Ava i l ab l e on l i ne a t www.sc i enced i r ec t . com

www.e l sev i e r . com/ loca te / sc r

Stem Cell Research (2012) 8, 141–153

Bmi1 marks intermediate precursors duringdifferentiation of human brain tumor initiating cellsChitra Venugopal a, 1, Na Li a, 1, Xin Wang a, Branavan Manoranjan a,Cynthia Hawkins f, Thorsteinn Gunnarsson b, Robert Hollenberg b,Paula Klurfan b, Naresh Murty b, Jacek Kwiecien d, Forough Farrokhyar c, g,John P. Provias d, Christopher Wynder a, Sheila K. Singh a, b, c, e,⁎

a McMaster Stem Cell and Cancer Research Institute, 1280 Main Street West, Hamilton, ON, Canada L8S 4K1b Department of Surgery, 1200 Main Street West, Hamilton, ON, Canada L8N 3Z5c Department of Biochemistry and Biomedical Sciences, 1200 Main Street West, Hamilton, ON Canada L8N 3Z5d Department of Pathology and Molecular Medicine, 1200 Main Street West, Hamilton, ON, Canada L8N 3Z5e Department of Neuroscience, 1200 Main Street West, Hamilton, ON, Canada L8N 3Z5f Arthur and Sonia Labatt Brain Tumor Research Centre and Department of Pediatric Laboratory Medicine,The Hospital for Sick Children and University of Toronto, ON, Canadag Department of Health Research Methodology, Faculty of Health Sciences, McMaster University,1200 Main Street West, Hamilton, ON, Canada L8N 3Z5

Received 5 May 2011; received in revised form 21 September 2011; accepted 28 September 2011

Available online 8 October 2011

Abstract The master regulatory gene Bmi1modulates key stem cell properties in neural precursor cells (NPCs), and has beenimplicated in brain tumorigenesis. We previously identified a population of CD133+ brain tumor cells possessing stem cell prop-erties, known as brain tumor initiating cells (BTICs). Here, we characterize the expression and role of Bmi1 in primary minimal-ly cultured human glioblastoma (GBM) patient isolates in CD133+ and CD133− sorted populations. We find that Bmi1 expressionis increased in CD133− cells, and Bmi1 protein and transcript expression are highest during intermediate stages of differenti-ation as CD133+ BTICs lose their CD133 expression. Furthermore, in vitro stem cell assays and Bmi1 knockdown show that Bmi1contributes to self-renewal in CD133+ populations, but regulates proliferation and cell fate determination in CD133− popula-tions. Finally, we test if our in vitro stem cell assays and Bmi1 expression in BTIC patient isolates are predictive of clinical out-come for GBM patients. Bmi1 expression profiles show a marked elevation in the proneural GBM subtype, and stem cellfrequency as assessed by tumor sphere assays correlates with patient outcome.

© 2011 Elsevier B.V. All rights reserved.

Introduction

⁎ Corresponding author at: McMaster University Stem Cell andCancer Research Institute, 1280 Main Street West, Hamilton, ON,Canada L8S 4K1. Tel.: +1 905 521 2100x75237; fax: +1 905 521 9992.

E-mail address: ssingh@mcmaster.ca (S.K. Singh).1 These authors contributed equally to this work.

1873-5061/$ - see front matter © 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.scr.2011.09.008

Using the cell surface marker, CD133 (Yin et al., 1997; Yuet al., 2002), we identified a subpopulation of cells from a va-riety of human brain tumors, termed brain tumor initiatingcells (BTICs), which exhibit stem cell properties in vitro

142 C. Venugopal et al.

(Singh et al., 2003; Reynolds and Weiss, 1992; Tropepe et al.,1999) and in vivo (Singh et al., 2004; Wang et al., 2010). Sim-ilar to neural precursor cells (NPCs), BTICs generate all neuralcell types through the process of differentiation, during whichCD133+ stem and early progenitor cells lose their CD133 ex-pression, giving rise to late progenitors, and finally differenti-ated progeny. Currently, there is a lack of definitive markersdefining NPC and BTIC early and late progenitors. However,molecular markers which capture cells during the intermedi-ate stages of differentiation, just as CD133 expression levelsbegin to decline, may improve our understanding of signalingpathways that promote BTIC cell fate determination.

The Bmi1 signaling pathway is not only required for execu-tion of key stem cell programs in mammals, but is also consis-tently dysregulated or overexpressed in numerous emergingcancer stem cell populations (Cui et al., 2007; Lessard andSauvageau, 2003; Liu et al., 2006a; Mihic-Probst et al., 2007;Song et al., 2006). The Bmi1 gene, first isolated as an onco-gene in mouse lymphoma (van Lohuizen et al., 1991), wasalso identified as a key regulator of hematopoietic (Lessardand Sauvageau, 2003) and neural stem cell self-renewal (Fasanoet al., 2007, 2009; Molofsky et al., 2005). Bmi1 has been impli-cated in the pathogenesis of brain tumors such as gliomas(Bruggeman et al., 2007) and medulloblastoma (Leung etal., 2004; Michael et al., 2008; Wiederschain et al.,2007), and as an important epigenetic regulator of fate de-termination in stem cell populations (Bracken et al., 2006;Valk-Lingbeek et al., 2004). However, a targeted study ofthe potential role of Bmi1 in further defining BTIC popula-tions has yet to be performed.

Here, we use pre-existing BTIC assays and Bmi1 siRNA-mediated knockdown to characterize the expression and stemcell regulatory role of Bmi1 in primary minimally culturedhuman glioblastoma (GBM) patient isolates prospectively sortedfor CD133. We show that Bmi1 governs key stem cell propertiesin different cellular compartments: self-renewal in CD133+stem and early progenitor cells, and proliferation, differentia-tion and cell fate determination in CD133− progenitors. Ourwork demonstrates that Bmi1 differentially regulates stemcell function and cell identity in BTIC-enriched (CD133+) andBTIC-depleted (CD133−) compartments of human glioblasto-ma. Interestingly, in our minimally cultured GBM BTICs, wefind that Bmi1 mRNA expression level is significantly higher inCD133− cell subpopulations, suggesting that Bmi1 may play apreviously uncharacterized role in CD133− late progenitorpopulations, in addition to its known role in CD133+ stem cells(Abdouh et al., 2009; Molofsky et al., 2003). While contributingto stem cell self-renewal in CD133+ BTIC populations, Bmi1mayalso play a role in regulating proliferation, differentiation andtumor maintenance in CD133− populations. Furthermore, invitro stem cell assays are predictive of clinical outcome forGBM patients, and Bmi1 expression in BTIC patient isolatesmay contribute to current molecular classifications of GBM.

Results

Bmi1 expression is higher in the CD133− fractions ofhuman GBMs

We prospectively sorted CD133+ and CD133− cells fromprimary human fetal brain (FB) specimens (n=34) and

primary human glioblastoma multiforme (GBM) (n=27,Table 1). CD133 expression across all human BT sampleswas 18.1±6.9%, and across human FB samples CD133 expres-sion was 2.6±1.1%. We show that 12 representative primaryGBMs (Fig. 1A) and 5 normal primary FB samples (Supple-mentary Fig. 1A), when assayed after minimal culture anddirectly after formation of primary spheres, yielded aCD133+ enriched fraction that housed significantly higherself-renewal capacity in primary limiting dilution analyseswhen compared to CD133− fractions (Pb0.0001 for both).We offer these data to substantiate the claim that, in mini-mally cultured primary human cells or BTIC patient isolates(Singh et al., 2003, 2004; Uchida et al., 2000), CD133 expres-sion reliably marks stem cell or early progenitor populations,although this claim may not necessarily be extended to BTICcell lines or human glioma stem cells propagated in long-term culture (Abdouh et al., 2009; Chen et al., 2010).

We hypothesized that since Bmi1 is essential for NPC self-renewal (Fasano et al., 2007; Molofsky et al., 2003, 2005;Bruggeman et al., 2005), it should be exclusively expressedin CD133+ BTICs. We assayed mRNA and protein expressionin CD133-sorted populations from human GBMs. Interesting-ly, we found that across all samples, Bmi1 mRNA (Fig. 1B)and protein (Fig. 1C) expression levels are significantlyhigher in CD133− populations compared to CD133+ popula-tions. Similar trends were observed in human NPC popula-tions (Supplementary Figs. 1B and C).

To investigate whether genes downstream of Bmi1 aresimilarly activated in CD133− populations, we performedRT-PCR and microarray experiments on both populations.Bmi1 downstream effector gene p16 is repressed by Bmi1and involved in proliferation control. RT-PCR experimentsshow that p16 levels are decreased in CD133− populations(Fig. 1D for GBMs and Supplementary Fig. 1D for NPCs).Comparative microarray analysis also shows selective si-lencing of p16 as well as other Bmi1 downstream targetssuch as HoxD9, HoxD8, p14 and p27 in CD133− populations(Fig. 1E).

Although Bmi1 expression and stem cell regulatory func-tion have been well studied in CD133+ cells, the role ofBmi1 in CD133− cells remains poorly characterized. Wetherefore decided to investigate the potential role of Bmi1in CD133− brain tumor cells.

Bmi1 expression is highest during intermediatestages of BTIC differentiation

Differentiation is a key stem cell property, during whichCD133+ stem and early progenitor cells lose their CD133 ex-pression to give rise to late progenitors, and finally fully dif-ferentiated neural cells (Singh et al., 2003). We performedmRNA expression analysis of Bmi1 in CD133+ and CD133−cells at days 0, 3 and 7 of differentiation (Fig. 2A). At day3, Bmi1 mRNA expression peaks in the CD133+ populationand diminishes at day 7 to equal that of the CD133− popula-tion. This led us to hypothesize that Bmi1 mRNA expressionmay mark CD133+ progenitors as they lose CD133 expressionduring differentiation.

In order to investigate this hypothesis, we quantified Bmi1protein expression over the time course of differentiation, bysubjecting sorted CD133+ BTICs to differentiation assays

Table 1 Patient data and experimental use of BTIC isolates.

Specimen # Brain tumor ID Age/gender of donor patient Pathology CD133% in unsorted tumor Experimental use in figures

1 BT36 M/67 GBM 24.2 1a, 4b2 BT47 M/59 GBM 12.2 1a, 1b, 1d, 4b, Supp Fig. 3a,b3 BT62 M/54 GBM 16.3 1a, 2a, 2c, 4c4 BT63 M/76 GBM 28.1 1a, 1d, 4b5 BT75 M/69 GBM 22.1 1a, 4b6 BT82 M/49 GBM 10.9 3c, 3d7 BT89 M/57 GBM 24.3 1b, 2b, 2c8 BT92 M/77 GBM 18.4 1b, 1e9 BT104 M/74 GBM 13.2 1c, 2c10 BT106 M/46 GBM 14.1 1a, 1c, 4b11 BT111 F/34 GBM 10.8 1a12 BT114 M/76 GBM 5.9 1d13 BT119 M/63 GBM 24.9 1e14 BT120 M/62 GBM 29.2 1a, 4b15 BT121 M/51 GBM 15.8 1a, 3a, 3b, 3e, 3f16 BT132 M/41 GBM 2.2 2a, 3a, 3b, 4b17 BT147 M/82 GBM 19.2 1a, 4b18 BT149 M/46 GBM 0.8 2a19 BT154 F/59 GBM 10.7 1a, 4b20 BT 54 M/47 GBM 17.0 1a, 1c, 4b21 BT 132 M/41 GBM - 3a22 BT 149 M/46 GBM 15.2 2b, 3c, 3d23 BT210 F/75 GBM 16.8 4b, Supp Fig. 224 BT213 M/67 GBM - Supp Fig. 2, Supp Fig. 3a,b25 BT237 M/51 GBM 11.2 3e, 3f26 BT239 F/65 GBM 17.8 3c, 3d, Supp Fig. 227 BT305 M/48 GBM 10.5 4b, 4c

143Bmi1 marks intermediate precursors in brain tumors

(Singh et al., 2003, 2004; Reynolds and Weiss, 1992; Tropepeet al., 1999) and flow cytometry on days 0, 3, 5, and 7. Resultsfor BTIC patient isolates (Fig. 2B) identified four cell popula-tions that emerge sequentially over the time course of differ-entiation. On day 0, 11% of CD133+ BTICs are Bmi1− and 87%are Bmi1+; on day 3, only 5% of cells remain CD133+Bmi1+and the majority of cells (94%) are CD133–Bmi1+. On day 5,the most prominent population is CD133–Bmi1+, and by day7, 99% of cells are CD133–Bmi1−.

To further characterize these four populations morpho-logically, we performed immunofluorescence analysis onCD133+ BTICs at days 0, 3, 5, and 7 of differentiation, stain-ing for Bmi1, CD133 and markers of fully differentiated neu-ronal and astrocytic lineages (TUJ1 and GFAP, respectively).This staining (Fig. 2C, replicates shown in SupplementaryFig. 2) showed that cellular morphology, combined withmarker expression over time, demonstrates the transitionof undifferentiated BTICs through progenitor stages to fullydifferentiated progeny. Bmi1 expression marks intermediateprecursors during this transition from CD133+ BTICs toCD133− differentiated cells.

We then evaluated the transcript levels of known regula-tors and effectors of Bmi1 throughout differentiation. TheNotch pathway regulates self-renewal of Bmi1+ BTICs (Fanet al., 2009), and p21 (Fasano et al., 2007) is a cell cycle reg-ulator repressed by Bmi1 in NPCs. We show that throughoutdifferentiation of BTICs, p21 transcript levels vary inverselywith Bmi1 levels, whereas Notch1 and Bmi1 transcript levelsdecrease concordantly (Supplementary Fig. 3).

Bmi1 siRNA knockdown exerts multiple effects onstem cell properties of CD133+ and CD133− braintumor cells

To determine a functional role for Bmi1 in our sorted cellpopulations, we carried out knockdown studies on thesecells and subsequently performed in vitro stem cell assaysfor self-renewal, proliferation and differentiation. We firstvalidated that siRNA targeted to Bmi1 decreased mRNA ex-pression of Bmi1 and increased levels of downstream targetp16 (Supplementary Figs. 4A and B).

Limiting dilution analysis estimates the clonogenic stemcell frequency of a whole cell population by quantifying sec-ondary sphere formation, which correlates directly withself-renewal capacity (Singh et al., 2003; Tropepe et al.,1999; Singh et al., 2004; Bellows and Aubin, 1989). Wefound that Bmi1 knockdown significantly reduced the num-ber of secondary spheres, and thus self-renewal capacity,in CD133+ BTICs (Fig. 3A, P=0.023). However, Bmi1 knock-down had no significant effect on secondary sphere forma-tion in the CD133− population (Fig. 3A, P=0.11).

Next, we chose to analyse proliferation on sorted knock-down cells. These assays showed that Bmi1 knockdown re-duced proliferative activity of both populations, especiallythe CD133− population (Fig. 3B, P=0.001 for CD133− cells,P=0.03 for CD133+ cells). Cell cycle analysis of BTICs showedthat Bmi1 knockdown reduced the number of cells found in Sphase (Figs. 3C and D), recapitulating Bmi1 knockdown ef-fects in leukemic stem cells (Lessard and Sauvageau, 2003)

Figure 1 Bmi1 expression is higher in CD133− cells in human brain tumors. (A) In 12 representative primary GBMs (BTs 1–5, 10, 11,14, 15, 17, 19, 20) assayed after minimal culture and directly after formation of primary spheres, CD133+ cells, when compared toCD133− fractions, showed significantly higher secondary sphere formation in primary limiting dilution analyses (Pb0.0001) (B) Bmi1mRNA expression levels are higher in the CD133− cells in human BTs (P=0.012, n=3, BTs 2, 7, 8). (C) Bmi1 protein levels are alsohigher in the CD133− cells (P=0.0004 for BTs, n=3, BTs 9, 10 and 20). U2OS cell lysate served as positive control. Bottom panelshows the immunoblot normalized to GAPDH as internal control. (D) mRNA expression levels of p16 are decreased in CD133− cells(n=3 for BTs 2, 4, 12). (E) Treeview representation of Affymetrix expression data depicting known Bmi1 downstream target genesthat are silenced in the CD133− population (n=2 for BTs 8 and 13). ⁎Pb0.05, **Pb0.01.

144 C. Venugopal et al.

145Bmi1 marks intermediate precursors in brain tumors

and further corroborating the role of Bmi1 in regulation ofproliferation (Lessard and Sauvageau, 2003; Molofsky etal., 2005; Bruggeman et al., 2005). These data show thatBmi1 knockdown affects self-renewal in the CD133+ popula-tion (as shown by decreased sphere number in Fig. 3A),whereas maintenance of proliferation is affected mostly inthe CD133− compartment (as shown by decreased prolifera-tion in Fig. 3B and significantly decreased number of CD133−cells in S phase in Fig. 3D).

To assess the impact of Bmi1 on cell fate determination,we plated Bmi1 knockdown cells for subsequent differentia-tion and staining. Control siRNA treated CD133− cells werefully differentiated with multiple processes after 5 days ofdifferentiation (Fig. 3E). However after Bmi1 knockdown,

Figure 2 Bmi1 expression is highest during intermediate stages ofjected to differentiation assays (n=3 for BTs 3, 16 and 18), Bmi1 mRN7, whereas, in CD133− cells, Bmi1 mRNA expression levels decreasetative flow cytometry analysis of Bmi1 and CD133 protein expressionThroughout differentiation, four distinct cell populations (as describmicroscopic images of unsorted spheres (from representative sampleprofile. Day 0: Undifferentiated spheres show central CD133 (red) anarrows) and CD133+/Bmi1+ (yellow arrows) cell populations emergarrow); Day 7: Few cells retain Bmi1 expression (thick green arro(red arrow, Day 7 left panel) and neuronal marker TUJ1 (red arrows

differentiation was partially blocked as indicated by spheri-cal cells lacking mature processes (Fig. 3E, black arrows).The restriction of neurogenic potential and default to aglial fate reported by others (Fasano et al., 2009; Bruggemanet al., 2007) was seen primarily with Bmi1 knockdown of theCD133− cells (Fig. 3F), whereas CD133+ cells showedcompletely blocked differentiation.

Bmi1 is enriched in the proneural subtype of GBM andtumor sphere assays correlate with patient outcome

We noted that throughout all experiments, Bmi1 expres-sion levels are uniformly higher in brain tumor cell populations

BTIC differentiation. (A) In CD133+ cells from brain tumors sub-A expression levels are increased at day 3 and decreased by dayd throughout differentiation (for day 3, *Pb0.05). (B) Represen-in sorted CD133+ cells from human BT 7 (n=2 for BTs 7 and 22).ed in text) were identified. Inset: isotype controls. (C) Confocal, BT 3; n=3 for BTs 3, 7, 9) show their time course differentiationd peripheral Bmi1 (green) staining ; Day 3: CD133+/Bmi1− (reded; Day 5: Only 50% of cells retain Bmi1 expression (thin greenw) whereas most now express mature astrocytic marker GFAP, Day 7 right panel) respectively.

Figure 3 Bmi1 siRNA knockdown has differential effects on stem cell properties of CD133+ and CD133− cells. (A) Bmi1 siRNA con-siderably decreased secondary sphere formation in CD133+ cells (P=0.023) but not in CD133− cells (P=0.11) (n=3 for BTs 15, 16and 21). (B) Bmi1 knockdown decreased cell proliferation in both populations (n=2 for BTs 15 and 16), although the effect wasmore marked in the CD133− population (P=0.001 for CD133− cells, compared to P=0.03 for CD133+ cells) (RFU=relative fluores-cence units). (C) Cell cycle analysis showed that Bmi1 knockdown decreased the S phase in both CD133+ and CD133− populations,(D) although the effect was more marked in CD133− populations (n=3 for BTs 6, 22, 26). (E) Control siRNA treated CD133− cellswere fully differentiated with multiple processes after 5 days of differentiation (n=2 for BTs 15 and 25; representative samplesshown from BT 15). However after Bmi1 knockdown, differentiation was partially blocked as indicated by spherical cells morpholog-ically representative of dedifferentiation (black arrows). (F) Bmi1 knockdown decreased the number of cells expressing mature neu-ronal markers (TUJ1, MAP2) in CD133− cells compared to control siRNA. ⁎Pb0.05, **Pb0.01, ***Pb0.001.

146 C. Venugopal et al.

compared to normal NPCs (Fig. 1B, Pb0.0001 and Supplemen-tary Fig. 1). This observation is confirmed by reports of Bmi1overexpression in brain tumors such as medulloblastomas(Leung et al., 2004; Michael et al., 2008; Wiederschain etal., 2007; Bracken et al., 2006), and the identification of aBmi1-driven gene signature through a mouse/human compar-ative translational genomics approach which has been

validated in numerous human cancers (Glinsky et al., 2005).Further reports have also suggested that Bmi1 confers radiore-sistance to GBM CD133+ cells through preferential activationof the DNA double-strand break response machinery, leadingto decreased survival (Facchino et al., 2010). Using the TCGAGBM database (http://tcga-data.nci.nih.gov/tcga/findArchives.htm), we analyzed Bmi1 and CD133 expression

147Bmi1 marks intermediate precursors in brain tumors

in 203 GBMs across the 4 GBM subgroups (Fig. 4A). Thesedata show a statistically significant enrichment of bothCD133 and Bmi1 in the proneural subgroup compared toother subgroups. The proneural subgroup is associatedwith younger age, secondary GBM presentation and a higherexpression of oligodendroglial and developmental genes(such as PDGFRA, which is often expressed in the subventri-cular zone where stem cells reside) (Verhaak et al., 2010).The markedly increased expression of BTIC and stem/

Figure 4 BTIC patient isolate stem cell assays correlate with patiedatabase (http://tcgadata.nci.nih.gov/tcga/findArchives.htm), shosubgroup compared to the other GBM subtypes. (B) We analyzed 1224-month period of documented clinical follow-up (BT 1, 2, 4, 5,isolate sphere formation rates against months of survival. There ismation rate and time to survival, indicating that as the numbe(R2=0.723, Pb0.01). (C) 12 patient GBMs (BT 1, 2, 4, 5, 10, 14, 1scored for high or low expression. Proneural GBMs identified by higpared to all other GBM subtypes. (D) Bmi1 (P=0.11) and CD133 (P=0other GBM subtypes (n=5).

progenitor markers such as CD133 and Bmi1 in the proneuralsubtype of GBMs supports its origin from a transformed neu-ral stem or progenitor cell.

To validate the application of in vitro stem cell assays topatient tumors, we undertook statistical comparison of BTICpatient isolate sphere formation rate (which approximatesstem cell frequency in a cell population) and patient survival.We analyzed 12 GBM patients from our patient population whohad at least a 24-month period of documented clinical follow-

nt outcome. (A) Expression analysis of Bmi1 using the TCGA GBMws a statistically significant enrichment of Bmi1 in the ProneuralGBM patients from our patient population who had at least a

10, 14, 16, 17, 19, 20, 23, 27), and plotted their BTIC patientan 85% negative correlation between the secondary sphere for-r of secondary spheres increases, time to survival decreases6, 17, 19, 20, 23, 27) were stained with PDGFRA antibody andh expression of PDGFRA show a lower self-renewal index com-.13) expression is higher in proneural GBMs (n=5) compared to

Table 2 RT-PCR primers.

Bmi1 FBmi1 R

5′GGAGGAGGTGAATGATAAAAGAT3′5′AGGTTCCTCCTCATACATGACA3′

GAPDH FGAPDH

5′TGCACCACCAACTGCTTAGC3′5′GGCATGGACTGTGGTCATGAG3′

p16 Fp16 R

5′CAGGTGGGTAGAGGGTCTGC3′5′GCCAGGAGGAGGTCTGTGATT3′

p21 Fp21 R

5′TGTCACTGTCTTGTACCCTTG3′5′GGCGTTTGGAGTGGTAGAA3′

Notch1 FNotch1 R

5′TGCCTGGACAAGATCAATGAG3′5′CAGGTGTAAGTGTTGGGTCC3′

28s rRNA F28s rRNA R

5′AAGCAGGAGGTGTCAGAAA3′5′GTAAAACTAACCTGTCTCACG3′

148 C. Venugopal et al.

up, and plotted their BTIC patient isolate sphere formationrates against months of survival (Fig. 4B). There is an 85% neg-ative correlation between the secondary sphere formationrate and time to survival, indicating that as the number of sec-ondary spheres increases, time to survival decreases(R2=0.723, Pb0.01). This correlation between BTIC sphereformation rates and patient survival demonstrates that invitro stem cell assays reliably define stem cell frequency andcan be predictive of patient outcome, and may thus serve asvalid surrogates of BTIC xenograft tumorigenicity.

We also wanted to validate the feasibility of implement-ing Bmi1 and CD133 expression studies to identify stem cellpopulations in patient tumors. We performed immunohisto-chemistry for antibodies against CD133, Bmi1 and PDGFRAon tissue sections from the 12 GBM patients representedin Fig. 4B (representative immunohistochemistry shown inSupplementary Fig. 5). We chose PDGFRA as according toTCGA analysis (Verhaak et al., 2010), its high-level expres-sion should specifically correlate with the proneural sub-type of GBMs, and would allow us to segregate proneuralGBMs from all other subtypes. Indeed, we found thatpatients with high levels of PDGFRA expression tended tobe younger patients (see Table 3), often with secondaryGBMs, all of whom were alive at the end of the survivalanalysis. Patient tumors with high levels of PDGFRA

Table 3 GBM patient data with Molecular subtype.

Specimen # Age atdiagnosis(years)

Outcome(at time ofanalysis)

Molecular subtype(as per PDGFRAexpression)

BT 1 67 Dead OtherBT 2 76 Dead OtherBT 3 34 Alive ProneuralBT 4 51 Alive ProneuralBT 5 69 Dead OtherBT 10 46 Alive ProneuralBT 11 59 Dead OtherBT 14 62 Dead OtherBT 15 81 Dead OtherBT 17 76 Dead OtherBT 19 59 Alive ProneuralBT 20 45 Alive Proneural

expression (i.e. proneural GBMs) showed lower rates of sec-ondary sphere formation compared to all other GBM sub-types (Fig. 4C), showing that in these patients, a lowerrate of self-renewal correlates with survival. We then ana-lyzed expression of Bmi1 and CD133 in proneural subtypes(n=5) vs other GBM subtypes (n=5), and found that Bmi1and CD133 expression are higher in the proneural GBMs(Fig. 4D), corresponding with a likely stem cell origin as postu-lated by Verhaak et al. (Verhaak et al., 2010). According toTCGA analysis data, although proneural GBM patients areyounger at diagnosis and have a longer overall survival, theyrespond to treatment poorly and have a relative survival disad-vantage once treated with aggressive therapy. ProneuralGBMs may be particularly refractory to post-surgical chemo-therapy and radiotherapy if their stem cell populationsevade these aggressive treatments, as suggested by the can-cer stem cell hypothesis (Bao et al., 2006; Liu et al., 2006b;McCord et al., 2009; Nakai et al., 2009). These data showthat Bmi1 and CD133 immunostaining combined with currentmolecular subtyping may offer prognostic or diagnostic valueif applied to routine pathological examination in GBM patientpopulations.

Discussion

We have analyzed Bmi1 expression and function in humanGBM patient isolates prospectively sorted for CD133, ratherthan in whole spheres, which are heterogeneous stem cellcolonies that contain a mixture of stem, progenitor and dif-ferentiated cells. We called our experimental model system“BTIC patient isolates” to emphasize the fact that thesecells are not cell lines, but rather minimally cultured underconditions to select for stem cell populations. BTICs areassayed immediately after primary sphere formation, andgrown briefly in human NPC media with LIF. Under theseconditions, we find that CD133 reliably marks stem or earlyprogenitor populations (Singh et al., 2003, 2004; Wang etal., 2010; Uchida et al., 2000) (Fig. 1A), although this findingmay not pertain to human BTICs propagated in long-termculture (Abdouh et al., 2009; Chen et al., 2010).

Bmi1 is known to act in a concentration-dependent man-ner to control a delicate balance between self-renewal anddifferentiation (Cui et al., 2006). We found that Bmi1 ex-pression is highest during intermediate stages of BTIC differ-entiation, when CD133+ progenitors transition to CD133−cells prior to acquisition of a fully differentiated state.These results are corroborated by functional data obtainedfrom Bmi1-GFP knock-in mice (Hosen et al., 2007). It is likelythat Bmi1 partners with other transcription factors in thetemporal control of self-renewal and differentiation. For ex-ample, Bmi1 and Mel18 (Ishida et al., 1993) are highly ho-mologous constituents of the PRC1 complex, and studieshave shown that Mel18 negatively regulates Bmi1 expression(Guo et al., 2007). It has also been observed that Bmi1 ex-pression is upregulated in the earlier stages of differentia-tion, while Mel18 levels increase in the later stages ofdifferentiation (Lessard and Sauvageau, 2003; Hosen et al.,2007). It is possible that when the CD133− cells fully differ-entiate, Mel18 levels increase, thereby decreasing Bmi1levels to give rise to CD133− Bmi1− fully differentiatedcells.

149Bmi1 marks intermediate precursors in brain tumors

Bmi1 activity could also be modulated by the presence ofalternatively spliced variants during self renewal and differ-entiation, hinting at an additional level of regulatory com-plexity in BTICs. Such is the case for Oct4, as Oct4 splicevariants are differentially expressed in human pluripotentand nonpluripotent cells: Oct 4A in stem cells and Oct4B indifferentiated cells (Atlasi et al., 2008). Sorting of braintumor cells for CD133 allowed us to make the novel observa-tion that Bmi1 governs key stem cell properties in differentcellular compartments. Bmi1 regulates self-renewal inCD133+ stem and early progenitor cells, and regulates prolif-eration, differentiation and cell fate determination inCD133− progenitors.

Bmi1 belongs to a family of polycomb group (PcG) pro-teins which together form multimeric complexes responsiblefor gene silencing via chromatin modification. Bmi1, in par-ticular, belongs to the PRC1 complex, which is more func-tionally redundant through paralogous genes, but has beenreported to be critical in cancer development (Margueronand Reinberg, 2011; Simon and Kingston, 2009). The currentmechanistic theory of gene silencing through chromatincompaction follows two steps. First, PRC2 proteins E(Z),ESC, and SU(Z)12 form the core responsible for the initiationof gene repression. Second, the PRC1 complex, which in-cludes PCGF, Ring1, PHC and CHX components, is recruitedand causes a covalent modification leading to chromatincompaction (Sauvageau and Sauvageau, 2008). In the cancerstem cell hypothesis, it is suggested that a sub-population ofcells have stem cell characteristics that lead to the genera-tion of a heterogeneous tumor. Stem cell pluripotency re-quires that lineage specific programs are repressed, andthis has been demonstrated to be partly due to PcG regula-tion. In BTICs, PcG proteins may drive the cells towards astem cell phenotype. Thus, in early BTIC progenitors, it ap-pears that PcG proteins, such as Bmi1, may be responsibleto maintain BTICs in a stem cell state primed for differenti-ation (Sauvageau and Sauvageau, 2008). As the stem cells/early progenitors begin to specify into late progenitors,PcGs appear to function to restrict certain lineage programsfor cell fate determination.

Cellular identity is maintained through chromatin modi-fications during cell division. The Polycomb group proteins,and specifically Bmi1, have been shown to guide cell-fatedecisions in a number of normal somatic and cancer stemcell populations. Initial work in hematopoietic (HSC) (Parket al., 2003), leukemic (LSC) (Lessard and Sauvageau,2003), and neural (NSC) (Molofsky et al., 2003) stem cellsidentified Bmi1 as a regulator of the key stem cell property,self-renewal. Paired daughter cell assays in CD34−c-Kit+Sca-1+lineage marker- HSCs have demonstrated overex-pression of Bmi1 to enhance in vitro symmetrical cell divi-sion of HSCs (Iwama et al., 2004). Our work demonstratesfor the first time that Bmi1 differentially regulates cellidentity in BTIC-enriched (CD133+) and BTIC-depleted(CD133−) compartments of glioblastoma. Although ourwork does not explore the mechanism by which Bmi1 main-tains symmetric and asymmetric cell division in BTICs, itmay be hypothesized that it regulates asymmetric divisionin CD133+ cells and then symmetrically maintains CD133−cells.

The critical role of Bmi1 in tumorigenesis has promptedinvestigations into genes upstream of this proto-oncogene.

Several candidate genes have been reported to be upstreamof Bmi1 in various cancer models. In neuroblastoma, MYCNand E2F-1 appears to be regulating Bmi1 expression (Ochiaiet al., 2010; Nowak et al., 2006), whereas in leukemia, thisrole appears to be played by SALL4 (Yang et al., 2007). Inmedulloblastoma, it appears that sonic hedgehog (Shh)may have a direct regulatory function, and Twist1 may bean important regulator upstream of Bmi1 (Wang et al.,2011). In GBMs however, several reports have shown the im-portance of miRNA-128 as a direct regulator of Bmi1 throughbinding to a 3′-untranslated region of Bmi1 mRNA. Overex-pression of miRNA-128 resulted in a decrease in Bmi1 expres-sion, suggesting that miRNA-128 may represent a potentialtherapeutic target for GBM BTICs (Cui et al., 2010; Godlewskiet al., 2008).

Finally, we sought to test if our in vitro stem cell assaysand marker expression in BTIC patient isolates were clinicallyrelevant for GBM patients. The markedly increased expres-sion of Bmi1 in the proneural GBM subtype reveals the poten-tial for stem cell markers to further elucidate currentmolecular GBM classifications. We also show that our markerscould feasibly be assessed by immunohistochemistry in pa-tient tumors, and further work assessing the functional roleof these presumed BTIC populations will offer validation oftheir future clinical utility. While xenograft models currentlyrepresent the gold standard for BTIC definition (Singh et al.,2004; Bonnet and Dick, 1997), we offer compelling evidencethat sphere formation in early passage BTIC patient isolatesis, in fact, a surrogate for tumorigenicity in xenografts, inthat higher BTIC sphere formation rate correlates stronglywith shorter patient survival. Together with other work link-ing high sphere formation rates to poor patient outcome(Kong et al., 2010; Panosyan et al., 2010; Yuan et al., 2004;Laks et al., 2009), these data may serve to validate sphereformation as a viable surrogate for the xenograft model in de-termining BTIC potency and frequency. Further diagnosticand prognostic information can be extracted from these out-come analyses; for example, if the line of least square linearfit shown in Fig. 4B is intersected at 15 months' survival (themean survival time of all patients with GBM who receivegold standard therapy of surgery, temozolomide chemother-apy and radiation (Stupp et al., 2005)), an intercept of 48.4spheres per 2000 cells is generated, which could be validatedin larger GBM patient cohorts as a prognostic indicator (i.e.,b48.4 spheres predicts longer survival, N48.4 spheres pre-dicts shorter survival).

In multiple cancer models, Bmi1 is an oncoprotein pre-requisite for transformation, metastatic migration, andmaintenance of malignancy (Liu et al., 2006a; Bruggemanet al., 2007; Michael et al., 2008; Glinsky et al., 2005;Dovey et al., 2008). In the context of brain tumorigenesis,our data suggest that Bmi1 may regulate tumor initiation inCD133+ BTICs and tumor maintenance in CD133− prolifera-tive progenitors. However, whether Bmi1 overexpression inhuman NPCs in vivo is necessary and sufficient for braintumor generation remains largely unexplored (Fasano etal., 2009; He et al., 2009). Conversely, the effect of Bmi1knockdown in human BTICs on in vivo brain tumorigenesishas not yet been fully elucidated (Abdouh et al., 2009). Itwill be important to investigate the role of Bmi1 in stemcell maintenance as it applies to in vivo models of braintumorigenesis.

150 C. Venugopal et al.

Materials and methods

Primary cell and sphere culture

We have called our experimental model system “BTIC pa-tient isolates” to emphasize the fact that these cells areonly minimally cultured (24 hours to b1 week) under condi-tions, which select for stem cell populations. These BTIC pa-tient isolates are not cell lines, and small cell numbers havenecessitated experimental repetition in multiple primarypatient isolates to attain statistical significance. Patient iso-lates are described in Table 1. Human brain tumor (BT) andwhole 13- or 14-week fetal brain (FB) samples were obtainedfrom consenting patients, as approved by the HamiltonHealth Sciences/McMaster Health Sciences Research EthicsBoard. As previously published (Singh et al., 2003, 2004),samples were dissociated in artificial cerebrospinal fluidcontaining 0.2 Wunisch Unit/mL Liberase Blendzyme® 3(Roche), and placed in a rocking incubator at 37 °C for15 minutes. The dissociated tissue was then filtered througha 70 μm cell strainer and collected by centrifugation at1500 rpm for 3 minutes. Tumor cells were resuspended inTumor Sphere Medium (TSM) consisting of a chemically de-fined serum-free neural stem cell medium (Singh et al.,2003; Singh et al., 2004), human recombinant EGF (20 ng/mL; Sigma), bFGF (20 ng/mL; Invitrogen), LIF (10 ng/mL;Chemicon), Neural Survival Factor (NSF) (1×; Clonetics), N-acetylcysteine (60 μg/mL; Sigma) and antibiotic antimycoticsolution (Wisent). Red blood cells were removed using am-monium chloride (STEMCELL Technologies). All experimentswere performed following minimal culture (b1 week). Via-bility was measured at N80% for inclusion in these experi-ments, and was ascertained by 7AAD staining by flowcytometry, or trypan blue dye exclusion.

Magnetic cell sorting and flow cytometry analysis

Primary spheres were dissociated with Liberase Blendzyme®3 to single cell suspension. CD133+ cells were purified usingmagnetic bead cell sorting (STEMCELL Technologies) as pre-viously published (Singh et al., 2003; Singh et al., 2004) orflow cytometric sorting. The percentage expression ofunsorted cells, CD133+ selected cells and the CD133-negative fraction was determined by flow cytometry (FACS-Calibur, BD Biosciences or MoFlo XDP, Beckman Coulter)using APC-labeled anti-CD133 antibodies (Miltenyi Biotec).The purity of CD133+ cell fractions from brain tumors was85±3.4%, and the CD133− cell fraction purity was 99.5±0.3%. For human NPCs, the purity of CD133+ cells was 89±4.6%, and CD133− cell fraction purity was 99.3±0.4%. Theappropriate isotype control served as the negative controlfor every experiment.

Limiting dilution assay (LDA)

After primary sphere formation was noted, spheres were dis-sociated to single cells and replated in 0.2 mL TSM as previ-ously published (Singh et al., 2003; Singh et al., 2004).Briefly, neurospheres were treated with Liberase Blend-zyme® 3 and plated at clonal density in 96-well microwellplates in 0.2 mL volumes of TSM. Final cell dilutions ranged

from 200 cells per well to 1 cell per well in 0.2 mL volumes.Cultures were fed 0.025 mL of TSM every 2 days until day 7,when the number of secondary spheres formed per 2000cells was counted and represented graphically in Fig. 1a.To rule out the possibility that primary spheres representedcell aggregates, we plated bulk GBM cultures at 12 hours(n=4) at single cell density, and through primary LDA,found that single cells possess sphere forming capacity. Forprimary LDA, on day 7, the percentage of wells not contain-ing spheres for each cell plating density (F0) is calculatedand plotted against the number of cells per well (x). Thenumber of cells required to form at least one tumor spherein every well is determined from the point at which theline crosses the 0.37 level (F0=e−x) (Tropepe et al., 1999;Bellows and Aubin, 1989).

Cell proliferation assay

Cell proliferation assay was performed on days 0, 3, 5 and 7using CyQuant Cell Proliferation Kit (Invitrogen) as per man-ufacturer's instructions. Cells were plated in 96 well micro-well plates in 0.1 mL TSM with 20% FBS at the density of100 cells/well. The media was changed every other day.The plates were read with FLUOstar Omega fluorescenceMicroplate reader (BMG LABTECH) at λEx=485, λEm=525.

Bmi1 siRNA treatment

Sorted CD133+ and CD133− cells from brain tumors were im-mediately treated with human Bmi1 siRNA, which is a pool of3 target specific siRNAs (Santa Cruz Biotechnology). Follow-ing treatment, Bmi1 mRNA expression was measured byreal-time PCR to determine the efficiency of knockdown.Cells treatedwith scrambled siRNA (Santa Cruz Biotechnology)were used as controls. Knockdown effects were validated bythe use of Bmi1 shRNA constructs and three other independentBmi1 siRNA (Santa Cruz Biotechnology). Secondary sphere for-mation, proliferation assay, cell cycle analysis, and differenti-ation assays were performed prior to and after Bmi1 siRNAtreatment.

Quantitative cell cycle experiments

Cell cycle experiments were performed using BrdU-FITC flowkit (BD Pharmingen). Sorted human BT cells were incubatedwith 10 μM BrdU for 1 hour after treatment with control orBmi1 siRNA. Then, cells were fixed and co-stained with 7-AAD for DNA content.

Differentiation assay and immunofluoresence staining

Cell fixation and immunofluorescence staining were per-formed as described previously (Singh et al., 2003, 2004).For the staining of undifferentiated spheres, unsortedspheres were plated onto glass coverslips coated with 1:10diluted Matrigel (BD Biosciences) in TSM without FBS. Fourhours later, the spheres were fixed and permeabilized withCytoperm/Cytofix buffer kit (BD Biosciences). For stainingof differentiated cells, unsorted spheres were cultured inTSM supplemented with 20% FBS which was replaced every

151Bmi1 marks intermediate precursors in brain tumors

other day. Immunofluorescence staining was performed7 days after differentiation with the appropriate antibodies.All primary antibodies were incubated overnight at 4 °C, inthe following the dilutions: CD133/1 (Miltenyi Biotec)1:200, Bmi1 (Upstate) 1:100, TUJ-1 (Sigma) 1:1000, Nestin(BD bioscience) 1:100 and GFAP (DAKO) 1:100. All secondaryantibodies (conjugated to Alexa 488 or Alexa 647) were di-luted 1:1000. After staining with Hoechst 33342 (MoleculeProbes), slides were coverslipped using Fluorescence Mount-ing Medium (DAKO). Images were taken using Olympus spin-ning disk confocal microscope.

Real-time PCR

Total RNA was isolated using Qiagen RNeasy Micro Kit. cDNAwas synthesized by qScript cDNA supermix followed by real-time PCR using iQSYBR Green Kit (Quanta) with GAPDH as aninternal control. Samples were quantified using Opticonsoftware. Data were presented as the ratio of the gene of in-terest to GAPDH. No changes were noted with 28srRNA as acontrol, validating that gene expression was not affectedby hypoxia (Zhong and Simons, 1999). Primer sequencesare provided in Table 2. To further confirm our resultsobtained in Fig. 1B, real-time RT–PCR was repeated withTaqMan universal PCR Master Mix (Applied Biosystems), andthe Bmi1 Taqman Gene Expression Kit.

Immunohistochemistry

Briefly, 5 μm sections were cut from paraffin-embedded pa-tient GBMs and mounted on positively charged microscopeslides. The sections were baked overnight at 60 °C, dewaxedin xylene, and hydrated through immersion in decreasingconcentrations of ethanol in distilled water. The tissue sec-tions were heat treated for the purpose of antigen retrievaland blocked for endogenous biotin and peroxidase. Tissuesections were stained using antibodies against CD133(Miltenyi Biotec), Bmi1 (R&D systems) and PDGFRA (Cellsignaling), and incubated overnight at 4 °C, according topreviously published protocols (Singh et al., 2003, 2004).10 high-power (400×) fields were blindly assessed by twoauthors (JP, BM) for each of our 12 patients. Average inten-sity of staining was measured using a 0 (negative), 1 (low),2 (intermediate), 3 (high) scale. Average percent of positivestaining was determined by counting positive stains in 100cells per field. Proneural subtypes were classified accordingto staining intensity as having an average score N1.8 and allother tumors with intensity scores b1.5 were identified as“other subtypes”.

Statistical analysis

All quantitative data presented are the mean±sem. Samplesused and the respective n values are listed in the figurelegends. The level of significance was determined byStudent's two-tailed t-test or ANOVA with α=0.05 (GraphPadSoftware).

Supplementary materials related to this article can befound online at doi:10.1016/j.scr.2011.09.008.

Acknowledgments

We express our gratitude to Rebecca Laposa, LaurieAilles, Tamra Werbowetski-Ogilvie, Peter Dirks, John Dickand Mick Bhatia for critical reviews of this paper. This workwas supported by funds from the Department of Surgery atMcMaster University, the Ontario Institute for Cancer Re-search (OICR) and the J.P. Bickell Foundation. N.L. held apost-doctoral fellowship from OICR.

References

Abdouh, M., Facchino, S., Chatoo, W., Balasingam, V., Ferreira, J.,et al., 2009. BMI1 sustains human glioblastoma multiforme stemcell renewal. J. Neurosci. 29, 8884–8896.

Atlasi, Y., Mowla, S.J., Ziaee, S.A., Gokhale, P.J., Andrews, P.W.,2008. OCT4 spliced variants are differentially expressed inhuman pluripotent and nonpluripotent cells. Stem Cells 26,3068–3074.

Bao, S., Wu, Q., McLendon, R.E., Hao, Y., Shi, Q., et al., 2006. Gli-oma stem cells promote radioresistance by preferential activa-tion of the DNA damage response. Nature 444, 756–760.

Bellows, C.G., Aubin, J.E., 1989. Determination of numbers ofosteoprogenitors present in isolated fetal rat calvaria cells invitro. Dev. Biol. 133, 8–13.

Bonnet, D., Dick, J.E., 1997. Human acute myeloid leukemia is orga-nized as a hierarchy that originates from a primitive hematopoi-etic cell. Nat. Med. 3, 730–737.

Bracken, A.P., Dietrich, N., Pasini, D., Hansen, K.H., Helin, K.,2006. Genome-wide mapping of Polycomb target genes unravelstheir roles in cell fate transitions. Genes Dev. 20, 1123–1136.

Bruggeman, S.W., Valk-Lingbeek, M.E., van der Stoop, P.P., Jacobs,J.J., Kieboom, K., et al., 2005. Ink4a and Arf differentially affectcell proliferation and neural stem cell self-renewal in Bmi1-deficient mice. Genes Dev. 19, 1438–1443.

Bruggeman, S.W., Hulsman, D., Tanger, E., Buckle, T., Blom, M., etal., 2007. Bmi1 controls tumor development in an Ink4a/Arf-in-dependent manner in a mouse model for glioma. Cancer Cell12, 328–341.

Chen, R., Nishimura, M.C., Bumbaca, S.M., Kharbanda, S., Forrest,W.F., et al., 2010. A hierarchy of self-renewing tumor-initiating cell types in glioblastoma. Cancer Cell 17, 362–375.

Cui, H., Ma, J., Ding, J., Li, T., Alam, G., et al., 2006. Bmi-1 regu-lates the differentiation and clonogenic self-renewal of I-typeneuroblastoma cells in a concentration-dependent manner.J. Biol. Chem. 281, 34696–34704.

Cui, H., Hu, B., Li, T., Ma, J., Alam, G., et al., 2007. Bmi-1 is essen-tial for the tumorigenicity of neuroblastoma cells. Am. J. Pathol.170, 1370–1378.

Cui, J.G., Zhao, Y., Sethi, P., Li, Y.Y., Mahta, A., et al., 2010. Micro-RNA-128 (miRNA-128) down-regulation in glioblastoma targetsARP5 (ANGPTL6), Bmi-1 and E2F-3a, key regulators of brain cellproliferation. J. Neurooncol 98, 297–304.

Dovey, J.S., Zacharek, S.J., Kim, C.F., Lees, J.A., 2008. Bmi1 is crit-ical for lung tumorigenesis and bronchioalveolar stem cell expan-sion. Proc. Natl. Acad. Sci. U. S. A. 105, 11857–11862.

Facchino, S., Abdouh, M., Chatoo, W., Bernier, G., 2010. BMI1 con-fers radioresistance to normal and cancerous neural stem cellsthrough recruitment of the DNA damage response machinery.J. Neurosci. 30, 10096–10111.

Fan, X., Khaki, L., Zhu, T.S., Soules, M.E., Talsma, C.E., et al.,2009. Notch pathway blockade depletes CD133-positive glioblas-toma cells and inhibits growth of tumor neurospheres and xeno-grafts. Stem Cells 28, 5–16.

Fasano, C.A., Dimos, J.T., Ivanova, N.B., Lowry, N., Lemischka, I.R., et al., 2007. shRNA knockdown of Bmi-1 reveals a critical

152 C. Venugopal et al.

role for p21-Rb pathway in NSC self-renewal during develop-ment. Cell Stem Cell 1, 87–99.

Fasano, C.A., Phoenix, T.N., Kokovay, E., Lowry, N., Elkabetz, Y.,et al., 2009. Bmi-1 cooperates with Foxg1 to maintain neuralstem cell self-renewal in the forebrain. Genes Dev. 23, 561–574.

Glinsky, G.V., Berezovska, O., Glinskii, A.B., 2005. Microarray anal-ysis identifies a death-from-cancer signature predicting therapyfailure in patients with multiple types of cancer. J. Clin. Invest.115, 1503–1521.

Godlewski, J., Nowicki, M.O., Bronisz, A., Williams, S., Otsuki, A.,et al., 2008. Targeting of the Bmi-1 oncogene/stem cell renewalfactor by microRNA-128 inhibits glioma proliferation and self-renewal. Cancer Res. 68, 9125–9130.

Guo, W.J., Datta, S., Band, V., Dimri, G.P., 2007. Mel-18, a poly-comb group protein, regulates cell proliferation and senescencevia transcriptional repression of Bmi-1 and c-Myc oncoproteins.Mol. Biol. Cell 18, 536–546.

He, S., Iwashita, T., Buchstaller, J., Molofsky, A.V., Thomas, D., etal., 2009. Bmi-1 over-expression in neural stem/progenitor cellsincreases proliferation and neurogenesis in culture but has littleeffect on these functions in vivo. Dev. Biol. 328, 257–272.

Hosen, N., Yamane, T., Muijtjens, M., Pham, K., Clarke, M.F., etal., 2007. Bmi-1-green fluorescent protein-knock-in mice revealthe dynamic regulation of bmi-1 expression in normal and leuke-mic hematopoietic cells. Stem Cells 25, 1635–1644.

Ishida, A., Asano, H., Hasegawa, M., Koseki, H., Ono, T., et al.,1993. Cloning and chromosome mapping of the human Mel-18genewhich encodes a DNA-binding protein with a new ‘RING-finger’motif. Gene 129, 249–255.

Iwama, A., Oguro, H., Negishi, M., Kato, Y., Morita, Y., et al., 2004.Enhanced self-renewal of hematopoietic stem cells mediated bythe polycomb gene product Bmi-1. Immunity 21, 843–851.

Kong, D., Banerjee, S., Ahmad, A., Li, Y., Wang, Z., et al., 2010. Epi-thelial to mesenchymal transition is mechanistically linked withstem cell signatures in prostate cancer cells. PLoS One 5, e12445.

Laks, D.R., Masterman-Smith, M., Visnyei, K., Angenieux, B., Orozco,N.M., et al., 2009. Neurosphere formation is an independent pre-dictor of clinical outcome in malignant glioma. Stem Cells 27,980–987.

Lessard, J., Sauvageau, G., 2003. Bmi-1 determines the prolifera-tive capacity of normal and leukaemic stem cells. Nature 423,255–260.

Leung, C., Lingbeek, M., Shakhova, O., Liu, J., Tanger, E., et al.,2004. Bmi1 is essential for cerebellar development and is overex-pressed in human medulloblastomas. Nature 428, 337–341.

Liu, S., Dontu, G., Mantle, I.D., Patel, S., Ahn, N.S., et al., 2006a.Hedgehog signaling and Bmi-1 regulate self-renewal of normal andmalignant humanmammary stem cells. Cancer Res. 66, 6063–6071.

Liu, G., Yuan, X., Zeng, Z., Tunici, P., Ng, H., et al., 2006b. Analysisof gene expression and chemoresistance of CD133+ cancer stemcells in glioblastoma. Mol. Cancer 5, 67.

Margueron, R., Reinberg, D., 2011. The Polycomb complex PRC2 andits mark in life. Nature 469, 343–349.

McCord, A.M., Jamal, M., Williams, E.S., Camphausen, K., Tofilon,P.J., 2009. CD133+ glioblastoma stem-like cells are radiosensi-tive with a defective DNA damage response compared withestablished cell lines. Clin. Cancer Res. 15, 5145–5153.

Michael, L.E., Westerman, B.A., Ermilov, A.N., Wang, A., Ferris, J.,et al., 2008. Bmi1 is required for Hedgehog pathway-driven me-dulloblastoma expansion. Neoplasia 10, 1343–1349 (1345p fol-lowing 1349).

Mihic-Probst, D., Kuster, A., Kilgus, S., Bode-Lesniewska, B., Ingold-Heppner, B., et al., 2007. Consistent expression of the stem cellrenewal factor BMI-1 in primary and metastatic melanoma. Int.J. Cancer 121, 1764–1770.

Molofsky, A.V., Pardal, R., Iwashita, T., Park, I.K., Clarke, M.F., et al.,2003. Bmi-1 dependence distinguishes neural stem cell self-renewal from progenitor proliferation. Nature 425, 962–967.

Molofsky, A.V., He, S., Bydon, M., Morrison, S.J., Pardal, R., 2005.Bmi-1 promotes neural stem cell self-renewal and neural deve-lopment but not mouse growth and survival by repressing thep16Ink4a and p19Arf senescence pathways. Genes Dev. 19,1432–1437.

Nakai, E., Park, K., Yawata, T., Chihara, T., Kumazawa, A., et al.,2009. Enhanced MDR1 expression and chemoresistance of cancerstem cells derived from glioblastoma. Cancer Invest. 27,901–908.

Nowak, K., Kerl, K., Fehr, D., Kramps, C., Gessner, C., et al., 2006.BMI1 is a target gene of E2F-1 and is strongly expressed in prima-ry neuroblastomas. Nucleic Acids Res. 34, 1745–1754.

Ochiai, H., Takenobu, H., Nakagawa, A., Yamaguchi, Y., Kimura,M., et al., 2010. Bmi1 is a MYCN target gene that regulates tu-morigenesis through repression of KIF1Bbeta and TSLC1 in neuro-blastoma. Oncogene 29, 2681–2690.

Panosyan, E.H., Laks, D.R., Masterman-Smith, M., Mottahedeh, J.,Yong, W.H., et al., 2010. Clinical outcome in pediatric glial andembryonal brain tumors correlates with in vitro multi-passageable neurosphere formation. Pediatr. Blood Cancer 55,644–651.

Park, I.K., Qian, D., Kiel, M., Becker, M.W., Pihalja, M., et al., 2003.Bmi-1 is required for maintenance of adult self-renewing haema-topoietic stem cells. Nature 423, 302–305.

Reynolds, B.A., Weiss, S., 1992. Generation of neurons and astro-cytes from isolated cells of the adult mammalian central nervoussystem. Science 255, 1707–1710.

Sauvageau, M., Sauvageau, G., 2008. Polycomb group genes: keep-ing stem cell activity in balance. PLoS Biol. 6, e113.

Simon, J.A., Kingston, R.E., 2009. Mechanisms of polycomb gene silen-cing: knowns and unknowns. Nat. Rev. Mol. Cell Biol. 10, 697–708.

Singh, S.K., Clarke, I.D., Terasaki, M., Bonn, V.E., Hawkins, C., etal., 2003. Identification of a cancer stem cell in human brain tu-mors. Cancer Res. 63, 5821–5828.

Singh, S.K., Hawkins, C., Clarke, I.D., Squire, J.A., Bayani, J., etal., 2004. Identification of human brain tumour initiating cells.Nature 432, 396–401.

Song, L.B., Zeng, M.S., Liao, W.T., Zhang, L., Mo, H.Y., et al., 2006.Bmi-1 is a novel molecular marker of nasopharyngeal carcinomaprogression and immortalizes primary human nasopharyngeal ep-ithelial cells. Cancer Res. 66, 6225–6232.

Stupp, R., Mason, W.P., van den Bent, M.J., Weller, M., Fisher, B.,et al., 2005. Radiotherapy plus concomitant and adjuvant temo-zolomide for glioblastoma. N. Engl. J. Med. 352, 987–996.

Tropepe, V., Sibilia, M., Ciruna, B.G., Rossant, J., Wagner, E.F., etal., 1999. Distinct neural stem cells proliferate in response toEGF and FGF in the developing mouse telencephalon. Dev. Biol.208, 166–188.

Uchida, N., Buck, D.W., He, D., Reitsma, M.J., Masek, M., et al.,2000. Direct isolation of human central nervous system stemcells. Proc. Natl. Acad. Sci. U. S. A. 97, 14720–14725.

Valk-Lingbeek, M.E., Bruggeman, S.W., van Lohuizen, M., 2004. Stemcells and cancer; the polycomb connection. Cell 118, 409–418.

van Lohuizen, M., Frasch, M., Wientjens, E., Berns, A., 1991. Se-quence similarity between the mammalian bmi-1 proto-oncogene and the Drosophila regulatory genes Psc and Su(z)2.Nature 353, 353–355.

Verhaak, R.G., Hoadley, K.A., Purdom, E., Wang, V., Qi, Y., et al.,2010. Integrated genomic analysis identifies clinically relevantsubtypes of glioblastoma characterized by abnormalities inPDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98–110.

Wang, R., Chadalavada, K., Wilshire, J., Kowalik, U., Hovinga, K.E.,et al., 2010. Glioblastoma stem-like cells give rise to tumour en-dothelium. Nature 468, 829–833.

Wang, X., Venugopal, C., Manoranjan, B., McFarlane, N., O'Farrell,E., et al., 2011. Sonic hedgehog regulates Bmi1 in human medul-loblastoma brain tumor-initiating cells. Oncogene. doi:10.1038/onc.2011.232.

153Bmi1 marks intermediate precursors in brain tumors

Wiederschain, D., Chen, L., Johnson, B., Bettano, K., Jackson, D.,et al., 2007. Contribution of polycomb homologues Bmi-1 andMel-18 to medulloblastoma pathogenesis. Mol. Cell. Biol. 27,4968–4979.

Yang, J., Chai, L., Liu, F., Fink, L.M., Lin, P., et al., 2007. Bmi-1 is atarget gene for SALL4 in hematopoietic and leukemic cells. Proc.Natl. Acad. Sci. U. S. A. 104, 10494–10499.

Yin, A.H., Miraglia, S., Zanjani, E.D., Almeida-Porada, G., Ogawa,M., et al., 1997. AC133, a novel marker for human hematopoieticstem and progenitor cells. Blood 90, 5002–5012.

Yu, Y., Flint, A., Dvorin, E.L., Bischoff, J., 2002. AC133-2, a novelisoform of human AC133 stem cell antigen. J. Biol. Chem. 277,20711–20716.

Yuan, X., Curtin, J., Xiong, Y., Liu, G., Waschsmann-Hogiu, S., etal., 2004. Isolation of cancer stem cells from adult glioblastomamultiforme. Oncogene 23, 9392–9400.

Zhong, H., Simons, J.W., 1999. Direct comparison of GAPDH, beta-actin, cyclophilin, and 28S rRNA as internal standards for quanti-fying RNA levels under hypoxia. Biochem. Biophys. Res. Commun.259, 523–526.