Established melanoma cell lines (Supplementary Table 1): Forty cell lines were selected from the European Searchable Tumour Line Database (ESTDAB; http://www.ebi.ac.uk/ipd/estdab) [29]. Of the 40 cell lines, 38 were metastasis-derived, and 2 were derived from primary-derived melanomas (EST-66 and EST-83). These cell lines have been certified by DNA fingerprinting and tested for mycoplasma contamination. They are currently also available from the European Collection of Animal Cell Cultures (ECACC, see https://www.culturecollections.org.uk/products/celllines/generalcell/search.jsp?searchtext=estdab&dosearch=true).

Early-passage melanoma cell strains: Four early-passage cell strains were derived from metastatic lesions from patients treated at the Tübingen University Hospital, according to the following protocol [30, 31]: “TüMel 39” (4th passage), “TüMel 49” (3rd passage), “PDX 25” (P2, P1) and “PDX 35” (P0, P1). The PDX early-passage cell strains were initially passaged in mice before being cultured in vitro.

Human epidermal melanocytes: Two sources of primary melanocytes were used: (1) Cells in their third passage derived from the circumcision of a healthy individual treated at the Tübingen University Hospital. (2) Human adult primary epidermal melanocytes purchased from ATCC (Manassas, VA, USA).

Fibroblasts: Neonatal fibroblasts NuFF1 were obtained commercially (Globalstem, Gaithersburg, MD, USA).

Skin sections: Five µm thick sections of formalin-fixed paraffin-embedded normal human skin were sourced commercially (Abcam, Cambridge, UK).

Melanoma tissue samples: Forty formalin-fixed paraffin-embedded metastatic lesions from patients treated at the Tübingen University Hospital Dermatology Department were used. Patients provided their written informed consent for the storage and scientific analysis of tissue samples. The use of these samples was approved by the University of Tübingen Ethics Committee (ethics approval number: 017/2016BO2).

Established melanoma cell lines were cultured in 40 mL RPMI 1640 medium (Life Technologies, Darmstadt, Germany) supplemented with 10% foetal bovine serum (FBS) (Sigma-Aldrich, Munich, Germany) in either 20% O${}_2$ or 2% O${}_2$ using the Concept 1000 InvivO${}_2$ hypoxic chamber (Ruskinn Technology (Pencoed, UK)) with media titrated to pH 6.7 with 37% hydrochloric acid (Merck, Darmstadt, Germany). Depending on the generation time of each cell line, between 2 and 4 $\times$ 10${}^6$ cells were seeded per flask. Cell culture and estimation of viability was performed as previously described, although, in the current study, cells were cultured for seven days and cultures were harvested with Accutase (Sigma-Aldrich) [32]. Selected cell lines (n = 2) were repeated to confirm the initial results. Early-passage melanoma cell strains were cultured until confluence in a medium containing 1% Pen–Strep (Biochrom, Berlin, Germany). Melanocytes were cultured in DermaLife M melanocyte growth medium (CellSystems, Troisdorf, Germany).

BD LSR II, multicolour flow cytometry was used to measure protein expression, as previously described [32], although with the following modifications. Automatic software compensation was performed to minimise spectral overlap between different fluorochromes, and CST beads were run prior to each sample measurement to control for consistency in machine performance. The following antibodies were used: ALDH1A1-PE (Lot: HG09MY1304, Clone: 03) (Sino Biological Inc., North Wales, PA, USA), Nestin-PE (Lot: 2524561, Clone: 10C2) (Merck Millipore, Temecula, CA, USA), ABCG2-PE (Lot: B143287, Clone: 5D3) (Biolegend, San Diego, CA, USA), CD44v7/8-FITC (Lot: 150715, Clone: VFF-17) (Acris Antibodies, San Diego, CA, USA), CD44v10-FITC (Lot: 9E08V1) (Bioss, Woburn, MA, USA), CD133-APC (Lot: 5150611303, Clone: AC133) and CD271-FITC (Lot: 5150609183, Clone: ME20.4-1.H4) (both from Miltenyi Biotec, Teterow, Germany). The DNA-binding dye ethidium monoazide bromide (Biotium, Hayward, CA, USA) was used to exclude dead cells (incubated on ice under bright light for 20 min) before antibody staining. Data were analysed using FlowJo software version 10.0.7 (Tree Star, Ashland, OR, USA). Cell viability was determined by a commercial viability kit, according to the manufacturer’s instructions (BD Biosciences, Heidelberg, Germany).

Melanoma tissue sections (5 µm thick) were prepared and stained with antibodies, as previously described [33], but with the following modifications: an EDTA- and SDS-based antigen retrieval solution containing 25 mM Tris–HCl (pH 8.5) (Sigma-Aldrich), 1 mM EDTA and 0.05% SDS (both from SERVA Electrophoresis, Heidelberg, Germany) were used to unmask antigens. The following antibodies were used: ALDH1A1 rabbit monoclonal (Lot: GR41450-6, Clone: EP1933Y) (Abcam), CD133 rabbit polyclonal (Lot: X13030523) (Fitzgerald, Acton, MA, USA), ABCG2 mouse monoclonal (Lot: D15KF02234, Clone: BXP-21) (Biolegend, San Diego, CA, USA), CD44v7/8 mouse monoclonal (Lot: 051114, Clone: VFF-17) (Bio-Rad, Hercules, CA, USA), Alexa Fluor 488 donkey anti-rabbit IgG (H+L), Cy3 donkey anti-mouse IgG (H+L) (both from Jackson ImmunoResearch Laboratories, West Grove, PA, USA). Fluorescence intensity for each antibody-stained tissue section was compared by fluorescence from a control tissue (secondary antibody only) mounted on the same slide. The software PixelStats (designed in-house by the University Hospital Tübingen, Tübingen, Germany) recorded the mean fluorescence intensity of microscopy images and was used to create a ratio between the stained and control tissue sections. A tissue was considered positive if it showed at least a 50% increase in fluorescence over the control tissue. An average of 12 images per tissue (i.e., 12 for stained and 12 for control) covering the entire tumour (including all regions of the tumour centre and at the invasive front) were captured at 20$\times$ magnification. Fluorescently stained tissue slides were measured by a Zeiss Axiophot fluorescence microscope (Carl Zeiss, Oberkochen, Baden-Wuerttemberg, Germany).

Statistical analyses were performed using Prism software version 6 (GraphPad, La Jolla, CA, USA). Changes of less than 10% between different culture conditions for the same cell line were not considered to be different. Cell line tumour-associated antigen and CD44 expression data were obtained from ESTDAB [29]. Correlations were assessed by a non-parametric two-tailed (Spearman) correlation test. Significance between the two groups was assessed by a two-tailed non-parametric (Mann–Whitney U) test. Trends across four grouping variables were assessed by a two-tailed Fisher’s exact contingency test. Significant relationships were considered as p 0.05.

We examined 40 different established melanoma cell lines to screen for the expression of seven putative CSC markers (ALDH1A1, ABCG2, CD44v7/8, CD44v10, CD133, CD271, and Nestin) and document similarities and differences between extensively proliferated cells, and those from short-term cultures or freshly isolated tumour tissue. Cell lines are commonly used in cancer research, and it is important to document their similarities in ex vivo situations. First, we identified the number of cell lines from the 40 that expressed each of these CSC markers. This was achieved by examining the average protein expression of the entire population for each cell line to determine a fluorescence index (FI). This approach showed that all 40 cell lines were positive for ALDH1A1, CD271, and Nestin, while around half expressed ABCG2, and only three were positive for CD133. Although all 40 cell lines expressed the CD44 molecule [29], we could only identify four (10%) which expressed the CD44 splice variant isoforms 7/8 and none that expressed the splice variant isoform 10 (Fig. 1A). In addition, we examined how many cells within each melanoma cell line were positive for each marker. The results, considering the frequency of positive cells, generally agreed with the results obtained from determining the average protein expression in the entire population (FI), described above: ALDH1A1, CD271, and Nestin were found to be expressed in the majority of cells in the population for all 40 cell lines. ABCG2 and CD44v7/8 were found to be present in 0 to 70% and in 0 to 40% of cells in the population, respectively, while CD133 and CD44v10 were found at a lower maximal frequency (Fig. 1B). We observed that when these proteins were expressed, they were usually not present in the discrete sub-populations of positive cells, i.e., a single population of expressing cells was observed rather than separate populations of positive and negative cells within a cell line (Fig. 1C). This was the case for both established cell lines and early-passage cell strains. Additionally, we noted the absence of discrete positive and negative populations for proteins that showed expression in less than 100% of the cells—for example, ABCG2 was expressed in between 0 and 70% of these melanoma cell lines; however, the majority of cells in the population expressed this protein similarly, with no discrete positive and negative populations (Fig. 1C, lower panel). However, we also occasionally observed heterogeneity in the expression of these proteins–for example, Nestin and CD133 were sometimes found to be expressed at different levels in a fraction of cells in the population (Supplementary Material 2). Interestingly, when comparing primary- and metastatic-derived cell lines, we observed no marked differences in the expression of the seven CSC markers (Fig. 1D), while we also obtained similar results when comparing early-passage cell strains with established cell lines, i.e., a common expression of CD271, ALDH1A1, Nestin, and ABCG2 (all 4/4), a rare expression of CD44v7/8 (1/4), and a lack of expression of CD133 and CD44v10 (data not shown). Finally, we investigated the potential relationships between the expression of the seven CSC markers. This analysis revealed correlations between Nestin and ALDH1A1 (p 0.0001, r = 0.5988), and ABCG2 (p = 0.0193, r = 0.364) expressions (Fig. 1E).

Fig. 1.

Expression patterns and correlations of putative cancer stem cell (CSC) markers in established melanoma cell lines. Forty melanoma cell lines were assessed for their expression of ABCG2, ALDH1A1, CD44v7/8, CD44v10, CD133, CD271, and Nestin by flow cytometry. (A) Average protein expression in the entire population shows Nestin, CD271, ALDH1A1, and ABCG2 to be commonly expressed in established cell lines, while CD133 and CD44 variants were less commonly found. Grey shading indicates positivity for the respective markers. (B) Assessing the frequency of positive cells within each cell line showed Nestin, CD271, ALDH1A1, and ABCG2-positive cells to be common in this panel of 40 established cell lines and in the four early-passage cell strains. Lower frequencies in cells were found for CD133 and CD44 variants. (C) Examples of flow cytometry plots showing similar expressions by most melanoma cells for CD271 and ABCG2. Although less than 100% of cells are positive for ABCG2, most cells in the population express the protein at similar levels. (D) Expression of CSC markers is similar between primary- and metastatic-derived cell lines. (E) Expression levels correlated between ALDH1A1 and Nestin (r = 0.5988; p 0.0001) and between ABCG2 and Nestin (r = 0.4235; p = 0.0058).

Since standard in vitro cell culture conditions (20% O${}_2$ and neutral pH medium) do not accurately reproduce in vivo growth conditions, we cultured the established cell lines under conditions designed to better reflect the tumour microenvironment (i.e., 2% O${}_2$ and pH 6.7, here designated “oxy low/pH low”). We were interested in understanding the relationship between the tumour microenvironment and cancer stemness because our previous study had shown that incorporating these features as part of an in vitro culture can have a dramatic impact on the behaviour of melanoma cells [32]. Compared to the conventional in vitro culture model, oxy low/pH low culture conditions slowed growth in all melanoma cell lines (p 0.0001) and reduced viability in the majority (p 0.01) (Fig. 2A). We also found the same trend toward reduced viability under oxy low/pH low conditions in a subset of cell lines (n = 15) using a commercial apoptosis kit, which was statistically significant when considering differences in late-apoptotic cells (p = 0.035) (data not shown).

Fig. 2.

Effects of hypoxic and acidic culture conditions on melanoma cell lines and the expression of putative CSC markers. Forty melanoma cell lines were cultured under conventional (20% O${}_2$ and neutral pH) or experimental conditions (2% O${}_2$ and pH 6.7) for seven days. Following the culture period, cell lines were harvested and living-to-dead cell ratios were determined using trypan blue. Changes in the expression of CSC markers under experimental culture conditions were tested for their association with cell line viability and growth. (A) Experimental culture conditions reduce the speed of growth by melanoma cell lines and decrease their viability. Cell numbers indicate the total number of living cells. (B) Experimental culture conditions that either up- or downregulate the expression (FI) of putative CSC markers ALDH1A1, Nestin, CD271, and ABCG2. (C) Changes in expression (FI) of Nestin and CD271 are associated with melanoma cell viability under experimental culture conditions, while ALDH1A1 correlated to cell growth (* = p 0.05; ** = p 0.01, **** = p 0.0001).

The expressions of the CSC markers were also altered by hypoxic and acidic conditions. In the 40 cell lines, ALDH1A1 and ABCG2 were downregulated in most, while CD271 was upregulated in the majority, and Nestin was up- or downregulated in roughly equal numbers (Fig. 2B). Each cell within all cell lines showed similar changes in expression under the test conditions. The cell lines that were negative for any protein under conventional conditions also remained negative in the oxy low/pH low model. The correlations between the expression of Nestin and ALDH1A1 and Nestin and ABCG2 (Fig. 1E) were observed under conventional conditions and remained retained when these cells were cultured in oxy low/pH low (p = 0.0002 for ALDH1A1 vs. Nestin and p = 0.0439 for ABCG2 vs. Nestin) (data not shown).

Changes in the expression of CD271 and Nestin between the conventional and oxy low/pH low culture models were found to correlate with the viability of the melanoma cell lines (p = 0.0063 and p = 0.0258, respectively) (Fig. 2C). Further, we observed that an improvement in viability was associated with the upregulation of these proteins under oxy low/pH low culture conditions, as indicated by the higher live/dead cell ratios. Despite the majority of cell lines tending to downregulate ALDH1A1 in the experimental model, we found improved cell growth in the cell lines where the expression of ALDH1A1 was upregulated (p = 0.0168) (Fig. 2C). No such associations were found for ABCG2. Since the CD44 variants and CD133 were rarely, or not at all, expressed by these cell lines, this analysis could not be performed for these proteins.

To validate the results obtained from the melanoma cell lines in vitro, we examined the expression of the selected CSC markers in situ, using an equal number of melanoma tissue deposits. Since CD133 and CD44v7/8 were rarely present in the melanoma cell lines, we considered it important to compare the results for these proteins obtained in vitro with those in situ. We additionally tested ALDH1A1 and ABCG2 because we found a number of interesting correlations in vitro and these proteins have rarely been studied in melanoma tissue, while CD271, Nestin, and CD44v10 have all been previously examined in situ by other investigators [34, 35, 36]. Consistent with the cell line results, all the CSC markers were commonly expressed in the majority of the melanoma cells in situ, although we did also observe a degree of heterogeneity, which revealed areas in the negative cells or more highly/weakly expressing cells within the tumour regions. In particular, CD133 was expressed more highly in a proportion of melanoma cells in some tissue samples, although in these cases most cells still exhibited positive staining. A common observation was that lower cell expressions resembled the tumour stroma. Notably, ALDH1A1 was an exception to our observation, whereby the common CSC markers were expressed by the majority of cells since we observed it to be either commonly expressed by all cells (17.5% of samples) or to be selectively expressed by a subset of cells (45% of samples) in situ (Supplementary Material 3).

When comparing tissue with established melanoma cell lines, the four putative CSC markers showed different expression patterns in situ (Fig. 3A,B). CD44v7/8 expression was more commonly observed in tissues (37.5% of samples were positive) compared to cell lines (10% of samples). The same was true for ABCG2, which was much more frequently expressed in the tissues (77.5%) rather than in the cell lines (46.34%), whereas ALDH1A1 was expressed only in 62.5% of the melanoma deposits compared to 100% of the cell lines. Interestingly, we found that CD133 was the most commonly expressed putative CSC marker in melanoma tissues, with 95% of the deposits expressing it. This stands in stark contrast to the established cell lines, whereby only a small proportion (7.5%) of cell lines expressed CD133, according to the FI. Representative immunofluorescence images stained for ALDH1A1, ABCG2, CD44v7/8, and CD133 are shown in Fig. 3C. Supplementary Material 3 shows a single representative stained and control image for all four proteins in two tissue samples (in the case of ALDH1A1, it shows one example, which demonstrates the expression exhibited in most cells, with the second example showing the expression in a subset of cells) and Supplementary Material 4 shows one example of a complete set of images comprising all fluorescence images captured throughout one tumour sample for all four proteins.

Fig. 3.

Expression of putative CSC markers in melanoma tissue, cell lines, and differentiated non-malignant cells. Immunofluorescence (melanoma tissue) and flow cytometry (cells and cell lines) were used to assess expression levels of putative CSC markers. (A) Almost all 40 melanoma tissues express CD133, and less frequently express ABCG2, ALDH1A1, and CD44v7/8. Grey shading indicates positivity for the respective markers. (B) Forty melanoma cell lines and an equal number of melanoma tissues were compared for their expression of CSC markers; tissues more frequently express CD133, CD44v7/8, and ABCG2, although ALDH1A1 is more common in cell lines. The FI was used to determine positive samples. (C) Representative images from melanoma deposits stained for ALDH1A1, ABCG2, CD44v7/8, and CD133. Scale bar indicates 100 µm. (D) Putative CSC markers are expressed at similar levels by established (n = 40) and early-passage (n = 4) melanoma cell lines, primary epidermal melanocytes (n = 2), and dermal fibroblasts (n = 1).

Since we observed that both melanoma tissues and cell lines expressed these putative CSC markers, next, we examined their expression in benign differentiated cells to test their specificity as CSC markers and to examine their potential use as therapeutic targets in melanoma. Thus, using flow cytometry, we investigated the expression of these molecules in control cells; human dermal fibroblasts and primary human epidermal melanocytes, the latter from two different sources. We additionally tested normal human skin sections for CD133 and ABCG2, using immunofluorescence. Our results demonstrate that these seven putative CSC markers are not specific for cancer or normal stem cells since they were also detected in all the tested benign differentiated cell types examined in this study. Interestingly, the benign samples were found to express these proteins at comparable or occasionally even higher levels than in malignant cell types. A comparison between the expression levels of all proteins tested in the benign and malignant cell types was conducted by flow cytometry and is shown in Fig. 3D. The test results for human skin using immunofluorescence can be found in Supplementary Material 5.