Based on the clonal evolution model and the assumption that the vast majority of tumor cells are able to propagate and drive tumor growth, the goal of cancer treatment has traditionally been to kill all cancerous cells. This theory has been challenged recently by the cancer stem cell (CSC) hypothesis, that a rare population of tumor cells, with stem cell characteristics, is responsible for tumor growth, resistance, and recurrence. Evidence for putative CSCs has been described in blood, breast, lung, prostate, colon, liver, pancreas, and brain. This new hypothesis would propose that indiscriminate killing of cancer cells would not be as effective as selective targeting of the cells that are driving long-term growth (ie, the CSCs) and that treatment failure is often the result of CSCs escaping traditional therapies.
The CSC hypothesis has gained a great deal of attention because of the identification of a new target that may be responsible for poor outcomes of many aggressive cancers, including malignant glioma. As attractive as this hypothesis sounds, especially when applied to tumors that respond poorly to current treatments, we will argue in this article that the proposal of a stemlike cell that initiates and drives solid tissue cancer growth and is responsible for therapeutic failure is far from proven. We will present the point of view that for most advanced solid tissue cancers such as glioblastoma multiforme, targeting a putative rare CSC population will have little effect on patient outcomes. This review will cover problems with the CSC hypothesis, including applicability of the hierarchical model, inconsistencies with xenotransplantation data, and nonspecificity of CSC markers.
The approach to cancer treatment has traditionally been to kill all of the cancerous cells to achieve a “cure” and is based on the idea that the vast majority of cancer cells have the potential to proliferate, self-renew, and drive tumor growth.1,2 This notion is derived from the stochastic or clonal evolution model, which posits that transformed single cells gain unlimited proliferative capability to cause a tumor.3 During early tumor development, a single or a few cells transform, producing uncontrolled growth with a resulting evolutionary accumulation of different mutations. These mutating cells drive tumor growth and result in heterogeneous subpopulations within the tumor.4–8 Importantly, over the lifetime of the tumor, any of the cancer cells can participate in tumor growth or develop resistance and cause recurrence. Hence, all cells are considered tumorigenic and therefore are targets for treatment. This approach has formed the backbone of a great deal of studies designed to eliminate tumor burden. However, for the majority of solid tissue cancers, cures are rarely obtained and death rates have changed very little over the past 50 years.9
This theory has been challenged by the recently resurrected hierarchical model or cancer stem cell (CSC) hypothesis.10–12 Similar to the stochastic model, this theory proposes that single mutated cells gain unlimited proliferative potential and create a tumor.13,14 The proliferative cells that exclusively drive tumor growth are rare and have some stem cell qualities.15 In contrast to the stochastic model, the multipotent nature of these cells results in heterogeneity within a tumor as a result of aberrant differentiation and epigenetic changes of the progeny. The majority of the cells (progeny) do not contribute to tumor growth, cannot form secondary tumors, and have the same genetic abnormalities of the hierarchical cell from which they originated. This rare population of stemlike cells are responsible for tumor growth, recurrence, and resistance, implying that indiscriminate killing of all cancer cells may be an inefficient and ineffective method to treat cancer because it is not able to target or eliminate the rare cell that is responsible for tumor growth. The development of a new approach to treating a disease that often has few options has sparked the imagination of scientists and activists.
As attractive as this hypothesis sounds, especially when applied to tumors that respond poorly to current treatments, we argue in this article that the proposal of a stemlike cell that initiates and drives solid tissue cancer growth and is responsible for therapeutic failure is far from proven. We present the point of view that for most advanced solid tissue cancers such as malignant glioma, targeting a putative rare CSC population will have little effect on patient outcomes. The discussion reviews the hierarchical model, the definition of somatic stem cells and CSCs, and the literature against the CSC hypothesis.
WHERE DID THE CANCER STEM CELL HYPOTHESIS COME FROM?
The CSC hypothesis challenges the stochastic model of tumor initiation, which proposes that tumor cells are heterogeneous but virtually all of them can function as a tumor-initiating cell (TIC).16 Alternatively, the CSC hypothesis follows a hierarchical model in which a small subpopulation of CSCs can proliferate extensively and sustain growth and progression of a neoplastic clone (Figure 1).
This is not a new concept. Rudolph Virchow (in 1855)17 and later Julius Cohnheim18 proposed, on the basis of observed similarities between fetal tissues and cancer both histologically and in terms of proliferation and differentiation, that cancer develops from activation of dormant embryonic-tissue remnants. More than 100 years later, as hematopoietic stem cells were being discovered, investigators found that tumor cells have a heterogeneous potential to self-renew in vitro and in vivo.19 In the 1960s, Alexander Brunschwig, Chester Southam, and Arthur Levin,20 in ethically questionable experiments, demonstrated a low frequency of tumor formation when tumor cells harvested from patients with malignancy were injected subcutaneously into the same or different patients. They found that tumors formed in these patients only when > 1 × 106 cells were injected. These observations, along with isolation of colony-forming lymphoma cells by injecting lymphoma in mice and collecting cells from the resultant splenic colonies,21 led to the concept that cancer growth may be initiated and maintained by a minority of the cancer cells, not the entire tumor population. Coming full circle, Virchow's original “embryonal-rest” hypothesis developed into the concept that a primitive embryonal-like cell may be responsible for tumor initiation and progression. This phenomenon was based on the discovery of a low-frequency, highly proliferative somatic stem cell and the subsequent observation that a minority of cancer cells are able to form secondary tumors.
This notion eventually became known as the CSC hypothesis, a premise that a relatively rare cell population within a tumor sharing common criteria with somatic stem cells represent the only cells with the capacity for limitless self-renewal and these cells initiate and drive tumor growth.22,23 Now, evidence for putative CSCs has been described in blood,24,25 breast,26 lung,27 prostate,28 colon,29,30 liver,31,32 pancreas,33,34 and brain35–38 (Table 1).
The CSC hypothesis can be used to describe the scenario when treatment kills the proliferating cell population and the patient seems to have a positive response to treatment. However, CSCs remain, as a result of treatment resistance related to their quiescence or ability to recover from treatment toxicity, which eventually leads to relapse. In this scenario, the relapsed tumor is identical to the original tumor because the progeny are produced by the same CSCs. A classic example of this phenomenon is leukemic response to Gleevec (Novartis Pharmaceuticals Corportation, East Hanover, New Jersey). Patient survival is increased with Gleevec, but relapse can take place after the drug is discontinued.39,40 This model rests on the principle of separation between proliferation and self-renewal.41 The CSCs have the ability to self-renew and to generate progenies that are responsible for tumor proliferation.
HOW DO WE DEFINE SOMATIC STEM CELLS?
The definition of a CSC is derived from the somatic, noncancerous stem cell definition based on functional criteria (Table 2). Somatic stem cells in adult tissue were first described in the hematopoietic system. A series of experiments were conducted in which bone marrow cells were taken from mice and then transplanted peripherally into irradiated hosts.42,43 Only a subset of cells resulted in clonally derived colonies in the spleen, with some colonies capable of producing additional colony-forming cells when isolated and retransplanted (Figure 2).
Subsequently, mouse hematopoietic stem cells were grown in cell culture and described as cells with the ability for extensive proliferation, self-renewal, differentiation, and production of multilineage progeny (Figure 3).44–46 Classically, somatic adult stem cells are relatively quiescent to preserve their self-renewal capacity over time47 and have the ability to generate a transiently amplifying non–stem cell population to replenish the cells in a particular tissue.11,48–50 A stem cell is able to maintain its phenotype while generating non–stem cell progeny (ie, the transiently amplifying cell) by dividing asymmetrically. Whereas stem cells conserve their proliferative ability, progenitor cells divide rapidly, are committed to differentiation, and undergo a more limited number of divisions before dying or terminally differentiating.51
SOMATIC NEURAL STEM CELL
The belief that the adult mammalian brain is unable to generate new brain cells was entrenched in the scientific and medical community for the better part of the last century, despite findings of mitotic cells in the postnatal mammalian brain.52–54 So strong was this dogma that reports in the 1950s to 1970s of new neurons being generated in the mature mammalian CNS were largely ignored,55–59 while demonstration of seasonal neurogenesis in adult song birds by Fernando Nottebohm and coworkers60,61 was thought to be restricted to birds and other nonmammalian species. One of the confounding problems for the acceptance of continuous neurogenesis in the mammalian central nervous system (CNS) was the lack of any evidence of a stem cell population that would sustain new cell genesis throughout the life of the animal. Although several technical advances contributed to studies that overturned the “no new neuron” dogma, the demonstration in 1992 by Reynolds and Weiss62 that the mature mammalian CNS contained a stem cell population supported the notion that the adult brain had the capacity to generate new cells.
This discovery opened a new area of research into the investigation of the location and purpose of these cells in the CNS. Cell culture experiments found that < 0.1% of the total cell population from the adult mouse striatum could proliferate and generate multipotent clones of cells called neurospheres when placed in serum-free culture with growth factors. Differentiated cells could not be sustained in this culture system. Conversely, neural stem cells (NSCs) responded to the growth factors, divided, and formed neurospheres, which could be dissociated and replated for secondary spheres, generating a large number of progeny. This could be repeated virtually indefinitely with NSCs compared with progenitor cells that could be sustained for only 1 or 2 passages (Figure 4).63–65 To achieve reliable and consistent results, investigators found that sphere overgrowth had to be avoided to prevent depletion of the medium and increased apoptosis and/or differentiation that can occur specifically in the core of the spheres.66 Additionally, when the growth factors were removed, the cells differentiated into neurons, astrocytes, and oligodendrocytes. Therefore, adult NSCs were found to possess all of the fundamental criteria for stem cells: self-renewal, multipotency, and the ability to generate many functional progenies67 (Table 2). After this study, adult NSCs were discovered lining the ventricular system (subventricular zone)68 and in the subgranular zone of the hippocampus.69 Other culture techniques have been developed to propagate and maintain an enriched NSC population. These include culturing with laminin to grow cells as a flat monolayer instead of spheres.70
The definition of somatic stem cells and NSCs is based on functional criteria and retrospective analysis. To measure these characteristics, the cell must be manipulated. Self-maintenance and extensive proliferative potential, properties associated with the future of the cell, require the cell to be placed in certain conditions in an attempt to measure its potential. It is only when the cell acts that we can measure its ability to perform. In doing so, we place cells under conditions they may not normally be exposed to in our attempt to observe their range of behaviors. We are always left with the unanswered question of whether the observed behavior occurs normally or whether we have induced a phenotype that does not actually exist in nature. This is often incorrectly referred to as the biological equivalent of the Heisenberg uncertainty principle. This principle posits that it is impossible to simultaneously determine both the position and momentum of a particle because as one property is calculated more precisely, the other becomes increasingly inaccurate. This phenomenon cannot be overcome by altering the method of measurement. However, the problem with defining stem cells is better characterized as the observer effect because it is a function of the technique used to make the measurement and may be overcome by changing the experiment. Hence, in measuring stem cell properties, we inevitably alter the characteristics of the cells, and caution is advised in interpreting those results in situ.16
CSCs were originally considered somatic stem cells that had undergone oncogenic mutations to initiate and drive tumor growth.25,71 However, Krivtsov et al72 demonstrated that tumors do not always originate from mutated somatic stem cells. They introduced mixed lineage leukemia–AF9 fusion protein into granulocyte macrophage progenitors and implanted them into sublethally irradiated mice. The mice formed acute myeloid leukemia (AML), and their leukemias were transplantable into secondary recipients, demonstrating the capacity for self-renewal of the progenitors. This experiment and others73–75 demonstrated that cancer could be initiated with transformed progenitor cells that gained self-renewal capabilities and did not always require mutations of somatic stem cells. These studies took advantage of the fact that genetic translocations in cancer cells could be matched to the same genetic abnormality in the cell of origin, and this cell was not always a somatic stem cell. This idea led to the use of the term CSC to describe TICs without inference of origin from somatic stem cells.48,76,77
After adult somatic hematopoietic stem cells were described using severe combined immunodeficient (SCID) mice and cell culture–based experiments, these methods were used to study leukemic blast cell populations in leukemia.71,78 Applying the methods described by Linder and Gartler79 to demonstrate clonal origins of leiomyomas, Fialkow et al80 demonstrated that leukemias also develop from a single cell. They studied chronic myelogenous leukemia patients who were heterogeneous for X-chromosome–linked enzyme glucose-6-phosphate dehydrogenase (G6PD) and found that erythrocyte and granulocyte preparations demonstrated only 1 type of G6PD enzyme, which again was evidence of clonal origin.80–82
Building on these data, Buick et al83 took AML tumor samples from 44 patients and demonstrated colony formation when the single cells were grown in methylcellulose. These cells were capable of proliferation, self-renewal, and generation of a large number of progeny, like somatic stem cells. Furthermore, they found that when they took the original colonies, dissociated them into a single cell suspension, and returned them into culture, the number of colonies formed varied on the basis of the patient's response to treatment. The samples from “remitters” had the formation of fewer colonies after the first pass compared with “nonremitters.”83 These results support the correlation between CSCs and disease outcome and led to the notion that cancer cells with stem cell qualities are responsible for tumor initiation, maintenance, and resistance. In addition, the identification of CSCs and experimentation using the soft-agar stem cell assay was modified and used for other types of tumors with a plating efficiency of 0.001% to 0.1%, suggesting that only a small population of cells represented CSCs, defined by the ability to proliferate in culture.84–87
On the basis of these findings, CSCs are now characterized by the ability to initiate cancer growth on implantation, extensive self-renewal in vitro or in vivo, genetic alterations, aberrant differentiation, capacity to generate tumorigenic and nontumorigenic differentiated cells, and multilineage differentiation capacity88 (Table 2). According to the CSC hypothesis, cells that initiate and drive cancer growth are rare and are resistant to treatment because of their stem cell qualities. Moreover, designing a therapeutic protocol targeting this population is necessary to ensure treatment success. The CSC hypothesis has now been extended from simply demonstrating the existence of CSCs to implicating these cells in tumor initiation, maintenance, and recurrence.89,90
EVIDENCE SUPPORTING THE CSC HYPOTHESIS
Identifying CSCs in vitro, while useful for experimentation, did not lend itself to identifying in vivo and targeting CSCs clinically. This problem subsequently became the focus of CSC research. In 1994, John Dick's laboratory reported the characterization of a human AML cell that could initiate AML in sublethally irradiated nonobese diabetic/SCID (NOD/SCID) mice.91 Dick and colleagues found that transplanting blood from AML patients into these mice resulted in complete engraftment of the animals' bone marrow, and the mice died of AML. Analyzing the leukemic cells alone in culture at various dilutions demonstrated that about 0.3% of them resulted in colonies in culture. Because CD34 is a marker of pluripotent cells in normal bone marrow and CD38 is a marker of lineage commitment, they subsequently separated the cells into CD34+CD38+ and CD34+CD38− (pluripotent with and without lineage commitment) AML cells by flow cytometry. Both groups formed colonies in culture, but only the CD34+CD38− group resulted in tumors in NOD/SCID mice. For these experiments, 1 to 4 × 107 cells were transplanted, and the mice were treated with growth factors for the duration of the experiment. Bonnet and Dick24 concluded that cells initiating tumor in vivo were immature (ie, possessed more “stemness”) compared with the cells that formed colonies only in culture. These articles identified an AML CSC based on the ability to form colonies in culture and to form tumors when transplanted in immunocompromised animals. Problematically, the CD34+CD38+ fraction in this experiment was approximately 75% pure, contaminated with CD34+CD38− cells. Although selection for CD34+CD38− does enrich for TICs, the described contamination brings into question the definitive conclusion that CD34+CD38− cells are exclusively responsible for tumor growth.
Similarly, Al-Hajj et al26 reported in 2003 the identification of the first solid tissue CSC for breast cancer. Initially, they found that tumors would form with transplants into NOD/SCID mice of human carcinoma breast cancer cells that either were immunoreactive for CD44 or B38.1 or had low to no CD24. Some of these markers have been used to identify mammary somatic stem cells. In 2006, Shackleton et al92 published results demonstrating complete reconstitution of the mammary gland in vivo with a single cell that was Lin−CD29hiCD24+. Al-Hajj et al found that cells that were negative for CD44 or B38.1 did not form tumors. They did find that approximately 16% of the mice developed small tumors when transplanted with CD24+ cells. The authors attributed this finding to a contamination with CD24− cells. Because all of the CD44+ cells were also CD38.1+, subsequent experiments focused on the CD44 and CD24 markers. They isolated 2 cell populations, CD44+CD24low/− and CD44−CD24+, and found that only the former exhibited the ability to form tumor when transplanted in NOD/SCID mice.26 By enriching the cell population even further for CD44+CD24low/−ESA+, they increased the ability to form tumors by 10- to 50- fold compared with the total cell population, which supports the CSC hypothesis. In contrast, the finding that 16% of the mice developed small tumors when transplanted with CD24+ cells suggests that CD24+ cells also have tumor-initiating capability. This challenges the validity of these markers to distinguish CSCs from cells that are not tumor initiating.
Phenotypic identification of CSCs was also accomplished in other solid tumors28,34,93–95 (Table 1). For colon cancer, a rare population (2.5%), CD133+ was discovered to have an increased tumor-initiating capability in NOD/SCID mice compared with the total cell population.93,94 Ricci-Vitiani et al94 found that CD133+ colon cancer cells formed tumors much more efficiently in immunocompromised animals compared with unseparated colon cancer cells. In this study, CD133− cells did not form tumors at all. Conversely, a subsequent study demonstrated that tumors did form with 2.5 × 105 injections of CD133− cells 11% of the time,93 creating doubt in the claim that CSC marker–negative cells do not initiate or drive cancer growth.
EVIDENCE OF BRAIN CSC
These concepts were first extended to brain tumors by Ignatova et al,37 who isolated clonogenic, neurosphere-forming stemlike cells from human glioblastoma multiforme (GBM). The neurosphere assay had previously been shown to be a robust method for the isolation and expansion of somatic NSCs.62 Subsequently, GBM CSCs in neurosphere culture (Figure 5) were shown to differentiate into multilineage progeny and had the ability to form tumors in immunocompromised mice,35 establishing the neurosphere assay as a reliable culture system to study and propagate brain CSCs. GBM tumor neurosphere formation in culture was found to correlate with poorer patient outcomes.96 Moreover, the rate of GBM tumor cell growth in culture (cellular fold expansion) was associated with disease progression.97 Using bone morphogenetic protein 4 to force GBM CSCs to differentiate, Piccirillo et al97 could significantly reduce cell growth in vitro and in vivo. Therefore, changes in cell growth measured by the neurosphere assay reflect changes in the CSC compartment. GBM CSCs have also been studied and propagated using alternative culture techniques such as growing the cells as a monolayer using laminin as a substrate.98
Using the neurosphere assay, Hemmati et al36 modeled the findings by Ignatova and colleagues37 by demonstrating that human medulloblastoma could form neurospheres in culture. Furthermore, these cells were positive for somatic stem cell markers such as CD133, Sox2, musashi-1, and bmi-1, and after transplantation into neonatal rat brains, the tumor cells would proliferate, migrate, and produce neurons and glia. Singh et al38 demonstrated that approximately 20% of brain tumor cells (medulloblastoma and GBM) could form neurospheres. Interestingly, about 20% of the tumor cells were also positive for CD133. Singh et al sorted the cells into CD133+ and CD133− groups and transplanted them into NOD/SCID mice. Sixteen of the 19 CD133+ transplants resulted in tumors. On the other hand, none of the 15 CD133− transplants resulted in tumors.99 CD133 was also found by immunohistochemistry in tumor samples and determined to be associated with poorer patient outcomes.100 Subsequently, other groups found that CD15+ brain tumor cells were tumorigenic in a xenograft model and may represent a more reliable marker of brain CSCs.101 Hence, brain tumor CSCs were identified on the basis of the ability to form tumors in immunocompromised mice, to form neurospheres, and to generate a large number of multilineage progeny.
The brain CSCs were further characterized by their relationship with surrounding vasculature. GBM, in particular, is known to have endothelial proliferation, and tumor cells are often found surrounding blood vessels (perivascular satellitosis).102 These histological findings were then extended to brain CSCs.103 Bao and colleagues104 demonstrated that CD133+ GBM cells produced a high level of vascular endothelial growth factor and that the CD133+ cells exclusively, not the CD133− cells, could form highly vascular tumors in a mouse model. When these cells were treated with bevacizumab, an antibody to vascular endothelial growth factor, they lost their ability to form tumors in vivo.104
Additionally, brain CSCs were found to have variable genetic expression profiles. First, GBM CSCs grown in the neurosphere assay were found to recapitulate the genotype, epigenetic pathways, and in vivo biology of the primary tumor sample.105 Subsequently, epigenetic regulation of CSCs (ie, epigenetic silencing of bone morphogenetic protein 4) was demonstrated to change tumor growth and disease outcome.106 Brain CSCs, like other CSCs, were hypothesized to be genetically unstable and epigenetically deregulated, making them the driving compartment for tumor initiation, growth, resistance, and recurrence.107
IMPLICATIONS FOR THE TREATMENT OF CANCER
A defining characteristic of somatic stem cells is their relative infrequent cell division, or quiescence. If general quiescence is present in the CSC population, then they would be hypothesized to be less sensitive to classic antiproliferative treatment regimens.108–110 CSCs from hematopoietic malignancy and solid tissue cancer have been shown to have high expression of drug transporters, potentially resulting in resistance to chemotherapeutic agents.111–115 In GBM, after radiation treatment to the entire tumor cell population, the fraction of CD133+ cells has been found to increase. Bao et al116 attributed this finding to relative radiation resistance of the CD133+ brain tumor cells that exhibit a more effective radiation-induced DNA damage repair process compared with the CD133− population. Gamma radiation has also been found to lack significant effect on the neurosphere-forming ability of CD133+ GBM cells compared with CD133− cells.117 Moreover, GBM CSCs (defined as CD133+ cells) were found to have decreased growth inhibition response to temozolomide in the neurosphere assay compared with the CD133− cells.118 However, in lung cancer, CD133+ and CD133− cell populations have had similar growth inhibition response to chemotherapy.119 The demonstration of CD133+ brain tumor cell resistance to therapy has been used to support the concept that CSCs are responsible for tumor resistance and recurrence after apparent clinical remission.
As part of the CSC hypothesis, indiscriminate therapy that targets all proliferating cells spares the resistant CSCs, resulting in tumor recurrence, even after tumor response to treatment.120 This theory has been described by the “dandelion hypothesis,” that similar to weeds, tumors recur after the cell progeny (leaves) have been killed but the CSCs (roots) have been left behind.121 Using this premise, the tumor would be indefinitely controlled because the progenitors would continue to be sensitive to treatment during every relapse. This theory has been built on a leukemia model but has not been applied in solid tumors. Unfortunately, brain tumors such as GBM very rarely demonstrate clinical remission, thereby implicating the majority of tumor cells, not just CSCs, in tumor resistance and recurrence.
PROBLEMS WITH THE CSC MODEL
For most cancers, including GBM, the dandelion hypothesis does not apply because only a small portion of the tumor responds to treatment instead of a small population surviving treatment.115 The stem cell model of drug resistance applies only to the limited cases in which therapy results in clinically perceived complete remission, with delayed relapse months or years later. In some tumors such as testicular cancer, the “CSCs” that drive tumor growth are actually more sensitive to cisplatin therapy compared with the differentiated cells,122 the opposite of what the CSC hypothesis posits. Applying concepts from hematologic malignancies to solid tissue cancer is problematic because of the lack of widely accepted cellular markers and developmental paradigms in solid tissue tumors. Moreover, in many cases such as GBM, therapy has no effect on tumor growth or has only limited effect with relapse resulting from therapy-resistant progenitors.51,123 A large number of tumor cells remain after treatment, not just a rare population of CSCs, and many of these cells may be capable of recapitulating the tumor.
The applicability of the hierarchical model struggles when the relapsed tumor is no longer sensitive to treatment. Chronic myelogenous leukemia will eventually result in “blast crisis” in which proliferation and self-renewal take place in one cell that drives tumor growth and cannot be controlled indefinitely.124 This phenomenon cannot be explained by a model in which the original CSCs self-renew and repopulate proliferating treatment-sensitive progeny. Either CSCs begin rapidly proliferating and forming the bulk of the tumor population, or progenitors become immortal or develop CSC features.11,48,115,124–127 If genetic and epigenetic changes in tumor cells can confer stem cell characteristics and growth advantage, any tumor cell has the potential to drive tumor growth and resistance. This concept supports the stochastic model in which any cell can transform (with genetic or epigenetic changes) and develop a clone of cells, so a tumor can consist of multiple different clones, each with their own growth advantage. Shipitsin et al found that breast cancer cells that expressed CD24 and CD44 (previously shown to be breast CSCs) did have stem cell characteristics.128 Conversely, they found that these cells were genetically different, suggesting the cells originated from different clones that had each undergone their own transformation to gain a growth advantage, supporting the stochastic model.
Cells Come in and Out of CSC “Pool”
Changes in oncogene expression, often induced by environmental changes, have been shown to transform relatively differentiated tumor cells into undifferentiated CSCs that also drive tumor growth.129,130 Zheng et al131 and Kelly et al132 found that they could increase the clone-forming capacity in culture of their malignant glioma cell line (C6) to almost 100% when supplementing the media environment with serum. They reported that the majority of their malignant glioma cell line (C6) consisted of cells with the capacity for self-renewal and proliferation. Immortal cell lines are an example of this population of self-renewing, proliferating cells. In optimal conditions, all of these cells proliferate, exhibiting exponential growth without spontaneous differentiation.41 This scenario is best described with the stochastic model in which all of the cells have the potential to be a tumor-founding cell.133 For all cells to have the potential to be a TIC, they may need to differentiate or dedifferentiate for tumor formation. Tumor cells, as a result of genetic or epigenetic differences, can demonstrate phenotypic differences that can confer growth advantage or treatment resistance to drive and maintain tumor growth.134,135 This is especially true for GBM tumor cells, which have highly variable genetic expression profiles.
Differentiation of TICs to confer growth advantage was demonstrated by Bruggeman et al.136 They showed that GBM TICs that lacked Bmi-1, a marker for TICs, failed to exhibit pseudopalisading and had reduced differentiation capacity. They hypothesized that TIC partial differentiation was necessary for GBM cells to migrate away from a thrombosed central vessel and necrosis, giving the cells a survival advantage.137 When this ability to differentiate was inhibited, the TICs demonstrated more benign histology. These findings provide evidence that more committed cells can initiate and drive tumor growth and support the notion that tumor cell proliferation and a CSC phenotype may not be determined solely by the intrinsic features of a cell but may be defined by temporal and spatial characteristics.
Measuring “Cancer Stemness”
The seemingly changing intrinsic features of a cell are further compounded by the influence of the assays we use to test the potential of a cell. Many biological processes can be described in probabilistic terms in which the likelihood of an event occurring is described as a probability (ie, > 0 or < 1.0). Under a particular set of circumstances, the probability of an event occurring may be increased, whereas in a different set of conditions, the probability of the same event may be reduced. Hence, isolation of a defined population of cells (ie, CDxx+/CDxy−) and placed under particular growth conditions may allow their tumor-initiating ability to be expressed, whereas placing the same cell in different conditions will inhibit or reduce this potential. The other side of this may also apply (ie, CDxx+/CDxy− cells in condition A exhibit a CSC phenotype, and CDxx+/CDxz+ cells do not demonstrate CSC features in condition A but do in condition B).
Viewing experiments from this perspective may explain contrasting results that have been published.38,138–141 It can be difficult to reconcile opposing findings when a CSC is defined as being an entity with a specific phenotype (ie, CD133+) but is easier to understand if we view the data from a probabilistic point of view. This approach fits well with the stochastic or clonal evolution model in which a large percentage of the tumor population have the ability to contribute to tumor growth and the potential of a particular population of cells to do so will be defined more by when they are captured and in what conditions they are tested. Nevertheless, the hierarchical and clonal stochastic models may not be mutually exclusive and may coexist to explain cellular behavior in a single tumor.142
PROBLEMS WITH XENOTRANSPLANTATION EXPERIMENTS
Xenotransplantation has been used to demonstrate stemness of CSCs. In most cases, only a small population of tumor cells have been shown to result in tumor growth when implanted into immunodeficient animals.143 For instance, 1 in 104 to 107 AML cells would result in tumor in NOD/SCID mice and had all of the features of the original tumor from which the cells were obtained.24,48 Similar findings were reported in solid tumors (breast, brain, and colon).26,30,38 With this assay, cells with certain phenotypes would be more likely to grow into a tumor (CD34+CD38− in leukemia, CD44+CD24− in breast, CD133+ in brain), resulting in the concept that certain cell surface markers indicated the ability for self-renewal and tumorigenicity.26,38,143
Yet, the growth of human tumor cells in the mouse environment may be more complex. The growth of tumor cells requires a group of support cells such as fibroblasts, endothelial cells, macrophages, and mesenchymal cells. The cytokines and receptors that coordinate these interactions are mostly incompatible between human and mouse.144 These support cells are even more important for solid tumor formation145 than hematologic malignancies. Furthermore, immunosuppressing animals results in more efficient tumor growth, demonstrating the environmental effects on the growth of a human tumor in a mouse model. Using NOD/SCID interleukin-2 receptor gamma chain null mice, Quintana et al146 found that 27% of melanoma cells could form a tumor with a single cell transplant. To remove the effect of foreign milieu, Kelly et al132 experimented with primary pre-B/B lymphoma cells from transgenic mice. They injected 10 to 105 cells into nonirradiated congenic animals. All recipients formed tumors, although those transplanted with fewer cells had slower tumor growth. They also showed that single cell transfer resulted in fatal lymphoma in 3 of 8 recipients within 33 to 76 days. These data suggest that the low frequency of TICs may be an artifact of the model instead of the reality of tumorigenesis in humans.
These results were supported by findings published by Kern and Shibata,147 who mathematically analyzed the CSC hypothesis. According to the hypothesis, after tumor response to treatment, only 0.5% of the original cells (CSCs) remain, representing 100% of the residual cells. Yet, in many cases such as infiltrative glioma, the tumor only partially responds to treatment, with a 50% decrease in diameter and 87.5% reduction in volume. Treatment failure cannot be attributed only to CSCs.41,147 Consider a tumor that consists of 200 cells, of which 1 cell is a CSC (0.5%). After treatment, the total tumor is reduced to 50 cells with the 1 CSC (2%). Although the CSCs are enriched after treatment (400%), they still represent a tiny fraction of the overall population and can be ignored therapeutically because within 2 effective cell doublings, the nonstem cells can recapitulate the original tumor size. If one were able to selectively target the CSC population, this would have little effect on the outcome, given that 98% of the tumor cells are remaining. Therefore, targeting CSCs would be relevant only in tumors in which almost all of the nontumor stem cells have been eliminated.
Phenotypic association with tumor formation in a xenograft model may represent the ability to obtain stromal support in a foreign environment, instead of the ability for self-renewal. The increased tumorigenicity of CD133+ cancer cells (brain and colon tumors)30,38 has led to the conclusion that CD133 identifies CSCs. Nevertheless, some of these CD133+ cells may include support cells that improve tumor growth, thereby explaining the 20-fold increase in tumorigenicity of CD133+ cells.145 Interestingly, CD133+ endothelial precursors can improve grafting of human cancer cells.148 In addition, samples purified in cells with certain phenotypes such as CD44+CD24low/−ESA+ breast cancer cells demonstrate more efficient tumor formation compared with the nonpurified tumor cells, pointing to an inhibitor of grafting in the nonpurified cell populations. This concept is supported by the finding that the addition of CD133− cells decreases the tumorigenicity of CD133+ cells by 95%.29 Similarly, as few as 100 fractionated CD133+ cells result in xenograft tumors of GBM and medulloblastoma compared with 105 or 106 of nonfractionated tumor cells.38 The presence of an inhibitor of tumor formation in nonfractionated tumor cells challenges the ability of xenograft models to identify CSCs.147
These examples demonstrate the practical problem in stem cell biology: the difficulty in identifying CSCs within tumors because of poor CSC markers.149
THE DIFFICULTY WITH CSC DEFINITION AND MARKERS
The “stem cell” properties of CSCs are experimentally demonstrated to be similar to somatic stem cells and are subject to the same uncertainty. To measure these characteristics, the cell must be manipulated. This principle leads to uncertainty, allowing only probability statements to be made about cell populations instead of absolutes.16 Potten and Loeffler16 proposed that “stemness” is not found in specific properties but a spectrum of capabilities that may be weighted differently. Potential stem cells are distinct from actual stem cells. As actual stem cells mature and differentiate, they have declining stem cell properties and are less likely to be able to self-renew, to generate large number of progeny, and to be multipotent (Figure 6). Still, they may still have the potential to regain these properties and under the right conditions become actual stem cells.16,150
In addition to the difficulty with defining CSC, poor CSC markers can lead to difficulty targeting CSCs with therapy. Collins et al28 found that CD44+α2β1+CD133+ prostate cancer cells had 3- to 4-fold increase in tumorigenicity and represented only 0.1% of the tumor cells. The purity of these populations after sorting has been variable. In fact, the first identification of GBM CSCs using CD133+ reported purities of the CD133+ population ranging from 70% to 91% and of the CD133− population of 87.5% to 99.5%.38 Li et al34 showed that CD44+CD24+ESA+ pancreatic cancer cells had a 100-fold increase in tumorigenicity but represented < 1% of the unfractionated tumor cells. In the same article, they reported that 1% to 3% of the nontumorigenic population (CD44−CD24−ESA−) were CD44+CD24+ESA+ as a result of contamination. These discrepancies have significant implications for the relevance of CSC-directed therapy.
Using the reported rates of contamination after sorting, Kern and Shibata147 calculated that the majority of the CSCs would rest in the negative phenotype cell population or the nontumorigenic cell fraction (Figure 7). Calculating a ratio of tumorigenicity between the CSC marker–positive and –negative cells allows an estimation of how many TICs are in each group. Imagine that we start with 1000 unfractionated tumor cells and separate them into 2 cells positive for CSC markers and 998 negative for CSC markers. The 2 cells that are positive have a 100-fold increased tumorigenicity compared with the 998 CSC marker–negative cells. To determine how many of the truly tumorigenic cells are represented by these CSC marker–positive cells, we can multiply tumorigenicity by the fraction of cells: 2 × 100 = 200 and 998 × 1 = 998. The CSC-positive population represents 200 / 998 = 0.17 or only 17% of the tumorigenic cells in the total population. To increase the number of tumorigenic cells in the CSC marker-positive group, either the total number of CSC marker-positive cells needs to be increased or the tumorigenicity of these cells needs to be higher. Their calculations demonstrate the weakness of selection strategies based on the current CSC markers and the inability to develop effective therapy if < 20% of tumorigenic cells are being targeted.
Discrepancies also exist in the study by Al-Hajj et al.26 They noted that approximately 23% of human breast cancer cells were CD44+CD24−/low with a 10- to 50-fold increased tumorigenicity compared with the unfractionated tumor cells.147 These values mean that the CD44+CD24−/low cells had a 230% (23% × 10-fold) to 1150% (23% × 50-fold) increase in tumorigenicity compared with the unfractionated samples. These findings implicate a strong inhibitor of grafting when unfractionated cells are transplanted such as autocrine or paracrine factors.151,152
PROBLEMS WITH CSC MARKERS IN GBM
Similarly, CD133 (prominin-1) has been used as a marker for brain CSCs. Initially, noncancerous neural cells that were sorted for CD133 were found to grow in neurospheres and demonstrated the qualities of somatic stem cells.99 These findings were extended to brain tumors, and researchers found that CD133+ GBM cells formed tumors in immunodeficient mice compared with no tumor growth from CD133− GBM cell implants.38 On the other hand, subsequent studies found that CD133− GBM cells not only can form tumors in immunodeficient mice but also can give rise to CD133+ cells.153 Other investigators found that bioenergetic stress–induced mitochondrial changes in glioma cells would change the expression of CD133.154 Environmental effects on CD133 expression in glioma cells bring into question the reliability of CD133 or any other marker as a CSC marker. Without being able to distinguish between tumorigenic and nontumorigenic cells with a reliable marker, difficulty arises in applying the CSC hypothesis to a particular tumor.
The CSC model is based on a rare subpopulation of cells that can be enriched with special markers. Even the property of multilineage differentiation of these cells is based on marker expression of differentiated cells (ie, neurons, astrocytes and oligodendrocytes). However, differentiated marker expression may be a result of dedifferentiation of differentiated cancer cells when fundamental genetic regulation is aberrant.155 These concepts may not be clinically relevant because this rare population is not solely responsible for tumor growth. Some tumor cells that do not exhibit experimental evidence of stem cell characteristics and cells that are not marker positive can also drive tumor growth. Although this rare “CSC” population may be an important consideration in a therapeutic paradigm, it cannot be to the exclusion of other tumor cells. Treatment focused on these particular “CSCs” may not change disease outcome. Furthermore, it has not yet been demonstrated in preclinical solid tumor cancers that targeting the CSC improves outcomes.
Current understanding of tumor biology continues to drive research for novel cancer therapy. With increased appreciation for tumor heterogeneity and growth dynamics, focus has shifted from cytotoxic therapy that indiscriminately kills tumor cells to targeted therapy based on patient factors and tumor genetic and cellular signaling pathway expression. In this environment, the CSC hypothesis has developed, resulting in therapeutic efforts targeted to a subpopulation of tumor cells. The CSCs have been described as a rare population of quiescent cells that demonstrate self-renewal, generation of a large number of multilineage progeny, the ability to recapitulate the original tumor, and resistance to conventional treatment. These definitions have been based on models that may not represent true tumor biology. Cell culture assays, SCID mice tumor models, and CSC markers each have their own limitations in identifying and describing the characteristics of cells that initiate and maintain tumor growth. These limitations have led to the redefinition of a CSC as a tumor cell that demonstrates self-renewal and can regrow the tumor from which it originated.156,157
This new definition may be a better description of TICs without inference of origin, quiescence, resistance to treatment, or other traditional characteristics of stem cells. Using the term TIC instead of CSC would also prevent confusion by removing the requirement of the cells to meet criteria that define somatic stem cells. The value of targeting a rare CSC in solid tumors has not been demonstrated. The biology of many tumors, especially solid tissue tumors, can be explained with a mixed model (combination of hierarchical and clonal stochastic) in which stemlike TICs can generate colonies and generate many cells with variable potential to drive tumor growth, given the right conditions.134,142
The CSC hypothesis has invigorated the research community to find novel approaches to cancer therapy. However, for many cancers, targeting a rare population of tumorigenic cells without consideration of the large bulk of proliferating cells may not change patient outcomes. Efforts to discover new and effective treatments may require more than just targeting CSCs.
This review by Rahman and colleagues raises important issues regarding the cancer stem cell (CSC) hypothesis, both in how useful it is to import the concepts of normal stem cell biology into cancer and whether these ideas can guide translational research. It comes from the group of one of the discoverers of stem cells in the adult mammalian central nervous system, Dr Reynolds, and offers a comprehensive and thoughtful consideration of the state of the field. The review gives balanced and thorough consideration to the published data in support of the idea that brain tumor CSCs drive the process of tumor initiation and recurrence, and it sets a context for them by explaining important limitations in the techniques used. It explores the risks in using a strong form of the CSC hypothesis as a basis for research seeking treatments, arguing that any therapy that simply eliminates the CSCs is unlikely to yield an effective therapy. Interestingly, experimental evidence in support of this concern was recently published in a model system that used forced epithelial-mesenchymal transition in breast cancer cells to model CSCs, which were then used to identify drugs that preferentially treated the CSC component.1 In the breast cancer models used, this new agent effectively eliminated the CSCs in xenografts, but the tumors grew just as fast as the controls that were treated with an agent that spared the CSCs. In other words, the modeled CSC and non-CSC components could be selectively eliminated, but in either case, the remaining tumor grew at similar rates. Gupta and colleagues concluded that both components need to be targeted, and Rahman and colleagues likewise conclude that one cannot leave any tumor component aside. The review from Dr Reynolds' group concludes that it may be better to discuss tumor-initiating cells; that the ability to initiate tumors, resist therapy, and grow may be available to the majority of cancer cells; and that their ability to access them may be stochastic and subject to environmental modulation.
This article provides a balanced view of this very important, intriguing, controversial, and unresolved hypothesis of the “cancer stem cell” as it applies to solid tumors such as gliomas. The camps of believers and nonbelievers of the hypothesis sometimes take extreme views of the interpretation of preclinical and clinical observations with an attempt to sometimes pigeonhole them into their preference. However, the truth is likely in between, varying from one tumor to another and probably during the evolution of the tumor as it responds to its microenvironment and therapies directed toward it. Cancers such as malignant gliomas have been resistant to cures despite our many attempts because of their flexibility and adaptability. Similarly, we should also keep an open mind about the “cancer stem cell” hypothesis as per this review article.
Toronto, Ontario, Canada
acute myeloid leukemia
cancer stem cell
X-chromosome–linked enzyme glucose-6-phosphate dehydrogenase
neural stem cell
severe combined immunodeficient