Dynamic cohesin-mediated chromatin architecture controls epithelial mesenchymal plasticity in cancer

Transcription

1 Article Dynamic cohesin-mediated chromatin architecture controls epithelial mesenchymal plasticity in cancer Jiyeon Yun,, Sang-Hyun Song, Hwang-Phill Kim, Sae-Won Han,, Eugene C Yi & Tae-You Kim,,, Abstract Epithelial to mesenchymal transition (EMT) and mesenchymal to epithelial transition (MET) are important interconnected events in tumorigenesis controlled by complex genetic networks. However, the cues that activate EMT-initiating factors and the mechanisms that reversibly connect EMT/MET are not well understood. Here, we show that cohesin-mediated chromatin organization coordinates EMT/ MET by regulating mesenchymal genes. We report that, a subunit of the cohesin complex, is expressed in epithelial breast cancer cells, whereas its expression is decreased in mesenchymal cancer. Depletion of in epithelial cancer cells causes transcriptional activation of TGFB and ITGA, inducing EMT. Reduced binding of changes intrachromosomal chromatin interactions within the TGFB and ITGA loci, creating an active transcriptional environment. Similarly, stem cell-like cancer cells also show an open chromatin structure at both genes, which correlates with high expression levels and mesenchymal fate characteristics. Conversely, overexpression of in mesenchymal cancer cells induces MET-specific expression patterns. These findings indicate that dynamic cohesinmediated chromatin structures are responsible for the initiation and regulation of essential EMT-related cell fate changes in cancer. Keywords cancer stem cell; cohesin; EMT; higher-order chromatin structure Subject Categories Cancer; Chromatin, Epigenetics, Genomics & Functional Genomics DOI /embr.8 Received December Revised June 6 Accepted June 6 Published online 7 July 6 EMBO Reports (6) 7: 9 Introduction Approximately 9% of human cancer cases are caused by metastasis [], a multi-step process in which primary tumor cells disseminate from their origin site and move into secondary locations to acquire a better environment for tumor proliferation [,]. Epithelial mesenchymal transition (EMT) associated with metastasis is known as an initiation process related to intravasation [,]. During EMT, epithelial cancer cells acquire migratory potential, loss of apical basal polarity, and resistance to apoptotic stimuli that promote detachment from the origin sites and neighboring cells through coordinate gains in EMT-related genes and losses of epithelial-related genes expression. Consequently, cancer cells gradually obtain migratory and invasive capabilities [6]. According to the cancer stem cell (CSC) hypothesis, tumor growth is only driven by a subpopulation of tumor-initiating cells that have self-renewal and multi-potency properties [7]. CSCs are known to have mesenchymal traits and be capable of mobility; accordingly, they have been proposed as the seeds of metastasis [7]. Additionally, CSC plasticity postulates that cancer has CSCs and non-cscs, and that these different cancer cells have a characteristic of bidirectional conversion. Metastases in patients with aggressive forms of cancer and CSCs are closely linked to cancer mortality. Thus, understanding the regulatory mechanisms of EMT and mesenchymal to epithelial transition (MET), a reversible process of EMT, is important to preventing tumor metastasis [8]. During cancer metastasis, EMT-MET is essential for the timely and accurate differentiation of epithelial (or mesenchymal) cancer cells into mesenchymal (or epithelial) cancer cells for tumor development [9]. Over the past few decades, many groups have studied EMT and EMT-related genes such as E-cadherin (CDH), vimentin (VIM), tumor growth factor b (TGFB), twist-related protein (TWIST), b-catenin (CTNNB), zinc finger E-box-binding homeobox / (ZEB/), and the mir family []. Recent studies have provided compelling evidence of epigenetic regulation of the EMT process including histone deacetylation and DNA methylation of CDH, expression of non-coding RNA (mir- a,b,c, mir-9, and mir-) that regulates EMT plasticity and post-translational modifications such as sumoylation of ZEB, and phosphorylation of snail family zinc finger (SNAI) [ ]. Based on these findings, epigenetic mechanisms appear to play a crucial role in the regulation of EMT-related genes and are important mechanical factors during the EMT process. Despite these data elucidating various factors that control dramatic changes during EMT, the upstream regulatory Cancer Research Institute, Seoul National University College of Medicine, Seoul, Korea Department of Molecular Medicine and Biopharmaceutical Sciences, Graduate School of Convergence Science and Technology, Seoul National University College of Medicine, Seoul, Korea Department of Internal Medicine, Seoul National University Hospital, Seoul, Korea Corresponding author. Tel: ; Fax: ; kimty@snu.ac.kr ª 6 The Authors EMBO reports Vol 7 No9 6

2 EMBO reports Dynamic chromatin loops control EMT in cancer Jiyeon Yun et al molecules responsible for initiating EMT, called cues, are still unclear [7]. Higher-order chromatin architecture is known to be important to gene regulation during hematopoiesis, erythropoiesis, and development. Therefore, chromatin architectural proteins such as CCCTC-binding factor (CTCF), cohesin, and LIM domain binding (LDB) have been actively investigated [,6]. In particular, the cohesin complex, which consists of four main subunits, two structural maintenance of chromosomes (SMC) molecules SMC and SMC, either stromal antigen (SA; also known as STAG) STAG or STAG, and the kleisin subunit [7,8], tightly and dynamically controls genomic organization to regulate gene transcription, transcript splicing, chromosomal instability, and gene amplification in cancer [9,]. Substantial evidence has shown that cohesin proteins are involved in tumorigenesis. While cohesin complex is overexpressed and mutated in some types of cancer such as breast, prostate, and colon, other malignancies such as oral squamous cell carcinoma, colorectal cancer, and myeloid leukemia express low levels or mutated forms of these proteins [7,8]. It has been reported that, a subunit of cohesin complex, is expressed at a low level in metastatic breast and oral squamous cancers [,]. However, the reason why is lower in metastatic than in epithelial cancer is still unclear. Although many distinct molecular mechanisms associated with the EMT process in cancer have been identified, the contributions of higher-order chromatin architecture to regulating EMT regulation and acquisition of mesenchymal traits are still unknown []. In the present investigation, we showed that deprivation of in epithelial cancer cells increases the transcription of two EMT-related genes: TGFB and ITGA by releasing the higher-order chromatin structure of the genes. Up-regulation of these genes leads directly to EMT. Furthermore, this result is confirmed by the same outcomes in cancer stem cell-like cells (CSLCs) which are known to have mesenchymal traits. The results of this study strongly suggest that cohesin dynamically regulates the three-dimensional chromatin architectures of the key genes responsible for EMT initiation. Moreover, it directly affects gene expression, causing a reversible transition between epithelial and mesenchymal states in a subpopulation of cancer. Results expression in mesenchymal breast cancer is relatively lower than that of epithelial cells was previously demonstrated to be expressed at low levels in metastatic cancer cells such as breast and oral squamous cell carcinoma [,]. Therefore, we investigated whether is important in determining the state of either EMT or MET. To accomplish this, we compared expression between epithelial (HCC7,, and ) and highly metastatic breast cancer cell lines (MDA-MB-, HCC, and MDA-MB-7) [] (Fig A). Consistent with the results of previous studies, we observed relatively low expression of protein in mesenchymal breast cancer cell lines compared with those in epithelial breast cancer cell lines. Although the other subunits of cohesin complex SMCA and STAG were also checked, there were no significant differences between the epithelial and mesenchymal breast cancer cell lines (Fig EVA). Next, to visually compare the relationships between expression and EMT state, we analyzed immunofluorescence staining with and E-cadherin, a well-known epithelial marker, or VIM, a marker of mesenchymal trait, in or HCC cells. Consistent with the Western blot data, we observed that HCC cells expressed a high level of VIM and low levels of E-cadherin and, while cells had a low level of VIM and high levels of E-cadherin and (Fig B). To understand why different levels of protein expression were observed when comparing mesenchymal and epithelial breast cancer cells, we analyzed mrna (Fig C). No significant difference in mrna expression was observed between the cells. The mrna expression of other cohesin complex subunits did not differ greatly between cells and was not correlated with the protein levels (Fig EVB). We further checked whether the gene copy number was correlated with the protein levels in epithelial and mesenchymal breast cancer cells. The results showed that the gene copy number did not appear to be relevant to its protein levels (Fig EVC). We then assessed expression in epithelial breast cancer cells and mesenchymal breast cancer HCC cells following treatment with lg/ml actinomycin D (Act D) for,, 6, 9,, and h (Figs D and EVD F). The levels of transcripts and protein in cells remained steady after h of Act D treatment. In contrast, a dramatic reduction of transcripts and protein was observed after 6 h of Act D exposure in HCC cells. We also observed a significant reduction of c-myc transcripts, which was used as a control [,6] in both cell lines (Figs D and EVF). These findings showed that low expression of protein in mesenchymal breast cancer cell lines might be due to less stable transcripts than in epithelial cell lines with higher expression. Taken together, our results demonstrated that the stability of transcripts in mesenchymal breast cancer cell lines was relatively low compared to those in epithelial cancer cell lines, leading to low expression of in mesenchymal breast cancer cells. Disruption of expression in epithelial cancer cells induces transcription of two EMT-related genes, TGFB and ITGA Since we found that expression levels correspond to the EMT state (Fig A), we investigated whether was critical for determining the fate of cellular EMT state in cancer. To accomplish this, we stably knocked down (KD) expression using -specific shrna (shr# and shr#) in epithelial breast cancer cells and. In addition, to rule out the possibility of breast cancer specific properties, we stably depleted using shrnas in epithelial gastric cancer cells SNU6 and SNU6 (Fig A). expressions decreased by approximately 7% in all cells used for knockdown of. A chromosome spread assay showed that the chromosomes in KD cells appeared to have no defective sister chromatid cohesion during mitosis and meiosis (Fig EVA). Moreover, the 7% reduction of did not appear to affect cell cycle (Fig EVB). Interestingly, sufficient knockdown of in the cells significantly decreased E-cadherin levels and simultaneously induced VIM EMBO reports Vol 7 No9 6 ª 6 The Authors

3 Jiyeon Yun et al Dynamic chromatin loops control EMT in cancer EMBO reports Epithelial-like Mesenchymal-like A E-cadherin Vimentin α-tubulin HCC7 MDA-MB- HCC MDA-MB-7 Protein levels/α-tubulin. HCC7 MB HCC MB7. HCC7 E-cadherin MB HCC MB Vimentin HCC7 MB HCC MB7 /VIM/DAPI /E-Cad/DAPI B μm DAPI μm DAPI VIM Merge E-Cad μm HCC DAPI μm DAPI VIM Merge E-Cad D (E) HCC (M) Act D (ug/ml) 6 9 (h) (m) (P) c-myc (m) 8s (m) Anti-α-tubulin (m) (P) c-myc (m) 8s (m) Anti-α-tubulin Merge Merge C mrna expression..... Figure. A B C D Unstable mrna leads to low levels of protein in mesenchymal breast cancer cells. Representative results of Western blot analysis of, E-cadherin, and VIM are shown for luminal and basal-like breast cancer cell lines. The images were analyzed using ImageJ (National Institutes of Health, USA). Protein levels from the Western blots were quantified and normalized relative to a-tubulin. Bars represent the mean SD from four independent experiments. Immunofluorescence staining for VIM (green) or E-cadherin (green) and (red) in a panel of or HCC cells. The level of mrna in breast cancer cell lines. Values were normalized relative to 8S transcription. The plotted data are the average from three independent experiments. Bars represent the mean SD. Stabilities of the transcript and protein in or HCC cells. Cells were treated with lg/ml actinomycin D for,, 6, 9,, and h, after which total RNA and proteins were extracted. c-myc and 8S were used as controls. The data represent three individual experiments (m: mrna, P: protein). ª 6 The Authors EMBO reports Vol 7 No9 6

4 EMBO reports Dynamic chromatin loops control EMT in cancer Jiyeon Yun et al A Relative expression.8.6 Breast cancer cells Gastric cancer cells SNU6 SNU6.8.6 C. 9. shr# shr#... ZEB CDH ZEB FN TWIST SNAIL B SNU6 SNU shr# shr# SNU6 E-cadherin Vimentin ZEB CDH ZEB FN TWIST SNAIL α-tubulin D _ VS Rad-KD SNU6_ VS Rad-KD ITGA TGFB TGFB ITGA E SNU6 TGFB F mrna expression mrna expression TGFB HCC7 MB HCC MB7 E M ITGA HCC7 MB HCC MB7 E M shr# shr# SNU6 6 ITGA shr# shr# X^(-)..8.6 X^(-).8.6 Figure. 6 EMBO reports Vol 7 No9 6 ª 6 The Authors

5 Jiyeon Yun et al Dynamic chromatin loops control EMT in cancer EMBO reports Figure. Reduction of induces TGFB and ITGA transcription. A qpcr was conducted to measure the expression of. The indicated cells were stably transduced with control GFP-shRNA or two different -specific shrna (shr# or shr#, day ). Data are presented as the mean SD from three independent experiments. P <.. Statistical significance was validated by Student s t-tests. B Results of the Western blot analysis for, E-cadherin, and VIM in KD-, -, -SNU6, and -SNU6 cells are presented. C qpcr was performed to measure the expression of EMT markers ZEB, CDH, ZEB, FN, TWIST, and SNAIL in KD- (upper graph) or -SNU6 cells (lower graph). Data are presented as the mean SD from three independent experiments. P <., P <.. Statistical significance was validated by Student s t-tests. D Up-regulated genes by KD (shr#) in or SNU6 cells, each of which showed more than a twofold change in KD cells compared to the. E Induction of TGFB and ITGA transcription in KD-, -, -SNU6, and -SNU6 cells was evaluated using qpcr. Data are presented as the mean SD from three independent experiments. P <.. Statistical significance was validated by Student s t-tests. F Expression of TGFB and ITGA mrna normalized to the 8S transcript in luminal and basal-like breast cancer cell lines, E: epithelial-like, M: mesenchymal-like breast cancer cell lines. Data are presented as the mean SD from three independent experiments. expression (Figs B and EVC). We also evaluated the expression of mesenchymal markers such as ZEB, ZEB, N-cadherin (CDH), fibronectin (FN), TWIST, and SNAIL (Slug) [7] to confirm the induction of EMT by depletion in and SNU6 cells (Fig C). As shown in Fig C, the expression of most EMT markers was significantly increased in the -knockdown (KD) cells although the expression of all EMT markers was not simultaneously induced in either types of cells. Based on these results, we investigated whether tightly regulated EMT plasticity in cancer cells. To identify the target genes of which trigger EMT, we analyzed microarray data by comparing rol () and KD (shr#) in and SNU6 cells (Fig D). We found that 9 and 88 genes were up-regulated (fold change > ) in KD- and -SNU6 cells, respectively (Datasets EV and EV). Next, we applied The Functional Annotation Clustering tool in Database for Annotation, Visualization, and Integrated Discovery (DAVID) ( to analyze enriched annotations on the up-regulated genes. As shown in Tables EV and EV, the functional annotation clustering analysis of each of the 97 and 67 DAVID IDs generated a total of 78 and 9 functional clusters (EASE score <.) using high classification stringency. The significant gene annotations included regulation of cell motion, regulation of cell migration, cell-substrate junction, and adherens junctions, suggesting that the cell movement and migration processes are the major processes induced by -depletion. Among the up-regulated genes, 9 were commonly increased in both KD cell lines. Using the functional annotation chart analysis of these genes, we found that the genes, included in the significant gene annotations (P-value <.), were only 9 genes: TGFB, ITGA, CAV, PRLR, LGALS, SEMAC, WLS, FERMT, and TGFBI. TGFB is known to be a key regulator of EMT in cancer progression, development, and fibrosis [8,9]. In vitro studies have shown that treating various epithelial cells with TGFB initiates EMT [9,], and it was reported that TGFB cross talks with other pathways that lead to EMT, especially ITGA []. ITGA encodes the integrin alpha- chain and forms a hetero-dimeric chain with one of various beta chains to form integrins, which act as receptors and transport environmental signals to cells by binding to the extracellular matrix (ECM) proteins []. A previous study showed that during the EMT process, ITGA expression is induced in transformed epithelial cells []. Therefore, among the commonly up-regulated genes, we explored whether the induced TGFB and ITGA genes led to EMT gene expression signatures in by KD-epithelial cells. To confirm the microarray data, we measured the mrna expression of TGFB and ITGA in KD-, -, -SNU6 and -SNU6 cells (Fig E). Consistent with the microarray data, quantitative real-time PCR (qpcr) results showed that both TGFB and ITGA mrna expression were induced in -depleted epithelial cells. We wondered if this result was simply an effect of deprivation in epithelial cells or if it was an outcome of long-term stable knockdown using lentiviral-shrna vectors. To address this question, we transiently depleted mrna using -specific sirna ( nm) (Fig EVD). Although the induction of TGFB and ITGA expression was more modest compared to stable knockdown of (shown in Fig E), we observed an increase of TGFB and ITGA mrna following transient depletion. Furthermore, CAV and PRLR, which were up-regulated in both KD- and -SNU6 cells, were induced (Fig EVD). These results suggested that the expression changes in the microarray data were not due to the long-term culture of the KD cells alone. The expression of TGFB and ITGA was induced along with transcription changes of EMT markers in KD epithelial cancer cells. These two factors were highly expressed in mesenchymal breast cancer cells compared to epithelial breast cancer cell lines (Fig F). Next, to determine which gene was more susceptible or responded more rapidly to KD, we transiently depleted in a time-dependent manner using nm sirna (Fig EVE). ITGA transcription was induced in the cells h after transfection and the induced levels lasted for over 8 h. Following the induction of ITGA expression, TGFB mrna levels increased only at 8 h after si transfection, suggesting that ITGA was more sensitive to changes in levels than TGFB. Taken together, our results indicated that depletion in epithelial cancer cells leads to a notable induction of EMT marker expression along with increased ITGA and TGFB transcriptions. Induced TGFB and ITGA expression leads directly to cellular morphological changes and acquisition of mesenchymal properties in -depleted cells As EMT is known to acquire the ability of cell movement, it usually occurs when the cell has undergone morphogenesis []. As mentioned above, in vitro experiments showed that treating different epithelial cells with TGFB promotes the acquisition of a clear fibroblast-like phenotype characterized by a loss of epithelial traits and gain of mesenchymal features []. ITGA knockdown also conferred an epithelial phenotype []. This led us to question whether TGFB and ITGA expression induced by KD in epithelial cancer cells could cause cellular or morphological conversion into an EMT-like phenotype with an elongated spindle shape. ª 6 The Authors EMBO reports Vol 7 No9 6 7

6 EMBO reports Dynamic chromatin loops control EMT in cancer Jiyeon Yun et al To address this issue, we first observed changes in KD cell morphology using a microscope (Figs A and EVA). With knockdown, the epithelial-like cells (represented as ) acquired a fibroblast-like appearance (represented as shr#) that correlated with EMT initiation. We also observed that the induction of VIM and b-catenin expression occurred along with nuclear accumulation of b-catenin in KD cells compared to control cells (Fig B and C). According to a recent study that showed TGFB activates canonical Wnt signaling [], we speculated that the results of this study suggested that the TGFB induced by KD might lead to the accumulation of b-catenin into the nucleus via the canonical Wnt signaling pathway, accelerating EMT-related gene activation. Similar to the epithelial and mesenchymal breast cancer cells shown in Fig B, reduction of E-cadherin levels and induction of VIM expression occurred with the transition from an epithelial to mesenchymal morphology in -depleted epithelial breast cancer cells (Fig EVB). Consistent with the transition from an epithelial to mesenchymal morphology, the expression of other EMT-related molecules, especially those involved in the TGFB- related signal pathway including TGFBR and TGFBR that are overexpressed in subpopulations with CSC features (CD + ) and metastatic tumors [6,7], along with b-catenin was significantly induced in KD cells (Figs D and EVC). During EMT, cancer cells gradually acquire a migration ability to metastasize to other sites [8]. To evaluate cell motility in addition to morphological changes, we performed a wound-healing assay. KD- cells migrated more rapidly toward the wound sites than control cells (Fig E). As mentioned above, it has been proposed that TGFB promotes and initiates EMT in cancer cells [9]. Additionally, another previous study showed that ITGA expression is induced through the TGFB intracellular Ca + signaling pathway in osteoblasts [9]. Contrary to this, others have demonstrated that ITGA activates TGFB signaling to initiate EMT [ ]. Similarly, many studies have argued that TGFB and integrins are associated and closely cross talk with each other in cells to modulate cellular metastasis [,]. Given these data, we wondered if the TGFB and ITGA expression induced by -depletion directly initiated the EMT process or if the expression of one or both was simply a result of the acquisition of EMT properties via depletion in epithelial cancer cells. Therefore, we knocked down either TGFB or ITGA in KD epithelial cancer cells with elevated TGFB and ITGA expression (Fig F I). Transient knockdown of TGFB in KD- cells with EMT traits showed that the levels of mesenchymal markers, such as VIM, ZEB, CDH, and FN, which were up-regulated in KD- cells, were significantly reduced with ITGA down-regulation (Fig F), while the epithelial marker CDH was unaffected. Although loss of E-cadherin expression is considered a critical marker of EMT plasticity, some studies have shown that mere depletion of E-cadherin cannot fully promote EMT, demonstrating that E-cadherin is not the sole pivotal molecule for EMT initiation [,6]. Similarly, ITGA depletion in KD- cells notably reduced not only the expression of TGFB but also that of other EMT markers (Fig G), indicating that TGFB and ITGA expression promoted by KD individually and cooperatively affected the expression of each other as well as the EMT process in epithelial cancer cells. These findings suggested that TGFB and ITGA are mutually regulated by each other rather than TGFB expression only being regulated by ITGA Overall, our results demonstrated that the induction of TGFB and ITGA expression in depleted epithelial cancer cells directly promoted EMT and conferred mesenchymal-associated traits. Gene-specific intrachromosomal architecture in the TGFB and ITGA genes is dynamically regulated by cohesin We next ultimately investigated how KD significantly induced TGFB and ITGA expression. To determine whether was involved in chromatin organization, we evaluated the physical interactions between and the TGFB or ITGA gene. We also explored the relationship between the levels of enrichment and gene transcription using a ChIP assay. For the ChIP assay, we used mesenchymal breast cancer HCC cells, which express high levels of both TGFB and ITGA, and epithelial breast cancer cells, in which TGFB and ITGA are expressed at low levels (Figs F and EVA). in the cells was strongly enriched on the TGFB promoter (indicated by amplicons, 6, and 7 of TGFB) and the far-upstream region from the ITGA gene promoter (indicated by amplicon 6 of ITGA). In contrast, weakly bound to the genes in mesenchymal HCC cells, suggesting that the enrichment of on the genes was negatively correlated with gene transcription levels. We also found that the strong binding of to both the TGFB and ITGA genes was significantly reduced with KD in epithelial and SNU6 cells (Fig A), implying that the induction of TGFB and ITGA transcription was inversely correlated with enrichment in the genes., one of the cohesin complex subunits, physically regulates the formation of the chromatin loop structure to control the gene-specific transcriptional environments [7]. Therefore, we speculated that on the gene might modulate gene-specific Figure. Disrupted causes morphological changes from epithelial-like to fibroblast-like shapes. A Imaging of KD- and -SNU6 cells to assess cell morphology. Depletion of caused a conversion from an epithelial-like to fibroblast-like morphology. B, C Immunofluorescence staining for (B) VIM and (C) b-catenin in KD-SNU6 cells. The indicated cells were immunostained with antibody specific for VIM or b-catenin, and DAPI. Each image represents three individual experiments. D Western blots for EMT markers TGFBR, TGFBR, TGFB, ITGA, and b-catenin in KD- and -SNU6 cells. E Stable transfection of shrna in cancer cells increased migration capabilities revealed by a wound-healing assay. Wound size was measured at,, and 8 h to calculate open wound rate (%). Data are presented as the means SD from three independent experiments. P <., P <.. Statistical significance was validated by Student s t-tests. F I qpcr was performed to measure the expression of EMT markers VIM, ZEB, CDH, ZEB, and FN along with the epithelial marker CDH in (F, G) shr#_ or (H, I) shr#_snu6 cells after transfection with nm negative control, TGFB (sitgfb# and sitgfb#) or ITGA (siitga# and siitga#) sirna for 8 h. Values were normalized relative to 8S transcription. Gene expression values in the samples were divided by those in the controls (i.e., each gene in the control is ). Data are presented as the means SD from three independent experiments. P <., P <.. Statistical significance was validated by Student s t-tests. 8 EMBO reports Vol 7 No9 6 ª 6 The Authors

7 Jiyeon Yun et al Dynamic chromatin loops control EMT in cancer EMBO reports A SNU6 B. μm. μm shr# C. μm shr#. μm shr#_ shr#_snu6 Vimentin DAPI shr#_ shr#_snu6 Merge β-catenin DAPI Merge D SNU6 E shr# _shr# h h h TGFBR TGFBR TGFB ITGA β-catenin h 8 h h 8 h h 8 h % of Open Wound 8 6 h h 8h α-tubulin shr# shr# F G _shr# sicont sitgfb# sitgfb# sicont siitga# siitga# H I SNU6_shR# sicont sitgfb# sitgfb# sicont siitga# siitga# Figure. ª 6 The Authors EMBO reports Vol 7 No9 6 9

8 EMBO reports Dynamic chromatin loops control EMT in cancer Jiyeon Yun et al CCDC97 Kb TGFB Kb ITGA CTCF binding sites A B C Pol AcH/H Enrichment Enrichment Enrichmen nt Amplicon TGFB.8.6 SNU TGFB ITGA SNU6 6 ITGA shr# _IgG shr#_igg D CCDC97 TGFB E ITGA MCF F7 SNU 6 Interaction fr requency Interaction fre equency cohesin Interaction fre equency Interaction fre equency shr# shr# Figure. EMBO reports Vol 7 No9 6 ª 6 The Authors

9 Jiyeon Yun et al Dynamic chromatin loops control EMT in cancer EMBO reports Figure. Released chromatin loop structure in TGFB and ITGA genes by KD leads to the enhancement of active transcriptional environment on the gene promoter. A C TGFB and ITGA loci on chromosome 9q. and chromosome q., respectively. The location of putative CTCF binding sites and representative amplicon sites used for qpcr is shown with names below. A ChIP assay was performed for KD- or -SNU6 cells using antibodies specific for (A), (B) Pol II, (C) AcH, or IgG (as a control). Enrichment was measured by qpcr and expressed relative to the total input (%). ChIP signals for AcH were normalized to total H. Results for at least three chromatin preparations are shown SEM. P <., P <.. Statistical significance was validated by Student s t-tests. D, E Relative cross-linking frequencies among CTCF/ binding sites on the genes were measured with a C assay in the control GFP-shRNA (blue line) or KD (red line)- or -SNU6 cells on day after lentiviral transduction. (D) BamHI or (E) XbaI restriction sites on the TGFB or ITGA gene, respectively, appear as gray shaded bars. Black shading indicates the anchor fragment. Each value was normalized to cross-linking frequency at the ERCC gene. The maximum crosslinking frequency was set at (means SEM, n = ). P <., P <.. Statistical significance was validated by Student s t-tests. chromatin architecture to regulate transcriptional activity of the gene. Thus, a chromatin conformation capture (C) assay for the TGFB and ITGA genes was performed [8] using KD- or -SNU6 cells. Based on the ChIP data showing a high enrichment of (Fig A), amplicon TGFB_6 and amplicon ITGA_6 were selected as the anchors for the TGFB and ITGA genes, respectively, in KD- or -SNU6 cells (Fig D and E). The results showed that strong binding of formed gene-specific chromatin interactions within the TGFB or ITGA gene in - or - SNU6 cells (blue line, Fig D and E). Consistent with the loss of binding on the genes due to KD shown in Fig A, we also found that gene-specific chromatin interactions were disrupted by KD in both and SNU6 cells (red line, Fig D and E). Similarly, another anchor (amplicon TGFB_ and amplicon ITGA_) also formed a chromatin loop structure that was disrupted on the TGFB and ITGA genes by KD (Fig EVD and E). We next determined why the gene-specific chromatin architecture was related to gene transcription levels. A ChIP assay was performed to analyze four active transcription markers: RNA polymerase II (Pol II), Ser-phosphorylated RNA polymerase II (SerP), acetyl-h (AcH), and HK9ac in - and KD- or -SNU6 cells (Figs B and C, and EVB and C). In both - and -SNU6 cells, low enrichment of Pol II, SerP, AcH, and HK9ac was detected on both genes. On the other hand, KD significantly induced both factors to bind to the promoters of the genes. These findings suggested that the release of gene-specific intrachromosomal architecture shown in Fig D and E might induce enrichment of these active transcriptional marks to the gene promoters, establishing active transcriptional environments. Taken together, our results showed that the gene-specific chromatin architecture in TGFB and ITGA affected by were closely associated with transcriptional activities. High-chromatin interactions on the genes indicated low expression of the genes. In contrast, low-chromatin interaction on the genes signified high expression (Fig D and E). Cancer stem-like breast cancer cells show a released higher-order chromatin structure of TGFB- and ITGA-mediated by, which correlated to the high transcriptional expressions and mesenchymal property Since it has been reported that most types of tumor overexpress [9 ], we speculated that only small population of tumor burdens with a low expression of might be responsible for cancer metastasis to a second metastatic site. To verify this, we analyzed transcript expression data obtained from The Cancer Genome Atlas (TCGA). A total of 8 breast tumor tissues and matched normal samples, including all subtypes, were analyzed to measure the transcript levels of, CDH, VIM, TGFB, ITGA, ZEB/, and SNAI/ (Fig EVA). The results revealed that the mrna levels in the tumor samples were higher than those in normal cells although we could not assess the protein expression levels. Moreover, the tumor samples showed obvious epithelial traits, consistent with our data showing the relationship between expression and the EMT (Fig EVA). In contradiction of our hypothesis, relatively high levels of TGFB mrna expression were observed in the tumor samples relative to normal samples. Even though it is known that TGFB has multiple functions, it primarily acts as a tumor suppressor via antiproliferative activity in normal cells. Furthermore, this factor contributes to the differentiation, proliferation, and migration of aggressive tumor cells [,]. These findings suggested that the high levels of TGFB expression in tumors with high expression, independent from transcriptional up-regulation caused by reduced involved in tumor metastasis, might be due to a proliferative effect of TGFB on tumor cells. This was supported by an absence of significant changes in ITGA mrna levels when comparing normal and tumor samples (P =.768) (Fig EVA). Since primary tumors are composed of heterogeneous tumor cells, there might be certain cells with initial migratory properties. However, we could not easily detect this type of population due to their small portion like cancer stem cells (CSCs). To understand this, we expanded our model to include CSCs since EMT has been shown to be highly linked to CSC [,7,]. While traditional models of tumor initiation suggest that cancer cells arise from environmental factors or genetic alterations, then gradually acquire aggressive properties, the CSC hypothesis argues that cancer development is attributed to CSCs which consist of a small population of heterogeneous tumors that can be responsible for tumor proliferation and initiation []. The most important feature of CSCs has been reported to be the characteristics of EMTs that allow cell dissemination, invasion, and migration. Therefore, we used MDA-MB- breast cancer cell derived cancer stem-like cells (CSLCs) [6,7] positive for CD/CD/ CD7 expression, which are well-known markers of CSCs. Consistent with previous studies [,6,8], high levels of mesenchymal-related genes including ZEB, FN, TWIST, VIM, TGFBR, TGFBR, and b-catenin were observed in the CSLCs, indicating that these cells had significant mesenchymal properties (Fig A and B). However, we unexpectedly discovered high levels of E-cadherin in the CSLCs (Fig A and B), suggesting that it can serve as an evidence that E-cadherin alone is not sufficient for EMT initiation, and that CSCs have partial EMT properties. To study whether CSLC cells had mesenchymal properties such as the ª 6 The Authors EMBO reports Vol 7 No9 6

10 EMBO reports Dynamic chromatin loops control EMT in cancer Jiyeon Yun et al A D B Fold chan nge Fold change ZEB P CSC MDA-MB- TGFB ITGA 8 FN 6 P CSC TWIST 9 6 P MDA-MB- CSC Interaction frequency Interaction frequency CCDC97 P CSC TGFB ITGA P CSC E-cadherin Vimentin ITGA TGFBR TGFBR E CCDC97 TGFB ITGA a-tubulin β-catenin C Figure. Parental Number of migrated cells/field Parental CSLC CSLC AcH/H Pol Enrichmen nt richment En Enrichment P -CSC -P_IgG -CSC_IgG TGFB ITGA migration ability, we performed a transwell assay. As shown in Fig C, when compared with the parental MDA-MB- cells, the CSLC cells showed significantly higher migration ability. We next checked and compared the levels of expression between MDM-MB- CSLCs and the parental cells. As shown in Fig A and B, the expression of both mrna and protein in EMBO reports Vol 7 No9 6 ª 6 The Authors

11 Jiyeon Yun et al Dynamic chromatin loops control EMT in cancer EMBO reports Figure. is responsible for metastatic characteristics in CSLCs. A, B and EMT-related molecules in parent MDA-MB- breast cancer cells (P) and the corresponding CSLCs (indicated by CSC or CSLC) sorted according to the CD + cancer stem cell markers were observed by (A) qpcr and (B) Western blot analysis. Data are presented as the mean SD from three independent experiments. P <.. Statistical significance was validated by Student s t-tests. C Cell migration assay of MDA-MB- parental and CSCL cells using Transwell chamber without Matrigel. Graphs indicate the average number of cells per field. Bars represent the mean SD for three independent experiments. P <.. Statistical significance was validated by Student s t-tests. D Relative cross-linking frequencies among CTCF/ binding sites on the genes were measured with a C assay in the parental cells (blue line) or CSLCs (red line). BamHI or XbaI restriction sites on the TGFB or ITGA gene, respectively, appear as gray shaded bars. Black shading indicates the anchor fragment. Each value was normalized to the cross-linking frequency at the ERCC gene. The maximum cross-linking frequency was set at (mean SEM, n = ). P <., P <.. Statistical significance was validated by Student s t-tests. E A ChIP assay was performed for the CSLCs (red bar) or parental cells (blue bar) using antibodies specific for, AcH, Pol II, or IgG (as a control). Enrichment was measured by qpcr and reported relative to the total input (%). ChIP signals for AcH were normalized to total H. Results for at least three chromatin preparations are shown SEM. P <., P <.. Statistical significance was validated by Student s t-tests. the CSLCs was reduced to less than.-fold compared to the levels found in the parental cells. Although the mrna levels of per se were lower in the CSLCs compared to the parental cells, which was unlike the patterns of mrna expressions found in mesenchymal and epithelial breast cancer cell lines (Fig C), the mrna in the CSLCs was relatively unstable compared to that in the parental cells. These findings (Fig EVB and C) were consistent with the results presented in Figs D and EVE. In addition to a low level of expression, we also observed high expression of TGFB and ITGA mrna in the CSLCs, corresponding to results for the KD cells. To determine whether high levels of TGFB and ITGA expression found in the CSLCs formed its chromatin architecture as in KD epithelial cells (Fig D and E), we conducted a C assay using the CSLCs and parental cells (Fig D). In CSLCs with high expression of TGFB and ITGA, lower interaction frequencies were observed within both the TGFB and ITGA genes compared to those in the parental cells. To determine whether the gene-specific chromatin architectures were directly mediated by and regulated gene transcription, we assessed the enrichment of, AcH, and Pol II on the genes using a ChIP assay (Fig E). Consistent with the C data, binding on the genes was lower in the CSLCs than the parental cells. Conversely, the markers corresponding to transcriptional activity, AcH and Pol II were notably enriched on the promoter of the genes, consistent with gene transcription levels, indicating that the CSLCs might also be regulated and maintained by TGFB and ITGA gene-specific intrachromosomal interactions. The loose chromatin loop structures in each gene easily recruit transcription factors and enhance the enrichment of active histone marks on the gene promoter to activate gene transcriptions like KD epithelial cancer cells. Collectively, these results strongly support our hypothesis that cohesin-mediated threedimensional chromatin structures are responsible for regulating and maintaining EMT plasticity in a small population of tumors or cancer stem cell model by controlling transcriptional activities of EMT-related genes associated with metastasis, thereby conferring migratory and invasive properties. Overexpression of in mesenchymal breast cancer cells reduces TGFB and ITGA expression, concomitant with enhancement of intrachromosomal interaction on the gene To confirm the function of on the transcriptional regulation of TGFB and ITGA through modulating the transcriptionally repressive gene-specific chromatin architecture in epithelial cancer cells, we overexpressed in MDA-MB- mesenchymal breast cancer cells and MDA-MB- CSLC cells with a low level of expression (Figs 6A D and EVD F, respectively). Although the transient overexpression of did not appear to be sufficient in MDA-MB- and CSLC cells, we clearly found that the overexpressed in mesenchymal cancer cells reduced TGFB and ITGA gene transcriptions (Figs 6A and B). Furthermore, most of the EMT markers that were induced in the KD-epithelial cells, such as VIM, ZEB, and SNAI, were decreased following overexpression. These gene expression changes induced by overexpression also led to the induction of a direct molecule binding and the decreased Pol II and HK9ac on the TGFB and ITGA genes (Figs 6C and EVF). Using a C assay, we determined that the induction of binding on each TGFB and ITGA genes enhanced the intrachromosomal interaction in each genes. Taken together, these results suggest that TGFB and ITGA gene transcription in cancer cells might depend on its three-dimensional chromatin structure mediated by cohesin complex, playing a crucial role in the epithelial to mesenchymal transition and mesenchymal to epithelial transition. During tumorigenesis, individual tumor cells responsible for metastasis might accurately regulate the transcriptional activity of upstream EMT-related genes such as TGFB and ITGA, which are thought to be responsible for EMT initiation, through dynamic changes in higher-order chromatin loop structures mediated by the cohesin complex (Fig 6E). Discussion In the current study, we addressed the role of the cohesin complex in the dynamic transition of epithelial to mesenchymal cancer cells by altering distinct chromatin interactions of TGFB and ITGA genes and inducing their transcriptional activities. These two factors are considered upstream molecules that influence the EMT in epithelial cancer cells and CSCs. We analyzed transcript expression data for primary breast tumor and normal tissue samples from the TCGA database. Interestingly, higher expression of was observed in the tumor tissues than the normal samples. Since primary samples or samples of the so-called metastatic tumor obtained by biopsy are already differentiated and have lost mesenchymal properties [,9], and the tumor samples might have a high level of expression similar to epithelial cancer cell lines. This suggests that only a small proportion of tumors or CSCs might play an important role in metastasis. ª 6 The Authors EMBO reports Vol 7 No9 6

12 EMBO reports Dynamic chromatin loops control EMT in cancer Jiyeon Yun et al A mrna expression X^(-). X^(-).8.6 X^(-). 6 TGFB ITGA.8. X^(-) X^(-) VIM ZEB ZEB SNAI X^(-) X^(-) CDH X^(-) X^(-). TWIST C HK9ac/H CCDC Mock Rad Mock Rad Enrichment Enrichment TGFB ITGA B D CCDC97 TGFB ITGA MDA-MB- Actin Mock Overexpressed Interaction frequency.8.6 P R E Epithelial Cancer Cells EMT Rad Mesenchymal Cancer Cells or Cancer Stem Cells? EMT-related genes Pol TGFB ITGA Inhibits the transcription of EMT-related genes Overexpression? Low transcriptional activity Figure 6. Rad MET MORPHOLOGICAL CHANGE Knockdown EMT-related genes Pol Ac High transcriptional activity EMBO reports Vol 7 No9 6 ª 6 The Authors

13 Jiyeon Yun et al Dynamic chromatin loops control EMT in cancer EMBO reports Figure 6. overexpression in MDA-MB- mesenchymal cancer cells decreases TGFB and ITGA expression with enhanced intrachromosomal interactions. A The relative mrna expression levels of EMT-related genes in -overexpressing MDA-MB- cells were quantified and normalized by 8S transcript. Empty vector-expressing MDA-MB- control cells are indicated by _P; -expressing MDA-MB- cells are indicated by _R. The data shown are the mean SD from three independent experiments. P <., P <.. Statistical significance was validated by Student s t-tests. B Western blots for in -overexpressing MDA-MB- cells. C A ChIP assay was performed for -overexpressing MDA-MB- and the empty vector-expressing cells using antibodies specific for, Pol II, and HK9ac, or IgG (as a control). Enrichment was measured by qpcr and expressed relative to the total input (%). ChIP signals for HK9ac were normalized to total H. Results for at least three chromatin preparations are shown SEM. P <., P <.. Statistical significance was validated by Student s t-tests. D Relative cross-linking frequencies among CTCF/ binding sites on the genes were measured with a C assay in the control (blue line) or -overexpressing (red line) MDA-MB- cells on day after lentiviral transduction. BamHI or XbaI restriction sites on the TGFB or ITGA gene, respectively, appear as gray shaded bars. Black shading indicates the anchor fragment. Each value was normalized to cross-linking frequency at the ERCC gene. The maximum cross-linking frequency was set at (means SEM, n = ). P <., P <.. Statistical significance was validated by Student s t-tests. E A proposed model of cohesin-mediated dynamic chromatin architecture of the TGFB and ITGA genes associated with EMT plasticity. Tumor cells that have undergone EMT are characterized by proposed chromatin architectures. In epithelial cancer cells, cohesin tightly mediates distant intrachromosomal interactions in the TGFB and ITGA genes. When mrna stability is lost, bound on the genes is reduced and intrachromosomal interactions in the genes are released, causing active recruitment of the transcriptional complex to the gene promoters. This enhances gene transcription and the cells acquire mesenchymal properties such as enhanced mobility and morphological changes. Differences in expression between cells with epithelial or mesenchymal traits were due to variations in mrna stability. However, we could not determine how the stability of mrna is regulated or the stage of tumorigenesis and mechanisms associated with the loss of mrna stability. Therefore, further study is required to understand the underlying mechanisms that govern dynamic during tumorigenesis. However, we can propose that chromatin dynamics mediated by the higher-order chromatin structural molecule cohesin actively confer epithelial or mesenchymal properties depending on the surrounding environment or cell fate. Interestingly, we observed significant induction of TGFB and ITGA expression immediately after knockdown in and SNU6 cells (Fig EVE). TGF-b is known to be a potent inducer of EMT in mammary cells that increases stem-like properties in human breast cancer cells [7,6]. ITGA is expressed during tumor development and has been implicated in EMT induction. This may explain why highly invasive cancer cells have higher levels of ITGA and overexpression of this factor increases the metastatic capacity [6,6]. Based on the data shown in Fig F and G, we postulated that either TGFB or ITGA expression induced by depletion might directly result in the induction of mesenchymal-related genes, most of which are well-known repressors of CDH, but not the epithelial-related gene CDH itself. Up-regulation of these genes clearly led to a morphological transition from epithelial to mesenchymal phenotypes and provided the cells with motility and invasive capabilities (Fig ). Interestingly, we observed a relatively high level of E-cadherin in MDA-MB-- derived CSLCs compared to the parental cells (Fig B). As previously mentioned, this finding suggests that E-cadherin may not be the sole molecule responsible for EMT initiation without orchestrating regulations or changes in other EMT-related factors or MDA-MB- derived CSLCs may not be completely transformed into mesenchymal and invasive cells. These results also suggest that changes in mesenchymal-related gene expressions are more susceptible to alterations in the expression of upstream factors for EMT initiation (such as TGFB and ITGA) than epithelial-related genes. We demonstrated that the expression of TGFB and ITGA closely relies on the gene-specific three-dimensional chromatin structure mediated by cohesin (Figs D and E, and EVD and E). Not surprisingly, enrichment of on the genes affected the formation of gene-specific chromatin loop structures. While strong binding of on the gene forms strong chromatin interactions within the genes, low enrichment of leads to a loss of the interactions. Interestingly, gene-specific chromatin interactions in CSLCs were very weak compared to those in the parental cells, despite the persistent expression in the CSCLs (Fig D). This significant difference in the chromatin interaction for TGFB and ITGA between CSLCs and the parental cells largely affected expression of the genes, suggesting that transcriptions related to the EMT gene may be dependent on chromatin interactions. According to recent studies, cohesin can bind to enhancer of genes, thereby stabilizing long-range enhancer promoter interactions [6,6]. Moreover, it binds to insulator of genes to prevent the spread of repressive transcription from near the transcriptional environment. Similarly, many other studies have shown that the major role of cohesin in chromatin architecture for gene transcriptional regulation is to activate the gene transcription (e.g., b-globin, interferon c locus, and H9-IGF). However, other studies have shown that the perturbation of cohesin complex causes up-regulation of gene transcriptions [7]. Consistent with our results, several studies have shown that knockdown of leads to significant recruitment of Ser-Pol, causing the activation of gene transcription [6]. Taken together, these findings demonstrate that inactivated TGFB and ITGA genes are maintained by the cohesin-mediated chromatin loop structure in epithelial cells, and release of the chromatin interactions by a loss of cohesin complex binding results in the genes becoming transcriptionally activated via active recruitment of transcriptional machineries to the gene promoter and formation of an active transcriptional environment. Based on this view, depending on where the chromatin interactions are formed, chromatin loop structures are categorized into five distinct patterns with transcriptional activity (Pol II) and several histone modification signatures [66]. One of these patterns features chromatin signatures of active characteristics with enrichment of Pol II, HKme, HK6me, or AcH, and depleted repressive markers such as HK9, and HK7 methylation inside the loops. Conversely, another pattern features chromatin loops with extensive methylation of HK9 and HK7 but loss of active markers within. As shown in Figs B and C, and EVB and C, gene-specific ª 6 The Authors EMBO reports Vol 7 No9 6