P. G. Pelicci, Department of Experimental Oncology, European Institute of Oncology, Via Adamello 16, 20139 Milan, Italy Fax: +39 02 9437 5990 Tel: +39 02 9437 5040 E-mail: firstname.lastname@example.org
Mutations can confer a selective advantage on specific cells, enabling them to go through the multistep process that leads to malignant transformation. The cancer stem cell hypothesis postulates that only a small pool of low-cycling stem-like cells is necessary and sufficient to originate and develop the disease. Normal and cancer stem cells share important functional similarities such as ‘self-renewal’ and differentiation potential. However, normal and cancer stem cells have different biological behaviours, mainly because of a profound deregulation of self-renewal capability in cancer stem cells. Differences in mode of division, cell-cycle properties, replicative potential and handling of DNA damage, in addition to the activation/inactivation of cancer-specific molecular pathways confer on cancer stem cells a malignant phenotype. In the last decade, much effort has been devoted to unravel the complex dynamics underlying cancer stem cell-specific characteristics. However, further studies are required to identify cancer stem cell-specific markers and targets that can help to confirm the cancer stem cell hypothesis and develop novel cancer stem cell-based therapeutic approaches.
In a healthy organism, adult stem cells (SCs) maintain tissue homeostasis throughout life and are thought to be rare, largely quiescent cells capable of both self-renewal, to maintain the stem cell pool, and differentiation, to ensure life-long production of all mature cells within a tissue [1,2].
Similarly to highly regenerative tissues, cancer can be considered as a malignant tissue organized along a comparable architecture with, at its apex, a cell with aberrant self-renewal and differentiation capacities. Evidence to support this paradigm was first observed in human acute myeloid leukaemia (AML) and then expanded to solid tumours [3–6]. This hierarchical organization of tumours is reflected in their intratumoral biological heterogeneity, as cells within individual tumours are phenotypically and functionally distinct [7,8]. Heterogeneity includes traits that are critical to transformation, such as differentiation and proliferation.
Almost half a century ago, it was demonstrated for the first time that only a restricted pool of cancer cells was responsible for the massive malignant proliferation in lymphoma . Indeed, it has now been shown that only a small subpopulation of cells in the cancer tissue has tumorigenic capacities; because these cells exhibit characteristics similar to normal SCs, such as self-renewal and differentiation, they have been named cancer stem cells (CSCs) .
However, despite an intense effort, the biological properties of normal and cancer SCs, such as mode of division, cell cycle, replicative potential and DNA-damage response, are still not fully characterized. In this review, we provide a synthesis of the latest insights in the regulation of normal and cancer stem cell self-renewal and the contribution from our group to this field. It is envisaged that a deeper understanding of the molecular and biological features that sustain CSCs will lead in the near future to the identification of alternative approaches to eradicate cancer.
Self-renewal in normal stem cells
Regulation of normal SC self-renewal is tightly controlled by cell intrinsic and extrinsic mechanisms. Integration of extrinsic cues from the microenvironment or niche with intrinsic signalling pathways helps to prevent premature exhaustion of the SC pool through the production of growth factors, cytokines, secreted proteins and cell–cell interactions . To date, the precise mechanisms controlling SC self-renewal in contexts such as homeostasis, stress or injury and malignancy have not yet been fully delineated. Developmental pathways such as Wnt, Notch and Hedgehog have been identified as core intrinsic signalling cascades controlling the balance between self-renewal and differentiation in neural, epidermal, intestinal, breast and haematopoietic SCs (for current reviews see 2,11–13). Similarly, cell survival and cell-cycle-regulating pathways, such as p53, Bmi-1 and cyclin-dependent kinase inhibitors (CDKI), represent an additional intrinsic regulatory mode of SC self-renewal [14–17]. In particular, it has been recently demonstrated that haematopoietic stem cell (HSC) self-renewal is governed by members of the Cip/Kip family of CDKI: p21, p27 and p57 [18–21]. However, the direct regulation of SC fate via CDKIs has not been extensively explored in other adult SCs. It should be noted that CDKI regulation of HSC fate decisions is context dependent because it has been shown that p21 and p27 act primarily under conditions of stress, whereas p57 loss affects HSC self-renewal in homeostatic conditions [20,21]. Not surprisingly, deregulation of SC self-renewal is thought to lead to defective homeostasis of the organ system, contributing to the development of malignancies.
The CSC hypothesis
The CSC hypothesis proposes that a small pool of low-cycling CSCs is necessary and sufficient to initiate and maintain the disease. Moreover, CSCs would function as a reservoir for the reappearance and metastatic evolution of the disease. Thanks to its unique biological features, this small pool of cancer cells is thought to be responsible for the clinical observations that nearly all tumours are heterogeneous and that relapse occurs in patients considered tumour-free for many years. Even though the CSC hypothesis is appealing and, on the whole, accepted by the scientific community, many aspects of this theory need to be conclusively proven. In fact, the existence of CSCs, and whether they are sufficient to maintain tumour growth in humans, has not yet been definitively confirmed.
In mice, only rare tumour cells, the so called tumour-initiating cells, possess the ability to propagate tumours upon transplantation in host animals [3,22,23]. Tumour-initiating cells, on clonal or multiclonal bases, are the only cells within a tumour with the capability to initiate new tumours in vivo. The transplanted tumours show the same biological heterogeneity as the parental tumours, even after single cell transplantation, suggesting that a single tumour-initiating cell is able to originate diverse tumour cell subpopulations . Because only SCs are endowed with the ability to both self-renew and differentiate to give rise to mixed cellular populations, the tumour-initiating cell might in fact be a cancer stem cell. However, it is still unclear whether tumour initiation is driven by a genomically advantaged stem cell or by a more differentiated cell which has reacquired stem cell properties, or if both events are possible. Within a cancer, tumorigenic and nontumorigenic cells can be distinguished by the expression of different lineage markers [24–26]. Indeed, fractionation of tumour cell subpopulations based on lineage markers demonstrated that not all cancer cells have the same stochastic probability to give rise or propagate a tumour. In particular, compared with the CSC population, the cancer cells that form the bulk of the tumour are nontumorigenic and are not able to maintain tumour outgrowth and propagate the disease , even though they represent the fast-cycling fraction of tumour cells. Moreover, the frequency of CSCs can vary considerably among different patients or animal models and it is generally higher in poor-prognosis tumours . Actually, the most severe tumour phenotypes are associated with increased proliferation and de-differentiation, suggesting that the abundance of CSCs, or, more generally, of cells endowed with self-renewal capacity, correlates with the disease outcome .
Tumour progression is promoted by the continuous selection of mutations that confer advantageous phenotypic traits. The target cell of transforming mutations can be a stem cell, as widely demonstrated for certain leukaemias [24,28] and other tumours [29–31], or a progenitor cell that acquires a gain of function mutation that endows it with self-renewal capability [5,10,32–34]. Misregulation of the signalling pathways that control the self-renewal of normal SCs can lead to oncogenesis . Thus, both the genesis and fate of a cancer reside in a ‘cancer cell of origin’ with aberrant self-renewal capabilities.
Identification and isolation of normal and cancer stem cells
The identification of markers expressed by SCs is essential for their purification and biological analyses; currently, normal SCs have been isolated to near homogeneity. When considering CSC subsets, this task appears to be increasingly more complicated, depending on the tissue analysed.
Landmark studies by McCullough and Weismann, as well as other groups, have described a hierarchical organization of the haematopoietic system [35–38]. In this model, the production of all the mature blood cells is carried out by a single HSC. Several combinations of cell-surface antigens are currently used to isolate subsets enriched in murine HSCs such as Lineage− (Lin−), cKit+, Sca-1+ and CD34− Flk2− and/or CD150+/CD48− [39–45]. An equivalent human population resides in the Lin− CD34+ CD38− fraction of the bone marrow. The work of Dick and colleagues demonstrated that human CD34+ CD38− AML cells are the only subset capable of propagating leukaemia in immune-deficient mice, suggesting that leukaemia and normal SCs express similar markers [3,24]. This has been corroborated in mouse models of chronic myeloid leukaemia; however, it remains more controversial in human chronic myeloid leukaemia and AML models harbouring specific translocations such as MLL-AF9 [5,10,46,47].
By contrast, the developmental organization of the mammary gland is less well characterized, consonant with other epithelial tissues such as in the brain and intestine. Unlike in the haematopoietic system, pinpointing unique markers which can identify normal mammary and cancer SCs has proven challenging. Two recent papers identified different candidates expressed on mouse normal mammary SCs such as Lin− (CD31−, CD45− and Ter119−), CD29high and CD24+  and Lin− CD24med CD49fHigh . However, human tumorigenic mammary SCs appear to express other markers such as CD44+ CD24−/low and ALDH1+ [25,50]. The lack of consistency in marker combinations could be due to the inherent heterogeneity observed between human tumours and in mouse tumour models.
Marker analysis is mainly used as a tool for the enrichment in SC populations, and is not informative for the study of the functional properties of SCs, particularly CSCs. Several assays are currently used to study the biological properties of normal and cancer SCs, including in vivo transplantation and label-retaining assays [14,39,40,51]. Indeed, whether or not the CSC population exhibits identical characteristics of normal SCs, such as quiescence, is still a matter of debate [14,27,52,53]. Limiting dilution transplantation experiments are the gold standard to assess SC frequency and have been extensively used in the haematopoietic system to evaluate normal and cancer samples. Label retaining assays exploit the notion that SCs are quiescent or slow-cycling cells compared with actively proliferating progenitors. Briefly, all cells in a tissue are labelled equally by a fluorescent marker (e.g. histone H2B–green fluorescent protein [GFP]), which becomes more dilute with each round of cell division. Functional evaluation of label retaining HSCs has been demonstrated using a histone H2B–green fluorescent protein transgenic mouse [54,55]. HSCH2B-GFP+ cells showed multilineage potential in primary and secondary transplantations in mice, proving long-term self-renewal potential in vivo.
These same strategies have been used to identify candidate SC populations in epithelial tissues such as the mammary gland [56–58]. Initial transplantation experiments using fragments of the mammary epithelium resulted in regeneration of a functional gland, suggesting the presence of SCs . Several groups have isolated mammary SCs based on Hoechst 33342 efflux [56–58]. Our laboratory has employed a different label-retaining approach to identify SCs in the breast. We showed that a lipophilic fluorescent dye, PKH-26, can be used to purify subsets of mammary epithelial cells [14,27]. The PKHHigh fraction consisted of an almost pure population of mammary SCs, because regeneration of a gland was achieved by single cell transplantation in 38% of cases. We demonstrated that tumour PKHhigh cells exhibit the most robust SC activity in normal and tumorigenic mammary glands. Notably, an increase in SC frequency was observed in ErbB2 transgenic mice compared with the normal gland (1 : 1.4 cells compared to 1 : 3). The reason for this observation is discussed in more detail in the following section. Finally, we observed a significant correlation between the frequency of PKHhigh cells and poorly differentiated human tumours (G3), suggesting that CSC frequency and their inherent biological properties could be a function of tumour grade .
Mode of division and cell-cycle properties in normal and cancer stem cells
SCs have characteristic cell-cycle properties that allow them to generate two cells with different developmental potentials: a quiescent cell that preserves the SC identity and a proliferating progenitor with the capacity to divide several times before committing to a fully differentiated state . This self-renewing division is mainly accomplished by SCs through an asymmetric cellular division that results in the differential distribution of structural components, DNA and protein determinants into the two daughter cells. Even if the mechanisms of asymmetric division in vertebrates are not well characterized, evidence of its existence has been collected for SCs of several tissues, such as skin, mammary gland, muscle, gut and haematopoietic system [14,61–64]. Asymmetric division is supposed to be a mechanism for the maintenance of self-renewal potential in SCs but it does not allow expansion of the number of SCs. Therefore, SCs can also divide symmetrically, as is the case in the embryo, during certain developmental stages, or in response to tissue injury. Symmetric division produces two daughter cells of the same stem fate and therefore can lead to an increase in the overall pool size, which can be both deleterious or beneficial given the context (Fig. 1).
A tight control on SC asymmetric cell divisions is important to prevent the formation of aberrant SC pools with unrestrained proliferation, which might result in overgrowing tissues. In Drosophila, the misregulation of asymmetric cell division switches neural SCs into tumour-initiating cells, and many regulators of asymmetric cell division in the fly are also tumour suppressors [65,66]. In vertebrates, many homologues of these fly genes also act as tumour suppressors, suggesting a close relationship between the symmetry of cell division and tumour formation [67–69]. Defects supporting the symmetric division of SCs result in tumorigenesis, because symmetric divisions generate highly proliferating SCs, promoting the expansion of SC number and exposing SCs to the risk of harmful mutational events.
Recent studies on breast SCs performed in our laboratory demonstrate that symmetric and asymmetric divisions coexist in both normal SCs and CSCs, but in different ratios: SCs mostly divide asymmetrically, whereas CSCs preferentially divide symmetrically . SCs deficient (p53−/−) or weakened (ErbB2-expressing) in their p53 response primarily underwent symmetric divisions, indicating that this pathway as a key component in determining the mode of SC division . Preliminary unpublished data from our laboratory also suggest the involvement of the oncogene Myc in the p53-mediated regulation of SC division (E. Pasi Cristina, unpublished data). Beyond the fact that Myc is a key mediator of self-renewal in other adult SCs [70,71], overexpression in mammary SCs phenocopies loss or attenuation of the p53 pathway (E. Pasi Cristina, unpublished data). In normal SCs and CSCs, Myc expression alone can sustain a pool of expanding SCs in the absence of transformation or apoptosis (E. Pasi Cristina, unpublished). Moreover, Myc acts through additional modalities such as the reprogramming of progenitors, albeit at a low frequency, and induction of genomic instability . Therefore, the p53–Myc axis is crucial in regulating the normal SC and CSC pool (Fig. 1).
Normal SCs function as a ‘reservoir’ of cells that the body can use during a lifetime, both for the maintenance of tissue homeostasis and under specific conditions, such as tissue regeneration or injury repair. For this reason, normal SCs are expected to divide asymmetrically only a limited number of times [20,73–75]. When a SC exceeds this established limit, it undergoes functional exhaustion, demonstrating that the self-renewal potential of normal SCs is intrinsically restricted. Normal and cancer SCs exist in two hierarchical subpopulations: quiescent and proliferating. Absolute numbers of quiescent SCs are comparable in normal and cancer SCs, however, normal SCs are mainly quiescent, whereas CSCs are mainly proliferating.
When SCs exit their quiescent state to enter cellular division they are exposed to genotoxic damage and begin to accumulate DNA damage. The physiological accumulation of DNA damage that occurs with ageing causes a decline in SC self-renewal potential and eventual functional exhaustion . The accumulation of DNA damage in SCs is a risk that has to be tightly controlled because potential mutations occurring in SCs can be transmitted to both the daughter SC and the downstream progenitors. In order to protect the SC compartment, they are kept in a noncycling state that prevents them from accumulating dangerous DNA damage. In the so-called quiescent or dormant state, SCs slow their general energy metabolism and stay ‘frozen’ in a semipermanent G0 phase of the cell cycle . More than a decade ago it was demonstrated that the CDKI p21 is the guardian of the quiescent state of HSCs . p21 restricts cell cycling of SCs, and loss of expression induces uncontrolled entry into the cell cycle, with consequent increased DNA damage accumulation and premature functional exhaustion [21,77]. In addition to p21, it has recently been shown that another CDKI belonging to the Cip/Kip family, p57, is required for HSC quiescence and self-renewal. Cytoplasmic p57 prevents cyclin-D1 shuttling to the nucleus thus preserving HSC quiescence [18,19].
We demonstrated that differently from their normal counterparts, CSCs divide symmetrically and possess an unlimited replicative potential that allows them to undergo an indefinite number of rounds of cell division, thus boosting the number of proliferating SCs in tumoral tissues . In spite of the fact that CSCs hyperproliferate during tumorigenesis and are more vulnerable to DNA-damage accumulation, they do not exhaust their replicative potential but rather become immortal. These findings suggest that there must be specific mechanisms activated during tumorigenesis that prevent the functional exhaustion of CSCs. In order to investigate the mechanisms that regulate CSC immortality, we dissected the role of p21 in the regulation of leukaemia SC self-renewal. We demonstrated that the expression of leukaemia-associated oncogenes in normal HSCs induces DNA damage and p53-independent upregulation of p21 . In turn, the cell-cycle inhibitor p21 activates a cellular response that forces cell-cycle restriction and activates repair of the damaged DNA, preventing the physiological breakdown of HSC self-renewal, which would normally occur, in time, due to accumulation of DNA damage. In the absence of p21, leukaemia SCs hyperproliferate, accumulate massive DNA damage and consequently exhaust their replicative potential . Therefore, the upregulation of p21 confers an advantage on HSCs when they hyperproliferate, as noted during malignant transformation. Moreover it has been suggested that p21 overexpression in CSCs is not transforming per se, however, it is ‘permissive’ for leukaemia to progress, because p21-deficient leukaemia cells are not transplantable . All these observations reveal that increased levels of p21 are indispensable for maintaining the quiescent pool of CSCs by preventing excessive accumulation of DNA damage and functional exhaustion (Fig. 1).
Recent findings on the regulation of SC division and replicative potential indicate that the tumour suppressor p53 plays a role both in tumour formation and normal tissue homeostasis (for a review see ref. ). In normal SCs, p53 imposes an asymmetric mode of division , possibly via myc downregulation (E. Pasi Cristina, unpublished observations). We recently showed that p53 signalling is attenuated in ErbB2-driven breast tumours, and that its pharmacological reactivation re-establishes asymmetric division in CSCs, suggesting that loss of p53 in epithelial cancers supports symmetric divisions of CSCs, contributing to tumour growth [14,72] (Fig. 1).
Thus both the tumour-suppressive mechanism by which p53 mediates asymmetric division of SCs and the tumour-promoting mechanism by which oncogene expression induces the activation of p21 are ideal targets for the development of new anti-CSC drugs. Using Nutlin 3 to induce pharmacological restoration of p53 in ErbB2 transgenic mouse breast tumours, we observed tumour regression associated with a selective reduction in CSC number . Our studies on the role of p21 in leukaemogenesis suggest that inhibition of DNA repair might be synthetic lethal with oncogene expression, therefore other new anti-CSC drug opportunities might arise from the direct inhibition of p21 or the usage of selected DNA repair inhibitors. Our laboratory is exploring the possibility of boosting the DNA damage of leukemic cells using an inhibitor of poly(ADP-ribose) polymerase in combination with conventional chemotherapeutic agents.
DNA-damage processing in SCs
SCs generate all the progeny of differentiated cells that give rise to a functional normal tissue; it is therefore fundamental that they are equipped with specific and effective DNA-damage response mechanisms in order to avoid propagation of genetic lesions to all their progeny. Indeed, adult SCs possess specific means of protection against genotoxic insults: they show increased activity of transporters that pumps genotoxic compounds out of the cells and they are mostly quiescent and metabolically inactive, minimizing replication errors and the production of reactive oxygen species [80–82]. Nonetheless, as any other cell in the organism, they are subjected to DNA damage and we are only now starting to elucidate their response mechanisms. In most adult tissues (one exception being the intestine), SCs appear to be more resistant to DNA damage than their differentiated progeny [83–86]. In fact, whereas differentiated and proliferating progenitors that harbour DNA damage are mostly eliminated by activation of an apoptotic response or undergo premature senescence, SCs survive by activation of specific prosurvival and DNA-repair responses. Upregulation of p53 in all these type of cells has been always observed during these responses to induced DNA damage, irrespective of their degree of differentiation [80,86,87]. However, it remains unclear why this does not result in apoptosis or senescence in the SCs, as observed in progenitors, and how p53 is governing these responses. In an effort to try to elucidate the mechanisms through which SCs respond to DNA damage, our group has studied the responses elicited by X-ray irradiation in highly purified SC populations derived from both the haematopoietic and the mammary systems (I. Alessandra and C. Angelo, unpublished data). Our data suggest that the quiescent long-term HSC (LT-HSC) population possesses specific regulatory mechanisms. Indeed, irradiation activates p21 independently of p53, and p21 activation results in the inhibition of p53-dependent apoptosis. Identical responses were obtained by irradiation of mammary stem cells, indicating that these specific responses to DNA damage are conserved in SCs of different origins. Because p21 is involved in cell cycle checkpoints that block cellular proliferation following DNA damage, in order to allow DNA repair  we investigated the effects of the upregulation of p21 on the cell-cycle properties of SCs. Surprisingly, we observed that irradiation induces a p21-dependent cell-cycle entry and expansion of absolute numbers of SCs. Thus, our observations (I. Alessandra and C. Angelo, unpublished) suggest that in adult SCs, upregulation of p21 results in downregulation of p53 activity and that this is required not only for the inactivation of apoptotic responses, but also for the induction of cell-cycle entry and symmetric self-renewing division. Moreover, they indicate that p21 activates DNA damage repair mechanisms, limiting its accumulation and maintaining self-renewal.
Oncogenes induce the accumulation of DNA damage and, therefore, activate a DNA-damage response that induces very potent tumour suppressor mechanisms through activation of surveillance systems, such as apoptosis and senescence . The initial observation of the existence of a permanent cell-cycle arrest induced by oncogene expression, defined as oncogene-induced senescence, derived from studies in vitro . However, these signalling responses are activated in the very early phases of cancer progression and rely on the activation of fundamental checkpoints in the cells. Indeed, since then, several groups have shown, both in mouse models and in human samples, that this process is activated in vivo in the premalignant lesions of several cancers, such as lung, breast, colon, prostate, bladder, melanoma and lymphoma. Aberrant activation of several oncogenes such as Ras, Myc, B-Raf and growth factor receptors results in oncogenic stress that induces DNA damage [91–99]. DNA damage is mostly due to stalling of replication forks and increased reactive oxygen species production, which, in turn, result in double-strand breaks. This triggers a DNA-damage response that engages Atm and downstream effectors, activating p53 and resulting in the induction of senescence via p21. Tumour progression from premalignant lesions to advanced cancer requires that the tumour cells overcome the barrier imposed by the DNA damage by loss, via mutations or deletions, of the functions of the genes involved in the checkpoints regulating these responses . As mentioned earlier, our group has recently shown the existence of an alternative tumour suppressor mechanism following DNA damage induced by oncogene expression. In the preleukemic phases of AML, PML/RAR and AML1/ETO induce DNA damage in the hematopoietic SCs and activate a p21-dependent response that results in the maintenance of a pool of preleukemic SCs that harbour a moderate degree of damaged DNA compatible with cell survival . These data indicate a paradox as different oncogenes, inducing the same kind of damage, can elicit tumoral protective mechanisms that rely on completely different cellular responses and have entirely different outcomes: apoptosis/senescence versus extended self-renewal. We believe that ours and other groups’ recent studies on DNA-damage responses in SCs may help explain this issue. Indeed, SC and progenitor cells from the same tissue must be regarded as completely different entities from this point of view. Depending on their differentiation stage, cells can adopt different mechanisms in order to respond to DNA damage. Our data (I. Alessandra and C. Angelo, unpublished) suggest that different DNA-damage responses exist in subpopulations of cells with different self-renewal potential. In particular, the progressive loss of self-renewal ability appears to correlate with a switch from a p21-dependent response that inhibits p53 functions and apoptosis in SCs to a p53-dependent response that activates p21-dependent apoptosis in progenitors. These specific ways of handling DNA damage can have important physiological and pathological consequences. In fact, under physiological conditions, following DNA damage, it is necessary to eliminate those damaged progenitors unable to properly execute their functions in the tissue. By contrast, it is essential to allow both the expansion of a functional SC pool able to respond to tissue injuries during the lifetime of an individual and the reconstitution of all the progeny of differentiated cells. However, suppression of apoptosis/senescence in SCs might favour the survival of cells that harbour genomic alterations; in fact, the repair of damaged DNA appears to be incomplete in SCs, because we observed residual DNA damage several months after cell irradiation (I. Alessandra and C. Angelo, unpublished data), and to rely mainly on nonhomologous end joining, an error-prone DNA repair mechanism typical of quiescent cells [83,100]. As a result, DNA mutations might become fixed in the genome of these cells leading to the acquisition of a mutator phenotype, responsible for increased cancer risk and tissue ageing. Indeed, genomic instability is a characteristic of almost all human cancers and cancerogenesis is a multistep process that implies the acquisition of sequential mutations that accumulate in a cell and lead to a progressively more transformed phenotype. Stratton and colleagues see cancer development as an evolutionary programme based on two processes: the continuous acquisition of heritable genetic variations in individual cells by random mutations and the natural selection that acts on the resultant phenotype . Indeed, as discussed in this review, oncogene expression in adult SCs has several fundamental consequences. Oncogenes are able to extend self-renewal of pretumoral SCs and increase the frequency of symmetric self-renewing divisions, thus expanding a stem cell pool which maintains the ability to grow indefinitely. Oncogenes can also cause genomic instability in the long-living SCs because of their capacity to induce DNA-damage and to the SC natural propensity to repair the damage with low efficiency [83,100,102]. We envision a multistep model of tumour onset and progression in which SCs initially acquire mutations that endow them with unlimited self-renewal capacity. This might allow the expansion of an immortal and genomically unstable pretumoral cell population predisposed to the acquisition of secondary mutations, which would be, in turn, responsible for the fully transformed phenotype. Therefore, during the lifetime of an individual, the imperfect repair of DNA damage in SCs, a response that has evolved in order to assure important tumour suppression functions, can instead become very risky and detrimental, contributing to tumour progression (Fig. 2).
New approaches to cancer treatment: targeting CSCs
Based on the CSC hypothesis, the therapeutic approach to a cancer cure is being revisited. Even though some canonical anticancer therapies have achieved satisfactory results in the remission of many malignancies, the problem of metastatic progression and cancer recurrence still exists and needs to be faced. Most of the traditional therapies are quite unselective and cytotoxic because they aim to kill all the cells in rapid expansion. Sometimes this approach brings a divergence between a good clinical response, with significant tumour size reduction, and a unsatisfactory survival response, probably due to the CSC-driven relapse that presumably follows the killing of the bulk cancer cells without eradication of CSCs . Recent studies on the effects of treating leukaemias with valproic acid (VPA), a histone deacetylase inhibitor, have highlighted the importance of eradicating CSCs. In mouse models, VPA treatment rapidly resulted in tumour regression and extended animal survival. However, it never led to eradication of the disease, which rapidly reestablished itself following withdrawal of the drug. The authors showed that this outcome was linked to VPA ability to increase the self-renewal capacity of the CSCs population and to induce its expansion. VPA effect on CSCs might explain the unsuccessful treatment of AML patients with VPA alone [104–106]. Therefore, the new generation of therapeutic approaches needs to take into account the genetic and behavioural differences between the various cell types that compose the tumour and combine efficient shrinking of the tumour bulk with specific targeting of CSCs. The ability of CSCs to escape canonical cancer therapies is associated with some critical intrinsic properties of normal SCs, such as their capacity to repair DNA damage, handle toxic injury, and enter a quiescent or dormant state . In particular, emerging evidence from our and other groups indicating that CSCs enter quiescence in order to prevent self-renewal exhaustion, implies that in order to eradicate the disease, it might be necessary to induce the dormant CSCs to enter the cell cycle. This, in principle, could be achieved using growth factors or cytokines in order to boost CSCs proliferation and render them sensitive to drugs that kill actively cycling cells. In mouse models it has been shown that consecutive treatments with interferon-α and 5′-fluorouracil determine loss of self-renewal potential, up to functional exhaustion of HSCs [73,108]. The role of cytokines in the treatment of AML has been tested in several clinical trials showing highly variable clinical outcomes. For example, the combination of G-CSF with chemotherapy for the treatment of AMLs has been analysed in a randomized clinical trial. This therapeutic approach resulted in reduced relapse rate without improving the rate of complete response or the overall survival . On the other hand, a renewed interest in the use of interferon-α in AML treatments, has been recently roused by the finding that this cytokine appears to induce antileukaemic responses when its level is maintained at high concentrations in the serum (reviewed in ref. 110). Although results from many trials indicate that awaking CSCs from dormancy could be a good therapeutic strategy, we still need to define what are the best therapeutic conditions and the right mode of administration.
Finally, several recent studies, using different mouse models, have focused on the identification of new putative therapeutic targets that specifically deplete CSCs. By RNAi screens, Zuber et al. identified Brd4 as a critical player in leukaemia maintenance in AML . Similarly, in chronic myeloid leukaemia the Musashi-Numb, Beta-catenin and AKT-Foxo3a signalling cascades were shown to regulate the survival of the leukaemia initiating subsets [112–114]. This suggests that a deeper understanding of the underlying molecular mechanisms of SC and CSC self-renewal could help in finding new druggable targets and developing more efficient anticancer strategies.
The idea that a restricted pool of cells can give rise to tumours is not new. In the last 10 years, much effort has focused on the characterization of normal and cancer SC self-renewal properties. Regulation of self-renewal in SCs allows tight control over total pool size and governs the concomitant generation of more differentiated progeny. Notably, CSCs have coopted this regulatory programme to enable unrestricted self-renewal capacity. Many elements appear to influence cell fate decisions, such as mode of division (symmetric vs. asymmetric), cell-cycle regulation and replicative potential. Recent studies from our group and others have begun to elucidate the mechanisms involved in controlling these features. More specifically, we have recently observed a p53-dependent regulation of the Myc oncoprotein that controls symmetric versus asymmetric self-renewing divisions in SCs and CSCs, with symmetric division being strongly favoured in CSCs (E. Pasi Cristina, unpublished data). Inappropriate regulation of this balance resulted in increased CSC frequency; an observation having considerable implications in human tumours because CSC content appears to correlate with poor prognosis in patients. Studies aimed at defining how Myc imposes its effect and if other pathways can influence SC mode of division are now required to understand if these are avenues that could be explored to find new therapeutic approaches. It has also been shown that SCs and progenitors respond differently to DNA damage. SCs are able to handle DNA damage without undergoing apoptosis or senescence but rather acquiring unrestricted self-renewal potential. This phenomenon appears to be mainly mediated by the p53-independent upregulation of p21 and p57. These two cell-cycle inhibitors pause the cell cycle to allow DNA repair, assuring the survival and expansion of cells prone to genomic instability, which might then lead to transformation. Thus, we need to elucidate what the pathways that control p21 regulation in SCs and CSCs are and how they block p53 tumour suppressive functions.
Even though deeper studies on CSC biology are needed, the discovery of the molecular mediators that regulate CSC properties has opened up the possibility to develop new and more efficient cancer therapies that directly target CSCs. In spite of the fact that most of the cancer treatments in use give a good remission rate, they do not prevent relapse of the neoplastic disease. Quiescent CSCs seem to be resistant to conventional therapies and to be responsible of disease reoccurrence after treatment [115,116]. Under certain antiproliferative treatments, such as cell-cycle-specific drugs, CSCs may be activated and selected for a more aggressive phenotype leading to metastatic behaviour . Whole-genome analysis by next generation sequencing has recently been performed on samples from primary/relapse tumour pairs in acute myeloid leukaemia . This is the first indication that chemotherapy can induce the selection of rare tumour subpopulations harbouring specific gene mutations (relapse-specific mutations). Tumour relapse may be linked to the clonal selection and expansion of rare cells that acquired mutations in genes responsible for chemoresistance. It would be fundamental for cancer treatment to understand if these mutations are indeed specifically acquired by the CSCs of the tumour and which are the genes targeted by these mutations.
We would like to kindly thank Paola Dalton for critical reading and editing of the manuscript.