Human melanoma is composed of distinct cell types reminiscent of neural crest derivatives and contains multipotent cells that express the neural crest stem cell markers CD271(p75NTR) and Sox10. When isolated from solid tumors by using a method that leaves intact cell surface epitopes, CD271-positive, but not CD271-negative, cells formed tumors on transplantation into nude or nonobese diabetic/severe combined immunodeficient (NOD/SCID) mice. These tumors fully mirrored the heterogeneity of the parental melanoma and could be passaged more than 5 times. In contrast, in more immunocompromised NOD/SCID/IL2rγnull mice, or in natural killer cell–depleted nude or NOD/SCID mice, both CD271-positive and CD271-negative tumor cell fractions established tumors. However, tumors resulting from either fraction did not phenocopy the parental tumors, and tumors derived from the CD271-negative cell fraction could not be passaged multiple times. Together, our findings identify CD271-positive cells as melanoma stem cells. Our observation that a relatively high frequency of CD271/Sox10-positive cells correlates with higher metastatic potential and worse prognosis further supports that CD271-positive cells within human melanoma represent genuine cancer stem cells. Cancer Res; 71(8); 3098–109. ©2011 AACR.
Tumors usually are heterogenic and comprise cells with different capacities to proliferate and differentiate. It has been proposed that this cellular heterogeneity depends on the presence of so-called cancer stem cells, which are defined as cells that can induce de novo tumor formation, self-renew in vivo, and reestablish the cellular composition of the parental tumor (1). Although the cancer stem cell concept is accepted for several types of tumors (2), data for human melanoma, the most aggressive skin cancer, are conflicting (3). For example, melanoma cells exhibiting stem cell properties in vitro displayed increased tumorigenicity as compared with melanoma cells lacking self-renewal potential on xenotransplantation into immunodeficient nude or nonobese diabetic/severe combined immunodeficient (NOD/SCID) mice (4). Later studies suggested that the expression of markers such as CD20, CD133, and MDR1 are associated with melanoma stem cells (4–6), but a strict correlation between marker expression, self-renewal in vitro and in vivo, multilineage differentiation, and high tumorigenicity remains to be established (7). More recently, ABCB5 has been identified as a marker of melanoma-initiating cells capable of self-renewal and differentiation and associated with clinical melanoma progression in human patients (8). Similarly, in fully immunocompromised mouse models, including NOD/SCID/IL2rγnull (NSG) mice, melanoma cells expressing the neurotrophin receptor CD271 (p75NTR) had a higher tumor-initiation capacity than CD271-negative cells, although the negative fraction was also able to generate tumors in these mouse models (9).
In contrast, other recent studies reported a generally high frequency of tumorigenic melanoma cells when NSG mice were used as recipients (10, 11). In addition, these studies did not find a correlation between the capacity to form a tumor and the expression of previously published markers, including CD271 and ABCB5, thus questioning the concept of cancer stem cells (10, 11).
Because cancer stem cells by definition must be able to reproduce the full heterogeneity of the parental tumor and to grow continuously even after multiple passages, we embarked to compare those aspects for the tumors arising in NOD/SCID, nude, and NSG mice after transplantation of CD271-positive versus -negative melanoma cell fractions. To this end, the cells were isolated from human melanoma samples applying an enzymatic tumor digestion protocol that does not harm surface epitopes. Using this gentle method of cell fraction preparation, we reveal that CD271-positive melanoma cells meet the definition of melanoma stem cells. Intriguingly, the incidence of such cells in patient biopsies is associated with poor prognosis.
Materials and Methods
Tissue microarray analysis
All analyses involving human melanoma tissue were carried out in accordance with the ethical committee in canton Zurich. The tissue microarray used here comprised 54 primary melanomas, 141 melanoma metastases, and 53 melanoma cell lines and was generated as described (12, 13). Coexpression of CD271 and Sox10 was identified by double immunostaining (Sox10, 1:800, R&D, MBA2864; CD271, 1:50, Miltenyi Biotec, 130-091-883). The staining procedure was done as previously described (12). The frequency of cells that express CD271 plus Sox10 is presented as mean ± SD and median with interquartile range (IQR). Values were compared by using the Mann–Whitney U test with exact P values. Tumor-specific survival was analyzed by using Kaplan–Meier curves and compared between groups by log-rank test. SPSS 15 software (SPSS Inc.) was used for statistical analyses. Two-tailed P values of 0.05 or less were considered statistically significant.
BALB/c-Swiss nude (CAnN.Cg-Foxn1nu/Crl, Nude) and NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/S, NSG) mice were purchased from Charles River. NOD/SCID (NOD.CB17/JHliHsd-Prkdcscid, N/S) mice were obtained from Harlan Laboratories. Mice were housed under standard conditions with free access to water and food. Experiments were carried out with male or female mice of 6 to 10 weeks of age in accordance with the Swiss federal and cantonal laws on animal protection.
Tumor cell isolation and xenotransplantation
All patients enrolled in the study were treated at the Dermatology Department of the University Hospital of Zurich. The study was approved by the ethics committee of canton Zurich and all patients gave informed consent. Immediately after surgical resection, the solid, metastatic lesions were dissociated into single-cell suspensions by using Hank's buffered salt solution (without Ca2+ and Mg2+; Invitrogen) containing collagenase III (1 mg/mL; Worthington Biochem) and dispase (0.5 mg/mL; Roche). Incubation at 37°C for 1 hour with concurrent mincing allowed complete digestion. In some cases, xenografts and cell lines were dissociated as indicated earlier in the text, followed by an additional digestion with trypsin (0.05% trypsin-EGTA; Gibco) for 5 minutes at 37°C, exactly as described (10, 11). The resulting cell suspension was filtered through 40-μm nylon mesh and single cells were harvested. Tumor xenografts (1 cm2) were harvested from euthanized mice and dissociated as described earlier in the text. A 1,000-bulk tumor, either in vitro cultured or as fluorescence-activated cell sorting (FACS)-sorted cells, was resuspended in Matrigel matrix (BD Biosciences) 1:1 diluted with RPMI-1640 (Invitrogen), and 200 μL were injected s.c. in the flank of mice with a 1-mL syringe with a 25-gauge hypodermic needle.
Paraffin-embedded, 4-μm formalin fixed tissue sections were deparaffinized in xylene and rehydrated. Following heat-induced epitope retrieval by using 10 mmol/L trisodium citrate buffer at pH 6.0 and a Microwave Histoprocessor (Milestone), the sections were incubated overnight at 4°C as described with antibodies for the following markers: Sox10 (1:800; R&D, MAB2864), CD271 (1:200; Alomone Labs, ANT-007), MelanA (1:50; Abcam, Ab785), S100P (1:400; DAKO, Z0611), HMB45 (1:100; DAKO, M0634), microphthalmia-associated transcription factor (MITF; 1:100; Acris, DM4625), neuron-specific class III beta-tubulin (TuJ1; 1:200; Sigma, T8660), neurofilament 160 (NF; 1:200; Sigma, N5264), neuron-specific enolase (NSE; 1:100; Abcam, Ab53025), Peripherin (1:100; Abcam, ab4666), smooth muscle actin (SMA; 1:400; Sigma, A2547), and green fluorescent protein (GFP; 1:500; Abcam, Ab290). Sections were subsequently incubated with following secondary antibodies for 1 hour at room temperature: Cy3-conjugated goat antimouse (1:500; Jackson ImmunoResearch Laboratories), Cy3-conjugated goat antirabbit (1:500; Jackson ImmunoResearch Laboratories), Alexa 488-conjugated goat antimouse (1:500; Invitrogen). All slides were counterstained with Hoechst 33342 (Invitrogen). In addition, sections were stained with hematoxylin and eosin (H&E). Stainings on cultured cells were done as described (14).
Quantification of positive cells was done by ImageJ software (NIH). Three images from each biopsy were taken randomly. Images were imported into ImageJ, converted to 8-bit formats, and subjected to automatic threshold. The “Analyze Particles” function was used to count positive cells.
Flow cytometry and FACS
All steps for flow cytometry were carried out in RPMI-1640 supplemented with 10% fetal calf serum (Invitrogen), 5% Pen/Strep (Invitrogen), and 2 mmol/L EDTA. For sorting or analysis, single cell tumor samples were stained with antihuman CD271 (Miltenyi Biotec; FITC-conjugated: 130-091-917, or APC-conjugated: 130-091-884), antihuman CD44 (1:100; BD Pharmingen, 555477), or anti-ABCB5 antibody (1:200; clone 3C2-1D12, a gift from Markus Frank). Samples were measured with FACScanto II (BD Biosciences) and analyzed with Diva software (BD Biosciences). Sorting was done with FACSAria (BD Biosciences), and the purity of sorted fractions was 95% to 99% of CD271-positive cells in the CD271-enriched fraction and less than 0.5% of CD271-positive cells in the CD271-depleted fraction.
Natural killer cell depletion
Mice were injected i.p. with 50 μL polyclonal rabbit anti-asialo GM1 antibodies (Wako Chemicals, catalogue no. 986-10001) in 150 μL PBS every 5 days starting at the day of melanoma cell engraftment, which resulted in a greater than 95% depletion of CD3− DX5+ natural killer (NK) cells for the duration of the experiment as measured by flow cytometry (15).
Sphere cultures were established as previously described for neural stem cells (16). Briefly, single cells were plated in flasks (Nunc) coated with Poly(2-hydroxyethylmethacrylate) (Poly-Hema; Sigma) at a density of 20,000 viable cells/mL in DMEM-F12 1:1 media (Gibco) containing 1× B-27 supplement (Invitrogen), 20 ng/mL FGF2 (PeproTech), 10 ng/mL EGF (PeproTech), and 5% Pen/Strep (Invitrogen). The resulting spheres were collected after 7 to 10 days by gentle centrifugation (800 rpm), dissociated by using PBS containing 2 mmol/L EDTA, and replated into Poly-Hema–coated culture flasks. To ensure plating of single, viable cells, the dissociated sphere cells were passed through a 40 μm nylon mesh followed by Trypan blue examination. To assess the relative sphere numbers over passage, spheres were grown for 7 days, counted, dissociated, and replated under the same conditions.
Melanoma cell lines
Cell lines M990115 and M010817 were previously described (17). The cell line M070302 was established from surplus material from a xenograft (patient 2481) generated in nude mice. Cell lines were grown in sphere culture condition as describe earlier in the text.
Transduction of melanoma with GFP-expressing lentivirus
The GFP-expressing lentivirus was produced as described (18). Lentivirus-containing supernatants were collected 48 hours after transfection, filtered through a 0.22-μm membrane, and transferred to target cells (see the following text). We xenografted a fresh human melanoma into nude mice, processed the xenograft into single cells and cultured them under sphere conditions. Cells were transduced with GFP-lentivirus at the eighth passage. After 1 week, resulting GFP-tagged cells were checked by FACS and injected s.c. into nude mice. The GFP-tagged xenografts were collected after 2 months, dissociated into single cells, and GFP-positive cells were purified by FACS sorting.
Generation of GFP-tagged clones
After in vitro transduction, GPF-expressing cells were stained for CD271 and single-cell sorting was done on a BD FACSAria sorting system. Fibronectin-coated (Sigma) 96-well flat-bottom tissue culture plates (Nunc) were used for 1 cell per well sorting into 200 μL of medium (RPMI-1640 supplemented with 10% FCS, 2 mmol/L glutamine, 5% Pen/Strep) per well. Each well was checked with a fluorescence microscope, and only wells containing a single cell were processed further. After 2 weeks in culture, clones were stained for CD271 expression and 3 selected clones were processed as follows. The clone was divided into 2 aliquots; 1 aliquot was cultured for 4 hours fixed, and stained for different markers as described earlier in the text, and the other aliquot was xenografted into nude and NSG mice.
Xenotransplantation into highly immunocompromised mouse models fails to phenocopy the cellular heterogeneity of parental human melanoma samples
Because melanoma derives from the neural crest cell lineage, we expect putative melanoma stem cells to exhibit features of neural crest stem cells (NCSC) and individual tumors to comprise cells expressing the NCSC markers CD271 and Sox10 (19, 20) as well as cells with features of different neural crest derivatives. To examine the cellular heterogeneity of a given patient's tumor, we analyzed 19 independent melanoma metastases (Supplementary Table S1) for the expression of 11 NCSC, melanocytic, neural, and mesenchymal markers and found that the majority of the samples were positive for most or all of these markers (Fig. 1A and B; Supplementary Fig. S1).
To investigate mechanisms of human melanoma formation and propagation, it is imperative to use a model that faithfully reproduces the parental phenotype, which we will call phenocopy hereafter. Melanoma cell xenotransplantation into different models of immunodeficient mice, including highly immunocompromised NSG mice, previously led to conflicting results (3, 7). To address which mouse model(s) for melanoma formation might best recapitulate the cellular composition of parental tumors, 1,000 cells from the bulk of various patient samples were injected s.c. into nude, NOD/SCID, and NSG mice, and sections of the resulting xenografts were stained for the expression of 11 markers. The transplanted bulk melanoma cells gave rise to tumors in all cases. Importantly, 100% of all xenografts in nude and NOD/SCID mice were exact phenocopies of the parental tumors with respect to the presence of all markers tested (Fig. 1; Supplementary Fig. S2; Table 1). In contrast, all xenografts in NSG mice (n = 7) differed from the parental tumors and lacked cells expressing one or more of the following markers: CD271, MITF, S100P, and neuronal markers (Fig. 1; Supplementary Fig. S2; Table 1). Quantification of cells expressing a given marker confirmed the consistency of CD271/Sox10, S100P, and MITF expression between nude and NOD/SCID xenografts and the respective parental tumors, and the discrepancies in marker expression between NSG xenografts and original patient tumors (Fig. 1C).
Because the major difference between NSG mice and nude or NOD/SCID mice is the absence of NK cells in NSG mice, we addressed whether NK cells influence the phenotype of xenografted tumors by depleting NK cells from nude and NOD/SCID mice (15). Intriguingly, xenotransplanted melanoma cells did not phenocopy the parental tumor in NK-depleted NOD/SCID mice (Fig. 1A–C; Supplementary Fig. S3; Table 1). Thus, the level of immunocompetence of the xenotransplant recipient crucially affects the capacity of human melanoma cells to form tumors resembling the respective patient melanoma.
CD271-positive melanoma cells are multipotent and able to establish the heterogeneity of the parental tumor
It has recently been debated whether CD271-positive melanoma cells isolated from patient tumors possess an increased tumorigenic capacity in fully immunocompromised mice as compared with CD271-negative cells (9–11). Notably, in these conflicting reports and other studies, different enzymatic procedures have been used to prepare cell fractions from solid melanoma samples. Tumors have been mechanically dissociated followed by enzymatic digestion with collagenase (8), a mix of collagenase and dispase (9), or collagenase and trypsin (10, 11, 21). Therefore, we addressed whether these different tumor digestion protocols might affect detectability of surface markers used to directly isolate putative tumorigenic melanoma cell populations. Strikingly, the percentage of retrievable CD271- and ABCB5-positive cells was consistently reduced in conditions that included a trypsin incubation step (11), whereas CD44 was not affected (Supplementary Fig. S4). These results show that a strong proteolytic activity can selectively damage surface protein epitopes and thus lead to reduced detectability by the corresponding antibodies. This unavoidably results in underestimation of surface marker expression in human melanoma cells and in negative fractions being contaminated by cells actually expressing the selected marker. For this reason, in this study human biopsies and xenografts were dissociated in the absence of trypsin activity.
Using this gentle protocol for tumor dissociation and cell isolation, FACS-sorted CD271-positive melanoma cells from primary xenografts consistently promoted tumor formation in nude and NOD/SCID mice, whereas the CD271-negative cell fraction never gave rise to tumors (Fig. 2A; 32 transplantations representing 7 distinct patient tumors). In contrast, we did not detect differences in tumor initiation in NSG mice between CD271-positive and CD271-negative cells from primary xenografts (12 transplantations each, representing 3 distinct patient tumors; Fig. 2A). Strikingly, in NK-depleted nude or NOD/SCID mice both CD271-positive and CD271-negative cells initiated tumor formation. Hence, the tumorigenic potential of melanoma cell subpopulations is influenced by the presence of NK cells in xenograft recipients.
Because a cancer stem cell by definition must display the differentiation capacity reflecting its parental tumor, we investigated the heterogeneity of different successfully xenografted tumors in NOD/SCID, nude, and NSG mice and of the corresponding parental tumor. The cellular composition of all nude and NOD/SCID xenografts derived from CD271-positive cells (tumors representing 7 distinct patients) was analogous to the corresponding parental tumors (Fig. 2B; Supplementary Fig. S5). In contrast, in NSG mice neither CD271-positive nor CD271-negative cells were able to fully phenocopy the cellular heterogeneity of the corresponding parental tumors with respect to all 11 markers tested (Fig. 2B; Supplementary Fig. S5; Table 2). Of importance, both the CD271-positive and -negative cell fractions failed to consistently generate xenografts expressing crucial melanoma markers such as MITF and S100P or neuronal markers present in the parental tumors. Moreover, none of the tumors produced by CD271-negative cells in NSG mice comprised CD271-positive cells, despite the presence of such cells in the parental tumors (Fig. 2B; Supplementary Fig. S5; Table 2; 12 transplantations representing 3 patients' tumors). In summary, faithful phenocopies of original patient tumors were only achieved on transplantation of CD271-positive melanoma cells into nude and NOD/SCID mice, but not into NSG mice.
The fact that CD271-positive cells generate the full heterogeneity of human melanoma on xenotransplantation into NOD/SCID or nude mice suggests that these cells are multipotent. To specifically address this issue, we infected melanoma cells from a xenograft of patient sample 2481 with a GFP-expressing lentivirus, followed by FACS of GFP/CD271 double-positive cells and plating of those cells at clonal density. Although 19 ± 4% of the CD271-positive cells gave rise to clones consisting of more than 10 cells after 2 weeks in culture, such clones were not observed in the CD271-negative cell fraction (Fig. 2C). Three clones derived from GFP/CD271 double-positive founder cells were selected and subjected to differentiation assays in cell culture, revealing the emergence of multiple cell types, such as cells expressing CD271/Sox10, neuronal and melanocytic markers, and SMA (Fig. 2D). Thus, the CD271-positive melanoma cell fraction contains a population that is clonogenic and multipotent in cell culture, similar to NCSCs.
To address the multipotency of CD271-positive melanoma cells in vivo, cells from the 3 selected clones were grafted s.c. into nude and NSG mice, and the xenografts were analyzed for expression of the 11 markers as described before. In nude mice, the 3 clones were able to establish a cellular hierarchy with all 11 markers present in the xenografts. In particular, 10.2 ± 2.0% of the xenograft cells derived from a CD271-positive cell expressed CD271. Moreover, although most smooth muscle cells present in the xenografts were host derived, all nude xenografts comprised cells double positive for SMA and GFP, revealing that some smooth muscle cells in the tumor originated from CD271-positive melanoma cells. In contrast, neither the NCSC markers CD271/Sox10 nor S100P, MITF, and neuronal traits were consistently expressed in xenografts derived from clones transplanted into NSG mice. Moreover, we did not find any GFP/SMA double-positive cells in the NSG grafts, suggesting that the NSG host environment either suppresses the in vivo multipotency of CD271-positive cells or supports the growth of melanoma cells with restricted developmental capacities (Fig. 2D; Supplementary Fig. S6).
CD271-positive melanoma cells have self-renewal capacity and sustain long-term tumor growth in vivo
To address whether CD271-positive cells have self-renewal capacity, in addition to being multipotent (Fig. 2B–D), we carried out sphere assays under conditions that are known to support sphere formation and self-renewal of normal skin-derived NCSCs (16). CD271-positive cells isolated from fresh melanoma tissue or from melanoma cell lines readily generated spheres that contained CD271/Sox10-positive cells and that could be propagated by serial passaging, whereas the few spheroid aggregates generated by the CD271-depleted melanoma cell fraction displayed a reduced passaging capacity (Fig. 3A and B). Determining the ratio between numbers of secondary over primary and quinternary over quaternary spheres showed that CD271-positive, but not CD271-negative, cells have an extensive self-renewal potential and expand their stem cell activity over time (Fig. 3B).
To confirm the self-renewal potential of CD271-positive melanoma cells in vivo, we isolated CD271-positive and CD271-negative cells from primary xenografts generated by bulk tumor cells in nude mice and retransplanted these cell fractions. We found that only CD271-positive cells were able to generate secondary tumors in nude mice (Fig. 3C and D; Supplementary Fig. S7, “first passage”; n = 4 for each biopsy). Likewise, CD271-positive cells reisolated from secondary xenografts consistently generated tertiary tumors, unlike CD271-negative cells (Fig. 3C and D; Supplementary Fig. S7, “second passage”). To test whether the cellular heterogeneity originating from CD271-positive cells is a stable trait in vivo, we analyzed first and second passage xenografts by immunofluorescence. Intriguingly, not only the cellular heterogeneity with respect to neural crest-specific differentiation markers (Fig. 3D; Supplementary Fig. S7) but also the frequency of CD271-positive cells was maintained in xenografts directly generated from unsegregated patients' melanoma cells and in tumors obtained by serial in vivo passaging of CD271-positive cells (Fig. 3C; percentage of CD271-positive cells in primary xenograft, first passage, and second passage, respectively: 10 ± 1%, 11 ± 3%, 8 ± 2%; n = 4). Hence, CD271-positive melanoma stem cells faithfully phenocopy the parental tumor in nude mice, even after multiple passages in vivo.
Our data suggest that the presence of CD271-expressing melanoma stem cells is a prerequisite for long-term tumor propagation. Therefore, although CD271-negative melanoma cells can generate tumors in NSG, NK-depleted NOD/SCID, or NK-depleted nude mice (Fig. 2; refs. 9, 10), we expected that such tumors could not be maintained in vivo over a prolonged time period. To test this, we serially passaged tumors generated in NSG mice from CD271-positive or CD271-negative cells, using tumors produced by CD271-positive melanoma cells in nude mice as positive control. In both nude and NSG mice, tumors originating from CD271-positive cells allowed propagation in vivo over 5 passages, in accordance with the earlier-described in vivo self-renewal capacity of CD271-positive cells in nude mice (Fig. 4). In striking contrast, tumors derived from CD271-negative cells in NSG mice exhausted with time and could only be propagated in vivo for 3 to 4 passages (Fig. 4). Quantification of CD271-expressing cells in consecutively produced NSG xenografts revealed that the capacity for tumor propagation over several passages was associated with the presence of CD271-positive melanoma stem cells in each xenograft (Fig. 4A–C). CD271-negative cells, however, were unable to produce CD271-positive cells, even after xenograft passaging (Fig. 4A and D). It should be noted, however, that although CD271-positive cells generate tumors in NSG mice that can be serially passaged, those tumors did not phenocopy the parental tumor (Fig. 2).
Thus, NSG mice, NK-depleted NOD/SCID mice, or NK-depleted nude mice provided a host environment permissive for tumor growth by CD271-negative cells, whereas such cells were unable to initiate tumorigenesis in NOD/SCID or nude mice. In contrast, CD271-positive melanoma stem cells were tumorigenic in all mouse models tested. In any case, only in nude or NOD/SCID mice CD271-positive melanoma stem cells produced tumors that completely phenocopied the parental melanoma even on serial xenotransplantation, whereas they failed to do so in NSG mice.
Expression of the NCSC markers CD271/Sox10 correlates with metastatic potential and poor prognosis of melanoma
To verify our findings with a large number of patient samples, we stained tissue microarrays containing more than 200 different melanoma biopsies of primary melanomas, melanoma metastases, and melanoma cell lines for cells coexpressing the NCSC transcription factor Sox10 (20) and CD271 that is found both in NCSCs (16, 19) and in melanoma-initiating cells (9). Consistent with its expression in NCSCs and melanocytes, nuclear Sox10 was found in the majority of cells in melanoma cell lines, primary melanomas, and metastases (Fig. 5A). In contrast, the number of cells with detectable CD271 expression was highly variable. Among primary melanoma biopsies, we found 31 biopsies without detectable CD271 expression (57%), 14 biopsies with less than 5% CD271-positive cells (26%), and 9 biopsies with more than 5% of CD271-positive cells (17%). Intriguingly, however, the proportion of CD271/Sox10 double-positive cells in primary tumors without evidence of metastasis was significantly less than that in primary tumors of patients, who developed metastases during 5-year follow-up (P = 0.01; Fig. 5B). In addition, there was a higher proportion of CD271/Sox10-positive cells in metastases as compared with primary tumors without evidence of metastasis (P = 0.04, Fig. 5C). Similarly, the proportion of CD271/Sox10-positive cells was significantly increased in cell lines derived from metastases as compared with cell lines derived from primary tumors (P = 0.01; Fig. 5C). Thus, both in primary tumors with evidence of metastasis and in metastatic lesions, the number of CD271/Sox10-positive cells was relatively increased, suggesting that their frequency is associated with the metastatic potential in human melanoma.
To specifically address this issue, we focused our analysis on 54 primary malignant melanomas of a sentinel lymph node study for which tumor-specific survival data were available (12). Within this cohort, a frequency of CD271/Sox10-positive cells greater than 5% was associated with poor tumor-specific survival (P = 0.03; Fig. 5D). Together, these findings support the hypothesis that an elevated frequency of melanoma cells expressing NCSC markers is a prognostic factor for the development of metastasis.
In this study we show that human melanoma contains cells that fulfill the definition of cancer stem cells, including the capacity for extensive in vivo self-renewal, maintaining long-term tumor growth, and faithfully recapitulating the cellular composition of the patient tumor, from which the cells have been derived. These melanoma stem cells share properties with normal NCSCs, the precursors of melanocytes in the skin. Indeed, human melanoma stem cells express the NCSC markers CD271 and Sox10, and, similar to NCSCs, have the capacity to self-renew and to generate multiple cell types in vitro and in vivo. CD271-expressing melanoma cells have most recently been shown to exhibit an increased tumor-initiating capacity as compared with CD271-negative cells in fully immunocompromised mice (9). In addition, as we show here, CD271-positive melanoma cells are not only able to initiate tumorigenesis, but invariably regenerate heterogeneous tumors analogous to the parental tumors in patients, even after serial retransplantation. Importantly, CD271-positive cells were required for continuous melanoma growth, as long-term passaging and expansion of tumors was dependent on the presence of a CD271-positive cell fraction in the tumor.
The recent debate on whether particular surface markers can be used to distinguish tumorigenic from nontumorigenic melanoma cells in human biopsies (8–11) has pointed out the importance of establishing appropriate methods for cancer cell isolation from solid tumors and for the study of tumorigenic properties in vivo (3, 11, 22). Our work shows that an excess of proteolytic trypsin activity during tumor digestion and cell fraction preparation can significantly reduce the percentage of cells positive for a given surface marker in immunolabeling assays. In particular, trypsin treatment substantially lowered the detectability of CD271- and ABCB5-positive cells by specific antibodies as compared with more gentle protocols of tumor digestion, presumably because of proteolytic cleavage of surface epitopes. Our findings conceivably explain the discrepancies between conflicting reports on the nature of melanoma subpopulations with tumor-initiation potential. Specific markers defining tumorigenic cells have been identified in studies avoiding trypsin during melanoma cell fraction preparation (refs. 8, 9; this study), whereas no such markers were identified when trypsin has been included in the tumor digestion protocol (10, 11). Thus, we propose that trypsin treatment of melanoma cells can yield cell fractions false-negative for a given cell surface marker. The resulting contamination of supposedly marker-negative cell fractions by cells actually expressing the marker might explain why in some experimental setups cells positive for CD271 or ABCB5 as well as cells seemingly negative for these markers seemed to be equally tumorigenic and to give rise to tumors reexpressing the markers (10, 11). Moreover, loss of surface epitopes might generally affect the capacity of a tumor cell to associate with the surrounding tissue on transplantation and to initiate tumor formation.
Strikingly, complete phenocopies of parental tumors were never obtained in NSG mice, irrespective of whether bulk tumor cells or isolated cell fractions were used for xenograft experiments. Furthermore, only fully immunocompromised NSG mice provided a host environment permissive for tumor growth by CD271-negative cells, whereas these cells were unable to initiate tumorigenesis in more immunocompetent models. In contrast, CD271-expressing melanoma stem cells were tumorigenic in all mouse models tested, indicating a specific capacity for immune evasion by the stem cell population. The apparent immunoselection of melanoma stem cells in xenografts involves the innate immune system, as NK cell depletion in nude or NOD/SCID mice restored the capacity of CD271-negative cells to form tumors. In agreement with these results, NK and NKT cells have been shown to play a central part during immune surveillance of chemically induced skin tumor in mice (23–25). In addition, there are many lines of evidence supporting the concept of immunoediting also in human tumors, including melanoma (26, 27).
There are a number of mechanisms by which tumor cells may escape or suppress an immune response (28). Melanoma cells might achieve immunogenic tolerance by promoting apoptotic cell death or inactivation of antigen-reactive cells (29) or by inducing an immunosuppressive environment as provided, for instance, by increased levels of immunosuppressive macrophages and neutrophils (30). Interestingly, it has recently been shown that ABCB5-positive melanoma cells have the capacity to inhibit interleukin-2–dependent T-cell activation and to induce tolerization by regulatory T cells (31). Reduced expression levels of specific tumor antigens, such as Melan-A/MART-1, tyrosinase, and gp-100, might present another mechanism for immune evasion by melanoma cells. Notably, CD271- and ABCB5-positive melanoma cells express low levels of melanoma-associated antigens such as MART-1, supporting the idea that melanoma cells expressing melanoma stem cell markers escape the immune system attack by the host (refs. 9, 29, 31; G. Civenni and L. Sommer, unpublished data). Of note, we identified a subpopulation of CD271-positive melanoma cells that also express ABCB5 (G. Civenni and L. Sommer, unpublished data), although the functional implication of this finding remains to be addressed. All together, these results suggest that in patients, melanoma stem cells might be able to evade or modulate the immune response, allowing these cells to promote tumorigenic growth and to provide resistance to immunotherapy. The hypothesis that these processes are relevant in patients is further supported by our data that only in mice with a certain level of immunocompetence, CD271-positive melanoma cells generated tumors fully phenocopying the original patient melanoma. However, because xenotransplantation models cannot accurately recapitulate the immune response induced by cancer in human patients, the interaction between cancer cell populations and the immune system should be addressed by using syngeneic melanoma mouse models, in which components of the anticancer immune system can be manipulated during tumor initiation and progression.
Our data are consistent with the idea that CD271-positive melanoma cells play a crucial role in tumor formation in human patients. This is further supported by our clinical data obtained with an extensive tissue array of melanoma samples. These show an association between the proportion of CD271/Sox10-positive cells in primary melanoma and metastatic disease as well as poor tumor-specific survival. Hence, high numbers of melanoma stem cells expressing NCSC markers might influence aggressiveness and the metastatic behavior of malignant melanoma. This conceivably reflects intrinsic, NCSC-like features of melanoma stem cells, given that normal NCSCs have the capacity to extensively migrate through embryonic tissue before differentiation into melanocytes and other cell types. In support of this, increased levels of CD271 expression in melanoma have been associated with enhanced invasive potential in culture (32). Interestingly, CD271/Sox10 expression in metastatic primary melanoma was higher than that in metastatic lesions, which could be explained by different patient groups. Whereas all primary melanomas (metastatic versus nonmetastatic) were primary tumors without chemotherapy or vaccination therapy, tissue of the metastases was obtained from patients after adjuvant therapies. Possibly, this therapy might result in a lower prevalence of cells with CD271/Sox10 expression in the metastases. Alternatively, the site of primary metastatic tumor formation might offer a microenvironment more favorable for stem cells than provided by distant metastases. In any case, however, the proportion of CD271/Sox10-positive cells can potentially be used as a predictor of metastases and consequently poor tumor-specific survival. This in turn suggests that monitoring melanoma cells with NCSC-like features in primary melanoma might be of great prognostic relevance.
Several biopsies of primary, mostly nonmetastatic melanoma did not contain CD271/Sox10-positive cells according to our TMA analysis. CD271/Sox10-positive cells might therefore be involved in the formation of metastatic melanoma, whereas less aggressive tumors might originate from other types of melanoma-initiating cells. By extrapolation one could argue that tumor aggressiveness depends on whether the key oncogenic mutation occurred in a normal stem cell or in a more restricted progenitor cell (33). However, although the melanoma stem cells identified in this study display similar marker expression and potential as normal NCSCs in the adult skin (16, 34), melanoma-initiating cells could also arise by dedifferentiation of more mature melanocytic cells (35). Similarly, we cannot exclude that depending on the context or the oncogenotype, CD271/Sox10-melanoma stem cells emerge from other tumor cells, for instance on a process reminiscent of an epithelial-to-mesenchymal transition during metastasis or by epigenetic modification (36, 37). Intriguingly, however, at least in the conditions chosen for this study, CD271 expression was a stable trait of human melanoma-initiating cells, and we did not find any evidence for CD271-positive cells originating from the CD271-negative tumor cell fraction even after prolonged incubation in vivo. Moreover, CD271-negative cells did not acquire properties of actual melanoma stem cells, in that they could neither reproduce the cellular heterogeneity of parental tumors nor sustain long-term tumor formation.
In conclusion, the identification of human melanoma stem cells required for continuous tumor growth points to potential culprits of tumor formation in patients. In the future, it might be possible to establish specific treatments that reduce tumorigenesis by elimination of these cells or by targeting “stemness” in melanoma (1). Thus, efforts should be made to develop drugs able to promote differentiation of melanoma stem cells, to selectively kill these cells, or to specifically block their self-renewal and expansion.
Disclosure of Potential Conflict of Interest
No potential conflicts of interest were disclosed.
This work was supported by the Swiss Cancer League, the Swiss National Science Foundation, the Vontobel Foundation, UBS Wealth Management and the Swiss NCCR “Neural Plasticity and Repair.”
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
We thank M. Storz, P. Frei, and S. Behnke for assistance and Dr. M. Frank for providing the antibody to ABCB5. We would also like to thank the technical staff of the Central Flow Cytometry Laboratory of the ETH and University of Zurich for their assistance.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received November 3, 2010.
- Revision received January 31, 2011.
- Accepted February 14, 2011.
- ©2011 American Association for Cancer Research.