Main

Glioblastoma (GBM) is the most frequent and lethal primary brain tumor with inevitable recurrence in the vast majority of cases after conventional therapy.1 Therefore, there is an urgent need to develop novel therapeutic options that effectively target therapy-resistant GBM cells. Cancer stem cells in GBM (glioma stem cells: GSCs) are a subpopulation of tumor cells that retains undifferentiated stem cell characteristics (stemness) and high tumorigenic potential.2 Evidence is accumulating that GSCs drive GBM initiation and propagation and contribute to the development of resistance to current treatment options.3, 4, 5, 6 Therefore, this provides a novel therapeutic rationale for targeting GSC in GBM. However, the clinical significance of GSCs is still controversial and the regulatory molecular mechanisms for GSCs remain elusive.

The LMO2 gene contains two zinc-binding LIM-domains that are essential for LMO2 as a bridging molecule in multiprotein complexes.7 Through binding of the LIM domain to various proteins including TAL/SCL, GATA-1, E47, and LDB1, they are able to regulate gene expression at the transcriptional level by recognizing a unique bipartite DNA sequence comprising an E box separated by about one helix turn from a GATA site.8

Transcriptional dysregulation of LMO2 is frequently observed in human acute T-cell lymphoblastic leukemia patients.9 Lmo2 transgenic activation in the thymus results in T-cell lymphoma/leukemia. Lmo2 overexpression in T-cell progenitors caused differentiation block, exit from quiescence, and increased self-renewal, all of which are the hallmarks of hematopoietic stem cells (HSCs).10 Indeed, Lmo2-expressing T-cell progenitors display an HSC-like transcriptional signature indicating de-differentiation. This reprogramming event caused by Lmo2 may indicate that LMO2 is a driver of cancer initiation in T-cell progenitors.

Despite these intensive studies of LMO2 in leukemia genesis in the past decade, pathophysiology of LMO2 in solid cancers remains largely undetermined. In this study, we sought to elucidate the physiological roles and mechanism of action of LMO2 in GBM and GSCs in mice and human.

Results

LMO2 is required for GSC growth both in vitro and in vivo

First, we investigated the expression of LMO2 in GSC and non-GSC derived from acutely dissociated primary xenografts propagated in vivo sorted for CD133-positive and -negative population. Real-time PCR and western blot analysis demonstrated that LMO2 expression was reduced, whereas an astrocyte differentiation marker, glial fibrillary acidic protein was increased in the non-GSCs in Figure 1a. Human GBM-derived GSCs showed significant decline of LMO2 expression upon induction of differentiation with serum-containing media (Supplementary Figure 1a). Transcriptome microarray data with 11 GCS samples and 5 normal astrocyte samples from Mao data set demonstrated that human GSCs have relatively higher LMO2 mRNA expression compared with differentiated normal astrocytes (Figure 1b). Elevated LMO2 expression in GSCs was also observed in two other data set (Schulte data set and Lee data set; Figure 1c and Supplementary Figure 1b). Among the two GSC subtypes, LMO2 expression was specifically higher in proneural subtype than mesenchymal one in both Mao data set and Bhat data set (Figure 1d). Altogether, LMO2 expression is enriched in GSCs with proneural identity. We then performed LMO2 knockdown by five different shLMO2 lentivirus clones in GSCs and selected the shLMO2 #5 with 90% reduction of LMO2 for the further study (Supplementary Figure 2a and b). In vivo limiting dilution assay shows that LMO2 depletion attenuated sphere-forming activity in GSC1T (Figure 1e and Supplementary Figure 2c). Conversely, LMO2 overexpression exhibited increased sphere-forming activity in the non-GSC, U87MG (Supplementary Figure 2d and e). These in vitro data appeared relevant in vivo, as the orthotopic mouse xenograft assays revealed that tumor growth of human GSCs is significantly diminished by LMO2 knockdown (Figure 1f).

Figure 1
figure 1

LMO2 is necessary for human glioma stem cell growth in vivo (a) mRNA and protein expression of LMO2, CD133, and glial fibrillary acidic protein (GFAP) in GSC and non-GSC acutely isolated from two different GBM-derived xenograft brain tumors based on the CD133 expression. (b and c) Relative LMO2 mRNA expression in indicated samples in Mao’s data set (b) and Schulte’s data set (c). (d) Relative LMO2 expression in proneural and mesenchymal GSC samples in indicated data set. (e) In vitro limiting dilution assays (LDAs) to determine the effect of shLMO2 on self-renewal activity in GSC1T cells. Wells without sphere formation were counted. P=0.009 with extreme LDA (ELDA) analysis; http://bioinf.wehi.edu.acu/software/elda/. (f) Representative H&E staining of whole-mouse brain xenografts derived from Scram- and shLMO2-GSCs (GST1T) and the Kaplan–Meier survival analysis. Scale bar represents 500 μm. Asterisk (*) indicates statistical significance by Student’s t-test. *P<0.05; **P<0.01

Full size image

LMO2 is sufficient to induce glioma stemness from premalignant astrocytes both in vitro and in vivo

Next we investigated the phenotypic consequences of Lmo2 overexpression in primary premalignant astrocytes derived from Ink4a/Arf null mice (hereby termed as Ink4a/Arf−/− astrocytes), which lack p16Ink4a and p19Arf genes—the most frequently altered tumor suppressor genes in human GBM.13 Overexpression of Lmo2 increased in vitro growth of Ink4a/Arf−/− astrocytes (Figure 2a). This increase of cell growth by Lmo2 overexpression was associated with elevated expression of CyclinD1 protein (Figure 2b). Consistent with these in vitro data, Lmo2 overexpression significantly accelerated in vivo growth of subcutaneous and intracranial mouse tumors derived from Ink4a/Arf−/− astrocytes (Figure 2c). Notably, Lmo2- Ink4a/Arf−/− astrocytes-derived tumors have been shown to have histological features of high-grade glioma, such as intratumoral hemorrhage, angiogenesis, and necrosis (Supplementary Figure 3).

Figure 2
figure 2

LMO is sufficient to induce stem cell phenotype in mouse premalignant astrocytes in vivo (a) Cell proliferation of Control-, and Lmo2-overexpressing Ink4a/Arf−/− astrocytes in vitro. Western blot showing the expression level of LMO2 in each sample. β-actin as internal control. (b) Protein expression levels of cell cycle-associated proteins in indicated samples. (c) An in vivo mouse subcutaneous (n=6 per group, left panel) and intracranial (n=4 per group, right upper panel) tumorigenic potential of Control- and Lmo2-Ink4a/Arf−/− astrocytes. Kaplan–Meier analysis of neurological deficit-free survival for the Lmo2 overexpression compared with the control (right lower panel). Scale bar represents 500 μm. (d) In vitro limiting dilution assay of Control-, Lmo2-, and Lmo2/shCyclin D1- Ink4a/Arf−/− astrocytes. (e) Proportion of CD133(+), CD15(+), OCT4(+), TuJ-1(+), and S100β(+) cells of Control- and Lmo2-Ink4a/Arf−/− astrocytes. (f) Representative immunofluorescence images and quantification of Ki-67 in subcutaneous tumors derived from Control- and Lmo2-Ink4aArf−/− astrocytes. (g) Representative immunofluorescence images of Nestin, CD133, and S100β expression in subcutaneous tumors derived from Control- and Lmo2-Ink4aArf−/− astrocytes. DAPI for nuclear staining. Scale bars represent 100 μm. Data are expressed as mean±S.D. **P<0.01

Full size image

Because Lmo2 promotes self-renewal of preleukemic thymocytes,10 we investigated if Lmo2 overexpression leads to stem cell phenotypes in brain cancers. When we performed in vitro limiting dilution assay, Lmo2-overexpressing Ink4a/Arf−/− astrocytes promotes sphere-forming cells that are markedly decreased by Cyclin D1 knockdown, suggesting Cyclin D1 is involved in Lmo2 mediated self-renewal activity (Figure 2d). With regard to the marker expression, Lmo2- overexpressing cells displayed a marked increase of stem cell marker, CD133;14 8% versus 35%, and OCT4; 3% versus 7% of CD133(+) cells in control versus Lmo2-ovexpressing cells as determined by FACS analysis (Figure 2e). In turn, expression of the differentiation markers, TuJ1 and S100β, was strongly decreased in Lmo2-overexpressing cells. Consistent with these in vitro data, immunohistochemistry demonstrated increases of Ki-67, Nestin(+), and CD133(+) cells in Lmo2-overexpressing subcutaneous mouse tumors (Figure 2f and g). On the other hand, these tumors diminished the proportion of S100β(+) cells (Figure 2g). Taken together, Lmo2 promotes tumor propagation by inducing undifferentiated stem cell phenotype of Ink4a/Arf−/− astrocytes.

LMO2 mediates GSC phenotype through transcriptional regulation of the Jagged1-Notch signaling axis

Previously we identified that the key signaling pathways activated in proneural GSCs include Notch-, Shh-, and Wnt-mediated pathways.12 Given that LMO2 is highly expressed in proneural but not mesenchymal GSCs (Figure 1d), we analyzed the effect of LMO2 on these signaling pathways using luciferase reporter vectors containing the specific binding sequences of either CSL (Notch signaling), Gli-1 (Shh signaling), or TCF/LEF (TOP; Wnt/β-catenin signaling). Lmo2 overexpression resulted in a 1.8-fold increase of the CSL/Notch-responsive promoter activity (Figure 3a). Quantitative RT-PCR revealed that one of the Notch ligands, Jagged1, is upregulated in Lmo2-Ink4a/Arf−/− astrocytes (Figure 3b). Jagged1 promoter region has one E-box (CANNTG) DNA motif that is known as a binding site for the LMO2 transcriptional complexes.15 Indeed, Lmo2 increased Jagged1 promoter activity in a dose-dependent manner (Figure 3c), and Jagged1 protein and cleaved Notch1 (Notch intracellular domain-NICD; Figure 3d and Supplementary Figure 4a). Lmo2 also increased Notch downstream target genes Hes1 and Hey1, and these increases were completely attenuated by shRNA-mediated depletion of Jagged1 (Figure 3e and Supplementary Figure 4b). In turn, depletion of LMO2 by shRNA greatly reduced Jagged1 expression in human GSCs (Figure 3f). LMO2 expression in GSCs exhibited statistically significant positive correlation with those of Jagged1, Hes1, Hey1, and Hey2 (Supplementary Figure 5). In addition, a gene set enrichment analysis (GSEA) of transcriptome in GBM patients exhibited a significant correlation between LMO2 and NOTCH target gene signature (Figure 3g). The sphere-forming ability of Lmo2-Ink4a/Arf−/− astrocytes was completely attenuated by Jagged1 depletion (Figure 3h and Supplementary Figure 6). Collectively, these data indicate that LMO2 causally contributes to the reprogramming of differentiated astrocytes into GSC-like tumorigenic cells in a Jagged1/Notch pathway-dependent manner.

Figure 3
figure 3

LMO2 maintains stem cell phenotypes by transcriptional activation of Jagged1-Notch signaling pathway (a) Relative activity of Notch (CLS), Shh (GLI-1), and Wnt (TCF-LEF) responses element luciferase reporters. (b) Relative mRNA levels of Jagged1, Jagged2, Dll1, Dll3, and Notch1. (c) The effect of LMO2 overexpression on the Jagged1 promoter-driven luciferase activity. Rectangle indicates increasing doses of the LMO2 expressing vector. (d) Western blot showing expression levels of Jagged1 and active Notch1 (NICD) proteins in indicated samples. (e) Hes1 and Hey1 mRNA levels of Control-, Lmo2-, and Lmo2-shJagged1-Ink4a/Arf−/− astrocytes. (f) Jagged1 protein levels in 2 GSC samples with or without LMO2 silencing (shLMO2). (g) GSEA analysis showing the high enrichment of Notch target gene signature in LMO2-overexpressng tumors. (h) The numbers of neurosphere forming cells in indicated samples. Data are mean±S.D. *P<0.05, **P<0.01, ##P<0.01

Full size image

LMO2 induces endothelial-like cells from GSCs

Next, we investigated the cellular identities of xenografted human GSCs in mouse brains by using a GFP-labeled GSC sample (GFP+ GSC1T). With a close investigation at the perivascular area in tumors, we identified vWF(+) cells contained either GFP(+) or (−) cells (Figure 4a), raising a possibility that tumor vasculature is composed of both host endothelial cells and GSC-derived cells that utilize vascular mimicry.16 Given that LMO2 is widely expressed in the vasculature of a variety of neoplasm,17 we sought to determine whether the LMO2-driven tumorigenicity is associated with facilitated angiogenesis in an autocrine or paracrine manner. To assess the paracrine effect, we treated human umbilical vein endothelial cells (HUVECs) with the conditioned medium derived from Lmo2-overexpressing cells under the endothelial cell growth-promoting culture condition. This treatment did not exhibit any significant changes in in vitro tube formation, suggesting that Lmo2 may not be involved in angiogenic induction in a paracrine manner (Figure 4b and Supplementary Figure 7a). Next, we directly induced tube formation of Control- and Lmo2-Ink4a/Arf−/− astrocytes using the same culture condition. Lmo2 overexpression markedly increased the tube-like structure in an eightfold in Ink4a/Arf−/ −cells compared with the control (Figure 4b and Supplementary Figure 7b). Furthermore, this phenotypic change by Lmo2 transduction was associated with enhanced immunoreactivity to two endothelial-specific markers, VEGFR2 and VE-cadherin (Figure 4c). Quantitative analysis by FACS showed the increases of VEGFR2, VE-cadherin, and CD34 were greater than sixfold (Figure 4d). Consistent with these mouse data, LMO2 knockdown in human primary GSCs (GSC1T) significantly reduced the number of tubes along with the decreased expression of VE-cadherin and VEGFR2 in vitro (Figure 4e,Supplementary Figure 7c and d). Furthermore, LMO2 knockdown in GFP-expressing human GSC1T brain xenografts resulted in a fivefold decrease in vWF-expressing GFP(+) cells compared with the control tumors (Figure 4f). In turn, Lmo2 overexpression displayed a substantial increase of vWF(+) cells in Lmo2-Ink4a/Arf−/− astrocytes-derived mouse tumors (Figure 4g).

Figure 4
figure 4

LMO2 modulates endothelial-like cell conversion of GSCs in vivo (a) Representative images of vWF and GFP in GFP expressing GSC-derived mouse brain tumors. DAPI for nuclear staining. Arrows indicate double-positive cells for GFP and vWF at perivascular area. Scale bar represents 10 μm. (b) Effects of conditioned media from control- and Lmo2-Ink4a/Arf−/− astrocytes on tube formation of HUVEC cells (left panel). Effect of Lmo2 overexpression on tube formation of Ink4a/Arf−/−astrocytes (right panel). (c) Immunofluorescence showing VE-cadherin and VEGFR2 expressions in indicated samples. Scale bar represents 10 μm. (d) FACS data showing the proportions of VE-cadherin(+), VEGFR2(+), and CD34(+) cells in the control- and Lmo2-Ink4a/Arf−/− astrocytes grown in the proangiogenic media. (e) An in vitro tube forming ability of indicated samples. (f) GFP/vWF double-positive cells in GFP-labeled GSCs (GSC1T) brain xenografts with indicated treatments. Arrow indicates GFP/vWF double-positive cells. Graph in right panel indicates quantification. DAPI for nuclear staining. Scale bar represents 20 μm. (g) Representative images and quantification of the vWF(+) cells in Control- and Lmo2-Ink4a/Arf−/− tumors. Scale bar represents 50 μm. (h) The occupancy of LMO2 to the cognate binding elements in the VE-cadherin promoter in Ink4a/Arf−/− astrocytes. Data are mean±S.D. *P<0.05, **P<0.01

Full size image

We next performed chromatin immunoprecipitation assay to determine whether LMO2 directly occupies the promoter regions of the VE-cadherin gene. Both of the two VE-cadherin promoter regions were significantly amplified by quantitative PCR from chromatin immunoprecipitated Ink4a/Arf−/− astrocyte samples using LMO2 antibody in the proangiogenic condition (EGM2 media; Figure 4h). Collectively, these data indicate that LMO2 binds and activates the promoter regions of the VE-cadherin gene thereby promoting endothelial-like cells from mouse premalignant astrocytes and human GSCs both in vitro and in vivo.

VE-cadherin(+) cells derived from GSCs promotes tumor aggressiveness

We then sought to determine whether the human GSC-derived VE-cadherin(+) cells contribute to tumor growth in vivo (Figure 5). To this end, we transplanted GSCs after infection with a lentiviral vector containing the herpes simplex virus thymidine kinase gene (tk) under the control of transcriptional regulatory elements of the VE-cadherin gene (VE-cadherin-tk) so that the GSC-derived VE-cadherin(+) cells would be selectively eradicated by ganciclovir. The ganciclovir treatment markedly diminished the vWF(+) cell population in VE-cadherin-tk expressing mouse tumors but not in the negative control tumors (Figure 5a). As a positive control (PGK-tk), we validated the cytotoxic effect of ganciclovir in PGK-tk expressing tumors. Consequently, elimination of GSC-derived VE-cadherin-expressing cells resulted in a statistically significant survival benefit for mice bearing GBM-like tumors derived from GSCs (Figure 5b).

Figure 5
figure 5

VE-cadherin(+) cells derived from human GSC1T give rise to endothelial-like cells and has a role in tumor growth in vivo (a) Immunofluorescence displaying vWF-expressing cells (red) in ganciclovir-treated mouse xenograft brain tumors derived from GST1T with indicated lentiviral infection. DAPI for nuclear staining. Scale bar represents 50 μm. (b) Kaplan–Meier curves showing the tumor-free survival of indicated tumor burden mice (n=4 mice/group) after ganciclovir treatment (arrows)

Full size image

Clinical relevance of LMO2 as a therapeutic target in GBM

Last, we examined the clinical relevance of these experimental data by using the glioma tumor samples. In order to do that, we interrogated tissue microarray of 93 clinical glioma samples containing WHO grade II–IV tumors, as well as adjacent normal brain tissues to determine its expression and a potential contribution of LMO2 to the patient survival. Immunohistochemistry displayed significantly higher LMO2 protein expression in high-grade glioma tissues compared with low-grade glioma tissues (Figure 6a and Supplementary Figure 8a). Furthermore, the expression of LMO2 is correlated with poor GBM patient survival (Supplementary Figure 8b). Among the 4 LMO genes, LMO2 was significantly elevated in astrocytoma, oligodendrocytoma, and GBM compared with nontumor tissues as analyzing by the Repository of Molecular Brain Neoplasia Data (REMBRANT) and Grzmil’s data set (Figure 6b and Supplementary Figure 8c). Significant elevation of LMO2 expression in GBM tumors compared with lower grade glioma tissues was also confirmed with two different data set (Phillips and Freije; Figure 6c). When GBM patients were separated based on the prognosis (cut off: 3-year survival), those with shorter survival had significantly elevated LMO2 expression (Figure 6d). This trend was also confirmed by the Kaplan–Meier survival analysis (Figure 6e and Supplementary Figure 7d; REMBRANT survival data). Interestingly, higher LMO2 levels were also strongly correlated with patient clinical characteristics such as age (>40 years), histopathological grade (WHO grade IV), and clinical status of affected patients (<80% Karnofsky performance score; Supplementary Table 1). Collectively, LMO2 is a clinically relevant molecular target in GBM.

Figure 6
figure 6

LMO2 is highly expressed in GBM and its expression is inversely correlated with post-surgical patient prognosis (a) Representative images of LMO2 immuohistochemistry in different WHO tumor grades of clinical glioma samples or normal brain tissues. (b) Heatmap analysis showing that elevated LMO2 mRNA levels in high-grade brain tumors. NT, nontumor; AST, astrocytoma; OLG, oligodendrocytoma; GBM, glioblastoma. (c and d) Relative expression levels of LMO2 in normal brain, astrocytoma, and glioblastoma in indicated patient groups in Phillips data set and Freije data set. (e) Kaplan–Meier survival analysis showing negative correlation of LMO2 mRNA levels with patient survival. Asterisk (*) indicates statistical significance by Student's t-test. **P<0.01

Full size image

Discussion

The significance of this study is related to our identification of a novel function of LMO2 in GBMs and GSCs. LMO2 promotes cell proliferation and the gain of cancer stem cell phenotypes via the Jagged1-Notch signaling axis. LMO2 also facilitates endothelial-like cell conversion of GSCs by the direct transcriptional activation of VE-cadherin. Therefore, the mechanism of LMO2 action may be dependent upon spatial and cellular context. LMO2, in particular under proangiogenic conditions, resulted in endothelial cell marker expression along with subsequent endothelial-like cell conversion of GSCs. In these conditions, LMO2 might comprise a transcriptional complex with TAL-1, E47, LMO2, GATA-2, and LDB1 to regulate VE-cadherin in agreement with a previous study with HUVEC cells.18 In erythroid cells, LMO2 is known to bind to basic-helix-loop-helix transcription factors including TAL1/SCL, LDB1, and GATA-1, thereby forming a large DNA-binding complex to drive the downstream target signals.8 An open question still remains as to how LMO2 coordinates to form a transcrption factor complex in order to mediate the tumorigenic activity of GSCs. Such candidates include Olig2, which is highly expressed in proneural GBM and has a critical role in maintaining stem cell phenotypes in GSCs.19, 20 In support of this hypothesis, several reports have shown a striking correlation between LMO2, TAL1, and Olig2 based on gene expression analysis in human and murine T-cell leukemia.21, 22, 23 In future studies, it will be interesting to explore whether LMO2 interacts with Olig2 to regulate GSCs and brain tumorigenesis.

Presence of tumor-derived vascular components sheds light on the plasticity of cancer cells and the potential involvement of cancer stem cells in cancer-associated angiogenesis. In normal brains, Lmo2 is highly expressed in mouse vascular endothelium and is necessary for angiogenic remodeling of the existing capillary network into mature vasculature.17, 24 GBMs display an enriched vasculature phenotype, which is strongly correlated with tumor aggressiveness and poor patient outcomes. Our studies presented the first evidence that LMO2 modulates endothelial-like cell conversion of GSCs by directly regulating VE-cadherin transcriptional activity. Prolonged survival of mice with the VE-cadherin promoter-driven Hsv tk suicide system further suggested the gain of the endothelial-like cell phenotype in GSCs might regulate tumor aggressiveness. Nonetheless, research on the contribution of GBM cells including GSCs to tumor angiogenesis is still evolving and many questions remain to be addressed. Recently, some studies demonstrated the evidence of transdifferentiation of some GSCs into endothelial cells, whereas others demonstrated that some GSCs are incapable of such transdifferentiation; instead they give rise to vascular stromal cells (e.g. smooth muscle cells, pericytes).16, 25, 26, 27, 28, 29 One possible explanation for this discrepancy is that the lineage specification may be linked to the GBM/GSC subtypes. Given that neural stem cells are known to transdifferentiate into endothelial cells,30 the proneural GSC subtype may have the potential to give rise to endothelial cells rather than stromal cells. To investigate whether mesenchymal GSCs are capable of giving rise to stromal cells is also an interesting question for future studies.

In conclusion, in this study, we identified that LMO2 acts as an oncogenic transcription factor by maintaining the GSC phenotypes and inducing conversion of GSCs into endothelial-like cells. Collectively, our data provide a set of evidence suggestive of a novel signaling mechanism underlying the tumorigenic and angiogenic traits of GSCs to regulate tumor growth in vivo. Future characterization of the LMO2-mediated pathways could elucidate novel molecular mechanisms of contribution of GSCs toward GBM initiation and propagation, and eventually lead to development of novel and effective targeted therapies for this devastating disease.

Materials and Methods

Ethics

Experiments using human tissue-derived materials were carried out under the approved institutional review board at Korea University, the Cleveland Clinic, and Ohio State University (OSU). All animal experimentation was performed with the approval of the OSU Animal Research Committee, following NIH guidelines.

Patient-derived tissue culture

GBM tissue-derived neurosphere cultures (GSC1T, GSC01, GSC13, and GSC83), GBM157, and CW3038 were established and maintained in the Kim, the Nakano, the Lathia laboratory as previously described.11, 12 These GSCs were cultured in Neurobasal medium (NBE; Invitrogen, Carlsbad, CA, USA) supplemented with modified N2, B27, EGF (20 ng/ml; R&D Systems, Minneapolis, MN, USA) and bFGF (20 ng/ml; R&D Systems) in suspension or in an adherent culture with laminin-coated flasks.

Cell lines and reagents

Primary HUVECs were purchased from American Type Culture Collection. Ink4a/Arf−/− astrocytes were maintained in DMEM high glucose medium enriched with 10% fetal bovine serum (Hyclone, Logan, UT, USA). HUVECs and Ink4a/Arf−/− astrocytes were maintained in endothelial cell basal medium-2 with growth supplements (EGM2; Clonetics, San Diego, CA, USA).

In vitro tube formation assay

In vitro tube formation of HUVECs, Ink4a/Arf−/− astrocytes, and GSCs was assessed using an in vitro angiogenesis assay kit (Chemicon, Temecula, CA, USA). Control- and Lmo2-Ink4a/Arf−/−astrocytes were cultured for 24 h (5 × 105 cells/10-cm plate) and then used as a conditioned medium, which had been filtered through a 0.2-μm filter (Sartorius Stedim Biotech, Goettingen, Germany). Primary Ink4a/Arf−/−astrocytes and HUVECs (1 × 104 cells/well in 96-well plate) were cultured in each type of conditioned medium at 37 °C. Six hours after incubation, the cultures were photographed ( × 40 magnification). Three random view-fields per well were examined, and tube numbers were counted.

Subcutaneous and orthotopic implantation assay

To establish subcutaneous xenograft models, cells (2 × 106) were subcutaneously injected into nude mice (BALB/c nu/nu mice). For the orthotopic implantation, 5 × 104 cells were stereotactically injected into the left striatum of nude mice (coordinates: anterior–posterior, +2 mm; medial–lateral, +2 mm; dorsal–ventral from the bregma, –3 mm from the dura). For ganciclovir treatment, mice received ganciclovir at 50 mg/kg intraperitoneally for 5 days. Ganciclovir-treated mice were monitored for survival and collected samples for histology and immunofluorescence.

Statistics

Data were analyzed by two-tailed Student’s t-test. P-values <0.05 and <0.01 were considered statistically significant. The correlations between gene expression levels were evaluated by Spearman’s rank correlation coefficient (ρ).

Other experimental materials and methods are detailed in Supplementary Materials and Methods.