|Home | About | Journals | Submit | Contact Us | Français|
Cancer is the second leading cause of death in the United States, and therefore remains a central focus of modern medical research. Accumulating evidence supports a ‘cancer stem cell’ (CSC) model - where cancer growth and/or recurrence is driven by a small subset of tumor cells that exhibit properties similar to stem cells. This model may provide a conceptual framework for developing more effective cancer therapies that target cells propelling cancer growth.
We review evidence supporting the CSC model and associated implications for understanding cancer biology and developing novel therapeutic strategies. Current controversies and unanswered questions of the CSC model are also discussed.
This review aims to describe how the CSC model is key to developing novel treatments and discusses associated shortcomings and unanswered questions.
A fresh look at cancer biology and treatment is needed for many incurable cancers to improve clinical prognosis for patients. The CSC model posits a hierarchy in cancer where only a subset of cells drive malignancy, and if features of this model are correct, has implications for development of novel and hopefully more successful approaches to cancer therapy.
Despite substantial research directed towards elucidating the physiological mechanisms of cancer growth, and developing and refining surgery, chemotherapy and radiation techniques, only marginal improvements have been made in the overall cancer survival rate from 50% in 1974–1976 to 66% . Cancer remains the second leading cause of death in the United States and will likely soon become number one , and has not exhibited the decline in deaths observed with heart disease stemming from modern treatments. The marginal improvements in cancer survival have also been mainly attributed to earlier diagnosis and prevention, and few examples of breakthrough cancer treatments exist from the last three decades of intense research . Survival of patients suffering from several varieties of cancer, including brain, liver, lung and pancreatic cancers, have failed to improve despite modern therapies. A modest increase in mean survival from 12.1 to 14.6 months for patients with gliobastoma multiforme, the most frequently diagnosed and incurable brain neoplasm, was considered a major breakthrough when Stupp and colleagues discovered a new chemo-radiation therapy strategy in 2005 . Likewise, chemotherapeutics are practically ineffective in hepatic carcinoma, with a median response below 15% . Even with initially effective treatments, overall survival following a cancer diagnosis remains low because many cancers recur after developing therapeutic resistance and metastasizing throughout the body. Clearly, increased understanding of cancer mechanisms, especially the biology of therapeutic resistance and clinically relevant patient-specific molecular variation in cancers will be crucial for improving outcomes with better therapies.
Given the variability of cancers and survival rates, it is clear that certain aspects of cancer are not adequately understood. The stochastic model, a widely accepted theory in the field of oncological research, states that tumors are a homogenous mass, arising from any cell expressing the requisite mutations . New research suggests a cellular hierarchy is involved in tumor development, with only a small subset of cells capable of recapitulating the cancer via tumor initiation and propagation (Figure 1) . A subset of tumor cells can initiate tumors with high efficiency (as few as 100 cells) compared to non-initiating cells from the same tumor specimen that rarely form tumors [7, 8]. These tumor-initiating cells also display stem cell properties such as self renewal and multipotent differentiation, possibly recapitulating and resulting in the heterogenity of cell phenotypes comprising most cancers . Because of their similarities to normal stem cells, these tumor initiating cells are often referred to as cancer stem cells (CSCs). The origin of CSCs is still being debated; they may result from oncogenic mutation of otherwise normal stem cells, or derive from adult neoplastic cells acquiring stem cell properties via mutations . The total CSC population may also correlate with malignant potential of cancer (Figure 1), although this hypothesis requires experimental verification and is only supported by anecdotal evidence from histologically “undifferentiated” tumors [11, 12]. This model challenges the assumptions of the stochastic theory and suggests that re-examination of cancer mechanisms and cell-targeted therapies would be informative for future therapies.
Although the concept of stem cell involvement in cancer development has been proposed in the past, evidence supporting the hierarchal nature of cancer development originated from experiments carried out with hematopoietic acute myeloid leukemia (AML) cells [13–15]. Hematopoietic cells were extracted from patients afflicted with AML and transplanted into immunodeficient mice with the intention of modeling leukemic disease in vivo. Results demonstrated that a core group of cells expressing CD34, a surface marker primarily associated with hematopoietic stem cells, were responsible for generation of leukemia in xenografts [13, 14, 16]. Discovering that only CD34+ cancer cells could initiate leukemia suggested that cells expressing stem cell-associated markers and properties play a role in cancer development. The discovery of cancer cells expressing stem cell properties also raised the possibility that cancers other than AML may be linked to CSCs [14, 15].
After their discovery in hematopoietic cancers, CSCs were hypothesized to exist within solid tumors as well, particularly in tissues subject to rapid self-renewal and proliferation . This theory gained support when Al-Hajj and colleagues (2003) isolated putative CSCs from breast cancer, identifying a small population of CD44+CD24− cells, typical cell surface markers of normal mammary stem cells. When these CSCs were re-injected into immunodeficient mice, breast cancer arose in eight out of nine cases . Significant populations of CD44+ CSC were not required to cause tumor growth, with as few as 100 CD44+ cells recapitulating breast tumors. In contrast, significantly higher numbers of other cell types derived from the tumor mass did not form tumors after injection [7, 17].
The hypothesis that cells sharing many properties of embryonic stem cells were responsible for the development of brain tumors, particularly medulloblastomas, originated from pioneering research conducted by James Homer Wright in the early 1900s . Evidence for the existence of brain tumor stem cells had been growing since the discovery of clonogenic, multipotent cells from brain tumors in vitro when cultured under neural stem cell conditions [19, 20]. These findings were verified in vivo in 2004 when tumor-initiating cells were isolated from brain tumors [8, 21]. Further study demonstrated the capability of these cells to self-renew and differentiate into multiple neural lineages . The highly tumorigenic nature of these cells was demonstrated through xenografts in NOD/SCID mice, establishing the ability of isolated brain tumor stem cells to generate tumors [8, 23]. Genetic analyses also demonstrated a strong correlation between the original patient tumors and derived brain tumor stem cells, implicating the CSCs as a better model for brain cancer and therapeutic testing [24, 25].
CSCs have also been implicated in other solid tumor cancers, such as colon cancer. Xenografts of human colon cancer in NOD-SCID mice have demonstrated that as few as 100 CD133+ cells were required to generate tumor growth, giving additional credence to the CSC theory . These data appear to support the conclusion that a small population of CD133+ cells which exhibit stem-like properties serve as the progenitors of colorectal cancer .
As further research is conducted and the CSC model becomes better accepted, the list of solid tumor cancers containing putative CSCs has only continued to grow (Table 1). As the list of tumors linked to CSCs has increased, so too has the base of knowledge surrounding CSCs.
Current cancer therapies are often limited by their lack of specificity and high incidence of adverse effects. Chemotherapies usually rely on a “carpet-bombing” strategy that results in toxicity to both cancerous and non-cancerous cells. Likewise, radiation therapy grossly targets tumor masses and cannot target specific tumor cells. Traditional therapies are also hampered by the inability to reliably expunge an entire tumor mass, with individual tumorigenic cells potentially evading destruction. The CSC theory underlies the new strategy of novel therapeutics aiming to overcome many of the limitations of current treatments, by suggesting the more nuanced option of designing therapies for the tumor-initiating cellular compartment, the CSC (Figure 2).
The efficacy of current chemotherapies and radiotherapies may be effective at destroying the majority of a tumor, but many cancers recur because these therapies are likely ineffective against the CSCs. Cancer stem cells, along with non-cancerous stem cells, express protective drug transport mechanisms that remove cytotoxic chemicals from the cell [28, 29]. Due to the presence of efflux mechanisms, including ATP binding cassette (ABC) and multi-drug resistance (MDR) transporters, CSCs are resistant to traditional chemotherapies compared to tumor mass or non-cancerous cells [30, 31]. Additionally, CSCs demonstrate enhanced radiation resistance through heightened activation of DNA repair mechanisms  or increased defenses against reactive oxygen species . Current therapies also fail to take into account the heterogeneous nature of tumors and the differences in tumors between patients, instead applying broad treatment principles rather than personalized treatment regimens.
One of the difficulties of targeting CSC is their chemo- and radiation resistance [33–35]. Efforts to target the chemoresistant mechanisms of CSCs have met with some success in vitro. Inhibitors have been identified which act on the ABC transporters ABCC1 and ABCG2, both critical to the development and normal function of stem cells. Some of the ABCC1-specific inhibitors are being tested in clinical trials . However, research has also suggested that ABC transporters play a critical role in several physiological processes, including the maintenance of the blood-brain barrier (BBB) . Although such transporter inhibitors might render the CSC population more sensitive to chemotherapeutic agents, it remains to be seen whether they also cause harm by inhibiting the BBB or indirectly by rendering normal stem cells more vulnerable to chemotherapeutic agents.
Pathways which regulate normal stem cell growth and division are also potential cancer therapeutic targets . Up-regulation of the TGFβ pathway has been correlated to tumor growth, as well as to regression of differentiated cells into progenitor and stem cell lineages . Bone morphogenic protein is a well-studied TGFβ inhibitor; knowledge of the interaction of these two proteins led to the discovery that BMP can limit tumor growth by ‘differentiating’ tumor-initiating brain cells . The Notch developmental pathway is also associated with cancer survival and proliferation, especially in various brain cancers [40, 41]. Inhibition of Notch signaling in medulloblastomas was shown to halt tumor cell proliferation and also to decrease the number of cells positively expressing CD133 both in vitro and in vivo . Dysregulated gene expression in the Sonic Hedgehog-homolog (SHH) pathway has also been implicated in the development of a wide variety of cancers [43, 44]. Specific small molecular SHH signaling inhibitors have shown success at halting tumor progression in glioblastomas and medulloblastomas [45–47], with some early clinical success [48, 49]. Likewise, the WNT pathway is often up-regulated in cancers . The biological WNT blocker, DKK1, was shown to dramatically increase tumor cell apoptosis levels and synergized with other chemotherapeutic agents . Preliminary work suggests that targeting these pathways for cancer treatment could be accomplished in a feasible and effective manner.
Although many of the initial results seemed promising, there are several limitations to using developmental pathways to target CSC populations. Because these pathways auto-regulate and interact with many other pathways, it is difficult to understand the full effects of inhibiting any one pathway [51–53]. Additionally, targeting these developmental pathways will affect both CSC and normal stem cell populations [44, 54–56]. Nonetheless, therapies designed to block or to interfere with these pathways may help to minimize CSC proliferation.
Another potential therapy for CSC is to target unique cell surface markers that are expressed in stem cells. Most CSCs are detected and isolated with specific cell surface markers that are typically found in stem cells (Table 1). Early therapy based on this concept targeted CD44, a cell surface marker expressed in AML and critical for residence in a stem cell niche . An antibody specific to CD44 was used to target the cell marker and induced two effects: tumor cells were unable to effectively engraft, leading to inhibition of tumor growth, along with loss of primitive cell types within tumor grafts . This study appears promising; however, a similar study indicated that treatment with antibodies had little effect on tumor formation in xenograft . However, both studies were able to show that blockade of CD44 limited AML CSC proliferation, suggesting that CD44 is a required element of the AML CSC niche . Just as CD44 may prove a useful molecular target in treating AML, several other cell surface markers linked to the stem cell niche have been identified, such as the prostate stem cell antigen (PSCA), a marker expressed in both normal prostate stem cells as well as prostate cancers .
Another method for targeting the CSC niche involves blocking the factors that CSCs use to produce and stabilize supporting tissues. Vascular endothelial growth factor (VEGF) has been observed to be up-regulated in tumors, causing angiogenesis that provides the tumor with necessary blood supply [61, 62]. Likewise, the role of VEGF in normal neural stem cell (NSC) function has also been well explored. VEGF and several other factors, involved in cross talk with the endothelium maintain the neural stem cell niche . This discovery has also been applied to CSC, where it was shown that VEGF and several other cell factors are responsible for maintaining the CSC niche . By interrupting the ability of CSCs to interact and regulate supporting cells to maintain the CSC niche, tumor engraftment and growth will be interrupted.
As with targeting the signaling pathways used by CSCs, one of the major risks to all of these therapeutic approaches is the similarities between the CSCs and other adult stem cells in treated patients . Because the targeted markers and factors are present in normal adult stem cells and CSCs, these methods may be inadequately specific and have associated toxicity to normal tissue.
Since the conception of the CSC hypothesis, many controversies have arisen that are currently the focus of intense research. Identification of specific surface markers is critical for CSC isolation and subsequent characterization and for future CSC-targeted therapies. Current research is demonstrating that a universal single marker may not exist for any specific cancer or among all patients. One of the earliest markers suggested for brain tumor CSCs was CD133, which was successfully used to enrich for tumor initiating CSCs . Additional work from multiple groups has clearly shown that CD133+ populations do not exist in every brain tumor [66, 67] and that CD133− cells are capable of initiating tumors . For this reason, other stem cell markers such as stage-specific embryonic antigen 1 (SSEA-1)/CD15 [66, 69, 70] have been proposed and successfully used to enrich for CSCs, but in each case these newly proposed markers were only expressed in a subset of brain tumors. Similar results have been obtained in breast CSCs, originally defined as CD44+CD24− . Upon further characterization, additional surface markers such as aldehyde dehydrogenase 1 (ALDH1), CD201/PROCR, ESA, CD133, and HER2/ERBB2, have been shown to enrich for breast CSCs in certain tumor subsets [71–74]. Accordingly, varying CSC enrichment markers have been prospectively identified for other forms of cancer as well (Table 1)
Adding to the complexity, it has recently been proposed that multiple CSC populations can exist within any given cancer. For example, distinct CSC subgroups may exist within a tumor that separately drive primary tumor growth and metastasis . Hermann et al. (2007) identified a metastatic CSC subgroup within the CD133+ cells of pancreatic cancer expressing chemokine (C-X-C motif) receptor 4 (CXCR4). Depletion of CXCR4+ cells within this population abrogated the metastatic phenotype without affecting tumorigenic potential . The existence of specific recurrence CSCs has also been recently postulated . In general, CSCs are likely a dynamic population subject to change and adaptation through clonal evolution  or epigenetic plasticity . The diversity of CSCs in any given tumor may be especially evident in highly malignant cancers or after chemotherapeutic and radiation interventions [78, 79]. It is likely that successful cancer treatments must eradicate all CSC sub-populations along with associated tumor cells.
The origin of CSCs and contribution to cancer growth remains a topic of hot debate. Some evidence suggests that cancer can arise from aberrant stem cells that acquired oncogenic characteristics [18, 80]. Human adult stem cells can rarely transform into cancerous cells following long-term culturing in an in vitro environment for 4–5 months . Expression of oncogenes driven by stem cell-specific promoters (Sca1 in HSCs, CD133 in intestinal goblet cells, nestin in multiple brain tumors) in mice consistently results in tumors that faithfully recapitulate the human condition [82, 83]. However, in certain instances, more differentiated progenitor cells may initiate and drive tumor progression as CSCs. Jaiswal et al. (2003) restricted BCR/ABL oncogene expression specifically to myeloid progenitors leading to AML in mice . Similarly, Krivtsov et al. (2006) described conversion of a committed granulocyte macrophage progenitor to leukemic stem cell through expression of MLL-AF9 fusion protein ; similar leukemia driving progenitors have been inferred in human leukemias through surface marker analysis . It is also likely that not every cancer is driven by stem cell-like cells. Joseph et al. (2008) analyzed the development of malignant peripheral nerve sheath tumors (MPNSTs) in NF1-deficient mice and found that although the neural crest stem cells hyperproliferate extensively in the fetal peripheral nervous system, these stem cells do not persist into adulthood and the MPNSTs are driven by cancerous differentiated glia . An interesting view on these results is that some forms of cancer may be stem cell driven following a reprogramming-like process. Oncogene expression from primitive HSCs (Sca1+) led to CML in mice and recapitulated the molecular and cellular diversity of the disease . Similar reprogramming has been demonstrated in the conversion of fully mature fibroblasts to embryonic stem-like cells, also termed induced pluripotent stem cells [88, 89], and in direct conversion of mature pancreatic exocrine cells to insulin-producing cells . In many of these instances, the initial reprogramming events are turned off again as the cells re-differentiate to more mature phenotypes; a similar process could occur in cancer where initial oncogenic / reprogramming hits occur in stem cells but further mutations or epigenetic modifications are required later during differentiation to initiate tumorigenesis. In these cases, the cell-of-origin of the cancer (the true CSC) may not be tumorigenic on its own and would differ from the identified CSC of the disease. Eradication of the originating CSC would still halt the disease, but attacking the tumorigenic progeny would do little . Some evidence for reprogramming exists in human childhood leukemia in which only 1 of a pair of monochorionic twins developed leukemia despite the shared altered self-renewal and survival properties of the twins’ preleukemic cells . The importance of the stem cell microenvironment, or niche, and supporting stromal cells in regulating oncogenic processes must also be acknowledged. CSC induction through disruption of the niche can lead to de-regulated proliferation and tumor initiation by adult stem cells . Obviously, much research needs to be done into the origin and mechanisms of propagation of CSCs, focusing on clinical relevance of CSC biology to improved therapeutics.
Finally, CSCs initially isolated were described as rare subsets of cells that drive cancer propagation and recurrence, with only approximately 1 CSC per every couple hundred thousand cells . Recent work using human melanoma and transplantation between syngeneic mice have challenged this assumption that CSC sub-populations are always rare. Kelly et al. (2007) in leukemia and Quintana et al. (2009) in melanoma have each demonstrated abundance of CSCs in these cancers, with CSCs estimated at 1 in 20 cells or lower [92, 93]. These authors point out currently used xenograft models into NOD/SCID mice as inefficient, and improved the rates of CSCs seen through improvements on this traditional model. Despite the increased frequency of CSCs found, the overall CSC hypothesis is not refuted, it is just important to remember that, especially in highly malignant cancers, a larger sub-population of CSCs may need to be extinguished than originally thought. Clinically, increased staining for stem cell markers in various forms of cancer has been correlated with aggressiveness and malignancy [11, 94], so in advanced cases or after non-CSC specific therapies self-renewing CSCs may have overtaken any differentiated tumor cells.
Although relatively new, evidence for the CSC hypothesis (Figure 1) is rapidly accumulating, and putative CSCs have been isolated from many human cancers (Table 1). The highly efficient tumor-initiating ability of CSCs in xenograft animal models [7, 8] strongly supports a critical role for CSCs in human cancer initiation, progression, and recurrence. In vitro and in vivo evidence also have demonstrated increased radiation and chemotherapeutic resistance of CSCs [32, 33, 95, 96]. Therefore, it is imperative that new therapeutic strategies focus on ablating the CSC sub-population, whether this is achieved through targeting key molecular pathways, overcoming CSC resistance mechanisms, or via other methods. Ideally, many of these targeted treatments would also result in fewer side effects, although rigorous testing of new treatments will still be required to ensure safety and efficacy.
Although the study of CSCs offers an array of new therapeutic strategies, current efforts have been limited by the lack of an identified, unique marker for CSCs. Currently, efforts to localize CSCs rely on normal stem cell surface markers such as CD133 [8, 26] or CD44 , among others. As discussed above, even this list of surface markers continues to grow, and since these markers are expressed in normal, adult human stem cells, it is likely that targeted therapies directed to these markers would ablate the normal as well as cancerous stem cell populations. These directed therapies may be particularly dangerous to the pediatric patients. Identification and validation of specific CSC markers through genetic, proteomic, or metabolomic screens may enable substantial progress in the development of targeted therapeutics, and possibly antibody-mediated therapy or CSC-specific vaccine-based therapies. Because of the multiple isolated populations of CSCs in various cancers and the potential for neoplastic transformation of progenitors as well as stem cells, universal markers of CSCs may not exist. Therefore, study of the shared signaling pathways of CSCs, such as Wnt, TGFβ, Notch, and Hedgehog as described above, may be a more useful focus and more efficient for drug targeting opportunities . Many of these pathways have already been demonstrated critical to CSC self-renewal and propagation [45, 47, 74, 79, 97, 98]; however, rigorous in vivo as well as in vitro tests may be performed as certain proteins, such as the polycomb BMI-1, may have profoundly different effects . It is important to note that proof-of-principle evidence in animal models and anecdotal clinical evidence already exists for the efficacy of killing CSCs to eradicate cancer. Unique elimination of the CSC population through conditional knockdown or drug treatment in mouse models of CML eradicated the CML, whereas drug treatments not eliminating the CSC population had little effect [82, 100]. Short-term treatment of glioblastoma multiforme (GBM, brain cancer) CSCs with Notch or Hedgehog inhibitors or treatment with BMPs depletes the CSC population and prevents engraftment in immunodeficient mouse models [39, 45, 101]. In certain forms of breast cancer, the HER2-expressing cells have CSC properties, and blockade with the pharmaceutical trastuzumab depletes these CSCs preventing primary and serial engraftment into mice. These results may help to explain the better than expected results seen in some patients [74, 102]. However, much of the complexity of targeting and eradicating CSCs must still be worked out before more broad successful clinical therapies.
Further characterization of adaptive and/or resistance mechanisms of CSCs to current treatment regimens will also be required for developing more effective future therapies. Although molecular-targeted therapies show early promise in treating cancer and its stem cells [48, 103, 104], CSCs are likely a dynamic and changing population through clonal evolution  or adaptive/resistance responses. For many complex cancer types, such as GBM, it is likely that molecular monotherapy will be inefficient, and knowledge of the resistance mechanisms of CSCs will inform rational selection of CSC-combinatorial medicine using molecular therapeutics combined with other modalities. Blockade of specific CSC resistance mechanisms to traditional therapies is also a useful approach, with early successes of IL-4 blockade to sensitize colon CSCs to chemotherapy  and CHK kinase inhibition to radiosensitize GBM CSCs . Different cancer CSCs may exhibit varied response mechanisms, such as breast vs. GBM CSCs in radioresistance [32, 33], so tissue-specific resistance mechanisms may need to determined in many cases.
Lastly, the heterogeneity of cancer and rapid improvement in diagnostic methods is making possible an age of customized medicine for individual patients. The power of genetic and other global arrays has demonstrated repeatedly the molecular variation among cancers, and genetic signatures can often predict outcome and treatment response better than histological analysis alone [106, 107]. Molecular variation is also quickly being demonstrated in the CSCs within cancer as well [67, 72, 108]. Ideally, CSCs will be rapidly isolated and screened for molecular defects sensitizing them to specific drugs or subjected to a direct drug screen (Figure 3). Some research has focused on developing such methods [109, 110]; however, much work must be done to optimize these methods and establish feasible protocols for clinical application as the time and effort required currently make them impossible to broadly apply. Isolation techniques of CSCs must also be further perfected before such protocols can be enacted in the clinic. One example pertains to GBM CSCs, which can be isolated from operating room specimens via flow sorting, neurosphere formation, or tissue culture on laminin [8, 109, 111, 112]; if these different techniques isolate the same or unique CSC populations is unknown and as discussed above all populations of CSCs must be eliminated for successful clinical therapies. Much of the improvement in cancer survival over the past decades is attributed to improved preventive and diagnostic measures ; along these lines, improved biomarker identification of CSC transformation either by non-invasive imaging or blood draw could help early detection and make it easier to eliminate CSCs before multiple and highly aggressive populations can arise. With these diagnostic tools in hand, doctors may eventually be able to create a tailored prescription for oncogenic pathway components from a wide array of options, achieving individualized, targeted therapies with enhanced efficacy to patient’s specific tumor with a lower toxic side effect profile.
The CSC field is a relatively new one with a great deal of research and exploration still necessary to yield greater practical benefits towards improved cancer treatment. The cancer stem cell hypothesis offers a new perspective for understanding tumorigenesis, and will likely lead to rational development of refined and novel therapies that yield tangible clinical benefits to patients.
We apologize to colleagues whose work was not cited due to space constraints. PAC was partially supported by a NIH T32 postdoctoral fellowship from the University of Wisconsin Stem Cell Program. MZ was supported by the NIH-supported Neuroscience Training Program at the University of Wisconsin – Madison. RGK is a University of Wisconsin Letters and Sciences Honors Scholar. JSK was supported by the HEADRUSH Brain Tumor Research Professorship, Loff Memorial Fund for GBM Research, AANS-NREF Young Clinician Investigator Award, and funding from the Dept. of Neurological Surgery, Graduate School, and Medical School at the University of Wisconsin.