Data preprocessing and microarray data were downloaded from Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo/) under the accession number GSE79973 (8). This dataset was acquired from the study by He et al (8). Totally, 20 samples were included in this dataset, which consisted of 10 gastric cancer samples and 10 normal samples. We used array Quality package to quality control and limma package to apply raw data in R software. The normalization criteria were quantile normalization. Genes having fold-changes ≥2 and FDR <0.05 were selected as of significantly differential expression (9).

To conduct enrichment analyses, the package clusterProfiler (version 3.5.5) of R (version 3.4.0) was used for analyzing the KEGG pathways and GO processes, and DOSE (version 3.3.2) of R (version 3.4.0) as was used for Gene Set Enrichment Analysis (GSEA) (10,11).

WGCNA is a typical systemic biological method for describing the correlation patterns among genes and identifying modules of highly correlated genes by using average linkage hierarchical clustering coupled with the topological overlap dissimilarity measure based on high-throughput chip data or RNA-Seq data (12). In current study, WGCNA package (version 1.60) in R was used to construct lncRNA-mRNA co-expression network and identify modules based on the expression levels of the differentially expressed mRNAs and lncRNAs. The gene modules were signified by different colors and the grey module showed the genes that cannot be merged (13).

Cis-regulation target genes were predicted based on the nearby genes of lncRNAs and we chose the promoters that located in the 1M bp regions around the lncRNAs as potential target genes by Bedtools and hg19 as reference genome (14). Trans-regulation target genes were predicted by the sequence of lncRNAs based on RBPDP, and relative score >80% was used to select potential target genes (15). Co-expression target genes were predicted by the psych package in R (version 3.4.0) to calculate Pearson correlation coefficient between genes and two lncRNAs. The absolute value of co-expression coefficient >0.7 and P-value <0.05 were used to select co-expression genes. The construction of ceRNA network was as followings: i) We predicted the potential target microRNAs of the lncRNAs through RegRNA 2.0 (16); ii) Then, we analyzed the differential expression of the microRNAs that got from 1 in gastric cancers through The Cancer Genome Atlas (TCGA); iii) Next, we found the potential target genes of the microRNAs based on miRTarBase (17); and iv) Finally, the target genes and lncRNAs were used to construct ceRNA network when the co-expression coefficient was >0.7 (9).

The SGC-7901 human gastric carcinoma cell line were cultured in RPMI-1640 medium (Gibco; Thermo Fisher Scientific, Inc., Waltham, MA, USA), supplemented with 10% fetal bovine serum (Gibco; Thermo Fisher Scientific, Inc.) and incubated in an atmosphere of 5% CO₂ at 37°C. SGC-7901 cell line was obtained from the cell bank of the Committee on Type Culture Collection of the Chinese Academy of Sciences.

For overexpression of LINC00982, the coding region was PCR-amplified from cDNA generated from the SGC-7901 cell line and was subcloned into a PINCO retroviral vector. For knocking down LL22NC03-N14H11.1, we firstly designed the lncRNA specific small interfering (si)RNAs (the sequence is 5′-GCACUCACCUACACGUUUAGG-3′), and cloned it into plko.1 vector. SGC-7901 cells (1×10⁵ per well) were inoculated in six-well plates and cultured for 24 h. The cells were transfected with LINC00982-overexpression and the corresponding negative control, or the siRNAs targeting LL22NC03-N14H11.1 and the corresponding negative control retroviral vectors using Lipofectamine 3000 transfection reagent (Invitrogen; Thermo Fisher Scientific, Inc., Waltham, MA, USA) following the manufacturer's instructions.

An MTT assay was used to determine the cell proliferative capacity after overexpression of LINC00982 or down-regulation of LL22NC03-N14H11.1. In brief, transfected cells (2×10⁴ cells/well) were seeded into 96-well culture plates and cultured in RPMI-1640 medium containing 10% FBS. After culturing the seeding cells for 12, 24, 48 or 72 h, MTT reagent (Sigma-Aldrich; Merck KGaA, Darmstadt, Germany) was added to each well, followed by incubation at 37°C for an additional 4 h. Subsequently, 150 µl dimethyl sulfoxide (Sigma-Aldrich; Merck KGaA) was added to dissolve the crystals for 10 min at 37°C. The spectrometric absorbance at 490 nm was measured by an EnSpire Multimode Plate Reader (PerkinElmer, Inc., Waltham, MA, USA).

Anti-mouse-ki-67-PE flow cytometric antibody (eBioscience; Thermo Fisher Scientific, Inc.) was purchased from eBioscience. And the intracellular staining of ki-67 was performed by the eBioscience™ Foxp3/Transcription Factor Staining Buffer Set (eBioscience; Thermo Fisher Scientific, Inc.), following the manufacturer's instructions. Briefly, cells were incubated with the mixture of fixation/permeabilization concentrate and diluent (at the ratio of 1:3) at 4°C for 2 h. Then after washing with fixation/permeabilization buffer, cells were incubated with anti-mouse-ki-67-PE flow cytometric antibody for 0.5 h in 4°C. Finally, the cells were washed and analyzed by BD FACSCanto^™.

Total RNA was extracted from cells with GenElute^™ Total RNA Purification kit (Sigma-Aldrich; Merck KGaA) according to the manufacturer's instructions. Then the RNA was reversely transcribed into cDNA with PrimeScript^™ RT reagent kit (Takara Bio, Inc., Otsu, Japan). RT-PCR reactions were carried out with SYBR^® Premix Ex Taq^™ II (Takara Bio, Inc.) using an ABI Prism 7700 Sequence Detector. Relative mRNA expression levels were calculated by normalizing the relative cycle threshold value to the control group after normalization to the internal control, β-actin and then the results of the semi-quantitative RT-PCR were quantified (18). The primer pairs used are as follows: Human-LL22NC03-N14H11.1-Foward: 5′-GAGTCTGGGGATCAGCATCG-3′, Human-LL22NC03-N14H11.1-Reverse: 5′-TCCAGGGGGCTGGATAATGA-3′; Human-LINC00982-Forward: 5′-AAGTCGTGCTGAGTGTCTGG-3′, Human-LINC00982-Reverse: 5′-CACAACGTGCCACGAACAAT-3′; Human-β-actin-Forward: 5′-CAGGGCGTGATGGTGGGCA-3′, Human-β-actin- Reverse: 5′-CAAACATCATCTGGGTCATCTTCTC-3′.

Statistical analyses were performed using R (version 3.4.0; http://www.r-project.org/). The GSE15459 dataset, which includes 200 gastric cancer samples, together with lncRNA and mRNA expression and clinical information, were used to analyze the associations between lncRNA expression signatures and the corresponding overall survival in patients with gastric cancer (19). Survival curves were constructed with the Kaplan-Meier method and the log-rank test was used to determine survival differences between groups. The survival data were evaluated by univariate and multivariate Cox regression analyses to search for independent prognostic factors. P<0.05 was considered to indicate a statistically significant difference (20). When two independent groups were compared, if the data was normally distributed, an unpaired Student's t-test was used.

Based on the qualified data from He et al study (8), we identified the lncRNAs related to gastric cancer through GenCode V24 (21). By the criteria of FDR <0.05 and fold-change >2, we identified 953 mRNAs and 50 lncRNAs differentially expressed in gastric cancer compared with para-carcinoma tissues. As a result, 499 mRNAs and 14 lncRNAs were upregulated, and 454 mRNAs and 36 lncRNAs were downregulated (Fig. 1A and B).

We assumed that the 50 differently expressed lncRNAs we identified might be involved in various cellular pathways in gastric cancer. To figure out the issue, we firstly run GSEA to identify the dysregulated pathways and BP in gastric adenocarcinoma. Through KEGG pathways, we found 9 upregulated enrichment pathways, including DNA replication, Proteasome, Nucleotide excision repair and 11 downregulated enrichment pathways, including Neuroactive ligand-receptor interaction, Pancreatic secretion, Gastric acid secretion and so on (Fig. 1C). Further, we also conducted GO BP analysis to assign pathways and functionally classified the dysregulated genes. As shown in Fig. 1D, 20 GO BP term, such as mitotic cell cycle phase transition, and ncRNA metabolic process, were significantly enriched.

Co-expression analyses of protein-coding RNAs and lncRNAs reflect the potential function of lncRNAs (22). To further confirm the functions of lncRNAs in gastric cancer, we used WGCNA to construct a co-expression network for both mRNAs and lncRNAs that were identified as differentially expressed. We identified 6 co-expression modules in which the highly co-expressed mRNAs and lncRNAs were clustered in the same module (Fig. 2A) and we specifically identified 14 differently expressed lncRNAs in all the modules (Fig. 2B and Table I). The co-expression network from WGCNA as showed in Fig. 2C and hub-genes in different modules were listed in Table II. In order to further explore the functions of the module, we performed GO enrichment and KEGG pathway analysis and found several dysregulated pathways, such as mitotic nuclear division, extracellular structure organization, extracellular matrix organization, ECM-receptor interaction or cell cycle, which were key signal pathways in cancer generation or development (Fig. 3A and B).

To further investigate the role of these lncRNAs in gastric cancer, we compared the expression of the lncRNAs from different modules, and chose the lncRNAs which changed similarly in different datasets for further analysis (Table III). Then we used GSE15459 to identify lncRNAs associated with clinical outcome (19). Univariate cox survival analysis was performed to analyze the association of clinico-pathological variables, including sex, age, clinical stage, and the lncRNAs' expression with clinical prognosis. Moreover, further multivariate cox analysis showed that LINC00982 expression (P=0.026) and LL22NC03-N14H11.1 expression (P=0.015) were independent prognostic indicators for gastric cancer patients' overall survival (Table IV). Kaplan-Meier survival analysis was used to show the relationship of the two lncRNAs expression level and pathology grade on patients' survival time (Fig. 4A and B).

Next, we explored the potential mechanisms of LINC00982 and LL22NC03-N14H11.1 in the regulation of gastric cancer cells. Firstly, we forecasted the two lncRNAs potential target genes by cis or trans regulation analysis (Tables V and VI). We found LINC00982 could regulate target genes through potential ceRNA network (Fig. 5), but not the same with LL22NC03-N14H11.1. The microRNAs that could bind to LL22NC03-N14H11.1 were not differently expressed in gastric cancers, and these microRNAs were not co-expressed with the target genes of LL22NC03-N14H11.1. This was not accord with the principles of ceRNA network and suggested that LL22NC03-N14H11.1 could not regulate gastric cancer through ceRNA regulation. Then we used enrichment analysis based on the potential target genes to forecast potential pathways that the two lncRNAs may be involved in. The results showed that both LINC00982 and LL22NC03-N14H11.1 may be involved in drug metabolism, ECM-receptor interaction, cell cycle, focal adhesion and other pathway in cancer (Table VII). According to previous studies, cell cycle, ECM-receptor interaction and focal adhesion are important pathways to regulate cell proliferation which play an important role in different cancers, so we further inquiried the ability of the two lncRNAs in regulating tumor cell proliferation (23,24).

To further confirm the functions of LINC00982 and LL22NC03-N14H11.1 in the regulation of gastric cancer cells' proliferation, we decided to construct overexpressed or down-regualted cell lines for the two lncRNAs. As our previous bioinformatics analysis revealed that LINC00982 was reduced while LL22NC03-N14H11.1 was increased in gastric cancer cells (Table III), we overexpressed LINC00982 and knocked down LL22NC03-N14H11.1 in gastric cell lines SGC-7901 respectively (Fig. 6A). Then we performed MTT assay and flow cytometric analysis and found that the proliferation as well as the expression of ki-67, which acts as a key marker for cell mitosis, were impaired in LINC00982 overexpressed and LL22NC03-N14H11.1 down-regulated SGC-7901 cell lines respectively (Fig. 6B and C). These experimental results provided solid evidence to support our conclusion that LINC00982 inhibited while LL22NC03-N14H11.1 promoted the proliferation of gastric cancer cells.