Three sets of primary GC tissue samples (the GC group) and paired, adjacent noncancerous tissues (the control group), which were ≥5 cm away from cancerous tissue, were obtained from patients who had undergone curative surgical resection at the Affiliated Hospital of Hainan Medical University (Haikou, China) from June 2014 to July 2014. Histological diagnoses were made from formalin-fixed, paraffin-embedded tissues by two pathologists. The clinicopathological characteristics were obtained from medical records, including gender, age, tumor size, tumor location, histological type, Lauren classification, differentiation grade and surgical record. None of the patients received neoadjuvant therapy. Written informed consent was obtained from the patients for the use of their samples for research, and the research protocols were approved by the Ethics Committee of the Affiliated Hospital of Hainan Medical University.

A total of 100 mg tissue was taken from each GC and paired normal mucosal tissues, and were homogenized using a TissueLyser II BioRobot Universal system (Qiagen GmbH, Hilden, Germany). Total RNA was isolated using TRIzol reagent (Invitrogen; Thermo Fisher Scientific, Inc., Waltham, MA, USA). RNA was purified using the RNeasy Mini kit (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol. Subsequently, RNA quality and quantity were measured using a NanoDrop spectrophotometer (ND-1000; NanoDrop; Thermo Fisher Scientific, Inc., Wilmington, DE, USA), and RNA integrity was determined by gel electrophoresis.

The Arraystar Human Circular RNA Microarray V2.0 (Arraystar, Inc., Rockville, MD, USA) is designed for global profiling of human circRNAs. This microarray is comprised of ~13,617 circRNAs with stringent experimental support, which was carefully and comprehensively collected from circRNA studies and landmark publications (10,11). Each circRNA is represented by a circular splice junction probe, which can reliably and accurately detect the circRNA, even in the presence of its linear counterparts. A random primer-based labeling system is coupled with RNase R sample pretreatment to ensure specific and efficient labeling of circRNAs. RNA spike-in controls were included to monitor labeling and hybridization efficiencies. Microarray hybridization and bioinformatic analysis were performed by KangChen Bio-tech, Inc. (Shanghai, China).

RNA from three pairs of selected samples was subjected to microarray analysis according to the manufacturers protocol (Arraystar, Inc.). Briefly, total RNA was digested with RNase R (Epicentre; Illumina, Inc., Madison, WI, USA) to remove linear RNAs and enrich for circRNAs. Subsequently, the enriched circRNAs were amplified and transcribed into fluorescent cRNA using a random priming method (Arraystar Super RNA Labeling kit; Arraystar, Inc.). The Cy3-labeled cRNAs were purified using the RNeasy Mini kit (Qiagen GmbH). Each labeled cRNA (1 µg) was fragmented by adding 5 µl 10X blocking agent and 1 µl 25X fragmentation buffer, after which the mixture was incubated at 60°C for 30 min and was then diluted in 25 µl 2X hybridization buffer. A total of 50 µl hybridization solution was dispensed into the gasket slide, which was assembled onto the circRNA expression microarray slide. The slides were incubated for 17 h at 65°C in a hybridization oven (Agilent Technologies, Inc., Santa Clara, CA, USA). The hybridized arrays were washed, fixed and scanned using an Agilent Microarray Scanner system (catalog no. G2505C; Agilent Technologies, Inc.).

Scanned images were imported into Agilent Feature Extraction software version 11.0.1.1 (Agilent Technologies, Inc.) for raw data extraction. Quantile normalization of raw data and subsequent data processing were performed using the R software package (www.arraystar.com/arraystar-human-circular-rna-microarray). Following quantile normalization of the raw data, low-intensity filtering was performed and circRNAs having the ‘P’ or ‘M’ flags (‘All Targets Value’) in ≥3 out of 6 samples were retained for further analyses. Subsequently, samples were clustered hierarchically with Cluster software version 2.0 (http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm) to evaluate the robustness of the formed clusters, using the correlation-centered metric and average linkage-clustering algorithm. When comparing profile differences between the GC and control groups, the fold-change between the groups for each circRNA was computed. The statistical significance of each difference was estimated by Student's t-test. CircRNAs exhibiting a ≥2-fold change in expression (P<0.05) were considered to be significantly differentially expressed.

The present study initially identified circRNAs that exhibited a >2-fold change in expression by microarray analysis. Subsequently, circRNA/miRNA interactions were predicted using Arraystar in-house generated miRNA target-prediction software (Arraystar, Inc.), based on TargetScan (16) and miRanda (17). Differentially expressed circRNAs were used as seeds to enrich for a circRNA-miRNA-gene network, based on a cutoff value determined using miRNA support-vector regression (mirSVR), as described by Wang et al (18). The predicted gene functions in the networks were annotated using GO (http://www.geneontology.org/) and the Database for Annotation, Visualization and Integrated Discovery (DAVID; https://david.ncifcrf.gov/).

Total RNA from three GC specimens and matched, adjacent normal tissues were reverse transcribed to cDNA using the FastQuant RT kit with gDNase (Tiangen Biotech Co., Ltd., Beijing, China) in 20-µl reactions. Triplicate qPCR assays were performed in 20-µl reactions using the FastFire qPCR PreMix (SYBR-Green) kit (Tiangen Biotech Co., Ltd.) according to the manufacturer's protocol. The thermal cycling conditions were as follows: Initial denaturation at 95°C for 60 sec, 40 cycles of amplification at 95°C for 20 sec, annealing and extension at 60°C for 30 sec. GAPDH was used as an internal control for PCR amplification. The sequences of circRNAs were obtained from the circBase database (http://circbase.org/), and PCR primers were designed in divergent orientation, in order to be capable of amplifying the reverse splice site of the circRNA, using Primer Premier software version 6.0 (Premier Biosoft International, Palo Alto, CA, USA) and were synthesized by Sangon Biotech Co., Ltd. (Shanghai, China) (Table I). The data were analyzed using the 2^−ΔΔCq method (19). Validation was performed by determining the ratio of differential cellular expression/differential microarray expression, and amplification products were analyzed by 1.5% agarose gel electrophoresis and stained with 0.1% GeneGreen Nucleic Acid gel stain (Tiangen Biotech Co., Ltd.) for band size consistency. The qPCR data were analyzed by Student's t-test using SPSS software version 18.0 (SPSS, Inc., Chicago, IL, USA). P<0.05 was considered to indicate a statistically significant difference.

The present study investigated the alterations in circRNA expression profiles between the GC group and the control group. To obtain consistent biological information, paired samples from 3 patients with similar clinical data were selected for microarray analysis. All GC cases were of the diffuse type (Lauren classification), male, stage IIIA (TNM staging system), and aged 55–58 (average age, 56.7 years). Total RNA was subsequently extracted, and circRNA expression in the groups was analyzed using an Arraystar microarray.

To determine whether circRNA profiles were informative with regards to tissue type, an unsupervised, hierarchical cluster analysis was conducted based on the circRNA expression levels in 3 GC specimens and adjacent normal tissues. The results of hierarchical clustering showed distinguishable circRNA expression profiling among 6 samples. Two main clusters were formed, the GC group cluster and the control group cluster (Fig. 1A). These data indicated that circRNAs have a different expression pattern in GC compared with in normal gastric mucosa.

The circRNA chip detected >2,000 circRNAs expressed in GC tissues and adjacent normal tissues, and 100 circRNAs (86 exonic circRNAs, 12 intronic circRNAs and 2 antisense circRNAs) that were ≥2-fold differentially expressed between the GC and the matching, adjacent normal tissues (Fig. 1B), which were distributed across all 22 autosomes. A total of 16 circRNAs were significantly upregulated and 84 were significantly downregulated in the GC group, compared with the control group. The circRNAs were ranked according to fold-changes in expression between the groups. The top 5 upregulated and downregulated circRNAs are presented in Table II.

To validate the expression profiles of circRNAs in three matched pairs of GC and normal specimens, the top 5 down- and upregulated circRNAs were further studied by qPCR (Fig. 2). The primers were specifically capable of amplifying the back splice sites of each circRNA, and the PCR results were validated by gel electrophoresis and melt-curve analysis (Fig. 2B and C); melting curves for qPCR assays of circRANs showed a single peak indicated the specificity of the PCR results, and the sizes of the PCR products were in accordance with the anticipated sizes, based on the primer designs. The expression levels of 7 selected circRNAs were consistent with those measured by microarray analysis; however, 2 circRNAs were not detected, perhaps due to the extremely low expression levels in the 6 samples, and differential expression was not detected with 1 circRNA (P>0.05; Fig. 2A). Of these, only circRNA-0026 expression was significantly different between the GC and control samples (2.8-fold, P=0.001), and all other differences were less than 1.5 fold.

It has previously been indicated that circRNAs regulate gene expression by targeting miRNAs and blocking their biological functions (7,10). In the present study, miRNA and circRNA sequences were aligned, and 5 miRNAs with the highest mirSVR were identified for each differentially expressed circRNA using miRNA target-prediction software, based on TargetScan and miRanda. All differentially expressed circRNAs were annotated in detail, with the circRNA/miRNA interaction information. As a result, 327 miRNAs (including 221 miRNA families) were identified and ranked according to their predicted number of interactions with different circRNAs, as described by Wang et al (18); the top 10 miRNA families consisted of the miR-29, miR-30, miR-15, miR-146, miR-16, miR-181, miR-135, miR-23, miR-19 and miR-let-7 families. Bioinformatics analysis revealed that these miRNAs were abnormally expressed in various malignant tumors, including GC (20), colon cancer (21), breast cancer (22), lung cancer (23) and others (24–26), thus suggesting their involvement in the occurrence and development of malignant tumors. Subsequently, candidate target genes were identified based on miRNA/mRNA sequence pairing (mRNA-dependent cutoff value, −0.50) using TargetScan online tools, and a circRNA-0026-targeted circRNA-miRNA-mRNA/gene network was constructed. It was demonstrated that 13 miRNAs and 578 genes were targeted in the network (miRNA-dependent cutoff value, −0.14; mRNA-dependent cutoff value, −0.50). To investigate the functions of the predicted network genes, GO analysis was performed using the DAVID tool. Genes in the top 10 annotated clusters were involved in regulating transcription, including RNA metabolic processes, gene expression, gene silencing and other biological functions (Table III).