40–80 DIV, the abundance of multipolar GFAP-expressing cells with the HoxB4⁺/Otx2⁻ profile characteristic of spinal cord increased ~10-fold to >70% of DAPI-labeled cells (Figure 3F and not shown) and by 80 DIV astrocyte markers CD44, S100β, CX43, vimentin, NF1A and aldolase C were all present in cultures from each of the four hESC/hiPSC lines (Figure 3G, H, I, J). Nearly 100% of the cells expressed the glial marker S100β (Figure 3F) consistent with their homogeneous morphology (Figure S4).

We next analyzed the functional capacity of astrocytes derived following early neuralization. EAAT2 (equivalent to rodent GLT1) was already detectable in a subset of cells (Figure 3I, J) and, accordingly, all human astrocyte cultures exhibited a basal level of Na⁺-dependent glutamate transport (Figure 4A). Mechanical stimulation of single astrocytes from all four lines generated calcium waves that propagated to adjacent astrocytes (Figure 4B and data not shown). Lastly, the astrocytes expressed both brain-derived neurotrophic factor (BDNF) and glial cell line-derived neurotrophic factor (GDNF), and secreted factors that significantly enhanced survival and neurite outgrowth from FACS-purified human ESC-derived motor neurons (Figure 4C, D, E). This correlates well with the growth-promoting activity of immature astrocytes reported in vivo (Filous et al., 2010; Goldshmit et al., 2012). Thus early neuralization generates human spinal cord astrocytes expressing many canonical biochemical and functional traits.

Figure 4. Human stem cell-derived astrocytes are functional but not completely mature.

A. All hESC- and hiPSC-derived astrocytes show Na⁺-dependent L-(³H)-glutamate transport activity (s.e.m. shown for n=3 independent experiments were performed).

B. Mechanical stimulation (white arrow) of a single astrocyte induces a wave of calcium influx that propagates to adjacent astrocytes. Scale bar: 75 μm.

C. Whole-well imaging of hESC-derived motor neurons live-stained with calcein following co-culture with astrocytes in the presence or not of exogenous growth factors (NTF). Green images show selected fields. Scale bar: 200 μm (top) and 75 μm (bottom).

D. Human astrocytes enhance survival and growth of hESC-derived motor neurons following co-culture. One-way ANOVA reveals an effect of treatment. For motor neuron survival, P and F values are: no NTF vs. astrocyte coculture P=0.0224, F_(4;8)=5.25; NTF vs. astrocyte coculture supplemented with NTF P=0.0239, F_(4;9)=4.79. For neurite outgrowth, P and F values are: no NTF vs. astrocyte co-culture P=0.0221, F_(4;8)=5.38; NTF vs. astrocyte supplemented with NTF co-culture P=0.0045, F_(4;9)=8.19. * P < 0.01 and ** P < 0.001 denote statistically significant differences.

E. qPCR analysis shows that hESC- and hiPSC-derived astrocytes express BDNF and GDNF at levels comparable to, or greater than, human adult spinal cord tissue (HASC). Error bars indicate s.e.m. for n=3 independent experiments (all in duplicate).

F. hESC- and hiPSC-derived astrocytes do not form teratomas in the striatum of adult rats even at 7 weeks post-transplantation, revealed by Nissl staining. Scale bars: 250 μm and 50 μm.

G. Even at 7 weeks post-transplantation, human ESC- and hiPSC-derived astrocytes maintain an immature/reactive phenotype. Scale bar: 50 μm.

See also Figure S4 and S5.

Nevertheless, as with mESC cultures, maturation was less than complete and varied between lines. For example, the glial progenitor marker NF1A was strongly expressed in cultures from all four lines (Figure 3J). Moreover, even among mature markers, the populations derived from different stem cell lines were not homogeneous. For example, EAAT2 expression was consistently lower in the H13 hESC cultures, as shown by immunocytochemistry and Western blotting (Figure 3I and J). We reasoned that this might reflect the absence of humoral maturation factors that astrocytes normally encounter in vivo. We therefore grafted 80- to 90-DIV hESC- and hiPSC-derived astrocytes into the striatum of adult rats (Figure 4F; Brederlau et al., 2006). After 2 and 7 weeks, staining for human nuclei in 4 animals per line revealed that grafted cells in all animals had survived (Figure S5). Survival in vivo is a characteristic of immature astrocytes (Filous et al., 2010) and, accordingly, at 7 weeks 86 ± 1% (mean ± s.e.m., n=16; no significant differences between lines or time points) of grafted cells were GFAP⁺ and 88 ± 6% expressed NF1A (Figure 4G). The presence of these immature markers demonstrated that the in vivo environment alone was not sufficient to induce quiescence and may have indeed stimulated astrocyte reactivity.

FGF1, but not TNFα, promotes maturation of human stem cell-derived astrocytes

We therefore tested candidate factors for their ability to promote maturation, focusing on FGF1 because of the potent effects we had found in the murine system. As a control, we used tumor necrosis factor-alpha (TNFα) since it has been implicated in triggering astrocytosis in neurodegenerative contexts (Cui et al., 2011) and was previously shown to activate primary human astrocytes in vitro (Croitoru-Lamoury et al., 2003; Meeuwsen et al., 2003). Human ESC and hiPSC lines were differentiated as above for 90 days and then thoroughly washed to remove any residual FBS. They were subsequently treated for 7 days with FBS (1% v/v), FGF1 (50 ng/ml) or TNFα (50 ng/ml), the latter two factors in the absence of serum. Although morphological changes were observed (Figure S4), levels of the canonical markers S100β and aldolase C were not affected by these treatments (Figure 5A and B), suggesting that the cells retained their astrocyte identity throughout. In striking contrast, Western blotting revealed that FGF1 induced a nearly complete loss of expression of both GFAP and NF1A in both cell lines, whereas cells treated with TNFα showed little change in either marker (Figure 5B). Correspondingly, immunostaining for NF1A and GFAP was markedly reduced in the cell aggregates induced by FGF1 (Figure 5C and S6). To exclude the possibility that this might be a secondary effect of cell aggregation, we performed the same experiments after dissociation of cell clumps followed by replating. Similar reductions were observed (Figure S6A and S6B), confirming that these are direct effects of FGF. Another marker whose levels in human astrocytes were significantly reduced by FGF1 as compared to FBS was CX43 (Figure 5D). This is consistent with a more quiescent phenotype following FGF treatment (Theodoric et al., 2012). Although mESC-derived astrocytes treated with FGF showed the opposite change (Figure 2D), Cx43 levels in rodents rise during perinatal development (Dermietzel et al., 1989) and so this change may reflect developmental maturation. Overall, therefore, human astrocytes are induced to adopt a mature quiescent phenotype by FGF1 but not by TNFα.

Figure 5. FGF, but not TNFα, triggers biochemical and functional maturation of human stem cell-derived astrocytes.

A. Astroglial purity at 90 DIV: virtually all cells express S100β in all conditions. Scale bar: 50 μm.

B. Dramatic reduction in GFAP and NF1A protein levels, but not aldolase C, following treatment with FGF1 at 80 DIV, whereas TNFα has little impact. α-tubulin was used as loading control. Results are typical of 3 independent experiments.

C. Reduced abundance of NF1A-expressing cells in 80-day hESC- and hiPSC-derived astrocytes following treatment for 7 days with FGF1 (50 ng/ml). FGF1 leads to cell aggregation. Scale bar: 50 μm.

D. Decrease in CX43 expression following treatment with FGF1 as compared to TNFα. Results are typical of 2 independent experiments. Scale bar: 75 μm

E. Two-fold increase in glutamate uptake in astrocytes treated with FGF1 but not TNFα. For line 18c, one-way ANOVA (P<0.0001; F_(2;12)=48.58, n=5) followed by Newman-Keuls multiple comparison post-hoc test shows FGF1 treatment significantly (***=P<0.0001) promotes uptake over control- (FBS) and TNFα-treated cultures. TNFα treatment decreases uptake (*=P<0.05). Similar results were obtained for line R1.

F. Absence of pronounced change in EAAT2 protein levels following FGF1 or TNFα treatment. Results are typical of 4 independent experiments.

G. qPCR analysis shows no major change in EAAT2 levels when astrocytes are cultured with FGF1 and TNFα.

H. qPCR analysis shows that hESC- and hiPSC-derived astrocytes strongly upregulate the alternative glutamate transporter EAAT1 seven days after treatment with FGF1 but not TNFα (mean ± s.e.m.; n=4 for R1 and n=3 for 18c). One-way ANOVA reveals an effect of treatment. P and F values are: R1 P=0.0028, F_(2;9)=12.18; 18c P=0.0019, F_(2;6)=21.94. **P < 0.001 denotes statistically significant difference between treatments. Bottom panel: Similar changes are observed after replating of dissociated clumps (n=1 for R1 and 18c).

See also Figure S5.

To confirm this conclusion we assayed functional changes triggered by each factor. Under pathophysiological conditions, reactive inflammatory astrocytes express low levels of glutamate transporters (Rothstein et al., 1995); this was potentially the case with our FBS cultures, which are rich in GFAP⁺ astrocytes. In agreement with this, FGF1 led to a 2-fold increase in Na⁺-dependent glutamate transport activity. Similar effects were seen with FGF2 (Figure S7) whereas TNFα had no or even a negative effect (Figure 5E), as reported for primary astrocytes (Fine et al., 1996). Unexpectedly, given the increased transport, FGF1 induced no marked changes in EAAT2 protein or mRNA levels (Figure 5F and G). As an alternative explanation for the increased glutamate transport we investigated EAAT1, a related transporter whose levels in gray-matter astrocytes in vivo also increase with postnatal age (Voutsinos-Porche et al., 2003a; Voutsinos-Porche et al., 2003b). We indeed observed a 4-fold increase in EAAT1 levels following treatment with FGF1, whereas TNFα had no effect (Figure 5H). Moreover, DHK (an inhibitor of EAAT2 transport) had no effect on glutamate uptake (Figure S7), further supporting the involvement of EAAT1. Moreover, similar effects of FGF were seen in aggregated or dissociated cultures, suggesting that the increased glutamate transport is not an indirect effect of cell clumping (Figure 5H). Overall, our results show that FGF1 or FGF2, but not TNFα, are sufficient to induce a mature biochemical and functional phenotype in human, as in mouse, stem cell-derived astrocytes.

TNFα and IL-1β, but not FGF1, trigger reactivity of human pluripotent stem cell-derived astrocytes

Since TNFα had proven inactive in driving astrocyte maturation we asked whether it was able to induce astrocyte activation. We monitored expression of the chemokines CXCL8 and RANTES (also known as IL8 and CCL5, respectively), which are strongly upregulated in human astrocyte primary cultures treated with TNFα (Croitoru-Lamoury et al., 2003; Meeuwsen et al., 2003), as well as expression of lipocalin-2 (LCN2) and tissue inhibitor of metalloproteinase-1 (TIMP1), which were recently shown to be upregulated in reactive astrocytes in vivo (Zamanian et al., 2012). Within 2 days of TNFα administration, levels of mRNA for CXCL8 and RANTES were increased >300-fold over those for cells maintained in FBS (Figure 6A), while 7 days after TNFα treatment LCN2 levels were also increased (>1,000 fold in R1 and 8-fold in 18c astrocytes; Figure 6B). Another proinflammatory cytokine, IL-1β, led to a 4,000-fold increase in LCN2 (Figure S8F). In contrast, FGF1 induced a modest down-regulation in both chemokines and LCN2 (Figure 6A and B). TIMP1 mRNA levels did not change more than 8-fold in any condition (not shown), suggesting that it is less strongly regulated than LCN2 in cultured astrocytes. Thus TNFα induces a molecular phenotype that is consistent with astrocyte reactivity.

Figure 6. TNFα, but not FGF1, triggers reactivity of hESC- and hiPSC-derived astrocytes.

A. Strong effects of TNFα but not FGF1 on Rantes and CXCL8 expression (mean ± s.e.m.; n=4 for R1 and n=3 for 18c). One-way ANOVA reveals an effect of treatment. P and F values are: Rantes R1 P = 0.0002, F_(2;6)=46.99; Rantes 18c P < 0.0001, F_(2;6)=111.7; CXCL8 R1 P = 0.0006, F_(2;6)=32.73; CXCL8 18c P < 0.006, F_(2;6)=13.03. ***P <0.0001, **P < 0.001, *P < 0.05 denote statistically significant differences between treatments.

B. hESC- and hiPSC-derived astrocytes strongly express lipocalin-2 (LCN2) after 7 days treatment with TNFα but not with FGF1 (mean ± s.e.m.; n=3). One-way ANOVA reveals an effect of treatment. P and F values are: R1 P = 0.0076, F_(2;6)=12.26; 18c P < 0.02, F_(2;6)=8.06. **P < 0.001 and *P < 0.05 denote statistically significant differences between treatments.

C. TNFα but not FGF1 triggers IL6 release over time. Two-way ANOVA reveals an effect of treatment for R1 line and treatment and time for 18c line. P and F values are: IL-6 R1 (treatment) P = 0.0035, F_(2;18)=7.88; IL-6 18c (treatment) P =0.0003, F_(2;18)=111.7; IL-6 18c (time) P = 0.01, F_(2;18)=5.86. ***P < 0.0001, **P < 0.001 denote statistically significant differences between treatments on day 7.

D. hESC- and hiPSC-derived astrocytes have an immature reactive phenotype when cultured in FBS but can be triggered by single factors to adopt a mature, quiescent phenotype (FGF1 or FGF2) or a reactive inflammatory one (TNFα or IL-1β).

See also Figure S6.

To test this more directly, we measured the production of the cytokine interleukin-6 (IL-6), which in primary astrocytes acts in an autocrine manner to stimulate its own production (Van Wagoner et al., 1999). No IL-6 could be detected in FBS control cultures, or in cells treated with FGF1 (Figure 6C). In marked contrast, TNFα led to an increase in IL-6 production that was already detectable after 1 day and increased to >300 pg/ml after 1 week (Figure 6C). Similarly strong responses to TNFα were observed in astrocytes generated from another hESC line (H13) and three other hiPSC lines (11a, 25b, 39d; data not shown) and IL-1β produced similar effects (Figure S8). Therefore both TNFα and IL-1β induce a reactive phenotype.

Overall, our data show that single factors can rapidly modify the phenotype of pluripotent stem cell-derived astrocytes in distinct and biologically relevant manners (Figure 6D and S8H).

DISCUSSION

The role of astrocytes in vivo varies widely according to their degree of developmental maturation or reactivity, and potentially their regional specificity, yet the distinction between these functional states has not been clearly established in in vitro models. Having generated human spinal cord astrocytes from hESCs or hiPSCs through early neuralization, we focused on biochemical and functional characteristics that distinguish quiescence from reactivity in the animal and used them to define methods through which each state can be selectively enhanced in culture. Strikingly, we found that single factors were sufficient to promote such transitions: FGF1 or FGF2 are sufficient to strongly enhance maturation/quiescence without affecting reactivity, whereas TNFα or IL-1β produces the converse effect. Our data show clearly how exogenous factors can modify astrocyte functional status and they provide a means of generating human patient-derived astrocytes that are functionally homogeneous for future studies of development or of disease.

One of the challenges in using human pluripotent stem cells for either basic or translational biology is the length and consequent variability of the protocols needed to generate specific cell types (Krencik and Zhang, 2011). One approach to the acceleration of protocols for astrocyte generation used in several recent publications is based on the initial generation of neural precursors, in some cases involving manual picking of rosettes (Emdad et al., 2012; Juopperi et al., 2012; Lafaille et al., 2012; Serio et al., 2013; Shaltouki et al., 2013). Given our interest in motor neuron biology and disease, we instead began by early neuralization of monolayer cultures followed by caudalization and ventralization, which we have shown to generate spinal neurons with high yield in shorter times (Amoroso et al., 2013; Boulting et al., 2011; Chambers et al., 2009). Our method allowed spinal cord astrocytes with the expected molecular and functional characteristics to be generated in half the time (80–90 days as compared to 180 days) needed for neurosphere-based approaches (Gupta and Kanungo, 2011; Krencik et al., 2011). However, as with all other protocols published to date the astrocytes were immature as judged by both marker expression and glutamate uptake.

We found that both FGF1 and FGF2 are capable of inducing a mature, quiescent astrocyte phenotype characterized by low GFAP and NF1A but high levels of glutamate transport. Our findings are distinct from other reported effects of FGFs in vitro but closely match developmental in vivo data. In vitro, FGFs have been reported to induce mitosis (Lin and Goldman, 2009) and gap junction coupling (Garre et al., 2010) but not maturation. Particularly striking is the diversity of responses reported to FGFs. Some studies report induction of stellate morphology (Cassina et al., 2005; Reilly et al., 1998) and even increased GFAP immunostaining (Cassina et al., 2005), whereas others observe de-differentiation and adoption of a bipolar morphology (Goldshmit et al., 2012). It is likely that the different outcomes reflect the presence of different astrocyte subtypes or culture conditions. In vivo, FGFR3, which can be activated by both FGF1 and FGF2 (Chen and Hristova, 2011; Ornitz et al., 1996), is expressed specifically by astrocytes and their precursors in postnatal spinal cord (Pringle et al., 2003). Strikingly, in null-mutant mice for FGFR3 there is a strong upregulation of GFAP in spinal gray-matter from 2 months of age onward, and signaling through FGFR3 can repress GFAP (Pringle et al., 2003). In agreement with our findings, this suggests that FGFs drive and maintain a quiescent mature phenotype in vivo. In FGF2 null mutants, in apparent contrast, GFAP is reduced in spinal gray matter but this was suggested by the authors to reflect an early developmental role for FGF2 (Irmady et al., 2011). Therefore, treatment of immature pluripotent stem cell-derived astrocytes with FGF1/2 is an effective and physiologically relevant means for inducing in vitro maturation.

Astroglial activation is associated with several nervous system pathologies (Glass et al., 2010). Focusing on motor neuron disease, specific astrocyte reactivity is observed in post-mortem cortex and spinal cord of ALS patients (Kamo et al., 1987; Murayama et al., 1991; Schiffer et al., 1996) and inflammatory gene expression is upregulated in astrocytes derived from sporadic ALS patients (Haidet-Phillips et al., 2011). GFAP and ALDH1L1 are upregulated following injury and in the ALS spinal cord (Anthony and Heintz, 2007; Yang et al., 2011) and a dramatic loss of GLT1 is observed in other pathological situations (Chao et al., 2010; Faideau et al., 2010; Rothstein et al., 1995). For reliable disease modeling it is therefore essential to reproduce the reactive state. We found that TNFα induces in human stem cell-derived astrocytes a phenotype characterized by high GFAP and production of inflammatory chemokines and cytokines, as well as very strong upregulation of Lcn2, the top gene in the list of markers shown by Zamanian et al. (2012) to be upregulated in two different models of reactivity. This is of particular interest since Lcn2 was recently reported to trigger neuronal death and to contribute to astrocyte toxicity (Bi et al., 2013). In parallel experiments, we extensively tested the effects of other potential activators. For example, IL-1β elicited several characteristic signatures of activation but also promoted GLAST expression, a marker of maturation (not shown). Opposite effects of TNFα and IL-1β on astrocytes have previously been reported (Sokolova et al., 2012). We therefore recommend for standard differentiations the use of the more selective activator TNFα.

The two populations of astrocytes we have generated will have multiple uses. We report for the first time that human stem cell-derived astrocytes are capable of an immunological response similar to that evoked by cytokines in primary human astrocytes (Meeuwsen et al., 2003) and this should be important for modeling human pathological processes in vitro. Mature, quiescent astrocytes will clearly be important for studying normal function but may also be relevant to disease conditions. In vitro, astrocytes from mutant SOD1 mouse models of ALS release toxic factors for motor neurons (Di Giorgio et al., 2007; Nagai et al., 2007) while in vivo, reduction of SOD1 levels in GFAP⁺ (and therefore reactive) cells leads to an increase in survival of the mice (Yamanaka et al., 2008). This might suggest that the reactive, GFAP⁺ astrocytes are the principal contributors to toxicity for motor neurons. However, the rescue in vivo was only partial, raising the possibility that if SOD1 had also been removed from quiescent astrocytes at earlier stages, the benefit would have been greater. Moreover, astrocytes lacking SOD1 retained GFAP expression although they were less toxic (Yamanaka et al., 2008). Therefore, astrocyte toxicity and reactivity can be uncoupled. Thus, in assaying the contribution of astrocytes to pathology in vivo, it is critical to evaluate even mature, quiescent populations. Moreover, for modeling neurodegenerative disease in vitro, it will be essential to distinguish between the effects of mature and activated astrocytes, and our protocol provides a means of doing this using both rodent and human cells.

EXPERIMENTAL PROCEDURES

Differentiation of spinal cord astrocytes from mouse and human pluripotent stem cells

Rodent motor neuron cultures were dissociated into single cell suspension and grown in NS medium supplemented with FGF2 and EGF. Neurospheres were dissociated and single cells were seeded onto 100 μg/mL poly-L-ornithine and 15 μg/ml laminin-coated surfaces, in NS medium supplemented with 10% FBS. Neuralized human motor neuron cultures (aged 31 DIV) were passaged onto 100 μg/mL poly-ornithine and 3–5 μg/ml laminin-coated surfaces in NS medium supplemented with 1% FBS. Rodent and human cultures were passaged when they reached confluency. The presence of astrocytes was detected by immunocytochemistry.

RNA isolation and quantitative PCR

The total RNA was isolated from cultured cells with RNeasy Kit (Qiagen). cDNA was generated using the Verso TM cDNA Kit and was used as template for the quantitative PCR (qPCR).

Measurement of glutamate uptake

Cells were incubated for 5 minutes at 37°C in Na⁺ buffer containing 0.5 μM L-glutamate and 0.3 μCi L-[³H]-glutamate per sample. Cells were then lysed with 0.1N NaOH solution and radioactivity was measured using a scintillation counter. Radioactive counts were normalized to total protein per culture well.

IL-6 ELISA

Astrocytes were treated for 1, 2 or 7 days with medium containing FBS, FGF1, IL1 and TNFα. Medium was then harvested and used for the human IL-6 ELISA assay following manufacturer's instructions (Invitrogen). Absorbance was measured at 450 nm.

Statistical analyses

For each independent experiment, duplicate sometimes triplicate cultures were analyzed from twelve 10× fields per sample. All quantitative data were analyzed using Prism 5.0c (GraphPad). Sample groups were subjected to one- or two-way ANOVA, with Newman-Keuls post hoc comparison when all conditions are compared to each other, and Dunnett post hoc comparison when all columns are compared to control condition (e.g FBS treatment). The null hypothesis was rejected at 0.05 for all ANOVAs and post hoc tests.

OTHER EXPERIMENTAL PROCEDURES

Full descriptions are provided in Supplemental information.

HIGHLIGHTS.

• Spinal cord astrocytes can be generated efficiently from human ES/iPSCs using early neuralization

• hESC/hiPSC-derived astrocytes are immature but FGF1 and FGF2 lead to rapid maturation

• TNFα and IL1β selectively and rapidly induce a reactive phenotype in human hESC/hiPSC-derived astrocytes

• Novel insights into human astrocyte biology and new methods for disease modeling