We first determined whether the ovarian malignancies showed LOH by comparing the restriction fragment length polymorphism (RFLP) patterns of normal lymphocyte DNA and 74 matching primary HOC DNA samples. Samples where RFLPs were present in the lymphocyte DNA sample but absent or with an altered ratio in the tumor sample were considered to exhibit LOH in the regions of 8 imprinted genes (H19, IGF2, KCNQ1, LIT1, GTL2, PEG1, PEG3 and NDN). The average percentage of heterozygosity was 48.0% (16.2-58.5%). We found only 14 cases of LOH in the 8 imprinted genes in the 74 HOC samples we analysed (Table 1). The most frequent gene with LOH was IGF2 (9.0%, 3/33), followed by PEG1 (8.1%, 3/37) and GTL2 (7.1%, 3/42). LOH of NDN and LOT1 was not detected (0/31 and 0/12). The samples with LOH were not from the same cases (Additional file 1: Table 1).

We next performed RT-PCR and RFLP analysis to identify the samples of LOI without LOH. The frequency of LOI was higher than that of LOH for all 8 imprinted genes and we found a total of 46 cases of LOI (Table 1, Additional file 1: Table S1). The most frequent sites of LOI were PEG1 (45.9%, 17/37), IGF2 (45.4%, 15/33) and H19 (29.2%, 12/41). NDN had the lowest frequency. In 19 of the 46 cases, the abnormal gene expression pattern was apparent at two or more imprinted loci. A normal imprinting pattern, maintenance of imprint (MOI), was most frequent in NDN (93.5%, 29/31). ND (not determined) means no amplification of RT-PCR at 3 times in several samples, perhaps indicating low expression of the genes. In 9 of the 14 LOH cases, LOI was also found in at least one gene. In HOC cell lines, LOI was found in 2 of 3 informative cases for IGF2, and 3 of 9 cases for PEG1. We did not find any LOH or LOI in 7 normal ovarian surface tissues and 4 normal cell lines. We compared patients' ages, progression, histology and tumor grades with imprinted gene expression pattern profiles. Patients with LOI had a tendency to be younger than patients with LOH (mean ages for LOH and LOI: 55.0 ± 7.4 and 47.7 ± 6.9, respectively), but the difference was not statistically significant by ANOVA, and no other correlations were apparent.

The proof-of principle experiment of the BPL method has been described in detail 18 . Briefly, bisulphite-DNA can be used to distinguish between methylation and non-methylation status in the genome, e.g. cytosine and uracil. The BPL method can determine one base substitution by specific hybridization and detect the ratio of methylation to non-methylation. We examined the quality of the BPL method in spermatic DNA, which should show 100% methylation of the paternally methylated DMRs: ZDBF2, H19 and GTL2, whereas the maternally methylated DMRs: PEG1, ZAC, SNRPN, PEG3 and LIT1 are non-methylated. We applied the classic methylation assay COBRA technique and our recently devised BPL method to the DNA of 7 normal ovarian surface epithelium tissues, 4 primary cultures of normal human ovarian surface epithelium (OSE1-4) and 21 HOC cell lines, and performed statistical analysis with Spearman's and Pearson's rank correlations. For all 8 DMRs a good correlation was found between these two methods (Figure 1, Table 2, Additional file 2: Figure S1).

We next determined the methylation status of the 8 DMRs from the 74 samples of primary ovarian cancer tissue by the BPL method. Overall, we compared the average DNA methylation status of cancer and normal samples for each DMR and found that PEG1 from ovarian cancers was significantly more hypermethylated than normal ovarian tissues (normal, 30.7% ± 15.1: HOC, 45.9% ± 15.5) (Table 3). The numbers of cancer tissue cases with hypermethylation above the range of the methylation rates in the normal ovarian surface epithelium were 17 for H19, 21 for GTL2, 21 for ZDBF2, and 14 for PEG1 (Table 3). On the other hand, hypomethylation below the methylation level of normal ovarian tissues was found in 15 cases for GTL2 and 23 for ZDBF2. We did not observe a significant difference in the DNA methylation between localized early-stage and advanced-stage tumor groups (Table 3). This suggested that the DNA methylation changes we detected occurred as early events of ovarian cancer.

To determine whether the DNA methylation status in these DMRs of the imprinted genes acts as an indicator for potential LOH and/or LOI, we evaluated the association between DNA methylation and LOH and/or LOI in the IGF2/H19 and PEG1 imprinted domains separately.

IGF2, which acts as a dominant oncogene, and H19, a physically and mechanistically linked gene on human chromosome 11, are reciprocally imprinted. In the paternal allele, H19 DMR is methylated and silenced, whereas the reciprocally imprinted gene IGF2 is transcribed. By contrast, in the maternal unmethylated allele, H19 is expressed but IGF2 is inactivated because of the binding of the repressor factor CTCF to the unmethylated H19 DMR, which then prevents the H19/IGF2 common enhancers from activating the IGF2 promoter 19 . IGF2 was found to have high frequencies of both LOH and LOI in HOC. H19 was also found to have high frequencies of LOI (29.2%, 12/41). Nine of 14 cases with both IGF2 and H19 heterozygosity showed LOH or LOI of both genes and only one case had MOI for one of the two. Thus relaxation of IGF2 and H19 imprinting is frequent. In the IGF2/H19 imprinted region, the samples with LOH and/or LOI at H19 was more methylated than those with MOI (Figure 2A, Additional file 3: Table S2). Our results for H19 were similar to a previously reported finding 20 . PEG1 was reported to be a TSG and was also found to have high frequency of LOH/LOI in HOC. We also found that the samples with LOH/LOI at PEG1 were more methylated than those with MOI with statistical significance (Figure 2B, Additional file 3: Table S2).

Twenty-one HOC cell lines were used in our study: 10 from serous adenocarcinoma (OVCAR3, CAOV3, JHOS2, HTOA, SKOV3, OV90, JHOS3, JHOS4, KF, MH), 2 from mucinous adenocarcinoma (OMC3, MCAS), 8 from clear cell adenocarcinoma (ES2, JHOC5, TOV21G, JHOC7, JHOC8, KM, HAC2, RMG) and 1 from endometrioid adenocarcinoma (TOV112D). The sources of these cells and culture methods were as described previously 38 39 . Four primary cultures of normal human ovarian surface epithelial (OSE1-4) cells were initiated from surface scrapings of normal ovaries as described 40 .

Seventy-four primary HOC tissues (36 serous, 9 mucinous, 10 endometrioid, 18 clear cell, and 1 other, Table 3) were obtained from patients presenting at our hospital. The mean ± standard deviation (SD) of the patients' ages for normal ovary and ovarian cancer tissues were 44.1 ± 6.6 and 52.9 ± 7.3, respectively. Seven specimens of normal ovarian surface epithelium were obtained from patients with benign non-ovarian disease. Histological diagnoses and clinical staging were performed according to the International Federation of Gynecologists and Obstetricians (FIGO) criteria. The numbers of cancer patients with localized tumors (stage I and II) and advanced tumor (stages III and IV) were 29 and 45, respectively. The samples were stained with hematoxylin and eosin to demonstrate > 85% of epithelial tumor cells. DNA and RNA were then extracted from the remaining samples 38 . DNA was also extracted from peripheral blood in matched patients. The study was performed after obtaining the patients' informed consent and with approval from the institutional ethics committee of the Tohoku University Graduate School of Medicine.

PCR was performed on patient blood and tumor genomic DNA using the primer sequences summarized in Additional file 4: Table S3. A PCR reaction mix containing 0.5 μM of each primer set, 200 μM dNTPs, 1 × PCR buffer, and 1.25U of EX Taq Hot Start DNA Polymerase (Takara Bio, Tokyo, Japan) in a total volume of 20 μl was used. The following PCR program was used: 1 minute of denaturation at 94°C followed by 35 cycles of 30 seconds at 94°C, 30 seconds at 60°C and 30 seconds at 72°C and a final extension for 5 minutes at 72°C. PCR products were digested by unique polymorphic enzymes to identify samples that were heterozygous for a single nucleotide polymorphism (SNP). For samples found to be heterozygous for a SNP, RNA was prepared from matched tumors, followed by reverse transcription-PCR (RT-PCR) and by restriction digestion 41 42 43 44 45 46 47 48 . The digested PCR products were electrophoresed on 2% agarose gel.

Bisulphite PCR-Luminex (BPL) methylation analysis was performed as described 18 . PCR primers sets, biotinylated at their 5'-end, were designed for gene amplification. PCR reaction mix contained 0.2 μM primer, 0.2 mM dNTPs, 1 × PCR buffer (50 mM KCl, 10 mM Tris-HCl, pH 8.3). 3 mM MgCl₂, 2% dimethyl sulfoxide (DMSO), 0.625U Taq DNA Polymerase (Roche, Tokyo, Japan) and 100-200 ng of bisulphite treated DNA in a total volume of 25 μl. PCR conditions: 40 cycles of 95°C for 20 s/60°C for 30 s/72°C for 30 s using a GeneAmp 9700 thermal cycler (Applied Biosystems, CA, USA). Oligonucleotide probe sequences (Additional file 2: Table S1 in Ref 18) were synthesized and covalently bound to carboxylated fluorescent microbeads using ethylene dichloride (EDC). These oligonucleotide-labeled microbeads (oligobeads) were mixed together to make an oligobeads mixture of 100 oligobeads/μl and hybridized to the 5'-biotin-labeled PCR amplicons in a total volume of 50 μl per well in a 96-well plate by adding 5 μl of the appropriate oligobead mixture and 5 μl of the PCR amplicons to 40 μl of hybridization buffer. This reaction mixture was first denatured at 95°C for 2 min and then hybridized at 48°C for 30 min. After hybridization, the oligobeads were washed in 100 μl of PBS-Tween and pelleted by microcentrifugation. Pelleted oligobeads were reacted with a 70 μl aliquot of a 100 × diluted solution of SA-PE in PBS-Tween. Hybridized amplicons were labeled with SA-PE at 48°C for 15 min. Reaction outcomes were measured by the Luminex 100 flow cytometer. Methylation assays were additionally performed for each DMR using the conventional bisulphite treatment PCR methylation assay and combined bisulphite PCR restriction analysis (COBRA) as described previously 21 .

Differences between groups were analysed by analysis of variance, followed by Post-hoc, Tukey's HSD test. Statistical analyses were performed using the JMP (v9.0.0, SAS Institute Japan, Tokyo, Japan). Statistically significant differences between groups are presented as *P < 0.05, and **P < 0.01. Results for BPL and COBRA were compared using Spearman's rank method and Pearson's product-moment correlation coefficients.