|Home | About | Journals | Submit | Contact Us | Français|
Unlike epigenetic silencing of tumor suppressor genes, the role of epigenetic derepression of cancer-promoting genes or oncogenes in carcinogenesis remains less well understood. The tight junction proteins claudin-3 and claudin-4 are frequently overexpressed in ovarian cancer and their overexpression was previously reported to promote the migration and invasion of ovarian epithelial cells. Here, we show that the expression of claudin-3 and claudin-4 is repressed in ovarian epithelial cells in association with promoter ‘bivalent’ histone modifications, containing both the activating trimethylated histone H3 lysine 4 (H3K4me3) mark and the repressive mark of trimethylated histone H3 lysine 27 (H3K27me3). During ovarian tumorigenesis, derepression of CLDN3 and CLDN4 expression correlates with loss of H3K27me3 in addition to trimethylated histone H4 lysine 20 (H4K20me3), another repressive histone modification. Although CLDN4 repression was accompanied by both DNA hypermethylation and repressive histone modifications, DNA methylation was not required for CLDN3 repression in immortalized ovarian epithelial cells. Moreover, activation of both CLDN3 and CLDN4 in ovarian cancer cells was associated with simultaneous changes in multiple histone modifications, whereas H3K27me3 loss alone was insufficient for their derepression. CLDN4 repression was robustly reversed by combined treatment targeting both DNA demethylation and histone acetylation. Our study strongly suggests that in addition to the well-known chromatin-associated silencing of tumor suppressor genes, epigenetic derepression by the conversely related loss of repressive chromatin modifications also contributes to ovarian tumorigenesis via activation of cancer-promoting genes or candidate oncogenes.
In addition to genetic DNA alterations, including mutations, deletions, amplifications, rearrangements and translocations, epigenetic abnormalities are now known to be intimately involved in multistep carcinogenesis (1,2). The role of epigenetic silencing of key tumor suppressors by DNA hypermethylation or histone modifications in their promoter regions is now well established in tumorigenesis (2–4). Although less well understood, the loss of epigenetic repression also plays a role in tumorigenesis, by facilitating the activation of oncogenes or cancer-promoting genes, and global DNA hypomethylation at repetitive sequences and imprinted genes also contributes to tumorigenesis by promoting chromosomal instability or loss of imprinting (2,5,6).
Claudins are a 24-member family of proteins that are the major components of tight junctions, epithelial cell–cell contacts that play crucial roles in cell polarity maintenance and control of paracellular ion flux (7). Although loss of claudins has been associated with tumorigenesis (probably by allowing cell detachment and migration), claudin-3 and claudin-4 have been demonstrated to be overexpressed in several cancers including those of the breast, prostate and uterus (7,8). More frequently, however, claudin-3 and claudin-4 have been shown to be overexpressed in ovarian cancer (7–9), a malignancy that is atypical in that it actually retains or gains (rather than loses) epithelial characteristics during tumor progression (10,11). It is believed that claudin-3 and claudin-4 overexpression in ovarian cancer enhances tumor cell motility, invasiveness and survival, possibly by enhancing proteolytic activation of basement membrane-degrading matrix metalloproteinases (12). Due to their consistent overexpression in ovarian cancer and the association of claudin-3 overexpression with poor prognosis, claudin-3 and claudin-4 are under investigation as diagnostic or prognostic biomarkers (7,8). Furthermore, as claudins are transmembrane proteins with two extracellular loops, they represent promising targets for therapeutic antibodies (7), and interestingly, claudin-3 and claudin-4 are receptors for the Clostridium perfringens enterotoxin (7,9,13), representing a possible targeted therapeutic using C.perfringens enterotoxin in ovarian cancer (7,9,13), whereas a CLDN3 small interfering RNA (siRNA) demonstrated potent suppression of tumor growth and metastasis of mouse and human ovarian tumor xenografts (14).
Despite the importance of the upregulation of claudin-3 and claudin-4 in ovarian cancer, the mechanism by which their overexpression occurs remains under investigation, although recent studies have suggested the crucial roles of epigenetic modifications, including DNA hypomethylation and histone H3 acetylation in the upregulation of two genes in ovarian cancer cells (9,15,16). Interestingly, CLDN3 and CLDN4 are adjacent genes on chromosome 7q11.23, located only ~60 kb apart and being transcribed in opposite directions, suggesting their possible coordinated regulation by sharing regulatory regions.
In addition to DNA methylation, the ‘histone code’ hypothesis posits that covalent modification of histone tail residues acts in concert to govern DNA packaging and thus access of transcription machinery to coding sequences (17,18). Histone modifications on specific residues correlate with either active or repressive transcription (3,19). Specifically, trimethylation of histone H3 lysine 9 (H3K9me3), H3 lysine27 (H3K27me3) and H4 lysine 20 (H4K20me3), facilitates transcriptional repression, whereas histone acetylation of histone H3 (H3Ac) and H4 (H4Ac), and trimethylation of H3 lysine 4 (H3K4me3) are associated with active transcription (3,19). Methylation of specific histone lysine residues is mediated by their cognate histone methyltransferases, and the recent discovery of histone lysine demethylases has indicated that the ‘histone code’ is highly signal responsive and dynamic (19,20). In embryonic stem cells, a ‘bivalent’ colocalization of the activating H3K4me3 and the repressive H3K27me3 of development-associated genes, followed by lineage-specific loss of the H3K4me3 or H3K27me3, has been reported to allow differentiated tissue silencing or expression (21,22). To fine-tune gene regulation, ‘crosstalk’ between various epigenetic modifications has also been reported (23–26). In particular, it is well known that histone H3K9 methylation and H3K27me3 are linked to DNA methylation (24,27–29).
In this study, we describe epigenetic mechanisms that regulate CLDN3 and CLDN4 expression during ovarian tumorigenesis and suggest that epigenetic derepression by loss of repressive histone modifications is one possible mechanism underlying the overexpression of CLDN3 and CLDN4. These results demonstrate that DNA methylation-independent epigenetic activation of previously repressed cancer-promoting genes, in addition to the better understood process of epigenetic silencing of tumor suppressor genes, may represent an additional mechanism of tumor progression.
Human ovarian cancer cell lines were obtained from the American Type Culture Collection (Manassas, VA) (Caov3, OV-90, SW626 and TOV-112D) or from the Korean Cell Line Bank (Seoul, Korea) (OVCAR3, SK-OV3, SNU8 and SNU119). Human immortalized ovarian surface epithelial (IOSE) cell lines were obtained from Dr Nelly Auersperg (University of British Columbia, Vancouver, Canada) (IOSE80UBC and IOSE385) and from Dr Michael Birrer (National Cancer Institute, Bethesda, MD) (IOSE80NIH and IOSE120) and grown in 1:1 mixture of Medium 199 and MCDB 105 supplemented with 5% fetal bovine serum. Primary cultures of human normal ovarian surface epithelial (NOSE) cells were obtained from Dr Yong-Sang Song of the Department of Obstetrics and Gynecology, Seoul National University College of Medicine, following Institutional Review Board approval, using resected healthy ovaries, followed by ovarian surface epithelium exofoliation and culturing as described previously (30).
Cells were treated with 5 μM 3-deazaneplanocin A (DZNep; obtained from Dr Victor E.Marquez of National Cancer Institute, MD) or 5 μM 5-aza-2’-deoxycytidine (5-aza-dC; Sigma, St. Louis, MO) for 72 h and 200 nM trichostatin A (TSA; Sigma) for 24 h as described previously (31). Twenty micromolar LY294002, a PI3K-Akt inhibitor (Sigma), was added to cells for 24 h (32).
Formalin-fixed, paraffin-embedded (FFPE) tissues of human NOSE, ovarian adenoma,borderline tumor and ovarian carcinoma which are de-identified and anonymized, were obtained from the Samsung Medical Center (Seoul, Korea), following approval by that institution’s review board. Immunohistochemistry was performed as described previously (8) using Zymed antibodies against EZH2 (36-6300), claudin-3, (18-7340) or claudin-4 (18-7341). We used the same ovarian carcinoma FFPE tissues that were used for tissue microarray analysis of the expression levels of claudin-3 and claudin-4 by Choi et al. (8) for EZH2 immunohistochemistry.
Total RNA was extracted from cells using Trizol (Invitrogen, Carlsbad, CA) or the Allprep (Qiagen GmbH, Hilden, Germany) kits and reverse transcribed to complementary DNA using the Superscript™ II First-Strand Synthesis System (Invitrogen) according to the manufacturer’s protocol. Following complementary DNA synthesis, quantitative polymerase chain reaction (PCR) was carried out as described previously (33) in a dual system LightCycler (Roche Applied Science, Mannheim, Germany) using the primers, Universal Probe Library (Roche Applied Science) probe sequences are listed in supplementary Table S1 (available at Carcinogenesis Online), with the HPRT1 Taqman probe (TIB MOLBIOL, Berlin, Germany) used as a ‘reference gene’ to normalize gene expression.
Whole cell lysates were prepared by cell incubation with RIPA lysis buffer and 20–30 μg of total proteins were used for immunoblotting with antibodies to claudin-3 (Zymed Laboratories, South San Francisco, CA; 18-7340), claudin-4 (Zymed, 18-7341), actin (Santa Cruz Biotechnology, Santa Cruz, CA; sc-1616), H3K27me3 (Millipore, Temecula, CA; 07-449), EZH2 (Cell signaling, Danvers, MA; 4905), SUZ12 (Abcam, Cambridge, UK; ab12073) or EED (Millipore, 09-927).
CLDN3 promoter luciferase reporter plasmids were constructed by PCR using a bacterial artificial chromosome clone as a template (AC093168, RP11-148M21/DH10B; Invitrogen) and the primers are listed in supplementary Table S2 (available at Carcinogenesis Online). The PCR product was then cloned into a pGL3-Basic (Promega, Madison, WI) and the correct product confirmed by DNA sequencing. Cells were harvested 16 h after cotransfection with luciferase reporter and pCMV-lacZ plasmids using Lipofectamine LTX (Invitrogen) according to the manufacturer’s instructions and firefly luciferase activity was measured using a luciferase assay system (Promega). All experiments were repeated three times.
CLDN3 and CLDN4 CpG islands were located using CpGIE (34) with the following criteria: GC% ≥55; observed to expected ratio ≥0.65; length ≥500. Genomic DNA was extracted from (i) cells, using G-spin (iNtRON Biotechnology, Seongnam-si, Korea) or Allprep (Qiagen) genomic DNA extraction kits or (ii) ovarian FFPE tissues, using the ChargeSwitch gDNA Micro Tissue Kit (Invitrogen) according to the manufacturer’s protocols. One microgram of genomic DNA was bisulfite modified using the EzWay™ DNA methylation Detection Express Kit (KOMABIOTECH, Seoul, Korea) and PCR amplified using the primers and Taqman probes for methylation-specific PCR (MSP), quantitative MSP and bisulfite sequencing PCR listed in supplementary Tables S2 and S3 (available at Carcinogenesis Online). For quantitative MSP, methylation levels were calculated as percentage of methylated reference (%) by dividing CLDN3 or CLDN4:ACTB ratio of a sample by the CLDN3 or CLDN4:ACTB ratio of SssI-treated (and thus fully methylated) human leukocyte genomic DNA (from the Samsung Medical Center) and multiplying this value by 100.
Chromatin immunoprecipitation (ChIP) assays were performed using the EZ-ChIP kit (Millipore) according to the manufacturer’s instructions with 5 μg of normal rabbit IgG (Cell signaling, 2729) as a negative control and 5 μg Millipore antibodies against H3K4me3 (05-745), H3K27me3 (07-449), H3K9me2 (07-441), H3K9me3 (07-442), H3Ac (06-599), H4Ac (06-598), H4K20me3 (Abcam, 9053). ChIP primers and Taqman probes are shown in supplementary Tables S2 and S3 (available at Carcinogenesis Online). Fold enrichment by quantitative ChIP was calculated as the value of the immunoprecipitated sample divided by the value of the negative control (IgG).
Control (siGENOME, non-targeting siRNA #1) and gene-specific (EZH2, SUZ12, On Target plus SMART pool) siRNAs were purchased from Dharmacon (Lafayette, CO). After one or more siRNA transfections using Oligofectamine (Invitrogen), cells were harvested at various time points and used for quantitative real-time reverse transcription–PCR, western blotting and ChIP assays.
TOV-112D cells stably expressing H3K27me3 dominant-negative mutation (TOV-K27R) or wild-type histone H3 (TOV-WtH3) were established through G418 (Invitrogen) selection after transfection with the plasmids described previously (29). Transformants were first selected by resistance to G418 (800 or 1000 μg/ml), and single-cell clones were then isolated by limiting dilution, with plasmid stable expression confirmed by enhanced green fluorescent protein fluorescence.
Using eight ovarian cancer and four IOSE cell lines, we first confirmed a positive correlation between CLDN3 and CLDN4 messenger RNA and protein levels, demonstrating both to be regulated at the level of transcription (Figure 1A and B). Based on previous studies suggesting the upregulation of CLDN3 and CLDN4 in ovarian cancer cells via hypomethylation of promoter CpG islands (9,15,16), we next performed MSP and quantitative MSP to further explore those previous correlations. MSP analyses of normal and malignant ovarian cell lines demonstrated that CLDN4 messenger RNA levels inversely correlate with CLDN4 promoter DNA methylation (Figure 1A), with DNA hypermethylation in CLDN4-repressed IOSE cells being consistent with the NOSE cells used by Litkouhi et al. (9). To further confirm CLDN4 methylation level, we performed bisulfite sequencing PCR to quantitatively examine the methylation levels of individual CpG dinucleotides. Bisulfite sequencing results were similar with the MSP data, showing CLDN4 hypermethylation in CLDN4-repressed cells (SK-OV3, OV-90, TOV-112D and IOSE80NIH) (Figure 1C). These findings confirm that CLDN4 derepression in ovarian cancer cell lines strongly associates with DNA hypomethylation, as compared with its promoter hypermethylation-induced silencing in primary NOSE or IOSE cells (Figure 1A).
However, unexpectedly, we observed little or no DNA methylation in the CLDN3 promoter of six cell lines lacking CLDN3 expression (OV-90, TOV-112D and four IOSE cells), in contrast to CLDN3 promoter hypermethylation in SK-OV3 and SNU119 cells, two cell lines that express CLDN3 at relatively low levels (Figure 1B). As shown in Figure 1D, bisulfite sequencing PCR analysis also corroborated the MSP data, further confirming a lack of DNA methylation in CLDN3-repressed cells, whereas CLDN3 expression (absent) and DNA methylation levels (little or none) in primary cultured NOSE cells were similar to IOSE cells. Taken together, our results demonstrate a lack of DNA hypermethylation in the CLDN3 promoter region in CLDN3-repressed cells, suggesting another mechanism(s) for gene repression.
Based on different DNA methylation pattern between CLDN3-repressed and CLDN4-repressed ovarian cells, we further compared CLDN3 and CLDN4 promoter DNA methylation levels in FFPE samples from NOSE, ovarian adenoma, borderline tumors and ovarian carcinomas. Similar to IOSE cells, CLDN3 was not methylated in NOSE tissues (0%, 0/9), despite undetectable claudin-3 expression in those normal tissues (Figure 2A; supplementary Table S4 is available at Carcinogenesis Online). Interestingly, in precancerous borderline tumors, which had little or no claudin-3 expression (supplementary Table S4 is available at Carcinogenesis Online), CLDN3 DNA promoter methylation (62.5%, 5/8) was significantly increased, as compared with NOSE tissues (P = 0.007) (Figure 2A). However, in ovarian carcinomas (26.8%, 15/56), CLDN3 methylation was significantly lower compared with that in borderline tumors (P = 0.039) (Figure 2A; supplementary Table S5 is availabe at Carcinogenesis Online). These results suggest that CLDN3 promoter methylation increases during ovarian premalignant tumorigenesis, but is lost in fully neoplastic tumors, in agreement with a previous study (16).
In contrast to the lack of association between DNA methylation and claudin-3 expression in premalignant tissues, the frequency and level of CLDN3 methylation in the claudin-3-non-expressing group were higher than in the claudin-3-expressing group (frequency: P = 0.07; level: P = 0.077) in ovarian carcinomas (Figure 2B; supplementary Table S5 is available at Carcinogenesis Online), demonstrating that CLDN3 promoter DNA methylation at least partially associates with protein expression in ovarian carcinoma tissues.
We also examined CLDN4 promoter DNA methylation and protein expression in normal and malignant ovarian FFPE tissues. Consistent with the cell line data, CLDN4 methylation was frequently found in NOSE tissues at frequencies and methylation levels similar to those found in ovarian adenoma and borderline tumor tissues (Figure 2C; supplementary Table S6 is available at Carcinogenesis Online). These FFPE analyses indicate that in contrast to CLDN3, CLDN4 methylation is maintained during early-stage tumorigenesis. Interestingly, there was no inverse correlation between claudin-4 expression and DNA methylation in ovarian adenoma and borderline tumor tissues (supplementary Table S4 is available at Carcinogenesis Online), despite the strong inverse correlation observed in the cell lines studies (Figure 1A), suggesting additional mechanisms of CLDN4 dysregulation during premalignant tumor progression. In ovarian carcinomas, CLDN4 was surprisingly methylated, with no significant difference in the levels and frequency of DNA methylation between the claudin-4-expressing and claudin-4-non-expressing groups (supplementary Table S6 is available at Carcinogenesis Online). However, we note the small number (n = 7) of claudin-4-repressed carcinomas, in addition to possible leukocyte contamination of the FFPE tissues, as CLDN4 was also found hypermethylated in human leukocytes (Figure 1A, ‘h.L’ bar).
To determine whether CLDN3 repression in IOSE cells and ovarian cancer cell lines was due to absent transcription factors, we examined luciferase reporter activity following deletion analysis of CLDN3 promoter in four cell lines possessing varying levels of endogenous CLDN3 expression. As shown in supplementary Figure S1 (available at Carcinogenesis Online), luciferase activity of the various CLDN3 promoter fragments was similar among the four cell lines examined (regardless of their CLDN3 expression status), in agreement with a similar study of the CLDN3 promoter (16). These promoter regulation studies confirm that the basal transcription machinery remains intact in both CLDN3-repressed and CLDN3-expressing cells, supporting the role of other possible mechanism(s) of CLDN3 regulation during ovarian tumorigenesis. Similarly with CLDN3, the same pattern of luciferase activity of the CLDN4 promoter was also reported in ovarian cancer cells regardless of their CLDN4 expression in previous study (15).
To determine whether histone modifications in the CLDN3 promoter mediate transcriptional repression, independently of DNA methylation, we performed ChIP studies of three distinct CLDN3 promoter regions and quantitative ChIP of one region. Interestingly, the CLDN3 promoter showed a ‘bivalent’ colocalization of both repressive H3K27me3 and activating H3K4me3 histone mark in CLDN3-repressed IOSE cells in addition to CLDN3-repressed ovarian cancer cell lines (OV-90 and TOV-112D), whereas H3K27me3 was lost in CLDN3-expressing cells (Caov3 and SNU119) (Figure 3A). The most disparately altered chromatin modification between CLDN3-repressed and CLDN3-expressing cells was the repressive H3K27me3 histone mark, although loss of two other repressive marks, H4K20me3 and H3K9me3, likewise associated with CLDN3 derepression, also independently of altered DNA methylation. Detectable H3K9me2 was not found in the CLDN3-repressed cells and active chromatin marks, H3K4me3 and H3Ac, were observed in the CLDN3 promoters of all cells, regardless of CLDN3 expression (Figure 3A). In contrast to H3Ac, however, only low levels of H4Ac were detected in the CLDN3 promoter of the three cell lines (Caov3, SNU119 and IOSE80NIH) having varying levels of CLDN3 expression.
Similar to the CLDN3 promoter, we found a bivalent pattern of CLDN4 promoter histone methylations in CLDN4-repressed IOSE cells, which were DNA hypermethylated (Figure 3B). In addition to DNA hypomethylation, loss of the repressive H3K27me3 and H4K20me3 histone mark was observed in CLDN4-expressing cells (SNU119 and Caov3), as compared with CLDN4-repressed cells. In contrast to CLDN3, the CLDN4 promoter possessed detectable H3K9me2, in addition to DNA hypermethylation, in cells lacking its gene expression (TOV-112D and IOSE80NIH) (Figure 3B), suggesting a link between these two repressive modifications in the CLDN4 regulation, in accord with previous studies supporting such an association (24,35). The levels of active histone mark H4Ac and H3K4me3 were correlated with CLDN4 expression (Figure 3B), whereas H3Ac levels were high regardless of CLDN4 expression similar to CLDN3. These results indicate that in normal and malignant ovarian epithelial cells, repressive histone methylations (H3K27me3, H3K9me2 and H4K20me3) in the CLDN4 promoter, together with DNA hypermethylation, strongly repress CLDN4, whereas activating histone marks (H3K4me3 and H4Ac), in conjunction with DNA hypomethylation, allow CLDN4 transcriptionally permissive.
Trimethylation of H3K27 is catalyzed by the histone methyltransferase EZH2, a member of the Polycomb repressive complex-2 that is also overexpressed in embryonic stem cells and in highly aggressive epithelial malignancies (36). To examine the role of EZH2 in the regulation of CLDN3 and CLDN4 expression, we utilized siRNA transfection to knock down EZH2 in TOV-112D cells, which normally repress CLDN3 and CLDN4 in association with promoter H3K27me3. Reduced levels of EZH2 messenger RNA and protein corresponded with significantly decreased H3K27me3 binding in both the CLDN3 and CLDN4 promoter region (P < 0.05) (Figure 4A). However, the loss of these repressive chromatin marks did not correspondingly induce CLDN3 or CLDN4 expression, as compared with the control siRNA treatment (Figure 4A). Similar results were observed in CLDN3- and CLDN4-repressed OV-90 cells (supplementary Figure S2 is available at Carcinogenesis Online). Further attempts to augment EZH2 knockdown by repeated EZH2 siRNA transfections or cotransfection of EZH2 siRNA with an siRNA against SUZ12 in TOV-112D cells also did not restore the expression of both genes (data not shown). We also treated TOV-112D cells with the PI3K-Akt inhibitor LY294002 to inhibit EZH2 serine-21 phosphorylation and thus upregulate its H3K27 methyltransferase activity (32). This treatment actually increased CLDN3 and CLDN4 expression (Figure 4B), rather than decreasing the expression. Additional evidence against direct EZH2 regulation of CLDN3 and CLDN4 expression was provided by a lack of correlation between EZH2 and claudin-3/claudin-4 expression by immunohistochemistry in ovarian carcinoma tissues (supplementary Figure S3 is available at Carcinogenesis Online).
We also established stable cell lines with a dominant-negative histone H3 lysine 27 mutation in TOV-112D cells (TOV-K27R) to rule out the possibility of incomplete H3K27me3 by transient siRNA transfection and the existence of other methyltransferases involved in the methylation of H3K27. In agreement with our EZH2 knockdown findings, the TOV-K27R cells did not derepress CLDN3 or CLDN4, as compared with untransfected TOV-112D or TOV-WtH3 cells (Figure 4C). Overall, these results suggest that the expression of CLDN3 and CLDN4 is not solely regulated by EZH2, or that H3K27me3 loss alone is insufficient to derepress their expression consistent with findings in previous reports (37,38), and lend support to the idea that additional mechanisms might be involved in their regulation.
Recently, DZNep was found to deplete components of the Polycomb repressive complex-2 complex and inhibit histone methylations, including H3K27me3 and H4K20me3, whereas it had little effect on the levels of H3K9me3 or H3Ac (39). To test whether DZNep, alone or in combination with another epigenetic modulator, could regulate CLDN3 expression in cells lacking its expression (TOV-112D, OV-90 and IOSE80NIH), we assessed CLDN3 transcript levels after mono- or co-treatment with the histone deacetylase inhibitor, TSA. By themselves, DZNep or TSA treatment alone only slightly induced CLDN3 expression in each cell line (Figure 5A). However, we detected a strong (3- to 5-fold) increase in CLDN3 expression in all three cell lines following treatment with DZNep combined with TSA (Figure 5A), suggesting possible crosstalk of histone methylations (H3K27me3 and/or H4K20me3) with histone acetylation. Interestingly, 5-aza-dC treatment induced CLDN3 expression in TOV-112D and IOSE80NIH cells with very low DNA methylation in contrast to no induction in OV-90 cells having little DNA methylation (supplementary Figure S4 is available at Carcinogenesis Online). However, it was difficult to detect the concomitant DNA demethylation at CLDN3 promoter in TOV-112D and IOSE80NIH cells after 5-aza-dC treatment (data not shown), suggesting that increased CLDN3 expression by 5-aza-dC might occur independently of DNA demethylation.
To examine possible mechanisms of DZNep/TSA-mediated CLDN3 upregulation, we assessed changes in histone modifications in the CLDN3 promoter following DZNep/TSA-mediated gene induction in TOV-112D and OV-90 cells. Polycomb repressive complex-2 proteins (EZH2, SUZ12 and EED) levels were further depleted when DZNep/TSA cotreated, as compared with DZNep alone (Figure 5B). As shown in Figure 5C, the combined treatment significantly reduced CLDN3 promoter-localized H3K27me3 and H4K20me3 (P < 0.05), with a slight decrease in H3K9me3, while concomitantly increasing the activating mark H3Ac and H4Ac (P < 0.05) in TOV-112D cells. These results demonstrate that in addition to loss of repressive histone methylations, histone acetylations on both H3 and H4 may also regulate CLDN3 expression. In OV-90 cells, both H3K27me3 and H4K20me3 were significantly decreased (P < 0.05) without a marked increase in histone acetylation (Figure 5C), suggesting that a reduction in multiple repressive histone methylation modifications alone can allow gene induction when high levels of active marks are already present.
As CLDN4 was repressed with DNA hypermethylation in ovarian cells (Figure 1A and C), in addition to H3K27me3 and histone acetylation, we also assessed the effects of a DNA methylation inhibitor, 5-aza-dC (alone or in combinations with DZNep or TSA), on CLDN4 expression in promoter-hypermethylated TOV-112D cells. We found that combined 5-aza-dC/TSA treatment resulted in robust CLDN4 induction (~16-fold), with only minimal induction by 5-aza-dC monotreatment (Figure 5D, top). DNA demethylation was observed in 5-aza-dC-treated cells (Figure 5D, middle), indicating that increased CLDN4 expression by 5-aza-dC/TSA treatment is related to DNA demethylation. Fairly similar to CLDN3, CLDN4 was induced nominally (~3-fold) by DZNep/TSA, although that combination induced simultaneously significant decreases in repressive marks (H3K27me3 and H4K20me3) and increases in active marks (H3Ac and H4Ac) in the CLDN4 promoter (P < 0.05; Figure 5D, bottom). Unlike CLDN3, marked increases in the H3Ac and H4Ac were observed, with H4Ac increased greater (5.1-fold) than H3Ac (2.5-fold). These results indicate that concurrent histone acetylation and DNA hypomethylation are required for maximal CLDN4 derepression. Interestingly, DZNep treatment of TOV-112D cells induced the loss of DNA methylation with a concurrent increase in demethylation of the promoter region of CLDN4 (Figure 5D, middle), suggesting an interaction between DNA methylation and Polycomb group proteins (PcG)-mediated histone methylation in the regulation of CLDN4 expression, in agreement with a previous report (27).
These results indicate that CLDN3 and CLDN4 repression can be robustly reversed by simultaneous changes in multiple epigenetic modifications using combined treatment targeting several chromatin modifications.
Several candidate oncogenes, including BORIS, SNGG, IGF2 and tumor progression-associated CLDN4, have been shown to be induced by the loss of promoter DNA methylation in ovarian and other cancers (40). However, little is known about activation of tumorigenesis-associated genes due to the loss of other repressive epigenetic marks, including silencing histone methylations.
In the present study, we first demonstrate that CLDN3 and CLDN4 overexpression in ovarian cancer cells is associated with decrease of the repressive histone methylation marks H3K27me3 and H4K20me3 in their promoter regions. Interestingly, we found CLDN3 to be repressed in NOSE cells and tissues, despite a lack of promoter DNA methylation. Promoter bivalency, simultaneously possessing activating (H3K4me3) and repressive (H3K27me3) histone mark, is found in numerous development-related genes in embryonic and hematopoietic stem cells (21,41). This histone pattern is associated with a low level of gene expression, keeping genes poised for activation (21,41). Similarly, we found that CLDN3 and CLDN4 repression is associated with promoter bivalency, with their upregulation associated with loss of H3K27me3 and H3K4me3 retention. It has been recently demonstrated that a H3K27me3 demethylase UTX forms a complex with the H3K4 trimethyltransferase MLL2 (42), suggesting counter-modulation of these two opposing histone marks in the loss of bivalency during gene upregulation (20,42,43), and further studies could investigate a role for this complex in tumorigenesis-associated CLDN3 or CLDN4 reexpression.
As expected, in highly gene-repressed cells, we observed the presence of additional repressive histone marks, including H3K9me3, H4K20me3 and hypoacetylated H4 in CLDN3-repressed cells and H3K9me2, H4K20me3 and hypoacetylated H4 in CLDN4-repressed cells, consistent with previous reports (44,45). Although CLDN3 was repressed in IOSE cells by histone modifications alone, CLDN4 additionally possessed DNA hypermethylation in CLDN4-repressed IOSE cells and NOSE tissues, consistent with previous cell line studies (9,15), indicating a more pronounced gene-suppressive chromatin environment for that gene.
Previous studies have linked bivalent chromatin pattern or H3K27me3 to mark genes for subsequent DNA methylation (28,46), and in CLDN4 promoter, we indeed observed the presence of both repressive marks. Although CLDN3 possessed H3K27me3, but not DNA methylation, we did observe CLDN3 DNA methylation in premalignant borderline tumor tissues consistent with the histone mark acting as an identifier for subsequent DNA methylation. Similarly, DNA methylation has been associated with both H3K9me2 and H3K9me3 (24); however, although DNA-methylated CLDN4-repressed cells possessed both histone marks, DNA methylation-independent CLDN3 repression associated with H3K9me3 alone, suggesting that H3K9me2, rather than H3K9me3, is more likely to be linked to DNA methylation.
In contrast to some genes whose expression was reactivated by EZH2 siRNA alone (47,48) or in H3K27 dominant-negative mutant cells (29), CLDN3 and CLDN4 expression was not induced by EZH2 knockdown or loss of H3K27me3, suggesting that CLDN3 and CLDN4 are not direct targets of EZH2 and PcG-mediated H3K27me3 depletion alone insufficient to derepress their expression, in agreement with previous studies (37,38). However, DZNep treatment in combination with TSA resulted in considerable CLDN3 induction with changes in multiple histone modifications in the CLDN3 promoter. These results indicate that simultaneous changes in multiple epigenetic modifications are required for CLDN3 expression induction, in agreement with previous reports (31,38).
Moreover, in CLDN4-repressed cells with DNA hypermethylation, 5-aza-dC treatment strongly induced CLDN4 expression in combination with TSA (~16-fold) or with DZNep (~4-fold), whereas inhibition of DNA methyltransferases or DZNep alone resulted in only minimal reexpression, indicating the need for loss or gain of other chromatin marks in addition to DNA hypomethylation for CLDN4 strong reexpression.
Although it needs to be biochemically determined whether strong synergy by DZNep/TSA for CLDN3 or 5-aza-dC/TSA for CLDN4 is generated by direct interaction between epigenetic modifications, simultaneous changes in epigenetic marks by combined treatment in our study reflect possible interplay between histone modifications in the regulation of CLDN3 expression or link between histone modification and DNA methylation in the regulation of CLDN4 expression. Further studies will be required to address mechanistically if CLDN3 and CLDN4 are regulated by multiple chromatin modifications via their crosstalk.
Although previous studies (15,16) reported the similar epigenetic mechanism regulating CLDN3 and CLDN4 expression in ovarian cancer, our study shows that different epigenetic mechanism is involved in the alteration of claudin-3 and claudin-4 expression during ovarian tumor progression. This might be due to different chromatin-binding proteins including transcriptional factors in their upstream regulatory elements because transcription regulatory factors are involved in recruiting multiple chromatin modifying proteins, which in turn regulate epigenetic modifications in the promoter region (49). These findings raise the possibility of more complex upstream mechanism regulating CLDN3 and CLDN4 and further allow different epigenetic strategy targeting claudin-3 or claudin-4 for cancer therapy.
Finally, our findings suggest that epigenetic therapies capable of targeting multiple histone modifications in the promoter regions of CLDN3 and CLDN4 may have the potential to treat ovarian cancer. Since the overexpression of CLDN3 and CLDN4 is related to ovarian malignancy, repression of these genes in ovarian cancer by an epigenetic approach could be used for treatment of ovarian cancer. Conversely, as claudin-3 and claudin-4 have been demonstrated to be receptors for the C.perfringens enterotoxin, it is possible that their further epigenetic upregulation could enhance the sensitivity to that compound, similar to currently studied methods of derepressing cell surface antigens for immunotherapy (50,51).
In conclusion, we have elucidated the epigenetic mechanisms underlying the overexpression of CLDN3 and CLDN4 in ovarian cancer. Although there exists a growing number of examples or DNA hypomethylation-related induction of candidate oncogenes, we further demonstrate the crucial importance of histone modifications in tumorigenesis-associated gene activation and suggest the epigenetic derepression by loss of repressive histone modifications as one possible mechanism underlying the overexpression of cancer-related genes during ovarian tumorigenesis. Our study offers novel insight into cancer epigenetics and the current understanding of ovarian tumorigenesis, and further suggests novel epigenetic approaches that could be used to target these proteins for ovarian cancer therapy.
National Research Foundation of Korea, Korea government (20090083533); Industry University Research Cooperation Program, Seoul Metropolitan Government (10541); Intramural Research Program of the National Institutes of Health, National Cancer Institute, Center for Cancer Research, USA; and National Cancer Institute Awards (CA085289 and CA113001).
We thank Dr Nelly Auersperg for access to the Canadian Ovarian Tissue Bank and Dr Michael Birrer for providing the IOSE cell lines. We are also very grateful to Dr Mi-Ock Lee (College of Pharmacy, Seoul National University, Korea) for her critical comments and Si Eun Kim for her technical assistance.
Conflict of Interest Statement: None declared.