Identification of candidate epigenetic biomarkers for ovarian cancer detection

Tumor specimens from patients with ovarian cancer

Table I summarizes the clinicopathological information of 77 patients diagnosed with stage III or IV epithelial ovarian cancer. These tumor samples were generously provided by the Western Infirmary and Stobhill General Hospital (Glasgow, UK) and the University Medical Centre Groningen (Groningen, The Netherlands). As controls, normal ovarian surface epithelia (nOSE) were obtained from 9 patients undergoing procedures for benign conditions. All experimental protocols were approved by local institutional review boards and ethics committee.

Bisulfite conversion, qMSP, and conventional real-time polymerase chain reactions (PCR) to assess the copy numbers of genes of interest. Genomic DNA was extracted from ovarian tumors or from cell lines, purified by using QIAamp DNA Mini Kit (Qiagen, Germantown, MD), and subjected to bisulfite conversion using the EZ DNA Methylation Kit (Zymo Research, Orange, CA). The resultant DNA was subjected to real-time qMSP using Applied Biosystems 7500 Fast Real-Time PCR System (Applied Biosystems, Foster City, CA) as described previously (37,38). The sequences of DNA oligonucleotide primers utilized in qMSP reactions are shown in Table II. PCR comprising SYBR-Green as the reporting molecule (Applied Biosystems), were carried out at 95°C for 10 min followed by 40 cycles of 95°C for 15 sec, annealing temperature (60-64°C) for 30 sec, and 72°C for 35 sec. To equalize the amount of DNA in each reaction, the threshold cycle number (Ct) for each sample was first normalized in a PCR amplification utilizing primers devoid of CpG dinucleotides but interrogating COL2A1. Because COL2A1 is frequently in low- or single-copy in human genomes, it hence commonly serves as a methylation-independent control to normalize the quantity of DNA specimens (39). Next, relative methylation level in the locus of interest was assessed by comparing Ct of qMSP from each sample to that from a methylation standard curve generated by SssI-treated sperm DNA (Chemicon, Temecula, CA) in which nearly all CpG dinucleotides are methylated by CpG methylase in this study (data not shown). The level of methylation in each sample is expressed as a percentage, in relation to SssI standard which was set as 100%. Thus, a percentage of methylation >100 may be partly due to an elevated copy number in the tumor samples compared to the sperm DNA. To assess the DNA copy number of a given locus in relation to the one of SssI, DNA was amplified by a conventional real-time PCR method (40), in a methylation-independent manner using specific primers to interrogate CYP39A1 locus: forward primer (GGAGTTTAGGAGGT TATTGGG) and reverse primer (CTCCCTCCCACCTC CACT).

Ovarian cancer cell lines and epigenomic treatments

Immortalized ovarian surface epithelial cells (IOSE), derived from the primary culture transduced with the catalytic subunit of hTERT and the human papiloma virus E7 subunit (41), were grown in Medium MCDB105:199 (1:1) (Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS). The human ovarian cancer A2780 cell line was originally isolated from an ovarian adenocarcinoma of an untreated patient and is relatively sensitive to cisplatin in vitro (42). Independent, multiple selected lines (MCP3 and MCP8) derived from A2780, were obtained by seven cycles of selection, with increasing concentrations of cisplatin (43,44). Nevertheless, A2780, MCP3 and MCP8 were cultivated in RPMI-1640 (Invitrogen) containing 10% FBS as previously described (44). To reverse the aberrant epigenomes, sub-confluent cells were seeded on cultivation vessels and were treated with 2.5 μM 5-aza-2′-deoxycytidine (ADC) (Sigma Inc., St. Louis, MO) alone, 0.5 μM trichostatin A (TSA) (Sigma Inc.) alone, or with both drugs. Mock control cells were grown concurrently, but treated with vehicle DMSO. Forty-eight hours after drug treatments, cells were harvested and subjected to RNA extraction.

Quantify mRNA by quantitative reverse transcription PCR (qRT-PCR)

Total RNA from ovarian cancer cell lines was purified by using TRIzol reagents (Invitrogen). The resultant RNA was first reverse transcribed into cDNA, and later annealed to the corresponding primers (Table II) prior to PCR amplification using the following conditions: 95°C for 10 min, 40 cycles of 95°C for 15 sec, 58-60°C for 30 sec, and 72°C for 35 sec. In each sample, the level of mRNA was determined by subtracting the threshold cycle number (C_t) of the internal control GAPDH from that of test locus. The relative expression of mRNA in the locus of interest is expressed as fold changes in the treated cells, in relation to the untreated mock control.

Statistical analysis

The significance of DNA methylation in tumors, in relation to nOSE samples, was assessed by real-time qMSP followed by statistical analysis using the Wilcoxon rank-sum test. Pearson's correlation was used to calculate concurrent hypermethylation among multiple genes. Difference in mRNA level between the untreated and treated cell lines was determined by the paired Student's t-test and was considered to be significant, if the P-value is <0.05.

The clinical application of these five genes as biomarkers for detecting ovarian carcinomas was appraised by their specificity and sensitivity using a 100-fold cross-validation method (23). For a given locus, 44 specimens from the cancer group and 5 from nOSE, were randomly selected using the statistical analysis package R (24), and whose qMSP values were used for the following calculation. In principle, by collecting: i) the nOSE samples with methylation levels greater than the minimum in the cancer group; and ii) the cancer samples with methylation degree less than the maximum in nOSE specimens, the threshold value was set to be the median of all qMSP values of samples from this collection. By repeating the analysis for additional 99 times and each run encompasses a different random sampling of the specimens, the final threshold value was set as the average of median across the 100 iterations, and thereby served as a cut-off value for a given gene above which the test specimen is counted as hypermetylated. Further, sensitivity for a given locus is the percentage of the true positives (hypermethylated) in all cancer specimens studied, whereas the specificity is that of the true negatives (unmethylated) in nOSE (shown in Table III). On the other hand, Cox proportional hazard regression model was used to assess the affect of hypermethylation on clinical parameters including progression-free survival, using a software package R (17,24).

Identification of hypermethylated loci in ovarian cancer

Aberrant epigenomic regulation is a frequent event associated with human malignancy. While DNA hypermethylation and histone deacetylation contributes to tumor suppressor gene silencing, our laboratory has previously examined methylation patterns of ovarian cancer using DMH and has identified potential CpG island loci that can be developed as biomarkers for cancer diagnosis and prognosis (12,13). Briefly, amplicons derived from ovarian tumors or from normal controls were labeled with fluorescent Cy5 and Cy3 dyes, respectively, and then were co-hybridized to a CpG island microarray slide. Hybridized slides were scanned and candidate genes with signal ratios (Cy5/Cy3) ≥2 were scored as potential hypermethylation in tumor samples compared with nOSE. To continue validating the loci identified from our previous DMH microarray findings (12,13), we developed qMSP assays to quantify the methylation levels of candidate CpG island loci.

For each given locus, a study was conducted to determine amplification specificity of each methylation-specific primer pair. Pairs of oligonucleotides were utilized in the qMSP reactions (sequences are listed in Table II) to interrogate the CG-rich regions within the vicinity of the promoters and the first exon, of these genes. PCR products were viewed on agarose gels as an initial discriminatory testing to differentiate pooled ovarian tumors from the combined controls (nOSE; representative examples are shown in Fig. 1A). qMSP was then used to assess the methylation level of 5 CpG island loci (CYP39A1, GTF2A1, FOXD4L4, EBP, and HAAO) in 77 ovarian tumors and 9 control nOSEs.

Due to limited amounts and the quality of DNA, not all patient specimens could be amplified by qMSP. However, the average methylation level of a given locus in 55 tumors successfully analyzed, in relation to the mean of 9 nOSE samples, was calculated as: 469.93 vs. 27.98 for CYP39A1; 182.02 vs. 29.43 for GTF2A1; 204.15 vs. 6.63 for FOXD4L4, 45.18 vs. 11.93 for EBP; and 17.35 vs. 0.05 for HAAO (Fig. 1B). Exceeding the 100% methylation level in four loci (CYP39A1, GTF2A1, FOXD4L4, and EBP) could likely be a result of increased genome copy number in ovarian tumors, as compared to the control DNA derived from the SssI-treated sperm sample and nOSE. For example, while we interrogated the meythylation levels of CYP39A1 by qMSP, we observed that hypermethylation was positively associated with increased DNA copy numbers. As the metyhylation levels in ovarian tumors samples named 5, Z112, and S05), were quantified to be 696, 210 and 10.61% of the SssI control (black bars in Fig. 1B), their corresponding DNA copy numbers were estimated as 4.00, 2.28, and 0.45, in relation to SssI derived from sperm DNA. In fact, growing body of evidence has demonstrated that elevated genome copy number is a common hallmark of advanced ovarian malignancy (14-16). Most importantly, the methylation levels of these five loci were significantly elevated in tumors (P<0.001). Interestingly enough, after adjusting for age, FIGO stage and response to chemotherapy, another advanced statistical method employing multivariate Cox regression analysis (17) has identified hypermethylation of CYP39A1 correlated with an increase rate of relapsing (P=0.032, hazard ratio >1). However, the same multivariate analysis cannot correlate hypermethylation of the remaining four genes with any of clinical parameters including progression-free survival.

Next, we determined whether concordant hypermethylation of multiple genes could be simultaneously observed in these tumors. Based on the qMSP data shown in Fig. 1B, a methylation heat map of four loci was generated from 55 tumors and 9 nOSE specimens (Fig. 2A). Simultaneous hypermethylation of equal or greater than three loci was observed in 50 out of 55 (91%) of the tumors examined. Interestingly, two of these loci, CYP39A1 and GTF2A1, displayed statistically significant concurrent hypermethylation in 52 out of 55 (95%) ovarian tumors examined (P<0.001; Pearson's correlation coefficient, r=0.6408; Fig. 2B), suggesting the likelihood of the so-called CpG island methylator phenotype (CIMP) (18) in these ovarian tumors. CIMP refers to a notion that a subset of tumors has widespread methylation of CpG islands that leads to epigenetic inactivation of tumor suppressor genes. This phenomenon is most apparent in colon cancer and was first characterized by a concordant methylation of promoter regions of two separate genes, CDKN2A and TBSI (19). Subsequent studies have added several loci to the list of genes and the minimal number of genes is commonly ranged between 3 and 8 in colorectal cancer (20). Similar CIMP, with concurrent hypermethylation in between three to six loci, has been observed in ovarian cancer and is associated with adverse prognosis (21) or response to chemotherapy (22).

qMSP assays for hypermethylated genes bestow diagnostic tests with high sensitivity and specificity

The application of these hypermethylated loci as biomarkers was accessed by their sensitivity and specificity. By using a 100-fold cross-validation method (23) and the statistical analysis package R (24), we have set off the threshold cut-off value for hypermethylation of a given locus (first column, Table III) and demonstrated that qMSP tests on all five genes can offer high sensitivity in detecting ovarian carcinoma (with sensitivity values ranged between 81 and 96%) (Table III). Among them, the highest specificity was identified in GTF2A1 (88%), while HAAO ranked the second (82%). Given the advantage of high sensitivity plus specificity, testing for the methylation in GTF2A1 alone or in combination with HAAO, may comprise novel targets for molecular diagnostics.

DNA demethylation and/or histone acetylation reactivated the expression of hypermethylated genes

Our data demonstrated elevated methylation of five loci in ovarian tumors. To validate this finding in the laboratory cultivation system, ovarian cancer cell lines (A2780, MCP3 and MCP8) and the immortalized ovarian surface epithelia (IOSE) were subjected to epigenomic studies. We demonstrated that all three ovarian cancer lines retained remarkable methylation in three loci (FOXD4L4, EBP, HAAO) (Fig. 3A). However, methylation of two other loci (CYP39A1 and GTF2A1) was unnoticeable (data not shown) and hence both genes were excluded from further studies.

Our ultimate goal is to decipher the epigenetic gene silencing in ovarian cancer and to advance our current understanding about this malignancy. Because both DNA methylation and histone modifications are apparent epigenetic phenomena, we next performed a series of treatments to reverse the epigenomic suppression and to reveal the epigenetic regulatory mechanisms underlying gene silencing. Interestingly, DNA demethylating agent (ADC) caused prominent reactivation of HAAO (>60 fold increase) in all 3 ovarian cancer cell lines. However, inhibition of histone de-acetylation (by TSA treatment) resulted in variable outcomes. TSA induced greater than 70-fold activation in parental A2780 line, but conducted suppression, rather than activation, in the drug-resistant lines (MCP3 and MCP8). Yet, dual treatments failed to convey synergistic effect, suggesting DNA hypermethylation is far more critical than histone deacetylation in commanding HAAO repression in the drug resistant lines (MCP3 and MCP8), as opposed to the parental A2780 line that requires both mechanisms for regulation.

Similar variability was observed in silencing of two loci, FOXD4L4 and EBP (Fig. 3B). DNA hypermethylation appeared to be more critical in two cases (FOXD4L4 in MCP8, and EBP in MCP3) while histone deacetylation seemed to be more prominant in three occurrence (FOXD4L4 in A2780 and MCP3; and EBP in A2780). As an exception, reactivation of EBP in MCP8 required combinatorial actions from both agents, suggesting both mechanisms simultaneously suppressed the expression of EBP. However, we cannot exclude the possibility that failed outcomes are due to ineffective uptake, increased efflux and metabolism, or unknown cytotoxicity that disabled their action.