The Comprehensive Analysis of Competitive Endogenous RNA Networks And Tumor-Inltrating Immune Cells In Intestinal and Diffuse Gastric Cancer

Background: Gastric cancer (GC) is one of the leading causes of cancer death worldwide. Increasing evidences have revealed different molecular characteristics between intestinal type GC and diffuse type GC. Results: We constructed Lauren subtype-specic competitive endogenous RNA (ceRNA) networks and identied common ceRNA network by comprehensive bioinformatics methods including differential expression analyses, RNA-RNA interaction prediction, negatively correlated RNA pairs and weighted genes co-expression network analysis. Besides, we detected the fraction of 22 immune cell types in GC by CIBERSORTx and investigated the correlation of immune cells and markers of ceRNA network. Ultimately, diagnostic performance of lncRNAs and mRNAs in ceRNA network were estimated by support vector machine (SVM). SNHG14 was identied as ceRNA of hsa-miR-429/ ZFPM2, hsa-miR-429/ZEB1 and hsa-miR-200b-3p/ZEB1 axes in all GC. Mast cells and Macrophages signicantly differed not only between the tumor and normal tissue, but also between the two subtypes. Mast cells and Macrophages are signicantly associated with multiple components in ceRNA network. Our results demonstrated lncRNAs and mRNAs of ceRNA network in the whole group (AUC=0.9152), the intestinal group (AUC=0.9670) and the diffuse group (AUC=0.9737) showed good performance in tumor diagnosis. Conclusion: Based on ceRNA networks and patterns of immune inltration, our study provided a valid bioinformatics basis in order to explore the molecular mechanism and estimated diagnosis performance of RNAs in ceRNA network.

expression of miRNA target messenger RNA (mRNA), thus playing a crucial role in tumor initiation as well as progression.
Increasing evidence suggests that lncRNA-miRNA-mRNA regulatory axes are highly correlated with GC [13,14] [15]. In our study, we performed weighted genes co-expression network analysis (WGCNA) to differentially expressed lncRNAs and mRNAs in The Cancer Genome Atlas (TCGA) and screened out the malignant stage-related RNAs for constructing the ceRNA network in two Lauren types patients and all TCGA patients. In addition, the proportions of immune cells were quali ed by CIBERSORTx algorithm. We assessed the association between immune cells and RNAs of ceRNA network and the prediction performance of them distinguishing the malignant tissue and normal tissue by supporting vector machine (SVM). The owchart of this research is shown in Fig. 1. Systematically identifying and comparing lncRNA biomarkers acting as ceRNAs, could contribute to elucidate the similarities and differences of the molecular mechanisms between intestinal-type gastric cancer (IGC) and diffuse-type gastric cancer (DGC), thereby providing valuable clues for investigating molecular mechanism. Result 2.1 Identi cation of differentially expressed RNAs: The gene expression pro les of 375 GC samples and 32 normal samples were obtained. We removed the miRNAs having zero expression more than 20% samples. The baseline characteristics of three patient groups available from the TCGA are described in Table 1  2.2 WGCNA: In order to select tumor-related RNAs, we did WGCNA to DElncRNAs and DEmRNAs of two groups. Nine modules related to pathological stage and T stage with p < 0.05 in IGC and ve modules in DGC were included in subsequent investigation ( Fig. 2A and 2B).

Differential immune cells and correlation analyses: Tumor-in ltration immune cells had been shown
to play a role in tumorigenesis, progression, metastasis and drug resistance. Immune cells estimated by CIBERSORTx algorithm are displayed in Table S1. The differences of immune cells by Wilcoxon rank-sum test were depicted in Fig. 5. Some immune cells (Macrophages M0, Macrophages M1 and Mast cells resting) were commonly differing in three groups, which may play important role in GC. But Macrophages M0 and Mast cells resting immune cells are also different between the two Lauren sub-types (Fig. 5D). To investigate the potential relationship immune cells with the RNAs of Lauren classi cation speci c ceRNA network, the correlation analyses were implemented (Fig. 6). Mast cells resting was evidently correlated with multi-markers of ceRNA network in three groups.
2.6 Diagnostic signi cance of lncRNA and mRNA biomarkers: Increasing evidence has demonstrated that lncRNAs and mRNAs can be used as biomarkers to guide decision on cancer diagnosis and therapy. The common RNAs of ceRNA network could represent the common molecular characteristics of two Lauren gastric cancer. Here, lncRNA and mRNA signatures were developed to distinguish tumor with normal tissues by SVM. The results in the test cohort showed that identi cation effect of SVM models was pretty excellent, whether distinguishing tumor in IGC (AUC = 0.9670) (

Discussion
It was reported that lncRNAs promotes tumorigenesis and development by competitively combining the shared miRNAs with the target gene sponge through the ceRNA mechanism [13,16,17]. Tumor-in ltration immune cells, which are an important part of tumors, had been revealed to play a role in tumorigenesis, progression, metastasis and drug resistance by many studies [18] [19]. GC is a heterogeneous disease. Herein, we divided GC samples into the whole group, the intestinal group and the diffuse group, which allowed us to study the common mechanisms and speci c mechanisms between two Lauren classi cation. In our study, we constructed Lauren classi cation speci c ceRNA network and GC ceRNA network based on differential expression, WGCNA, negatively correlated RNAs pairs. Then, correlation analyses markers of ceRNA networks with immune cells were carried out. The high AUC values of these markers of ceRNA networks identifying tumors proved their clinical application.
Based on ceRNA theory and supplementary theory [12,20,21], We use p < 0.01 and log 2 FC > 1 to identify differentially expressed RNAs because this minimizing interferential RNA pairs and avoiding missing important RNA pairs. WGCNA can make us to further screen out disrelated RNAs with tumor. Then, we performed differential expression analysis between two Lauren classi cations and many cancer-related and immune-related KEGG enrichment pathways (corrected p < 0.05) indicated that there are also differences in these pathways between IGC and DGC (detailed pathways results in Table S2). The differences in cancer-related pathways prove that there are differences in many cancer characteristics between two sub-types, and also support the strong heterogeneity of GC. The DERNAs between two subtypes were took into consideration to further identify subtype-speci c network and common network. we thought that even the common networks in two subtypes must have differences in the level of expression and subtype-speci c networks were even more so. The violin plot illustrated some immune cells, such as T cells gamma delta, NK cells resting, Macrophages M0, Macrophages M2 and Mast cells activated, differed in IGC and DGC. The Lauren classi cation should be considered into immune therapy of GC.
The mRNAs in common ceRNA network were enriched in Regulation of actin cytoskeleton and MicroRNAs in cancer, which manifested that targeted mRNAs of common network were regulated by relevant miRNAs. In the ceRNA network, CFL2, FGF2 and ZEB1 were reported to be miRNA's target associated with GC [22] [23] [24]. ZFPM2 and ZEB1 consisting in "MicroRNAs in cancer" pathway had common upstream lncRNA, namely SNHG14. It was reported to contribute to GC development through targeting miR-145/SOX9 axis [25]. However, it is possible that there are other regulation axes.
It was reported that Mast cells and Macrophages made effect in GC development[26] [27]. Mast cells and Macrophages signi cantly differed not only between the tumor and normal tissue, but also between the two subtypes. Like the expression level of two subtypes' cancer-related pathways, a similar pattern exists for two types of immune cells. And these two kinds of immune cells are signi cantly associated with multiple components in ceRNA network. Further studies on these two types of immune cells are necessary, which can help us to discover new therapeutic method or mechanism of GC.
LncRNAs and mRNAs were reported to be prognostic and diagnostic biomarkers of many cancers. Our results demonstrated lncRNAs and mRNAs of ceRNA network in all three groups showed good performance in tumor diagnosis. Perhaps they also guide clinical decision-making.
It is important to note that our results have not been validated by well-designed experiment and that is our prospective work. The public data we use are western demographic data, a conclusion should be made with caution in Asian countries. The number of DGC was relatively small, rendering the results less reliable.

Conclusion
we constructed Lauren subtype ceRNA networks and common network based on differential expression analyses, negatively correlated RNAs pairs and WGCNA. SNHG14, Mast cells and Macrophages maybe have crucial function in GC. LncRNAs and mRNAs had excellent effect in tumor diagnosis.

Data collection and differential gene expression analysis: GC gene expression data of fragments per
kilobase of transcript per million mapped reads (FPKM) and count were obtained from TCGA database by using the "TCGAbiolinks" package in R [28], a total of 375 GC samples and 32 adjacent tissue samples. The RNA-seq data was loaded with FPKM, which were converted to transcripts per million (TPM) after removing duplicated genes and zero expression genes. We matched and selected LncRNA and mRNA using GENCODE Release 22 (https://www.gencodegenes.org/human/release_22.html) [29] as gene annotation, which consisted with the o cial pipeline of TCGA data portal. Based on the Lauren classi cation information, all GC patients were divided into the whole group, the intestinal group and the diffuse group. The miRNA data transferred by log2(Reads per million mapped reads + 1) was downloaded in UCSC Xena (http://xena.ucsc.edu/) [30]. We removed the miRNAs having zero expression more than 20% samples.
The Limma package in R was utilized to screen the differentially expressed miRNAs (DEmiRNAs) [31] and the DESeq2 and EdgeR package in R was used to identify the differentially expressed lncRNAs (DElncRNAs) and mRNAs (DEmRNAs) between tumor and adjacent tissues [32]. DESeq2 and EdgeR algorithm were used for identifying DElncRNAs and DEmRNAs between IGC and DGC. The nal differentially expressed RNAs (DERNAs) were obtained based on the two algorithm results. The lncRNAs and mRNAs with the false discovery rate (FDR) P < 0.01 and the log (fold change) > 1.0 or < − 1.0 were only regarded as differentially expressed RNAs and the miRNAs with false discovery rate (FDR) < 0.05 and the log 2 fold change (log 2 FC) > 1.0 or < − 1.0 as differentially expressed miRNAs.

WGCNA based on DElncRNAs and DEmRNAs:
The lncRNAs and mRNA data pro les identi ed above were merged into one pro le. Weighted genes co-expression network analysis was performed using the WGCNA R package [33]. We set minModuleSize as 30 and no mixModuleSize. Firstly, we removed the signi cant outliers and screened out appropriate soft threshold power. Then, the Pearson's correlation matrices were calculated for all the paired RNAs and a weighted adjacency matrix was constructed. Finally, we performed the correlation analysis to the module eigengenes as the rst principal component and clinical traits including pathologic stage and tumor stage (T stage). All processes above were carried out in intestinal group and diffuse group.

5.3
Prediction of lncRNA-miRNA and miRNA-mRNA interactions: The lncRNA-miRNA interaction pairs were predicted by DEmiRNAs in LncBase Predicted v.2 with the threshold of 0.9 [34]. The miRNA-mRNA interaction pairs were predicted by DEmiRNAs in miRTarBase [35] and TarBase V.8 [36] in which there are experimental validation miRNA-mRNA interaction pairs. The correlation analyses were performed on all DElncRNA-DEmiRNA pairs and DEmiRNA-DEmRNA pairs, the reason is that their expression is negatively related. We thought the RNA-RNA pairs with R < -0.3 and p < 0.05 are negatively related and taken in consideration for subsequent analyses. Then, we took the intersection of DElncRNAs and DEmRNAs of tumor and normal tissue, DElncRNAs and DEmRNAs of intestinal group and diffuse group, RNAs related with stage from WGCNA and negatively related RNA interaction pairs. Finally, the ltered RNA-RNA interaction pairs were obtained. All processes above were carried out in intestinal group and diffuse group. The constructed ceRNA networks were visualized by Cytoscape V3.7.2 [37]. . The Wilcoxon test was used to look for signi cantly differential immune cells between adjacent normal tissues and tumor.
5.6 Correlation analysis: We merged lncRNA and mRNA pro les of ceRNA network into one pro le. Then, the signi cantly differential immune cell pro les were added into above pro le. The correlation analysis of lncRNAs, mRNAs and immune cells with pearson was performed in R.

5.7
Performance of distinguishing tumor of members in the ceRNA network: In order to avoid over tting of predicting model, the samples were divided two groups on the basis of 7:3, one in front was named as the train cohort and one in back was named as the test cohort. We trained classi ers in the train cohort using lncRNAs and mRNAs of ceRNA networks and examined the performance of classi ers in the test cohort. The receiver operating characteristic curves (ROC) were used to visualize the prediction effect and Ethics approval and consent to participate: Not applicable.
Consent for publication: Not applicable.
Competing Interests: The authors declare no potential con icts of interest.