According to data published in 2021, gastric cancer (GC), among all cancers, ranked fourth in cancer-related deaths (Sung et al., 2021). Stomach adenocarcinoma (STAD), the most common histological type, accounts for more than 90% of GC (Ajani et al., 2017). Although endoscopy or histological detection has developed a lot in recent years, the majority of GC patients are diagnosed at their late and advanced stage due to an insidious onset, resulting in high morbidity and mortality (Chen et al., 2020; Wang W. et al., 2020). However, advances in molecular biology techniques allow us to approach precise molecular mechanisms of carcinogenesis and enable us to find potential diagnostic and prognostic biomarkers for GC.

Previous bioinformatic studies resulted in different biomarkers due to different screening criteria and different datasets from Gene Expression Omnibus (GEO) (Sun et al., 2017; Zheng H.-C. et al., 2017; Shi and Zhang, 2019). In the present study, we identified DEGs based on three datasets from GEO, GSE19826, GSE54129, and GSE118916. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis were performed subsequently. Afterwards, we constructed the protein–protein interaction (PPI) network to identify hub genes using the STRING database and Cytoscape software. Then, we performed the survival analysis, including overall survival (OS), disease-free survival (DSS), and progress-free interval (PFI) to identify candidate genes. The expression of candidate genes and their correlation with the pathological stage were further analyzed along with the diagnostic value. A total of four genes were identified as potential biomarkers for GC in our study.

GC is one of the most diagnosed cancers and has brought great burden to global health. Patients were likely to be diagnosed in their late stage due to the lack of specific clinical symptoms at an early stage. Thus, patients with GC have poor prognosis. It is urgent to identify relevant biomarkers that are valid for both diagnostic and prognostic evaluation. Bioinformatics analysis enables us to explore the genetic alterations in GC and has been proved to be a useful approach to identify new biomarkers in plenty of diseases. An initial objective of the project was to identify appropriate biomarkers of GC using bioinformatics analysis.

In the current study, we identified 607 DEGs meeting the criteria. GO enrichment suggested those genes were significantly associated with extracellular structure and matrix organization, collagen fibril organization, and ECM and integrin binding. KEGG was mainly enriched in protein digestion and absorption, ECM-receptor interaction, and focal adhesion. In accordance with the present results, previous studies have reported that cancer-associated fibroblasts are essential in creating extracellular matrix structure and metabolism and account for the adaptive resistance to chemotherapy caused by immune reprogramming of the tumor microenvironment (Quante et al., 2011; Kalluri, 2016). Extracellular matrix plays a significant part in the creation of tumor microenvironment and promotes malignancy (Madsen and Sidenius, 2008; Najafi et al., 2019; Mohan et al., 2020; Piersma et al., 2020; Wang W. et al., 2020). Integrins coordinate ECM–cell and cell–cell interactions, signal transmission, gene expression, and cell function. The interaction between integrin and the cancer glycol microenvironment plays a significant part in regulating cancer progression (Marsico et al., 2018).

The results of this study showed that COL1A1, COL4A1, COL5A2, P4HA3, and SPARC were associated with the OS of GC. COL5A2 was associated with DSS and PFI. Further analysis of the correlation between these key genes and the pathological stages of GC showed that COL4A1 showed no significance in the correlation analysis to pathological stages. Therefore, we identify four genes as potential biomarkers of GC including COL1A1, COL5A2, P4HA3, and SPARC. Also, the diagnostic value of these genes was confirmed in the following analysis.

COL1A1 is an important member of the type-I collagen family, the main fibrillar collagen and an essential structural component of the ECM (Li J. et al., 2016). Many bioinformatic analyses identified COL1A1 as a biomarker of GC (Wang W. et al., 2020; Wang Y. et al., 2021; Zhao et al., 2021). Abnormal expression of COL1A1 has been reported in several cancers, including hepatocellular carcinoma, ovarian cancer, and colorectal cancer, as well as in GC (Li J. et al., 2016; Zhang et al., 2018; Ma et al., 2019; Li et al., 2020). In vitro, enhanced expression of COL1A1 promotes the invasion and migration of GC cells, while knocking out COL1A1 inhibits the increase in cell metastasis ability (Li et al., 2021). It plays an important role in promoting tumor cell proliferation, migration, invasion, epithelial–mesenchymal transformation (EMT), and chemotherapy resistance (Armstrong et al., 2004; Koenig et al., 2006; Shintani et al., 2008; Yang et al., 2014; Zheng X. et al., 2017; Yamazaki et al., 2018; Shi et al., 2021). ROC analysis showed high diagnostic value of COL1A1 (AUC = 0.916) based on 414 GC samples and 210 normal gastric tissues. This finding is consistent with that of Zhao et al. (2021) (AUC = 0.917) based on 375 GC samples and 32 normal samples (Zhao et al., 2021). The diagnostic and prognostic values of COL1A1 were confirmed with extra data (more samples than others) from the present work.

COL5A2 is a member of the type-V collagen family which is also a significant structural component of the ECM. COL5A2 was reported to promote proliferation and invasion in colon cancer and prostate cancer (Ren et al., 2021; Wang J. et al., 2021). Also, it has a strong correlation to the prognosis of renal cancer and gastric cancer (Ding et al., 2021; Tan et al., 2021). The overexpression of COL5A2 promoted the migration of GC cells in vitro and in vivo, and the knockdown of COL5A2 could significantly decrease the migration of cell (Tan et al., 2021). A previous study had demonstrated that patients with higher COL5A2 levels were more likely to suffer from renal metastasis (AUC = 0.878). Among all those genes we identified as potential biomarkers, COL5A2 was the unique gene that was associated with the OS, DSS, and PFI of GC, which had not been reported in previous studies. The value of AUC in our current study is 0.802 based on data from TCGA and GTEx. Therefore, COL5A2 could serve as a novel biomarker of GC. Also, we would perform biological experiments to verify the result.

Previous research showed that P4HA3 was up-regulated in head and neck squamous cell carcinoma (HNSCC) tissue, and it was demonstrated to promote HNSCC cell proliferation, invasion, and migration in vitro (Wang T. et al., 2020). A recent study showed that the de-regulation of P4HA3 was associated with increased metastasis and poor prognosis of GC (Song et al., 2018). In the present work, the value of AUC of P4HA3 is 0.875, which indicated high value of diagnosis and has not been reported in previous studies. The result suggests that P4HA3 is a potential biomarker of GC.

SPARC is one of the first-known matricellular protein that modulates interactions between cells and the ECM. It has divergent actions due to different categories of tumors. It shows anti-tumor or tumor-promoting effects in different cancers (Tai and Tang, 2008). What is surprising is that previous research results are inconsistent. As Zhang et al. (2012) and Zhang et al. (2014) reported, “SPARC expression is negatively correlated with the clinicopathological factors of gastric cancer and inhibits malignancy of gastric cancer cells,” and they confirmed the anti-tumor activity of SPARC in vivo and in vitro. The anti-tumor activity was also reported by Wang L. et al. (2012) in a clinical trial involving 80 gastric cancer samples and 30 normal samples. On the contrary, the tumor-promoting effect of SPARC was also reported in GC. Over expression of SPARC promoted GC progression, including serosal invasion, lymph node, and distant metastasis, and tended to poor prognosis of patients (Zhao et al., 2010; Sato et al., 2013; Wang et al., 2014). Also, the invasion and proliferation ability was inhibited in SPARC knockdown MGC803 and HGC 27 gastric cancer cell lines, which demonstrated the tumor-promoting activity of SPARC. Increased expression of SPARC in this study corroborates these earlier findings (Li Z. et al., 2016; Liao et al., 2018; Li et al., 2019). ROC analysis showed high diagnostic value of SPARC, and the value of AUC was 0.895 in the current study. Biological experiments in different cell lines and clinical samples are necessary to verify the result.

The correlation between these four genes was analyzed, and we found that all of these genes were highly correlated with each other, which enhanced their possibility as potential biomarkers of GC.

In summary, previous studies have identified COL1A1 as a biomarker for GC diagnosis and prognosis. COL5A2, P4HA3, and SPARC were reported to be associated with poor prognosis (OS and DSS, but not PFI); however, the diagnostic value has not been recognized. In the present study, the prognosis values of COL1A1, COL5A2, P4HA3, and SPARC were confirmed. The ROC analysis showed that they could distinguish between GC samples and normal samples effectively. Thus, we suggest COL1A1, COL5A2, P4HA3, and SPARC as biomarkers for both diagnosis and prognosis of GC. Each of the biomarkers identified in the present work plays a significant role in the ECM, which highlights the importance of the tumor microenvironment in GC. Compared with similar studies, we suggested those genes as both diagnostic and prognostic biomarkers for GC. Nevertheless, the current results are all derived from bioinformatics analysis and are limited by the absence of confirmation. Due to different screening criteria, previous bioinformatics research produced different biomarkers. Many of the biomarkers have been verified, and the combination of those results might be more rigorous. Further clinical experiments are underway to verify their value in GC.