Hypoxia and HIF1α Repress the Differentiative Effects of BMPs in High-Grade Glioma§


  • Author contributions: F.P.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing; H.-L.C.: collection and/or assembly of data; B.R.: administrative support, provision of study material, manuscript writing; H.-Z.Z.: provision of study material; D.D.: provision of study material; L.D.: provision of study material; M.G.: provision of study material; G.t.K.: data analysis and interpretation; P.H.S.: provision of study material, manuscript writing; E.F.: collection and/or assembly of data; S.I.: collection and/or assembly of data, data analysis and interpretation; G.B.: financial support, data analysis and interpretation; D.M.P.: conception and design, financial support, data analysis and interpretation, manuscript writing, final approval of manuscript.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLSExpress October 2, 2008.


Hypoxia commonly occurs in solid tumors of the central nervous system (CNS) and often interferes with therapies designed to stop their growth. We found that pediatric high-grade glioma (HGG)-derived precursors showed greater expansion under lower oxygen tension, typical of solid tumors, than normal CNS precursors. Hypoxia inhibited p53 activation and subsequent astroglial differentiation of HGG precursors. Surprisingly, although HGG precursors generated endogenous bone morphogenetic protein (BMP) signaling that promoted mitotic arrest under high oxygen tension, this signaling was actively repressed by hypoxia. An acute increase in oxygen tension led to Smad activation within 30 minutes, even in the absence of exogenous BMP treatment. Treatment with BMPs further promoted astroglial differentiation or death of HGG precursors under high oxygen tension, but this effect was inhibited under hypoxic conditions. Silencing of hypoxia-inducible factor 1α (HIF1α) led to Smad activation even under hypoxic conditions, indicating that HIF1α is required for BMP repression. Conversely, BMP activation at high oxygen tension led to reciprocal degradation of HIF1α; this BMP-induced degradation was inhibited in low oxygen. These results show a novel, mutually antagonistic interaction of hypoxia-response and neural differentiation signals in HGG proliferation, and suggest differences between normal and HGG precursors that may be exploited for pediatric brain cancer therapy. STEM CELLS2009;27:7–17


Central nervous system (CNS) tumors represent 22% of all malignancies in children up to 14 years of age and 10% of tumors in 15- to 19-year olds; they are one of the leading causes of cancer death in children. Grade III anaplastic astrocytoma and grade IV glioblastoma multiforme make up the high-grade gliomas (HGGs), which are commonly referred to as malignant gliomas to reflect their invasive nature and rapid proliferation [1]. In addition to intrinsic differences traced to the cell of origin, tumor properties are strongly regulated by variations in their microenvironment [2].

One aspect of the microenvironment that differs in normal versus tumor tissue is oxygen tension. Normal oxygen tensions in cortical gray matter are generally in the range of 2.5%–5.3%, with readings as high as 13% [3]. In contrast, oxygen tensions in solid tumors can range from these physiological levels to <0.1% in necrotic regions [4]. Hypoxia actually correlates positively with tumor aggressiveness [5–7]. Although this may be, in part, a consequence of aggressive tumors outstripping their blood supply, there is evidence that hypoxia may play a contributing role in tumor growth. Overactivity of hypoxia inducible factor 1α (HIF1α) and related proteins, which are rapid mediators of hypoxic signaling, is implicated in tumor progression [8–13]. Hypoxia has been causally linked to increased genomic instability [14] and neuroblastoma cell dedifferentiation [7], suggesting that genomic changes in mature cells underlie this phenomenon. Alternatively, hypoxia and HIF1α can repress p53 [15], a major effector of mitotic arrest and apoptosis [16], and HIF1α activity can block normal CNS precursor differentiation [17]. These findings suggest that hypoxia selectively suppresses mitotic arrest or apoptosis in a pre-existing primitive tumor cell population, rather than promoting the dedifferentiation of mature cells. This model would complement evidence that cancer is caused by dysregulated stem cells [18, 19].

During normal brain development, stem cells are regulated by bone morphogenetic proteins (BMPs), which signal through the canonical proteins Smad1, Smad5, and Smad8. BMPs are well-characterized inducers of CNS stem cell differentiation, astroglial fate, mitotic arrest, and apoptosis; in contrast, the endogenously secreted BMP antagonist Noggin limits glial differentiation and redirects normal postnatal stem cells to generate neurons (reviewed in [20]). Consistent with this role, BMPs promote the differentiation of glioma-derived precursors; however, BMP application during tumor cell engraftment slowed but did not stop the growth of brain tumors or eventual animal death [21, 22]. This suggests that the brain microenvironment counteracts the actions of BMPs in ways not present under culture conditions. Indeed, the continued widespread expression of BMPs in the brain after birth [23–25] begs the question of how newly forming tumors escape their differentiating effects.

Here, we investigated whether the low oxygen tensions typically found in the normal brain and tumors act in opposition to prodifferentiation signals such as BMPs. We show a potential mechanism by which hypoxia and its signal transducers can allow tumors to escape the effects of both endogenous and therapeutically administered growth-inhibition factors.


Isolation and Gas-Controlled Expansion of Cells

Approval for the acquisition of human brain tissue was obtained from the institutional review boards of Children's National Medical Center, Children's Hospital of Orange County, and the Italian Association for Pediatric Hematology and Oncology of the University of Padova. Normal precursor cells were derived from brain subventricular zone (SVZ) tissue of three premature neonates who died shortly after birth (HuSC23, HuSC27, and HuSC30, previously characterized [26, 27] and hereafter termed normal SVZ precursors) (supporting information Table 1). HGG precursors were derived from tumors taken at surgery (supporting information Table 1); the initial pathological review was followed by a secondary external neuropathological review (Elisabeth Rushing, Armed Forces Institute of Pathology, Washington, DC; Prof. Felice Giangaspero, Neurosurgery Department, Umberto I Hospital, Rome, Italy) to reconfirm the diagnosis. We dissociated [28] and cultured cells on fibronectin-coated dishes in Dulbecco's modified Eagle's medium/Ham's F-12 (Irvine Scientific, Santa Ana, CA, http://www.irvinesci.com) supplemented with BIT9500 (1% bovine serum albumin, 10 μg/ml insulin, 200 μg/ml transferrin; Stem Cell Technologies, Vancouver, Canada, http://www.stemcell.com), 20 ng/ml basic fibroblast growth factor, and 20 ng/ml epidermal growth factor (both human and from R&D Systems Inc., Minneapolis, http://www.rndsystems.com), as previously described [27], in an atmosphere of 2%, 5%, or 20% oxygen, 5% CO2, and the balance nitrogen. For continuous expansion, one half of the medium was replaced every day and cultures were passaged every 7–10 days using TrypLE (Invitrogen, Carlsbad, CA, http://www.invitrogen.com). In some experiments, cultures were supplemented with BMP2 (10 ng/ml; R&D Systems) or Noggin (200 ng/ml; R&D Systems) for 3–5 days prior to analysis. Cobalt chloride (CoCl2; Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com), which mimics the effect of hypoxia on HIF1α, was used at a concentration of 100 μM and was added to cells after 1 hour of acute exposure to 20% oxygen.


Cells were fixed in cold 4% paraformaldehyde for 15 minutes, rinsed, and stored prior to analysis. Primary antibody staining was performed for Ki67 (mouse, 1:100; Dako, Fort Collins, CO, http://www.dako.com), nestin (mouse, 1:200; Millipore, Billerica, MA, http://www.millipore.com), activated caspase-3 (rabbit, 1:1,000; Cell Signaling Technology, Danvers, MA, http://www.cellsignal.com), glial fibrillary acidic protein (GFAP) (mouse, 1:1,000; Sigma), β-III-tubulin (rabbit, 1:2,000; Covance, Princeton, NJ, http://www.covance.com) or Tuj-1 (mouse, 1:1,000, Covance), p21cip1 (mouse, 1:800; Lab Vision, Fremont, CA, http://www.labvision.com), phospho-p53 Sampler Kit (as directed, Cell Signaling Technology, Beverly, MA, http://www.cellsignal.com), phospho-Smad1, Smad5, and Smad8 (rabbit, as directed, Cell Signaling Technology), or Id1 (rabbit, 1:50; Santa Cruz Biotechnology Inc., Santa Cruz, CA, http://www.scbt.com). After incubation, cells were washed and incubated with species-specific secondary antibodies conjugated to Alexa dyes (Invitrogen). Cells were counterstained with 4′,6-diamidino-2-phenylindole to measure total cell number. Staining was visualized by epifluorescence (BX60 upright microscope, Olympus, Tokyo, http://www.olympus-global.com) and images were compiled for figures using Adobe Illustrator (Adobe Systems Inc., San Jose, CA, http://www.adobe.com).

Western Blot and Densitometric Analysis

Total protein extracts were isolated in lysis buffer as described [29]. Equal amounts of protein (10–20 μg) were resolved using an SDS-PAGE gel and transferred to a polyvinylidene difluoride Hybond-p membrane (GE Healthcare, Milano, Italy, http://www.gehealthcare.com). Membranes were blocked with ECL Advance Blocking (2%; Amersham Biosciences, Piscataway, NJ, http://www.amersham.com) overnight under rotation at 4°C. Membranes were then incubated with primary antibodies against HIF1α (mouse, 1:250, BD Biosciences, San Diego, http://www.bdbiosciences.com), Id1 (rabbit, 1:1,000; Santa Cruz Biotechnology), phospho-Smad1, Smad5, and Smad8 (rabbit, 1:1,000; Cell Signaling), BMP2 (mouse, 1:100; R&D Systems Inc.), prolyl hydroxylase domain (PHD) proteins PHD1, PHD2, and PHD3 (goat, 1:300; Santa Cruz Biotechnology), or β-actin (mouse, 1:10,000; Sigma) for 2 hours. Membranes were next incubated with peroxidase-labeled goat anti-rabbit IgG (1:100,000; Sigma), peroxidase-labeled goat anti-mouse IgG (1:100,000; Sigma), or peroxidase-labeled donkey anti-goat IgG (1:100,000; Santa Cruz Biotechnology) for 60 minutes. All membranes were visualized using ECL Advance (GE Healthcare) and exposed to Hyperfilm MP (GE Healthcare). Densitometric analysis of the film was performed using Scion Image densitometer software (Scion Corporation, Frederick, MD, http://www.scioncorp.com).

Flow Cytometry

Cultured cells were passaged and resuspended in flow cytometry buffer [28]. Cells (4 × 106 cells/ml) were incubated with anti-human CD133 (clone AC133/2-PE, as directed, Miltenyi Biotec, Bergisch Gladbach, Germany, http://www.miltenyibiotec.com) at 4°C for 30 minutes, washed in buffer, and resuspended in 400 μl buffer. Viability was assessed by adding 7-amino-actinomycin-D (50 ng/ml; BD Biosciences) prior to analysis. Cells were analyzed on a Cytomics FC 500 flow cytometer (Beckman Coulter, Fullerton, CA, http://www.beckmancoulter.com). For the cell cycle analysis, 1 × 106 cells were resuspended in 1× phosphate-buffered saline (PBS) and fixed in cold ethanol, 70%, then washed in 1× PBS and stained with propidium iodide as directed (Beckman Coulter protocol) prior to analysis. Fluorescence intensities were point-plotted on two-axis graphs or histograms using Expo32 software (Beckman Coulter).

Real-Time Polymerase Chain Reaction Analysis

RNA was isolated from cells using Trizol (Invitrogen), and 1 μg of total RNA was reverse-transcribed using SuperScript RNAse H-Reverse Transcriptase (Invitrogen). Real-time quantitative polymerase chain reactions (RQ-PCRs) were run in duplicate using the Brilliant® SYBR® Green QPCR Core Reagent Kit (Stratagene, La Jolla, CA, http://www.stratagene.com). Fluorescent emission was recorded in real time (Sequence Detection System 7900HT; Applied Biosystems, Foster City, CA, http://www.appliedbiosystems.com). Gene expression profiling was completed using the comparative Ct method of relative quantification. Relative RNA quantities were normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) or β-glucuronidase (GUSB) (giving similar results) as a control, and 2% oxygen, untreated, was used as the calibrating condition. PCR amplification conditions consisted of 35 cycles with primers annealing at 56°C. The specificity of primers (supporting information Table 2) was confirmed for every PCR run by dissociation curve analysis.

Transduction of HGG Cells Using Lentiviral Vectors

The lentiviral plasmids containing HIF1α small interfering (si)RNA and Luciferase (LUC) siRNA target sequences, termed pLSLG-HIF1α-siRNA and pLSLG-Luciferase-siRNA, respectively, were a kind gift of Dr. O.V. Razorenova (Department of Molecular Cardiology, Lerner Research Institute, Cleveland) [30]. The lentiviral vectors were produced as previously described [31] and used to infect HGG cells. Target HGG cells were incubated with stocks for at least 12 hours. Transduced cells were cultured for 5 days and analyzed by immunocytochemistry and Western blot.

Statistical Analysis

Graphs and statistical analyses were prepared using Prism 3.03 (GraphPad Software Inc., La Jolla, CA, http://www.graphpad.com). All values are presented as the mean ± standard error of the mean (SEM). Statistical significance was measured by simple paired t-tests or one-way analysis of variance with post hoc Newman-Keul's test, with *, p < .05, **, p < .01, and ***, p < .001. For all graphs, an asterisk directly above a bar indicates a significant difference with its 2% oxygen counterpart; an asterisk over a bracket indicates a significant difference with another variable as indicated.


HGG Precursors Require a Lower Oxygen Tension than Normal Precursors for Optimal Expansion

Our previous work demonstrated that normal human postnatal SVZ-derived precursor cells expanded more extensively over longer times under a physiological (5%) oxygen tension than under the 20% oxygen tension typically used experimentally [27]. Using three normal SVZ cell isolates (supporting information Table 1), we found that lowering the oxygen tension to 2%, below reported values for normal cortical gray matter, did not further increase the expansion of normal SVZ precursors (Fig. 1A–1D). Since the oxygen tension in brain tumors is often lower [4] than the 2.5%–5.3% range measured in normal gray matter [3], we examined whether the expansion of HGG precursors was differentially sensitive to oxygen tension, as compared with normal precursors. Brain tumors (supporting information Table 1) were collected at surgery, dissociated, and expanded in 2%, 5%, or 20% oxygen. Despite some variability in growth rates among tumors, all responded qualitatively the same; the expansion of both pediatric and adult HGG precursors increased as oxygen tension decreased. However, only in 2% oxygen was it significantly higher than in 20% oxygen (Fig. 1E–1H). This indicates that HGG precursors require lower oxygen tensions for maximal in vitro expansion than normal precursors.

Figure 1.

HGG precursors (pediatric and adult) maximally expand at a lower oxygen tension than normal SVZ precursors. (AD) Normal SVZ precursors plated at medium density (49 cells/mm2) and expanded for 5 days in 2% (A), 5% (B), or 20% (C) oxygen. (D): Histogram showing mean ± SEM, n = 12 (SC23, n = 3; SC27, n = 3; SC30, n = 6). (EH): HGG precursors, plated at medium density (49 cells/mm2) and expanded 5 days in 2% (E), 5% (F), or 20% (G) oxygen. (H): Bar graph showing mean ± SEM, n = 7, using five different tumors. *, p < .05; **, p < .01; ***, p < .001. Bars = 50 μm. Abbreviations: HGG, high-grade glioma; SEM, standard error of the mean; SVZ, subventricular zone.

Acute Increases in Oxygen Tension Activate a Differentiative Response in HGG Precursors

Normal SVZ precursors show limited expansion in 20% compared with 5% oxygen. This involves the rapid activation of p53 and the induction of p21cip1, an effector of mitotic arrest, within 48 hours [27]. We performed an analogous assay on pediatric and adult HGG precursors by initially expanding them in 2% oxygen, to account for the lower oxygen tension required for enhanced expansion, but then acutely exposing them to 20% oxygen for 24 and 48 hours. This acute exposure led to an increase in total p53 expression within 24 hours, which decreased at 48 hours (Fig. 2A–2C). Examination of several phosphorylation sites in p53 revealed a strong activation of p53 at Ser37 with additional phosphorylation at Ser20. This phosphorylation was seen most strongly within 24 hours and persisted at a lower level at 48 hours (Fig. 2C). This is consistent with p53-mediated mitotic arrest and/or apoptosis in response to an acute increase in oxygen tension. In support of this mechanism, we also found a decrease in proliferating cells as determined by Ki67 staining (Fig. 2D–2F) and mitotically arrested cells as determined by p21cip1 staining (Fig. 2G-2I). However, no significant change was seen in apoptotic cells, as determined by proteolytically cleaved caspase-3 (Fig. 2D–2F).

Figure 2.

Increasing oxygen tension induces p53 activation, leading to glial differentiation of HGG precursors. HGG precursors (pediatric and adult) were initially expanded in 2% oxygen, followed by acute exposure to 20% oxygen. (AC): Expression of p53 using a pan-p53 antibody (green) in 2% oxygen (A) or after 24–48 hours of acute exposure to 20% oxygen (B). (C): Bar graph comparing total p53 expression with activation of p53 via phosphorylation at serine residues 20, 37, or 392. (DF): Expression of Ki67 (red) and activated caspase-3 (green) in 2% oxygen (D) or 48 hours after acute exposure to 20% oxygen (E). (F): Quantitation. (GI): Expression of p21cip1 (red) in 2% oxygen (G) or 48 hours after acute exposure to 20% oxygen (H); inset shows higher magnification of p21/DAPI colocalization. (I): Quantitation. (JL): Expression of nestin (green) and GFAP (red) in 2% oxygen (J) or 48 hours after acute exposure to 20% oxygen (K). (L): Quantitation. Bar graphs show mean ± SEM, n = 3–7. *, p < .05. Bar = 8 μm for (A, B) and inset in (H). Bar = 50 μm for (D, E, G, H, J, K). Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; GFAP, glial fibrillary acidic protein; HGG, high-grade glioma; SEM, standard error of the mean.

We found that increased oxygen caused an increase in the number of differentiated cells expressing GFAP, a marker of astroglia and their progenitors [32, 33], and β-III-tubulin, a marker of neurons and their committed progenitors [34] (Fig. 2J–2L). A concomitant decrease in nestin staining, indicative of undifferentiated precursors, was also seen (Fig. 2J–2L). O4, a marker of immature oligodendrocytes [35], was seen in few cells under either condition (not shown). Although the percentages of each cell type varied according to tumor, increases in oxygen tension always promoted the differentiation of HGG precursors with this short 48-hour period.

Hypoxia Represses Both Exogenously and Endogenously Generated BMP Activity

BMPs are important regulators of mitotic arrest and differentiation of CNS precursors [20]. We previously found that lower oxygen represses BMP signaling via Smad phosphorylation in normal SVZ precursors [27]. Here, we investigated whether oxygen tension regulated the response of HGG precursors to BMPs, including Smad activation and transcriptional induction of target genes such as Id1 [36], which is involved in gliogenesis [37–39]. We started with HGG precursors that had been cultured previously in 2% oxygen and switched to 20% oxygen upon beginning BMP2 treatment. By measuring BMP-stimulated serine phosphorylation of Smad1, Smad5, and Smad8, a key activation step in BMP signal transduction [40], we found that activation occurred more rapidly, in more cells, and for a longer duration in 20% oxygen than in 2% oxygen (Fig. 3A, 3B). We compared BMP4 and BMP7 treatment and found that Smad1, Smad5, and Smad8 activation was modest compared with BMP2 (not shown). Treatment with BMP2 in 2% oxygen had little effect on Id1 induction, whereas in 20% oxygen it led to strong Id1 induction after 24 hours (Fig. 3C, 3D).

Figure 3.

Lower oxygen inhibits BMP signaling from both exogenous and endogenous sources. (AD): Pediatric HGG precursors were initially expanded in 2% oxygen, then treated with 10 ng/ml BMP2 in 2% oxygen or after switching to 20% oxygen (2→20%). Immunofluorescent staining for phospho-Smad1, -Smad5, and -Smad8 (A) and Id1 (C) in untreated and BMP2-treated cells, along with histogram showing percentage of cells with nuclear-activated Smad1, Smad5, and Smad8 (B) or Id1 expression (D) at varying times after BMP2 treatment in 2% versus 20% oxygen. Values are expressed as mean ± SEM, n = 3 from two different tumors. Bar = 8 μm. (EF): HGG precursors were initially expanded in 2% oxygen, followed by acute exposure to 20% oxygen for 30, 60, or 120 minutes without BMP treatment. (E): Western blot showing Smad1, Smad5, and Smad8 phosphorylation and Id1 expression, along with β-actin as a loading control. (F): Histogram showing mean densitometric intensity of phospho-Smad1, -Smad5, and -Smad8 ± SEM, n = 3 from two tumors. (G): Histogram showing mean densitometric intensity of Id1 ± SEM, n = 3 from two tumors. (HJ): Measurements of endogenous BMP2 in HGG precursors in 2% oxygen or after 72 hours of acute exposure to 20% oxygen. (H): Western blot with quantitation (I), n = 3 from two tumors. (J): BMP2 RQ-PCR quantitation, n = 4 from two tumors. *, p < .05; **, p < .01; ***, p < .001. Abbreviations: BMP, bone morphogenetic protein; HGG, high-grade glioma; RQ-PCR, real-time quantitative polymerase chain reaction; SEM, standard error of the mean.

Surprisingly, we found that acutely increasing oxygen tension from 2% to 20% led to a rapid activation of Smad1, Smad5, and Smad8 even in the absence of added BMPs (Fig. 3E, 3F). Id1 expression in HGG precursors was also induced simply by increasing oxygen tension (Fig. 3D, 3E, 3G), whereas in normal precursors little induction occurred (not shown) except after BMP treatment [27]. One possibility for this rapid response would be the presence of endogenously secreted BMPs, which could immediately activate Smads under permissive conditions. Indeed, Western blot analysis demonstrated the presence of immature unprocessed BMP2 with small amounts of mature BMP2 in HGG precursors expanded in 2% oxygen. An acute increase to 20% oxygen for 72 hours led to a further increase in the active (mature) form of BMP2 (Fig. 3H, 3I). Since the medium was aspirated prior to cell lysis, it is possible that additional mature BMP was secreted into the medium even in 2% oxygen. BMP2 mRNA was also present (Fig. 3J). RQ-PCR analyses of BMPR-IA, BMPR-IB, and BMPR-II (supporting information Fig. 1) mRNA showed that each of these genes was upregulated in HGG precursors after 72 hours of exposure to 20% oxygen, suggesting that low oxygen tension also exerts long-term BMP pathway regulation. Together, these results indicate that HGG precursors are a source of their own endogenous BMP signaling, which is actively repressed by hypoxia in vitro.

BMP2 Promotes a Stronger Antimitotic Effect on HGG Precursors than Normal Precursors, but This Effect Is Repressed by Hypoxia

To determine how oxygen regulates cell fate in response to BMP signaling, we treated both normal and HGG precursors for 48 hours with 10 ng/ml BMP2 or 200 ng/ml Noggin, an endogenous competitive inhibitor of BMP ligand-receptor binding [41]. These treatments were performed in either 2% oxygen (Fig. 4A–4C, 4G) or 20% oxygen (Fig. 4D–4G). Whereas pediatric and adult HGG precursors responded similarly to changes in oxygen tension, we found differences in their responses to BMPs (described below). Treatment with BMP2 promoted a greater reduction in the pediatric HGG precursor number at 20% oxygen than at 2% oxygen; this reduction was greater than that seen in normal SVZ precursors (Fig. 4G, 4H). Interestingly, Noggin had no effect on HGG precursor numbers under either oxygen tension, but increased the number of normal SVZ cells in 2% oxygen (but not 20% oxygen) (Fig. 4C, 4F–4H); this may be due to higher secretion of BMPs from HGG precursors than from normal precursors. Interestingly, from one tumor (HuTu14), we saw a rapid loss of HGG precursors in response to BMP2 treatment, suggesting apoptosis (supporting information Fig. 2). To measure proliferation, we performed cell cycle analysis using flow cytometry and the fluorescent DNA intercalating dye propidium iodide to measure DNA content. We found that BMP2 treatment increased the proportion of HGG precursors in G0-G1 with a corresponding decrease in the S phase, indicating a lower rate of proliferation. Furthermore, this antiproliferative response was stronger in 20% oxygen (Fig. 4I). BMP2 treatment of normal SVZ precursors led to no change in proliferation at 2% oxygen, but slightly lower proliferation in 20% oxygen (Fig. 4J), consistent with a permissive effect of higher oxygen on BMP signaling. Interestingly, the antiproliferative effect of BMP2 on pediatric HGG precursors was stronger in 2% oxygen than on normal SVZ precursors in 20% oxygen. These results indicate that low oxygen attenuates the antimitotic effect mediated by BMPs to a greater extent in HGG than in normal SVZ precursors.

Figure 4.

Lower oxygen inhibits the antiproliferative effect of BMP2 on pediatric HGG precursors. (AF): HGG precursors were initially expanded in 2% oxygen, passaged, and replated at medium density (49 cells/mm2), and expanded for 3 days at either 2% or 20% oxygen in the presence of bFGF and EGF alone (A, D) or with BMP2 (10 ng/ml; R&D Systems Inc., Minneapolis, http://www.rndsystems.com) (B, E) or Noggin (200 ng/ml; R&D Systems) (C, F). (G, H): Total number of pediatric HGG (G) or normal (H) SVZ precursors after expansion in 2% or 20% oxygen with bFGF and EGF alone (control) or with the addition of BMP2 or Noggin; mean ± SEM, n = 3–8. (I, J): Cell cycle analysis of precursors from pediatric HGG (I) and normal SVZ precursors (J), showing change in percentage in each cell cycle phase as a result of BMP2 treatment (compared with 2% oxygen control) in 2% versus 20% oxygen, n = 2 for both cell types. *, p < .05; **, p < .01; ***, p < .001. Bar = 50 μm. Abbreviations: bFGF, basic fibroblast growth factor; BMP, bone morphogenetic protein; EGF, epidermal growth factor; HGG, high-grade glioma; SEM, standard error of the mean; SVZ, subventricular zone.

Increased Oxygen Accelerates BMP2-Induced Astroglial Differentiation

Since BMPs are known to promote astroglial differentiation, we measured whether this response was regulated by oxygen tension after a longer (72-hour) period of treatment. Whereas pediatric HGG precursors in 2% oxygen consisted predominantly of tripolar nestin+ cells with smaller numbers of nestin+GFAP+ and nestinGFAP+ cells (Fig. 5A), BMP2 treatment yielded an increased frequency and intensity of GFAP+ cells (Fig. 5B), which also assumed a characteristic pleiomorphic, flattened, phase-dark morphology. This effect was amplified in 20% oxygen (Fig. 5C, 5D). Nestin+ cells were reduced by BMP2 treatment in 20% oxygen more strongly than in 2% oxygen (Fig. 5E) and the number of GFAP+ cells was twofold higher in 20% oxygen than in 2% oxygen (Fig. 5F). The inclusion of faint-staining cells as GFAP+ likely overestimated the abundance of differentiated astrocytes in untreated cultures in 2% oxygen, since the faint GFAP staining and morphology of these cells is also consistent with radial glia, another multipotent cell type [19, 42]. In contrast, BMP promoted astroglial differentiation of normal SVZ precursors only in 20% oxygen (not shown), consistent with our previous observations [27]. We also found a slight increase in the percentage of β-III-tubulin+ cells after BMP2 treatment in both oxygen tensions, whereas the percentage of this subpopulation was comparable to the control in the presence of Noggin (not shown), even though it has been reported that Noggin modulates neurogenesis in the normal adult SVZ [25]. These results indicate that, in normal SVZ precursors, gliogenic responsiveness to BMP2 is repressed at lower oxygen tension, whereas, in HGG precursors, this effect is present under both oxygen tensions, but it is partially inhibited at 2% oxygen.

Figure 5.

Lower oxygen inhibits HGG stem cell differentiation caused by BMP2. Pediatric HGG precursors were initially expanded in 2% oxygen, then treated with 10 ng/ml BMP2 in 2% oxygen or after switching to 20% oxygen (2→20%). (AD): Fluorescent staining for nestin (green) and GFAP (red) in untreated (A, C) and BMP2-treated (B, D) cells in 2% (A, B) or 20% (C, D) oxygen; note faint staining and simple morphology of GFAP+ cells in nontreated cells compared with BMP2-treated cells. Bar graphs show mean ± SEM, n = 3–4, for nestin (E) and GFAP (F). (G, H): Flow cytometric analysis of live cells after labeling with antibody to the stem cell surface marker CD133. Histograms show mean percentage ± SEM of CD133+ and CD133 cells in pediatric HGG (G), n = 3, or normal SVZ precursors (H), n = 9, from three lines. *, p < .05; **, p < .01; ***, p < .001. Abbreviations: BMP, bone morphogenetic protein; DAPI, 4′,6-diamidino-2-phenylindole; GFAP, glial fibrillary acidic protein; HGG, high-grade glioma; SEM, standard error of the mean; SVZ, subventricular zone.

We also performed an analysis of the cell surface antigen CD133, which marks human CNS stem cells [43]. Glioblastoma, medulloblastoma, and ependymoma cells expressing high levels of CD133 show higher tumor reinitiating capacity than do CD133 cells isolated from the same tumor [18, 19], operationally defining them as cancer stem cells. We analyzed HGG precursors for expression and intensity of this marker by flow cytometry. A pediatric glioblastoma contained a high proportion of both nestin+ and CD133+ cells (Fig. 5G), higher than that seen in normal SVZ precursors from three lines grown in low oxygen (Fig. 5H). Treatment of pediatric HGG precursors with BMP2 led to a decrease in CD133+ cells and a corresponding increase in CD133 cells, which correlated with the increase in GFAP+ astrocytes (Fig. 5H). This response was amplified in 20% oxygen compared with 2% oxygen. In contrast, although we had previously shown that longer-term exposure to 20% oxygen depleted the normal SVZ pool of CD133+ cells [27], the 72-hour exposure was insufficient to cause this depletion. Thus, high oxygen tension potentiates the BMP-induced differentiation of HGG stem cells, which show greater sensitivity to depletion relative to normal SVZ stem cells.

When we tested adult HGG precursors, we found that even in 2% oxygen they had greater glial characteristics than pediatric HGG precursors, as measured by morphology and GFAP expression (supporting information Fig. 3A–3F). Although there was no difference in the effect of oxygen alone in the expansion of adult versus pediatric HGG precursors, BMP2 treatment had less effect on total cell number, Ki67 staining, or cell cycle distribution (supporting information Fig. 3G–3I). However, treatment with BMP2 did further reduce the proportion of CD133+ cells in both 2% and 20% oxygen. The lower proportion of CD133+ adult HGG precursors (supporting information Fig. 3J), compared with pediatric HGG precursors (Fig. 5G), suggests that part of the diminished response of adult HGG precursors to BMPs may be due to their more strongly glial phenotype.

BMP2 Treatment Downregulates HIF1α Expression in HGG Precursors

The best-described mediator of oxygen response is HIF1α [44]. HIF1α and its downstream target genes, such as VEGF, GLUT-1, and CAIX, are overexpressed in high-grade gliomas, which supports a role for hypoxia-response signals in tumor aggressiveness [45]. HIF1α is targeted for degradation by PHD1, PHD2, and PHD3, which utilize oxygen as a cofactor for enzymatic activity; high levels of oxygen lead to higher levels of HIF1α hydroxylation and degradation [46]. We found that HIF1α protein expression was reduced by increased oxygen in HGG precursors (Fig. 6A, 6C) and also in normal SVZ precursors (Fig. 6A, 6E). The decrease, especially in HGG precursors, was more modest than that reported after more extreme variations in oxygen tension [47] and may also reflect the daily addition of mitogens, which may independently induce HIF1α [10, 48–50] during the 72-hour period of the experiment, and/or a spontaneous reconstitution mechanism of the HIF1α protein level. All three PHD proteins were expressed in 2% oxygen, but PHD3 was expressed at the highest levels (Fig. 6B). Notably, in HGG precursors cultured under hypoxic conditions, 72-hour treatment with BMP2 also caused a similar decrease in HIF1α and an increase in PHD1, PHD2, and PHD3. The combination of increased oxygen and BMP2 further downregulated HIF1α (Fig. 6A–6C). In contrast, BMP2 treatment did not have any additive effect on HIF1α downregulation by increased oxygen in normal SVZ cells, suggesting that the downregulation of HIF1α by BMPs was specific to primary HGG precursors. Analysis of RNA expression in HGG precursors showed little change in HIF1α expression except after BMP2 treatment in 20% oxygen (Fig. 6D); this agrees with the literature showing that HIF1α is regulated principally at the post-translational level [44]. These results indicate that BMP stimulation reciprocally downregulates low oxygen-response signaling, suggesting a mutual antagonism between signals that promote proliferation and those that promote differentiation. This mutual antagonism appears to be specific for primary HGG precursors, being absent in normal SVZ cells.

Figure 6.

BMP2 treatment downregulates HIF1α expression in HGG precursors. (A): Example of Western blot of HIF1α using total protein extracts from HGG precursors (HuTuP01) or normal SVZ precursors, collected after 3–5 days culture under the following conditions: lane 1, untreated cells in 2% oxygen; lane 2, untreated cells in 20% oxygen; lane 3, treated with 10 ng/ml BMP2 in 2% oxygen; lane 4, treated with 10 ng/ml BMP2 in 20% oxygen.(B): Analysis of PHD1, PHD2, and PHD3 expression in HGG precursors using same treatment paradigm. (C): Bar graph showing mean intensity ± SEM (n = 5) of HIF1α protein expression from two different HGGs, normalized to β-actin expression to control for protein loading. (D): Bar graph showing corresponding HIF1α mRNA levels by RQ-PCR (n = 5). (E): Bar graph showing mean intensity ± SEM (n = 3) of HIF1α protein expression from normal SVZ precursors, normalized to β-actin expression to control for protein loading. *, p < .05; **, p < .01. Abbreviations: BMP, bone morphogenetic protein; HGG, high-grade glioma; HIF, hypoxia-inducible factor; PHD, prolyl hydroxylase domain; RQ-PCR, real-time quantitative polymerase chain reaction; SEM, standard error of the mean; SVZ, subventricular zone.

HIF1α Is Required to Repress Smad Activation in Hypoxic Tumor Cells

To understand if HIF1α mediates the repressive effect of low oxygen on BMP signaling, we silenced HIF1α expression using a lentiviral vector containing siHIF1α along with enhanced green fluorescent protein (siHIF1α-EGFP) to indicate the efficiency of transduction. We also used a siLuciferase-EGFP vector (siLUC-EGFP) as a negative control to test for nonspecific effects [30]. Transduction of adult HGG precursors with siHIF1α-EGFP promoted strong differentiation and eventually cell death after 1 week (Fig. 7B); these effects were not observed with siLUC-EGFP (Fig. 7A).

Figure 7.

Silencing of HIF1α promotes Smad activation in HGG precursors. (A, B): Cells were transduced using a lentiviral vector containing siLuciferase along with enhanced green fluorescent protein (siLUC-EGFP) (A) to test for nonspecific effects and efficiency of transduction, whereas siHIF1α-EGFP was transduced to test the effects of HIF1α silencing (B). (C): Western blot analysis of HIF1α protein level, along with β-actin as a loading control. (D): RQ-PCR analysis of HIF1α, normalized to β-glucuronidase (GUSB) and then normalized to siLUC-EGFP negative control (ΔΔCt method). (E, F): Immunofluorescent analysis of activated Smad1, Smad5, and Smad8 in HGG precursors that were transduced with siLUC-EGFP or siHIF1α-EGFP. (G): Bar graph showing total activated Smad as a percentage of total DAPI+ cells in transduced HGG cells compared with the untransduced 2% oxygen control group. (H): Activated Smad+/EGFP+ HGG precursors as a percentage of total DAPI+ cells. (I): Western blot analysis of HIF1α protein level, activated Smad and Id1, along with β-actin as a loading control in HGG precursors grown in 2% oxygen and acutely exposed for 1 hour to 20% oxygen, then treated with CoCl2 (100 μM; Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com), a HIF1α inducer, following a time course, compared with untreated HGG precursors acutely exposed to only 20% oxygen (J). *, p < .05. Bars = 50 μm. Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; HGG, high-grade glioma; HIF, hypoxia-inducible factor; RQ-PCR, real-time quantitative polymerase chain reaction; SEM, standard error of the mean; SVZ, subventricular zone.

This indicates a requirement for HIF1α in preserving the proliferation and survival of undifferentiated HGG precursors. Western blot analysis showed a reduction in the HIF1α protein level (Fig. 7C). RQ-PCR analysis (Fig. 7D) showed only a modest reduction in the HIF1α mRNA level (20% less than with siLUC/EGFP), consistent with previous siRNA studies [51]. By measuring Smad1, Smad5, and Smad8 phosphorylation as an indicator of Smad activation, silencing of HIF1α in primary HGG cells revealed a stronger activation of Smads, even at 2% oxygen (Fig. 7E–7H), compared with the negative control and with untransduced cells. Total Smad1, Smad5, and Smad8 protein expression was not affected by HIF1α silencing (not shown), indicating that HIF1α exerted its actions at the level of Smad activity.

We also performed a time course incubation on HGG precursors initially grown in 2% oxygen, then treated with the hypoxia-simulating drug CoCl2 (100 μM, Sigma), which stabilizes HIF1α [52], after 1 hour of 20% oxygen acute exposure (Fig. 7I). We found that stabilized HIF1α inhibited Smad activation even at the earliest time points and promoted a transient Id1 reduction, when compared with untreated HGG precursors acutely exposed to 20% oxygen. We compared these results with those from HGG precursors acutely exposed to 20% oxygen without subsequent CoCl2 treatment (Fig. 7J) and found again that increased oxygen caused initially high HIF1α levels to decrease, followed by increased Smad activation and Id1 expression. In total, these results indicate that HIF1α is required for the full repressive effect of low oxygen on BMP signaling.


Hypoxia is positively correlated with higher aggressiveness in brain tumors and limited effectiveness of anticancer treatments [53]. It is unclear whether hypoxia plays a causal role in this relationship. In the present study, we found that HGG precursors require a lower (2%) oxygen tension for maximal expansion than their normal counterparts (5%, Fig. 1). This is consistent with the lower oxygen tensions found in tumors compared with normal tissue. Furthermore, hypoxia and HIF1α blocked activation of Smad1, Smad5, and Smad8 and glial differentiation of HGG precursors in response to BMPs, which is a critical pathway for precursor differentiation during CNS development [20]. Importantly, we found that BMPs reciprocally downregulated HIF1α in primary HGG precursors, but not in normal SVZ precursors, and that this downregulation was facilitated by increasing oxygen tension. These results provide a possible explanation for the altered response of HGG precursors, compared with normal SVZ precursors, to BMP treatment and suggest a selective vulnerability of HGG to the combined effects of HIF1α inhibition and BMP pathway activation.

Oxygen tensions in tumors can range from physiological levels to anoxia in highly necrotic areas [4], and tumor aggressiveness inversely correlates with oxygen tension [5–7]. Although hypoxia has primarily been thought to be a consequence of inefficient angiogenesis in aggressively growing tumors [4], there is evidence that hypoxia and HIF proteins have an active role in promoting tumorigenesis [7, 10]. It is possible that low oxygen tension promotes tumorigenesis by different mechanisms depending on whether it is very low (i.e., close to anoxia) or near physiological levels. Severe hypoxia has been shown to promote HIF1α-mediated genomic instability [14]. However, our present study indicates that oxygen tension in the physiological range promotes a selective suppression of mitotic arrest or apoptosis. This is critical in interpreting how anticancer agents perform under standard laboratory culture conditions (20% oxygen) versus those found in vivo. Our work also shows that increasing oxygen tension promotes activation of p53 through Ser37 phosphorylation. Additionally, normal SVZ precursors show selective Ser15 phosphorylation whereas HGG precursors show Ser20 phosphorylation of p53. Differential phosphorylation may, in part, explain the distinct thresholds for enhanced expansion of normal and HGG precursors under low oxygen tensions.

These results suggest that hypoxia may enhance the maintenance and expansion of cancer stem cells [18, 19]. We found that the percentage of CD133+ stem cells from pediatric HGG precursors declined after acute exposure to 20% oxygen for 72 hours and was most strongly reduced when the cells were simultaneously exposed to BMP2 in 20% oxygen. There was a corresponding increase in CD133 cells, which are enriched in GFAP+ astrocytes [27], consistent with a differentiative effect. A decrease was seen also in CD133+ stem cells from adult HGG precursors (supporting information Fig. 3J), in which the percentage of this immature subpopulation was lower. This age difference is consistent with other evidence that malignant gliomagenesis is molecularly distinct in children and adults [1].

In normal neural precursors, low physiological oxygen tensions repress differentiation whereas higher oxygen tensions promote differentiation [27, 29, 54]. This may reflect a normal developmental program by which precursors respond to oxygen tension. Lower oxygen, mediated by HIF1α, promotes the activity of Notch [17], which is an important antineurogenic, progliogenic, and stem cell self-renewal factor. Conversely, we show here that lower oxygen represses the differentiative responses to BMPs in pediatric HGG precursors (Figs. 3–5). Whereas BMPs promote the differentiation of HGG precursors when applied in vitro under high oxygen tensions, BMP application during tumor cell grafting delays, but does not prevent, mortality [21, 22]. This suggests that the BMP-treated cells were able to resume proliferation once returned to the brain, or that a subset of cells escaped the effect of BMPs due to the antagonistic effects of the brain microenvironment. BMPs are expressed in the choroid plexus [55] and have been detected in adult cerebrospinal fluid [56], suggesting that their differentiative actions persist throughout life. One way these antimitotic actions can be blocked is through endogenous BMP inhibitors such as Noggin, which allows stem cells to proliferate and generate neurons [25]. Importantly, our work indicates that low oxygen, via HIF1α, can also antagonize BMP activation, both from exogenous sources as well as from endogenous activity arising from the cells themselves.

We previously found that distinct BMP receptor signaling pathways are paradoxically capable of promoting proliferation or differentiation/apoptosis in normal stem cells [57]. Lee et al. [22] showed that methylation of the BMP receptor BMPR-IB gene inhibited its expression and led to a proliferative response to BMPs, consistent with our earlier observations. In the present study we found that BMP2 exerted an antimitotic effect on all tumors tested, consistent with BMPR-IB expression (supporting information Fig. 1), and selectively promoted apoptosis of one tumor (supporting information Fig. 2). This suggests that altered regulation of BMP signaling molecules in HGG cells may confer either resistance or selective vulnerability to death or differentiation.


Recent advances in cancer therapy are based on a more detailed understanding of the signaling molecules that regulate normal proliferation and that are dysregulated in cancer [58]. Similar insights are now being made with signaling pathways that control development and cell fate [10, 59, 60]. Our results show that HGG precursor proliferation is controlled by a novel and mutually antagonistic interaction between two developmentally important signals, low oxygen and its effector, HIF1α, and BMPs. Due to the technical difficulty in manipulating local oxygen tensions in vivo and the potential influence of nonoxygen-related signals on HIF activity [10], it is not yet known if this hypoxia-regulated mechanism is operative in vivo. However, the differential sensitivity of HGG and normal precursors to these signals suggests that titrating HIF1α inhibition and BMP activation may allow selective targeting of brain tumors while sparing normal stem cells.


This study was supported by funds from the Frank and Nancy Parsons Fund, the Zickler Fund, the Georgia Derrico and Rod Porter Fund for the CNMC/University of Padova Sister Program, the Italian Association AIRC regional grant, the FIRB grant, the Italian Association for the Fight against Neuroblastoma (Pensiero Project), the Italian Ministry of Health Oncology Program 2006, the CNMC Research Advisory Council, the CHOC Foundation for Children, and grants from the CNMC Board of Visitors. We thank Dr Elisabeth Rushing, Armed Forces Institute of Pathology, Washington, DC, and Prof. Felice Giangaspero, Neurosurgery Department, Umberto I Hospital, Rome, Italy, for providing the external secondary neuropathological review.


The authors indicate no potential conflicts of interest.