EZH2 endorses cell plasticity to non-small cell lung cancer cells facilitating mesenchymal to epithelial transition and tumour colonization

ARTICLE EZH2 endorses cell plasticity to non-small cell lung cancer cells facilitating mesenchymal to epithelial transition and tumour colonization Amador Gallardo 1,2,3,13 , Aldara Molina 1,2,3,13 , Helena G. Asenjo 1,2,3 , Lourdes Lopez-Onieva 1,3,4 , Jordi Martorell-Marugán 1,5,6 , Mencia Espinosa-Martinez 1,2,3 , Carmen Griñan-Lison 1,3,7 , Juan Carlos Alvarez-Perez 1,3,4 , Francisca E. Cara 1 , Saul A. Navarro-Marchal 3,8,9,10,11 , Pedro Carmona-Sáez 1,5 , Pedro P. Medina 1,3,4 , Juan Antonio Marchal 3,9,10,12 , Sergio Granados-Principal 1,2,3 , Antonio Sánchez-Pozo 1,2 and David Landeira 1,2,3 ✉ © The Author(s), under exclusive licence to Springer Nature Limited 2022 Reversible transition between the epithelial and mesenchymal states are key aspects of carcinoma cell dissemination and the metastatic disease, and thus, characterizing the molecular basis of the epithelial to mesenchymal transition (EMT) is crucial to find druggable targets and more effective therapeutic approaches in cancer. Emerging studies suggest that epigenetic regulators might endorse cancer cells with the cell plasticity required to conduct dynamic changes in cell state during EMT. However, epigenetic mechanisms involved remain mostly unknown. Polycomb Repressive Complexes (PRCs) proteins are well-established epigenetic regulators of development and stem cell differentiation, but their role in different cancer systems is inconsistent and sometimes paradoxical. In this study, we have analysed the role of the PRC2 protein EZH2 in lung carcinoma cells. We found that besides its described role in CDKN2A-dependent cell proliferation, EZH2 upholds the epithelial state of cancer cells by repressing the transcription of hundreds of mesenchymal genes. Chemical inhibition or genetic removal of EZH2 promotes the residence of cancer cells in the mesenchymal state during reversible epithelial–mesenchymal transition. In fitting, analysis of human patient samples and tumour xenograft models indicate that EZH2 is required to efficiently repress mesenchymal genes and facilitate tumour colonization in vivo. Overall, this study discloses a novel role of PRC2 as a master regulator of EMT in carcinoma cells. This finding has important implications for the design of therapies based on EZH2 inhibitors in human cancer patients. Oncogene (2022) 41:3611–3624; https://doi.org/10.1038/s41388-022-02375-x INTRODUCTION Epithelial to mesenchymal transition (EMT) is a cellular program crucial for embryogenesis, wound healing and malignant progression. EMT mediates dispersion of cells in embryos, formation of mesenchymal cells in injured tissues and initiation of the invasive and metastatic behaviour of epithelial cancers [1,2]. EMT is a dynamic and reversible process between the epithelial and the mesenchymal cell states, and therefore, different types of metastable quasi-mesenchymal cells can revert back to an epithelial state in a process known as mesenchymal–epithelial transition (MET) [3,4]. The current view is that EMT is governed by the activity of the EMT-inducing transcription factors (EMT-TFs), TWIST1/2, SNAI1/2 and ZEB1/2, that coordinate the repression of genes that maintain the epithelial state and the activation of the ones that induce the acquisition of mesenchymal features [3,5]. Expression of EMT-TFs is co-ordinately regulated by several intracellular signalling pathways of which the most prominent is the transforming growth factor-β(TGF-β) pathway that can on its own induce entrance into the EMT program [6]. The molecular mechanisms facilitating MET remain mostly unknown. Although EMT–MET is in essence a dynamic change between cellular states that must be regulated at the epigenetic level, chromatin modifiers involved and their mode of action remain to be identified [3,4]. In the context of cancer, activation of EMT imparts features that are essential to the formation of metastasis by carcinoma cells, including cell motility, ability to disseminate, tumour-initiating properties and elevated resistance to chemotherapy [7,8]. In addition, MET is associated with more efficient colonization of new tissues by carcinoma cells at metastatic sites [3,7]. Thus, characterizing the molecular basis of Received: 2 November 2021 Revised: 25 May 2022 Accepted: 31 May 2022 Published online: 9 June 2022 1 Centre for Genomics and Oncological Research (GENYO), Avenida de la Ilustración 114, 18016 Granada, Spain. 2 Department of Biochemistry and Molecular Biology II, Faculty of Pharmacy, University of Granada, Granada, Spain. 3 Instituto de Investigación Biosanitaria ibs.GRANADA, Granada, Spain. 4 Department of Biochemistry and Molecular Biology I, Faculty of Sciences, University of Granada, Granada, Spain. 5 Department of Statistics, Faculty of Sciences, University of Granada, Granada, Spain. 6 Data Science for Health Research Unit, Fondazione Bruno Kessler, Trento, Italy. 7 UGC de Oncología Médica, Complejo Hospitalario de Jaen, 23007 Jaen, Spain. 8 Department of Applied Physics, Faculty of Science, University of Granada, 18071 Granada, Spain. 9 Biopathology and Regenerative Medicine Institute (IBIMER), Centre for Biomedical Research (CIBM), University of Granada, 18016 Granada, Spain. 10 Excellence Research Unit Modelling Nature (MNat), University of Granada, 18016 Granada, Spain. 11 Cancer Research UK Edinburgh Centre, Institute of Genetics and Cancer, University of Edinburgh, EH4 2XU Edinburgh, UK. 12 Department of Human Anatomy and Embryology, School of Medicine, University of Granada, Granada, Spain. 13 These authors contributed equally: Amador Gallardo, Aldara Molina. ✉email: [email protected] www.nature.com/onc Oncogene 1234567890();,: EMT and MET is crucial to better understand cancer progression and design more precise therapeutic intervention approaches. Polycomb group (PcG) proteins are chromatin regulators of human development, tissue homeostasis and cancer [9,10]. PcG proteins associate to form multimeric complexes termed Polycomb Repressive Complex 1 and 2 (PRC1 and PRC2) of which the catalytic subunits are RING1A/B and EZH1/2 respectively. In stem cells, RING1A/B monoubiquitinates lysine 119 on histone H2A (H2AK119ub) while EZH1/2 tri-methylates lysine 27 on histone H3 (H3K27me3) co-ordinately to establish chromatin domains that maintain transcriptional repression of hundreds of lineage-specific genes, reenforcing maintenance of the current gene expression program and upholding cell identity [9,10]. Importantly, although mutations of genes that encode for PcG proteins are recurrent and associated with poor prognosis in a large number of human cancers, the role of PRCs during malignant progression remains unclear [10,11]. Initial studies showed that PcG proteins acted as oncogenes by repressing the tumour suppressor INK4A/ARF (CDKN2A) locus, and therefore inhibiting senescence and favouring cell proliferation [12,13]. However, subsequent studies revealed senescence-independent prooncogenic as well as tumour suppressor functions, highlighting unanticipated complexity in the role of PcG proteins in cancer [11,14]. This apparently conflicting observations are probably a consequence of the ability of PRC2 to regulate hundreds of functionally distinct target genes, together with the fact that PRC2 function during cancer progression can be dramatically affected by the genetic alterations accumulated by the cancer cell [15–18]. The initial oncogenic role of EZH2 led to the development of several small molecule inhibitors that are currently being tested in clinical trials [19]. Thus, given the context-dependent function of EZH2 in cancer, it is urgent a much better understanding of its molecular role in this disease. Emerging studies have led to hypothesize that the function of PcG proteins during cancer progression does not only rely on the regulation of cell proliferation through the CDKN2A locus, but also on the ability of PRC2 to maintain the repression of hundreds of genes in the tumour cell of origin, which in turn might regulate epigenetic plasticity and impact cancer dissemination [11,20]. In fitting with this idea, PRC2 can affect changes in cell state during EMT, but current studies reveal apparently contradictory results lacking a unifying molecular mechanism [18,21,22]. Here, we set out to determine the molecular mechanism by which PRC2 regulates EMT–MET in human non-small cell lung carcinoma cells. We established a TGF-β-inducible reversible in vitro system and we found that EZH2 facilitates the maintenance of the epithelial identity by binding and repressing the transcription of a large set of mesenchymal genes during EMT–MET. These include well-known regulators of the mesenchymal state such as SNAI2, MMP2 and ITGB3. In agreement, analysis of human patient samples and xenograft experiments support that EZH2 is required to efficiently repress mesenchymal genes and facilitate tumour colonization in vivo. Overall, our results support that EZH2 is required to repress mesenchymal genes during MET and facilitate tumour colonization in lung cancer. RESULTS EZH2 binds and represses mesenchymal genes in lung carcinoma cells To dissect the function of EZH2 in lung cancer we first focused on the human A549 cell line, which is a well-established system to study the molecular basis of non-small cell lung cancer. This cell line is homozygous null for CDKN2A and thus it is suitable to study the CDKN2A-independent role of EZH2 in cancer. In addition, A549 cells harbour a homozygous point mutation in the KRAS gene (G12S) encoding a hyperactive isoform of KRAS protein. First, we analysed the genome-wide binding profile of EZH2 by chromatin immunoprecipitation followed by high throughput sequencing (ChIP-seq). We found that, in agreement with the role of this protein in stem cell biology, there were 1969 EZH2 binding peaks that mapped to the promoter of 1237 genes (Fig. 1A, Fig. S1A), suggesting that PRC2 is repressing transcription of this set of genes. In pluripotent stem cells, promoters bound by EZH2 display bivalent chromatin and thus they are not only enriched for H3K27me3 but are also marked by H3K4me3 [23,24]. Likewise, analysis of publicly available H3K27me3 and H3K4me3 ChIP-seq datasets confirmed that most of EZH2 target promoters in A549 cells displayed a bivalent state (Fig. 1A, B, Fig. S1B), and that there is a positive correlation between the level of EZH2 and H3K27me3 at individual target promoters (Fig. 1C). As expected, transcriptome profiling of A549 cells by mRNA-high throughput sequencing (mRNA-seq) showed that EZH2 target genes were transcriptionally repressed or expressed at very low level (Fig. 1D). EZH2-repressed bivalent domains were readily detectable at individual target regions (i.e. GREM1 gene in Fig. 1E). Importantly, gene ontology analysis revealed that many EZH2 target genes are related to mesenchymal cells (Fig. 1F). Thus, similarly to the role of EZH2 in stem cells, we concluded that EZH2 binds to hundreds of bivalent genes that are transcriptionally repressed in A549 cells. Interestingly, EZH2-target genes were enriched in mesenchymerelated terms leading us to hypothesize that EZH2 is required to repress the mesenchymal gene expression program and maintain an epithelial cell state in A549 cells. To address whether EZH2 was upholding the epithelial state in A549 cells we analysed the effect of inhibiting EZH2 methyl transferase activity. We used a highly specific small molecule inhibitor of EZH2 (GSK126/EZH2i) [25]. Titration experiments confirmed that 5 µM EZH2i drastically reduces global levels of H3K27me3 upon four days of treatment, leaving only a residual H3K27me3 signal that is probably a consequence of EZH1 activity (Fig. S1C). Importantly, growing A549 cells in the presence of EZH2i for twelve days, consistently reduced global H3K27me3 levels without affecting cell growth (Fig. 1G, H). Strikingly, A549 cells plated at low density and grown in the presence of the EZH2i for 18 days displayed a very clear phenotype as compared to untreated cells: they formed unpacked colonies with more elongated and fusiform cell morphology typically associated with mesenchymal states (Fig. 1I, Fig. S1D). In fitting, EZH2i-treated cells displayed reduced levels of the epithelial marker E-CADHERIN and augmented expression of the mesenchymal markers such as N-CADHERIN and EMT-TFs including SNAI1, SNAI2 and ZEB1 (Fig. 1J, K, Fig. S1E). Among the analysed EMT-TFs, upregulation of SNAI2 protein was associated with increased abundance of its mRNA transcript, suggesting that EZH2 directly represses transcription of SNAI2 (Fig. S1E). In fitting, ChIP-qPCR analysis confirmed that SNAI2 is a direct target of EZH2 activity (Fig. S1F). Next, we asked whether in addition to the upregulation of mesenchymal molecular markers, treatment with EZH2i led to acquisition of mesenchymal cellular features. In agreement with our molecular analyses, cultured wound healing assays showed that EZH2i-treated cells healed faster that control cells (Fig. 1L). Complementarily, inhibition of EZH2 resulted in cells with reduced capacity to grown as colonies in soft agar (Fig. S1G) and as tumourspheres (Fig. S1H). Thus, we concluded that inhibition of EZH2 in A549 cells leads to the expression of EMT-TFs and acquisition of mesenchymal cellular features. To confirm that these findings were not a exclusive feature of the A549 cell line, we analysed whether inhibition of EZH2 induced activation of mesenchymal genes in three additional lung cancer cell lines that are genetic null for CDKN2A (cell lines A427, H1944 and H2122). In fitting with our results in A549 cells, treatment with 5 µM EZH2i during eight days had only minor effects on cell survival in all tested cell lines (Fig. S1I). Importantly, plating cells at low density in the presence of EZH2i promoted the upregulation of mesenchymal genes (Fig. 1M) and A. Gallardo et al. 3612 Oncogene (2022) 41:3611 – 3624 the acquisition of a typically mesenchymal cellular morphology in the four tested cell lines; unpacked colonies composed by fusiform cells (Fig. S1J). Treatment of H1944 cells with 5 µM EZH2i during fourteen days induced upregulation of N-CADHERIN and TWIST1 couple to increased wound-healing capacity (Fig. S1K, L). We concluded that EZH2 binds and represses mesenchymal genes, and that inhibition of EZH2 promotes spontaneous transition into a quasi-mesenchymal state in NSCLC cell lines. TGF-β-induced EMT is reversible in A549 cells To analyse the function of EZH2 during EMT in lung cancer we setup an in vitro reversible system to dissect the molecular mechanisms regulating EMT and MET. We titrated TGF-βand epidermal growth factor (EGF) and determine that treatment of A549 cells with 10 ng/ml TGF-βand 50 ng/ml EGF rapidly induced downregulation of E-CADHERIN mRNA and increased woundhealing capacity after 48 h of treatment in A549 cells (Fig. S2A, B). Treatment of cells with TGF-β+EGF during six days induced clear Fig. 1 EZH2 binds and represses mesenchymal genes in lung carcinoma cells. A Venn diagram comparing gene promoters bound by EZH2, H3K27me3 and H3K4me3 in ChIP-seq analyses in A549 cells. BChIP-seq average binding profile of EZH2, H3K27me3 and H3K4me3 around the TSS of 1237 gene promoters bound by EZH2 in A549 cells. CSpearman correlation analysis of the binding of EZH2 and H3K27me3 at EZH2- target genes. DBoxplot of mRNA transcribed from EZH2-target or control promoters measure by RNA-seq. EGenome browser view of H3K4me3, H3K27me3, EZH2 binding and mRNA level at the GREM1 locus. FGene Ontology analysis using Jensen tissues database of EZH2- bound bivalent targets. GSchematic diagram of the treatment of A549 cells with 5 µM EZH2i and western blot analysis of whole-cell extracts comparing the level of H3K27me3 during the course of the experiment. ACTIN B provides a loading control. HGrowth curve of A549 cells untreated or treated with EZH2i. IBrightfield images of A549 cells plated at low density and treated with EZH2i or untreated during 14 days. JAnalysis of E-CADHERIN and N-CADHERIN mRNA expression by RT-qPCR during EZH2i treatment in A549 cells. Expression is calculated relative to housekeeping genes GAPDH and ACTIN B.KWestern blot analysis of whole-cell extracts in EZH2i-treated A549 cells. ACTIN B was used as a loading control. LBrightfield images and plot comparing A549 cells untreated or pretreated with EZH2i for four days and plated in the presence of EZH2i during the course of a wound healing assay. MHeatmap representing the fold change in the expression of epithelial and mesenchymal markers (treated vs untreated with EZH2i) in the indicated NSCLC cell lines (A549, A427, H2122, H1944). Mean and SEM of three experiments are shown in H,Jand L. Asterisks indicate statistical significance using a Mann–Whitney test (*p< 0.05, ****p< 0.0001). A. Gallardo et al. 3613 Oncogene (2022) 41:3611 – 3624 morphological changes in cell culture: cells stopped forming tight colonies and acquired a fusiform shape (Day6, Mesenchymal (M)) (Fig. 2A). Removal of both cytokines from the culture media prompted a reversion of cell morphology to the epithelial state within the following six days (Day 12, epithelial reverted (ER)) (Fig. 2A). In agreement, time course analysis of the expression of epithelial (E-CADHERIN) and mesenchymal (N-CADHERIN,NRP2) markers showed expected opposite patterns of expression (Fig. 2B, C). To address whether the presence or absence of TGF-β+EGF in the culture media led to extensive reorganization of epithelial–mesenchymal gene expression programs, we compared the transcriptome of cells at days 0, 6 and 12 by mRNA sequencing (mRNA-seq). Cells at day 0 and 6 differentially expressed 1485 genes of which most of them (936 genes) were upregulated upon EMT on day 6 (Fig. S2C, D). Gene ontology analysis confirmed that these genes were enriched in mesenchymal-related terms including mesenchymal tissues (stromal cells, mesenchymal stem cells), cellular processes (focal adhesion, cell migration) and signalling pathways (Ras and PI3K- Akt signalling) (Fig. S2E). As expected, comparison of cells in days Fig. 2 TGF-β-induced EMT is reversible in A549 cells. A Scheme of the experimental design used to induce EMT and MET in A549 cells during 12 days. Brightfield images of the representative morphology at the different stages of the transitions are shown. Scale bar 100 µM. BExpression of E-CADHERIN by western blot during the course of the experiment. ACTIN B serves as loading control. CAnalysis of mRNA expression by RT-qPCR of indicated epithelial or mesenchymal genes during the course of the experiment. Relative expression level against GAPDH and ACTIN B is shown. DHeatmap analysis of mRNA expression of 829 reversibly regulated genes (FC > 2, p< 0.05) at day 0 (E), day 6 (M) and day 12 (ER) by RNA-seq of two independent replicates (R1 and R2). Yellow and blue indicate higher and lower expression respectively. Gene ontology analyses of genes in cluster II are shown. ETable showing reversible genes included in the GO category “Mesenchyme” identified in D.FHeatmaps representing the fold change expression measured by mRNA-seq of rationally selected epithelial and mesenchymal markers during EMT (upper strip) or MET (lower strip) GAnalysis of mRNA expression by RT-qPCR of indicated EMT-TFs during EMT–MET. Expression level relative to GAPDH and ACTIN B is shown. HAnalysis of the expression of SNAI1, SNAI2 and ZEB1 proteins during EMT–MET by Western blot. ACTIN B was used as a loading control. IBrightfield images and quantification plot comparing the wound healing capacity of cells at day 0, 6 and 12 of EMT–MET. Mean and SEM of three experiments are shown in B,Gand I. Asterisks indicate statistical significance using a Mann–Whitney test (*p< 0.05). A. Gallardo et al. 3614 Oncogene (2022) 41:3611 – 3624 6 and 12 showed the opposite trend: most differentially expressed genes were downregulated (586 out of 990 genes) and these were again associated with a mesenchymal phenotype (Fig. S2F, G, H). Importantly, a large set of the differentially expressed genes were reversibly regulated: they were either downregulated or upregulated on day 6, but restored to their initial expression level on day 12 (829 reversible genes, Fig. 2D, Table S1). Reversible genes upregulated on day 6 were again enriched in ontology terms associated with mesenchyme (Cluster II, Fig. 2D) and included key genes with important roles in a wide range of cellular processes (i.e. cell surface, ligands, extracellular matrix organization and transcription factors) (Fig. 2E). Analysis of key rationally selected epithelial (i.e. E-CADHERIN, CLDN2, OCLN) and mesenchymal markers (N-CADHERIN, VIMENTIN, MMP2/9/15/25, EMT-TFs) also displayed a reversible gene expression pattern during the course of the experiment (Fig. 2F, G). In fitting, analysis of protein expression by Western blot confirmed that the level of EMT-TFs (SNAI1/2 and ZEB1) peaked on day 6 and was reduced on day 12 (Fig. 2H). Importantly, comparison of wound-healing capacity at different time points showed that M cells on day 6 heal faster than E and ER cells on day 0 and 12 (Fig. 2I). Taken together, these analyses indicate that addition of TGF-βand EGF to the culture media induces the activation of a mesenchymal gene expression program and mesenchymal cellular features, and that removal of these cytokines facilitates the reversion of the cells to an epithelial state. We concluded that transient addition of TGF-βand EGF to A549 cells provides a well-suited system to analyse the molecular basis of EMT–MET in lung cancer. EZH2 directly regulates a large set of mesenchymal genes during EMT–MET in lung cancer cells To dissect the role of EZH2 during epithelial–mesenchymal transition we first focused our analysis in the expression level of PRC2 components during EMT. EZH2, other PRC2 subunits and H3K27me3 were detected at similar levels at days 0 (E) and day 6 (M) (Fig. 3A–C, Fig. S3A, B), suggesting that PRC2 is functional in both epithelial and mesenchymal cell states. Quantitative ChIP-seq (cChIP-seq) analysis of EZH2 binding at day 0 (E) and day 6 (M) revealed that EZH2 can bind to 1237 gene promoters in epithelial and/or mesenchymal states (Table S1). Strikingly, despite widespread changes in gene expression were detected upon EMT (Fig. 2D), EZH2 displayed only minor changes in the pattern of binding of EZH2 around the transcription start site (TSS) of target genes in both cell states (Fig. 3D, E, Fig. S3C). To confirm this observation, we focused our analysis on genes that were bound by EZH2 and were transcriptionally up-regulated (FC > 2, p< 0.05) during EMT Fig. 3 Binding of EZH2 to target promoters remains constant during EMT in A549 cells. A Expression of EZH2 mRNA by RT-PCR at different times points during EMT. BWestern blot analysis at day 0 (E) and day 6 (M) during EMT. ACTIN B was used as loading control. CImmunofluorescence analyses of EZH2 and H3K27me3 at day 0 (E) and day 6 (M) during EMT. Scale bar is 10 µm. DHeatmap of the EZH2 cChIP-seq signal around the TSS of 1267 genes bound by EZH2 in cells at day 0 (E) and/or day 6 (M). ECorrelation analysis of EZH2 signal at target promoters (−0.5 +1.5Kb around the TSS) in cells at day 0 (E) and/or day 6 (M). FVenn diagram comparing genes marked by H3K27me3 at day 0 (E) and upregulated (FC > 2, p< 0.05) on day 6 (M). GAverage binding profile of indicated proteins at 386 gene promoters identified in (F). HAnalysis of mRNA expression from 386 gene promoters identified in (F) at day 0 (E) and day 6 (M). Asterisks indicate statistical significance using Mann–Whitney test (**p< 0.01). IAverage binding profile of EZH2 at day 0 (E) and day 6 (M) at 386 gene promoters identified in (F). JGenome browser view of EZH2 binding and mRNA expression at the BMP2 locus at day 0 (E) and day 6 (M). A. Gallardo et al. 3615 Oncogene (2022) 41:3611 – 3624 (386 genes, Fig. 3F, Table S1). Albeit these genes showed clear binding of EZH2, H3K27me3 and H3K4me3 around the TSS at day 0 (Fig. 3G, Fig. S3D), and were transcriptionally induced on day 6 (Fig. 3H), binding of EZH2 to the promoter region remained mostly unaltered during the six days of EMT induction (Fig. 3I). This suggests that although EZH2 is repressing mesenchymal target genes in the epithelial state, their transcriptional upregulation during EMT does not require eviction of EZH2 from target promoters (i.e. BMP2, Fig. 3J). A change in the binding of EZH2 was detected for a few genomic regions (i.e. RNF182) (Fig. S3E), demonstrating that stable binding of EZH2 to target promoters is not a technical caveat of our cChIP-seq approach. In fitting, stable binding of EZH2 to target gene promoters (FGF3,LGR5,NPTX1 and CCND2) was also detected by ChIP-PCR (Fig. S3F). Previous reports show that PRC2 represses the E-CADHERIN gene during EMT in prostate and colon cancer cell lines [21,22]. However, we only found a marginal increased binding of EZH2 to the promoter region of E-cadherin on day 6 in A549 cells (Fig. S3F, G). Thus, we concluded that EZH2 binds to 386 mesenchymal genes that are transcriptionally induced upon TGF-β-stimulation, and that their activation does not require eviction of EZH2 from target promoters. To address how the loss of function of EZH2 affects the regulation of its target genes during EMT–MET, we compared the transcriptome of EZH2i-treated and control cells during EMT–MET (Fig. 4A). A549 cells were pretreated with 5 µm EZH2i for four days to reduce H3K27me3 to background levels and EZH2i was refreshed every two days during the course of the experiment, leading to sustained reduction of H3K27me3 levels (Fig. S4A). TGF- β-treated cells acquired mesenchymal morphology on day 6, irrespectively of being treated or not with EZH2i (Fig. S4B). In contrast, while untreated cells reverted back to an epithelial morphology on day 12, EZH2i-treated cells retained a clear mesenchymal morphology (Fig. S4B). In agreement, examination of the expression of key epithelial and mesenchymal genes confirmed that cells treated with EZH2i express higher level of key mesenchymal genes at days 6 and 12 (Fig. 4B). Therefore, we settled that inhibition of EZH2 enhances EMT and hinders MET in A549 cells. To identify which are the genes directly regulated by EZH2 during EMT–MET we determined which promoters were bound by EZH2 and differentially regulated in cells treated with EZH2i. Out of the 829 reversibly regulated genes (identified in Fig. 2D), we found that 335 genes (Clusters I +II +II +IV) were overexpressed on day 6 (Fig. 4C, Table S1) or day 12 (Fig. 4D, Table S1) in cells treated with EZH2i. Of these, 140 genes were marked by EZH2, H3K27me3 and H3K4me3 at their promoter region at day 0 (Fig. 4E, F). These were very significantly enriched for EMT genes (Fig. S4C) and included established regulators (i.e. SNAI2,MMP2,MMP9, ITGB3 and GREM1) of different EMT processes (Table S2). In fitting with our previous analyses in Fig. 3, binding of EZH2 to these genes was similar in epithelial and mesenchymal states, although their RNA expression was increased as cells acquired mesenchymal features on day 6 (Fig. 4G, H). Direct EZH2 targets (140 genes H3K27me3 positive and EZH2i-responsive) could show different levels of H3K27me3 and H3K27ac at their associated CpG islands (Fig. S4D–F). Importantly, inhibition of EZH2 did not hinder repression of epithelial genes at day 6 (Fig. 4C), including E-CADHERIN (Fig. S4G), indicating that EZH2 is not critical to downregulate expression of epithelial genes during EMT. RT-qPCR analysis confirmed that inhibition of EZH2 hindered repression of EZH2-target mesenchymal genes (SNAI2 and MMP2) at day 12, and led to expected secondary changes in the expression of E-CADHERIN,N-CADHERIN and SNAI1 (Fig. 4I, Fig. S4H). Likewise, inhibition of EZH2 during EMT–MET in H1944 cells obstructed repression of mesenchymal genes such as TWIST1 and N-CADHERIN at day 12 (Fig. S4I). To address whether inhibition of EZH2 led to the acquisition of mesenchymal cellular features during EMT–MET, we measured the behaviour of EZH2i-treated cells in wound healing and soft agar assays. Inhibition of EZH2 led to increased healing ability and reduced capacity of anchorageindependent growth at days 6 and 12 (Fig. 4J, K). Taken together, these experiments demonstrate that EZH2 represses a large set of mesenchymal genes in lung cancer cells, and that blocking EZH2 methyltransferase activity favours the acquisition of molecular and cellular mesenchymal features during EMT-MET. EZH2 modulates the transcriptional upregulation of mesenchymal genes in response to TGF-βstimulation To confirm that the phenotype observed in EZH2i-treated cells was a consequence of inhibiting EZH2 and not due to unspecific inhibition of another methyl transferase enzyme, we analysed the effect of another highly specific EZH2 inhibitor (EPZ6438) [26]. In agreement with previous results using the EZH2i GSK126, treatment of A549 cells with 2 µM EPZ6428 dramatically reduced the level of H3K27me3 on chromatin (Fig. S5A) and promoted higher mRNA expression of EZH2 target genes (SNAI2 and MMP2) during EMT–MET (Fig. S5B). Concomitantly, cells treated with EPZ6428 displayed reduced expression of E-CADHERIN but increased number of N-CADHERIN transcripts (Fig. S5B). Importantly, treatment with EPZ6428 also led to increased healing capacity in wound healing assays at both day 6 and 12 during EMT–MET assays (Fig. S5C, D). To obtain solid proof that EZH2 represses mesenchymal genes during EMT–MET in lung cancer cells, we used a genetic approach and derive A549 cells knockout for EZH2 using CRISPR/Cas9 (Fig. S5E). Upon transfection and flow cytometry sorting of the cells that had incorporated the GFP-expressing CRISPR/Cas9 plasmid, we derived both bulk and genetically clonal stable populations of A549 cells that expressed very low (bulk) or undetectable (clone) levels of EZH2 protein and H3K27me3 (Fig. 5A). In fitting with the phenotype observed in experiments using EZH2 inhibitors, EZH2-null cells did not longer form tight epithelial colonies in culture, but organized as non-packed fusiform cells (Fig. 5B). This was particularly evident in the clonal population of cells, suggesting that the bulk population harbours a low proportion of non-edited cells. In consonance with EZH2i-treated cells, EZH2-null cells displayed a proliferation capacity comparable to A549 parental cells (Fig. 5C, Fig. S5F). The loss of EZH2 led to overexpression of EZH2-target genes SNAI2 and MMP2 coupled with changes in the expression of E-CADHERIN and N-CADHERIN (Fig. 5D). In addition, loss of EZH2 repression induced acquisition of mesenchymal cellular features such as enhanced cell motility in wound healing assays, reduced anchorage-independent growth and decreased ability to form tumour spheres compared to parental cells (Fig. 5E, Fig. S5G, H). Taken together, these experiments confirm that EZH2 activity is dispensable for cell growth but required to repress mesenchymal genes and uphold an epithelial identity in A549 lung cancer cells. We wondered whether, given that EZH2-null cells already showed obvious mesenchymal molecular and cellular features, they could still respond to TGF-βstimulation and become more mesenchymal. We compared the expression of EZH2-target mesenchymal genes (SNAI2,MMP2 and ITGB3) as well as E-CADHERIN and N-CADHERIN during in vitro EMT–MET in EZH2- null and parental cells. Non-stimulated EZH2-null cells already displayed a level of expression of mesenchymal genes similar to the one observed in TGF-β-treated parental cells on day 6 (see SNAI2 and MMP2 in Fig. 5F). Interestingly, addition of TGF-β further enhanced transcriptional activity of EZH2-target genes SNAI2 and MMP2 (Fig. 5F), and this was accompanied by the acquisition of an extreme fusiform morphology of EZH2-null cells at day 6 (Fig. 5G). Importantly, removal of TGF-βfrom the culture media led to the restoration of the initial transcriptional level in both parental and EZH2-null cells for all genes tested (Fig. 5F). Thus, we concluded that EZH2 binds and represses mesenchymal A. Gallardo et al. 3616 Oncogene (2022) 41:3611 – 3624 genes at different stages of the epithelial–mesenchymal spectrum where it seems to coordinate and buffer the transcriptional response of the cells to TGF-βstimulation (Fig. 5H). To determine whether the phenotype of EZH2 loss of function was orchestrated by SNAI2 protein we used CRISPR/Cas9 (Fig. S5I) to derive a SNAI2-null bulk population of A549 cells. Inhibition of EZH2 activity induced the expression of SNAI2 protein in parental cells, but was barely detected in SNAI2-null cells (Fig. S5J). SNAI2- null cells treated with EZH2i acquired a mesenchymal state similar to parental cells, as judged by their fusiform cell morphology as well as expression of E-CADHERIN and N-CADHERIN (Fig. S5K, L). Importantly, in fitting with the role of SNAI2 as an inducer of cell motility [27], loss of function of SNAI2 blocked the acquisition of wound healing upon EZH2i-treament (Fig. S5M). Thus, we interpreted that the mesenchymalization observed in cells with EZH2 loss of function depends on the combinatorial activity of several EZH2-target genes of which SNAI2 is responsible for changes in cell motility. EZH2 protein level inversely correlates with expression of mesenchymal genes in human lung tumors To obtain insights into whether EZH2 represses mesenchymal genes in human lung tumours, we analysed genome-wide datasets from the cancer genome atlas repository of protein and mRNA I Fig. 4 EZH2 represses a large set of mesenchymal genes during EMT–MET in A549 cells. A Schematic diagram of the experimental conditions used to study the effect of inhibiting EZH2 activity (EZH2i 5 µM) during reversible EMT–MET (TGF-β+EGF). BHeatmaps showing the fold change expression of epithelial and mesenchymal marker genes measured by cChIP-seq in cells treated with EZH2i vs untreated at day 6 and day 12. C,DExpression analysis of genes that are reversibly regulated (FC > 2, p< 0.05) during EMT–MET (n=829). Heatmaps show the expression of two biological duplicates (R1 and R2) at day 0 (E), day 6 (M) and day 12 (ER). Genes that are upregulated in EZH2i-treated cells are identified in cluster I–IV. EVenn diagram comparing H3K27me3 ChIP-seq signal at day 0 and genes in clusters I–IV. Significant association between clusters I–IV and H3K27me3 (Chi-square test, odds ratio =2.8, p< 0.00001). FAverage binding plots of EZH2, H3K27me3 and H3K4me3 to the 140 overlapping genes identified in E.GAverage binding plots of EZH2 at day 0 and day 6 at the 140 overlapping genes identified in E.HBox plot showing mRNA expression at day 0 and day 6 of the 140 overlapping genes identified in E.IRT-qPCR analysis measuring mRNA level of indicated genes during EMT–MET in the absence or presence of EZH2i. Expression level is calculated relative to GAPDH and ACTIN B.JGraph comparing the kinetics of wound closure of cells at day 6 or day 12 of the EMT–MET treatment in the presence or absence of EZH2i. KPlot showing the number of colonies formed in soft agar by cells at day 6 or day 12 of the EMT–MET treatment in the presence or absence of EZH2i. Mean and SEM of three experiments are shown in I,Jand K. Asterisks indicate statistical significance using a Mann–Whitney test (*p< 0.05). A. Gallardo et al. 3617 Oncogene (2022) 41:3611 – 3624 expression of resected tumours from 427 NSCLC patients. Expression of EZH2 protein showed a robust inverse correlation with the level of mRNA of mesenchymal genes identified in our in vitro system (Fig. 6A). This correlation was solid for the set of H3K27me3- positive genes that were transcriptionally induced during EMT (386 genes identified in Fig. 3F), and for the genes that were H3K27me3 positive and overexpressed in EZH2i-treated cells during EMT–MET (140 genes identified in Fig. 4E). Inverse correlation was also evident for individual EZH2-targets including SNAI2 and MMP2 (Fig. 6B). To further comprehend the role of EZH2 in lung cancer, we analysed whether the expression level of EZH2 was a marker of survival probability using available datasets of a cohort of 1925 NSCLC patients. In agreement with previous reports [18,19,28], expression of EZH2 mRNA was higher in lung tumours as compared to control healthy lung tissue (Fig. S6A), supporting an oncogenic role of EZH2 in lung cancer. In fitting, high expression of EZH2 associated with poor survival (median survival EZH2, high: 45.27 months, low: 79.54 months) (Fig. 6C). In contrast, other PRC2 subunits did not behave as markers of negative prognosis (Fig. S6B), indicating that the effect of different PRC2 subunits in cancer progression is complex and must be individually analysed. We concluded that expression of EZH2 correlates with reduced expression of EZH2- target mesenchymal genes in lung cancer tumours and it is associated with poor survival. Fig. 5 EZH2-null A549 cells acquire mesenchymal features but remain responsive to TGF-βstimulation. A Western blot analysis of whole cell extracts comparing the levels of EZH2 and H3K27me3 in parental and EZH2-null bulk (EZH2KO) and clonal (EZH2KO#1) populations. TUBULIN serves as loading control. BBrightfield images of parental, EZH2KO and EZH2KO#1 cells. Scale bar is 100 µm. CGrowth curve of parental, EZH2KO and EZHKO#1 cells. DAnalysis of mRNA expression by RT-qPCR in parental, EZH2KO and EZH2KO#1 cells. Expression level relative to GAPDH and ACTIN B is shown. EKinetics of wound closure of parental, EZH2KO and EZH2KO#1 cells. FAnalysis of mRNA expression by RT-qPCR in parental and EZH2KO#1 cells during EMT–MET. Expression level relative to GAPDH and ACTIN B is shown. GBrightfield images of parental and EZH2KO#1 cells during EMT–MET. Scale bar is 100 µm. HSchematic diagram proposing the mechanism by which EZH2 regulate mesenchymal genes during EMT–MET. EZH2, RNA Polymerase II (RNAPII) and a hypothetical transcriptional activator regulating mesenchymal genes in the presence or absence of TGF-βstimulation are depicted. Low or robust gene transcription is represented by discontinuous and solid arrows respectively. Mean and SEM of three experiments are shown in C,D,Eand F. Asterisks indicate statistical significance using a Mann–Whitney test (*p< 0.05). A. Gallardo et al. 3618 Oncogene (2022) 41:3611 – 3624 EZH2-null A549 cells display reduced tumor colonization capacity in xenograft assays Our analyses in cell lines and human patients support that EZH2 is required to transcriptionally repress the mesenchymal gene expression program during MET in lung cancer. Thus, we conjectured that EZH2 is required for tumour colonization and disease progression in vivo. To test this, we compared the tumour colonization capacity of A549 cells that were either wild-type or EZH2-null. Using lentiviruses, we transduced a luciferase reporter into three independent EZH2 knockout clones (EZH2KO#1, EZH2KO#2 and EZH2KO#3) and parental cells (parental pool, parental clone #2 and parental clone #3), injected these luciferaseexpressing cells intravenously in the tail vein of immunocompromised mice, and analysed tumour colonization kinetics by measuring luciferase activity in vivo for 30 days. Mice injected with A549 cells displayed robust luciferase signal in their thoracic cavity that increased during the course of the experiment (Fig. 6D, E, Fig. S6C, D). In contrast, mice injected with EZH2-null cells showed very low luciferase activity that was sometimes indistinguishable from the background (Fig. 6D, E, Fig. S6C, D). Examination of organs upon necropsy on day 30 showed that luciferase expression was evident in the lungs of mice injected with parental cells, while very low luciferase activity was detected in lungs from mice that had been injected with EZH2-null cells (Fig. 6F, luciferase). No luciferase activity was found in brain, liver nor kidneys in any of the two conditions (not shown). Although nodules on the surface of lungs were not evident on day 30, we could readily detect them 10 days later in mice injected with parental cells, but not in EZH2KO-injected mice (Fig. 6F, visible). Haematoxylin-eosin and anti-Ki-67 staining of lung tissue samples Fig. 6 EZH2-null A549 cells display reduced tumor colonization capacity in xenograft assays. A Spearman’s correlation heatmaps of EZH2 protein and indicated sets of mRNAs in 426 human samples of NSCLC available at TCGA database. EMT UP and EZH2i responsive genes that are bound by H3K27me3 were identified in Fig. 3F and Fig. 4E respectively. The pattern of a set of randomly selected genes is shown for comparison purposes. BSpearman´s correlation of EZH2 protein and MMP2 (top panel) and SNAI2 (bottom panel) mRNA. Each dot represents one patient. CKaplan–Meier plot showing survival probability of 1925 NSCLC patients depending on the expression of EZH2 mRNA. DKinetics of tumor growth inferred by luciferase activity upon tail intravenous injection of luciferase-expressing control or EZH2KO#1 A549 cells in immunocompromised mice. Mean flux signal in eight mice was normalized to day 0 for every time point. Statistical significance was measure using an ANOVA test. ERepresentative bioluminescence images of animals at day 30 after injection. FBioluminescence signal of surgically extracted mouse lungs 30 days post-injection (left panel). Scale bar is 10 mm. Image of lungs on day 40 post-injection. Scale bar is 5 mm. Insets highlight the presence of tumor-like nodules at the surface of the lungs. Scale bar is 2.5 mm. GHematoxylin & eosin staining of fixed ex vivo lungs obtained from mice on day 30 after injection. Scale bar is 250 µm. Insets identify regions positive for H&E and Ki67 staining. Arrows highlight the position of Ki67-positive regions. Scale bar is 200 µm. HHistogram showing the bioluminescence signal of animals at day 4 after injection. Representative pictures are shown. Asterisks indicate statistical significance using a Mann–Whitney test (****p< 0.0001). A. Gallardo et al. 3619 Oncogene (2022) 41:3611 – 3624