Introduction

Well over 40 years ago, President Richard Nixon signed the National Cancer Act and officially started the war on cancer [1]. Despite this act and resulting significant improvements in therapy, the war has not ended and cancer is still one of the leading causes of death worldwide. To illustrate, in 2012, 14.1 million adults in the world were diagnosed with cancer and, in the same year, an estimated 8.2 million people died from the disease [2]. The main problem that we face in cancer treatment is the presence or development of resistance to therapy for which a multitude of distinct reasons have been defined. For example, acquisition of mutations in key signaling molecules, enhanced anti-apoptotic protein expression, the presence of quiescent and/or resistant tumor cells or high expression of drug efflux pumps are all potential means that impair therapy efficacy [3, 4]. In the last decade, attention has focused on the role of a specific subset of cancer cells: cancer stem cells (CSCs). In analogy with their normal counterparts (i.e. the stem cells), these cells display a high level of therapy resistance and can effectively repopulate the tumor.

CSCs are the tumorigenic core of tumors

CSCs are defined based on their tumor-forming capacity in xenograft studies [5]. These cells normally represent a minority of the tumor cells and can be identified by a long list of markers, although most of these are not restricted solely to the CSC population [6]. CSCs can be selected in vitro using spheroid growth in suspension and defined media compositions. Upon injection in mice, CSCs, but not their more differentiated counterparts, can very efficiently form tumors that resemble the original tumor from which they were derived, including all differentiated cells. Moreover, re-isolation of the CSCs from xenografts allows for serial transplantation to secondary and tertiary mice, which is the gold-standard assay to confirm that tumor cells are indeed CSCs [5].

CSCs were first defined in acute myeloid leukemia (AML) in 1994 [7]. CD34⁺/CD38⁻ expression has long been used to mark progenitor and pluripotent stem cells in the bone marrow. Intriguingly, a similar subpopulation was detected in AML and xenotransplantation of specifically CD34⁺/CD38⁻ leukemia cells resulted in leukemia in mice that was observed to reproduce many features of human AML [7]. A decade later, CSCs were detected in solid tumors. Breast, glioblastoma, prostate and colorectal tumors are only some of the tumor types where CSCs were identified [6, 8-12]. The variety of tumors in which CSCs were identified suggests that it is a common feature in most cancers, although some observations indicate that it may not occur in all tumor types or, alternatively, at all stages of disease [13-16].

As noted above, CSCs are highly tumorigenic and therefore are also referred to as tumor-initiating cells. The term ‘CSC’ does not refer to the fact that CSCs can be derived from normal stem cells but rather points to the idea that CSCs display properties normally attributed to stem cells. First, stem cells have the capacity to self-renew (i.e. to form a new stem cell upon division) and, second, stem cells can differentiate into the more specialized cell types that make up a tissue [17]. Self-renewal and differentiation of stem cells is regulated by morphogenic pathways and, interestingly, these signaling pathways are also highly active in many CSCs, suggesting that equal regulatory principles exist in CSCs. One of the morphogenic pathways that is active in stem cells is the Wnt signaling pathway [18]. This pathway determines self-renewal and cell fate of hematopoietic stem cell (HSCs) [19]. High activity of this pathway is also observed in stem cells of other tissues, such as breast and colon [18, 20, 21]. Next to Wnt signaling, Notch signaling is shown to be essential for stem cell maintenance. Notch signaling is highly active in HSCs compared to more differentiated cells and inhibition of Notch signaling promotes differentiation of HSC [22]. Similarly, Hedgehog (HH) signaling regulates proliferation and self-renewal of stem cells and activation of HH signaling is able to expand HSC and brain stem cells both in vitro and in vivo [23, 24]. By contrast to Wnt, Notch, and HH signaling, bone morphogenetic protein (BMP) signaling inhibits stem cell expansion. Activation of BMP signaling results in the suppression of Wnt signaling and this controls stem cells numbers [25-27]. Combined, these morphogenic pathways regulate stem cell fate and differentiation cues. Intriguingly, this regulatory network appears to extend to CSCs. In a high throughput screening in breast CSCs, salinomycin was identified as a compound that eliminates CSCs [28]. This antibiotic was shown to inhibit Wnt signaling and, as a result, is capable of differentiatinh breast CSCs [28]. High Wnt pathway activity is also shown to be important for cell fate of CSCs from many tumors, such as chronic lymphocytic leukemia, breast, colorectal cancer (CRC), squamous cell carcinoma and lung cancer [28-32]. In these tumors, inhibition of Wnt pathway activity (e.g. with salinomycin) is detrimental for CSCs [28-32]. Beside Wnt signaling, Notch signaling can also regulate CSC self-renewal. Inhibition of Notch signals can be achieved by neutralizing antibodies against DLL4, or treatment with a γ-secretase inhibitor (dibenzazepine). Glioblastoma multiforme (GBM), CRC and breast cancer stem cells require high Notch activity and inhibition of Notch results in loss of CSCs [33-36]. Furthermore, HH signaling is highly active in CSCs, which was shown to be required for self-renewal of CSCs in various cancers, such as breast, lung and CRC [37-39]. Morphogenic pathways and inhibitors are depicted in Fig. 1.

In addition to the analogous usage of morphogenic pathways, stem cells and CSCs also appear to share high activity of DNA repair pathways. For example, HSC can repair UV induced single-strand breaks faster than more differentiated cells [40] and this was reported in CSCs as well [41]. There is increasing evidence indicating that CSCs are not a fixed cell population but rather represent a state of tumor cells that appears to be inducible. The appropriate signals from the micro-environment or the introduction of new mutations can result in de-differentiation of more differentiated tumor cells into CSCs [31, 42-44].

Not only signaling pathways, but also cell surface molecules are similarly expressed in stem cells and CSCs. The pentaspan membrane glycoprotein CD133, also known as prominin-1, is expressed in normal stem cells (e.g. hematopoietic, neural and intestinal stem cells) [45-47], although CD133 was also used to identify CSCs from different tumor types [8, 9, 41, 48]. Moreover, in the last decade, the G-protein coupled receptor Lgr5 has received much attention because high Lgr5 expression was reported to mark stem cells in various organs. The gene for this target of the Wnt signaling pathway is exclusively expressed by stem cells of various organs [49-52] and it was shown that Lgr5 also marks CSCs in various tumors [53, 54]. Most of the currently used markers have no known role in CSC biology. By contrast, aldehyde dehydrogenase isoform 1 (ALDH1) oxidizes aldehydes to carboxylic acids and thus, for example, catalyzes the conversion of retinol (vitamin A) to retinoid acid. Inhibition of ALDH1 reduces retinoic acid levels and thereby promotes HSC and breast CSCs self-renewal [55, 56]. ALDH1 is highly expressed in many stem cells and CSCs. Accordingly, ALDH1 expression and its activity were used to isolate stem cells and CSCs [57-59]. Taken together, it appears that stem cells and CSCs are wired in the same way and share expression of several cell surface markers. Unfortunately, this similarity extends to a more detrimental property, namely therapy resistance.

The extreme survivors: CSCs and their therapy resistance

The concept that CSCs selectively resist therapy stems from a multitude of observations in cell culture, animal models and cancer patients. In cell culture, direct analysis of apoptosis revealed that differentiated colon cancer cells are induced to die upon chemotherapy treatment, whereas colon CSCs from the same cultures survive the toxic insults [60]. This differential sensitivity was not the result of proliferation differences between CSCs and more differentiated cells because it was also observed when using treatments that are not dependent on cycling cells [60]. Moreover, these surviving CSCs can re-establish the culture, confirming that they are responsible for therapy failure [60]. Chemotherapy-resistant CD133⁺ CSCs were also described in liver and lung cancer [61, 62]. Similar observations were obtained in pancreatic cancer where CSCs were isolated from patient specimens and subsequently treated with gemcitabine. Also in this tumor type, in vitro cell death was more pronounced in the differentiated CD133⁻ cells compared to CD133⁺ cells [48]. Finally, GBM CSCs and breast CSCs isolated from patient specimens, displayed selective resistance to various chemotherapies [63, 64].

Next to the in vitro evidence, escape from therapy was also evident from xenograft studies. Chemotherapy treatment of xenotransplanted CRCs resulted in an increase in CD133⁺ in the tumor [65]. This indicates that CD133⁺ CSCs are more resistant to oxaliplatin in vivo compared to differentiated CD133⁻ cells. In vivo resistance of CRC CSCs is not restricted to oxaliplatin because mice bearing human CRC tumors treated with irinotecan show an increase in cells that express ESA⁺/CD44⁺/CD166⁺, which are distinct markers for CSCs [66]. Moreover, in vivo gemcitabine treatment of xenotransplanted pancreatic cancer induced the CD133⁺ fraction, suggesting a gemcitabine-resistant CSC population [48]. Finally, also in xenotransplants of AML enhanced CD34⁺/CD38⁻, CSCs were observed in vivo upon cytosine arabinoside treatment [67].

Intriguingly, therapy resistance appears to be a general feature of these cells and is not restricted to chemotherapy, but also is observed with radiotherapy [68]. Irradiated glioblastoma, either implanted subcutaneously or intracranially, showed an increase of CD133⁺ cells compared to non-irradiated tumors. This distinction was suggested to be clinically relevant because these findings were extended to ex vivo irradiation of surgically removed glioblastomas [41].

Also, examples for irradiation encompass other tumor types. MMTV-Wnt1 mice bearing breast tumors showed an increase in the CSC (Thy1⁺ CD24⁺ Lin⁻) fraction after irradiation and, in the same study, CSCs from primary human head and neck cancers were found to be radioresistant [69]. Combined, these data indicate that cell line or primary tumor-derived cells with CSC markers display decreased sensitivity to chemo- and radiotherapy. One potential caveat with this conclusion is that CSC markers are heavily debated, suggesting that the increases observed in marker expression may not represent enhanced stemness [6]. Moreover, xenotransplantation models may not adequately represent the normal situation in patients [6] and select for distinct traits and/or markers. Nevertheless, a first direct hint that this CSC resistance concept could explain minimal residual disease and therapy failure in patients came from a study using a genetically modified mouse model to investigate endogenously growing tumors in which CSCs could be traced using a Nestin reporter construct. Nestin⁺ tumor cells, which represent a quiescent CSC population, could fully repopulate the tumor after temozolomide chemotherapy, whereas selective deletion of these cells prevented tumor outgrowth [70]. These data indicate that CSCs resist therapy and might be a potential cause of tumor relapse. In line with this observation, an increasing list of observations made in patients supports the crucial role of CSCs in tumor relapse after therapy. In patients with GBM, CRC or breast cancer, increased CSC fractions using marker expression were measured after chemotherapy treatment [71-74]. More direct evidence for increases in true functional CSCs came from a study in breast cancer. By contrast to other reports, such studies investigated patient samples and performed functional assays. An increase in mammosphere formation capacity was seen after chemotherapy treatment [72, 73], confirming that stemness showed a relative increase, rather than decrease, upon therapeutic intervention. The growing body of evidence pointing to a role for CSCs in resistance warrants a more detailed survey to increase our understanding of the mechanisms that determine resistance to enable the targeting of these survivors of therapy.

Mechanisms behind therapy resistance

Normal stem cells contain multiple mechanisms to control cell death, which aids to protect these crucial cells from cytotoxic insults. Elevated apoptosis resistance, drug-efflux pumps, enhanced DNA repair efficiency, detoxification enzyme expression and quiescence are all identified as pro-survival mechanisms. Intriguingly, all these mechanisms appear to be hijacked by CSCs. For example, mitochondrial apoptosis is associated with a loss of mitochondrial membrane integrity, which is maintained by a strict balance of anti-apoptotic BCL2 proteins (e.g. BCL2, BCLXL and MCL1), pro-apoptotic BCL2 family members (BAX and BAK) and BH3 proteins (e.g. BIM, BAD and NOXA). A cytotoxic insult-induced imbalance in the ratio of these molecules results in permeabilization of the mitochondrial outer membrane and subsequent activation of a caspase cascade [75]. In stem cells, as well as in CSCs, elevated anti-apoptotic protein expression increases the threshold for apoptosis induction and thereby directly protects the cells against apoptosis. For example, in breast and AML CSCs, BCL2 and BCLXL are highly expressed [76, 77]. Similarly, in primary GBM cultures, CD133⁺ CSCs had elevated expression of BCL2 and BCLXL compared to their more differentiated CD133⁻ progeny [78]. In agreement with a role for apoptosis regulation in CSCs, direct proteomic analysis of CRC CSCs and differentiated cells revealed ‘apoptosis’ as one of the main molecular pathways affected, involving differential expression of key anti-apoptotic proteins, including BIRC6 [79]. Combined, this suggests that CSCs have an elevated anti-apoptotic threshold. Recent data confirmed this idea using so-called BH3 profiling, an assay to directly measure the apoptosis priming state of cells [80]. This revealed that CRC CSCs were less-primed compared to differentiated cells, which at least in part explains their resistance to conventional chemotherapy [60]. In agreement, sublethal doses of BH3 mimetics can change this threshold and strongly sensitize CSCs to chemotherapy.

Besides an elevated apoptotic threshold, CSCs display high expression of drug efflux pumps, such as ATP-binding cassette (ABC) transporter family proteins [71, 81-83]. These proteins are important for efflux of chemotherapy across the plasma membrane [4]. Various ABC transporter proteins are highly expressed in HSC and in AML CSCs (CD34⁺/CD38⁻) compared to the nonstem (CD34⁺/CD38⁺) cells [81]. Also in GBM and melanoma, high expression of drug efflux pumps in CSCs is reported [82]. In the latter, expression of ABC transporter ABCB5 serves as a marker for CSCs [83]. Unexpectedly, in CRC, a different scenario is reported where differentiated cells, rather than CSCs, express high levels of the drug efflux pump ABCB1. It was suggested that differentiated cells protect CSCs from chemotherapy treatment by forming a protective rim around the CSCs [84].

All of the above suggests that CSCs employ means to avoid the impact of therapy, which we can potentially circumvent using combination therapy. However, a potentially more challenging problem is the recent observation that CSCs may exist that display quiescent properties. Selectivity of chemotherapy for cancer cells relies on the fact that chemotherapy mainly kills cells that are highly proliferative. Because rapid uncontrolled proliferation is a standard feature of many tumor cells, chemotherapy is assumed to target tumor cells selectively over nonproliferating normal cells, consistent with the observed toxicity in organs with rapidly dividing cells, such as bone marrow, digestive tract and hair follicles. By contrast, slow proliferating or quiescent normal cells are largely protected from chemotherapy treatment. Importantly, this resistance also extends to quiescent tumor cells. In ovarian cancer, CD24⁺ CSCs are less proliferative and more resistant to chemotherapy compared to CD24⁻ cells [85]. Recent data point to the existence of CSCs that are quiescent. These can be identified using the dye PKH26, which dilutes out when cells proliferate and therefore only low or nonproliferative cells will retain the label. In primary melanoma cultures, label-retaining cells were detected with a very low doubling time of approximately 4 weeks in vitro. Although these cells are slow dividing, they have increased sphere-forming capacity in vitro, suggesting that these label-retaining cells are enriched in CSCs [86]. Such quiescent cells are also identified in pancreatic adenocarcinoma and shown to be enriched for CSC markers such as CD133, CD24⁺/CD44⁺ and ALDH. In agreement with this notion, these label-retaining cells are more tumorigenic, indicating that cancer is not only a disease of homogeneously rapidly proliferating cells, but also contains quiescent cells that can escape classical chemotherapy and subsequently induce regrowth of the tumor [87].

Quiescent cells not only display decreased sensitivity to chemotherapy, but also contain enhanced potential and/or time to repair the damage that is inflicted to them. Because many chemotherapeutic agents as well as radiotherapy work by inducing DNA damage, cells that effectively repair DNA damage can potentially survive chemotherapy. Various studies have shown that CSCs (e.g. from GBM) possess high DNA repair activity, which makes them resistant to radiation and chemotherapy [41]. Similarly, in breast CSCs, there is increased expression of DNA repair genes, indicating that high DNA repair pathway activity may aid in making CSCs resistant to tumor therapy. In conclusion, there are many ways for CSCs to resist tumor therapy. Figure 2 illustrates the reasons for therapy resistance in CSCs.

Killing CSCS: magic bullets or combination cocktails?

Although considered as bad news, the efficient DNA repair of CSCs may also point to a dependency for these mechanisms and, as such, offer a means for targeting these cells. For example, CSCs in GBM show elevated activity of Chk1 and ATM and survive irradiation, although inhibition of the cell cycle checkpoint kinases Chk1 and Chk2 is sufficient to sensitize CSCs towards irradiation [41]. Recently, it was reported that combined Chk1 and PDK1 inhibition is required to kill CSCs in GBM [88]. Similarly, nonsmall cell lung cancer CSCs can be sensitized to chemotherapy by combining treatment with the Chk1 and Chk2 inhibitors, SB218078 or AZD7762 [89]. Mechanistically, inhibition of Chk1 results in an active Cdc2-cyclin B complex that is followed by mitotic catastrophe [90]. Although effective, these compounds are also relatively toxic and combination of the Chk inhibitor AZD7762 with gemcitabine demonstrated cardiac toxicity [91]. To overcome this toxicity, an inhibitor of a downstream target of Chk1 was developed, namely Wee1. In the presence of DNA damage Wee1 arrests cells in G2 phase and allows cells to repair DNA before entering into mitosis. Interestingly, Wee1 is reported to be overexpressed in GBM CSCs. The same study showed that inhibition of Wee1 with PD0166285 sensitizes GBM CSCs towards irradiation [92].

Besides targeting the core of the repair machinery, much effort is made with respect to targeting the execution machinery in cancer cells. Previously, we used an inducible caspase-9 to target colon CSCs. Upon activation of caspase-9, colon CSCs were killed efficiently in vitro and in vivo suggesting that activation of caspases is sufficient to efficiently kill CSCs [93]. As described, anti-apoptotic proteins are highly expressed in various cancers and especially in CSCs. Targeting these anti-apoptotic proteins using small molecules that have been developed therefore forms an attractive mechanism. For example, ABT-737, a small molecule inhibitor that targets BCL2, BCLXL and BCLW, tips the apoptotic balance to a more pro-apoptotic state and reverts the resistance of colon CSCs [60, 94]. Unfortunately, Navitoclax (ABT-263), an orally bioavailable variant of ABT-737 with the same specificity, appears to exert high toxicity for platelets, which depend on BCLXL for survival [95]. Although more selective inhibitors have been developed to overcome this problem, such as ABT-199 targeting only BCL2, our recent data indicate that colon CSCs are dependent on BCLXL for survival. In agreement, inhibition of BCLXL with ABT-737 or a BCLXL specific inhibitor (WEHI-539) is sufficient to kill colon CSCs. Similarly, lung CSCs are shown to be dependent on BCLXL and, also in these cells, inhibition eliminates lung CSCs in vitro and in vivo [96]. Although this still raises the problem of toxicity, it is possible to use sublethal amounts of BCLXL inhibition, which is sufficient to strongly sensitize CSCs towards chemotherapy [60]. It remains unclear why colon CSCs acquire this dependency on BCLXL. One possible explanation is the observation made by Todaro et al. [65], where an autocrine loop of interleukin (IL)-4/IL-4R in colon CSCs appears to maintain BCLXL levels and protect CSCs from chemotherapy.

The presence of several cell surface markers allows for the identification of CSCs. Various groups and companies have therefore developed immunotoxins that directly target such CSC markers. For example, antibodies against, for example, the stem cell marker CD133 conjugated to paclitaxel or cytolethal distending toxin target CD133 expressing cells and show in vitro and in vivo elimination of tumors [97, 98]. Similarly, targeting of CD133⁺ cells can be achieved by generation of CD133-specific measles viruses. These oncolytic viruses infect CD133 expressing cells and destroy them by lysis [99]. Moreover, selective killing of CD133⁺ GBM cells was shown using CD133 antibodies coupled to single-walled carbon nanotubes. These anti-CD133-single-walled carbon nanotubes induce the thermal destruction of cancer cells when combined with near-IR laser light [100]. However, CD133 expression is not only specific for CSCs, but also is expressed in normal stem cells, which should be protected from such therapies at all times. To minimize toxicity and to deliver drug to cancers selectively, photochemical internalization was developed. This technique makes it possible to release the drug in the tumor area specifically [101]. Next to CD133, there is an increasing effort to target other cell surface molecules, including the stem cell marker Lgr5. Nevertheless, as is the case for CD133, toxicity with such an approach can be expected. Unpredictably, antibodies without toxins targeting other cell surface molecules are shown to be efficient in killing CSCs. Antibodies against CD47 show promising effects in various cancers. CD47 is a receptor that is involved in inhibition of so called ‘eat-me’ signals and is highly expressed on CSCs compared to more differentiated cells. The blocking of this CD47 receptor with an antibody enables the phagocytosis of AML CSCs and thereby blocks tumor growth [102]. In addition, CD47 inhibition also blocks tumor growth in solid cancers, such as breast cancer, CRC, ovarian cancer and GBM, which is also suggested to depend on facilitating the phagocytosis of CSCs [103].

Next to the induction of phagocytosis, several antibodies delete essential signals from CSCs. For example, direct targeting of breast CSCs can be achieved by using an antibody against CXCR1. The IL-8 receptor CXCR1 is expressed almost exclusively on CSCs and repertaxin, an inhibitor of CXCR1/2, or anti-CXCR1 treatment induces cell death in CXCR1⁺ breast CSCs, which appears to be mediated by Akt signaling inhibition [64]. Intriguingly, phosphoinositide 3-kinase (PI3K)/Akt signaling addiction in colon CSCs was also reported in colon CSCs, where a CD44v6-positive subset was identified that is exclusively metastatic. These cells express high levels of PI3K, which, if inhibited, alters the viability of the cells and impedes their capacity to migrate [104], suggesting that PI3K signaling is crucial for CSCs.

CSCs require signaling through morphogenic pathways for their maintenance, suggesting that these may be attractive targets for therapy as well. In agreement, a screen for CSC-sensitizing compounds identified salinomycin, which inhibits Wnt signaling and eliminates breast and chronic lymphocytic leukemia CSCs [28, 32]. Because CSCs in chronic myeloid leukaemia, AML and skin tumors are dependent on the Wnt pathway, inhibition can be clinically relevant [29, 105-107]. Furthermore, inhibition of Notch signaling pathway using a neutralizing antibody against DLL4 results in less tumor engraftment in secondary tumors, suggesting in vivo differentiation of CSCs. Importantly, DLL4 antibody was also able to sensitize the tumor to irinotecan in vivo [33]. Inhibition of Notch signaling was also sufficient to deplete GBM CSCs and sensitize ovarian CSCs to chemotherapy [35, 108]. Lastly, HH signaling can be inhibited by using cyclopamine and this smoothened antagonist sensitizes AML CSCs to Ara-c treatment [109]. Similarly, in GBM and in pancreatic cancer, decreases in CSCs are observed after treatment with smoothened inhibitors cyclopamine or CUR199691 [110, 111]. These data point to a crucial role for HH signaling in cancer stemness and this is confirmed by knockdown of smoothened, which results in loss of chronic myeloid leukaemia CSCs [112]. Antibodies can also be used to target the CSC niche. Blood vessels maintain GBM CSCs in a stem-like state. Targeting of the microenvironment with bevacizumab, an antibody against vascular endothelial growth factor, is able to differentiate GBM CSCs [113]. Jin et al. [114] showed that using an antibody against CD44 decreased homing of AML cells, thereby promoting the AML CSCs to differentiate to a more mature cancer cell progeny. This antibody inhibited AML growth in mice [114].

Forcing CSCs to nontumorigenic differentiated cells is clinically very relevant and is reported to occur with BMP4 as well. In CRC, BMP4 expression is exclusively expressed by differentiated cancer cells and is shown to induce the differentiation of CSCs and sensitization to oxaliplatin in vivo [115]. In addition, BMP4 also forces GBM CSCs to differentiate and thereby inhibits their tumorigenicity [116].

Not only the inhibition of morphogenic pathways, but also the activation of signaling pathways can change the fate of CSCs. Activation of the unfolded protein response (UPR) induces the differentiation of stem cells in the mouse intestine [117]. In line with this, salubrinal induces UPR in colon CSCs and forces them to differentiate. In addition to inducing differentiation, UPR sensitizes cells to chemotherapy in vitro and in vivo (M. C. B. Wielenga and S. Colak, unpublished observations). Although the targeting of CSCs by forcing them to differentiate or by the induction of apoptosis appears to be an attractive therapeutic option, the suggested flexibility of the system is a clear caveat. Even when CSCs are eliminated within a tumor, differentiated cells can de-differentiated and take the place of the CSCs that were deleted [31, 43]. Targeting the cues that induce de-differentiation or simply attacking both CSCs and more differentiated cells needs to be achieved to eradicate a tumor [6]. The direct targeting of CSCs and CSC differentiation therapy represent promising methods for improving tumor therapy. Further studies are needed to investigate the most promising combination treatments that do not result in severe toxicities.

Summary

The identification of CSCs in many tumors has enabled a better understanding of why tumors can relapse even many years after therapy. There is increasing evidence indicating that the targeting of these CSCs is important for improving therapies. We have reviewed the mechanisms that make CSCs resistant to therapy. A better understanding of such mechanisms and the way that CSCs retain their tumorigenic stem cell capacities is crucial. The exciting new insights obtained will undoubtedly provide new therapeutic tools in the years to come.