Cerebellar granule cell precursors can differentiate into astroglial cells

Abstract

During CNS development, multipotent neural stem cells give rise first to various kinds of specified precursor cells, which proliferate extensively before terminally differentiating into either neurons or glial cells. It is still not clear, however, whether the specified precursor cells are irreversibly determined to differentiate into their particular cell types. In this study, we show that isolated mouse cerebellar granule cell precursors from the outermost, proliferative zone of the external germinal layer can differentiate into astroglial cells when exposed to sonic hedgehog (Shh) and bone morphogenetic proteins. These induced cells initially expressed both glial fibrillary acidic protein and neuronal markers, but they then lost their neuronal markers and acquired S100-β, a marker of differentiated astroglial cells. These results indicate that at least some granule cell precursors are not irreversibly committed to neuronal development but can be induced to differentiate into astroglial cells by appropriate extracellular signals.

Although neural stem cells (NSCs) are studied intensely by researchers interested in either regenerative medicine or developmental neurobiology, the detailed pathways by which NSCs give rise to neurons, astrocytes, or oligodendrocytes are still uncertain (1). It is generally accepted, for example, that NSCs first give rise to neuronal precursors, which proliferate and then terminally differentiate into postmitotic neurons. It is still not clear, however, whether such neuronal precursors are irreversibly committed to become neurons or whether they have the ability to differentiate into glial cells or even to revert to NSCs.

In this study, we have examined whether granule cell precursors (GCPs) are irreversibly determined to differentiate into granule cells, the most abundant class of CNS neurons. GCPs arise from the rhombic lip, the dorsal part of the neural tube at the boundary of the mesencephalon and the metencephalon (2, 3). They migrate rostrally up the lip and onto the surface of cerebellar anlage, where they form the external germinal layer (EGL) (3, 4). In rodents, GCPs proliferate rapidly in the EGL for 2–3 wk after birth. They then exit the cell cycle, extend axons, and migrate inward to their final destination in the granule layer (GL) (4, 5). We show that some GCPs in the EGL are not irreversibly determined to differentiate into granule cells; when treated with sonic hedgehog (Shh) and bone morphogenetic proteins (BMPs) in culture, a proportion of the GCPs lose their neuronal markers and differentiate into astroglial cells (6).

Materials and Methods

Animals and Materials. C57BL/6 mice were purchased from SLC (Hamamatsu, Japan). The recombinant N-terminal-active fragment of mouse Shh (Shh-N) and recombinant human BMP2 were purchased from Genzyme. Shh-N was expressed also in Escherichia coli transformed with GST-Shh-N plasmid (provided by P. A. Beachy, The Johns Hopkins University, Baltimore). The DNA synthesis inhibitors aphidicolin and 1-β-d-arabinofuranosylcytosine (Ara-C) were purchased from Wako Pure Chemical (Osaka, Japan) and Sigma, respectively.

Preparation of Immature GCPs by Immunopanning. Immature GCPs were prepared by immunopanning methods using the anti-human natural killer 1 (HNK-1) antibody. Cerebella from postnatal day 7 (P7) mice were cut into small pieces and incubated at 37°C for 30 min in papain solution (16.5 units/ml papain/200 μg/ml l-cysteine/0.008% DNase). Tissue was rinsed in Dulbecco's PBS containing 1.5 mg/ml ovomucoid, 1.5 mg/ml BSA, and 0.008% DNase and triturated in the same solution containing rabbit anti-mouse macrophage antibodies to obtain a single-cell suspension. Cells were centrifuged at 1,000 rpm for 10 min at room temperature and suspended in Dulbecco's PBS containing 10 mg/ml ovomucoid and 10 mg/ml BSA and centrifuged again. Cells were resuspended in panning buffer (Dulbecco's PBS containing 0.02% BSA and 5 μg/ml insulin) and passed through a cell strainer (Falcon). To obtain a fraction enriched in GCPs (7), the cell suspension was loaded onto a step gradient of 35% and 60% Percoll (Amersham Biosciences) and centrifuged at 3,000 rpm for 20 min at room temperature. GCPs were recovered from the 35%/60% interface and washed twice in panning buffer. To remove contaminating microglial cells, the cell suspension was plated onto a 100-mm tissue culture dish precoated with affinity-purified goat anti-rabbit IgG antibodies (Jackson ImmunoResearch) and incubated for 20 min at room temperature. The dish was shaken vigorously, and then nonadherent cells were plated onto a Petri dish precoated with affinity-purified goat anti-mouse IgM antibodies (Jackson ImmunoResearch) and supernatant from HNK-1 hybridoma (American Type Culture Collection) for 30 min. The dish was rinsed with Dulbecco's PBS to remove nonadherent cells completely, and strongly adherent cells (i.e., HNK-1-positive immature GCPs) were harvested by trypsinization (0.125% trypsin solution; Sigma). Immature GCPs were suspended in neurobasal medium (GIBCO/BRL) containing 100 units/ml penicillin, 100 μg/ml streptomycin, 1 mM sodium pyruvate, 2 mM l-glutamine, 2% B-27 (all obtained from GIBCO/BRL), 5 μg/ml insulin, 100 μg/ml apotransferrin, 100 μg/ml BSA, 62 ng/ml progesterone, 16 μg/ml putrescine, 40 ng/ml sodium selenite, and 30 μM N-acetyl cysteine (all obtained from Sigma) and plated on 8-well slide glasses (Matsunami, Osaka) precoated with 10 μg/ml poly-d-lysine at a density of 3 × 10⁴ cells per well (≈6 × 10⁴ cells per cm²).

To identify glial fibrillary acidic protein (GFAP)-positive cells that coexpressed neuronal markers, immature GCPs were plated at a low density (6 × 10³ cells per well) on poly-d-lysine-coated 8-well slide glasses.

Immunostaining. Cells were fixed with 4% paraformaldehyde for 20 min, washed with PBS, and then incubated for 30 min in blocking buffer [Tris-buffered saline (50 mM Tris/150 mM NaCl, pH 7.4) containing 50% goat serum, 1% BSA, 100 mM l-lysine, and 0.04% sodium azide]. For intracellular antigens, blocking buffer containing 0.4% Triton X-100 was used to permeabilize cells. The cells were incubated with primary antibody for 1 h at room temperature. Primary antibodies used for immunostaining included anti-HNK-1 (as described above), anti-TAG-1 (hybridoma obtained from the Developmental Studies Hybridoma Bank, Iowa City), mouse monoclonal antibodies directed against neuronal nuclei (NeuN), microtubule-associated protein 2 (MAP2), and GFAP (all obtained from Chemicon; diluted 1:100, 1:400, and 1:400, respectively), β-tubulin class III (Research Diagnostics, Flanders, NJ; diluted 1:200), S100-β (Sigma; diluted 1:100), nestin (BD Biosciences; diluted 1:100), rabbit polyclonal antibodies directed against MAP2 (Chemicon; diluted 1:200), GFAP (DAKO; diluted 1:200), and brain lipid-binding protein (BLBP) (provided by N. Heintz, The Rockefeller University, New York; diluted 1:1,500). Cells were washed with PBS and subsequently incubated with fluorescein- or rhodamine-conjugated secondary antibodies [for detection of mouse and rabbit IgG (Jackson ImmunoResearch; diluted 1:100) or for detection of mouse IgM (Cappel; diluted 1:100)] for 1 h. Propidium iodide (2 μg/ml) or TO-PRO-3 (2.5 μM; Molecular Probes) was used as a counterstain. Samples were mounted by using Vectashield mounting medium (Vector Laboratories) and visualized by using confocal laser scanning microscopy (MRC-1024, Bio-Rad; or LSM 510 Meta, Zeiss).

For detection of 5-bromodeoxyuridine (BrdUrd) incorporation, cells were pulsed with 10 μM BrdUrd (Roche Diagnostics) for the last 24 h of culture. At the end of culture, cells were fixed with 4% paraformaldehyde for 10 min and incubated with primary and fluorescein- or rhodamine-conjugated secondary antibodies as described above. Cells were then postfixed with 4% paraformaldehyde for 5 min, incubated with 2 M hydrochloric acid for 20 min to denature DNA, and then neutralized with 0.1 M sodium tetraborate (pH 8.5). Cells were incubated with the rat monoclonal anti-BrdUrd antibody (Oxford; diluted 1:200) and then with fluorescein- or rhodamine-conjugated donkey anti-rat IgG antibodies (Jackson ImmunoResearch; diluted 1:100).

At least 500 cells were assessed in each sample, and the fraction of GFAP-positive or BrdUrd-positive cells was determined.

Results

Separation of Proliferating GCPs from Postmitotic GCPs. It has been shown that oligodendrocyte precursor cells can be reprogrammed by extracellular signals to resemble NSCs, which can give rise to both neurons and glial cells (8). To examine whether neuronal precursors have the potential to differentiate into glial cells, we isolated immature GCPs from P7 mice. We used a protocol developed by Hatten and coworkers (7, 9, 10), which exploits the small size of GCPs to separate them from the other cell types in the developing cerebellum. This protocol does not, however, separate proliferating immature GCPs from postmitotic differentiating GCPs. To do this, we used the HNK-1 monoclonal antibody, which has been reported to label cells in the outermost, proliferative zone of the EGL (11).

First, we stained frozen sections of developing mouse cerebellum with the HNK-1 antibody and confirmed that it labeled cells strongly in the outer zone of the EGL and labeled cells only weakly in the deeper, postmitotic zone (Fig. 1A). As shown by Wechsler-Reya and Scott (12), the antibody also strongly labeled most of the proliferating cells (BrdUrd-positive cells) in cultures of P7 cerebellar cells (Fig. 1B).

Fig. 1.

Immature GCPs strongly express HNK-1 both in vivo and in vitro.(A) Frozen sections from P7 mouse cerebellum were stained with the HNK-1 monoclonal antibody, followed by FITC-conjugated goat-anti-mouse IgM (green). The sections were counterstained with propidium iodide (red). (B) GCPs were cultured for 3 days in the presence of 3 μg/ml Shh-N. BrdUrd was added during the last 24 h of culture. Cells were fixed and double stained with antibodies directed against HNK-1 (green) and BrdUrd (red). Cells were counterstained with TO-PRO-3 (blue). (C) HNK-1-positive and unsorted GCPs were cultured for 3 days in the presence (Shh-N) or absence (None) of Shh-N. BrdUrd was added during the last 24 h of culture. Cells were fixed and stained with an anti-BrdUrd antibody. BrdUrd-positive cells were counted, and results are shown as mean ± SD of three areas. (Scale bars: A, 20 μm; B, 10 μm.)

To separate proliferating GCPs from postmitotic GCPs, we used the HNK-1 antibody for sequential immunopanning. We first collected the small cell fraction on a Percoll density gradient. We then removed the microglial cells by negative selection with rabbit IgG and positively selected the HNK-1-positive cells, as described in Materials and Methods. We measured the ability of the cells to incorporate BrdUrd when stimulated by Shh-N, which has been shown to be a potent mitogen for GCPs (12–15). As shown in Fig. 1C, the HNK-1-positive population contained a higher proportion of BrdUrd-positive cells than the unsorted cells.

The HNK-1-positive population also contained a higher proportion of cyclin D1-positive, p27-negative, and NeuN-negative cells than the unsorted cells (data not shown), indicating that it was enriched for immature cells (15). When we cultured the HNK-1-positive GCPs on poly-d-lysine-coated slide glass in serum-free medium in the absence of Shh-N, they stopped dividing and differentiated into granule neurons within 2 days, strongly expressing β-tubulin III (see Fig. 3A). Even when Shh-N was added to the culture medium, the HNK-1-positive GCPs stopped dividing within 1 wk and differentiated into granule neurons.

Fig. 3.

GFAP-positive cells in GCP cultures treated with Shh-N and BMP2 do not express nestin or BLBP. HNK-1-positive GCPs were treated with (B–D) or without (A) Shh-N and BMP2 for 2.5 days. Cells were fixed and double stained for GFAP (red in A, B, and C, green in D) and either β-tubulin III (green in A and B), nestin (green in C), or BLBP (red in D). TO-PRO-3 (blue) was used as a counterstain. A GFAP-positive cell that also expresses nestin (an immature Bergmann glia) is shown in C Inset; a GFAP-positive cell that also expresses BLBP (an astroglial cell or an immature Bergmann glia) is shown in D Inset. (Scale bar: 50 μm.)

Shh and BMP2 Increase GFAP-Positive Cells in GCP Cultures. To determine whether the HNK-1-positive GCPs could differentiate into astrocytes also, we cultured the cells under various conditions and stained them with anti-GFAP antibodies. BMP2, which has been shown to induce oligodendrocyte precursor cells to differentiate into astrocytes (8, 16), failed on its own to induce the HNK-1-positive cells to express GFAP (Fig. 2A). Similarly, Shh-N alone did not promote differentiation into GFAP-positive cells (Fig. 2 A). When we added Shh-N and BMP2 simultaneously, however, the number of GFAP-positive cells in the culture increased substantially (Fig. 2 A and D), although most of the cells still differentiated into β -tubulin III-positive neurons (Fig. 3B). Although we did not examine the survival rates under the various conditions, the number of total cells within the counted areas was not significantly different between the untreated cultures and the Shh-N- and BMP2-treated cultures. Thus, the increase in proportion of GFAP-positive cells resulted from the net increase in the number of GFAP-positive cells (see Fig. 2 legend for details). In the presence of a constant concentration of Shh-N, BMP2 increased GFAP-positive cells in a dose-dependent manner, with an effect at all concentrations >10 ng/ml (data not shown). BMP4 and BMP7 had the same effect as BMP2 (data not shown).

Fig. 2.

Shh-N and BMP2 dramatically increase GFAP-positive cells in GCP cultures. HNK-1-positive GCPs were cultured for 4 days without Shh-N or BMP2 (None), with 3 μg/ml Shh-N (Shh-N), with 80 ng/ml BMP2 (BMP2), or with both Shh-N and BMP2 (Shh-N + BMP2). BrdUrd was added during the last 24 h of culture. Cells were fixed and double stained with antibodies directed against GFAP (red) and BrdUrd (green). TO-PRO-3 (blue) was used as a counterstain. The percentage of GFAP-positive (A) or BrdUrd-positive (B) cells was counted, and results are shown as mean ± SD of three areas. The number of total cells within the counted areas was 752 ± 166 in the untreated culture and 819 ± 202 in the culture treated with Shh-N and BMP2. The number of GFAP-positive cells was 1.7 ± 1.2 in the untreated culture and 28.7 ± 3.1 in the culture treated with Shh-N and BMP2. The results were confirmed in three additional experiments. A single, typical GFAP-positive cell found in BMP2-treated cultures is shown in C; note that it is BrdUrd-negative. GFAP-positive cells induced by Shh-N and BMP2 are shown in D; note that most of these cells are BrdUrd-positive. (Scale bar: 20 μm.)

Our immature GCP preparation contained <1% GFAP-positive cells when assessed at 2 h after plating, and GFAP-positive cells remained <1% for >1 wk under all examined conditions except when they were cultured with Shh-N and BMP2. Most of these GFAP-positive cells had the typical morphological characteristics of astroglial cells (Fig. 2C) and expressed BLBP (Fig. 3D Inset), which is expressed in both astrocytes and Bergmann glia in developing mouse cerebellum (17). Because they also expressed nestin, a marker for NSCs, some of these GFAP-positive cells were probably developing Bergmann glia (18). Nestin has been reported to be expressed in immature Bergmann glia in vivo (19), and we confirmed this result by staining frozen sections from P7 mouse cerebellum with an antibody directed against nestin (data not shown). In contrast to these GFAP-positive cells, the GFAP-positive cells that developed in the presence of Shh-N and BMP2 had a distinctive morphology (mostly bipolar with fine processes) and did not express either nestin or BLBP (Fig. 3 C and D).

Ciliary neurotrophic factor has been shown to induce GFAP-negative astrocyte precursor cells to differentiate into GFAP-positive astrocytes (20). In our cultures, however, ciliary neurotrophic factor did not increase the number of GFAP-positive cells either in the absence or presence of Shh-N (data not shown). In addition, although Shh has been reported to induce the expression of BLBP in immature glial cells (12), we did not see an increase of BLBP-positive cells in our GCP cultures treated with either Shh-N or Shh-N and BMP2.

When we prepared the astrocytes from P7 mouse cerebellum according to the protocol by McCarthy and de Vellis (21) and examined the effect of Shh-N on BrdUrd incorporation by these astrocytes, Shh-N did not stimulate their BrdUrd incorporation either in the presence or absence of BMP2.

Taken together, these results suggest that the GFAP-positive cells induced by the combination of Shh-N and BMP2 did not develop either from contaminating astrocytes or astrocyte precursors or contaminating immature Bergmann glia but instead developed from the immature GCPs themselves.

To determine whether the GFAP-positive cells could have developed from contaminating NSCs, we searched for nestin-positive cells in our cultures. Except for contaminating immature Bergmann glia, which were nestin-, BLBP-, and GFAP-positive from the start of the culture, as described above, all other cells in our cultures in all conditions were nestin-negative, suggesting that contaminating NSCs were not the source of the GFAP-positive cells that developed in cultures treated with Shh-N and BMP2.

BMP2 Acts on Proliferating GCPs to Induce GFAP Expression. Simultaneous treatment with BMP2 and Shh-N significantly decreased the percentage of BrdUrd-positive cells in our purified GCP cultures compared with treatment with Shh-N alone (Fig. 2B; compare Shh-N with Shh-N + BMP2). This finding suggested that BMP2 opposed the mitogenic activity of Shh-N, thereby helping to induce the terminal differentiation of GCPs into GFAP-positive cells. To test this suggestion, we first cultured GCPs with Shh-N alone and then added BMP2 at various times after plating. The later we added BMP2 to the Shh-N-treated cultures, the fewer GFAP-positive cells developed (Fig. 4A). The decrease in the percentage of GFAP-positive cells paralleled a decrease in the percentage of BrdUrd-positive cells (Fig. 4; compare A and B), suggesting that BMP2 induced GFAP expression only in GCPs that were still proliferating. To confirm this suggestion, we used inhibitors of DNA synthesis, Ara-C or aphidicolin. Simultaneous addition of either of these reagents with BMP2 almost completely suppressed the appearance of GFAP-positive cells in GCP cultures that had been treated with Shh-N alone for 2 days (Fig. 4C). These reagents had no effect on cells that had already expressed GFAP (see Fig. 6B, and data not shown). As described above, treatment of fresh GCP cultures with BMP2 alone did not increase GFAP-positive cells significantly (see Fig. 2 A), even though immature GCPs had a high proliferative rate at the start of culture in the absence of any added signaling molecule (data not shown, see ref. 15). Thus, Shh-N signaling apparently is required for BMP2 to induce immature GCPs to differentiate into GFAP-positive cells.

Fig. 4.

BMP2 induces GFAP expression only if the GCPs are proliferating. In A and B, HNK-1-positive GCPs had been treated with Shh-N alone for various times before BMP2 addition. BMP2 was added to the cultures, and the cells were cultured for 2 additional days. BrdUrd was added during the last 24 h of culture. Cells were fixed and double stained for GFAP and BrdUrd and counterstained with TO-PRO-3. GFAP-positive (A) or BrdUrd-positive (B) cells were counted. In C, HNK-1-positive GCPs were treated with Shh-N alone for 2.5 days and then either BMP2, BMP2 and 2 μM Ara-C (BMP2 + Ara-C), or BMP2 and 2 μg/ml aphidicolin (BMP2 + Aph) were added. After 2 additional days of culture, cells were fixed and stained for GFAP. GFAP-positive cells were counted. The results are shown as mean ± SD of four to eight areas and were confirmed in two additional experiments.

Fig. 6.

Induced GFAP-positive cells express S100-β. HNK-1-positive GCPs were cultured in the presence (B) or absence (A) of Shh-N and BMP2 for 2.5 days. Ara-C (2 μM) was then added to the cultures to suppress the proliferation of contaminating astroglia or immature Bergmann glia. The cells were then cultured for an additional 5 days, fixed, and double stained for S100-β (green) and GFAP (red). TO-PRO-3 (blue) was used as a counterstain. The results were confirmed in two additional experiments. (Scale bar: 50 μm.)

Some “Switching Cells” Coexpress Neuronal and Astrocyte Markers Initially. To confirm that GFAP-expressing cells were derived from immature GCPs, we treated the cells with Shh-N and BMP2 for 2.5 days and then double stained them with antibodies directed against GFAP and TAG-1, a cell-surface glycoprotein expressed by some CNS neurons (22). In frozen sections, TAG-1 appeared to be expressed transiently by postmitotic GCPs because it was seen on cells in the deeper zone of EGL but not elsewhere in the EGL, in the molecular layer, or in the GL (Fig. 5D). When purified GCPs in culture were treated with Shh-N and BMP2, some of the GFAP-positive cells also expressed TAG-1 (Fig. 5A). We further examined whether the GFAP-positive cells expressed other neuronal markers. In frozen sections of P7 mouse cerebellum, β-tubulin III, NeuN, and MAP2 were all expressed strongly by GCPs in the deeper zone of the EGL, in the molecular layer, and in the GL (Fig. 5 E, F, and I, respectively). As was the case with TAG-1, when purified GCPs in culture were treated with Shh-N and BMP2, some of the induced GFAP-positive cells expressed these three neuronal markers also (Fig. 5 B, C, G, and H). In these immunostaining studies, we used mouse monoclonal antibodies directed against the neuronal markers and rabbit antibodies directed against GFAP. When we stained the cells with rabbit anti-MAP2 antibodies and a mouse monoclonal anti-GFAP antibody, the results were the same (Fig. 5H). We searched for the cells that coexpressed neuronal and astroglial markers in our cultures in all conditions and found these cells only in the cultures treated with both Shh-N and BMP2; even in these cultures, such cells were seen only in a narrow window between days 2 and 7. These results strongly suggest that the induced GFAP-positive cells developed from immature GCPs.

Fig. 5.

Induced GFAP-positive cells coexpress various neuronal markers. HNK-1-positive GCPs were treated with Shh-N and BMP2 for 2.5 days, fixed, and double stained with rabbit antibodies directed against either GFAP (red in A–C and G) or MAP2 (red in H) and with mouse monoclonal antibodies directed against either TAG-1, β-tubulin III, NeuN, MAP2, or GFAP (green in A–C, G, and H, respectively). TO-PRO-3 (blue) was used as a counterstain. Frozen sections from P7 mouse cerebella were stained with rabbit antibodies directed against GFAP (red in D–F and I) and mouse monoclonal antibodies directed against either TAG-1, β-tubulin III, NeuN, or MAP2 (green in D–F and I, respectively). (Scale bars: A–C, G, and H, 20 μm; D–F and I, 50 μm.)

To confirm that the induced GFAP-positive cells went on to become bona fide astroglial cells, we continued the cultures for 7.5 days before staining them for other astroglial markers. The induced GFAP-positive cells cultured in Shh-N and BMP2 for 7.5 days had longer processes and stained more strongly for GFAP than the induced cells after only 2.5 days of culture. Furthermore, some of them also expressed S100-β, a marker for differentiated astroglial cells (Fig. 6B). None of GFAP-positive cells at either time point expressed BLBP (data not shown). By 7.5 days, all of the induced GFAP-positive cells had lost the expression of all of the neuronal markers (data not shown). Taken together, these results suggest that immature GCPs treated with Shh-N and BMP2 differentiate first into the GFAP- and neuronal marker-positive cells and then finally into GFAP-positive, S100-β-positive, neuronal marker-negative, and differentiated astroglial cells.

Discussion

BMPs and Shh are known to play crucial roles in the development of cerebellum. Alder et al. (2) demonstrated that BMPs initiate the program of granule cell specification, and Angley et al. (23) provided evidence that BMP4 might participate in regulating postnatal granule cell and astroglial cell differentiation. Several laboratories have shown that Shh produced and released by Purkinje cells is a potent mitogen for GCPs (12–15, 24). Also, Dahmane and Ruiz-i-Altaba (13) reported that Shh induces the differentiation of Bergmann glia. Because both BMPs (23) and Shh (12–14) are expressed in the postnatal cerebellum, we examined the effect of both factors administered together to cultured GCPs. We find that a small proportion of the GCPs differentiate into astroglial cells in the presence of both BMPs and Shh but not in the presence of either protein alone.

Is it possible that the GFAP-positive cells that develop in response to BMPs and Shh are actually NSCs rather than bona fide astroglial cells? There is increasing evidence that at least some NSCs express GFAP (25, 26). Furthermore, Kondo and Raff (8) reported that oligodendrocyte precursor cells can be reprogrammed by extracellular signals to resemble NSCs, and the first signals required for reprogramming include BMPs. We think it is unlikely, however, that the GFAP-positive cells induced by BMPs and Shh in our cultures are NSCs for the following reasons. First, when we remove Shh and BMP2 from our cultures and add basic fibroblast growth factor and/or epidermal growth factor to promote NSC proliferation (1), the GFAP-positive cells do not divide or incorporate BrdUrd (data not shown). Second, when we remove Shh and BMP2 and add both platelet-derived growth factor and thyroid hormone to promote oligodendrocyte differentiation of NSCs (8, 27), we do not see oligodendrocytes developing in our cultures (data not shown). Third, some of the GFAP-positive cells induced in our cultures also expressed S100-β, a marker of differentiated astroglial cells. Together, these results strongly suggest that the GFAP-positive cells are astroglial cells rather than NSCs.

It was thought originally that rapidly dividing cells in the EGL give rise to various cell types in the cerebellar cortex, including granule cells, interneurons (basket cells and stellate cells), and some glial cells (24). Recent studies, however, using quail-chick chimeras (28, 29), transplantation of EGL cells (30), or retroviral labeling of EGL cells (31, 32) indicate that cells in the EGL give rise to granule cells exclusively in vivo. Here, we show that some EGL GCPs can differentiate into GFAP-positive glial cells in vitro also. It is unknown whether some GCPs also develop into astroglial cells in vivo. Although it has been shown that both Shh (12–14) and BMPs (23, 33) are expressed in Purkinje cells and in the EGL, it is not known whether the levels of these proteins are sufficient to induce some GCPs to develop into astroglial cells. Such differentiation could easily be missed because only a small proportion of GCPs are induced to become astroglial cells in culture by Shh and BMPs.

We cannot completely exclude the possibility that the induced GFAP-positive cells could have developed from the progenitors in the white matter of the cerebellum, which are reported to give rise to cortical interneurons (stellate cells and basket cells in the molecular layer and Golgi cells in the GL), astroglial cells, and oligodendrocytes (31, 32). We think that the possibility is unlikely, however, for the following two reasons. First, the induced GFAP-positive cells express TAG-1. In developing cerebellum, only the cells in the deeper zone of the EGL express TAG-1; the cells in the molecular layer or in the GL or in the white matter do not express TAG-1 (Fig. 5D). Second, it is reported that none of the progenitors in the white matter was doubly labeled with any combination of markers characteristic to different cell lineages (34), whereas the induced GFAP-positive cells transiently express both neuronal and astroglial markers (Fig. 5).

To our knowledge, this article is the first to show that neuronal precursor cells can differentiate into glial cells and that they are, therefore, not irreversibly committed to differentiate into neurons. Our findings add to the expanding list of examples where precursor cells, or even differentiated cells, can be diverted from their expected developmental fate by extracellular signals (for review, see ref. 35). It seems that such cells are specified rather than irreversibly committed or determined.