Integrative Analysis of Dysregulated lncRNA-Associated ceRNA Network Reveals Functional lncRNAs in Gastric Cancer

Mounting evidence suggests that long noncoding RNAs (lncRNAs) play important roles in the regulation of gene expression by acting as competing endogenous RNA (ceRNA). However, the regulatory mechanisms of lncRNA as ceRNA in gastric cancer (GC) are not fully understood. Here, we first constructed a dysregulated lncRNA-associated ceRNA network by integrating analysis of gene expression profiles of lncRNAs, microRNAs (miRNAs), and messenger RNAs (mRNAs). Then, we determined three lncRNAs (RP5-1120P11, DLEU2, and DDX11-AS1) as hub lncRNAs, in which associated ceRNA subnetworks were involved in cell cycle-related processes and cancer-related pathways. Furthermore, we confirmed that the two lncRNAs (DLEU2 and DDX11-AS1) were significantly upregulated in GC tissues, promote GC cell proliferation, and negatively regulate miRNA expression, respectively. The hub lncRNAs (DLEU2 and DDX11-AS1) could have oncogenic functions, and act as potential ceRNAs to sponge miRNA. Our findings not only provide novel insights on ceRNA regulation in GC, but can also provide opportunities for the functional characterization of lncRNAs in future studies.


Introduction
Gastric cancer (GC) is one of the leading causes of cancer-related death worldwide, and imposes a considerable global health burden [1,2]. In 2012, an estimated 723,100 deaths from GC and nearly half of the burden occurred in Eastern Asia, particularly in China, Korea, and Japan [3]. In China, GC was the second most common cancer, as reported by the National Central Cancer Registry (NCCR) [4].
Previous studies have demonstrated that coding RNAs and non-coding RNAs (ncRNAs) are closely related to cancer [5]. MicroRNAs (miRNAs) are an important subclass of endogenous non-coding RNAs with a typical length of 23 nucleotides [6], and play crucial roles in cancer by regulating gene expression at the post-transcriptional level [7]. Long non-coding RNAs (lncRNAs), Genes 2018, 9,303 2 of 17 another important subclass of endogenous non-coding RNAs, are longer than 200 nucleotides, and lack protein-coding capability [8]. Similar to coding RNAs and miRNAs, some lncRNAs can drive cancer phenotypes [9]. However, the complete spectrum of mechanisms by which lncRNAs affect cancer development and pathology remains to be fully elucidated.
Recent studies have demonstrated that lncRNA can act as a miRNA sponge via competing endogenous RNA (ceRNA) activity [10,11]. Understanding this novel RNA crosstalk will lead to significant insights into gene regulatory networks and have implications in human development and disease [12]. For example, PTENP1, a pseudogene of the tumor-suppressor gene PTEN, is among the first reported miRNA sponges with a function in cancer [13]. BRAF is an oncogene and its overexpression leads to enhanced activity of the MAPK signaling pathway [14]. Overexpression of the BRAF pseudogene results in upregulation of BRAF. HOTAIR acts as an miRNA sponge to regulate HER2 expression by sponging hsa-miR-331-3p in GC [15]. GAPLINC acts as a ceRNA for hsa-miR-211-3p, and mediates cell migration and proliferation in GC [16]. KRTAP5-AS1 and TUBB2A participate in the regulatory network of CLDN4 based on ceRNA activity in GC [17]. MEG3 can acts as a ceRNA to regulate GC progression [18].
Based on the ceRNA hypothesis, comprehensive analyses of lncRNA-associated ceRNA networks have been reported in ovarian cancer [19], glioblastoma [20], prostate cancer [10], thyroid carcinoma [21], and breast cancer [22]. Xia et al. [23] constructed a cancer-associated ceRNA network in GC based on lncRNA microarray data. Li et al. [24] showed the GC-specific lncRNA expression patterns, and constructed a ceRNA network based on The Cancer Genome Atlas (TCGA). Although previous studies have identified ceRNAs in GC, the functional roles and regulatory mechanisms of lncRNAs as ceRNAs in GC are not fully understood.
In this study, we first systematically integrated gene expression profiles of GC, miRNA target interactions, and significantly co-expressed gene pairs based on the ceRNA hypothesis. We then constructed dysregulated lncRNA-associated ceRNA network. After the network topological analysis, we determined that the hub lncRNAs play crucial roles in GC development. By reconstructing and functionally analyzing the hub lncRNA-associated subnetworks, we identified three hub lncRNAs (RP5-1120P11, DLEU2, and DDX11-AS1) that are involved in cell cycle-related processes and cancer-related pathways. Finally, we validated that DLEU2 and DDX11-AS1 are significantly upregulated in GC tissues by using qRT-PCR. Cellular experimental results indicated that the two lncRNAs promote GC cell proliferation and negatively regulate miRNA expression. Through an integrative analysis of dysregulated lncRNA-associated ceRNA network ( Figure S1), we determined that hub lncRNAs (DLEU2 and DDX11-AS1) have oncogenic functions, and act as potential ceRNAs to sponge miRNA. Our findings can be used to reveal ceRNA regulation in GC, and provide novel insights on functional characterization of lncRNAs.

GDC Data Collection and Processing
Gastric cancer gene expression data were obtained from Genomic Data Commons (GDC, https: //cancergenome.nih.gov/) [25]. The upper quantile normalized FPKM (fragments per kilobase per million) values and the normalized count in reads-per-million-miRNA-mapped values were used. Patients with follow-up time exceeding three years were excluded. Overall, a total of 364 samples, including 337 GC samples and 27 normal samples, were included in our study and each sample consisted of the corresponding RNA-seq data and miRNAseq data.
To assemble a lncRNA-associated ceRNA data, we downloaded the RNA gene annotation and sequence data from GENCODE v26 (https://www.gencodegenes.org/), which aims to identify all gene features in the human genome using a combination of computational analysis, manual annotation, and experimental validation [26]. Transcripts longer than 200 nt and categorized as "lincRNA", "antisense", "processed_transcript", and "non_coding" were identified as lncRNAs. Genes categorized Genes 2018, 9, 303 3 of 17 as "protein_coding" were considered messenger RNAs (mRNAs). In total, 19,799 mRNAs and 13,545 lncRNAs in 364 samples were annotated Genes expressed in more than 90% of samples were retained, and the values of zero or NA in the expression profile were replaced with the (minimum non-zero FPKM value)/2 of the corresponding genes. All the expression profile values were then log2 transformed. Therefore, the expression profiles of 16,363 mRNAs, 440 miRNAs, and 4211 lncRNAs in 364 samples were selected for further analysis.

Identifying Differentially Expressed mRNAs, lncRNAs, and miRNAs Related to GC
To identify differentially expressed lncRNAs (DELs), mRNAs (DEMs), and miRNAs (DEMis), we used the two-tailed Mann-Whitney U test to compare the expression values in cancer with those in normal. For all p-values, FDR (false discovery rate) was used to correct the statistical significance of multiple testing by the Benjamini-Hochberg method. In this paper, we mainly focused on the upregulated lncRNAs, upregulated mRNAs, and downregulated miRNAs, which could have oncogenic functions based on ceRNA activity. Therefore, we defined differentially expressed lncRNAs/mRNAs with fold change ≥1 and adjusted p-value <0.001 as DELs/DEMs. In order to avoid insufficient number of downregulated miRNAs, we defined differentially expressed miRNAs with fold change ≤−0.6 and adjusted p-value <0.001 as DEMis.

Identification of Potential ceRNA Triples Based on Gene Expression
The competition of lncRNAs with mRNAs to sponge miRNAs is reflected at the gene expression level [31]. These lncRNA-miRNA-mRNA interactions were defined as ceRNA triples, and the potential ceRNA triples were identified as follows: (1) expression correlations among all mRNAs, lncRNAs, and miRNAs in GC were evaluated by the Pearson correlation coefficient (PCC). The PCC of gene pair (X and Y) is defined as where n is the number of GC cancer samples; Exp(X, i) (Exp(Y, i)) is the expression level of X (Y) in sample i; Exp(X) (Exp(Y, i)) denotes the average expression level of X (Y), and σ(X) (σ(Y)) represents the standard deviation of expression level of X (Y). In order to obtain the distribution of the overall PCC in GC, we calculated the PCC between 16,363 mRNAs, 440 miRNAs, and 4211 lncRNAs, respectively. We then obtained the PCC of 68,904,593 mRNA-lncRNA pairs, 1,852,840 lncRNA-miRNA pairs, and 7,199,720 mRNA-miRNA pairs. (2) To obtain the significant co-expressed pairs, PCC higher than the threshold of 95% was considered a significant positively co-expressed pair, and lower than the threshold of 95% was considered as a negatively co-expressed pair. Therefore, DEL-DEM pairs with PCC > 0.292 and p-value < 0.0001 was considered as significantly co-expressed DEL-DEM pairs ( Figure S2A). DEM-DEMi pairs with PCC <−0.275 and p-value <0.0001 were considered as significantly negatively co-expressed DEM-DEMi pairs ( Figure S2B). DEL-DEMi pairs with PCC < −0.233 and p-value < 0.0001 were considered as significantly negatively DEL-DEMi pairs ( Figure S2C). (3) For a given significant co-expressed DEM-DEL pair, both DEM and DEL interacted with the same DEMi, and were negatively co-expressed with the same DEMi, and this DEL-DEMi-DEM interaction was defined as a potential ceRNA triple.

Construction Network and Topological Analysis
To clarify the roles of the dysregulated lncRNA-associated ceRNA network, we constructed a co-expression network of DEL-DEMi-DEM interactions based on the potential ceRNA triples, and visualized this ceRNA network by Cytoscape 3.5.1 [32]. Betweenness centrality (BC) is a measure of centrality in a network based on the number of shortest paths from every node to all others passing through this node. In a network, a node with a higher BC is implied to act as a bridge to connect different nodes and control the network communications. In this study, BC was considered as the topological analysis measure to select the hub lncRNAs. The BC of a node i is defined as where p st is the total number of the shortest paths from node s to node t, and p st (i) is the number of these paths that pass through node i. We assumed that the hub lncRNAs play critical roles in the ceRNA network. Accordingly, we reconstructed the hub lncRNA-associated subnetwork with the nodes, which is the hub lncRNA related to the first-neighbor miRNAs and the second-neighbor mRNAs in the ceRNA network.

Functional Enrichment Analysis
To reveal the function of lncRNA-associated ceRNA network, the DEMs were subjected to functional enrichment analysis via gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) with the R package cluster Profiler [33]. GO is a structured standard biological model established by the GO consortium, which aims to establish a standard system for computing the functions of genes, covering biological processes (BP), molecular functions (MF), and cellular components (CC) [34]. KEGG is widely used as a reference for integrating large-scale molecular datasets generated by sequencing and high-throughput experimental technologies [35]. In this study, we considered functional categories with a p-value < 0.05.

Patient Samples and GC Cell Lines
Gastric cancer tissues and adjacent normal tissues were collected from patients who underwent surgery treatment at the Xijing Hospital (Xi'an, China). Written informed consents had been obtained from all patients who donated tissue samples, which was under the supervision of the Hospital's Protection of Human Subjects Committee. None of our study patients had received chemotherapy or radiotherapy before the surgery, and all tissues were paraffin-embedded after RNA and protein isolation. Human GC cell lines BGC823 and MKN45 were purchased from the Cell Bank of the Chinese Academy of Sciences. Cells were incubated at 37 • C in a humidified atmosphere containing 5% CO 2 and cultured in Dulbecco's Modified Eagle's Medium (ThermoFisher Scientific, Waltham, MA, USA) combined with 10% fetal bovine serum (FBS).

RNA Isolation and Real-Time PCR Analysis
Total RNA was isolated using a Qiagen miRNeasy kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol. Then, reverse transcription of RNA to cDNA was performed using the the Prime Script RT Master Mix (TaKaRa Bio Inc., Kusatsu, Japan). qRT-PCR was carried out using SYBR premix Ex Taq (Takara Bio Inc., Kusatsu, Japan). RNA qRT-PCR PrimerSets specifc for DLEU2, DDX11-AS1, hsa-miR-30-5p, and hsa-miR-145-5p were designed by RiboBio (Guangzhou, China). The expression levels of the RNA and internal control (U6) were measured using qRT-PCR (Bio-Rad, Hercules, CA, USA).

CCK-8 and Colony Formation Assays
Cells were seeded in 96-well plates (1 × 10 4 cells per well), cultivated at 37 • C in a 5% CO 2 humidified incubator for 24 h. Ten microliters of Cell Counting Kit-8 solution (CCK-8) (Dojindo Laboratories, Kumamoto, Japan) were added to each well and incubated at 37 • C in a 5% CO 2 humidified incubator for 2 h. Spectrometer Varioskan Flash (ThermoFisher Scientific) was used to measure the 450 nm absorbance. The CCK-8 assay was performed every 24 h for 5 days. A proliferation curve was drawn with the time as the abscissa, and average absorbance value in each group as the ordinate. For the colony formation assay, Cells were seeded in 6-well plates (1 × 10 3 cells per well), and were incubated at 37 • C in a 5% CO 2 -humidified incubator. After 2 weeks, the cells were stained with Gentian violet (Beyotime Biotechnology, Shanghai, China). Triplicate reactions were performed for the experiment.

Statistical Analysis of Biological Experimental Results
Biological experimental results were analyzed using paired or unpaired t-test (GraphPad Software, La Jolla, CA, USA). Values of p < 0.05 were considered statistically significant (*). ALL biological experimental results were presented as mean ± SD.

Identification of Differentially Expressed lncRNAs, mRNAs, and miRNAs
We obtained the gene expression profiles of 337 GC samples and 27 normal samples from GDC and identified differentially expressed genes. A total of 757 lncRNAs, 2441 mRNAs, and 256 miRNAs were detected. Among them, 571 lncRNAs, 1097 mRNAs, and 227 miRNAs were significantly upregulated, while 186 lncRNAs, 1344 mRNAs, and 29 miRNAs were significantly downregulated in GC (Table S1). In this paper, we mainly focused on the upregulated lncRNAs, upregulated mRNAs, and downregulated miRNAs, which could have oncogenic functions based on ceRNA activity. Therefore, we selected 571 DELs, 1097 DEMs, and 29 DEMis from the significantly differentially expressed set for further analysis. To determine if the DELs, DEMs, and DEMis can be used to separate normal samples from cancer samples, we utilized "ward.D" algorithm and Euclidean distance to cluster these genes with the R package pheatmap. The heatmap indicated that all samples could be clearly separated into two groups, and less than 3% of the cancer samples were incorrectly classified ( Figure 1). These differentially expressed genes were therefore considered as potential key regulators to construct the ceRNA network in GC.

Construction and Functional Analysis of ceRNA Network in GC
To construct the ceRNA network in GC, we focused on the relationships among the DELs, DEMs and DEMis. First, we identified miRNA-target interactions by integrating reliable online databases and prediction software. We then selected 1146 reliable DEM-DEMi interactions from 148,459 miRNA-mRNA interactions and 2794 candidate DEL-DEMi interactions from 29,523 miRNA-lncRNA interactions for further analysis. To reduce false positives, we used the PCC to filter the significant co-expressed DEL-DEMi-DEM interactions. After filtering the interactions, a total of 442 potential ceRNA triples were identified (Table S2). Based on the potential ceRNA triples, we constructed and visualized a dysregulated lncRNA-associated ceRNA network by Cytoscape ( Figure  2A). The ceRNA network was a fully connected network, which consisted of 174 nodes (including 28 DELs, 130 DEMs, and 16 DEMis) and 173 edges. Then, we utilized the DEMs to reveal the gene functions of the dysregulated lncRNA-associated ceRNA network via the functional enrichment analysis. The ceRNA function analysis revealed 250 GO terms in BP and 12 KEGG pathways ( Figure  2B,C, and Table S3). Cancer is characterized by aberrant cell cycle activity [36], and most of these GO terms are cell cycle process related in the ceRNA network. Among the KEGG pathways, cell cycle, p53 signaling pathway, Fanconi anemia pathway, mismatch repair, DNA replication, and platinum drug resistance are cancer-related pathways [36][37][38][39][40].

Construction and Functional Analysis of ceRNA Network in GC
To construct the ceRNA network in GC, we focused on the relationships among the DELs, DEMs and DEMis. First, we identified miRNA-target interactions by integrating reliable online databases and prediction software. We then selected 1146 reliable DEM-DEMi interactions from 148,459 miRNA-mRNA interactions and 2794 candidate DEL-DEMi interactions from 29,523 miRNA-lncRNA interactions for further analysis. To reduce false positives, we used the PCC to filter the significant co-expressed DEL-DEMi-DEM interactions. After filtering the interactions, a total of 442 potential ceRNA triples were identified (Table S2). Based on the potential ceRNA triples, we constructed and visualized a dysregulated lncRNA-associated ceRNA network by Cytoscape ( Figure 2A). The ceRNA network was a fully connected network, which consisted of 174 nodes (including 28 DELs, 130 DEMs, and 16 DEMis) and 173 edges. Then, we utilized the DEMs to reveal the gene functions of the dysregulated lncRNA-associated ceRNA network via the functional enrichment analysis. The ceRNA function analysis revealed 250 GO terms in BP and 12 KEGG pathways ( Figure 2B,C, and Table S3). Cancer is characterized by aberrant cell cycle activity [36], and most of these GO terms are cell cycle process related in the ceRNA network. Among the KEGG pathways, cell cycle, p53 signaling pathway, Fanconi anemia pathway, mismatch repair, DNA replication, and platinum drug resistance are cancer-related pathways [36][37][38][39][40].

Topological Analysis of the lncRNA-Associated ceRNA Network in GC
To clarify the roles of the dysregulated lncRNA-associated ceRNA network, we computed BC, which are topological features of ceRNA network, as shown in Table S4. A node with a higher BC plays a more important role in the ceRNA network, and is more likely to be associated with cancer [19,41]. Among the 174 nodes in the ceRNA network, the top 17 highest BC nodes (including 3 DELs, 3 DEMs, and 11 DEMis) were shown in Figure 3. We selected the three DELs (RP5-1120P11, DLEU2, and DDX11-AS1) with the highest BC values as the hub lncRNAs, and calculated the hub lncRNAs related to the first-neighbor DEMis, the second-neighbor DEMs, and the ceRNA triples. We determined that the hub lncRNAs were related to 87.5% of the DEMis, 87.69% of the DEMs, and

Topological Analysis of the lncRNA-Associated ceRNA Network in GC
To clarify the roles of the dysregulated lncRNA-associated ceRNA network, we computed BC, which are topological features of ceRNA network, as shown in Table S4. A node with a higher BC plays a more important role in the ceRNA network, and is more likely to be associated with cancer [19,41]. Among the 174 nodes in the ceRNA network, the top 17 highest BC nodes (including 3 DELs, 3 DEMs, and 11 DEMis) were shown in Figure 3. We selected the three DELs (RP5-1120P11, DLEU2, and DDX11-AS1) with the highest BC values as the hub lncRNAs, and calculated the hub lncRNAs related to the first-neighbor DEMis, the second-neighbor DEMs, and the ceRNA triples. We determined that the hub lncRNAs were related to 87.5% of the DEMis, 87.69% of the DEMs, and 53.85% of the ceRNA triples in the ceRNA network (Table 1). These results indicated that the hub lncRNAs are potential key regulators controlling the ceRNA network in GC.
Genes 2018, 9, x FOR PEER REVIEW 8 of 17 53.85% of the ceRNA triples in the ceRNA network (Table 1). These results indicated that the hub lncRNAs are potential key regulators controlling the ceRNA network in GC.

Reconstruction and Functional Analysis of Hub lncRNA-Associated Subnetworks in GC
We assumed that the hub lncRNAs (RP5-1120P11, DLEU2, and DDX11-AS1) play critical roles in the GC-related ceRNA network. Accordingly, we reconstructed the hub lncRNA-associated subnetworks with the nodes, which were the hub lncRNAs related to the first-neighbor miRNAs and the second-neighbor mRNAs in the ceRNA network. To reveal the functions of the hub lncRNAs, we performed functional enrichment analysis for each associated subnetwork (Figure 4).

Reconstruction and Functional Analysis of Hub lncRNA-Associated Subnetworks in GC
We assumed that the hub lncRNAs (RP5-1120P11, DLEU2, and DDX11-AS1) play critical roles in the GC-related ceRNA network. Accordingly, we reconstructed the hub lncRNA-associated subnetworks with the nodes, which were the hub lncRNAs related to the first-neighbor miRNAs and the second-neighbor mRNAs in the ceRNA network. To reveal the functions of the hub lncRNAs, we performed functional enrichment analysis for each associated subnetwork (Figure 4).
DDX11-AS1 is upregulated in multiple cancer types and affects sister chromatid cohesion [50]. The subnetwork related to DDX11-AS1 consisted of 6 DEMis and 45 DEMs ( Figure 4C). The function analysis of the DDX11-AS1-associated subnetwork revealed 204 GO terms in BP and 8 KEGG pathways ( Figure 4F, Table S3). We determined that most of these GO terms are cell cycle process-related and most pathways are cancer-related pathways, such as cell cycle, Fanconi anemia pathway, mismatch repair, DNA replication, and p53 signaling pathway.
All these results will provide us important cell cycle processes and cancer-related pathways information for understanding lncRNA-associated ceRNA network in GC.
All these results will provide us important cell cycle processes and cancer-related pathways information for understanding lncRNA-associated ceRNA network in GC.   The scatter plots showed the correlation of expression between DLEU2 and hsa-miR-30-5p; (C) The scatter plots showed the correlation of expression between DLEU2 and DLEU2 related DEMs; (D) The scatter plots showed the correlation of expression between DLEU2 related DEMs and hsa-miR-30-5p. All the data were obtained from GDC.

DLEU2 and DDX11-AS1 Are Upregulated in GC Tissues, Promote GC Cell Proliferation, and Inhibit miRNA Expression, Respectively
To validate the findings, we first examined DLEU2 and DDX11-AS1 expression level in a cohort of 12 paired GC tissues, and compared with those of the adjacent normal tissues by using qRT-PCR. The DLEU2 and DDX11-AS1 expression levels were significantly upregulated in GC tissues ( Figure  6A). siRNA-mediated silencing of lncRNAs were validated by qRT-PCR, and the knockdown efficiencies were achieved in BGC823 and MKN45 cell lines ( Figure 6B). Next, we investigated the potential effect of DLEU2 and DDX11-AS1 in both cell lines. Knockdown of DLEU2 and DDX11-AS1 suppressed cell proliferation in BGC823 and MKN45 cells, as determined by CCK-8 assays ( Figure  6B). The results of colony formation assay showed that BGC823 and MKN45 cells with DLEU2 and DDX11-AS1 knockdown formed significantly less colonies than control cells ( Figure 6C). By miRNA target prediction, hsa-miR-30a-5p and hsa-miR-145-5p were noticed, as they have one putative target on DLEU2 and DDX11-AS1. DLEU2 and DDX11-AS1 complementarily matched the sequences on miRNA seed position, respectively. hsa-miR-30-5p regulated DLEU2 [43] and is a tumor suppressor To validate the findings, we first examined DLEU2 and DDX11-AS1 expression level in a cohort of 12 paired GC tissues, and compared with those of the adjacent normal tissues by using qRT-PCR. The DLEU2 and DDX11-AS1 expression levels were significantly upregulated in GC tissues ( Figure 6A). siRNA-mediated silencing of lncRNAs were validated by qRT-PCR, and the knockdown efficiencies were achieved in BGC823 and MKN45 cell lines ( Figure 6B). Next, we investigated the potential effect of DLEU2 and DDX11-AS1 in both cell lines. Knockdown of DLEU2 and DDX11-AS1 suppressed cell proliferation in BGC823 and MKN45 cells, as determined by CCK-8 assays ( Figure 6B). The results of colony formation assay showed that BGC823 and MKN45 cells with DLEU2 and DDX11-AS1 knockdown formed significantly less colonies than control cells ( Figure 6C). By miRNA target prediction, hsa-miR-30a-5p and hsa-miR-145-5p were noticed, as they have one putative target on DLEU2 and DDX11-AS1. DLEU2 and DDX11-AS1 complementarily matched the sequences on miRNA seed position, respectively. hsa-miR-30-5p regulated DLEU2 [43] and is a tumor suppressor in GC [44,45]. hsa-miR-145-5p inhibits gastric cancer cell migration and metastasis [51], and the relationship between DDX11-AS1 and hsa-miR-145-5p has not been reported in previous studies, but the correlation coefficient between DDX11-AS1 and hsa-miR-145-5p was the most significant (Table S2). Therefore, we selected hsa-miR-30-5p and hsa-miR-145-5p to verify the relationships between lncRNAs and miRNAs. By using qRT-PCR, we observed that knockdown of DLEU2 and DDX11-AS1 significantly increased the hsa-miR-30-5p and hsa-miR-145-5p expression level, respectively ( Figure 6D). These results indicated that DLEU2 and DDX11-AS1 might have oncogenic function and act as potential ceRNAs to sponge miRNAs in GC.

Discussion
Gastric Cancer is a leading cause of cancer-related death worldwide, and the second most common cancer in China. Many studies have demonstrated that coding RNAs and ncRNAs are closely related to GC, but the mechanism of GC remains to be elucidated. According to the ceRNA hypothesis, lncRNAs can act as miRNA sponges to construct a complex ceRNA network with miRNAs and mRNAs, as reported in many cancer types [10,19,20,22,23,31]. Thus ceRNAs play an important role in multilayered gene regulation and contribute to cancer development and pathology [52]. Some studies have constructed lncRNA-associated ceRNA networks in GC [23,24], but the mechanism of lncRNAs as ceRNAs in GC remains poorly explored.
In this paper, we mainly focused on upregulated lncRNAs in GC samples based on ceRNA activity. Since the expression of lncRNAs are generally lower expressed than protein-coding genes [64], it is difficult to confirm the functions of the low expressed lncRNA in normal samples by experimental methods. On the other hand, the number of GC samples is much larger than controls. It is very difficult to compare a GC-related ceRNA network to a control ceRNA network. To construct the dysregulated ceRNA network in GC, we first identified miRNA-target interactions by integrating reliable online databases and prediction software. To reduce false positives, we used the PCC to filter the significantly co-expressed DEL-DEMi-DEM interactions. We then constructed a ceRNA network based on 442 potential ceRNA triples. The gene expression data and bioinformatics methods used differ from those employed in previous studies, the ceRNA triples were no overlap with others. After network topological analysis, we determined that the hub lncRNAs play crucial roles in GC. By reconstructing and functionally analyzing each hub lncRNA-associated subnetworks, we identified that three hub lncRNAs (RP5-1120P11, DLEU2, and DDX11-AS1) are involved in cell cycle related processes and cancer-related pathways. Although only a few studies of the three lncRNAs were available, some indirect evidence might support their potential regulatory function in GC. RP5-1120P11 might regulate many cancer-related signaling pathways and is associated with the poor prognosis in ovarian cancer [65]. DLEU2 is a lymphocytic leukemia-related gene and is the host gene promoter for miR-15a/16-1 [42], and acts as ceRNA to sponge miRNA in clear cell renal cell carcinoma (ccRCC) [43]. DDX11-AS1, which is upregulated in multiple cancer types, affects sister chromatid cohesion [50], and could play an important role in hepatocellular carcinoma (HCC) [66]. However, in previous studies, none of them has been reported as functional lncRNAs or act as ceRNAs in GC.
These findings indicate that the hub lncRNA plays complex and critical roles in the development and pathology of GC. Although our study provides multistep computational and experimental methods to reveal functional lncRNAs in GC, biological experimental methods are still needed.

Conclusions
In summary, we employed multistep computational and experimental methods to reveal functional lncRNAs in GC based on the ceRNA hypothesis. By constructing and functionally analyzing lncRNA-associated network, we identified that three hub lncRNAs (RP5-1120P11, DLEU2, and DDX11-AS1) that are involved in cell cycle-related processes and cancer-related pathways. Furthermore, we confirmed that DLEU2 and DDX11-AS1 have oncogenic function and act as potential ceRNAs to sponge miRNAs. Our findings not only provide novel insights on ceRNA regulation in GC, but can also provide opportunities for the functional characterization of lncRNAs in future studies.

Conflicts of Interest:
The authors declare no conflict of interest.