Search tips
Search criteria 


Logo of hhmipaabout author manuscriptssubmit a manuscriptHHMI Howard Hughes Medical Institute; Author Manuscript; Accepted for publication in peer reviewed journal
Cancer Cell. Author manuscript; available in PMC 2011 May 16.
Published in final edited form as:
PMCID: PMC3031091

Phenotypic heterogeneity among tumorigenic melanoma cells from patients that is reversible and not hierarchically organized


We investigated whether melanoma is hierarchically organized into phenotypically distinct subpopulations of tumorigenic and non-tumorigenic cells, or whether most melanoma cells retain tumorigenic capacity, irrespective of their phenotype. We found 28% of single melanoma cells obtained directly from patients formed tumors in NOD/SCID IL2Rγnull mice. All stage II, III, and IV melanomas obtained directly from patients had common tumorigenic cells. All tumorigenic cells appeared to have unlimited tumorigenic capacity upon serial transplantation. We were unable to find any large subpopulation of melanoma cells that lacked tumorigenic potential. None of 22 heterogeneously-expressed markers, including CD271 and ABCB5, enriched tumorigenic cells. Some melanomas metastasized in mice, irrespective of whether they arose from CD271- or CD271+ cells. Many markers appeared to be reversibly expressed by tumorigenic melanoma cells.


In cancers that follow a stem cell model, phenotypically distinct tumorigenic cells form abundant and phenotypically diverse non-tumorigenic progeny in a hierarchical manner that resembles normal stem cell differentiation. In contrast to this model, our results indicate that primary cutaneous or metastatic melanomas from patients have common and phenotypically diverse tumorigenic cells that undergo reversible phenotypic changes in vivo. Most of the phenotypic heterogeneity in melanoma is therefore not associated with a loss of tumorigenic potential or organized in stable hierarchies. These data suggest a phenotypic plasticity model in which phenotypic heterogeneity is driven largely by reversible changes within lineages of tumorigenic cells rather than by irreversible epigenetic or genetic changes.


  • Tumorigenic cells in melanomas from patients are common and phenotypically diverse
  • Irrespective of their phenotype, many melanoma cells recapitulate tumor heterogeneity
  • Melanoma cells exhibit phenotypic differences that are not hierarchically organized
  • Heterogeneity arises from reversible phenotypic changes among tumorigenic cells


Cancer is a heterogeneous disease, involving differences between tumors as well as between cancer cells within the same tumor. Clonal evolution contributes to this heterogeneity as cancer cells undergo irreversible genetic changes over time, leading to functional and phenotypic differences (Nowell, 1976). Another explanation for heterogeneity within tumors comes from the cancer stem cell model, which posits that tumors are hierarchically organized, with a small subpopulation of tumorigenic cells that generates phenotypically diverse non-tumorigenic progeny in a manner similar to normal stem cell differentiation (Kleinsmith and Pierce, 1964; Lapidot et al., 1994; Reya et al., 2001). These models are not mutually exclusive in that cancers that follow the stem cell model would be expected to undergo clonal evolution.

Evidence supports the cancer stem cell model in some acute myeloid leukemias (Bonnet and Dick, 1997; Lapidot et al., 1994), chronic myeloid leukemias (Eisterer et al., 2005; Neering et al., 2007; Oravecz-Wilson et al., 2009), teratocarcinomas (Kleinsmith and Pierce, 1964), breast cancers (Al-Hajj et al., 2003), brain tumors (Read et al., 2009; Singh et al., 2004), and colon cancers (O'Brien et al., 2007; Ricci-Vitiani et al., 2007). In each cancer, markers have been identified that distinguish small, often rare, subpopulations of cancer cells that are greatly enriched for tumorigenic/leukemogenic activity as compared to unfractionated cancer cells. The same markers were concluded to distinguish tumorigenic from non-tumorigenic cells in multiple patients, suggesting these cancers adopt reproducible cellular hierarchies. Nonetheless, the robustness of some cancer stem cell markers has been questioned (Joo et al., 2008; Ogden et al., 2008; Wang et al., 2008) and it remains to be determined how generalizable the model is.

Cancer stem cell studies have consistently found that cells from non-tumorigenic/non-leukemogenic cancer cell populations are rarely able to form tumors/leukemias, even when assayed under conditions permissive for tumorigenesis by small numbers of cancer stem cells (Al-Hajj et al., 2003; Bonnet and Dick, 1997; Lapidot et al., 1994; O'Brien et al., 2007; Oravecz-Wilson et al., 2009; Read et al., 2009; Ricci-Vitiani et al., 2007; Singh et al., 2004). In cancers that follow this model, non-tumorigenic cells have therefore irreversibly lost tumorigenic capacity or only regain this capacity under rare circumstances. The cancer stem cell and clonal evolution models have thus emphasized the role of irreversible epigenetic and genetic changes in determining heterogeneity among cancer cells.

On the other hand, recent studies performed in cancer cell lines have suggested that some phenotypic and functional attributes of tumorigenic cells can reversibly turn on and off (Mani et al., 2008; Pinner et al., 2009; Roesch et al., 2010; Sharma et al., 2010). This raises the question of whether reversible changes are observed in primary cancers from patients and whether many or few cells in these cancers can undergo such changes. If most cells in a cancer can reversibly gain and lose competence to form a tumor, then this is a transient state rather than a hierarchically determined attribute possessed only by rare cancer stem cells. These studies also raise the separate question of whether phenotypic heterogeneity in patient tumors is driven mainly by irreversible or reversible phenotypic changes.

The growth and metastasis of melanomas has been proposed to be driven by a small subpopulation of melanoma stem cells that can be distinguished from non-tumorigenic melanoma cells based on the expression of ABCB5 (Schatton et al., 2008) or CD271 (Boiko et al., 2010). However, the xenotransplantation experiments that were the basis for these conclusions were performed with assays in which an average of only 1 in 50,000 (see Suppl. Table 3 in (Boiko et al., 2010)) to 1 in 1,090,000 (Schatton et al., 2008) unfractionated melanoma cells form tumors. We have found that the frequency of tumorigenic melanoma cells that can be detected after xenotransplantation is highly assay-dependent (Quintana et al., 2008). By making a series of changes in assay conditions, we increased the detectable frequency of tumorigenic melanoma cells by several orders of magnitude. We found that 1 in 4 cells obtained directly from patients with primary cutaneous or metastatic melanomas are able to form tumors after xenotransplantation into NOD/SCID IL2Rγnull (NSG) mice (see Figure 3c in (Quintana et al., 2008)). The conclusion that many melanoma cells are capable of forming tumors has been independently confirmed in primary mouse melanomas (Held et al., 2010) and in human melanoma cell lines (Roesch et al., 2010).

Although many melanoma cells are able to form tumors, it is critical to assess whether melanoma is hierarchically organized into phenotypically distinct subpopulations of tumorigenic and non-tumorigenic cells. We have found that both CD133+ and CD133- melanoma cells form tumors that exhibit similar heterogeneity in CD133 expression, suggesting that CD133 is reversibly expressed by tumorigenic melanoma cells rather than distinguishing cells at different levels of a hierarchy (Shackleton et al., 2009). The JARID1B histone demethylase regulates the tumorigenic activity of melanoma cell lines and can also reversibly turn on and off, raising the possibility that competence to form a tumor is reversible (Roesch et al., 2010). Brn2 and pigment are also reversibly expressed by cells from melanoma cell lines as they metastasize in vivo (Pinner et al., 2009). These studies raise the question of whether reversible phenotypic plasticity is observed among many, or few, markers in melanomas obtained from patients. We addressed this by performing extensive functional and phenotypic analyses of melanoma cells obtained from patients with stage II, III, and IV disease.


We previously reported that 69 of 254 (27%) single, unselected melanoma cells, isolated from four patient melanomas that had been passaged as xenografts in mice, formed tumors after injection into NSG mice (Quintana et al., 2008). We have now extended these studies to evaluate an additional 210 single, unselected cells from melanomas that were either obtained from other xenografts (n=3 patient tumors, up to two passages in mice) or directly from patients with stage III disease (n=5 patient tumors). In each case, 15–50% of single cells formed subcutaneous tumors in NSG mice. Overall, 62 of 210 (30%) single cells formed tumors, including 28% of single cells obtained directly from patients (Table 1). These are minimum estimates of the frequency of tumorigenic melanoma cells as additional assay improvements could further increase the frequency of tumorigenic cells that can be detected.

Table 1
Summary of tumor formation by single cells or by limit dilution analyses of melanoma cells from 16 stage II, III, and IV patients

We also performed limit dilution assays on cells obtained directly from 11 additional patients with primary cutaneous (stage II or III) or metastatic (stage III or IV) melanomas (Table 1). The stage II melanoma (Breslow depth 3.3 mm, no ulceration, 2 mitoses/mm2), obtained directly from a patient, contained a high frequency of tumorigenic cells, with 5 of 6 injections of 10 unselected cells forming tumors. Primary cutaneous melanomas obtained directly from early stage patients with a good prognosis can therefore contain a high frequency of tumorigenic cells, though additional studies will be required to determine whether some stage I or II melanomas contain less frequent tumorigenic cells. The melanomas obtained directly from stage III (n=9) and IV (n=1) patients also contained high frequencies of tumorigenic cells. We have thus examined the frequency of cells with tumorigenic potential in melanomas obtained from 18 different patients (see Figure S1 for patient details). We observed high frequencies of tumorigenic cells in every case, irrespective of whether the melanomas were obtained directly from patients (n=16 cases) or xenografted (n=3), and irrespective of whether they were primary cutaneous melanomas (n=3), cutaneous or subcutaneous metastases (n=4), or regional lymph node metastases (n=12).

No correlation between tumorigenic cell frequency and tumor growth rate

A prediction that is commonly made based on the cancer stem cell model is that aggressively growing tumors contain a higher frequency of tumorigenic cells. Nonetheless, this prediction has rarely, if ever, been tested. To evaluate this, we transplanted single cells from melanomas obtained from 12 patients (see Figure S1) into NSG mice then compared the growth rates of the resulting tumors with the percentage of single cells that were tumorigenic. Tumors that arose from some patients grew quickly (patients 205, 214 and 491), while tumors from other patients grew slowly (patients 405, 481, 487, 501, 526, 528, 530, 534 and 600) (Figure 1A). Although fast-growing tumors grew an average of 3 times faster than slow growing tumors in NSG mice (1.52 ±0.64 vs 0.50±0.30 mm/week; p<0.0001), there was no correlation with the frequency of tumorigenic cells in NSG mice (Figure 1B). The melanoma that gave rise to the fastest growing tumors (205) also had the lowest frequency (13%) of tumorigenic cells, whereas a melanoma that gave rise to slow-growing tumors (481) had a remarkably high frequency (70%) of tumorigenic cells. Tumors derived directly from patients tended to grow more slowly than tumors derived from xenografts, but these tumors did not tend to have lower frequencies of tumorigenic cells (Figure 1A). Differences in the growth rates of melanomas from different patients in NSG mice are not determined by differences in tumorigenic cell frequency.

Figure 1
Lack of correlation between growth rate and tumorigenic cell frequency in human melanomas

All tumorigenic melanoma cells appear capable of indefinite proliferation

To test whether the proliferative capacity of some tumorigenic melanoma cells may be limited, as observed in some human leukemia-initiating cells (Hope et al., 2004), we formed melanomas in NSG mice from single cells derived from 6 patients (32 tumors arose from the injection of 58 single cells). We then serially transplanted 100 cell aliquots from each tumor into secondary and tertiary NSG mice. All 32 single-cell derived tumors gave rise to secondary tumors, and all 31 of the tumors from which tertiary transplants were attempted gave rise to tertiary tumors (Figure 2A, B; one secondary tumor could not be retransplanted due to the unexpected death of the mouse). In all cases, the single melanoma cells appeared to be capable of unlimited growth, necessitating euthanasia of the mice. Tumor growth rate significantly increased in a few cases between secondary and tertiary transplants (see * in Figure 2E-H, p<0.05; primary tumor growth rates could not be compared to secondary tumors because primary tumors were initiated with single cells while secondary and tertiary tumors were initiated with 100 cells). In no case did the rate of tumor growth decrease during serial passaging (Figure 2C-H). We have thus been unable to detect any tumorigenic melanoma cells that have limited tumorigenic potential in vivo.

Figure 2
Tumorigenic human melanoma cells exhibit indefinite proliferative potential upon serial transplantation

Both ABCB5- and ABCB5+ melanoma cells can form tumors

We previously published (Quintana et al., 2008) that four markers (CD133, CD166, L1-CAM and CD49f) were not able to distinguish tumorigenic from non-tumorigenic melanoma cells. We have now tested 18 additional markers (ABCB5, CD271, MCAM, E-Cadherin, N-Cadherin, c-kit, CD29, CD44, CD49d, CD49b, A2B5, HNK1, CD54, CD9, CD151, CD10, L6 and CD49e) that we found to be heterogeneously expressed among human melanoma cells out of 85 markers that we examined (for a summary of all tumors and markers see Figure S1 and Table S2). We present data in greatest detail for ABCB5 (Figure 3) and CD271 (Figure 4).

Figure 3
Staining with an antibody against ABCB5 is variable among melanomas and does not distinguish tumorigenic from non-tumorigenic cells in NSG mice
Figure 4
CD271 expression does not correlate with the frequency of tumorigenic cells and does not enrich tumorigenic melanoma cells

We evaluated ABCB5 expression by flow-cytometry in melanomas from 9 patients using dissociation conditions and antibody provided by Schatton et al. (2008). We did not independently test the specificity of this antibody but rather relied upon published results (Frank et al., 2003). We detected staining with this antibody in a minority of cells from 4 of these tumors, ranging from 2.9 to 5.3% of cells (Figure 3A). Since the detectable frequency of ABCB5+ cells (0 to 5.3%) was much lower than the frequency of tumorigenic cells (11 to 70%) in all tumors, cells with tumorigenic potential could not reside exclusively within the ABCB5+ subpopulation (Figure 3A). We observed no significant difference in the frequency of tumorigenic cells between melanomas with ABCB5+ cells and melanomas without ABCB5+ cells (27±13% and 27±25% respectively; mean±SD, p=0.97).

To compare the tumorigenic potential of ABCB5- and ABCB5+ melanoma cells, we separated these cells by flow cytometry from tumors derived from 3 different patients (Figure 3B, C) and transplanted them into NSG mice with Matrigel. All injections of 50 cells, and most injections of 10 cells, formed tumors, irrespective of whether unfractionated, ABCB5- cells, or ABCB5+ cells were injected (Figure 3D). We were therefore unable to detect any difference in the tumorigenic capacity of ABCB5- and ABCB5+ melanoma cells.

Both CD271- and CD271+ melanoma cells can form tumors

We also examined CD271 expression in primary cutaneous and metastatic melanomas from 20 patients (Figure 4A). CD271 expression was analyzed in some tumors immediately after removal from the patient (n = 13), and in other tumors after ≤2 passages as a xenograft (n = 9; tumors from some patients were examined both ways). CD271 was heterogeneously expressed in every tumor tested (Figure 4A, C), but the range of expression was large: 1% to 91% of cells were CD271+. Indeed, CD271+ cells accounted for >50% of the cells in 4 of 13 tumors obtained directly from patients (Figure 4A). Overall, we observed no correlation between the frequency of CD271+ cells and the frequency of tumorigenic cells (r2=0.004, Figure 4B). In some cases, the frequency of CD271+ cells was much lower than the frequency of tumorigenic cells (see xenografts from patients 481, 487, 526, 534 and tumor directly from patient 610), while in other cases CD271+ cell frequency was much higher than the frequency of tumorigenic cells (see xenografts from patients 491 and 530). This suggests that neither the CD271+ nor the CD271- cell fraction can consistently contain all tumorigenic melanoma cells.

To directly compare the tumorigenic potential of CD271- and CD271+ melanoma cells, we isolated these cells from xenografted tumors derived from three patients (491, 526, and 534; ≤2 passages; Figure 4C, D), and injected into NSG mice with Matrigel. In each case, tumors arose from only ten CD271- or ten CD271+ cells and were at least as likely to arise from CD271- as from CD271+ cells (Figure 4D). We also compared the tumorigenicity of CD271- and CD271+ cells from xenografts from three other patients in NSG mice, but this time without Matrigel. Again, both CD271- and CD271+ cells readily formed tumors and neither fraction was enriched for tumorigenic activity (Table S1).

We performed similar experiments using CD271- and CD271+ cells from six melanomas obtained directly from patients, including two primary cutaneous tumors, one from a patient with good prognosis stage II disease (Figure 4 A,C,D). Similar to xenografted tumors, tumors arose readily from CD271- and CD271+ cells from the four metastatic melanomas (patients 600, 608, 631, 641), (Figure 4D). In the two primary cutaneous melanomas (patients 597 and 610), tumors were more likely to arise from CD271- cells (Figure 4D). This is reminiscent of observations from primary mouse melanomas, which also exhibited tumorigenic activity from both the CD271- and CD271+ fractions, but more tumorigenic activity from CD271- cells (Held et al., 2010). In our experiments, the less-tumorigenic CD271+ cells were a minor subpopulation (2 to 12% of cells) in primary cutaneous melanomas, rather than the bulk of tumor cells. It remains unclear whether the reduced tumorigenic capacity of CD271+ cells in these experiments was determined by genetic, epigenetic, or environmental differences.

Both CD271+ and CD271- cells form tumors that metastasize

We extended our studies to evaluate spontaneous metastasis by melanoma cells. Transplanted melanoma cells could disseminate as a result of inadvertent intravascular injection rather than from spontaneous metastasis; therefore, we first tested whether metastases arose in mice injected subcutaneously with single melanoma cells. NSG mice with subcutaneous tumors that arose from single cells from melanoma 481 were euthanized and examined as tumors approached 20 mm diameter. Macroscopic metastases, involving lymph nodes and/or visceral organs, were visible in most mice (see Figure 5A-H). This demonstrates that metastases can develop spontaneously in NSG mice with xenografted human melanomas.

Figure 5
Spontaneous metastasis of human melanomas in NSG mice irrespective of whether they derived from CD271- or CD271+ melanoma cells

To test whether CD271- and CD271+ melanoma cells form tumors with different metastasis potentials, we established subcutaneous tumors in NSG mice from CD271- or CD271+ melanoma cells isolated directly from patient 608. Tumors were allowed to grow until they approached 20 mm in diameter or until the mice became ill. When the mice were euthanized, the tumors derived from CD271- and CD271+ cells had mean±s.d. diameters of 16±8 mm and 17±9 mm (p=0.66), respectively, and mean times since injection of cells of 26±6 weeks and 26±5 weeks (p=0.85). Metastases were similarly evident in the kidneys and lungs of all mice that developed subcutaneous tumors, irrespective of whether the tumors derived from CD271- or CD271+ cells (Fig 5I-Q). CD271- and CD271+ melanoma cells from a xenograft derived from patient 205 were also injected in NSG mice, in this case without Matrigel. In this experiment, distant metastases were evident in abdominal structures (mesentery, ovaries, fallopian tubes, para-renal tissues) of 3/6 mice with tumors arising from CD271- cells and in 2/5 mice with tumors arising from CD271+ cells (Figure 5R). Some human melanomas spontaneously metastasize in NSG mice, irrespective of whether they arise from CD271- or CD271+ cells.

Other markers fail to identify non-tumorigenic melanoma cells

To identify markers that are heterogeneously expressed by melanoma cells, we screened antibodies against 85 cell surface markers in melanomas from 3 to 19 patients per marker (see Table S2). We were unable to detect the expression of 32 antigens, including CD34 and SSEA-1 (CD15). Other markers appeared to be nearly uniformly expressed by melanoma cells, including CD63, which was only heterogeneously expressed in 4 of 13 tumors. Beyond ABCB5 (Figure 3), CD271 (Figure 4), and the four heterogeneously-expressed markers we tested previously (CD133, CD166, L1-CAM and CD49f; (Quintana et al., 2008)), we also compared the tumorigenic potential of melanoma cells that differed in the expression of MCAM, E-Cadherin, N-Cadherin, c-kit, CD29, CD44, CD49d, CD49b, A2B5, HNK-1, CD54, CD9, CD151, CD10, L6, and CD49e (Figure 6). In each case, marker-negative/low and marker-positive/high cells were separated by flow cytometry from tumors derived from two to three different patients (Figure 6A), and 10 cell aliquots were injected subcutaneously into NSG mice. In no case did we observe clear enrichment of tumorigenic activity in any fraction of cells (Figure 6B). In 2 of 3 experiments evaluating L6, tumors were more likely to arise from the L6- fraction of cells; however, the less-tumorgenic L6+ fraction accounted for only 9% of cells. We have therefore been unable to identify any large subpopulation of melanoma cells that lacks tumorigenic activity or any small subpopulation that is enriched for tumorigenic activity.

Figure 6
None of sixteen heterogeneously expressed markers distinguished tumorigenic from non-tumorigenic melanoma cells

We also compared the growth rates of tumors to test whether some markers distinguished tumorigenic fractions with intrinsically different growth rates. We compared the rate at which tumors grew from 10 unfractionated cells, 10 marker-negative/low cells, or 10 marker-positive/high cells for 22 different markers: MCAM, E-Cadherin, N-Cadherin, ckit, CD29, CD44, CD49d, CD49b, A2B5, HNK-1, CD54, CD133, CD166, L1-CAM, CD49f, ABCB5, CD271, CD49e, CD9, CD10, CD151, and L6 (Figure S2). For each marker, a total of 6 to 28 tumors per cell fraction were generated from cells isolated from 2 to 4 different patients per marker. In no case did we observe a statistically significant difference in the rate at which tumors grew from marker-negative/low versus marker-positive/high fractions of cells. These data indicate there is extensive phenotypic heterogeneity among melanoma cells that does not correlate with differences in tumorigenic capacity or tumor growth rate.

Many phenotypically diverse melanoma cells can recapitulate tumor heterogeneity

The capacity of tumorigenic cells to recapitulate the phenotypic heterogeneity of a parental tumor after transplantation is considered a defining characteristic of cancer stem cells. In contrast, phenotypic heterogeneity that arises from clonal evolution might lead to cells that form phenotypically distinct tumors that do not recapitulate the parental tumor. Given that melanoma cells with many different phenotypes were capable of forming tumors (Figure 3, ,4,4, ,6),6), we tested whether these cells recapitulated the heterogeneity of parental tumors.

We evaluated the phenotypes of tumors that arose from marker-defined fractions of cells in the experiments described in Figures 3, ,4,4, and and6.6. For every marker tested, we found that secondary tumors were phenotypically similar to the parent tumor, irrespective of whether they arose from marker-negative/low cells or marker-positive/high cells. For example, melanoma cells were isolated based on ABCB5 expression from a tumor in which 3.8% of cells were ABCB5+ (Figure 7A). Both ABCB5+ and ABCB5- cells formed phenotypically similar tumors in NSG mice that recapitulated the heterogeneity of the parent tumor, with 4.2 to 5.4% of cells that were ABCB5+ (Figure 7A). Similar results were observed in experiments that involved CD166 (Figure 7B), A2B5 (Figure 7C), CD151 (Figure 7D), CD54 (Figure 7E), CD44 (Figure 7F), CD9 (Figure 7G), CD29 (Figure 7H), N-Cadherin (Figure 7I) and CD271 (Figure 7J). This indicates that many melanoma cells are capable of recapitulating tumor heterogeneity after transplantation, irrespective of their phenotype at the time of transplantation. The capacity to recapitulate melanoma heterogeneity is therefore widely shared by many phenotypically diverse melanoma cells.

Figure 7
Many phenotypically distinct fractions of melanoma cells can recapitulate the heterogeneity of the tumors from which they derive

In some experiments the marker-negative/low fraction and the marker-positive/high fraction both gave rise to heterogeneous tumors, but the tumors that arose from the positive fraction of cells contained a higher percentage of marker positive cells. This was observed in some experiments performed with CD271 (data not shown), CD49e (Figure 7K), CD49f (Figure 7L), L1CAM (Figure 7M), E-cadherin (Figure 7N), and c-kit (Figure 7O). This suggests that there are sometimes genetic or epigenetic differences among phenotypically distinct human melanoma cells that bias the phenotype of the progeny they generate without preventing tumorigenesis or the re-establishment of heterogeneity. However, this was only observed in only a minority of experiments, suggesting that such biases are not a major driver of phenotypic heterogeneity.

The capacity of tumorigenic cells to recapitulate the heterogeneity of parental tumors has generally been tested with only a few (1 to 3) surface markers. We wondered whether analyses of larger numbers of surface markers might reveal phenotypic differences among tumors formed by phenotypically distinct pairs of cells from the same parental tumor. To test this, we analyzed some of the pairs of daughter tumors shown in Figure 7 with panels of 13 to 16 heterogeneously expressed surface markers. ABCB5- and ABCB5+ cells not only formed secondary tumors that were indistinguishable with respect to ABCB5 expression, but these tumors were also similar with respect to CD166, CD54, L1-CAM, CD49b, CD49d, CD49f, A2B5, CD271, HNK1, MCAM, E-Cadherin, c-kit, CD44, CD133, and N-Cadherin expression (Figure S3A). Similar results were obtained with pairs of tumors formed by CD271- and CD271+ cells (Figure S3B), CD54- and CD54+ cells (Figure S3C), c-kit- and c-kit+ cells (Figure S3D), L1-CAM- and L1-CAM+ cells (Figure S3E), CD44- and CD44+ cells (Figure S3F). In a few cases, we observed differences in the percentage of cells that stained with individual markers. However, most markers exhibited similar staining patterns in secondary tumors, despite being derived from the transplantation of phenotypically distinct cells. These results suggest many markers are reversibly turned on and off within lineages of tumorigenic melanoma cells.


Our experiments suggest that melanoma does not adopt a hierarchy consisting of a minor subpopulation of tumorigenic cells and a majority population of non-tumorigenic cells. Melanomas consistently contained high frequencies of tumorigenic cells, irrespective of whether they were primary cutaneous or metastatic melanomas, whether they were from stage II, III, or IV disease, and whether they were obtained directly from patients or after xenografting (Table 1, Figures 1A, ,3A,3A, ,4A).4A). All tumorigenic cells appeared to have the capacity to proliferate indefinitely upon serial transplantation assay (Figure 2).

We have not been able to identify any marker that robustly distinguishes tumorigenic from non-tumorigenic melanoma cells despite examining 85 markers and carefully studying the tumorigenic potential of cells that differ in their expression of 22 heterogeneously expressed markers, including ABCB5 (Figure 3), CD271 (Figure 4; Table S1) and CD133 (Shackleton et al., 2009). For all of the markers we studied, tumors with similar growth rates readily arose from all fractions of cells (Figure 6; Figure S4). Despite subdividing melanoma cells using many markers we have been unable to identify any large subpopulation of melanoma cells that lacks tumorigenic potential. Thus, tumorigenic capacity is not restricted to a small subpopulation of melanoma cells but is widely shared among phenotypically diverse cells.

Many phenotypically distinct melanoma cells had the capacity to form tumors that recapitulated the phenotypic diversity of the tumors from which they derived (Figure 7; Figure S5-10). This suggests that tumorigenic cells appeared to undergo reversible changes in the expression of many markers in vivo. This contrasts with models that attribute phenotypic heterogeneity to the hierarchical differentiation of cancer stem cells into non-tumorigenic progeny or to irreversible genetic changes that arise through clonal evolution.

Our results are compatible with the idea that tumorigenic competence might reflect a reversible state in melanoma. Studies of breast cancer cell lines have suggested that tumorigenic activity correlates with the capacity to undergo an epithelial to mesenchymal transition and that cells might reversibly undergo such transitions (Mani et al., 2008). Studies of other cell lines have suggested that therapy resistance can also reflect a reversible state (Sharma et al., 2010). Recent studies of melanoma cell lines have indicated that the JARID1B histone demethylase, Brn2, and pigment are reversibly turned on and off within lineages of melanoma cells in a manner related to cell function (Pinner et al., 2009; Roesch et al., 2010). Transient exposure of glioblastoma cells to perivascular nitric oxide confers tumorigenic competence and stem cell properties, raising the possibility that these attributes reflect a reversible state in brain tumor cells (Charles et al., 2010). These studies make the prediction that in some cancers many cells will be capable of forming phenotypically diverse tumors, without robust hierarchical organization. Our study comprehensively tests this prediction in tumors from patients in vivo, finding that many phenotypic differences among melanoma cells reversibly change within lineages of tumorigenic cells rather than being hierarchically organized.

Although no marker robustly distinguished tumorigenic from non-tumorigenic melanoma cells, we observed little tumorigenic activity among CD271+ cells from two primary cutaneous melanomas. In this regard, our data are similar to results from primary mouse melanomas (Held et al., 2010) in that both studies found tumorigenic activity in a high percentage of single cells, but CD271- cells were more likely to form tumors. However, neither our results nor the results from Held et al. were consistent with the cancer stem cell model because Held et al. found that the tumorigenic cells they studied often did not recapitulate the heterogeneity of parental tumors, and the CD271+ cells with limited tumorigenic activity in two tumors in our study represented only 2 to 12% of tumor cells, a minor subpopulation of cancer cells.

Our results with CD271 are different from the results reported by Boiko et al. (2010) even though both studies used the same anti-CD271 antibody, both studied a similar spectrum of melanoma stages (mainly stage III), and both studied a combination of xenografted tumors and tumors obtained directly from patients. The most obvious potential explanation for the difference in results lies in the different assays used in the two studies: different enzymatic dissociation conditions (22 minutes in our study versus up to 3 hours in their study), different injection sites (subcutaneous versus intradermal) and different recipient mice (NSG versus Rag-/-IL2Rγ-/-). An average of 1 in 50,000 unfractionated melanoma cells formed tumors in the study performed by Boiko et al. (using 4 stage III, 1 stage IV, and 1 stage II melanoma, see Suppl. Table 3 in Boiko et al., 2010). In contrast, when we transplanted unfractionated melanoma cells directly from 6 stage III patients in our prior study, an average of 1 in 4 cells formed tumors (Quintana et al., 2008). In our current study, an average of 28% of single, unfractionated melanoma cells obtained directly from 5 stage III melanoma patients (Table 1) and 17% (1 in 6) of melanoma cells obtained directly from a stage II melanoma patient (Table 1) formed tumors. Thus, the assay we used appears to be approximately 10,000-fold more sensitive than the assay used by Boiko et al. Using this more sensitive assay, we find that CD271- cells have at least as much ability to form tumors as CD271+ cells. It will now be critical for other labs to independently assess whether they also observe tumor formation by CD271- melanoma cells.

While our results argue against the cancer stem cell model in melanoma, they do not mean that other cancers do not follow a stem cell model. We and others have found that most chronic myeloid leukemias (Eisterer et al., 2005; Jamieson et al., 2004; Neering et al., 2007; Oravecz-Wilson et al., 2009) and acute myeloid leukemias (Bonnet and Dick, 1997; Lapidot et al., 1994; Yilmaz et al., 2006) do follow a cancer stem cell model in which leukemogenic cells are rare, phenotypically distinct from the vast majority of other leukemia cells, and robustly hierarchically organized. It will be critical to determine which cancers follow a stem cell model and which do not, so therapies designed to target rare subpopulations of cells are not inappropriately tested in patients whose disease is driven by many diverse cancer cells.

If a marker is identified in future that robustly distinguishes tumorigenic from non-tumorigenic melanoma cells, melanoma would still be quite different from cancers, such as myeloid leukemia, that follow a stem cell model. Our observation that many markers are reversibly expressed by tumorigenic melanoma cells contrasts with the obvious morphologic and phenotypic differences between leukemogenic and non-leukemogenic cells. Thus, even if some melanomas do contain a hierarchy, it would be a shallow hierarchy that includes abundant and diverse tumorigenic cells rather than a steep hierarchy driven by rare tumorigenic cells, as described so far in cancers found to follow a stem cell model.

Some have suggested that cancer stem cells might be distinguished from non-tumorigenic cancer cells by reduced immunogenicity, allowing them to proliferate more extensively by escaping immune detection (Schatton and Frank, 2009). However, this hypothesis is not testable in human cancers because they cannot be transplanted autologously into patients or into immunocompetent mice. Immunocompetent mice mount a powerful xenogeneic immune response against human cells, making it impossible to assess whether a failure to engraft reflects immune rejection or an intrinsic lack of tumorigenic capacity. Furthermore, no xenotransplantation model recapitulates the autologous anti-cancer immune response that occurs in some patients against their own cancers because the xenogeneic immune response is far more powerful and driven by very different mechanisms than autologous immune responses. For these reasons, thirty years of research has found that the preferred system for studying the potential of normal human hematopoietic stem cells (Ito et al., 2002; Shultz et al., 2005) and human leukemic stem cells (Agliano et al., 2008; Sanchez et al., 2009) is NSG mice that are not only highly immunocompromised, but also irradiated to further promote the engraftment of human cells. The failure of human hematopoietic stem cells and leukemic stem cells to engraft in immunocompetent mice does not mean theses cells are normally regulated by autologous immune responses in patients because the xenogeneic immune response that rejects these cells from mice uses very different mechanisms.

To identify the spectrum of human cancer cells that have the potential to contribute to disease, these cells must be studied in highly immunocompromised mice. Once the spectrum of cells capable of contributing to disease is identified, a separate and context-dependent question concerns which of these cells are actually fated to contribute to disease in a patient. This question can only be addressed in mouse cancers because no xenograft model reflects the anti-cancer immune response, or certain other aspects of the environment, in patients.

Our results demonstrate that phenotypic heterogeneity in melanomas obtained from patients is largely driven by reversible changes in a broad range of markers that turn on and off within lineages of tumorigenic cells. This phenotypic plasticity contrasts with both the cancer stem cell and clonal evolution models, which largely attribute heterogeneity to irreversible epigenetic and genetic changes. So while clonal evolution occurs in many cancers, including melanoma, and some cancers follow a stem cell model, our results raise the possibility that pervasive phenotypic plasticity is an independent source of heterogeneity in many cancers.

Experimental Procedures

Tumor cell preparation

Melanoma specimens were obtained with informed consent from all patients according to protocols approved by the Institutional Review Board of the University of Michigan Medical School (IRBMED approvals 2004-1058 and 2000-0713). Tumors were mechanically dissociated with a McIlwain tissue chopper (Mickle Laboratory Engineering Co., Guilford, U.K.) prior to sequential enzymatic digestion in 200 U/mL collagenase IV (Worthington, Lakewood, NJ) for 20 min followed by 0.05% trypsin-EGTA for 5 min, both at 37°C. DNase (50-100 U/mL) was added to reduce clumping of cells during digestion. Cells were filtered (40μm cell strainer) to obtain a single cell suspension. Dead cells and debris were reduced by density centrifugation (1.1g/ml Optiprep™, Sigma, St Louis, MO) when necessary. To test ABCB5 expression, single cell suspensions were derived after incubation of mechanically dissociated tumor tissue in 10 ml sterile PBS containing 0.1 g/L calcium chloride and 5 ug/ml Collagenase Serva NB6 (SERVA Electrophoresis GmbH) for 3 h at 37 °C (Schatton et al., 2008).

Cell labeling and flow cytometry

All antibody staining was performed for 20 min on ice, followed by washing and centrifugation. A list of primary antibodies is in Table S2. Anti-ABCB5 antibody (clone 3C2-1D12) was a gift from Markus Frank. Secondary antibodies were conjugated to phycoerythrin (goat anti-mouse IgG or IgM, goat anti-rat IgG or goat anti-rabbit IgG; Jackson ImmunoResearch, West Grove, Pennsylvania). Primary isotype controls followed by the same secondary antibodies were used to set background. Cells were subsequently stained with directly conjugated antibodies to human CD45 (HI30-APC, BD Biosciences, San Jose, CA), human CD31 (WM59-APC, eBiosciences, San Diego, CA) and Glycophorin A (HIR2-APC, Biolegend, San Diego, CA) (for tumors obtained directly from patients) or mouse CD45 (30-F11-APC, eBiosciences), mouse CD31 (390-APC, Biolegend), mouse Ter119 (TER-119-APC, eBiosciences) and human HLA-A,B,C (G46-2.6-FITC, BD Biosciences) (for xenograft tumors) to select live human melanoma cells and to exclude endothelial and hematopoietic cells. Cells were resuspended in 10μg/ml DAPI (Sigma) and analyzed and/or sorted on a FACSAria Cell Sorter (Becton-Dickinson, San Jose, CA). After sorting, an aliquot of sorted cells was always reanalyzed to check for purity, which was usually >95%.

Transplanting melanoma cells

After sorting, cells were counted and resuspended in staining medium (L15 medium containing 1mg/ml BSA, 1% penicillin/streptomycin and 10mM HEPES (pH7.4)) with 25% high protein Matrigel (product #354248, BD Biosciences). Subcutaneous injections of human melanoma cells were performed in each flank and the inter-scapular region of NOD.CB17-Prkdcscid Il2rgtm1Wjl/SzJ (NOD/SCID IL2Rγnull, NSG) mice (Jackson Laboratories) according to protocols approved by the Committee on the Use and Care of Animals at the University of Michigan (protocol #9055). Tumor formation was evaluated regularly after injection by palpation of injection sites, and tumor diameters were measured with calipers. The presence of human melanomas was confirmed at necropsy by gross appearance, histology, and immunohistochemistry.

Injection of single melanoma cells

Single cells were isolated and identified using methods described previously (Quintana et al., 2008). Briefly, sorted melanoma cells were diluted and aliquoted into 10 μl microwells (Thermo Fisher Scientific, Roskilde, Denmark). Plates were centrifuged at 450×g for 30 seconds, and wells containing single cells were identified by phase microscopy. Cell doublets could be identified easily, were rare, and were discarded. Each single cell was drawn into a syringe containing high protein Matrigel (product #354248, BD Biosciences), the well was visually confirmed to no longer contain the cell, and the cell was injected into a NSG mouse as described above.

Histopathology and immunostaining

Portions of melanoma tumors and mouse organs used in experiments were fixed in 10% neutral buffered formalin, paraffin embedded, sectioned and stained with haematoxylin and eosin for histopathology analysis. Paraffin-embedded tumours were confirmed as melanomas by staining for S100 expression after quenching endogenous peroxidase activity. Binding of anti-S100 antibody (DAKO) was performed for 30 min at room temperature, detected by anti-rabbit secondary (30 min at room temperature) and revealed using DAB chromagen. S100-stained slides were counterstained with haematoxylin.


Tumor growth rates were determined by maximum tumor diameter (in mm) at euthanasia divided by elapsed time (in weeks) from injection. Differences between mean growth rates were compared using unpaired Student's T-tests. Correlations between tumor growth rates (Figure 1B) or CD271+ cell frequencies (Figure 4B) and tumorigenic cell frequencies were performed by linear regression analysis using Graphpad Prism 3.0 software. Limiting dilution analyses were performed using ELDA: Extreme Limiting Dilution Analysis (Hu and Smyth, 2009). Melanoma-initiating cell frequencies were compared using likelihood ratio tests. Statistical significance was defined as p<0.05.

Supplementary Material



This work was supported by the Howard Hughes Medical Institute and by the Allen H. Blondy Research Fellowship. The University of Michigan (UM) Melanoma Bank was supported by a gift from Lewis and Lillian Becker. Flow cytometry was partially supported by UM-Comprehensive Cancer NIH CA46592. Thanks to David Adams and the UM Core Facility for flow cytometry, to Melissa Voutsalath for tissue collection, and to Daniel Weinberg for technical assistance. Elsa Quintana was supported by the Marie Curie Outgoing International Fellowship from the European Commission. Mark Shackleton was supported by the Australian National Heath and Medical Research Council, the Human Frontiers Science Program, and Australia Post.


Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


  • Agliano A, Martin-Padura I, Mancuso P, Marighetti P, Rabascio C, Pruneri G, Shultz LD, Bertolini F. Human acute leukemia cells injected in NOD/LtSz-scid/IL-2Rgamma null mice generate a faster and more efficient disease compared to other NOD/scid-related strains. International Journal of Cancer. 2008;123:2222–2227. [PubMed]
  • Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF. Prospective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA. 2003;100:3983–3988. [PubMed]
  • Boiko AD, Razorenova OV, van de Rijn M, Swetter SM, Johnson DL, Ly DP, Butler PD, Yang GP, Joshua B, Kaplan MJ, et al. Human melanoma-initiating cells express neural crest nerve growth factor receptor CD271. Nature. 2010;466:133–137. [PMC free article] [PubMed]
  • Bonnet D, Dick JE. Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nature Medicine. 1997;3:730–737. [PubMed]
  • Charles N, Ozawa T, Squatrito M, Bleau AM, Brennan CW, Hambardzumyan D, Holland EC. Perivascular nitric oxide activates notch signaling and promotes stem-like character in PDGF-induced glioma cells. Cell Stem Cell. 2010;6:141–152. [PMC free article] [PubMed]
  • Eisterer W, Jiang X, Christ O, Glimm H, Lee KH, Pang E, Lambie K, Shaw G, Holyoake TL, Petzer AL, et al. Different subsets of primary chronic myeloid leukemia stem cells engraft immunodeficient mice and produce a model of the human disease. Leukemia. 2005;19:435–441. [PubMed]
  • Frank NY, Pendse SS, Lapchak PH, Margaryan A, Shlain D, Doeing C, Sayegh MH, Frank MH. Regulation of progenitor cell fusion by ABCB5 P-glycoprotein, a novel human ATP-binding cassette transporter. The Journal of Biological Chemistry. 2003;278:47156–47165. [PubMed]
  • Held MA, Curley DP, Dankort D, McMahon M, Muthusamy V, Bosenberg MW. Characterization of melanoma cells capable of propagating tumors from a single cell. Cancer Research. 2010;70:388–397. [PMC free article] [PubMed]
  • Hope KJ, Jin L, Dick JE. Acute myeloid leukemia originates from a hierarchy of leukemic stem cell classes that differ in self-renewal capacity. Nature Immunology. 2004;5:738–743. [PubMed]
  • Hu Y, Smyth GK. ELDA: extreme limiting dilution analysis for comparing depleted and enriched populations in stem cell and other assays. J Immunol Methods. 2009;347:70–78. [PubMed]
  • Ito M, Hiramatsu H, Kobayashi K, Suzue K, Kawahata M, Hioki K, Ueyama Y, Koyanagi Y, Sugamura K, Tsuji K, et al. NOD/SCID/gamma(c)(null) mouse: an excellent recipient mouse model for engraftment of human cells. Blood. 2002;100:3175–3182. [PubMed]
  • Jamieson CH, Ailles LE, Dylla SJ, Muijtjens M, Jones C, Zehnder JL, Gotlib J, Li K, Manz MG, Keating A, et al. Granulocyte-macrophage progenitors as candidate leukemic stem cells in blast-crisis CML. The New England Journal of Medicine. 2004;351:657–667. [PubMed]
  • Joo KM, Kim SY, Jin X, Song SY, Kong DS, Lee JI, Jeon JW, Kim MH, Kang BG, Jung Y, et al. Clinical and biological implications of CD133-positive and CD133-negative cells in glioblastomas. Laboratory Investigation. 2008;88:808–815. [PubMed]
  • Kleinsmith LJ, Pierce GB. Multipotentiality of single embryonal carcinoma cells. Cancer Research. 1964;24:1544–1551. [PubMed]
  • Lapidot T, Sirard C, Vormoor J, Murdoch B, Hoang T, Caceres-Cortes J, Minden M, Paterson B, Caligiuri MA, Dick JE. A cell initiating human acute myeloid leukemia after transplantation into SCID mice. Nature. 1994;17:645–648. [PubMed]
  • Mani SA, Guo W, Liao MJ, Eaton EN, Ayyanan A, Zhou AY, Brooks M, Reinhard F, Zhang CC, Shipitsin M, et al. The epithelial-mesenchymal transition generates cells with properties of stem cells. Cell. 2008;133:704–715. [PMC free article] [PubMed]
  • Neering SJ, Bushnell T, Sozer S, Ashton J, Rossi RM, Wang PY, Bell DR, Heinrich D, Bottaro A, Jordan CT. Leukemia stem cells in a genetically defined murine model of blast-crisis CML. Blood. 2007;110:2578–2585. [PubMed]
  • Nowell PC. The clonal evolution of tumor cell populations. Science. 1976;194:23–28. [PubMed]
  • O'Brien CA, Pollett A, Gallinger S, Dick JE. A human colon cancer cell capable of initiating tumour growth in immunodeficient mice. Nature. 2007;445:106–110. [PubMed]
  • Ogden AT, Waziri AE, Lochhead RA, Fusco D, Lopez K, Ellis JA, Kang J, Assanah M, McKhann GM, Sisti MB, et al. Identification of A2B5+CD133- tumor-initiating cells in adult human gliomas. Neurosurgery. 2008;62:505–514. [PubMed]
  • Oravecz-Wilson KI, Philips ST, Yilmaz OH, Ames HM, Li L, Crawford BD, Gauvin AM, Lucas PC, Sitwala K, Downing JR, et al. Persistence of leukemia-initiating cells in a conditional knockin model of an imatinib-responsive myeloproliferative disorder. Cancer Cell. 2009;16:137–148. [PMC free article] [PubMed]
  • Pinner S, Jordan P, Sharrock K, Bazley L, Collinson L, Marais R, Bonvin E, Goding C, Sahai E. Intravital imaging reveals transient changes in pigment production and Brn2 expression during metastatic melanoma dissemination. Cancer Research. 2009;69:7969–7977. [PMC free article] [PubMed]
  • Quintana E, Shackleton M, Sabel MS, Fullen DR, Johnson TM, Morrison SJ. Efficient tumour formation by single human melanoma cells. Nature. 2008;456:593–598. [PMC free article] [PubMed]
  • Read TA, Fogarty MP, Markant SL, McLendon RE, Wei Z, Ellison DW, Febbo PG, Wechsler-Reya RJ. Identification of CD15 as a marker for tumor-propagating cells in a mouse model of medulloblastoma. Cancer Cell. 2009;15:135–147. [PMC free article] [PubMed]
  • Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and cancer stem cells. Nature. 2001;414:105–111. [PubMed]
  • Ricci-Vitiani L, Lombardi DG, Pilozzi E, Biffoni M, Todaro M, Peschle C, De Maria R. Identification and expansion of human colon-cancer-initiating cells. Nature. 2007;445:111–115. [PubMed]
  • Roesch A, Fukunaga-Kalabis M, Schmidt EC, Zabierowski SE, Brafford PA, Vultur A, Basu D, Gimotty P, Vogt T, Herlyn M. A temporarily distinct subpopulation of slow-cycling melanoma cells is required for continuous tumor growth. Cell. 2010;141:583–594. [PMC free article] [PubMed]
  • Sanchez PV, Perry RL, Sarry JE, Perl AE, Murphy K, Swider CR, Bagg A, Choi JK, Biegel JA, Danet-Desnoyers G, Carroll M. A robust xenotransplantation model for acute myeloid leukemia. Leukemia. 2009;23:2109–2117. [PMC free article] [PubMed]
  • Schatton T, Frank MH. Antitumor immunity and cancer stem cells. Annals of the New York Academy of Sciences. 2009;1176:154–169. [PMC free article] [PubMed]
  • Schatton T, Murphy GF, Frank NY, Yamaura K, Waaga-Gasser AM, Gasser M, Zhan Q, Jordan S, Duncan LM, Weishaupt C, et al. Identification of cells initiating human melanomas. Nature. 2008;451:345–349. [PMC free article] [PubMed]
  • Shackleton M, Quintana E, Fearon ER, Morrison SJ. Sources of heterogeneity in cancer: cancer stem cells versus clonal evolution. Cell. 2009;138:822–829. [PubMed]
  • Sharma SV, Lee DY, Li B, Quinlan MP, Takahashi F, Maheswaran S, McDermott U, Azizian N, Zou L, Fischbach MA, et al. A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell. 2010;141:69–80. [PMC free article] [PubMed]
  • Shultz LD, Lyons BL, Burzenski LM, Gott B, Chen X, Chaleff S, Kotb M, Gillies SD, King M, Mangada J, et al. Human lymphoid and myeloid cell development in NOD/LtSz-scid IL2R gamma null mice engrafted with mobilized human hemopoietic stem cells. J Immunol. 2005;174:6477–6489. [PubMed]
  • Singh SK, Hawkins C, Clarke ID, Squire JA, Bayani J, Hide T, Henkelman RM, Cusimano MD, Dirks PB. Identification of human brain tumour initiating cells. Nature. 2004;432:396–401. [PubMed]
  • Wang J, Sakariassen PO, Tsinkalovsky O, Immervoll H, Boe SO, Svendsen A, Prestegarden L, Rosland G, Thorsen F, Stuhr L, et al. CD133 negative glioma cells form tumors in nude rats and give rise to CD133 positive cells. International Journal of Cancer. 2008;122:761–768. [PubMed]
  • Yilmaz OH, Valdez R, Theisen BK, Guo W, Ferguson DO, Wu H, Morrison SJ. Pten dependence distinguishes haematopoietic stem cells from leukaemia-initiating cells. Nature. 2006;441:475–482. [PubMed]