Recent findings have demonstrated that malignant tumors, including glioblastoma multiforme, contain cancer-initiating cells (also known as cancer stem cells), which self-renew and are malignant, with features of tissue-specific stem cells. As these cells are resistant to irradiation and anti-cancer drugs, it is important to characterize them and find targeting therapies. In this study, we established two primary human glioma cell lines from anaplastic oligodendroglioma and glioblastoma multiforme. These lines were enriched in glioma-initiating cells, as just 10 cells formed malignant glioma when injected into mouse brain. We used these cell lines to examine the roles of the Notch, Hedgehog and Wnt signaling pathways, which are involved in stem-cell maintenance and tumorigenesis, to determine which of these pathways are crucial to glioma-initiating cells and their regulation. Here we show that the Hedgehog pathway is indispensable for glioma-initiating cell proliferation and tumorigenesis; the Hedgehog signaling inhibitors prevented glioma-initiating cell proliferation, while signaling inhibitors for Notch or Wnt did not. Overexpression of Gli2ΔC, a C-terminal-truncated form of Gli2 that antagonizes Gli transcription factor functions, blocked glioma-initiating cell proliferation in culture and tumorigenesis in vivo. Knockdown of the Gli downstream factor Cdc2 also prevented glioma-initiating cell proliferation. Taken together, these results show that the Hedgehog→ Gli→ Cdc2 signaling cascade plays a role in the proliferation and malignancy of glioma-initiating cells. (Cancer Sci 2011; 102: 1306–1312)
Cancer-initiating cells (CIC) are capable of indefinite self-renewal and generate the amplifying cancer cells that make up the majority of cells in a tumor.(1–4) The CIC-enriched cell populations can be obtained from various types of cancers including glioma, and from cancer cell lines with characteristics in common with tissue-specific stem cells such as side population cells, cell-surface antigens such as CD133, floating sphere formation or a combination of these features.(2–4) However, CIC and their specific markers and targets are not well characterized.
There are several lines of evidence that glioma cells and normal neural stem cells (NSC) share many mechanisms, including signaling pathways activated by Notch, Wnt and Hedgehog (Hh).(1,5) Notch signaling is strongly activated in both primary human gliomas and in many glioma cell lines.(6) Depleting Notch1 or the Notch ligands delta like 1 and Jagged1 by RNAi blocks glioma cell proliferation in vivo and in vitro.(7)
Ectopically activated Hh signaling in the central nervous system (CNS) causes brain tumors to form.(8,9) Gli1, a Hh-signaling effector molecule, is highly activated in many brain cancers(8) including glioblastoma multiforme (GBM). Overexpressing Gli1 in the developing tadpole CNS induces brain tumors,(10) whereas cyclopamine, which specifically inhibits Hh signaling, blocks the growth of primary gliomas and glioma cell lines.(10,11)
Finally, the canonical Wnt signaling pathway is involved in gliomagenesis; one of its essential mediators, β-catenin, is upregulated along with Wnt2 and Wnt5a in human gliomas and glioma cell lines. Overexpressing a stabilized form of β-catenin in NSC causes lateral ventricle hyperplasia,(12) whereas β-catenin knockdown inhibits glioma cell proliferation in vitro and in vivo.(13,14)
The Notch, Hh and Wnt signaling pathways all appear to be important for gliomagenesis, but it is uncertain whether they are equally essential for the proliferation and maintenance of the glioma-initiating cell (GIC). Using human GIC lines established from anaplastic oligodendroglioma (AO) and GBM, we show that the Hh signaling cascade Hh→Gli→Cdc2 is indispensable for GIC proliferation and malignancy, whereas inhibiting either the Notch or Wnt signaling pathway has little effect.
Materials and Methods
Animals and chemicals. Mice were obtained from Charles River Japan, Inc. (Yokohama, Japan). All mouse experimental protocols were approved by the RIKEN Center for Developmental Biology Animal Care and Use Committee. Chemicals and growth factors were purchased from Sigma (St. Louis, MO, USA) and Peprotech (Rocky Hill, NJ, USA), respectively, except where indicated.
Primary human glioma cell culture and human brain tumors. Two primary human glioma samples, AO and GBM, were obtained at Kumamoto University Hospital with patients’ consent, according to the Research Ethics Committee guidelines. Tumor samples were washed twice with PBS and dissociated with an enzymatic solution containing 0.25% Trypsin in 0.1 mm EDTA at 37°C for 60 min. Dissociated cells were then cultured as tumor spheres in serum-free NeuroBasal Medium-A (Gibco, Invitrogen, Carlsbad, CA, USA) containing human basic fibroblast growth factor (bFGF; 20 μm), human epidermal growth factor (EGF; 20 μm), human leukemia inhibitory factor (LIF, 20 μm; Chemicon, Millipore, Billerica, MA, USA), heparin (5 μm), insulin (10 μm), N2 supplement (1%, Gibco), B27 supplement (1%, Gibco) and GlutaMAX1 (Gibco) (NSC medium). The GIC were cultured on poly-d-lysine (PDL; 15 μg/mL)- and fibronectin (1 μg/mL)-coated eight-well chamber slides (Nunc) for immunostaining. To assess GIC multipotentiality, the cells were cultured in DMEM/F12 (Gibco) with 1% fetal calf serum alone (differentiation medium) for 2 days. Two GIC and four GBM were used in accordance with the research guidelines of the RIKEN CDB and the Kumamoto University Graduate School of Medical Science.
Intracranial cell transplantation into the brain of nude mice. The GIC were suspended in 5 μL of culture medium and injected into the brain of 5- to 8-week-old female nude mice, as described previously.(15)
Mouse brain fixation and histopathology. Mouse brains were dissected and fixed in 4% paraformaldehyde overnight, transferred to 70% ethanol, processed on Tissue-Tek VIP (Sakura Finetek Japan, Tokyo, Japan) and embedded in paraffin. Coronal sections (6 μm thick) from the cerebral cortex were prepared on a Microtome and stained with hematoxylin–eosin (HE).
Immunostaining. Immunostaining was carried out as previously described.(16,17) The following antibodies were used to detect antigens: rabbit anti-CD133 (1:200; Abcam, Cambridge, MA, USA), mouse anti-Nestin (1:200; Chemicon), mouse anti-beta tubulin isotype III (βIII tubulin) (1:200; Sigma), rabbit anti-glial fibrillary acidic protein (GFAP) (1:400; Dako, Copenhagen, Denmark), rat anti-galactocerebroside (GC) (1:4; hybridoma supernatant; ATCC, Manassas, VA, USA) and rat anti-green fluorescent protein (GFP) (1:400; Nacalai Tesque, Kyoto, Japan). Antibodies were detected with goat anti-rat IgG-Alexa488 (1:400; Molecular Probes, Eugene, OR, USA), goat anti-rabbit IgG-Cy3 (1:400; Jackson ImmunoResearch, West Grove, PA, USA) and goat anti-mouse IgG-Alexa488 (1:400; Molecular Probes). The cells were counterstained with DAPI (1 μg/mL) to label the nuclei.
Paraffin-embedded tumors were sectioned at 6 μm thickness and immunostained as previously described.(15) The following antibodies were used to detect antigens: mouse monoclonal anti-Bmi1 (1:200; Abcam), mouse monoclonal anti-Notch (1:200; Abcam), rabbit polyclonal anti-Hes1 (1:200; Santa Cruz Biotechnology, Santa Cruz, CA, USA), rabbit polyclonal anti-Hes5 (1:200; Chemicon), rabbit polyclonal anti-β-catenin (1:200; Abcam), rabbit anti-Sox2 (1:500; StemCell Technologies, Vancouver, BC, Canada), guinea pig anti-Gli2 (1:100),(16) mouse anti N-myc (1:200; Calbiochem, Merck, Darmstadt, Germany) and mouse anti-Cdc2 (1:500; Abcam).
For TUNEL assays, the In Situ Cell Death Detection Kit, TMR red was used according to the supplier’s instructions (Roche, Basel, Switzerland).
RT-PCR. Total RNA was prepared with the RNeasy mini kit (Qiagen, Hilden, Germany), the cDNA was synthesized with the Transcription First Strand cDNA Synthesis kit (Roche) and RT-PCR was carried out as previously described.(16,17) Primer sequences are provided in Table S1.
Vector construction. Vector construction was carried out as previously described.(15,16) In brief, full-length human sox2, bmi1, n-myc and cdc2 were amplified from human fetal brain cDNA libraries (Clontech, Mountain View, CA, USA) using RT-PCR and KOD plus DNA polymerase ver.2 (Toyobo, Osaka, Japan) according to the manufacturer’s instructions, and were cloned into a pMOSBlue vector (GE Healthcare, Little Chalfont, UK). To construct the FLAG-tagged expression vector, cDNA was inserted into a p3xFLAG-CMV10 vector (Sigma) (FLAG-cont), resulting in p3xFLAG-sox2 (FLAG-Sox2), p3xFLAG-bmi1 (FLAG-Bmi1), p3xFLAG-n-myc (FLAG-Nmyc) and p3xFLAG-cdc2 (FLAG-Cdc2). Primer sequences for cDNA amplication are provided in Table S2.
To construct a tetracycline-dependent Gli2ΔC expression vector, Gli2ΔC cDNA was inserted into a pTRE2-hyg vector (Clontech), producing pTRE2-hyg-Gli2ΔC.
To knockdown human bmi1, n-myc, sox2 and cdc2, their hairpin sequences were inserted into a psiRNA-h7SKhygro G1 expression vector (InvivoGen, San Diego, CA, USA), producing psiRNA-h7SKhygro-bmi1sh, psiRNA-h7SKhygro-nmycsh, psiRNA-h7SKhygro-sox2sh and psiRNA-h7SKhygro-cdc2sh, respectively. The siRNA target sequences for human n-myc, sox2 and cdc2 were 5′-GAAGAAATCGACGTGGTCACT-3′, 5′-GAAGGAGCACCCGGATTATAA-3′, and 5′-GAATCTTTACAGGACTATA-3′, respectively. The siRNA target sequence for bmi1 was kindly provided by Dr Maarten van Lohuizen of The Netherlands Cancer Institute (Amsterdam, The Netherlands).
Transfection. Transfection was performed as described previously.(15,16) In brief, the cells were transfected with the vector using the Nucleofector device (Lonza, Cologne, Germany) according to the supplier’s instructions, and were cultured for 24 h, after which the GFP-positive cells were collected by JSAN flow cytometry (Bay Bioscience, Kobe, Japan) and immunostained as described above.
For shRNA experiments, transfected cells were selected in an optimized medium with Hygromycin-B (1 mg/mL; Wako Chemical, Osaka, Japan) for 10 days. Stable hygromycin-resistant clones, of which there were more than 50, were mixed and used for experiments.
To establish inducible Gli2ΔC-bearing GIC lines, GIC were transfected with pTet-On vector (Clonetech) as described above, and selected with G418 sulfate (1 mg/mL; Nacalai Tesque). The G418-resistant GIC were then transfected with the pTRE2-hyg-Gli2ΔC vector and selected with Hygromycin-B. Inducible Gli2ΔC-bearing GIC clones were established by limiting dilution assays, and their proliferation was assayed in the presence of Doxycycline hyclate (Dox, Clontech). The selected clones were transplanted into the brain of nude mice, and tumorigenesis was examined by daily intraperitoneal Dox injection (100 μL of 150 μg/mL) for up to 30 days.
Cell proliferation assay. To examine cell proliferation, MTT assays were performed as described previously.(15,16) Various concentrations of γ-Secretase Inhibitor (Calbiochem), Cyclopamine (Toronto Research Chemicals, Toronto, Canada) and Wnt inhibitor (WIF-1) (R&D Systems, Minneapolis, MN, USA) were added to the medium to examine the effects of blockading the Notch, Hh and Wnt signaling pathways, respectively.
Statistical analysis. Survival data were analyzed for significance using Kaplan–Meier methods with GraphPad Prism version 5 software (P-values were calculated with the Log-rank test).
Characterization of human GIC lines. We established two human GIC lines, GIC1 and GIC2, from AO and GBM, respectively, as described previously.(18,19) Both GIC lines formed floating spheres in NSC medium and could be expanded and maintained thereafter.
We first examined the tumor-forming ability of each GIC line in vivo by injecting 103 cells into the brain of nude mice. All mice (n = 5 for each cell line) transplanted with GIC1 or GIC2 developed malignant glioma with hypercellularity, nuclear pleomorphism, mitosis, microvascular proliferation, hemorrhage and necrosis (Fig. 1A) and died within 53 or 42 days, respectively (Fig. 1B). Detailed histopathological examination revealed that GIC1-derived tumor was composed of a dense population of mitotically active cells with rounded, hyperchromatic nuclei and abundant microcysts, while GIC2 formed hypercellular tumors composed of mixed large and small anaplastic cells, consistent with their original pathological features. We then evaluated the frequency of GIC in these lines. All mice (n = 5 each) injected with 100 and 10 GIC1 cells died within 70 and 108 days, respectively, while all mice (n = 5 each) injected with 100 and 10 GIC2 cells died within 49 and 118 days, respectively (Fig. 1B). Using limiting dilution assays we further found that 15 of 24 single clones of GIC1 and 19 of 24 single clones of GIC2 were expandable in culture and formed colonies in soft agar. In addition, all of 10 GIC sublines, which were picked up randomly, retained their tumorigenicity in vivo (data not shown), suggesting that our GIC lines are enriched in GIC.
To characterize these GIC lines further, we immunolabeled them for NSC markers CD133 and Nestin, and for the neuronal, astrocyte and oligodendrocyte differentiation markers βIII tubulin, glial fibrillary acidic protein (GFAP) and galactocerebroside (GC). Both GIC lines prominently expressed the NSC markers, and only a small number of cells were positive for differentiation markers in NSC medium (Fig. 1C). In contrast, when the cells were cultured in differentiation medium for 2 days, they lost their NSC marker expression and expressed differentiation markers. As GIC1 was established from AO, we examined whether GIC1 predominantly expresses mature oligodendrocyte markers, myelin basic protein and glutathione S-transferase-pi in the presence of the inducers, IGF1 and thyroid hormone,(20) compared with GIC2. However, we could not find such a tendency (data not shown). Together, these data suggest that both GIC lines are multipotent.
Notch, Hh, Wnt and other NSC-related signaling factors are expressed in GIC. We used RT-PCR to determine whether NSC-related genes, including genes involved in the Notch, Hh and Wnt signaling pathways, are expressed in GIC. As shown in Figure 2A, both GIC1 and GIC2 expressed a number of these genes, although GIC2 did not express wnt3a, wnt5a or wnt10b, and neither GIC line expressed sox1 nor jagged (jag) 1. We could not detect the expression of jag2, deltex (dtx) 1-4, wnt4, wnt7, wnt8, wnt11, wnt receptor frizzled homolog (fzd) 2, fzd4, fzd5, fzd8, secreted frizzled-related protein (sfrp) 1, sfrp2, sfrp4, wnt inhibitory factor (wif), seven-in-absentia (siah) 1 and 2 or norrin in either GIC line (data not shown). We confirmed the expression of Sox2, Bmi1, Gli2, N-myc, Notch1, Hes1, Hes5 and β-catenin in the paraffin-embedded GIC xenografts by immunohistochemical analysis (Fig. 2B). These data suggest that GIC hijack the molecular machineries that work in NSC.
The Hh signaling pathway is indispensable for GIC proliferation, survival and tumorigenesis. While there is much evidence that Notch, Wnt and Hh are all involved in gliomagenesis,(1,5) it is not clear which of these signaling pathways are essential for GIC proliferation and malignancy. To address this question, we examined the effect of a γ-secretase inhibitor, cyclopamine, and Wnt inhibitory factor 1 (WIF1), which specifically block the Notch, Hh and Wnt signaling pathways, respectively, in GIC. We found that cyclopamine strongly prevented cell proliferation in both GIC lines, whereas WIF1 or the γ-secretase inhibitor did not (Fig. 3). We confirmed that other Hh signaling inhibitors, JK418, Jervine, SANT1 and U18666A (Data S1) could block GIC proliferation (Fig. S1).
To verify that the Hh pathway is critical for GIC proliferation, we constructed a truncated form of Gli2 (Gli2ΔC) that antagonizes the functions of all three Gli transcription factors, overexpressed it in the cells and examined their proliferation. We found that Gli2ΔC overexpression prevented both GIC lines from proliferating and induced their death, just as it does in NSC(16) (Fig. 4).
We then examined whether Gli2ΔC overexpression prevents GIC tumorigenesis in vivo. Because overexpressing Gli2ΔC induced cell cycle arrest and apoptosis, we established the tetracycline (Tet)-inducible Gli2ΔC-expressing GIC lines Tet-On/Gli2ΔC-GIC1 and Tet-On/Gli2ΔC-GIC2. We confirmed that Tet-On/Gli2ΔC-GIC proliferation was blocked by the presence of Dox in culture (Fig. 5A). We then transplanted Tet-On/Gli2ΔC-GIC into the brain of nude mice and injected either saline (control) or Dox intraperitoneally daily for 30 days. All mice treated with saline died with malignant glioma within 40 days, whereas the mice treated with Dox had better survival rates (median survival: GIC1, 38 [control] vs 59 [Dox] days; GIC2:36 [control] vs 69 [Dox] days) (Fig. 5B). In addition, we found that tumors in the Dox-treated mouse brains did not show necrosis or microvascular proliferation, essential features of GBM, although they still showed hypercellularity and infiltrated normal parenchyma, suggesting that Gli2ΔC overexpression reduced the malignancy of the cells (Fig. 5C).
The Gli → Cdc2 pathway is essential for GIC proliferation and tumorigenesis. Although there is increasing evidence that Gli transcription factors regulate many important factors in both NSC and cancer cells,(21,22) it has not been clear which of Gli’s downstream factors are important for GIC proliferation. We were interested in the Gli-downstream factors Sox2, Bmi1, N-myc and Cdc2 as they play essential roles in both NSC and human glioma,(16,23–27) and we began by examining their expression in human GBM and GIC lines by RT-PCR. We found that human primary GBM samples and GIC lines express all of these factors (Fig. S2A). In addition, immunolabeling confirmed that Gli2, Sox2, Bmi1, N-myc and Cdc2 were present in all four GBM tissues (Fig. S2B) and in GIC xenografts (Fig. 2B and Fig. S2C), although the tissues also contained marker-negative cells.
To examine the functions of these candidate factors in GIC lines, we constructed shRNA expression vectors that knocked down each factor (Fig. S3), transfected them into the GIC lines, and examined the cell proliferation by MTT assay. We found that Cdc2 knockdown inhibited the proliferation of both GIC lines, but knockdown of other factors had little effect on proliferation (Fig. 6), even though it inhibited NSC marker expression (data not shown). As it was shown that inhibition of Sox2 or Bmi1 resulted in impaired proliferation of GIC,(28–30) we further examined whether Sox2 and Bmi1 are generally essential for GIC proliferation. We cultured four tumorigenic GBM cell lines, U251, T98G, U118 and C6, in the NSC medium to enrich for GIC and examined the expression of sox2, bmi1 and cdc2 in the cells. We found that all cells express cdc2, whereas bmi1 expression was at an undetectable level in these cells. In addition, sox2 was expressed in U251 and C6 but not in the others (Fig. S4). Together with the immunolabeling data showing that GBM consists of both these marker-positive and -negative cells (Fig. 2B and Fig. S2B), this suggests that all GIC might not express Bmi1 and Sox2. Thus, these data indicate that the Gli downstream factor Cdc2 is essential for GIC proliferation.
In the present study we established two human GIC lines, GIC1 and GIC2, from AO and GBM, respectively. Both GIC lines not only retained characteristics of NSC, but also formed malignant glioma with their original pathological features when transplanted in vivo, suggesting that they were enriched with CIC. Although GIC1 and GIC2 showed similar characteristics, we found in a parallel experiment that tumorigenesis of GIC1 was prevented by a combination of Cox2 and epidermal growth factor receptor (EGFR) signaling inhibitors, whereas that of GIC2 was not.(31) There are some possible explanations for this result. First, EGFR signaling is known to be involved in the proliferation and tumorigenesis of oligodendrocyte precursor cells.(32,33) Second, we found that GIC1 expressed EGFRvIII, which was shown by Mellinghoff et al.(34) to cause “pathway addiction” of the tumor cells. Third, the accumulation of Cox2-expressing astrocytes is more often observed in high-grade oligodendroglioma than in GBM.(35,36) Thus, these findings suggest that the therapeutic targets for malignant glioma are different according to the cell-of-origin of the GIC.
Hh signaling inhibitors blocked CIC proliferation, whereas either WIF or a γ-secretase inhibitor had little effect, suggesting that the Hh signaling pathway is crucial for GIC proliferation. Since many papers have shown that inhibitors of Notch or Wnt signaling block cancer cell proliferation, the inhibitors used in the present study might not have effectively prevented these pathways, or alternative pathways such as LIF and transforming growth factor might directly activate functional molecules in the Notch and Wnt signaling pathways in GIC lines, as shown previously.(37,38) Therefore, the question of whether Notch and Wnt signaling pathways are involved in GIC proliferation and maintenance still needs to be investigated further.
Our findings clearly indicate that the Hh/Gli signaling pathway is essential for gliomagenesis, as both Gli2ΔC overexpression and Hh inhibitors inhibited proliferation in these GIC lines. These findings suggest that inactivating the Hh/Gli pathway using RNAi or small chemical inhibitors is a potential therapeutic strategy. In addition, because in the absence of Hh both Gli2 and Gli3 are processed at their C-terminus and converted to dominant-negative regulators that block all Gli transcription factor functions,(39,40) it is of interest to determine which protease is involved in this processing and evaluate it as a potential therapeutic target.
The Hh/Gli signaling pathway is essential for the maintenance and proliferation of both NSC and brain tumor cells.(41) Indeed, Gli2ΔC overexpression induces cell cycle arrest and apoptosis not only in GIC (the current study) and in a C6 rat glioma cell line (data not shown), but also in NSC,(16) indicating the importance of finding drugs that kill GIC but not NSC. This also raises the question of whether GIC hijack the Hh/Gli signaling network in the process of transforming from non-NSC. Alternatively, NSC, as the glioma cell-of-origin, might transform into GIC. Both hypotheses seem to be true, as we have successfully induced GIC from both NSC and oligodendrocyte precursor cells when the cells lost p53 and expressed oncogenic H-Ras.(31)
Among the potential Gli target factors Bmi1, N-myc, Sox2 and Cdc2, we found Cdc2 to be indispensable for GIC proliferation. This is consistent with a previous finding that inhibiting Cdc2 induces cell cycle arrest and apoptosis at the G2/M phase in glioma cells.(42) Because Cdc2 is a critical regulator for the G2/M transition of every proliferating cell, it cannot serve as a direct target for therapy. However, the mechanism by which Gli regulates Cdc2 expression involves many transcription factors, including E2F and CREB, and that mechanism might be a therapeutic target. The findings that knockdown of Bmi1 and Sox2 did not inhibit GIC proliferation and that human GBM cell lines, T98G and U118, expressed neither bmi1 nor sox2 suggest that Bmi1 and Sox2 might not be essential for GIC proliferation. Nonetheless, as Bmi1, N-myc or Sox2 knockdown prevented NSC marker expression in our GIC (data not shown), it will be interesting to investigate whether their knockdown blocks the tumorigenesis of our GIC lines.
The authors thank Maarten van Lohuizen (The Netherlands Cancer Institute, Amsterdam, The Netherlands) for the Bmi1 shRNA sequence, K. Ichimura and P. Collins (University of Cambridge, Cambridge, UK) for U251, T98G and U118 human GBM cell lines, and the members in Dr Kondo’s Laboratory at the RIKEN CDB for their helpful comments and discussion. This work was supported in part by a Grant-In-Aid for Scientific Research on Priority Areas from the Ministry of Education, Culture, Sports, Science, and Technology of Japan to T.K.