p63 Attenuates Epithelial to Mesenchymal Potential in an Experimental Prostate Cell Model

The transcription factor p63 is central for epithelial homeostasis and development. In our model of epithelial to mesenchymal transition (EMT) in human prostate cells, p63 was one of the most down-regulated transcription factors during EMT. We therefore investigated the role of p63 in EMT. Over-expression of the predominant epithelial isoform ΔNp63α in mesenchymal type cells of the model led to gain of several epithelial characteristics without resulting in a complete mesenchymal to epithelial transition (MET). This was corroborated by a reciprocal effect when p63 was knocked down in epithelial EP156T cells. Global gene expression analyses showed that ΔNp63α induced gene modules involved in both cell-to-cell and cell-to-extracellular-matrix junctions in mesenchymal type cells. Genome-wide analysis of p63 binding sites using ChIP-seq analyses confirmed binding of p63 to regulatory areas of genes associated with cell adhesion in prostate epithelial cells. DH1 and ZEB1 are two elemental factors in the control of EMT. Over-expression and knock-down of these factors, respectively, were not sufficient alone or in combination with ΔNp63α to reverse completely the mesenchymal phenotype. The partial reversion of epithelial to mesenchymal transition might reflect the ability of ΔNp63α, as a key co-ordinator of several epithelial gene expression modules, to reduce epithelial to mesenchymal plasticity (EMP). The utility of ΔNp63α expression and the potential of reduced EMP in order to counteract metastasis warrant further investigation.


Introduction
A family of transcription factors is constituted by p53, p63 and p73. Unlike p53, which is expressed in response to environmental stress, p63 (TP63) is constitutively expressed at high levels in a variety of epithelial tissues including prostate. There are two different promoters leading to two N-terminal variants DNand TAp63 [1] and five reported splice variants of the C-terminal (a, b, c, d and e) [2]. Originally, only the TA isoforms were considered to be transcriptional activators, but the consensus is now that DNp63 isoforms that carry a truncated aminoterminal domain have transcriptional activity not only restricted to the TAp63 controlled genes [1].
Mice with p63 knock-out fail to develop a prostate, highlighting p63's important role in development. In adult tissues p63 is normally expressed in basal cells of the prostate and other stratified epithelia, but the expression is lost in more differentiated luminal cells. In prostate cancer cells p63 is typically undetectable and this is the basis for routine immunohistochemistry diagnostics of this cancer type. In benign prostate hyperplasia p63 positive cells are still numerous, leading to high cancer sensitivity for negative p63 staining of suspected prostate cancers [3]. Despite these insights, the role of p63 in differentiation of prostate cells is poorly understood, although mouse studies have shown that secretory cells in the prostate gland are derived from previously p63 positive cells [4]. There are also many unresolved questions regarding the cell of origin of prostate cancer and the role of p63, with some reports suggesting p63-positive basal cells [5] and others Nkx3-1positive luminal cells [6] as the cell of origin. Additionally, the DNp63a isoform has been implicated as an oncogene in several types of squamous cell carcinomas, while other lines of evidence suggest that p63 functions in a tumor suppressive manner [1].
The epithelial to mesenchymal transition (EMT) is a preserved cellular program used extensively during embryogenesis. It has also been shown to be utilized in the adult organism notably in the processes of fibrosis and carcinogenesis. In carcinogenesis EMT is implicated in gain of invasiveness and escape from the primary site, an early event in metastasis [7].
We have previously published that the immortalized primary epithelial cells, EP156T, underwent EMT during long-term culture and selection for cells with loss of cell contact inhibition [8]. The EPT1 and EPT2 cells that were derived from EP156T cells following EMT have maintained their mesenchymal like morphology for more than 100 subsequent sub-confluent passages during the last 4 years [9].
EMT can be induced by forced expression of certain classical mesenchymal transcription factors such as ZEBs, SNAIs and TWISTs or knock-down of E-Cadherin [10]. Much less investigaton has been done regarding putative epithelial transcription factors, but recently transcription factors responsible for maintaining and inducing the epithelial state, such as GRHL2, ELF3 and ELF5 have been acknowledged [11]. Recently, p63 has been shown to play a role in EMT of breast cells [12]. In the present work we have investigated the potential of p63 as an epithelial transcription factor in prostate cells. We found that p63 is lost during EMT of prostate EP156T cells and that re-expression of its predominant isoform DNp63a in the mesenchymal type cells re-introduces several epithelial characteristics, suggesting an important role during EMT.

DNp63a is the predominant isoform in EP156T cells and is shut down during EMT
We have previously published that EMT is one part of the stepwise transformation model based on human primary immortalized, epithelial prostate cells [8,9]. Genome-wide gene expression analyses revealed that p63 was one of the most downregulated transcription factors when epithelial EP156T cells underwent EMT to become EPT1 cells. This led us to question if p63 has a direct role in the process of EMT and its reversibility in prostate cells. To further investigate the role of p63, we performed isoform-specific reverse transcriptase quantitative PCR (RT-qPCR) of EP156T cells. This demonstrated that DNp63a was the predominant isoform ( Figure 1A), in line with earlier reports showing that DNp63a is the most expressed isoform in basal prostate cells [13].
Based on these data we decided to express DNp63a in EPT1, EPT2 and EPT1B8 cells, the latter is a clone of the published EPT1 cells. The DNp63a gene was recombined into a retroviral vector ( Figure S1B) and transduced into EPT mesenchymal type cells, lacking detectable endogenous p63. Re-expression of p63 to levels comparable to expression in EP156T cells was verified using RT qPCR and Western blot analyses, while indirect immunofluorescence analysis of p63 showed nuclear staining (shown for EPT1B8 in Figure 1B, 1C, 1D). The possible mesenchymal-toepithelial transition (MET) was assayed using whole-genome gene expression microarray data and evaluation of cell morphology.
Re-expression of DNp63a did not lead to a complete reversion of morphological features comparable to those seen in EP156T in monolayer cultures, but the mesenchymal EPT cells re-expressing DNp63a were markedly changed from the mock transduced cells, particularly at confluent growth, as the DNp63a transduced cells failed to organize like long spindle-shaped cells, but organized themselves towards an epithelial cobblestone pattern ( Figure 1E). Western blot analysis of E-Cadherin and vimentin expression is often used as a marker of either an epithelial or mesenchymal state. EPT1B8 cells re-expressing DNp63a did not show any altered E-Cadherin or vimentin protein expression compared to EPT1B8 cells ( Figure 1F).

Re-expression of DNp63a in mesenchymal cells induces expression of genes involved in cell-to-cell junctions and extracellular matrix (ECM) contacts
We previously published that multiple gene module switches were observed when EP156T cells underwent EMT and became EPT1 cells during long-term confluent culture, including cell junction, cell polarity, cytoskeleton, receptor tyrosine kinase, transcription factor and microRNA modules [8] ( Figure S1C). This model therefore offers an excellent opportunity to examine the requirements for reversal of EMT, i.e. to examine if single master regulators have the power to co-ordinate the entire EMT/ MET program or whether reversal of selected modules can create intermediate stages between the epithelial and the mesenchymal states. To investigate if DNp63a can coordinate gene modules relevant to EMT, we performed genome-wide microarray gene expression analysis of EPT1B8 cells transduced with DNp63a compared with mock-transduced cells. Significance Analysis of Microarray (SAM) showed that 163 genes were down-and 581 genes were up-regulated at least 2 fold at the highest False Discovery Rate (FDR ,5%). Still, the number of genes differentially expressed in EPT1B8 cells after DNp63a reexpression were 744 compared to 5229 genes differentially expressed after EMT from EP156T cells to EPT1B8 cells, i.e. 14.2% of the genes are re-expressed in mesenchymal EPT1B8 DNp63a re-expressing cells ( Figure 2A). 398 genes were shared between the groups, showing highly significant (p,0.0001, Chisquare test) enrichment of genes regulated by DNp63a reexpression among differentially expressed genes during EMT. We then wanted to see if DNp63a regulates specific gene modules. For this purpose we evaluated gene ontology terms and the differential expression of their constituent genes based on microarray findings. 72 of the 737 genes in the biological adhesion group (GO:0022610) that were represented on the array were differentially regulated compared to a total of 744 differentially expressed genes out of 25988 genes in the dataset (p,0.0001, Fisher's exact test). The same relation was found for ''cell junction'', ''gap junction'', ''desmosome'', ''adherens junction'' and ''cell-cell junction'' (Table 1). Using RT qPCR we verified the differential regulation of several cell adhesion genes based on the microarray data. Although multiple cell adhesion genes were upregulated, the expression levels did not reach that of EP156T cells ( Figure 2B), as also confirmed at the protein level ( Figure 2C).
To functionally examine the possibility that DNp63a expression affects cell adhesion we grew the cells on an array of ECM components and found stronger binding to ECM components for DNp63a re-expressing cells although only differential binding to collagen IV was statistically significant ( Figure 2D).
DNp63a re-expression reverses several phenotypic traits associated with EMT We further characterized the EPT1B8DNp63a cells concerning traits associated with epithelial cells. Following EMT, cells have increased migratory ability. We therefore assessed the ability of EPT1B8 cells to migrate in a Boyden chamber assay after DNp63a re-expression. Indeed, the EPT1B8DNp63a cells had a markedly and statistically significant (p,0.0001, Student's t-test) decreased ability to migrate ( Figure 3A). We also investigated if the ability to invade was altered, by a similar assay where the cells had to pass a pore with a membrane of extracellular matrix (ECM). The EPT1B8 cells re-expressing DNp63a did, however, not show any significant difference of invasive behavior ( Figure S2A). p63 has previously been implicated in resistance to anoikis in mammary epithelial cells [14], we therefore sought to find out if p63 Attenuates EMT in Prostate Cells PLOS ONE | www.plosone.org DNp63a plays a similar role in prostate cells. EPT1B8DNp63a and EPT1B8mock cells were seeded in wells covered with a hydrogel coating, which inhibits adhesion to the substrate, after 48 hours we observed a statistically significant higher amount of living cells among DNp63a expressing cells based on the MTT proliferation assay. Cells growing on a normal substrate served as control. Here DNp63a expressing cells yielded the same values as the control cells, thus ruling out that difference in proliferation could cause the observed difference in the anoikis assay ( Figure  3B). Increased resistance to apoptosis is another trait associated with EMT. When cells were treated with the apoptosis-inducing agent staurosporine we could not detect any significant difference in the amount of activated caspase 3/7, as a marker of apoptosis ( Figure S2B). Cells with a mesenchymal phenotype are characterized by actin stress fibers as opposed to epithelial cells that display a cortical actin organization. DNp63a re-expression in EPT1B8 cells (and EPT1 and EPT2) did not reverse the stress fibers observed in mesenchymal cells ( Figure S2C).
As p63 has been implicated in carcinogenesis we also wanted to investigate the effect of p63 on growth in soft agar, which has been considered an in vitro transformation assay [15]. DNp63a has been found to be over-expressed in several carcinomas [1]. EPT1 cells have already gained several traits associated with cancer cells as compared to the epithelial EP156T cells [8], and we further wanted to explore if expression of DNp63a could endow EPT1B8 cells with the ability to grow anchorage independently, or abolish the ability of EPT2 cells to do so [9]. Compared to EPT2 cells that    have this ability, DNp63a did not provide EPT1B8 cells with this ability. EPT2DNp63a cells did exhibit a small but significantly decreased ability to grow in soft-agar ( Figure 3C).

Knock-down of p63 in EP156T cells
To complement over-expression studies of DNp63a in EPT1B8 cells we wanted to see if abrogation of p63 expression could induce EMT or reverse selected epithelial features in EP156T cells that contain high levels of endogenous p63. We transduced EP156T cells with shRNA particles targeting three different regions of the DNA-binding domain of p63, thereby targeting all isoforms. Knock-down was confirmed by RT qPCR and Western blot assays ( Figures S3A, S3B).
Following p63 knock-down we could not observe any alterations in cell morphology in monolayer cultures (data not shown), but when analyzed by genome-wide gene expression analysis we found more genes differentially expressed than we found after DNp63a re-expression in EPT1B8 cells. 1408 genes were down-regulated and 1220 genes up-regulated at least 2 fold at a FDR ,5%. Several of the cell adhesion gene modules were significantly differentially expressed, for example ''Cell Junction'', ''Focal adhesion'' and ''Tight Junction'' (Table S1). We also validated differentially expressed adhesion genes by RT qPCR ( Figure S3C). The ECM adhesion assay showed that EP156T p63 knock-down cells had significantly decreased binding to fibronectin, laminin and collagen IV while collagen I also exhibited decreased binding (p = 0.05, Student's t-test), demonstrating complimentary results compared to the DNp63a over-expression in EPT1B8 cells ( Figure  2D).
We then performed phenotypic assays on the EP156T cells with p63 knock-down as for the p63 re-expression in EPT1B8 cells. We did not find any significant changes in the ability to migrate ( Figure S3D), but the EP156T p63 knock-down cells had a small, but statistically significant decreased ability to invade compared to EP156T cells ( Figure S3E). The EP156T p63 knock-down cells also displayed a significantly decreased sensibility to the apoptosisinducing agent staurosporine ( Figure S3F). We also observed a significant decrease in MTT-based proliferation activity of EP156T p63 knock-down cells compared to control when grown on a hydrogel coating, suggesting increased anoikis-sensibility, although this could also be the result of higher proliferation as the MTT value was higher for vector-transduced EP156T cells ( Figure  S3G).
p63 is able to co-ordinate induction of cell junction modules in mesenchymal prostate cells In EPT1B8 cells the re-expression of DNp63a led to the induction of multiple cell junction components. The co-ordinated gene expression induction is exemplified by hemidesmosome components, including the extracellular LAMC2, its integrins ITGA6 and ITGB4 and its associated cytoplasmic cytokeratins, such as KRT5 ( Figure 2B). It is interesting to know if these genes are direct target genes of p63 in order to understand how p63 can execute a co-ordinated gene expression program in both the epithelial and mesenchymal contexts. Carroll et al. used ChIP-PCR to demonstrate direct binding of p63 in the promoters of ITGA6, ITGB4 and LAMC2 in breast epithelial cells [14].
In order to identify p63 binding genes, we performed genomewide profiling of p63 binding in our prostate cell model, EP156T cells using the ChIP-seq technology [16,17]. ChIP-grade anti-p63 antibody was used for chromatin immunoprecipitation followed by deep sequencing. In the end, we obtained 20,511,592 raw reads from the ChIP with anti-p63 antibody and 19,455,454 raw reads from ChIP with the mouse IgG antibody. The sequence tags were mapped to the human genome (hg19) with the bowtie algorithm [18] using default parameters. After mapping, we obtained 14,868,166 and 13,494,179 uniquely mapped reads. We used MACS (model based analysis for ChIP-seq) [19] to find p63 binding peaks and identified 7,021 peaks. Table S2 lists the 7021 peaks in BED format. We used CisGenome [20] to add gene annotations associated with the peaks if a peak is within 50kb from the TSS (transcription start site) of a gene. We were able to annotate 2,487 peaks within 50 kb from TSS of genes. Comparison with a genome-wide p63 ChIP-seq study in human keratinocytes (HFK) [21] showed 46.9% overlap (Table S3) between the two analyses. We also performed functional annotations of the p63 peaks associated genes using DAVID (http://david.abcc.ncifcrf.gov/) [22,23] for KEGG pathways and Gene Ontology (GO) terms. We found that the GO term cytoskeletal protein binding (GO:0008092) is significantly enriched in the p63 binding targets with FDR ,5% (Table 2) with 82 genes related to the cytoskeletal protein binding containing p63 binding sites (Table S4). Other significant enriched biological processes include epidermis development, angiogenesis, ectoderm development and regulation of cell motion. There are 39 p63 targeted genes that are related to regulation of cell motion (Table S5), which could explain p63's effect on cell migration ( Figure 3A) as we observed.
In order to identify consensus binding motifs for p63 binding in prostate cells, we employed the MEME web server (http://meme. nbcr.net/meme/). We only chose those p63 peaks with FDR equal to zero. After MEME analysis, we identified a consensus motif, which is the same as previously reported for primary human keratinocytes [24]. This motif ( Figure 4A) was found to exist in 351 of the 526 selected peaks (FDR = 0) and has a p-value of 1.5e-717.
Among 4028 genes changed at least 2 fold between EP156T and EPT1 cells in a t-test, 366 genes (9.1%) contained significant p63 peaks in EP156T cells. SAM analysis showed that 70 of the 366 genes (19%) were re-induced at least 2-fold with an FDR ,5% in EPT1DNp63a cells compared to EPT1 mock transduced cells (EPT1mock), including LAMC2 and ITGA6, and both are p63 targets (Table S3). Additional analyses of genome-wide gene expression data files of both EPT1B8 and EPT2 cells transduced with DNp63a and parallel mock-transduced cells showed that DNp63a induced re-expression of 57 of the 366 genes (16%) in EPT1B8DNp63a cells and 72 of the 366 genes (20%) in EPT2DNp63a cells (Table S6, S7, S8). Figure 4B shows the p63 binding peaks found in putative regulatory regions of LAMC2.
Finally, Gene Ontology (GO) analyses were performed on genes that were up-regulated after re-expression of DNp63a in EPT1B8, EPT1 and EPT2 and that were in the group of the 366 genes defined above. These analyses clearly showed that terms associated

Examination of the potential of E-cadherin to contribute to MET in the present model
In several published EMT models, re-expression of E-cadherin (CDH1) is sufficient to reverse EMT (induce MET) alone or together with co-factors [25]. Interestingly, CDH1 was among the cell adhesion genes that seemed not to be induced by DNp63a expression in EPT1B8 cells. As published previously, CDH1 was down-regulated more than 10 fold, but was still easily detectable by Western blot in EPT1 cells following EMT of EP156T cells [8]. In EPT2 cells, derived by their ability to form foci in monolayers and to grow in soft agar in contrast to EPT1 cells, CDH1 production was not detectable [9]. Examination of the clone EPT1B8 showed substantial and variable endogenous expression of CDH1. But this residual endogenous expression of CDH1 was evidently not sufficient in addition to exogenous DNp63a expression in order to induce a complete MET. For this reason CDH1-expression was increased to similar levels in EPT1 cells as in EP156T cells by lentiviral transduction of the CDH1 gene. Although both RT qPCR and Western blot analyses verified that CDH1 expression in EPT1 cells reached comparable levels to those detected in EP156T cells ( Figure S4A, S4B), neither CDH1 expression alone nor co-expression with DNp63a resulted in more complete MET induction than the one obtained by isolated DNp63a re-expression (data not shown). The same panels of cell junction genes were induced in both EPT1 cells and EPT2 cells following re-expression of DNp63a as in the EPT1B8 subclone (Table S10, S11).
The role of ZEB1 knock-down in EMT regulation in prostate cells.
ZEB transcription factors are known to repress CDH1 expression following direct binding to E-boxes in the CDH1 promoter [26]. In turn, ZEB1 is targeted by the microRNA 200 family [27] and by miR-205 [28,29] to constitute reciprocal feedback loops. Examination of our p63 ChIP-seq data files revealed peaks corresponding to p63 binding sites within the miR-205 regulatory regions ( Figure 4C), suggesting that p63 might induce miR-205 expression. We analysed our microRNA profiling data files of EP156T, EPT1 and EPT1DNp63a cells (Table S12). Pronounced down-regulation of all miR-200 family and miR-205 was observed following EMT. In particular, miR-205 was abundantly expressed in EP156T cells, but undetectable in EPT1 cells. A significant induction of miR-205 was observed in EPT1DNp63a cells, although to a much lower level than in EP156T cells.
When EP156T cells underwent EMT to become EPT1 cells, the expression of several classical EMT associated transcription factors, in particular ZEB1, ZEB2, SNAI1 and TWIST2, were strongly increased [8]. In order to examine the MET-inducing potential in our model ZEB1 was knocked down in EPT1B8 and in EPT1B8DNp63a cells. Gene expression profiling showed 3.1 fold and 2.6 fold decrease in ZEB1 expression in EPT1B8 and in EPT1B8DNp63a cells, respectively, according to RT qPCR validation of microarray results. The knock-down of ZEB1 was associated with an increase of CDH1, ITGB4 and LAMC2 ( Figure  S4C). ZEB1 knock-down in EPT1B8 cells was additionally associated with specific re-expression of miR-141 and miR-200c, but neither with re-expression of other miR-200 family members nor with re-expression of miR-205 (Table S12), consistent with previous studies in different models [30]. Following ZEB1 knock-down miR-141 and miR-200c were found to be re-expressed in EPT1B8 cells on the average to 14.4% and 13.2%, respectively, of EP156T expression levels ( Figure S4D). Still a complete MET based upon cell morphology criteria, was not achieved.

Epigenetic patterns of DNp63a target gene promoters
We have previously published that extensive epigenetic alterations occurred when EP156T cells underwent EMT to become EPT1 cells [31]. As we observed that DNp63a re-expression to cell-physiological level was not sufficient to make EPT1B8 cells undergo an EMT and that several of the p63 target genes involved in cell adhesion did not reach expression levels seen in EP156T cells, we speculated that there was an epigenetic restriction reducing the effect of p63 on these promoters. Taking advantage of this dataset, we compared the pattern of transcriptionally stimulatory and inhibitory epigenetic marks on selected p63 target gene promoters. LAMC2 and ITGB4 displayed H3K4me3 and H3K27me3 patterns in EPT1 cells similar to that observed in EP156T cells suggesting that the genes are open for transcription also in EPT1 cells. Still, p63 re-expression could not bring the expression of these genes to levels observed in EP156T cells. KRT5 and CDH3 have a more repressive H3K4me3/H3K27me3 status in EPT1 cells and interestingly these genes also were clearly upregulated in EPT1 cells ( Figure 4D).

Discussion
We have examined the role of p63 in EMT and its reversal, MET, since p63 was the most down-regulated transcription factor during EMT in our cell culture model of step-wise accumulation of malignant features in immortalized primary prostate cells. Although loss of p63 was a significant event during EMT, questions remain as to whether it has a functional role in this process or if the expressional changes are secondary to other processes. In order to further elucidate this, we performed an overexpression experiment of the predominant epithelial isoform DNp63a in mesenchymal EPT1B8, EPT1 and EPT2 cells and knock-down of p63 in epithelial EP156T cells.
Re-expression of DNp63a in EPT1B8, EPT1 and EPT2 mesenchymal like cells and knock-down of p63 in EP156T cells led to widespread and concerted gene expression changes, in particular concerning cell adhesion associated genes. Interestingly, this applies to both cell-to-cell adhesion genes and to genes involved in cell-to-extracellular-matrix (ECM) contact. Earlier studies have shown that p63 plays an important role in regulating cell-adhesion networks in other epithelia [14,32]. We here show that p63 is an important regulator of such genes in benign immortalized prostate epithelial cells, with ChIP-seq data corroborating our gene expression findings by defining genes directly regulated by p63 promoter binding. Additionally we show that p63 has the ability to regulate these genes in prostate cells with mesenchymal features indicating that the epigenetic context in mesenchymal prostate cells is permissive to p63 binding or that DNp63a itself can alter the epigenetic permissibility and result in subsequent p63-dependent gene regulation. We also found evidence of a functional role as EPT1B8 cells' ability to bind certain ECM components was increased after DNp63a reexpression. Additionally, when we knocked down p63 expression in EP156T cells we found that adhesion to the same ECM components decreased. Furthermore, we observed that cell migration was strongly inhibited by forced DNp63a expression in EPT1B8 cells, suggesting loss of a mesenchymal trait.
Taking into account that forced expression of DNp63a reaches cell-physiological levels and is similar to the p63 expression level p63 Attenuates EMT in Prostate Cells PLOS ONE | www.plosone.org seen in EP156T cells, it is striking that the p63 target genes and miR-205 are up-regulated but do not reach the expression levels observed in epithelial EP156T cells, indicating a quantitative rather than qualitative restriction of target gene expression. The limitation of target gene expression is underscored by the observation that neither expression/co-expression of physiological levels of E-cadherin nor complementing knock-down of ZEB1 was sufficient for complete MET induction in EPT cells. The reason for the limited expression of p63-induced cell adhesion and miR-205 genes is unclear. In fact, potential positive feedback circuits can be envisaged according to which p63-mediated induction of miR-205 would lead to ZEB1 reduction (as observed) that next could be expected to alleviate ZEB1 mediated repression of the p63 gene promoter [28]. Among additional reported targets of miR-205, however, is a group of de-ubiquitinating enzymes whose inhibition is associated with increased ubiquitination and proteasomal degradation of p63 [28]. It is conceivable that a balance between these positive feedback and negative feedback mechanisms (or additional undefined ones) could lock DNp63a expressing mesenchymal cells at an intermediate stage of MET, such as is observed in our experimental model.
One alternative explanation of quantitative limitation of p63 target gene expression in EPT cells could be related to changes in the epigenetic promoter patterns. Indeed, we have previously shown that EMT results in widespread epigenetic changes between epithelial EP156T cells and mesenchymal EPT1 and EPT2 cells and that these changes are correlated to changes in gene expression, for example for a number of cell adhesion genes [31]. There is limited knowledge concerning which epigenetic backgrounds are favorable to EMT/MET reprogramming, but evidence for this connection has recently emerged also in the field of induced pluripotency [33,34].
p63 has also been implicated in control of EMT/MET in other studies. Recently, Coppola et al. showed that RWPE-1 cells, that are heterogeneous immortalized prostate epithelial cells and display both basal and luminal characteristics, gained mesenchymal features including increased vimentin and decreased Ecadherin expression, when all p63 isoforms where knocked down. In RWPE-2 cells that are more mesenchymal, forced DNp63 expression led to decreased vimentin expression [35]. This study suggests that p63 is important to maintain the epithelial phenotype in prostate epithelial cells and that over-expression of DNp63 isoforms can reverse mesenchymal features supporting our findings. Even though p63 knock-down led to lower levels of CDH1 (data not shown) we did not observe gain of a mesenchymal phenotype suggesting that EP156T cells are more refractory to induction of mesenchymal features. Another p63 isoform, DNp63c, has been implicated in induction of EMT in human mammary cells [12]. In cells devoid of DNp63a and DNp63b, DNp63c expression was sufficient to induce EMT in a TGFb-dependent manner, through DNp63c induced transcription of TGFb-isoforms. This finding does not seem to apply to our prostate cells as strong re-expression of DNp63a did not induce MET, suggesting DNp63c plays a minor role in maintaining the EMT phenotype of the EPT1B8 cells and that p63 isoform function is highly context-dependent.
As we have different passages of EPT1 and EPT1B8 cells with different CDH1 expression levels as well as EPT1 cells with CDH1 over-expression alone and in combination with DNp63a, we have a good model to investigate the role of CDH1 in MET. In contrast to other studies [25], CDH1 expression levels did not seem to alter the phenotype further toward an epithelial state, as well as showing that the effect of p63 on cell adhesion genes is CDH1 independent.
To conclude, DNp63a re-expression in EPT1B8 cells leads to several functional and expressional changes indicating a switch towards an epithelial phenotype, but lack of re-expression of central epithelial markers and of a clear epithelial morphology suggest that DNp63a is only able to induce a partial MET in EPT1B8 cells. Knock-down of p63 in epithelial EP156T cells further supports these findings. Additionally, we show that p63 is functional in a mesenchymal context (i.e. EPT1B8, EPT1 and EPT2 cells) and that this is independent of CDH1 expression. These data suggest that EMT and MET seem to be the extremes of a continuum of states and that certain phenotypes associated with EMT/MET can be controlled without committing fully to this switch.

Cell Lines
The prostate cell lines EP156T [36], EPT1, EPT1B8 and EPT2 were grown in MCDB153 medium (Biological Ind. Ltd., Israel) with 1% for EP156T and 5% fetal calf serum (FCS) for the others, and supplemented with growth factors and antibiotics as described elsewhere [8,36]. DNA microsatellite validation of progeny identity of EP156T, EPT1 and EPT2 cells has been published previously [9]. EPT1B8 is a clone obtained from EPT1 cells by limiting dilution to single cells in 96-well plates. The cells were grown in tissue culture plates in a humidified atmosphere containing 5% CO2 at 37uC. The medium was changed every third day.

Cloning and Transduction
DNp63a-FLAG in a pcDNA3 vector was obtained from L. Ellisen, Harvard. The L335 N792 attSfipolylinkIREStdTom plasmid was digested with XhoI and blunt ends were generated using T4 DNA polymerase (Fermentas Cat# EP0061) and dephosphorylated with calf intestine alkaline phosphatase (Fermentas Cat# EF0341). pcDNA3DNp63a was excised using BamHI and XhoI, and blunted as above. Ligation was performed using HC DNA Ligase (New England Biolabs, Cat# M0202M). Correct insertion was verified with restriction analysis using SpeI (New England Biolobs, Cat# R0133S) and sequencing. Pseudotyped retroviral particles were prepared in HEK293 Phoenix cells. Following retroviral transduction Fluorescence Activated Cell Sorting (FACS) was used to select cells expressing exogenous genes.
For p63 knock-down, a pool of three shRNA lentiviral particles (Open Biosystems, clone IDs: V3LHS_397883, V3LHS_397884 and V3LHS_397885) were used, and similarly for ZEB1 knockdown (Open Biosystems, clone IDs: V3LHS_356184, V3LHS_356186 and V3LHS_356187), viral particles with clone ID RHS-4348 were used as negative control. Cells were transduced with a MOI of 1, and subsequently sorted by FACS using the turboGFP reporter in the inserted sequence. The CDH1 gene in a Gateway pDONR vector (Cat# GC-M0954, GeneCopoeia, Rockville, MD, USA) was cloned into pLenti7.3/V5-DEST Gateway Vector Kit (Invitrogen, Carlsbad, CA, USA). The vector p63 Attenuates EMT in Prostate Cells PLOS ONE | www.plosone.org was then sequenced to confirm correct insertion. EPT1 and EPT1DNp63a cells were transduced according to the manufacturer's protocol and cells expressing CDH1 were selected using FACS based on EmGFP expression.

Real-time quantitative PCR (qPCR)
Total RNA was extracted using the miRNeasy kit from Qiagen (Cat# 217004). The total RNA was DNase treated, ss-cDNA was synthesized and the RT qPCR was run and analyzed as previously described [37], using pre-designed Taqman probes (Table S13).

Global Gene Expression Analysis
The Agilent Human Whole Genome (4644 k) Oligo Microarray with Sure Print Technology (Agilent Technologies, Design G4112-60520 G4845-60510, Palo Alto, CA, USA), was used to analyze samples in the present study. Total RNA purification, cDNA labeling, hybridization and normalization have been described previously [37,38]. Following normalization, significance analysis of microarray (SAM) of the J-Express program package (http://www.molmine.com) [39] was used for identification of differentially expressed genes. Only genes that changed at least 2.0 fold with FDR below 5% were considered as differentially expressed genes in cell lines. For GO analysis, cell adhesionrelated GO terms were searched for and a Fischer's exact test was performed to obtain nominal p-values.

Micro-RNA profiling
Total RNA was extracted using the miRNeasy kit from Qiagen (Cat# 217004). The miRNA expression was profiled by Toray (Tokyo, Japan) platforms. According to the Toray protocol, 500 ng total RNA was labeled using the miRCURY LNA microRNA Hy5 Power labeling kit (Cat# 208030-A, Exiqon) and hybridized to the 3D-Gene TM Human miRNA oligo chip. The microarray was spin-dried followed by image scanning using the 3D-Gene TM scanner at 635 nm excitation. The obtained images were numerated by 3D-Gene TM Extraction. The detected spots were defined as the ones that had the signal intensity over 2SD plus mean of background signal intensity. The mean of the background-subtracted signal intensity of duplicate spots was used for the further analysis.

Chromatin immunoprecipitation and ChIP-chip
ChIP was performed according to the Agilent ChIP-chip protocol with modifications as previously described [31]. To immunoprecipitate chromatin, 6610 7 cells were treated with 1% formaldehyde at room temperature for 10 min followed by quenching with 0.125 M glycine. Cells were lysed and the nuclei were sonicated under conditions yielding DNA fragments ranging from 200 to 800 base pairs. Five percent of the sonicated material was saved as whole cell extract. Sonicated lysate was divided into three equal volumes and immunoprecipitated with specific or nonspecific antibody bound to protein A magnetic beads (Invitrogen) overnight at 4uC with rocking. Antibodies used were against: H3K4me3 (Abcam, Cat# ab8580) or H3K27me3 (Abcam, Cat# ab6002) or mouse IgG (Sigma). Five mg of antibody was used per 2610 7 cells. Immunoprecipitated complexes were collected, washed and eluted using the Dynal Magnetic Particle Concentrator (Invitrogen). Eluted DNA and whole-cell extracts were incubated at 65uC in a rotating incubator for 8 hours to reverse cross-links. DNA samples were sequentially treated with RNase A and proteinase K and then purified by phenol/chloroform extraction. The immunoprecipitated (ChIP'ed) and purified DNA was ethanol precipitated using glycogen as a carrier and resuspended in nucleic acid grade water. Analysis of promoter H3K4me3/H3K27me3 ratio was performed as previously described [31].

ChIP sequencing (ChIP-seq)
The Chromatin immunoprecipitation was essentially as described for ChIP-chip above. For immunoprecipitation 5 mg of ChIP-grade p63 antibody (Abcam, Cat# ab3239) was used per 2610 7 cells.

Indirect immunofluorescence (IF) and Western blot assays
For IF cells were grown on 10 mm Assistant glass coverslips in 24 well plates, then washed with PBS, fixed (4% fresh formaldehyde in PBS for 20 min. at room temperature), permeabilized (0.5% Triton X-100 for 10 min.), blocked (100 mM glycin for 10 min) and stored (in PBS at 4uC) with PBS wash between each step. Following blocking with 0.5% BSA/PBS for 15 min. primary mouse monoclonal antibodies were added at room temperature for 1hour at indicated dilutions in 0.5% BSA/PBS. The FITClabelled secondary anti-mouse Ig (SouthernBiotech, Birmingham, AL, USA, Cat# 4050-02) was added for 1 h at room temperature in 0.5% BSA/PBS. Coverslips were mounted in Prolong Gold with DAPI (Molecular Probes, Invitrogen, Carlsbad, CA, USA) on glass slides and analysed using Leica TCS SP5 confocal microscopy or Leica DM IRBE fluorescence microscopy.

Migration and invasion assay
For trans-well migration assays and invasion assays, 1.8610 5 cells starved overnight in serum-free medium were added to the p63 Attenuates EMT in Prostate Cells PLOS ONE | www.plosone.org top chambers of 24-well trans-well plates (8 mm size, Cat# CBA-100-C, Cell Biolabs Inc.) and media containing 10% fetal calf serum (FCS) were added to the bottom chambers. Incubation was for 12 h in the migration assay and 24 h in the invasion assay. Top (non-migrating) cells were removed, bottom (migrating) cells were stained and absorbance was recorded at 560 nm. All of these assays were done in at least triplicate and the data are presented as the average absorbance of migrating cells and invading cells 6s.d.
Cell adhesion, anoikis and apoptosis assay 1.0610 5 cells were seeded in each well of a 48 well plate, incubated at 37uC and 5% CO 2 for 90 minutes. The cells were then washed and treated according to the manufacturer's protocol. CytoSelect TM 48-Well Cell Adhesion Assay (ECM Array, Colorimetric Format, Cat# CBA-070, Cell Biolabs Inc., San Diego, CA, USA). Anoikis was assayed by seeding 20.000 EPT1B8 cells and 10.000 EP156T cells per well in a 96 well plate covered with hydrogel and a in normal control plate. The cells were incubated for 48 h before the wells were assayed by MTT solution for viability according to the manufacturer's protocol. CytoSelectTM 96-Well Anoikis Assay (Cat# CBA-081, Cell Biolabs Inc., San Diego, CA, USA), Apoptosis was determined using the Caspase-Glo 3/7 Assay (G8090, Promega, Madison, WI, USA) according to the manufacturer's protocol. Briefly, 10.000 EPT1B8 cells and 5000 EP156T cells were plated in 96-well plates and incubated for 12 h followed by staurosporine (Sigma, S6942) treatment at indicated concentrations for 6 h at 37uC to induce apoptosis. Equal volumes of Caspase-Glo 3/7 Reagent was added directly to each well and the plates were incubated at room temperature for 1 h before recording luminescence in a Synergy H1 reader (BioTek Instruments Inc., Bad Friedrichshall, Germany).

Anchorage-independent growth assay
The anchorage independent growth was examined in soft agar. 50 ml of base agar matrix (CytoSelect TM Cell, Colorimetric kit, Cat# CBA-135, Cell Biolabs Inc., San Diego, CA, USA) was added in the bottom of each well of a 96-well plate. When the agar was solid, 75 ml of cell suspension/soft agar matrix containing 2500 cells was layered on top followed by 50 ml of complete medium. After 8 days of incubation, the agar matrix was solubilized and cells stained and absorbance was recorded at 570 nm. Data show the quantification of proliferation of EPT1B8 cells and positive control EPT2 cells in the soft agar assay.
Accession numbers of gene expression, ChIP-chip and ChIP-seq data Performed as described previously [31]. Annotated microarray data were uploaded in the BASE database, formatted and exported to ArrayExpress (http://www.ebi.ac.uk/arrayexpress) at the European Bioinformatics Institute in agreement with the MIAME guidelines, accession number: E-MTAB-1471. The ChIP-chip study has ArrayExpress accession numbers: E-TABM-635; E-TABM-983 [31]. The ChIP-seq data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus [40] and are accessible through GEO Series accession number GSE43111 (http://www.ncbi.nlm.nih.gov/geo/query/ acc.cgi?acc = GSE43111). Figure S1 (A) Schematic overview of p63 isoforms adopted from Mangiulli et.al. [2]. There are two different 5' variants DN and TA which can combine with five 3' variants a,b,c,d and e, giving 10 isoforms all together. (B) Schematic overview of the construct used to over-express DNp63a in EPT cells. (C) Schematic figure of the cell culture model. EP156T cells are immortalized cells derived from primary benign prostate epithelial cells. EPT1 cells are EP156T cells that have undergone an EMT [8]. EPT2 cells have acquired ability to grow anchorage independently and are derived from EPT1 cells [9]. (TIFF) Figure S2 (A) Invasion of EPT1B8 DNp63a and EPT1B8 as measured by invasion through a Boyden chamber inserted with extracellular matrix. Student's t-test was used for statistical analysis (p = 0.098). (B) Induction of apoptosis in EPT1B8 DNp63a and EPT1B8 by staurosporine measured by caspase 3/7 activity. Error bars show 6s.d (C) Immunofluorescence by phalloidin staining actin filaments in EPT1B8 DNp63a and EPT1B8 cells.    Table S6 366 genes that were differentially expressed (.2 fold between EP156T and EPT1) and had significant p63 peaks in EP156T found by ChIP-seq analysis, were compared to differentially expressed genes (.2-fold) between EPT1DNp63a and EPT1mock cells (Table S6). Genes that were in both groups were analysed by functional annotation by DAVID (http://david.abcc.ncifcrf.gov/). (XLS)

Supporting Information
Table S7 366 genes that were differentially expressed (.2 fold between EP156T and EPT1) and had significant p63 peaks in EP156T found by ChIP-seq analysis, were compared to differentially expressed genes (.2-fold) between EPT1B8DNp63a and EPT1B8mock. Genes that were in both groups were analysed by functional annotation by DAVID (http://david.abcc.ncifcrf.gov/).

(XLS)
Table S8 366 genes that were differentially expressed (.2 fold between EP156T and EPT1) and had significant p63 peaks in EP156T found by ChIP-seq analysis, were compared to differentially expressed genes (.2-fold) between EPT2DNp63a and EPT2mock. Genes that were in both groups were analysed by functional annotation by DAVID (http://david.abcc.ncifcrf.gov/). (XLS)