A novel multi-scale bioinformatics strategy for identifying the molecular mechanism by insighting into ceRNA network integrating WGCNA and meta-analysis: compound kushen injection for treating gastric carcinoma as a proof

predicted targets to systematically identify genes associated with GC progression. Afterwards, we here undertake a systematic study of the molecular mechanism of CKI in the treatment of GC using a network pharmacology analytical model. Drawing on the above research, we conducted a meta-analysis of key targets to verify their expression changes in GC and conduct immune inltration to explore their prognostic impact on GC patients. At last, in order to better analyze and predict the molecular mechanism of CKI on GC, enrichment analysis and molecular docking were exploited to discover the involved pathways and the binding of CKI components to key targets. Figure owchart the in this study.


Abstract Background
Gastric carcinoma (GC) is a severe digestive system tumor with high morbidity and mortality and poor prognosis, of which the novel treatments are urgently needed. Compound kushen injection (CKI), a classic Chinese medicine injection, has been widely used to treat a variety of tumors in clinical for decades. In recent years, a growing number of studies have con rmed that CKI has a favorable therapeutic effect on GC, but there are few reports on the potential molecular mechanism of action.

Methods
Here, using network pharmacology as the core concept, we identi ed the ceRNA network and key targets of CKI in the treatment of GC. In order to further explore the impact of key targets, we conducted a meta-analysis of them and compared the expression differences between GC tissues and normal tissues. Functional analysis was utilized to understand the biological regulation pathways involved in key genes. Moreover, we further detected the signi cance of key genes for the prognosis of GC through survival analysis and immune in ltration analysis. Finally, molecular docking simulation was adopted so as to verify the combination of CKI components and key targets.

Results
Analysis of the ceRNA network of CKI on GC illustrated that the potential molecular mechanism of CKI was possible to regulate PI3K-AKT and Toll-like receptor signaling pathways by intervening hub genes including AKR1B1, MMP2 and PTGERR3.

Conclusions
In conclusion, this study not only partially highlighted the molecular mechanism of CKI in GC therapy but also provided a novel strategy for exploring the effective mechanisms of traditional Chinese medicine formulations.

Background
Gastric carcinoma (GC) is a malignant tumor of the digestive system that seriously threatens human health [1]. According to the statistics of the International Agency for Research on Cancer, there were about 1 million new cases of stomach cancer in the world in 2018, and about 783,000 deaths due to stomach cancer, ranking fth in the incidence of malignant tumors and third in mortality [2]. The morbidity and mortality of GC have declined sharply in some Western countries in the past few decades, while it is still relatively high in East Asia and imposes a substantial medical burden [3,4]. Among the factors that increase the risk of human GC, Helicobacter pylori gastric infection plays a particularly important role, and 75% of GC cases worldwide are caused by Helicobacter pylori infection [5]. Although GC is highly treatable in the early stages, the median survival time of advanced GC is only 9-10 months [6]. Unsatisfactorily, the global 5-year survival rate of patients is still less than 30% [7]. Combining different forms and different drugs of chemotherapy and radiotherapy and surgery are common methods of treating GC [8]. However, because of the internal metastasis and changes of the tumor, the heterogeneity of different patients and the side effects of radiotherapy and chemotherapy, the options of patients in clinical practice are very limited [1].
Traditional Chinese medicine (TCM) as a complementary therapy has gradually entered the world's horizon in recent years. Compound kushen injection (CKI) which is also named compound Kushen injection is composed of Kushen (Radix Sophorae avescentis) and Baituling (Rhizoma Smilacisglabrae) [9]. CKI has been adopted clinically to treat various types of solid tumors, including GC, liver cancer, lung cancer, breast cancer, and other cancer types for decade years [10][11][12]. It is worth noting that CKI can also relieve cancer pain, regulate immunity and improve conventional chemotherapy to reduce tumor e cacy and reduce chemotherapy toxicity [13,14]. The anti-tumor effect of CKI has been con rmed whereas its underlying molecular mechanism remain poorly understood.
Molecular studies have yielded a vast quantity of new information for potential mechanisms for cancer treatment exploitation. Microarray and high-throughput sequencing technologies provide a reliable guarantee for deciphering key genetic or epigenetic changes in carcinogenesis and discovering potential biomarkers for cancer diagnosis, treatment, and prognosis [15]. MicroRNA (miRNA) and long noncoding RNA (lncRNA) are the two most common subtypes of noncoding RNA (ncRNA). Their abnormality leads to the inability of mRNA to be transcribed normally, and may contribute to unrestricted growth and invasion of cancer cells [16,17]. At present, studies have con rmed that the lncRNA-miRNA-mRNA network plays an important role in the occurrence and development of cancer, which may have huge clinical prospects for identifying potential biomarkers and therapeutic targets of various tumors [18,19].
Thus, in this study, we rstly analyzed the microarray dataset in the Gene Expression Omnibus (GEO) database and The Cancer Genome Atlas (TCGA) to nd miRNAs that are differentially expressed in GC compared to normal tissues. Secondly, weighted gene co-expression network analysis (WGCNA) was applied and merged with differentially expressed miRNAs (DEMs) predicted targets to systematically identify genes associated with GC progression. Afterwards, we here undertake a systematic study of the molecular mechanism of CKI in the treatment of GC using a network pharmacology analytical model. Drawing on the above research, we conducted a meta-analysis of key targets to verify their expression changes in GC and conduct immune in ltration to explore their prognostic impact on GC patients. At last, in order to better analyze and predict the molecular mechanism of CKI on GC, enrichment analysis and molecular docking were exploited to discover the involved pathways and the binding of CKI components to key targets. Figure 1 depicts a owchart of the technical strategy used in this study.

Materials And Methods
Construction of CKI ingredient prediction target network a. Candidate compound screening for CKI In order to obtain the active ingredients of CKI, we have conducted a literature research [10,20]. Sixteen active ingredients in CKI were selected for the next study, and the three-dimensional structures of active ingredients were derived through the Pubchem database [21]. b. Prediction of CKI putative target For a more profound and comprehensive study, we input the 3D structure of the chemical composition into the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) [22], Search Tool for Interactions of Chemicals (STITCH) [23], SuperPred [24], SwissTargetPrediction [25] for target prediction. Furthermore, the predicted multiple target information of the compounds and the obtained information were introduced into Cytoscape 3.6.1 [26] (http://www.cytoscape.org/) to obtain an intermolecular interaction network and carry out complex network analyses.
Differentially expressed miRNA analysis and target prediction of GC a. Differential Expression of miRNAs in GC Microarray data on gene expression GSE23739 was downloaded from the GEO database. A total of 80 samples were obtained, including 40 primary tumors and 40 normal samples. After the Raw data has undergone background correction and standardization, the Limma [27] R package was applied to analyze the difference between cancer and normal tissues. The miRNA-seq data in TCGA contains 446 tumor samples and 45 normal samples. In order to verify and get DEMs, the edgeR [28] package was used to analyze the difference between groups.
b. DEMs target genes prediction MiRWalk2.0 [29] is a comprehensive archive that fully integrates the interactions of multiple existing miRNA target prediction databases and provides predictable and experimentally veri ed miRNA target prediction. On the one hand, the interactions between miRNAs-genes were speculated by 12 servers and only those genes projected by more than six of the servers were identi ed as target genes. On the other hand, 5 servers with miRWalk, miRanda, PITA, RNAhybrid and Targetscan were utilized to prognosticate miRNA-lncRNA targets.
Weighted gene co-expression network analysis for GC mRNA a. Data collection and preprocessing The TCGA-STAD RNA-seq data includes 407 samples of its HTSeq -Counts data and related clinical information, which was downloaded in February 2020. After removing samples containing incomplete analytical data and/or other malignancies, 375 samples were retained. Since some genes without signi cant changes in expression between samples we chose the top 5000 genes that are most important for differential expression for the next WGCNA analysis.
b. Weighted gene co-expression network analysis and module preservation The gene co-expression networks were constructed by the WGCNA package [30]. The similarity between gene expression pro les was used to construct a similarity matrix based on pairwise Pearson correlation coe cient matrices. In order to improve co-expression similarity and achieve a scale-free topology,an appropriate soft threshold power β was selected by using the integration function (pickSoftThresshold) in the WGCNA software package [31]. Besides, we have reconstructed the topological overlap matrix by the calculated the Topological Overlap Measure (TOM) which is a robust measure of network interconnectedness [32,33]. At last, the Dynamic Tree-Cut algorithm method was adopted to identify the module of gene co-expression with the maxBlockSize of 6000, minModuleSize of 30 and mergeCutHeight of 0.2.

c. Identi cation of Clinical Signi cant Modules
Module eigengene (ME) means the rst principal component of each gene module and the expression of ME is considered as a representative of all genes in one module. The Module Membership (MM) is the correlation between the ME and the gene expression pro le. Gene Signi cance (GS) is the absolute value of the correlation between a speci c gene and a clinical trait. According to ME, GS, MM, we can associate modules with clinical traits, not only to calculate the correlation between ME and clinical traits, but also to analyze clinically vital modules [30].

Prediction of ceRNA network of YS intervention in GC
In order to systematically describe GC-associated underlying molecular mechanism, a competing endogenous RNA (ceRNA) network was conducted by merging the prediction correlation of DEMs and key modules in WGCNA. The CKI active component predicted target network was combined with the ceRNA network of GC for CKI intervention in GC ceRNA network prediction, while the overlapping proteins in the two networks are likely to be the potential key gene for the treatment of GC by CKI active ingredients. A potential ceRNA network for CKI treatment of gastric cancer was constructed by Cytoscape, and the potential targets for CKI treatment of gastric cancer were systematically analyzed.

Functional enrichment analysis
In order to analyze the enrichment of key proteins, we rst used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) 10.5 (https://string-db.org/) database to construct a protein-protein interaction (PPI) network [34] for key proteins. We performed the Gene Ontology (GO) Functional and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis for the predicted key targets of the CKI compounds applied in GC therapy in order to identify their biological functions. Besides, R package clusterPro ler was used to perform GO and KEGG functional enrichment analysis. Particularly, the function and pathway enrichment analyses of the validated target genes of miRNAs, were used by the DIANA tool which is based on the cooperation of the previously-mentioned database (TarBase v7.0) and the mirPathv3.0 (a miRNA pathway analysis web server deciphering miRNA function with experimental support) [35,36].
Comprehensive meta-analysis of the hub gene a. Data collection A microarray search of hub genes was conducted in the GEO database with the following terms: ("stomach neoplasms"[MeSH Terms] OR gastric cancer[All Fields]) AND "Homo sapiens"[porgn] AND ("gse"[Filter] AND "Expression pro ling by array" [Filter]) and the latest searching time was April 5, 2020. The criteria for inclusion were as follows: (1) patients diagnosed with stomach cancer were investigated; (2) cancerous and noncancerous samples were involved; (3) datasets samples were no less than 20. Additionally, the following conditions caused the exclusion of a study: (1) lack of original data; (2) the patients with stomach cancer were accompanied by other tumors (3) the interventions included surgery, radiotherapy or other cancer treatment.

b. Statistical analysis and comprehensive meta-analysis
The expression pro ling information of the datasets were exploited to calculate mean (M) and standard deviation (SD) of each hub gene in control experimental group. Thereafter, the meta package of R software was brought into play the standardized meta-difference (SMD) and 95% con dential interval (CI) analysis.
Furthermore, in order to determine a reasonable choice of random effects and xed effects models and heterogeneity evaluation, the chi-squared test of Q and the I 2 statistic were calculated.

Survival analysis of hub genes
The correlation between hub gene expression and overall survival was assessed using the Kaplan-Meier estimation method, based on the "survival" package in R. A signi cant difference of survival curves was assessed by a log-rank test. P value less than 0.05 was considered as statistically signi cant.
Immune in ltrates analysis TIMER (https://cistrome.shinyapps.io/timer/) is a database that can comprehensively study the molecular characterization of tumor-immunity interactions. Not only can the association between immune in ltrates and a variety of factors be explored interactively but the dynamic analysis and visualization of these associations can be done with a TIMER. In this study, we evaluated the hub gene expression in GC and its correlation with the abundance of tumor-in ltrating immune cells, via gene modules [37].

Molecular docking simulation
Molecular docking can re ect the binding energetics of drug molecules to protein receptors by calculating the binding a nity between ligands and receptors and the corresponding intermolecular interactions [38,39]. The crystal structure of the key gene was downloaded from the Protein Data Bank (PDB) (https://www.rcsb.org/) database. The 3D protein crystal structure needed to be determined by X-ray crystallography and the crystal resolution was less than 3 Å. The protein receptor and ligand les were pre-processed by AutoDock Tools and then Autodock venue was used for molecular docking [40,41]. In addition, Pymol and Ligplot were used to visualize the results so as to show the intermolecular interaction and docking more clearly [42,43].

CKI-predicted target network
Basic information on the 16 active ingredients in CKI is shown in Table 1. The active compound-predicted target network (Figure 2) consists of 301 nodes (16 compound points and 285 gene points) that constitute 635 active compound-predicted target linkages.

Construction and Screening of WGCNA Key Modules
After normalization, no outlier samples were eliminated in present study. In order to build a scale-free network, the power of β = 6 (scale free R 2 = 0.85) was selected as the soft-thresholding parameter ( Figure 4A and B). A total of 9 modules were identi ed via the average linkage hierarchical clustering. Clinical traits including vital status, new tumor events, cancer status, pathologic T, pathologic N, pathologic M, stage, H pylori, barretts esophagus were selected to calculate the correlation between the module and the Pearson test. Assessed by the Pearson test when P <0.05, the module and clinical characteristics were considered statistically signi cant. As shown in Figure 4C, the blue, turquoise and brown modules were highly correlated with clinical traits and were identi ed as key modules. Figure 4D Table S3 to 5)

Prediction of ceRNA network of CKI intervention in GC
The intersection of the WGCNA key module network and the hub DEMs prediction target constitutes the ceRNA Network of lncRNA-miRNA-mRNA Axis in GC. Furthermore, the ceRNA network and CKI-predicted target network were merged through Cytoscape to obtain prediction of ceRNA network of CKI intervention in GC ( Figure 5A). As is shown in Figure 5B Go and Kegg pathway enrichment analysis A total of 14 putative targets were uploaded to the STRING database to identify the functional partnerships and interactions between them. The key genes and their interacting proteins form the PPI network for functional enrichment analysis (Supplemental Fig.1). For further interpretation of the function of the key gene, KEGG and GO annotations were performed in R software. A total of 127 GO entries were identi ed, including 93 biological process (BP), 25 molecular function (MF), and 9 cellular component (CC) (FDR < 0.01 and P < 0.01). The top ten GO terms were tissue homeostasis, anatomical structure homeostasis, toll-like receptor signaling pathway, bone resorption, intracellular receptor signaling pathway, multicellular organismal homeostasis, integrin complex, protein complex involved in cell adhesion, tissue remodeling, response to ketone (Figure6A-C). The KEGG results demonstrated that 37 entries satisfy FDR < 0.05 and P < 0.05 ( Figure   6D). These targets were signi cantly enriched in many pathways related to cancer and signaling pathways, such as the PI3K-Akt signaling pathway. In addition, there was also Toll-like receptor signaling pathway related to immunity signi cantly enriched (Figure 7).
We also conducted a modular analysis of lncRNA-miRNA-mRNA Axis intervened by CKI in Cytoscape by Mcode. A total of key modules were analyzed including CTSK, ITGB3, PTGER3, hsa-miR-20a-5p and hsa-miR-30a-5p ve targets ( Figure 8A). To gain insights into the pharmacological mechanisms of CKI on GC, we performed KEGG analysis for two key miRNAs. The results illustrated that the validated targets of hsa-miR-20a-5p and hsa-miR-30a-5p were both associated with the pathways closely related to the occurrence and development of cancer, such as Pathways in cancer, Hippo signaling pathway and p53 signaling pathway ( Figure 8B).

Results of meta-analysis
A total of 8 microarrays from the GEO database met the entry criteria. The features of the included GEO datasets are depicted in Table 2. The expression data from the tumor and control groups were collected on the basis of the GEO database. A meta-analysis was conducted on the basis of expression data of the 8 included microarrays. The results (Figure 9) demonstrated that 7 of the key genes were remarkable abnormal regulation in the stomach cancer groups. Given the apparent heterogeneity, a random effects model was the tumor proto-oncogene. A sensitivity analysis was later conducted to explore whether a particular microarray played a vital role in signi cant heterogeneity. No study was found to have played a crucial role in any of the enrolled studies. A funnel plot was generated to estimate publication bias ( Figure 10). The points in the funnel were distributed asymmetrically on both sides of the midline, indicating that the bias was mainly related to publication bias, but there might also be other reasons such as the lack of included literature ( Figure  11). Survival analysis of key genes A Kaplan-Meier curve was later used to identify the effects of the expression of hub genes on survival time. As is shown in Figure 12, AKR1B1 (p=0.000988), AR (p=0.0102), ITGB3 (p=0.0389), MMP2 (p=0.0465), PTGER3 (p=0.0449) and PTGFR (p=0.0439) with the p values were all less than 0.05, thus indicated that these genes may be the key targets affecting the survival of GC patients

Immunohistochemical analysis of hub genes
Through the above analysis, it was found that AKR1B1, MMP2 and PTGER3 were statistically signi cant in GEO chip meta-analysis and prognostic survival analysis. Therefore, we considered these three genes as the hub genes of CKI in the treatment of GC. Immunohistochemical (IHC) images of human stomach tissue samples stained with antibody were also obtained from the human protein atlas (http://www.proteinatlas.org/) [44]. Furthermore, the IHC pro ler Plugin was utilized to automatically score the staining of the sample through the spectral deconvolution method [45]. According to the analysis results, AKR1B1 and MMP2 were negative in normal gastric tissue and low positive in tumor tissue, whereas PTGER3 was low positive in normal tissue and negative in tumor tissue ( Figure 13). These results approved our nding.

Results of immunoin ltration analysis
Analysis using TIMER showed that hub genes was negatively associated with purity, and ADRB2 (cor=-0.275) was most negatively correlated with tumor purity. In addition, the key genes were strongly correlated with macrophages and dendritic cells. Wherein PTGER3 correlation of macrophages (cor=0.637) and CTSK (cor=0.624) for the relevance of dendritic cells was the strongest correlation (Table 3, Figure 14A-C).
Univariate Cox survival analysis showed that among the six types of immune cells, only macrophages were associated with the survival of GC patients, which was an indicator of the survival of GC patients ( Figure   14D). Docking studies were carried out between CKI and hub genes, and the 3D protein structures of PTGFR and PDE1C were not found in the PDB database. The molecular docking results were shown in Table 4. AR, ITGB3, AKR1B1, ADRB2 and PTGER3 were top ve genes of a nities predicted for the interaction between each of the ve protein targets and corresponding CKI components.
Moreover, the results in Figure 15E show that oxymatrine can bind to ITGB3 by forming a hydrophobic interaction with the surrounding residues Glu536, Arg515, Asn508, Phe547, Tyr571 and Lys548. Oxymatrine could form H-bonds with Tyr571 and Ser507.

Page 12/50
The docking results in this study demonstrate that the receptor-ligand interaction between liriodendrin and AKR1B1 involves both hydrophobic interactions and polar interactions. As shown in Figure 15F, their interaction is centered on a stable hydrophobic core consisting of several nonpolar residues in AKR1B1 (Pro24, Gln49, Ala212, Lys21, Val47, Trp20, Cys298, Leu301, Trp219, Pro23 and Asn50). In addition, the hydroxyls within the main chains of Leu300, Ala299, Lys211, Trp20 and Ser22 form ve hydrogen bond contacts with the liriodendrin, which further stabilizes the entire interaction region.

Discussion
Gastric cancer is one of the most common cancers and its mortality rate remains high [46]. Since GC is di cult to detect in the initial stage, the delay in diagnosis always occurs in patients with gastric cancer [47]. Therefore, there is an urgent need to develop new treatments due to the crisis situation of GC with less cure and poor prognosis. CKI is a prescription approved by the Chinese Medicine Administration of China (NMPA). Besides, it has passed the standardized Good Manufacturing Process (GMP) certi cation and is often used clinically to treat gastric cancer [9]. After multiple systematic reviews and meta-analysis studies, it was found that CKI can not only improve the clinical e cacy of gastric cancer patients but also alleviate the adverse effects of radiotherapy and chemotherapy [48][49][50]. Cancer is a complex disease arising from changes in multiple biological networks [51]. Give this, integrated bioinformatics combined network pharmacology strategy were used to reveal the mechanism underlying the effects of CKI on GC. The high-throughput data analysis method was used to nd miRNAs closely related to gastric cancer for target prediction, and intersects with the key modules of WGCNA analysis in TCGA to identify ceRNA networks closely related to GC. Taking network pharmacology as the core concept, nally obtained the key target of CKI on gastric cancer, and 14 intersection genes were identi ed as hub genes. In order to further explore the impact of CKI on gastric cancer, we conducted a meta-analysis of key targets to compare the differential expression of key genes in gastric cancer tissues and normal tissues. Second, we performed functional analysis to understand the biological regulation pathways involved in key genes. In addition, survival analysis and immune in ltration analysis were used to analyze the relationship between key genes and the prognosis of gastric cancer. Finally, molecular docking simulation was used to verify the binding of CKI components to key targets.
Network pharmacology can explain the impact of drugs on the disruptions of biological networks from the perspective of macro or overall regulation, and explain the treatment of diseases from the perspective of multi-component-multi-target-multi-pathway [52]. We discovered through network pharmacology that 14 intersection genes may be the key targets for CKI treatment of GC. After a meta-analysis of the GC gene expression pro le chip, it was found that 7 of these genes, including, AKR1B1, CTSK, MMP2, TLR4, ADRB2, PDE1C and PTGER3 had signi cant differences in gastric cancer tissues. Besides, AKR1B1, MMP2 and PTGER3 were found to be meaningful in the analysis of gastric cancer survival, so the above three genes are considered to be the most signi cant hub genes for CKI to treat gastric cancer and improve the prognosis of GC. AKR1B1 as a common high-expressed gene in cancer, including gastric cancer, may lead to increased proliferation, metastasis and invasion of tumor cells by driving the epithelial-tomesenchymal transition (EMT) [53,54]. Studies have shown that in pancreatic cancer cells, ADRB2 can directly interact with and upregulate AKR1B1, promote cell proliferation and inhibit apoptosis through the ERK1/2 pathway [55]. In addition, the expression of MMP2 in AKR1B1 knockdown cancer cells also decreased signi cantly compared with the control group [54]. MMP2 has been implicated in the development and morphogenesis of tumors [56]. It has been demonstrated that MMP-2 can regulate the degradation of the extracellular matrix (ECM), which plays an important role in cancer development [57]. Previous studies have con rmed that matrine, an important component of CKI, can downregulate the abnormal expression of MMP2 and thus inhibit the invasion and metastasis of tumor cells [58][59][60]. Whole-genome analysis showed that PTGER3 is abnormally low in gastric cancer. PTGER3 can inhibit the secretion of gastric parietal cells and gastric acid [61]. The lack of PTGER3 leads to abnormal secretion of gastrin and gastric acid and accelerates the occurrence of gastric cancer [62].
In addition, PTGER3 can also up-regulate the expression of related MMP2 and AR to promote the proliferation of cancer cells and the deterioration of gastric cancer [63,64]. Although there is no research showing the direct effect of CKI on PTGER3, there have been studies that matrine and oxymatrine can inhibit gastric acid secretion and protect the stomach, so we think this may be also an important way for CKI to treat gastric cancer [65,66].
In this study, we performed a GO enrichment and KEGG pathway analysis to clarify the multiple mechanisms of CKI against GC from a systematic level. The key genes of CKI on GC were enriched in the PI3K-AKT signaling pathway, the Toll-like receptor signaling pathway and other pathways which was indicated by the functional enrichment analysis. With frequent alterations identi ed in GC, the PI3K/AKT pathway is signi cantly involved in gastric carcinogenesis and progression [67]. The PI3K/AKT pathway can be activated by a variety of factors including hormone and ECM pathways, thereby regulating multiple fundamental cellular activities such as cell proliferation, apoptosis and metastasis [68]. As the most common dysregulation pathway in cancer, the PI3K/AKT pathway has received increasing attention due to its potential for target therapy in many malignancies [69]. We proposed that CKI plays a therapeutic role in the treatment of GC that is mediated by PI3K-AKT pathway, and it has been experimentally con rmed earlier. Previous studies have con rmed that the active ingredients of matrine, oxymatrine and sophoridine in CKI can treat various tumors by inhibiting the PI3K-AKT signaling pathway [70][71][72]. Peng etal [73] found that matrine can inhibit the proliferation and metastasis of gastric cancer cell SGC7901 through PI3K/Akt pathway. GC is a progressive process triggered by H. pylori-induced in ammation [74]. The initial recognition of H. pylori involves Toll-like receptors, the central molecule in the host's in ammatory response [75]. Most studies have concluded that TLR4 is the rst innate immune response against H. pylori [76]. The innate immune response of Toll-like receptor signaling pathway in gastric tumor cells against luminal microorganisms may be involved in the destruction of host defenses and cancer progression leading to DNA damage, cell proliferation and apoptosis regulation [77]. A recent study found that matrine can down-regulate the expression of TLR4 and regulate the Toll-like receptor signaling pathway to reduce gastric mucosal damage in rats [78].
From a comprehensive analysis of the above results, we also found that immunization may be an important potential in uencing factor in CKI treatment of GC, so we conducted an analysis of immune invasion of key genes for gastric cancer. Analysis found that the degree of macrophage in ltration affects the prognosis of GC patients and the expression of key genes is positively correlated with the degree of macrophage in ltration. Consistent with this, sophoridine has been proved that can educate tumor-associated macrophages (TAMs) polarize to M1-TAMs and suppressed M2-TAMs polarization through TLR4/IRF3 axis. In addition, it can inhibit the migration ability of macrophages and reshape the immune microenvironment of gastric cancer [79]. Taken together, we propose that CKI can not only directly inhibit the proliferation and metastasis of gastric cancer tumor cells, but also improve the prognosis of cancer patients through immunotherapy.
In addition to key mRNAs, we also found two important miRNAs that may be involved in CKI regulation of gastric cancer through module analysis, which are hsa-miR-20a-5p and hsa-miR-30a-5p. Analysis of miRNA's KEGG pathway demonstrated that most of the target genes are enriched in Hippo signaling pathway and p53 signaling pathway, which are closely related to the proliferation and metastasis of cancer cells. Matrine plays an important role in regulating p53 signaling pathway to inhibit cell proliferation in liver cancer, lung cancer and esophageal cancer [80][81][82]. Besides, matrine can also promote colorectal cancer cell apoptosis via Hippo signaling pathway [83]. It is worth mentioning that the study found sophoridine can signi cantly activate Hippo and p53 signaling pathways and inhibit lung cancer progression and enhance the effects of the anticancer drug cisplatin against lung cancer cells [84]. However, there are few studies on the direct effect of CKI and its active ingredients on miRNA in the treatment of gastric cancer, which needs to be con rmed by experiments in vivo and in vitro.

Conclusions
In conclusion, this study explored the potential molecular mechanism of action of CKI in the treatment of GC. This research established a new multidisciplinary strategy. On the basis of traditional network pharmacology, through the analysis of high-throughput chip data and WGCNA, the potential molecular mechanism of CKI in the treatment of GC was initially obtained. And chip meta-analysis methods and survival analysis were used to verify the expression and prognosis of key genes in GC. Functional enrichment analysis and immune in ltration analysis focus on the functional impact of key genes. Finally, molecular docking was employed to verify the tightness of the binding between the target and the components. By using network pharmacology combined with multiple integrated bioinformatics method framework, we systematically revealed that CKI may be involved in regulating the ceRNA network for the treatment of GC. Among them, mRNA including AKR1B1, MMP2 and PTGER3 and miRNA including hsa-miR-20a-5p and hsa-miR-30a-5p may play an important role in the therapy. Our preliminary conclusion is that CKI can be used for GC therapy by activating signaling pathways such as PI3K/Akt and Toll-like receptor signaling pathways to inhibit cancer cell proliferation and regulate immunity. Based on multidisciplinary approach, this study might provide a new perspective for the profound exploration and provide a reference for multicomponent-multitargetmultipathway clinical research.