Epigenetic Mechanisms Underlying the Dynamic Expression of Cancer-Testis Genes, PAGE2, -2B and SPANX-B, during Mesenchymal-to-Epithelial Transition

Cancer-testis (CT) genes are expressed in various cancers but not in normal tissues other than in cells of the germline. Although DNA demethylation of promoter-proximal CpGs of CT genes is linked to their expression in cancer, the mechanisms leading to demethylation are unknown. To elucidate such mechanisms we chose to study the Caco-2 colorectal cancer cell line during the course of its spontaneous differentiation in vitro, as we found CT genes, in particular PAGE2, -2B and SPANX-B, to be up-regulated during this process. Differentiation of these cells resulted in a mesenchymal-to-epithelial transition (MET) as evidenced by the gain of epithelial markers CDX2, Claudin-4 and E-cadherin, and a concomitant loss of mesenchymal markers Vimentin, Fibronectin-1 and Transgelin. PAGE2 and SPAN-X up-regulation was accompanied by an increase in Ten-eleven translocation-2 (TET2) expression and cytosine 5-hydroxymethylation as well as the disassociation of heterochromatin protein 1 and the polycomb repressive complex 2 protein EZH2 from promoter-proximal regions of these genes. Reversal of differentiation resulted in down-regulation of PAGE2, -2B and SPANX-B, and induction of epithelial-to-mesenchymal transition (EMT) markers, demonstrating the dynamic nature of CT gene regulation in this model.


Introduction
Cancer-testis (CT) or cancer-germline genes are expressed in tumors originating from various tissues, as well as in normal germline and trophoblast cells, but are generally silent in other normal tissues of the adult [1][2][3]. More than 100 different CT genes can be grouped according to homology into families [2]. Despite the lack of sequence similarity between CT genes from different families, re-expression of all CT genes in tumors has been associated with demethylation of CpG residues within their promoter-proximal regions [4]. This shared mechanism of expression regulation is most likely the reason for their coordinate expression in cancer [5][6][7]. However, the exact mechanism by which DNA demethylation occurs at CT gene promoter-proximal regions in cancers is unknown. CT genes show mostly a heterogeneous expression pattern in tumors [8][9][10]; in contrast to their expression in testis, which is demarcated and orderly [11]. A study in which stem-like and non-stem like cells of breast cancer were selectively killed, revealed that CT gene expression is generally a feature of more differentiated, non-stem cells [12]. Similarly, in melanoma, a subgroup of cells with more epithelial features express CT genes, when cells with mesenchymal features don't [13]. Interestingly, melanoma cells can switch between these two classes in vivo, suggesting that tumor heterogeneity, as defined by CT gene expression, might represent a transitional phase similar to the switch between epithelial and mesenchymal phenotypes. Indeed, mesenchymal-to-epithelial transition (MET) is associated with the induction of CT gene expression [14]. To define mechanisms involved in regulating CT gene expression in cancer and during MET, we chose to study the Caco-2 spontaneous differentiation model which demonstrates features of MET and EMT during differentiation and de-differentiation, respectively. Our data reveal that the dynamic regulation of the two CT genes, PAGE and SPANX-B in this model system, involves alterations of polycomb repressive complex 2 (PRC2) and heterochromatin protein 1 (HP1) occupancy within their promoter-proximal regions, with concordant changes in TET expression and cytosine hydroxymethylation (hmC) levels.

Cell lines, induction of differentiation and de-differentiation
The Caco-2 cell line was obtained from the SAP Enstitusu (Ankara, Turkey). HCT116 (colorectal) and Mahlavu (hepatocellular) cancer cell lines were obtained from LGC Standards,   [15]. To induce dedifferentiation, cells at the 20th day of differentiation were detached and replated at about 50% confluence and RNA and protein were harvested 5 days following replating.
In silico analysis of CT and EMT gene expression Expression data contained within GSE1614 [16] was GC-RMA normalized using GeneSpring v. 11.0. CT gene expression was analyzed based on 31 probesets in corresponding to 23 CT genes from 7 families ( Figure S2). An interpretation was generated with an entity list composed of EMT related genes as defined by Loboda et al. [17], at three different time points. Genes for validation were selected among those for which significant differences of expression (p,0.05) was observed by one way ANOVA test and Bonferroni FWER correction, when proliferating cells were compared to those at day 15.

Promoter methylation analysis
Genomic DNA was isolated by Proteinase K treatment, following a phenol-chloroform extraction protocol. Bisulphite treatment of 200 ng genomic DNA was performed using Zymo DNA Methylation Gold Kit (Zymo Research, Irvine, CA, USA). Bisulphite modified DNA was stored at 220uC and used for PCR within 2 months. Two rounds of DNA amplification were performed using One Taq Hot Start DNA polymerase (New England Bioscience/NEB, Ipswich, MA, USA) using a Perkin Elmer 9700 thermal cycler (Applied Biosystems, Carlsbad, CA, USA). Primers used are given in Table S1. PCR products were gel extracted using the QIAGEN gel extraction kit (Qiagen, Hilden, Germany) and cloned into pCR2.1 (Invitrogen, Carlsbad, CA, USA). Plasmid DNA was purified using the QIAprep Spin Miniprep Kit (Qiagen, Hilden, Germany) from at least ten clones, and sequence analyzed by IONTEK (Istanbul, Turkey).

5-hydroxymethyl cytosine analysis
Caco-2 genomic DNA (gDNA) was sheared by probe sonication (30 sec. on, 30 sec. off, 5 cycles) to obtain 200-600 bp. fragments assessed by 1% agarose gel electrophoretic analysis. Immunoprecipitation was performed using the hMEDIP kit (Abcam, Cambridge, UK) according to manufacturer's instructions. 5 pg of control DNA was spiked into 500 ng of gDNA to use as an internal control. Positive and negative controls of the kit were included in all experiments. 2 ml from the eluted DNA was used as template for quantitative RT-PCR using 2 X SYBR Green master mix with ROX reference dye (Applied Biosystems, Carlsbad, CA, USA) with the primers given in Tables S1 and S2. Primer efficiencies were controlled. Cycling conditions were 50uC for 2 min., 95uC for 10 min. followed by 40 cycles of 94uC for 15 sec., 60uC for 1 min. Shared genomic DNA was included in quantitative RT-PCR to calculate % input.

Immunofluorescence microscopy
Cells attached to glass slides by centrifugation using the Shandon CytoSpin3 (Thermo Scientific, Waltham, MA, USA) were immediately fixed in 2% formaldehyde/PBS at room temperature for 15 min. Fixed cells were permeabilized in 0.2% Triton X-PBS for 10 min. followed by blocking with 1% BSA in 0.1% PBS-Tween for 1 hour. Incubations with the primary antibody, diluted at 1:50, were performed overnight at 4uC. Secondary antibody was added at 1:200, following washing in 0.1% PBS-Tween for 5 min. for 3 times, and incubated with cells for 45 min. at room temperature (see Table S3 for the complete antibody list). Stained samples were mounted with mounting medium (Santa Cruz Biotechnology, Santa Cruz, CA, USA) containing DAPI solution. Cell lines used as positive controls were Mahlavu (PAGE-2,-2B and SPANXB), MDA-MB 231 (VIM), MCF-7 (TAGLN) and SW620 (CDX2, FN1). Negative controls were combinations of primary antibodies with un-related secondary antibodies. All images were obtained using an AxioCam MRc5 image capture device (Carl Zeiss, Oberkochen, Germany).

Chromatin Immunoprecipitation
Chromatin Immunoprecipitation (ChIP) was performed as previously described [15]. Briefly, formaldehyde cross-linked cell constituents were precipitated by proten A sepharose beads coupled to antibodies against EZH2, HP-1 or H3K27m3 (Abcam, Cambridge, UK), as well as isotype-specific control. Precipitated DNA was amplified using primers specific for PAGE2, -2 or SPANX-B promoter sequences (Tables S1 and S2), following de-crosslinking.

CT gene expression during Caco-2 spontaneous differentiation (Caco-2 SD)
The undifferentiated colorectal cancer cell line Caco-2 undergoes enterocytic differentiation upon reaching confluence in vitro [19,20]. Gradual differentiation has been observed up to 30 days post-confluence as evidenced by the up-regulation of various differentiation-associated genes including sucrase-isomaltase, alkaline phosphatase and carcinoembryogenic antigen (CEA), ( Figure  S1) [15]. An in silico analysis of CT gene expression as defined by 31 probesets in the GSE1614 dataset, which contains gene expression data for the Caco-2 SD model obtained during differentiation (proliferating (2 nd day), post-proliferation-undifferentiated (8 th day), and post-proliferation-differentiated (15 th day)) [16], revealed modest up-regulation of almost all CT genes during differentiation ( Figure S2). We chose to validate the change in expression of six CT gene/gene families by quantitative RT-PCR in differentiating Caco-2 cells in vitro. GAGE, MAGE-A3, NY-ESO-1 and SSX4 transcripts were undetectable on the first day of confluence, as well as at later time points (data not shown). However, significant up-regulation of PAGE2 (2 and 2B), and SPANX-B genes was evident (Figure 1).

PAGE2 and SPANX-B expression follow MET in the Caco-2 SD model
Spontaneous differentiation of Caco-2 in vitro has been reported to result in MET [21,22]. To determine if this occurred in parallel to the up-regulation of PAGE2 and SPANX-B, we analyzed the GSE1614 dataset for the expression of genes representing EMT in colorectal cancer [17], and selected 6 genes to be validated in our model. Analysis of mRNA and protein expression of these revealed a decrease in mesenchymal genes (vimentin, fibronectin 1 and transgelin) with a concomitant increase in expression of epithelial genes (CDX2, claudin-4 and E-cadherin) as the cells differentiated, demonstrating that the increase in CT gene expression occurs simultaneously with MET in this model (Figure 1 & Figure S3).

PAGE2, SPANX-B and EMT gene expression in situ
To study if the changes in protein expression of CT and EMT genes occurred simultaneously in the same cells, we performed double immunofluorescence staining during differentiation. A gradual loss of mesenchymal markers was observed as cells differentiated, with a concomitant increase in epithelial genes and CT genes. SPANX-B and PAGE-2 were frequently co-expressed with the epithelial marker CDX2 in the same cells but almost never with VIM or FN1 (Figures 2, 3 and Figures S4, S5, S6, S7).

PAGE2 and SPANX-B expression correlates with increased hmC and ten-eleven translocation methylcytosine dioxygenase (TET) up-regulation
Expression of all CT genes studied thus far including PAGE2 and SPANX-B have been associated with the demethylation of CpG residues within regions proximal to the transcription start site [1,2,[23][24][25][26]. In this line, both PAGE2 and SPANX-B can be upregulated by 5-aza 29-deoxycytidine treatment ( Figure S8). However, bisulfite sequencing of promoter-proximal regions of both PAGE2 and SPANX-B revealed no differences at different stages of Caco-2 SD ( Figure 4). As bisulfite sequencing is unable to distinguish methyl cytosine (mC) from hmC, we asked whether the change in CT gene expression could be related to altered hmC/mC ratios within their promoters. In fact, chromatin immunoprecipitation (ChIP) with a hmC specific antibody revealed an increase in hmC during differentiation in both PAGE2 and SPANX-B2 promoters ( Figure 5). We next asked if the increase in hmC was related to an increase in TET1, -2, and -3 expression as these proteins are responsible for converting mC to hmC [27,28]. Indeed, the increase in hmC of PAGE2 and SPANX-B2 promoters were correlated with an up-regulation of TET2 mRNA expression, together with modest increases in TET1 and 3 ( Figure 6A). Double immunofluorescence staining revealed that the majority of cells expressing PAGE2 or SPANX-B were positive for TET2 staining; indicating these two events occurred in the same cells (Figure 7). It is therefore, likely that the increase in TET2 expression causes increased hmC in these genes. Interestingly, only a low molecular weight translation product (,25 kD) of TET2 was increased in the differentiating cells, when no clear difference in levels of the full-length TET2 protein was observed ( Figure 6B & C). The peptide used for generating the commercial TET2 antibodies could specifically inhibit recognition of the 25 kD product confirming its identity with TET2 (data not shown). TET2 has recently been shown to undergo proteolytic cleavage by calpain 1 and 2, generating a 25 kD product in vitro [29]. To test if a cation-dependent protease is responsible for the generation of the 25 kD protein, we treated differentiating cells with an intracellular Ca 2+ chelator (BAPTA-AM). This resulted in a modest reduction in the 25 kD TET2 protein with the concomitant generation of a larger molecular weight product (,50 kD), suggesting that the 25 kD TET2 protein is a Ca +2 dependent protease cleavage product with a 50 kD intermediate ( Figure 6D).

EZH2 and HP-1 occupancy of PAGE2 and SPANX-B promoter proximal regions decrease during differentiation
Hydroxymethylation has been reported to prevail in promoters with dual H3K4 and H3K27 trimethylation that also bind PRC2 proteins [30]. The PRC2 complex protein EZH2 has been implied in the repression of GAGE, another CT gene [31]. We therefore, asked whether increased hmC within CT gene promoters resulted in altered EZH2 binding to the same sites. Indeed, ChIP experiments demonstrated a decrease in EZH2 occupancy, as well as a decrease in H3K27m3 in both PAGE2 and SPANX-B promoters during Caco-2 SD (Figure 8). The PRC2 component SUZ12 has been reported to regulate H3K9 methylation and in turn, heterochromatin protein 1 (HP1a) binding. In fact, we observed a simultaneous decrease in HP1 binding to both PAGE2 and SPANX-B promoters during differentiation, that correlated with PAGE2 and SPANX-B upregulation ( Figure 8). Thus, our data suggest that both PRC2 and HP-1 contribute to maintaining PAGE2 and SPANX-B in a transcriptionally silent state when the  cells have a mesenchymal phenotype, and that increased TET2 expression and hmC mediated transcriptional activation are related to PRC2 and HP-1 dissociation from the promoters of these CT genes during differentiation.

PAGE2 and SPANX-B up-regulation is reversed during EMT
We hypothesized that if the epigenetic alterations underlying CT gene expression happened in parallel to MET, that this process could be reversed if cells entered EMT. To test this hypothesis, differentiated Caco-2 cells were detached and allowed to proliferate for 5 days. This resulted in their rapid dedifferentiation as evidenced by down-regulation of SI. Dedifferentiated cells down-regulated PAGE2 and SPANX-B, as well as CDX, as they up-regulated TAGLN, in line with ongoing EMT (Figure 9). Although transcription of all three TET genes decreased during de-differentiation (Figure 9), we did not observe a decrease in hmC during this period (data not shown). We, therefore, conclude that the up-regulation of CT gene expression is reversible in this model.

Discussion
Previous studies revealed that CT gene expression correlated with an epithelial rather than a mesenchymal phenotype, and showed the up-regulation of CT genes during MET [12,14]. To our knowledge, this is the first report describing alterations in several epigenetic mechanisms within promoters of two CT genes during MET-like differentiation concordant with a dynamic change in gene expression. As bisulfite sequencing of PAGE2 and SPANX-B promoters revealed no change upon differentiation, the increased hmC must strictly involve methylated CpG residues. This is in line with the fact that TET enzymes are responsible for the conversion of 5-methyl cytosine to 5hydroxymethyl cytosine [27,28]. Conversion of hmC to mC is a far more complex process and might not happen with similar kinetics [32,33]. This is likely the reason why we did not observe a change in hmC during the 5 day de-differentiation process of Caco-2 cells despite the decrease observed in global TET levels. Our finding that PAGE2, SPANX-B and TET2 induction is reversible is similar to another study in embryonic stem cells where Vitamin C was shown to induce TET2 expression, which in turn, resulted in up-regulation of CT genes. Both events were reversible upon Vitamin C withdrawal [34].
Although hydroxymethylation within gene promoters has been reported to decrease during differentiation of normal cells, a recent study revealed that about 20% of all modified cytosines in most CT genes in human brain, where they are not expressed, consist of hmC [35]. Up-regulation of TET2 expression in cancer has been associated with MET [36,37]; and therefore, a more differentiated state [30]. Similar to the inverse correlation between EZH2 and CT/TET2 expression we report here, others have shown EZH2 and TET enzymes to repress and induce differentiation of neuronal precursors, respectively [38]. CT genes are up-regulated during the initial stages of development in the human embryo, but decrease as tissues differentiate further [39]. As adult colon tissue does not show PAGE2 or SPANX-B expression (data not shown), had Caco-2 cells the capability of differentiating further, both genes might have been down-regulated completely. On the other hand, the fact that we could not demonstrate up-regulation of GAGE, MAGE-A3, NY-ESO-1 or SSX4 expression in this model might be because these genes are expressed at earlier stages of differentiation. We believe this because SPANX-B expression is primarily in post-meiotic cells of the testis (i.e. spermatocytes, spermatids, or sperm), whereas GAGE, MAGE-A3, NY-ESO-1 or SSX expression is primarily in spermatogonia [11].
Our data and that of several others' indicate that cancer cells that express CT genes have more of an epithelial rather than a mesenchymal phenotype. We suggest that CT genes PAGE2 and SPANX-B are induced during a window of differentiation that correlates with up-regulation of epithelial markers of differentiation. The Caco-2 SD model has made it possible to observe the actively changing epigenetic landscape within the promoters of these CT genes. However, as CT gene expression in tumors has closely been related to the methylation state of their promoter, the process that leads to CT gene induction in vivo might ultimately result in ''fixing'' of the epigenetic state which would in turn result in CpG methylation. Yet, via dynamic MET in tumors [40], it is conceivable that even this might change over the course of the disease.
From a clinical perspective, data from our lab as well as from others reveal that sub-grouping of tumors based on gene expression profiles can clearly identify cells with different chemosensitivity profiles [12,41,42]. In this line, we predict future studies will reveal distinct drug sensitivity profiles for colorectal cancer subtypes as possibly defined by PAGE2 and SPANX-B expression, for which the Caco-2 SD model could be used.