Human melanoma contains a side population

In a recent study by Luo et al., a Hoechst dye-excluding side population (SP) was identified in 8 clinical melanoma samples [10]. Meanwhile, we analyzed the SP in a larger series of patient melanoma tumors covering different progression phases (n=38; Table S1 in File S1), and examined the correlation with disease stage and tumor thickness (Breslow’s depth), a strong prognostic factor and key parameter in melanoma staging [1,24].

A SP was detected in all melanoma samples analyzed, representing 0.1-2.2% of the viable tumor cells (median SP: 0.4%; Figure 1A-B). Verapamil, an inhibitor of efflux pumps, strongly reduced the SP percentage, thereby confirming the SP phenotype (Figure 1A). The SP proportion did not significantly change between the various melanoma progression stages (Figure 1A-B); primary melanomas (cutaneous malignant form), in-transit metastases, lymph-node metastases and visceral metastases harbor a median SP of 0.4% (n=13), 0.4% (n=8), 0.5% (n=14) and 0.5% (n=3), respectively. Of note, a SP was not detected in pre-malignant nevus (data not shown). Interestingly, the SP proportion in the primary melanomas was found to correlate with Breslow’s depth (Figure 1C; p=0.01).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. The side population (SP) in human melanoma of various progression phases.

A) Representative examples of FACS density plots of Hoechst-incubated cells from primary melanoma (cutaneous malignant form; top left), in-transit metastasis (top right), metastasis to a regional lymph node (bottom left) and visceral metastasis (bottom right) with indication of SP percentage, and the corresponding controls with verapamil.

B) Boxplot of the SP proportions from all human melanoma samples taken together (All, n=38), and from the different progression phases as indicated (primary melanomas, n=13; in-transit metastases, n=8; regional lymph-node metastases, n=14; visceral metastases, n=3).

C) SP proportion in primary (cutaneous malignant) melanomas as outlined against tumor’s thickness (Breslow’s depth). A significant correlation is observed (Spearman r=0.68; p=0.01; dashed line, 95% confidence interval of the best-fit line).

D) Representative example of FACS density plots of Hoechst-incubated cells from the human malignant melanoma cell line A375 with indication of SP percentage, and the corresponding verapamil control (n=18).

https://doi.org/10.1371/journal.pone.0076550.g001

Because of the restricted availability of patient-derived melanoma tissue, we also used the human malignant melanoma A375 cell line [25] for further extended SP characterization. In the cell line we identified a small but clear SP (Figure 1D; range SP: 0.1-0.3%; median: 0.1%; n=18).

The melanoma SP is enriched in tumorigenic activity

In several types of cancer, the SP is enriched in cells that are more tumorigenic than the other cancer cells and that can regrow the tumor [3,26–28]. These cells are generically designated as CSC.

We selected two melanoma progression phases (i.e. primary and lymph-node metastasis) from which to assess the in vivo tumorigenic activity of SP cells. Because primary melanomas were typically too small to sort a sufficient number of cells, they were first expanded in immunodeficient (SCID) mice. SP and bulk tumor “main population” (MP) cells were then sorted and 10,000 cells subcutaneously (sc) injected into SCID mice. The SP generated a new tumor in every experiment (n=3) whereas the MP produced a tumor at lower frequency (1 from n=3). To explore tumorigenic activity of a further melanoma progression stage and at the same time enabling the assessment of clinical melanoma without intervening expansion in a mouse, SP and MP cells were sorted from a larger lymph-node metastasis and sc transplanted into SCID mice. High cell numbers of SP and MP (50,000) both generated tumors at a similar pace, with visible appearance after 6-7 weeks. The re-formed tumors from both primary melanomas and lymph-node metastasis were histologically comparable to the original human lesions (Figure 2A). In particular, the SP from a pigmented melanoma generated a pigmented tumor (arrows in Figure 2A; no tumor from the MP). Of note, the tumors grown from the SP expanded more than from the MP, i.e. they were larger at the time point of analysis (18 weeks after injection; Figure 2B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. In vivo tumorigenic activity of the melanoma SP.

A) Sections (H&E staining) of the xenograft tumors developed in SCID mice 18 weeks after sc injection of SP and MP cells from a primary melanoma (pigmented, see arrows) (upper row) and from a lymph-node metastasis (lower row), together with sections of the original tumors (scale bar, 150µm).

B) Volume of the xenograft tumors grown in SCID mice 18 weeks after sc injection of the melanoma SP and MP cells (10,000-50,000 cells). Bars represent mean ± SEM (n=6). *, p<0.05.

C) Photographs and sections (H&E staining) of the xenograft tumors developed in SCID mice 18-32 weeks after sc injection of SP and MP cells from the SP-derived first-generation xenograft tumors (F1, see A), together with sections of the original tumors (scale bar, 150µm). Again note the pigmentation (upper row) as clear from the picture of the dissected tumor.

D) Volume of the F2 xenograft tumors grown in SCID mice 32 weeks after sc injection of the SP and MP cells from the SP-derived F1 tumors (10,000 SP cells, 50,000 MP cells). Bars represent mean ± SEM (n=2).

E) Tumors grown in SCID mice 5-7 weeks after sc injection of 2500, 1000 and 500 SP cells (upper row) or MP cells (lower row) from the A375 melanoma cell line (n=3). Representative examples are shown.

F) Summary of tumor volume (relative to volume of the tumors grown from 2500 SP cells) after sc injection of the indicated number of A375 SP cells (black bar) and MP cells (white bar). Bars represent mean ± SEM (n=3). *, p<0.05; ***, p<0.001.

https://doi.org/10.1371/journal.pone.0076550.g002

To investigate whether tumorigenic capacity is sustained, we again sorted the SP and MP cells from the SP-derived first-generation xenograft tumors (F1, see Figure 2A). Injection of 10,000 SP cells resulted in a visible tumor (F2) as early as 10 weeks after injection (n=2). In contrast, MP cells – injected at even 5-fold higher numbers (50,000) -, induced palpable F2 tumors only after 18-22 weeks. The F2 xenograft tumors again were histologically comparable to the original tumors (Figure 2C). In addition, the SP-derived F2 tumors were larger than the MP-derived tumors as analyzed at a late time point (32 weeks after injection; Figure 2C-D).

Together, the data above provide support that the melanoma SP contains higher tumorigenic activity than the MP.

Finally, to examine tumorigenic activity in conditions that are more permissive to standardization, we analyzed tumorigenesis by SP and MP cells from the A375 cell line and at the same time tested smaller numbers of cells. After sorting by FACS, 2500, 1000 and 500 SP or MP cells were sc transplanted into SCID mice. Irrespective of cell numbers injected, SP-derived tumors were significantly larger than the tumors grown from the MP (Figure 2E-F; n=3). Moreover, at the lowest number of cells tested, the MP became incapable of generating tumors whereas the SP remained tumorigenic. In agreement with this higher in vivo tumorigenic activity, SP cells (from A375) generated more colonies (~2.5-fold) than MP cells when living (propidium iodide-negative) cells were seeded in vitro at equal low density (Figure 3A). Furthermore, the A375 SP was able to gradually reconstitute the cell line’s heterogeneity when seeded in culture, i.e. to re-generate the MP with concurrent decrease of the SP proportion (Figure 3B). Of note, the MP still contained some SP cells (~0.1%) after sorting which did not change during culture (Figure 3B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Clonogenic and reconstituting capacity of the melanoma SP.

A) Clonogenic capacity of A375 SP cells (relative to MP cells) as assessed at day 9 after seeding at low density. Representative examples of colony-forming activity by SP (bottom left) and MP (bottom right) are shown. Bars represent mean ± SEM (n=3). ***, p<0.001 versus MP.

B) SP analysis (representative FACS density plots) 5, 10 and 11 days after seeding A375 SP cells (upper row) and MP cells (lower row) in standard culture medium (n=4 except for day 11 where n=2; numbers indicate the mean SP proportion).

https://doi.org/10.1371/journal.pone.0076550.g003

Taken together, the melanoma SP is enriched in cells with higher clonogenic and tumorigenic activity than the MP, and seems able to reconstitute the tumor’s heterogeneity in vivo and in vitro. These findings support the idea that the SP encompasses melanoma CSC or CSC-like cells.

Gene expression profiling of melanoma SP

In the recent study by Luo et al., SP gene expression was determined starting from xenograft tumors of lymph-node metastases (n=2) [10]. Here, we explored whole-genome expression in SP (versus MP) from primary melanomas obtained directly from the patient (n=4). Microarray analysis revealed 462 differentially expressed genes (≥1.5-fold up or down, p<0.05; see Table S2 in File S1). Of these genes, 264 were differentially expressed in at least 3 of the 4 individual samples (bold in Table S2 in File S1), corresponding to 251 human ENSEMBL IDs of which 142 and 109 genes were up- and downregulated, respectively, in the SP compared to the MP. The upregulated expression of a few genes (DDX17, MARCH8, TNRC6A) was confirmed by RT-qPCR (using the limited residual RNA/cDNA of 2 of the microarrayed melanoma samples; Figure S1A in File S2). The 251 differentially expressed genes were submitted to Gene Set Enrichment Analysis (GSEA) which revealed a significant overlap of the genes upregulated in the melanoma SP with gene sets upregulated in other cancers such as colon carcinoma (as compared to normal mucosa [29]) and bladder cancer with high recurrence rate (versus low recurrence rate [30]) (Table S3 in File S1). In addition, a significant overlap of genes downregulated in the melanoma SP was observed with gene sets downregulated in the breast cancer cell line MCF7 SP cells (versus the MP cells) [31], of quiescent cells of chronic myeloid leukemia (as compared to proliferating cells [32]) and of melanomas from patients that develop distant metastases within 4 years (as compared to non-metastasizing melanomas [33]) (Table S4 in File S1). Visualization of the interaction network of SP-upregulated genes by STRING analysis revealed that genes involved in regulation of apoptosis (e.g. BAG4, CASP2, DDX17, IGF1 and MDM2), chemoresistance (e.g. IGF1, IGF1R, MAPK14 and MDM2), and epithelial-mesenchymal transition (MDM2, SETD8 and TWIST1) occupy a central position (Figure S1B in File S2). A further detailed and more focused analysis of a selection of genes of the microarray expression data revealed upregulated expression (≥1.5-fold in half or more of the 4 tumors, p<0.05) in the SP of genes implicated in “cancer stemness”, chemoresistance, anti-apoptosis, and cell invasion and migration; and downregulated expression of cell adherence and pro-apoptotic genes (Table 1 and Table S5 in File S1).

The SP is considered as a subpopulation that can efflux drugs, and that is enriched in candidate CSC in several cancers [3,4,26–28]. The higher tumorigenic activity as described above supports a CSC-like phenotype of the SP also in melanoma. To further investigate the drug-effluxing and CSC-like character, we analyzed an additional and larger set of melanoma samples (primary, n=3; metastases, n=4), obtained directly from the patients, by RT-qPCR for expression of ABC multidrug transporters and of CSC-associated markers (see Table S6 in File S1). Compared to the MP, SP cells showed a significantly higher expression of ABCB1, CXCR4, DNMT3B, EPAS1 (or hypoxia-inducible factor 2A, HIF2A), FOSL2, FOXC1, JARID1B, LEF1, MYC, SOX10 and TERT (Figure 4A). Amongst these, ABCB1, DNMT3B, EPAS1, JARID1B and TERT have previously been designated as melanoma CSC(-associated) markers ( [8,22] and reviewed in 13). Of note, these transcripts were not revealed as significantly upregulated in the microarray analysis, which may be accounted for by the limited number of samples (n=4), causing low statistical power and differences in expression levels that may have been missed (i.e. not revealed as significant). Indeed, a number of the genes including ABCB1 and JARID1B showed similar folds of upregulation in the microarray analysis, but the differences were not statistically significant (see Table S7 in File S1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Gene expression in melanoma SP as analyzed by RT-qPCR.

A) Expression ratios of the indicated genes in the SP versus the MP from patient melanomas. Bars represent mean ± SEM (n= 3 primary melanomas and 4 melanoma metastases). *, p<0.05; **, p<0.01.

B) Expression ratios of the indicated genes in the SP versus the MP from the A375 cell line. Bars represent mean ± SEM (n=4). *, p<0.05; **, p<0.01; ***, p<0.001.

https://doi.org/10.1371/journal.pone.0076550.g004

Furthermore, in the RT-qPCR analysis, ABCG2, KLF4, POU5F1/OCT4, SNAI2 and TDGF1 displayed upregulated expression, but also not statistically significant (Figure S2 in File S2). Other proposed CSC-associated (melanoma) markers like ABCB5, CD44, MAGE-C2, NANOG, NES, NGFR and SOX2 did not show different expression levels in SP and MP (Figure S2 in File S2). Of note, expression of ABCB5 was higher in the SP than the MP in 3 out of the 7 tumors (0.5- to 7.8-fold), and lower in the 4 other tumors (-1.7- to -9-fold), making the overall difference statistically non-significant. Finally, a selection of these factors was further validated using the A375 cell line (Figure 4B). The majority (except FOXC1 and LEF1) was also found upregulated in the SP of the melanoma cell line.

The melanoma SP shows resistance to dacarbazine chemotherapy and to hypoxia

In general, the SP phenotype is considered to enrich for chemoresistant tumor cells [2–4]. In agreement, our gene expression analysis of melanoma SP revealed upregulated expression of chemoresistance genes (see Table 1 and Table S5 in File S1). Therefore, we tested the effect of dacarbazine (DTIC) - the current standard chemotherapeutic agent for melanomas [23] -, in the A375 cell line. First, time- and dose-response of cell toxicity was determined (Figure S3 in File S2). Cell death was most prominent after 3-day exposure (with an IC₅₀ of 75µg/ml), in accordance with other reports [34]. A375 cells were then treated during 3 days with double the IC₅₀ dose of DTIC (150µg/ml) to obtain a profound toxic effect. As shown in Figure 5A, the SP increased ~10-fold in proportion suggesting that the SP cells are more resistant to DTIC than the non-SP cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. The melanoma SP after exposure to dacarbazine and hypoxia.

A) Representative example of FACS density plots of Hoechst-incubated A375 cells treated with dacarbazine (+DTIC; 150µg/ml for 3 days) (left; verapamil control not shown) and summary of the SP proportions (right). Bars represent mean ± SEM (n=3). *, p<0.05 versus control (–DTIC).

B) Representative example of FACS density plots of Hoechst-incubated A375 cells cultured in hypoxic conditions (1.5% O₂ 3 days) (left; verapamil control not shown) and summary of the SP proportions (right). Bars represent mean ± SEM (n=3). *, p<0.05 versus standard cell-culture condition (20% O₂).

https://doi.org/10.1371/journal.pone.0076550.g005

Gene expression analysis further revealed upregulated expression of anti-apoptotic factors in the SP suggesting resistance to severe conditions such as hypoxia, typically occurring in cancer growth [4,35]. The A375 cell line was grown in 1.5% O₂ (meaning hypoxia in cell cultures that are standardly kept in 20% O₂). Analysis of the SP showed a ~45-fold increase in proportion (Figure 5B), indicating that SP cells are more resistant to hypoxic conditions. This finding concurs with the upregulated expression of EPAS1/HIF2A in the SP (Figure 4).

ABCB1⁺ cells are localized in the tumor’s invasive component

Gene expression profiling identified upregulated expression of ABCB1 in the melanoma SP, in agreement with findings by Luo et al. [10]. In order to localize ABCB1-expressing cells in melanomas in situ, a series of 20 additional patient tumor samples (7 primary melanomas and 13 corresponding metastases), different from the 38 samples analyzed above, was examined using immunohistochemistry. ABCB1 expression was detected in a subset of melanoma cells, displaying a punctuated cytoplasmic or a membranous staining pattern (Figure 6A). No obvious differences in proportion of ABCB1-immunopositive (ABCB1⁺) were observed between primary melanomas and metastases (Figure 6A-B). Remarkably, ABCB1⁺ cells in the primary melanomas were predominantly found at the invasive front of the tumors (Figure 6C), reminiscent of putative CSC in other cancer types [36]. These ABCB1⁺ cells frequently occurred in small clusters, whereas the ABCB1⁺ cells found scattered within the tumors mostly presented as individual cells. In melanoma metastases, single or clustered ABCB1⁺ cells were only found disseminated over the tumor (and not localized in a front; Figure 6A). In some of the tumors analyzed, similar cells also expressed JARID1B (Figure S4 in File S2). Although low-signal ABCB1 immunoreactivity was also observed in a few endothelial cells and some (tumor-infiltrating) lymphocytes, the tumoral stroma mostly lacked ABCB1 expression. ABCB1 was also absent from most (normal) epidermal cells, although the basal layer of the epidermis showed sporadic immunoreactivity (data not shown).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. ABCB1 expression in primary human melanomas and corresponding metastases.

A) Representative examples of primary melanoma (left) and melanoma metastasis (right) immunostained for ABCB1 (scale bar, 150µm; +, cytoplasmic staining; *, membranous staining).

B) Summary of semi-quantitative analysis, showing estimated percentage of ABCB1⁺ cells and signal intensity (no staining = 0, weak signal = +, moderate signal = ++, strong signal = +++; N/A, not applicable).

C) Primary melanoma immunostained for ABCB1 (top) and higher magnification of the boxed area (bottom) (scale bar, 300µm).

https://doi.org/10.1371/journal.pone.0076550.g006

Ethics Statement

In agreement with the admission rules of the University Hospitals, Leuven, Belgium, patients undergoing surgical removal of a primary or metastatic melanoma provided written consent to use their left-over diagnostic material for scientific purposes in an anonymized way; the research project was approved by the local Ethical Committee of the University Hospitals Leuven, and the use of the biological materials was formally approved by the Biobank of the University Hospitals, Leuven. The animal studies were approved by the local Ethical Committee of the KU Leuven.

Human melanoma specimens

Melanoma samples were obtained after surgical resection in the University Hospitals, Leuven. Clinical information on the samples analyzed (n=38) is shown in Table S1 in File S1. Part of the tumor was used for (immuno-)histological and pathological evaluation, and part (cut into 3-mm³ pieces) was dissociated into single cells using collagenase type IV (180U/ml; Life Technologies, Ghent, Belgium) for 2.5 h at 37°C. Cells were subjected to further analysis as described below.

The human melanoma A375 cell line

The malignant (metastatic) human melanoma cell line A375 [25] was obtained from Dr. L. Van Kempen (Department of Pathology and Oncology, McGill University, Montreal, Canada). Cells were cultured in Dulbecco’s modified Eagle’s medium/Ham’s F12 (DMEM/F12 1:1; Lonza, Verviers, Belgium) supplemented with 4mM L-Glutamine (Life Technologies), 10% fetal bovine serum (FBS; Lonza) and antibiotic/antimycotic (Life Technologies). Cell cultures were regularly tested for mycoplasma contamination by PCR, and were found negative (data not shown). For the analyses described below, cells were released from the culture vessel and dissociated into single cells using trypsin (0.05%) with EDTA (Life Technologies).

SP analysis by FACS

We adapted the original SP protocol [5] for optimal resolution of SP from human melanoma. Briefly, cell density was adjusted to 1x10⁶ cells/ml and cells were incubated with 5µg/ml Hoechst33342 (Sigma, Bornem, Belgium) for 90 min at 37°C. Verapamil (100µM; Sigma), added 20 min before in control samples, was used to confirm the SP phenotype because it reduces efflux of the dye [5,67]. Cells were finally resuspended in ice-cold PBS (Life technologies) with FBS (2%; Lonza) and propidium iodide (2µg/ml; Sigma) for subsequent analysis by flow cytometry using a FACSVantage or FACSAriaIII (BD Biosciences, Erembodegem, Belgium). The SP is visualized as a Hoechst^low population after UV excitation of Hoechst and detection of blue emission with a BP 424/44 filter and of red emission with a BP 630/22 filter. Analysis was done, and proportions calculated, within the living (propidium iodide-negative) cell population. Statistical analysis of SP percentages was performed using one-way ANOVA (Dun’s Multiple Comparison test).

Exposure to dacarbazine and hypoxia

To obtain time- and dose-response curves of cell toxicity by dacarbazine (DTIC, activated by light; see 68), A375 cells were treated with different concentrations of DTIC (25-200 µg/ml; Sigma) or vehicle (PBS). Viable cells were counted after 1 to 3 days using the 4-methylumbelliferyl heptanoate (MUH) assay (according to [69]). For SP analysis, A375 cells (triplicate) were treated with 150 µg/ml DTIC for 3 days and cells analyzed by FACS as described above.

To investigate the effect of hypoxia on the SP, A375 cells (triplicate) were cultured in 1.5% O₂/5% CO₂ using the “In Vivo₂ 400” hypoxic incubator (Ruskinn Technology, Bridgend, UK) for 3 days, and the SP compared with the SP of A375 cells cultured under standard cell-culture conditions (20% O₂/5% CO₂). Statistical analysis of SP percentages was performed using two-tailed unpaired Student’s t-test.

Culture and colony formation

SP and MP were sorted from subconfluent A375 cell cultures, and equal numbers seeded in culture at 200-1000 cells/ml in the standard culture medium. Both cell populations displayed comparable viability after sorting (Trypan Blue exclusion test; data not shown). Five, 10 and 11 days later, cell cultures were analyzed by FACS to determine the SP proportion.

To assess colony-forming capacity, A375 SP and MP cells were cultured at low density (1000 cells/60-mm dish) in the standard medium. At day 9, cultures were washed twice with PBS, stained for 10 min at room temperature with 1,9-dimethyl-methylene blue (0.25% in 50% ethanol; Sigma) and the number of colonies was counted.

In vivo tumorigenesis

To assess tumorigenic activity in vivo, SP and MP cells from melanoma or A375 were sorted by FACS into DMEM/F12 (supplemented with 10% FBS). Cells were mixed with Matrigel (1:1; BD Biosciences), and varying numbers sc injected into “severe combined immunodeficiency” (SCID) mice (males, 6-10 weeks old; bred in-house). Tumor volume was calculated using the formula [(large side)x(small side) ²]x0.52, and tumor sizes from SP and MP compared at identical time points in the individual experiments.

Microarray analysis

SP and MP cells from primary (cutaneous malignant) melanomas were sorted by FACS into cold lysis solution of the RNeasy Micro Kit (Qiagen, Venlo, The Netherlands). RNA was extracted according to the manufacturer’s protocol (comprising DNase treatment), and quality and concentration determined using Picochips on a BioAnalyzer 2100 (Agilent Technologies, Diegem, Belgium). Only samples with RNA Integrity Number ≥8.0 were used for microarray analysis (n=4 primary melanomas), which was performed in the VIB Nucleomics Core (Leuven, Belgium). RNA (20ng) was subjected to 2 successive rounds of linear amplification by in vitro transcription while incorporating Cy3 label [70]. cRNA probes were hybridized onto Agilent whole human genome 4x44K oligonucleotide arrays (Ag-Hu-WG-G4112F). The microarray data have been deposited in the NCBI’s Gene Expression Omnibus (GEO) and are accessible through GEO Series accession number GSE48838. Data were further processed using the Agilent’s Feature Extraction Software (version 10.1.1.1). Raw signal intensities were background-corrected, subjected to quantile normalization and log₂-transformed. The obtained data sets of SP and MP were compared in a pair-wise fashion, and p-values calculated by paired Students t-test performed on the log₂ values of the probe sets. Genes were considered differentially expressed when p<0.05 and fold change≥1.5. To explore the underlying functional categories, clustering analysis of differentially expressed genes was performed using GSEA [71]. Finally, we narrowed the gene sets by focusing on factors shown to be important in “stemness”/CSC, cellular migration, chemoresistance, apoptosis and cell adherence (≥1.5-fold in ≥ half of the samples, p<0.05). To further visualize gene networks, the Search Tool for the Retrieval of Interacting Genes (STRING; http://string-db.org/) was used.

Reverse-transcription quantitative PCR (RT-qPCR)

For real-time qPCR using SYBR Green, RNA was amplified and converted into cDNA with the Ovation Pico SL WTA kit (NuGEN, Bemmel, The Netherlands). Oligonucleotide primers for qPCR were designed with Perlprimer [72] (see Table S6 in File S1) and validated for comparable amplification efficiency (data not shown). Multiple housekeeping genes were tested for normalization (GAPDH, HPRT, B2M, RPL19). Based on the level and stability of expression in all samples (relative to the other housekeeping genes), the most appropriate set of housekeeping genes (i.e. with uniform expression), being RPL19 and B2M, was selected. qPCR was performed on an ABI 7900HT Fast PCR system using Fast SYBR Green Master Mix (Applied Biosystems/Life Technologies) according to the manufacturer’s protocol. Negative controls and dissociation curves were used to confirm specific amplification.

For RT-qPCR using Taqman technology, SP and MP cells were sorted in lysis solution of the Taqman Preamp Cell-to-Ct Kit (Applied Biosystems/Life Technologies) and RNA isolated according to the manufacturer’s protocol (including DNase treatment). After examination of quality and concentration, RNA was amplified and converted to cDNA with the Taqman Preamp Cell-to-Ct Kit. To measure expression of selected genes, corresponding Taqman GEX Assays (containing primers and TaqMan probe) were used (see Table S6 in File S1). ACTB was included as housekeeping gene for normalization, as recommended by the manufacturer. qPCR was performed on an ABI 7900HT Fast PCR system using Taqman Gene Expression Master Mix (Applied Biosystems/Life Technologies) following the manufacturer’s recommendations.

Normalized, relative expression data were calculated using the comparative threshold cycle (2^ddCt) method [73]. Statistical analysis was performed using two-tailed one sample t-test (if Gaussian distributed) or two-tailed Wilcoxon rank sum test.

Immunohistochemistry

Five-µm sections of formalin-fixed paraffin embedded melanoma specimens were immunostained using a Bond-max fully automated staining system (Leica Microsystems, Wetzlar, Germany), including onboard heat-induced antigen retrieval during 20 min in EDTA-Tris buffer, pH 9.0 (for ABCB1) or citrate buffer, pH 6.0 (for JARID1B) and alkaline phosphatase-based Bond Polymer Refine Red Detection. The primary antibodies were used at optimized dilutions, i.e. 1/200 for ABCB1 (Clone JSB-1; MON9011 from Monosan/Sanbio, Uden, The Netherlands) and 1/100 for JARID1B (HPA027179 from Sigma). Specificity of the ABCB1 antibody [74] has been amply validated before in our lab, using multiple types of normal tissues and tumors [75–77]. The anti-JARID1B antibody [78] was manufactured by the Human Protein Atlas (HPA), which validates the affinity-purified antibodies by immunohistochemistry in a multitude of tissues and cells assembled in tissue microarrays, by immunofluorescence and confocal microscopy in human cell lines, by Western blot and by protein array analyses (http://www.proteinatlas.org/ [79]). In addition, we routinely perform immunohistochemistry using these antibodies on control specimens (human tissue with known expression pattern such as liver for ABCB1 and skin for JARID1B) with positive results.

Sections were evaluated for estimated abundance of immunopositive cells and signal intensity (no staining = 0, weak signal = +, moderate signal = ++, strong signal = +++) by 2 investigators.