Insight into the complex regulation of CD133 in glioma


  • Benito Campos,

    1. Division of Neurosurgical Research, Department of Neurosurgery, University of Heidelberg, Heidelberg, Germany
    Search for more papers by this author
  • Christel C. Herold-Mende

    Corresponding author
    1. Division of Neurosurgical Research, Department of Neurosurgery, University of Heidelberg, Heidelberg, Germany
    • Sektion Neurochirurgische Forschung, Neurochirurgische Universitätsklinik, INF 400, 69120 Heidelberg, Germany
    Search for more papers by this author
    • Tel.: +49−6221−566405, Fax: +49-6221-5633979


The transmembrane protein CD133 and its extracellular epitope AC133 are controversial cancer markers. In glioma, AC133 demarcates a subpopulation of stem-like tumor cells, so-called cancer stem cells (CSCs), which seem to drive tumor formation and are highly resistant to conventional chemo- and radiotherapy. Lately, experimental evidence for the existence of AC133-independent CSCs has challenged the importance previously attributed to AC133-positive glioma cells. These findings either imply that (i) AC133-positive and AC133-negative glioma cells comprise different, independent CSC populations, (ii) AC133-positive glioma cells are derived from primordial AC133-negative CSCs or (iii) AC133-negative CSCs have lost AC133 expression, while retaining their stem-like features and tumor initiation capacity, and can reacquire AC133 expression in vivo. In our article, we review evidence for and against each of the possible tumor models in glioma and will discuss technical hurdles in the AC133 detection process. In addition, we will outline new insights into CD133 regulation, which suggest certain degree of plasticity between some AC133-positive and AC133-negative CSC populations.

Antibodies directed against AC133, an extracellular epitope of CD133, have been used to enrich so-called cancer stem cells (CSCs) in different types of brain tumors.1, 2 The term “cancer stem cell” alludes to certain functional properties common to normal neural stem cells and CSCs. Both cell types are able to self-renew extensively and to express lineage-specific surrogate markers under differentiation-permissive (culture) conditions.1–3 In addition, in gliomas, the most common primary brain tumors, AC133-expressing CSCs were characterized as a highly tumorigenic and therapy-resistant tumor cell population.1, 2, 4, 5 In line with this, increased AC133 expression in situ was associated with a poor patient prognosis independently of tumor grade, patient age at diagnosis and extent of tumor resection.6 However, despite of these findings, the relevance of CD133 and AC133 to brain tumor pathobiology is a matter of ongoing scientific debate. Although a large number of reports corroborated the initial characterization of AC133-positive CSCs in gliomas, recent studies have reported the existence of AC133-negative CSCs.7–10 In addition, some studies demonstrated that xenotransplantation of AC133-negative glioma cells led to the formation of tumors, which unexpectedly contained a fraction of AC133-positive tumor cells.8, 10 In light of these findings, various different but not necessarily mutually exclusive tumor models (Fig. 1) can be envisioned: (i) Given the molecular heterogeneity of gliomas it is likely that AC133-positive and AC133-negative CSCs isolated from different tumors comprise independent and unrelated CSC types. At the same time, there is reason to believe that in individual tumors AC133-positive and AC133-negative CSCs constitute two different CD133 phenotypes in a continuum of cellular differentiation, and (ii) AC133-positive CSCs are derived from primordial AC133-negative CSCs or (iii) AC133-negative CSCs have lost AC133 expression, while retaining their stem-like features and tumor initiation capacity, and can reacquire AC133 expression in vivo. Finally, because of the complexity of the CD133 molecule caused by alternatively spliced and glycosylation-dependent epitopes, interpretation of AC133/CD133 expression levels may vary substantially depending on the analytic approach, making it likely that some CSCs are in fact AC133 false negative. In our article, we review evidence for and against each of this possible tumor scenarios and will discuss how new insights into CD133 regulation raise the intriguing possibility that the existence of AC133-positive and AC133-negative CSC might reflect some degree of CSC plasticity.

Figure 1.

Illustration of three hypothetical CSC models: (a) AC133-positive glioma cells are derived from primordial AC133-negative CSCs. (b) AC133-positive and AC133-negative glioma cells comprise different, independent CSC populations. (c) AC133-negative CSCs are derived from primordial AC133-positive CSCs (red circles = AC133-negative CSCs; green circles = AC133-positive CSCs; gray circles = non-CSC progeny).

Occurrence of AC133/CD133-Positive Stem-Like Cells in Normal and Tumor Tissues

There is a large body of data associating CD133 expression, especially in its glycosylated form, with the occurrence of stem and progenitor cells in various body tissues. Antibodies directed against glycosylation-dependent epitopes of the CD133 molecule, namely AC133, and to some extent AC141, have been particularly useful to enrich stem-like cells in the CNS: AC133-positive neural stem cells could be isolated from human fetal and postmortem brain tissue.3, 11, 12 After orthotopic xenotransplantation in mice, these cells successfully engrafted, migrated and underwent neural differentiation. In vitro AC133-positive cells differentiated into neurons and glial cells and demonstrated extensive self-renewal potential.3 Analogously, in the hematopoietic system, expression of the glycosylation-dependent epitope AC133 was confined to a small subpopulation of CD133-positive hematopoietic stem and progenitor cells.13 In other tissues, expression of the glycosylation-dependent epitope AC133 marked endothelial progenitor cells,14 myogenic cells15 and prostatic epithelial stem cells.16 Accordingly, for some of these cell types, the ability to regenerate the tissue of origin was proven. For example, differentiation of AC133-positive prostate basal layer epithelial cells restored a fully differentiated prostatic epithelium with expression of prostatic secretory products.16 Leong et al. even showed that a single Lin/Sca-1+/AC133+/CD44+/CD117+ cell could generate a fully developed, functional, secretion-producing prostate when transplanted in vivo.17

First evidence for AC133/CD133 expression in other cells but normal stem cells came from studies on blood malignancies. A rare fraction of leukemic cells reminiscent of normal hematopoietic stem and progenitor cells was characterized as stem-like tumor population bearing the AC133 epitope and was endowed with tumor-initiating capacity.18 Subsequently, other tumor cell types expressing AC133/CD133 were identified and successfully enriched from different malignancies including prostate cancer,19 colon cancer,20 hepatocellular carcinoma21 and medulloblastoma.1 With regard to brain tumors, Singh et al. were the first to describe an AC133-positive tumor subpopulation in gliomas.1 AC133-positive tumor cells grew as neurospheres in stem cell-optimized medium reminiscent of normal neural stem cells and were able to self-renew extensively and to express surrogate markers of neuronal, astroglial and oligodendroglial lineage upon differentiation. As few as 100 AC133-positive cells were capable to induce tumors after xenotransplantation in mice, whereas corresponding AC133-negative cells of the same tissue failed to engraft.2 First data corroborating a clinical relevance of AC133-positive tumor cells in solid malignancies came from studies on brain tumors. In a study involving 95 astrocytic gliomas, AC133 immunoreactivity of tumor cells proved to be an independent prognostic marker for adverse progression-free and overall survival.6 Consistent observations from Beier et al. on high-grade oligodendroglial tumors further substantiated these results22: Primary tumor cell cultures derived from tumors with poor prognosis showed a distinct AC141-positive population (4–34%) and retained their proliferative potential in vitro, whereas tumors with favorable prognosis were devoid of AC141-positive cells. In a further study involving six low-grade and 17 high-grade gliomas, Rebetz et al. reported that the content of AC133-positive cells was significantly increased in high-grade tumors and most importantly, AC133 expression was negatively associated with patient survival.23 At present, observations made by these studies have been confirmed by additional reports.24, 25

Altogether, antibodies directed against AC133 and AC141 have been used successfully to enrich stem-like cell populations in normal and cancerous tissues, suggesting that these epitopes mark cells, which would allocate at early steps in a model of cellular hierarchy. At the same time, several independent studies have substantiated a prognostic influence of AC133/AC141-positive cancer cells.

Technical Hurdles in the CD133 Detection Process

Focus on CD133 expression varies among studies ranging from mRNA expression and protein levels to glycosylation of CD133.13, 26–28 It still remains a matter of investigation to which extent these different expression levels can be linked: The CD133 protein bears eight possible glycosylation sites on its two extracellular loops (Fig. 2a) raising the possibility of differential glycosylation as a complicating factor in the CD133 detection process.13 This assumption is supported by the observation that expression of the glycosylation-dependent AC133 epitope is lost upon differentiation.13, 27, 29 These data are further complemented by the observation that AC133 immunoreactivity is reduced in CaCo-2 colorectal adenocarcinoma cells upon differentiation resulting from epitope masking rather than from reduced CD133 protein expression.28, 30 Consistently, AC133 immunoreactivity seems to be restricted to undifferentiated cell types in both, embryonic and adult tissues,13, 26, 27 whereas CD133 protein can be found in a broader range of tissues, including the human eye, salivary glands, lacrimal glands, sweat glands, liver and uterus.31 Moreover, as an additional factor of complexity, transcription of CD133 can be initiated at five tissue-restricted promoters, yielding several alternatively spliced transcripts (Fig. 2b).32, 33 Accordingly, CD133 isoforms containing facultative exon 3 are predominantly found in fetal brain, whereas most CD133 transcripts found in the adult brain lack exon 3.32 Taking into account all these factors influencing CD133 expression, it is likely that unconstrained use of antibodies directed against different types of epitopes will lead to inconsistent results (Figs. 2b–2d). Therefore, it seems important to mention that flow cytometric characterization and isolation of CD133-positive CSCs in glioma have been mostly performed with antibodies recognizing the extracellular epitopes AC133 and AC141. Although at present there are no specific studies on the comparability of antibodies recognizing these two different epitopes, several reports have documented a proper overlap of AC133- and AC141-positive cell populations.13, 27, 34 Immunohistochemical staining of AC133 and AC141 epitopes, however, poses a special challenge leading some researchers to use alternative antibodies recognizing CD133 protein epitopes,22, 25 and it is still a matter of investigation how AC133, AC141 and various protein epitopes compare in their biological significance. Altogether, most in vitro work on CD133-positive CSCs has been performed with two, likely interchangeable antibodies (AC133/AC141), whereas only few glioma studies relied on CD133 protein expression for in vitro characterization and/or isolation of potential CSCs.

Figure 2.

(a) Schematic overview of the pentaspan transmembrane protein CD133. Potential n-glycosylation sites are indicated by green circles. “Y”s point to predicted binding sites of five different commercially available antibodies. Note: The exact binding site of AC133/AC144 has not yet been established in detail, although recent data suggest localization of these epitopes to the second extracellular loop.30 (b) Schematic overview of the CD133 mRNA transcript indicating constitutive exons (blue), facultative exons (yellow) as well as the 3′ and the 5′ untranslated regions (red). (c) Example of distinct immunofluorescent staining patterns in mouse ependymal cells using three different antibodies directed against alternative epitopes of CD133: brush-like staining at the apical cell surface of ependymal cells after staining with ANC9C5 antibody (Santa Cruz Biotechnology, Santa Cruz, USA) directed against an unknown CD133 epitope. (d) Granular intracellular staining of ependymal cells after staining with C24B9 antibody (New England Biolabs, Frankfurt, Germany) directed against a peptide epitope surrounding Asp562 of CD133. (e) Absent immunoreactivity after staining with the glycosylation-dependent AC133 antibody (Miltenyi Biotec, Bergisch-Gladbach, Germany). Scale bar: 30 μm (ac).

An even greater source of inconsistency in experimental results may derive from indiscriminative use of cell material and cell culture conditions. Studies on CSCs in glioma have been performed on many different cell types including freshly isolated tumor tissues,4, 35, 36 briefly cultured tumor cells37, 38 and a large variety of tumor cell lines.8, 9 In addition, putative glioma CSCs have been cultivated in the presence of fetal calf serum8 or under serum-free conditions,1 either as suspended neurospheres1, 39 or as adherent monolayer cultures.40 Particularly, studies involving cell lines, which are cultivated in serum-containing growth media, should be considered with great care because serum has a well-known prodifferentiation effect that can alter the CSC phenotype and not least AC133 expression.39 However, even if technical hurdles in the CD133 detection process are taken into account, there is still growing experimental evidence suggesting coexistence of several types of CSCs in glioma.

Occurrence of AC133-Negative Stem-Like Cancer Cells in Glioma

Although many publications corroborated the initial characterization of AC133-positive CSCs, several recent publications reported the existence of AC133-negative CSCs.7–10 Beier et al. generated primary cell cultures with varying AC133 content from glioblastomas tissues.7 Upon replating sorted AC133-positive and -negative cells derived from the same cell culture, a fraction of AC133-positive cells (2–5%) formed new tumor spheres, showed extensive self-renewal and expressed surrogate markers of astrocytes, neurons and oligodendrocytes consistent with multilineage differentiation capacity previously ascribed to these cells. In the same experimental setting, however, unsorted cells from constitutively AC133-negative glioblastoma cultures also showed sphere-forming capacity, albeit to a lower extent (0.5–2% clonogenic cells), were able to self-renew over a large period of time and displayed markers of all three neural cell lineages. In addition, xenotransplanted AC133-negative glioblastoma cultures formed large circumscribed tumors displaying similar tumorigenicity as sorted AC133-positive tumor cells.

Further support for the existence of AC133-negative CSCs comes from research on alternative CSC surface markers and specific, marker-independent isolation procedures. In a recent study, Clément et al. reported a marker-independent method to enrich glioma CSCs based on their specific autofluorescence in flow cytometric analyses.37 Hereby, the authors were able to enrich CSCs with high nuclear:cytoplasmic ratio capable of self-renewing over several passages in clonogenic assays, differentiating in the presence of serum and initiating tumors in mice even if 103 tumor cells were injected. Although these CSCs expressed several stem cell-associated proteins such as NANOG, Sox2 and Nestin, frequencies of AC133/AC141-positive cells were not increased in these cells when compared to the remaining tumor bulk and most importantly, tumor formation was independent of AC133/AC141 expression levels. In addition to the results described by Clément et al., it has been known for many years that gliomas contain a fraction of highly tumorigenic cells, the so-called side population (SP).41 SP cells show exclusive ability to extrude Hoechst 33342 dye and can be isolated as dye-negative cells from the remaining tumor bulk. Sorted SP cells quickly reconstitute the initial population of SP and non-SP cells, whereas sorted non-SP cells never give rise to SP progeny, suggesting some sort of cellular hierarchy. It is important to mention that despite of the well-documented characterization of SP cells in glioma, data on the overlap between SP- and AC133-positive cells are scarce and rely on a single study using the murine glioma cell line GL261. Although this study reported that AC133-positive cells were not enriched in the SP fraction, further studies on human cell lines cultivated in the absence of serum are needed to support this claim.9

Although Hoechst-dye extrusion and autofluorescence-based selection constitute marker-independent approaches to isolate CSCs in glioma two additional studies reported alternative and AC133-independent surface markers to enrich stem-like glioma cells with the capacity to initiate tumor growth in rodents. Son et al. reported expression of the neural stem and progenitor marker stage-specific embryonic antigen 1 (SSEA-1) in a study sample involving cell lines and tissue samples from 12 primary glioblastomas.35 Although AC133 was heterogeneously expressed, SSEA-1 expression was confirmed in all but one sample and SSEA-1-positive tumor cells showed high levels of the stem cell-associated markers Sox2, Bmi1, Ezh2, L1Cam, Olig2 and integrin beta 1, when compared to SSEA-1-negative cells. Further, SSEA-1-positive tumor cells showed increased self-renewing capacity and were capable of rebuilding a cellular hierarchy as opposed to their SSEA-1-negative counterparts. At the same time, tumor initiation capacity in SSEA-1-positive tumor cells was enriched more than 100-fold compared to SSEA-1-negative cells. Most importantly, tumor formation capacity was unrelated to AC133 expression and even 104 SSEA-1-positive/AC133-negative cells were able to grow large tumors. Searching for alternative surface markers to enrich glioma cells with tumor-initiating capacity, Ogden et al. characterized tumor cells expressing the glial progenitor marker A2B5.36 Analogously to SSEA-1, A2B5-positive cells were expressed in most glioma samples whereas AC133/AC141-expressing cells were undetectable in several tumors. In a series of subsequent xenograft experiments, Ogden et al. went on to show that A2B5-positive glioma cells were able to grow tumors in nude rats as opposed to A2B5-negative cells and that tumor engraftment was unrelated to AC133/AC141 expression levels. Finally, tumorigenic cell lines, such as U87, U373 and U251, have been passaged over many years and were engrafted successfully innumerable times, while being virtually devoid of AC133 immunoreactivity.9 Wu et al. reported persistence of AC133-positive cells in established glioma cell lines U87, U373, CRL2365, CRL1620 and SF295 and the murine glioma cell line GL261, with contents of positive cells ranging from merely 0.02 to 0.5%.9 Further analyses of GL261 cells revealed that both AC133-positive and AC133-negative glioma cells formed tumors, although tumorigenicity was increased in AC133-positive cells and the corresponding xenografts were invasive as opposed to the demarcated tumors originating from AC133-negative counterparts. Again, both cell types were endowed with self-renewal capacity although AC133-negative cells were less clonogenic, reminiscent of the observations made by Beier et al.7

Although the aforementioned studies clearly speak for the existence of AC133-negative tumor-initiating glioma cells, it is important to mention that tumor-initiating capacity is also dependent on the xenotransplantation model used and does not seem to comprise an exclusive CSC trait, e.g., Quintana et al. were able to increase frequencies of tumorigenic melanoma cells by several orders of magnitude through modifications to their xenotransplantation assays including the use of highly immunocompromised NOD/SCID interleukin-2 receptor gamma chain null mice.42 A second important concern relates to the use of established glioma cell lines, cultivated in the presence of serum, to study AC133-negative CSCs. Because serum is known to induce differentiation and to impact on AC133 expression great care should be exercised in the evaluation of the aforementioned studies.

Bearing these limitations in mind, it can still be summarized that several studies characterized AC133-positive and AC133-negative CSCs, which could be differentiated in vitro and were capable of extensive self-renewal as well as tumor formation upon xenotransplantation. Hence, it is likely that in different tumors AC133-positive and AC133-negaitve CSCs comprise independent CSC types. Strengthening this conclusion, Lottaz et al. recently reported a comparison of gene expression profiles of 17 glioblastoma-derived CSC lines.43 On the basis of these expression profiles, the authors were able to group all 17 CSCs lines into two main categories. Type 1 CSCs preferentially expressed AC141 and grew as suspended neurospheres as opposed to Type 2 CSCs, which were mostly AC141-negative and grew adherently or semiadherently. Lottaz et al. were able to identify a 24-gene signature, which discriminated between Type 1 and Type 2 CSCs and reflected among others differences in the expression of TGF-beta pathway-related transcripts (upregulated in Type 2 CSCs). Interestingly, gene expression profiles of Type 1 CSCs were reminiscent of fetal neural stem cells, whereas Type 2 CSCs genetically resembled adult neural stem cells. At the same time, neither Type 1 nor Type 2 was closely related to other putative founder cells like mature astrocytes, neurons and mesenchymal stem cells. Taking together, despite of the descriptive nature of these results, the study of Lottaz et al. provides strong evidence for the existence of independent AC133-negative and AC133-positive CSCs. At the same time, these findings impose the question whether AC133 and AC141 comprise phenotypic or functional markers in CSCs. As mentioned in previous chapters, these epitopes have been associated with increased tumorigenicity, neurosphere-like growth and augmented radioresistance, although the exact function of AC133 and AC141 remains elusive. As a complicating issue, future studies would have to distinguish between the relevance of CD133 in general and the relevance of AC133 and AC141 in particular, which will certainly represent a major experimental challenge.

Bridging the Gap Between AC133-Negative and AC133-Positive Glioma Cells

In the midst of the ongoing discussion, several recent publications have shed new light into the subject and may help to bridge the gap between AC133-positive and AC133-negative CSCs in a subset of tumors. Griguer et al. reported that the established glioma cell line U251MG, which had previously been reported to be AC133-negative and to contain only marginal levels of CD133 mRNA transcripts,22 became AC133-positive under hypoxic conditions.44 Although AC133 immunoreactivity in these cells was undistinguishable from background under normoxic conditions (21% oxygen), a large fraction of U251MG glioma cells acquired a strong membrane staining for AC133 when cultured in 1% oxygen atmosphere. After only 24 hr of exposure AC133/CD133 expression could be detected in more than 20% of cells and even reached 40–60% after 72 hr. Reoxygenation of U251MG glioma cells quickly restored original AC133 levels, and after 6 days of subsequent cultivation under normoxic atmosphere AC133 expression had again reached levels close to background. Griguer et al. went on to show that the plasticity of AC133 expression was associated with alterations in mitochondrial function. Consistently, inhibition of the mitochondrial electron transport chain with rotenone or through depletion of mitochondria recapitulated increments in AC133 expression reported under hypoxic conditions and revealed an inverse correlation between mitochondrial membrane potential and AC133 expression. In addition, mitochondria-depleted U251MG cells displayed both significantly increased colony formation capacity and augmented cell migration. Concordantly, reinsertion of mitochondrial DNA restored original AC133 levels. In light of these discoveries, the authors propounded a tumor scenario, in which glioma growth is controlled by nutrient and oxygen supply: under stringent hypoxic conditions some tumor cells might respond to their microenvironment with a loss of mitochondrial oxidative function and subsequently trigger expression of stem cell markers such as AC133 and multiple functional changes that would result in increased migration and survival.44 Such a process could enable tumor cells to migrate away from hypoxic microenvironments and to coopt new blood vessels that would eventually lead to reoxygenation and subsequent reconstitution of the initial phenotype (Fig. 3). This model of glioma progression would involve many of such cycles occurring in the course of tumor formation and be largely dictated by local microenvironmental conditions, rather than by a specific cell type. Interestingly, this hypothesis predicts a close spatial association of undifferentiated glioma cells with hypoxic regions in the vicinity of necrotic tumor areas on the one hand and with newly formed capillaries on the other hand. Although, in terms of oxygen and nutrient supply, these constitute completely remote and seemingly disparate extremes, several studies have reported such physical relationship between undifferentiated glioma cells, necrotic tumor areas and proliferating endothelial capillaries.45, 46

Figure 3.

Illustration of a possible relationship between stem-like tumor cells and oxygen tension in glioma. (a) Under average oxygen levels, like in close vicinity of blood vessels, tumor cells proliferate normally. (b) As the tumor volume increases oxygen levels in the center of the tumor mass decrease and trigger phenotypic changes in those tumor cells, which are close to the most stringent hypoxic conditions. (c) As tumor growth continues, most cells in the tumor center succumb to the hypoxic microenvironment, generating necrotic tumor islands, whereas “activated” tumor cells migrate away to coopt new blood vessels. Eventually, reoxygenation of these tumor cells will lead to a reversion of the changes induced by hypoxia and restore the initial phenotype.

Although experiments performed by Griguer et al. were carried out with an established glioma cell line cultivated under serum conditions, at present, several independent studies have corroborated and substantiated their observations using primary glioma cultures growing under serum-free conditions.44, 47–49 McCord et al. demonstrated that already moderate levels of hypoxia (corresponding to 7% oxygen) resulted in increased AC133 expression in three glioma cell cultures, coinciding with augmented proliferation, and potentiated colony formation capacity.47 In addition, expression levels of prominent stem cell-associated markers Oct4, Sox2 and Nestin increased under hypoxic cultivation conditions. McCord et al. could show that this increase in stem-like features reversed under normoxic conditions within a few days and was partially dependent on the activity of the hypoxia-inducible transcription factor HIF2α. In agreement with this data, a recent publication by Li et al. demonstrated that HIF2α was highly expressed in AC133-positive primary glioma cell cultures compared to their AC133-negative counterparts and was even expressed in some tumor cells under normoxic conditions, thus rendering these cells highly sensitive to even modest levels of hypoxia.46 Further immunohistochemical analyses revealed that Hif2α immunoreactivity in tumor tissues was associated with CD133 protein expression, and Hif2α-positive tumor cells were found mostly in close vicinity of necrotic tumor areas and around some proliferating blood vessels. Using mRNA data bases, Li et al. substantiated their results demonstrating a significant association between increased Hif2α-transcript levels and poor patient survival in glioblastoma.46 In agreement with previous reports, Soeda et al. corroborated expansion and increased self-renewal of three AC133-positive glioma cell lines under hypoxic conditions (1% oxygen).50 Interestingly, however, rather than Hif2α, a second hypoxia-inducible factor, Hif1α, seemed to be mediating these effects. Concordantly, knockdown or inhibition of Hif1α attenuated expansion and increased self-renewal of AC133-positive cells, whereas knockdown of Hif2α did not show any significant effects.

Summarizing all the aforementioned findings, it must be reasoned that microenvironmental cues such as hypoxia and mitochondrial function seem to modulate the stem-like phenotype in glioma to an unforeseen extent, and that these findings broaden the “classical” concept of hypoxic regions as locations for radio-resistant glioma cells.51, 52

However, the issue of molecular key players mediating these effects seems to be controversial and requires careful consideration of the experimental circumstances. One possible factor of complexity might be the use of varying oxygen tension levels in the different studies. In neuroblastoma, Hif2α can already be stabilized under moderate hypoxia (5% oxygen), whereas Hif1α stabilization requires an additional decrease in oxygen tension (1% oxygen).53 Thus, moderate oxygen tension levels used by McCord et al. (corresponding to 7% oxygen) could explain the predominant role attributed to Hif2α, while the exclusive role of Hif1α documented by Soeda et al. might have resulted from comparatively lower oxygen levels (corresponding to 1% oxygen). It is also important to mention that the aforementioned experimental work was performed in vitro, and further studies are needed to elucidate the predominant role of particular Hifs in vivo. A recent study by Pistollato et al. could shed new light into this controversy.38 Through multiple sampling of tumor tissue from a single glioblastoma, the authors were able to identify a well-vascularized outer tumor layer, an intermediate zone with moderate hypoxia and a predominantly avascular core. This stratification was associated with a concentric gradient of Hif1α expression, which correlated with tumor proliferation, expression of AC133 and Nestin as well as increased levels of the chemoresistance-associated repair enzyme MGMT. At the same time, expression of Hif2α was similar among different tumor layers. Although results from a single tumor sample do not specifically exclude a biological role of Hif2α, larger study samples could help to clarify the importance of Hif1α and Hif2α in glioma.

Finally, it seems important to mention a series of independent findings reported by Matsumoto et al.54 Contrary to previous observations made by Soeda et al.,50 Matsumoto et al. were able to show that activation of Hif1α in established gastric, colorectal and lung cancer cell lines resulted in downregulation of AC133 expression as well as CD133 mRNA levels. Concordantly, microarray data from 40 gastric cancers revealed an inverse correlation of CD133 and Hif1α mRNA levels. In light of these findings, future studies should probably exercise great care when comparing data on CD133 regulation generated from experiments involving intra- and extracranial tumors.

Although the aforementioned studies suggest that microenvironmental cues may bridge the gap between some AC133-negative and AC133-positive CSC types, two independent studies reported the existence of AC133-negative CSCs with the ability to generate AC133-positive progeny in vivo and in vitro. In the first study, Wang et al. xenotransplanted shortly cultured glioblastoma spheroids into nude rats.10 Inoculation of shortly cultured glioblastoma spheroids gave rise to highly invasive tumors with moderate angiogenic activity and undetectable contrast enhancement on MRI scans. Although no AC133 expression was measurable in the initial xenografts, in vivo passaged tumors acquired AC133 immunoreactivity as well as increased CD133 mRNA levels. The increase in CD133 expression coincided with a shortened survival of xenotransplanted rats as well as with aberrant vessel growth, necrosis and contrast enhancement in tumor tissues. Both AC133-positive and AC133-negative cells isolated from these in vivo passaged tumors showed extensive proliferation, stained positive for the neural stem cell marker Nestin and upon xenotransplantation led to tumor formation with similar engraftment rates. Finally, flow cytometric analyses revealed that AC133-positive cells (1–5%) had formed in tumor tissues derived from AC133-negative glioblastoma cells. In line with this data, a second study by Chen et al. documented the occurrence of three different, coexisting types of CSCs in a subset of glioblastoma specimens.55 These different CSCs, termed Types 1, 2 and 3, seemed to be related in a lineage hierarchy, where AC133-negative Type 1 CSCs could give rise to AC133-positive Type 2 CSCs or constitutively AC133-negative Type 3 CSCs. Type 2 CSCs would only generate Type 2 and Type 1 cells, whereas Type 3 CSCs only propagated Type 3 AC133-negative cells. Corresponding to these phenotypic differences, Type 1 and Type 2 CSCs were enriched for transcripts of stem cell-associated genes like the radial glia marker FABP7, nestin, SOX2 and vimentin, whereas Type 3 CSCs preferentially expressed transcripts for genes associated with progenitor cell types, namely TBR2, CULX2, DLX2 and DLX1. Finally, xenografted Type 1 and Type 2 CSCs derived from the same CSC cell line resulted in large, invasive tumors with infiltration and tumor cell migration along white matter tracts, whereas Type 3 CSCs only generated small noninvasive tumors, with delayed growth onset and perivascular tumor infiltrates. These experiments together with data reported by Wang et al. clearly show that in some tumors, primordial AC133-negative CSCs can give rise to AC133-positive daughter cells, thus constituting a lineage hierarchy of CSCs.


Since the original identification of AC133-positive CSCs in gliomas, several groups have produced strong experimental evidence for the existence of AC133-independent CSCs. The characterization of different CSC types in glial tumors challenges the pathobiological relevance previously attributed to AC133-positive CSCs, giving rise to speculations about primordial CSCs. Taking into account the cellular and molecular heterogeneity of glial tumors and the numerous findings on bona fide AC133-negative CSCs, these findings strongly suggest the existence of different and independent types of CSCs in gliomas.

However, at a stage, where apparently opposed, AC133-dependent and AC133-independent CSCs are dividing the scientific community, several pioneering studies have enriched our knowledge of stem cell plasticity in glioma. The authors of these studies purport that—to some extent—absence and presence of AC133 may represent two sides of the same coin. The tantalizing, inherent consequences of the aforementioned discoveries may presage a paradigm shift in glioma research and merit particular scientific attention. Especially, cross-disciplinal knowledge on normal stem cells, which are also tightly regulated by oxygen levels and mitochondrial function, may inspire research on the putative cancerous counterparts.56

Despite the fact that AC133 remains a controversially debated cancer marker, there is a large, growing set of data linking the expression of AC133 with poor patient prognosis.6, 23–25, 57, 58 In these studies, AC133 expression is associated with an immature tumor phenotype, which, in turn, is linked to the aggressiveness of glial tumors. New insights into the complex regulation of CD133 might help us to understand why, in some cancers, AC133-positive tumor cells exert a prognostically relevant influence. On the one hand, AC133 expression might mirror the amount of highly migratory tumor cells, which will ultimately limit surgical tumor resection. On the other hand, AC133 expression might reflect hypoxic microenvironmental conditions within the tumor tissue and could serve as additional explanation for increased radioresistance in tumors with high AC133 content.4 Finally, key roles of hypoxia, mitochondrial function and hypoxia-inducible factors may uncover new therapeutic targets in the treatment of glioma, especially once mechanisms orchestrating these different factors are revealed.

Coming back to the three different tumor scenarios put forth in the introductory chapter, i.e. (i) independent CSC populations, (ii) primordial AC133-negative and (iii) primordial AC133-positive CSCs, new studies suggest the existence of (i) at least two diverse CSC types and (ii) within a subgroup of CSC, a lineage hierarchy with primordial AC133-negative CSCs, which can give rise to AC133 progenitors. Some reports even suggest the possibility of an intriguing, fourth tumor model: in this setting, a portion of glioma cells shows large phenotypic plasticity and can acquire stem-like features and AC133 positivity in response to a hypoxic microenvironment, whereas the remaining portion of AC133-negative cells appears irresponsive to oxygen tension levels and probably represents more differentiated progeny. One future challenge will certainly consist in bringing together these different tumor scenarios in a new, all-encompassing theoretical frame that would have to stand the test of future research on AC133-negative CSCs.


C.H.M. was supported by Verein zur Förderung der Krebsforschung in Deutschland e.V., Tumorzentrum HD-Ma, BMBF NGFNplus Brain Tumor Network and Deutsche Krebshilfe e.V.