The cell of origin in brain tumors, including gliomas, is still a matter of debate [17]. However, there has been intense speculation that brain tumors may arise from neurogenic niches in the brain [18–27]. Since the discovery of adult neurogenesis, new insights have emerged about the mechanisms by which the brain maintains a small population of NSCs that can replenish both neurons and glia [28, 29]. Given the intrinsic ability of both BTSCs and NSCs to self-renew and differentiate, it is important to consider how cancer stem cells’ ability to regulate their own function in the tumor deviates (or remains the same) from that of normal NSCs. NSCs in the adult brain actively generate both neurons and glia that contribute to the brain’s cellular and functional homeostasis, as well as plasticity, remodeling, and response to injury [30–32]. In contrast to post-mitotic neurons, it is the cell types undergoing mitoses that are thought to harbor the greatest potential to give rise to brain tumors. Such mitotically active cells are found in neurogenic niches. Cells in the neurogenic niche perturbed by mutagenic events can theoretically serve as starting points for brain tumor formation and may harbor intrinsically higher oncogenic potential than other dormant neural cell types. Understanding neurogenesis and gliogenesis will allow us to explore concepts relevant to the cell-of-origin question in brain tumors and the regulation of BTSCs during tumorigenesis.

NSCs have been identified in at least two regions of the adult brain: the subventricular zone (SVZ) of the lateral ventricles, and the subgranular zone (SGZ) of the hippocampus [28, 33–35]. NSCs may also exist within the subcortical white matter [36]. During fetal life, radial glia, which are derived from neuroepithelium, are responsible for neurogenesis. Radial glia participate in neural organization; immature neurons that arise from radial glia move along their transcortical extensions for migratory guidance to their respective areas in the cortex, contributing to its stratified organization during the late embryonic stages [37]. With the transition to postnatal life, radial glia differentiate into many different cell types, including neurons, astrocytes, oligodendrocytes, ependymal cells, and the SVZ NSCs, which contribute to adult neurogenesis in the mammalian brain in postnatal life [28, 32, 34, 38, 39]. The adult SVZ NSCs, or type B cells, line the subependymal zone of the lateral ventricles. B cells give rise to intermediate progenitors termed transit-amplifying cells (or type C cells), which proliferate and have the ability to form immature neurons termed neuroblasts (type A cells). Neuroblasts migrate through the rostral migratory stream (RMS) to the olfactory bulb in rodents [40] and, additionally, through the medial prefrontal migratory stream in humans [41], eventually becoming mature post-mitotic neurons. Type B cells additionally give rise to oligodendrocyte precursor cells (OPCs) and astroglia depending on growth or inhibitory signals ([31, 42], see reviews [29, 43]).

In the hippocampal formation, the dentate gyrus SGZ houses radial astrocytes, which serve as a source of neurogenesis in the region [28, 33]. Radial astrocytes (type 1 cells) form intermediate progenitor cells (type 2 cells), which become immature granule cells (type 3 cells) and subsequently mature granule neurons [44–47]. The SGZ and SVZ house the two identified groups of stem cells responsible for formation of new neuronal and glial cell types in adult mammals. The majority of what we have learned about neurogenesis has emerged from studies conducted using rodent models. Some of these findings have been validated in postmortem human brain samples [30, 35, 41], although such studies are limited in number. The discovery of active neurogenesis in the adult human brain has numerous implications for the pathobiology of brain tumor formation, neurodegenerative disorders, and response to injury.

In the very beginning of brain tumorigenesis, the cellular composition is thought to differ drastically from that of a mature tumor. Given the known heterogeneity of fully developed tumors and the overwhelming possibility that the tumor began from a single cell type, an open question remains: How does this heterogeneity arise? To address this question, we will focus on gliomas, whose developmental biology is perhaps better understood than any other brain tumor.

We know that BTSCs from mature gliomas can phenocopy the tumors they were derived from in animal models. The BTSCs found in gliomas must have arisen from a tumor that lacked cells with stem cell properties or a tumor that arose from an endogenous NSC (Fig. 2.2) The cell of origin is defined as the cell type that has accumulated the correct combination of mutations that induces proliferation and eventually tumor formation (oncogenic transformation). Although not mutually exclusive, the cell of origin differs from the cell type that acquires mutations, which is referred to as the cell of mutation. This is because the cell of mutation, after acquiring the first oncogenic hit, may differentiate or dedifferentiate into another cell type before proliferating, so a distinction must be made for the cases where this happens (see reviews [17, 48]). An important concept to clarify is the difference between the cell of origin and BTSCs. Both can theoretically form the mature tumor, but the cell of origin is responsible for initiating the tumor and may or may not be a stem cell. The cell of origin is of interest in the discussion of BTSCs because evidence from mouse models has pointed to the possibility that NSCs within the brain may serve as the cells of origin. Given that BTSCs have stem cell properties, identifying the cell of origin may provide insight into how BTSCs are derived and how their ability to self-renew or differentiate differs from that of normal stem cells and the cell of origin. It is still unclear if multiple cell types within the brain can give rise to the same type of tumor and whether different types of tumors share the same cell of origin harboring different genetic mutations. To answer some of these questions, glioma models have been developed and used successfully to uncover key aspects of glioma biology.

Models of glioma aim to recapitulate human glioma pathology most commonly through two model systems: genetically engineered rodent models with mutations found in human gliomas, or xenografts of primary human glioma lines derived from patients. Other model systems also exist, including in vitro cultures of glioblastoma multiforme (GBM) cell lines, allografts of rodent glioma lines and virally mediated oncogenesis. These approaches have been used to provide reproducible platforms to study many aspects of tumor biology, including BTSCs, the cell of origin, and therapeutic implications [19, 21, 22, 49–54].

Primary lines derived from patient tumors carry a distinct advantage due to their genetic make-up, which most faithfully represents the disease [49]. The drawbacks of primary lines include: (1) the necessity to inject the mouse brain to create a xenograft, thus potentially altering the microenvironment; (2) the possibility that, in derivation of each culture, a subpopulation of the tumor is selected for; and (3) the fact that these tumors are grown in an immunodeficient background. Murine models have the disadvantage of a rodent genetic background and are limited in recapitulating the order and number of mutations that occur in a sporadic human glioma. Modeling efforts will likely continue to sample the phenotype produced by different combinations of mutations, location of the tumor, and the developmental point of induction of these mutations. In both cases, the tumor is grown in the mouse brain microenvironment, which raises an additional degree of separation from human glioma biology for both the mouse and xenograft model systems.

Experimental evidence from murine models of glioma suggests the cell of origin to be adult NSCs or proliferating progenitor/precursor cells, but not mature glia. However, this is highly controversial and remains an open question for the many different subtypes of glioma (see reviews [53, 55]). Following the discovery of adult neurogenesis, there was a paradigm shift in thinking about the origins of glioma, as the discovery of NSCs and their progenitor/precursor cell progeny became new candidates (Fig. 2.2) In a landmark effort to link the differentiation stage to tumor initiation, Holland et al. found that NSCs or neural progenitor cells expressing Nestin in the mouse brain were preferentially forming tumors with GBM characteristics after activation of the K-Ras/Akt pathway [25]. The same oncogenic insult did not produce tumors under the control of a GFAP promoter, suggesting that not all cell types within the same lineage could serve as the cell of origin [25]. Parada and colleagues have developed multiple tumor suppressor knockout models of gliomagenesis via inducible loss of p53, PTEN, and NF1, which are some of the most commonly mutated tumor suppressors in GBM [56, 57]. Analysis of the high-grade gliomas generated from these models revealed that Nestin-expressing cell types found in the SVZ are likely to contain the cell of origin. More recent studies with PDGF-driven tumor formation and p53/NF1 knockout have shown that oligodendrocyte precursor cells (OPCs) are capable of giving rise to GBM tumors in mice [20, 58].

Mouse models using Ink4a-ARF loss, K-ras activation, or PDGF signaling give rise to gliomas in areas and cell types found outside of the neurogenic niches, suggesting that mature glial cells can produce malignancy when given the correct combination of oncogenic mutations [54, 59–62]. Interestingly, it has also been reported that mature neurons, in addition to GFAP-positive astrocytes, are capable of acting as the cell of mutation in a p53/NF1 model of glioma by undergoing dedifferentiation [24]. Some consideration should be given to the fact that some of these murine models represent functional genetic alterations that may or may not be the initiating events in glioma formation despite their oncogenic transforming abilities in this context. It remains an open question as to what the initial events in the different subtypes of glioma are, and how restricted the cell of origin truly is for any given tumor type, considering the unique combination of microenvironment, genetic changes, and organism.

It should be highlighted that in many of the aforementioned murine models, there is a propensity of early events in murine gliomagenesis to occur near the SVZ when the NSC population is targeted [56, 57, 63]. There is clinical evidence in humans that initiating events in glioma formation occur in or near the neurogenic zones of the brain. Human GBM has a propensity to occur most frequently in the periventricular area and less so in the surrounding cortex, albeit this evidence is controversial [63–65]. GBM also occurs infratentorially, but with much lower frequency [66]. The tendency for GBM to occur in the periventricular area within the cerebrum suggests that a cell-of-origin also resides within the same region; however, this correlation has not been directly linked to human neurogenesis. It is possible that tumors found far from neurogenic regions may be initiated by migrating cells that originated in the neurogenic niche. This is an interesting but unexplored hypothesis.

Grade II/III gliomas mutated for isocitrate dehydrogenase (IDH) tend to arise in different anatomic locations as compared to their grade IV GBM counterparts. Although IDH1 normally functions to convert isocitrate to α-ketoglutarate, the mutation leads to the production of oncometabolite 2-hydroxyglutarate (2HG) which stereo-chemically resembles α-ketoglutarate and is hypothesized to cause tumorigenic epigenetic changes [67–70]. In IDH-mutated gliomas, the cytosolic variant IDH1 is most frequently mutated, whereas mutations in IDH2 can be found less commonly [71]. Strikingly, the IDH1-mutated gliomas are most commonly found in the frontal lobe in an area that overlaps with the rostral and medial migratory streams used by neuroblasts to replenish interneurons in the olfactory bulb and frontal cortex, respectively [30, 63]. Nevertheless, IDH1-mutated tumors can also be found in other regions of the brain, albeit at a lower frequency. The fact that low- and high-grade tumors tend to arise in different anatomic locations raises the possibility of differing cells of origin. Alternatively, it may signify that IDH1 mutations promote gliomagenesis only in restricted cell types or lineages. The lack of mouse models that produce IDH1-mutated tumors has inhibited the dissection of the cell of origin in this class of tumors [72–74].

Despite the possibility that the cell of origin may originate from a stem cell or a more restricted progenitor/precursor cell, there is evidence that the tumor either gains (via dedifferentiation) or maintains a portion of its population as cells with stem cell properties. The continuum between the cell of origin and BTSCs is not well understood (Fig. 2.2) Murine models have allowed the detection and study of BTSCs in the context of gliomas and medulloblastomas primarily. Much of our understanding of BTSC biology has derived from the study of primary human GBM.

Functional similarities between BTSCs and NSCs directed researchers to analyze the expression of markers that were shown to be important for NSC biology and neural development. One of the best-defined molecular markers for brain tumors, including pediatric tumors, ependymoma, and especially GBM, is the cell surface marker CD133. CD133 is a pentaspan, glycosylated transmembrane protein. Apart from being expressed in fetal brain NSCs during embryonic development, its expression is highly associated with other tissue-specific stem cells and cancer stem cells of blood and solid malignancies [14, 81–85]. CD133-knockout murine models show photoreceptor degeneration, but the signaling functions of CD133 remain unknown [86]. In the context of GBM and medulloblastoma, it was shown that when injected into animals in limiting dilutions, CD133+ cells generate tumors more efficiently than their CD133− counterparts, suggesting that they have BTSC properties [87]. Furthermore, downregulation of CD133 via short hairpin RNA (shRNA) suppresses self-renewal of BTSCs in GBM [88].

Although CD133 is one of the best-studied markers in brain tumors, it is now well established that some CD133− cells do possess tumorigenic potential, suggesting that CD133 is not a universal marker and that CD133− BTSC populations exist as well [11, 89–91]. Furthermore, the fact that not all GBM tumors have CD133+ cells supports the hypothesis that CD133− BTSCs exist in these tumors [92–94]. Another important marker associated with BTSCs is the intermediate filament protein Nestin, a well-established NSC marker. Nestin + cells were shown to have tumor-initiating ability in animal models and to generate tumor recurrence after chemotherapy [10, 61, 95]. BTSCs are also enriched for other NSC and stem cell markers, such as Nanog, Musashi-1, Bmi-1, Sox2, and Oct4 [96–98]. In addition, some BTSCs were shown to express other surface markers and transmembrane proteins, such as CXCR4, integrin α6, SSEA-1/CD15, L1CAM, and A2B5 [11, 99–101]. Finally, the side population (SP), defined as cells with the ability to extrude Hoechst dye via ABC-type transporters on the cell surface on flow cytometric analysis, has been shown to contain stem-like cells in a variety of brain tumors [102, 103].

Besides from traditional coding genes, the importance of noncoding RNAs in the regulation of gene expression has been increasingly recognized in recent years, making them important biomarkers in cancer biology. In particular, microRNAs are responsible for the posttranscriptional fine-tuning of gene expression by binding the 3′ UTR of messenger RNAs (mRNAs) and causing their translational arrest or degradation. MicroRNAs have been implicated in the regulation of stem cell self-renewal and differentiation, as well as the control of cell cycle and apoptosis [104].

MicroRNAs that are upregulated in gliomas are mostly associated with antiapoptotic, pro-proliferative, pro-invasion, and antidifferentiation pathways [105, 106]. On the other hand, some microRNAs, which are important for neural differentiation, were shown to be downregulated in gliomas, functioning as tumor suppressors [107]. An example of a microRNA critical to brain tumor biology is miR-124. Normally, miR-124 is known to promote neural differentiation in the brain. In gliomas, however, it was shown to be downregulated and its overexpression promotes differentiation [108, 109].

Many other microRNAs, whose targets include mRNAs encoding important survival and oncogenic molecules, such as PI3K, AKT, EGFR, and MAPK, were shown to play important roles in glioma and medulloblastoma [105, 108, 110].

Dissecting major signaling pathways that are important for BTSC biology have been, and will continue to be, a challenge due to the complex interplay and cross talk between different signaling pathways. Besides molecular markers shared between NSCs and BTSCs, signaling pathways are also conserved between these two populations. This conservation has highlighted several signaling pathways in BTSC biology, some of which are described below.

NSCs were shown to reside within a vascular niche, adjacent to endothelial cells, which are believed to provide signals required for self-renewal [143]. BTSCs were also shown to reside closer to endothelial cells [144]. In medulloblastoma, CD133+ cells where shown to be in proximity to endothelial cells [121, 122]. Similarly, glioma BTSCs acquire a vascular niche, in which CD34+ endothelial cells present Notch ligands to BTSCs, keeping them in an undifferentiated state via activation of Notch signaling [122]. However, the complex architectural features of GBM suggest that BTSCs may also reside in relatively avascular microenvironments.

The importance of hypoxia in promoting self-renewal in embryonic stem cells and NSCs suggests that it may also regulate the self-renewal of BTSCs, especially in GBM, which is a highly hypoxic tumor. When considering this possibility, the question that arises is whether there are hypoxic areas within GBM. One such plausible histologic area is represented in pseudopalisading necrosis (PPN), in which densely packed cells surround necrotic regions [145]. The etiology and biological significance of PPNs is not well understood. However, others and we speculate that they may represent areas of active tumor growth and revascularization. A tantalizing hypothesis to explain such tumor growth and angiogenesis is, in turn, the presence of BTSCs within a hypoxic niche. Indeed, some studies have shown enriched CD133 immunoreactivity in PPNs, supporting this hypothesis [141]. Importantly, the putative presence of BTSCs in hypoxic areas devoid of blood vessels raises doubts about the effectiveness of systemic drug delivery methods.

Invasion of glioma cells through the brain parenchyma represents perhaps their most malignant feature [146]. Single GBM cells have been shown to infiltrate normal brain tissue and travel more than several centimeters from the bulk of the tumor [147–149]. After surgical resection, the recurrent tumor occurs within the borders of the resection cavity, suggesting that these infiltrating cells have the capacity to regenerate tumors.

Although the mechanisms of invasion of BTSCs are not clear, C–X–C chemokine receptor type 4 (CXCR4) and its ligand, stromal-derived factor-1α (SDF-1α), which are highly important for vascularization, were shown to be crucial for the invasive behavior of GBM cells [140, 150]. Enrichment of SDF-1α/CXCR4 expression in glioma BTSCs highlights their invasive properties. This signaling was also shown to mediate recruitment of BTSCs toward endothelium, leading to further invasion and differentiation. Furthermore, this chemoattractant signaling induces endothelial proliferation via attracting tumor cells and inducing VEGF expression in gliomas and other systems, such as the gastrointestinal tract [151].

Because of their central role in promoting growth and recurrence of primary brain tumors, BTSCs are major candidates for targeted therapeutic approaches. In particular, signaling pathways promoting BTSC self-renewal and inducing therapy resistance represent critical drug targets.