EZH2 and KDM6B Expressions Are Associated with Specific Epigenetic Signatures during EMT in Non Small Cell Lung Carcinomas

Simple Summary Epithelial to Mesenchymal Transition (EMT) has been linked to multiple cancer features including invasion, metastasis, immune escape, and treatment resistance, and has therefore become a promising target mechanism for improving cancer treatment. Since it has also been described that EMT can be reversed by modulating its epigenetic regulation, our study was designed to decipher this epigenetic regulation of EMT, and more particularly its regulation by the H3K27me3 (trimethylation of lysine 27 of H3) mark and the two main enzymes which modulate it: EZH2 and KDM6B. Our results showed that these two enzymes present paradoxical roles during EMT in two cancer cell lines since both overexpression and inhibition of these two proteins led to the induction of EMT. We then identified three new target genes of EZH2 or KDM6B during EMT (INHBB, WNT5B and ADAMTS6) and demonstrated that the modulation of expression of these genes modulated the invasion capacities of the cells. These findings are therefore promising to prevent cancer cells from acquiring an invasive phenotype via the modulation of these new biomarkers of EMT regulated at the epigenetic level. Abstract The role of Epigenetics in Epithelial Mesenchymal Transition (EMT) has recently emerged. Two epigenetic enzymes with paradoxical roles have previously been associated to EMT, EZH2 (Enhancer of Zeste 2 Polycomb Repressive Complex 2 (PRC2) Subunit), a lysine methyltranserase able to add the H3K27me3 mark, and the histone demethylase KDM6B (Lysine Demethylase 6B), which can remove the H3K27me3 mark. Nevertheless, it still remains unclear how these enzymes, with apparent opposite activities, could both promote EMT. In this study, we evaluated the function of these two enzymes using an EMT-inducible model, the lung cancer A549 cell line. ChIP-seq coupled with transcriptomic analysis showed that EZH2 and KDM6B were able to target and modulate the expression of different genes during EMT. Based on this analysis, we described INHBB, WTN5B, and ADAMTS6 as new EMT markers regulated by epigenetic modifications and directly implicated in EMT induction.


Introduction
The crucial role of Epigenetics in Epithelial Mesenchymal Transition (EMT) has recently emerged. Indeed, type III EMT, which has been linked to metastasis and tumor immune escape [1], two hallmarks of cancer, is a reversible process, conferring migration and invasive abilities to cancer cells [2]. EMT is a conserved cellular mechanism which leads to the transition from an epithelial phenotype to a mesenchymal one [3]. The loss of E-Cadherin, together with an increase in Vimentin expression, is a feature of EMT. Different biological or chemical molecules such as TGF-β, hypoxia, cytokines or growth factors are able to induce EMT in specific models [4][5][6][7][8][9][10][11][12]. This process is accompanied by various modifications of the expression of a wide variety of additional epithelial/mesenchymal markers, depending on the EMT model and the inducer used. These modifications are controlled by specific transcriptional factors, so-called EMT-TFs, belonging to the SNAIL, ZEB (Zinc Finger E-Box Binding Homeobox) and TWIST families.
Epigenetic marks, whose roles have been largely demonstrated in cancers, are reversible and control gene expression. It has, for example, been demonstrated that the hypermethylation of the CDH1 (Cadherin 1) promoter is involved in the decrease in expression of the gene during EMT [13]. On the contrary, DNA demethylation has been shown to induce the upregulation of the microRNA, miR-200c, and the subsequent inhibition of EMT in clear cell renal cell carcinoma [14]. Similarly, the demethylation of the PD-L1 (Programmed Cell Death 1 Ligand 1) promoter in TGFβ-induced EMT cell models favored the expression of the immune checkpoint inhibitor PD-L1 and led to the inhibition of immune response against cancer cells during EMT [15]. Interestingly, two epigenetic enzymes with paradoxical roles have previously been associated to EMT: (i) the histone methyl transferase (HMT) EZH2 (Enhancer of Zeste 2 Polycomb Repressive Complex 2 (PRC2) Subunit also called ENX-1 or KMT6 for Lysine methyltranserase 6) which is able to add the H3K27me3 mark and, (ii) the histone demethylase (HDM) KDM6B (Lysine Demethylase 6B also called JMJD3) which is able to remove the H3K27me3 mark. In previous studies, it has been shown that the miR101-dependent inhibition of expression of EZH2 has been associated with EMT and poor prognosis in oral tongue squamous cell carcinoma [16]. EZH2 expression has also been correlated to invasiveness in laryngeal or breast carcinoma [17,18]. Similarly, KDM6B, although far less studied, has been positively associated with EMT or metastasis in renal cancer or hepatocarcinoma [19,20]. Although these paradoxical roles have been recently discussed [21], it remains unclear how these proteins, with apparent opposite activities, could both promote EMT and/or cancer aggressiveness. To address this question, and in order to better understand the respective roles of EZH2 and KDM6B during EMT, we decided, for the first time, to simultaneously study these proteins in in vivo and in vitro models.

Both EZH2 and KDM6B Are Overexpressed during EMT in Patients with NSCLCs
Since EZH2 and KDM6B have been both associated with EMT and aggressiveness in independent cancer models, we first analyzed their expressions in NSCLCs (Non Small Cell Lung Carcinomas). FFPE (Formalin-Fixed Paraffin-Embedded) biopsies were sorted in EMT-and EMT+ groups according to their Vimentin status (Figure 1), as already described in Reference [15]. An increase in both EZH2 and KDM6B stainings was observed in EMT+ tumors compared to EMT-tumors ( Figure 1A-C). Increases in the mRNA level of EZH2 and KDM6B are also obtained using an EMT-inducible A549 cell model (using both TGFβ/TNFα) ( Figure S1A). Moreover, these expressions were also strongly correlated ( Figure 1D), independently of the Vimentin status ( Figure S1B). These results were confirmed using data issued from the retrospective KM-plotter database with JetSet best probes [22]. We showed that lung cancer patients presenting high EZH2 mRNA levels or high KDM6B mRNA levels presented a significantly higher risk of early death (p = 3.4 ×10 −5 and p = 4.8 × 10 −5 ) compared to patients with lower EZH2 or KDM6B mRNA levels ( Figure 1E-G). This risk was further discriminated (p = 1.8 × 10 −5 ) when patients were compared in regard to the mean expression of EZH2 and KDM6B (high score: high mean EZH2/KDM6B vs. low score: low mean EZH2/KDM6B).

Effects of Overexpression of EZH2 and KDM6B on EMT Induction
Next, we determined, using an EMT-inducible A549 cell model (using both TGFβ/TNFα), if the overexpression of EZH2 (pMSCV-EZH2-IRES-GFP) or KDM6B (pMSCV-JMJD3 expressing an HA tag) was able to induce EMT ( Figure 2). Overexpression of both proteins in stable transfected cells (Figure 2A, Figure S2) did induce morphological changes characterized by a decrease in cell-cell adhesion and an increase in cell length ( Figure 2B). Increased cell length was confirmed using phalloidin-rodhamin staining ( Figure 2C). EZH2 and KDM6B overexpression also induced an increase in EMT markers at the mRNA level ( Figure 2D) characterized by VIMENTIN, ADAM19 or MMP2 upregulation and a CDH1 decrease, as already described using a TGFβ/TNFα treatment. These effects were accentuated in cells overexpressing EZH2. Western-blot experiments show a significant Vimentin increase in cells overexpressing EZH2 and a decrease in E-Cadherin expression in cells overexpressing both EZH2 or KDM6B ( Figure 2E). In order to clarify the role of these proteins on cancer phenotypes, we performed migration and invasion assays ( Figure 2F-G). As expected, EZH2 and KDM6B overexpression increased these EMT-related phenotypes. To validate the role of EZH2 and KDM6B on EMT induction, we also performed transient transfection of HA-EZH2 and HA-KDM6B ( Figure S3) and observed the same increase in EMT markers compared to the levels quantified in the untransfected control cells.

Inhibition of EZH2 and KDM6B Activities Regulated EMT Signaling
We next decided to clarify the function of these enzymes during EMT using specific inhibitors of EZH2 and KDM6B (EZH2i: GSK343 and KDM6Bi: GSKJ4) in EMT cell models. As expected, the inhibitor EZH2i decreased H3K27me3 levels in A549 cells, but according to the data found in the literature [23], we could not observe a significant increase in H3K27me3 signals in these cells when we used the inhibitor KDM6Bi ( Figure 3A and Figure S4A). Inhibition of KDM6B activity alone, using the specific inhibitor, led to a slight induction of the mesenchymal phenotype ( Figure 3B). Indeed, we observed an increase in cell length following an EZH2i or KDM6Bi/TGFβ/TNFα co-treatment compared to the TGFβ/TNFα treatment alone in the A549 cell model ( Figure 3C). We next targeted molecular markers to confirm these results. Using RT-qPCR, WB, IF, or flow cytometry, we showed that KDM6Bi induced Vimentin expression ( Figure 3D,E and Figure S4). Moreover, a decrease in E-Cadherin and an increase in N-Cadherin stainings were observed after a KDM6Bi treatment in the EMT-induced A549 cell model ( Figure 3F,G). Since GSK343 is more specific to EZH2 than to EZH1 (about 60-fold and 1000-fold for others methyltransferases), we next used the chemical inhibitor UNC1999 (EZH1/2i) which targets both enzymes, EZH1 and EZH2. This inhibitor EZH1/2i decreased H3K27me3 levels and promoted Vimentin and N-Cadherin stainings ( Figure S5). A phalloidin-rodhamin staining confirmed an increase in cell lengthening following the inhibition of both EZH1 and 2 activities. We finally analyzed whether EZH2i and KDM6Bi could also modulate EMT-related phenotypes, such as cell migration and invasiveness. As expected, KDM6Bi potentiated both cell migration and invasion in EMT-induced A549 cells by TGFβ/TNFα, whereas EZH2i did not ( Figure 3H,I).     EZH2 and KDM6B_CRISPR/Cas9 A549 models also confirmed that the invalidation of both enzymes favored EMT phenotypes ( Figure S6). However, in agreement with the results obtained with the chemical inhibitors, CRISPR/Cas9_KDM6B cells presented a more detectable mesenchymal phenotype than CRISPR/Cas9_EZH2 cells, with a higher expression of Vimentin and a significant increase in cell migration ability compared to control cells.
To confirm that our results obtained using the A549 model induced in EMT using TGFβ/TNFα were not dependent of the cell line, or the inducer, we also used a second model of EMT induction, the MDA-MB-468 cells in which EMT was induced using EGF. In this second model, overexpression of EZH2 or KDM6B was also associated with a higher Vimentin staining. Moreover, EZH2i, and KDM6Bi at a lower level, promoted EMT-related phenotypes in this second model, as well ( Figure S7).

The Genes Targeted by EZH2 and KDM6B during EMT Were Different
Since we demonstrated that both enzymes were able to promote EMT, even if they present an apparent opposite activity, we speculated that these surprising results might be linked to the targeting of different genes during EMT. So, to answer this question, we performed ChIP-seq (chromatin immunoprecipitation-sequencing) experiments targeting EZH2, KDM6B, and the H3K27me3 mark in A549 cells treated with TGFβ/TNFα. We then sorted 9493 loci presenting an enrichment of both EZH2 and the H3K27me3 mark and 13,122 loci associated with an enrichment of only KDM6B ( Figure 4A). As expected, the number of loci targeted by both EZH2 and KDM6B was very low (<3%), suggesting that the recruitment of these enzymes was indeed exclusive. Overlap and downstream positions of these loci were then crossed with previous transcriptome results obtained in our laboratory [24], and we then identified 169 down-regulated genes during EMT which might be controlled by H3K27me3 methylation in an EZH2 manner, and 566 up-regulated genes during EMT which could be regulated by KDM6B ( Figure 4A). A list of the 40 most down-regulated genes presenting an increased enrichment of both EZH2 and the H3K27me3 mark on the gene or the promoter was summarized in Table 1. Similarly, the 40 most up-regulated genes presenting an increased enrichment of KDM6B alone was summarized in Table 2. Amongst these genes, many have been previously associated with EMT features, such as adhesion, migration, invasion, cytoskeleton remodeling, or differentiation.
Interestingly, our ChIP-seq results revealed enrichment of KDM6B on the SNAI1, ADAM19, (ADAM Metallopeptidase Domain 19), and MMP9 genes, but not on the ZEB-1 gene. Moreover, an increase in SNAI1, ADAM19, and MMP9 expression was observed when A549 cells were treated with EZH2i or KDM6Bi combined with a TGFβ/TNFα treatment compared to a TGFβ/TNFα treatment alone ( Figure 4B). But, similarly to our previous results, the induction of expression was higher when we used KDM6Bi rather than EZH2i. As expected, the expression of the ZEB-1 gene was not modified by EZH2i or KDM6Bi, the two enzymes being unable to bind to this core promoter. We then confirmed, using single ChIP experiments, that a decrease in the H3K27me3 mark, which was potentiated by EZH2i, was observed on ADAM19 and MMP9 promoters in cells treated with TGFβ/TNFα ( Figure 4C).

INHBB, WNT5B and ADAMTS6 Established as New Epigenetic Biomarkers of EMT
In order to identify new target genes of EZH2 and KDM6B during EMT which could be later used as epigenetic biomarkers during EMT, we next confirmed our transcriptome and ChIP-seq results using single H3K27me3-targeted ChIP experiments and RT-qPCR analysis. Our data showed that, in agreement with the data presented in Table 1, INHBB (Inhibin Subunit Beta B) expression was significantly decreased during EMT and associated with a gain of the H3K27me3 mark on its promoter in EMT-induced A549 cells. On the contrary, in agreement with the data presented in Table 2, WNT5B (Wnt Family Member 5B) and ADAMTS6 (ADAM Metallopeptidase with Thrombospondin Type 1 Motif 6) expressions were increased during EMT and associated with a loss of the H3K27me3 mark on their respective promoters ( Figure 5A,B). To determine whether the modulation of the levels of H3K27 methylation was inversely correlated to the level of H3K27 acetylation (H3K27ac, a permissive mark [25]) in EZH2/KDM6B-targeted genes and to characterize the timing of these epigenetic modifications during EMT, we performed kinetic ChIP experiments ( Figure 5C). The data obtained validated our hypothesis that an increase in the H3K27ac content was correlated with a decrease in the H3K27me3 level, and inversely, on INHBB, ADAMTS6, or WNT5B promoters. Our data, therefore, validated the idea that the H3K27 mark was tightly regulated during EMT and might be a determinant of controlling gene expression. Moreover, these kinetic analyses indicated that the modulation of H3K27 methylation levels was progressive during EMT induction, with the highest effect obtained after 72 h of induction.
Next, we analyzed the expression of INHBB, ADAMTS6, and WNT5B at the protein level in NSCLCs ( Figure 5D,E). The expression of these markers was assessed in EMT-(n = 25) and EMT+ (n = 24) tumors using IHC. Our in vivo data were in agreement with our in cellulo results since we observed a significant increase in WNT5B and ADAMTS6 expressions in EMT+ vs. EMT-tumors while the expression of INHBB was significantly decreased in EMT+ compared to EMT-tumors.
We next wondered whether our targets were directly involved in EMT-related phenotypes following the invalidation of INHBB, WNT5B, and ADAMTS6 using specific siRNAs ( Figure 5F). Interestingly, the invalidation of INHBB expression, without the use of an EMT inducer, was sufficient to increase invasion ( Figure 5G). As expected, since a TGFβ/TNFα treatment completely abolished INHBB expression, no additive effect of the combination of siINHBB and TGFβ/TNFα could be observed. On the opposite, the combination of TGFβ/TNFα with a siWNT5B, or with a siADAMTS6, decreased invasion. These data strongly supported the idea that the invalidation of INHBB expression mediated by H3K27 methylation and the promotion of WNT5B and ADAMTS6 expressions mediated by H3K27 demethylation favored invasion during EMT. Altogether, our data demonstrated that INHBB, ADAMTS6, and WNT5B could be considered as three new EMT biomarkers, epigenetically regulated during EMT.

Effect of EZH2 and KDM6B Inhibition on INHBB, WNT5B or ADAMTS6 Expression
To analyze whether the inhibition of EZH2 and KDM6B was able to modulate INHBB, WNT5B, or ADAMTS6 expression, we performed RT-qPCR experiments in the presence of the different enzyme inhibitors ( Figure 6A). As expected, EZH2i decreased the levels of the H3K27me3 mark on the three promoters ( Figure 6B). Neither EZH2i ( Figure 6A) nor siEZH2 ( Figure S8) were sufficient to induce INHBB expression during EMT, even if H3K27me3 levels were decreased after EZH2i treatment. However, EZH2i was able to induce an increase of WNT5B expression in cells co-treated with TGFβ/TNFα ( Figure 6A). Similar results were observed when EZH2 expression was silenced ( Figure S8). Regarding ADAMTS6, unexpected results were observed with EZH2i since we observed a decreased expression, whereas, as expected, siEZH2 increased its expression in the presence of TGFβ/TNFα. These results indicated that the modulation of these enzymes partially modulated the expression of our targeted genes, probably due to additional post-translational histone marks, such as deacetylation, which could not be reversed by these treatments. Our results ( Figure 5C) might also suggest that H3K27me3 is a late mark modified to stabilize the effects on gene expression, rather than to induce the initial modulations.

Discussion
Both EZH2 and KDM6B, two proteins with opposite catalytic activities, have previously been associated with cancer aggressiveness or EMT. Indeed, a high expression of EZH2 was associated with metastasis and a poor clinical outcome in clear cell renal cell carcinoma, esophagus squamous cell carcinoma, bladder urothelial carcinoma, cutaneous melanoma, hormone-refractory and metastatic prostate cancer, gastric carcinoma, breast carcinoma, lung carcinoma and ovarian carcinoma, as well as in pediatric brain tumors (for a review, see Lachat et al. [21]). However, KDM6B expression has also been correlated with metastasis in clear cell renal cell carcinoma, ovarian cancer, myeloma, hepatocellular carcinoma, malignant pleural mesothelioma, and Hodgkin's lymphoma or invasive breast tumors. To clarify this apparent paradox or to understand whether these proteins were involved in metastasis in specific models, we decided to analyze the role of both EZH2 and KDM6B in the same EMT cellular model. Indeed, the majority of previous investigations were focused on only one epigenetic enzyme, and our study highlighted the complexity of simultaneously analyzing two epigenetic modulators with opposite functions. Therefore, for the first time, we observed that EZH2 and KDM6B expressions were strongly correlated in NSCLCs and both enzymes correlated with EMT in these tumors ( Figure 1) and with poor clinical outcomes. These data strongly supported the idea that both enzymes were involved in EMT and activated pro-EMT signaling pathways and that this induction only required one of these proteins as most people suggested in the literature. Our data also suggested that the expression of these two enzymes was co-regulated in lung cancers, but this last point would need further investigation to be confirmed.
To clarify the roles of EZH2 and KDM6B during EMT, we modulated the expression and/or activity of these enzymes separately. As expected, both overexpression of EZH2 or KDM6B promoted Vimentin expression, a well-known EMT marker ( Figure 2). Interestingly, the inhibition of KDM6B activity alone, using a specific inhibitor, was sufficient to promote EMT signaling (Figure 3). Nevertheless, the EZH2 or KDM6B inhibitor potentiated the effect of TGFβ/TNFα to induce EMT, and a more pronounced phenotype was observed when the activity of KDM6B was inhibited compared to the inhibition of EZH2. These data confirmed that the modulation (increase or decrease) of expression of each enzyme promoted EMT. Das and Taube [26] indeed highlighted the difficulty of addressing the precise contribution of these enzymes in a recent review.
The results of Figure 1 and Figure S1 suggest that both enzymes were strongly co-regulated and might be each responsible for the regulation of a specific panel of genes during EMT. Indeed, EZH2 would be involved in the repression of a specific subset of genes, whereas KDM6B would activate different targets. Our data also highlighted that the down-or up-regulation of one enzyme might be compensated by an increase, or inhibition, of the other. For example, it has been shown that PRC2, whose EZH2 is the catalytic member, could regulate different subsets of genes depending on the cell type and could function, at least partially, independently of the EMT-TFs. Indeed, EZH2 did not regulate SNAI1 or SNAI2 gene expression in head and neck squamous cell carcinoma cell lines, and the inhibition of its expression repressed the expression of mesenchymal markers and inhibited migration and invasion capacities without modulating EMT-TF levels [27]. Similarly, KDM6B has also been associated with the direct activation of the EMT-TF gene SNAI1, or SNAI2, in different cancer models (breast cancer cells [28], hepatocarcinoma cells [20], and CRCCs [19]) but KDM6B might also be recruited on specific promoters involved in downstream EMT signaling. Therefore, to identify the promoters which could be targeted by EZH2 or KDM6B, we performed a ChIP-seq analysis in A549 EMT-induced cells (Figure 4). We first confirmed that EZH2 and KDM6B occupied distinct sites in the genome and that most sites targeted by EZH2 also presented the H3K27me3 mark, while KDM6B was mostly associated with genomic sites in which the H3K27me3 mark was absent. Although CDH1 is a well-known repressed gene by EZH2 [29,30], our study suggested, for the first time, that numerous additional genes were also controlled by EZH2 and KDM6B during EMT. Interestingly, it has been previously reported in liver cancers that EZH2 repressed, in an H3K27me3-dependent manner, the APC promoter leading to the activation of the Wnt/β-Catenin pathway [31]. Interestingly, our study confirmed the presence of both EZH2 and the H3K27me3 mark on the APC promoter (data not shown) demonstrating the validity of our model and experimental scheme.
Amongst the genes specifically targeted by EZH2 and KDM6B during EMT, we confirmed a decreased expression of INHBB and an increased expression of WNT5B and ADAMTS6 ( Figure 5). INHBB was the gene presenting the most important decrease in expression (FC = 31) associated with concomitant recruitment of EZH2 and the presence of the H3K27me3 mark. Interestingly, this gene was previously identified as a regulator of hormone secretion from the pituitary gland (for a review, see Reference [32]) but, more recently, INHBB expression has been associated with cancer, although its role in EMT remains controversial. Indeed, high Activin B levels, a homodimer of INHBB, have been associated with a higher risk of metastasis in oral squamous cell carcinoma [33] but, INHBB expression was more recently associated with an inhibition of cell migration in nasopharyngeal carcinomas [34]. ADAMTS6 and WNT5B, whose expressions were increased (FC = 60 and 49, respectively) and associated with the presence of KDM6B on their promoters, have been associated with EMT in recent publications, although their roles are still poorly investigated in this process [35][36][37]. ADAMTS6 is one of the 19 members of the metallopeptidase ADAMTS family, but its physiological role is unknown. For example, its expression has been associated with invasion and cancer recurrence in pituitary tumors and poor prognosis (clinical stage, lymph node metastasis and recurrence) in esophageal squamous cell carcinoma [37,38]. Similarly, WNT5B has been very recently identified as a key regulator governing the aggressive basal-like phenotype in breast cancers [39]. Moreover, WNT5B appeared necessary to promote invasion and secondary EMT in a non-canonical WNT pathway manner [35]. We confirmed, by selectively inhibiting the expression of these genes, that the inhibition of INHBB promoted cell invasion, whereas the inhibition of WNT5B or ADAMTS6 decreased cell invasion in EMT-induced cells ( Figure 5). Moreover, an increased expression of ADAMTS6 and WNT5B and a decreased expression of IHNBB were confirmed in EMT-positive compared to EMT-negative NSCLCs ( Figure 5). Although EZH2 or KDM6B were found on the promoters of these genes, the regulation of their expression seemed only partially regulated by the H3K27me3 mark. Indeed, the inhibition of EZH2 or KDM6B partially controlled the expression levels of these new targets. These partial effects might be due to additional epigenetic modifications or the fact that H3K27me3 might only stabilize previous modulations of gene expression rather than to initiate these modulations. To further characterize the list of genes, directly and fully, controlled by EZH2 and KDM6B, new ChIP-seq experiments combined with transcriptomic analysis will be performed following the inhibition of EZH2 or KDM6B.

Cell Culture, Transfections, and Inhibitors Used in Cell Culture
The

Invasion and Migration Assays
Migration assays were performed using a 2-well culture-insert in 35 mm microdishes (81176, ibidi). A total of 18,000 pretreated cells were plated in the 2 wells, and the inserts were removed 24 h later. Migration was monitored for 8 h. For Invasion assays, 50 µL of ECM Gel from Engelbreth-Holm-Swarm murine sarcoma (diluted 8 fold in free serum medium) (E1270, Sigma-Aldrich) was added in the upper chamber (140629, Thermo Scientific, Waltham, MA, USA), positioned in 24-multiwell dishes, and incubated for 4 h at 37 • C for polymerization. Then, 50,000 pretreated cells were seeded into the upper chamber in presence of the desired treatment, diluted in free serum medium, and 500 µL of complete medium containing the same treatment was added into the lower chamber. The insert was removed 24 h later, rinsed with PBS, and invasive cells present in the well (cells which migrated through the matrigel and attached to the bottom of the well) were fixed with ethanol 100% for 5 min, stained with crystal violet for 10 min, and then counted.

Quantitative RT-PCR
RNA was isolated from cells using the Tri Reagent (TR118, Molecular Research Center, Cincinnati, OH, USA) as described by the manufacturer. Reverse transcription was performed using 12 U of M-MLV (M-1302, Sigma-Aldrich) reverse transcriptase, 0.25 µM oligodT (Eurogentec, Liège, Belgium), 1.25 µM random hexamers (C118A, Promega), and 1.5 µg of total RNA according to the manufacturer's instructions (Sigma-Aldrich). Quantitative PCR (qPCR) was performed in duplicate using the Step one plus Real-Time PCR system (Applied Biosystems, Foster City, CA, USA), the TB Green Premix Ex Taq (Tli RNase H Plus) kit (R420L, Takara, Ann Arbor, MI, USA), and specific primers (Eurogentec) according to the manufacturer's instructions. The primers were designed using the Primer 3 software and are listed in the Table S1.

ChIP/ChIP-seq
Washed cells were fixed with 1 % formaldhehyde at RT for 8 min with gentle agitation and then quenched with the addition of 125 mM Glycine for 5 additional minutes. Cells were centrifuged for 5 min at 500 g and washed twice with 1× cold PBS. Chromatin was prepared using the truChIP™ Chromatin Shearing Kit (520154, Covaris, Woburn, MA, USA) according to the manufacturer's instructions. Nuclei were submitted to 10 min of sonication using the M220 Covaris sonicator. ChIP was performed using the IP-Star Compact Automated System (B03000002, Diagenode, Denville, NJ, USA) with the Auto iDeal ChIP-seq kit for Histones (C01010171, Diagenode) and 1.5 µg of ChIP-grade antibody or IgG (IG07-2, P.A.R.I.S.; C15410206, Diagenode). Protein inhibitor cocktails were added in all buffers. Inputs and ChIP products were purified using the Auto Ipure kit V2 (C03010010, Diagenode) and IP-Star Compact Automated System (Diagenode). For ChIP-seq, libraries were prepared from 5 ng of ChIP DNA and Input DNA with the NEBNext®Ultra™ DNA Library Prep Kit for Illumina (E7370S, New England Biolabs, Ipswich, MA, USA) according to the manufacturer's instructions. From each library, 50 bp single-reads were sequenced using an Illumina Hiseq 1500 system (Illumina, San Diego, CA, USA). Reads were filtered according to their quality (Q Score ≥ 30), and adapter sequences were removed using Cutadapt [41]. Reads were aligned to the human genome (hg19) using the BWA (version 0.7.10). H3K27me3, EZH2, KDM6B differentially enriched regions (peaks) were identified using SICER-df (g = w = 200 bp). Primers used for classical ChIP were designed using the Primer 3 software (http://primer3.ut.ee/) and listed in the Table S1. Antibodies used for ChIP are listed in the  Table S2.

Immunohistochemistry
FFPE biopsies of NSCLCs were collected (in collaboration with the Tissue Biobank of the University of Liege (Liege, Belgium)) and stained with the listed antibodies (Table S2). The protocol was approved by the Ethics Committee of the University Hospital of Liege and the diagnosis of each case was confirmed by experienced histopathologists. IHC analysis was performed with a standard protocol detailed previously [42]. Samples were classified into two groups: EMT-positive vs. EMT-negative depending on their Vimentin staining. According to Reference [43], IHC were scored by multiplying two values: intensity (0−3) and extent (0−3), leading to a global score ranging between 0 and 9. All immunolabelled tissues were evaluated by two experienced histopathologists.

Statistics
Mean comparisons were analyzed using a Student t-test performed with the GraphPad Prism5 software (San Diego, CA, USA). Correlation indexes were measured using a Pearson test performed with the GraphPad Prism5 software. Significant values were highlighted in bold in each figure.

Conclusions
The epigenetic proteins EZH2 and KDM6B, two enzymes controlling the methylation level of the histone H3 at the position K27 (H3K27), are major components of cancer aggressiveness and EMT. Altogether, our results demonstrated, for the first time, that both EZH2 and KDM6B were involved in EMT. Moreover, we showed that the H3K27me3-mediated repression of INHBB and the activation of WNT5B or ADAMTS6 expression, linked to the loss of the H3K27me3 mark, could be considered as new biomarkers of EMT in vitro and in vivo, which could be later quantified to better stratify patients. Moreover, the targeting of EZH2 and/or KDM6B might offer new perspectives of therapies to fight metastasis apparition and formation in the future.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6694/12/12/3649/s1, Figure S1: Expression of EZH2 and KDM6B during EMT in A549 cellular model and correlation between EZH2 and KDM6B expression in NSCLCs, Figure S2: Validation of stable transfection of EZH2 and KDM6B in A549 cells, Figure S3: Effects of transient overexpression of EZH2 and KDM6B on EMT induction, Figure S4: EZH2i and KDM6Bi regulated the expression of Vimentin and the levels of H3K27me3 in A549 cells, Figure S5: EZH2/1i (UNC1999) promoted an EMT-related phenotype in A549 cells, Figure S6: Invalidation of EZH2 and KDM6B using CRISPR/Cas9 favored EMT-like phenotypes in A549 cells, Figure S7: Overexpression and chemical inhibition of EZH2 and KDM6B regulated EMT-like phanotype in MDA-MB-468 cells, Figure S8: Effects of siEZH2 or siKDM6Bi on target genes. Table S1: List of primers, Table S2: List of antibodies.