The hitchhikers guide to cancer stem cell theory: Markers, pathways and therapy


  • Ákos Fábián,

    1. Department of Biophysics and Cell Biology, Medical and Health Science Center, University of Debrecen, Hungary
    Search for more papers by this author
  • György Vereb,

    Corresponding author
    1. Department of Biophysics and Cell Biology, Medical and Health Science Center, University of Debrecen, Hungary
    2. MTA-DE Cell Biology and Signaling Research Group, Medical and Health Science Center, University of Debrecen, Hungary
    • Department of Biophysics and Cell Biology, Medical and Health Science Center, University of Debrecen, P.O. Box 39, Nagyerdei krt. 98, H-4012 Debrecenm Hungary
    Search for more papers by this author
  • János Szöllősi

    Corresponding author
    1. Department of Biophysics and Cell Biology, Medical and Health Science Center, University of Debrecen, Hungary
    2. MTA-DE Cell Biology and Signaling Research Group, Medical and Health Science Center, University of Debrecen, Hungary
    • Department of Biophysics and Cell Biology, Medical and Health Science Center, University of Debrecen, P.O. Box 39, Nagyerdei krt. 98, H-4012 Debrecenm Hungary
    Search for more papers by this author


Cancer stem cell (CSC) biology is a rapidly developing field within cancer research. CSCs are postulated to be a unique cell population exclusively capable of infinite self renewal, multilineage differentiation and with ability to evade conventional cytotoxic cancer therapy. These traits distinguish CSCs from their more differentiated counterparts, which possess only limited or no potential for self renewal and tumor initiation. Therefore, CSCs would be the driving motor of malignant growth and therapy resistance. Accordingly, successful cancer treatment would need to eliminate this highly potent group of cells, since even small residual numbers would suffice to recapitulate the disease after therapy. Putative CSCs has been identified in a broad range of human malignancies and several cell surface markers have been associated with their stem cell phenotype. Despite all efforts, a pure CSC population has not been isolated and often in vitro clonogenic and in vivo tumorigenic potential is found in several cell populations with occasionally contradictory surface marker signatures. Here, we give a brief overview of recent advances in CSC theory, including the signaling pathways in CSCs that also appear crucial for stem cells homeostasis in normal tissues. We discuss evidence for the interaction of CSCs with the stromal tumor environment. Finally, we review the emerging potentially effective CSC-targeted treatment strategies and their future role in therapy. © 2012 International Society for Advancement of Cytometry

Cancer stem cell (CSC) theory hypothesizes that heterogeneity within tumors is not a mere consequence of random mutation and clonal evolution, but results from an intrinsic hierarchy of cells, with the putative CSC at the apex of the hierarchy (1, 2). The CSC is thought to share several key features with normal stem cells: unlimited capacity for self renewal, including maintenance of the CSC population through asymmetric division, the ability to differentiate into several cell lineages and intrinsic resistance against cytotoxic therapies through drug-efflux mechanisms and slow cell cycling (2). CSC theory postulates that not all tumor cells are equal with regard to self-renewal, tumor initiation, and maintenance potential, these traits being reserved for the CSC population. Additionally, cellular heterogeneity and hierarchy within the tumor originates from CSCs, which give rise to daughter cells that proliferate and differentiate into the cell mass that compromises a significant portion of the bulk tumor (1). Further, CSCs are thought to be responsible for therapy resistance, minimal residual disease and relapse after initial successful therapy, their stem-like features making them able to evade conventional treatment modalities (3). It has also been implied that these stem cell traits allow CSCs to play a leading role in metastasis (4–6).

Since the first experiments verifying that a specific phenotype within AML cells enriches for cells that recapitulate the original disease upon transplantation into NOD/SCID mice (7, 8), putative CSCs have been identified in several human malignancies such as CML (9, 10), brain (11–15), breast (14, 16–21), colon (22–26), endometrial (27), head and neck (28, 29), gastric (30, 31), liver (32–36), lung (19, 37–39), melanoma (40–42), ovarian (43–47), pancreatic (48–51), prostate (14, 52), renal cell (53, 54) and thyroid carcinomas (55). Despite the large body of work in CSC research, there is still a considerable amount of confusion and controversy surrounding the CSC hypothesis. Part of this stems from the fact, that to date, the CSC phenotype is a functional one, based on the cell's behavior in an array of in vitro and in vivo experiments designed to test for renewal and differentiation capacity, as well as tumorigenic potential. Recent addition of testing for stemness-related gene expression patterns (31, 36, 38, 39, 53, 54, 56, 57) has helped to further verify stem-like properties at the molecular level, but there is still not a set of surface markers in any malignancy that not only enriches, but is exclusive for cells displaying CSC behavior. Often, cells lacking the surface marker phenotype of putative CSCs still retain various degrees of tumor initiation capacity (14, 20, 27, 31, 41, 49, 58). In extremis, mutually exclusive surface marker phenotypes have been shown to harbor cells with CSC traits (12, 59–61). To further obscure the picture, a significant amount of evidence is mounting, that CSC behavior is substantially influenced by the tumor microenvironment (62) and CSCs can alter their expression profiles, resulting in a “moving target” for CSC research and raising questions about the adequacy of the assays used in vitro and in vivo. Whether CSCs have exclusive tumorigenic potential within the tumor population and are responsible for all the heterogeneity and the establishment of all cell lineages, remains to be seen.

Nevertheless, there is clearly a functionally distinct population within most malignancies with an intrinsic resistance against conventional cytotoxic therapy (63, 64). This in vitro and in vivo resistance is in line with recent findings that CSC-associated expression patterns in bulk tumors have clinical significance and are independent prognostic markers for patient survival (65–70). Further, treatment modalities developed along the lines of CSC theory and targeting putative CSCs show promising results and are effective in preliminary studies (71, 72). Taken together, the CSC population isolated with the admittedly limited functional assays and surface markers has biological relevance and broadens our appreciation for the complexity of tumor biology.


The CSC theory, although widely accepted, has had to face justified criticism. Several caveats are known, starting with the identification of putative CSCs (discussed in detail later). CSCs were originally thought to arise from normal stem cells of tissues through malignant transformation, but little irrevocable evidence exists linking CSCs to their normal counterparts. One major reason for CSCs having remained elusive is that CSC isolation techniques only select an enriched, nonhomogeneous population. Therefore, a direct investigation of CSCs and their possible evolution from nonmalignant tissue has largely not been possible. At the same time, the dysregulation of several key pathways that are important in the maintenance and differentiation of normal stem cells were shown to play crucial roles in carcinogenesis. Amongst others, pathways regulated by Wnt (73), Hedgehog (74, 75), Notch (76), Nestin (77), Nanog (78, 79), and transforming growth factor beta (TGF-β) (80, 81) were shown to be functionally active and relevant in cancer. More and more active stemness-associated pathways are also identified in putative CSCs (82–84). Not only are these pathways active in CSCs, but they also seem to be essential as evidenced by experiments in which selective inhibition led to loss of CSC function. Invasion and migration of prostate CSCs was not possible after knockdown of sex determining region of Y box 2 (SOX-2) and octamer-binding transcription factor 3/4 (OCT3/4) (85), while small interfering RNA (siRNA) knockdown of SOX-2 eliminates tumorigenicity of side population (SP) cells in lung adenocarcinoma (86). Knockdown of β-catenin led to tumor regression and loss of CSCs in chemically induced murine skin tumors and in human squamous cell carcinoma, and reduced the growth of xenografted human tumor cell lines (87). Combined inhibition of sonic hedgehog and mammalian target of rapamycin (mTOR) together with gemcitabine depleted CSCs in human pancreatic cancer cell lines (88), while antagonism of hedgehog signaling induced apoptosis in CSCs (89). Downregulation of SOX-2 or twist-related protein 1 (TWIST-1) lead to differentiation of tumorsphere forming CD133+ glioblastoma cells (90). Blocking hedgehog signaling at the level of Smoothened by cyclopamine inhibited chronic myeloid leukemia (91) and prostate CSCs (92). Although active stemness-associated pathways do not necessarily prove stem cell origin of CSCs, they do show a common functional heritage.

Biological tumor behavior does not always conform to the linear-hierarchical organization dictated by the strictly taken CSC theory. For instance, Quintana et al. showed that the cells in various types of melanoma tumors are not hierarchically organized and have a reversible phenotype with tumorigenic potential equal among different phenotypes (93). Similarly, in BCR-ABL1 lymphoblastic leukemia, functionally defined CSC populations proved to be genetically diverse, often with multiple genetically distinct CSC subclones in the same patient, indicating a branching multi-clonal evolution (94). While these cases demonstrate that not all tumors adhere to the CSC model, there is mounting evidence that clonal expansion and multi-lineage differentiation is sufficient to drive tumor growth and heterogeneity (95, 96). Lathia et al. separated CSCs and non-CSCs from glioblastomas and then lentivirally labeled the cells with either GFP or YFP (97). After inoculating mice with a cell suspension containing CSCs and non-CSCs in a 1:10 ratio, they monitored tumor growth with intravital microscopy. The CSCs initially localized in the perivascular region and then outgrew the non-CSC population, with cells of dominantly CSC origin constituting the bulk tumor. Since tumors evolving in this manner eventually resembled the parent tumors at the histological level, the case can be considered exemplary of the potential of CSCs for intratumor evolution and diversification. The same group also found CSCs capable of symmetric or asymmetric cell divisions, depending on the environmental conditions (98). The coexistence of these division types is a long standing cornerstone of CSC theory and was thought to be a prerequisite for CSC derived tumor heterogeneity and CSC renewal. Interestingly, both differentiated progeny and CSCs could be generated by either division type, questioning the dogma, that one type is responsible for self-maintenance and the other for cellular expansion. In a truly groundbreaking work, Zhu et al. first identified Prominin-1 (Prom1 or CD133) cells as normal tissue stem cells in the mouse small intestine (99). Then endogenous Wnt signaling was activated in the mice using a mutant beta-catenin allele, leading to gross dysplasia and intraepithelial neoplasia. Through lineage tracking with a YFP encoding reporter allele, they showed a disproportionate expansion of the Prom1+ cells and that cells within the neoplastic lesion were progeny of Prom1+ cells. In line with CSC theory, only a fraction of cells retained the Prom1+ phenotype and only a fraction of these cells were proliferating. This, to our knowledge, was the first direct proof that a solid tumor can arise from malignant transformation of a tissue stem cell and that all tumor cells are descendents of CSCs. Previously, leukemia CSCs were found to be heterogeneous in self renewal potential and to follow a hierarchy reminiscent of the normal hematopoietic stem cell compartment (100). While this can be interpreted as indirect evidence of stem cell origin, a recent study suggests that CSCs need not originate from the pinnacle of cellular hierarchy. Goardon et al. showed that in human acute myeloid leukemia, two molecularly distinct CSC harboring populations are present (101). These populations are hierarchically ordered, in so far as one can give rise to the other, but not vice versa. Global gene expression profiling showed that the CSC populations resembled normal progenitors and not hemopoetic stem cells. Therefore, stem cell origin is not necessary for CSC functionality. Based on these data it is highly probable, that depending on the type and origin of malignancy, a spectrum of cells can be responsible for giving birth to the common CSC phenotype. Newly developed in vivo imaging modalities could aid in tracking and identifying putative CSCs (102–104) and thereby bring us closer to the origins of CSCs. Recent advances in detecting circulating tumor cells (105–108) and monitoring minimal residual disease (109, 110) should also help establish the role of CSCs in metastasis formation and disease propagation.


The in vitro and in vivo assays for identifying cell populations harboring CSCs have been extensively discussed elsewhere (2, 111–113). Briefly, colony forming capacity and tumorsphere formation in specific nonattached culturing conditions, maintained through serial passaging, has been associated with a stem-like phenotype. Activity of pathways associated with stemness is increasingly being used to verify stem-like behavior. Additionally, loss of stemness-associated molecular and surface markers parallel to induced differentiation is also used. In this respect, growth of daughter cell lineages with differentiated marker signatures distinct from that of parent cells and/or other progeny are hallmarks of multi-lineage differentiation capacity.

Xenotransplantation has become the gold standard of in vivo testing. Putative CSCs should be tumorigenic in vivo, reproducibly giving rise to tumors when transplanted into immune-compromised mice. The cell numbers required for tumor initiation are preferentially lower than when inoculating unsorted or non-CSC phenotype cells. The tumors should ideally recapitulate the morphology and histological characteristics of parent tumors, verifying multi-lineage differentiation and hierarchical heterogeneity. The tumors should also be capable of serial passaging, that is, CSCs derived from primary xenograft tumors should initiate tumor growth and give rise to secondary tumors. While exemplifying the special capabilities of CSCs, xenotransplantation has also come under the most scrutiny as an adequate method for assessing tumorigenic potential (114–116). The question that frequently arises is whether we are indentifying all cells capable of tumor initiation or only a subset that has the unique capacity to propagate in mice. Experiments have shown that even non-CSCs can engraft and remain viable in the mouse environment, but fail to proliferate and form tumors (11, 23, 97). As more and more information surfaces about the active role of the tumor microenvironment in tumor growth, concerns amount that otherwise tumorigenic cells may lack stromal and interstitial factors in the xeno-environment and therefore fail to thrive (115). While one can argue that CSCs identified under these circumstances still have survival advantage and growth autonomy over the non-CSC population, whether this automatically translates into the same biological behavior and tumorigenic potential in the human body is an open question. The efficacy of the xenotransplant system was also stressed in an elegant work by Quintana et al. (117), who showed that the addition of Matrigel lowers the number of melanoma cells needed for tumor initiation in NOD/SCID mice, indirectly lending support to the significance of the inoculation environment. More importantly, by using NOD/SCID IL2Rγnull mice, which lack the natural killer activity of wild type NOD/SCID mice, the number of melanoma cells required for tumor initiation was drastically reduced to almost the single cell level and tumor initiation capability did not show an association with any particular cell surface phenotype. While this might be a unique property of melanoma, which is a highly aggressive malignancy known to metastasize early and at small tumor size; it clearly signifies that even in the immune-compromised NOD/SCID mice tumorigenic cells are lost due to immune-surveillance. Again, the biological significance and specific human relevance of cells with tumorigenic potential that cannot escape the immune-surveillance of NOD/SCID mice can be debated. One way to circumvent the potential negative influence of the xeno-environment is by using syngeneic murine model systems of human disease (118, 119). Of course, in these cases care needs to be taken to verify that the relevant cellular markers and signaling pathways are also present and implicated in the human counterpart.

Using the methods mentioned above, CSC populations with a characteristic cell surface expression profile were discovered in a plethora of human malignancies (for a summary and references, please see Table 1). Several of the most commonly used CSC markers are CD133, CD44, CD24, and ALDH1 in solid tumors and CD34 in hematological malignancies. Recently, several new markers have been identified, including CD105 (53), granulin-epithelin precursor (GEP) (36), c-Met (50), and integrin α6 (120). It is important to realize that all the markers used only enrich for the putative CSC population and the signature is not unique to CSCs. Often, the population negative for one or more CSC markers also has the potential for tumor initiation, although usually a very limited capability. One of the most controversial markers is CD133 (Prominin-1). Initial studies proved CD133 as a marker for brain and colon CSCs and later on also liver, lung, ovarium, pancreatic and prostate cancer. Several authors also found CSC activity in CD133 populations of glioblastoma and colon tumors (12, 61, 121), in some instances CSC functionality being exclusive to the CD133 population (122, 123). Günther et al. showed that under stem cell culturing conditions, cell lines derived from patient glioblastomas can form tumorspheres which, depending on the cell line, are either formed by CD133+ or CD133 cells (60). Although the cell lines with CD133+ tumorspheres proved to be more tumorigenic, the CD133 tumorspheres were also capable of initiating tumor growth. This controversy of CD133 expression in glioblastoma CSCs can be reconciled through the work of Chen et al., whom from PTEN-deficient glioblastomas isolated three subtypes of neurosphere forming cells, all capable of multi-lineage differentiation (59). Type I cells were CD133, produced CD133+ and CD133 daughter cells, showed the most aggressive tumor growth in xenotransplants and gave rise to histologically less differentiated tumors; Type II cells were CD133+, produced CD133+ and CD133 daughter cells, showed intermediate tumor growth and differentiation; Type III cells were also CD133, but produced only CD133 daughter cells, showed the slowest tumor growth and produced the most differentiated tumors. The authors concluded that in PTEN-deficient glioblastomas, several CSC lineages may be present, which are hierarchically organized and have different cell surface markers. An example of CSC marker plasticity was supplied by the work of Bae et al. (85). They showed that E-cadherin, which is expressed by prostate CSCs, was transiently downregulated during the migration of CSCs in a Matrigel assay and only returned to premigration levels several hours afterwards. It should also be mentioned that CSC markers are not necessarily universal in a given tumor type. For instance, the CD44+CD24 phenotype thought specific for CSCs in breast tumors has been shown to be only present in basal-type breast cancers and ALDH1+ is only applicable as a CSC marker in basal-type and HER2+ breast tumors (124).

Table 1. Cancer stem cell markers and associated tumor types
Tumor typeCSC marker
  1. GEP, granulin-epithelin precursor; SP, side population.

Brain+/–   + SP, integrin-α6(11–15, 120)
Breast++++ SP(14, 16–21)
Colon/colorectal++ +   (22–26)
Endometrium+      (27)
Gastric ++    (30, 31)
Head and Neck + +   (28, 29)
Leukemia (AML, CML)     +CD38(7–10)
Liver++ ++ GEP(32–36)
Lung+  +++SP, CD117(19, 37–39, 86)
Melanoma   +  CD20, ABCB5(40–42, 117)
Ovarian++    SP, CD117(43–47)
Pancreatic++++  c-Met, EpCam(48–51)
Prostate++    SP(14, 52)
Renal cell      CD105(53–54)
Thyroid      SP(55)

It also remains to be seen whether all CSC markers have an essential part in maintaining the functional CSC phenotype or their expression is just a byproduct of stemness. The role of CD44 is crucial for cell viability of hepatocellular carcinoma CSCs (33), intestinal carcinogenesis of mice (125), and bone marrow homing of chronic myeloid leukemia (CML) cells (126). Du et al. found that in colorectal cancer, CD44+ cells were enriched for CSCs with clonogenic and in vivo tumorigenic potential and both properties were impaired after siRNA knockdown of CD44 (58). Similarly, Pham et al. showed in primary cultures of malignant breast tumors that small hairpin RNA (shRNA) knockdown of CD44 in CD44+/CD24 breast CSCs reduced expression levels of Wnt, Hedgehog and Akt pathway associated markers to that seen in non-CSCs (57). The in vivo tumorigenicity of CSCs was also impaired with CD44 knockdown. In gastric CSCs, lentiviral knockdown of CD44 was shown to impair tumorsphere formation and tumor growth (30). Recently, Li et al. reported that they identified c-Met, a receptor tyrosine kinase, as a marker for CSCs in human pancreatic adenocarcinomas (50). In their study, lentiviral knockdown of c-Met inhibited in vivo tumor formation and pharmacological blockade of c-Met inhibited tumorsphere formation, induced CSC apoptosis and decreased tumor growth while concurrently reducing the CSC population of the tumor. In glioblastoma multiforme, integrin α6 was shown to enrich for CSCs (120). The shRNA knockdown of integrin α6 inhibited in vitro CSC growth and tumorsphere formation as well as in vivo tumor formation and tumor growth. The same effects were achieved using a blocking antibody against integrin α6. It has also been suggested that the ABCG2 multidrug-transporter not only causes the SP phenotype of putative CSCs through dye-efflux mechanisms, but also regulates Hedgehog signaling in these cells (127).

In general, whenever feasible, knockdown or blockade of a new CSC marker should be attempted to determine whether it influences CSC behavior and function. Ideally, gene expression should be monitored simultaneously, preferentially focusing on known stemness-associated pathways.


While initial research focused on the autonomous tumor initiation and growth properties of CSCs, recent studies have highlighted the important influence of the tumor environment not only on tumor behavior, but on CSCs themselves. Early on it was hypothesized that similar to their normal stem cell counterparts, a niche may exist that provides CSCs with environmental cues that help maintain stemness and may partially regulate CSC fate (128). Support for this theory was gained from histological and immunohistochemical slides showing cells with a CSC-associated marker phenotype preferentially enriched in perivascular regions of tumors (15, 120, 129, 130). As previously mentioned, initial perivascular localization of xenografted CSCs was recently visualized with intravital microscopy (97). There is mounting evidence that this is not a mere spatial coincidence, but also harbors a functional relationship. Beck et al. found that vascular endothelial growth factor receptor 2 (VEGFR-2) blockade decreased tumor microvascular density and also proportionally reduced CSC pool size in papillomas (130). Most remarkably, this was not solely owed to the reduced availability of nutrients, but also to a direct inhibition of CSC proliferation. They successfully verified a neurolipin-dependent autocrine VEGF loop, in which VEGF regulates CSC stemness-associated gene expression and proliferation. Similarly, underscoring the crucial interaction of tumor cells with surrounding stroma, paracrine hedgehog signaling with stromal elements was found to be essential for carcinogenesis and tumor angiogenesis (131–134). Even the normal stem cell compartment is implicated, as mesenchymal stem cells were shown to affect the phenotype of tumor cells, as reviewed by Grisendi et al. (135). In a mouse breast cancer model, Malanchi et al. verified that putative CSCs are the only cell population capable of forming distant metastasis (136). This function of CSCs, however, was dependent on inducing periostin (POSTN) expression in the host stroma via transforming growth factor β3 (TGF-β3). POSTN was shown to augment Wnt signaling in CSCs, and loss of POSTN or blockade of TGF-β3 action inhibited metastasis formation. Further evidence that CSCs are susceptible to environmental factors was provided by a study showing that stemness of glioma CSCs can be inhibited by human umbilical cord blood stem cells (HUCBSC), resulting in reduced in vitro migration and in vivo reduction of tumor volumes (90). The inhibitory effect of HUCBSCs was found to arise from reducing SOX-2 and TWIST1 expression in CSCs, which induced differentiation of the cells. The tumor environment can also significantly contribute to therapy resistance of tumors. Lonardo et al. found that despite successful targeting of Nodal signaling in CSCs in a pancreatic cancer cell line leading to diminished in vivo tumor growth in combination with gemcitabine, initial results with primary tumor xenografts were discouraging (137). They discovered that efficient accumulation of the therapeutic agent was inhibited; however, blocking the hedgehog pathway in stromal cells increased drug delivery by a factor of 10 and restored therapeutic efficacy, in line with a previous study showing stromal depletion as an effective measure for enhancing target site delivery of chemotherapeutics (138).

Hypoxia within tumors has been suggested to be a determinant of CSC behavior (139). It was shown that in glioblastomas, CSCs differentially express hypoxia inducible factors (86), with the HIF2α subtype preferentially expressed in CSCs (140). While knockdown of HIF1α inhibited both CSCs and non-CSCs, knockdown of HIF2α selectively decreased in vitro CSC cell survival and impaired in vivo tumor formation. Another group identified a set of signature genes differentially expressed in SP cells from glioblastoma cell lines and primary glioblastomas (141). The expression of signature genes as well as of CD133 was upregulated by culturing under hypoxic conditions. Overexpression of HIF2α had a similar effect, while knockdown of HIF2α inhibited hypoxic upregulation of genes. In human pancreatic cancer cell lines, hypoxia also elevated the expression of CD133 and increased invasiveness of cancer cells (142). In this study, overexpression of HIF1α increased metastatic and tumorigenic ability. Finally, in another study with prostate cancer cell lines, culturing under hypoxic conditions induced expression of HIF1α and HIF2α with an upregulation of Oct 3/4 and Nanog (143). Hypoxia also induced expression of CD44 and ABCG2 and led to higher clonogenicity and increased sphere formation. The hypoxia induced changes were already evident at a moderate oxygen level of 7%. This oxygen level, thought to be representative of physiological concentrations in human tissues, was also shown to increase HIF2α expression in glioblastoma CSCs (144). Upregulation of Oct 4and Sox2, and an increase in proliferation rate and self-renewal potential of CD133+ cells were also witnessed. An interesting signaling pathway was identified by Feldman et al., who found leptin receptor to be expressed via an OCT4/SOX2 mechanism on hepatocellular carcinoma cell line CSCs (145). In CSCs, leptin induced phosphorylation of STAT3 and induction of Oct4 and Sox2. Additionally, intraperitoneal injection of leptin into xenografted mice led to faster tumor growth and CSCs formed tumors faster in obese mice than in lean littermates. Immunostaining of hepatocellular carcinoma specimens demonstrated leptin receptor expression in putative CD133+ CSCs. Leptin driven Notch-1 signaling has also been identified in breast cancer cell lines (146). The clinical relevance of leptin augmented tumor development could be especially high in developed countries with obesity-prone populations. The diversity of identified environmental factors shows that CSC-niche interactions are at least as complex as the CSCs themselves. It has also been hypothesized, that several different CSC niches may coexist, with different pathways active in perivascular, hypoxic, and metastatic environments (62).

In summary, the stromal compartment may play as much a pivotal role in determining tumor growth and metastasis, as the intrinsic properties of CSCs. However, the CSC niche need not necessarily evolve from nontumorous cell populations. In two studies focusing on angiogenesis in glioblastomas, vasculature of tumors was shown to be of partially neoplastic origin. Ricci-Vitiani et al. showed a portion of endothelial cells in glioblastomas carry the same genetic alterations as tumor cells (147). They also found that in xenograft tumors grown from glioblastoma CSCs, the endothelial lining of vessels is primarily composed of cells of human origin. Wang et al. also showed somatic mutations to be shared by tumor cells and a fraction of endothelial cells within the tumor (148). Through lineage analyses they were able to trace the endothelial cells to a special subset of CD133+ glioblastoma CSCs, which can differentiate into endothelial progenitors which in turn give rise to the endothelial population. They found the transition from CSC to endothelial progenitor to be VEGF dependent, whereas further maturation to endothelial cells relied on effective Notch-1 signaling. A new layer of complexity is added if we consider that hedgehog signaling was shown to induce VEGF production in fibroblasts (131), which in this context may not only promote angiogenesis per se, but also the cross-differentiation of CSCs. Results similar to those with glioblastoma were seen in xenografts of human renal cell carcinoma, where tumor vasculature was also of human origin (53). These studies shed light on a new facet of CSC function and hint that the role of CSCs in tumor biology may go beyond maintaining the strictly taken tumor cell population. Interestingly, the capacity of CSCs to differentiate into endothelial cells seems to be a trait shared with normal pluripotent- and stem cells, which were shown to be capable of cross-lineage differentiation (149–152). Therefore, cross-differentiation might be a further normal stem cell property conserved in CSCs.


The clinical significance of CSCs has been a subject of much debate. It was speculated that stemness features of CSCs would allow them to evade conventional anticancer therapy and maintain minimal residual disease, eventually leading to tumor relapse (153, 154). While the therapy resistance in vitro, and in xenotransplanted tumors has been well documented (64, 155, 156), there was little proof that poor therapy outcomes in patients can be directly linked to CSCs. Recently, it has been shown that immunohistochemically verifiable CSC marker expression in histological slides from patient tumors has prognostic value and is associated with a poorer clinical outcome (65–70, 157, 158). Similarly to circulating cancer cells (159–162), the presence of circulating CSCs was proven to be a prognostic factor for faster disease progression (163–165). Further, a relationship was established between myeloma CSC numbers in patients and progression free survival after treatment with rituximab (153). In breast cancer, the number of breast CSCs in patient's core needle biopsies increased after chemotherapy, demonstrating a survival advantage of CSCs over the bulk tumor population (166). While CSCs were thought to resist therapy through drug efflux mechanisms, antiapoptotic ability, efficient DNA repair and slow cycling (167), two recent studies have suggested other coping mechanisms that may be active in the CSC population. Chaterjee et al. found that breast CSCs can enter dormancy upon treatment with farnesyl transferase inhibitors, a class of drugs designed to inhibit proper maturation of Ras (168). This induced dormancy was maintained by CSCs through autophagy and the cells were able to reenter the cell cycle upon withdrawal of the drugs. Viale et al. showed p21 dependent cell cycle restriction to be functional in CSCs in a mouse leukemia model (169). The cell cycle restriction not only prevented DNA damage accumulation, but also functional exhaustion of CSCs and therefore contributed to maintenance of stemness. The role of CSCs in minimal residual disease and relapse was supported by the work of Hamilton et al., who showed that a subpopulation of Bcr-Abl positive CML CSCs are able to survive the withdrawal of growth factors in addition to combined Bcr-Abl knockdown and pharmacological inhibition by dasatinib, a tyrosine kinase inhibitor (170). These cells were quiescent under the harsh conditions and although they did not proliferate, remained viable. Upon withdrawal of dasatinib and addition of growth factors, the cells reentered the cell cycle and began to proliferate.

It has become clear that traditional treatment regiments are inefficient at eliminating the CSC population and new therapeutic approaches are needed that take into account the special properties of CSCs. Basically, a reduction in the CSC population can be achieved by inducing differentiation/apoptosis in CSCs by targeting receptors or signaling pathways crucial for CSC self-maintenance and viability. This approach has started to show promising results. The paracrine signaling of colon CSCs through interleukin 4 (IL-4) was successfully blocked by anti-IL-4 antibody treatment, reducing in vitro and in vivo chemotherapy resistance (24). In human renal cell carcinomas, IL-15 treatment of CSCs abolished tumorigenic potential and blocked CSC self-renewal (54). A heterodimer of bone morphogenetic protein efficiently inhibited TGF-β dependent breast CSC function, decimating CSCs in vitro and inhibiting metastases formation in vivo (171). Reduction in Nanog-1 expression in breast CSCs by 3-O-methylfunicone inhibited sphere formation and colony formation (172). An off-target shRNA against neostemin, lentivirally transfected into glioblastoma cells, led to preferential apoptosis of CD133+ putative CSCs and significantly slowed xenotransplanted tumor growth (173). Another study found shRNA knockdown of ubiquitine ligases to induce apoptosis and differentiation in cell line derived glioblastoma CSCs (174). As previously discussed, blockade of TGF-β3 action with a secreted decoy receptor inhibited metastasis of mouse breast CSCs (136) and antibody targeting of integrin α6 in glioblastoma inhibited CSC driven in vivo tumor initiation and growth (120). Recent development of subtype-specific Notch antibodies demonstrated efficacy by inhibiting Notch-dependent xenograft tumor growth without the side effects of pan-Notch blockade (175). A selective antibody targeting Notch-1 induced apoptosis in a breast cancer cell line, reduced the CSC population and irreversibly impaired CSC mammosphere formation (176). Oxymatrine was also shown to diminish the side population in a breast cancer cell line by decreasing β-catenin levels and activation (177).

The successful targeting of CSCs may, however, not be sufficient to achieve adequate disease control. A computational modeling of CSC and non-CSC population dynamics concluded that differentiation therapy of slow cycling, self-renewing populations or cytotoxic therapy of a larger, faster proliferating population alone cannot eliminate tumorous growth (178). These calculations did not utilize any specific assumptions about tumor growth regulation, they only hypothesized that relative population densities regulate proliferation and asymmetric cycling of CSCs. A recent study demonstrated the viability of this assumption, with CSC densities reaching the same level in in vitro mixed cultures of CSCs and non-CSCs, irrespective of the initial ratio of the populations after plating (179). Therefore, efficient tumor therapy is suggested to be dependent on parallel administration of cytotoxic agents in combination with targeted CSC elimination. Several studies have found a systematic synergistic effect between traditional chemotherapeutics and newer, CSC-targeted agents. Checkpoint kinase 1 inhibition in combination with chemotherapy greatly enhanced in vitro cytotoxity and in vivo efficacy of chemotherapy against non-small cell lung CSCs (180). Acting through ataxia telangiectasia mutated, an oncolytic virus and the alkylating agent temozolomide synergistically killed glioblastoma CSCs while sparing neurons (181). In pancreatic cancer, inhibition of c-Met reduced the CSC fraction in xenograft tumors; however, full growth inhibition was only achieved after adding gemcitabine to the treatment regimen (50). Similarly, stable disease and growth inhibition of primary pancreatic cancer xenografts was only accomplished with triple therapy consisting of gemcitabine, knockdown of Nodal in CSCs and blockade of smoothened in stromal elements (137). Also in pancreatic cancer, combined gemcitabine, sonic hedgehog- and mTOR inhibition abolished metastatic activity and reduced xenograft tumor growth (88). Finally, retinoid compounds blocked the increase in ALDH1+ cells after carboplatin therapy of ovarian cancer cell lines and in combination with carboplatin significantly reduced in vivo tumor growth (182). While the novel combination therapies show great potential, tolerability in the clinical setting and especially the long term effects on normal stem cells still need to be determined. In a special setting, the evolving approach of tumor therapy with the patient's reprogrammed and expanded own T cells (183) can be ideally exploited to target CSC specific surface antigens (72). This approach has first been confirmed for targeting CD20 on CSCs in a xenograft model of human melanoma cells (184) and then successfully applied in a clinical case (185).


Our understanding of CSC biology is rapidly expanding. Attention is gradually shifting from merely identifying markers that enrich for CSCs to untangling the complex web of signaling pathways that determine the stem-like behavior. These pathways are often dysregulated counterparts of networks functionally important in the maintenance of normal tissue stem cells. We are also beginning to recognize that CSCs are not lonely fighters, they often interact with and modulate signaling of stromal elements to elucidate a paracrine environment favoring tumor growth and invasion. To what extent stromal factors influence CSC behavior and how essential these factors are for tumor propagation still remains to be seen. An intriguing new aspect of CSC biology is their potential to not only give rise to the various malignant cell lines within a tumor, but also to produce progeny that can differentiate into various supporting tissues and structures. The clinical significance of CSCs is growing, with the CSC phenotype representing an independent prognostic factor for poorer patient outcome. Indirect and direct evidence for therapeutic consequences of CSC behavior, such as therapy resistance, minimal residual disease and relapse are mounting. Luckily, every identified marker and pathway represents a new potential target for effective therapy. Initial experience with targeted CSC therapy is promising and hopefully the preliminary in vitro and in vivo results, whether alone or in combination with traditional cytotoxic agents, can be carried over into clinical trials and eventually lead to successful cancer therapy.