The EMT Transcription Factor ZEB2 Promotes Proliferation of Primary and Metastatic Melanoma While Suppressing an Invasive, Mesenchymal-Like Phenotype

Cutaneous melanoma is the most aggressive form of skin cancers and remains a major clinical challenge. Malignant transformation of melanocytes is often associated with constitutive MAPK signaling, mostly driven by activating mutations in BRAF or NRAS, and inactivation of tumor suppressors such as PTEN, CDKN2A, or TP53. Melanoma is a highly heterogeneous tumor, composed of phenotypically distinct subpopulations (1). While the mechanisms that contribute to intratumor heterogeneity remain poorly understood, it is without a doubt a major contributor to both intrinsic and acquired therapy resistance and metastatic dissemination. Identification of mediators of intratumor heterogeneity may therefore lead to new therapeutic avenues that exhibit longer-lasting antitumor responses, prevent metastatic dissemination and identification of prognostic markers for therapy stratification.

Melanoma progression and therapy resistance does not rely solely on mutation-driven mechanisms. There is increasing evidence that nongenetic mechanisms that modulate the epigenetic landscape and/or transcriptional and metabolic state, also contribute to progression and drug resistance (2). It is increasingly recognized that phenotypic heterogeneity stems, at least partly, from the ability of melanoma cells to switch back and forth between a proliferative cell state and a mesenchymal-like, invasive state. Both cell-intrinsic and -extrinsic (i.e., microenvironmental) factors such as oxygen, cytokines, and growth factors modulate this reversible phenotype switch. Over recent years, the reported reversible mechanisms of melanoma invasion have supported this model (3–7). Phenotype switching is reminiscent to epithelial-to-mesenchymal transition (EMT). This phenotype switching event was also shown to promote therapy resistance and is a key driver of metastatic spreading. In epithelial tissues, EMT-inducing transcription factors, such as SNAIL and ZEB family members, in addition to promote invasion and metastasis may also contribute to de novo carcinogenesis by bypassing cellular senescence and expanding the cancer stem cell pool. The functional role of EMT drivers has thus been expanded beyond their ability to induce a mesenchymal transcriptional program. The regulation and role of EMT-inducing transcription factors in nonepithelial contexts such as melanocytes and melanoma may even be more diverse and pleiotropic. We have, for instance, shown that ZEB2 is an important driver of normal melanocyte development and differentiation and established correlations between nuclear ZEB2 expression in the primary melanoma tumor and recurrence-free survival (8). We also showed that ZEB2 positively regulate MITF levels and/or activity during melanocyte homeostasis. These data raised the possibility that ZEB2 may contribute to melanoma plasticity and, as such, modulate melanoma growth and progression and metastatic dissemination in a manner that is unconventional for a EMT-TF. To test this possibility, we examined the contribution of ZEB2 to melanoma development and metastatic spreading. Our data establish a critical role for ZEB2 as a key modulator of melanoma heterogeneity and cell population dynamics in vivo.

Tissue samples were obtained from the Department of Dermatology, Universitair Ziekenhuis Gent (Ghent, Belgium), from the Department of Pathology, University Hospital Leuven, KU Leuven, Belgium and from archival paraffin-embedded patient samples from St. Vincent's University Hospital (Dublin, Ireland). All patient specimens were used in accordance with institutional and national policies at the respective locations, with appropriate approval provided by the relevant Ethics committees at the respective institutions. All patient-related information was anonymized.

M-series primary melanoma cell cultures have been established from surplus material from cutaneous melanoma metastases. Written informed consent was approved by the local institutional review board (IRB; EK647 and EK800). Melanoma cells were brought into culture as described previously (9). SK-MEL28 and 501MEL were obtained from ATCC. TGFβ treatment of cell lines M000921 and M010817 was performed with recombinant hTGFβ, 5 ng/mL (PeproTech, 100-21). All cell lines were screened periodically for Mycoplasma contamination.

Mice were kept in accordance with the institutional guidelines regarding the care and use of laboratory animals and all procedures and experiments were approved by the institutional review ethics committee. Mice with the following alleles have been described elsewhere: conditional Zeb2^fl/fl, Rosa-YFP, Rosa-Zeb2TG/TG-IRES-GFP, Tyr::CreERT2 and Tyr:NRAS p53 (10–13). Conditionally, Rosa-Zeb2TG/TG-IRES-GFP–overexpressing mice were generated using G4 hybrid ES cells and the Gateway-compatible Rosa26 locus targeting vector, as described previously (14). For TyrCreERT2 induction, the dorsal hair of anesthetized mice was removed with a (50:50 w/w) mix of beeswax and gum rosin (Sigma-Aldrich, 243221 and 60895-250G). 4-hydroxytamoxifen (4-OHT; 70% Z-isomer, Sigma Sigma-Aldrich, H-6278) was dissolved at 50 mg/mL in DMSO at 37°C under gentle agitation. 4-OHT was further diluted at 5 mg/mL in ethanol and 5 μL of the 4-OHT solution was applied using a pipette on the depilated area directly. All Tyr::CreERT2 animals including control animals received 4-OHT treatment. Mice were examined at a regular basis and sacrificed at an endpoint defined by adverse clinical symptoms, such as weight loss (>15%), a hunched posture or multiple skin tumors (diameter > 5 mm). Quantification of the percentage of mice bearing macro-metastases in each organ was visually examined under the binocular at necropsy.

Tumors, organs, and skin were isolated and fixed overnight in 4% paraformaldehyde solution, dehydrated, embedded in paraffin and cut into 5-μm sections. For histology, samples were stained with hematoxylin and eosin. For IHC staining, antigen retrieval was done in citrate buffer and endogenous peroxidases were blocked with 3% H₂O₂ in methanol. The sections were incubated with primary antibodies and stained with biotin-conjugated secondary antibodies followed by Streptavidin-HRP based development (substrate development with DAB). When necessary, the signal was amplified using the Tyramide Signal Amplification (TSA) kit (Perkin Elmer). TMA microarray slides containing 178 primary melanoma samples were digitally scanned on an Aperio ScanScope scanner and analyzed using Aperio algorithms. Correlation between the IHC data was compared with clinical parameters using the two-tailed Fisher exact test and the χ² test. Immunostaining against ZEB1 and ZEB2 on mouse sections was performed using in-house generated rabbit and mouse mAb, respectively, GFP was detected using a rabbit monoclonal antibody (clone D5.1 XP Rabbit mAb #2956, Cell Signaling Technology). Rabbit anti-ZEB2 (HPA003456, Sigma) and rabbit anti-ZEB1 (Clone H-102, sc-25388, Santa Cruz Biotechnology) was used for human sections.

For generation of melanoma cells overexpressing ZEB1, virus production was performed in HEK293T cells using calcium phosphate transfection with pMD2.G (envelope plasmid), psPAX2 (packaging plasmid), and pSIN-TRE-GW-3xHA ZEB1. Transduced cells were selected by puromycin (2 μg/mL for all cell lines except 501MEL and SK-MEL28: 1 μg/mL). The shRNA vectors are pLKO1.5-based from Sigma, Mission shRNA series.

For RNAi, we used siRNA pools (Dharmacon) and a scrambled siRNA pool (Dharmacon) as a control. Transfections were performed with HiPerfect (Qiagen) according to the manufacturer's instructions (siZeb1: Dharmacon, M-051513-01-0005; siZeb2: M-059671-01-0005; siZEB1: M-006564-02-0005; siZEB2: M-006914-00-0005). Plasmids carrying myc-tagged ZEB2, HA-tagged MITF, or empty cDNA expression cassettes were transiently transfected. 501MEL cells were transfected using FugeneHD (Promega E-231): cells were seeded in a 6-well to reach 75% confluency prior to the transfection in 3 mL growth medium with a 4:1 ratio DNA:Lipofectant. Subsequently, for each well, 3.3 μg total DNA was diluted in 160 μL OptiMEM, 13 μL Fugene HD reagent was added, and vortexed briefly, and the solution was incubated at room temperature during 5 minutes, and 150 μL of the mixture was added dropwise on the cells (in 3 mL growth medium). MITF-VP16 chimera, a transcriptionally more active MITF-derivative, and MITF cDNA expression constructs were kindly provided by Prof. Vachtenheim (Department of Transcription and Cell Signaling, Charles University, Prague, Czech Republic).

8W10E array plates (Applied Biophysics) were washed and treated with 10 mmol/L cysteine for 30 minutes at room temperature. Next, for HUVEC assays, array plates were coated with laminin before cell seeding. HUVECs (Lonza) were seeded and challenged with melanoma cells once stable impedance (and visible confluency of the HUVECs) was achieved. As a control, heat-inactivated cells (10 minutes at 55°C) in conditioned medium were included. Next, arrays were mounted onto the array holder connected to the electric cell surface impedance sensing (ECIS) module and placed in a standard incubator at 37°C and 5% CO₂ throughout the experiment. All impedance measurements were performed in the multifrequency measurement mode.

RNA was extracted from cell cultures by using RNeasy extraction columns (Qiagen) or TRIsure lysis buffer (BioLine) according to the manufacturer's guidelines. For tumors, after collecting, the tissue is subsequently disinfected in a povidone–iodine solution (iso-Betadine) and washed twice in 70% ethanol and three times in PBS. Skin preparations were carefully dissected, followed by removal of fat tissue using a scalpel. Tissue pieces were incubated in RNAlater (Ambion, AM7021; BioLine, BIO-38032) at 4°C overnight. Subsequently, the pieces were put in a petridish in 1 mL TRISure on ice and cut into small pieces. Subsequently, the pieces were homogenized using a mechanical mixer followed by passing the sample five times through a 21-gauge needle. To reduce the carryover of RNases during further processing, the homogenized sample is centrifuged for 15 minutes, 20,000 × g, 4°C to eliminate debris. The supernatant is transferred to a new RNAse-free Eppendorf tube and incubated for 5 minutes at room temperature. After phase separation by adding 200 μL chloroform and vigorous shaking, 250 μL from the aqueous upper phase was added to 500 μL cold RNase-free ethanol. Samples were snap-frozen in liquid nitrogen and stored at −80°C.

RNA was treated with 1 U of RNAse-free DNase RQ1 (Promega) per μg RNA for 30 minutes at 37°C in appropriate buffer. DNAse was inactivated by incubation in Promega stop solution for 10 minutes at 65°C. Bulk Mg2⁺ was removed by using Amicon ultra 0.5-mL centrifugal filters (Millipore, UFC510096) in two consecutive diluting washes. cDNA synthesis was performed with iScript advanced cDNA synthesis kit (Bio-Rad, 172-5038) following the manufacturer's instructions. Quantitative PCR was done using the SYBR Master Mix Kit SensiFast SYBR No-Rox Kit (Bioline, CSA-01190) for the genes of interest and reference genes. Plates were run on the LightCycler 480 (Roche). The average threshold cycle of triplicate reactions was used for all subsequent calculations using the ΔC_t method, calculated in qBasePLUS (BioGazelle). The following reference genes were used for a first read-out: Calm2, Matr3, Hmbs, Rpl13a, Oaz1, Gapdh, Ywhaz, Tbp, Cox4i1, and Eef1a for mouse, EEF1A1, DCAF6, OAZZ1, CALM2, MATR3, COX4I1, RPL37A, FTL, SH3KBP1, TPT1 for human. Subsequently, the most stable reference (housekeeping) genes from this set were determined via qBase+ (geNorm) prior to repeating the experiment.

Melanoma cell cultures were established from patient biopsies. Patients gave their informed consent and was approved by the local IRB (EK647 and EK800). RNA from the melanoma cell cultures were isolated with the QIAGEN RNeasy Kit. RNA capture was performed with TruSeq RNA Library Prep Kit v2 (Illumina) and sequenced on a HiSeq4000. RNA counts were quantified from single-end reads using STAR aligner (15). Normalization of RNA counts was performed with edgeR (16). RNA sequencing (RNA-seq) data of TGFβ treatment human melanoma cell lines available at GEO database (GSE148767).

ZEB1 and ZEB2 expression values (TPMs) were inferred from single-cell RNA-sequencing analyses (scRNA-seq) data (GSE115978) from human malignant melanoma cells and log-normalized expression values represented for n = 1,951 cells.

Seurat R package (version 3.1.3) was used to log normalize the data, perform dimensionality reduction analysis, clustering of the cells and performing differential gene expression. We extracted the melanoma cells that were annotated by Jerby-Arnon and colleagues (17). However, we deleted a small cluster of contaminating CD3E⁺ CD45⁺ T cells that were initially annotated as melanoma cells in the original dataset. 526 of 1,951 cells were double negative for ZEB1 and ZEB2. Remaining cells came from 23 patients. 1,373 cells showed ZEB2 and 238 cells ZEB1 expression. ZEB1 and ZEB2 expression per single cells are significantly anticorrelated, as shown by Spearman correlation (ρ = −0.28; P < 2.2E-16).

The proliferative capacity of melanoma cells was analyzed by performing the Sulforhodamine B (SRB) proliferation assay. Briefly, cells were seeded at 3,000 to 7,000 cells per 96-well and at day 0 and 4, the cells were fixed with 30% (wt/vol) trichloroacetic acid for 1 hour, and stained for 30 minutes at room temperature with 0.4% SRB in 1% acetic acid. Excessive dye was washed off with 1% acetic acid and dissolved in 10 mmol/L Tris buffer (pH 10.5). The plates were measured with a microplate reader (Bio-Rad, Eke, Belgium) at 570 nm. Day 0 was used as a reference sample.

Tissues were mechanically separated from the dermis and ground into small pieces prior to lysis, culture cells were immediately lysed in Laemmli-lysis buffer (50 mmol/LTris-HCl pH 6.8, 10% glycerol, 2% SDS) using standard practices. After sonicating and centrifuging the samples, 20 μg of protein was separated on a acrylamide gel and transferred to a polyvinylidene difluoride membrane. Membranes were incubated with primary antibodies and appropriate HRP-labeled secondary antibodies (GE Healthcare). Primary antibodies were diluted in 3% milk. Detection was performed with the Western Lightning Chemiluminescence Reagent Plus Kit (PerkinElmer) or the Immobilion Western HRP Substrate (Millipore).

Petri dishes (60-mm diameter) were coated with 0.5% Bacto agar (Difco) in growth medium. A suspension of 1,104 cells in growth medium containing 0.35% Noble agar (Difco) was seeded on top of the agar coating and then covered with growth medium. Fresh medium was applied weekly. Cell colonies were allowed to form over a time period of 55 days, after which, they were stained for 1 hour with 0.005% crystal violet (Sigma) in PBS and photographed.

An QCM ECMatrix 24-well (8 μm) Fluorimetric Cell Invasion Assay kit (Chemicon; Sigma-Aldrich) was used according to the manufacturer's protocol. An insert polycarbonate membrane with a pore size of 8 μm) was coated with a thin layer of ECMatrix. Cells were seeded in the insert (top chamber) at a density of 25 × 10³ cells/well in serum-free RPMI. RPMI medium supplemented with 10% FCS was added to the bottom chamber. Following 24-hour incubation, invading cell numbers were determined via fluorescent read-out. For migration assays, similar inserts were not coated with ECMatrix.

Results were evaluated using the statistical tests and indicated with P values in the figure legends. P < 0.05 was considered statistically significant. Survival curves and metastasis-free survival were analyzed by Kaplan–Meier analysis, which were compared by log-rank (Mantel–Cox–Kaplan Meyer Plot). GraphPad Prism 7 was used to perform log-rank (Mantel–Cox), Student t test, two-way ANOVA followed by post-hoc Tukey HSD (SRB assays). R (http://www.cran.r-project.org) was used to perform Fisher exact test and the χ² test (for TMA staining), Benjamini–Hochberg Multiple Testing Correction (adjusted) P value (for bulk RNA-seq data). Differential gene expression during single-cell analysis was performed via a nonparameteric Wilcoxon rank sum test in Seurat.

ZEB2 expression in human melanoma is heterogenous (Fig. 1A). Such a heterogeneous pattern was also observed in lesions from the previously established NRAS^Q61K-driven mouse model of melanoma (Fig. 1B). Note that NRAS is consistently altered in 20% of melanomas and is the second most common oncogenic driver mutation (18). Constitutive melanocyte-specific expression of one copy of the activated form of human NRAS (NRAS^Q61K) triggers melanoma development in a small fraction of Tyr::NRAS^Q61K mice. The mouse skin is hyperpigmented and reminiscent of patients with a giant congenital nevus (GCN). These nevi are large pigmented lesions that frequently progress to melanoma. On average 15%–20% of Tyr::NRAS^Q61K mice develop melanomas within 1 year in the absence of secondary genetic alterations (11). Concomitant loss of p53 or p16^INK4A functionally increases the incidence to nearly 100% and shortens the latency significantly (11). The biological significance of the varying ZEB2 levels remains unclear. We hypothesized that graded levels of ZEB2 may associate with distinct melanoma phenotypic properties. To test this possibility, we engineered a conditional ZEB2 allelic series onto the NRAS^Q61K-driven mouse melanoma background (Fig. 1C). The Tyr::NRAS^Q61K;TyrCreERT²;Trp53^lox/lox mice (henceforth NRAS^Q61K p53^−/− mice) were intercrossed with mice carrying a Zeb2 conditional transgenic allele, Rosa26-(loxP-STOP-loxP)-Zeb2. Note that the Zeb2 open reading frame is followed by an IRES-EGFP reporter sequence, GFP expression can therefore be used to fate map transgenic cells. Mice carrying two copies of this allele are thereafter referred to as Rosa-Zeb2. In addition, the Tyr::NRAS^Q61K;TyrCreERT²;Trp53^lox/lox mice have also been intercrossed with mice carrying a conditional Zeb2 knockout allele (Zeb2^fl/fl; refs. 10, 14). To modulate ZEB2 expression the compound mice were exposed to topical application of 4-hydroxytamoxifen (4-OHT) onto their back skin. Nevus formation and melanoma development was monitored weekly.

Strikingly, Zeb2 ablation dramatically attenuated nevus and melanoma outgrowth (Fig. 2A and B). Moreover, targeting Zeb2 in a NRAS^Q61K p53^−/− background delayed melanoma initiation (Fig. 2C). Interestingly, all analyzed pigmented tumors that arose in NRAS^Q61K p53^−/− Zeb2^fl/fl mice contained only Zeb2 wild-type melanoma cells, indicating that these lesions originated from melanoma cells that escaped Zeb2 ablation (Supplementary Fig. S1A and S1B). As such, no genuine Zeb2-knockout melanoma tumors are formed. These results indicate that ZEB2 is required for the growth and expansion of primary NRAS-driven melanoma. In contrast, ectopic ZEB2 overexpression accelerated melanoma initiation, decreased the latency and increased melanoma penetrance. ZEB2 expression did not specifically affect metastatic spread on a p53-null–mutated background.

To assess whether ZEB2 can promote melanomagenesis in the presence of functional p53, we conditionally overexpressed ZEB2 in a Tyr::NRAS^Q61K (p53 wild-type) background. Ectopic expression of transgenic ZEB2 and GFP was confirmed via immunoblot analysis (Supplementary Fig. S2A). Forced expression of ZEB2 in Tyr::NRAS^Q61K Rosa26-Zeb2 mice (NRAS^Q61K Rosa-Zeb2) did not significantly affect tumor latency or incidence (Supplementary Fig. S2B). Ectopic ZEB2 expression is therefore not sufficient to transform NRAS^Q61K–expressing melanocytes. Molecular analysis of NRAS^Q61K and NRAS^Q61K Rosa-Zeb2 tumor cells showed an increased expression in genes that belong to the previously described melanoma “proliferative” gene signature including Mitf, Mc1r, and Sox10 (19, 20). This increase occurred concomitantly to a decrease in expression of Zeb1 expression (Fig. 3A). RNAi-mediated Zeb2 knockdown in melanoma cell lines Mel3 and Mel6 derived from the Rosa-Zeb2 melanoma mouse model attenuated and reversed this gene signature, as illustrated by a decrease in Mitf levels (Fig. 3B). This was accompanied by an increase in Zeb1 levels. As soon as cells regained Zeb2 expression (6 days after transfection of the siRNA molecules), the “proliferative” gene signature reemerged, highlighting the reversibility and dynamic nature of the ZEB2-driven transcriptional program (Fig. 3C). Importantly, Zeb2 depletion led to a decrease in the proliferation potential of primary melanoma cell lines derived from 5 different NRAS^Q61K Rosa-Zeb2 mice and the B16 melanoma cell line (Supplementary Fig. S3A) and induces a melanoma “invasive/mesenchymal-like” gene expression in B16 cells (Supplementary Fig. S3B; refs. 19, 20). Zeb2 depletion, Mitf depletion or ectopic Zeb1 expression in B16 cells promoted an invasive and migratory phenotype in vitro (Supplementary Fig. S3C and S3D). A transendothelial migration assay as described by Tarantola and colleagues (21) and Keese and colleagues (22) showed how ZEB1 promotes melanoma cells to attach to and cause retraction of the endothelial monolayer (Supplementary Fig. S3E). We validated our conclusions in a panel of human melanoma cell lines. Lu1205 melanoma cells show increased or attenuated invasive capacity upon respectively enhanced or reduced ZEB1 expression (Supplementary Fig. S3F). Upon overexpression of ZEB1 in the human 501MEL melanoma cell line, we readily observed a dramatic growth attenuation (Supplementary Fig. S3G). Three days following doxycycline administration, the 501MEL iZEB1 cell line altered predominantly genes associated with the proliferative phenotype, whereas the invasive gene repertoire is not induced yet to the fullest. ZEB1 induction shifts the cells toward an invasive gene signature after 6 days, similar to ZEB2 depletion (Supplementary Fig. S3G). ZEB1-mediated depletion of MITF protein levels and increased migratory capacities were confirmed in the differentiated melanoma cell lines 501MEL and SK-MEL28 (Supplementary Fig. S3H and S3I).

We next assessed whether the ectopically induced Zeb2 transcriptional program could be spontaneously reversed in the in vivo setting. We took advantage of the GFP-reporter gene expression to fate map the Zeb2-transgenic cells in primary NRAS^Q61K Rosa-Zeb2 melanoma (Fig. 3D). Remarkably, an attenuation of ZEB2 transgenic protein levels was observed in a fraction of the melanoma cells thus recapitulating the naturally occurring cellular heterogeneity observed in the NRAS^Q61K wild-type Zeb2 melanoma lesions. Because, the Zeb2 transgene does not contain untranslated regions, we concluded that silencing of its expression is likely due to a (post)translational regulatory event. Interestingly, tumor areas that had lost Zeb2-transgenic expression became strongly positive for Zeb1. This was in keeping with the above in vitro and with our previous observation that Zeb2-deficient melanocytes acquire a dedifferentiated stem cell phenotype exhibiting elevated Zeb1 levels (8).

Complete necropsy of NRAS^Q61K and NRAS^Q61K Rosa26-Zeb2 tumor-bearing mice revealed the presence of large pigmented pulmonary metastases that had colonized the lung parenchyma, the liver, and muscle tissue in a large fraction of Zeb2 transgenic mice (n = 26), whereas no macrometastases could be observed in the control animals (n = 19; Fig. 4A). In the lung parenchyma of NRAS^Q61K control animals, we did observe small clusters of melanoma cells, only visible through microscopic analysis of hematoxylin and eosin (H&E) sections. Although such pulmonary micrometastases were detected in 88% of the control animals, none of them was able to succesfully grow to a visible metastatic lesion seen with the naked eye (Fig. 4B). Such micrometastases were not observed in other organs. From the NRAS^Q61K Rosa26-Zeb2 mice, 46% of the mice did have many pulmonary macrometastases (>50 lesions), in the remaining 54%, only micrometastases were detected through the analyses of H&E sections (Fig. 4C). Collectively, these findings indicated that ectopic Zeb2 expression facilitates the outgrowth of “dormant” disseminated melanoma cells and promote the formation of successful metastases.

RNA-seq analysis of a large cohort of human primary melanoma cultures revealed a strong inverse correlation between expression levels of ZEB2, MITF, and the MITF-target gene TYR, and an inverse correlation between MITF and ZEB1 expression. ZEB1 expression was correlated with AXL, WNT5A, TCF4, CDH2, and FN1 expression (Fig. 5A). These correlations were confirmed via qRT-PCR on a selection of cultures and a panel of human melanoma cell lines (Fig. 5B). ZEB2-positive melanomas exhibit a differentiated and proliferative gene signature, whereas ZEB1 melanomas express the invasive gene signature, indicating that melanoma switch between a proliferative MITF^highZEB2^highZEB1^low state and an invasive and metastatic ZEB1^highZEB2^lowMITF^low state. Upon RNAi-mediated silencing of ZEB2, using two independent shRNAs (sh1 and sh2 ZEB2), SK-MEL28 melanoma cells and 501MEL cells loose MITF expression and exhibit, a severe growth retardation, while gaining expression of invasive markers including ZEB1 (Fig. 5C and D; Supplementary Fig. S4A and S4B). ZEB2 depletion also hampers the ability of these melanoma cells to form colonies in vitro (Fig. 5E; Supplementary Fig. S4C). Upon restoration of Zeb2 expression following doxycycline withdrawal, the cells recovered a proliferative phenotype, further confirming the reversibility of the phenotype switch (Fig. 5E). RNAi-mediated depletion of ZEB2 or MITF in 501MEL cells, results in a strong proliferation defect (Supplementary Fig. S4D). The decreased proliferation in siZEB2-treated 501MEL cells, but not siMITF-treated cells, can partially be rescued by siZEB1 treatment (Supplementary Fig. S4D and S4E). Reexpression of the MITF-VP16 chimera, a transcriptionally more active MITF-derivative is sufficient to restore growth in ZEB2/MITF-depleted cells. In contrast, ZEB2 reexpression is not able to rescue growth inhibition upon MITF depletion. This supports the model that ZEB2 is required for MITF functionality and suppression of the alternative ZEB1 state but is unable to compensate for MITF loss, while ZEB1 may act downstream of the ZEB2/MITF interplay and contributes to the phenotype switch when ZEB2/MITF functionality is compromised. Taken together, our data identified ZEB2 as a key modulator of the proliferative to invasive phenotype switch both in mouse and human melanoma cells. To further substantiate these findings, we examined the correlation between ZEB2 expression and Breslow Depth of Invasion index. This is used as a standard prognostic factor in melanoma pathology measuring how deep the tumor cells have invaded the dermis and the surrounding tissue in human melanoma biopsies. A robust association between intense nuclear ZEB2 immunoreactivity and lower Breslow score was identified, thus confirming that loss of ZEB2 in primary melanoma is associated with increased invasiveness in human patients (Fig. 6A and B). Moreover, IHC expression analyses on human melanoma samples made clear that ZEB2 is attenuated and ZEB1 nuclear expression is enhanced in the invasive front at the deepest margins of the tumor, strongly associating ZEB1 with the invasive behavior of human primary melanoma cells (Fig. 6C). To assess whether the ZEB1- and ZEB2-expressing cells are distinct subpopulations in human melanomas, we analyzed single-cell expression data of human melanoma (GSE115978; ref. 17). This analysis made clear that the majority of the melanoma cells (1,951 cells coming from 33 patients analysed by scRNA-seq) are ZEB2^high ZEB1^low and that ZEB2 and ZEB1 are clearly anticorrelated as shown by Spearman correlation (ρ = −0.28, P < 2.2E-16; Fig. 6D). Interestingly, ZEB2^low ZEB1^high cells have a clear loss of proliferation/differentation markers such as MITF, TYR, PMEL, and SOX10 and upregulation of invasion markers such as COL1A2, TPM2, ALDH1A3, and TCF4 (Fig. 6E).

TGFβ1 has been shown to promote melanoma switching from a proliferative to an invasive state (23, 24). Accordingly, exposure of two early-passage melanoma cultures (M010817 and M000921), expressing a high ZEB2/ZEB1 ratio to TGFβ1 increased ZEB1 and fibronectin expression, while decreasing MITF levels, both at protein and RNA levels (Fig. 7A and B). This TGFβ1 treatment shifted the gene expression profiles of these cells toward an invasive/mesenchymal-like phenotype, as illustrated by the upregulation of genes such as AXL and WNT5A (Fig. 7A; Supplementary Fig. S5). Interestingly, while ZEB2 mRNA levels were not affected by this treatment in the M010817 cells 48 hours after TGFβ1 exposure, a clear drop in ZEB2 protein levels was observed, suggesting the involvement of a posttranslational mechanism in ZEB2 downregulation. Withdrawal of TGFβ1 after 96 hours, in combination with inhibitor SB431542 (inhibiting kinase activity of the TGFβ1 receptors ALK4, -5 and -7) to block all positive autocrine feedback loops, was able to revert this phenotype after 7 days. These data show that well-established phenotype switching inducer, TGFβ1, is sufficient to revert the ZEB2/ZEB1 ratio, an event that is likely to directly contribute to phenotype switching.

We previously showed that ZEB2 is a gatekeeper of melanocyte development and differentiation (8). Here, we provide evidence that ZEB2 promotes the growth of both primary and metastatic melanoma lesions, while suppressing ZEB1 expression and thereby an invasive/mesenchymal-like transcriptional program. Genetic ablation of Zeb2 indeed hampered outgrowth of primary melanoma in mice, whereas ectopic expression enhanced melanoma proliferation and the growth of melanoma at both primary and secondary sites. Therefore, Zeb2 cannot be regarded as a genuine tumor suppressor, as suggested previously (25). In support of our findings, ZEB2 was instead recently identified as an “Achilles heel” of melanoma growth in a genome-wide RNAi-based loss-of-function screen aiming at establishing a “cancer dependency map” (26). Here, we show how strong nuclear ZEB2 expression in human primary melanoma was associated with lower Breslow index and increased metastasis-free survival. An increased metastatic burden in mice ectopically expressing Zeb2 has, at first, been difficult to reconcile with these observations. However, we noticed that although presence of increased Zeb2 mRNA levels, ZEB2 protein levels were not elevated in all GFP-tagged transgenic cells, indicating that specific microenvironmental cues were able to destabilize ZEB2 protein expression in a fraction of the cells within the primary tumor.

We hypothesize that the transgenic cells that lose ZEB2 expression are the cells that leave the primary lesions and eventually contribute to the seeding of metastasis-initiating cells in various distant organs. Interestingly, in control Tyr::NRAS^Q61K mice, only microscopic clusters of tumor cells were observed at distant sites. In contrast, multiple macroscopic colonies could be detected in Zeb2 transgenic mice. We therefore propose that increased Zeb2 expression facilitates the awakening of otherwise dormant ZEB2-negative cells. As such, the observed oscillation of both endogenous and transgenic ZEB2 levels within the primary tumor suggests that the increased metastatic burden may be due to enhanced outgrowth at distant sites, rather than the promotion of primary tumor invasion and metastasis. The ZEB2-mediated growth effect at distant sites is reminiscent of the enhancement of melanoma cell growth at primary cutaneous sites. When melanoma cells reach growth-permissive microenvironments, Zeb2 expression is reinstated to enable expansion. Melanoma cells that acquire elevated Zeb2 expression have a growth advantage outcompete Zeb2-negative cells and contribute to the development of macrometastases, an event that is generally considered as one of the severe rate-limiting steps during metastasis and that is likely more efficiently bypassed via the ZEB2 transgenic expression (27). Although ZEB2 reinforces a differentiated and proliferative phenotype, forced expression of Zeb2 did not necessarily eliminate phenotype-switching abilities in our mouse model of melanoma. Indeed, to achieve local invasion, tumor cells may temporarily abolish ZEB2 protein levels, which results in an oscillation from ZEB2 toward ZEB1, even in tumors with transgenic ZEB2 expression. Phenotype switching involving a dual role of ZEB2 in melanoma development shows striking parallels with the need for reversibility of EMT in metastasis of carcinomas (28, 29). Metastases of the most common human carcinomas often exhibit a redifferentiation. Reexpression of differentiation markers such as E-cadherin, induced by microenvironmental signals, confers a selective advantage and enhances communication with the neighboring parenchyme (30). In regards to this, it is generally accepted that organ colonization by circulating tumor cells is the most complex and rate-limiting step in the metastatic cascade. Our data also establish antagonistic roles for ZEB2 and ZEB1, two members of one EMT TF family. We provide evidence that the ZEB2/ZEB1 ratio is a critical determinant of the melanoma phenotypic state. A high ratio is associated with proliferation, whereas a low ratio favors invasion and migration. The strict dichotomy between proliferation and invasion can be nuanced, as it has been reported that proliferation and invasion are not always mutually exclusive states in melanoma and that several intermediate states exist (31, 32). Also, our single-cell analyses indicate that ZEB1/ZEB2 are coexpressed in a subset of melanoma cells. As elevated ZEB1 levels have been associated with drug resistance of many cancer types and increased survival of circulating tumor cells after extravasation (33), it is tempting to speculate that coexpression of ZEB2 may favor the growth of drug-resistant ZEB1^high melanoma cells or the survival of ZEB1^high-invasive circulating tumor cells. While we hypothesize that dormant, disseminated melanoma cells need to restore ZEB2 expression to a certain minimal level to regain proliferation potential, survival of circulating tumor cells prior to extravasation might also be enhanced by ZEB2 levels, contributing to the successful organ colonization of ZEB2-positive melanoma cells. While the role of oncogene signaling to direct the reversible reprogramming of EMT-inducing transcription factors in melanoma was characterized thoroughly by Caramel and colleagues, the microenvironmental signals that contribute to this ZEB1/ZEB2 switch in melanoma are largely unknown (25). Changes in the microenvironment are presumed to direct melanocyte plasticity and melanoma phenotype switching (2). In line with this, phenotype switching can initially be triggered by genetic instability that sensitizes the cells toward microenvironmental cues. TGFβ, which is abundant in the tumor microenvironment, is known to regulate phenotype switching (24). The extensive interplay between ZEB transcription factors and members of the TGFβ family in the context of epithelial cells is well established, but has been poorly examined in melanocytes and melanoma (34, 35). Our findings indicate that TGFβ may affect melanoma phenotype switching, at least by partly modulating the ZEB2/ZEB1 ratio. This antagonstic effect on ZEB1/ZEB2 levels is in contrast with the classic view that TGFβ activates both EMT inducers in epithelium-derived malignancies. However, because the role of ZEB2 in carcinoma progression is unrelated to the functions of ZEB2 in neural crest–derived malignancies, the regulation and interplay with TGFβ is also likely to be different. From a therapeutic standpoint, our data indicate that approaches aiming at targeting ZEB2 expression/function may be beneficial to limit the growth of metastatic lesions. However, such an approach is expected, in the same time, to promote the transition towards an invasive state and as such, favor metastatic spreading. Such an approach should therefore only be considered very carefully, in meticulously selected cases, or in combination with drugs that compromise the viability of invasive melanoma cells as well. Our data also call for the careful (re-)examination of the individual biological and pathologic functions of the EMT TFs as the repertoire of their activities keeps expanding beyond what was previously anticipated.

W.M. Gallagher has a paid consulting position and ownership interest with OncoMark Limited, and also has received commercial research funding and has an advisory relationship with Carrick Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N. Vandamme, G. Denecker, G. Berx

Development of methodology: N. Vandamme, G. Denecker, G. Blancke, Ö. Akay, G. Berx

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Vandamme, G. Denecker, K. Bruneel, G. Blancke, Ö. Akay, J. Taminau, E. De Smedt, N. Skrypek, W. Van Loocke, J. Wouters, D. Nittner, C. Köhler, P.F. Cheng, M.I.G. Raaijmakers, M.P. Levesque, F. Rambow, V. Andries, B. Balint, W.M. Gallagher, J.J. Haigh, P. Van Vlierberghe, S. Goossens, J.J. van den Oord, J.C. Marine, G. Berx

Analysis and interpretation of data (e.g. statistical analysis, biostatistics, computational analysis): N. Vandamme, J. De Coninck, F. Rambow, G. Berx

Writing, review, and/or revision of the manuscript: N. Vandamme, J.C. Marine, G. Berx

Administrative, technical, or material support (i.e. reporting or organizing data, constructing databases): U. Girish Mallya, M. Rafferty, B. Balint, W.M. Gallagher, D.S. Darling, L. Brochez, D. Huylebroeck, J.J. van den Oord

Study supervision: G. Berx

N. Vandamme was supported by a personal PhD fellowship at the Institute for the Promotion of Innovation by Science and Technology in Flanders (IWT) (2012-2016) and a personal PhD fellowship from Kom op tegen Kanker (2016-2017). K. Bruneel and E. de Smedt are predoctoral fellows with the FWO. G. Berx's laboratory is supported by the Fonds Wetenschappelijk Onderzoek (3G050217W), Strategic Basic Research (SBO; S008518N), the Geconcerteerde Onderzoeksacties Ghent University (GOA-01GB1013W), Vlaamse Liga tegen Kanker (365U8914U), and the Stichting tegen Kanker (FAF-F/2016/814).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.