Stem cells in cancer: instigators and propagators?

There is growing realization that many – if not all – cancer-cell populations contain a subpopulation of self-renewing stem cells known as cancer stem cells (CSCs). Unlike normal adult stem cells that remain constant in number, CSCs can increase in number as tumours grow, and give rise to progeny that can be both locally invasive and colonise distant sites – the two hallmarks of malignancy. Immunodeficient mouse models in which human tumours can be xenografted provide persuasive evidence that CSCs are present in human leukaemias and many types of solid tumour. In addition, many studies have found similar subpopulations in mouse tumours that show enhanced tumorigenic properties when they are transplanted into histocompatible mice. In this Commentary, we refer to CSCs as tumour-propagating cells (TPCs), a term that reflects the assays that are currently employed to identify them. We first discuss evidence that cancer can originate from normal stem cells or closely related descendants. We then outline the attributes of TPCs and review studies in which they have been identified in various cancers. Finally, we discuss the implications of these findings for successful cancer therapies.


Introduction
Cancer begins in normal somatic cells with a mutation that is fixed by a round of cell proliferation -a process termed initiation. If the mutated cell has a selective growth advantage over its normal cell neighbours owing to, for example, enhanced proliferation or resistance to apoptosis, then a clone of cells carrying the same mutation will emerge; in the colon this would manifest as a small adenoma. Further tumour progression can be viewed as an evolutionary process: if a second mutation in one cell of the nascent cell clone gives that particular cell a further growth advantage, then a new clone will outgrow the preceding clone; in the colon, this would manifest as a larger adenoma with more cellular irregularity (dysplasia). Further rounds of mutation and clonal expansion give rise to progressively more abnormal clones until an additional mutation produces the malignant cell phenotype that can invade beyond normal tissue boundaries and, subsequently, spread to distant sites -a phenomenon known as metastasis.
We now believe that most tumours contain a subpopulation of malignant cells with stem cell properties. These stem cells can give rise to new tumours when xenografted, usually in small numbers, into immunodeficient mice -hence, we refer to them as tumourpropagating cells (TPCs). However, it is currently unclear whether TPCs arise directly from the transformation of normal stem cells or whether they are derived from aberrant de-differentiation of more mature neoplastic cells at a later stage of cancer progression.
Microscopic inspection of most tumours reveals a complex heterogeneous picture. Tumours have a hierarchical nature, in part because TPCs can give rise to more TPCs as well as to transitamplifying cells (TACs) and terminally differentiated (TD) cells (Fig. 1). However, phenotypic and behavioural heterogeneity is also generated by variation within different parts of a tumour with respect to their proximity to the vascular network, and because not all of the cells are cancer cells. Non-cancer cells within a tumour can include inflammatory cells, cancer-associated fibroblasts and immature myeloid cells, all of which influence tumour behaviour and often facilitate invasion and metastasis (Coussens and Werb, 2002;Alison et al., 2009b). The microenvironment surrounding blood vessels is conducive to the highest rates of tumour-cell proliferation, but it can also serve as the local stem cell niche, a specialised microenvironment of nerves, mesenchymal cells and extracellular matrix molecules that regulate aspects of stem cell behaviour, particularly the choice between asymmetric division (which underlies self-renewal) and symmetric division (which leads to an increase in the number of stem cells) (Alison and Islam, 2009).
In this Commentary, we discuss various lines of evidence that illustrate how stem cells are central to cancer biology. First, mutation in the DNA of normal stem cells appears to be the initiating event in many (although not all) types of cancer. Second, cancers themselves have stem cells (referred to here as TPCs). Third, TPCs can be prospectively isolated based on the expression of specific surface markers or intracellular enzyme activity. Fourth, TPCs can be remarkably resistant to therapies that kill the majority of cancer cells. Fifth, there are several new and, hopefully, moreeffective therapies currently being designed to target the unique attributes of these cells.

Normal stem and progenitor cells and the origins of cancer
It is widely believed that tumorigenesis commonly begins in normal adult stem cells or progenitor cells that have recently descended from them. This concept is supported first by the fact that stem cells have -or can re-acquire -the self-renewal mechanisms needed for maintaining and expanding stem cell numbers. Second, stem cells are anchored to the niche and are therefore not swept away in the prevailing cell flux, allowing time to acquire further mutations (Fig. 1); this is particularly pertinent to cancers of continually renewing populations, such as cells of the haematopoietic system, gut and skin.

Small intestine
In the mouse small intestine there is both a slowly cycling stem cell population found ~4-5 cell positions above the base of the crypt (Sangiorgi and Capecchi, 2008), and a rapidly cycling stem cell population composed of slender-shaped cells [so-called crypt base columnar cells (CBCCs)] that are sandwiched between Paneth cells at the base of the crypt (Barker et al., 2007). The stem cell characteristics of the two compartments have been established using elegant genetic-lineage tracking techniques. The former cell population expresses Bmi1, a component of the polycombrepressing complex 1 (PRC1) that prevents stem cell senescence; this Bmi1-expressing population can give rise to clones that contain all intestinal lineages. CBCCs, by contrast, express the Wnt target gene Lgr5, which encodes an orphan G-protein-coupled receptor (Barker and Clevers, 2010); these Lgr5-expressing cells are also multipotential stem cells (Barker et al., 2007). Targeted deletion of the tumour suppressor gene Apc (adenomatous polyposis coli) in Lgr5-expressing CBCCs resulted in rapidly growing adenomas, whereas targeted deletion of Apc in the higher-positioned TACs that are the direct descendants of either stem cell population failed to induce substantial adenoma growth (Barker et al., 2009). These data illustrate that a mutation in a stem cell population is most effective for tumour initiation in the short term. However, small adenomas derived from TACs persisted for nearly a year in this model, so the possibility that cancer later arises from these adenomas cannot be excluded, particularly given the fact that human colorectal cancers grow very slowly (Camplejohn et al., 1973). Intriguingly, 36 days after induction of the stem-cell-derived adenomas, both colonic and small intestinal adenomas contained 6.5% Lgr5-positive (Lgr5 + ) cells; it would be interesting to know whether these cells become TPCs. The expression of Lgr5 in mice can occur in the same cells that express prominin 1 (PROM1, also known as CD133, a family member of glycosylated pentaspan membrane proteins; Box 1). Activation of Wnt signalling specifically in these CD133-positive (CD133 + ) cells demonstrated that all neoplastic tissue came from CD133 + cells (Zhu et al., 2009). Remarkably, the frequency of CD133 + cells in the adenomas (7%) was the same as the frequency of Lgr5 + cells in the adenomas generated in the study in which Apc was deleted from the Lgr5expressing population. Intestinal tumours may also be initiated from the stem cell population at positions 4-5 above the base of the crypt. These cells overexpress Wip1 phosphatase, an enzyme that turns off the DNA-damage response through inactivation of stress-induced kinases such as Chk2. Apc Min/+ mice are a model of familial adenomatous polyposis; the absence of Wip1 in Apc Min/+ Wip1-null mice severely reduces adenoma burden through p53dependent apoptosis of damaged stem cells because the DNAdamage response is extended (Demidov et al., 2007). The absence of the hyaluronan receptor CD44 also reduces adenoma burden in Apc Min/+ mice, probably because of increased apoptosis of initiated cells. As CD44 is expressed only by CBCCs and early TACs in intestinal tissues, this further supports the idea that these cells can initiate adenomas (Zeilstra et al., 2008).

Prostate
It was previously thought that all tumours in the human prostate gland arose from androgen receptor (AR)-negative (AR -) basal stem cells (Maitland and Collins, 2008). Indeed, androgen inhibition does not reduce the clonogenicity of prostatic cancer cell lines but does reduce the size of individual colonies, consistent with the finding that androgens are unnecessary for the maintenance of stem cells but required for proliferation of TACs (Bisson and Prowse, 2009). However, in some human prostate cancer cell lines, CD133 + TPCs are also AR-positive (AR + ), suggesting that they originated from differentiated luminal cells (Vander-Griend et al., 2008). Likewise, in mice, a rare population (<1% of total) of castration-resistant luminal cells that expresses the transcription factor Nkx3-1 (a negative regulator of prostatic epithelial proliferation) was found to have self-renewing and bipotential properties. Furthermore, targeted deletion of the phosphatase Pten 2358 Journal of Cell Science 123 (14)

Box 1. Defining markers of human TPCs ALDH
The ALDH gene superfamily encodes detoxifying enzymes; high expression of ALDH can be detrimental to tumour eradication (Dylla et al., 2008).

CD24
A sialoglycoprotein that acts as a ligand for P-selectin. It enables cells to bind to platelets -tumour-platelet thrombi protect cells in the bloodstream and in turn facilitate tumour invasion through interacting with endothelia.

CD44
A transmembrane glycoprotein that binds hyaluronan (HA); often expressed in various isoforms.

CD105
A type I integral membrane protein.

CD117
Also known as c-kit; a tyrosine kinase receptor for SCF.

EpCAM
A glycosylated type I integral membrane protein expressed by many tumour cells (Munz et al., 2009). Involved in components of Wnt signalling that stimulate cell proliferation after cleavage of EpCAM intracellular domain.

SP
Side population; defines cells that are fluorescent-dull by flowcytometric analysis. SP cells efflux fluorescent dyes, such as Hoechst 33342 and DyeCycle Violet, which is a phenotype that usually depends on expression of the ABC superfamily of membrane transporters. The 49 human ABC transporters are organized into seven families (A-G; http://www.nutrigene.4t.com/humanabc.htm). In particular, ABCG2 (also known as breast cancer resistance protein 1, BCRP1) is expressed in many stem cells and is upregulated in hypoxic environments (such as stem cell niches) (Teramura et al., 2008).
(phosphatase and tensin homologue; a known tumour suppressor in several contexts) in these cells after androgen-stimulated regeneration of the epithelium promoted the rapid development of prostatic intraepithelial neoplasia (PIN) and microinvasion . Thus, in the prostate, there might be two distinct cell types from which tumours arise -basal and luminal. In particular, the indication that these cancer-initiating cells are of luminal origin is consistent with the absence of basal cells in human prostatic adenocarcinoma.

Mammary and gonadal tissues
There might also be more than one cell type in breast tissue from which cancer develops. Many breast cancers are believed to originate from undifferentiated oestrogen-receptor-negative (ER -) multipotential basal stem cells, the numbers of which can be hugely expanded by exposure to progesterone (Joshi et al., 2010), perhaps explaining why more reproductive cycles (early menarche, late menopause) are a strong risk factor for breast cancer. However, in basal-like breast cancers, gene-expression profiling suggests that tumours originate from luminal progenitors. Moreover, luminal progenitors are expanded in pre-neoplastic tissue from carriers of mutations in the breast cancer 1 gene (BRCA1) (Lim et al., 2009). Two distinct cell types (one enriched for myoepithelial transcripts) that were differentiated from human breast epithelium and transfected with the same defined genetic elements produced very different tumour types (Ince et al., 2007). This observation suggests that clinical differences between the various subtypes of breast cancer reflect different populations of founder cells. In the testis, pluripotent germ-cell tumours (teratomas) probably arise from maturation-arrested gonocytes (foetal germ cells), which are phenotypically similar to the probable precursor lesion of teratomas, known as carcinoma in situ (CIS) (Kristensen et al., 2008); the expression of stem cell factor (SCF) detects CIS cells (Stoop et al., 2008).

Liver
Two distinct cell populations can give rise to hepatocellular carcinoma (HCC). Because oncogenic transgenes can be driven by albumin promoters, hepatocytes are implicated in many animal models of HCC. However, hepatic progenitor cells (HPCs) that arise from a potential stem cell compartment in the small biliary ducts proliferate in response to chronic liver damage such as cirrhosis, which usually precedes HCC. Therefore, HPCs might be particularly susceptible to mutation (Alison et al., 2009a), and maturation arrest of HPCs might contribute to tumorigenesis in the liver. Four prognostic subtypes of HCC have been identified that correspond to a hierarchy of liver-cell lineages (Yamashita et al., 2008). Patients with tumours comprising a sizeable proportion of either foetal-like liver cells or HPCs had the poorest prognosis, whereas patients with tumours composed of cells resembling either mature hepatocytes or biliary cells had a more favourable clinical outcome. Moreover, gene-expression profiling identified a subset of HCCs with a poor prognosis and in which patient tumour-cell profiles suggested an HPC origin (Lee et al., 2006). Finally, identifying tumour cells on the basis of either high or low expression of the HPC marker cytokeratin 19 (CK19, also known as KRT19) can be used to identify an HCC patient group with a shorter time to recurrence -that is, high CK19 expression is predictive of local or distant recurrence after whole-liver resection and transplantation (Durnez et al., 2006).

Lung
The phenotypic diversity of lung tumours, which vary depending on their location, is probably due to region-specific variation in stem cells in the tracheobronchial tree (Alison et al., 2009c). The idea that pro-oncogenic stem cell niches exist in the lung is supported by mouse models in which tissue-wide knockdown of a tumour suppressor gene (e.g. p53) or upregulation of a protooncogene under the regulation of a widely expressed lung-specific promoter (in effect inducing a 'field cancerization' effect), does not induce a wide distribution of tumours. Instead, tumours are geographically linked to stem cell niches. For example, mutations in KRAS are common in human non-small-cell lung cancer (NSCLC) (Riely et al., 2009), but mouse models that have widespread Kras mutations only exhibit adenocarcinoma in the bronchioalveolar region (Jackson et al., 2001) that involves the expansion of self-renewing, multipotent bronchioalveolar stem cells at the bronchioalveolar duct junction (Kim et al., 2005). Furthermore, human small-cell lung cancers (SCLCs) arise from mid-level bronchioles where neuroendocrine precursors are located, whereas squamous-cell carcinomas (SCCs) generally occur in the proximal airways and appear to develop from cytokeratin-14positive basal cells found either in the submucosal gland ducts or intercartilagenous boundaries (Barth et al., 2000).

Haematopoietic cells
The cell of origin of most haematological malignancies is probably a multipotential haematopoietic stem cell (HSC). For example, in chronic myeloid leukaemia (CML), in which 95% of affected patients have the Philadelphia chromosome (which is likely to be the founder mutation), the BCR-ABL oncogenic fusion transcript can be found in otherwise normal mature blood cells that have differentiated from an HSC carrying this translocation (Cobaleda et al., 2000). However, leukaemias might develop from more committed progenitors that have reacquired the stem cell property of self-renewal (Passegue et al., 2003;Dalerba et al., 2007a). This idea is supported by the finding that transfection with oncogenic fusion genes such as MOZ-TIF2 can confer granulocyte-monocyte progenitors (GMPs) with leukaemic properties (Chan and Huntley, 2008). Other examples in which leukaemia is derived from progenitors include CML patients in blast crisis (the final phase in the evolution of leukaemia), in which GMPs can become self-renewing TPCs; this change is associated with elevated Wnt signalling in the GMPs (Jamieson et al., 2004). Likewise, in childhood acute lymphoblastic leukaemia, clonal T-cell receptor rearrangements have been reported to occur in CD133 + , CD19 + TPCs, suggesting dedifferentiation of a cell that has already gone through the first stages of differentiation (Cox et al., 2009). Recent studies in mice have also highlighted the importance of the niche for stem cell homeostasis, showing that stromal cell dysfunction can lead to leukaemogenesis (Raaijmakers et al., 2010).
Overall, experimental and clinical evidence support the idea that, in humans, the process of tumorigenesis begins in an adult stem cell, although other more committed cells, particularly in the haematopoietic system, might also be founder cells of malignancy. Whether the cells that are operationally referred to as TPCs (on the basis of characteristics determined in an in-vivo assay, see Fig. 2 and discussion below) are directly derived from asymmetric or symmetric division of mutated adult stem cells, or whether they evolve from further-differentiated cancer cells is presently unclear. In the following sections, we explore the characteristics of TPCs and highlight the unique characteristics that differentiate them from normal stem cells.

Functional assays for TPCs
TPCs are believed to self-renew and to give rise to a hierarchy of progenitor and differentiated cells, albeit in an unorthodox manner that gives rise to more TPCs through symmetric divisions (Boman et al., 2007;Cicalese et al., 2009) (Fig. 1). Operationally, TPCs are a population of cells that can be defined by the expression of specific molecules (see Box 1), and that are more tumorigenic than the bulk tumour population. An example of a 'tumorigenicity assay' for human TPCs that reveals their tumorigenic potential is the transplantation of human cells into either nude or non-obese diabetic/severe combined immunodeficient (NOD/SCID) mice, which lack major elements of the immune system and therefore do not reject the human cells (Fig. 2C). Such assays can be complemented by two types of ex vivo assay known as clonogenicity assays ( Fig. 2A,B); these assays also seem to be useful surrogates for in vivo TPC assays.
It was previously thought that TPCs were present in only very small numbers, as large numbers of human tumour cells had to be xenotransplanted into immunodeficient mice to generate tumours. However, this might be because the human cells in this assay are in a foreign microenvironment, as transplantation of mouse tumour cells into mice indicates that TPCs can be quite common in some tumours. For example, as few as ten mouse lymphoma or acute myeloid leukaemia (AML) cells can propagate tumours when they are transplanted into immunocompetent histocompatible mice (Kelly et al., 2007). Furthermore, using standard immunodeficient NOD/SCID mice, the frequency of TPCs in human melanoma was estimated to be 1 in 10 6 (Schatton et al., 2008), but even single human melanoma cells were found to form tumours in more highly immunocompromised mice [NOD/SCID mice that were also null for the interleukin-2 receptor -chain (IL2R)] (Quintana et al., 2008). Clearly, estimated frequencies of human TPCs in cancer crucially depend on the immune status of the recipient mouse.

Mouse models of cancer support the concept of TPCs
Several studies have reported subpopulations of TPCs in mouse models of cancer, overcoming the initial skepticism of experiments involving xenografting human cells into immunodeficient mice. In a mouse model of HCC (involving targeted deletion of Pten in hepatocytes), only a CD133 + CD45population was found to have tumour-propagating activity in nude mice (Rountree et al., 2009). In mammary tumours developing in p53-null mice, TPCs that were able to form tumours which expressed both myoepithelial and luminal markers had the phenotype of normal mammary-gland stem cells (CD29 high CD24 high ) (Zhang et al., 2008a), but their frequency was only 1:300, supporting the notion that TPCs can be a minority subpopulation. A subset of TPCs has also been identified in mouse medulloblastomas occurring in Ptc +/mutant mice (Read et al., 2009).

EMT can generate TPCs
Epithelial-to-mesenchymal transition (EMT) is a process by which the largely E-cadherin (CDH1)-dependent cell contacts between contiguous epithelial cells break down, resulting in epithelial cells becoming fibroblast-like and motile. For example, a main source of fibroblasts in renal fibrosis is through EMT of kidney epithelial cells, which is mediated by transforming growth factor 1 (TGF1).
Transcription factors such as Twist and Snail downregulate CDH1 expression; accordingly, it has been shown that, when Twist or Snail expression is upregulated in breast cancer epithelial cells, EMT results. Remarkably, the resulting mesenchymal-like cells acquire a breast TPC phenotype (CD44 + CD24 low ) (Mani et al., 2008). In ovarian cancer, transfection with Snail and Snail2 (also known as Slug) leads to de-repression of 'stemness' genes, including Nanog and KLF4, and four-to fivefold increases in the size of a CD44 + CD117 + TPC population that is more resistant to chemo-and radiotherapy (Kurrey et al., 2009).
Undifferentiated tumours that have features of EMT and contain large numbers of stem cells can also be induced in mouse mammary glands by overexpressing the developmentally regulated homeobox protein SIX1 (McCoy et al., 2009). SIX1 is more commonly expressed in human metastatic lesions compared with primary breast tumours. Another protein associated with aggressive breast cancer is YB-1 (also known as YBX1). YB-1 activates the translation of mRNAs that encode proteins such as Snail, Twist and ZEB1, and other transcription factors that coordinate EMT (Evdokimova et al., 2009). Although the rarity of TPCs has hampered efforts to identify drugs that selectively kill them, the acquisition of the breast TPC phenotype by transformed breast cancer epithelial cells (CD44 high CD24 low ) upon EMT has made 2360 Journal of Cell Science 123 (14)

Fig. 1. Tumour evolution and heterogeneity. TPCs probably originate from normal stem cells (S, yellow), can be derived from transit-amplifying cells (TACs, blue) but not terminally differentiated (TD) cells (green)
. TPCs can undergo self-renewal by asymmetric cell division and increase in number by symmetric divisions. TPCs might have a single identity (red outline), but further genetic and epigenetic changes can give rise to other TPCs with a different phenotype (TPC2, green outline). In addition, EMT might give rise to more TPCs. This hierarchical organization, which gives rise to intratumoral heterogeneity, might be supplemented by the clonal evolution of a population of cells with a selective growth advantage (TAC, blue/yellow) (Odoux et al., 2008;Shipitsin et al., 2007). high-throughput screening possible (Gupta et al., 2009). In this study, short hairpin RNA (shRNA)-mediated inhibition of CDH1 was used to generate large numbers of TPCs, and the antibiotic salinomycin was identified as a highly effective agent to reduce their viability, clonogenicity and tumorigenicity.
The development of metastasis might involve the dissemination of TPCs, in particular cells at the tumour margins that have undergone EMT (Brabletz et al., 2009). This process might be aided by the expression of chemokine receptors on TPCs, as observed in pancreatic cancer (Hermann et al., 2007). Overall, EMT in epithelial tumours appears to be an adverse prognostic factor and can be associated with acquisition of a TPC phenotype.

Are TPCs multipotential?
Are tumours heterogeneous because of distinct clones that arise from different TPCs, or are the TPCs in a particular tumour multipotential -similar to many normal stem cells? Clonal populations have been derived from patients with colorectal cancer (CRC) that recapitulated the heterogeneity of the original tumours when they were xenografted, exhibiting enterocytic, neuroendocrine and goblet-cell differentiation -all from a single cell (Kirkland, 1988;Odoux et al., 2008;Vermeulen et al., 2008). In glioblastoma multiforme (GBM), single TPCs can generate all three lineages (neural, astrocytic and oligodendrocytic; often in different proportions), underscoring TPC multipotentiality (Borovski et al., 2009). However, these observations do not preclude the possibility that there are also different TPCs in a tumour. In support of this idea, the centre and peripheral areas of human glioblastomas have been found to contain two populations of cytogenetically diverse TPCs: both were multipotential, but they had distinct tumorigenic potentials (Piccirillo et al., 2009b) (although they probably both derived from a common clone). Cytogenetically diverse clones have also been found in human breast cancer (Shipitsin et al., 2007) and metastatic colon cancer (Odoux et al., 2008). Each tumour contained consistent (clonal) karyotypic abnormalities, but also contained clones with unique additional abnormalities, suggesting chromosomal instability that could contribute to the generation of alternative TPCs -this clonal evolutionary process gives rise to intratumoral heterogeneity together with that of the TPC hierarchy (Fig. 1). Indeed, chromosomal instability directly generates increased numbers of tumorigenic side population (SP) cells (see Box 1) in nasopharyngeal carcinoma (Liang et al., 2010).

The TPC niche
Normal adult stem cells seem to be regulated by molecular cues that are provided by neighbouring connective-tissue cells, mainly mesenchymal (fibroblast-like) cells and vascular cells; these stromal cells contribute to what is known as the stem cell niche (Watt and Hogan, 2000;Alison and Islam, 2009). TPCs can be present in and influenced by a similar microenvironment. In both SCC (Atsumi et al., 2008) and bladder cancer , putative TPCs are located at the tumour-stroma interface. In colorectal cancer, the promotion of Wnt signalling in TPCs requires co-stimulation by hepatocyte growth factor (HGF) produced by stromal fibroblasts, illustrating the importance of the microenvironment in maintaining the tumorigenic capacity of cells within tumours (Vermeulen et al., 2010). Brain TPCs identified on the basis of the expression of nestin and CD133 have been observed congregated close to capillaries in a niche; thus, therapeutic targeting of the vasculature could destroy the niche as well as achieve tumour debulking (Calabrese et al., 2007). In leukaemic mice, leukaemic TPCs outcompete normal HSCs for their normal osteoblastic and vascular niches and, as a consequence, normal HSCs lack support and decline in number (Colmone et al., 2008). There is increasing evidence that the leukaemic TPCs have maintenance requirements similar to those of HSCs, which are provided by the niche (Lane et al., 2009); therefore therapeutic targeting of the niche occupied by leukaemic TPCs is an attractive yet difficult proposition. Such strategies could include inhibiting leukaemic TPC homing and engraftment by antibody blocking of molecules such as CD44 and CXCR4, which are expressed by tumour cells, and CXCL12 (also known as SDF-1), which is expressed by niche stromal cells (Fig. 3C).

Markers of TPCs in human tumours
TPCs have been prospectively isolated on the basis of their expression of particular markers -often CD133, but also celladhesion molecules, cytoprotective enzymes (e.g. aldehyde

2361
Stem cells in cancer Fig. 2. Assays for TPCs. Selected cells (e.g. CD133 + cells) can be assessed either for their ability to form a large family of descendants, i.e. whether they are clonogenic in an in-vitro assay for TPCs, or by their ability to initiate new tumour growth after xenotransplantation into immunocompromised mice. (A)Cells can be plated at low density directly onto plastic or in Matrigel and, after several weeks, the plating efficiency (in percent) can be calculated as the number of macroscopically visible clones per total number of cells plated; this assay probably does not distinguish between stem and progenitor cells. (B)Cells can be grown at low density in non-adherent conditions in semiliquid medium, the resulting spheres (e.g. neurospheres, colonospheres, mammospheres) should generate both CD133 + and CD133cells. The resulting CD133 + cells can be serially passaged, generating secondary and tertiary spheres with a cellular composition resembling that of the primary spheres. (C)The gold-standard assay for TPCs in which the selected cells are orthotopically (common for brain) or ectopically (e.g. subcutaneous site, common for colon) xenotransplanted into an immunocompromised mouse. dehydrogenase, ALDH) and drug-efflux pumps (e.g. ABC transporters) (Box 1). As this field is still evolving, there is not yet an apparent consensus about the best marker by which to identify TPCs in any particular cancer (Table 1). Brain TPCs have mainly been studied in medulloblastoma, a paediatric tumour of the cerebellum and gliomas: in this case, CD133 has been the best marker to enrich for TPCs (Piccirillo et al., 2009a). By contrast, there has not been a similar consensus on the best marker for identifying TPCs in other cancers. This is particularly true for gastrointestinal carcinomas, as several permutations and 2362 Journal of Cell Science 123 (14) Fig. 3. Possible therapeutic targets in TPCs. Bold blocking arrows indicate points of therapeutic intervention. (A)Strategies based on targeting intracellular pathways active in TPCs. The active Wnt, EpCAM, Hedgehog and Delta/Notch pathways have all been implicated in TPC self-renewal and proliferation, and can be inhibited to therapeutically target these cells. (B)Strategies based on targeting molecules that are highly expressed by TPCs. Targeting ABC transporters and ALDH activity might sensitise TPCs to currently available drugs. CD133 is another potential target (although CD133 expression can be ubiquitous in some tissues). Manipulation of mRNA expression levels through microRNAs (miRNAs) is also a promising strategy by which to target TPCs. (C) Strategies based on targeting molecules involved in invasion and metastasis. Blocking CXCR4-CXCL12 and CD44-hyaluronan (HA) interactions might target TPCs. Inhibiting EMT might slow the generation of more TPCs with metastatic potential. The direct destruction of endothelial cells that maintain vascular niches also has the potential to eliminate TPCs. -Cat, -catenin; Dsh, dishevelled; ECM, extracellular matrix; EpEX, EpCAM extracellular domain; Fz; Frizzled; GSK3, glycogen synthase kinase 3; HH, Hedgehog; Lrp5/6, low-density lipoprotein receptor-related protein 5 and/or 6; PTC, Patched; SMO, Smoothened. apparent contradictions have been proposed regarding the use of various markers, particularly CD133 (Shmelkov et al., 2008).

TPCs in tumour invasion and metastasis
If TPCs are involved in tumorigenesis, it follows that their frequency in primary tumours is correlated with the extent of tumour invasion and metastasis and, in turn, to patient prognosis (their clinical outlook). The chemokine receptor CXCR4 is often expressed by tumour cells and by TPCs in particular. Hence, these cells can migrate along a gradient of the CXCR4 ligand, CXCL12, which can originate from haematopoietic niches and many other tissue sites, thereby facilitating metastasis. For example, CXCR4 is highly expressed by human endometrial cancer, whereas CXCL12 is expressed by normal tissues; a CXCR4-neutralizing antibody was effective in blocking the seeding of cancer cells at ectopic sites in nude mice, highlighting the potential use of antagonists of chemokine receptors as a therapeutic strategy (Gelmini et al., 2009) (although it was not clear if TPCs were specifically targeted in this study). Similarly, blocking CXCR4 expression in sphere-forming cells (putative TPCs) in breast cancer cell lines using antibodies or shRNA inhibited invasive behaviour in vitro (Krohn et al., 2009). CXCR4-expressing CD133 + TPCs are also directly involved in the metastasis of human pancreatic cancer cells: when CD133 + TPCs were partitioned into CXCR4 + and CXCR4fractions, only the CXCR4 + cells formed metastases after orthotopic xenografting. Notably, the metastases could be blocked by AMD3100, a small-molecule inhibitor of CXCR4 (Hermann et al., 2007).
The invasion of ALDH + TPCs in breast cancer cell lines involves interleukin-8 (IL-8) and its receptor CXCR1; these TPCs expressed high levels of CXCR1 and showed markedly enhanced invasion through Matrigel in response to IL-8 (Charafe-Jauffret et al., 2009). The expression of various integrins and of the transmembrane glycoprotein CD44 on gastric cancer SP cells can also be important for metastasis (Nishii et al., 2009).
In the metaplastic type of breast cancer that is characterised by aggressive growth, chemoresistance and poor patient outcome, the gene signature of the tumour generally closely matches that of CD44 + CD24 -/low TPCs found in most human breast tumours (Hennessey et al., 2009). As well as illustrating the link between TPCs and metastatic tumours, these data also highlight that stemness traits are usually a poor prognostic indicator.

TPCs and patient prognosis
Overall, we believe that, for a given tumour type, a high number of stem cells indicates a poor prognosis. In breast cancer, for example, the most poorly differentiated tumours have the highest burden of TPCs (Pece et al., 2010). In brain tumours, high CD133 expression is a prognostic marker for reduced time of diseasefree survival and overall survival, a measure that is independent of tumour grade, the extent of resection and the age of the patient (Cheng et al., 2009). Another study reported that combined nestin and CD133 expression is a marker of poor prognosis (Zhang et al., 2008b). Neurosphere formation (an assay to identify TPCs) has also been found to be an independent predictor of death in patients with malignant gliomas, suggesting that the ability to propagate neurospheres in culture is a clinically relevant parameter (Laks et al., 2009). High CD133 expression in colorectal carcinoma has also been found to be an independent marker of poor prognosis (Horst et al., 2009). In pancreatic cancer, in which 60% of tumours express CD133 (with <15% positive cells per case), CD133 expression was found to be an independent adverse prognostic factor for 5-year survival. In addition, CD133 expression was significantly associated with lymph-node metastases (Maeda et al., 2008). However, despite the large body of evidence showing that TPCs in a wide range of tumour types express CD133, the functional role of this molecule in TPC activity is not clear. Although high CD133 expression is tightly correlated with colorectal liver metastasis, small interfering RNA (siRNA)-mediated knockdown of CD133 expression in cultured colon cancer cell lines affected neither overall cell migration nor invasion (Horst et al., 2009). High ALDH expression in tumours has also been associated with poor prognosis in a number of tumour types including breast Tanei et al., 2009), pancreas (Rasheed et al., 2010) and early-stage lung cancer (Jiang et al., 2009a). High activity levels of the ABC transporters have also been reported to be a sign of poor prognosis; in patients with AML, there is reduced overall 4-year survival in patients with tumours expressing high levels of ABCC11 (also known as MRP8) (Guo et al., 2009). In summary, patients with tumours that express high levels of molecules associated with TPCs have a poorer prognosis than patients with tumours that express low levels of these markers.

Therapeutic targeting of TPCs
If TPCs are the roots of cancer, then these are the cells that must be specifically eliminated for a successful therapy. However, it is well known that many cancers are resistant to currently available therapies. For example, CD133 + TPCs in glioblastomas are highly resistant to irradiation owing to upregulation of the DNA-damage response and can be sensitised by inhibition of Chk1 and Chk2 (Bao et al., 2006), whereas the entire pool of CD133 + TPCs can be reduced by forcing them to differentiate into astrocytes with bone morphogenetic proteins (Piccirillo et al., 2006). Therapy can also enrich for TPCs in the residual tumour, as seen with gemcitabine treatment of pancreatic cancer (Mueller et al., 2009), cyclophosphamide treatment of colorectal cancer (Dylla et al., 2008), doxorubicin and fluorouracil treatment of HCC (Ma et al., 2008) and cisplatin, doxorubicin and methotrexate treatment of lung cancer (Levina et al., 2008). Therefore, highly specific therapies must be developed to target TPCs.
TPCs and other cancer cells express several potential therapeutic targets (Fig. 3). Targeting key signalling pathways that are active in TPCs is one approach to therapy (Fig. 3A): for example, inhibiting the release of Notch intracellular domain (NICD) with -secretase inhibitors, and by blocking the Hedgehog and Wnt pathways with cyclopamine and DKK1, respectively. A new approach to antagonize Wnt signalling has been to stabilize axin, thereby preserving the -catenin destruction complex . Components of the Wnt signalling pathway also link up with the intracellular domain of EpCAM (EpICD) to promote cell proliferation; thus, similar to the Notch pathway, targeting the proteases involved in intracellular cleavage are promising approaches (Munz et al., 2009).
Small-molecule inhibitors of various other molecules expressed by TPCs is an alternative avenue for therapy (Fig. 3B). As the drug resistance of TPCs can often be directly attributed to the activity of ALDH (Dylla et al., 2008) or ABC transporters (Loebinger et al., 2008;Sung et al., 2008), these are both promising targets for small-molecule therapy (Fig. 3B). In addition, given that many types of cancer cells have a specific microRNA-expression profile (Nicoloso et al., 2009), the use of microRNA-based therapeutic tools to target TPCs is an area of increasing interest. For example, low expression of microRNA-199b-5p is a marker of poor patient survival in medulloblastoma: this miRNA affects the Notch pathway by downregulating Hes1 expression, and its introduction reduces proliferation and engraftment of CD133 + medulloblastoma TPCs in athymic mice (Garzia et al., 2009). In GBM, transfection of microRNA-124 and microRNA-137 causes cell-cycle arrest and apparent differentiation of CD133 + cells (Silber et al., 2008). MicroRNA-128 is markedly downregulated in glioblastomas, and introducing microRNA-128 into glioma cells reduces their proliferation by directly binding to the 3Ј-untranslated region of BMI1 mRNA (Godlewski et al., 2010).
Other therapeutic strategies might involve inhibiting invasion and metastasis by blocking integrins or CD44 (which binds to hyaluronan) at the cell surface (Fig. 3C). For example, CD44 + ovarian tumour cells that express markers of activated pluripotent stem cells might have a selective advantage for spreading through adhering to the hyaluronic-acid pericellular coat of surrounding mesothelial cells (Bourguignon et al., 2008). Finally, smallmolecule inhibitors of CXCR4 were shown to successfully block the metastasis of xenografted human pancreatic cancer (Hermann et al., 2007).
As discussed above, EMT of cancer cells can generate moreinvasive cells and give rise to new TPCs; therefore, developing therapies that inhibit this process is an important aim. EMT is a prominent feature of basal-like breast cancer; here, inhibiting Wnt signalling was shown not only to block stem cell self-renewal but also represses the expression of the CDH1 repressors Twist and Slug, which, in turn, blocks metastasis (DiMeo et al., 2009).

Conclusions and perspectives
Normal adult multipotential stem cells are self-renewing. If normal adult stem cells are the founder cells of many tumours, TPCs probably inherit many of the attributes of normal stem cells. We believe that most human tumours -both solid tumours and haematological malignancies -have a subpopulation of TPCs, although recent data challenge the once-popular belief that they are always a minority subpopulation. For example, in malignant melanoma, they can make up 25% of the total cell population. The fact that mouse tumours also contain subpopulations that are enriched for TPCs supports the notion that malignancies contain specific cells with stem cell attributes. Collective evidence indicates that markers such as CD133, CD44 and ALDH are characteristically expressed by TPCs, and that this population can be expanded within tumours by both symmetrical division of TPCs and/or through other cancer cells undergoing EMT. Many therapeutic approaches are on the horizon by which to target TPCs in cancer, which is a challenging prospect given that these cells seem to be particularly resistant to current therapies.