As one of the top five common malignancies, gastric cancer (GC) is the fourth leading cause of cancer-associated death around the globe, according to Global cancer statistics 2020. It accounted for more than one million new cases in 2020 and is responsible for one in every 13 deaths globally (Sung et al., 2021). The incidence and mortality of GC vary among different populations and geographical locations. GC is the leading cause of cancer death in several South Central Asian countries. Incidence rates are highest in Eastern Asia and Eastern Europe, while the incidence rates are generally low in Northern America, Northern Europe, and African regions (Sung et al., 2021). These differences could be attributed to the various environmental risk factors, such as the different prevalence of Helicobacter pylori (H. pylori) infections, alcohol consumptions, tobacco smoking, consumption of preserved salty food and processed meat, ingestion of grilled or barbecued meat and fish, and viral infection (Hooi et al., 2017; Research. WCRFAIfC, 2018; Palrasu et al., 2021; Sung et al., 2021). In addition to these environmental factors, genetic factors were also thought to affect the carcinogenesis of GC, as less than 5% of H. pylori infected hosts will develop GC, and evidence of genetic alterations, such as aberrantly expressed activation-induced cytidine deaminase (AID), has emerged (Nagata et al., 2014; Nakanishi et al., 2021). Significant progress in the diagnosis and treatment of GC has been made, such as development of novel human epidermal growth factor receptor 2 (HER2)-targeted drugs for GC (Zhu et al., 2021), the development of minimally invasive surgery, and endoscopic mucosal resection (EMR) and endoscopic submucosal dissection (ESD) (Roh et al., 2020; Wang et al., 2021). Moreover, several biomarkers have been found for the target therapy of GC, such as programmed death 1 (PD-1), HER2, and MNNG HOS transforming gene (MET) (Choi et al., 2022). PD-1 is an inhibitory checkpoint receptor protein expressed on cytotoxic T cells and other immune cells (Pardoll, 2012). Some tumor cells express high levels of PD-L1 to evade from immune system, as PD-1/PD-L1 interaction induces cytotoxic T cell inactivation and downregulation of immune responses (Schreiber et al., 2011), and PD-L1 expression was proposed to be a potential biomarker of response to pembrolizumab (Ghidini et al., 2021). HER2 overexpression is particularly important in GC, as targeted therapy trastuzumab has been widely used to treat HER2+ GC (Bang et al., 2010). MET activation triggers a downstream cascade of phosphoinositide 3-kinases (PI3K) and Rat sarcoma virus (RAS) signaling and regulates cell survival and proliferation (Zhang et al., 2018), Thus, the over-activation of MET plays a critical role in cancer development and is frequently identified in various types of tumors, including GC (Ariyawutyakorn et al., 2016). However, the prognosis of GC remains quite unsatisfactory due to its low early diagnosis rate, with a 5-year overall survival (OS) of less than 40% (Tan, 2019). Therefore, exploration of novel biomarkers that are sensitive and specific for early diagnosis, as well as predictors of prognosis and response to potential targeted treatment, is pivotal in the management of GC.

With the development of next-generation sequencing (NGS) and other techniques, the availability of information related to these potential biomarkers and knowledge of the relevant gene expressions available for various tumors have increased significantly (Behjati and Tarpey, 2013; Levy and Myers, 2016; Giunchi et al., 2021; Shirdarreh et al., 2021). As a result, the mechanisms of various cancers and other diseases have become more widely studied based on bioinformatic analysis, a field combining molecular biology and information technology. Bioinformatics methods, such as data-mining, are now commonly used to explore the carcinogenesis at the molecular level, and to explore biomarkers for potential diagnostic markers, prognostic predictors, and therapeutic targets (Li et al., 2018a; Shen et al., 2019; Yang et al., 2019). As several microarray profiling studies have been performed in GC, this study integrating publicly available data of some of these existing studies, to search for the differentially expressed genes (DEGs) and ultimately biomarkers that could show potential diagnostic values, predict prognosis, and those that might become therapeutic targets.

As one of the most common malignant tumors and top leading causes of cancer-related death, the GC is induced by a variety of factors (Nagata et al., 2014; Hooi et al., 2017; Research. WCRFAIfC, 2018; Nakanishi et al., 2021; Palrasu et al., 2021; Sung et al., 2021). In this study, 9 up-regulated hub genes (COL1A2, THBS2, COL11A1, COL8A1, COL10A1, ADAMTS2, CTHRC1, FAP, and WISP1) and 3 down-regulated hub genes (TFF2, GIF, and CYP2C9) were identified through integrating 4 gene expression profiles to screen for DEGs and perform PPI network analysis. Previous studies have already identified a number of potential hub genes in GC by using similar research methods. For example, a bioinformatics analysis by Liu et al. (Liu et al., 2018b) found nine hub genes, including DNA Topoisomerase II Alpha (TOP2A), collagen type I alpha 1 chain (COL1A1), COL1A2, NDC80 Kinetochore Complex Component (NDC80), Collagen Type III Alpha 1 Chain (COL3A1), Cyclin Dependent Kinase Inhibitor 3 (CDKN3), Centrosomal Protein 55 (CEP55), Targeting protein for Xklp2 (TPX2), and TIMP Metallopeptidase Inhibitor 1 (TIMP1), that might be associated with the pathogenesis of GC. In addition, the results of the article by Wang et al. (Wang et al., 2020b) identified that Fibronectin 1 (FN1), COL1A1, Inhibin beta-A (INHBA), and cystatin SN (CST1) might be potential biomarkers and therapeutic targets for GC patients. Moreover, Lu et al. (Lu et al., 2021) found 5 key genes, namely, Hyaluronan-Mediated Motility Receptor (HMMR), Cyclin B1(CCNB1), C-X-C motif chemokine ligand 8 (CXCL8), mitotic arrest deficient 2 like 1 (MAD2L1), and Cyclin A2 (CCNA2), in GC patients with poor prognosis using the datasets from GEO database. Additionally, Liu et al. suggested that the expression levels of (ATPase H+/K + Transporting Subunit Alpha (ATP4A), carbonic anhydrase 9 (CA9), Fibrinogen Alpha Chain (FGA), Aldehyde Dehydrogenase 1 Family Member A1 (ALDH1A1), and Ghrelin And Obestatin Prepropeptide (GHRL) were reduced, whereas those of TIMP1, Secreted Phosphoprotein 1 (SPP1), CXCL8, Thy-1 Cell Surface Antigen (THY1), and COL1A1 were increased in GC. Another study demonstrated that COL1A1, Collagen Type V Alpha 2 Chain (COL5A2), Prolyl 4-Hydroxylase Subunit Alpha 3 (P4HA3), and Secreted Protein Acidic And Cysteine Rich (SPARC) showed vital values in prognosis and diagnosis of GC (Niu et al., 2022). Furthermore, another study conducted by Dalkilic to analyze GC transcriptomic data revealed that Secreted Frizzled Related Protein 2 (SFRP2), Early Growth Response 1 (EGR1), Chitinase 3 Like 1 (CHI3L1), COL8A1, Nuclear Enriched Abundant Transcript 1 (NEAT1), INHBA, CXCL8, and Myosin Light Chain 9 (MYL9) were highly expressed, while expression of Gastrin (GAST), GIF, Gastrokine 1 (GKN1), Gastrokine 2 (GKN2), Secretoglobin Family 2A Member 1 (SCGB2A1), and HRAS-like suppressor 2 (HRASLS2) were downregulated (Dalkilic, 2020). Although a variation of hub genes were found in previous studies, our study added additional evidence of potential hub genes in GC that could be served as important biomarkers.

As expected, GO analysis found that all the hub genes were related to the digestion pathway for the biological process. Collagen catabolic process and extracellular matrix were also involved significantly in the biological process. For cellular components, the extracellular space, and extracellular regions, as well as proteinaceous extracellular matrix, were mostly involved. In regards to molecular functions, extracellular matrix structural constituent, inward rectifier potassium channel activity, and extracellular matrix binding were greatly involved. KEGG analysis identified that the gastric acid secretion pathway was significantly involved, which was unsurprising given that interference with gastric acid secretion may have caused damage to gastric mucosa. In addition, chemical carcinogenesis and ECM-receptor interaction were also found to be involved. Most of the pathways found in GO and KEGG analyses were closely related to either gastric functions or extracellular activities and structures. A summary of these biological functions and pathways that the hub genes were mostly involved in are shown in Table 4.

Further validation of these hub genes found that all up-regulated hub genes were expressed higher in GC tissues, while among the down-regulated hub genes, the expression level of CYP2C9 was not significantly lower in GC samples. Further survival analysis of the up-regulated hub genes found that COL8A1, COL10A1, CTHRC1, and FAP were associated with poor OS. COL8A1 was known to encode the alpha 1 chain of collagen, type VIII, and thus may modulate migration, proliferation, and adherence of various cells (Zhao et al., 2009a). COL8A1 was also proposed to promote the migration of certain cancer cells by mediating ECM-receptor interaction (Peng et al., 2020). Zhao et al. showed that knockdown of COL8A1 induced the inhibition of hepatocellular carcinoma growth and invasion (Zhao et al., 2009a). Previous studies also found that COL8A1 was associated with poor prognosis of GC, consistent with our findings (Wu et al., 2020a; Chen et al., 2020; Wang et al., 2020c; Zhou et al., 2020). COL10A1, which encodes Collagen type X alpha 1, was found to be overexpressed in different types of cancer, such as esophageal cancer and breast cancer, and promoted the malignant progression by upregulating the expression of Prolyl 4-hydroxylase beta polypeptide (P4HB) (Song et al., 2021; Yang et al., 2021). As a member of the collagen family, COL10A1could activate ECM remodeling and the epithelial-to-mesenchymal transition (EMT), vascular endothelial growth factor receptor 3 (VEGFR3) and Wnt signaling pathway, and its aberrant expression might affect the development of cancer (Song et al., 2021). A recent study identified COL10A1 as a potential inducer of EMT (Li et al., 2018b). Silencing of COL10A1 was found to induce inhibition of cell proliferation, migration, and invasion in GC (Li et al., 2020). Consistent with previous studies, we also found that COL10A1 was associated with adverse outcomes in GC (Chen et al., 2021). Collagen triple helix repeat containing-1 (CTHRC1) was known as a cancer-related protein, and overexpression of CTHRC1 was believed to be involved in tumorigenesis, proliferation, invasion, and metastasis in various gastrointestinal malignancies, including gastric cancer, as well as other non-gastrointestinal malignant tumors (Wang et al., 2012b; Mei et al., 2020). A recent in vitro study found that CTHRC1 increased CXCR4 expression through upregulating hypoxia-inducible factor 1-alpha (HIF-1α) expression, leading to the promotion of cell migration and invasion in GC (Ding et al., 2020). Another study found that repression of CTHRC1 protein activity could inhibit cell proliferation, migration, and invasion in GC (Yu et al., 2015). In addition, a previous clinical study also concluded that higher expression of CTHRC1 was associated with worse prognosis (Gu et al., 2014). Moreover, the results of our meta-analysis further confirmed that CTHRC1 overexpression was associated with worse OS. As for FAP, it is one of the active members of the S9b protease family and known to promote EMT of oral squamous cell carcinoma (OSCC) (Wu et al., 2020b; Huang et al., 2021). However, comprehensive research of FAP on GC is limited. Our finding of higher expression of FAP in GC and its association with worse OS suggests that more studies are needed to investigate the mechanism and effects of FAP in GC. It also has to be noted that the verification by Kaplan–Meier plotter only found marginal association with worse OS, and thus more original studies on its prognostic role in GC are also needed.

As CTHRC1 was found to have the highest HR among other up-regulated hub genes, a nomogram was created based on CTHRC1, age, gender, reflux history, Barrett’s esophagus, H. Pylori infection, pathologic stage, histologic grade, and resident tumor. This showed a relatively high accuracy of prediction given its C-index above 0.7. Thus, CTHRCI demonstrated its utility as not only a diagnostic biomarker, but also as one of the predictors in the nomogram developed in this study.

Further analysis of the up-regulated hub genes also identified the potential diagnostic markers found that ADAMTS2, COL10A1, COL11A1, and CTHRC1 as potential diagnostic markers given that their AUC value and 95% CI were all above 0.9. There were a few studies on the prognostic value of ADAMTS2 in GC, but the evidence of its diagnostic value on GC is limited (Jiang et al., 2019; Liang et al., 2020), while previous study showed that high levels of COL10A1 in plasma could provide diagnostic value in GC with AUC of 0.9171 (p = 0.0002) (Necula et al., 2020). As for COL11A1, a previous study suggested it could be used for differentiating the malignant lesions from premalignant tissues in stomach cancer based on 42 tissues samples (Zhao et al., 2009b), and our result further supports their conclusions though more studies are still warranted. Regarding CTHRC1, although its prognostic predicting value was more extensively studied, its diagnostic value has not been fully investigated and our results provide new evidence of its potential use in diagnosis of GC.

To further explore the potential drugs that may target the hub genes that are associated with worse prognosis, analysis of drug interactions was performed. For COL8A1, the compound or drug that showed the highest interaction was not investigated previously on GC. The second strongest interactor was lovastatin, an agonist of Src homology-2 domain-containing protein tyrosine phosphatase-2 (SHP2) that was found to significantly enhance the efficacy of chemotherapy in colon cancer (Wei et al., 2021). Other in vitro studies found that lovastatin inhibited gastric cancer cells (Cheng-Qian et al., 2014; Zhang et al., 2019b). As for COL10A1, although the top interacting compound, 2-amino-4-(2-hydroxy-4-methylphenyl)-5-phenylpyrimidine, was not previously investigated, the second drug on the list was sendanin. This compound was shown to inhibit the cancer cell lines according to previous in vitro study (Kim et al., 1994). In regards to CTHRC1, the first four compounds on the list were not well studied on cancer, while PF2341066, an inhibitor of anaplastic lymphoma kinase and c-Met later named as crizotinib, was previously shown to have antitumor activity of PF-2341066 in experimental models of anaplastic large-cell lymphoma. Its clinical use is increasingly reported in gastric cancer patients (Christensen et al., 2007; Sabree et al., 2018; Hou et al., 2019; Aparicio et al., 2021). Regarding FAP, although the first few drugs on the list were not well studied, kinetin riboside was found to inhibit colon cancer cells or even stimulated apoptosis (Cheong et al., 2009; Dudzik et al., 2011; Rajabi et al., 2012). Despite a number of studies investigating the compounds or drugs found in our analysis, more research studying on their effects on GC patients is still needed.

Further analysis on the relationship between COL8A1, COL10A1, CTHRC1, and FAP, and immune cell infiltrations in GC found that macrophages, neutrophils, and dendritic cells are positively correlated with these genes. Tumor-infiltrating macrophages were known to play a vital role in tumorigenesis by promoting tumor growth, migration, and invasion, as well as suppression of anti-tumor activity and progression (Van Overmeire et al., 2014; Rihawi et al., 2021). Moreover, tumor-associated macrophages (TAMs) were proposed to have the ability of significantly interfering with treatment response to chemotherapy, immune checkpoint inhibitors (ICIs), antiangiogenic drugs, and even radiotherapy or other treatment methods, leading to failure of treatment (Mantovani and Allavena, 2015; Qiu et al., 2018; Rihawi et al., 2021). Previous studies based on gastric cancer samples found increased neutrophil infiltration in GC, which is consistent with our finding (Kim et al., 2017). It was also shown that neutrophils activated by granulocyte-macrophage colony-stimulating factor (GM-CSF) could express CD54 and B7-H4 that are associated with reduced overall survival of GC patients following surgery (Shan et al., 2021). Tumor-associated neutrophils (TANs) were proposed to induce lymphangiogenesis and angiogenesis, and it was shown that local infiltration of certain types of TANs may play a role in the metastasis in GC (Hiramatsu et al., 2018). Infiltration of dendritic cells in gastric cancer and lymph nodes of GC patients are well known (Tsujitani et al., 1992; Tsujitani et al., 1995), and its infiltrating level in GC was also found to be closely correlated with macrophage infiltration (Xiang et al., 2020). One study found that higher level of dendritic cell infiltration was associated with longer OS (Higgins and Thompson, 2002; Hu et al., 2014). Given that our findings of the roles of the tumor-infiltrating immune cells and the selected up-regulated hub genes in the tumor microenvironment, further investigation and more comprehensive studies on the associations of tumor-infiltrating immune cells and these genes in GC are needed.

This bioinformatic study also has some limitations: First, all data were retrieved from online databases; therefore, the results need to be validated with other cohorts and experiments. Second, as this study mainly aimed to explore the potential clinical values of selected hub genes in the diagnosis and therapy of GC, the details of their mechanisms were not comprehensively explored, especially FAP with very limited number of previous studies of its effect on GC. Third, the TIMER database was mainly based on the TCGA database; therefore, the results need to be verified in the future with other cohorts and experiments.

In conclusion, among the nine up-regulated hub genes (COL1A2, THBS2, COL11A1, COL8A1, COL10A1, ADAMTS2, CTHRC1, FAP, and WISP1) and three down-regulated hub genes (TFF2, GIF, and CYP2C9), ADAMTS2, COL10A1, COL11A1, and CTHRC1 have demonstrated potential as diagnostic markers. COL8A1, COL10A1, CTHRC1, and FAP are associated with worse prognosis and could be potential prognostic biomarkers and therapeutic targets. The infiltration of macrophages, neutrophils, and dendritic cells are positively correlated with these COL8A1, COL10A1, CTHRC1, and FAP, suggesting the need for exploration of their roles in the tumor microenvironment of GC. Among these hub genes, CTHRC1 was found to have the highest prognostic accuracy and associated with worst prognosis as compared with other hub genes, and could be the next research hot spot. A nomogram based on CTHRC1 and other clinical predictors might be useful in clinical decision-making.