Bmi-1 Regulates the Differentiation and Clonogenic Self-renewal of I-type Neuroblastoma Cells in a Concentration-dependent Manner*

Human neuroblastoma I-type cells have been proposed as a population of malignant neural crest stem cells, based on their high tumorigenic potential, expression of stem cell markers, and ability to differentiate into cells of neural crest lineages, including neuroblastic (N-type) and Schwann/glial (S-type) cells. Here, we demonstrate at single cell levels that a subpopulation of I-type cells possess clonogenic self-renewal capacity that requires the Polycomb group family transcription repressor Bmi-1. We further show that Bmi-1 expression levels exert an instructive influence on lineage commitment by I-type cells. Spontaneous and induced differentiation of I-type cells into S-type cells is accompanied by a marked reduction in the level of Bmi-1 expression, and enforced down-regulation of BMI-1 facilitates spontaneous differentiation of I-type cells into S-type cells. By contrast, N-type neuroblastoma cell lines and differentiated N-type cells express higher levels of Bmi-1 relative to I-type cells, and overexpression of BMI-1 promotes the differentiation of I-type cells along the neuronal lineage. Thus, Bmi-1 acts in a concentration-dependent manner in the control of the delicate balance between the self-renewal and differentiation of neuroblastoma I-type cells. These observations suggest that graded activation of a master regulator within individual tumors could trigger divergent developmental programs responsible for both tumor growth and heterogeneity.

Neuroblastoma is a common childhood malignant tumor of the sympathetic nervous system (1), arising in sympathetic ganglia and adrenal medulla that are derived from neural crest stem cells (2,3). Clinically, neuroblastoma is a heterogeneous group of tumors, displaying histopathological features that range from tumors with predominant undifferentiated neuroblasts to those largely consisting of fully differentiated neurons surrounded by a dense stroma of Schwann cells (1,4). Cell lines established from human neuroblastomas also show the same cellular heterogeneity. Based on morphological appearances, biochemical properties, and growth pat-terns, three major cell types have been identified in neuroblastoma cell lines, which were designated as N-(neuroblastic), S-(substrate-adherent and non-neuronal), and I-type (intermediate) neuroblastoma cells (5).
N-type cells, the most common cell type in neuroblastoma cell lines, have small and rounded cell bodies, scant cytoplasm, and neurite-like processes. They attach poorly to the substrate and often form focal aggregates in culture. These cells express neurofilaments and enzymes specific to noradrenergic neurons. S-type cells have large and flattened bodies with abundant cytoplasm, display an epithelial-or glial-like morphology, and grow in culture as a monolayer. These cells express no neuronal markers, but show a protein expression pattern consistent with Schwann/melanoblastic cell lineages of the neural crest (6 -11). Of particular interest are I-type cells that have a morphology intermediate to those of N-and S-type cells. These cells have small but flattened cell bodies with or without neurite-like processes, attach modestly to the substrate, and express low levels of both N-and S-type cell marker proteins (12). Several lines of evidence suggest that I-type cells may represent a population of neuroblastoma stem cells or malignant neural crest stem cells. First, I-type cells are multipotent and can differentiate to either N-or S-type cells when induced by specific agents (13). Second, I-type cells express stem cell markers CD133 and c-kit (14). Finally, in comparison with N-and S-type cells, I-type cells exhibit significantly higher clonogenic activity in soft agar culture and tumorigenic potential in immunodeficient mice, and their numbers in primary neuroblastoma tumors correlate with disease progression (5,14).
Bmi-1 is a member of the Polycomb group family of transcription repressors that was originally identified as an oncogenic partner of c-Myc in murine lymphomagenesis (15,16). It is a component of the Polycomb repressive complex 1, which represses gene expression through chromatin modifications (17). Bmi-1 has diverse biological functions as demonstrated by phenotypic analysis of mice lacking Bmi-1. These mice are defective in axial skeleton patterning and hematopoiesis, and have neurological abnormalities (18). Further studies revealed that Bmi-1 is required for the selfrenewing proliferation of several types of normal and cancer stem cells, including neural crest stem cells (19 -23). Bmi-1deficient mice displayed progressive loss of neural crest stem cells, and Bmi-1 Ϫ/Ϫ neural crest stem cells showed a reduced ability to self-renew in culture (21,22).
In this study, we characterized the stem cell activities of neuroblastoma I-type cells and their regulation, using the human neuroblastoma cell line BE(2)-C as a model. BE(2)-C cells have a typical I-type phenotype, and show consistent morphological and biochemical responses to differentiation-inducing agents (13). Our study demonstrates the existence in the BE(2)-C cell line of a subpopulation of cells with stem cell-like activities and a critical role of Bmi-1 expression levels in regulation of their clonogenic self-renewal and multilineage differentiation.

EXPERIMENTAL PROCEDURES
Cell Culture and Differentiation Assays-The human neuroblastoma and glioblastoma cell lines were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Invitrogen). The human neuroblastoma I-type BE(2)-C cells (ATCC CRL-2268) were subcultured at the split ratio of 1:5 every 3 days. Unless indicated, all of the studies were conducted with BE(2)-C cells and their derivatives that had been cultured for less than 12 passages. For differentiation assays, BrdUrd 3 and RA (Sigma) were dissolved in Me 2 SO and 10 mM stock solutions were prepared. BE(2)-C cells were treated with 10 M RA for 1-2 weeks or with 10 M BrdUrd for 2-3 weeks. BE(2)-C cells treated with Me 2 SO were used as negative control.
Soft Agar Clonogenic Assays-BE(2)-C cells were mixed in 0.3% Noble agar (in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum) and plated at 1000 cells/well onto 6-well plates containing a solidified bottom layer (0.6% Noble agar in the same growth medium). After 14 days, colonies were either unstained or stained with 5 mg/ml MTT (Sigma), and photographed. For serial soft agar assays, individual colonies were removed from the agar with sterile Pasteur pipettes and pooled. After brief treatment with trypsin, colonies were mechanically dissociated into single cell suspensions and replated at 1000 cells/well (6-well plates).
Immunofluorescence-For spontaneous differentiation assays, individual primary and secondary BE(2)-C colonies in soft agar were transferred onto laminin-coated glass coverslips and allowed to adhere for 24 h. For agent-induced differentiation assays, secondary colonies were dissociated into single cell suspensions and plated on coverslips, followed by 10 M RA or BrdUrd treatment. The colonies and cells were fixed with a solution of 95% ethanol and 5% glacial acetic acid at Ϫ20°C for 20 min, permeabilized with 0.3% Triton X-100 in phosphatebuffered saline at room temperature for 5 min, and blocked with 5% milk for 1 h. Cells were incubated at 4°C overnight with primary antibodies as follows: 1:100 HNK-1 (IgM) mouse monoclonal antibody (BD Pharmingen); 1:400 peripherin rabbit antiserum (Chemicon); 1:200 GFAP rabbit polyclonal antibody (Sigma); 1:200 vimentin mouse monoclonal antibody (sc-6260, Santa Cruz); and 1:100 S100 rabbit polyclonal antibody (Dako Cytomation). After washing with phosphate-buffered saline, cells were incubated with 1:1000 Alexa Fluor 568 goat anti-mouse IgG, 1:1000 Alexa Fluor 594 goat anti-rabbit IgG, 1:1000 phycoerythrin rat anti-mouse IgM, or 1:800 fluorescein goat anti-rabbit IgG (Molecular Probes) at room temperature for 30 min. Cells were stained with DAPI (300 nM in phosphatebuffered saline) to visualize nuclei and photographed with a Nikon Eclipse E800 microscope with Image-Pro Plus software for image analysis.

RESULTS
A Subpopulation of BE(2)-C Cells Possess Clonogenic Self-renewal Capacity-Self-renewal is an essential property of stem cells, which, by definition, is the ability to generate progeny cells with the same developmental potential as their parental cells. Two in vitro clonogenic assays have been widely used to determine the self-renewal capacity of individual stem cells. In the sphere formation assay, cells are plated at a clonal density so that individual cells will form spatially distinct spheres (26). In the colony formation assay, soft agar or methylcellulose is used as the semisolid support media to prevent the migration of cells, which also leads to the formation of spatially distinct colonies (27)(28)(29). Moreover, recent studies suggest that the capacity of anchorage-independent growth is a property of stem cells (30). In both clonogenic assays, a primary sphere or colony is dissociated into single cells and replated. Formation of new spheres or colonies that maintain a multilineage potential demonstrates self-renewal. BE(2)-C cells did not form spheres under a variety of culture conditions (data not shown), therefore, we examined their capacity to undergo clonogenic self-renewal in serial soft agar assays. BE(2)-C cells were plated at 1000 cells per well (6-well culture plates). Examination of the wells under a microscope immediately after plating revealed mostly individual cells (Ͼ95%, data not shown). Colonies, defined as a collection of more than 50 cells, appeared between 10 and 14 days after plating (Fig. 1A, 1 0 well). Based on the calculation by dividing the number of colonies by the number of cells plated, we estimated that ϳ20% of BE(2)-C cells were able to give rise to colonies in soft agar (cloning efficiency). Individual primary colonies containing more than 100 cells were removed from the agar with sterile Pasteur pipettes, pooled, and dissociated into single cell suspensions. The cells were replated at 1000 cells per well. The single cell nature of the suspensions was again confirmed by microscopy. After 2 weeks, secondary (2 0 ) colonies were formed that were similar in size to the primary colonies ( Fig. 1, 2 0 well). Notably, cells replated from 2 0 , 3 0 , and 4 0 colonies showed a gradual increase in cloning efficiency (from ϳ20% in passage 2 0 to ϳ60% in passage 5 0 ), indicating an expansion of clonogenic BE(2)-C cells during the successive soft agar assays. The largest colonies contained more than 5000 cells, which required at least 12 cell divisions for a single cell to achieve the number. Thus, at the end of the successive soft agar assays, a subpopulation of BE(2)-C cells must have undergone at least 60 cell divisions. We also performed serial soft agar assays with BE(2)-C cells from individual colonies. At each passage, single colonies of similar size (containing more than 500 cells) were removed from soft agar and dissociated into single cells, and the entire progeny of each colony was plated into one well of 6-well plates. In all cases, we observed the formation of colonies within 2 weeks (data not shown). Together, these data demonstrate that a subpopulation of BE(2)-C cells can maintain and expand themselves over extended clonal passages in soft agar.
To ascertain that the colonies were clones of individual BE(2)-C cells, we mixed equal numbers of BE(2)-C cells that were infected with either vector control (pBabe-puro) or green fluorescent protein-expressing (pBabe-GFP) retroviruses. We examined at least 100 primary and secondary colonies in two independent experiments, and observed no colonies of mixed nature (Fig. 1B), demonstrating that the colonies arose from single cells and not from cell aggregates.
Clonally Derived BE(2)-C Progeny Cells Retain the Capacity to Undergo Spontaneously Multilineage Differentiation-We next determined whether the clonally derived BE(2)-C progeny cells retain the differentiation potential of their parental cells by immunofluorescence staining of primary and secondary colonies for lineage markers HNK-1, peripherin, and GFAP. HNK-1 is a marker for avian and human neural crest stem cells (31,32). It is also expressed in histologically immature neuroblastoma cells, with subsequent loss of expression in more differentiated cells (33). Peripherin is expressed by neural crest-derived peripheral neurons (34,35), and GFAP is a marker for Schwann cells (36). Individual primary (n ϭ 20) and secondary colonies (n ϭ 20) in soft agar were transferred separately onto laminincoated coverslips, allowed to adhere for 24 h, and processed for dual immunofluorescence staining for HNK-1 and peripherin or HNK-1 and GFAP (Fig. 2). All of the primary and secondary colonies examined contained 20 -60% of cells that were stained strongly for HNK-1. We also found that 20 -40% of cells expressed high levels of peripherin and 5-20% of cells were stained positively for GFAP. We wanted to point out that these were approximate numbers because a significant portion of the cells remained in aggregates even after a 24-h culture on coverslips. Nonetheless, these data demonstrate that clonally derived BE(2)-C cells can spontaneously differentiate into both neurons and Schwann/glial cells.
Clonally Derived BE(2)-C Progeny Cells Retain the Ability to Undergo Agent-induced Multilineage Differentiation-We further determined whether the clonally derived BE(2)-C cells, like their parental cells, can undergo agent-induced differentiation into the cells of neural crest lineages. Individual secondary col-  Clonally derived BE(2)-C progeny cells retain the capacity to undergo spontaneous multilineage differentiation. Immunofluorescence staining of representative primary and secondary colonies for the expression of lineage markers HNK-1, peripherin, and/or GFAP. Individual colonies were removed from soft agar, plated separately onto laminin-coated glass coverslips, and processed for immunofluorescence staining 24 h after plating. Scale bars, 100 m.
onies were dissociated into single cells, plated onto glass coverslips, and were either untreated or treated with 10 M of RA, which induces differentiation of BE(2)-C cells into neurons (13,37). Most of the untreated cells showed a typical I-type cell morphology, with a round and prominent nucleus, and relatively abundant cytoplasm (Fig. 3A, Un). Cells treated with RA for 7 days exhibited the neuronal morphology, with many long neuritic processes and small cell bodies that often formed aggregates (Fig. 3A, RA-7d). To confirm the phenotype of RAtreated BE(2)-C cells, we performed immunofluorescence staining of the cells with antibodies against HNK-1 and peripherin. Compared with untreated cells (Fig. 3, B and C, Un), RA treatment resulted in a marked decrease in the number of cells expressing HNK-1 (from an average of 45 to 21%), which was accompanied by a significant increase in the number of cells expressing peripherin (from an average of 41 to 90%) (Fig. 3, B and C, RA-7d). The up-regulation of peripherin was further confirmed by immunoblotting, which showed a progressive increase in the cellular levels of peripherin during the course of RA treatment (Fig. 3D). We noticed that a subpopulation of either untreated or RA-treated cells expressed both HNK-1 and peripherin (stained yellow in Fig. 3B), which probably represent neuronal precursors or partially differentiated neurons. It is also possible that a subpopulation of differentiated neurons continued to express HNK-1.
We next treated the clonally derived BE(2)-C cells with BrdUrd, an agent inducing differentiation of BE(2)-C cells into Schwann-like S-type cells (13,38). Cells treated with BrdUrd for 14 days displayed the morphology of S-type cells with enlarged and flattened bodies, and abundant cytoplasm (Fig.  4A, BrdU-14d). To confirm the phenotype of BrdUrd-treated BE(2)-C cells, we performed immunofluorescence staining of the cells with antibodies against GFAP, vimentin, and S100. Vimentin is an intermediate filament highly expressed in S-type neuroblastoma cells (13), and S100, like GFAP, is a common marker for Schwann/glial cells (36). Compared with untreated cells (Fig. 4, B-F, Un), BrdUrd-treated cells showed marked down-regulation of HNK-1 expression, which was accompanied by a significant increase in the number of cells that expressed GFAP, vimentin, and S100 (Fig. 4, B-F,  BrdU-14d). The most dramatic change was the increase in the number of cells that expressed high levels of vimentin following BrdUrd treatment (Fig. 4C). Quantitative analysis revealed that BrdUrd treatment reduced the proportion of HNK-1-expressing BE(2)-C cells from an average of 45 to 22% and increased the vimentin-expressing cell population from an average of 12 to 92% (Fig. 4E). Consistent with the results of immunofluorescence studies, immunoblot analysis revealed a progressive increase in the cellular levels of vimentin during the course of BrdUrd treatment (Fig. 4F). These differentiation analyses, combined with the above serial soft agar clonogenic assays, demonstrate that a subpopulation of BE(2)-C cells can self-renew, generating progeny cells with the same multilineage differentiation potential of their parental cells.
Differentiation is associated with exiting from the stem cell state. As expected, BE(2)-C cells treated with either RA or BrdUrd were unable to grow into colonies in soft agar (Fig. 4G), indicating a loss of their clonogenic activity.
Down-regulation of BMI-1 in BE(2)-C Cells Promotes Spontaneous Differentiation Along the Schwann/Glial Lineage-It has been shown recently that Bmi-1 is essential for maintaining the self-renewal capacity of neural crest stem cells (21). As neuroblastoma originates from cells of neural crest origin and BE(2)-C cells show significant levels of Bmi-1 expression (Fig.  5A), we investigated the possibility that Bmi-1 may also regulate the clonogenic self-renewal of BE(2)-C cells. We generated two BMI-1 siRNA-expressing retroviral constructs that target different regions of the human Bmi-1-coding sequence. Retroviruses produced from either construct were effective in down-regulating BMI-1 expression in BE(2)-C cells and, on average, BE(2)-C cells infected with BMI-1 siRNA-expressing retroviruses showed a 70% reduction in Bmi-1 levels (Fig.  5A, Bmi-1si). No such effect was observed in BE(2)-C cells infected with the vector control retroviruses (Fig. 5A, pSuper) or retroviruses expressing siRNA against human MYCN or p53 (data not shown), demonstrating the specificity of the BMI-1 siRNA constructs. Down-regulation of BMI-1 abrogated the ability of BE(2)-C cells to grow in soft agar (Fig. 5B), suggesting that Bmi-1 is required for the clonogenic capacity of BE(2)-C cells. The inability of the BE(2)-C cells with reduced levels of Bmi-1 to undergo clonogenic self-renewal in soft agar could result from defects in regulation of proliferation, apoptosis, and/or differentiation. We detected no significant effect of BMI-1 knockdown on the cell cycle status of BE(2)-C cells under standard culture conditions (data not shown), in agreement with previous findings in hematopoietic stem cells from Bmi-1 Ϫ/Ϫ mice (19). Also, we found no significant difference in the basal levels of apoptosis between BE(2)-C cells infected with the vector control or BMI-1 siRNA-expressing retroviruses by annexin-V and trypan blue assays (data not shown). These results suggest that Bmi-1 maintains the clonogenic self-renewal capacity of BE(2)-C cells through a mechanism that is largely independent of cell cycle and apoptosis regulation. We wanted to point out that although BMI-1 knockdown had no significant effect on the cell cycle status and survival of BE(2)-C cells, longterm culture apparently selected for the subpopulation of BE(2)-C/Bmi-1si cells that retained normal levels of Bmi-1 expression. Whereas early passage (passage 9) BE(2)-C/Bmi-1si cells showed a significant reduction in the Bmi-1 level, late-passage (passage 18) BE(2)-C/Bmi-1si cells expressed Bmi-1 at a level similar to that of control BE(2)-C/pSuper cells (supplemental Fig. S1A), and regained the ability to grow in soft agar (supplemental Fig. S1B). Thus, Bmi-1 is required for sustaining long-term growth of BE(2)-C cells in the culture, most likely by maintaining their self-renewal capacity.
During the study, we noticed that about 20 -30% of early passage BE(2)-C/Bmi-1si cells displayed the morphology of S-type cells, with large and flattened bodies (Fig. 5C, BE(2)-C/Bmi-1si, Phase). To assess their differentiation state, immunofluorescence staining of cell lineage markers was performed. Compared with BE(2)-C/pSuper cells (Fig. 5C, pSuper), BE(2)-C/Bmi-1si cells showed a marked decrease in the expression of HNK-1 and peripherin, which was accompanied by an increase in the expression of the S-type cell markers, GFAP, vimentin, and S100 (Fig. 5C, Bmi-1si). Quantitative analysis revealed that BMI-1 knockdown resulted in a reduction in the proportion of HNK-1-expressing BE(2)-C cells from an average of 41 to 21% and an increase in the vimentin-expressing cell population from an average of 13 to 66% (Fig. 5D). The up-regulation of vimentin was also confirmed by immunoblot analysis, which revealed a 6.3-fold increase in the vimentin level in BE(2)-C/Bmi-1si cells (Fig. 5E). Together, these results suggest that down-regulation of Bmi-1 in BE(2)-C cells promotes spontaneous differentiation into S-type cells. Thus, Bmi-1 maintains the stem cell state of BE(2)-C cells at least partly by inhibiting their differentiation.
We further examined the effect of BMI-1 down-regulation on the ability of BE(2)-C cells to undergo neuronal differentiation in response to RA. After a 7-day treatment with RA, most of BE(2)-C/pSuper cells displayed the neuronal morphology with a concomitant increase in the level of peripherin expression (Fig. 5F, BE(2)-C/pSuper). In contrast, most of BE(2)-C/Bmi-1si cells failed to undergo neuronal differentiation, as revealed by both their morphology and high-level expression of vimentin (Fig. 5F, BE(2)-C/ Bmi-1si). Thus, down-regulation of BMI-1 not only instructs BE(2)-C cells to differentiate into S-type cells, but also inhibits their neuronal differentiation potential.
Up-regulation of Bmi-1 in BE(2)-C Cells Promotes Spontaneous Differentiation Along the Neuronal Pathway-Given the critical role of Bmi-1 in maintaining the stem cell state of BE(2)-C cells, we asked whether up-regulation of BMI-1 would enhance their clonogenic capacity. The human BMI-1 gene was introduced into BE(2)-C cells by retroviral-mediated gene transfer, and overexpression of BMI-1 was confirmed by immunoblotting (Fig. 6A). Much to our surprise, BMI-1-overexpressing BE(2)-C cells formed much smaller colonies in soft agar than did their vector control cells (Fig. 6B), indicating that BMI-1 up-regulation actually inhibited the clonogenic activity of BE(2)-C cells. We found no difference in the basal levels of apoptosis and the cell cycle status between the vector control (pBabe) and Bmi-1-overexpressing cells (data not shown). We next examined the effect of BMI-1 up-regulation on the differentiation state of BE(2)-C cells. Compared with the control BE(2)-C/pBabe cells (Fig. 6C, BE(2)-C/pBabe, Phase), BE(2)-C/Bmi-1 cells had a tendency to form focal aggregates and neurite-like processes (Fig. 6C, BE(2)-C/Bmi-1, Phase). Moreover, BE(2)-C/Bmi-1 cells displayed an expression pattern of lineage markers similar to that of BE(2)-C cells undergoing neuronal differentiation, with a reduction in the level of HNK-1 expression and a concomitant increase in the level of peripherin expression (Fig. 6C, HNK-1/Peripherin). Quantitative analysis revealed that Bmi-1 expression resulted in a reduction in the proportion of HNK-1-expressing cells from an average of 43 to 23% and an increase in the peripherin-expressing cell population from an average of 41 to 88% (Fig. 6D). The up-regulation of peripherin was further confirmed by immunoblotting, which revealed a 3.1-fold increase in the peripherin level in BE(2)-C/ Bmi-1 cells (Fig. 6E). Together, these results suggest that BMI-1 overexpression instructs BE(2)-C cells to exit from the stem cell state and to differentiate along the neuronal pathway.
We further examined the effect of BMI-1 overexpression on the ability of BE(2)-C cells to differentiate into S-type cells in response to BrdUrd treatment. After a 17-day treatment with BrdUrd, nearly all of the control BE(2)-C cells displayed the morphology of S-type cells and expressed high levels of vimentin (Fig. 6F, BE(2)-C/ pBabe). In contrast, most of the BE(2)-C/Bmi-1 cells failed to differentiate into S-type cells. Instead, these cells showed the morphology of N-type cells and expressed high levels of peripherin (Fig. 6F, BE(2)-C/Bmi-1). Thus, overexpression of BMI-1 not only instructs BE(2)-C cells to differentiate into N-type cells, but also inhibits their differentiation along the S-type cell pathway.
Cell Type-dependent Regulation of BMI-1 Expression in Human Neuroblastoma Cell Lines-Given that enforced overexpression or down-regulation of BMI-1 in BE(2)-C cells facilitated spontaneous differentiation into N-or S-type cells, respectively, we addressed the question of whether such changes in Bmi-1 levels occur in spontaneously or agent-in- C, analyses of BE(2)-C/pSuper and BE(2)-C/Bmi-1si cells for their morphology and expression of the lineage markers HNK-1, peripherin, GFAP, vimentin, and S100. Nuclei were stained with DAPI. Images represent similar results from at least three independent experiments. Scale bar, 100 m. D, quantification of HNK-1-or vimentin-expressing populations in BE(2)-C/pSuper and BE(2)-C/Bmi-1si cells. Cells were counted from at least 5 randomly selected fields (ϳ200 cells/field), and data are presented as mean Ϯ S.E. E, immunoblot analysis of vimentin levels in BE(2)-C/pSuper and BE(2)-C/Bmi-1si cells. Relative levels of vimentin are indicated. ␣-Tubulin levels are shown as loading control. F, immunofluorescence staining of BE(2)-C/pSuper and BE(2)-C/Bmi-1si cells for the expression of the lineage markers vimentin and peripherin following 7-day RA treatment. Nuclei were stained with DAPI. Images are representatives of at least three independent experiments. Scale bar, 50 m. NOVEMBER 10, 2006 • VOLUME 281 • NUMBER 45 duced differentiated N-and S-type neuroblastoma cell lines. We first examined the effects of RA or BrdUrd treatment on BMI-1 expression in BE(2)-C cells. After 7-or 14-day RA treatments, BE(2)-C cells showed an average of 2-fold increase in the Bmi-1 levels (Fig. 7A, RA). By contrast, BE(2)-C cells displayed a progressive down-regulation of BMI-1 during the course of BrdUrd treatment (Fig. 7A, BrdU).

Bmi-1 Regulation of Neuroblastoma Stem Cell Activities
We next examined Bmi-1 expression levels in several N-and S-type neuroblastoma cell lines that had been well characterized in terms of morphology and expression of lineage markers, such as neurofilaments 68, 160, and 200, neuronal specific RNA-binding protein Hu, neurotransmitter biosynthetic enzymes, vimentin, and S100A6 (6 -8, 10 -12, 14, 39). Also, in agreement with the previous report that HNK-1 expression is down-regulated in more differentiated neuroblastoma cells relative to histologically immature neuroblastoma cells (33), several N-and S-type neuroblastoma cell lines examined by immunofluorescence staining displayed either undetectable or significantly lower levels of HNK-1 expression compared with BE(2)-C cells (data not shown). The neuroblastoma cell lines SH-SY5Y (N-type) and SHEP1 (S-type) were clonally derived from the neuroblastoma cell line SK-N-SH (7,39). Despite their common cellular origin, the N-type SH-SY5Y cells showed a 1.8-fold increase in the expression level of Bmi-1 compared with the S-type SHEP1 cells (Fig.  7B). Similarly, all of the 7 N-type human neuroblastoma cell lines expressed higher levels of Bmi-1 than BE(2)-C cells (Fig. 7C). Together, these analyses revealed a pattern of cell type-dependent regulation of Bmi-1 expression in human neuroblastoma cell lines, with S-, I-, and N-type cells expressing low, intermediate, and high levels of Bmi-1, respectively.
Similar to the S-type SHEP1 neuroblastoma cells, human glioblastoma cell lines also expressed very low or undetectable levels of Bmi-1 (Fig. 7C). Thus, down-regulation of BMI-1 appears to be a common feature of Schwann/glial tumor cells. Interestingly, these glioblastoma cells were able to maintain themselves over extended clonal passages in soft agar (data not shown), suggesting a Bmi-1-independent mechanism for maintaining their clonogenic self-renewal capacity.

DISCUSSION
The present study provides direct evidence that the BE(2)-C I-type neuroblastoma cell line contains a subpopulation of cells with stem cell-like activities. These cells are capable of maintaining and expanding themselves over extended clonal passages in soft agar, and, based on both morphology and expression of lineagespecific markers, their clonally derived progeny cells retain the capacity to differentiate into neurons and Schwann/glial cells, two major cell types of the sympathetic nervous system. We further show that the Polycomb protein Bmi-1 is a key regulator of the stem cell activities of I-type neuroblastoma cells. Bmi-1 is not only required for their clonogenic self-renewal, but also regulatestheirlineage-specificdifferentiationinaconcentrationdependent manner.
Our findings have a number of implications in understanding the development of neuroblastoma. Although the cell of origin of neuroblastoma remains to be defined, there is evidence suggesting that neuroblastoma originates from primitive sympathetic precursor cells with multilineage differentiation potential characteristic of neural crest stem cells. Fluorescence in situ hybridization demonstrates that neuroblasts and Schwann cells within individual neuroblastomas are derived from a common multipotent tumor cell (40). Morphological and biochemical analyses also indicate that human neuroblastoma cells are capable of differentiation along multiple lineages that recapitulates neural crest development (11,13). Notably, recent studies have presented convincing evidence for the presence of neural crest stem cells in the peripheral nervous system, including the sympathetic nervous system (21,(41)(42)(43). Our study suggests that the BE(2)-C cells with stem cell-like activities are likely within the HNK-1 ϩ subpopulation, as differentiation is associated with down-regulation of HNK-1 and loss of self-renewing clonogenic capacity. Given that HNK-1 is a marker for human neural crest stem cells (32), our data are consistent with the suggestion that I-type neuroblastoma cells represent malignant neural crest stem cells (13). Together, these observations raise the possibility that sympathetic neural crest stem cells may be the target for initial oncogenic transformation during neuroblastoma development. Further studies, including isolation and molecular characterization of neuroblastoma stem cells, may shed light on their cellular origin.
Our data further suggest a molecular link between the development of the sympathetic nervous system and the pathogenesis of neuroblastoma. Bmi-1 is essential for the maintenance of neural crest stem cells (21,22), which give rise to, among other cell types, cells of the sympathetic nervous system (2,3). In this study, we found that Bmi-1 expression at an intermediate level is required for the clonogenic activity of BE(2)-C cells, which correlates closely with the tumorigenicity of cancer cell lines (44). Taken together, these observations suggest that neuroblastoma development depends on an inappropriate activation of pathways that normally regulate the self-renewal of neural crest stem cells. We speculate that aberrant regulation of BMI-1 expression during early development of the sympathetic nervous system may lead to expansion of the pool of neural crest stem cells, and some of these cells may acquire additional genetic mutations, such as MYCN amplification, leading to neuroblastoma genesis.
The process of tumorigenesis is characterized by the generation of tumor cells with heterogeneous morphologies and biochemical activities (45). The cancer stem cell model suggests that the cellular heterogeneity of tumors results from aberrant differentiation of cancer stem cells (46). It is unclear at molecular levels how the divergent developmental programs are activated and regulated within a tumor. We found that neuroblastoma cell lines representing different cell types within individual neuroblastomas display graded activation of Bmi-1, with the Schwannlike S-type cells, the intermediate I-type cells, and the neuroblastic N-type cells expressing low, intermediate, and high levels of Bmi-1, respectively. More importantly, we demonstrate that Bmi-1 expression levels exert an instructive influence on lineage commitment by I-type neuroblastoma cells (Fig. 7D). Whereas Bmi-1 expression at an intermediate level appears to be essential for maintaining their stem cell state, lower or higher levels of Bmi-1 facilitate their spontaneous differentiation along the Schwann/glial or neuronal pathway, respectively. Our findings are also consistent with genetic evidence. A recent study has shown that Bmi-1-deficient mice display a significant increase in the number of astroglial cells in the central nervous system and neural stem cells from these mice preferentially generate astroglial cells in culture (47). Thus, down-regulation of Bmi-1 also instructs neural stem cells to differentiate into glial cells. Based on these observations, we suggest that neuroblastoma stem cells within individual neuroblastomas choose between various cell fates specified by different threshold levels of Bmi-1, leading to the generation of neuroblastoma heterogeneity. We further speculate that such graded activation of master regulator oncogenes may be a common mechanism underlying tumor heterogeneity.
The mechanism that controls the graded activation of Bmi-1 in neuroblastoma cells remains to be elucidated. However, mechanisms that regulate graded activation of transcription factors during normal development are well documented. For example, in the neural tube of mammals the morphogen Shh is secreted by two ventral midline signaling centers, the notochord and the floor plate, leading to the formation of an extracellular, ventral-to-dorsal concentration gradient of Shh. This gradient of Shh signaling is transduced into graded activation of the Gli family of transcription factors, which determines alternative neuronal phenotypes in the ventral half of the neural tube (48). Given the striking similarity between the development of normal organs and tumors, it may not be too difficult to envision the presence of multiple signaling centers within individual neuroblastomas that would secrete a morphogen-like factor, leading to the formation of multifocal concentration gradients of the factor. Depending on their positions around a particular signaling center, neuroblastoma stem cells would receive different amounts of the signal and, in response, express different levels of Bmi-1. In this respect, it is interesting to note that BMI-1 has been shown recently to be a downstream target of Shh signaling (49). Our preliminary studies indicate that the Shh pathway is also activated in some neuroblastoma cell lines and primary neuroblastomas (data not shown). Thus, the Shh signaling gradient could be one of the mechanisms for graded BMI-1 expression in neuroblastoma. Another potential regulator is RA, which, as shown in this study, can induce BMI-1 in neuroblastoma cells. Like Shh, RA has been shown to act in a concentration-dependent manner in the specification of neuron types along the dorsoventral axis of the neural tube (50,51).
In summary, our study presents an example of transcription regulators acting in a concentration-dependent manner in the control of stem cell activities of cancer cells. This finding suggests a molecular mechanism for initiating divergent developmental programs within individual tumors that are responsible for both tumor growth and heterogeneity. Thus, graded activation of transcription factors, which has been shown previously to govern various normal developmental processes, may also be fundamental to the development of human cancers.