HIF1α regulates single differentiated glioma cell dedifferentiation to stem-like cell phenotypes with high tumorigenic potential under hypoxia

The standard treatment for Glioblastoma multiforme (GBM) is surgical resection and subsequent radiotherapy and chemotherapy. Surgical resection of GBM is typically restricted because of its invasive growth, which results in residual tumor cells including glioma stem cells (GSCs) and differentiated cells. Recurrence has been previously thought to occur as a result of these GSCs, and hypoxic microenvironment maintains the GSCs stemness also plays an important role. Summarizing traditional studies and we find many researchers ignored the influence of hypoxia on differentiated cells. We hypothesized that the residual differentiated cells may be dedifferentiated to GSC-like cells under hypoxia and play a crucial role in the rapid, high-frequency recurrence of GBM. Therefore, isolated CD133–CD15–NESTIN– cells were prepared as single-cell culture and treated with hypoxia. More than 95% of the surviving single differentiated CD133–CD15–NESTIN– cell dedifferentiated into tumorigenic CD133+CD15+NESTIN+ GSCs, and this process was regulated by hypoxia inducible factor-1α. Moreover, the serum also played an important role in this dedifferentiation. These findings challenge the traditional glioma cell heterogeneity model, cell division model and glioma malignancy development model. Our study also highlights the mechanism of GBM recurrence and the importance of anti-hypoxia therapy. In addition to GSCs, residual differentiated tumor cells also substantially contribute to treatment resistance and the rapid, high recurrence of GBM.


INTRODUCTION
Glioblastoma multiforme (GBM) is the most aggressive brain tumor in adults and Standard treatment comprises surgical resection followed by radiotherapy and chemotherapy. However, these strategies have minimal impact on extending the life expectancies of GBM patients [1,2]. To protect brain function, complete surgical resection of GBM is typically impossible because of its invasive growth. This issue may cause residual tumor cells including glioma stem cells (GSCs) and differentiated tumor cells according to cancer stem cells (CSCs) theory [2]. GBM is driven by a series of GSCs with the capacity of self-renewal [3]. However, because the GSC frequency is only 11.4% [4] to 20% [5], thus we speculate that it is unlikely this small population of cells is the only causative factor that contributes to the rapid and high recurrence of GBM. As a result, differentiated cells may also promote relapse and malignant progression in glioma.
Both GSCs and differentiated glioma cells are located in a hypoxic microenvironment [6,7]. For GSCs hypoxia plays an important role in stemness maintenance [8], and their phenotype maintenance, enrichment and tumorigenic capacity are also regulated therein [6,9]. Unfortunately the influence of these niches on differentiated glioma cells is unclear. In 2002, Jögi et al [10] demonstrated hypoxia

Research Paper
Oncotarget 28075 www.impactjournals.com/oncotarget up-regulated GSC markers such as c-Kit and Notch-1 in neuroblastoma both in vitro and in vivo. Using an in vitro approach, Li [9] demonstrated hypoxia induced the "dedifferentiation" of differentiated glioma cells. Based on these reports, we hypothesize glioma stem-like cells may be induced through dedifferentiation under hypoxic conditions. However, studies have traditionally used cell populations (typically hundreds of cells or more) instead of single cell and have cultivated them with stem cell medium. Thus, the exact role that residual differentiated tumor cells play in the resistance and recurrence of GBM remains unclear.
Three basic features of GSCs are neurosphere formation, stemness marker expression in vitro and tumorigenesis in vivo [11]. We performed validation assays that included all the factors. In brain regions, normoxia is close to 3% O 2 [12], and oxygen concentration in glioma becomes more serious [7]; thus, in our study, we used 1% low oxygen to investigate the effects of hypoxia on differentiated tumor cells in vitro, as well as their tumorigenicity in vivo. Because hypoxia inducible factor-1a (HIF1α) is steadily expressed under hypoxic conditions and plays a significant role in angiopoiesis and stemness maintenance [7,13], we also determined whether HIF1α affects the process of "dedifferentiation" under 1% hypoxia.

Hypoxia induced neurospheres formation from single differentiated glioma cell
We first investigated whether hypoxia induced single differentiated CD133 -CD15 -NESTINglioma cell into neurosphere. Under hypoxia more than 60% cells survived at 3 days (d). Most surviving cells (95.38% ± 5.83 from GL261 and 99.10% ± 0.78 from U87) formed neurospheres at 21 d. However, under normoxia only 5.05% ± 1.12 from GL261 and 1.31% ± 1.32 from U87 cells formed sparse, irregular and non-adherent aggregates. Most surviving cells remained a single cell and died at 21 d ( Figure 1B Table 1). Besides, the results of Trypan blue showed almost all the cells in neurospheres kept alive ( Figure 1D). Asymmetric division experiments demonstrated that newly formed neurospheres grew in suspension and were maintained as spheres in stem cell medium; however, adherent growth and morphology was induced with 10% FBS administration ( Figure 1E).
Firstly, we detected and found stem cell markers were highly expressed in neurospheres originating from single CD133 -CD15 -NESTIN -GL261 cell after hypoxia 21 d ( Figure 2A); and primary GBM CD133 -CD15 -NESTINcells exposed to hypoxia 48h also increased the expression of stem cell markers compared with control in normoxia through immunofluorescence ( Figure 2B).To improve the accuracy and set the detection in the same background in immunofluorescence, we did double immunofluorescent labeling of f-actin and stem cell markers for U87 CD133 -CD15 -NESTINcells cultured in hypoxia and normoxia and the results showed there were no difference for the expression of f-actin between hypoxia and normoxia group; however, significant higher expression were demonstrated for SOX-2, OCT-4, KLF-4, Nanog, CD133, CD15, NESTIN and ABCG2 in the cells under hypoxia compared with the expression of stem cell markers of normoxia treated cells (Supplementary Figure 2).
Compared with normoxia controls, RT-PCR showed the expression of stem cell markers increased significantly in a time-dependent manner following hypoxia treatment for 3, 6, 9, 12 and 24 h. After 6 h of hypoxia, a significant up-regulation was identified in U87 cells, and the peak expression was detected at 9-12 h. The expression subsequently slightly decreased at hypoxia 24 h but remained statistically significant compared with control normoxia treated cells ( Figure 3A). Similar results were identified with GL261 and GBM cells (data not shown).
We subsequently used Western blot to examine the expression of these markers in U87 cells exposed to hypoxia for different times (12,24,48 and 72 h). The protein levels of these markers sharply increased in hypoxia treatment cells and ranged from two-to six-fold higher compared with controls ( Figure 3B-3C). For the GL261 and GBM cells, time-dependent increases under hypoxia were also identified for the stem cell markers (data not shown).
We cultured CD133 -CD15 -NESTINglioma cells under hypoxia and detected the percentages of CD133, CD15 and NESTIN positive cells at days 0, 3, 6, 9, 12 and 15 with flow cytometry. The CD133 expression in primary glioma cells was 7.03% at day 0 (normoxia); however, its rate increased to 17.4% after hypoxia for 3 d and 62.9% after 12 d of hypoxia. The CD133 expression subsequently decreased but remained significantly increased compared with the control normoxia-treated cells. Similar results were identified in two other cell lines. The rate of CD15 positive U87 cells increased to 69.3% after 15 d of hypoxia and substantial changes were identified at hypoxia 12 d (59.055%). The CD15 expression was also significantly increased in hypoxia compared with the control in the GL261 and primary glioma cells, which increased from 1.152% to 44  Oncotarget 28077 www.impactjournals.com/oncotarget time-dependent manner. Similar results were identified in other two cell lines ( Figure 3D).

Effect of hypoxia on cell proliferation, cycle and apoptosis
Increased proliferation [9], cell cycle arrest [22] and apoptosis resistance [15] are also stem cell-like characteristics. We used CCK-8 to evaluate cell growth under hypoxia condition, and the results showed hypoxia greatly promoted cell growth compared to that of normoxia after 5d exposure ( Figure 4A). Cell cycle analysis showed hypoxia induced more cells arrested in G 0 /G 1 but a decrease in G 2 /M+S before 5d exposure, and the rate of G 0 /G 1 stage became stable after 5d exposure ( Figure 4B). Annexin-V assay revealed lower apoptosis tendency under hypoxia exposure. Early apoptosis represented cell injury and remarkable difference came after treatment 1d, but  dependent changes of stem cell markers before (con) and after hypoxia in U87 glioma cells. In general, 6 h after hypoxia, there was a significant increase of stem cell markers, which reached peak values at 9-12 h. ( * P < 0.05, One-sample T Test). (B) Western blot analysis indicated a higher expression of stem cell markers after hypoxia for 12-48 h in U87 glioma cells. (C) Gray value analysis of Western blot in B by Quantity One indicated the expression of stem cell markers (SOX-2, OCT-4, KLF-4, Nanog, CD133, CD15, NESTIN and ABCG2) increased at least two-fold compared with control ( * P < 0.05, One-sample T Test). (D) An increase expression of CD133, CD15 and NESTIN with a time-dependent manner after hypoxia ( * P < 0.05, One-sample T Test).

Influence of hypoxia on tumorigenic through xenografts
To investigate the tumorigenic ability of hypoxiainduced GSC-like cells in vivo, 14 d hypoxia (1%)-induced neurosphere-derived cells (Group 1, 10 4 ) or normoxia (21%)-derived CD133 -CD15 -NESTIN -GL261-luc cells (Group 2, 10 4 ) were injected into the brains of adult female C57 mice. These mice were raised under normoxia (21% O 2 ) as indicated in Figure 5A (STRATEGY 1). There were no differences on tumor volume between the two groups on day 5. After 25 d, tumors were detected in group 1 with magnitude orders more than 10 6 . There was fewer tumor formation in the second group and the order of magnitude was even less than 10 4 ( Figure 5B-5C and Supplementary Table 2). Furthermore, tumor samples and H&E staining indicated significant tumor formation in group 1 but not group 2 ( Figure 6A). The mortality of the mice seeded with hypoxia cultured cells was significantly higher compared with control, and the mice without tumor cells raised in 21% O 2 remained alive after 30 d ( Figure 5D).
In the alterative strategy ( Figure 5A), the mice in group 3 (injected GL261-luc cells at 3 × 10 4 ) underwent similar changes as the normoxia group in strategy 1, and group 4 was injected with differentiated CD133 -CD15 -NESTIN -Gl261-luc cells (3 × 10 4 ) but raised under 10% O 2 . At 25 d, tumors were detected in group 4 with magnitude orders greater than 10 6 . However, the order of magnitude decreased to 4,161 in group 3 (Figure 5B-5C and  Supplementary Table 2). Tumor samples and H&E staining demonstrated similar results ( Figure 6A). The mortality of the mice raised in hypoxia was significantly increased compared with group 3. The mice raised in 10% O 2 without tumor cells remained alive after 30 d ( Figure 5D). The expressions of stem cell markers in tumor sample were firstly determined via immunohistochemistry ( Figure 6B); and RT-PCR analysis indicated there were higher expression (> 2-fold) of stem cell markers in the tumor sample collected from mice raised in 10% O 2 ( Figure 7A). Similar results were identified via western-blot ( Figure 7B-7C).

Influence of serum on neurosphere formation
We used different strategies to investigate the influence of serum on neurosphere formation ( Figure 8A). The clone formation rates (spheres/d3 viable cells) were approximately 5% from both GL261 and U87 cells suspended in medium without FBS following hypoxia treatment for 21 d (Figure 8B and Supplementary  Table 3). However, the clone formation rates (spheres/d3 viable cells) of GL261 and U87 suspended with DMEM/ F12+10% FBS were more than 95% ( Figure 8B and Supplementary Table 3). The d3 survival rate of glioma cells without FBS was lower than that with FBS in 1% hypoxia; however, there was no difference between the normoxia and hypoxia treated cells with FBS ( Figure 8C). The mortality of glioma cells suspended without FBS under hypoxia increased in a time-dependent manner and approached to 100% for both GL261 and U87 following 7 d of treatment ( Figure 8D). We used immunofluorescence to examine VEGF and found its expression increased in the U87 cells after hypoxia ( Figure 8E). The VEGF mRNA and protein levels of primary GBM were significantly increased in a time-dependent manner after hypoxia ( Figure 8G-8H). According to the data from Koh et al [23], we analyzed and found VEGF mRNA level sharply increased and was ranked at 16th among the up-regulated mRNAs ( Figure 8F).

Influence of HIF1α on neurosphere formation
High HIF1α expression was identified through immunofluorescence in hypoxia-treated GL261 neurospheres or primary GBM cells after hypoxia for 48 h, which expressed in the cell cytoplasm, not in nuclear as usual ( Figure 9A). RT-PCR indicated the peak HIF1α expression of more than 5.5-fold occurred after 12 h of hypoxia ( Figure 9B). The same trend was demonstrated via Western blot ( Figure 9C). Next, we successfully silenced HIF1α through HIF1α-ShRNA or digoxin in U87 cells ( Figure 9D). To determine the influence of HIF1α on dedifferentiation, we examined neurosphere formation rate of U87 HIF1α-ShRNA CD133 -CD15 -NESTINcells under hypoxia conditions. And we found the neurosphere formation rate of the HIF1α-ShRNA-or digoxin-treated U87 cells was substantially lower than controls. The numerical value was 63.55 ± 3.59% from the single CD133 -CD15 -NESTIN -HIF1α-ShRNA U87 cells and 1.94 ± 0.65% from digoxin treatment ( Figure 9F and Supplementary Table 4). And the proportion of stem cell markers including CD133, CD15 and NESTIN decreased significantly in CD133 -CD15 -NESTIN -HIF1α-ShRNA U87 cells after hypoxia 12d ( Figure 9E). HIF1α-ShRNA had no statistically effect on cell survival rates (d3 surviving cells/d0 seeded cells; d7 surviving cells/d3 surviving cells) compared with normal HIF1α expression group; however, the cell survival rate reduced under digoxin ( Figure 9G). We then detected and found the VEGF expression of HIF1α-ShRNA and digoxin treated U87 cells was lower than controls at both mRNA and protein levels ( Figure 9H-9I).   The quantitative value of bioluminescence was significantly increased in hypoxia compared with normoxia ( * P < 0.001, Paired-samples T Test). (D) Lethality analysis indicated there were significant higher death rates in hypoxia-exposed cell injection than normoxia-exposed cell injection, the same trend was found between the mice with normoxia-exposed cells raised under normoxia or hypoxia (Log-rank Test).

The influence of HIF1α on tumor formation
Oncotarget 28082 www.impactjournals.com/oncotarget Western-blot and RT-PCR were used and we found HIF1α expression of the tumor sample raised in hypoxia was about three to four times higher than control including normal brain tissues and tumor sample obtained from normoxia (21% O 2 ) raised mice (Supplementary Figure 3B-3C). Then we implanted 3 × 10 4 GL261-luc HIF1α-ShRNA cells into C57 mice brain and observed that the tumor size was much smaller than control groups with the same tumor cells seeded. The orders of magnitude all reached to 3 × 10 6 in the control group, however the magnitude orders were less than 1 × 10 6 in GL261 HIF1α-ShRNA cells seeded group (Supplementary Figure 3D-3E). Besides, the GL261 HIF1α-ShRNA cells seeded group showed higher survival rate (Supplementary Figure 3F).

DISCUSSION
Hypoxic microenvironment exists in glioma and this becomes more serious for residual glioma cells after surgery because of vascular damage [24,25]. It is widely accepted that lower oxygen is a major reason for chemoradioresistance through stemness maintenance in glioma patients [8]. For example, 7% oxygen concentration induced higher levels of CD133 and other stem cell markers such as Sox2 and Oct4, and increased the ability of GSCs to form neurospheres compared with an atmospheric oxygen level (20%) [6]. For gliomaderived CD133cells, both Blazek [26] and Mathieu [27] demonstrated hypoxia significantly increased CD133 expression and induced neurosphere formation. However, these studies cultured hundreds of cells in stem cell medium (DMEM/F12+EGF+FGF2+B27); thus, it is not clear whether the enhanced CD133 originated from natural CD133 + cells or differentiated tumor cells. Moreover, the up-regulation of CD133 may be influenced by EGF and FGF2 in stem cell medium, so what's the influence of hypoxia microenvironment on differentiated glioma cells is not clear. To further validate our study, we sorted single CD133 -CD15 -NESTINdifferentiated glioma cells and performed our experiments under 1% hypoxia without growth factors in DMEM/F12 culture medium (conditioned medium).
GSCs are capable of neurosphere formation and clone proliferation [19], which is a vital prerequisite for GSCs. We first examined the neurosphere formation rate of differentiated glioma cells under 1% hypoxia in conditioned medium. When single CD133 -CD15 -NESTINcell was cultured under hypoxia for 21 days, more than 95% of surviving cells were induced to form neurospheres. Even if 20% of the single glioma cells were GSCs as reported [4,5], the other 75% of the newly formed neurospheres identified in our studies should be dedifferentiated from differentiated non-GSCs. Moreover, these CD133 -CD15 -NESTINdifferentiated cells were sorted through MACS three times. Therefore, the GSCs rate should be less than 20%. Another basic feature of GSCs is biomarker expression. CD133-positive cells as GSCs have been studied in many traditional studies [4,5,28]. However, the reliability of CD133 as a stem cell marker remains questionable because CD133cells may be tumorigenic in some cases [29,30]. Although some researchers attributed this to CD133 + cell contamination as a result of missorting [31], we fully considered this issue and sorted with other GSC markers including SSEA-1/ CD15 [16], NESTIN [21,32] and ABCG2 [17], as well as several known dedifferentiation stem cell markers including SOX-2, OCT-4, KLF-4 and Nanog [18,21]. Both induced neurospheres and cells cultured at 1% hypoxia for 48 h expressed all the stem cell markers, which indicated hypoxia induced neurospheres possessed of stem cell features. The most important biological function of GSCs is their tumor formation ability in vivo [3]. Singh et al [5] developed a xenograft assay and demonstrated that brain tumor initiating cells initiated tumors in vivo. We used two different strategies to identify the tumorigenic ability of glioma cells under hypoxia exposure and found glioma cells derived from hypoxia exposure or mice injected with CD133 -CD15 -NESTINcells raised under hypoxia induced significant tumor formation with increased lethality and GSC markers expression. This in vivo study further demonstrated hypoxia dedifferentiated tumor cells into functional GSCs.
To further investigate the stem cell biology of these hypoxia induced cells, we employed asymmetric division [33], cell proliferation [9] and cycle arrest [22] as well as apoptosis analysis [15]. Beier et al [34] reported that GSCs displayed neurosphere-like and asymmetrical divisions. Our results showed these newly-formed neurospheres presented self-renewal and extensive proliferation in serum free medium containing EGF and FGF2 but the differentiation could be induced with serum. Additionally, proliferation analysis with CCK-8 showed hypoxia greatly promoted cell growth after 5d exposure and cell cycle analysis showed hypoxia induced more cells arrested in G 0 /G 1 but a decrease in G 2 /M+S. Finally, Annexin-V assay revealed lower apoptosis tendency under hypoxia exposure. These results further demonstrated the induced neurospheres were stem cells, indicating single CD133 -CD15 -NESTINcell was successfully dedifferentiated into CD133 + CD15 + NESTIN + GSCs by hypoxia and these neurospheres possessed the characteristics of stem cells.
Oncotarget 28087 www.impactjournals.com/oncotarget identified a dynamic interconversion between CSCs and differentiated cells, which indicated the differentiation of CSCs was not the only way in which cells grew and a reversible dedifferentiation process may occur through special signals [30,32,35,36]. These findings were in accordance with our results as we found cancer stem cells can be induced through dedifferentiation under hypoxia conditions. We also challenge the traditional glioma cell heterogeneity model that includes a stochastic model as all tumor cells may develop mutations to promote tumor growth, and a hierarchical model in which only GSCs contribute to tumor growth [37]. In a new theory based on our results, we agree with the stochastic model that each cell is associated with tumor maintenance; however, differentiated glioma cells participate this through a special dedifferentiation way in some microenvironments such as hypoxia. Moreover, differentiated tumor cells are formed by GSCs in specific situations such as serum presence. This viewpoint is in accordance with two recent studies which demonstrated GSC-like cells may be induced by non-GSCs in response to therapeutic stress including TMZ or ionizing radiation treatment [30,36]. Another interesting and significant question is whether we should only target GSCs in the treatment of glioma. We suggest this strategy is ineffective, and instead, we should target both GSCs and differentiated glioma cells simultaneously.
Recent studies demonstrated there were many GSCs in the vascular niche, which are important to maintain the GSC phenotype and promote glioma malignant progression [31,37,38]. Calabrese and colleagues [39] demonstrated compared with GSC mono-culture, the co-culture of glioma and endothelial cells contributed to GSCs proliferation and malignant tumor formation. VEGF expression in the hypoxic microenvironment is important to vessel formation [37,40], which is regulated by hypoxia inducible factor-1α (HIF1α). Studies have shown that HIF1α plays an important role in stemness maintenance, angiogenesis and tumorigenesis [2,9,13,37,41], and we therefore analyzed our results and determined they were in accordance with previous studies. HIF1α was highly expressed in hypoxia and the neurosphere formation rates sharply decreased by approximately 30% when HIF1α was silenced. And then we implanted Gl261 HIF1α-ShRNA cells into mice brain and found the tumor size in this group was much smaller than control with the same tumor cells seeded. We also demonstrated that serum played an important role in the dedifferentiation process because few neurospheres formed without serum. We subsequently detected the VEGF expression in 1% hypoxia and demonstrated that the VEGF expression significantly increased at both mRNA and protein levels. Moreover, the VEGF mRNA and protein levels decreased after HIF1α knockdown. We conclude that HIF1α expresses at first in hypoxia and then promotes the dedifferentiation and vessel formation of glioma cells. These newly formed vessels transport serum and oxygen to the GSCs, which subsequently acquire differentiated features and grow the tumor stepwise. Through the dedifferentiation and differentiation regulated by HIF1α and VEGF, both cell types are dynamic and thus promote glioma growth and increase malignancy. However, another interesting and strange phenomenon why HIF1α expressed at cell cytoplasm not nucleus after hypoxia in this study needs further studies.
Recently, additional studies have demonstrated stemness increased in different microenvironments, and transcription factors such as Sox-2, Oct-4, Klf-4 and Nanog were reprogrammed into both normal tissues including human dermal fibroblasts [42] and Müller glia cells [43], and tumors including lung cancer [44], colon cancer [45] and ovarian cancer [46]. Here, we demonstrated that hypoxia may induce the dedifferentiation of differentiated glioma cells which then acquire the stemness features [9], suggesting that development of therapeutic strategies targeting GSCs should take oxygen tension in full consideration [47].

CD133-CD15-NESTIN-cell isolation and Cell culture
GL261 and U87 cells, as well as primary glioma cells isolated from surgical waste were used as glioma cell lines. Magnetic cell sorting (MACS; Miltenyi Biotech, Bergisch-Gladbach, Germany) was used to isolate CD133 -CD15 -NESTINcells [34]. In brief, primary glioma, GL261 or U87 cells were incubated in DMEM/ F12+10% FBS at 37°C for 3 days, and cell suspensions were prepared following 0.25% trypsin digestion. The cells were subsequently counted and re-suspended in PBS that contained 0.08% EDTA and 0.5% BSA (PBSE; 10 8 cells/500 μl), followed by incubation for 15 min at 4°C with polyclonal rabbit anti-human CD133 + IgGs (Miltenyi Biotech, Germany) or polyclonal rabbit antimouse CD133 + IgGs (Miltenyi Biotech, Germany). After washing with PBS that contained 1% BSA, the cells were re-suspended in PBSE (10 8 cells/300 μl), labeled with goat anti-rabbit IgG MicroBeads (Miltenyi Biotech, Germany), incubated for 15 min at 10°C, washed twice with PBSE and re-suspended in 500 μl PBSE. A cell separation column with a flow resistor was subsequently placed in a miniMACS magnet fixed on the MACS multistand and flushed with 500 μl PBSE. The 500 μl cell suspension was poured into the column reservoir. The CD133 + cells were retained within the magnet, and unlabeled nonmagnetic CD133cells passed through the column and were collected. To increase the purity of the CD133cells, we repeated the previous steps three times. The same methods were used to sort CD15cells from CD133cells and NESTINcells from CD133 -CD15www.impactjournals.com/oncotarget cells. The CD133 -CD15 -NESTINcells were subsequently cultured in DMEM/F12+10% FBS at 37°C to maintain their differentiated status.

Clonogenicity and asymmetric division assay
The CD133 -CD15 -NESTINglioma cells were diluted to 1,500 cells/1 ml DMEM/F12+10% FBS. 1 µl of medium was subsequently transferred to one well of a 96well plate that had previously been pre-coated with 200 μl serum-free DMEM/F12 culture medium ( Figure 1A). Three 96-well plates were incubated at 37°C with 1% O 2 and 5% CO 2 , and other three plates were incubated with 21% O 2 and 5% CO 2 . The numbers of surviving cells and newly formed neurospheres were recorded at 0, 3, 7, 14 and 21 days (d). Then we digested neurospheres to single cell suspension with 0.25% trypsin and added 1ml Trypan blue into the cell suspension and observed whether the colour of cell cytoplasm became blue. For asymmetric division, the neurospheres were subsequently transferred to six 24-well plates and half of the samples were cultured with serum-free medium that contained EGF and FGF2, and the other samples were cultured with DMEM/ F12+10% FBS. All 24-well plates were incubated at 37°C with 21% O 2 and 5% CO 2 and recorded neurospheres morphology at 1, 3 and 5 d.

Protein detection through immunofluorescence
Newly formed neurospheres and differentiated cells after 48 h of hypoxia were fixed with 4% paraformaldehyde for 10 min at 4°C, washed with PBS and blocked with 10% normal serum for 20 min in PBS that contained 0.5% Triton X-100. The cells were incubated for 24 h at 4°C with primary antibodies against SOX-2 Real-time quantitative polymerase chain reaction Cells exposed to hypoxia for 0, 3, 6, 9, 12 and 24 h were collected and used to examine the mRNA expression of stem cell markers with RT-PCR. The melting temperature was 94°C, 5 min; the denaturing temperature was 94°C, 30 s; the annealing temperature was 57°C, 30 s; the extending temperature was 72°C, 30 s; in total, 40 cycles were completed. The primer sequences were in Supplementary Table 5.

Cell growth assay
CCK-8 was used to examine cell growth. Logarithmic phase glioma cells were collected and digested to prepare single cell suspension. The cells were counted and seeded into four 96-well plates (2000 cells/ well). Two plates were cultured with normoxia (21% O 2 ) and others exposed to hypoxia (1% O 2 ). Cell growth was examined on day 1, 3, 5, 7, 9 and 11, respectively. 10µl of CCK-8 solution and 90µl culture medium were mixed and added into each well. The resulting samples were cultured for 2 h respectively under hypoxia or normoxia condition. The absorbance at 450nm was measured with ELISA reader (Varioskan Flash, Thermo Scientific, USA). www.impactjournals.com/oncotarget

Construction lentiviral luciferase reporters and induced the formation of GL261-Luc
We used the pGL3 reporter plasmid as a template. PCR was conducted using FX-KOD HiFi enzyme constructed with Bgl II-Sal I restriction enzyme cutting sites (as subsequently described). The primers were used to amplify the luciferase gene and for sequencing.
upstream 5′-GAAGATCTCACCATGGAAGACG CCAA-3′ downstream 5′-GCGTCGACTTACACGGCGATC TTTCCGC-3′ The sequenced plasmid was cut by the restriction enzymes. The fragment was sub-cloned into the plenti6-V5-D-TOPO plasmid and packaged into virus. The target Gl261 cells were subsequently infected with virus for 2 days and screened using blasticidin; stably infected Gl261-Luc cells were obtained at approximately 2 weeks.

Xenograft analysis
We used two different strategies ( Figure 5A) to investigate the effect of hypoxia-induced GSCs on tumorigenesis in vivo.
Strategy 1: 5 × 10 3 differentiated CD133 -CD15 -NESTIN -GL261-Luc cells were seeded into each well of 6-well plates that had previously been pre-coated with 2 ml serum-free DMEM/F12 culture medium and cultured under hypoxia conditions 14d; then collected and digested neurospheres with 0.25% trypsin to single cell suspension. Counted and injected 10 4 cells into the right frontal lobe of 6-week-old C57 female mice (n = 25) that were raised under normoxia. Fifteen mice were used to examine the survival rate, and the remaining mice were used to investigate tumor growth by the bioluminescence of cancer cells at 5, 10, 15, 20 and 25 d. To determine the amount of bioluminescence, 200 μl (15 mg/1 ml) of luciferins were administered via peritoneal injection to mice under anesthesia, and the bioluminescence was detected using the NightOWL Macro Imaging system (LB983 NC320, Berthold Technologies, Germany). For control group, cells cultured under normoxia (21% O 2 ) were injected into the same brain region of the mice (n = 21) and examined using identical procedures.
All animals were raised for 30 days, and the 30 day survival rate was assessed. To rule out the influence of other factors on mice, 30 additional mice were raised in 21% O 2 or 10% O 2 without tumor cells. Immunohistochemistry, RT-PCR and Western blot were used to examine the stem cell markers and HIF1α expression.

Influence of serum on clone formation
Cell suspensions were diluted to 1,500 cells/1 ml DMEM/F12+10% FBS or 1,500 cells/1 ml DMEM/F12 ( Figure 8A). Single glioma cell sphere formation assay were defined as previously described. All 96-well plates were incubated under 1% O 2 ; the numbers of surviving cells and newly formed neurospheres were recorded at 0, 3, 7, 14 and 21 d. Cell death rate was examined at 1, 3, 5 and 7 d.

HIF1α interfere assays
The HIF1α-ShRNA lentivirus was purchased from Cyagen, Guangzhou, China. CD133 -CD15 -NESTINglioma cells were counted, and 10 5 CD133 -CD15 -NESTINcells were seeded into 6-well plates. HIF1α-ShRNA-lentivirus (MOI=1) was added to 6-well plates following cell attachment for 24 h. 1 μg/ml of puromycin was subsequently added to the 6-well plates following lentiviral infection for 48 h. This process lasted one week. The cells were then collected, and HIF1α expression was detected via Western blot. Successful HIF1α-interference cells were cultured in DMEM/F12+10% FBS to maintain growth. Digoxin (TOCRIS, 4583) was dissolved at 100 mM in DMSO, and the effective concentration was 100 nm/L [24,48]. Successful HIF1α-interference cells were detected via Western blot. To determine the influence of low HIF1α expression on neurosphere formation in hypoxia, we performed single glioma cell sphere formation assays as previously described. Moreover, we also determined VEGF expression of HIF1α interference cells following 1% hypoxia using RT-PCR and Western blot. For in vivo study, we implanted 3 × 10 4 HIF1α-ShRNA CD133 -CD15 -NESTIN -GL261 cells (Group1, n = 30), HIF1α-Vector CD133 -CD15 -NESTIN -GL261 cells (Group 2, n = 30) and CD133 -CD15 -NESTIN -GL261 cells (Group 3, n = 30) respectively into C57 mice brain. Fifteen mice of each group were used to examine survival rate, and the remaining mice were used to investigate tumor growth by the bioluminescence of cancer cells at 20 d as identical procedures above. SPSS 19.0 was used for statistical analysis. Data were expressed as the means ± standard deviations (SDs). Differences were evaluated with T test and one-way analysis of variance (ANOVA) when necessary. A log-rank test was used to assess the survival rate and P < 0.05 was considered statistically significant.