Differential DNA methylation profiles in gynecological cancers and correlation with clinico-pathological data

The study was approved by the Institutional Review Board. Patients of 127 cases of cervical cancer, 60 cases of endometrial cancer and 50 cases of ovarian cancer were recruited for this study. Those patients were diagnosed and treated in Queen Mary Hospital, Hong Kong from January, 1990 to June, 2003. Primary tumor tissues and paired normal tissues were obtained from biopsy or surgical specimens. The paired normal tissues taken from cervical cancer patients are usually blood. From endometrial cancer patients, the normal tissues are mainly cervix. Patients with ovarian cancer, the normal tissues taken are either from the cervix or the endometrium. Finally, the tissues were maintained frozen in our tissue bank. Informed consent was obtained from all individuals for the collection of tissue or blood.

The histological types and grades of tumor were classified according to WHO criteria. The stage of each cancer was established according to the International Federation of Gynecology and Obstetrics (FIGO) criteria. Table 1 lists the clinical characteristics and treatment modality for those patients. All the patients were followed up after the last treatment every 3 months for the first 2 years, every 6 months for the next 3 years, then yearly thereafter. The end of the observation period was August, 2003. Response to treatment was evaluated by physical examination, appropriate imaging studies or biopsy. Persistent disease was defined as detection of disease within 6 months after completion of treatment. Recurrent disease was defined as local (within pelvis) or distant (outside the pelvis) lesions appearing after a disease-free interval of more than 6 months after completion of treatment.

Genomic DNA from tumor and paired normal tissues was extracted using the standard Proteinase K treatment followed by phenol/chloroform extraction. One μg of tumor DNA from each of the 50 patients with same cancer type (for cervical and endometrial cancer, these 50 patients were randomly selected from the total 127 and 60 patients, respectively) was mixed together as the test DNA pool (M2, M4 and M6 for cervical, endometrial and ovarian cancers, respectively). Similarly, the paired normal DNA from those patients was also pooled to form the controls (M1, M3 and M5 for cervical, endometrial and ovarian cancers, respectively).

CpG island methylation status for a total of 34 loci in the three cancers was first assessed in the differential DNA pools representing tumor and normal DNA by MSP. The methylation status of each locus has already been reported in one or multiple cancer types (see Additional file 1). A pair of primers for each locus, which only recognizes methylated sequence, was used to amplify the sequence of interest. Methylated alleles detected in tumor DNA pool, but not in normal DNA pool, were indicative of aberrant methylation in tumors. Furthermore, nine loci, which were aberrantly methylated in tumors suggestive by the above analysis, were selected for subsequent methylation analysis in individual samples including 127 cervical cancers, 60 endometrial cancers and 49 ovarian cancers (DNA was not sufficient in one case). Another set of primer pair for each locus, which only recognizes unmethylated alleles, was added to discriminate the methylated and unmethylated sequences for each sample. A placenta DNA sample treated in vitro with SssI methyltransferase (New England Biolabs Inc., Beverly, MA) followed by bisulfite treatment was used as a positive control for methylated allele and as a negative control for the unmethylated allele. A placenta DNA without pre-treatment of SssI methyltransferase prior modification, and a placenta DNA without any treatment were used as negative controls for the methylated allele.

The bisulfite treatment of DNA, MSP conditions and sequencing of PCR products have been described in our previous report [16]. Primer sequences and annealing temperature were same as those published reports. Details of the primers, annealing temperature, product size are listed [see Additional file 1].

The association between aberrant hypermethylation and clinico-pathological parameters was tested by the χ² test or Fisher's exact test. The t-test was used to compare the mean age of different groups. All variables (age, stage, grade, histological type, and CpG island hypermethylation) were also subjected to multivariate analysis using Logistic regression. Disease-free survival (DFS) was defined as the period from the completion of the last treatment to the date of the first documented evidence of recurrent disease. Patients without recurrent disease were censored at their last follow-up visit or death. Overall survival (OS) was calculated from the date of diagnosis to death; patients who survived until the end of the observation period were censored at their last follow-up visit. Patients who died because of other causes than cancer were censored at their date of death. Survival analysis was performed by the method of Kaplan-Meier, and differences between groups were determined by log-rank test. The Cox proportional hazards model was used for multivariate survival analysis. CpG island hypermethylation, patients' age, FIGO stage, tumor grading and histological type were included in the regression model. All the statistical analysis was performed using the software SPSS for Windows version 13.0. P < 0.05 was taken as statistically significant.

Figure 1 shows the methylated alleles for some gene loci produced by MSP in the tumor and normal DNA pools of the three cancers. The methylation status for the 34 loci in the three cancers was summarized in Figure 2. Among those 34 loci, (1) aberrant methylation of APC and p16 was detected across the three cancers; (2) MINT31 and PTEN were aberrantly methylated in both cervical cancer and ovarian cancer; (3) specifically, aberrant methylation for CDH1, DAPK, MGMT and MINT2 was only present in cervical cancer; CASP8, CDH13, hMLH1 and p73 were just methylated in endometrial cancer; aberrant methylation for BRCA1, p14, p15, RIZ and TMS1 was only detected in ovarian cancer. For the other loci, aberrant methylation was not indicated in any of the three cancers because methylated alleles were present in both tumor and normal DNA (i.e. methylation occurred in that particular gene in both tumor and normal tissues) e.g. FHIT gene (Figure 2), or methylated allele was not detected in both (i.e. methylation did not occur in that particular gene in both tumor and normal tissues) e.g. ALX3 gene (Figure 2).

For further confirmation, 9 loci with aberrant methylation in the three cancers were further studied by MSP in individual primary tumors. The incidence of methylation for these loci in the three cancers and their association with clinico-pathological parameters and prognosis were also determined. Four loci including DAPK, MGMT, p16 and PTEN were selected for methylation analysis in 127 cervical cancers; APC, CDH13, p16 and hMLH1 were examined in 60 endometrial cancers; and BRCA1, p14, p16 and PTEN in 49 ovarian cancers. Figure 3 shows the methylated and unmethylated alleles produced by MSP in representative samples. Methylation for at least one gene was detectable in 72.4% (92/127) of cervical cancers (Table 2), 66.7% (40/60) of endometrial cancers (Table 3) and 51.0% (25/49) of ovarian cancers (Table 4). Incidence of hypermethylation for DAPK, MGMT, p16 and PTEN in cervical cancer was 56.7%, 25.2%, 16.5% and 15.7% respectively. Incidence for APC, CDH13, p16 and hMLH1 in endometrial cancer was 41.7%, 35.0%, 25.0% and 13.3% respectively. Incidence for BRCA1, p14, p16 and PTEN in ovarian cancer was 24.5%, 18.4%, 16.3% and 8.2% respectively. Although we found genes specific for each type of the three tumors, we deliberately chose p16 and PTEN to determine if the frequencies of methylation are different between different tumors. From the above results, no obvious difference was observed in the methylation frequencies of p16 and PTEN in the different tumors.

Tables 2, 3 &4 show the association between CpG island hypermethylation of the nine loci and the clinico-pathological parameters in the three cancers. Hypermethylation for some specific loci was related to particular tumor subtypes. In cervical cancer (Table 2), DAPK gene hypermethylation was more frequently detected in squamous cell carcinoma (SCC) than in adenocarcinoma (AC) (63.3% versus 34.5%, P = 0.006), whereas, PTEN was more frequently detected in AC than in SCC (34.5% versus 10.2%, P = 0.002). The two findings were confirmed by multivariate analysis (P = 0.004, and 0.002, respectively). A significant correlation was observed between MGMT hypermethylation and the early-stage tumor (P = 0.03), but was not confirmed by multivariate analysis (P = 0.733).

In endometrial cancer (Table 3), hMLH1 gene hypermethylation was more common in moderately or poorly differentiated tumors (G2 & G3) than in well-differentiated tumors (G1) (28% versus 2.8%, P = 0.007), which was confirmed by multivariate analysis (P = 0.032). p16 gene hypermethylation only existed in 15 (32.6%) of 46 stage I carcinomas and none of the 14 stages II-IV carcinomas was positive (P = 0.014). But, the association between p16 gene hypermethylation and early-stage tumor was not confirmed by multivariate analysis (P = 0.748).

In ovarian cancer (Table 4), hypermethylation of BRCA1 was detected at a significantly higher frequency in serous carcinomas than in tumors of the other histological types (35.3% versus 6.2%, P = 0.015], whereas, PTEN were more frequently detected in mucinous carcinoma than in tumors of the other histological types (30% versus 2.6%, P = 0.023). Hypermethylation of p16 gene was detected in 33.3% (9/27) of serous or mucinous carcinomas, but not in any of the tumors of other histological types (P < 0.001). Additionally, p14 hypermethylation was associated with low-grade tumor (40.9% in G1 + G2 versus 12.0% in G3, P = 0.023) and early-stage cancer (37.5% in stage I+II versus 12.0% in stage III+IV, P = 0.038). Hypermethylation of PTEN was also associated with low-grade tumor, whereas, BRCA1 was associated with high-grade tumor (P = 0.026, 0.033 respectively). However, only the association between BRCA1 gene hypermethylation and serous carcinoma was confirmed by multivariate analysis (P = 0.045). More cases are needed for the confirmation of the other findings.

Among 127 cervical cancer patients, there were a total of 31 deaths and 8 patients defaulted follow-up at 3 to 22 months with a median of 11.6 months. The remaining 88 patients had no evidence of disease on last follow-up at 3 to 153 months with a median of 83.7 months.

In survival analysis, the mean of overall survival (OS) was 114.3 months (95% confidence interval: 102.8–125.8 months). Using univariate survival analysis, FIGO stage was significantly associated with OS with estimated 5 year survival of 89% for stage I, 72% for stage II and 40% for stage III and IV (log rank test, P < 0.001). Modalities of treatment also was significantly associated with OS with estimated 5 year survival of 93% for patients treated by radical surgery, 86% for patients treated by radical surgery and radiotherapy and 66% for those treated by radiotherapy with or without chemotherapy (log-rank test, P = 0.022). Women younger than 51 had better survival than older women (5 year survival rate 83 vs 67%, log-rank test P = 0.015).

No significant association of survival was found in relation to grade 1 vs grades 2 & 3 (P = 0.066), histology types, methylation status of DAPK (P = 0.16), MGMT (P = 0.504), p16 (P = 0.258) and PTEN (P = 0.541). On multivariate analysis using Cox forward logistic regression with age, grade, stage, histology types, methylation status of DAPK, MGMT, p16 and PTEN, treatment modalities, only stage (P = 0.003, odds ratio = 1.9) remained as independent significant factor.

To evaluate the role of CpG island hypermethylation in the prediction of response to radiotherapy, 63 cervical cancer patients underwent primary radiotherapy were analyzed. Tumors were divided into radiosensitive and radioresistant groups based on the histological findings of residual tumor cells in the cervical biopsy specimens taken after the completion of external radiotherapy and brachytherapy. If there are no residual viable tumor cells in cervical biopsies, the tumor is defined as radiosensitive. Forty-four tumors were radiosensitive and 19 tumors were radioresistant. Tumors with methylation for any of the four genes were more sensitive to radiotherapy than tumors without (76.6% vs 50.0%, P = 0.045) (Table 2).

Previous methods for DNA methylation analysis, such as Southern-blot analysis, bisulfite genomic DNA sequencing, and restriction enzyme digestion, require large amounts of DNA. The number of CpG islands investigated and sample sizes are also limited. Recently, some high-throughput approaches are developed for the large-scale analysis of multiple CpG islands including restriction landmark genomic scanning, differential methylation hybridization [17], and methylation specific oligonucleotide microarray [18]. However, there are some practical problems with these methods in dealing with clinical specimens. The discriminative ability of each probe is hampered by its different hybridization efficiency between methylated and unmethylated alleles. Moreover, these techniques need the most advanced chip technology and informatics approaches, which might not be widely available.

Compared to bisulfite sequencing, MSP is much simpler, requires less time and avoids the use of expensive sequencing reagents. Furthermore, simultaneous detection of unmethylated and methylated products in a single sample by MSP confirms the integrity of DNA as a template for PCR and allows a semi-quantitative assessment of allele types. Thus, in the present study, the pooled DNA approach and MSP were employed to screen those loci with aberrant methylation in tumors by comparing the methylation status in pooled tumor DNA and normal DNA. Only those loci with aberrant methylation in tumor were recognized as the potential epigenetic markers. Those loci with the same methylation status (presence or absence of methylated allele) in tumor and normal DNA pools were excluded for further study. In the present study, DNA methylation profiles in the three gynecologic cancers were compared easily and rapidly using the method introduced here.

Specific pattern of CpG island hypermethylation existing in each human cancer was first reported by Costello et al. [5], and confirmed by Esteller et al. [6]. In the present study, 34 loci were randomly chosen for investigation. Those genes involve in many cellular pathways and have critical biological functions. Multiple genes were aberrantly methylated in the three cancers; in which some genes have been confirmed in other laboratories studying a single tumor type, whereas, others have not been reported before [15, 16] and [19‐25]. For each cancer type, aberrant methylation for several pathways happened simultaneously. For example, genes involved in cell cycle (p16), apoptosis (DAPK), cell adherence (CDH1), DNA repair (MGMT), and APC/β-cateninroute (APC) were found to be aberrantly methylated in cervical cancer. However, the profile of CpG island hypermethylation for the 34 loci differed among the three types of gynecologic cancers (Figure 2).

Hypermethylation of the p16 gene has been suggested to be a shared epigenetic alteration in multiple human cancers [6]. It was also observed across the three gynecologic cancers in the present study. The association between DAPK hypermethylation and SCC has been reported in our previous study [16] other studies [20, 24]. In this study, aberrant methylation for the DAPK gene was only observed in cervical cancer, of which 80% of the samples were SCC. It was not detected in endometrial and ovarian cancers, both of which are AC. Thus, DAPK gene is selectively methylated in the development of SCC. hMLH1 is one of the DNA mismatch repair genes. Inactivation of hMLH1 results in microsatellite instability (MSI) in tumors [25]. MSI is predominant in tumors associated with HNPCC including endometrial cancer. Our previous study has found that a higher frequency of MSI was detected in endometrial cancer than in ovarian and cervical cancers (unpublished data), which could explain the presence of aberrant methylation of hMLH1 gene in endometrial cancer, but absence in cervical cancer and ovarian cancer. Thus, hMLH1 is susceptible to methylation in the development of endometrial cancer. BRCA1 is critical in the development of breast and ovarian cancers. Germline mutations account for most of the cases with hereditary breast and/or ovarian cancers [26]. However, in sporadic cancer, promoter hypermethylation, not somatic mutation, is the cause for BRCA1 inactivation [27]. In the present study, BRCA1 was specifically methylated in ovarian cancer, but not in the other two cancers. These observations have been confirmed in other laboratories studying a single tumor type. Thus, our results consistently demonstrate that differential DNA methylation profiles exist in the three gynecologic cancers.

Finally, we would like to raise a query concerning the FHIT gene. Our own result indicates that methylated alleles were found in all tumor and normal DNA (Figure 1) i.e. FHIT methylation was not tumor-specific. In contrast, it has been demonstrated that aberrant methylation of FHIT occurred specifically in lung and breast cancers [28]. However, in a study of FHIT methylation in cervical cancer, the methylation pattern identified by bisulfite genomic sequencing did not correlate with the MSP [29]. The authors suggested that false-positive results might be resulted from mispriming in the MSP assay. In this study, we used the primers exactly the same as those published information. Thus, one should be cautious when using biomarker for cancer. The recent publications on frequencies of FHIT methylation in human cancer have been tabulated [see Additional file 2] as references for those readers particularly interested in studying FHIT.

Esteller et al. [6] analyzed a series of promoter CpG island hypermethylation changes for 12 genes in DNA from over 600 primary tumor samples representing 15 major tumor types and concluded that a panel of three to four markers could define an abnormality in 70–90% of each cancer type. Thus, in the present study, 4 markers were chosen from all genes with aberrant methylation in each of the three cancers to determine the incidence of methylation in that particular cancer type and to define their role in cancer diagnosis and prognosis. Hypermethylation in any of the four genes investigated in each cancer type was detected in 50–70% of the patients. In each cancer type, methylation frequency varied among the four genes investigated, which indicated that CpG islands might differ in their susceptibility to de novo methylation.

Furthermore, association analysis demonstrated that hypermethylation of particular CpG islands was correlated with clinico-pathological parameters (tumor histology, grade or stage) and treatment response, which produced distinct epigenetic signatures for particular tumor subtypes. Thus, this epigenetic event has the potential to be used as a molecular marker for cancer diagnosis and prognosis in gynecologic cancers. The most significant finding is the DAPK gene hypermethylation in cervical cancer. DAPK, a pro-apoptotic serine/threonine kinase, involves in apoptosis and plays a role in tumor pathogenesis and metastasis when inactivated [30, 31]. Inactivation of DAPK, mainly by promoter hypermethylation, has been demonstrated in some tumor types and found to be associated with aggressive and metastatic phenotype [32‐35]. In cervical cancer, the correlation between DAPK gene hypermethylation and SCC demonstrated by other studies was further confirmed in this study. Our in-vitro study has shown that methylated promoter of DAPK was present in the radio-resistant cervical cancer cell line, SiHa; while absent in the radio-sensitive cell line, C4–1 (unpublished data). Moreover, the expression of DAPK was down regulated in SiHa cell line, but reactivated by exposure to the demethylation reagent (5-Aza-2' deoxycytidine), which was consistent to the findings by Narayan et al. [24]. Furthermore, our recent study has found that DAPK gene hypermethylation was detected in 50% of plasma samples of cervical cancer patients [16]. Thus, DAPK gene hypermethylation might be a valuable marker for tumor diagnosis, and evaluation of treatment outcome in cervical cancer. Because of its high frequency of methylation in cervical cancer, it might also be potentially used as a therapeutic target for cervical cancer treatment.