Nature recently published a paper by Sean Morrison and others finding that melanoma stem cells are not rare and that standard assays to identify tumorigenic cells fail to detect a large portion of them. This prompted two letters describing an earlier study by David Taussig and others which found that the antibodies used to detect the leukemogenic cells first identified by John Dick changed their behaviour. Another letter pointed out the role that the extracellular matrix plays in shielding transplanted cells from the immune response, and suggested that this could provide insight in developing immune-based therapies to cancer.
Here, we publish that correspondence along with replies from David Taussig, which describes evidence for that cancer stem cell hypothesis, including his own evidence that leukemia-initiating cells are less than 1 in 100 cells. Finally, a reply by John Dick and colleagues says that the effects described by Taussig do not apply to a key leukemogenic cell marker and goes on to describe misconceptions about the cancer stem cell model.
Read those letters below. Here are links to NPG’s research and other articles on cancer stem cells.
Cell-sorting techniques skew cancer stem cell results
I read your news article about cancer stem cells in the latest issue of Nature  with considerable interest as an agnostic bystander. The recent paper you discussed from Sean Morrison’s laboratory calls into question the relative rarity of ‘cancer stem cells’ in human melanoma . This is of particular interest because Morrison himself was one of the chief proponents of the cancer stem cell hypothesis . This hypothesis initially gained favour because of careful experiments in John Dick’s laboratory, in which human acute myelogenous leukaemia (AML) cells were transplanted into immunocompromised mice . The observed frequency of cancer-initiating cells in these experiments was remarkably low (~1 in 250,000). However, when murine AML cells were transplanted into murine recipients in the laboratories of Mike Cleary and Andreas Strasser, the observed frequency of cancer-initiating cells was very high (~1 in 3) [5,6]. Together these results suggested that inefficient xenografting of human AML cells into mice might explain this discrepancy, raising some doubts about the cancer stem cell hypothesis.
Further evidence for a rare cancer stem cell subpopulation in human AML was provided by fluorescence-activated cell sorting (FACS) using antibodies directed against cell-surface proteins to prospectively enrich for cancer-initiating cells . This type of experiment has since been used again and again to identify so-called cancer stem cells in a variety of different types of human cancer, including breast cancer . However, a very interesting set of recently published experiments have now called the entire FACS-then-transplant methodology into serious question . In particular, antibodies used in FACS have themselves been shown to dramatically affect the in vivo survival of transplanted leukemic cells, as well as normal haematopoietic cells. Remarkably, pretreatment of recipient mice with immunosuppressive antibodies permitted ‘non-cancer stem cells’ to now behave as ‘cancer stem cells’. This paper has not received the attention it properly deserves, presumably because it went against the current ‘wisdom’ of the field and therefore wasn’t published in a ‘major’ journal. In particular, neither the recent paper from Morrison’s lab nor the two commentaries in the same issue of Nature cited this interesting and important study.
It has been known for decades that cancer cells have stem cell–like properties of self-renewal (the artist formerly known as ‘immortality’). The initial formulation of the cancer stem cell hypothesis therefore held that most cancers must arise from normal stem cells . This is clearly not the case in a variety of human cancers and animal models. It now appears likely that the remainder of the cancer stem cell hypothesis may also fall by the wayside. In an era of limited resources for biomedical research, one therefore wonders how much time, effort and money should continue to be spent in this area.
Professor of Pathology and of Genetics
Department of Pathology, Room L217
Stanford University School of Medicine
1. Baker, M. Nature 456, 553 (2008).
2. Quintana, E., Shackleton, M., Sabel, M. S., Fullen, D. R., Johnson, T. M. & Morrison, S. J. Nature 456, 593–598 (2008).
3. Reya, T., Morrison, S. J., Clarke, M. F. & Weissman, I. L. Nature 414, 105–111 (2001).
4. Lapidot, T. et al. Nature 367, 645–648 (1994).
5. Somervaille, T. C. & Cleary, M. L. Cancer Cell 10, 257–268 (2006).
6. Kelly, P. N., Dakic, A., Adams, J. M., Nutt, S. L. & Strasser, A. Science 317, 337 (2007).
7. Al-Hajj, M. et al. Proc. Natl Acad. Sci. USA 100, 3983–3988 (2003).
8. Taussig, D. C. et al. Blood 112, 568–575 (2008).
FACS-sorting finds false negatives for cancer stem cells
I read with interest the commentaries by Connie J. Eaves  and Monya Baker  on the article entitled “Efficient tumour formation by single human melanoma cells” by Quintana et al. 
The article points out the problems involved with the NOD/SCID mouse system, which unfortunately has become the standard for isolating and studying the so-called cancer stem cells (CSCs) from different solid tumours including breast , brain , prostate , pancreas  and colon. By using the more highly immunocompromised NOD/SCID IL-2 receptor-gamma chain–null (Il2rg–/–) mice as the xenotransplantation system, Quintana et al. have unequivocally demonstrated the higher percentage of cancer-initiating cells in human melanoma tumours by several orders of magnitude, constituting an increase from about 0.1–0.001% in NOD/SCID mice to about 25–27% in the NOD/SCID Il2rg–/– mouse test system. In addition, the authors have also presented data to support the heterogeneous nature of the tumour-initiating cells as opposed to the concept of the CSC hypothesis, which assumes that there is only one type of CSC in a given tumour . It is evident that this test system is very inefficient in detecting the number of cancer-initiating cells from human tumour samples. Yet the authors, as well as the commentators, are not willing to let go of the CSC hypothesis in their discussions.
The main reason for this is the original data on the human acute myeloid leukaemia (hAML) system with NOD/SCID mice . Even in the hAML system, the standard procedure for isolating CSCs is to stain the cells by coating them with fluorescence-conjugated antibodies based on the expression of antigens in order to sort the cells and transplant the different fractions into NOD/SCID mice to test for their tumour-repopulating ability. Such studies indicated that the CD34+CD38– fraction of cells were enriched with human repopulating haematopoietic stem cells (hHSCs) and hAML cells in NOD/SCID mice. In hAML samples, the CD34+CD38– fraction of cells (0.1–0.0001%) showed leukemic graft [9,11]. This original data became the basis of the CSC hypothesis, which was interpreted to mean that CD34+CD38– cells were the only cells capable of yielding leukaemia in the transplanted mice, whereas the other leukemic cells did not have this ability.
These antibodies are generally assumed to be neutral in terms of their effects on cells. Contrary to these expectations, Taussig et al. have recently shown that anti-CD38 antibodies have a profound inhibitory effect on normal and leukemic repopulating cells ; this was due to an Fc-mediated effect. When this was overcome by treating mice with immunosuppressive antibodies, CD34+CD38+ cells of certain hAML samples contained all or most of the leukaemia-initiating cells. In addition to the increase in the frequency of leukaemia-initiating cells, their data, just like the data on melanoma cells from Quintana et al., indicate a greater heterogeneity in the leukemic stem cell composition consisting of at least three different phenotypes of leukaemia-initiating cells that were capable of repopulating leukaemia in the NOD/SCID mouse system. The presence of heterogeneity in the leukaemia-initiating cell population not only suggests a potential differential origin and progression of hAML, but also has important implications for the development of new therapies aimed at eradicating the leukaemia-initiating cells. These data indicate that even the hAML system, upon which the CSC hypothesis was based, turned out to consist of multiple cancer-initiating cell phenotypes that could function as leukaemia-initiating populations in transplanted mice.
These two articles put together address the inadequacies of the CSC hypothesis, including the inability of the NOD/SCID mice to support the growth of human tumour cells , the error in the design of a test system that did not foresee the effect of the Fc portion of the antibodies on cancer-initiating cells  and the heterogeneity in the cancer-initiating cells within a given tumour . It appears that the CSC hypothesis was based on an artefact of unexpected error in the experimental design. It is very unfortunate that the studies on CSCs in different solid tumour systems, mentioned above, are subject to scepticism and, therefore, need to be validated. Thus, the question arises: are we back to the stochastic nature of cancer-initiating cells as opposed to the hierarchical nature of cancer cells, a concept supported by the CSC hypothesis .
Department of Medicine, Division of Haematology
1. Eaves, C. Nature 456, 581 (2008).
2. Baker, M. Nature 456, 553 (2008).
3. Quintana, E. et al. Nature 456, 593 (2008).
4. Al-Hajj, M. et al. Proc. Natl Acad. Sci. USA 100, 3983 (2003).
5. Singh, S. K. et al. Nature 492, 396 (2004).
6. Collins, A. T. et al. Cancer Res. 65, 10946 (2005).
7. Li, C. et al. Cancer Res. 67, 1030 (2007).
8. Dalerba, P. et al. Proc. Natl Acad. Sci. USA 104, 10158 (2007).
9. Wang, J. C. & Dick, J. E. Trends Cell Biol. 9, 494 (2005).
10. Bonnet, D. & Dick, J. E. Nature Med. 3, 730 (1997).
11. Blair, A., Hogge, D. E., Ailles, L. E., Lansdorp, P. M. & Sutherland, H. J. Blood 89, 3104 (1997).
12. Taussig, D. C. et al. Blood 112, 568 (2008).
13. Dick, J. E. Proc. Natl Acad. Sci. USA 100, 3547 (2003).
David Taussig response: Results suggests heterogeneity of leukaemogenic cells
The cancer stem cell hypothesis states that tumour cells are functionally heterogeneous with a subpopulation capable of initiating and driving the disease. The cells of this subpopulation have been termed cancer stem cells or cancer-initiating cells. The rest of the tumour cells are thought to be functionally inert with a limited ability to replicate. (The cancer stem cell hypothesis does not say that all cancer stem cells are derived from normal stem cells.) This hypothesis has been called into question by a number of recent studies that suggest the cancer-initiating cell frequency is much higher than estimates derived from studies in NOD/SCID mice previously indicated. Sean Morrison’s paper on the efficient induction of melanoma tumours with small numbers of unfractionated primary melanoma cells provides strong evidence that a significant minority (approximately 25%) of melanoma cells are tumour-initiating cells . They improved the xenotransplant model by employing a more immunodeficient mouse than the NOD/SCID mouse and using Matrigel to support the transplanted cells. We have recently published a paper showing that some antibodies used to sort tumour cells into fractions before transplanting them into NOD/SCID mice can eliminate leukaemia-initiating cells from acute myeloid leukaemia (AML) samples . Of note, anti-CD38 antibody (that had been used in key publications to isolate leukaemia-initiating cells ) profoundly reduced the number of engrafted cells in the bone marrow.
Two commentators suggest that these two publications put a nail in the coffin of the cancer stem cell hypothesis. However, there is evidence from other sources to support the hypothesis. First, morphological analysis of some primary human tumours supports the existence of nonreplicating terminally differentiated tumour cells. In chronic myeloid leukaemia, the majority of tumour cells are well-differentiated neutrophils and myelocytes. I am unaware of any evidence to suggest that these neutrophils and myelocytes have the ability to self-renew. In addition, as mentioned in the paper by Quintana et al., the differentiated cells from germ cell tumours can survive curative chemotherapy.
Second, we have unpublished data derived from limiting dilution analysis of primary AML samples in the more immunodeficient NOD/SCID/beta2-microglobulin null mice (this strain has a similar immune defect to the mice used in Sean Morrison’s study) that suggests that less than 1 in 100 AML cells can act as leukaemia-initiating cells.
Third, most patients with AML attain a complete remission following induction chemotherapy, and yet many patients relapse, indicating that the bulk of tumour cells are sensitive to chemotherapy while resistant subpopulations exist. Although some relapses occur due to the presence of resistant subclones, some relapses occur with the same genetic lesions as the predominant presentation clone . This latter situation strongly supports functional heterogeneity within a single clone and is consistent with the cancer stem cell hypothesis. In contrast with leukaemia, melanoma is highly resistant to systemic therapy, with very few remissions observed. One would predict this situation when cancer stem cells make up a significant proportion of tumour cells.
Fourth, both commentators indicate that our data are against the cancer stem cell hypothesis, and yet the effects of anti-CD38 antibody were also seen on normal haematopoietic tissue. Does this mean that normal haematopoietic stem cells do not exist either? We interpret our data to indicate that the phenotype of leukaemic stem cells is more heterogeneous than the original publication suggested . This is entirely consistent with the genetic and clinical heterogeneity of AML.
Finally, I would like to add that although there are problems with the original NOD/SCID model, a number of improvements to the model have been developed (e.g., the use of more immunodeficient variants, IVIG (intravenous immunoglobulin) conditioning, etc. ) that allow the better study of primary tumours in vivo and will answer some of these important questions.
Department of Medical Oncology
St. Bartholomew’s Hospital
1. Quintana, E et al. Nature 456, 593 (2008).
2. Taussig, D. C. et al. Blood 112, 568 (2008).
3. Bonnet, D. & Dick, J. E. Nature Med. 3, 730 (1997).
4. Ley, T. J. et al. Nature 456, 66–72 (2008).
John Dick et al Response: Cancer stem cell model misunderstood
The recent study by Quintana et al.  has renewed the debate regarding the validity of the cancer stem cell (CSC) model. Evidence supporting the existence of CSCs distinct from bulk, nontumorigenic cancer cells was first provided for haematologic malignancies, [2,3] and since then many groups have used those same approaches to look for the existence of CSCs in solid tumours as well. These approaches by necessity involve xenotransplantation of human cancer cells to assay their ability to initiate and sustain tumour growth in vivo. These assays have limitations, including compromise of cancer cell function by pretransplantation manipulations, inability to provide an ideal (natural) microenvironment for tumour growth and prevention of tumour growth by the residual immunity of recipient mice; thus, study results must be interpreted with caution. However, some have gone a step further, taking results such as those from the Quintana report to indicate that the seeming existence of CSCs in human cancer is simply an artefact of the experimental system. One must be careful, however, not to generalize results from one tumour type (in this case, melanoma) to all forms of cancer. Furthermore, arguments against the existence of CSCs are commonly rooted in misconceptions regarding the core tenets of the model and often raise issues with concepts that are not central to the CSC hypothesis.
It is often assumed that the CSC model requires CSCs to be rare and that the high frequency of cancer-initiating cells in some cancer models [1,4] raises doubts about the CSC model’s validity. In fact, the CSC hypothesis was originally proposed as an alternative to the stochastic model to explain the observed functional heterogeneity within tumours,  and it does not make any predictions as to frequency. Indeed, it is becoming clear that the frequency of cells with cancer-initiating ability varies significantly with tumour type (for example, they appear to be more frequent in lymphoid versus myeloid haematologic malignancies ). The CSC model proposes as its fundamental principle that CSCs are biologically distinct from bulk tumour cells that do not have the ability to initiate and sustain tumour growth, and therefore is not applicable to tumours in which there is a minimal cell function hierarchy, or no hierarchy at all. The steepness of the cell hierarchy within tumours will likely reflect the characteristics of the tissue from which the tumour arose. If nearly every cell in the tumour can act as a CSC, then there is no need to invoke a model to explain heterogeneity — such tumours can be considered functionally and biologically homogeneous. The results of the Quintana study, in which the authors were unable to separate melanoma cells into tumorigenic and nontumorigenic fractions, highlight the possibility that not all human cancers follow the CSC model.
Results in human cancers have been paralleled by studies in mice. In a recent report of three different murine breast cancer models, the frequency of tumorigenic cells in bulk (Lin–) tumours, assessed by syngeneic transplantation of cancer cells and thus bypassing many of the obstacles presented by xenotransplant assays, varied from 1/112 (MMTV-neu) to 1/1090 (p53/–) . Furthermore, cancer cells could be fractionated based on CD61 expression into tumorigenic and nontumorigenic fractions for MMTV-Wnt-1 and some p53/– tumours, but not for MMTV-neu tumours. Similar data have been obtained in murine leukaemia models: some follow the CSC model and some do not [4,6]. Overall, these results demonstrate the inherent variability of tumour biology and indicate that the CSC model should not be considered universal, either in mice or humans.
Another oft-mentioned misconception is that the CSC hypothesis proposes the existence of only one type of CSC with a static phenotype in a given tumour. On the contrary, the concept of clonal evolution of the cancer-initiating cell subset is not excluded by the CSC hypothesis — rather, evolution of the CSC phenotype during tumour progression and/or relapse may present a ‘moving target’ for therapy and thus an obstacle to easy eradication. A recent report of an experimental model of human leukaemia provided evidence that CSCs can evolve during tumour progression . It should be noted that the CSC immunophenotype may vary within a given tumour subtype, as we have observed in recent experiments with acute myeloid leukaemia samples. This is not unexpected given the aberrant developmental pathways present in malignant cells. The critical test is whether cancer-initiating cells can be identified and prospectively separated from nontumorigenic cells — an affirmative answer negates the stochastic model.
A recent report by Taussig et al.  has raised questions regarding the validity of experiments performed using monoclonal antibodies to fractionate tumour cells prior to transplantation into mice. Residual immunity in recipients appears to induce an antibody-dependent cell-mediated cytotoxicity–type reaction to anti-CD38 antibodies, preventing antibody-coated cells (specifically CD34CD38 acute myeloid leukaemia cells) from engrafting and thus reading out as cancer-initiating cells, although this effect appears to be attenuated following intraosseous versus intravenous transplantation. This study obliges a re-examination of previous data and careful design of future experiments. However, this immune-mediated clearance was not observed with anti-CD34 antibodies; thus, the original data from human acute myeloid leukaemia demonstrating a clear hierarchy between CD34 and CD34– leukemic cells  is not contradicted by the Taussig report. In addition, this argument cannot be invoked in the many tumour types in which CSCs have been found in the antigen-positive cell fraction (e.g. CD133 or CD44 CSCs in colon cancer, CD133 CSCs in brain cancer, CD44 CSCs in head and neck squamous cell tumours).
As a final point, we would like to reiterate the distinction between the concepts of CSCs (cells within a tumour that possess stem cell–like properties) and ‘cells of origin’ (the normal cell type in which the tumorigenic process is initiated, whether a stem cell or progenitor). The CSC model does not make any assumptions regarding the origins of CSCs — indeed, there is now abundant evidence that CSCs in mice may arise from either normal stem or progenitor cells depending on the specific transforming events, although evidence in human cancers is still lacking. Insight into the cellular origins of CSCs is not a prerequisite for their identification and characterization, but it may ultimately lead to a better understanding of their biology.
Overall, many questions about CSC biology remain unanswered, and future insights will need to come from both murine models and human xenograft assays.
Department of Medicine
Division of Medical Oncology and Hematology
Division of Developmental Biology and Stem Cells
University Health Network
Department of Surgery
Division of General Surgery
University Health Network
John E. Dick
Department of Molecular and Medical Genetics
University of Toronto
1. Quintana, E. et al. Nature 456, 593–598 (2008).
2. Lapidot, T. et al. Nature 367, 645–648 (1994).
3. Bonnet, D. & Dick, J. E. Nature Med. 3, 730–737 (1997).
4. Kelly, P. N., Dakic, A., Adams, J. M., Nutt, S. L. & Strasser, A. Science 317, 337 (2007).
5. Wang, J. C. & Dick, J. E. Trends Cell Biol. 15, 494–501 (2005).
6. Kennedy, J. A., Barabe, F., Poeppl, A. G., Wang, J. C. & Dick, J. E. Science 318, 1722 (2007).
7. Vaillant, F. et al. Cancer Res. 68, 7711–7717 (2008).
8. Barabe, F., Kennedy, J. A., Hope, K. J. & Dick, J. E. Science 316, 600–604 (2007).
9. Taussig, D. C. et al. Blood 112, 568–575 (2008).
Tumour-initiating cells vary in vulnerability to immune system
A gloomy message was published in Nature that melanoma tumour–initiating cells are far more abundant than so far anticipated . Based on previous data, fewer than 1 in 1 million cells was a reasonable estimate of tumour-initiating cells. This number of tumour-initiating cells has increased spectacularly due to an assay using more immunocompromised mice and embedding the tumour cells in Matrigel. Now, one out of four melanoma cells is able to initiate tumours. This study clearly demonstrates that heterogeneity exists in tumours: one population of cells initiates tumours due to lack of immune surveillance whereas a less abundant population resists a better equipped immune system.
Embedding melanoma cells into a matrix shields them from immediate immune elimination because the immune cells do not easily penetrate the matrix. This has been shown for collagen scaffolds. Trapping melanoma cells in a matrix shields them from inflammation and thereby from immune cell invasion, giving them more opportunities for outgrowth. An important and hopeful message to patients from this study is that tumour initiation is highly dependent on the status of the immune system and the accessibility of the aberrant cells by immune cells.
By improving the immune status of patients, a spectacular decrease in tumour-initiating cells can be foreseen. It remains to be seen if the most resistant tumour-initiating cells are vulnerable to immune attack or whether those cells escape by being shielded from the immune system.
Department of Tumor Immunology
Radboud University Nijmegen Medical Centre
1. Quintana, E. et al. Nature 456, 593 (2008).