Normal and tumor regions of cervical tissues were biopsied from four patients, two pre-invasive cancer subjects (CIN III and CIS), and two patients with cervical cancer (CxCa Ib and CxCa IIa), and stored at the Kangdong Sacred Heart Hospital (Hallym University, Seoul, South Korea) following international ethical guidelines. All the patients provided written informed consent. This study was approved by the Kangdong Sacred Heart Hospital Institutional Review Board. Genomic DNA and RNA were purified from frozen tissues. HeLa cells were obtained from the American Type Culture Collection (Manassas, VA, USA) and cultured in Dulbecco’s modified Eagle’s medium (Gibco-BRL, Grand Island, NY, USA) supplemented with 10% fetal bovine serum, 100 U/ml penicillin and 100 μg/ml streptomycin at 37°C in a humidified CO₂ incubator.

Total RNA was extracted using TRIzol^® (Invitrogen Life Technologies, Carlsbad, CA, USA) and further purified using RNeasy columns (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. RNA purity and integrity were evaluated by gel electrophoresis and the ratio of absorbance at 260 and 280 nm using the NanoDrop^® ND-1000 (Thermo Scientific, Waltham, MA, USA). Samples were further analyzed on an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Microarray analysis using human HT-12 expression v.4 bead array (47,000 probes; Illumina, Inc., San Diego, CA, USA) was carried out by Macrogen Co. (Seoul, South Korea) following the manufacturer’s instructions. Raw data were extracted using the software provided by the manufacturer [Illumina GenomeStudio (v2011.1) Gene Expression Module (v1.9.0)] and filtered if the P-value was <0.05, similar to signal to noise ratio, in ≥50% of the samples analyzed. Signal values of the selected genes were transformed using logarithms and normalized based on the quantile method. The statistical significance of the expression data was determined using an independent Student’s t-test and fold change, in which the null hypothesis was such that no difference existed between the two groups. The false discovery rate was controlled by adjusting the P-value using the Benjamini-Hochberg algorithm. Gene enrichment and functional annotation analysis was performed using Panther DB (http://www.pantherdb.org/). All data analysis and visualization of differentially expressed genes was conducted using The R project (www.r-project.org).

Genomic DNA was purified using the DNeasy mini-kit (Qiagen), and sample quality was evaluated using the NanoDrop ND-1000. Intact genomic DNA was diluted to 50 ng/μl based on Quant-iT Picogreen (Invitrogen Life Technologies) quantitation. Microarray analysis using Infinium Human Methylation 450K BeadChip (Illumina, Inc.) was carried out by Macrogen Co., followed by the manufacturer’s instructions. Image processing and raw data extraction were performed using GenomeStudio software [Illumina GenomeStudio (v2011.1) Methylation Module (v1.9.0)] following the manufacturer’s instructions. The β-value was calculated by subtracting the background signals using negative controls on the array and determining the ratio of the methylated signal intensity against the sum of the methylated and unmethylated signals. A β-value of 0–1.0 was reported as a significant percentage of methylation, 0–100%, respectively, for each CpG site (16). Array CpG probes with a detection P-value of ≥0.05, similar to the signal to noise ratio, in >50% samples were filtered out. A filtering criterion was applied for data analysis; a good signal value was required to obtain a detection P-value of <0.05. Filtered data was normalized using the quantile method. Differential methylation was determined based on |Δ_mean| ≥0.2 (difference in methylation signal = average β of case - average β of control).

Statistically significant mRNA and CpGs were identified by |fold change| ≥2 and |Δ_mean| ≥0.2, respectively. Next, putative mRNA-CpG target pairs with regulatory associations were extracted. This approach assumed that DNA methylation is negatively correlated with the mRNA expression of its targets. Negative correlation was identified between putative pairs of mRNA and CpGs. The novelty of selected genes with cervical cancer was confirmed in cervical cancer gene database (CCDB; http://crdd.osdd.net/raghava/ccdb) (17). To reduce the bias of the present data, which came from a limited number of samples, the expression of selected genes was confirmed in a publicly accessible microarray database, the gene expression database across normal and tumor tissues (GENT; http://medicalgenome.kribb.re.kr/GENT) (18). All statistical data analysis was conducted using R 2.15.1 (www.rproject.org).

Hierarchical clustering of DNA methylation and RNA expression profiles was conducted using Cluster 3.0 (19). Input data was analyzed by array mean centering and followed by average linkage clustering using correlation (centered) similarity metrics. Clustering results were viewed using TreeView version 1.60 (Eisen Lab, Berkeley, CA, USA). Functional annotation clustering and pathway analysis were conducted using DAVID (20). Functional clustering analysis of all selected genes was performed using medium classification stringency (κ similarity: Similarity term overlap, 3; threshold, 0.5; Classification: Initial and final group membership, 3 and multiple linkage threshold, 0.5).

HeLa cells (4.0×10⁴) were plated in a six-well plate and treated with various concentrations (0, 5, 10 and 20 μM) of 5-aza-2′-deoxycytidine as decitabine (Selleckchem; Houston, TX, USA) for 96 h. Total DNA and RNA were isolated using the DNeasy and RNeasy mini kits, respectively.

RT was performed with 2 μg total RNA using the First-Strand cDNA Synthesis kit (Invitrogen Life Technologies) following previously described procedures (21). To quantify the expression levels of the selected genes, quantitative PCR analysis was performed using the first cDNAs as templates and a specific primer set for each gene, ALDH1A3 (P290199), PPP1R3C (P131676) and CAMK2N1 (P229340) (Bioneer Co., Daejeon, Korea) following the manufacturer’s instructions. Gene amplification was performed using the ABI^® PRISM 7900HT Sequence Detection System (Applied Biosystems Life Technologies, Foster City, CA, USA) in a 20 μl reaction mixture containing 2 μl cDNA template (80 ng), 2.5 μl of each primer, 3 μl distilled water and 10 μl 2X Power SYBR Green PCR Master mix (Applied Biosystems Life Technologies), which included a dNTP mixture. PCR conditions were as follows: Initial denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 10 sec and 60°C for 30 sec. Sequence Detection System (SDS) Software version 2.4 (Applied Biosystems Life Technologies) was used to calculate Ct values for all genes. Relative expression was calculated using the 2^−ΔΔCt method.

Genomic DNA was modified using the EpiTect Bisulfite kit (Qiagen), which converts cytosine residues to uracil, but does not affect 5-methylcytosine, thus allowing identification of methylated sequences in the DNA by subsequent MS-PCR analysis. For MS-PCR analysis, MS primers (MSP) and unmethylation-specific primers were designed using methylation identification software (22). PCR was then performed in a final volume of 50 μl using 100 μg bisulfite modified DNA and 10 pmol of each primer using the EpiTect MSP kit (Qiagen) under the following conditions: Initial denaturation at 95°C for 5 min, 35 cycles of 95°C for 30 sec, 55°C for 30 sec and 72°C for 30 sec, and a final extension step of 72°C for 5 min. PCR products (20 μl) were visualized on 1.8% agarose gels using the Gel Documentation System (AE-9000 E-graph; ATTO Corporation, Tokyo, Japan).

Data from DNA methylation and RNA expression was analyzed using an integrated approach and strong cut-off values to overcome the limited number of samples. For differentially-methylated genes in cervical cancer tissues, methylation loci with an absolute value ≥0.2 of the Δ_mean in the methylation profile were selected from the data set. To select genes showing differential expression in cancer tissues, genes with ≥2-fold changes in cancer tissues were selected. The data from selected methylated loci and RNA expression were then merged. To prevent the selection of genes with individual variations, genes showing no methylation in the DNA methylation profiles and no change in the expression profiles in one of two individuals were removed. Finally, nine genes that were downregulated by hypermethylation and four genes that were upregulated by hypomethylation in the cancer tissues were selected. Five genes that were downregulated by hypermethylation and one gene that was upregulated by hypomethylation in the pre-cancer tissues were also selected (Table I).

Next, the selected genes were analyzed using a publicly available database. Firstly, it was determined whether the genes had been reported as cervical cancer-related genes in the CCDB. As shown in Table I, only one of the downregulated genes in the tumor tissues, PPP1R3C, had previously been reported to be downregulated in cervical cancer. Almost all the selected genes were identified as novel genes associated with cervical cancer. Secondly, to elucidate whether the selected genes are expressed differentially in cervical cancer, the RNA expression patterns of genes in the tissue samples from patients with cervical cancer were analyzed using GENT. Four genes, PPP1R3C, PHF21B, TPM1 and PPP1R14A, were highly downregulated in the tumor tissues compared with the normal tissues. In addition, three genes, MAP1LC3A, PRRX1 and SYT11, were moderately downregulated in the tumor tissues. Other genes, CAMK2N1 and ALDH1A3, showed no change or only minimal upregulation in the tumor tissues. These expression patterns were similar to the RNA expression data. The same analysis was conducted using genes upregulated by hypomethylation in cancer and using genes up- or downregulated in pre-cancer. Seven genes were determined as novel genes associated with cervical cancer, since they had not been reported in the CCDB. The majority of those genes, excluding FAT1, which appeared as not applicable in GENT, were up- or downregulated in the tumor tissue samples of patients with cervical cancer. GENT and CCDB data supported the observed association of the selected genes with cervical cancer.

To confirm the negative correlation between DNA methylation and RNA expression profiles associated with the progression of cervical cancer, 19 genes that were selected on the basis of integrated analysis were clustered. Common methylation and expression patterns were visualized by unsupervised hierarchical clustering using an average linkage clustering method. As shown in Fig. 1, a separate cluster was formed by methylation and expression data. A negative correlation of methylation and expression was observed using TreeView. Functional annotation clustering was conducted using DAVID with default conditions to analyze the biological significance of the 19 selected genes. This showed that the selected genes were classified into five clusters to be enriched in the cell junction, membrane/plasma membrane, cytoskeletal part, glycoprotein and ion/metal binding (Table II).

To further confirm whether the selected genes were regulated by DNA methylation, the HeLa cells were treated with a demethylating reagent, and the methylation status and RNA expression level was measured by MS-PCR using each primer set (Table III) and RT-quantitative PCR analysis, respectively. Methylation of selected genes, CAMK2N1, ALDH1A3 and PPP1R3C, was found to be decreased, while demethylation was increased following treatment with the demethylating reagent, as expected. Additionally, RNA expression was increased dose-dependently as a result of the decreased methylation status (Fig. 2). These results indicated that the genes were originally methylated in cervical cancer cells and showed increased demethylation followed by increased RNA expression following treatment with a demethylating reagent.