Glioma Stem Cell Maintenance: The Role of the Microenvironment

Abstract

Glioblastomas are highly lethal cancers for which conventional therapies provide only palliation. The cellular heterogeneity of glioblastomas is manifest in genetic and epigenetic variation with both stochastic and hierarchical models informing cellular phenotypes. At the apex of the hierarchy is a self-renewing, tumorigenic, cancer stem cell (CSC). The significance of CSCs is underscored by their resistance to cytotoxic therapies, invasive potential, and promotion of angiogenesis. Thus, targeting CSCs may offer therapeutic benefit and sensitize tumors to conventional treatment, demanding elucidation of CSC regulation. Attention has been paid to intrinsic cellular systems in CSCs, but recognition of extrinsic factors is evolving. Glioma stem cells (GSCs) are enriched in functional niches—prominently the perivascular space and hypoxic regions. These niches provide instructive cues to maintain GSCs and induce cellular plasticity towards a stem-like phenotype. GSC-maintaining niches may therefore offer novel therapeutic targets but also signal additional complexity with perhaps different pools of GSCs governed by different molecular mechanisms that must be targeted for tumor control.

Introduction

Despite many advances in tumor biology, glioblastoma (GBM) has remained the most common and deadly adult brain tumor [1]. Recent advances in clinical treatment have only modestly improved the average survival of a newly diagnosed GBM patient to less than 15 months [1]. Insights into GBM pathogenesis via genetically engineered mouse models [2] and global genetic analyses [3] have increased our understanding of this malignant disease. Contributing to GBM pathogenesis is the hierarchy of the heterogeneous neoplastic compartment, which was first described more than a century ago by Rudolf Virchow. At the pinnacle of the neoplastic hierarchy are glioma stem cells (GSCs, also known as tumor propagating cells or tumor initiating cells) that are defined through their functional ability to self-renew and propagate tumors in immunocompromised mouse models [4]. While studies have refuted the GSC paradigm in certain cancers, the presence of a stem-like subpopulation within GBMs has been well established [5]. The existence of GSCs remains a point of debate due to the unresolved nature of the cell(s) of origin, their immunophenotypes (markers), and their rarity within the neoplastic cell population. Rigorous functional studies have been able to characterize the GSC population [4, 6, 7] and provide evidence that GSCs are resistant to radiation and chemotherapy [8, 9]. Identifying and understanding the intrinsic GSC regulators driving these phenotypes has been a focus of interest, but the importance of extrinsic factor regulation is increasing. Recent experimental evidence has demonstrated that GSCs are enriched in specific niches around tumor vessels and areas of necrosis [10, 11], the latter associated with restricted oxygen. Interrogation of the extracellular matrix (ECM) associated with the perivascular niche has revealed an important relationship between ECM components, such as laminin, and cell-surface proteins, such as the integrins found on GSCs [12]. Propagation of the GSC population as well as tumor growth was dependent on the communication between the extracellular matrix. Additionally, studies support the hypoxic niche as important to GSCs: GSC maintenance requires both hypoxia inducible factor-1α (HIF1α) and hypoxia inducible factor-2α (HIF2α; [11, 13–15]), two canonical signaling pathways that respond to oxygen tension. The downstream pathways that hypoxia, and specifically HIF2α, utilizes to promote a stem-like state are not well characterized, and the mechanisms driving HIF2α expression in GSCs are unknown. Additional studies have further demonstrated that microenvironmental conditions such as hypoxia and acidic stress actively promote GSC function in multiple cell populations [13–16]. Collectively, these data suggest that there is plasticity in the GSC phenotype that can be regulated by the microenvironment. In order to develop more effective routes of clinical treatment, the contribution of the microenvironment to overall GSC growth and tumor propagation must be appreciated. However, GSCs must first be prospectively identified and enriched from bulk tumor populations in order to interrogate their function within tumors.

Identification of Cancer Stem Cells

Many studies have provided evidence for the existence of a stem-like subpopulation in several tumor types. Cancer stem cells (CSCs) were first identified in acute myeloid leukemia (AML) following sorting for expression of cell surface markers previously associated with normal hematopoietic stem cells (Bonnet and Dick, 1997) [17]. Although prior studies indicated that human AML cells had finite replication capacity when assessed in clonogenic assays in vitro, a subset of AML cells was capable of engrafting the bone marrow of immunocompromised mice to give rise to de novo leukemia that bore many histologic similarities to the parental disease. In the xenograft model, only leukemia cells bearing the CD34+/CD38− immunophenotype were able to reliably initiate AML (Bonnet and Dick, 1997) (17). These studies clearly demonstrated that a fraction of leukemic cells identified by specific marker expression possessed leukemogenic capacity. Similar ectopic xenograft assays in immunocompromised mouse models were utilized to identify CSCs in other solid tumors, including breast, brain, bone, liver, colon, prostate, pancreas, head and neck squamous cell carcinoma, and melanoma, although controversy still exists over the CSC theory (Singh et al., 2003; Hemmati et al., 2003; Galli et al., 2004; Singh et al., 2004; Fang et al., 2005; Vescovi et al., 2006; Bao et al., 2006; Schatton et al., 2008; Boiko et al., 2010; Dou et al., 2009; Quintana et al., 2008) [18–34].

Required Functional Characteristics of Glioma Stem Cells

The real time identification of adult tissue specific stem cells presents continued challenge and remains functional. Besides requiring the functional characteristics of self-renewal and multi-lineage differentiation, normal stem cells in different organ systems possess dramatically varied proliferation profiles as well as lineage potency. Although GSCs share many similarities with normal stem cells, such as marker expression and self-renewal, glioma inherently possesses aberrant cell behavior as well as unique gene profiles among the heterogeneous subpopulations. Intertumoral variability is reflected in variance of the GSC phenotypes. Keeping in mind this phenotypic diversity, care must be taken when defining the GSC population.

Given the dramatic variance in tumors between patients, the definition of GSC is necessarily functional. A GSC can be defined as a cell that is able, through continuous proliferation and self-renewal, to maintain and propagate a glioma that is a phenotypic copy of the parental tumor. In order to maintain a tumor, a GSC must also have to ability to maintain its own subpopulation via self-renewal, meaning reproducing non-differentiated stem cell-like copies following cell division. A GSC that cannot self-renew will consequently exhaust the stem-like fraction of the tumor responsible for propagation. In addition, GSCs should exhibit the most important characteristic first enumerated by Koch: the cell should have the ability, when introduced into a suitable environment, to recapitulate the phenotype of the original tumor. These concepts are in accordance with the commonly accepted current definition of a CSC: a cell that can self-renew, demonstrate sustained proliferation, and is capable of tumor propagation [59].

As there is currently no prospective way of selecting for cells that demonstrate self-renewal or sustained proliferation, this must be observed retrospectively. The most widely used test for surrogate in vitro self-renewal is the serial passage of GSCs in suspended culture, in which single cells should form three dimensional structures termed neurospheres. This assay allows for measurement of the self-renewal through long-term generation of spheres, and neurosphere formation capacity in vitro may be an important indicator of glioma biology in vivo (Laks et al., 2009). A retrospective study of the relationship between neurosphere formation, tumorigenic capacity, and patient outcome determined that sustained neurosphere formation in vitro was associated with tumor propagation in xenograft models and poor clinical outcome (Laks et al., 2009). However, there are limitations to this approach (reviewed in Wan et al., 2010). GSCs should be cultured as single cells in suspension otherwise the assay may simply be demonstrating anchorage-independent growth and cell aggregation rather than self-renewal. Neurospheres are also an artifact of cell culture as no in vivo correlate exists in either glioma or normal brain physiology. As the spheres expand, internal cellular heterogeneity increases, most likely due to diffusion limitations of oxygen, growth factors, and metabolic factors. Thus, the growth of a neurosphere does not definitively prove that a glioma cell is a GSC.

Additional consideration must also be given to the cell culture conditions of neurospheres. The typical culture media for GSCs contains supplemental epidermal growth factor (EGF) and basic fibroblast growth factor (bFGF) which has been used to support the growth of normal neural stem cells [60–63], although proliferation of GSCs has been shown to occur independent of growth factor addition (Kelly et al., 2009). Typically, EGF and bFGF are included in culture media to support the growth of GSCs, inhibit spontaneous differentiation, and help to maintain genotypic similarity to the parental primary tumor. However the presence of strong pro-proliferative signals can eventually lead to selection for cells that possess high levels of the receptors (such as EGFR) or abnormal sensitivity to growth factors. The requirement of growth factors in media has raised concerns of cell culture bias and how this could alter in vitro data collection. The proper use and concentration of EGF and bFGF is a contested issue and it is still not entirely known what long-term effect EGF and bFGF can have on GSCs in culture. This in combination for the potential for selection makes it important to limit passage in cell culture and avoid the use of CSC lines, which have been passaged long term.

The gold standard for the functional demonstration of a GSC remains tumor propagation. In this assay, a limited number of cancer cells are introduced to an orthotopic host location such as the brain of immunocompromised mice. More accurately, a limiting-dilution assay is performed in which a decreasing number of putative GSCs are intracranially injected to determine the minimal number of cells required to form tumors, which then serves as a measure of the frequency of tumor-propagation capable cells [5]. The theoretical ideal would be injection of a single cell that would then generate a tumor, however this has not yet been demonstrated. In practice, efficient cell sorting and subsequent survival of solid tumor cells following flow cytometry varies widely. Currently, reliable tumor formation has been demonstrated with only a few hundred cells (Singh et al., 2004) [64]. In addition to the technical limitations of flow sorting, the difficulty found in tumor propagation could also be due to a requirement for support from non-stem cells [65]. Intracranial tumor formation, however, remains the only definitive way of determining the presence of functional GSCs, and as such, is absolutely required for any experimental interrogation that utilizes GSCs.

Other Functional Characteristics of Glioma Stem Cells

In addition the required functional characteristics of GSCs, there are several pro-tumorigenic properties of GSCs which contribute to the GSC phenotype but are not necessarily common for all isolated CSC subsets. Analysis of GBM cells positive for the GSC marker CD133 has suggested a molecular profile associated with invasion and angiogenesis (Garcia et al., 2010), and both promotion of tumor angiogenesis and invasion are suggested as additional functional characteristics of GSCs. Tumors derived from GSCs are highly vascular (Bao et al., 2006) with more infiltration of normal tissue compared to standard glioma cell lines (Inoue et al., 2010; Brehar et al., 2010; Wakimoto et al., 2009; Cheng et al., 2011). The angiogenic properties of GSCs are due, at least in part, to elevated production of VEGF and stromal-derived factor 1 (SDF1) (Garcia et al., 2010; Folkins et al., 2009; Bao et al., 2006; Oka et al., 2007; Yao et al., 2008), and recent evidence suggests that GSCs can transdifferentiate to endothelial cells (Ricci-Vitiani et al., 2010; Wang et al., 2010). Although the precise mechanisms responsible for differential GSC invasion are not clear, GSCs may express differential activity of matrix metalloproteinases (Inoue et al., 2010) or AKT which contribute to invasion (Molina et al., 2010; Eyler et al., 2008). In addition, some cell surface markers known to enrich for GSCs, such as L1CAM (Bao et al., 2006) and integrin α6 (Lathia et al., 2010), can regulate invasion in glioma (Goldbrunner et al., 1996; Cheng et al., 2011). These data suggest that angiogenesis and invasion may be driven by specific molecular pathways in GSCs within gliomas.

GSCs are also characterized by an ability to resist chemo- and radiotherapy. Although surgery and radiation are a standard and effective therapy for GBM, CD133 positive GBM cells retain the ability to propagate tumors in immunocompromised mouse models post irradiation (Bao et al., 2006; Ma et al., 2011). Irradiation actually increases the percentage of GSCs increases in tumors due to the preferential survival of this tumor subset (Bao et al., 2006). GSC survival is due to the ability to more readily repair radiation-induced DNA damage than non-stem glioma cells through mechanisms which involve the checkpoint kinases (Bao et al., 2006). In addition to radiation and surgery, GBM patients are treated with the oral alkylating agent temozolomide (Stupp et al., 2009). However, GSCs were less sensitive to temozolomide induced cell death than normal neural stem cells (Gong et al., 2011), leukemia cells (Eramo et al., 2006), or (non-stem glioma cells (Fu et al., 2009; Liu et al., 2006), although there are reports of preferential targeting of GSC lines by temozolomide (Beier et al., 2008). Resistance to temozolomide in GBM cells was also associated with a stem cell-like gene signature that included expression of the GSC marker CD133 (Gaspar et al., 2010). CD133+ or neurosphere forming GBM cells also displayed resistance to the type II topoisomerase inhibitors etoposide (Jin et al., 2010; Nakai et al., 2010; Liu et al., 2006; Eramo et al., 2006) and teniposide (Qin et al., 2010). In converse experiments, overexpression of CD133 alone was suggested to promote resistance to other chemotherapies including the topoisomerase I inhibitor camptothecin and the DNA synthesis inhibitor doxorubicin (Angelastro and Lame, 2010). This chemo- and radioresistance may be overcome by targeting of GSC molecular pathways. A cell surface protein known to enrich for GSCs, L1CAM (Bao et al., 2006), regulates the DNA damage checkpoint response of GSCs (Cheng et al., 2011). Targeting of L1CAM through directed shRNAs reduced DNA repair and sensitized GSCs to irradiation (Cheng et al., 2011). The CSC marker and polycomb group protein BMI1 has also been suggested to recruit DNA damage response proteins after irradiation to promote repair (Facchino et al., 2010). Although the importance of cyclooxygenase-2 (COX-2) signaling for the GSC phenotype is not well characterized, treatment of GSCs with the selective COX-2 inhibitor celecoxib potently decreased the ability to propagate tumors in vivo and improved the efficacy of radiotherapy (Ma et al., 2011). Treatment with the mTOR inhibitor rapamycin also induced radiosensitivity through a mechanism that may involve activation of autophagy (Zhuang et al., 2011a). Autophagy induced by loss of DNA-dependent protein kinase catalytic subunit (DNA-PKCs) was also associated with radiosensitization of GSCs (Zhuang et al., 2011b), although other data has suggested that autophagy inhibitors would be beneficial for improving radiotherapies (Lomonaco et al., 2009). Together these data suggest that tumor recurrence is mediated by GSCs which survive therapy and should be targeted for better tumor control.

One final property of GSCs contributing to tumor propagation which is only beginning to be understood is their ability to suppress the immune system. Multiple studies now demonstrate that GSCs can preferentially inhibit the proliferation of T-cells (Di Tomaso et al., 2010; Wei et al., 2010a; Wei et al., 2010b). GSCs also produce cytokines known to recruit and promote immunosuppression by microglia, suggesting modulation of innate immunity (Wu et al., 2010). Thus, GSCs may have several mechanisms through which they could escape immune surveillance. Taken together, all of these data suggest that many of the known properties of cancer which contribute to progression and make treatment difficult--the ability to drive angiogenesis, invade or metastasize, survive cytotoxic therapies, and suppress the immune system—are elevated in the CSC subset. Thus, targeting of GSCs within GBM may represent a significant advance in our ability to treat this devastating disease.

Immunophenotypic Characteristics of Glioma Stem Cells

In addition to the functional characteristics of GSCs mentioned above, GSCs commonly express normal stem cell markers. Assaying the immunophenotype by flow cytometry is a useful tool for enrichment of GSCs in a laboratory setting, but there is no common antigenic profile that is representative of all GSCs. This may due to the inherent instability of glioma cells, as well as heterogeneity in the microenvironment of the tumor. The expression of markers within the GSC subpopulation of a single patient’s tumor may be wildly varied. This variation is exacerbated when comparing multiple patients who may present with different stages of the disease as well as different treatment regimens. In addition, the diagnosis of GBM is now recognized to include many different subtypes with different genetic profiles (Phillips et al., 2006; Verhaak et al., 2010; Zheng et al., 2011). Despite this variability, there are several markers that have been successfully utilized for GSC enrichment in the laboratory setting as later discussed in detail. These markers are not without limitations, and may only be reliable when utilized immediately following tumor resection, as exposure to an artificial culture environment may dramatically alter expression profiles of these markers (Lee et al., 2006) [66]. As a further complication, cell surface markers do not completely segregate as distinct populations with separate cytometric peaks. Rather, the population profile of these markers is often continuous, with the CSC population enriched in distinct levels of marker expression. Ideally, each marker will also be functionally verified in each tumor using neurosphere formation and in vivo tumor propagation as discussed above. Segregation for GSC enriched populations should also be confirmed through verification of enrichment for other intracellular stem cell markers such as Olig2, Sox2, Bmi1, or Musashi (Hemmati et al., 2003; Bao et al., 2006; Ligon et al., 2007; Godlewski et al., 2008; Abdouh et al., 2009; Gangemi et al., 2009). These studies are important as a marker which efficiently separates a tumor-initiating fraction from one patient-derived tumor may not efficiently segregate in another patient-derived sample, due largely to the vast heterogeneity of the disease.

Although the percentage of cells expressing GSC markers within a tumor is usually a minority of the total tumor, CSCs are not necessarily vary rare (Sing et al., 2004; Kelly et al., 2007; Quintana et al., 2008; reviewed in Girouard and Murphy, 2011). The frequency of GSCs expressing stem cell markers is not fixed, and may depend on the region of the tumor used for GSC isolation. For example and as later further discussed, the perivascular niche may regulate the GSC phenotype and vascularity is regional (Calabrese et al., 2007; Lathia et al., 2010; Charles et al., 2010). Indeed, experimental evidence suggested that the proportion of neoplastic cells that can possess a GSC phenotype may be significantly influenced by the microenvironment [13, 15, 16, 67]. In addition, the resistance of GSCs to radio- and chemotherapy facilitates enrichment for the GSC population in clinically treated patients (liu et al., 2006) [8]. These data suggest that the observed portion of a tumor that is GSC marker-positive may be arbitrarily determined at the time of sorting and not reflective of the in vivo tumorigenic fraction.

While GSCs express stem cell markers, cells derived from GSCs can display lineage markers. The ability of CSCs to derive cells expressing markers of multiple lineages is commonly observed in some tumor types, but is not necessarily required of CSCs. In glioma, GSCs have been shown to have the capacity to generate cells which express markers of astrocyte, neuronal, and oligodendrocyte lineages (Singh et al., 2003; Bao et al., 2006). This ability of GSCs is critical for recapitulating the phenotype of the parental tumor which is characterized by histological variance as the original name of grade IV glioma, Glioblastoma Multiforme, suggests. The presence of mutations, genetic instability, and cellular deregulation means that the differentiated cells in a tumor do not have counterparts in the normal physiological lineages. A differentiated glioma cell may express markers of different lineages simultaneously (for example the neuronal lineage marker β3-Tubulin and the astrocytic lineage marker Glial Fibrillary Acidic Protein, GFAP could be expressed on the same glioma cell) which would not occur in normal brain cells (Martinez-Diaz et al., 2003; Perry et al., 2009). Thus, it is critical to recognize how the immunophenotype of GSCs and their derivatives are regulated to continue to develop our understanding of the GSC phenotype.

CD133 as a Marker for Glioma Stem Cells

Immunophenotypes enrich rather than identify CSCs, and there is debate regarding the best marker(s) to use to enrich for GSCs. GSCs were first identified in human specimens using positivity for the cell surface neural stem cell marker, CD133 (Prominin1) (Uchida et al., 2000), to enrich for tumorigenic potential in immunocompromised mouse models [5, 35]. Their self-renewing potential was also evaluated in the neurosphere assay [5, 35], and CD133 is now a frequently used GSC enriching marker. Although some papers have failed to show GSC enrichment with CD133, these studies have generally used extensively cultured cells which limits their interpretation (Beier et al., 2007; Patru et al., 2010). In studies with CD133 positive GBM cells, gene expression profiles were similar to that of embryonic stem cells suggesting the involvement of a stem cell-like molecular profile [37, 38]. CD133 positive GBM cells are resist to radiation or chemotherapeutic treatment due to their capability to activate cell cycle check point, DNA repair mechanisms, and anti-apopototic processes, as compared to the CD133 negative counterpart [8, 9]. Further, gene expression profiling of chemo-resistant glioma cells revealed that resistant cells are enriched for CD133 expression [39, 40]. These data regarding the preferential survival of GSCs may explain findings which suggest that CD133 expression is associated with poor clinical outcome, tumor re-growth, and malignant progression or therapeutic resistance in gliomas (Zeppernick et al., 2008; Pallini et al., 2008; Murat et al., 2008). Even though there is not a complete consensus on the correlation between stem cell marker expression and glioma patient prognosis (Neuropathology, 2011), these data do suggest that CD133 is a useful marker to enrich GSCs and cells sorted for CD133 expression can be used to interrogate clinically relevant questions.

Although CD133 has been proven as a useful GSC enrichment tool, its biological function is not well understood. CD133 is a pentaspan transmembrane glycoprotein enriched in the cholesterol-rich membrane microdomain and localized to plasma membrane protrusions (Corbeil et al., 2010). Its expression in epithelial cells is localized to apical membrane as observed in neuroepithelial cells. Asymmetric distribution of CD133 in neuroepithelial cells is postulated to be critical for neurogenesis [41]. Polarization of CD133 has been observed during cytokine-stimulated migration of hematopoietic cells [42]. Overexpression of CD133 in glioma cells has also been demonstrated to activate MAP kinase pathway [43] and enhance drug resistance [44]. Down regulation of CD133 decreased proliferative activity of glioma [45]. These data demonstrate that CD133 does have functional significance in glioma cells. Further investigation into the molecular regulation of CD133 could yield clues as to its mechanistic function.

CD133 expression is subject to multiple levels of regulation. The CD133 gene can be transcribed from multiple promoter sites and the resulting transcripts are edited to yield multiple splice variants, some of which are not recognized by anti-CD133 antibodies due to improper protein localization [46–48]. CD133 protein can be post-translationally modified through glycosylation, which is required for recognition by some anti-CD133 antibodies [49]. The epitope availability is further modified by interaction between CD133 and sphingolipids, whose levels fluctuate according cell proliferation status [50]. With these complexities of CD133 expression regulation and its interaction with other molecule, a careful assessment is required to interpret immunological detection of CD133. In addition, several other markers are being utilized as effective GSC markers.

Alternative Cell Surface Markers for Glioma stem cells

Several studies have demonstrated tumorigenic capacity of CD133 negative cells that could be due to GSC populations that escaped antibody labeling and sorting. Indeed, several cell surface markers such as stage-specific embryonic antigen (SSEA-1/CD15/LeX), integrin α6, A2B5, and CD44 and have been reported to enrich for GSCs independent of CD133 (Ogden et al., 2008; Anido et al., 2010; Mao et al., 2009; Tchoghandjian et al., 2010) [12, 36]. These markers provide the benefit of being present on the cell surface providing the opportunity for prospective isolation of live cells, whereas other GSC markers (Olig2, Sox2, etc) are intracellular and cannot be directly utilized for sorting without cellular manipulations.

SSEA-1 selection was shown through functional assays to enrich for GSCs in that SSEA-1 positive cells were highly tumorigenic compared to SSEA-1 negative cells in immunocompromised mouse models (Son et al., 2009; Mao et al., 2009). In addition, SSEA-1 positive GBM cells were enriched for neurosphere formation capacity and cells derived from these neurospheres under differentiation-inducing conditions expressed markers of both the astrocyte and neuronal lineages (Son et al., 2009; Mao et al., 2009). Interestingly, in tumors in which CD133 was expressed, SSEA-1 selection enriched for CD133 as well suggesting some overlap between these GSC markers in certain tumors (Son et al., 2009). In addition to these data in human cells, SSEA-1 also enriches for CSCs in medulloblastoma mouse models (Ward et al., 2009) [51]. Although SSEA-1 may be secreted from neural stem cells to modulate Wnt signaling through its binding affinity for Wnt-1 [52], the biological function of SSEA-1 in GSCs remains to be fully elucidated.

Another cell surface marker which enriches for GSCs is integrin α6, as integrin α6 high cells are more tumorigenic than integrin α6 low cells (Lathia et al., 2010). Integrin α6 is co-expressed with conventional GSC markers including CD133 and Olig2, but integrin α6 positive cells are enriched for neurosphere formation capacity regardless of CD133 status. These data indicate that integrin α6 recognizes a broader pool of GSCs than that characterized by CD133 alone. Further demonstrating the biological importance of integrin α6 to the GSC phenotype, antibody-mediated neutralization or ShRNA mediated knockdown of integrin α6 significantly impeded neurosphere formation in vitro and tumor formation in vivo. These data demonstrated functional significance of a cell surface GSC marker useful for prospective sorting and suggested targeting of integrin α6 as a potential therapy [12].

Additional studies have utilized a cell surface ganglioside epitope expressed in white matter cells with neural stem cell properties, A2B5 (Dubois et al., 1990), to enrich for GSCs. A2B5 positive cells sorted from GBM specimens formed tumors in immunocompromised mice at high frequency whereas A2B5 negative cells only rarely formed tumors (Ogden et al., 2008; Tchoghandjian et al., 2010). A2B5 positive cells were also enriched for the capacity for form neurospheres, and cells from those spheres had the capacity for multi-lineage differentiation in vitro (Tchoghandjian et al., 2010). A2B5 appeared to co-segregate with CD133 as no A2B5−/CD133+ cells were able to be isolated via flow cytometry (Tchoghandjian et al., 2010; Ogden et al., 2008). However, A2B5 positive CD133 negative cells were still able to initiate tumors indicating that A2B5 could segregate for tumorigenic potential independent of CD133 (Ogden et al., 2008; Tchoghandjian et al., 2010).

Markers for GSCs will continue to be identified as our understanding of the proteins expressed at the cell surface of GSCs rapidly expands. When interpreting these results, evaluation of cell isolation conditions and functional GSC characteristics must be incorporated. In many cases, flow sorting based on surface marker is not efficient and the harsh conditions of sorting can force phenotypic changes in the cell. It is therefore critical to utilize additional methods to verify functional enrichment for the GSC population independent of surface marker expression.

Identification of Glioma Stem Cells Independent of Cell Surface Markers

GSCs can be also enriched using other approaches that do not rely on cell surface markers. Neural stem cells can be highly enriched by cell sorting of a side population capable of excluding a DNA-staining fluorescent dye, Hoechst 33342 (Lechner et al., 2002; Murayama et al., 2002) [53]. Similar side population cells sorted from embryonic stem cells resemble pluripotent embryonic cell lineages [54], demonstrating Hoechst 33342 exclusion is a good functional characteristic that can be utilized to enrich stem cells out of culture. When applied to a glioma population, the side population approach enriched for tumorigenic glioma cells [55, 56]. Cells in the side population were enriched for the ability to form neurospheres as well as tumorigenic potential (Bleau et al., 2009; Patrawala et al., 2005). Interestingly, loss of the tumor suppressor PTEN was associated with an increase in the side population phenotype (Bleau et al., 2009), suggesting that this common genetic alteration in glioblastoma may contribute to a GSC phenotype. As the ABC transporter protein ABCG2/Bcrp1 can mediate the side population phenotype (Zhou et al., 2001) and may be elevated in GSCs (Liu et al., 2006), ABCG2 was also evaluated in the glioma samples. One paper indicated that ABCG2 identified the side population in glioma (Bleau et al., 2009), whereas another indicated the tumorigenic capacity of the side population and ABCG2+ cells were distinct (Patrawala et al., 2005). A more recent study also suggests that side population cells had decreased tumorigenic potential (Broadley et al., 2011). While some of the differences in results may be due to differences in cell types evaluated (primary specimens, established human glioma lines, or mouse cells), these studies are complicated by the fact that the dye can be toxic and the viability of cells without efflux pumps reduced.

A recent report has also described an additional novel method to enrich for glioma growth which may indicate GSC enrichment. When autofluorescent (excitation at 480 nm, emission at 520 nm), large nongranular (high in forward scatter and low in side scatter) cells were freshly collected from human glioma samples by fluorescence activated cell sorting (FACS), these cells were tumorigenic in xenograft assay. Interestingly, CD133 did not co-segregate with this autofluorescent and large nongranular phenotype [58]. Together these data suggest that enrichment for GSCs may be achieved through methods based on functional characteristics independent of cell surface marker expression, but there is no method identified to date which has no caveats. Evaluation of enrichment efficiency is therefore vital regardless of the utilized method.

Cell Surface GSC Regulators

Characterization of the molecules differentially expressed or active in GSCs is anticipated to yield novel targets for cancer therapy and define unique mechanisms for regulating cell fate. Although intrinsic GSC signals can be stimulated by microenvironmental conditions and secreted factors that could be targeted for therapy (as will be further discussed in later sections), we will presently focus on molecular mediators of GSC maintenance known to be present on the cell surface.

One of the core ingredients in media supporting GSC growth is EGF; [60]. EGF binds to epidermal growth factor receptor (EGFR) to initiate a signaling cascade known to regulate a variety of cellular behaviors including growth, survival, and differentiation [68]. For example, overexpression of wildtype EGFR or a constitutively active mutant form, EGFRvIII, in NSCs promotes cell proliferation and survival while decreasing differentiation towards a neuronal lineage [69]. An important role for EGFR signaling in glioma has been established with multiple prior studies in both humans and animal model systems [68]. It is therefore not surprising that recent evidence demonstrates that EGFR is also critical for the GSC phenotype (Kelly et al., 2009; Howard et al., 2010; Mazzoleni et al., 2010). EGFR positive glioma cells form tumors at an increased rate compared to EGFR negative cells (Mazzoleni et al., 2010). Gain of EGFR function through overexpression increased tumorigenic potential. Among EGFR positive GSCs, targeting EGFR with shRNA or inhibitors reduced GSC mediated tumor propagation or cell growth respectively [70]. Thus, stimulation of EGFR signaling through EGF is likely to be important for GSC maintenance in vivo.

The transforming growth factor beta (TGF-β) superfamily of proteins is another set of growth factor receptors involved in regulation of the GSC phenotype. The TGF-β superfamily includes TGF-β itself as well as the bone morphogenic proteins (BMPs) and is a well established regulator of stem cell self-renewal and differentiation (reviewed in Watabe and Miyazono, 2009). TGF-β and BMP signal through receptor serine/threonine kinases which phosphorylate intracellular mediators called SMADs. Autocrine TGF-β signaling has been shown to maintain GSCs via regulation of the expression of Sox2/Sox4 and LIF (Ikushima et al., 2009; Penuelas et al., 2009). Addition of exogenous TGF-β promoted neurosphere formation from primary GBM specimens, antagonized FBS induced differentiation, and promoted the growth of tumors in vivo (Penuelas et al., 2009). TGF-β receptor inhibitors may target GSCs, as TGF-β receptor inhibitors decreased neurosphere formation capacity in vitro (Penuelas et al., 2009), and tumors from TGF-β receptor inhibitor treated mice had decreased expression of some GSC markers including CD44 (Andio et al., 2010). In contrast to these results suggesting TGF-β mediates GSC maintenance and similar to other biologies in which TGF-β and BMPs are antagonistic, BMPs have been shown to promote GSC differentiation (Piccirillo et al., 2006; Lee et al., 2008). Exogenous BMP treatment decreased GSC marker expression, including CD133, while increasing differentiation marker expression (Piccirillo et al., 2006). Importantly, both in vitro exposure of GSCs to BMP or in vivo delivery of BMP reduced the growth of GSC-derived tumors and increased survival of mice bearing human glioma xenografts (Piccirillo et al., 2006). In these cells in which BMP promotes differentiation and reduces the GSC phenotype, BMP receptor signaling is intact. If, however, the BMP receptor is epigenetically silenced, BMP can inhibit GSC differentiation (Lee et al., 2008). Restoration of BMP receptor expression through either overexpression or promoter demethylation reduces glioma cell tumorigenic capacity and enables BMP mediated differentiation (Lee et al., 2008). Together these results stress the potential significance of growth factor receptor signaling pathways, epigenetic modifications, as well as differences between GSC subsets.

Another growth factor receptor shown to be highly expressed on GSCs is erythropoietin (EPO) receptor (Cao et al., 2010). Erythropoietin is a glycoprotein regulating red blood cell production, so erythropoietin stimulating agents (ESAs) were given to cancer patients to counteract anemia resulting from chemo- and radiotherapy with sometimes adverse effects (reviewed in Arcasoy, 2008). The enhanced tumor progression in ESA treated cancer patients could be a result of the promotion of CSC maintenance (Cao et al., 2010; Phillips et al., 2007). GSCs were recently found to have an autocrine loop with elevated EPO production and EPO receptor expression (Cao et al., 2010), and EPO was shown to promote tumorsphere formation in CSCs isolated from both the brain and the breast (Cao et al., 2010; Phillips et al., 2007). Prospective enrichment for EPO receptor positive glioma cells selected for a tumor cell subset with increased capacity for neurosphere formation, and targeting EPO receptor with directed shRNAs reduced this ability. EPOR knockdown also reduced the capacity of GSCs to propagate tumors in immunocompromised mouse models in association with reduced GSC growth and survival (Cao et al., 2010). These data strongly suggest the potential clinical relevance of understanding the effects of growth factor receptor mediated signals on tumor cell subsets such as CSCs.

One cell surface molecule highly expressed in GSCs is the neural cell adhesion molecule, L1CAM [71]. L1CAM is overexpressed in glioma and other solid tumors in comparison to normal tissue [72–74] and is important for the regulation of neural cell growth, survival, and migration [75]. L1CAM is overexpressed in GSCs relative to non-stem glioma cells, where it plays a preferential role in GSC survival [71] Targeting of L1CAM inhibits GSC growth and neurosphere formation potential in association with the induction of apoptosis. This biological effect is attributed to a decrease in expression of the GSC marker Olig2 with a subsequent up-regulation of the growth regulator and tumor suppressor p21^WAF1/CIP1. Furthermore, L1CAM shRNA is sufficient to delay the growth of GSC initiated xenografts in vivo. These data demonstrate that L1CAM is required for GSC maintenance [71].

Notch signaling is another cell surface signaling pathway known to regulate neural stem cells (reviewed in Louvi et al., 2006) and now implicated in GSC maintenance (Hu et al., 2011; Hovinga et al., 2010; Fan et al., 2010; Zhang et al., 2008). Notch signaling is activated through cell-cell contacts in which ligand present on a neighboring cell binds to the extracellular receptor promoting Notch intracellular domain (NICD) release after γ secretase mediated protease cleavage. Notch activation promoted neurosphere formation (Zhang et al., 2008). Similarly, γ secretase inhibitors which block Notch signaling inhibited the proliferation and neurosphere formation capacity of isolated GSCs as well as promote the expression of differentiation markers (Hu et al., 2011; Fan et al., 2010). Inhibition of Notch signaling may also promote sensitivity to chemo- and radiotherapies (Ulasov et al., 2011; Wang et al., 2010), suggesting there may be benefit for γ secretase inhibitors for patient therapies. Indeed, γ secretase inhibitors are now in clinical trials at multiple major institutions (see ClinicalTrials.gov).

Similar to Notch signaling, the Sonic Hedgehog (SHH) pathway is a highly conserved mechanism of regulating neural stem cell biology (reviewed in Briscoe, 2009) which has recently been shown to regulate GSCs (Xu et al., 2008). In this pathway, SHH binding to the Patched receptor relieves Patched repression of Smoothened leading to activation of the GLI transcription factors. Although data suggests that there may be SHH dependent and independent GSC subtypes, targeting of GLI via shRNA decreased the proliferation and survival of SHH dependent GSCs in vitro with a concomitant decrease in GSC tumorigenic potential in vivo (Xu et al., 2008). These data are supplemented by recent findings indicating that SHH signaling is important for the infiltrative growth of GBM cell lines derived from CD133 positive cells (Uchida et al., 2011), and that inhibition of SHH reduces chemo- and radioresistance (Ulasov et al., 2011). Based on these preclinical data, it will be interesting to determine the outcome of initiated SHH pathway inhibitors in clinical trials (see ClinicalTrials.gov) and determine if there are any in vivo effects on CSCs in solid tumors such as glioma.

Intracellular Glioma Stem Cell Regulators

In addition to cell surface regulators of the CSC phenotype, there are multiple intracellular molecules and pathways now recognized to be important for GSC maintenance. Although we cannot comprehensively review all of the potential intracellular mediators of GSCs, we will seek to introduce some key pathways which several publications have indicated are critical for GSC biology.

One promising molecular target is the PI3K/PTEN/AKT/mTOR pathway, a mediator of cell survival and invasion signaling pathways which is often activated in gliomas through either loss of the tumor suppressor PTEN or growth factor mediated AKT phosphorylation (reviewed in Hambardzumyan et al., 2008). Suggesting involvement of the PI3K/PTEN/AKT/mTOR pathway in GSCs, activation of AKT was associated with acquisition of an invasive GSC phenotype (Molina et al.,k 2010). AKT also regulates the ABCG2 transporter involved in generation of the side population (discussed above; [55]) demonstrating AKT could directly modulate a potential GSC isolation technique. When AKT inhibitors were used to treat GSCs, GSCs display preferential sensitivity to AKT inhibition relative to the non-stem glioma cell fraction [78, 79]. Treatment of GSCs with multiple AKT inhibitors results in similar effects of decreased growth and neurosphere formation, and direct targeting Akt with shRNA improved the survival of mice bearing intracranial gliomas [78, 79]. In addition to effects of AKT, targeting mTOR with the inhibitor rapamycin or directed shRNA reduced the expression of GSC markers and impaired the ability of to generate neurospheres (Sunayama et al., 2010). Dual targeting of mTOR and PI3K with either rapamycin in combination with the PI3K inhibitor LY294002 or the use of the mTOR and PI3K inhibitor BEZ235 significantly increased the expression of a neuronal differentiation marker. BEZ235 was also capable of inhibiting the tumorigenic potential of GSCs in xenograft models (Sunayama et al., 2010). These data strongly suggest that targeting of the PI3K/PTEN/Akt/mTOR pathway may represent one anti-GSC based approach.

Another pathway now recognized to play an important role in the regulation of neural stem cells (reviewed in Widera et al., 2006) and GSCs involves nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB). In the canonical version of this pathway, a variety of stimuli can lead to NF-κB activation through degradation of cytoplasmic inhibitors of κB (IκB) which then permits nuclear localization of NF-κB proteins. Neural stem cells derived from the rat subventricular zone in the absence of exogenous growth factors that could form colonies in a soft agar assay had a high level of NF-κB activity [76]. NF-κB elevation appeared to correlate with adoption of a GSC-like phenotype in that there was an elevation of stem cell markers, prolonged proliferation, and elevated expression of VEGF [76]. These data suggested that constitutive activation of NF-κB was associated with the acquisition of a GSC-like phenotype [76]. Indeed, several proteins shown to be elevated in GSCs and mediators of the GSC phenotype are known to be targets of NF-κB signaling. For example, it was recently found that the NF-κB target gene Tumor Necrosis Factor inducible protein 3 (TNFAIP3), or A20, is expressed at higher levels in GSCs in comparison to matched non-stem tumor cells phenotype [77]. Targeting A20 reduces GSC growth in association with decreased neurosphere formation. The decreased growth of GSCs in the absence of A20 is due to increased apoptosis and may be related to the ability of A20 to confer resistance to TNF alpha induced apoptosis.. Furthermore, decreased A20 expression in GSCs reduces their tumorigenic potential demonstrating a critical role for A20 in GSC maintenance [77].

One transcription factor being investigated as a potential target for glioma therapy (reviewed in Liu et al., 2010) and recently recognized as an important regulator of GSCs is signal transducers and activators of transcription 3 (STAT3). STAT3 is activated through phosphorylation by a variety of growth factor receptor mediated signals, and is implicated in the regulation of neural stem cell differentiation (Cheng et al., 2011; Cheng et al., 2010). Elevated phosphorylation of STAT3 was observed in GSCs relative to non-stem glioma cells (Villalva et al., 2011; Sherry et al., 2009; Wang et al., 2009; Cao et al., 2010). Inhibition of Stat3 through small molecule inhibitors or directed shRNAs decreased GSC proliferation and neurosphere formation capacity (Villalva et al., 2011; Sherry et al., 2009; Wang et al., 2009; Cao et al., 2010). Targeting STAT3 in combination with temozolomide treatment also appears to sensitize GSCs to this standard of care chemotherapy (Villalva et al., 2011), suggesting the potential for STAT3 based combinatorial therapies.

Another important GSC regulating transcription factor is the oncoprotein c-myc which becomes active in response to mitogenic signals and has also has been distinguished as a significant regulator of stem cell biology, linking normal stem cell and cancer fields [80]. In human tumors, the presence of overexpressed c-myc is often linked to poor prognosis [81]. Studies have demonstrated that c-myc expression correlates with the grade of malignancy of GBMs. Furthermore, data has shown that not only is c-Myc required for glioma cancer cell proliferation, growth, and survival, but also, higher levels of c-Myc are expressed in GSCs in comparison to matched non-stem glioma cells [82]. In addition, when co-staining GSCs with neural stem cell marker Nestin and c-Myc, results demonstrated that more than 90% of the Nestin positive glioma cells were also c-Myc positive, suggesting that GSCs express high levels of c-Myc. These studies also demonstrated that upon knockdown of c-Myc, GSC growth and proliferation was significantly reduced, whereas the cell cycle progression of non-stem glioma cells was minimally disturbed. In fact, upon knockdown of c-Myc the expression of numerous cell cycle regulators downstream of c-Myc where altered in GSCs, however not in the non-stem cells. These data suggest that c-Myc along with its downstream regulators, play a specific role in the regulation of GSC proliferation and growth.

Perivascular Microenvironmental Regulation of Glioma Stem Cells

Although intrinsic regulation of the GSC phenotype is important for tumor growth and maintenance, recent studies have revealed that the extrinsic influence of the tumor microenvironment plays an critical role in maintaining the GSC population. It has been established that stem cells from multiple tissues reside in specified locations, or niches [83]. These niches, containing various differentiated cell types and secreted factors, direct the homeostasis of the stem cells by regulating the balance between self-renewal and differentiation. In the case of normal neural stem cells, regions rich in blood vessels, or perivascular niches, are critical for their maintenance [84–88]. More recently, a perivascular niche has been described for GSCs [10]. Although niche dependence between normal neural stem cells and CSCs is likely to differ greatly, insights have been made into components of the tumor perivascular niche that modulate GSC maintenance.

The first report of GSC residence near vascularized areas within the tumor came from the Gilbertson lab (10). Using co-immunostaining for CD34 (to mark the vascular endothelial cells) and Nestin (to mark the GSCs) in primary brain tumors, they were able to demonstrate direct contact of the GSCs with the tumor capillaries and quantify that the majority of the Nestin positive cells were localized in close proximity to the vasculature. Moving in vitro, they found that CSCs preferentially associated with co-cultured endothelial cells as compared to the majority of the tumor cells. A direct effect on CSC maintenance through factors secreted by the endothelial cells was demonstrated through the use of a transwell assay. The results indicated that both self-renewal and an undifferentiated state were maintained and highlighted potential niche regulation of a CSC phenotype. However, to validate this, in vivo systems were explored. Co-transplantation of brain tumor stem cells and endothelial cells into immunocompromised mice resulted in higher capacity of the propagation and growth of tumors in the brain by endothelial-derived factors. This ability to maintain brain tumor stem cells was not exhibited by any other cell types, highlighting the specificity of the functional relationship between endothelial cells and brain tumor stem cells. Together, these results suggest that support by the endothelium can promote self-renewal of GSCs and growth of the tumor in vivo.

In their next series of experiments, the Gilbertson lab demonstrated that disruption of the tumor vasculature, and hence the perivascular microenvironment supporting the GSCs, could prevent tumor growth. To achieve this, they treated mice bearing orthotopic xenograft tumors with anti-angiogenic therapy. Importantly, the percentage of self-renewing GSCs was significantly reduced, indicating that without the vascular support, there is reduced maintenance of the stem-like population. Bidirectional interplay between the tumor vasculature and GSCs was elucidated when it was shown that GSCs could release vascular endothelial growth factor (VEGF) to promote endothelial cell migration and tube formation in vitro [64]. Therapeutic targeting of GSC-expressed VEGF with the humanized neutralizing anti-VEGF antibody bevacizumab (Avastin) in a xenograft system resulted in decreased tumor growth with a less vascular phenotype (64). Further in vivo evidence for GSC contribution to tumor vasculature was shown using a VEGF-overexpressing mouse model whereby increased production of VEGF by the GSCs resulted in greater tumor burden that was more vascular and hemorrhagic [89].

The complex relationship between GSCs and tumor vasculature was made more so with recent reports identifying endothelial cells of tumor origin incorporated into the GBM microvessels [90, 91]. Ricci-Vitiani and colleagues showed that 20–90% of endothelial cells in each glioblastoma carries the same genomic alteration as tumor cells, suggesting that a significant portion of the vascular endothelium of tumors has a neoplastic origin. In vitro culture of GSCs in conditions similar to that of endothelial cells enhanced phenotypic and functional features characteristic of endothelial cells. In addition, tumor xenografts generated in immunocompromised mice by means of either orthotopic or subcutaneous injection of GSCs contain significant portions of human endothelial cells (90). Wang et al. similarly demonstrated that a subpopulation of endothelial cells within glioblastomas has the same somatic mutations as tumor cells, such as amplification of EGFR and chromosome 7. In their study, a fraction of GSCs displayed vascular endothelial-cadherin (CD144) expression, known to be characteristic of endothelial progenitors giving rise to mature endothelial cells. In vitro and in vivo lineage traces further support that the fraction of GSCs is multipotent and capable of differentiation along tumor and endothelial lineages, possibly via an intermediate CD133⁺/CD144⁺ progenitor cell. Interestingly, while treatments of GSCs with clinical anti-angiogenesis agent bevacizumab or knockdown shRNA demonstrate that blocking VEGF or silencing VEGFR2 inhibits the maturation of tumor endothelial progenitors into endothelium but not the differentiation of CD133⁺ cells into endothelial progenitors, γ-secretase inhibition or NOTCH1 silencing blocks the transition into endothelial progenitors [92]. These findings highlight that GSCs may not only depend on a perivascular niche for maintenance of their phenotype, but they may actually influence the establishment/maintenance of the niche.

The perivascular niche has also been shown to regulate GSC phenotype through regulation of the Notch pathway in these cells [93]. Blockade of this pathway has been demonstrated to deplete the GSC population through reduced proliferation and increased apoptosis as well as increase the sensitivity of GSCs to radiation induced cell death, underscoring the importance of Notch in the regulation of GSCs [92, 94–96]. Using a mouse model of PDGF-induced gliomas, it was shown that stem-like Nestin positive cells, which also expressed Notch1 and sGC, the nitric oxide (NO) receptor, were located near vascular endothelial cells that expressed the NO producer endothelial nitric oxide synthase (eNOS) [93]. Notch signaling in the stem-like cells was shown to be activated by NO in vitro and NO drove tumors in this system to have a more stem-cell like phenotype.

An additional molecule recently identified to regulate GSCs within the perivascular niche is integrin α6 (discussed above). In these studies, the authors identified integrin α6 expression adjacent to CD31 expressing endothelial cells in primary patient GBM specimens. They then went on to demonstrate the stem cell nature of integrin α6-expressing cells through in vitro analysis. Importantly, the ability of integrin α6 cells to form tumors in an orthotopic xenograft model was demonstrated. Interestingly, targeted inhibition of integrin α6 with RNA interference or with a blocking antibody reduced the stem cell phenotype and tumor formation. Although additional studies will likely elucidate additional pathways involved, these findings underscore the importance of perivascular niche regulation on GSC maintenance. In addition to the perivascular niche, the prevalence of hypoxic niches in GBM is being better understood as an important factor in GSC maintenance.

Hypoxia in Gliomas

Despite the recruitment of vasculature by tumors through the process of angiogenesis, the blood vessels that are feeding a tumor are often chaotic and disorganized [97]. Furthermore, the rapid expansion characteristic of glioma limits the diffusion of oxygen throughout the tumor mass. The resulting irregular blood flow creates regions within a solid tumor that experience cyclical periods of hypoxia, with oxygen tensions that range from mild (2–5% O₂) to severe (<1% O₂; [98]). As such, hypoxia is an essential part of the tumor microenvironment. While physiologic levels of oxygen in the body range from 1% to 7%, the oxygen level within a tumor can fluctuate to significantly lower levels. Gliomas, in particular, have characteristic regions of pseudopallisading necrosis that develop due to hypoxic regions (97).

Restricted oxygenation levels have been demonstrated to correlate with many aspects of tumorigenicity. Hypoxia is related to patient survival, therapeutic resistance, and tumor aggression in many solid cancers (Evans et al. 2010). Hypoxia is associated with resistance to traditional radiotherapies and chemotherapies that are used to treat GBM patients (98). For instance, resistance to temozolomide, the primary chemotherapeutic agent in clinical use for GBMs, has been linked to hypomethylation and increased expression of the DNA repair protein, O6-methylguanine-DNA-methyltransferase (MGMT). Hypoxia has been demonstrated to mediate MGMT expression and MGMT has been observed in hypoxic regions within gliomas [99]. These data suggest that restricted oxygen may be critical for the GSC response to DNA-damage inducing agents and cells that reside in hypoxic niches may be better suited to evade treatment.

Due to the chaotic vascular architecture and fluctuating oxygen state, it is very challenging to directly disrupt the hypoxic niche. A more effective approach would be to target the canonical hypoxia responsive signaling pathways. The hypoxic response in both tumors and nonmalignant tissue are mediated through the hypoxia inducible factors (HIFs). The HIFs are transcription factors that exist as heterodimers comprised of an alpha and beta subunit. The beta subunit (the most common is called aryl hydrocarbon receptor nuclear translocator or ARNT) is constitutively present in the nucleus. The alpha subunit (of which there are three isoforms) is oxygen labile and degraded very rapidly in the presence of oxygen. Under low oxygen conditions, the alpha subunit is stabilized, translocates to the nucleus, binds the beta subunit, and subsequently activates transcription. Although stable at low oxygen conditions, hypoxic cycling within a tumor has been suggested to promote the upregulation of HIF proteins to levels above those present in chronic hypoxic conditions (98). Hypoxic cycling also causes an increase in free radical generation, which is thought to also increase HIF protein translation [100].

HIF alpha subunits are regulated by the prolyl hydroxylase (PHD) family. In well-oxygenated environments, PHDs hydroxylate specific proline residues on HIFα subunits (100). The hydroxylation requires free iron and alpha-ketogluterate. Chemical mimetics of hypoxia (such as deferoxamine mesylate) function via PHD’s dependency on free iron. By chelating free iron in the cell, the alpha subunit cannot be hydroxylated and is then stabilized. However under normal conditions, this hydroxylation acts as a substrate for the E3 ligase, Von Hippel Lindau factor (pVHL). pVHL binds the alpha subunit and targets it for degradation by the proteasome.

Of the three isoforms, HIF1α and HIF2α are well characterized in terms of their function. HIF1α and HIF2α have overlapping, yet very distinct roles in biology [11]. For instance, HIF2α is induced at a physiologic level of oxygen (around 7%), while HIF1α is only responsive to more severe hypoxic conditions (<1% O₂). While HIF1α and HIF2α share 75% homology, HIF2α is regulated at the transcriptional level as well as by traditional PHD hydroxylation [101]) while HIF1α is only regulated at the protein level. The two HIF isoforms share several similar transcriptional targets, however there are specific genes such as Oct4 and TGFα that are regulated only by HIF2α [102].

Induction of anaerobic metabolism and increased angiogenesis are important tumor functions regulated by the HIFs. HIF1α induces anaerobic metabolism by promoting glycolytic enzyme and glucose transporter expression [100]. It also promotes glucose flux through a nonoxidative arm of the pentose phosphate pathway. HIFs also mediate tumor invasion and metastasis as well as the cellular resistance to oxidative stress [103].

In addition, in metastatic tumors such as breast cancer, HIFs upregulate the expression of genes that promote the process of endothelial-mesenchymal transition (EMT), including genes Twist1 and lysyl oxidase (LOX). LOX is an enzyme involved in the remodeling of extracellular matrix to facilitate EMT (100). Moreover, knockdown of HIF1α caused a reduction of tumor migration in vitro and also impaired the invasive tumor phenotype in vivo [104].

HIF mediation of angiogenesis is particularly interesting in GBM, which are classically tumors that are extremely vascularized. Vascular endothelial growth factor (VEGF) is a downstream target of HIF proteins, as are other pro-angiogenic factors such as angiopeitins[105, 106]. HIF induction will also lead to the downregulation of anti-angiogenic factors such as thrombospondin[107–109]. Viewing through mouse skin fold window chambers, Dewhirst and colleagues observed angiogenesis as a response to hypoxia by expressing GFP under the control of the VEGF HRE [98]. In their study, GFP induction was initially found in regions lacking extensive blood vessels. In glioma, it has also been shown that GSCs regulate angiogenesis through induction of VEGF [64]. The VEGF inhibitor, Bevacizumab, has been shown to be useful in clinical trials to decrease glioma growth [110]. The HIF signaling pathways in glioma present several interesting therapeutic targets. However, recent experimental evidence has demonstrated that the relationship between HIF2α and the GSC population goes beyond canonical signaling pathways.

Glioma stem cells, Hypoxia, and HIF2α

The role of hypoxia in normal stem cell biology is well established. Hematopoietic stem cells, for example, colonize hypoxic niches within the bone marrow, where they are maintained quiescent by hypoxia-induced proteins such as osteopontin [111, 112]. Severe hypoxia inhibits the differentiation of normal NSCs [113]. Also, hypoxic conditions reduce differentiation in human embryonic stem cell cultures while not affecting their proliferation [114]. Hypoxia is also known to improve the generation of induced pluripotent stem (IPS) cells [115].

Restricted oxygen has a significant role in GSC function. Hypoxia promotes the formation of neurospheres in vitro of both GSCs and non-stem cells [116]. In addition, stem cell genes such as Sox2 and Oct4 are upregulated in glioma cells under moderate hypoxia (5%). These data suggest that the hypoxic niche may act to maintain the GSC phenotype and may be able to induce changes in the functioning of GSC and non-stem glioma cells.

Current studies have revealed that both HIF1α and HIF2α are critical for GSC function. HIF1α is involved in maintaining the GSC population as stabilization of the protein expands the GSC population in a bulk tumor [14, 117, 118]. This effect is mediated in part by the PI3K/Akt and the Extracellular signaling related kinase (ERK) 1/2 pathways [14]. HIF1α knockdown depletes the self-renewal capacity of GSCs, as measured by neurosphere formation [14, 104]. HIF1α also appears to antagonize the effects of the bone morphogenetic proteins [119]. Unfortunately, HIF1α is critical for the normal function of neural progenitor cells as well as normal endothethelial cells, thus limiting its therapeutic index [11, 120].

In contrast, HIF2α has recently emerged as a potential therapeutic target for GSCs. Specific stem cells factors such as Nanog, Oct4, and Sox2 are upregulated under moderate hypoxic conditions at which only HIF2α is stabilized [116]. A recent study revealed that HIF2α is preferentially expressed in the GSC population [11, 15]. HIF2α is controlled at both transcriptional and post-translational levels whereas HIF1α mRNA is not. Specific knockdown of the HIF2α gene reduces self-renewal of the GSCs, as measured by the formation of neurospheres in vitro. In vivo, HIF2α and HIF1α knockdown increases the survival of mice bearing intracranial gliomas. In neuroblastoma, HIF2α is also upregulated in CSCs, which have a more immature neural crest-like phenotype, indicating a link between HIF2α expression and a more stem-like cell state [11, 15, 121]. In fact, later studies found an important role for hypoxia and HIF2α in the phenotype of GBM cells.

HIF2α can also promote a more tumorigenic phenotype in non-stem glioma cells. It has been previously shown that genes specifically regulated by HIF2α, such as Oct4, Serpin B9, and Glut1, are preferentially expressed in GSCs (11). Heddleston et al. demonstrated that expression of non-degradable HIF2α in non-stem glioma cells is able to drive the expression of several stem cell genes, including Oct4, c-myc and Nanog [13]. Furthermore, the non-stem cells exhibited morphological changes from adherent astrocyte-like cells to neurospheres. Expression of non-degradeable HIF2α also increased the ratio of GSCs to non-stem cells, as evidenced by cell surface marker expression. Lastly, HIF2α overexpression in non-stem glioma cells conferred tumorigenic potential in vivo in mouse flank xenografts. In addition to its role in maintaining the GSC population, HIF2α has a limited expression in normal cells and has very low expression in normal neural progenitors or other neural stroma [11]. This supports HIF2α as a promising and novel clinical target for glioma.

Because of the clear reliance of GSCs on HIF proteins for their survival, there is compelling evidence for HIF2α as a therapeutic target that could impair GSC responses to the hypoxic microenvironment thereby disrupting the niche. There are several approaches being explored to target HIF proteins. Drugs such as the aminoglycoside digoxin are currently used as a therapy for atrial fibrillation and other cardiac pathologies. In gliomas, digoxin has been shown to decrease HIF protein levels in vitro and to inhibit subcutaneous xenograft growth in mice [118]. However the caveat to broad spectrum drugs as digoxin is that they indiscriminately effect both HIF1α and HIF2α, thereby leading to off-target effects in normal neural progenitors. More effective pharmacological inhibitors are currently in development that specifically target HIF2α. First reported by the Iliopoulos group, the use of a chemical compound that is able to specifically bind iron responsive elements (IREs) in the HIF2α transcript results in effective inhibition of HIF2α protein translation [122]. This inhibition disrupted downstream HIF2α functions including promotion of angiogenesis in a renal cell carcinoma model. These data suggest that it is possible to develop specific inhibitors for HIF2a that could lead to effective clinical treatment and increased patient survival. However, targeting of transcription factors, such as HIF2α, has been difficult and development of HIF2α-targeted drugs is still a field in its infancy. As more studies examine the specific role of the HIFs in GSCs, it is becoming clear that hypoxia is able to influence cell phenotype on a more global scale by regulating cancer epigenetics.

Epigenetics

Cancer is thought to be a genetic disease and significant research efforts have been devoted to understanding how alterations in the genome can result in cancer [123, 124]. However, it was discovered that an additional level of regulation and complexity lie within the physical structure of the chromosome. Epigenetics, or the study of changes in cell phenotype that involve mechanisms other than the underlying DNA sequence, revealed what was required in addition to transcription factors for specific genes to be transcribed. It was found that during active transcription regions of the chromosome physically unwound in order to allow access to DNA sequences. DNA is further wound around proteins called histones. These histone-DNA complexes are referred to as nucleosomes. Histones were originally identified in 1884 by Albrecht Kossel but were thought to only be packing material for cellular DNA. It was not until 1990 that histones were appreciated as an additional mode of cell phenotype regulation [125]. Core histones are comprised of four subunit proteins: H3, H4, H2a, and H2b; two other histone proteins, H1 and H5, act as linker proteins between nucleosomes. Each core histone protein possess amino acid “tails” that are post-translationally modified and ultimately regulate the physical state of the histone and its associated DNA. Modifications include methylation, acetylation, phosphorylation, and sumylation. Histone methylation is the best understood modification, which is regulated by histone methyltransferases (HMTs) and demethylases (HDMs).

All histone-modifying enzymes recognize a specific substrate or amino acid on the core histone tail. Upon binding their target amino acid, catalytic domains add the post-translational modification. The modification then allows for changes in the nucleosome superstructure in order to permit or inhibit transcription factor binding of the DNA. For example, methylation of histone 3 at lysine 4 (H3K4) by SET-domain (Su(var), enhancer of Zeste, trithorax) proteins causes a separation of the nucleosomes which then allows for recruitment of transcription factors as well as additional modification on other amino acid moieties. In opposition, methylation of H3K27 causes a closure of open chromatin and inhibits DNA transcription. It is hypothesized that the balance of H3K4 and H3K27 methylation is a major contributor to whether a gene is actively transcribed [126–129].

Approximately 147 base pairs (bp) of DNA are wound around each set of core histone proteins. Upon modification of the histone amino tail, DNA can be released from the histone and accessed by transcription factors. However, specific base pairs within the DNA sequence can be epigenetically modified to prevent transcription as well. Specific families of proteins called DNA methyltransferases (DNMTs) recognize repeats of C-G base pairs (also referred to as CpG islands) and add a methyl group to the cytosine pyrimidine ring [130]. These methyl groups must be removed from the DNA in order for transcription to proceed; however no DNA demethylating proteins have yet been identified. The epigenetic phenotype is now recognized as a defining characteristic of cell state. Recent experimental evidence has described how modulation of histones and DNA methylation can control the pluripotency of cells in the body [131].

Induced Pluripotency

The pinnacle of regenerative medicine is to utilize a patient’s cells to generate new organs for replacement therapy. However somatic cells, which contribute to nearly all of an adult’s cellular makeup were thought to be terminally differentiated. Experimental evidence in the past five years has demonstrated that differentiated cells maintain the capacity to become pluripotent when stimulated with the appropriate factors. The Yamanaka group demonstrated that mouse and human fibroblasts were capable of induced pluripotency when transduced with stem cell factors Oct4, Sox2, c-Myc, and Klf4 [132–136]. These transcription factors promote expression of other stem cell genes and induced a pluripotent phenotype, creating a new cell type referred to as induced pluripotent stem cells (iPSCs). However efficiency of transformation is very low (<1%) [136]. Further studies revealed that the influence of the epigenetic state of the target cells play a vital role in determining efficiency. Microenvironmental factors, such as oxygen tension, were found to promote cell reprogramming largely due to the effect on histone modifications [115]. Subsequent studies have further suggested that reprogramming cells in the epithelial cell state increased efficiency. Induced pluripotency of fibroblasts required a mesenchymal-to-epithelial transition (MET). However reprogramming cells of epithelial origin significantly improved efficiency [137]. As MET likely involved chromatin remodeling, these studies suggest that changes in cell phenotype require concomitant modifications to the epigenetic phenotype. In the burgeoning field of GSC research, many similarities have been observed between GSCs and normal stem cells. Recent experimental evidence has demonstrated that non-stem glioma cells can be made tumorigenic following culture under specific conditions [13, 15, 16]. These data suggest that, like iPSCs, cancer cell epigenetics could play an important role in determining cell phenotype.

Epigenetic Modifiers in Glioma

The importance of CSC epigenetic pathways is not well understood. Initial studies focused on the role of DNA methylation in mediating gene transcription in glioma [138, 139]. These data revealed that many tumor suppressor genes were hypermethylated (inhibiting transcription) and pro-proliferation genes were hypomethylated (promoting transcription) at their DNA promoter sequence. For example, DNA repair gene O(6)-methylguanine-DNA-methyltransferase (MGMT) was found to be hypermethylated in many forms of glioma. This could serve as a prognostic indicator for the widely used clinical drug, Temozolomide, as sensitivity to the drug had been dependent on MGMT status. Cells lacking MGMT activity underwent apoptosis when treated with Temozolomide whereas cells that had high levels of MGMT were more resistant to treatment [140–142]. In general, gliomas that possessed low MGMT DNA methylation were resistant to DNA-damage inducing chemotherapeutics. Although DNA methylation status provided promising biomarkers of tumor treatment resistance, it is still unknown which proteins are responsible for methylating CpG islands on gene promoters.

The polycomb genes (e.g. Bmi1, EZH2 [Enhancer of Zeste-2]) repress transcription by methylating Histone 3 at Lysine 27 (H3K27). Previous studies have demonstrated the role of Bmi1 as a regulator of leukemic stem cell proliferation [143, 144]. Continuing studies have identified a significant role for Bmi1 in gliomas as well [145, 146]. However, the specific function of Bmi1 in glioma is not understood. In leukemia, Bmi1 signaling through Ink4a/Arf promotes cancer cell proliferation. Loss of Bmi1 or Ink4a/Arf can prevent tumor formation and cells lacking Bmi1 demonstrate reduced differentiation potential. Recent experimental evidence has demonstrated a particular importance of EZH2 in GSC self-renewal [147, 148]. Inhibition of EZH2 by short-hairpin RNA (shRNA) or pharmacologic disruption by S-adenosylhomocysteine hydrolase inhibitor 3-deazaneplanocin A (DZNep) caused a loss of in vitro self-renewal and in vivo tumor propagation by the GSC population. Although the complete mechanism is not known, it is thought that EZH2 silences c-Myc [81], a known GSC regulator [82]. Loss of EZH2 allows for unrestricted transcription of c-Myc, which then induces tumor formation. These data demonstrate that the function of epigenetic modifier, EZH2, in GSCs is a crucial regulator of the cell phenotype and suggest that the heterogeneous nature of glioma may be due in part to the function of epigenetic modifying proteins.

Opposite the polycomb proteins, the trithorax group activates transcription by methylating histone 3 at lysine 4 (H3K4). Trithorax proteins have been less well studied in the context of glioma but observations from leukemia indicate a potential role for these enzymes in promoting a GSC phenotype. In particular, the Mixed Lineage Leukemia 1 (MLL1) gene is involved in chromosomal rearrangements in a variety of leukemias [149, 150]. MLL1 encodes a large protein that includes a catalytic SET-domain [151, 152]. The loss of regulatory domains and aberrant downstream activity is common among MLL1 fusion proteins [153, 154] in leukemia, while alterations in MLL2 and MLL3 have been implicated in glioma [124, 155]. MLL1 along with its binding partners may regulate a large numbers of genes during embryonic development and hematopoiesis, however the number of target genes may be much smaller in adult tissues [152]. MLL1 function has been primarily investigated in leukemia models utilizing gene fusions that can aberrantly promote expression of tumorigenic target genes such as HoxA9 [156]. Little is known about MLL1’s wild-type function but a recent report demonstrated that DNA damage leads to MLL1 phosphorylation and checkpoint regulation [157]. These data implicate MLL1 and its family members as critical cellular regulators of normal stem cell biology (particularly in the brain) and leukemogenesis.

Lysine-specific demethylase 5B (JARID1B) is an epigenetic modifying protein that is localized to the CSC fraction within melanoma [158]. Recent studies demonstrated that JARID1B was present in a fraction of melanoma cells and sorting based on JARID1B expression enriched for a slow-cycling population. This population was able to give rise to highly proliferative and tumorigenic progeny. Furthermore, JARID1B expression patterns did not confirm to previously observed CSC markers. Increased JARID1B has been observed in other solid tumors such as prostate and breast cancer [159, 160], however its function in glioma is still unknown. Another epigenetic protein family, the jumonji-domain (JMJ) proteins have been shown to coordinate with JARID proteins and have also been implicated in neural cells and normal stem cells.

The JMJ-containing proteins are a family of histone demethylases that have been implicated in normal stem cell maintenance [161]. Recent studies have described the relationship between JMJ and JARID families and how they coordinate target gene occupancy in normal pluripotent cells [162, 163]. Additionally, JMJ family members have been shown to be required for demethylation of H3K27 and subsequent commitment of neuronal lineages [164, 165]. In several of these normal organ systems, the JMJ family is modulated by hypoxia [166–169]. These data suggest that functioning of specific epigenetic factors can be modulated by microenvironmental conditions. This is particularly interesting in the context of glioma, where clinical studies have demonstrated strong correlation of hypoxia to poor patient prognosis [170]. No studies to date have elucidated the contribution of JMJ proteins in neural malignancies. Furthermore, the already established role of hypoxia in maintaining the GSC population suggests that epigenetics could be critical for their function and further investigation is warranted. Considering the previous data that has demonstrated the intricate relationship between the microenvironment, tumorigenicity, and GSC phenotype, advances in clinical therapy may come from targeting the microenvironmental niche.

Opportunities for Targeting Glioma Stem Cells Via the Tumor Microenvironment and Secreted Factors

While the majority of novel molecular based approaches have focused on targets present on tumor cells, the ability of the microenvironment to promote GSC maintenance provides new avenues for therapeutic intervention. Malignant gliomas display increased angiogenesis as well as increased expression of VEGFs, supporting the creation of blood vessels through endothelial precursors [64]. VEGF is a protein secreted by tumor cells that promote endothelial cell survival, migration and proliferation by binding to specific high affinity transmembrane receptors expressed predominantly on endothelial cells. Studies demonstrate that using bevacizumab (Avastin), a neutralizing anti-VEGF antibody approved by the United States Food and Drug Administration for the treatment of GBMs, reduce GBM growth [110, 171, 172]. Studies have shown that GSCs generate highly vascularized tumors in immunocompromised mice in association with increased levels of VEGF [64, 173]. Treating endothelial cells with conditioned media from GSCs treat with bevacizumab inhibited the endothelial cell migration and tube formation (64). Treating mice bearing GSC initiated xenografts with bevacizumab or other anti-angiogenic agents (such as anti-SDF1 drug AMD3100) delayed tumor growth in vivo due in part to a decrease in tumor blood vessels as well as the percentage of GSCs [64, 173–176] However, several recent studies have called into question the long-term efficacy of anti-VEGF treatments [177, 178]. These studies have raised concerns that system anti-VEGF treatment may improve short-term patient outcome, but may ultimately lead to more aggressive malignancies. By “pruning” leaky vessels, anti-VEGF drugs like bevacizumab may in fact improve overall tumor vasculature. This could lead to increased invasion and more aggressive growth.

Another secreted factor known to support angiogenesis and can be produced by the tumor microenvironment is the cytokine interleukin 6 (IL6). IL6 production and signaling are highly correlated with tumor propagation as well as poor patient survival in many types of cancers, including GBM. Higher levels of IL6 protein are found in GBM samples in comparison to those of normal brains, and higher levels of IL6 mRNA are directly linked to poor patient survival [65]. Interestingly, within the neoplastic compartment the majority of IL6 is produced by the non-stem glioma cells whereas the IL6 receptors, gp130 and IL6Rα, are preferentially expressed on GSCs. These data suggest the importance of a paracrine loop within the glioma cell populations. However, IL6 can also be secreted by endothelial and immune cells and its expression can be stimulated by hypoxia, all of which are important components of the tumor microenvironment (65). Recent data demonstrate that directly targeting IL6 or IL6Rα by shRNA impairs GSC growth and survival in vitro, suggesting the significance of IL6 autocrine signals in GSC maintenance (65). Importantly, administration of anti-IL6 antibody delayed the growth of tumors initiated with GSCS. These data strongly suggest that targeting IL6 may be useful as anti-glioma therapies. The potential of anti-IL6 therapy is strengthened by data from clinical laboratories where anti-IL6 receptor antibodies have been approved for Rheumatoid arthritis treatment, which suggest that these therapies are tolerated well in patients [179].

Conclusions

Recent advances in clinical therapeutics such as the drugs Temozolomide and bevacizumab have been viewed as dramatic improvements in the treatment of GBM. However, these drugs still fail to significantly increase patient survival. The difficulty in effectively treating GBM is largely due to the heterogeneous nature of the disease. Although controversial, the GSC paradigm suggests that a subpopulation of glioma cells exist that exhibit enhanced proliferation, tumorigenicity, and the ability to evade typical chemo- and radiotherapy. The GSC population is thought to contribute to tumor recurrence. However, recent experimental evidence has called into question the hierarchical nature of GSCs in some cancer types (e.g. melanoma).

Several groups have demonstrated that microenvironmental conditions such as acidic stress and hypoxia maintain a stem-like phenotype. These data suggest that the GSC hierarchy contains a degree of plasticity, reminiscent of iPSCs. When faced with extrinsic stress, certain populations of non-stem glioma cells can modify their phenotype in order to ensure the continued propagation of the tumor. This has important ramifications when considering glioma treatment. Modulation of glioma cell phenotype increases the difficulty of effectively targeting the neoplastic population responsible for tumorigenesis. Even therapies that are purported to target GSCs may prove ineffective if non-stem glioma cells can then become GSCs at a later time post-treatment. In order to effectively treat GBM and other gliomas, the microenvironment must be considered. Novel therapeutics that inhibit the glioma cell-microenvironment interaction are vital for clinical efficacy. In addition to adjuvant chemotherapy, disruption of the microenvironmental niche would prevent glioma cell adaptation and could lead to better long term patient survival.

Abbreviations

GSC

Glioma stem cell

CSC

Cancer stem cell

GBM

Glioblastoma multiforme

HIF

Hypoxia inducible factor

ECM

Extracellular matrix

EGFR

Epidermal growth factor receptor

EGF

Epidermal growth factor

bFGF

basic Fibroblast growth factor