The most common primary brain tumors are derived from genetic anomalies in glial development in the brain, comprising of astrocytoma and oligodendroglioma [1, 2]. These are graded on a scale with increasing malignancy as lower grade astrocytoma (WHO I and II), anaplastic astrocytoma (WHO III) and glioblastoma multiforme (GBM; WHO IV), also termed glioblastoma (GBM). GBM is the most aggressive of these tumors and ranks among the most desperate of all human cancers. Survival of GBM patients, although individually variable, has improved from an average of 10 to only 14 months after diagnosis in the last 5 years, in spite of significant improvements in the standard care with more targeted therapy (reviewed in [3–5]). In addition to several clinical and histopathological parameters, as well as specific therapeutic approaches, patients' responses to therapy are heterogeneous. This is related to insufficient data for prognostic and survival stratification. More information about molecular fingerprints of tumors as well as to better understanding of cellular origin of the tumor-initiating cells will help to develop individualized therapies for patients.

The classification of malignant gliomas is changing rapidly, and novel schemes are being developed based on their genetic landscapes [6] from next generation sequencing [7, 8], followed by epigenetics, proteomics and oncogenic miRNA data [9]. These are superimposed on the histological origin of brain tumor-initiating cells (BTIC) and their precursors evolving into glioma stem-like cells, also termed glioma stem cells (GSC). Thus, in about 90% of cases, GBMs arise de novo (primary GBM) without any evidence of less malignant precursor lesions. Secondary GBMs progress from low-grade diffuse astrocytoma or anaplastic astrocytoma [2], occurring in younger patients and having a lesser degree of necrosis and significantly better prognosis. Although primary and secondary GBMs are nearly indistinguishable in their histology, they differ in their genetic and epigenetic profiles providing the state-of-the-art GMB classification. Decisive genetic markers for secondary GBMs are iso-dehydrogenase 1 (IDH1) mutations, which are absent in primary GBMs and associated with a hypermethylation (particularly of histones) phenotype. IDH1 mutations are the earliest detectable genetic alterations in precursors of low-grade diffuse astrocytomas and in oligodendrogliomas. The former acquire further TP53 and ATRX mutations, whereas the latter lose the 1p/19q arm and are mutated in CIC and FUBP1 genes. In contrast, primary GBM glial progenitor cells overexpress epidermal growth factor (EGF) receptors and exhibit TP53 and PTEN mutations as wells as LOH 10p and LOH 10q deviations. Therefore, recent classification of glioma is based on the tumor metabolome and epigenetic changes. As an example, methylation of the O-6-methylguanine-DNA-methyltransferase (MGMT) promoter silences expression of this gene by 40% in the primary tumor, and more than 70% in the secondary GBM [7]. As this enzyme acts by de-methylating alkyl groups, epigenetic silencing of its expression increases GBM sensitivity to treatment by DNA-alkylating agents, such as commonly used temozolomide (TMZ) [10].

GBMs are histologically similar and featured by atypical mitotic nuclei often associated with necrosis and/or angiogenesis. The standard treatment against GBM includes surgery, radiotherapy and, in the last 5 years, chemotherapy with TMZ [11]. However, since a population of the invasive cells is dispersed through surrounding normal tissue, and GBMs reveal resistance against radio- and chemotherapy, the effectiveness of the multimodal treatment is limited, resulting in GBM reoccurrence.

The observed diversity of molecular fingerprinting might originate from the genetically different glioma initiating cells [3] also called cancer stem-like cells, here termed glioblastoma stem cells (GSCs). Similar cells have been found in many, although not in all human tumors. These “roots of cancer” operate in a hierarchical fashion such as conceptualized for normal stem cells [12]. The authors of this work reviewed the evolution of the cancer stem cell (CSC) model, which is going back to the 19^th century to Virchow's embryonic-rest hypothesis, where cancer is a consequence of the activation of ‘dormant’ stem-like cells in adult tissues. The basic understanding of the current cancer stem cell theory is that heterogeneity within the tumors is not a mere consequence of random mutations and selection, leading to well recognized clonal evolution of cancer [13], but results from an intrinsic hierarchy of tumor cells [14–16]. CSCs have unlimited capability to divide symmetrically and asymmetrically. Asymmetric cell division occurs for originating progenitor cells together with a reduction in the differentiation potential. The theory of a hierarchical model of tumorigenesis has been widely accepted, thus stating that only a small fraction of tumor cells - the CSCs, are capable of initiating tumor growth or renewing the tumor in the same or other organ after incomplete surgical removal [17, 18]. When injected orthotopically, CSCs are capable of forming tumors in animal. Likewise, CSCs can directly or indirectly contribute to generation of metastasis [19]. However, the existence and identification of these cells remains an open question, as all the existing in vitro and in vivo using animal models in fact represent artificial environments. Consequently it cannot be ruled that the so named CSCs are just cells which have adapted to altered growth conditions.

The above-described CSC model has a “static” hierarchical structure, largely based on the notion that normal stem cells undergo oncogenic transformation. However, in the past few years a number of publications have challenged this concept by demonstrating that in the tumor environment progenitors, differentiated cells and even transformed tumor cells may acquire the ability of self-renewal through their de-differentiation, e.g. a reversal of differentiation [20, 21], reviewed in [22], or by other mechanisms, discussed in the following [23]. Moreover, the finding that epithelial to mesenchymal transition (EMT), which is a reversible process and can transiently induce stem-like cells phenotypes, was crucial to expand static the CSC model to the fluid CSC model [12], also termed the CSC plasticity model. This model proposes CSC plasticity, meaning that by dedifferentiation of a variety of cell types within growing tumor mass even various populations of CSCs may evolve by acquiring additional mutations or epigenetic modifications. Several authors propose that various CSC subclones can exist as the result of evolutionary pressure of therapies, giving rise to more aggressive secondary CSCs [24]. Taken together, the clonal evolution and CSC hierarchical theory merge into a so called fluid plasticity or convergence CSC model. This results in several subclones of CSC within the same tumors, depending on epigenetic and mutational events in specific cells with self-renewal capabilities. These secondary aggressive CSCs may become dominant and drive tumor formation or simply co-exists with other phenotypes of CSCs. Such concepts represent challenges in order to develop novel therapeutic approaches that would more efficiently target the cancerous clonal population with stem cell-like characteristics.

Recent research has revealed that GBM stem-like cells play important roles in GBM pathogenesis. The first evidence that GBM display a cellular hierarchy with self-renewing and tumor-initiating cell types located at the apex was provided by Singh et al [25, 26] and Galli et al [27] followed by further relevant studies on this topic (reviewed in [28–30]. The origin of GBM stem cell has been widely discussed and is now based on the fact that genome analyses revealed the transcriptional fingerprints of neural stem cells (NSC), their progenitors, as well as of mesenchymal cells [31, 32]. Tumor initiating cell based classification comprises four subtypes, such as classical, mesenchymal, neural and proneural transcription profiles, as summarized by Van et al. [3]. These GBM stem-like cells (GSCs) self-renew, their hijacking molecular mechanisms and expressing markers of NSCs [7, 33]. Over the past decade, a wide range of studies have shown that several signaling proteins involved in neural development also play important roles in GBM pathogenesis, as highlighted and discussed in the context of developing treatments for GBM [34]. GSCs express the intermediate filament protein nestin, the transcription factors Sox2 and Oct 3/4, the RNA-binding protein musahi, transcriptional repressor Bmi1 of the polycomb of proteins, proteins of the Notch signaling pathway receptor, aldehyde dehydrogenase 1 (ALDH1) and among a few others also the extensively studied transmembrane protein, prominin-1 (CD133) (reviewed in [35, 36]). It has been suggested that the resistance of GBMs towards radiation and chemotherapy is attributed to the GSC-like population [28, 37, 38]. Hence, GSCs are proposed to persist as a distinct population involved in tumor relapse and propagation and significantly impact the poor prognostics of the GBM patients. Accordingly, GSCs have been pursued as a promising target for the development of novel specific diagnosis/prognosis therapies against GBM [39].

Although GSCs are increasingly studied for molecular mechanisms, immunophenotyping or targeting purposes, the GSC markers, functional assays and the culture conditions remain controversial challenging further investigations [40]. Different methods of GSC enrichment have been proposed, either from GBM cell lines and/or primary cultures [41], resulting in GSCs with rather different phenotypes. Thus, the enrichment of GSCs by itself may represent an artifact and contribute to misidentification of these cells and to the discrepancies in their identification and characterization.

GSCs can be also identified within side population (SP) assays using flow cytometry. This SP cell population is obtained by the efflux of Hoechst fluorescent dyes revealing stem cell-like characteristics. This dye-efflux, presumably due to the exclusion of the dye triggered by an ATP binding cassette (ABC) transporter, was also identified in cells from several tumors exhibiting stem-like genes, tumorigenicity and multi-drug resistance [49, 50]. In the C6 glioma cell line, SP cells are described as enriched in stem-like cells [51]. To date, the SP is employed as gating criteria for GSC enrichment by flow cytometry with following demonstration of self-renewal, multi-lineage differentiation and tumorigenicity [52] or decreased migration [53]. However, as a drawback to GSC enrichment and purification based on Hoechst exclusion, recent evidences indicate that, in fact, the GBM SP cells do not contribute to self-renewal or tumor initiation and thus is not enriched in GSC and not tumorigenic [54] but consists of brain endothelial cells [55]. In fact, recent evidence shows that the three present ABC transporters ABCA1, MRP4 and MRP5 are differentiation instead of stemness markers [56]. Taken together, SP flow cytometry is not recommended as GSC enrichment methodology.

Since the first experimental studies supporting in vitro self-renewal and capacity of GSCs to differentiate into neuronal and glial cells, GSC enrichment has been essentially performed in vitro using defined medium. Originally, GSCs were identified by their capacity to form three dimension cellular clusters, termed tumorspheres, similar to neurosphere cultures of NSCs and NPCs [101, 102]. Spheres culture is considered to keep the undifferentiated state of the cells by minimizing stimulation from the environment, such as adhesion to substrates, but also avoiding the access of the cells in the core of the sphere to differentiation factors [103]. Classically, GSCs arive in adherent cultures with defined medium following 20 to 40 days after plating [25, 27, 104, 105]. Based on this, tumorspheres formation has been obtained faster through anchorage-independent culture, such as in dishes previously treated with poly-Hema (poly2-hydroxyethyl methacrylate) [105, 106], gelatin [83] and agar [107]. In fact, tumorigenicity correlates with suspension growth [108]. In addition, three-dimensional culturing models are also available, including gelatin foam cultures, for the study of GSCs [109]. However, special care should be taken regarding suspension cultures [110], since poly-Hema derived spheres might not be truly three dimensional clone derived, but rather reflect cell aggregates. In fact, the formation of spheres as predicative property of CSC enrichment has been questioned, as no correlation was found with the expression of GSC markers [111]. Indeed, the technique of tumorsphere cultivation presents some limitations, such as differentiation and cell death, occurring in the sphere environment [112]. Alternatively, GSC enrichment has also been performed using adhesive cultures, which would provide uniform access to growth factors, thus avoiding differentiation [113, 114]. Phenotypic differences are observed within adhesive surfaces [115], such as in dishes previously treated with matrigel [116], laminin [113], collagen [117, 118] or extracellular matrix [82]. On the other hand, organotypic cultures in the presence of ECM and vascular elements are also available, especially for migratory assays, as this model represents a more accurate brain matrix micro-environment, in which cells migrate [119].

GSC enrichment performed in adhesive as well as non-adhesive cultures supports a major role for medium composition in this process [113, 118]. GSC culture medium, similar to that for NSCs, is supplemented with 20 ng/ml of EGF, 20 ng/ml of FGF-2 and sometimes with the 5 ng/ml of leukemia inhibitory factor (LIF). The mitogenes EGF and FGF are for long known for inducing renewal and expansion and notably increasing the frequency of spheres formed by undifferentiated cells [120, 121]. On the other hand, LIF has also been shown to enrich GSCs based on the transforming growth factor-beta (TGF-β)-mediated promotion of symmetric expansions inducing self-renewal and preventing differentiation [122], also increasing the expression of core stemness transcription factors and thus the number of NPCs [123]. Furthermore, medium supplemented with B27 and/or N2 enhances proliferation of multi-potent NPCs [124].

Interestingly, the basis behind the use of the defined medium is that self-renewal and multipotency of stem-like cells would be induced and sustained with a supplement medium switch from serum to EGF and FGF [125, 126]. EGF induces proliferation of adult mouse brain NSCs, generating spheres responsive to FGF-2. Then, upon stimulation by FGF-2, undifferentiated cells proliferate undergoing symmetric or asymmetric divisions and give rise to either unipotent (neuronal) or bipotent (neuronal/astroglial) progenitors [127]. The EGF-responsive mammalian embryonic neural precursor is a stem cell [121]. Later, other evidences supported that FGF-2 responsiveness was sufficient to isolate progenitors found in the adult mammalian spinal cord [128] and that NSCs responsive to FGF-2 arise earliest in embryonic development [129]. GSC marker expression was increased upon cultivation stimulation with FGF-2 compared to EGF [130]. In fact, blockade of EGF receptors reduced sphere growth, even in the absence of EGF in the culture medium. Both EGF and FGF enhanced GSC proliferation and sphere size. However, the absence of both mitogens in GBM culture was already demonstrated to originate multipotent spheres that were capable of self-renewal and also forming highly invasive tumors [131].

Lowering oxygen concentrations promotes GSC enrichment. Although the standard normoxic condition uses 20% of oxygen, this oxygen level is considered as hyperoxic compared to physiological brain oxygenation. In fact, hypoxia is the physiological environment of adult brain tissue, ranging from 1% to 5% of oxygen [132, 133]. Solid tumors including GBM are less oxygenated, ranging from physiological levels to below 0.1% in necrotic areas [134, 135]. Hypoxia regulates gene expression of hypoxia-inducible factor 1 (HIF-1), a transcription factor composed by an oxygen-regulated HIF-1α and a constitutive HIF-1β subunit. HIF-1α has a half-life of few minutes and destabilizes in increasing oxygen levels by the action of proline hydroxylase which utilize O₂ in the range of 0.1 to 21% as a cofactor [136]. HIF post-translational modified within several domains interacts with the von Hippel-Lindau (pVHL) followed cytoplasmic translocation and ubiquinin-mediated for proteasome degradation [137, 138]. On the other hand, HIF-1α stabilized in hypoxic conditions translocates to the nucleus interacting with co-activators, such as cAMP, and binding to the hypoxia response element located (HRE) in the promoter or the enhancer regions of target genes [139]. To date, three HIF isoforms have been described, including HIF-2α and HIF-3α.

HIF-1α and HIF-2α expression was increased in GSCs exposed to hypoxic conditions together with other stem cell markers such as CD133, Bmi-1 and nestin [140]. Whereas HIF-1α is ubiquitously expressed, including by all hypoxic tumor cells, HIF-2α participates in cell reprogramming towards a cancer stem cell phenotype maintaining the GSC. HIF-1α target genes are regulated in a tissue-specific fashion. More than 100 downstream genes were identified as regulators of multiple physiological responses to oxygen deprivation, in general shifting the metabolic balance from oxidative phosphorylation toward glycolysis. HIF-1α is required for the proliferation, survival and angiogenesis [126]. Overstabilization of HIF-1α supports GBM resistance to TMZ [141]. Moreover, HIF-1α interaction with the activated Notch intracellular domain attenuates differentiation-promoting effects of bone morphogenetic proteins (BMPs), in particular of BMP-2. In NSCs and NPCs, increasing O₂ levels degrade HIF-1α, thus promoting differentiation or apoptosis [142]. In GBM, knock down of HIF-1α at the RNA expression level reduced migration/invasion phenotypes beyond the ability to form tumorspheres [143]. A proposed mechanism is that HIF-1α drives the initial response to hypoxia whereas HIF-2 α drives the chronic response. The switch from HIF-1α- to HIF-2α-dependent signaling pathways plays divergent, however complementary roles during the hypoxic response reprograming GBM cells towards stemness, increasing aggressive tumor growth and invasiveness [144, 145]. However, HIF-2α stability is unaffected in physiological oxygen levels. Moreover, evidences support that HIF-2α is essential only by GSCs expressed and not by NPCs [146]. HIF-2α is overexpressed by GSCs acting as a mediator of tumor plasticity and tumorigenesis [126, 146]. A underlying mechanism could be that HIF-2α upregulates expression of stem-related genes such as Oct4 [147].

Remarkably, hypoxic microenvironment contributes to the GSC phenotype and increases the number of stem-like cells in tumor populations. Low 1% oxygen level enriches the percentage of CD133-expressing cells [148] without affecting N-glycosylation of this epitope [149]. A culture in hypoxia for 48 hours resulted in a four-times increase of the number of CD133⁺ cells [150]. Hypoxia-mediated HIF-1α activation results in a time-dependent increase in CD133 mRNA and protein levels with the peak of protein expression after 48 hours [149, 150]. The enrichment of GSCs from GBM primary cultures or the cell line U-87 under 1% oxygen condition results in tumorspheres smaller than the ones obtained in conditions of 20% oxygen. In agreement, expression of Ki-67, a marker of proliferation, is reduced in conditions of 1% oxygen, whereas expression of stemness markers, such as CD133, podoplanin, Bmi-1 and nestin, is increased. Sox-2 expression, in particular, is increased only in tumorspheres from primary cultures [140]. In terms of enrichment, 1% of oxygen increases the stem-like cells over fivefold with the percentage of CD133-positive cells being threefold or more enriched. In the same line, oxygen levels reduced to 7% enhance the stem-like phenotype of CD133-positive cells [151]. In addition to promoting proliferation, hypoxic conditions induced angiogenesis known to be a prerequisite of tumor expansion. Angiogenesis is favored by the association of GSCs and endothelial cells in perivascular niches [152]. Regulatory functions of such niche on GSCs is proposed in analogy to the ones in the control of NSC proliferation [153].

The effects of hypoxia on in vitro cell culture have gained scientific interest. Precise control of oxygen levels is essential for accurately interpreting obtained results. To date, hypoxic cell culture might be performed by different strategies. The most widely used incubator is based on a chamber flushed with 1% O₂/5% CO₂/94% N₂ [154]. However, this chamber suffers frequently from the leakage of hypoxic conditions. As a main drawback of this approach, every time the incubator door is opened, a certain amount of oxygen needs to be consumed for re-establishing initial conditions. Alternatively, a hypoxic environment might be created by putting the cell dishes inside a subchamber with oxygen controller, which allows studies exploring multiple oxygen levels for each chamber to be placed inside a regular cell culture incubator. However, this alternative also implies in discontinuous oxygen levels and difficulties of reproducing exact hypoxic conditions.

For improving experimental conditions, hypoxic glove boxes emerged allowing cell manipulation without affecting oxygen levels. However, as the incubation and manipulation cells in the same humid station implies risk of contamination of cultures, the glove box station allowing hypoxic manipulation might be enclosed in a hypoxic incubator. Since exposure to room air also occurs, disrupting hypoxic condition during regular microscopy, a hypoxia microscope chamber should be added the workstation. In this sense, appropriate modular interconnected closed-hoods accommodating microscopes, centrifuges, cell sorters, bioreactors are receiving more attention as these ensure unprecedented stable and continuous hypoxia, completely isolated from the outside environment. In parallel, easy to use and low-cost alternatives have been developed, as, for instance, inflatable chambers made of transparent plastic materials allowing real-time studies of hypoxia under microscopy [155]. Finally, as the oxygen level is typically controlled in the gas phase differs from that supplied to the cells submerged in the growth medium; pericellular oxygen level monitoring might be used for better accuracy. This can be achieved with sensor chips embedded in a conventional tissue culture flask [156].

Vascularization is a diagnostic hallmark of GBM. At least five interlinked pathways by which GBM achieves neovascularization, have been described. A step of vascular co-option between brain and tumor vasculature, is followed by angiogenesis as second step. Vasculogenesis is the third mechanism sustained by the differentiation of circulating bone marrow-derived cells (BMDCs), known as endothelial progenitor cells (EPCs). These cells are suggested to promote a supportive role and might be incorporated into tumor vasculature. The fourth mechanism of neovascularization, termed vascular mimicry, is defined as the ability of tumor cells to form functional vessel-like networks. The fifth mechanism is based on the trans differentiation of GBM cells into an endothelial phenotype [157]. However, tumor cells involved in vascular mimicry are endowed with stemness plasticity, providing a source for transdifferentiation of GSCs towards the endothelial phenotype. The presence of GBM-derived vasculatures highlights the plasticity of GSCs, although the underlying mechanisms remain to be elucidated [158, 159].

Hypoxia is involved in all steps of the neovascularization process. HIF activation has been described to up-regulate VEGF expression and thus to promote angiogenesis [160], but also to increase levels of SDF-1, which in turn promote vasculogenesis recruiting CXCR4 positive BMDCs to the tumor site. However, unlike in normal vascularization, the emerging vasculature is often abnormal. In these pathological angiogenesis, the excess of vascular proliferation together with the lack of structure-giving pericytes contributes to the formation of tortuous and dilated blood vessels that are poorly organized and hyperpermeable (reviewed in [161]). Hence, vaso-occlusive and plasma coagulation impairs the blood supply perpetuating the hypoxic microenvironment, which in turn regulates HIF expression and activity thus leading to a vicious circle of induction of abnormal vasculature growth in GBM [162]. At the same time, hypoxic exosomes secreted by GBM cells induce endothelial cells to liberate growth factors and cytokines accelerating tumor growth [163].

In this context, GSCs are maintained within the vascular niche, in which the endothelial cells play an important role to sustain the GSC-like phenotype and GBM propagation. The endothelial secreted factors, which activate self-renewal pathways, including Hedgehog [164] and the Notch pathway [165, 166], at the same time sustain mTOR activation, which in turn suppresses autophagy and apoptosis of GSCs [167]. Both, endothelial cells and GSCs, express nestin and the cell surface antigen CD133 [168]. Almost 75% of nestin⁺ cells co-express CD133 while less than 0.1% of nestin⁺ cells coexpress the endothelial marker CD34 [116]. In addition, GBM SP cells, characterized by their efflux properties using flow cytometry were identified as brain endothelial cells [55].

Co-culture of GBM and endothelial cells results in a GSC population with higher proliferation rates and subsequently formation of bigger spheres and more aggressive tumors than observed with GSCs in monoculture [116]. On the other hand, co-culture of GSCs and endothelial cells also augments number and proliferation of endothelial cells [126]. The majority of GSC-enrichment studies employing co-culture conditions were done with immortalized human brain endothelial cells (hCMEC) [167, 169], umbilical vein endothelial cells (HUVEC) [164], microvascular endothelial cells (HMVEC) [126], primary endothelial cells [116] and primary bovine endothelial cells [117]. In general, trans-well inserts are used to co-culture endothelial and GSCs ensuring free diffusion of signaling molecules without direct cell contact [116, 117, 126, 164]. Alternatively, conditioned media from 72 h endothelial cell culture in serum-free medium might also be employed [167]. Endothelial cells cultivated in three-dimensional system enhanced secretion of IL-8, upregulating IL-8 cognate receptor CXCR1 and CXCR2 activities with the consequence of enhanced GSC migration, growth and stemness properties [169].

Pericytes are perivascular contractile stromal cells that surround the wall of endothelial cells regulating blood flow [170]. In GBM, it was demonstrated that pericytes mediate co-option of modified pre-existing blood vessels supporting the expansion of the tumor margin [161]. Besides, pericytes present mesenchymal stem cell (MSC) phenotypes, exhibiting neural multi potential activity [171, 172], also mediating immunosuppression effects on GBM. [173]. Interestingly, although GSCs have been suggested to differentiate into endothelial cells, recent evidence indicates that GSC preferentially differentiate into pericytes, attempting to support vasculature function and tumor growth [174, 175]. Besides, endothelial cells have been proposed to promote paracrine stimulation of pericytes, specifically in the context of hypoxia. In this context, hypoxic exosomes secreted by GBM cells enhance pericyte vessel coverage and induce endothelial cells to secrete factors stimulating pericyte PI3K/Akt pathway activation and migration [163]. Endosialin (CD248), not expressed by endothelial cells, is a marker of closely associated pericytes in GBM [176].

One of the initially described characteristics of GSCs is tumorigenicity in vivo [26]. Considering that GSCs present a distinct capacity for tumor growth and propagation, these cells have been enriched by serial passages in immune-compromised animals. The enrichment of human GSCs in vivo is based on tumor xeno-engrafting into immunosupressed animals (generally mice and rats). The congenitally athymic and hairless nude mouse immunodeficient in functional T cells has been routinely used in tumorigenesis assays and shown to be suitable for GSC transplantation due to its impaired capability of rejecting human xenografts tumors. On the other hand, the severe combined immunodeficient (SCID) mouse carries a genetic defect preventing functional development of T and B cells due to dysfunction in the TCR and BCR recombination machinery. However, the SCID mouse retains normal numbers of natural killer (NK) cells [177], which might interfere in the rejection of xenografts, as NK cells promote selective lysis of particular tumor targets [178]. An appropriate model for analysis of GSCs has been created by the inclusion of the non-obese diabetic background (NOD-SCID) with reduced NK cell function. Properly, injection of anti-NK cell antibodies before transplantation enhances the engraftment of human xenografts [179]. Followed by the NOD/SCID/γ c^null (NOG) strain, with the above-noted deficiencies including cytokine production incapability, the NOG strain has been considered a more appropriate animal recipients and especially suitable for GSC transplantation [180].

Intracranial GSC grafting has been widely performed using stereotactic approach, whereas few studies have explored subcutaneous or intraperitoneal engrafts. Engrafting cells into a mouse brain is a critical step as the site of injection determines the location of the tumor bulk and requires special care for in fact reaching the brain tissue [181]. Alternatively to stereotactic procedures, an implantable guide screw was developed to establish intracranial xenografts in mice, allowing a large number of animals to be engrafted [182]. Later, the guide screw method was modified incorporating an infusion pump that allows up to 10 animals to be simultaneously intracranial injected with tumor cells [181]. In general the number of intracranial injected cells was 10⁵ cells per animal, although tumor formation was also observed with injection of 2 × 10² to 10³ cells after 12 weeks [183]. Injection of 5 × 10³ to 10⁴ cells resulted in tumor establishment after 8 weeks. Tumor formation in vivo, a functional method for defining GSCs, is widely used since for enrichment of these cells. Cells are replicated by serial passages re-injecting them into secondary mice [26, 159], which also serve as a validation assay. Xenograft tumor excision later allows further histological and molecular characterization.

Tumor cells have less requirements for proliferation, in general being less independent from growth factor stimulation and anchorage when compared to normal cells. However, not every cell within the tumor present the same independence in order to survive and propagate. On the other hand, GSCs are considered to be capable of clonogenic proliferation as anchorage independent spheres under serum withdrawal, as similarly observed for NSCs. Also similar to NSCs, the frequency of GSCs has been predominantly assessed by primary spheres assays. The studies on GSCs usually refer to GBM primary tumors cultured with defined medium supplemented with EGF and FGF. Ignatova et al. [106] cultured tumor cells using non-adhesive dishes previously treated with poly-Hema and detected after 18 days a yield of GSC frequency from 0.05% to 1.26% (n = 10) in the GBM population. Moreover, the authors of this paper reported that the size of formed tumorspheres is heterogeneous (<50 cells or 50 to 2000 cells), and some tumors do not present no clonogenic cells at all. In another study published by Galli et al. [27], tumor cells cultured on adhesive surfaces for 20 to 40 days yield a GSC frequency of 0.5% to 31% (n = 12).

Another work reports that GSCs could by isolated following only 7 days of culture on adhesive surfaces with a yield of nearly 20% based on FACS separation of CD133-positive cells [184]. On the other hand, a subpopulation of 2–5% of CD133⁺ GSCs isolated by marker formed spheres after 42 days in culture whereas no sphere formation was observed in CD133⁻ cells by Beier et al [104]. The work of Wang et al. (2010) agreed with the observation that CD133⁻ cells were not capable of forming tumorspheres, while one of CD133⁺ cells actually contributed to tumorsphere formation. Another study resulted in the observation that small spheres (∼50 cells) were frequent and that 5% of plated cells were able to form secondary spheres, with around 70% of the sphere-forming cells being CD133⁺ [104]. Qiang et al. [105] confirmed that CD133⁺ cells presented higher colony-formation capacity including more G₀/G₁ phase cells.

GBM cell lines cultured using non-adhesive surfaces and presenting no ALDH activity (ALDH⁻) do not yield tumorspheres whereas GBM cell lines presenting ALDH activity (ALDH⁺) exhibit tumor sphere formation capacity [83]. A recent study supports this observation as ALDH⁺ cells isolated by FACS from primary GBM cultures on adhesive surfaces, previously treated with extracellular matrix, yield tumorspheres after only 1 day whereas ALDH⁻ cells did not. Regarding hypoxic conditions, tumor cell culture in 1% of O₂ for 5 days increased almost by three-fold the capacity of the tumor cells to form spheres [150]. Furthermore, hypoxia affects spheroid diameters and numbers forming tumorspheres. Sphere diameters were in average 50 μm in 1% O₂ and 100 μm in 7%. However, hypoxia favored a higher proliferation rate [140]. On the other hand, conditioned media from immortalized human brain endothelial cells yield larger tumorspheres size [169].

Neurosphere assays for quantification of NSCs in vivo [185] remain an useful and low-cost tools for assessing GSC populations. However, the capacity to form tumorspheres has been criticized as a prediction for CSC enrichment [111]. Tumorsphere formation is not predictive for CSC expression markers, at least not for all types of cancer cells. Pastrana and co–workers critically evaluated the sphere-formation as an assay for stemness and pointed out that this assay may not detect quiescent stem cells. In contrast, the proposed assay detects cells with either tumorigenic properties in vivo or already dividing ones [186]. It is noteworthy that the source of multipotencial NSCs relies on a relatively quiescent cell and not on the proliferating population [187]. For the determination of self-renewal capacity, cultures should be plated as single cells to ensure that tumorsphere formation is only due to truly clonogenic proliferation instead of cell fusion and/or aggregation, which might occur even at very low culture density. On the other hand, as cell survival and proliferation are directly affected by paracrine stimulation, very low culture density conditions suffer from poor dispersion of signaling molecules in the medium [186]. In this case, a good option to improve the sphere-formation assay is to plate cells with previously conditioned and filtered medium. Additionally, a software specially suitable for comparing populations depleted and enriched in stem cells is available [188].

GSCs are usually functionally defined by their tumorigenic property in immunosupressed animals. As self-renewal is a key feature of stemness, GSC identity relies on serial transplantation capability. Singh and co-workers showed that after 5 weeks of reinjection of 1,000 CD133⁺ cells in NOD-SCID mouse brains (n = 3) the generated tumors recapitulated the phenotype of the original patient primary tumor whereas CD133⁻ cells took 12 weeks after transplantation to exhibit tumor formation [26]. Of note, the first sign of neurologic impairment in mice was observed 8–10 weeks after transplantation of GSCs. Galli and co-workers observed that GSCs gave rise to tumors that mimic GBM, which did not occur in the U-87 MG cell line. Moreover, GSCs presented migratory capacity typical for GBM, which was also not observed with U-87 MG cells [27]. In contrast, Pollard et al. [113] demonstrated tumor engraftment using a minimum number of 100 cells, whose isolation was not based on CD133 expression, but on exploring adhesive culturing. Indeed, Qiang et al. pointed out that seven weeks of culture on non-adhesive surfaces generated larger subcutaneous tumor xenografts than compared to those obtained in cultures on adhesive surfaces [105]. Compared with CD133⁺ cells enriched in the presence of primary endothelial cells, xenografts were generated within 7 days of injection of 1,000,000 cells [116]. Co-culture with endothelial cells increases tumor formation by a factor of six compared to the monoculture. After 11 weeks, animals intracranially injected with 100,000 cells formed tumors with a size of 110 mm³, whereas previous co-cultures with endothelial cells resulted in 600 mm³ tumors [169]. In mice, co-injection of GL261glioma and b.END3 endothelial cells resulted in larger allografts when those observed following transplantation of glioma cells alone [164].

Alternative methods for animal experimentation include GBM growth in semi-solid medium employing agar or agarose suspension [189, 190]. Agarose-suspension culture is for long known to be one of the best and low cost in vitro assays correlating with tumorigenicity [108, 189, 190], still used for GSC validation [191–193]. Alternatively, GSC enrichment has also been performed using agarose as non-adhesive surface [194].

GSCs can be validated based on expression patterns of classical NSC markers, correlating with the progression of WHO grades astrocytomas [195]. Among them, the transcription factor Octamer 4 (OCT-4), a marker of stem cell pluripotency [196], has been employed. OCT-3/4 promotes migration and invasion of GBM [197], whose knockdown was described to result in CSC apoptosis [114]. In stem cells, OCT-4 expression was shown to be upregulated by HIF-2α during hypoxia, increasing its mRNA level four times [150]. However, in another studied OCT-4 expression could not be detected in GSC, although performed with GSC enriched in adhesive cultures previously treated with laminin [126]. On the other hand, SRY (sex determining region Y)-box 2, also known as SOX-2, which is a encoding group of transcription factors, is also found in NPCs and are also used as GSC validation marker [83]. Musashi encodes for various RNA-binding proteins which contribute to the maintenance of stemness during CNS development and are also applied in the validation of the GSC phenotype [83, 150]. Nestin is an intermediate filament protein specific for glia, usually applied as a multipotent marker in NSC characterization and also in the validation of the GSC phenotype [198]. Nestin-knockout embryos revealed reduced self-renewal with no overt defects in cell proliferation or differentiation, surprisingly uncoupled from nestin's structural involvement in the cytoskeleton [199]. Nestin has been found to be highly expressed in invasive GBM xenografts delineating tumor infiltration [200]. However, as neurogenesis precedes gliogenesis during development, the more undifferentiated stem cell-like phenotypes are thus negative for nestin expression and found nestin⁺ later [112]. In line with this observation, nestin⁻ mouse NPCs gave raise to neurons and glia [201]. In addition, 75% of nestin⁺ cells in GBM coexpressed CD133 [116]. However, in GSCs after 5 days of differentiation with serum, nestin expression levels were still at a high level, whereas CD133 was much lower expressed [105]. CD133 is the most used marker for GSC phenotype validation. However, a controversial discussion remains about the suitability of CD133 expression for validating and targeting GSCs. Beier and co-workers supported evidences that CD133⁺ GSCs maintain only a subset of primary GBMs [104]. At the same time, CD133⁻ derived GBM cells formed tumors in nude rats beyond the efficiency of CD133⁺ cells [61]. In addition, CD133 has been presented as marker of bioenergetic stress in GBM rather than for being a marker protein of their stem-like properties [202]. However, considering that surface phenotype of stem cells may remain intact despite decay in functional activity, a more functional assay would be appropriate, as those employing ALDH expression. According to Choi and co-workers ALDH⁺ cells range from 0.3–28.9% in GBM primary cultures. Furthermore, near 40% of ALDH⁺ cells overlapped with CD133 expression whereas only near 0.3% of CD133⁺ cells exhibit ALDH activity [82].

GSCs are proposed to be chemoresistant, a feature explicating GBM recurrence after conventional therapy [38]. Attempting to predict this chemoresistant phenotype, in vitro chemotherapeutic drug assays assess the chemosensitivity of CSCs [194]. Temozolomide (TMZ) is a chemotherapeutic alkylating agent included in the standard therapy against GBM. Although experimental data demonstrate that TMZ preferentially depletes GSCs [203], and that primary GBM cell response to therapy occurs in patient-specific fashion and independent of GSC phenotype [204], the majority of studies regarding the susceptibility of GSCs points at a chemoresistant phenotype of GSCs towards alkylating agents [205]. TMZ induces apoptosis and senescence in GBM cells cultured as tumorspheres, a phenotype also used for GSC validation [206]. In fact, conversion of differentiated GBM cells into GSCs could take place upon treatment with TMZ [207].

Recently, TMZ was described to down-regulate P-glycoprotein expression, subsequently promoting survival of GSCs by the Wnt3a/glycogen synthase-3 kinase/β-catenin signaling pathway [208]. In line with this observation, BMP2-induced differentiation sensitized GSC to TMZ therapy [141]. However, GSCs are not uniformly resistant to TMZ. In fact, CSCs are neither resistant nor susceptible to chemotherapy per se. DNA repair mechanisms restore the integrity of alkylated DNA bases thus contributing to drug resistance and subsequent tumor reoccurrence.

O6-methylguanine-DNA-methyltransferase (MGMT), a DNA repair enzyme, possesses detoxifying functions during TMZ treatment, when the MGMT gene is non-methylated, conferring a strong intrinsic resistance to GSC [209]. However, MGMT methylatio stauts alone sill does not predict the TMZ response with high precision as atypical TMZ-resistant GSC presented both methylated and unmethylated forms [210]. TMZ resistance of GSCs via regulation of MGMT expression is promoted by c-Jun N-terminal kinase (JNK) and Ras-Raf-MEK-ERK signaling pathways, whose inhibition enhances cytotoxicity of TMZ on GBM [211–213]. In addition to MGMT, another DNA repair protein, ALKBH2, was shown to mediate TMZ resistance in human GBM [214], however its specific effects on GSCs remain unknown. Recently, microRNA-125b has been shown to play a role in the resistance of GSCs against TMZ through down regulation of Bak1 and PIAS3 (protein inhibitor of activated STAT3) expression [215, 216]. In view of that, detection of microRNA-125b expression levels may be useful for validation of the GSC phenotype.

GSCs are postulated to recapitulate the heterogeneity of the parental tumor in immunosupressed animals due to their multipotency capacity. As the differentiation property is a key feature of stemness, differentiation assays attempt to evaluate the functional GSC phenotype. Similar to assessing stemness in NSCs, the assay is done by cultivating GSC under pro-differentiation conditions. Thus, upon induction of differentiation, cells are going to express specific genes and proteins of the three cell types of neural lineage (neurons, oligodendrocytes and astrocytes). These results suggest that initial cells have stem cell-like characteristics in vitro. Differentiation progress is usually evaluated based on expression determination of cytoskeleton-associated proteins such as neuron-specific β3-tubulin (TuJ1) and MAP2, or glial cell-specific GFAP [217]. Moreover, glycoproteins, postulated to determine stem cell fates, are differentially expressed during differentiation, and used to determine the differentiation status of GSCs [218]. The loss of the CD133 epitope, ABC transporters, and ALDH-1 also are evidence for CSC differentiation [56, 60, 73].

The glial differentiation of GSCs, evidenced by increased expression of GFAP, can be induced by all-trans retinoic acid (RA) GSC cultures were cultured in growth factor-free medium [70]. Similarly, BMPs promote also astro-glial differentiation of GSCs [219]. Interestingly, BMPs were already demonstrated to sensitize GSC for TMZ by affecting HIF-1α stability and MGMT expression [141]. Lee et al. [217] showed that GBM-derived GSCs cultured with N2 supplement, RA or 10% fetal bovine serum alone, within a month progressively differentiate losing their NSC markers nestin, SSEA-1 and Sox2 and developing morphologies and immunohistochemical staining patterns (GFAP, TuJ1 and MAP2), consistent with phenotypes of glial and neuronal lineages. Finally, in vivo experiments with immunosupressed animals provide a proper environment for differentiation of GSCs, recapitulating GBM mass with a variety of cell types [26]. Therefore, since differentiation property is a key feature of stemness, this process might be applied in the validation of GSC phenotypes.