Is cancer a stem cell disease? Theory, evidence and implications


  • T. M. Blacking,

    1. Royal (Dick) School of Veterinary Studies, The University of Edinburgh, Hospital for Small Animals, Easter Bush Veterinary Centre, Midlothian, UK
    Search for more papers by this author
    • *

      These authors contributed equally to this review.

  • H. Wilson,

    1. School of Veterinary Medicine, University of Wisconsin-Madison, Madison, WI, USA
    Search for more papers by this author
    • *

      These authors contributed equally to this review.

  • D. J. Argyle

    Corresponding author
    1. Royal (Dick) School of Veterinary Studies, The University of Edinburgh, Hospital for Small Animals, Easter Bush Veterinary Centre, Midlothian, UK
    Search for more papers by this author

David J. Argyle
Royal (Dick) School of Veterinary Studies
The University of Edinburgh
Hospital for Small Animals
Easter Bush Veterinary Centre
Midlothian EH25 9RG, UK


Cells of muticellular organisms form part of a specialized society that cooperates to promote survival of the organism. In this, cell division, proliferation and differentiation are strictly controlled and a balance exists between normal cell birth and cell death. Inextricably linked to this fundamental concept is the role of the adult stem cell (ASC) whose progeny and microenvironment make up the anatomical structure that coordinates normal homeostatic production of functional mature cells. ASCs have been best characterized in the haematopoietic system but exist in all major organ systems. These cells are characterized by a capacity for self-renewal, being undifferentiated but capable of multilineage differentiation, slowly cycling cells but clonogenic, and capable of asymmetric division. Further, ASC reside in particular ‘niche’ environments that support an appropriate spatiotemporal dialogue between ASC and microenvironmental cells in order to fulfil the lifelong demands for normal differentiated cells.1

For decades, advances in molecular techniques have allowed us to dissect the mechanisms of carcinogenesis, most work focusing on the accepted model of multistage carcinogenesis underpinned by progressive genetic changes that drive malignant transformation. In this model, any cell in the body has the potential for malignant transformation. More recently, attention has focused on an alternative model where the tumour is maintained by a cancer stem cell (CSC) which gives rise to a tumour composed mostly of daughter cancer cells and a small number of CSCs that drive tumour growth and expansion.

Cancer stem cells

The CSC theory probably represents a modern day interpretation of a similar proposal made by Virchow and Cohnheim nearly 150 years ago, proposing that cancer resulted from activation of dormant embryonic tissue. This theory was reawakened in the 1960s and 1970s with suggested theories of maturation arrest in tissue-specific stem cells,2,3 and then ultimately with the identification of the leukaemic stem cell in seminal experiments performed by Fialkow et al. in the late 1960s.4 In the current context, the CSC can be considered a cell that has the ability to self renew and is capable of asymmetric cell division, giving rise to another malignant stem cell and a cell that gives rise to the phenotypically diverse tumour cell population (Fig. 1).

Figure 1.

Two general models of cancer presented. In both (A) and (B) there is tumour cell heterogeneity. However, in (A) many different cancer cell phenotypes have the potential to proliferate extensively to cause a tumour. In (B) we predict that a small subset of cancer cells [the cancer stem cell (CSC)] are the only population that can form a new tumour upon transplantation.

Proof that CSC exists as a phenotypically different population of cancer cells requires isolation of different populations of cancer cells and demonstration that one or more groups are efficient at producing tumours while other groups lack this ability.5 However, these cells cannot be definitively called CSC until it is possible to show that a single transplanted cancer cell can give rise to a diverse population of cancer cells within a tumour.5

Evidence for CSC

Many parallels can be drawn between normal ASC and CSC in terms of clonality and asymmetric division (Fig. 2). Further, it has been shown that when cancer cells of different types are subjected to both in vitro and in vivo assays, that only a small minority of cells are able to proliferate extensively.6 This has given rise to the concept that tumours are composed of both CSC, which have a large proliferative capacity, and a daughter population of cells, with a limited proliferative potential.

Figure 2.

Parallels can be drawn between normal stem cell development and cancer stem cell development. Both pathways share properties of self-renewal and asymmetric division. Ultimately both pathways give rise to cells with limited proliferative potential.

The evidence for true CSC was first documented for haematopoietic malignancies such as acute myeloid leukaemia (AML) and multiple myeloma. In seminal studies, it was found that when mouse myeloma cells were obtained from mouse ascites, and subjected to in vitro colony-forming assays, only 1 in 10 000 to 1 in 100 cancer cells were able to form colonies.7 The normal haematopoietic stem cell (HSC), whose nature has still not been fully characterized, was first shown according to its ability to reconstitute the bone marrow of lethally irradiated mice. Mirroring this work, experiments using severe combined immunodeficient (SCID) mice indicated that normal haematopoietic stem cells become engrafted in the bone marrow of these immunodeficient animals (which lack B and T lymphocytes) such that all mature lineages, other than T-cells, were produced. Furthermore, transplantation of cells from human AML, chronic myeloid leukaemia (CML) and acute lymphoblastic leukaemia (ALL) could be performed in the same way and recapitulate the human cancer in the recipient mouse.8–11

However, the SCID mouse model had a number of limitations, which prevented definitive demonstration that the ‘SCID leukaemia-initiating cell’ (SL-IC) could recapitulate tumours; similarities between HSC and SL-IC cell-surface markers could only be shown for one leukaemia subtype.10 It was only after the development of the non-obese diabetic (NOD)/SCID mouse, which has additional immune response deficits (e.g. natural killer cell activity, complement activation), that the engraftment experiments were refined to allow Dick and colleagues to show properties in the SL-ICs which met the definition of CSC. One problem with the SCID model was the need to transplant large numbers of host cells to ensure engraftment in the recipient. It was shown that 10–20 times fewer cells were required in the NOD–SCID mice to achieve the same level of engraftment.12

The resulting experiments showed that, for a range of AML subtypes, both unfractionated bone marrow samples and purified CD34+ CD38 cells could reproduce the phenotype of the original human tumour in the recipient animal. This was seen even when the CD34+ purified fraction represented a tiny (0.2%) proportion of blast cells. The heterogeneous human-derived blast population, with CD38 expressed in most, even when transplanted cells had been purified according to the absence of this cell-surface marker, seemed to evoke normal haematopoietic differentiation. Cells could be further transplanted into a secondary recipient, recapitulating the tumour; it was calculated that the SL-ICs population must have expanded 30-fold, providing further evidence for self-renewal properties. These findings were compelling evidence for AML being a stem cell disorder.12

CSCs and solid cancers

The CSC theory has only been applied to solid cancers in recent years, although similar studies had been performed to show that cells of solid tumours are phenotypically heterogeneous and only a small proportion of cells are clonogenic in culture and in vivo.13–17 For example, only 1 in 1000 to 1 in 5000 lung cancer, ovarian cancer and neuroblastoma cells were found to form colonies in soft agar in studies performed in the 1970s. Despite this, the extension of the CSC hypothesis to solid tumours has been more challenging experimentally. Normal haematopoietic differentiation is better understood than the corresponding process in most solid tissues – importantly, cell-surface markers for normal haematopoietic stem cells and their progeny have been identified, allowing their isolation. For many solid tissue stem cells, assays have yet to be developed. A further problem is the physical nature of the tissues. Cells from solid tissues are often larger and more fragile than blood cells, cells are often less accessible for sampling, and creation of viable single-cell suspensions is challenging.

In 1992, Reynolds showed that a <1% subpopulation of embryonic striatal cells were viable after 5 days in vitro when plated in serum-free, low-density culture in the presence of the mitogen epidermal growth factor (EGF). These EGF-responsive striatal progenitors initially divided to form clusters of cells (neurospheres), with most of the constituent cells showing immunoreactivity for nestin (a cytoskeletal intermediate filament protein expressed in neuroepithelial stem cells). After 14 divisions, the clusters had continued to expand, and now two distinct populations of daughter cells stained positive for neurone-specific enolase and glial fibrillary acidic protein, markers for the neuronal and astrocyte lineages, respectively. Thus, the assay conditions and presence of EGF seemed to select for multipotent stem cells, defined by their capacity for self-renewal and multilineage differentiation.18

Experiments using modifications of this system, some exerting additional selection pressure by suspending cells in methylcellulose in low-adherence culture wells have been widely used to test the CSC hypothesis in solid tumours.19,20 Experiments inducing over-expression of the proto-oncogene Myc, Ras, Akt and platelet-derived growth factor (PDGF) showed that neural progenitors underwent transformation (increased proliferation21 and tumour formation22,23) more readily than more differentiated cell types.

Subsequently, reports were published of the isolation of CNS CSC from a variety of human cancers including astrocytomas, glioblastomas, medulloblastomas and ependymomas, by virtue of the ability of the cells to form neurosphere clones in serum-free culture in the presence of pleiotropic growth factors, and to differentiate into cells phenotypically similar to the lineages seen in the initial tumour.19,24–26

The foetal neuronal stem cell marker CD133 was widely expressed on these multipotent tumour cells.24,25 Karyotypic abnormalities and aberrant differentiation profiles indicated that they were not normal neural stem cells migrating within the tumour,25 but part of the tumour cell population. The in vivo ability of putative tumour stem cells to recapitulate the primary tumour mass was tested, by implanting cells isolated by sorting for CD13326 or growth in serum-starved, clonal density conditions27 into immunosuppressed mice, subcutaneously and/or intracranially. The resulting xenografts showed striking similarities to the tumours from which the progenitors were derived26,27 albeit with a ‘peculiar histomorphology’ in some cases.27

Dontu et al. in 2003 reported a low-adherence, clonal density serum-free culture system, which could be used to propagate ‘mammospheres’, enriched in mammary stem/progenitor cells, in vitro from normal human mammary tissue. As with the neurosphere assay, most primary mammary epithelial cells died under these conditions, but a few generated colonies of cells capable of self-renewal (passage) and differentiation into the three cellular lineages seen in adult mammary tissue.28

Mammosphere initiating cells were contained within the side population (SP) of cells capable of excluding Hoechst 33342 dye and showed upregulation of genes coding for membrane and cytoskeletal proteins, transcription factors, signalling and cell adhesion molecules, cell cycle regulators and metalloproteinases when compared with cells grown in differentiating conditions.28 The mammosphere system was used to show that activation of the Notch signalling pathway (Notch has been implicated as a proto-oncogene) affected lineage specification, and promoted self-renewal and proliferation of mammosphere cells. Conversely, alterations in signalling did not have any significant effect on fully committed mammary epithelial cells.29

Al-Hajj et al. used flow cytometry to separate cells from human primary and metastatic breast carcinomas according to the expression of cell-surface markers. In some cases, tumour cells were derived directly from patients and in others they had undergone one or two passages in mice. In all but one tumour, it was shown that CD44+ CD24−/lowLineage cells required initial cell inoculums of 2–10% of those required for unsorted cells to form tumours in immunosuppressed mice. CD24+Lineage cells were unable to form tumours except in one subject. The CD44+ CD24−/lowLineage population showed similar cell cycle distribution to the nontumorigenic cells, excluding this as the cause of tumorigenicity. CD44+ CD24−/lowLineage cells could be serially passaged in mice, forming tumours from which further CD44+ CD24−/lowLineage cells could be isolated (i.e. self-renewal capacity) as well as the other nontumorigenic cell populations found in the original tumour (i.e. multilineage differentiation).30

It is interesting, given the use of metastatic breast cancer cells (pleural effusion) in eight out of the nine tumour types investigated, that in another study the prevalence of the putative CD44+ CD24−/low tumour stem cell phenotype in breast tumours did not correlate with tumour progression or prognosis. A greater prevalence was associated, however, with a tendency for distant metastasis upon recurrence.31 The presence of disseminated tumour cells in breast cancer can be detected with immunohistochemistry for cytokeratins (CK). Balic et al. assessed CK+ bone marrow samples from 50 patients with early breast cancer and found that all specimens had detectable CD44+ CD24−/low cells, with prevalence (33–100%, median 65%) much greater than that seen in primary tumour masses, again associating the putative breast CSC population with a tendency towards metastasis.32

Gibbs et al. describe ‘sarcospheres’ created in low-adherence, serum-free culture from a variety of osteosarcomas and chondrosarcomas20– these phenotypically distinct tumours occur primarily in childhood and adulthood, respectively, yet the expression patterns and behaviour of the putative bone sarcoma stem cells were very similar. Self-renewal properties and multilineage gene expression were demonstrated, along with expression of the proteins Oct 3/4 and Nanog, associated with pluripotency and self-renewal in embryonic stem cells, and shared attributes with normal mesenchymal stem cells.20

Seaberg and van der Kooy showed the significance of serial passaging in the identification of stem cells.33 Early progenitor cells have limited self-renewing capabilities; however, they are able to form secondary and even tertiary spheres in non-adherent serum-starved-media conditions typically reserved for isolation of stem cell populations. It is important to differentiate these early progenitor cells from true CSC by serially passaging them many times. CSCs should, theoretically, be capable of indefinite self-renewal under these conditions with very little change in their ability to self renew as well as form differentiated progeny.34

Two groups of investigators have found that the minority CD133+-cell population from human colon carcinomas can reproduce the primary tumour in immunusuppressed mice, whereas CD133 cells do not have this capacity. Dick and colleagues showed that all human colon cancer-initiating cells (CC-IC) were CD133+, and that while 1 in 57 000 unsorted tumour cells were CC-IC, this was enriched to one in less than 300 CD133+ cells.35 Unlimited self-renewal potential and capacity for differentiation into all cell types seen in the primary tumour was established in vivo with serial transplantation in mice and also in vitro with a serum-free sphere culture system.35,36

CSC regulators and markers


Many molecular mechanisms have been identified in association with stem-like behaviour in cancer cells. Often these mirror the differential expression patterns that mark out normal tissue stem cells. The SP phenotype was first identified in haematopoietic stem cells, when a subset of cells with low uptake of the dye Hoechst 33342 was isolated by fluorescence-activated cell sorting (FACS) and found to be enriched for HSCs.37 The dye–efflux SP phenotype for haematopoietic and several other tissue progenitor cells has been shown both in vitro and in vivo to be conferred by the ATP-binding cassette (ABC) transporter ABCG2,38,39 which has also been identified for its role in multiple drug resistance,40 and shown to confer to breast cancer cells the ability to efflux chemotherapeutic drugs.41

This would tie in with the high frequency of cancer relapse following initial remission after chemotherapy, and a SP was detected in approximately 30% of cultured tumour cell lines from a variety of tissues including breast, colon, ovary and glioma. SP cells from breast, prostate and brain cancers were shown to be more tumorigenic than non-SP cells. However, when sorted according to ABCG2 expression, both ABCG2+ and ABCG2 cells were tumorigenic to similar degrees. It was postulated that ABCG2+ cells are tumour progenitors with more rapid turnover but that they themselves arise from more primitive, slow-cycling cells within the ABCG2 population with more long-term self-renewal capacity.42

Signalling pathways

Mutations in many of the signalling pathways and genetic mechanisms regulating normal stem cells have been shown in human cancers. Polycomb genes, particularly Bmi-1, HOX transcription factors, and the Wnt–β-catenin, Notch and Sonic Hedgehog pathways, are important in self-renewal and other functional stem cell properties; alterations in expression have been implicated in tumours of both blood and solid tissues.8,43–45


Bmi-1 is a member of the Polycomb group of transcriptional repressor proteins, which acts through the ink4a locus to downregulate the tumour suppressors encoded there – p16INK4a and p19ARF (Fig. 3). Bmi-1 over-expression in mouse embryonic fibroblasts leads to their immortalization, and in co-operation with ras can cause neoplastic transformation.46 Correspondingly, under-expression is associated with reduced proliferative capacity in both normal haematopoietic precursors derived from foetal liver cells and leukaemic stem cells in a mouse model of AML.47 Bmi-1−/− leukaemias were not transplantable into secondary recipients, although this capacity could be rescued by introduction of a retroviral Bmi-1 provirus. Interestingly, this Bmi-1-mediated rescue was also seen in Bmi-1−/− clones with defects in the expression of p16INK4a and p19ARF, indicating additional pathways through which the molecule exerts its effects.47 Glinsky et al. investigated the role of Bmi-1 in human prostate cancer using microarray analysis; elevation in expression was seen in all cancer cell lines, with greater increases seen in more metastatic tumour types. An 11-gene signature associated with Bmi-1 function in normal stem cells was found to be expressed in 11 different types of cancer, and consistently to predict metastasis and poor prognosis.48

Figure 3.

Bmi-1 signalling: Bmi-1 inhibits the transcription of two cyclin-dependent-kinase inhibitors, INK4A (p16) and ARF (p14). If INK4A is blocked, Retinoblastoma Protein (RP) becomes phosphorylated and inactivated by a complex of Cyclin-dependent kinase 4 (CDK4) and Cyclin D. This allows cells to enter the cell cycle. If ARF is blocked, MDM2 inhibits p53-dependent expression of genes that cause apoptosis.


The four Notch transmembrane receptors found in mammals are activated by their ligands Delta, Jagged and other members of the DSL (Delta, Serrate and Lag-3) family. Binding initiates a signalling pathway, which leads to the activation of the CSL transcription factor, along with mastermind-like (MAML) co-activators. This leads to transcription of genes associated with processes such as cell fate determination during development and self-renewal in adult tissues.49

Notch signalling has been shown to be oncogenic in mouse models of T-cell acute lymphoblastic leukaemia (T-ALL),50 and can collaborate with the c-neu/erbB2 oncogene in the development of mammary tumours.51 Dontu et al. showed that activation of Notch signalling promoted self-renewal and proliferation of normal mammary stem/progenitor cells cultured in mammospheres, but had no effect on fully committed mammary epithelial cells, suggesting that it exerts its oncogenic potential at the progenitor level.29 Interestingly, loss-of-function mutations have been shown to contribute to neoplastic transformation,52 showing that notch-activated gene expression has varied roles depending on the context.49

This may account for the differential expression of the Notch ligands, Delta and Jagged-1 and 2 between neurosphere clones derived from normal brain and glial tumour tissue. Delta expressed by normal tissue and neurosphere clones which had been allowed to attach to a substrate, but not suspended neurospheres; Jagged-2 was expressed by normal clones but not those originating from tumours.19 Conversely, recent microarray analysis of CD34+ CD38 leukaemic stem cells (LSC) from AML has indicated over-expression of Jagged-2, with inhibition of Jagged and Notch signalling reducing LSC growth in colony-forming assays.50

Wnt and β-catenin

Wnt signalling influences cell migration and developmental patterning, proliferation and survival, through the binding of β-catenin to the LEF/Tcf transcription factors and activation of downstream genes (Fig. 4). The binding of Wnt proteins to their Frizzled cell-surface receptors inhibits the cytoplasmic ‘destruction complex’ in which β-catenin is normally held, allowing it to accumulate in the cytoplasm and translocate to the nucleus. Over-expression of β-catenin in transgenic murine HSCs increased their self-renewal capacity; over time, while controls appeared to differentiate down myelo-monocytic lineage, an increased proportion of the β-catenin-transduced HSC population remained as lineage negative, proliferative cells. Inhibition of Wnt-signalling suppressed HSC growth between four- and seven-fold, suggesting that the pathway is required for normal HSC function.51 It remains possible, however, that this is an in vitro phenomenon and that in vivo, alternative mechanisms are involved and Wnt is less significant.44

Figure 4.

Wnt binds to Frizzled (FRZ) receptors and activates Dishevelled (DSH). This disrupts a β-catenin destruction complex and allows β-catenin to accumulate and translocate to the nucleus. In the nucleus β-catenin binds to LEF/Tcf transcription factors leading to expression of target genes (e.g. c-MYC) that promote survival, proliferation and cellular migration.

Wnt signalling is also involved in self-renewal of epithelial cells in other tissues [e.g. skin, intestine and central nervous system (CNS)]; over-activation of the pathway and increased nuclear β-catenin has been implicated in colon, prostate, ovarian, CNS and skin tumours as well as haematological malignancies.44 The Adenomatous Polyposis Coli (APC) tumour suppressor gene mutated in familial adenomatous polyposis is part of the β-catenin destruction complex inhibited by Wnt signalling. It has been proposed that both germline and somatic mutations of APC confers a selective advantage to a cell, with increased proliferation in response to the Wnt-pathway dysregulation, but that APC function must not be so impaired so as to lead to apoptosis. Evidence exists for both the ‘top down’ model of colon cancer, whereby somatic APC mutation occurs first at the luminal surface and spreads down into the crypts, and the ‘bottom up’ model, where the mutation is propagated in, and spreads from the stem cell located in the crypt base – obviously the latter being more easily reconciled with the CSC hypothesis.52

Telomeres and telomerase

When grown in tissue culture, the phenomenon of senescence limits the replicative potential of cells, in that eventually the cells will stop dividing.53 This is partly governed by the gradual loss of the protective telomeres, tandem repeats of a 6-bp sequence, which are present at the ends of chromosomes. DNA polymerises cannot fully replicate the 3' end of the DNA strand, such that upon each cell division there is a loss of 50–100 bp at each end of every chromosome. The telomeres buffer this loss but are of finite length, so they gradually become shorter through the life of a dividing cell – this protective mechanism means that the cell will be directed to senescence or apoptosis after a certain number of divisions.54,55

Stem cells circumvent this ‘end replication problem’ with the expression of low levels of Telomerase, an enzyme which catalyses the addition of telomeric repeat sequences on to the ends of chromosomes, such that they have lifelong replicative potential. Telomerase is also expressed in over 80% of human cancers, and has been presumed to confer unlimited cell cycling ability. Conversely, many tumours show foreshortened telomeres. A two-step process has been proposed, whereby early telomere shortening promotes chromosomal instability and mutation, and then telomerase activity stabilizes the telomeres and allows uncontrolled replication. In most haematological malignancies, short telomeres and telomerase activity are detected; telomerase activity levels of between 2 and 50 times that of normal haematopoietic precursors has been shown.55

Protein expression

The protein products of normal stem cells have been seen to be differentially expressed within tumours, with populations enriched for putative CSC showing higher levels of expression. Nestin is an intermediate filament protein and a marker for neuroepithelial precursors56; in undifferentiated clonal neurospheres from both normal brain18 and brain tumour tissue24,25 increased immunoreactivity is seen. Nanog and Oct 3/4 are homeoproteins involved in self-renewal and pluripotency of embryonic stem cells, with immunoreactivity demonstrable on tumour sections. Sarcospheres grown from osteosarcomas and chondrosarcomas show preferential expression of these proteins; the proportion of positively staining cells decreases as the spheres grow out and cells begin to differentiate.20

Which cell is the target for transformation?

Despite the evidence supporting the theory that cancer is initiated and propagated by cells with stem-like characteristics, it remains unclear whether the CSC is a normal tissue stem cell, which has undergone malignant transformation, or a more differentiated cell, which has acquired more primitive, stem-like characteristics as a result of mutation or dedifferentiation.

As discussed, many of the attributes of normal stem cells make them attractive candidates for malignant transformation into CSC – they are programmed for self-renewal and multilineage differentiation; they persist and continue to divide for the lifetime of the host, allowing them more opportunity to accrue transforming mutations; isolated tumour-initiating cells show many phenotypic similarities to the corresponding normal tissue stem cell (e.g. cell-surface markers, protein expression and telomerase activity).

At the same time, pluripotent stem cells represent a small minority of cells within a tissue. An individual stem cell would be a very small target population for the accumulation of sufficient mutations to confer a neoplastic phenotype.57 Also, mutations of pathways which control normal stem cell function, such as the Wnt–β-catenin pathway, are seen in many cancers57,58– this would perhaps be unexpected if normal stem cell processes had simply been co-opted to lead to tumorigenesis, rather than independent acquisition of such functions through mutation. It has been suggested that normal stem cells may be protected from the effects of mutations, but that these may manifest in their more downstream progeny.8

Certainly evidence exists that tumour-initiating cells may lie in a more committed cell population than the CSC. Both committed myeloid progenitors and haematopoietic stem cells produce a transplantable murine model of AML when transduced with a mixed lineage leukaemia fusion gene. Moreover, the phenotypic characteristics and stage of maturation arrest was identical in leukaemias derived from transplanted HSCs, common myeloid progenitors or granulocyte–macrophage progenitors (GMP).59 Analysis of the different cellular compartments in CML showed increased nuclear β-catenin in myeloid progenitors as the disease advanced to blast crisis, while levels in HSCs remained stable. Also in vitro self-renewal capacity was shown both by GMPs from leukaemic patients and normal GMPs with forced β-catenin expression.60 (It is of note that the cellular compartments in the leukaemic samples were defined according to the surface marker phenotypes of the normal haematopoietic hierarchy – this may be inappropriate, given that markers of normal differentiation would be expected to show derangement in leukaemia, particularly at blast crisis stage.)

Polyak and Hahn57 propose three models for development of malignancy involving stem cells: (1) a mutation causes dysregulation of asymmetric division in a CSC and is passed on to all progeny; progression to full transformation occurs in this population as further mutations are acquired; (2) the CSC itself acquires mutations sufficient for malignancy, and passes these on to all progeny; and (3) the transit-amplifying cells or more differentiated progeny accrue mutations leading to dedifferentiation and acquisition of stem-cell-like properties; CSCs themselves are not involved.

Although the dedifferentiation of a committed cell into one with more primitive, stem-like properties seems an unlikely event, it has been shown in Drosophila melanogaster that cells differentiating in four- to eight-cell cysts during early development can be induced to revert to germinal stem cells.61 Autocrine signalling of PDGF has been implicated for its role in the development of human gliomas. Over-expression in not only neural progenitors but also (albeit with less efficiency) differentiated astrocytes in vitro leads to increased proliferation, and in vivo can induce tumour formation in mice; this is enhanced by loss-of-function of the Ink4a locus.23

It has been proposed that perhaps, rather than unlimited self-renewal capacity being conferred on CSC by gain-of-function mutations, that it is in fact a ‘default’ pathway (seen, for example, in most single-celled organisms). If tissue specialization relies on a balance between self-renewal, differentiation and cell death, then could any cell in which apoptosis is prevented or differentiation is blocked (the effects of many of the mutations seen in cancer) act as a self-renewing CSC?62

Recently, interesting findings have suggested that there exists an even more primitive population of precursors within the bone marrow, the putative ‘haemangioblast’. One hallmark feature of CML, present in 95% of patients, is the Philadelphia chromosome – this truncated chromosome 22 results from a reciprocal translocation with chromosome 9, and produces the BCR–ABL fusion protein. The protein is found in multiple haematopoietic lineages in CML, suggesting that the translocation event arises in an HSC; however, demonstration of BCR–ABL expression in endothelial cells from a patient with CML may point to the mutational event having occurred in a cell preceding the HSC in the haematological hierarchy, with greater differentiation potential.63 Certainly, as more is understood about the stem cell hierarchies present in normal tissues, understanding of the role of stem or stem-like cells in cancer will be enhanced.

The importance of the niche

The potent effects imparted by the environment in which a cell exists, or ‘niche’, cannot be disregarded when considering the evolution of a tumour.57,64,65 The tumour interacts with its surrounding stroma in a reciprocal manner similar to the communications between cells in normal tissues, and can influence the stroma such that it is more conducive to tumour growth. Indeed, epithelial tumour formation can be enhanced by inducing mutations in cleared (epithelial cell-free) stromal environments, and then introducing normal untreated epithelial cells.64

Normal bone marrow stem cells possess the capacity to mobilize and migrate in the circulation to distant sites in response to tissue damage and stress; complex, co-ordinated homing mechanisms are involved.66 Stem cells from different tissues share common genetic programs,67 and bone marrow stem cells display plasticity allowing them to differentiate into a variety of cell types.65,68 There are many similarities between the mechanisms governing the migration of normal stem cells and the metastatic dissemination of tumour cells, such as the interaction between the cell-surface CXCR4 receptor and its ligand, stromal-derived factor 1, secreted by the niche.69

Cancer cells may disseminate to distant sites; however, while this may lead to the formation of metastases (whose cellular heterogeneity frequently resemble that of the primary tumour70), some of these migrating cells never develop further but remain detectable in remote tissues.6 It has been proposed that it is the migration of the CSC population which allows the expansion of the tumour mass – manifesting as local invasion over short distances and metastatic spread for longer migrations.70

The cell fusion hypothesis

An alternative mechanism proposed for the apparent plasticity shown by stem cells is cell fusion. Fusion of bone marrow stem cells with a variety of different cell types has been shown in vivo and in vitro. If this hypothesis is extended to cancer, the fusion of a stem cell with a somatic cell carrying mutations could result in a cell with genetic and karyotypic abnormalities, which has the properties of a CSC.65 The case for cell fusion remains equivocal. It has been best shown in models involving severe tissue injury; it remains to be established whether it occurs in diploid, adult cells in vivo in the absence of cell damage.71

Implications for cancer therapy

If a population of CSC is responsible for the propagation of a tumour, then these must be eliminated to effect a cure (Fig. 5). This may not be achieved by conventional strategies, which target rapidly dividing cells – stem cells may enter periods of quiescence during which they will be resistant to strategies aimed at eradicating cycling cells. This is evident in the treatment of acute leukaemias, in which total bone marrow ablation followed by reconstitution is required in a significant proportion of patients.71 Guzman et al. report the use of MG-132, a proteasome inhibitor, and idarubicin, an anthracycline, in preferential targeting of leukaemic stem cells in in vitro and in vivo models of AML – the cells are driven to apoptosis, but normal HSCs are spared.72

Figure 5.

Currently conventional cancer therapies are directed at non-stem cell populations. Consequently, any tumour has the capacity to regrow and recapitulate the original cancer. Therapies targeted at the cancer stem cell and/or the daughter cells would have the capacity to cause either involution or complete tumour eradication.

Other treatment modalities might also preferentially target putative CSC. Transiently inactivating the causative oncogene in transgenic mouse model of osteogenic sarcomas can cause the tumours to regress, with reactivation leading to apoptosis.73 It would seem logical that forcing CSC down a symmetrical division pathway, whereby two more committed daughter cells are produced, would deprive a tumour of its self-renewal potential and effect a cure. Differentiation therapy with retinoids is effective in a majority of cases of human acute promyelocytic leukaemia, although responses in other malignancies have been variable.71

Treatments directed solely at CSCs, however, may not have an effect on the differentiated progenitor and daughter cells. Therefore, the bulk of the tumour may remain intact while the CSCs are being destroyed. Jones et al. describes this theory as the dandelion phenomenon.74 This theory states that cutting off a dandelion at the roots, or treating the bulk of the tumour, takes away the disease that one can see, however, the weed will still regrow because the root has not been destroyed. Conversely, destroying the root, or CSC, of the weed leaves the flower above soil initially. However, as the root dies, the portion above ground will eventually wither and die without the root.

According to this dandelion phenomenon of CSC therapy, a treatment regimen may be abandoned too quickly if the only judge of a response to therapy is a remission of the bulk of the tumour. This response is likely to lag behind the destruction of CSC. Conversely, treatment of chronic phase CML with imatinib mesylate, a tyrosine kinase inhibitor, has achieved high rates of complete remission despite the fact that BCR–ABL-expressing progenitors are spared.75 This may indicate the significance of the niche provided to a stem cell by its surrounding daughter cell population – removal of paracrine factors, in targeting the tumour bulk, may sometimes be sufficient to arrest the progress of disease.57

Given that clinical response may not be an ideal way to monitor for response to therapies that target CSC, overall survival is left as common monitoring criteria. This requires long study times that can be impractical. Animal models are an ideal way to continue to use survival as criteria for determining effectiveness of therapy. Animal models of spontaneously occurring tumours often progress much faster than the human form of the disease. Many canine tumour models are very similar or even identical to their human counterparts,76,77 and the natural canine model may represent an ideal testing ground for novel compounds directed at the CSC.