Epigenetic changes within the promoter region of the HLA-G gene in ovarian tumors

Background Previous findings have suggested that epigenetic-mediated HLA-G expression in tumor cells may be associated with resistance to host immunosurveillance. To explore the potential role of DNA methylation on HLA-G expression in ovarian cancer, we correlated differences in HLA-G expression with methylation changes within the HLA-G regulatory region in an ovarian cancer cell line treated with 5-aza-deoxycytidine (5-aza-dC) and in malignant and benign ovarian tumor samples and ovarian surface epithelial cells (OSE) isolated from patients with normal ovaries. Results A region containing an intact hypoxia response element (HRE) remained completely methylated in the cell line after treatment with 5-aza-dC and was completely methylated in all of the ovarian tumor (malignant and benign) samples examined, but only variably methylated in normal OSE samples. HLA-G expression was significantly increased in the 5-aza-dC treated cell line but no significant difference was detected between the tumor and OSE samples examined. Conclusion Since HRE is the binding site of a known repressor of HLA-G expression (HIF-1), we hypothesize that methylation of the region surrounding the HRE may help maintain the potential for expression of HLA-G in ovarian tumors. The fact that no correlation exists between methylation and HLA-G gene expression between ovarian tumor samples and OSE, suggests that changes in methylation may be necessary but not sufficient for HLA-G expression in ovarian cancer.


Background
Classic and non-classic HLA (human leukocyte antigen) class I genes play a central role in the regulation of the immune response. The non-classic HLA-G gene is expressed in a variety of tissues but perhaps most notably in the fetal-maternal interface on the extravillous cytotrophoblast and has been postulated to help protect the fetus from maternal allorecognition [1]. This hypothesis is supported by subsequent studies demonstrating that HLA-G proteins can suppress a variety of immune functions including natural killer (NK) cell-mediated cytolysis and the T-cell proliferative response [2,3]. Recent findings indicate that HLA-G antigens are present in ovarian and various other types of malignant cells and tissues [4][5][6][7].
These findings and others have led to the hypothesis that induction of HLA-G expression in tumor cells may contribute to their avoidance of immunosurveillance by the host [8,9] (but disputed by [10]).
Sequences known to be involved in the transcriptional regulation of most HLA class I genes are disrupted in the HLA-G gene raising questions as to the mechanisms underlying HLA-G expression [11][12][13]. Studies conducted in a variety of human cancer cell lines suggest that epigenetic mechanisms may play an important role in HLA-G expression [14,15]. To explore the potential role of DNA methylation on HLA-G expression in ovarian cancer, we tested the effect of the methylation inhibitor 5-aza-deoxycytidine on methylation within the CpG-enriched regulatory region of the HLA-G gene and correlated changes in expression in an ovarian cancer cell line. The results demonstrate that 5-aza-dC treatment results in hypomethylation of putative control sequences within the 5' regulatory region of HLA-G and that these changes in methylation correlate with a significant increase in expression. A notable exception was a region (-211 to -290) containing a hypoxia response element (HRE; [16]) that remained completely methylated.
Methylation within the regulatory region of the HLA-G gene also was examined in eighteen malignant and benign ovarian tumor samples and in ovarian surface epithelial cells (OSE) isolated from four patients with normal ovaries. A number of significant differences in levels of methylation of sequences within the 5' regulatory region were detected between the tumor samples and the normal surface epithelial cells. Interestingly, the region containing the HRE (-211 to -290) that remained methylated in 5-aza-dC treated BG-1 cells was also completely methylated in all ovarian tumor samples, but not in OSE controls, suggesting strong selection against accessibility to the HRE in ovarian tumor cells. Although the highest levels of HLA-G expression were associated with tumor samples, no significant overall correlation between methylation and expression levels was detected by real time RT-PCR. Our results indicate that alterations in methylation may be necessary but not sufficient for HLA-G expression in ovarian tumors.

5-aza-dC treatment of ovarian cancer cells (BG-1) results in hypomethylation of sequences within the HLA-G regulatory region and correlates with an increase in gene expression
Previous studies have shown that 5-aza-deoxycytidine treatment of a variety of tumor (glioma, choriocarcinoma, B-lymphoma and melanoma) cell lines results in significant hypomethylation of a CpG-rich region located within 450 bp 5' of the HLA-G start codon and correlates with a significant increase in HLA-G expression [15]. To determine if ovarian tumor cells would show a similar response, we selected an ovarian cancer cell line (BG-1) that did not display HLA-G expression prior to treatment ( Figure 1). Sodium bisulfite modification and subsequent sequencing was carried out on 10 clones each of genomic DNA isolated from 5-aza-dC treated and untreated BG-1 cells. The results demonstrate that while all 19 CpG sites located within the 450 bp region 5' of the HLA-G start codon were methylated in untreated cells, there was a 36% overall decrease of methylation in treated cells (Figure 2). The approximately 200 bp region immediately 5' to the transcriptional start site (-8 to -188) that encompasses HLA-G cis-regulatory sequences was significantly (p = 0.001) hypomethylated, but a 79 bp region (-211 to -290) remained methylated after 5-aza-dC treatment in all of the clones examined. Correlated with the 5-aza-dC induced changes in methylation patterns, there was an increase in HLA-G expression ( Figure 1).

Methylation within the HLA-G 5' regulatory region significantly differs between ovarian surface epithelial cells (OSEs) and benign and malignant ovarian tumors
Epithelial ovarian carcinomas are believed to derive from ovarian surface epithelial cells (OSE) or epithelial cells lining the inner surface of inclusion cysts [17,18]. To determine if methylation within the 5' regulatory region of the HLA-G gene differs among OSE and benign and malignant ovarian tumors, we performed sodium bisulfite genomic sequencing of DNA isolated from laser capture microdissected (LCM) tumor cells from 9 adenomas and 9 adenocarcinomas and OSE brushings from the Activation of HLA-G gene transcription in the BG-1 cell line treated with 5-aza-deoxycytidine (5-aza-dC) Figure 1 Activation of HLA-G gene transcription in the BG-1 cell line treated with 5-aza-deoxycytidine (5-aza-dC).

in a 399 bp region of HLA-G gene extending from the transcriptional start site (TSS) into Exon II (-7 to +392) in malignant (CA), benign (BN) and normal (NL) ovarian tissue samples. (Continued)
ference overall in levels of methylation between OSE and malignant or benign ovarian tumor samples was detected within this region (p < 0.05; Table 2).

No significant difference in HLA-G expression was detected between OSE and malignant and benign ovarian tumor samples
To determine if levels of HLA-G expression are significantly different between ovarian tumors and OSE cells, we determined relative expression levels (real time RT-PCR) of 3 OSE samples and tumor cells isolated by LCM from 4 malignant and 4 benign ovarian tumor samples. The results presented in Table 3 indicate that HLA-G expression levels are highly variable among all samples and no consistent or significant differences in HLA-G expression were detected between the OSE and tumor samples. Our results are consistent with previous reports of high variability in HLA-G expression within and between tumors [7,10]. Thus, the significant difference in methylation levels within the promoter region observed between OSE and the ovarian tumor samples was not correlated with differences in HLA-G expression.

Discussion
Contrary to other HLA molecules, HLA-G regulation is only partially understood. Studies of the methylation levels of 450 bp of the promoter region of HLA-G in melanoma and choriocarcinoma cell lines showed a correlation between expression and low levels of methylation [15]. In our study we looked at the same region in the ovarian cancer cell line BG-1, which showed almost complete methylation of all CpG dinucleotides and no expression of HLA-G. Treatment of the cells with the methylation inhibitor, 5-aza-dC, resulted in a significant decrease in levels of methylation at the HLA-G promoter region that correlated with an increase in levels of HLA-G expression (Figures 1 and 2). Consistent with previously published studies conducted with other cancer cell lines [14,15], our results suggest that the methylation status of the HLA-G promoter region may be important for the control of HLA-G expression in ovarian cancer.
To test this hypothesis in tissues, we analyzed methylation of the HLA-G 5' regulatory region in 4 OSE samples and 18 malignant and benign ovarian tumor samples. In addition, we measured the relative levels of HLA-G expression in 11 OSE and ovarian tumor samples from which cancer cells were isolated using laser capture microdissection (LCM). We found that there were significant differences in levels of methylation within the 5' HLA-G regulatory region between OSE and the malignant and benign ovarian tumor samples, but no significant difference between the malignant and benign tumor samples. This observation is consistent with the view that any putative adaptive advantage imparted to tumor cells by HLA-G apparently does not distinguish between malignant and benign tumor cells.
We found that many of the CpG sites displaying a significant decrease in methylation in the tumor samples were located in proximity to intact binding sites of regulatory proteins known to serve as activators of gene expression (CCAAT box -76 and an intact SP1 and RFX binding site within the X1 box at -130 [13,19]; Figure 3). Hypomethylation of these sites in ovarian tumors may serve to potentiate HLA-G expression. In contrast, we observed a significant increase in levels of methylation in tumor samples at CpG sites located in proximity to a hypoxia response element [16] located 243 bp upstream of the HLA-G transcriptional start site. Interestingly, this same region remained methylated in 5-aza-dC treated BG-1 cells, suggesting that strong selection may operate to prevent binding to the HRE in ovarian tumor cells.
HRE is the binding site of HIF-1, a protein that plays a critical role in the cellular response to hypoxia, with the potential to act as either a positive or negative regulatory factor [20][21][22]. Evidence has recently been reported that HIF-1 may act as a negative regulator of HLA-G expression in some human melanoma (FON) and choriocarcinoma (JEG-3) cell lines [22] The functional significance of the methylation of the CpG island surrounding the HIF-1 binding site in the ovarian tumor samples remains to be determined. However, preventing the binding of a poten- It is important to keep in mind that the postulated significance of methylation changes in gene regulatory regions is the increased (hypomethylation) or decreased (hypermethylation) access it provides for the binding of proteins or protein complexes that regulate gene expression [23]. Thus, although methylation changes in gene regulatory regions may be necessary for subsequent changes in levels of expression, these changes in methylation alone may not be sufficient to effect expression changes. The fact that the significant differences in methylation we observed between OSE and ovarian tumor samples do not correlate with differences in HLA-G expression may serve to underlie this distinction. Our results are consistent with a scenario whereby changes in methylation in ovarian tumor cells potentiate them for HLA-G transcriptional activation (or avoidance of transcriptional repression) by regulatory proteins induced by micro-environmental or mutational changes in specific tumor cell lineages.

Conclusion
Differences in methylation within the 5' regulatory region of HLA-G were detected between normal ovarian surface epithelial cells and ovarian cancers. The most striking difference was in a region (-211 to -290 from HLA-G TSS), containing an intact hypoxia response element (HRE). This region was completely methylated in all of the ovarian tumor (malignant and benign) samples examined, but was only variably methylated in normal surface epithelial cells. Interestingly, this same region remained completely methylated in an ovarian cancer cell line (BG-1) after treatment with 5-aza-dC suggesting that there is strong selection against loss of methylation in this region in these ovarian cancer cells. Since HRE is the binding site of a known repressor of HLA-G expression (HIF-1), we hypothesize that methylation of the region surrounding the HRE may help maintain the potential for expression of HLA-G in ovarian tumors. The fact that no correlation exists between methylation and HLA-G gene expression between ovarian tumor samples and OSE, leads us to conclude that changes in methylation may be necessary but not sufficient for HLA-G expression in ovarian cancer.

Tissue samples
Ovarian tumor samples were collected at Northside Hospital (Atlanta, GA) during surgery and snap frozen in liquid nitrogen within 1 minute of removal from patients. Brushings of normal ovarian surface epithelial cells (OSE) were preserved in RNAlater (Ambion, Austin, GA) within 1 minute of removal from patients. Patient consent and approval from the Institutional Review Boards of the University of Georgia, Georgia Institute of Technology, and Northside Hospital were obtained. Pathological and clinical information of samples is presented in Table 4.

Cell treatments
BG-1 cells were plated at low density 16 hours before treatment with 50 μM 5-aza-deoxycytidine (Sigma, St. Louis, MO) for 5 days. On the third day of treatment, the media was replaced with fresh media containing 5-aza-dC. Control cells treated with PBS were grown for the same amount of time.

Laser capture microdissection (LCM)
Fresh frozen tissues from tumors were cut into sevenmicron sections, applied to non-charged slides, then fixed in 75% ethanol for 30 seconds, stained and dehydrated using the HistoGene LCM Frozen Section Staining Kit (Arcturus, Mountain View, CA). LCM was performed with an AutoPix Automated Laser Capture Microdissection System using the CapSure Macro Caps (Arcturus, Mountain View, CA). Approximately 10,000 cells were captured on each of 3-4 caps per sample. Subsequently the DNA was extracted from captured cells using the PicoPure DNA Extraction Kit (Arcturus, Mountain View, CA). RNA extraction from captured cells was carried out using the PicoPure RNA Extraction Kit and amplified with the RiboAmpHS RNA Amplification Kit (Arcturus, Mountain View, CA).

DNA and RNA extraction from ovarian surface epithelial cells
Epithelial brushings were obtained from normal ovaries and kept in RNAlater (Qiagen, Valencia, CA). For DNA extraction, cells were spun down and re-suspended in lysis buffer (10 mM Tris HCl, 10 mM EDTA, 0.2% SDS and 50 mM NaCl), with 50 μg/ml RNAse A (Invitrogen, Carlsbad, CA) and 100 μg/ml Proteinase K (EM Science, Gibbstown, NJ) overnight at 37°C. DNA was then phenol-extracted and ethanol precipitated. For RNA extraction, cells were spun down and Trizol was added, following the manufacturer's recommended protocol (Invitrogen, Carlsbad, CA). cDNA was synthesized from RNA using the Ribo-