Cancer Stem Cells

The cancer stem cell (CSC) model has provided a new paradigm in understanding what predominant cellular mechanisms drive tumor growth. Similar to the organization of growth processes within normal tissue such as bone marrow or intestinal epithelium, the CSC concept postulates the presence of a functional and cellular hierarchy within the heterogeneous tumor body. Within this hierarchical model, the vast majority of the tumor bulk comprises rapidly dividing, partially or terminally differentiated cells with limited replicative potential, while neoplastic growth is driven by a small population of cancer cells with stemlike properties. Similar to the properties of normal stem cells, CSCs represent a population of neoplastic cells that have the capacity to initiate and maintain tumors and are characterized by self-renewal, resistance to chemical insults and radiation, and ability to produce new tumors even after periods of dormancy.

To be defined as CSCs, the cells must have the following characteristics: (1) self-renewal and proliferation, (2) multilineage differentiation into mature fates resembling tissue of origin, and (3) capacity to form new tumors resembling the original tumor. To date, studies have used similar procedures to identify and characterize the stemness of putative CSCs. A differentially expressed marker or set of markers is used to isolate a small subset of tumor cells within the tumor bulk. The capacity to form nonadherent tumorspheres by limiting dilution in culture conditions permissive for stem cell proliferation is assessed to demonstrate properties of clonal expansion and self-renewal and is used to identify and expand potential CSCs. In vivo demonstration of CSC behavior is confirmed by the ability of putative CSCs transplanted into immunodeficient mice to form tumors resembling the original tumor.

Gliomas

Initial evidence suggesting the presence of brain tumor cells with stemlike potential came from the isolation of a clonogenic subpopulation of cells from resected gliomas through neurosphere assays. These neurospheres possessed the capacity for self-renewal and demonstrated intraclonal neuronal and glial cell lineage heterogeneity. Among the several markers associated with CSCs, CD133 is one of the most widely studied in brain tumors. Singh and colleagues first identified a subpopulation of brain tumor cells within medulloblastomas and gliomas coexpressing CD133 and nestin, a marker for undifferentiated neural stem cells. Unlike the CD133 ⁻ population, CD133 ⁺ cells, representing 0.3% to 25.1% of the tumor bulk, were capable of neurosphere formation in serum-free media following addition of the stem cell growth factors bFGF and EGF. These tumorspheres had greater proliferative capacity relative to neural stem cell controls and formed secondary tumorspheres with immunoreactivity for nestin and CD133. Tumorspheres resembled a more primitive state and were devoid of glial fibrillary acidic protein (GFAP) and β-tubulin III, markers of differentiated neural cells and glial cells, respectively. However, CD133 ⁺ cells were capable of multilineage differentiation into oligodendrocytes, astrocytes, and neurons and expressed differentiated markers mirroring that of the parent tumor. Although many xenograft assays require up to 10 ⁶ tumor cells for successful engraftment, as few as 100 purified CD133 ⁺ tumor cells have been reported to be sufficient for intracranial tumor formation when transplanted into NOD-SCID mice. In contrast, injection of 10 ⁵ CD133 ⁻ cells failed to form tumors. Engrafted CD133 ⁺ GBM cells produced tumors with classic GBM features that were phenotypically identical to the patient’s original tumor and could be serially transplanted. In addition, engrafted tumors were widely infiltrating and comprised differential expression of CD133 and GFAP, suggesting that CD133 ⁺ cells were capable of self-renewal and differentiation in vivo.

Subsequently, 2 groups identified tumorsphere-forming subpopulations within tumor specimens obtained from patients with primary and secondary GBMs. Spheres arising from less than 1% of all GBM cells coexpressed nestin and CD133. Unlike the progeny of normal neural stem cell (NSC)-derived neurospheres, differentiation of GBM tumorspheres led to progeny that closely mirrored the parental tumor phenotype, comprising 80% β-tubulin III ⁺ cells and 25% oligodendrocytes. In addition, tumorspheres demonstrated greater proliferative potential and the ability to retain stemlike features following differentiation, suggesting a role in the maintenance of the stem cell pool and production of differentiated cells within the tumor bulk. Unlike the non–sphere-forming population, the sphere-forming GBM cells were able to form tumors following transplantation in nude mice, and the new tumors expressed both neural and glial markers. Expression of CD133 is restricted to a subpopulation of tumor cells positive for the neural stem and progenitor cell marker nestin, which suggests that this subpopulation may represent a less-differentiated subset of tumor cells. Consistent with CSC model, which proposes that CSCs represent a rare fraction of tumor cells, the extent of CD133 expression within gliomas is typically low or barely detectable, as demonstrated by quantitative analysis using flow cytometry. However, some immunohistochemical reports have suggested that the CD133 ⁺ glioma fraction comprises up to 25% of the tumor bulk.

Recently, transcriptome profiling of purified CD133 ⁺ and CD133 ⁻ GBM cells identified a CD133 gene expression profile comprising 214 differentially expressed genes. Computational comparison with established stem cell and cancer cell profiles demonstrated a close association of the CD133 gene expression signature with that of human embryonic stem cells. The CD133 gene signature distinctly differentiated between NSC-like GBM cells cultured in stem cell medium and GBM cells cultured in serum. Enrichment of the CD133 gene signature was closely associated with increasing glioma grade, with greatest resemblance in grade IV GBMs. The CD133 gene signature was associated with a more aggressive GBM subtype and significantly shorter median patient survival. In addition, the GBMs enriched for the CD133 gene signature were associated with a greatly increased number of genetic mutations. Overall, the CD133 gene signature, obtained from sorted CD133 ⁺ populations, is characteristic of a stemlike cell population of tumor cells as well as more clinically aggressive and hypermutated subtypes of GBM cells.

There exists a functional hierarchy within the heterogeneous GBM cell population consisting of a small CD133 ⁺ fraction of GBM cells that formed tumorspheres in stem cell media, generated serial tumorspheres on dissociation, was enriched for several stem cell markers (nestin, Musashi 1, and SOX2), and gave rise to progeny-expressing neuronal, oligodendrocytic, and astrocytic markers on culturing under prodifferentiation conditions. Consistent with in vitro features of stem cells, CD133 ⁺ GBM cells demonstrated a heightened capacity for proliferation, self-renewal, and multilineage differentiation. Transplantation led to the formation of highly invasive and angiogenic tumors that histologically and morphologically resembled the original patient GBM. Altogether, these findings suggest that the CD133 ⁺ GBM subpopulation contains putative stemlike cells that can initiate and maintain tumors that phenotypically resemble the original GBM and are capable of driving tumor growth even when challenged with serial transplantations.

Limitations

The variable protocols used in the purification of CD133 ⁺ cells and the assessment of stemness make comparing study results difficult. Tumor samples have been obtained from a variety of sources (eg, fresh resections from patients, glioma cell lines ) and cultured in variable conditions, including serum-free media and in the presence of fetal calf serum. Because serum has been shown to induce differentiation, culture media should be taken into account when assessing results from studies examining CD133 expression.

Furthermore, many of the CD133 studies use AC133 monoclonal antibodies, which recognize glycosylated CD133 epitopes. Accurate detection of CD133 may be limited because of the unknown specificity of AC133 binding to differential glycosylation patterns of the 8 potential glycosylation sites on CD133. Furthermore, studies have characterized several different tissue-dependent isoform tissue produced by alternative splicing of the CD133 transcript within human and murine models. It is possible that AC133 monoclonal antibodies, unable to uniformly recognize different isoforms, would provide an underestimate of true CD133 protein expression levels within gliomas. Recently, Osmond and colleagues reported that GBM cells found to be AC133 ⁻ contained a truncated and possibly functional CD133 variant that was localized within the cytoplasm. It is unclear whether the isoform type, the extent of protein expression, or the specific glycosylation status is the more biologically relevant marker of CD133 expression. Tissue distribution of CD133 messenger RNA is much more prevalent than expression of AC133, as the downregulation of the AC133 epitope is independent of intracellular levels of CD133 messenger RNA. Thus, the presence of AC133 epitopes is not necessarily equivalent to CD133 expression.

Although transplantation of sorted cells into immunodeficient mice remains the gold standard of assessing tumorigenicity, it has several limitations. The process of sorting and transplanting tumor cells into mice subjects the cells to several procedural insults and places them in a foreign environment vastly different from the original tumor niche. Inherent differences between the patient and mice microenvironment due to various species barriers (eg, differences in cytokines and growth factors important in tumor growth) may drastically alter proliferation potential. In addition, the xenographic immune response in mice receiving transplants is significantly more robust than the native immune response within the patient. Fundamentally, the issue is the applicability of tumor xenograft assays in immunocompromised mice in studying the clinical behavior of human cancers.

The Multiple CSC Model

Recent data have suggested that CD133 ⁺ and CD133 ⁻ tumorigenic glioma cells may represent distinct CSC populations associated with different GBM subtypes. Following the initial characterization of CD133 ⁻ glioma cells with stemlike properties, Lottaz and colleagues compared gene expression profiles in various GBM CSC lines and identified a 24-gene signature that faithfully distinguished 2 distinct subgroups of GBM CSC cells. Compared with a previously established gene signature differentiating different GBM subtypes, type I CSCs showed a proneural transcriptional profile, whereas type II CSCs displayed a mesenchymal profile. Type I CSCs resembled fetal NSCs and were strongly CD133 positive, whereas type II CSC lines resembled adult NSCs and were mainly CD133 negative. In accord with previous studies reporting different growth capacities and extent of stemlike phenotypes between the 2 CSC types, CD133 ⁺ type I CSCs formed tumorspheres in culture, whereas type II cells displayed semiadherent growth. Key molecular differences among the 2 types included differences in TGF-β/BMP pathway activation and expression of extracellular matrix and adhesion molecules. Compared with their presumed cells of origin, both CSC types demonstrated heightened metabolic and proliferative activity as well as greatly impaired differentiation capacity. As suggested in previous studies of genetically engineered medulloblastoma and glioma models, tumorigenic CSCs are derived through distinct mechanisms from cells that have either preserved or gained (eg, trough dedifferentiation) features reminiscent of fetal NSCs and adult NSCs. The findings, corroborated by an earlier array-based classification system by Gunther and colleagues, suggest that CD133 ⁺ and CD133 ⁻ CSCs represents at least 2 different GBM subtypes with distinct molecular and phenotypic characteristics.

The Hierarchical Model

Other studies have demonstrated the ability of CD133 ⁻ glioma cells to give rise to both CD133 ⁺ and CD133 ⁻ progeny. Intracranial transplanted glioma spheroids obtained from human GBM biopsies in nude rats gave rise to invasive CD133-negative tumors with minor signs of angiogenesis. Successive serial transplantation of these initial tumors gave rise to tumors with increasing CD133 immunoreactivity, which was closely correlated with an increased angiogenic phenotype and decreased survival in the host rats. Transplantation of CD133 ⁻ cells, purified via FACS of the serially transplanted tumors, into nude rat brains gave rise to tumors with both CD133 ⁺ and CD133 ⁻ cells. Tumors derived from CD133 ⁻ GBM cells contained up to 5% CD133 ⁺ cells, suggesting that a subset of CD133 ⁻ cells can not only initiate and support tumor growth but also recapitulate the initial tumor heterogeneity.

Similarly, Chen and colleagues demonstrated that both CD133 ⁺ and CD133 ⁻ GBM cell fractions are capable of neurosphere formation and also demonstrate varying degrees of tumorigenicity in vivo. Accounting for all clonogenic GBM cells within neurosphere cultures, the investigators proposed 3 categories of GBM CSCs organized within a lineage hierarchy representing different stages of differentiation. CD133 ⁻ type I cells give rise to a mixture of CD133 ⁺ and CD133 ⁻ cells, CD133 ⁺ type II cells also give rise to a mixture of CD133 ⁺ and CD133 ⁻ cells, but CD133 ⁻ type III cells only give rise to a population of self-renewing CD133 ⁻ cells. All 3 cell types expressed the NSC marker nestin, were capable of multilineage differentiation, and formed tumors following serial transplantation in mice. Much of the histologic and molecular differences between the 3 types place types I and III at 2 extremes, whereas type II cells possessed intermediate features. Although CD133 ⁻ type III cells were restricted to only producing CD133 ⁻ tumors, type I cells could produce type I, II, and III cells. Tumors derived from type III cells grew the slowest and gave rise to well-circumscribed tumors, whereas type I cells were more elongated and generated more aggressive and invasive tumors with diffuse borders. Consistent with the observed histologic characteristics, type I and II clones expressed significantly higher levels of the radial glial marker, FABP7, among several other NSC markers in comparison with type III clones. FABP expression has been established as a GBM marker for increased invasion and shorter survival and is associated with maintaining the stem cell features during neural development. In contrast, type III grafts displayed higher levels of intermediate progenitor markers such as TBR2, DLX1, DLX2, and CUTL2. Within this model, type I and III CD133 ⁻ cells represent the most primitive and differentiated states along a spectrum. Taken together, CD133 ⁻ progenitor cells are capable of forming both CD133 ⁺ and CD133 ⁻ cells, supporting the presence of a lineal hierarchy of self-renewing cells that support GBM growth.

Enrichment of neurosphere-forming GBM cells have identified 2 lineally related but distinct populations of CD133 ⁻ cells (type I and III) with unique gene signatures and in vivo phenotypes representing differing stages of differentiation. These differences within the CD133 ⁻ population, comprising both type I and type III CD133 ⁻ cells, may account for the discordant results of previous studies examining the ability of CD133 ⁻ cells to demonstrate in vitro or in vivo stem cell properties.

The Dynamic Model

In addition, the cancer stem–like phenotype associated with CD133 expression may actually represent a dynamic and plastic trait that responds to the changing signals and stresses. Oxygen has been the well-characterized signaling molecule involved with mediating various signaling pathways and regulating gene expression, and availability of oxygen within the tumor microenvironment has been thought to influence the proliferative phase underlying neoplastic growth. Although oxygen tensions within the normal brain range from 5% to 10% and likely lower within the tumor bulk, in vitro studies typically culture glioma cells under normoxic conditions of 20% O ₂ . Platet and colleagues demonstrated that culturing of GBM resection specimens at 3% O ₂ , representing a more physiologically relevant concentration, was associated with a significant increase in CD133 expression in comparison with GBM cells cultured at atmospheric concentrations of 20% O ₂ . McCord and colleagues demonstrated similar results with more mild reductions of oxygen concentration and that disaggregated GBM spheres derived from surgical resection samples had a 2-fold increase in percentage of CD133 ⁺ cells when recultured and allowed to form neurospheres in 7% O ₂ compared with normoxic 20% O ₂ conditions. CD133 ⁺ cells cultured in 7% O ₂ had a higher frequency of colony formation, shorter doubling time, and enhanced ability to differentiate along glial and neuronal lines, suggesting that hypoxic conditions not only increase the CD133 ⁺ cell composition but also modify and enhance the associated tumorigenic and stemlike phenotype. CD133 ⁺ neurospheres cultured in hypoxic conditions demonstrated increased expression of other stem cell markers, including nestin, Oct4, and SOX2. Levels of HIF-2α were increased in CD133 ⁺ cells cultured in 7% O ₂ , and consequent small interfering RNA silencing of HIF-2α led to decreased Oct4 and SOX2. Consistent with the induction of the stemlike phenotype, growth in 7% O ₂ induced alterations in global gene expression patterns with the upregulation of critical stem cell–associated genes including those involved with the notch and frizzled-2 signaling pathway, angiogenesis, and transforming growth factor β. When CD133 ⁺ cells were moved from 7% to 20% O ₂ , rates of colony formation and expression of HIF-2α and stem cell markers reversed to levels originally observed in normoxic conditions. Oxygen-induced stemlike features, partially mediated by HIF-2α, are reversible and mediated by epigenetic changes and are not the result of the hypoxic selection of a CD133 ⁺ subpopulation of tumorigenic cells.

In their studies with the established human glioma cell line U251MG, Griguer and colleagues further proposed that mechanisms underlying hypoxia-induced CD133 upregulation and functional changes partly involved loss of mitochondrial function. In accord with previous studies, U251MG glioma cells, containing undetectable levels of CD133 above background in 21% O ₂ culture conditions, became strongly CD133 positive when changed to 1% O ₂ . Following exposure, 20% of U251MG cells were CD133 ⁺ within 24 hours and up to 60% were CD133 ⁺ within 72 hours. On return to 21% O ₂ , levels of CD133 expression decreased to original normoxic levels. In addition, treatment of U251MG glioma cells with rotenone, an electron transport chain blocker, resulted in a significant and dose-dependent increase in CD133 ⁺ expression, reminiscent of exposure to hypoxic conditions. Depletion of mitochondrial DNA similarly led to constitutive and substantial increases in CD133 expression that persisted through multiple cell passages. Relative to controls, mitochondrial DNA–depleted glioma cells demonstrated a more aggressive phenotype of increased anchorage-independent growth and invasiveness. These cells readily initiated and expanded as tumorspheres in serum-free media that expressed the stem cell markers nestin and CD133 and had multilineage differentiation potential. In addition, rescue of mitochondrial function by transfer of parental mitochondrial DNA reversed the elevated CD133 expression levels. In their proposed model of glioma progression, Griguer and colleagues suggest that tumor growth is driven by a dynamic and adaptive biological response to oxygen and metabolic demands in reaction to a changing tumor microenvironment. Stringent nutrient and oxygen barriers selectively signal a switch within glioma cells in hypoxic regions to gain phenotypic changes associated with increased survival and migration.

Overall, the data support a critical role of reduced oxygen tension and disruption of mitochondrial function in mediating an in vitro response within glioma cells characterized by upregulated CD133 expression. These hypoxia-induced changes are associated with global alterations in the expression of stem cell–associated genes and promote CSC phenotypes, including increased clonogenicity, proliferation, invasiveness, capacity for multilineage differentiation, and tumorigenicity in xenograft models. While it is unclear if the extent of fluidity is present in vivo, the sensitivity of glioma cells toward changes in oxygen levels and the reversibility of hypoxia-induced characteristics suggest a dynamic regulatory component of CD133 expression and stemlike features. In addition, the low frequency of CD133 ⁺ glioma cells reported in literature may be linked to normoxic culture oxygen concentrations that do not reflect physiologic levels.

Recent studies have begun to provide further insight into the complex mechanisms regulating the expression of CD133 and its associated phenotype in gliomas. Differential methylation patterns have been identified as transcriptional regulators of CD133 in several cancers. CD133 promoter methylation, not present within normal brain tissue, represents an abnormal epigenetic regulator of differential CD133 expression in glioma cells. Yi and colleagues reported a high frequency of CD133 promoter hypermethylation in both GBM and colon cancer cells lines and primary tumor samples; however, such methylation patterns were absent in CD133 ⁺ cells. Upregulation of CD133 expression was strongly associated with genetic and drug-induced inhibition of DNA methyltransferase activity. This mode of regulation is unique to CD133, as hypermethylation profiles of other genes do not vary between CD133 ⁺ and CD133 ⁻ populations. In addition, Jaksch and colleagues demonstrated that the extent of AC133 immunoreactivity is cycle dependent in embryonic stem cells, colon cancer, and melanomas. Specifically, AC133 immunoreactivity was highest in the actively dividing cell subpopulation with 4N DNA content and lowest in cells in the G ₀ and G ₁ phases with 2N DNA content. MELK protein expression, which has been demonstrated to be cell cycle dependent, mirrored AC133 immunoreactivity with respect to cell cycle status. Prolonged culturing of purified CD133 ⁻ and purified CD133 ⁺ cells produced similarly heterogeneous cell populations comprising both CD133 ⁺ and CD133 ⁻ cells. The ability of cells at either extremes of CD133 immunoreactivity to produce CD133-heterogeneous populations suggests that CD133 immunoreactivity may not be unique to a discrete and stable population. Overall, although it does appear that CD133 expression may be used to enrich a glioma population with increased tumorigenicity, it is possible that CD133 expression and the associated phenotype are fluid traits responding to dynamic extracellular and intracellular processes.