Identification of hub pathways and drug candidates in gastric cancer through systems biology

Gastric cancer is the fourth cause of cancer death globally, and gastric adenocarcinoma is its most common type. Efforts for the treatment of gastric cancer have increased its median survival rate by only seven months. Due to the relatively low response of gastric cancer to surgery and adjuvant therapy, as well as the complex role of risk factors in its incidences, such as protein-pomp inhibitors (PPIs) and viral and bacterial infections, we aimed to study the pathological pathways involved in gastric cancer development and investigate possible medications by systems biology and bioinformatics tools. In this study, the protein–protein interaction network was analyzed based on microarray data, and possible effective compounds were discovered. Non-coding RNA versus coding RNA interaction network and gene-disease network were also reconstructed to better understand the underlying mechanisms. It was found that compounds such as amiloride, imatinib, omeprazole, troglitazone, pantoprazole, and fostamatinib might be effective in gastric cancer treatment. In a gene-disease network, it was indicated that diseases such as liver carcinoma, breast carcinoma, liver fibrosis, prostate cancer, ovarian carcinoma, and lung cancer were correlated with gastric adenocarcinoma through specific genes, including hgf, mt2a, mmp2, fbn1, col1a1, and col1a2. It was shown that signaling pathways such as cell cycle, cell division, and extracellular matrix organization were overexpressed, while digestion and ion transport pathways were underexpressed. Based on a multilevel systems biology analysis, hub genes in gastric adenocarcinoma showed participation in the pathways such as focal adhesion, platelet activation, gastric acid secretion, HPV infection, and cell cycle. PPIs are hypothesized to have a therapeutic effect on patients with gastric cancer. Fostamatinib seems a potential therapeutic drug in gastric cancer due to its inhibitory effect on two survival genes. However, these findings should be confirmed through experimental investigations.


Methods
The approach used in this study is as follows in brief: gastric cancer was modeled by network analysis tools based on gene expression data, and potential compounds that can affect patients' survival were detected. Besides, non-coding RNA interactions, viral causes, and pathways were analyzed to better understand the pathological mechanisms of gastric cancer development. A summary of the method used is shown in Fig. 1.
Identification of samples and differentially expressed genes in gastric cancer. The gene expression omnibus (GEO) database (https:// www. ncbi. nlm. nih. gov/ geo) was used to download a microarray dataset related to gastric cancer. This database provides a great number of microarrays, RNA seq, and other data freely, so it can be a very useful database for working in the genetics and bioinformatics field. In the advanced search section, all microarray profiles with "gastric cancer" and "u133" keywords in their title or abstract were searched. Forty-five studies about Homo sapiens were found. Among them, GSE79973 consists of 10 pairs of gastric cancer tissue and adjacent non-tumor samples with HG-U133_Plus_2 platforms (a single array representing 14,500 well-characterized human genes, which can be used to explore human biology and disease processes), and was selected for further analyses 25 . The samples were analyzed by using transcriptome analysis console (TAC) software with the help of affymetrics U133-plus2 library to identify differentially expressed genes (DEGS). RMA algorithm was applied for data normalization. Then, PCA was used to reduce the dimensions of the dataset, so it could be evaluated if normal versus tumor was separated well. Comparison analysis (t-test) was also done on the dataset to identify differentially expressed genes with a significant p-value (p-value ≤ 0.05). Finally, gene set enrichment analysis helped to discover the possible related pathways associated with each group of genes. Two outlier samples were excluded. After that in PCA, normal and tumor samples were well separated. Finally, for further analysis, DEGS were detected based on their fold changes (FC ≥ 2.3 or FC ≤ -2.3) at a significance level of (p-value ≤ 0.05) (Table S1).
Gene set enrichment analysis. DEGs were imported in STRING v11.0 (https:// string-db. org/) 26 and ranked based on their fold change (proteins with value/rank section). The STRING database can help to find the relation between different proteins in different species, including homo sapiens. Enrichment analysis was done using the gene ontology (GO) database (http:// geneo ntolo gy. org/), which has a good potential for finding Non-coding RNA-protein interaction network construction. In this study, non-coding RNA network construction was considered to get a better view on the pathogenesis of gastric cancer. Non-coding RNA expression data were separated from DEGs at the significance level of (p-value ≤ 0.05, FC ≥ 2.3 or FC ≤ -2.3). The non-coding versus coding RNA network was constructed using the RNA Interactome Database (https:// www. rna-socie ty. org/ rnain ter/) 34 . A python script was also employed. The network was constructed between 116 and 37 coding hub and non-coding genes, respectively. STRING v11.0 protein-protein interaction network was also added to obtain a complete interaction network.
Gene-disease network. The 116 hub genes were exported to ENRICHR, and their correlated diseases were extracted from the DisGNet plugin to identify the best matched diseases with these hub genes, based on the p-value. The DisGNet plugin in Cytoscape was then used to find the interaction of our 116 hub genes with www.nature.com/scientificreports/ diseases and their involvement in several diseases 35 . The collected data can suggest the genes that might have role in the metastasis of gastric cancer or other related cancers based on the networks.
Statistical survival analysis. The GEPIA database (http:// gepia. cancer-pku. cn/) was employed for survival analysis 36 . GEPIA is an online tool that can be used for survival, correlation analysis, and gene expression analysis in different types of cancer and normal tissues. We hypothesized survival analysis can discover more reliable drug targets among our hub genes, because it can show the patient outcome in association with the underexpression or overexpression of genes, so a gene with both high centrality parameters in graph and significant impact on the patient survival at the same time might have stronger and more valid evidence for tumor growth progression or inhibition. The GEPIA uses TCGA and GTEx project data to perform overall survival or disease-free survival analysis using log-rank t-test with adjustable thresholds. It can also calculate the cox proportional hazard ratio and shows 95% confidence interval on survival plots. Log-rank < 0.05 was assumed as the significance level, and the threshold was set on 50 percent.
All the survival plots of 116 hub genes were checked, and their cox proportional hazard ratio was calculated to find the hub genes that had a significant effect on the survival rate. The survival gene expression between tumor and normal data in GEPIA were also checked, and the survival genes without significance differential expression based on GEPIA analyses were excluded. Correlation analysis. GEPIA was also used to identify the linear correlation between the discovered survival genes and the rest of hubs. Pearson method was used, and the significance level was set on "R coefficient ≥ 0.5, p-value ≤ 0.05" to discover the linear correlation. Again, the genes without significant expression in the tumor and normal samples were excluded based on the GEPIA analysis. The pathways in which a survival gene and its correlations were involved, were found using KEGG and GO in the ENRICHR database. It allowed to identify all possible drug targets with a high effect on tumor growth in order to use them in the experiments for validation.

Survival-correlation weighted network analysis.
To have a better analysis that can be attributed to all gastric cancer populations, the box plot of the GEPIA database was used. In our study, an interaction network was constructed between survivals and their correlations to understand the hypothetical mechanisms that result in poor prognosis in gastric cancer. The weight of edges in this network was the correlation coefficient retrieved from GEPIA. Drug discovery. The Targetmine database (https:// targe tmine. mizug uchil ab. org/) was used to find drugs and compounds that have interactions with genes/proteins 37 . Targetmine can search multiple genes in Drug bank and other databases at the same time and show the results. Based on the aforementioned database, the survival genes whose overexpression resulted in better patient survival rates could be targeted with cognate agonists. In this study, to discover any possible synergistic effects of drugs, the genes that had a positive linear correlation with survivals needing agonists were detected. Antagonists were selected against the genes whose underexpression either was related with better survival rates or had a positive linear correlation with this type of survival genes.
Ethical considerations. The human gene data obtained in the study was derived from a publicly available repository. All experiments were conducted in accordance with relevant guidelines and regulations in the main manuscript.

Results
Differentially expressed genes (DEGs) analysis. The normalized data of gastric adenocarcinoma samples versus paired normal tissue and the box plot gene expression distribution in each patient are shown in Fig. 2A. The microarray gene expression distribution before normalization is shown in blue boxes (CEL). A CEL file is a data file created by Affymetrix DNA microarray image analysis software. It contains the data extracted from "probes" on an Affymetrix GeneChip and can store thousands of data points, which may make the file size large. CEL files can be processed by software algorithms and visualized on a 2D grid as part of an overall genome experiment.
The gene expression distribution after normalization is represented in red boxes (CHP). CHP is a file format generated by Transcriptome analysis console (TAC) after normalization of the expression data by RMA, MSA5 or DABG algorithms. It saves algorithm parameters and summary statistics.
Gene set enrichment analysis. The DEGs were ranked based on their fold change and were imported to STRING v11.0 for gene set enrichment analysis based on the GO biological pathways. Filtering of pathways based on false discovery rate (FDR ≤ 0.05) showed the involvement of the overexpressed genes in the extracel- Protein-protein interaction network analysis. Fibronectin 1 (fn1), collagen type I alpha 1 chain (col1a1), matrix metallopeptidase 2 (mmp2), collagen type I alpha 2 chain (col1a2), secreted protein acidic and cysteine-rich (sparc), biglycan (bgn) were the first five hub genes based on eigenvector. These genes possibly can control most of the feedback loops involved in tumor cell survival, and interventions against their function may result in dramatic changes in the gastric cancer tumor cells' growth. The first 116 hub genes had four modules, and module enrichment analysis based on the KEGG pathways showed the involvement of these modules in pathways such as cell cycle, focal adhesion, platelet activation, and gastric acid secretion. It also suggested the HPV infection pathway, which was nearly correlated with extracellular matrix organization. All these pathways were upregulated except gastric acid secretion. Targeting these genes may result in numerous side effects, so  www.nature.com/scientificreports/ we decided to target the hub genes that specifically affected gastric cancer patients' survival (Supplementary  Table S3, Fig. 4A).

Survival analysis. The survival analysis performed on the overexpressed and underexpressed hub genes
represented the significant role of hub genes in the survival of patients suffering from gastric cancer. The overexpression of some hub genes such as fn1, platelet-derived growth factor receptor beta (pdgfrb), col1a1, versican (vcan), plasminogen activator inhibitor-1 (serpine1), immunoglobulin superfamily containing leucine rich repeat (islr), cluster of differentiation 36 (cd36), fibrinogen gamma chain (fgg), ghrelin and obestatin prepropeptide (ghrl), and coagulation factor II thrombin receptor (f2r) showed a negative impact on the survival of patients suffering from gastric cancer. In this group, all of the genes except cd36, ghrl, and kit proto-oncogene (kit) displayed overexpression in gastric cancer, which might suggest their oncogenic effect. On the other hand, the overexpression of some hub genes such as centrosomal protein 55 (cep55), centrosome-associated protein E (cenpe), and epithelial cell transforming 2 (ect2) indicated a positive effect on the survival of gastric cancer cases. Moreover, these genes were overexpressed in the tumor tissue, which might show their role in tumor growth inhibition. Checking the hub genes in GEPIA revealed that they all had significant roles in the prognosis of gastric cancer.
Correlation analysis. It is speculated that the linear correlated expression between the hub genes might have a synergistic effect on tumor progression, especially if the correlation between non-survival and survival hub genes is of a casual type. Although most of the genes in the network with high centrality parameters were not survival-affecting genes, they had a linear correlation with the survival genes. Thus, they might have a causal effect on the expression of the survival-affecting hub genes. However, it must be investigated in an experimental approach. Our results showed that some genes namely col3a6, pdgfrb, col1a2, fn1, thbs2, sulf1, and vcan had a positive linear correlation with col1a1 gene. Genes col1a2, col6a3, thbs2, fibroblast activation protein (fap), sulf1, and bgn showed a positive linear correlation with vcan gene, while genes col1a2, wnt5a, col6a3, thbs1, thy1, vcan, sparc, and bgn indicated a positive linear correlation with pdgfrb gene. Genes mki67, anln, ccnb1, cenpf, chek1, dtl, and ndc80 displayed a positive linear correlation with cep55 gene. It was also found that the genes with a significant role in survival were correlated with each other. Several paired genes including (f2r, islr), (fn1, col1a1), (islr, col1a1), (islr, f2r), (pdgfrb, col1a1), (pdgfrb, f2r), and (vcan, col1a1) showed a positive correlation with each other, so that using antagonists against both of them could have a synergistic effect on the gastric cancer survival. Moreover, as mentioned above, the overexpression of two other paired genes, i.e. (cep55, cenpe) and (cep55, ect2), revealed a good impact on survival, so that using their agonist may have a synergistic effect on gastric cancer survival. Interestingly, in this analysis, cep55 overexpression was found to relate to a positive prognosis in gastric cancer patients.
Non-coding RNA-protein interaction network construction. Using the complete constructed noncoding RNA protein interaction network, network centrality parameters were calculated for both hub coding and non-coding genes using Gephi. Fn1, growth arrest-specific transcript 5 (gas5), and metastasis-associated lung adenocarcinoma transcript 1 (malat1) were detected as hubs based on degree and betweenness in the network. It showed the central role of fn1 and malat1 in the gastric cancer pathogenesis (Fig. 4B). www.nature.com/scientificreports/ Gene-disease network. It was found that 116 hub genes were also involved in liver carcinoma, breast carcinoma, liver fibrosis, prostate cancer, ovarian carcinoma, lung cancer, brain ischemia, atherosclerosis, arteriosclerosis, aortic aneurysm, cardiovascular diseases, pulmonary fibrosis, Ehler-danlos syndrome, Marfan syndrome, osteogenesis imperfecta, esophageal neoplasm, osteosarcoma, and endometriosis. Beyond that, this network showed the hub genes with higher betweenness, such as hgf, metallothionein 2a (mt2a), mmp2, fibrillin-1 (fbn1), col1a1, and col1a2 could not only play a role in several diseases but also cause metastasis [38][39][40] . Moreover, it showed that survival hub genes, such as fn1 and serpine1, could be involved in several diseases. The genes acting in the components of the extracellular matrix (ECM) pathway and HPV infection were also identified in this network (Fig. 5).
Survival-correlation weighted network analysis. Interestingly, the agonist target network and the antagonist target network were separated, meaning that targeting one of the genes in each network did not have a significant impact on other networks. This can potentially reduce the chance of drug side effects (Fig. 4C). Based on this network, the antagonist targets mostly showed involvement in the ECM-receptor interaction and focal adhesion, while the agonist targets represented participation in the cell cycle pathway.
Drug discovery. The genes with a negative correlation with better survival rates were searched in the Targetmine database to identify their antagonists. Some of the genes included fn1, K + /H + ATPase transporting subunit alpha (atp4a), serpine1, ctsk, pdgfrb, and kit. There were also some genes that needed to be overexpressed, Figure 5. Gene-disease network. It shows not only the diseases and syndromes that have common genes and pathways with gastric adenocarcinoma, but also the common genes participating in different diseases and syndromes. Node size represents the degree (the bigger size shows the higher degree). The node color exhibits betweenness (the darker color indicates the higher betweenness). www.nature.com/scientificreports/ probably through using an agonist. Some of these genes were cep55, cenpe, ect2, anillin (anln), and checkpoint kinase 1 (chek1). Moreover, some candidate drugs for repositioning were found. Some of them were sorafenib, fostamatinib, troglitazone, quercetin (a natural flavonol found in vegetables preventing cancer and inflammation), and pantoprazole (Table 1).

Discussion
Recent studies showed that not only genetic factors but also some non-genetic factors contribute to gastric cancer 41 . Some pathways, including cell cycle, WNT/β-catenin signaling, focal adhesion, and nucleotide excision repair, are among the most important pathways in gastric cancer 42 . Currently, the main way of treating gastric cancer is still surgery, which might be followed by adjuvant therapy. However, surgery and adjuvant chemotherapy, as the gold standard therapy in gastric cancer, can increase median survival by only seven months 43 . Hence, understanding tumor biological pathways can be useful in finding new drugs and therapeutic methods for gastric cancer. In this study, using different bioinformatics tools and databases, the gene expression data in gastric cancer was analyzed and compared to a normal condition to identify the genes that are upregulated or downregulated in gastric cancer. The focus was given to survival genes, and the potential compounds that can affect patients' survival were detected by performed analyses. To further understand pathological mechanisms of gastric cancer development, non-coding RNA interactions, viral causes, and pathways were also analyzed. Cell cycle, which regulates cell proliferation and tumor growth, is one of the key pathways in all cancers including gastric cancer. According to our analysis, several genes including cdc20, ccnb1, chek1, and mad2l1 participate in the cell cycle pathway. This pathway interacts with most pathways in gastric cancer, as discussed in the following.
Based on our analysis, focal adhesion is a promising pathway, which was found to be upregulated in our analysis. It participates in multiple activities in tumor microenvironment (TME), immunosuppression, and metastasis. According to our analysis, col1a1, col1a2, col6a3, col6a1, thbs1, and thbs2 are the main hub genes of this pathway, which mostly produce ECM and have an initial role in the activation of focal adhesion pathway [30][31][32] . Focal adhesion kinase (FAK) is an important part of the focal adhesion pathway, whose activation by these ECM compounds leads to an increase in cancer cell migration and survival, angiogenesis, cytokine production, and abnormal ECM accumulation 44 . Some of these cytokines and chemokines, namely CCL1, CCL5, CCL7, CXCL10, and TGF-β2. are responsible for T-reg cell recruitment, leading to CD 8+ (cytotoxic T cell) exhaustion, which results in the reduction of tumoricidal cell function and helps cancer cells to escape from the immune system 45 . On the other hand, FAK activation can be induced by cancer-associated fibroblasts (CAFs) via β1 integrin, which can www.nature.com/scientificreports/ lead to proliferation, migration, and invasion potential in gastric cancer cells 46 . CAFs are crucial cells in tumor formation process. They can promote tumor cell growth via the promotion of cancer stemness or prevention of cancer cell recognition by T-cells. They are activated by pro-inflammatory cytokines such as TGF-α, TGF-β, FGF-2, and EGF, which are mainly secreted from tumor cells. Additionally, CAFs can release TGF-β leading to activation of STAT3 signaling pathway in tumor cells. CAFs and tumor cells can also secrete proteases, including matrix metalloproteinase, resulting in the breakdown of the cell's basement membrane, which has a key role in cancer metastasis 47 . CAFs can also activate fap transcription, which was found in our analysis as a hub gene in the antagonist network (Fig. 4C). Since CAFs are overexpressed in the gastric cancer tissue and have a crucial role in tumor cell migration and invasion 48 , it can be concluded that focal adhesion and CAFs can be activated via different pathways, and they play key roles in cancer in different ways (Fig. 6). The current study found that platelet activation is an important pathway in gastric cancer. Col1a1, fga, fgg, f2r, col3a1, and col1a2 are the important genes of this pathway, which are upregulated in gastric cancer. According to KEGG, collagen fibers (col1a1, col3a1, col1a2) can activate GPVI, which leads to the initiation of platelet activation [30][31][32] .
Thrombin is another molecule with multiple functions, which can not only attach to its receptor (PAR 1, coded by f2r gene) and activate platelets but also support angiogenesis by activating VEGF. Likewise, it could activate fibrinogen, which may cause cell proliferation and tumor growth in collaboration with PAR 1 49 . Platelet activation can help cancer development via different mechanisms. It may result in the release of transforming growth factor beta (TGF-β), VEGF, and platelet-derived growth factor (PDGF), which can cause tumor cell growth, angiogenesis, and neovascularization. Platelet and fibrinogen could coat tumor cells in the vascular and lymphatic systems, thereby supporting the escape of tumor cells from natural killer (NK) cells 49,50 . Platelets release TGF-β leads to the impairment of NK cell function 50 . This may imply that these mechanisms can hide tumor cells from the immune system and help them to metastasize into other organs and tissues (Fig. 7).
Another finding is that platelet activation by tumor tissue can cause a hypercoagulable state; consequently, patients have a higher risk of thrombotic conditions, namely deep vein thrombosis (DVT) 51 . Therefore, aspirin, as an anti-thrombotic agent, could be used in targeting platelet activation via acetylation of cyclooxygenase (COX) 52 . COX enzyme has two isoforms (COX1 and COX2), which are both inhibited by aspirin. COX1 inhibition leads to the anti-thrombotic effect of aspirin, while aspirin inhibition on COX2 enzyme results in anti-inflammatory effects. Its inhibitory effect on COX2 is 200 folds less potent than COX1 52 . The results of this study suggest that aspirin could act as a potential therapeutic agent in cancer. According to recent studies, the long-term use of aspirin was associated with a reduced risk of having gastric cancer, especially non-cardiac type gastric cancer 53,54 . In addition, aspirin could inhibit the growth of gastric cancer cell lines via suppressing the survivin protein and www.nature.com/scientificreports/ induction of apoptosis 55,56 . Thus, although aspirin is associated with stomach mucosal damage 57 , it might prevent gastric cancer occurrence via inhibition of platelet activation and survivin, as well as apoptosis induction. It was also found that upregulation of the HPV infection pathway plays an important role in gastric adenocarcinoma. HPV is one of the most important carcinogenic viruses in humans. Different HPV serotypes have been discovered, among which HPV-16 and HPV-18 are the most prevalent mucosal high-risk serotypes, associated with nearly 90% of human cervical cancers and 20% of oral cancers [58][59][60] . The E6 and E7 genes of HPV play a causative role in cancer progression; the former promotes degradation of p53 through its interaction with E6AP, an E3 ubiquitin ligase, while the latter binds to the retinoblastoma protein (prb) and disrupts its complex formation with E2F transcription factors so that p53 is suppressed and rb is promoted causing a higher rate of cell division 61 . According to our study, pdgfrb, which is a mitogen for the cells with a mesenchymal origin, is an important survival gene whose underexpression is associated with better outcomes in gastric cancer patients. On the other side, according to VIRHOSTNET 3.0, available at https:// virho stnet. prabi. fr/, the E6 protein of HPV can directly induce pdgfrb gene expression 62 .
The result of our study is consistent with a meta-analysis of 30 studies, including 1,917 gastric cancer patients and 576 controls. It was revealed that HPV prevalence was 28% higher in patients with gastric cancer, which was significant compared to the control group. It was found that HPV-positive patients with gastric cancer were higher in the Chinese than non-Chinese population, which may be the reason for the higher prevalence of gastric cancer in China 63 . Interestingly, the microarray database used in this study was prepared from people of Hangzhou in China 25 . Therefore, it is hypothesized that HPV may be a risk factor for gastric adenocarcinoma due to its direct effect on survival genes, as it is associated with a higher risk of cervical, colorectal, esophageal, and oropharyngeal cancers 64 . Nevertheless, there are other studies against our findings, which did not detect HPV in the gastric cancer samples 65 . Hence, more studies in various countries are needed to verify this hypothesis.
Survival and drug analysis. The results indicated that gastric acid secretion, which is a crucial pathway to digest food in normal stomach physiology, plays a key role in gastric adenocarcinoma progression. Atp4a had a strong positive linear correlation with ghrl (among survival genes), and both of them were underexpressed in gastric cancer according to GEPIA boxplot, which resulted in gastric acid secretion downregulation. Therefore, ghrl and possibly atp4a have a protective role in gastric cancer.
In our study, PPI drugs were shown to target the atp4a gene, so they might help patients to have a better prognosis. However, some studies showed PPIs as a risk factor in developing gastric cancer 66 . According to these contrary results, we decided to study the mechanism of gastric acid secretion to gain a better insight into PPIs' effect on atp4a and ghrelin interaction. Since a few studies have shown the relationship between atp4a and ghrelin, we were not certain if atp4a expression was the result of ghrelin expression. On the other hand, gastrin www.nature.com/scientificreports/ can stimulate atp4a, which can code for potassium hydrogen ATPase channels and increase acid secretion, as an important risk factor for gastric cancer 67 .
Gastrin hormone induces tumor cell growth, migration, autophagy, and survival 68 , and its secretion by G-cells is inhibited by somatostatin (SST) release through D cells 69 . Gastrin also can stimulate cholecystokinin β receptors (CCKβR), leading to the growth of gastric cancer tumor cells 70 . Thus, it is hypothesized that a lower level of SST leads to a higher level of gastrin, which increases gastric cancer occurrence. Downregulation of SST was also found in the DEGs of our study. Based on the aforementioned findings, SST decrease and gastrin increase have a synergistic effect on gastric cancer development, while ATP4A and GHRL decrease act oppositely so that their net effect on gastric cancer progression depends on the influences of each axis of gastric cancer development. Based on the results of our study, gastrin upregulation can also increase histamine release, and the downregulation of SST has the same effect. The result may imply that the underexpression of ghrl gene possibly affects enterochromafin cells (ECL) through decreasing histamine release; the effect of ghrl on histamine is contrary to gastrin and somatostatin 71 . Histamine can act through histamine receptor h2 (hrh2), which plays a stimulatory effect on acid secretion 72 . Based on GEPIA, hrh2, the same as ghrl, is a survival gene in gastric cancer whose underexpression is associated with better survival in patients (downregulated based on our DEGs) (Fig. 8). These data suggest the same mechanisms of gastric cell protective response against cancer development.
Based on these findings, PPIs may have therapeutic effects on gastric cancer. According to Cheung and Leung, the long-term use of PPIs could increase the risk of gastric cancer by nearly twofold 66 , while in other studies investigating PPI association with gastric cancer cell lines, rabeprazole, a second-generation PPI drug, could reduce gastric cancer cell survival in vitro 73 . The probable interpretation could be that although the long-term use of PPIs can lead to higher rates of gastric cancer in the normal population, PPI drugs could reduce cancer progression in patients diagnosed with gastric adenocarcinoma.
According to our analysis, pdgfrb is mostly involved in pathways such as focal adhesion, PI3K-AKT signaling pathway, the pathways in cancer, and HPV infection. Meanwhile, kit mostly participates in PI3K-AKT signaling pathway and the pathways in cancer, and both pdgfrb and kit are survival genes whose overexpression leads to poor survival for patients. Fostamatinib, which is used for rheumatoid arthritis and immune thrombocytopenic purpura (ITP) treatment 74,75 , can also be used as a potential inhibitor of pdgfrb and kit 76 . Thus, fostamatinib may be a probable small molecule for treating gastric cancer to reduce cell proliferation, metastasis, and cancer progression. A recent clinical trial for fostamatinib performed on multiple malignancies (advanced colorectal, non-small cell lung, head and neck, thyroid, and renal cell carcinomas, as well as pheochromocytoma) showed its anti-tumor activity in several patients 77 . However, its effect on gastric adenocarcinoma has not been studied yet.
Fibrinolytic system can also be a potential target for cancer therapy. Plasmin is the key compound of this system, which is derived from plasminogen. Plasminogen is an inactive enzyme, which is converted to plasmin by tissue-or urokinase-type plasminogen activators (tPA or uPA). uPA is coded by plau gene. Fibrinolytic system takes part in cancer in different ways. Plasmin can activate MMPs, which have a crucial role in ECM degradation, cancer cell migration, and metastasis. The uPA system can induce cell proliferation via increasing www.nature.com/scientificreports/ growth factors such as VEGF, EGF, FGF2, and TGF-β. It also causes a reduction in apoptosis, leading to cancer cell immortality 78,79 . Hence, using uPA inhibitors could be useful in cancer treatment. Amiloride is a selective uPA inhibitor commonly used for inhibiting sodium reabsorption through sodium channels in the renal epithelial cells 80 . It can reduce gastric cancer peritoneal metastasis, which could be useful at the end stages of gastric cancer 81 . The serpine1 gene codes plasminogen activator inhibitor 1 (PAI-1). Thus, PAI-1 overexpression, as the inhibitor of uPA system, must have a positive effect on patients' survival by the inhibition of cell proliferation, metastasis, and apoptosis induction 79 . However, in our analysis, we found that plau has a positive linear correlation with serpine1, so inhibiting both of them may have a synergistic effect in gastric cancer treatment. The PAI-1 level is higher in gastric cancer tissue than in normal tissue, and its overexpression leads to poor survival in gastric cancer patients via increasing tumorigenicity and inhibition of cancer cells apoptosis [82][83][84] , as was confirmed by GEPIA. Therefore, using a PAI-1antagonist as well as uPA antagonists, can be useful in gastric cancer treatment. Troglitazone, a PAI-1inhibitor, was once used as an anti-hyperglycemic agent, but it was withdrawn in the year 2000 because of its hepatotoxicity. Thus, finding a way to reduce its toxicity can candidate it as a potential drug for gastric cancer 80 .

MicroRNA analysis.
MicroRNAs are small non-coding RNAs with important effects on gene regulation.
Their regulation is being interrupted in a variety of cancers. Many different mechanisms are involved in micro-RNA dysregulation, such as amplification or deletion of microRNA genes, epigenetic changes and defects in the microRNA biogenesis machinery, and abnormal transcriptional control of microRNAs. They also have a considerable role in cancers by regulating cell proliferation signaling pathways, cell resistance to apoptosis, invasion, metastasis, and inactivating of tumor suppressor mechanisms, which help tumor cells to grow and build a tumor mass. They also affect the clinical features and therapeutic outcomes of tumors, which could be sensitive and specific biomarkers for diagnostic processes and therapeutic goals 85 . MicroRNAs also have a key role in gastric cancer. According to our studies, MALAT1 is a hub microRNA among others in gastric cancer (Fig. 4B). According to recent studies, MALAT1 function varies in different tumors, which may act as a tumor suppressor gene or have an oncogenic effect on different cancers. MALAT1 directly binds to SOX2 mRNA, which enhances its stability so that it can have a positive effect on the regulation of stemness of gastric cancer cells 86 . Likewise, MALAT1 can be a powerful candidate for prognostic goals in gastric cancer. Some studies reveal that MALAT1 level is higher in patients with a distant gastric cancer metastasis than those without distant metastasis in the control group 87 . As discussed above, the role of microRNAs in different cancers is proved 88 , but still more investigations are needed to reveal other dimensions of their effect on cancers. They can be very good biomarkers to solve challenges in finding specific and sensitive biomarkers for different types of cancers.
Gene disease network analysis. Gastric cancer is a complicated disease, and different pathways take part in its pathogenesis. Gastric cancer can lead to secondary metastatic cancers in other tissues. According to our analysis, different diseases and syndromes are associated with gastric cancer, such as liver carcinoma, malignant neoplasia of the prostate, atherosclerosis, liver fibrosis, myocardial infarction (MI), and lung carcinoma. Among these diseases, liver carcinoma, gastric cancer, and lung carcinoma are correlated via the baculoviral inhibitor of apoptosis repeat-containing 5 (birc5) gene. Birc5 is a member of the inhibitor of apoptosis (IAP) gene family, which codes proteins that can prevent cell apoptosis 89 . According to GEPIA, birc5 is overexpressed in these three cancers 36 , so that it can lead to the activation of the anti-apoptotic pathway, which is a crucial pathway in cancer progression. Gastric cancer and atherosclerosis are connected to each other via the serpine1 gene. On the other hand, gastric cancer and MI are similar in plau gene. As discussed above, these two genes participate in the complement coagulation pathway. In fact, the complement coagulation pathway is an interplay between the complement system as a part of innate immunity and the coagulation pathway. Inflammation is one of the key factors that can activate the complement coagulation pathway 90 . In both MI and atherosclerosis, inflammation is an important part of the pathogenesis, so that this pathway can interconnect MI and atherosclerosis with gastric cancer. The malignant neoplasias of prostate and gastric cancers also share many genes, one of the most important of which is mt2a. The mt2a gene is a member of the metallothionein family of genes, which code proteins that have a substantial role in the hemostatic control of metals in cells and detoxification of heavy metals influences apoptotic and autophagy pathways 89 . According to GEPIA, mt2a is underexpressed in both of these cancers 36 , which leads to less detoxification of heavy metals and increased risk of these cancers in the same way. These data show an interesting correlation between gastric cancer and other diseases, which can help in predicting other diseases induced by gastric cancer tumor or other diseases with a predisposing role in gastric cancer.
Limitations and perspectives for future studies. 1-Although the microarray dataset was paired data, and its PCA graph was well separated, it was not mentioned which part of the stomach was used for collecting the tumor samples.
2-For a better analysis of each type of cancer, it is better to have samples that are collected from the same disease stage. However, it was not mentioned in the source of microarray data.
3-In this study, all the linear relationships that existed between gene expression were used. Further inclusion of nonlinear relationships in future studies might generate more accurate data on the relationship between proteins, and probably more drugs could be suggested.
4-Using RNA-Seq technology to measure gene expression might lead to more reliable results. However, due to the unavailability of paired RNA-Seq data in this study, paired microarray data was used, which matched better and could produce more valid results. Besides, to increase accuracy in this study, key genes with significant differential expression in both microarray dataset and TCGA data (analyzed by GEPIA) were employed for drug target discovery.

Conclusion
Totally based on a multilevel systems biology analysis, the hub genes in gastric adenocarcinoma showed participation in the pathways such as focal adhesion, platelet activation, gastric acid secretion, HPV infection, and cell cycle. In the survival and drug analysis, fostamatinib and troglitazone were found as potential drugs targeting survival and hub genes in gastric adenocarcinoma. The fibrinolytic system and gastric acid secretion are two important pathways for drug analysis. PPIs are hypothesized to have a therapeutic effect on patients with gastric cancer, but their long-term administration can induce cancer in the normal population. Although microRNA analysis showed the potential role of MALAT1 in gastric cancer pathogenesis, there are presently few studies available on this subject. More future studies may help in finding novel therapies or new biomarkers. Moreover, gene-disease network showed many mechanisms and genes being shared in gastric cancer and other diseases. Therefore, a drug or therapeutic approach might be useful for more than one of these diseases due to their similar pathways. Through these analyses, a new window is opened to achieve a better understanding of gastric cancer as one of the most complicated tumors. Further research in this field can deepen our knowledge and drive advancement in developing novel therapeutic approaches for gastric cancer as well as some other related diseases.

Data availability
The dataset used in this study was obtained from a publicly available repository, gene expression omnibus (GEO) at https:// www. ncbi. nlm. nih. gov/ geo/. All primary expression files are deposited in the CEL files.