Single-Cell Gene Expression Signatures Reveal Melanoma Cell Heterogeneity

ORIGINAL ARTICLE Single-cell gene expression signatures reveal melanoma cell heterogeneity

M Ennen1, C Keime1, D Kobi1, G Mengus1, D Lipsker1,2, C Thibault-Carpentier1 and I Davidson1

It is well established that tumours are not homogenous, but comprise cells with differing invasive, proliferative and tumour- initiating potential. A major challenge in cancer research is therefore to develop methods to characterize cell heterogeneity. In melanoma, proliferative and invasive cells are characterized by distinct gene expression profiles and accumulating evidence suggests that cells can alternate between these states through a process called phenotype switching. We have used microfluidic technology to isolate single melanoma cells grown in vitro as monolayers or melanospheres or in vivo as xenografted tumours and analyse the expression profiles of 114 genes that discriminate the proliferative and invasive states by quantitative PCR. Single-cell analysis accurately recapitulates the specific gene expression programmes of melanoma cell lines and defines subpopulations with distinct expression profiles. Cell heterogeneity is augmented when cells are grown as spheres and as xenografted tumours. Correlative analysis identifies gene-regulatory networks and changes in gene expression under different growth conditions. In tumours, subpopulations of cells that express specific invasion and drug resistance markers can be identified amongst which is the pluripotency factor POUF51 (OCT4) whose expression correlates with the tumorigenic potential. We therefore show that single-cell analysis can be used to define and quantify tumour heterogeneity based on detection of cells with specific gene expression profiles.

Oncogene (2015) 34, 3251–3263; doi:10.1038/onc.2014.262; published online 18 August 2014

INTRODUCTION slow cycling, MITF-low cells have tumour-initiating properties and Tumours are not homogenous, but comprise cells with differing that such cells can arise spontaneously in the cultures of MITF- 10 invasive, proliferative and tumour-initiating potential. Classical expressing cells. anticancer agents target highly proliferative cells; however, the We have used single-cell quantitative (q) PCR to address limited success of this approach with initial tumour regression heterogeneity in melanoma cells in vitro and as xenografted followed by relapse attests the presence of residual, slow-growing, tumours in mice. We show that this method can be used to define therapeutically resistant tumour cells. These nonproliferative, melanoma cell heterogeneity. drug-resistant cells, sometimes termed cancer stem cells or tumour-initiating stem cells, may initiate new tumours when the local environment becomes favourable.1 RESULTS In melanoma, heterogeneity can arise from accumulation of Heterogeneous gene expression in melanoma cells in vitro genetic lesions that promote therapeutic resistance if a small To determine whether it is possible to observe and quantify subpopulation of cells harbour mutations that confer resistance tumour cell heterogeneity from gene expression in single cells, we V600E 2,3 towards drugs specifically targeting oncogenic BRAF . used two melanoma cell lines. MITF-high 501Mel cells proliferate Superimposed on this is phenotype switching where the tumour rapidly in vitro, but are poorly invasive/motile and poorly microenvironment can induce melanoma cells to adopt an tumorigenic when injected subcutaneously in nude mice. MITF- invasive, proliferative or stem-like phenotypes.4,5 Meta-analysis negative 1205Lu cells proliferate slowly in vitro, and are invasive/ of gene expression in more than 500 melanoma samples motile and highly tumorigenic in nude mice.8 RNA-seq identified identified specific signatures involving a group of around 100 3655 genes preferentially expressed in 1205Lu compared with genes whose differential expression correlates strongly with either 501Mel and 2505 genes show the opposite profile (Supplementary the proliferative or invasive cell states.6,7 The strongest marker of Table 1). The Widmer7 proliferative signature genes are preferen- the proliferative phenotype is MIcrophthalmia-associated Tran- tially expressed in 501Mel cells and invasive signature genes scription Factor (MITF) strongly expressed in over 90% of preferentially expressed in 1205Lu cells. We designed primer pairs proliferative cells, but low to undetectable in slow cycling invasive for invasive and proliferative signature genes, for POU3F2/BRN2 cells. MITF is heterogeneously expressed in human tumours and and GLI2, the ‘stemness’ markers POU5F1 (OCT4) and Nanog low-MITF-expressing cells are characterized by high expression of whose expression has been reported to be induced in melanoma the transcription factors POU3F2/BRN28 and/or GLI2,9 that appear cells under conditions that potentiate their tumour-initiating as markers for invasive cells, independent of the BRAF or NRAS potential,11 cell cycle markers such as CCNB1, CCND1 and several oncogenic mutation status. There is also compelling evidence that other genes preferentially or commonly expressed in the two cell

1Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS/INSERM/UDS, Illkirch, France and 2Faculté de Médecine and Service de Dermatologie, Hôpital Civil, Hôpitaux Universitaires de Strasbourg, Strasbourg, France. Correspondence: Dr I Davidson, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS/INSERM/UDS, 1 Rue Laurent Fries, 67404 Illkirch, France. E-mail: [email protected] Received 27 May 2014; revised 8 July 2014; accepted 10 July 2014; published online 18 August 2014 Melanoma heterogeneity M Ennen et al 3252 types, for example the drug resistance marker ABCB512 Increased heterogeneity in cells grown as melanospheres (Supplementary Table 1), to determine whether the differential Melanoma cells can be grown as spheres showing heterogeneous gene expression is represented at the single-cell level. expression of MITF and other markers.13,14 We grew 501Mel and We analysed 501Mel and 1205Lu cells from exponentially 1205Lu cells as spheres for 8 days and assessed the heterogeneity growing monolayer cultures. Heatmaps representing gene expres- of their gene expression profiles. Populations expressing high, low sion generated from qPCR performed on 501Mel cells and or no MITF can be observed in 501Mel-derived spheres (Figure 2a). clustered using the unsupervised Unweighted pair group method Expression of many MITF targets correlates with that of MITF with arithmetic mean identified two major populations expressing although a population expressing high levels of MITF is either high or intermediate levels of MITF expression (Figure 1a). distinguished from the others by lower and more variable High MITF-expressing cells also strongly express many known expression of its target genes. Thus compared with monolayers, MITF target genes (Figure 1a, Supplementary Table 1). In cells with growth as spheres increases cell heterogeneity. Immunostaining intermediate MITF, many 501mel-enriched genes display a more also identified sphere-derived cells with very different MITF levels heterogeneous profile of expression. We also identified a small and many MITF-negative cells (Figure 2b). Similarly, heteroge- number of cells showing strongly reduced MITF where expression neous expression of ITGA4 is also observed both by qPCR and of most of its target genes is also strongly diminished. In contrast, immunostaining. invasive signature and 1205Lu-expressed genes show low to Analysis of 1205Lu cells grown as spheres reveals cells undetectable expression in all cells. with high, low or no detectable ZEB1 expression (Supplementary In analysis of a second set of cells with a smaller number of Figure 3A). In addition, a subpopulation of ZEB1 high cells can be genes, cells with high and intermediate MITF levels are detected, observed that express generally higher levels of many of but are clustered together although separated from the MITF-low the tested genes (*** in Supplementary Figure 3A). Under these population (Supplementary Figure 1A). As seen above, MITF target conditions, the expression of ITGA4 and BIRC3 strongly correlate genes cluster together with MITF separately from invasive with that of ZEB1. Thus, similar to 501Mel cells growth as spheres signature genes. Importantly two cells with undetectable levels leads to increased heterogeneity of the 1205Lu population. of MITF can be seen (** in Supplementary Figure 1A) confirming The cell specificity and heterogeneity of expression are also previous reports that rare MITF-negative cells arise spontaneously underlined by comparisons of the two cell types (Figure 2c) that 10 in 501Mel cultures. Immunostaining confirmed the existence of also highlights a subpopulation of 1205Lu cells showing low 501Mel cells expressing high and lower levels of MITF in expression of many transcripts and thus may correspond to a proportions similar to those seen in the qPCR analysis, whereas different cell state (*** in Figure 2c). no MITF expression was detected in 1205Lu cells (Figure 1b). fi Together these data de ne heterogeneity in the 501mel mono- Correlation of gene expression identifies MITF-regulated genes layer cell population. Overall 1205Lu cell monolayers appear more homogeneous We used the single-cell data to make correlation maps of genes than the 501Mel cells as most 1205lu-expressed genes show showing co-regulated expression. In 501Mel-derived spheres, a rather homogenous expression (Supplementary Figure 1B). MITF group of highly expressed genes including MITF show strongly and its targets show low expression and no clearly distinct correlated expression (Figure 3a). A small sub-cluster shows populations are defined. A small number of cells with much lower particularly strong correlation with MITF. Integration with MITF expression of almost all assayed transcripts are however observed. ChIP-seq data shows that most of these genes are associated with MITF binding sites15 (Supplementary Table 1). This analysis The apparent homogeneity and lack of clusters of co-regulated fi genes may reflect the fact that no master regulators of the therefore identi es genes that are directly regulated by MITF in physiology of these cells, analogous to MITF in 501Mel cells, have 501Mel melanospheres. been identified and thus the transcripts we test may not be Several other clusters of co-regulated genes are observed. appropriate to reveal heterogeneity. Expression of several genes was strongly co-regulated with fi POU3F2/BRN2, however, examination of our previous ChIP-chip A comparison of the pro les of the two lines highlights both the 16 cell-specific expression of many transcripts and also the hetero- data did not reveal these genes to be enriched in BRN2 targets, suggesting their co-regulation by another, as yet, geneity in their expression within cells of each line (Figure 1c). fi Together these data indicate that it is possible to observe cell- unidenti ed factor. specific gene expression at the single-cell level and to identify This analysis is less informative for 1205Lu cells, since as gene expression signatures that define subpopulations of cells. mentioned above no master regulators of expression in this line We also assayed the reproducibility of the expression levels of are known. Nevertheless, several clusters of co-regulated genes selected genes. First, we analysed the expression of several genes can be discerned around ZEB1 and TLE2 or around TLE4 in duplicate reactions for the cDNA collection from the isolated (Supplementary Figure 3B). cells within the same qPCR plate. These replicates show a high correlation for the entire spectrum of expression levels detected Identification of cell subpopulations in tumours (Supplementary Figure 2A). Second, we re-analysed the expression We next performed subcutaneous injection of 501mel and 1205Lu of selected genes by qPCR reactions on the cDNA collection on an cells in nude mice to generate xenografted tumours. 501mel cells independent qPCR plate (Supplementary Figure 2B) with repro- are poorly tumorigenic and out of nine injections only five ducible although with slightly diminished correlations compared tumours were formed. Mice were killed at 12 weeks and tumours with that seen on the same plate. For the most part the were recovered that showed a heterogeneous appearance diminished correlations are due to the presence of a small with both pigmented and nonpigmented cells (Supplementary number of outliers rather than a general change. Note however Figure 4A). In contrast, all mice injected with 1205Lu cells rapidly that the cDNA collections cannot be frozen and must be kept on formed tumours that necessitated killing of the animals after ice and used as soon as possible. In our experiments the two qPCR 5 weeks. Tumours from these mice had a more homogeneous plates were performed one after the other on the same day. These appearance with no trace of pigmentation. results indicate the reproducibility of the signals for a given cDNA Cell from the 501Mel-derived tumour can be divided into three collection in multiple qPCR reactions and allow integration of major classes based on high or intermediate/low MITF expression results for genes analysed on separate qPCR plates and thus for a and differential expression of many of its target genes (Figure 4). larger number of genes. Within each major population many genes show heterogeneous

Figure 1. Gene expression in 501Mel cell monolayers. (a) Heatmap illustrating expression of 59 genes in 96 single cells from 501Mel monolayers. The colour key showing the log2 expression values is shown to the right of the panel. (b) Immunostaining of 501Mel and 1205Lu cell monolayers for MITF expression. Red, yellow and white arrows indicate cells with high, intermediate or low MITF expression, respectively in 501Mel cells, whereas no signiﬁcant expression is seen in 1205Lu cells (×40 magniﬁcation). (c) Heatmap illustrating comparative expression values of 59 genes in 96 cells from 501Mel and 95 cells from 1205Lu cell monolayers. Clusters of genes showing preferential expression in one or the other cell types are indicated.

Figure 2. Gene expression in cells from 501Mel-derived spheres. (a) Heatmap illustrating expression of 81 genes in 88 cells from 501Mel- derived spheres. (b) Immunostaining of cells derived from 501Mel spheres for MITF and ITGA4 expression. Red, yellow and white arrows indicate cells with high, low or no MITF expression or higher and lower ITGA4 expression. (×20 magniﬁcation for MITF and × 40 magniﬁcation for ITGA4). (c) Heatmap illustrating comparative expression values of 81 genes in 88 cells from 501Mel and 95 cells 1205Lu-derived spheres. Clusters of genes showing preferential expression in one or the other cell types are indicated. The asterisks *** indicate a group of 1205Lu cells showing little or no expression of the tested transcripts.

Figure 3. Identification of co-regulated genes in 501Mel cells grown as spheres or xenografted tumours. (a) Heatmap illustrating correlation of gene expression in single cells from 501Mel-derived spheres. The median expression of each gene relative to quartiles of expression is shown above the heatmap. (b) Heatmap illustrating correlation of gene expression in single cells from 501Mel-derived tumours. The median expression of each gene relative to quartiles of expression is shown above the heatmap. The colour key showing the Pearson correlation coefficient is shown to the right of the figure.

Figure 4. Gene expression in cells from 501Mel-derived tumours. Heatmap illustrating expression of 113 genes in 92 cells from 501Mel- derived tumours. Cells marked with dots, squares and asterisks represent subpopulations expressing invasion markers as described in text.

expression. Two clusters of genes can be distinguished, one strong expression of invasive markers. 501Mel tumours are comprising a set of genes showing strongest correlation with therefore highly heterogeneous composed of cells with different MITF, and a second that displays generally higher expression in gene expression profiles. Importantly, small subpopulations cells with lower MITF, but segregates into three broad populations expressing invasive and drug resistance genes that have under- with higher, lower or weak levels in the higher MITF-expressing gone a partial phenotype switch can be detected. cells. Among the second set are the cell cycle markers CCND1 and In cells from 1205Lu-derived tumours, two broad populations CCNB1 that are strongly diminished in the cells at the left of the can be distinguished based on high or low ZEB1 expression panel suggesting that these cells are growth arrested despite the (Supplementary Figure 4B). Within these, subpopulations can be high MITF expression. observed with differential expression of smaller clusters of genes, More strikingly, small subpopulations (** in Figure 4) expressing notably, a population of cells (marked with *) with generally invasion (ZEB1, GLI2, MYOF) or drug resistance (ABCB5) markers are higher expression of the majority of transcripts. Comparison of the clearly observed. Similarly, three high MITF-expressing cells profiles from the 501Mel- and 1205lu-derived tumours further (squares and star) display expression of another set of invasive highlights genes showing preferential of specific expression in one markers (BIRC3, TLE2, FGF2) and a single cell (spot) shows a unique of the two types and the heterogeneity in their expression in the profile, with high expression of MITF and its targets along with tumour cells (Supplementary Figure 5).

Oncogene (2015) 3251 – 3263 © 2015 Macmillan Publishers Limited Melanoma heterogeneity M Ennen et al 3257 Correlative analysis in 501Mel-derived tumour cells identified grown in ES cell medium. Expression of NR4A3, FOXD1 and several groups of co-regulated genes including a group whose RAB27A, a known MITF target gene,19 on the other hand is expression tightly correlates with MITF. With the exception of enriched in tumours (Figure 7a). These observations were TPBG and POU3F2/BRN2, all the other genes in this group are confirmed comparing only tumours with spheres that identified direct MITF targets with associated MITF-occupied sites as additional genes preferentially expressed in one or the other determined by ChIP-seq (Figure 3b, Supplementary Figures 2C conditions (Supplementary Figure 6A). The differential expression and D and Supplementary Table 1). Comparison with melano- of these genes was confirmed by qPCR on RNA from cell spheres revealed genes co-regulated with MITF in both conditions, populations grown under the different conditions. TLE2, TGFA, whereas others show better correlation with MITF only in tumours NRP1, RELB, CYR61, SLIT and HSP4 are all enriched in monolayers or in melanospheres. compared with tumours, whereas the opposite is seen for TPBG, Sections from the tumours described above were further NR4A3, FOXD1, RAB27A and TLE4 (Supplementary Figure 7A). analysed by immunostaining and immunohistochemistry. Staining Similarly, TLE2, TGFA and NRP1, are higher expressed in mono- of the 501Mel tumour showed a mosaic heterogeneous expres- layers than in spheres as seen by single-cell analysis. These qPCR sion of both MITF and CEACAM1 (Figures 5a and 6a). In contrast, experiments were performed using RNA from the same mono- no significant MITF staining could be seen in the 1205Lu-derived layers, spheres and tumour used for the single-cell analysis, thus tumour. We also determined if more rare 501Mel tumour cells confirming that the changes seen at the single-cell level reflect expressing high levels of BIRC3, as identified by single-cell qPCR, what can be observed in the population. Moreover, generally could be detected in this way. Generally high BIRC3 expression comparable, but not identical gene expression profiles were was observed in the 1205Lu tumour, although some hetero- observed with RNA prepared from independent monolayers, geneity was seen, with a subpopulation of cells showing lower spheres and tumours indicating that the differential expression of expression (Figure 5b). In contrast, only a small population of these genes is a general property of growth under different 501Mel tumour cells strongly expressed BIRC3, with the majority conditions (Supplementary Figure 7B). showing low or no staining (Figure 5a). Hence, consistent with the A similar analysis of 1205Lu cells showed that DAPK1 and NRP1 single-cell qPCR analysis, immunostaining confirms the differences are strongly depleted in tumours, whereas BIRC3, ZEB1 and a in expression between the two cell types and identifies a small larger group of genes are preferentially expressed in tumours subpopulation of 501Mel tumour cells strongly expressing BIRC3. compared with spheres (Figure 7b and Supplementary Figure 6B). An interesting observation from this analysis concerns the QPCR confirmed the changes seen at the single-cell level with expression of the POU5F1 (OCT4) pluripotency gene in tumour preferential expression of DAPK1 in monolayers, stronger expres- cells. Single-cell analysis shows that POU5F1 is strongly and rather sion of ZEB1, GLI2, TNC and WDR91 in tumours and a preferential homogeneously expressed in 1205Lu tumour cells. In 501Mel expression of INPP4B in spheres (Supplementary Figure 7C). Again tumour cells, its expression is more heterogeneous and correlates comparable although not identical expression profiles were seen with that of the invasive subpopulation (Figures 3b and 4 and in RNA from independent 1205Lu monolayers, spheres and Supplementary Figure 4B). Immunostaining confirmed that tumours (Supplementary Figure 7D). Together the above data POU5F1 is strongly expressed in a majority of 1205Lu tumour highlight that averaging changes in gene expression seen at the cells, but that its expression is lower and heterogeneous in the single-cell level recapitulates changes in the overall population. 501Mel tumour (Figures 5a and b and 6a). We analysed the expression of several genes in primary human melanomas and/or cutaneous melanoma metastases. Consistent DISCUSSION 8 with previous reports, MITF is heterogeneously expressed in Single-cell expression analysis identifies differentially and co- melanoma, where intermixing of cells with high and lower regulated gene networks expression levels can be observed (Figure 6b). CEACAM1 We show that single-cell gene expression profiling can be used to expression is limited to cells that have traversed the basal 17 investigate melanoma cell heterogeneity under different culture epidermal membrane and is also highly heterogeneous. ZEB1 conditions. To our knowledge, this is the first study describing the expression has previously been shown to be upregulated in fi 18 extensive use of single-cell pro ling to investigate melanoma and metastasis. As can be observed in the published data and in one of the few20 to address tumour cell heterogeneity in general additional sections here, ZEB1 displays a heterogeneous expres- in such a comprehensive manner. By using two melanoma cell sion pattern (Figure 6b). Expression of BIRC3 on the other hand is lines, we show that single-cell analysis can faithfully reproduce the limited to a restricted number of cells in cutaneous metastases, differential expression of a large number of genes and discrimi- similar to what is observed in 501Mel-derived tumours. We also nate cell subpopulations with distinct expression profiles. observed POU5F1 expression in a subset of metastatic cells as The power of the approach is also illustrated by identification of evidenced by their strong nuclear staining. Together these data coordinately regulated genes. Correlative analysis of gene show that expression of these markers is heterogeneous in human expression indicates that many changes are not simply stochastic melanoma and thus that analysis of human tumours by single-cell events, but rather reflect coordinated regulatory networks. This is techniques should identify other genes coexpressed with BIRC3 most clearly seen with MITF-regulated genes where correlative and POUF51. analysis clusters not only many previously known target genes together with MITF, but also many novel genes, the majority of Identification of genes differentially expressed in monolayers, which are associated with MITF-occupied sites therefore defining spheres and tumours the MITF-regulated gene network in the cell population. In We determined whether it was possible to identify genes showing addition to the MITF network, we identified other coordinately preferential expression under different culture conditions from regulated gene groups although the transcription factor(s) comparative analysis of the single-cell data. In 501mel cells, TLE2, responsible for their regulation remain to be identified. As noted TGFA and NRP1 display preferentially expression in monolayers, above, analysis of 1205Lu cells has been less informative in this RELB, CYR61 and CDK14 in melanospheres, whereas TPBG is regard. This may reflect an inherent lack of heterogeneity in these expressed only in three-dimensional conditions in spheres and cells, or that we have not analysed the most appropriate tumours (Figure 7a). POU5F1 expression is also more strongly and transcripts to reveal their heterogeneity. homogeneously expressed in melanospheres than under other We also show that the differential gene expression seen in growth conditions, consistent with the fact that the spheres were single cells can be used to identify changes observed at the

Figure 5. Immunostaining of cells from 501Mel and 1205Lu-derived tumours. (a) Immunostaining of sections from 501Mel-derived tumours for MITF, CEACAM1, BIRC3 and POU5F1. Red, yellow and white arrows indicate representative cells with high, intermediate or in some instances no detectable expression of the indicated protein (×40 magniﬁcation). (b) Immunostaining of sections from 1205Lu-derived tumours for BIRC3 and POU5F1. The asterisks in the Hoechst panel of BIRC3 indicate murine stromal cells readily distinguished from the surrounding human 1205Lu cells by the presence of visible heterochromatin foci.

Figure 6. Immunohistochemistry on 501Mel and 1205Lu-derived tumours and human melanomas. (a) Detection of MITF and POU5F1 by immunohistochemistry on 501Mel and 1205Lu-derived tumours as indicated. Red, yellow and white arrows illustrate representative cells expressing high, low or no detectable expression of these proteins. (b) Detection of the indicated proteins by immunohistochemistry in primary human melanomas (MITF, CEACAM1) or cutaneous metastases. Magniﬁcation × 20. Red, yellow and white arrows illustrate representative cells expressing high intermediate or no detectable expression of these proteins. In the MITF panel, arrows also indicate MITF-expressing cells that have invaded the dermis. E, epidermis; D, dermis; AD, adipocytes; IL, inﬁltrating lymphocytes.

Figure 7. Changes in gene expression under different culture conditions. (a) Heatmaps illustrating comparative expression values of 26 or 55 genes in 276 501Mel cells grown as monolayers (ML, 96 cells), melanospheres (MS, 88 cells) or tumours (TUM, 92 cells). Genes showing preferential or depleted expression in one or the other culture conditions are indicated. M, mixed clusters comprising cells from several growth conditions. (b) Heatmaps illustrating comparative expression values of 26 or 55 genes in 286 1205lu cells grown as monolayers (96 cells), spheres (95 cells) or tumours (95 cells). Genes showing preferential or depleted expression in one or the other culture conditions are indicated.

population level as a comparison of single-cell data identiﬁed TPBG also known as tumour antigen 5T4/WAIF1 is a cell surface genes whose expression is enriched under different growth protein that inhibits Wnt signalling and is targeted in multiple conditions. Interestingly, we found that TPBG is expressed only cancer immunotherapy clinical trials.21 Similarly, we found a under conditions of three-dimensional growth in 501Mel cells. striking preferential expression of the PI3K/AKT signalling pathway

Oncogene (2015) 3251 – 3263 © 2015 Macmillan Publishers Limited Melanoma heterogeneity M Ennen et al 3261 inhibitor INPP4B in 1205Lu melanospheres.22 In contrast, expres- phenotype-switching models implied that cells expressing invasion of DAPK1 is depleted in 501Mel tumours consistent with the sive markers showed a concomitant loss of MITF.4 Note, however, observation that its promoter is often hypermethylated in that we measure MITF mRNA expression in these cells, not melanoma and other cancers.23 whether they express high levels of MITF protein, in particular of Together the above data indicate that single-cell gene profiling transcriptionally active MITF protein whose properties may be can be used to quantify differences in gene expression to define modulated by posttranslational modifications. The cells expressing heterogeneity in cell populations, to identify gene-regulatory invasive markers may therefore have downregulated the tran- networks, and identify genes whose expression is modified under scriptional activity of MITF at the translational and/or posttransla- different growth conditions. tional levels. On the other hand, it has recently been reported that melanoma cells can invade while proliferating and that invasion 25 Identification of cells with invasive gene expression profiles in a and proliferation are not necessarily mutually exclusive states. 501Mel cell-derived tumour The MITF-high cells expressing invasion markers may thus represent such a cell population since many of them also express Applying single-cell analysis to 501Mel cell monolayers revealed high levels of CCND1 and CCNB1, although one of these cells no their heterogeneous expression of MITF and its target genes longer expresses these cell cycle genes and may be growth and a small population of cells with low-to-undetectable MITF arrested. It is also worth noting that we did not observe activation expression. The mechanisms underlying the variations in MITF of the invasive signature genes in the lowest MITF-expressing cells expression in these monolayers are, as yet, unknown. One suggesting that modulation of MITF levels alone may not be possibility is that the MITF-high and MITF-intermediate cells sufficient to elicit activation of invasion genes. Irrespective of the correspond to cells in the different stages of the cell cycle. Little is underlying mechanism, the important point here is, however, known about how MITF mRNA levels vary during the different that our approach allows the detection of such rare cells within phases of the cell cycle, thus MITF-high and MITF-intermediate the population. cells may correspond to those in the G1, S or G2 phases. More The gene expression data generated here also revealed a importantly, however, it remains to be determined whether the potential role of the POU5F1 pluripotency factor in melanoma. small population of MITF-low/null cells detected in these Several lines of evidence indicate that POUF51 expression populations correspond to cells that have undergone growth 24 correlates with a more tumorigenic and invasive state. POU5F1 arrest. As MITF knockdown leads to a G1 growth arrest, whether is more strongly expressed in 1205Lu than in 501Mel cells, and in the spontaneous growth arrest is the consequence or the cause of 501Mel tumours, its expression is correlated with the subpopula- the low MITF levels remains to be determined. tion expressing invasive markers. Similarly, its expression is Growth as melanospheres increases the heterogeneity in the upregulated in melanospheres, a growth condition that has population with an increased number of MITF-low/negative cells. previously been reported to dramatically enhance the tumorigenic These observations are in accordance with previous reports of 13 14 capacity of melanoma cells and when cells are exposed to the increased heterogeneity in MITF expression in melanospheres senescence associated secretome that also increases their and in correlation with the observation that growth as spheres 11 13 tumorigenic potential. We also observed POU5F1 expression in enhances tumour-initiating properties. In spheres, cells on the a subset of cells in human cutaneous metastases. The correlation outer layers proliferate actively, whereas those in the centre are 25 of POU5F1 expression with the invasive state suggests that these exposed to partial hypoxic conditions and enter a G1 arrest. It POU5F1-positive cells may define a subpopulation of invasive cells has previously been shown that hypoxia represses MITF expres- within the tumour. In the future it will be important to determine 10 sion through an HIF1a-dependent transcriptional mechanism. As which genes are regulated by POU5F1 in tumour cells. discussed above, therefore, the MITF-low/negative cells are likely In conclusion, we show that single-cell gene expression can be G1 arrested and originate from the central regions, but it is not used to define tumour cell heterogeneity and detect subpopula- clear whether the G1 arrest is the cause or consequence of tions expressing specific gene expression profiles. Heterogeneous repressed MITF expression. expression of several of the markers was also observed in human Interestingly, however, strong downregulation or loss of MITF melanomas confirming that their heterogeneity is not just a occurring in subpopulations both in monolayers and in melano- feature seen in xenografts. It should thus be possible to extend spheres does not lead to upregulation of the invasive signature this approach to human melanoma where, in particular, molecular genes. This contrasts with siRNA silencing of MITF in 501Mel analysis of the primary tumour is often hampered by the necessity 15 cells that upregulates for example ZEB1 expression (see to preserve intact biopsies for histopathology analysis. The Supplementary Table 1), and with the strong inverse correlation ability to analyse gene expression in small numbers of cells of GLI2 and MITF expression seen among different melanoma cell without compromising evaluation of the Breslow index, opens 9 lines and their ability to repress each others expression. Similarly, the possibility to investigate cell heterogeneity and identify many MITF target genes whose expression levels are positively novel marker genes and thus gain a better understanding of correlated in our data are not deregulated upon MITF silencing. melanomagenesis. These observations reveal that artificial si/shRNA silencing experiments do not necessarily reproduce the effects elicited by the changes in MITF expression occurring in response to MATERIALS AND METHODS physiological stimuli. High-throughput single-cell qPCR In contrast, single-cell analysis of 501Mel tumour cells allows the Single cells were isolated using the C1 Single-Cell Auto Prep System identification of small subpopulations of cells that have under- (Fluidigm, South San Francisco, CA, USA) followed by reverse transcription gone a ‘phenotype switch’ to express genes of the invasive and pre-amplification according the Fluidigm’s instructions. Single-cell signature as well as ABCB5 that has been shown to have a critical gene-expression experiments were performed using Fluidigm’s M96 role in drug resistance in melanoma and possibly also in the quantitative PCR (qPCR) DynamicArray microfluidic chips. A 2.25 μl aliquot fi μ tumour-initiating capacity of melanoma cells.12 More intriguingly, of ampli ed cDNA was mixed with 2.5 l of 2X Ssofast EvaGreen Supermix with Low ROX (Bio-Rad, PN 172-5211, Hercules, CA, USA) and 0.25 μlof invasive genes segregate into two signatures with distinct ‘ ’ – fi sample loading agent (Fluidigm, PN 100 3738), then inserted into one of expression pro les with some cells expressing a group comprising the chip ‘sample’ inlets. A total of 100 μM of mixed forward and reverse ZEB1 and GLI2, whereas others express a distinct group comprising primers were diluted at 1:10 ratios with Tris-Ethylenediaminetetraacetic BIRC3 and FGF2. Surprisingly, cells expressing the invasive acid. Then 2.5 μl of diluted primers were mixed with 2.5 μl of Fluidigm signature genes express high levels of MITF, whereas previous ‘Assay Loading Reagent’ and individually inserted into the chip ‘assay’

© 2015 Macmillan Publishers Limited Oncogene (2015) 3251 – 3263 Melanoma heterogeneity M Ennen et al 3262 inlets. Samples and probes were loaded into M96 chips using a HX IFC Freemont, CA, USA), ITGA4 d1/500 (clone EPR1355Y 04–1131 Millipore, Controller (Fluidigm) and then transferred to a Biomark real-time PCR Billerica, MA, USA), CEACAM1 d1/100 (sc-59812 Santa Cruz, Santa Cruz, CA, reader (Fluidigm) following the manufacturer’s instructions. In total, we USA), for 24 h at 4 °C followed by three washes in PBS, cells were incubated performed eight C1 capture experiments and 16 Biomark qPCR plates. with the Alexa Fluor 594–conjugated secondary antibodies or with the CY3–conjugated secondary antibodies, diluted at 1:100 in PBS-containing Single-cell qPCR data analysis BSA and Triton X-100 for 1 h at room temperature. Nuclei were stained with 4,6-diamidino-2-phenylindole (DAPI) diluted at 1:1000 in PBS- Initial data analysis of the cycle threshold (C ) values was done with the t containing BSA and Triton X-100. The cells were then mounted in anti- ‘Fluidigm Real-time PCR analysis’ software and further data analysis and graphics were performed using R software. We worked on the complement fading medium (FluorSafe; Merck, Whitehouse Station, NJ, USA) and observed with an epifluorescence microscope (Carl Zeiss, Göttingen, of Ct values defined as expression threshold (et), where et =Cmax − Ct =30 26 Germany). − Ct as described. Absent values were replaced by 0. We excluded from the analysis cells in which no genes are detected or in which the spike R. spk1mid is not detected and to compare two independent qPCR plates, we Immunohistochemistry considered only cells selected in the two experiments. We retained one Sections (4 μm) of formalin-fixed paraffin embedded tumours were measure for each gene corresponding to that with the least number of rehydrated, heated in citrate buffer (pH = 6) in a 750 W microwave oven absent values. Then e values were displayed on heatmap images following t at full power for 15 min and allowed to cool down for 30 min at room clustering of both genes and cells calculated with the unsupervised temperature. Sections were incubated in blocking solution (PBS-containing Unweighted Pair Group Method with Arithmetic Mean and the Euclidean distance measure, that are not displayed on the figure to save space. To 2% BSA/0.2% Triton X-100) and incubated at 4 °C overnight in the presence compare experiments performed on different cDNA collections, we only of primary antibody and imaged as described above. retained genes analysed in both experiments. As data from two different experiments cannot be directly compared, we then converted et values by Xenografts their ranks using the following method. In each cell, the values xi were fi o o…o o…o All experiments were carried out with immunode cient (nu/nu) mice sorted in ascending order x1 x2 xi xN (with N the number (6-week-old) from Charles River Laboratories (Willmington, MA, USA) in of genes). Then each value x 40 was replaced by its rank r :r=N, i i N accordance with National Animal Care Guidelines (European Commission r = N-1, … r = i. All values x = 0 (absent values) were kept as 0. These N-1 i i directive 86/609/CEE; French decree no.87–848). Mice were injected ranks were used to perform clustering, whereas original e values were t subcutaneously into both hind legs with 1 × 106/flank 1205LU cells and displayed on the heatmap images. For correlation of gene expression, 6 fl 3 Pearson correlation coefficients were calculated on each pair of genes and with 5 × 10 / ank 501MEL. When tumours reached about 1000 mm ,mice displayed on heatmap images. Clustering was performed with the method were killed and their tumours removed. Tumours were minced into small fragments and incubated in HBSS (Sigma-Aldrich, St Louis, MO, USA) described above. Above each heatmap are displayed the median of the et values for the corresponding gene, with four different colours correspond- supplemented with collagenase IV (0.6 Wunsch unit/ml, Eurobio, Courta- boeuf, France), dispase II (1 mg/ml, Sigma), DNAse I (200 UI/ml, Roche), 75 μM ing to the quartiles of the distribution of all et values. CaCl2 and 125 μM MgCl2, for 30 min at 37 °C. Cells were filtered through a 100 μmporesizefilter (Dutscher, Brumath, France). After centrifugation at Cell culture and melanoma three-dimensional spheroid growth 100 g for 7 min and at 4 °C, pellets were dissolved with an HBSS buffer Melanoma cell lines 501Mel and 1205LU were maintained as a monolayer containing 200 UI/ml of DNAse I and 125 μM MgCl2. Then, cells were in RPMI 1640 medium (Sigma, St Louis, MO, USA) supplemented with 10% centrifugated (100 g, 7 min, 4 °C) and filtered before single-cell analysis. fetal calf serum. Melanoma sphere growth was induced by placing 1 × 106 cells in a 10-cm plate in Knockout-DMEM medium (Sigma) supplemented with 20% KSR (knockout serum). Cultures were grown for a minimum of CONFLICT OF INTEREST 6 days and media changed every third day. Floating spheres were fl dissociated by enzymatic dissociation (TrypLE, Life Technologies, Carlsbad, The authors declare no con ict of interest. CA, USA) into single-cell suspensions. ACKNOWLEDGEMENTS mRNA preparation, quantitative PCR and RNA-seq We thank D Dembélé for help with data analysis, all the staff of the IGBMC common mRNA isolation was performed according to standard procedure facilities, as well as F Kardouz and the staff of the Strasbourg Hospital Dermatology (Qiagen kit, Qiagen, Venlo, Netherlands). qPCR with reverse transcription Clinic, This work was supported by institutional grants from the Centre National de la was carried out with SYBR Green I (Qiagen) and Multiscribe Reverse Recherche Scientifique, the Institut National de Sante et de la Recherche Médicale, Transcriptase (Invitrogen, Carlsbad, CA, USA) and monitored by a Light- the Université de Strasbourg, the Association pour la Recherche contre le Cancer, the Cycler 480 (Roche, Basel, Switzerland). Detection of Actin gene was used to Ligue Nationale contre le Cancer, the Institut National du Cancer PAIR-melanoma normalize the results. All qPCR reactions were performed in triplicate and grant, the ANR-10-LABX-0030-INRT French state fund through the Agence Nationale used only when variations of no more than once cycle of Ct were obtained. de la Recherche under the frame programme Investissements d’Avenir labelled Primer sequences for each cDNA were designed using Primer3 Software or ANR-10-IDEX-0002-02. The IGBMC high throughput sequencing facility is a member were designed by the Fluidigm Company. Messenger RNA-seq was of the ‘France Génomique’ consortium (ANR10-INBS-09-08). ID is an ‘équipe labellisée’ performed essentially as previously described.27 For analysis, RNA-seq of the Ligue Nationale contre le Cancer. ME was supported by a fellowship from the reads were mapped onto the hg19 assembly of the human genome by using Tophat v1.4.128 and the bowtie v0.12.7 aligner.29 Quantification of Ligue Nationale contre le Cancer. gene expression was performed using HTSeq v0.5.3p5 using gene annotations from Ensembl release 67. Read counts were normalized AUTHOR CONTRIBUTIONS among libraries with the method proposed by Anders and Huber30 and implemented in DESeq v1.10.1 Bioconductor package.30 Comparison ID, ME and CK designed the experiments. ME, CK, CT-C developed between 1205Lu and 501Mel was performed using the statistical method methodologies and performed the experiments. CK and ME performed the 30 proposed by Anders and Huber and implemented in the DESeq v1.10.1 bioinformatics and computational analyses. DL provided the human melanoma Bioconductor library. Resulting P-values were adjusted for multiple testing. samples and analysed the images. ID, ME and CK wrote the paper. Genes with |log2 fold-change|41 and adjusted P-value o0.05 were considered as significantly differentially expressed. REFERENCES Immunofluorescence 1 Meacham CE, Morrison SJ. Tumour heterogeneity and cancer cell plasticity. Nature Melanoma cells were cultured on glass slides for 24 h, fixed in 4% 2013; 501:328–337. paraformaldehyde, then blocked and permeabilized with phosphate- 2 Nazarian R, Shi H, Wang Q, Kong X, Koya RC, Lee H et al. Melanomas acquire buffered saline (PBS)-containing 2% BSA/0.2% Triton X-100. After incuba- resistance to B-RAF(V600E) inhibition by RTK or N-RAS upregulation. Nature 2010; tion with primary antibodies (MITF d1/400 (ab-1 (C5) from Neomarkers, 468:973–977.

Oncogene (2015) 3251 – 3263 © 2015 Macmillan Publishers Limited Melanoma heterogeneity M Ennen et al 3263 3 Somasundaram R, Villanueva J, Herlyn M. Intratumoral heterogeneity as a therapy Kitl as a novel regulated target gene. Pigment Cell Melanoma Res 2010; 23: resistance mechanism: role of melanoma subpopulations. Adv Pharmacol 2012; 404–418. 65: 335–359. 17 Thies A, Moll I, Berger J, Wagener C, Brummer J, Schulze HJ et al. CEACAM1 4 Hoek KS, Goding CR. Cancer stem cells versus phenotype-switching in melanoma. expression in cutaneous malignant melanoma predicts the development of Pigment Cell Melanoma Res 2010; 23:746–759. metastatic disease. J Clin Oncol 2002; 20: 2530–2536. 5 Quintana E, Shackleton M, Foster HR, Fullen DR, Sabel MS, Johnson TM 18 Caramel J, Papadogeorgakis E, Hill L, Browne GJ, Richard G, Wierinckx A et al. et al. Phenotypic heterogeneity among tumorigenic melanoma cells from A switch in the expression of embryonic EMT-inducers drives the development of patients that is reversible and not hierarchically organized. Cancer Cell 2010; 18: malignant melanoma. Cancer Cell 2013; 24: 466–480. 510–523. 19 Chiaverini C, Beuret L, Flori E, Busca R, Abbe P, Bille K et al. Microphthalmia- 6 Hoek KS, Eichhoff OM, Schlegel NC, Dobbeling U, Kobert N, Schaerer L et al. In vivo associated transcription factor regulates RAB27A gene expression and controls switching of human melanoma cells between proliferative and invasive states. melanosome transport. J Biol Chem 2008; 283: 12635–12642. Cancer Res 2008; 68:650–656. 20 Dalerba P, Kalisky T, Sahoo D, Rajendran PS, Rothenberg ME, Leyrat AA et al. 7 Widmer DS, Cheng PF, Eichhoff OM, Belloni BC, Zipser MC, Schlegel NC et al. Single-cell dissection of transcriptional heterogeneity in human colon tumors. Nat Systematic classification of melanoma cells by phenotype-specific gene expres- Biotechnol 2011; 29: 1120–1127. sion mapping. Pigment Cell Melanoma Res 2012; 25: 343–353. 21 Castro FV, Al-Muftah M, Mulryan K, Jiang HR, Drijfhout JW, Ali S et al. Regulation of 8 Goodall J, Carreira S, Denat L, Kobi D, Davidson I, Nuciforo P et al. Brn-2 represses autologous immunity to the mouse 5T4 oncofoetal antigen: implications for microphthalmia-associated transcription factor expression and marks a distinct immunotherapy. Cancer Immunol Immunother 2012; 61: 1005–1018. subpopulation of microphthalmia-associated transcription factor-negative 22 Perez-Lorenzo R, Gill KZ, Shen CH, Zhao FX, Zheng B, Schulze HJ et al. A tumor melanoma cells. Cancer Res 2008; 68: 7788–7794. suppressor function for the lipid phosphatase INPP4B in melanocytic neoplasms. 9 Javelaud D, Alexaki VI, Pierrat MJ, Hoek KS, Dennler S, Van Kempen L et al. GLI2 J Invest Dermatol 2013; 134:1359–1368. and M-MITF transcription factors control exclusive gene expression programs and 23 Liu S, Ren S, Howell P, Fodstad O, Riker AI. Identification of novel epigenetically inversely regulate invasion in human melanoma cells. Pigment Cell Melanoma Res modified genes in human melanoma via promoter methylation gene profiling. 2011; 24: 932–943. Pigment Cell Melanoma Res 2008; 21:545–558. 10 Cheli Y, Guiliano S, Botton T, Rocchi S, Hofman V, Hofman P et al. Mitf is the key 24 Carreira S, Goodall J, Denat L, Rodriguez M, Nuciforo P, Hoek KS et al. Mitf reg- molecular switch between mouse or human melanoma initiating cells and their ulation of Dia1 controls melanoma proliferation and invasiveness. Genes Dev differentiated progeny. Oncogene 2011; 30: 2307–2318. 2006; 20:3426–3439. 11 Ohanna M, Cheli Y, Bonet C, Bonazzi VF, Allegra M, Giuliano S et al. Secretome 25 Haass NK, Beaumont KA, Hill DS, Anfosso A, Mrass P, Munoz MA et al. from senescent melanoma engages the STAT3 pathway to favor reprogramming Real-time cell cycle imaging during melanoma growth, invasion, and drug of naive melanoma towards a tumor-initiating cell phenotype. Oncotarget 2013; 4: response. Pigment Cell Melanoma Res (e-pub ahead of print 5 June 2014; 2212–2224. doi:10.1111/pcmr.12274). 12 Chen KG, Valencia JC, Gillet JP, Hearing VJ, Gottesman MM. Involvement of ABC 26 McDavid A, Finak G, Chattopadyay PK, Dominguez M, Lamoreaux L, Ma SS et al. transporters in melanogenesis and the development of multidrug resistance of Data exploration, quality control and testing in single-cell qPCR-based gene melanoma. Pigment Cell Melanoma Res 2009; 22:740–749. expression experiments. Bioinformatics 2013; 29:461–467. 13 Perego M, Tortoreto M, Tragni G, Mariani L, Deho P, Carbone A et al. Hetero- 27 Herquel B, Ouararhni K, Martianov I, Le Gras S, Ye T, Keime C et al. Trim24- geneous phenotype of human melanoma cells with in vitro and in vivo features repressed VL30 retrotransposons regulate gene expression by producing of tumor-initiating cells. J Invest Dermatol 2010; 130: 1877–1886. noncoding RNA. Nat Struct Mol Biol 2013; 20: 339–346. 14 Thurber AE, Douglas G, Sturm EC, Zabierowski SE, Smit DJ, Ramakrishnan SN et al. 28 Trapnell C, Pachter L, Salzberg SL. TopHat: discovering splice junctions with Inverse expression states of the BRN2 and MITF transcription factors in melanoma RNA-Seq. Bioinformatics 2009; 25: 1105–1111. spheres and tumour xenografts regulate the NOTCH pathway. Oncogene 2011; 30: 29 Langmead B, Trapnell C, Pop M, Salzberg SL. Ultrafast and memory-efficient align- 3036–3048. ment of short DNA sequences to the human genome. Genome Biol 2009; 10:R25. 15 Strub T, Giuliano S, Ye T, Bonet C, Keime C, Kobi D et al. Essential role of micro- 30 Anders S, Huber W. Differential expression analysis for sequence count data. phthalmia transcription factor for DNA replication, mitosis and genomic stability Genome Biol 2010; 11: R106. in melanoma. Oncogene 2011; 30: 2319–2332. 31 Denecker G, Vandamme N, Akay O, Koludrovic D, Taminau J, Lemeire K et al. 16 Kobi D, Steunou AL, Dembele D, Legras S, Larue L, Nieto L et al. Genome-wide Identification of a ZEB2-MITF-ZEB1 transcriptional network that controls analysis of POU3F2/BRN2 promoter occupancy in human melanoma cells reveals melanogenesis and melanoma progression. Cell Death Differ 2014; 21: 1250–1261.

Supplementary Information accompanies this paper on the Oncogene website (http://www.nature.com/onc)