To analyze the differentially expressed genes between HG-EOCs and LG-EOCs, EOC datasets were downloaded from the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) database, using the keywords ‘ovarian cancer’ and ‘GPL570’. Subsequently, four public microarray datasets, including GSE26193 (14), GSE63885 (15), GSE30161 (16) and GSE9891 (17), were selected based on the following criteria (11): i) Expression profiling by array; ii) samples obtained from from Homo sapiens; iii) availability of raw CEL files; iv) GPL570 platforms; and v) EOC samples associated with EOC grades. As a result, the GSE26193, GSE63885, GSE30161 and GSE9891 datasets, consisting of 107, 80, 54 and 280 EOC samples associated with EOC grades, respectively, were included in the present study. The GSE26193 dataset consisted of 40 LG-EOCs and 67 HG-EOCs samples, GSE63885 contained 10 LG-EOCs and 70 HG-EOCs samples, GSE30161 included 21 LG-EOCs and 33 HG-EOCs samples, and GSE9891 was composed of 119 LG-EOCs and 161 HG-EOCs samples.

The publicly available raw CEL files downloaded from the GEO database pre-treated by robust multichip average (RMA) analysis in the affy package (version 3.9; http://www.bioconductor.org/packages/release/bioc/html/affy.html). In order to analyze dysregulated genes in each dataset, the limma package (version 3.9; http://bioconductor.org/packages/release/bioc/html/limma.html) was used. The upregulated or downregulated probes, where P<0.05 and log2 fold change (FC)>1 (upregulated genes) or <-1(downregulated genes), were listed. Venn diagrams (bioinfogp.cnb.csic.es/tools/venny) were used to analyze the consistently differentially expressed genes in the datasets in the present study. To further expand the sample size, the InSilicoMerging (18) approach was used to merge the normalized datasets selected for inclusion in the current study and the ‘RankProd’ (19) approach was applied to identify the dysregulated genes in the merged datasets.

Gene Ontology analysis for the list of differentially expressed genes identified by RankProd was performed to identify their prevalence in biological processes and in molecular functions and pathways, using the Database for Annotation, Visualization and Integrated Discovery (DAVID Bioinformatics Resources; version 6.8; http://david.ncifcrf.gov) (20). The ggplot2 package (21) was used to visualize the main functional pathways of dysregulated genes.

A total of 82 EOC tissue samples (including 36 LG-EOCs and 46 HG-EOCs) were collected from patients with an age range of 35–73 years who had received surgery at The First Affiliated Hospital of Nanjing Medical University (Nanjing, China) between July 2011 to December 2018. The resected tissues were assessed by histological analysis. The patients enrolled had been histopathologically diagnosed with primary ovarian cancer, and had not received any other treatment prior to surgical resection. The tissue samples were immediately stored at −80°C until subsequent analysis. Patients were followed up every 3 months after surgery, during which no patients were lost to follow-up. The follow-up information was recorded comprehensively. Written informed consent for the collection and analysis of tissue specimens in the present study was obtained from every patient; the study was approved by the Research Ethics Committee of Nanjing Medical University. The clinical and pathological characteristics of the patients with EOC are presented in Table I.

Total RNA in the EOC tissues was extracted by TRIzol^® reagent (Invitrogen, CA, USA) according to the manufacturer's protocol. The cDNA reactions prepared using the reverse transcriptase kit (Takara Bio, Inc.) according to the manufacturer's protocol. The mRNA expression of CCCTC-binding factor like (CTCFL) was detected using a standard SYBR Green permix Ex Taq kit (Takara Bio, Inc.) on the 7900 HT real-time instrument (Applied Biosytems; Thermo Fisher Scientific, Inc.). The amplification of CTCFL was performed with an initial step at 94°C for 30 sec, followed by 40 cycles of denaturation at 95°C for 5 sec, annealing at 60°C for 30 sec, and extension at 95°C for 15 sec. The sequences of the primers used were as follows: CTCFL forward, 5′-GTACTCCCCGCAAGAGATGG-3′ and reverse, 5′-TCACCGCTAACTTACTGTCTTCA-3′; and GAPDH forward, 5′-CCCACTCCTCCACCTTTGAC-3′ and reverse, 5′-GGATCTCGCTCCTGGAAGATG-3′. CTCFL mRNA levels were quantified using the 2^−ΔΔCq method and normalized to GAPDH (22).

Immunohistochemical analysis was performed to detect CTCFL protein expression in the tissue specimens. Analysis revealed that CTCFL was the most upregulated gene in the four datasets and was subsequently selected for immunohistochemical analysis. Briefly, paraffin-embedded tissue blocks were cut into 4-µm thick sections and then placed in a constant temperature box at 65°C for 30 min to deparaffinize. The sections were submerged in the ethylenediaminetetraacetic acid buffer and microwaved for 8 min for antigenic retrieval. 3% hydrogen peroxide in methanol was used to quench the endogenous peroxidase activity. Then 1% goat serum albumin (Abcam) was incubated at room temperature for 5 min to block nonspecific binding. The sections were subsequently stained with an anti-BORIS (CTCFL) primary antibody (1:200; cat. no. ab187163; Abcam) and incubated overnight at 4°C, and horseradish peroxidase-conjugated goat anti-rabbit IgG secondary antibody (1:1,000; cat. no. ab6721; Abcam) were then incubated at room temperature for 1 h. Finally, images were captured using a high-capacity digital slide scanner (Pannoramic SCAN, 3DHISTECH) at ×200 magnification. The sections were evaluated independently by two experienced pathologists. A total of 12 patients with paired serous ovarian cancer patients, were selected from patients enrolled in our study to conduct this experiment.

SPSS software (version 18.0; SPSS, Inc., Chicago, IL, USA) and GraphPad Prism software (version 5; GraphPad Software, Inc., La Jolla, CA, USA) were used for all statistical analysis. Data are presented as the mean ± standard error of mean (SEM). The differences between groups were analyzed by the Student's t-test. For the analysis on the clinicopathological parameters, χ² test and Fisher exact probability method were applied. And Kaplan-Meier analysis and the log-rank test were used for survival analysis. P<0.05 was considered to indicate a statistically significant difference. All experiments were repeated in triplicate.

The present study initially analyzed the differentially expressed genes between HG-EOCs and LG-EOCs in each dataset with the limma software package. The three datasets (GSE26193, GSE63885 and GSE30161) were then employed to analyze the consistently differential expressed genes (Fig. 1A). As illustrated in Fig. 1B, there were 90, 733 and 1,489 upregulated genes in the GSE26193, GSE30161 and GSE63885 datasets, respectively. CTCFL (fold change/FC=2.676; percentage of false prediction/pfp<0.01), EGFL6 (FC=2.140; pfp<0.01), radical S-adenosyl methionine domain containing 2 (FC=1.776; pfp<0.01) and SAM and HD domain containing deoxynucleoside triphosphate triphosphohydrolase 1 (SAMHD1; FC=1.639; pfp<0.01) were consistently upregulated among these three datasets. Furthermore, 415, 79 and 41 genes were identified as downregulated in the GSE26193, GSE30161 and GSE63885 datasets, respectively. Only ALK and LTK ligand 2 (FC=0.504; pfp<0.01) was downregulated among the three aforementioned datasets (Fig. 1C). In order to decrease the number of differentially expressed genes, and therefore identify more reliable potential molecular markers for EOC grade, the present study superadded an additional dataset (GSE9891). As presented in Fig. 2, only one overlapping candidate probe, CTCFL, was identified to be upregulated in the four datasets and no probes were consistently downregulated in the four examined datasets.

To further enlarge the sample size, the present study initially merged the three datasets by using the InSilicoMerging method. The ‘RankProd’ approach was subsequently used to analyze the differentially expressed genes in the merged datasets. As a result, a total of 6,103 upregulated probes corresponding with 5,766 genes (FC>1; pfp<0.01), and 4,004 downregulated probes corresponding with 3,707 genes (FC<1; pfp<0.01) were identified from the GSE26193, GSE63885 and GSE30161 datasets. The top 20 upregulated and downregulated genes are presented in Tables II and III, respectively.

Consistent with the intersection results among the aforementioned four datasets, CTCFL was the most upregulated gene in HG-EOCs from the merged GSE26193, GSE63885 and GSE30161 datasets (Table I). Therefore, CTCFL may serve as a potent biomarker and CTCFL was subsequently selected as a candidate gene for distinguishing HG-EOCs from LG-EOCs.

The top 100 dysregulated probes in the merged datasets are presented in Fig. 3, and the hierarchical cluster analysis revealed that the expression profiles of HG-EOCs were similar to those of LG-EOCs.

To effectively analyze the function of the dysregulated genes in the merged datasets, DAVID was utilized to process the functional enrichment analysis. The biological process (BP) terms associated with the top 3,000 dysregulated genes were downloaded. There were 370 significant BP terms associated with the genes upregulated in HG-EOCs and 380 significant BP terms associated with the genes that were downregulated in HG-EOCs. The upregulated genes were enriched in the ‘positive regulation of transcription from RNA polymerase II’, whilst the downregulated genes were enriched in ‘signal transduction’ and ‘positive regulation of transcription from RNA polymerase II promoter’. The top 20 BP terms are presented in Fig. 4A and B.

In order to verify the differential expression of CTCFL in LG-EOCs and HG-EOCs, RT-qPCR was performed in 36 LG-EOCs and 46 HG-EOCs tissue samples. As illustrated in Fig. 5A, the mRNA level of CTCFL was significantly upregulated in HG-EOCs compared with LG-EOCs (P=0.0007). This results indicated that upregulated CTCFL may be implicated in HG-EOCs. In addition, the present study divided the 82 patients with EOC into two groups on the basis of CTCFL expression in tumor tissues (Fig. 5B; cut-off, 0.1152275). By using the log-rank test, patients with high expression levels of CTCFL were observed to have a poor outcome compared with patients with low expression levels of CTCFL (P=0.0084, Fig. 5C). Notably, the association between clinical pathological characteristics and CTCFL expression revealed that high expression levels of CTCFL were significantly associated with tumor size (P=0.0344), the International Federation of Gynecology and Obstetrics stage (P=0.0343) and histological grade (P=0.0004). However, highly-expressed CTCFL was not associated with the other examined clinical characteristics (Table I).

For further validation, the protein expression of CTCFL in LG-EOCs EOCs and HG-EOCs was analyzed by immunohistochemistry. As hypothesized, the protein expression of CTCFL in HG-EOCs samples was markedly higher than LG-EOCs samples (Fig. 6A and B). Taken together, these results suggested that CTCFL was associated with EOC grade and may serve as a promising biomarker and therapeutic target for HG-EOCs.