Michael F. Clarke,Irving L. Weissman
SUMMARY OF KEY POINTS
Common cancers arise in tissues that contain a large subpopulation of proliferating cells that are responsible for replenishing the short-lived mature cells. In such organs, cell maturation is arranged in a hierarchy in which a rare population of stem cells, which perpetuate themselves through a process called self-renewal, gives rise to intermediate progenitors and then mature cells, neither of which self-renew.            Because of their rarity, stem cells must be isolated prospectively to study their biologic, molecular, and biochemical properties. Although it is likely that each tissue regenerates from tissue-specific stem cells, stem cells have been rigorously identified and purified in only a few. The stem cells that give rise to the lymphohematopoietic system, called hematopoietic stem cells (HSCs), have been isolated from mice and humans and are the best-characterized stem cells. The utility of tissue containing HSCs has been demonstrated in cancer therapy with its extensive use for bone marrow transplantation to regenerate the hematolymphoid system after myeloablative protocols. The prospective isolation of HSCs from patients can result in a population of cancer-free cells for autologous transplantation.     
Understanding the cellular biology of the tissues in which cancers arise, and specifically that of the stem cells that reside in those tissues, could provide new insights into cancer biology. Several aspects of stem cell biology are relevant to cancer. First, both normal stem cells and cancer stem cells undergo self-renewal, and emerging evidence suggests that similar molecular mechanisms regulate self-renewal in normal stem cells and their malignant counterparts. Next, it is quite likely that mutations that lead to cancer accumulate in normal stem cells. Finally, as was stated previously, it is likely that tumors contain a minority “cancer stem cell” population with indefinite proliferative potential that drives the growth and metastasis of tumors.          
PROPERTIES OF NORMAL STEM CELLS
HSCs are the most studied and best understood somatic stem cell population and serve as a model for stem cells from other tissues.      Hematopoiesis is a tightly regulated process in which a pool of HSCs eventually gives rise via oligo lineage intermediates     to the lymphohematopoietic system consisting of the formed blood elements (e.g., red blood cells, platelets, granulocytes, macrophages, and B and T lymphocytes). These cells are important for oxygenation, prevention of bleeding, immunity, and fighting infections. In the adult, HSCs have three fundamental properties. First, HSCs need to self-renew to maintain the stem cell pool. Self-renewal is not synonymous with proliferation. Self-renewal is a cell division in which one or both of the daughter cells remain undifferentiated and have the ability to give rise to another stem cell as well as to the spectrum of more differentiated progenitors. Second, HSCs must undergo differentiation to maintain a constant pool of mature cells in normal conditions and to produce increased numbers of a particular lineage in response to stresses such as bleeding or infection. Third, the total number of HSCs is under strict genetic regulation.
In the mouse hematopoietic system, multipotent cells constitute 0.05% of bone marrow cells and are heterogeneous with respect to their ability to self-renew. There are three different populations of multipotent cells: long-term self-renewing HSCs, short-term self-renewing HSCs, and multipotent progenitors without detectable self-renewal potential.   These populations form a hierarchy in which the long-term HSCs give rise to short-term HSCs, which in turn give rise to multipotent progenitors ( Fig. 7-1 ). As HSCs mature from the long-term self-renewing pool to multipotent progenitors, they become more mitotically active but lose the ability to self-renew. Only long-term HSCs can give rise to mature hematopoietic cells for the lifetime of the animal, whereas short-term HSCs and multipotent progenitors reconstitute in lethally irradiated mice for fewer than 8 weeks.
Figure 7-1 Blood development hierarchy. All of the diverse mature blood cells arise from the hematopoietic stem cells (HSCs). The cells that are capable of multilineage reconstitution of a lethally irradiated mouse are contained within two identifiable and separate populations of cells: the long-term hematopoietic stem cells HSCs (LT-HSCs) and the short-term hematopoietic stem cells HSCs (ST-HSCs). Only the LT-HSCs are capable of self-renewal for the lifetime of the animal. In contrast, other cells, even the ST-HSCs that can give rise to large numbers of mature blood cells, have very limited life spans (measured in hours to 1 or 2 months). GMP, granulocyte-macrophage progenitors; MEP, myeloid-erythroid progenitors.
Despite the fact that the phenotypic and functional properties of mouse and human HSCs have been extensively characterized, understanding of the fundamental stem cell property, self-renewal, is minimal.    In most cases, HSCs differentiate when exposed to combinations of growth factors that can induce extensive proliferation in long-term cultures. Although recent progress has been made in identifying culture conditions that maintain HSC activity in culture for a limited period, it has proven to be exceedingly difficult to identify tissue culture conditions that promote a significant and prolonged expansion of progenitors with transplantable HSC activity.
GENETIC REGULATION OF SELF-RENEWAL IN NORMAL STEM CELLS AND CANCER CELLS
Maintenance of a tissue or a tumor is determined by a balance of cell proliferation and cell death. As would be expected, many of the mutations that drive tumor expansion regulate either cell proliferation or survival. For example, the prevention of apoptosis by enforced expression of the oncogene Bcl-2 promotes the development of lymphoma and also results in increased numbers of HSCs in vivo, suggesting that cell death plays a role in regulating the homeostasis of HSCs; enforced expression of bcl2 does not endow short-term HSCs or multipotent progenitors with self-renewal properties.   In fact, the progression to experimental acute myelogenous leukemia (AML) in mice requires at least four independent events to block the several intrinsically triggered and extrinsically induced programmed cell death pathways of myeloid cells. Proto-oncogenes such as c-myb and c-myc that drive proliferation of tumor cells are also essential for HSC development     as well as increased expression of the telomere regenerating enzyme and RNA.
Self-renewal is critical for both normal stem cells and cancer stem cells. In a normal tissue, stem cell numbers are under tight genetic regulation, resulting in the maintenance of a constant number of stem cells in the organ.    In contrast, cancer stem cells have escaped this homeostatic regulation, and the number of cells within a tumor with the ability to self-renew is constantly expanding, resulting in the inevitable growth of the tumor. Because cancer cells and normal stem cells share the ability to self-renew, it is not surprising that a number of genes classically associated with cancer may also regulate normal stem cell development.   In combination with other growth factors, sonic hedgehog (Shh) signaling has also been implicated in the regulation of self-renewal by the finding that cells that are highly enriched for human HSCs (CD34+Lin-CD38-) exhibit increased self-renewal in response to Shh stimulation in vitro. Several other genes related to oncogenesis have been shown to be important for stem cell function. For example, mice that are deficient for tal-1/SCL, which is involved in some cases of human acute myeloid leukemia, lack embryonic hematopoiesis, suggesting thattal-1/SCL is required for intrinsic or extrinsic events necessary to initiate hematopoiesis, for maintenance of the earliest definitive blood cells, or for the formation of blood cells downstream of embryonic HSCs.   Members of the Hox family have also been implicated in human leukemia, and enforced expression of HoxB4 can affect stem cell functions.   One of the major targets of the p53 tumor suppressor gene is p21cip1. Bone marrow from p21cip1-deficient mice has a reduced ability to serially reconstitute lethally irradiated recipients. Failure at serial transfer could result from exhaustion of the stem cell pool, loss of telomeres, or loss of transplantability.    Thus, many genes that are involved in decisions about stem cell fate are also involved in malignant transformation.
The notion that the function of certain oncogenes is to regulate self-renewal is perhaps best illustrated by studies of the oncogene bmi-1. In mice, bmi-1 cooperates with c-myc to induce lymphoma.  The number of HSCs is markedly reduced in postnatal bmi-1-/- mice, and transplanted bmi-1-/- fetal liver and bone marrow cells are able to contribute only transiently to hematopoiesis, indicating a cell autonomous defect of HSC self-renewal in bmi-1-/- mice. The expression of stem cell-associated genes, cell survival genes, transcription factors, and genes that modulate proliferation, includingp16Ink4a and p19Arf, is altered in bone marrow cells of bmi-1-/- mice. This suggests that the function of bmi-1 is to regulate a cascade of genes that modulate stem cell self-renewal. In a mouse model of leukemia, leukemic cells that lack expression of bmi-1 eventually undergo proliferation arrest associated with signs of differentiation and apoptosis when they are transplanted in syngenic hosts. Infection of the cells with a bmi-1 retrovirus completely rescues the proliferative defect of the bmi-1-/- leukemic stem cells. Along a similar line, the regulated postnatal deletion of junB, an AP1 transcription factor, results in HSC expansion and then the development of a chronic myeloproliferative disorder that resembles human chronic myelogenous leukemia (CML) in its chronic phase. Transduced overexpression of junB in HSCs inhibits or prevents their self-renewing proliferations. In the junB knockout HSCs, p16ink4a and p19Arf levels are lowered, and the antiapoptotic proteins bcl2 and bclx are increased, while the reverse is true in HSCs that have junB overexpression. Only HSCs transplant this chronic phase of the disease, but later, in blast crisis, cells at a more differentiated stage emerge that transplant the blast crisis disease.   These studies conclusively demonstrate that malignant transformation requires not only activation of proliferation pathways, as well as inactivation of cell death and cell cycle arrest pathways, but also activation of self-renewal pathways.
Two other signaling pathways that are implicated in oncogenesis in both mice and humans, the Wnt/β-catenin and Notch pathways, may play central roles in the self-renewal of both normal and cancer stem cells. The Notch family of receptors was first identified in Drosophila species and has been implicated in development and differentiation. In Caenorhabditis elegans, Notch plays a role in germ cell self-renewal. In neural development, transient Notch activation initiates an irreversible switch from neurogenesis to gliogenesis by embryonic neural crest stem cells. Notch activation of HSCs in culture with either of the Notch ligands Jagged-1 or Delta transiently increases the primitive progenitor activity both in vitro and in vivo, suggesting that Notch activation promotes either the maintenance of progenitor cell multipotentiality or HSC self-renewal.   Although the Notch pathway plays a central role in development and the mouse oncogene int-3 is a truncated Notch 4, the role of Notch in de novo human cancer is complex and less well understood. Various members of the Notch signaling pathway are expressed in cancers of epithelial origin, and activation of the Notch pathway by chromosomal translocation is involved in some cases of leukemia.      Microarray analysis has shown that members of the Notch pathway are often overexpressed by tumor cells.   A truncated Notch 4 messenger RNA is expressed by some breast cancer cell lines. Overexpression of Notch1 leads to growth arrest of a small cell lung cancer cell line, whereas inhibition of Notch1 signals can induce leukemia cell lines to undergo apoptosis.    Elegant work by Weizen and colleagues showed that activation of Notch1 signaling maintains the neoplastic phenotype in Ras-transformed human cells. They also found that in de novo cancers, cells with an activating Ras mutation also demonstrate increased expression of Notch1 and Notch4.
Wnt/β-catenin signaling also plays a pivotal role in the self-renewal of normal stem cells and malignant transformation.    The Wnt pathway was first implicated in mouse mammary tumor virus-induced breast cancer in which deregulated expression of Wnt-1 caused by proviral insertion resulted in mammary tumors.   Subsequently, it has been shown that Wnt proteins play a central role in pattern formation. Wnt-1 belongs to a large family of highly hydrophobic secreted proteins that function by binding to their cognate receptors, members of the Frizzled and low-density lipoprotein receptor-related protein families, resulting in activation of β-catenin.      In the absence of receptor activation, β-catenin is marked for degradation by a complex consisting of the adenomatous polyposis coli, Axin, and glycogen synthase kinase-3b proteins.       Wnt proteins are expressed in the bone marrow, and activation of Wnt/β-catenin signaling by Wnt proteins in vitro or by expression of a constitutively active β-catenin expands the pool of early progenitor cells and enriched normal transplantable HSCs in tissue culture and in vivo.    Inhibition of Wnt/β-catenin by ectopic expression of Axin, an inhibitor of β-catenin signaling, leads to inhibition of stem cell proliferation both in vitro and in vivo. Addition of Wnt3a to purified HSC leads to their expansion, presumably self-limited usually by the expression of the β-catenin-induced transcription of axinII; transduction of purified HSC with axin prevents their expansion in vitro or in vivo.   Other studies suggest that the Wnt/β-catenin pathway mediates stem or progenitor cell self-renewal in other tissues.     The level of β-catenin in a particular keratinocyte directly correlates with its proliferative capacity.   As in their normal HSC counterparts, enforced expression of an activated β-catenin in epidermal stem cells increases their ability to self-renew and decreases their ability to differentiate. Mice that fail to express TCF-4, one of the transcription factors that is activated when bound to β-catenin, soon exhaust their undifferentiated crypt epithelial progenitor cells, further suggesting that Wnt signaling is involved in the self-renewal of epithelial stem cells.  
Activation of β-catenin in colon cancer by inactivation of the protein degradation pathway, most frequently by mutation of adenomatous polyposis coli, is common.     Expression of certain Wnt genes is increased in some other epithelial cancers, suggesting that activation of β-catenin might be secondary to ligand activation in such cancers.        There is evidence that constitutive activation of the Wnt/β-catenin pathway might confer a stem/progenitor cell phenotype to cancer cells. Inhibition of β-catenin/TCF-4 in a colon cancer cell line induced the expression of the cell cycle inhibitor p21cip-1 and induced the cells to stop proliferating and to acquire a more differentiated phenotype. Enforced expression of the proto-oncogene c-myc, which is transcriptionally activated by β-catenin/TCF-4, inhibited the expression of p21cip-1 and allowed the colon cancer cells to proliferate when β-catenin/TCF-4 signaling was blocked, linking Wnt signaling to c-myc in the regulation of cell proliferation and differentiation.
The implication of roles for genes such as Notch, Wnt, c-myc, and Shh in the regulation of self-renewal of HSCs, and perhaps of stem cells, from multiple tissues suggests that there might be at least some common self-renewal pathways in many types of normal somatic stem cells and cancer stem cells. It will be important to identify the molecular mechanisms by which these pathways work and to determine whether the pathways interact to regulate the self-renewal of normal stem cells and cancer stem cells.
TARGET CELLS FOR MALIGNANT TRANSFORMATION
If oncogenic mutations often target signaling pathways that regulate proliferation and self-renewal, then are stem cells, highly proliferative progenitor cells, or both the target of neoplastic transformation? Several lines of evidence suggest that stem cells might be involved in the evolution of a cancer. First, the fact that multiple mutations are necessary for a cell to become cancerous   suggests that in many cases, mutations accumulate in a stem cell. Progenitor cells have a very limited life span, making it less likely that all of the mutations occur during the life of these relatively short-lived cells.         Second, the regulation of stem cell expansion and self-renewal is under strict genetic regulation by multiple genes, and unregulated expansion of stem cells could, in essence, result in a cancer.    Third, most cancers arise in tissues that contain stem cells that have the intrinsic ability to self-renew. Because cancer cells must undergo self-renewal, this suggests that stem cells might more easily undergo steps in the progression to malignant transformation than will progenitor cells that lack this fundamental property and must therefore activate these self-renewal pathways to become malignant. In the hematopoietic system, the only cells that have the ability to self-renew are HSCs and mature lymphocytes. The common blood cancers, acute leukemias and lymphomas, may arise from the HSCs or lymphocytes, respectively, via constitutive activation of mitogenic pathways associated with the proliferation of normal cells.     Although stem cells may undergo steps toward malignant transformation, it is possible, if not likely, that in many cases, their progenitor cells that inherit the changes in stem cells, and in addition add the ultimate transforming event, give rise to cancer. For example, the initial mutations that occur in the stem cell could permit a single mutation to transform a progenitor cell, or perhaps events that shut down self-renewal could occur in stem cells, but only progenitors outside of the stem cell regulatory niche expand as cancer stem cells. It is also possible that certain oncogenic mutations such as bmi-1 or its downstream targets could confer the property of self-renewal on a progenitor cell.
The target cells for transformation are best understood in hematopoietic malignancies because the developmental hierarchy of the blood is well established.     One of the most frequent mutations in AML in elderly patients is the t(8;21) translocation, which results in the expression of a chimeric AML-ETO transcript in the leukemic cells.    Marrow samples from patients with early onset 8;21 leukemia in Hiroshima Hospital had CD34+Thy1(CD90)+CD38-Lin- HSCs, which, when isolated from patients in clinical remission, had up to 90% incidence of the chimeric AML-1-ETO transcript. When these HSCs were analyzed by means of in vitro differentiation assays, the HSCs gave rise to normal myeloerythroid progeny, demonstrating that the mutation was present in the otherwise normal stem cells. In these patients, the CD90 neg subset of CD34+CD38-Lin- cells gave rise to leukemic colonies in vitro; this could represent HSCs that have lost Thy1 expression or downstream multipotent progenitors that have gained self-renewal capacity. Taken together, these observations support the notion that mutations accumulate in stem cells and that subsequent mutations in either the stem cells or their progeny result in overt leukemia.
Although stem cells are frequently the target of mutations that are on the path to malignant transformation, it is likely that their clonal progenitor cells may be transformed by subsequent genetic events that confer immortality, self-renewal potential, or both to these normally non-self-renewing cells ( Fig. 7-2 ). In patients with CML, the BCR-ABL mutation is present in both normal and leukemic stem cells. In otherwise normal hematopoietic cells, the BCR-ABL mRNA is expressed solely by the progenitor cells.    In a mouse model of CML, BCR-ABL expression targeted to myeloid progenitor cells by the hMRP-8 promoter resulted in CML-like disease in a subset of the mice. Furthermore, when hMRP8p210BCR/ABL mice were crossed with hMRP8bcl-2 mice, a proportion of the mutant mice developed a disease resembling AML or myeloid blast crisis. Although the expression of transforming genes was targeted to early progenitor cells, the appearance of the leukemia cells and clinical course resembled human CML and AML in the hMRP8p210BCR/ABL mice and the hMRP8p210BCR/ABL/hMRP8bcl-2 mice, respectively. In the chronic phase of human CML, the CML phenotype of the CML stem cell is identical to that of a normal HSC. β-catenin signaling is active in both the normal HSCs and the CML stem cells but not in the progenitor cells. When CML progresses to “myeloid blast crisis,” the patients develop an acute leukemia-like disease. The leukemic stem cell in CML blast crisis phenotypically is at the developmental stage of a granulocyte-macrophage progenitor cell that transplants the disease to immunodeficient mice and that self-renews replating potential in vitro. Furthermore, the CML blast crisis stem cells appear to have activated the β-catenin signaling pathway and are inhibited in their in vitro self-renewal by transduced axin. Thus, disease progression in human CML appears to result from activation of self-renewal pathways in a progenitor cell population or, more likely, failure to shut down this pathway in the HSC to granulocyte-macrophage transition. These studies strongly suggest that the leukemia stem cell in CML blast crisis is derived from a progenitor cell, not from a normal stem cell, although the initial event and likely several progression events are in successive subclones of the initial bcr-abl HSC stage cells.
Figure 7-2 Target cells for neoplastic transformation. In many tissues in which cancers arise, the stem cells are the only long-lived cells and are the only cells capable of self-renewal. Because they are already capable of extensive self-renewal, they are good targets for neoplastic transformation. Dysregulation of the self-renewal process may be simpler in these cells than in progenitor cells that lack this ability. For progenitor cells to undergo malignant transformation, they must acquire the ability to undergo extensive self-renewal as a result of oncogenic mutations.
In a mouse model of high-grade glioblastoma, enforced expression of the epidermal growth factor receptor-enriched populations in either Ink4a/Arf null neuronal stem cells or Ink4a/Arf null astrocytes led to malignant glioblastomas when the cells were injected orthotopically into mice. Notably, in the majority of the cases, the transformed astrocytes appeared to acquire an immature phenotype in the brains of the mice, suggesting to some researchers that there was “dedifferentiation,” although we have never documented a case of dedifferentiation with purified normal hematopoietic progenitor cells. There are two other possible explanations for these results. First, the astrocyte tissue culture cells could have contained a rare population of neuronal stem cells that were transformed, and these stem cells were responsible for generating the tumors. Second, it is possible that the tissue culture conditions could have caused the dedifferentiation of the astrocytes and that unless the astrocytes are grown in tissue culture, they cannot give rise to glioblastomas in an animal. These observations in humans and mice support the notion that oncogenic mutations accumulate in the stem cells, but expression of the mutated gene by progenitors downstream of the stem cells can lead to their neoplastic transformation of progenitor cells. These observations have implications for targeted therapies.
It is possible that only a minority of the mice whose progenitors express BCR-ABL develop leukemia because the progenitors must acquire an additional mutation or epigenetic change that causes deregulated self-renewal. Two lines of evidence support this notion. First, expression of bmi-1 is necessary for the self-renewal of adult HSCs, and the blast cells of patients with AML express large amounts of this protein. Although expression of HoxA9 and Meis1 induces transplantable AML in normal mice, expression of these genes in the absence of bmi-1 does not.   This suggests that both normal HSCs and leukemic stem cells require bmi-1 to self-renew. Second, deregulated β-catenin signaling occurs in many de novo human cancers and causes cancer in transgenic mouse models. Because expression of a constitutively active β-catenin can promote the self-renewal of normal HSCs, as well as stem cells from other tissues, it is quite plausible that activation of this pathway promotes self-renewal of the cancer cells.           From these results, it is evident that future studies focusing on the molecular regulation of the self-renewal of normal stem cells and cancer cells will likely lead to more effective therapies for cancer.
EVIDENCE FOR CANCER STEM CELLS
It has long been known that cancers consist of phenotypically heterogeneous populations of cancer cells.         These phenotypically distinct cell populations could arise in part from sequential mutations caused by genetic instability, environmental factors, or both ( Fig. 7-3A ). Alternatively, a tumor can be viewed as an aberrant organ containing a tumorigenic (stem cell) population that drives tumor growth. These tumorigenic cells would have acquired oncogenic mutations and epigenetic changes that result in unregulated self-renewal, extended cell survival, and avoidance of innate and adaptive immune surveillance and would also give rise to phenotypically diverse populations of tumor cells that lack the ability to self-renew ( Fig. 7-3B ). Several lines of evidence suggest that the latter model accounts for some of the cellular heterogeneity that is seen in tumors, although genetic instability and environmental factors could also contribute to the variability in phenotypes.   It is well documented that many types of cancer contain heterogeneous populations of cells that variably express differentiation markers that reflect the tissues from which the tumors originate, as well as cancer cells that have an immature appearance.    Examples of this include the variable expression of milk proteins by some breast cancers and the variable expression of myeloid markers, lymphoid markers, or both in CML and AML. Perhaps the most striking example of abnormal differentiation in cancer is the variable expression of diverse tissues in some germ cell tumors. Mature tissues such as teeth, skin, and hair are present in some cases of teratocarcinomas ( Fig. 7-4 ). In contrast, in some tumors, only a minority of the cancer cells express immature cell markers such as α-fetoprotein (see Fig. 7-4 ). Because the terminally differentiated cells that form the teeth and hair in the tumors are unlikely to be able to proliferate and form new tumors, these data suggest that the minority population of α-fetoprotein-expressing cancer cells has the exclusive ability to form new tumors consisting of more tumorigenic cells, as well as the phenotypically diverse populations of non-self-renewing abnormally differentiated cells. If this is true, these cells can thus be considered cancer stem cells.
Figure 7-3 The two most likely models of heterogeneity of the cancer cells, shown as different colored cells within a tumor. A, Heterogeneity is due to environmental factors (gold, red, green, and blue cells) or due to ongoing mutations in the cancer cells (magenta cells). In this model, all of the cancer cells have the intrinsic ability to form tumors. B, Cancer stem cells (yellow cells) have the exclusive ability to self-renew. As in normal tissues, the stem cells would give rise to more stem cells with the capacity to form new tumors, as well as the other heterogeneous populations of cancer cells that lack the ability to form new tumors. To date, therapeutic and diagnostic strategies have been based on model A, but these strategies might be limited because they might not target the rare population of cancer stem cells depicted in model B.
Figure 7-4 Clinical evidence for the stem cell model. The clinical and radiographic information for a patient with metastatic teratocarcinoma of the testis is shown. A, In the original testicular biopsy specimen, expression of α-fetoprotein by only rare cancer cells (brown cells) was detected by immunohistochemistry of the original testicular tumor. The original histologic finding in this patient was believed to represent a teratoma. B, Computed tomographic scans before treatment (upper panel) demonstrated large retroperitoneal masses that were still present after four courses of platinum-based chemotherapy (lower panel). C, Biopsy specimen of the residual mass revealed only mature teratoma. No cells expressed the immature marker α-fetoprotein. The patient has survived for more than 10 years without recurrence of his germ cell cancer. This suggests that in some patients, therapies that selectively eliminate the rare stem cell population while sparing the “nontumorigenic” cancer cells could be curative.
If a tumor is viewed as an abnormal organ, then the principles of stem cell biology can be applied to better understand the biology of these diseases.    It was first shown in hematopoietic malignancies and subsequently in solid cancers that only a subset of cancer cells were clonogenic when placed in tissue culture or injected into immunodeficient mice.         For example, only 1 in 100 to 1 in 10,000 mouse myeloma cells obtained from ascites fluid formed in in vitro colony-forming assays. Similarly, only 1% to 4% of leukemic cells formed spleen colonies when transplanted into mice. In solid cancers, only 1 in 1000 to 1 in 5000 ovarian cancer or lung cancer cells formed colonies in soft agar. Because only a minority of normal bone marrow cells was also clonogenic, the clonogenic cancer cells were described as cancer stem cells, implying that only a distinct population of cancer cells was able to proliferate extensively in these assays. However, an alternative explanation is that all the cancer cells had an intrinsic ability to proliferate extensively but only a minority of cells did so in a particular assay. Variable plating efficiency in a variety of in vitro cells and cell lines could have accounted for the poor performance of such cells, and in none of the assays did the investigators prospectively purify the plating cell to show that it was, or was not, a cancer stem cell.
To prove that a phenotypically distinct population of cancer cells is solely responsible for perpetuating the disease, it is necessary to isolate different populations of cancer cells and demonstrate that one or more groups are enriched for the ability to initiate disease and other populations lack this ability. This was done in the case of AML, when it was shown that in most cases of human AML, a leukemic tumor initiating subpopulations of cells could be identified prospectively and enriched from the bone marrow of multiple patients. In most cases of AML, the minority population of CD34+CD38- cells was the only group of cells that was capable of establishing human AML in the bone marrow of nono-bese diabetic/severe combined immunodeficient (NOD/SCID) mice.   Remarkably, within the CD34+CD38- population are Thy1+CD34+CD38- Lineage- normal HSCs        as well as Thy1-CD34+38-Lineage- progenitors. Because normal HSCs, but not their leukemic initiating cell counterparts, express Thy1, it is likely that the early mutations occurred in the HSCs and the final transforming mutations occurred either in early downstream progenitors or in HSCs if, as a consequence of neoplastic transformation, Thy1 expression was lost.
Recently, tumorigenic and nontumorigenic subsets of cancer cells have been isolated from human breast cancer tumors. When a similar model for human breast cancer was used in which isolated cells were grown in immunocompromised mice, a minority population of breast cancer cells that had the ability to form new tumors was identified. Tumorigenic cells could be distinguished from nontumorigenic cancer cells on the basis of surface marker expression. In eight of nine patients, tumorigenic cells could be prospectively identified and isolated as CD44+CD24-/lowLineage- cells. As few as 100 CD44+CD24-/lowLineage- cells were able to form tumors, whereas tens of thousands of cells from other populations of cells within the tumor failed to form tumors in NOD/SCID mice. These tumorigenic cells could be serially passaged in mice, and each time, cells within this population generated new tumors containing additional CD44+CD24-/lowLineage- tumorigenic cells, as well as phenotypically mixed populations of other nontumorigenic cancer cells. These data demonstrate the presence of a hierarchy of cells within a breast cancer tumor in which only a fraction of the cells have the ability to proliferate extensively and other cells have only a limited proliferative potential, suggesting that the tumorigenic cells can both self-renew and differentiate. The phenotype of the tumorigenic breast cancer cells may be similar to that of normal breast epithelial stem or progenitor cells, because early multipotent epithelial progenitor cells have been reported to express epithelial cell antigen and CD44.   
The CD44+CD24-/lowLineage- tumorigenic breast cancer cell and the CD34+CD38- CD90- leukemia-initiating cells share with normal stem cells the abilities to proliferate extensively and to give rise to diverse cell types with reduced developmental or proliferative potential.   The extensive proliferative potential of the tumorigenic breast cancer cell population was demonstrated by the ability of as few as 200 tumorigenic breast cancer cells or several thousand leukemia-initiating cells to give rise to tumors that could be serially transplanted in NOD/SCID mice. This extensive proliferative potential contrasts with the bulk of the breast cancer cells that lack the ability to form detectable tumors. Not only was the CD44+CD24-/lowLineage- population of cells able to give rise to additional tumorigenic CD44+CD24-/low Lineage- cells, it was also able to give rise to phenotypically diverse nontumorigenic cells that made up the bulk of the tumors. Thus, both tumorigenic breast cancer cells and leukemia-initiating cells from most tumors appear to exhibit properties of cancer stem cells. However, before these cells can definitively be called cancer stem cells, new assays are needed to demonstrate that a single transplanted cell gives rise to all of the diverse populations of cancer cells within a tumor.
Cancer stem cells have since been identified in multiple tumors, including cancers arising in the brain, head and neck, pancreas, colon, and prostate.      Interestingly, CD44 seems to be useful as a marker for isolation of cancer stem cells from multiple types of tumors of epithelial origin, including head and neck cancer. Importantly, in histology sections, the cancer stem cells in well-differentiated or moderately differentiated tumors showed that the cancer stem cells, but not the nontumorigenic cancer cells, expressed CD44 and bmi-1, which had previously been shown to be involved in self-renewal in some types of stem cells. However, the nontumorigenic cells expressed mature cell markers, while the cancer stem cells did not. These studies demonstrate that the differential expression of the stem cell markers was not an artifact of flow cytometry.
IMPLICATIONS OF CANCER STEM CELLS FOR THE DIAGNOSIS AND TREATMENT OF CANCER
Although the immunocompromised mouse model provides compelling evidence in support of the stem cell model of cancer, the ultimate confirmation of the hypothesis requires proof in humans. If the growth of solid cancers is driven by cancer stem cells, this would have profound implications for the diagnosis and treatment of cancer. At present, all of the phenotypically diverse cancer cells are treated as if they possess the ability to form tumors and the ability to metastasize. However, if in most tumors, only a small population of cancer cells has the ability to self-renew and other populations of cancer cells have only limited ability to proliferate, then this would explain several conundrums of cancer biology. For example, for many years, it has been recognized that disseminated cytokeratin-positive breast cancer cells can be detected in the bone marrow of patients who never experience relapse, even without adjuvant therapy.         One possibility is that the cancer cells lie dormant until some unknown event triggers them to proliferate. Another explanation is that the cancer cells in the bone marrow in this group of patients arose from the spread of nontumorigenic cancer cells, and only when the cancer stem cells metastasize and subsequently self-renew will frank tumors form. Thus, the development of diagnostic reagents that allow cancer stem cells to be identified may have prognostic significance for patients with breast cancer.
The ability to prospectively isolate cancer stem cells and nontumorigenic cancer cells makes it possible to do molecular analyses of each population of cancer cells in a tumor. A 186-gene signature, called the invasiveness gene signature (IGS), was derived by identifying genes that are differentially expressed by breast cancer stem cells and normal breast epithelial cells. The IGS was used to stratify patients with early stage breast cancer on the basis of the similarity of the gene expression of the whole tumor to the cancer stem cell-derived signature. Remarkably, the IGS was associated with both survival and the risk of developing metastasis. The prognostic power of the IGS was greater than gene signatures derived from the nontumorigenic cancer cells, indicating that the IGS contained both cancer-specific and cancer stem cell-specific elements. The prognostic power of the IGS was even greater when combined with a wound repair signature derived from serum-stimulated fibroblasts. This suggests that cancer stem cells interact with normal tumor stromal elements. Such an interaction is further supported by the observation that the cancer stem cells are adjacent to normal stromal elements in well-differentiated and moderately differentiated head and neck tumors.
Another example of the implications of cancer stem cells is the observation that in most solid cancers, such as breast cancers, chemotherapy can frequently shrink tumors, but in most patients, the tumors rapidly recur, and there is only a small impact on patient survival.    Most of the cancer therapeutic agents that are in current use have been developed largely for their ability to shrink a tumor. If only a minority of the cancer cells are tumorigenic and are responsible for driving tumor growth and metastasis, then tumor shrinkage must reflect primarily the elimination of the bulk population of nontumorigenic cells. If a substantial number of tumor stem cells were spared, then the tumors would regenerate from these cells. In support of this model, many patients who are treated with chemotherapy experience an initial shrinkage of their tumors, but tumors recur in sites of prior disease.
The similarities of the AML tumor-initiating cells and normal HSCs suggest that the AML tumor-initiating cells may be more resistant to chemotherapy than the bulk population of leukemic blasts. Compared with their differentiated progeny, normal HSCs express high levels of genes that make them more resistant to cytotoxic agents including antiapoptotic members of the bcl-2 family, as well as members of the ABC transporters that pump many drugs out of the cell.       If the same is true for their cancer stem cell counterparts, then these cells may be significantly more resistant to cytotoxic agents than their nontumorigenic progeny. In support of this possibility, although the chemotherapeutic agent cytosine arabinoside very efficiently killed leukemic blast cells isolated from many patients, the leukemia-initiating cells were selectively spared. In glioblastoma, the cancer stem cells are more resistant to radiation than are their nontumorigenic cancer cell counterparts. These observations suggests that the effect of a particular therapeutic agent on the cancer stem cell population must be taken into account when its curative potential is evaluated.
Because therapeutic agents are selected on the basis of their ability to shrink tumors rapidly, agents that selectively target the cancer stem cells could be overlooked in screens to identify potential therapeutic agents. Initially, such agents would be expected to slow the growth of a tumor only modestly. However, the elimination of the cancer stem cells would eventually halt the spread of the tumor. Perhaps the best clinical evidence of this model occurs in patients with teratocarcinoma. Platinum-based chemotherapy is curative in the majority of these patients; however, many patients are left with residual masses (see Fig. 7-4 ). After surgical resection, the immature cancer cells have been eliminated, leaving only differentiated cancer cells in a mature teratoma (see Fig. 7-4 ). Patients with mature teratomas only occasionally have metastases, and most are cured, demonstrating that the elimination of the presumed stem cell population by the chemotherapy is sufficient for curing this solid cancer. Box 7-1discusses alternative models for cancer cell heterogeneity.
ALTERNATIVE CANCER CELL HETEROGENEITY MODELS
An alternative explanation for the ability of a single, phenotypically unique population of AML cells or breast cancer cells to engraft in NOD/SCID mice is that all cancer cells are tumorigenic in humans but that only the CD34+CD38-Thy1-Lineage- AML cells or the CD44+CD24-/lowLineage- breast cancer cells are able to proliferate in mice. However, there are several reasons that this appears unlikely. First, NOD/SCID mice have previously been validated as in vivo models for the growth of normal human HSCs and human neural stem cells.       Second, tumors that are passaged in mice contain heterogeneous cancer cells that are phenotypically similar to the cancer cells that were present in the original tumors from patients, including both tumorigenic and nontumorigenic fractions.     This demonstrates that the mouse environment is not incompatible with the survival of the nontumorigenic cell fractions. Third, in the case of breast cancers, the tumorigenic and nontumorigenic fractions of cancer cells exhibit a similar cell cycle distribution in mouse tumors, demonstrating that the nontumorigenic cells are able to divide in mice. Thus, on the basis of these data and the data obtained for other types of normal and malignant human stem cells, the NOD/SCID mouse model reliably supports the engraftment of clonogenic human progenitors. However, human data are required to completely exclude the possibility that different populations of cancer cells are clonogenic in mice than are clonogenic in humans. Because ethical issues preclude the injection of cancer cells into humans, unequivocal proof of the stem cell model will require clinical studies that confirm that therapeutic agents that effectively target cancer stem cells in the immunodeficient mice also eliminate cancer stem cells in patients and result in clinical cures.
FUTURE IMPLICATIONS OF CANCER STEM CELLS
The ability to prospectively identify cancer stem cells should have a major impact on the development of new diagnostic and therapeutic agents. At present, all of the cancer cells within a tumor are treated as if they had the ability to drive tumor growth, invasion, and metastasis. The ability to identify these crucial cells will allow efforts to develop new diagnostic markers and therapies to be focused on the cells that are responsible for the maintenance of the malignancy—the cancer stem cells. For example, in efforts to identify the genes and proteins expressed by cancer cells, either whole tumors or all of the phenotypically diverse cancer cells within a tumor are currently used. Because the cancer stem cells represent only a minority of the cancer cells in most tumors, it is nearly impossible to identify diagnostic markers or therapies that target these cells. However, directing expression analyses to enriched populations of cancer stem cells should allow the identification of novel diagnostic markers and novel therapeutic targets that can be exploited to more effectively diagnose and treat cancer. This principle is illustrated by the observation that BCR/ABL oncogene mRNA is not expressed by HSCs that carry the mutation in their DNA;   such an approach may have implications even when oncogenic mutations are targeted.
The ability to prospectively identify the cancer stem cells should also improve the ability to evaluate the curative potential of new therapeutic agents. Although cancer cell lines are useful for evaluating particular biologic pathways, they have proven to be somewhat unreliable when used in attempts to predict the clinical efficacy of a particular therapeutic agent in patients.   Because the tumors that arise in immunodeficient mouse models of human cancer appear to more closely recapitulate the phenotypic diversity of patients’ original tumors, including the generation of tumorigenic and nontumorigenic cells, these models might more effectively predict the potential usefulness of a particular drug. New agents could be tested for their ability to eliminate the tumorigenic (cancer stem cell) component of tumors from multiple patients, allowing the agents that have the greatest curative potential to proceed to human clinical trials.