KDM2B is involved in the epigenetic regulation of TGF-β-induced epithelial–mesenchymal transition in lung and pancreatic cancer cell lines

Polycomb repressive complex-1 (PRC1) induces transcriptional repression by regulating monoubiquitination of lysine 119 of histone H2A (H2AK119) and as such is involved in a number of biological and pathological processes including cancer development. Previously we demonstrated that PRC2, which catalyzes the methylation of histone H3K27, has an essential function in TGF-β-induced epithelial–mesenchymal transition (EMT) of lung and pancreatic cancer cell lines. Since the cooperative activities of PRC1 and PRC2 are thought to be important for transcriptional repression in EMT program, we investigated the role of KDM2B, a member of PRC1 complex, on TGF-β-induced EMT in this study. Knockdown of KDM2B inhibited TGF-β-induced morphological conversion of the cells and enhanced cell migration and invasion potentials as well as the expression changes of EMT-related marker genes. Overexpression of KDM2B influenced the expression of several epithelial marker genes such as CDH1, miR200a, and CGN and enhanced the effects of TGF-β. Mechanistic investigations revealed that KDM2B specifically recognized the regulatory regions of CDH1, miR200a, and CGN genes and induced histone H2AK119 monoubiquitination as a component of PRC1 complex, thereby mediating the subsequent EZH2 recruitment and histone H3K27 methylation process required for gene repression. Studies using KDM2B mutants confirmed that its DNA recognition property but not its histone H3 demethylase activity was indispensable for its function during EMT. This study demonstrated the significance of the regulation of histone H2A ubiquitination in EMT process and provided the possibility to develop novel therapeutic strategies for the treatment of cancer metastasis.

Polycomb repressive complex-1 (PRC1) induces transcriptional repression by regulating monoubiquitination of lysine 119 of histone H2A (H2AK119) and as such is involved in a number of biological and pathological processes including cancer development. Previously we demonstrated that PRC2, which catalyzes the methylation of histone H3K27, has an essential function in TGF-β-induced epithelial-mesenchymal transition (EMT) of lung and pancreatic cancer cell lines. Since the cooperative activities of PRC1 and PRC2 are thought to be important for transcriptional repression in EMT program, we investigated the role of KDM2B, a member of PRC1 complex, on TGF-β-induced EMT in this study. Knockdown of KDM2B inhibited TGF-β-induced morphological conversion of the cells and enhanced cell migration and invasion potentials as well as the expression changes of EMT-related marker genes. Overexpression of KDM2B influenced the expression of several epithelial marker genes such as CDH1, miR200a, and CGN and enhanced the effects of TGF-β. Mechanistic investigations revealed that KDM2B specifically recognized the regulatory regions of CDH1, miR200a, and CGN genes and induced histone H2AK119 monoubiquitination as a component of PRC1 complex, thereby mediating the subsequent EZH2 recruitment and histone H3K27 methylation process required for gene repression. Studies using KDM2B mutants confirmed that its DNA recognition property but not its histone H3 demethylase activity was indispensable for its function during EMT. This study demonstrated the significance of the regulation of histone H2A ubiquitination in EMT process and provided the possibility to develop novel therapeutic strategies for the treatment of cancer metastasis.
Metastasis is the leading cause for mortality of cancer patients. A developmental process termed epithelialmesenchymal transition (EMT) has been shown to play an essential role in cancer metastasis (1). During EMT, cells lose their epithelial characters such as cell polarity and cell contacts and gain migratory and invasive properties of mesenchymal cells. Physiological activation of EMT can be triggered by extrinsic signals including transforming growth factor-beta (TGF-β) secreted from tumor microenvironment (2). EMT program is orchestrated by the dynamic and reversible changes in epithelial and mesenchymal gene expression (3). One of the most-characterized hallmarks of EMT is the functional loss of E-cadherin, a defining marker of epithelial cells. A cohort of transcription factors such as ZEB and SNAIL family function as molecular switches to activate EMT-inducing transcriptional program through the downregulation of E-cadherin. Furthermore, owing to the plastic nature of EMT, epigenetic regulations including histone modification and DNA methylation are considered as one of the critical mechanisms for EMT (4,5). Thus, for example, several groups including us have investigated the functional connection between transcriptional repression of CDH1/E-cadherin and posttranslational modifications of histones (6)(7)(8).
Posttranslational covalent histone modifications such as methylation, acetylation, ubiquitination, and phosphorylation are central to epigenetic control of gene expression by modulating chromatin structure (9). Lysine (K) methylation of histone H3 (K4, K9, K27, and K36), in particular, has attracted much attention because of its dynamic outcomes of gene expression (10,11). Methylation of H3K4 and H3K36 is a hallmark of transcriptional active genes, whereas methylation of H3K27 and H3K9 is generally linked to gene repression. These methylations are controlled by the distinct enzymes, namely histone lysine methyltransferases (KMTs) and lysine demethylases (KDMs) called "writers" and "erasers," respectively. Strict regulation of these enzymes is essential for the biological output, and the deregulated expression of them has been associated with the developmental defects and the pathogenesis of cancer (10)(11)(12). We have previously identified many histone KMTs and KDMs as candidate cancer genes by retroviral insertional mutagenesis in mice and reported that some of them are implicated not only in cancer initiation but also in malignant progression such as cell invasion and EMT (12)(13)(14). Especially, the regulation of histone H3K27 methylation by polycomb repressive complex-2 (PRC2) was shown to be indispensable for EMT-inducing gene expression program in lung and colon cancer cells (15)(16)(17). We have also discovered that target gene-specific recruitment of PRC2 in EMT was regulated by the specific long noncoding RNA (lncRNA) through its "initiator" function (18,19).
Polycomb group (PcG) proteins are major epigenetic regulators in cell fate, development, and cancer and assembled into two complexes, PRC1 and PRC2 (20,21). The PRC2 complex, which comprises of EZH2, EED, SUZ12, and other associated proteins, conducts histone H3K27 methylation, thereby adding a repressive mark on the chromatin (22,23). On the other hand, PRC1 complex catalyzes monoubiquitination of histone H2A on lysine 119 (H2AK119Ub). It is subdivided into two subfamilies of canonical PRC1 and variant PRC1, which include a heterogenous group of several complexes (24). The catalytic core of PRC1 consists of an E3 ubiquitin ligase, RING1A or RING1B, and one of six PCGF proteins. The cooperative activities of PRC1 and PRC2 are thought to be important for PcG-mediated transcriptional repression (25). However, the molecular mechanism by which PRC1 and PRC2 recognize their target genes and induce repressive chromatin state seems to be different between canonical and variant PRC1. The canonical PRC1 containing CBX proteins can recognize H3K27me3 and occupy the chromatin modified by PRC2 to mediate gene repression. In contrast, variant PRC1 complexes have been proposed to contribute to transcriptional repression of the target genes in the absence of pre-existing H3K27me3 (26,27).
In this study, we tried to clarify the involvement of PRC1 components in the regulation of TGF-β-induced EMT of A549 lung cancer and Panc1 pancreatic cancer cell lines. We found that KDM2B, one of the members of variant PRC1 complex, was indispensable for EMT as an essential regulator of transcriptional repression of several epithelial marker genes including CDH1, miR200a, and CGN. Mechanistically, KDM2B was shown to associate with the regulatory regions of these epithelial genes and epigenetically repress the expression through histone H2A ubiquitination, indicating the importance of KDM2B-mediated PRC1 activity during EMT process.

KDM2B is indispensable for TGF-β-induced EMT process in A549 lung cancer and Panc1 pancreatic cancer cells
In order to find the PRC1 components involved in TGF-βinduced EMT of cancer cells, we tried to perform a candidate gene approach. Namely, we examined the effect of downregulation of each PRC1-related gene by CRISPR/Cas9 system on the changes in EMT character of A549 lung cancer cell line (Fig. S1). We picked up 23 PRC1-related genes encoding the core subunits (RING1A/1B and six PCGF proteins) and the accessary proteins based on the previous report (28). Five different sgRNAs for each candidate gene were introduced into the doxycycline-inducible Cas9-expressing A549 cells (A549/ Cas9h, see experimental procedures) by lentiviral infection. The infected cells were treated with doxycycline and the knockdown efficiency of each gene was examined by quantitative RT-PCR (QRT-PCR). The expressions of all 23 PRC1related genes were significantly decreased after doxycycline treatment (Fig. S2A). For several genes that showed relatively small reduction at mRNA levels, we performed immunoblotting to see the knockdown effects at protein levels. We confirmed that RING1A, RING1B, PHC1, SKP1, KDM2B, and RYBP proteins were efficiently downregulated by the sgRNAs (Fig. S2B). To find the potential involvement of each PRC1 member in EMT process, we analyzed the expression of CDH1/E-cadherin, an epithelial marker gene, in the knockdown cells by QRT-PCR (Fig. S2C). The expression of CDH1 was significantly affected by the sgRNAs for five PRC1-related genes, PCGF1, BCOR, SKP1, KDM2B, and RYBP (Fig. S2C). Among them, we decided to focus on KDM2B gene as a candidate in this study because its downregulation caused the most remarkable changes in CDH1 expression in A549 cells.
To validate the involvement of KDM2B in TGF-β-induced EMT process, we used two different cancer cell lines, A549 lung cancer and Panc1 pancreatic cancer cells, and treated them with TGF-β. QRT-PCR revealed that KDM2B expression was increased by TGF-β in both cells, suggesting its possible involvement (Fig. S3, A-B). We also detected the increase of endogenous KDM2B protein by TGF-β with the immunoprecipitation followed by immunoblotting (Fig. S3, E-F), although we did not detect endogenous KDM2B protein with our KDM2B antibody by simple immunoblotting (See Fig. S4, C-D). Then we examined the effects of KDM2B knockdown in A549 and Panc1 cells to clarify its function in EMT process. We used two different shRNAs (KDM2B shRNA#1 and #2) and confirmed the downregulation of KDM2B even in the presence of TGF-β in both cells by QRT-PCR and immunoprecipitation (Fig. S3, A, B, E and F). We also observed that both KDM2B shRNAs caused the same effects in the expression of CDH1 gene in both cells (Fig. S3, C-D), and therefore the data of KDM2B shRNA#2 were shown as a representative result.
During EMT, epithelial cells acquire a mesenchymal phenotype through downregulating epithelial cell markers such as CDH1, miR200a (microRNA-200a), and CGN/Cingulin and upregulating mesenchymal cell markers including VIM, FN1, ZEB1, ZEB2, SNAI1, SNAI2, and CDH2 (3,29). We examined the expression levels of them by QRT-PCR. KDM2B knockdown significantly increased the expression of epithelial marker genes such as CDH1, which was consistent with the above results in Fig. S1, and inhibited its transcriptional repression mediated by TGF-β in A549 and Panc1 cells (Fig. 1,  A-B). For the mesenchymal markers other than ZEB family, KDM2B knockdown by itself had no influence on their expression (Fig. 1, A-B), although FN1 expression was too low to be analyzed in Panc1 cells. For ZEB1 and ZEB2, the targets of miR200a, KDM2B knockdown itself slightly but significantly reduced their expressions, which was correlated well with the upregulation of miR200a (Fig. 1, A-B). Importantly, KDM2B knockdown prevented TGF-β-dependent increase of the expression of mesenchymal marker genes in the cells (Fig. 1,  A-B). We could also confirm these inhibitory effects of KDM2B knockdown on the protein expression of E-cadherin, Vimentin, and Fibronectin during TGF-β-induced EMT in A549 and Panc1 cells (Fig. 1, C-D). To examine the possibility Epigenetic regulation of EMT by KDM2B that KDM2B knockdown would influence TGF-β signaling pathway in the cells, the status of phosphorylated SMAD3 protein was analyzed (2). The phosphorylation levels of SMAD3 in the presence of TGF-β were similar in the control cells and the KDM2B knockdown cells (Fig. 1, C-D), indicating that TGF-β signaling pathway would not be impaired by KDM2B knockdown. These results suggested that KDM2B might be involved in TGF-β-dependent transcriptional regulation of EMT-related genes in lung and pancreatic cancer cells without influencing the direct activation of transcription factors by TGF-β.
Next we examined the cell morphologies and the state of E-cadherin and actin in A549 and Panc1 cells associated with EMT phenotypes. Control cells showed scattered, elongated, or enlarged cell shapes, lost E-cadherin staining on the cell membrane, and caused well-organized formation of actin stress fibers after TGF-β treatment (Fig. 1, E-F). KDM2B knockdown itself seemed to enhance cell adhesion and increased E-cadherin staining slightly (Fig. 1, E-F), which was consistent with the results of QRT-PCR and immunoblotting ( Fig. 1, A-D). In addition, KDM2B knockdown markedly suppressed the ability of TGF-β to induce mesenchymal morphology, E-cadherin downregulation, and actin remodeling ( Fig. 1, E-F). These results indicated that KDM2B knockdown antagonized TGF-β-induced EMT phenotypes of A549 and Panc1 cells such as morphological changes and cytoskeletal rearrangements.
We further measured cell motility and cell invasion by the transfilter assays to observe the effects of KDM2B knockdown in EMT-associated phenotypes. In line with the inhibition of EMT, KDM2B knockdown significantly attenuated TGF-βinduced cell migration (Fig. 1G) and cell invasion (Fig. 1H) in A549 and Panc1 cells. Taken together, we concluded that KDM2B was indispensable for TGF-β-induced EMT process in A549 lung cancer and Panc1 pancreatic cancer cells.
Overexpression of KDM2B partially influenced the gene expression program in EMT.
The findings above triggered us to further investigate the effects of KDM2B overexpression in EMT process of A549 and Panc1 cells. Both cells were infected with the retroviruses expressing human full-length KDM2B (WT), and its overexpression was confirmed by QRT-PCR (Fig. S4, A-B) and immunoblotting ( Fig. S4, C-D). Then we assessed the expression of EMT marker genes in A549 and Panc1 cells. Overexpression of KDM2B led to significant decrease in the expression of epithelial genes such as CDH1, miR200a, and CGN and significant increase in ZEB family expression possibly regulated by miR200a (Fig. 2, A-B). However, KDM2B overexpression had no effect in the expression of mesenchymal genes other than ZEB family (Fig. 2, A-B). In the presence of TGF-β, KDM2B overexpression potentiated the effects of TGF-β in the expression of these EMT marker genes (Fig. 2, A-B). We also confirmed the similar effects of KDM2B overexpression in protein expression of E-cadherin, Vimentin, and Fibronectin in both cells (Fig. 2, C-D). These results indicated that KDM2B overexpression by itself could repress the expression of several epithelial marker genes but not affect the expression of most of mesenchymal markers in A549 and Panc1 cells.
For the cell morphologies of A549 and Panc1 cells, cells overexpressing KDM2B revealed slight scattering compared with the control cells, but the effect was not significant (Fig. 2, E-F). More importantly, KDM2B overexpression reduced Ecadherin staining but failed to induce well-organized actin stress fiber formation (Fig. 2, E-F). We could observe that KDM2B overexpression slightly enhanced EMT-related morphological changes of the cells in the presence of TGF-β ( Fig. 2, E-F). The observed cell phenotypes caused by KDM2B overexpression were closely correlated with its effects on gene expression (Fig. 2, A-D). Cell motility and invasion assays in A549 and Panc1 cells also revealed that KDM2B overexpression showed no significant change but enhanced the effect of TGF-β (Fig. 2, G-H). These results suggested that KDM2B overexpression by itself could induce the downregulation of several epithelial marker genes, which is one of the critical events in EMT, but might not proceed EMT process completely.
Overexpression of KDM2B influenced histone H2A ubiquitination and H3 methylation on the regulatory regions of several epithelial genes such as CDH1, miR-200a, and CGN For the transcriptional repression of these epithelial genes by KDM2B, we hypothesized that KDM2B function as a component of PRC1 complex might be important. Since PRC1 is involved in monoubiquitination of lysine 119 of histone H2A (H2AK119) for gene repression (20,21), we first performed chromatin immunoprecipitation (ChIP) assay to analyze the status of histone H2A ubiquitination on the regulatory regions of CDH1, miR200a/b, and CGN genes in A549 cells. TGF-β treatment resulted in a clear increase of ubiquitinated H2AK119 (H2AK119Ub) on these regulatory regions and overexpression of KDM2B enhanced H2AK119Ub in the presence or absence of TGF-β (Fig. 3, A-C). In addition, we could detect the recruitment of endogenous and exogenously expressed KDM2B on these regulatory regions, and it was stimulated by TGF-β treatment (Fig. 3, A-C). These states of H2AK119 ubiquitination and KDM2B recruitment on these regulatory regions were well correlated with the transcriptional repression of these genes ( Fig. 2A). Interestingly, we could also observe the changes in histone H3K27 trimethylation (H3K27me3) and EZH2 recruitment induced by KDM2B overexpression. As reported previously, TGF-β caused the increase of H3K27me3 and EZH2 binding on the regulatory regions of these epithelial genes, which was closely correlated with transcriptional repression (Fig. 3, A-C) (15,16). KDM2B overexpression by itself increased H3K27me3 levels and EZH2 binding and further enhanced them in the presence of TGF-β (Fig. 3, A-C). These results suggested that KDM2B or KDM2B-mediated H2A ubiquitination might enhance the EZH2 recruitment for H3K27 methylation on these regulatory regions. On the regulatory region of unrelated GAPDH gene, we did not observe any changes in H2AK119Ub, KDM2B, H3K27me3, and EZH2 signals by KDM2B overexpression (Fig. 3D). These results indicated that KDM2B could interact with the chromatin regions of the specific target genes including CDH1, miR200a/b, and CGN and mediate H2A ubiquitination, EZH2 recruitment, and H3K27 methylation. Thus we concluded that KDM2B function as a component of PRC1 complex was involved in the transcriptional repression of these epithelial genes.
Combined knockdown of RING1A and RING1B inhibited the expression changes of the epithelial genes similarly in the case of KDM2B knockdown during EMT Since a variant PRC1 complex contains KDM2B and RING1A or RING1B, an essential catalytic subunit of PRC1, we tried to examine the involvement of RING1A and RING1B during TGF-β-induced EMT (24,26,27). The expression of RING1A and RING1B was shown significantly upregulated in response to TGF-β in A549 and Panc1 cells (Fig. S5, A-D), suggesting their possible involvement in EMT. Then we

Epigenetic regulation of EMT by KDM2B
performed knockdown experiments for RING1A and RING1B during EMT process. We showed the data using RING1A shRNA#1 and RING1B shRNA#2 as a representative result (See Methods and Fig. S5, A-D). When the expression of either RING1A or RING1B was knocked down in A549 and Panc1 cells, we did not see any changes in CDH1 expression during TGF-β-induced EMT (Fig. S6, A-B). These results were consistent with the results of CRISPR/Cas9 screening showing that CDH1 expression was not significantly changed by the sgRNAs of either RING1A or RING1B (Fig. S2C). This might be due to the compensatory effect of redundant function of RING1A and RING1B. Thus, we tried to knock down both RING1A and RING1B by the coinfection of lentiviruses expressing each shRNA, and the reduced expression of each gene was confirmed by QRT-PCR (Fig. S7). With the combined knockdown of RING1A and RING1B (indicated as RING1A/1B), we could detect the clear effects on EMT-related gene expression in A549 and Panc1 cells (Fig. 4, A-B). Combined knockdown of RING1A/1B increased the expression of epithelial genes such CDH1, miR200a, and CGN and inhibited its transcriptional repression mediated by TGF-β similarly in the case of KDM2B knockdown (Fig. 4, A-B). RING1A/1B knockdown itself decreased ZEB family expression significantly and had no influence on the expression of other mesenchymal genes. However, RING1A/1B knockdown counteracted the effect of TGF-β on all these mesenchymal gene expressions in both cells (Fig. 4, A-B). These results indicated that RING1A and RING1B, catalytic subunits of PRC1, were also indispensable for the downregulation of these several epithelial marker genes during TGF-β-induced EMT. The similar phenotypes of RING1A/1B and KDM2B knockdown suggested the important role of a variant PRC1 complex containing RING1A/1B and KDM2B in EMT.
Next we examined how KDM2B and RING1A/1B affected the expression of epithelial genes such as CDH1, miR200a, and CGN during EMT of A549 cells by the ChIP experiment ( Fig. 4, C-E). KDM2B or RING1A/1B knockdown by itself caused significant decrease of H2AK119Ub on these regulatory regions and antagonized TGF-β-effect on the increase of H2AK119Ub (Fig. 4, C-E), indicating the essential function of KDM2B and RING1A/1B in histone H2A ubiquitination during EMT. KDM2B recruitment on these regulatory regions was clearly reduced by KDM2B knockdown, but not significantly affected by RING1A/1B knockdown (Fig. 4, C-E). More importantly, TGF-β-dependent increase of H3K27me3 levels and EZH2 binding on these epithelial genes was inhibited by either KDM2B or RING1A/1B knockdown (Fig. 4, C-E). These changes were specific to the regulatory regions of CDH1, miR200a, and CGN genes and not observed at the regulatory region of unrelated GAPDH gene (Fig. 4F). These results suggested that the ubiquitination of H2AK119 on the regulatory regions of these epithelial genes by KDM2B and RING1A/ 1B was indispensable for the subsequent increase of EZH2 binding and H3K27me3 levels, which resulted in the transcriptional repression of these genes during TGF-β-induced EMT.
DNA recognition property but not histone demethylase activity was indispensable for KDM2B function in transcriptional regulation during EMT To gain further insight into the epigenetic regulation of the genes by KDM2B, we constructed two types of KDM2B mutants for the overexpression experiments. The JmjC mutant is defective in histone H3 demethylase activity (30) and the CXXC mutant lacks DNA recognition property of KDM2B (31). A549 and Panc1 cells were infected with the control retrovirus or the retroviruses expressing wild-type KDM2B (WT) or the mutants, and the expression levels of them were confirmed by QRT-PCR (Fig. S8, A-B) and immunoblotting (Fig. S8, C-D). Then we analyzed the expression of EMT marker genes in A549 and Panc1 cells. In the case of epithelial genes such CDH1, miR200a, and CGN, JmjC mutant by itself decreased the expression similarly with wild-type KDM2B, but CXXC mutant increased the expression significantly in A549 and Panc1 cells (Fig. 5, A-B). JmjC mutant also enhanced the effect of TGF-β in the expression of these genes while CXXC mutant reduced it (Fig. 5, A-B). For ZEB family, JmjC mutant itself increased the expression similarly with wild-type, but CXXC mutant decreased the expression. Both mutants had no effect on other mesenchymal marker genes in the absence of TGF-β, but the effects of TGF-β were potentiated by JmjC mutant and counteracted by CXXC mutant (Fig. 5, A-B). Similar effects of these mutants were confirmed in protein expression of E-cadherin, Vimentin, and Fibronectin in both cells (Fig. 5, C-D). These results indicated that DNA recognition property but not histone H3 demethylase activity of KDM2B was indispensable for the downregulation of epithelial genes such as CDH1, miR200a, and CGN during EMT. In addition, it was suggested that DNA recognition defective mutant (CXXC mutant) might behave like a dominant negative mutant in the transcriptional regulation.
Next we examined the state of histone H2A ubiquitination and H3 methylation on the regulatory regions of CDH1, miR200a, and CGN genes in A549 cells expressing KDM2B mutants. The JmjC mutant showed the same properties with wild-type KDM2B for its recruitment, H2AK119Ub increase, EZH2 recruitment, and H3K27me3 increase on these regulatory regions in the presence or absence of TGF-β (Fig. 5, E-G). In contrast, CXXC mutant significantly reduced H2AK119Ub and H3K27me3 modification on these epithelial genes with or without TGF-β (Fig. 5, E-G). The decreases of KDM2B and EZH2 recruitment on these regulatory regions in the absence of TGF-β were not statistically significant, but TGF-β-dependent increases of their recruitment were significantly reduced by the overexpression of CXXC mutant (Fig. 5, E-G). We did not detect any changes in H2AK119Ub, KDM2B, H3K27me3, and EZH2 signals by KDM2B mutants on the regulatory region of unrelated GAPDH gene (Fig. 5H). These results indicated that DNA recognition property but not histone H3 demethylase activity of KDM2B was indispensable for its function regulating histone H2AK119Ub and H3K27me3 modification on the specific target genes including CDH1, miR200a/b, and CGN. Furthermore, the results of CXXC mutant suggested that this DNA recognition defective mutant of KDM2B might partially inhibit the function of endogenous PRC1 complex for transcriptional regulation. These results together suggested that DNA recognition property of KDM2B might play an important role in the selection of target genes for regulating histone H2AK119 ubiquitination and thereby affecting the subsequent EZH2 recruitment and H3K27me3 modification.
We further tried to examine whether KDM2B histone demethylase activity would be involved in the gene expression changes during EMT. Since KDM2B is a histone H3K36 demethylase (32), we first analyzed the effects of KDM2B knockdown or KDM2B overexpression on the overall levels of histone H3K36 methylation in A549 and Panc1 cells. We also examined the levels of H3K27 methylation because our results indicated KDM2B changed H3K27me3 levels at the local regulatory regions of several epithelial genes. The immunoblotting revealed that KDM2B knockdown did not change the overall levels of H3K36me2 and H3K27me3 in A549 and Panc1 cells (Fig. S9). This might be due to the compensatory effect of redundant function of other H3K36 demethylases. On the other hand, KDM2B overexpression decreased the overall levels of H3K36me2 significantly, but had no effect on H3K27me3 (Fig. S10), which was consistent with the results of the previous paper (32). These results suggested that KDM2Bmediated changes of H3K27 methylation were only observed at the specific target genes and possibly caused by KDM2B function as a component of PRC1 complex. Next we examined the state of H3K36 methylation at the local level on the regulatory regions of the epithelial genes such as CDH1, miR200a, and CGN and unrelated GAPDH gene in A549 cells. The ChIP  analysis revealed that the levels of H3K36me2 were originally very low on the regulatory regions of the epithelial genes and not changed significantly by KDM2B knockdown or KDM2B overexpression (Figs. S11 and S12). In contrast, we observed that the levels of H3K4me3 on these regulatory regions were dynamically changed with KDM2B knockdown or KDM2B overexpression, which were proportional to the expression levels of these genes as reported previously (Figs. S11 and S12) (15,16). There were no significant changes observed at the regulatory region of unrelated GAPDH gene (Figs. S11D and S12D). These results suggested that H3K36 methylation was not responsible for the transcriptional regulation of the epithelial genes including CDH1, miR200a, and CGN by KDM2B.

Knockdown of EED, a core component of PRC2, inhibited KDM2B-mediated expression changes of EMT-related genes in EMT process
To clarify the relationship of KDM2B-containing PRC1 complex and PRC2 complex in the regulation of EMT-related genes, we examined the knockdown effects of EED, one of the core components of PRC2, in EMT process of A549 and Panc1 cells. We showed the data using EED shRNA#1 as a representative result because it caused efficient decrease of EED expression in the cells as reported previously (15). EED knockdown significantly increased the expression of epithelial marker genes such as CDH1, miR200a, and CGN and inhibited its transcriptional repression mediated by TGF-β in A549 and Panc1 cells (Fig. 6, A-B). For the mesenchymal markers, EED knockdown by itself had no influence on their expressions and prevented TGF-β-dependent increase of them as described previously (Fig. 6, A-B) (15). In the KDM2B-overexpressing cells, EED knockdown canceled KDM2B-induced downregulation of these epithelial genes, KDM2B-mediated upregulation of ZEB family, and TGF-β-dependent expression changes of all EMT-related genes (Fig. 6, A-B). These results indicated that EED was indispensable for KDM2B-dependent transcriptional regulation of EMT-related genes during EMT.
Next we examined how EED affected the KDM2B-mediated expression of epithelial genes such as CDH1, miR200a, and CGN in A549 cells by ChIP experiment (Fig. 6, C-E). As reported previously, EED knockdown by itself caused significant decrease of EZH2 recruitment and H3K27 methylation on these regulatory regions and antagonized TGF-β-effect on the increase of them (15). Importantly, KDM2B-dependent increase of EZH2 recruitment and H3K27 methylation was inhibited by EED knockdown (Fig. 6, C-E). These results indicated that EED was indispensable for KDM2B-mediated increase of EZH2 recruitment and H3K27 methylation on the regulatory regions of CDH1, miR200a, and CGN genes. For KDM2B recruitment, EED knockdown had no influence on the effects of TGF-β treatment and KDM2B overexpression (Fig. 6, C-E). EED knockdown could not cancel the effects of TGF-β or KDM2B overexpression on the increase of H2AK119Ub, but showed significant inhibition on its increase (Fig. 6, C-E). These results suggested that KDM2B-independent mechanism for H2A ubiquitination might be partly involved in this process. These observed changes were specific to the regulatory regions of CDH1, miR200a, and CGN genes and not detected at the regulatory region of unrelated GAPDH gene (Fig. 6F). The EED knockdown studies suggested that PRC2 function indicated by EZH2 recruitment and H3K27 methylation was indispensable for KDM2B-mediated transcriptional repression of CDH1, miR200a, and CGN genes. Together with the ChIP results shown in Figures 3 and 4, it was concluded that KDM2B-mediated H2A ubiquitination resulted in the subsequent increase of EZH2 binding and H3K27me3 levels on the regulatory regions and that these PRC1 and PRC2 activities were both indispensable for the transcriptional repression of the epithelial genes during EMT process.

Discussion
In this study, we found that KDM2B was indispensable for TGF-β-induced EMT of A549 lung cancer and Panc1 pancreatic cancer cell lines. KDM2B knockdown inhibited EMT phenotypes by antagonizing TGF-β-dependent changes in the expression of EMT-related genes. Overexpression of KDM2B and its mutants revealed that KDM2B was involved in transcriptional repression of several epithelial marker genes such as CDH1, miR200a, and CGN through its function as a variant PRC1 component. From the mechanistic investigations, it was strongly suggested that KDM2B could specifically recognize the regulatory regions of these epithelial genes and mediate histone H2AK119 monoubiquitination, thereby activating the EZH2 recruitment and histone H3K27 methylation for gene repression. This study revealed a novel functional significance of KDM2B in the epigenetic control of EMT process of lung and pancreatic cancer cells.
The polycomb group (PcG) proteins constitute PRC1 and PRC2 complex that function as epigenetic regulators of gene expression in cell fate, development, and cancer (20,21). While PRC2 induces histone H3K27 methylation by its catalytic subunit EZH2, PRC1 monoubiquitylates H2AK119 through its catalytic subunit, RING1A or RING1B ubiquitin ligase, for transcriptional repression. A growing body of evidence indicated that increased activity of PRC2 or EZH2 was responsible for malignant phenotypes of cancer cells (23,33). Especially, PRC2 was shown to be involved in the critical step during EMT. It has been reported that EZH2 downregulated CDH1/ E-cadherin expression by inducing histone H3K27 methylation on its promoter region (6). In addition, we found that EED, a core component of PRC2, and JARID2, an accessary factor of PRC2, were essential for TGF-β-induced EMT through CDH1 repression in cancer cells (15,16). On the other hand, several studies have demonstrated the role of PRC1 members in EMT process. For example, it was shown that BMI1, a component of PRC1, acted cooperatively with TWIST1, one of EMT-related transcription factors (EMT-TFs), to repress CDH1 during EMT in head and neck squamous cell carcinoma (34). RING1A/1B was reported to interact with SNAI1, another EMT-TF, and contribute to SNAI1-mediated repression of CDH1 in pancreatic cancer cells (35). In this report, we Epigenetic regulation of EMT by KDM2B  discovered that KDM2B, a member of variant PRC1 complex and also a histone H3 demethylase, epigenetically regulated the downregulation of several epithelial marker genes including CDH1, which was indispensable for TGF-β-induced EMT of lung and pancreatic cancer cells. Since the JmjC mutant of KDM2B exhibited the similar properties with wild-type KDM2B in the regulation of EMT-related genes, histone H3 demethylase activity of KDM2B was suggested to be dispensable for its function in EMT. The ChIP experiments (Figs. S11 and S12) also revealed that H3K36 methylation was not responsible for the transcriptional regulation of these epithelial genes by KDM2B, although KDM2B is a histone H3K36 demethylase. Therefore, our studies validated the importance of KDM2B function that regulates the PRC1 activity in EMTinducing transcriptional program.
Currently, PRC1 complex is divided into two subfamilies of canonical PRC1 and variant PRC1. Both PRC1 exist in different forms with distinct subunit compositions influencing the mechanistic properties of the complexes (24). The canonical PRC1 complex can recognize H3K27me3 and is recruited in a hierarchical manner to the sites modified by PRC2 for gene repression (21). The above two papers reporting the involvement of BMI1 and RING1A/1B in EMT have shown that prior modification of H3K27 by PRC2 was important for TWIST-or SNAI1-mediated recruitment of canonical PRC1 containing BMI1 and/or RING1A/1B on the CDH1 promoter. In contrast, our experiments showed that knockdown of KDM2B or RING1A/1B (Fig. 4) inhibited TGF-β-dependent increase of not only H2AK119Ub, but also H3K27me3 and EZH2 binding on the regulatory regions of the epithelial genes including CDH1 for transcriptional repression. These results suggested that the H2AK119 ubiquitination on these regulatory regions by the variant PRC1 complex including KDM2B and RING1A/1B was necessary for the subsequent PRC2 activation indicated by the increase of EZH2 binding and H3K27me3 levels. This is consistent with the proposed model of variant PRC1 function in gene repression in the absence of pre-existing H3K27me3 reported in the previous papers (See the possible mechanism in Fig. 7) (26,27). The variant PRC1 complex can recognize specific target genes to induce H2A ubiquitination, and then PRC2 is recruited to the sites in a hierarchical manner for H3K27 methylation. Furthermore, knockdown experiments of EED, a core component of PRC2, suggested that PRC2 function was indispensable for KDM2B-mediated transcriptional repression of CDH1, miR200a, and CGN genes. These results together indicated that PRC1 and PRC2 activities were both necessary for the transcriptional repression of these epithelial genes during EMT process. However, further studies will be required to address how PRC1 and PRC2 complexes interact with each other on the target genes to induce transcriptional repression.
In our experiments, DNA recognition defective mutant (CXXC) of KDM2B reduced TGF-β-dependent increases of endogenous KDM2B recruitment and H2AK119Ub modification on the regulatory regions of the epithelial genes including CDH1, miR200a, and CGN and might function as a dominant negative mutant in the transcriptional repression of these genes. This mutant also decreased the subsequent increases of EZH2 binding and H3K27 trimethylation on the genes. Taken together, the DNA recognition property of KDM2B was indispensable for its function to recognize the specific target genes and regulate gene repression through PRC1 activity during EMT (Fig. 7). These results are consistent with the previous reports showing that KDM2B recognized the genomic DNA in CpG islands (CGIs) via its CXXC domain and contributed to recruit variant PRC1 complex for gene repression independently of pre-existing H3K27 methylation in mouse embryonic stem (ES) cells (36,37). The CGIs were found in the regulatory regions of CDH1, miR200a, and CGN genes and were included in the probes used in our ChIP experiments. These regulatory regions revealed KDM2B binding, H2AK119 ubiquitination, subsequent EZH2 recruitment, and H3K27 methylation in the ChIP assays, which was correlated well with the transcriptional repression. Therefore, it was considered that the recognition of CGIs by KDM2B played a crucial role in the determination of target specificity. However, in contrast to ES cells, the KDM2B-binding CGIs have not been extensively analyzed in cancer cells such as A549 and Panc1 cells. Further examinations are warranted to clarify how KDM2B selects the specific target genes in cancer cells.
Among the mesenchymal marker genes, ZEB1 and ZEB2 were downregulated by KDM2B knockdown and upregulated by KDM2B overexpression (Figs. 1 and 2). Since ZEB family is a target of miR200a, the expression levels of ZEB family were inversely correlated with the levels of miR200a. Therefore, it was suggested that KDM2B affected the expression of ZEB1 and ZEB2 indirectly through the epigenetic regulation of miR200a. This is also consistent with our results in the previous papers (15,16). For other mesenchymal marker genes, KDM2B knockdown or KDM2B overexpression by itself had no effect in the expression of them but potentiated the effects of TGF-β ( Figs. 1 and 2). We tried to examine the effects of KDM2B on the mesenchymal genes such as FN1 and VIM by ChIP assays. We analyzed the state of H2AK119Ub, KDM2B recruitment, H3K27me3, EZH2 binding, H3K36me2, and H3K4me3 on the regulatory regions of FN1 and VIM genes in A549 cells with KDM2B knockdown or KDM2B overexpression (Figs. S13 and S14). We did not detect any significant levels of signals in H2AK119Ub, KDM2B, H3K27me3, EZH2, and H3K36me2 to see the effects of KDM2B. These results suggested that KDM2B was not directly involved in the epigenetic regulation of FN1 and VIM genes. We could observe that the levels of H3K4me3 on these regulatory regions were dynamically changed with TGF-β treatment, KDM2B knockdown, or KDM2B overexpression (Figs. S13 and S14). These changes of H3K4me3 appeared to be proportional to the expression levels of these genes. In mammalian cells, H3K27 methylation is catalyzed by polycomb (PcG) family and H3K4 methylation is mostly regulated by COMPASS family proteins (38). These two families are well known for their opposing roles in balancing gene expression. In this study, we mainly focused on the function of PcG family, PRC1 and PRC2, in transcriptional repression during EMT. However, it should be the next important subject to investigate the regulation of H3K4 methylation in EMT process.
A previous report showed that KDM2B knockdown and overexpression influenced the expression of both epithelial and mesenchymal genes in HCT116 colon cancer cell line (39), indicating a relationship between KDM2B expression and EMT. However, in colon cancer cells, it remains unclear whether KDM2B-induced EMT was dependent on its histone demethylase activity or PRC1 activity. Furthermore, a lot of evidences have revealed that histone modifying enzymes such as KDM2B are involved in the initiation and progression of human cancer (11,12). These epigenetic regulatory factors have been sometimes reported to act as oncogene products in some types of cancers or to reveal tumor suppressive function in other cancers, suggesting that their contribution to cancer is highly dependent on many factors and cellular contexts (40). Therefore, careful analyses and discussions are critical to understand their roles in different types of cancers. In this paper, we have found an essential function of KDM2B as a component of variant PRC1 complex during TGF-β-induced EMT of A549 lung cancer and Panc1 pancreatic cancer cell lines. KDM2B played a crucial role in EMT process by regulating the expression of several epithelial marker genes such as CDH1, miR200a, and CGN through the modulation of histone H2A ubiquitination. Our study provides novel insights into the involvement of variant PRC1 complex for epigenetic mechanism of cancer malignancies.

Plasmids, cell culture, transfections, and antibody
For the knockdown experiments, lentiviral vectors expressing small hairpin RNAs (shRNAs) were constructed as described previously (41). The pLKO.1-Puro plasmid (Sigma-Aldrich, St Louis, USA) and its derivative, pLKO.1-Neo plasmid were used for the combined knockdown. The oligonucleotide sequences for shRNAs were described previously (16) and are listed in Table S1. We usually used two different shRNAs (shRNA#1 and shRNA#2) for each gene. Two shRNAs have similar efficiencies for knockdown of each gene (Figs. S3, A-B and S5) and have similar effects on EMT phenotypes judged from the marker gene expression (Fig. S3, C-D). Thus, the data of one representative shRNA for each gene were shown in the main figures and the text. For cloning of human KDM2B cDNA, the primer sets described in Table S1 were designed based on the reference sequence (ENST00000636052.1) in GENCODE database. The two amplified cDNAs were connected with HincII site and cloned into pDON-5 Neo plasmid (Takara, Ohtsu, Japan) to produce retrovirus expressing KDM2B. KDM2B JmjC mutant (H242A) that is catalytically deficient in histone H3 demethylase activity (30) was constructed by PrimeSTAR Mutagenesis Basal kit (Takara, Ohtsu, Japan) using JmjC mutant primer set described in Table S1. KDM2B CXXC mutant (C613A, C616A, C619A) abolishing the DNA recognition property (31) was also constructed by the same kit. The C613A mutation was introduced to wild-type KDM2B using CXXC mutant1 primer set (Table S1) and then the C616A and C619A mutations were simultaneously introduced with CXXC mutant2 primer set (Table S1). The nucleotide exchanges of the mutants were confirmed by DNA sequencing (Fig. S15). The sequenceverified PCR products were used for cloning and plasmid construction. Human lung cancer and pancreatic cancer cell lines, A549 and Panc1, were obtained from ATCC and maintained in Dulbecco's modified Eagle's medium (DMEM) with 10% FBS, 2 mM glutamine, and penicillin/streptomycin (Sigma-Aldrich, St Louis, USA) at 37 C in 5% CO2. A549 and Panc1 cells are a good model system for EMT showing rapid and clear changes for cell morphology and EMT-related gene expression induced by treatment of 1 ng/ml of TGF-β (R&D Systems, Minneapolis, USA) for 24 to 72 h. The methods for the production and infection of shRNA-expressing lentiviruses or cDNA-expressing retroviruses were essentially the same as described previously (8). For the preparation of anti-KDM2B rabbit polyclonal antibody, histidine-tagged N-terminal KDM2B (M1 to Y396) protein purified from genetically engineered Escherichia coli was used.

Epigenetic regulation of EMT by KDM2B
lentiviruses were essentially the same as described previously (8). Doxycycline-inducible Cas9-expressing A549 cells were established by the infection of pCW-Cas9 (Addgene plasmid #50661)-derived virus (42), and the cell clone (A549/Cas9h) with the highest Cas9 induction was isolated (Fig. S1B). The A549/Cas9h cells were infected with lentiviruses expressing five sgRNAs for each gene and treated with 10 μg/ml doxycycline (Wako, Osaka, Japan) for 48 h and then used for the screening experiments (Fig. S1C). To confirm the knockdown efficiencies by the sgRNAs, quantitative RT-PCR (QRT-PCR) was performed as described below (Fig. S2A). Primers used for QRT-PCR are listed in Table S3.

Quantitative PCR
Total RNA was extracted with RNAiso plus (Takara, Ohtsu, Japan) using a standard method and transcribed to cDNA using SuperScript Vilo cDNA synthesis kit (Invitrogen, Waltham, USA). Quantitative RT-PCR (QRT-PCR) was performed as described previously (8). PCR data were normalized with control human GAPDH expression. The averages from at least three independent experiments are shown with the standard deviations. p-values were calculated between control and the samples using Student's t-test. Primers used for the quantitative PCR were described previously (8,16,41) and are listed in Tables S1 and S3.

Cell staining, immunofluorescence, immunoprecipitation, and immunoblotting
To observe cell morphologies, A549 or Panc1 cells were fixed in 4% paraformaldehyde and stained with 0.4% crystal violet. To visualize actin cytoskeleton, the cells were stained with 0.25 μM tetramethylrhodamine isothiocyanate (TRITC)conjugated phalloidin (Sigma, St Louis, USA). For indirect immunofluorescence, the specimens were treated with anti-Ecadherin antibody (#610181, BD Biosciences, San Jose, USA) and incubated with Alexa546-conjugated anti-mouse IgG antibody (Invitrogen, Waltham, USA). Nuclei were stained with 4',6-diamidino-2-phenylindole (DAPI). For immunoprecipitation and immunoblotting, cells were lysed in RIPA buffer as described previously (19). The lysates were immunoprecipitated with anti-KDM2B antibody coupled with Protein G sepharose 4 fast flow (GE Healthcare, South East England, UK). The precipitates or lysates were separated on SuperSep Ace 10% running gel (Wako, Osaka, Japan) and transferred to Hybond-LFP membrane (GE Healthcare, USA). The antibodies that were used in this study include anti-E-cadherin

Cell migration and invasion assays
Cell migration and invasion assays were carried out by using modified Boyden chambers consisting of Transwell membrane filter inserts (#3422, Corning, Corning, USA) as described previously (17). For cell invasion, the upper surfaces of the membranes were coated with 1 mg/ml Matrigel matrix (BD Bioscience, USA). Briefly, serum-starved cells (2 × 10 5 ) were suspended in the upper chamber in DMEM containing 1 mg/ml BSA and 0.5% FBS. The lower chamber was filled with DMEM containing 10% FBS. The chambers were incubated for 24 h (A549) and 36 h (Panc1) for cell migration, 36 h (A549) and 48 h (Panc1) for cell invasion. For TGF-β-treated cells, cells were pretreated with 1 ng/ml of TGF-β for 4 days and then allowed to migrate. Cells remaining on the upper surface of the membrane were removed by wiping with a cotton bud, and cells on the underside of the filter were fixed with 4% paraformaldehyde and stained with 0.4% crystal violet. Cells were counted under a microscope in at least eight different fields and two independent experiments.

Chromatin immunoprecipitation (ChIP) assays
ChIP assays were carried out essentially the same as described previously (41,44). In brief, cells were cross-linked with 1% paraformaldehyde for 10 min and chromatins were fragmented by Bioruptor II ultrasonicator (BM Equipment Co, Tokyo, Japan). The cross-linked cell lysates were immunoprecipitated with the antibodies (anti-H2AK119Ub (#8240, Cell Signaling Tech., Danvers, USA), anti-KDM2B, anti-H3K27me3 (44), anti-H3K36me2 (44), anti-H3K4me3 (44), and anti-EZH2 (#17-662, Millipore, Billerica, USA)). Quantitative PCR was performed to detect the enrichment of specific amplified region. Percentage enrichment over input chromatin was presented. Primers used for the QPCR corresponding to the regulatory region h of CDH1 gene, region a of miR200a/b gene, and region a of GAPDH gene are described previously (8). Other primers for the regulatory regions were listed in Table S1.

Data availability
All data are included within the article and supporting information. The materials and methods in this study are available from the corresponding author upon reasonable request.